kern_descrip.c revision 236911
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 236911 2012-06-11 19:51:27Z pjd $"); 39 40#include "opt_capsicum.h" 41#include "opt_compat.h" 42#include "opt_ddb.h" 43#include "opt_ktrace.h" 44#include "opt_procdesc.h" 45 46#include <sys/param.h> 47#include <sys/systm.h> 48 49#include <sys/capability.h> 50#include <sys/conf.h> 51#include <sys/domain.h> 52#include <sys/fcntl.h> 53#include <sys/file.h> 54#include <sys/filedesc.h> 55#include <sys/filio.h> 56#include <sys/jail.h> 57#include <sys/kernel.h> 58#include <sys/limits.h> 59#include <sys/lock.h> 60#include <sys/malloc.h> 61#include <sys/mman.h> 62#include <sys/mount.h> 63#include <sys/mqueue.h> 64#include <sys/mutex.h> 65#include <sys/namei.h> 66#include <sys/selinfo.h> 67#include <sys/pipe.h> 68#include <sys/priv.h> 69#include <sys/proc.h> 70#include <sys/procdesc.h> 71#include <sys/protosw.h> 72#include <sys/racct.h> 73#include <sys/resourcevar.h> 74#include <sys/signalvar.h> 75#include <sys/socketvar.h> 76#include <sys/stat.h> 77#include <sys/sx.h> 78#include <sys/syscallsubr.h> 79#include <sys/sysctl.h> 80#include <sys/sysproto.h> 81#include <sys/tty.h> 82#include <sys/unistd.h> 83#include <sys/un.h> 84#include <sys/unpcb.h> 85#include <sys/user.h> 86#include <sys/vnode.h> 87#ifdef KTRACE 88#include <sys/ktrace.h> 89#endif 90 91#include <net/vnet.h> 92 93#include <netinet/in.h> 94#include <netinet/in_pcb.h> 95 96#include <security/audit/audit.h> 97 98#include <vm/uma.h> 99#include <vm/vm.h> 100 101#include <ddb/ddb.h> 102 103static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 104static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 105 "file desc to leader structures"); 106static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 107 108MALLOC_DECLARE(M_FADVISE); 109 110static uma_zone_t file_zone; 111 112 113/* Flags for do_dup() */ 114#define DUP_FIXED 0x1 /* Force fixed allocation */ 115#define DUP_FCNTL 0x2 /* fcntl()-style errors */ 116 117static int do_dup(struct thread *td, int flags, int old, int new, 118 register_t *retval); 119static int fd_first_free(struct filedesc *, int, int); 120static int fd_last_used(struct filedesc *, int, int); 121static void fdgrowtable(struct filedesc *, int); 122static void fdunused(struct filedesc *fdp, int fd); 123static void fdused(struct filedesc *fdp, int fd); 124static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif); 125static int fill_socket_info(struct socket *so, struct kinfo_file *kif); 126static int fill_pts_info(struct tty *tp, struct kinfo_file *kif); 127static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif); 128static int fill_procdesc_info(struct procdesc *pdp, 129 struct kinfo_file *kif); 130static int fill_shm_info(struct file *fp, struct kinfo_file *kif); 131 132/* 133 * A process is initially started out with NDFILE descriptors stored within 134 * this structure, selected to be enough for typical applications based on 135 * the historical limit of 20 open files (and the usage of descriptors by 136 * shells). If these descriptors are exhausted, a larger descriptor table 137 * may be allocated, up to a process' resource limit; the internal arrays 138 * are then unused. 139 */ 140#define NDFILE 20 141#define NDSLOTSIZE sizeof(NDSLOTTYPE) 142#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 143#define NDSLOT(x) ((x) / NDENTRIES) 144#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 145#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 146 147/* 148 * Storage required per open file descriptor. 149 */ 150#define OFILESIZE (sizeof(struct file *) + sizeof(char)) 151 152/* 153 * Storage to hold unused ofiles that need to be reclaimed. 154 */ 155struct freetable { 156 struct file **ft_table; 157 SLIST_ENTRY(freetable) ft_next; 158}; 159 160/* 161 * Basic allocation of descriptors: 162 * one of the above, plus arrays for NDFILE descriptors. 163 */ 164struct filedesc0 { 165 struct filedesc fd_fd; 166 /* 167 * ofiles which need to be reclaimed on free. 168 */ 169 SLIST_HEAD(,freetable) fd_free; 170 /* 171 * These arrays are used when the number of open files is 172 * <= NDFILE, and are then pointed to by the pointers above. 173 */ 174 struct file *fd_dfiles[NDFILE]; 175 char fd_dfileflags[NDFILE]; 176 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 177}; 178 179/* 180 * Descriptor management. 181 */ 182volatile int openfiles; /* actual number of open files */ 183struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 184void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 185 186/* A mutex to protect the association between a proc and filedesc. */ 187static struct mtx fdesc_mtx; 188 189/* 190 * Find the first zero bit in the given bitmap, starting at low and not 191 * exceeding size - 1. 192 */ 193static int 194fd_first_free(struct filedesc *fdp, int low, int size) 195{ 196 NDSLOTTYPE *map = fdp->fd_map; 197 NDSLOTTYPE mask; 198 int off, maxoff; 199 200 if (low >= size) 201 return (low); 202 203 off = NDSLOT(low); 204 if (low % NDENTRIES) { 205 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 206 if ((mask &= ~map[off]) != 0UL) 207 return (off * NDENTRIES + ffsl(mask) - 1); 208 ++off; 209 } 210 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 211 if (map[off] != ~0UL) 212 return (off * NDENTRIES + ffsl(~map[off]) - 1); 213 return (size); 214} 215 216/* 217 * Find the highest non-zero bit in the given bitmap, starting at low and 218 * not exceeding size - 1. 219 */ 220static int 221fd_last_used(struct filedesc *fdp, int low, int size) 222{ 223 NDSLOTTYPE *map = fdp->fd_map; 224 NDSLOTTYPE mask; 225 int off, minoff; 226 227 if (low >= size) 228 return (-1); 229 230 off = NDSLOT(size); 231 if (size % NDENTRIES) { 232 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 233 if ((mask &= map[off]) != 0) 234 return (off * NDENTRIES + flsl(mask) - 1); 235 --off; 236 } 237 for (minoff = NDSLOT(low); off >= minoff; --off) 238 if (map[off] != 0) 239 return (off * NDENTRIES + flsl(map[off]) - 1); 240 return (low - 1); 241} 242 243static int 244fdisused(struct filedesc *fdp, int fd) 245{ 246 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 247 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 248 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 249} 250 251/* 252 * Mark a file descriptor as used. 253 */ 254static void 255fdused(struct filedesc *fdp, int fd) 256{ 257 258 FILEDESC_XLOCK_ASSERT(fdp); 259 KASSERT(!fdisused(fdp, fd), 260 ("fd already used")); 261 262 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 263 if (fd > fdp->fd_lastfile) 264 fdp->fd_lastfile = fd; 265 if (fd == fdp->fd_freefile) 266 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 267} 268 269/* 270 * Mark a file descriptor as unused. 271 */ 272static void 273fdunused(struct filedesc *fdp, int fd) 274{ 275 276 FILEDESC_XLOCK_ASSERT(fdp); 277 KASSERT(fdisused(fdp, fd), 278 ("fd is already unused")); 279 KASSERT(fdp->fd_ofiles[fd] == NULL, 280 ("fd is still in use")); 281 282 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 283 if (fd < fdp->fd_freefile) 284 fdp->fd_freefile = fd; 285 if (fd == fdp->fd_lastfile) 286 fdp->fd_lastfile = fd_last_used(fdp, 0, fd); 287} 288 289/* 290 * System calls on descriptors. 291 */ 292#ifndef _SYS_SYSPROTO_H_ 293struct getdtablesize_args { 294 int dummy; 295}; 296#endif 297/* ARGSUSED */ 298int 299sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 300{ 301 struct proc *p = td->td_proc; 302 uint64_t lim; 303 304 PROC_LOCK(p); 305 td->td_retval[0] = 306 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 307 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 308 PROC_UNLOCK(p); 309 if (lim < td->td_retval[0]) 310 td->td_retval[0] = lim; 311 return (0); 312} 313 314/* 315 * Duplicate a file descriptor to a particular value. 316 * 317 * Note: keep in mind that a potential race condition exists when closing 318 * descriptors from a shared descriptor table (via rfork). 319 */ 320#ifndef _SYS_SYSPROTO_H_ 321struct dup2_args { 322 u_int from; 323 u_int to; 324}; 325#endif 326/* ARGSUSED */ 327int 328sys_dup2(struct thread *td, struct dup2_args *uap) 329{ 330 331 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 332 td->td_retval)); 333} 334 335/* 336 * Duplicate a file descriptor. 337 */ 338#ifndef _SYS_SYSPROTO_H_ 339struct dup_args { 340 u_int fd; 341}; 342#endif 343/* ARGSUSED */ 344int 345sys_dup(struct thread *td, struct dup_args *uap) 346{ 347 348 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 349} 350 351/* 352 * The file control system call. 353 */ 354#ifndef _SYS_SYSPROTO_H_ 355struct fcntl_args { 356 int fd; 357 int cmd; 358 long arg; 359}; 360#endif 361/* ARGSUSED */ 362int 363sys_fcntl(struct thread *td, struct fcntl_args *uap) 364{ 365 struct flock fl; 366 struct oflock ofl; 367 intptr_t arg; 368 int error; 369 int cmd; 370 371 error = 0; 372 cmd = uap->cmd; 373 switch (uap->cmd) { 374 case F_OGETLK: 375 case F_OSETLK: 376 case F_OSETLKW: 377 /* 378 * Convert old flock structure to new. 379 */ 380 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 381 fl.l_start = ofl.l_start; 382 fl.l_len = ofl.l_len; 383 fl.l_pid = ofl.l_pid; 384 fl.l_type = ofl.l_type; 385 fl.l_whence = ofl.l_whence; 386 fl.l_sysid = 0; 387 388 switch (uap->cmd) { 389 case F_OGETLK: 390 cmd = F_GETLK; 391 break; 392 case F_OSETLK: 393 cmd = F_SETLK; 394 break; 395 case F_OSETLKW: 396 cmd = F_SETLKW; 397 break; 398 } 399 arg = (intptr_t)&fl; 400 break; 401 case F_GETLK: 402 case F_SETLK: 403 case F_SETLKW: 404 case F_SETLK_REMOTE: 405 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 406 arg = (intptr_t)&fl; 407 break; 408 default: 409 arg = uap->arg; 410 break; 411 } 412 if (error) 413 return (error); 414 error = kern_fcntl(td, uap->fd, cmd, arg); 415 if (error) 416 return (error); 417 if (uap->cmd == F_OGETLK) { 418 ofl.l_start = fl.l_start; 419 ofl.l_len = fl.l_len; 420 ofl.l_pid = fl.l_pid; 421 ofl.l_type = fl.l_type; 422 ofl.l_whence = fl.l_whence; 423 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 424 } else if (uap->cmd == F_GETLK) { 425 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 426 } 427 return (error); 428} 429 430static inline struct file * 431fdtofp(int fd, struct filedesc *fdp) 432{ 433 434 FILEDESC_LOCK_ASSERT(fdp); 435 436 if ((unsigned)fd >= fdp->fd_nfiles) 437 return (NULL); 438 439 return (fdp->fd_ofiles[fd]); 440} 441 442static inline int 443fdunwrap(int fd, cap_rights_t rights, struct filedesc *fdp, struct file **fpp) 444{ 445 446 *fpp = fdtofp(fd, fdp); 447 if (*fpp == NULL) 448 return (EBADF); 449 450#ifdef CAPABILITIES 451 if ((*fpp)->f_type == DTYPE_CAPABILITY) { 452 int err = cap_funwrap(*fpp, rights, fpp); 453 if (err != 0) { 454 *fpp = NULL; 455 return (err); 456 } 457 } 458#endif /* CAPABILITIES */ 459 return (0); 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; 468 struct proc *p; 469 char *pop; 470 struct vnode *vp; 471 int error, flg, tmp; 472 int vfslocked; 473 u_int old, new; 474 uint64_t bsize; 475 476 vfslocked = 0; 477 error = 0; 478 flg = F_POSIX; 479 p = td->td_proc; 480 fdp = p->p_fd; 481 482 switch (cmd) { 483 case F_DUPFD: 484 tmp = arg; 485 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 486 break; 487 488 case F_DUP2FD: 489 tmp = arg; 490 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 491 break; 492 493 case F_GETFD: 494 FILEDESC_SLOCK(fdp); 495 if ((fp = fdtofp(fd, fdp)) == NULL) { 496 FILEDESC_SUNLOCK(fdp); 497 error = EBADF; 498 break; 499 } 500 pop = &fdp->fd_ofileflags[fd]; 501 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 502 FILEDESC_SUNLOCK(fdp); 503 break; 504 505 case F_SETFD: 506 FILEDESC_XLOCK(fdp); 507 if ((fp = fdtofp(fd, fdp)) == NULL) { 508 FILEDESC_XUNLOCK(fdp); 509 error = EBADF; 510 break; 511 } 512 pop = &fdp->fd_ofileflags[fd]; 513 *pop = (*pop &~ UF_EXCLOSE) | 514 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 515 FILEDESC_XUNLOCK(fdp); 516 break; 517 518 case F_GETFL: 519 FILEDESC_SLOCK(fdp); 520 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 521 if (error != 0) { 522 FILEDESC_SUNLOCK(fdp); 523 break; 524 } 525 td->td_retval[0] = OFLAGS(fp->f_flag); 526 FILEDESC_SUNLOCK(fdp); 527 break; 528 529 case F_SETFL: 530 FILEDESC_SLOCK(fdp); 531 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 532 if (error != 0) { 533 FILEDESC_SUNLOCK(fdp); 534 break; 535 } 536 fhold(fp); 537 FILEDESC_SUNLOCK(fdp); 538 do { 539 tmp = flg = fp->f_flag; 540 tmp &= ~FCNTLFLAGS; 541 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 542 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 543 tmp = fp->f_flag & FNONBLOCK; 544 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 545 if (error) { 546 fdrop(fp, td); 547 break; 548 } 549 tmp = fp->f_flag & FASYNC; 550 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 551 if (error == 0) { 552 fdrop(fp, td); 553 break; 554 } 555 atomic_clear_int(&fp->f_flag, FNONBLOCK); 556 tmp = 0; 557 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 558 fdrop(fp, td); 559 break; 560 561 case F_GETOWN: 562 FILEDESC_SLOCK(fdp); 563 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 564 if (error != 0) { 565 FILEDESC_SUNLOCK(fdp); 566 break; 567 } 568 fhold(fp); 569 FILEDESC_SUNLOCK(fdp); 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 FILEDESC_SLOCK(fdp); 578 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 579 if (error != 0) { 580 FILEDESC_SUNLOCK(fdp); 581 break; 582 } 583 fhold(fp); 584 FILEDESC_SUNLOCK(fdp); 585 tmp = arg; 586 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 587 fdrop(fp, td); 588 break; 589 590 case F_SETLK_REMOTE: 591 error = priv_check(td, PRIV_NFS_LOCKD); 592 if (error) 593 return (error); 594 flg = F_REMOTE; 595 goto do_setlk; 596 597 case F_SETLKW: 598 flg |= F_WAIT; 599 /* FALLTHROUGH F_SETLK */ 600 601 case F_SETLK: 602 do_setlk: 603 FILEDESC_SLOCK(fdp); 604 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp); 605 if (error != 0) { 606 FILEDESC_SUNLOCK(fdp); 607 break; 608 } 609 if (fp->f_type != DTYPE_VNODE) { 610 FILEDESC_SUNLOCK(fdp); 611 error = EBADF; 612 break; 613 } 614 flp = (struct flock *)arg; 615 if (flp->l_whence == SEEK_CUR) { 616 if (fp->f_offset < 0 || 617 (flp->l_start > 0 && 618 fp->f_offset > OFF_MAX - flp->l_start)) { 619 FILEDESC_SUNLOCK(fdp); 620 error = EOVERFLOW; 621 break; 622 } 623 flp->l_start += fp->f_offset; 624 } 625 626 /* 627 * VOP_ADVLOCK() may block. 628 */ 629 fhold(fp); 630 FILEDESC_SUNLOCK(fdp); 631 vp = fp->f_vnode; 632 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 633 switch (flp->l_type) { 634 case F_RDLCK: 635 if ((fp->f_flag & FREAD) == 0) { 636 error = EBADF; 637 break; 638 } 639 PROC_LOCK(p->p_leader); 640 p->p_leader->p_flag |= P_ADVLOCK; 641 PROC_UNLOCK(p->p_leader); 642 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 643 flp, flg); 644 break; 645 case F_WRLCK: 646 if ((fp->f_flag & FWRITE) == 0) { 647 error = EBADF; 648 break; 649 } 650 PROC_LOCK(p->p_leader); 651 p->p_leader->p_flag |= P_ADVLOCK; 652 PROC_UNLOCK(p->p_leader); 653 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 654 flp, flg); 655 break; 656 case F_UNLCK: 657 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 658 flp, flg); 659 break; 660 case F_UNLCKSYS: 661 /* 662 * Temporary api for testing remote lock 663 * infrastructure. 664 */ 665 if (flg != F_REMOTE) { 666 error = EINVAL; 667 break; 668 } 669 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 670 F_UNLCKSYS, flp, flg); 671 break; 672 default: 673 error = EINVAL; 674 break; 675 } 676 VFS_UNLOCK_GIANT(vfslocked); 677 vfslocked = 0; 678 /* Check for race with close */ 679 FILEDESC_SLOCK(fdp); 680 if ((unsigned) fd >= fdp->fd_nfiles || 681 fp != fdp->fd_ofiles[fd]) { 682 FILEDESC_SUNLOCK(fdp); 683 flp->l_whence = SEEK_SET; 684 flp->l_start = 0; 685 flp->l_len = 0; 686 flp->l_type = F_UNLCK; 687 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 688 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 689 F_UNLCK, flp, F_POSIX); 690 VFS_UNLOCK_GIANT(vfslocked); 691 vfslocked = 0; 692 } else 693 FILEDESC_SUNLOCK(fdp); 694 fdrop(fp, td); 695 break; 696 697 case F_GETLK: 698 FILEDESC_SLOCK(fdp); 699 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp); 700 if (error != 0) { 701 FILEDESC_SUNLOCK(fdp); 702 break; 703 } 704 if (fp->f_type != DTYPE_VNODE) { 705 FILEDESC_SUNLOCK(fdp); 706 error = EBADF; 707 break; 708 } 709 flp = (struct flock *)arg; 710 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 711 flp->l_type != F_UNLCK) { 712 FILEDESC_SUNLOCK(fdp); 713 error = EINVAL; 714 break; 715 } 716 if (flp->l_whence == SEEK_CUR) { 717 if ((flp->l_start > 0 && 718 fp->f_offset > OFF_MAX - flp->l_start) || 719 (flp->l_start < 0 && 720 fp->f_offset < OFF_MIN - flp->l_start)) { 721 FILEDESC_SUNLOCK(fdp); 722 error = EOVERFLOW; 723 break; 724 } 725 flp->l_start += fp->f_offset; 726 } 727 /* 728 * VOP_ADVLOCK() may block. 729 */ 730 fhold(fp); 731 FILEDESC_SUNLOCK(fdp); 732 vp = fp->f_vnode; 733 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 734 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 735 F_POSIX); 736 VFS_UNLOCK_GIANT(vfslocked); 737 vfslocked = 0; 738 fdrop(fp, td); 739 break; 740 741 case F_RDAHEAD: 742 arg = arg ? 128 * 1024: 0; 743 /* FALLTHROUGH */ 744 case F_READAHEAD: 745 FILEDESC_SLOCK(fdp); 746 if ((fp = fdtofp(fd, fdp)) == NULL) { 747 FILEDESC_SUNLOCK(fdp); 748 error = EBADF; 749 break; 750 } 751 if (fp->f_type != DTYPE_VNODE) { 752 FILEDESC_SUNLOCK(fdp); 753 error = EBADF; 754 break; 755 } 756 fhold(fp); 757 FILEDESC_SUNLOCK(fdp); 758 if (arg != 0) { 759 vp = fp->f_vnode; 760 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 761 error = vn_lock(vp, LK_SHARED); 762 if (error != 0) 763 goto readahead_vnlock_fail; 764 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 765 VOP_UNLOCK(vp, 0); 766 fp->f_seqcount = (arg + bsize - 1) / bsize; 767 do { 768 new = old = fp->f_flag; 769 new |= FRDAHEAD; 770 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 771readahead_vnlock_fail: 772 VFS_UNLOCK_GIANT(vfslocked); 773 vfslocked = 0; 774 } else { 775 do { 776 new = old = fp->f_flag; 777 new &= ~FRDAHEAD; 778 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 779 } 780 fdrop(fp, td); 781 break; 782 783 default: 784 error = EINVAL; 785 break; 786 } 787 VFS_UNLOCK_GIANT(vfslocked); 788 return (error); 789} 790 791/* 792 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 793 */ 794static int 795do_dup(struct thread *td, int flags, int old, int new, 796 register_t *retval) 797{ 798 struct filedesc *fdp; 799 struct proc *p; 800 struct file *fp; 801 struct file *delfp; 802 int error, holdleaders, maxfd; 803 804 p = td->td_proc; 805 fdp = p->p_fd; 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 & DUP_FCNTL ? EINVAL : EBADF); 816 PROC_LOCK(p); 817 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 818 PROC_UNLOCK(p); 819 if (new >= maxfd) 820 return (flags & DUP_FCNTL ? EINVAL : EBADF); 821 822 FILEDESC_XLOCK(fdp); 823 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) { 824 FILEDESC_XUNLOCK(fdp); 825 return (EBADF); 826 } 827 if (flags & DUP_FIXED && old == new) { 828 *retval = new; 829 FILEDESC_XUNLOCK(fdp); 830 return (0); 831 } 832 fp = fdp->fd_ofiles[old]; 833 fhold(fp); 834 835 /* 836 * If the caller specified a file descriptor, make sure the file 837 * table is large enough to hold it, and grab it. Otherwise, just 838 * allocate a new descriptor the usual way. Since the filedesc 839 * lock may be temporarily dropped in the process, we have to look 840 * out for a race. 841 */ 842 if (flags & DUP_FIXED) { 843 if (new >= fdp->fd_nfiles) { 844 /* 845 * The resource limits are here instead of e.g. 846 * fdalloc(), because the file descriptor table may be 847 * shared between processes, so we can't really use 848 * racct_add()/racct_sub(). Instead of counting the 849 * number of actually allocated descriptors, just put 850 * the limit on the size of the file descriptor table. 851 */ 852#ifdef RACCT 853 PROC_LOCK(p); 854 error = racct_set(p, RACCT_NOFILE, new + 1); 855 PROC_UNLOCK(p); 856 if (error != 0) { 857 FILEDESC_XUNLOCK(fdp); 858 fdrop(fp, td); 859 return (EMFILE); 860 } 861#endif 862 fdgrowtable(fdp, new + 1); 863 } 864 if (fdp->fd_ofiles[new] == NULL) 865 fdused(fdp, new); 866 } else { 867 if ((error = fdalloc(td, new, &new)) != 0) { 868 FILEDESC_XUNLOCK(fdp); 869 fdrop(fp, td); 870 return (error); 871 } 872 } 873 874 KASSERT(fp == fdp->fd_ofiles[old], ("old fd has been modified")); 875 KASSERT(old != new, ("new fd is same as old")); 876 877 delfp = fdp->fd_ofiles[new]; 878 879 /* 880 * Duplicate the source descriptor. 881 */ 882 fdp->fd_ofiles[new] = fp; 883 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 884 if (new > fdp->fd_lastfile) 885 fdp->fd_lastfile = new; 886 *retval = new; 887 888 /* 889 * Save info on the descriptor being overwritten. We cannot close 890 * it without introducing an ownership race for the slot, since we 891 * need to drop the filedesc lock to call closef(). 892 * 893 * XXX this duplicates parts of close(). 894 */ 895 holdleaders = 0; 896 if (delfp != NULL && td->td_proc->p_fdtol != NULL) { 897 /* 898 * Ask fdfree() to sleep to ensure that all relevant 899 * process leaders can be traversed in closef(). 900 */ 901 fdp->fd_holdleaderscount++; 902 holdleaders = 1; 903 } 904 905 /* 906 * If we dup'd over a valid file, we now own the reference to it 907 * and must dispose of it using closef() semantics (as if a 908 * close() were performed on it). 909 * 910 * XXX this duplicates parts of close(). 911 */ 912 if (delfp != NULL) { 913 knote_fdclose(td, new); 914 /* 915 * When we're closing an fd with a capability, we need to 916 * notify mqueue if the underlying object is of type mqueue. 917 */ 918 (void)cap_funwrap(delfp, 0, &fp); 919 if (fp->f_type == DTYPE_MQUEUE) 920 mq_fdclose(td, new, fp); 921 FILEDESC_XUNLOCK(fdp); 922 (void) closef(delfp, td); 923 if (holdleaders) { 924 FILEDESC_XLOCK(fdp); 925 fdp->fd_holdleaderscount--; 926 if (fdp->fd_holdleaderscount == 0 && 927 fdp->fd_holdleaderswakeup != 0) { 928 fdp->fd_holdleaderswakeup = 0; 929 wakeup(&fdp->fd_holdleaderscount); 930 } 931 FILEDESC_XUNLOCK(fdp); 932 } 933 } else { 934 FILEDESC_XUNLOCK(fdp); 935 } 936 return (0); 937} 938 939/* 940 * If sigio is on the list associated with a process or process group, 941 * disable signalling from the device, remove sigio from the list and 942 * free sigio. 943 */ 944void 945funsetown(struct sigio **sigiop) 946{ 947 struct sigio *sigio; 948 949 SIGIO_LOCK(); 950 sigio = *sigiop; 951 if (sigio == NULL) { 952 SIGIO_UNLOCK(); 953 return; 954 } 955 *(sigio->sio_myref) = NULL; 956 if ((sigio)->sio_pgid < 0) { 957 struct pgrp *pg = (sigio)->sio_pgrp; 958 PGRP_LOCK(pg); 959 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 960 sigio, sio_pgsigio); 961 PGRP_UNLOCK(pg); 962 } else { 963 struct proc *p = (sigio)->sio_proc; 964 PROC_LOCK(p); 965 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 966 sigio, sio_pgsigio); 967 PROC_UNLOCK(p); 968 } 969 SIGIO_UNLOCK(); 970 crfree(sigio->sio_ucred); 971 free(sigio, M_SIGIO); 972} 973 974/* 975 * Free a list of sigio structures. 976 * We only need to lock the SIGIO_LOCK because we have made ourselves 977 * inaccessible to callers of fsetown and therefore do not need to lock 978 * the proc or pgrp struct for the list manipulation. 979 */ 980void 981funsetownlst(struct sigiolst *sigiolst) 982{ 983 struct proc *p; 984 struct pgrp *pg; 985 struct sigio *sigio; 986 987 sigio = SLIST_FIRST(sigiolst); 988 if (sigio == NULL) 989 return; 990 p = NULL; 991 pg = NULL; 992 993 /* 994 * Every entry of the list should belong 995 * to a single proc or pgrp. 996 */ 997 if (sigio->sio_pgid < 0) { 998 pg = sigio->sio_pgrp; 999 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 1000 } else /* if (sigio->sio_pgid > 0) */ { 1001 p = sigio->sio_proc; 1002 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 1003 } 1004 1005 SIGIO_LOCK(); 1006 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 1007 *(sigio->sio_myref) = NULL; 1008 if (pg != NULL) { 1009 KASSERT(sigio->sio_pgid < 0, 1010 ("Proc sigio in pgrp sigio list")); 1011 KASSERT(sigio->sio_pgrp == pg, 1012 ("Bogus pgrp in sigio list")); 1013 PGRP_LOCK(pg); 1014 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 1015 sio_pgsigio); 1016 PGRP_UNLOCK(pg); 1017 } else /* if (p != NULL) */ { 1018 KASSERT(sigio->sio_pgid > 0, 1019 ("Pgrp sigio in proc sigio list")); 1020 KASSERT(sigio->sio_proc == p, 1021 ("Bogus proc in sigio list")); 1022 PROC_LOCK(p); 1023 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1024 sio_pgsigio); 1025 PROC_UNLOCK(p); 1026 } 1027 SIGIO_UNLOCK(); 1028 crfree(sigio->sio_ucred); 1029 free(sigio, M_SIGIO); 1030 SIGIO_LOCK(); 1031 } 1032 SIGIO_UNLOCK(); 1033} 1034 1035/* 1036 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1037 * 1038 * After permission checking, add a sigio structure to the sigio list for 1039 * the process or process group. 1040 */ 1041int 1042fsetown(pid_t pgid, struct sigio **sigiop) 1043{ 1044 struct proc *proc; 1045 struct pgrp *pgrp; 1046 struct sigio *sigio; 1047 int ret; 1048 1049 if (pgid == 0) { 1050 funsetown(sigiop); 1051 return (0); 1052 } 1053 1054 ret = 0; 1055 1056 /* Allocate and fill in the new sigio out of locks. */ 1057 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1058 sigio->sio_pgid = pgid; 1059 sigio->sio_ucred = crhold(curthread->td_ucred); 1060 sigio->sio_myref = sigiop; 1061 1062 sx_slock(&proctree_lock); 1063 if (pgid > 0) { 1064 proc = pfind(pgid); 1065 if (proc == NULL) { 1066 ret = ESRCH; 1067 goto fail; 1068 } 1069 1070 /* 1071 * Policy - Don't allow a process to FSETOWN a process 1072 * in another session. 1073 * 1074 * Remove this test to allow maximum flexibility or 1075 * restrict FSETOWN to the current process or process 1076 * group for maximum safety. 1077 */ 1078 PROC_UNLOCK(proc); 1079 if (proc->p_session != curthread->td_proc->p_session) { 1080 ret = EPERM; 1081 goto fail; 1082 } 1083 1084 pgrp = NULL; 1085 } else /* if (pgid < 0) */ { 1086 pgrp = pgfind(-pgid); 1087 if (pgrp == NULL) { 1088 ret = ESRCH; 1089 goto fail; 1090 } 1091 PGRP_UNLOCK(pgrp); 1092 1093 /* 1094 * Policy - Don't allow a process to FSETOWN a process 1095 * in another session. 1096 * 1097 * Remove this test to allow maximum flexibility or 1098 * restrict FSETOWN to the current process or process 1099 * group for maximum safety. 1100 */ 1101 if (pgrp->pg_session != curthread->td_proc->p_session) { 1102 ret = EPERM; 1103 goto fail; 1104 } 1105 1106 proc = NULL; 1107 } 1108 funsetown(sigiop); 1109 if (pgid > 0) { 1110 PROC_LOCK(proc); 1111 /* 1112 * Since funsetownlst() is called without the proctree 1113 * locked, we need to check for P_WEXIT. 1114 * XXX: is ESRCH correct? 1115 */ 1116 if ((proc->p_flag & P_WEXIT) != 0) { 1117 PROC_UNLOCK(proc); 1118 ret = ESRCH; 1119 goto fail; 1120 } 1121 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1122 sigio->sio_proc = proc; 1123 PROC_UNLOCK(proc); 1124 } else { 1125 PGRP_LOCK(pgrp); 1126 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1127 sigio->sio_pgrp = pgrp; 1128 PGRP_UNLOCK(pgrp); 1129 } 1130 sx_sunlock(&proctree_lock); 1131 SIGIO_LOCK(); 1132 *sigiop = sigio; 1133 SIGIO_UNLOCK(); 1134 return (0); 1135 1136fail: 1137 sx_sunlock(&proctree_lock); 1138 crfree(sigio->sio_ucred); 1139 free(sigio, M_SIGIO); 1140 return (ret); 1141} 1142 1143/* 1144 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1145 */ 1146pid_t 1147fgetown(sigiop) 1148 struct sigio **sigiop; 1149{ 1150 pid_t pgid; 1151 1152 SIGIO_LOCK(); 1153 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1154 SIGIO_UNLOCK(); 1155 return (pgid); 1156} 1157 1158/* 1159 * Close a file descriptor. 1160 */ 1161#ifndef _SYS_SYSPROTO_H_ 1162struct close_args { 1163 int fd; 1164}; 1165#endif 1166/* ARGSUSED */ 1167int 1168sys_close(td, uap) 1169 struct thread *td; 1170 struct close_args *uap; 1171{ 1172 1173 return (kern_close(td, uap->fd)); 1174} 1175 1176int 1177kern_close(td, fd) 1178 struct thread *td; 1179 int fd; 1180{ 1181 struct filedesc *fdp; 1182 struct file *fp, *fp_object; 1183 int error; 1184 int holdleaders; 1185 1186 error = 0; 1187 holdleaders = 0; 1188 fdp = td->td_proc->p_fd; 1189 1190 AUDIT_SYSCLOSE(td, fd); 1191 1192 FILEDESC_XLOCK(fdp); 1193 if ((unsigned)fd >= fdp->fd_nfiles || 1194 (fp = fdp->fd_ofiles[fd]) == NULL) { 1195 FILEDESC_XUNLOCK(fdp); 1196 return (EBADF); 1197 } 1198 fdp->fd_ofiles[fd] = NULL; 1199 fdp->fd_ofileflags[fd] = 0; 1200 fdunused(fdp, fd); 1201 if (td->td_proc->p_fdtol != NULL) { 1202 /* 1203 * Ask fdfree() to sleep to ensure that all relevant 1204 * process leaders can be traversed in closef(). 1205 */ 1206 fdp->fd_holdleaderscount++; 1207 holdleaders = 1; 1208 } 1209 1210 /* 1211 * We now hold the fp reference that used to be owned by the 1212 * descriptor array. We have to unlock the FILEDESC *AFTER* 1213 * knote_fdclose to prevent a race of the fd getting opened, a knote 1214 * added, and deleteing a knote for the new fd. 1215 */ 1216 knote_fdclose(td, fd); 1217 1218 /* 1219 * When we're closing an fd with a capability, we need to notify 1220 * mqueue if the underlying object is of type mqueue. 1221 */ 1222 (void)cap_funwrap(fp, 0, &fp_object); 1223 if (fp_object->f_type == DTYPE_MQUEUE) 1224 mq_fdclose(td, fd, fp_object); 1225 FILEDESC_XUNLOCK(fdp); 1226 1227 error = closef(fp, td); 1228 if (holdleaders) { 1229 FILEDESC_XLOCK(fdp); 1230 fdp->fd_holdleaderscount--; 1231 if (fdp->fd_holdleaderscount == 0 && 1232 fdp->fd_holdleaderswakeup != 0) { 1233 fdp->fd_holdleaderswakeup = 0; 1234 wakeup(&fdp->fd_holdleaderscount); 1235 } 1236 FILEDESC_XUNLOCK(fdp); 1237 } 1238 return (error); 1239} 1240 1241/* 1242 * Close open file descriptors. 1243 */ 1244#ifndef _SYS_SYSPROTO_H_ 1245struct closefrom_args { 1246 int lowfd; 1247}; 1248#endif 1249/* ARGSUSED */ 1250int 1251sys_closefrom(struct thread *td, struct closefrom_args *uap) 1252{ 1253 struct filedesc *fdp; 1254 int fd; 1255 1256 fdp = td->td_proc->p_fd; 1257 AUDIT_ARG_FD(uap->lowfd); 1258 1259 /* 1260 * Treat negative starting file descriptor values identical to 1261 * closefrom(0) which closes all files. 1262 */ 1263 if (uap->lowfd < 0) 1264 uap->lowfd = 0; 1265 FILEDESC_SLOCK(fdp); 1266 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) { 1267 if (fdp->fd_ofiles[fd] != NULL) { 1268 FILEDESC_SUNLOCK(fdp); 1269 (void)kern_close(td, fd); 1270 FILEDESC_SLOCK(fdp); 1271 } 1272 } 1273 FILEDESC_SUNLOCK(fdp); 1274 return (0); 1275} 1276 1277#if defined(COMPAT_43) 1278/* 1279 * Return status information about a file descriptor. 1280 */ 1281#ifndef _SYS_SYSPROTO_H_ 1282struct ofstat_args { 1283 int fd; 1284 struct ostat *sb; 1285}; 1286#endif 1287/* ARGSUSED */ 1288int 1289ofstat(struct thread *td, struct ofstat_args *uap) 1290{ 1291 struct ostat oub; 1292 struct stat ub; 1293 int error; 1294 1295 error = kern_fstat(td, uap->fd, &ub); 1296 if (error == 0) { 1297 cvtstat(&ub, &oub); 1298 error = copyout(&oub, uap->sb, sizeof(oub)); 1299 } 1300 return (error); 1301} 1302#endif /* COMPAT_43 */ 1303 1304/* 1305 * Return status information about a file descriptor. 1306 */ 1307#ifndef _SYS_SYSPROTO_H_ 1308struct fstat_args { 1309 int fd; 1310 struct stat *sb; 1311}; 1312#endif 1313/* ARGSUSED */ 1314int 1315sys_fstat(struct thread *td, struct fstat_args *uap) 1316{ 1317 struct stat ub; 1318 int error; 1319 1320 error = kern_fstat(td, uap->fd, &ub); 1321 if (error == 0) 1322 error = copyout(&ub, uap->sb, sizeof(ub)); 1323 return (error); 1324} 1325 1326int 1327kern_fstat(struct thread *td, int fd, struct stat *sbp) 1328{ 1329 struct file *fp; 1330 int error; 1331 1332 AUDIT_ARG_FD(fd); 1333 1334 if ((error = fget(td, fd, CAP_FSTAT, &fp)) != 0) 1335 return (error); 1336 1337 AUDIT_ARG_FILE(td->td_proc, fp); 1338 1339 error = fo_stat(fp, sbp, td->td_ucred, td); 1340 fdrop(fp, td); 1341#ifdef KTRACE 1342 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1343 ktrstat(sbp); 1344#endif 1345 return (error); 1346} 1347 1348/* 1349 * Return status information about a file descriptor. 1350 */ 1351#ifndef _SYS_SYSPROTO_H_ 1352struct nfstat_args { 1353 int fd; 1354 struct nstat *sb; 1355}; 1356#endif 1357/* ARGSUSED */ 1358int 1359sys_nfstat(struct thread *td, struct nfstat_args *uap) 1360{ 1361 struct nstat nub; 1362 struct stat ub; 1363 int error; 1364 1365 error = kern_fstat(td, uap->fd, &ub); 1366 if (error == 0) { 1367 cvtnstat(&ub, &nub); 1368 error = copyout(&nub, uap->sb, sizeof(nub)); 1369 } 1370 return (error); 1371} 1372 1373/* 1374 * Return pathconf information about a file descriptor. 1375 */ 1376#ifndef _SYS_SYSPROTO_H_ 1377struct fpathconf_args { 1378 int fd; 1379 int name; 1380}; 1381#endif 1382/* ARGSUSED */ 1383int 1384sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1385{ 1386 struct file *fp; 1387 struct vnode *vp; 1388 int error; 1389 1390 if ((error = fget(td, uap->fd, CAP_FPATHCONF, &fp)) != 0) 1391 return (error); 1392 1393 /* If asynchronous I/O is available, it works for all descriptors. */ 1394 if (uap->name == _PC_ASYNC_IO) { 1395 td->td_retval[0] = async_io_version; 1396 goto out; 1397 } 1398 vp = fp->f_vnode; 1399 if (vp != NULL) { 1400 int vfslocked; 1401 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1402 vn_lock(vp, LK_SHARED | LK_RETRY); 1403 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1404 VOP_UNLOCK(vp, 0); 1405 VFS_UNLOCK_GIANT(vfslocked); 1406 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1407 if (uap->name != _PC_PIPE_BUF) { 1408 error = EINVAL; 1409 } else { 1410 td->td_retval[0] = PIPE_BUF; 1411 error = 0; 1412 } 1413 } else { 1414 error = EOPNOTSUPP; 1415 } 1416out: 1417 fdrop(fp, td); 1418 return (error); 1419} 1420 1421/* 1422 * Grow the file table to accomodate (at least) nfd descriptors. 1423 */ 1424static void 1425fdgrowtable(struct filedesc *fdp, int nfd) 1426{ 1427 struct filedesc0 *fdp0; 1428 struct freetable *fo; 1429 struct file **ntable; 1430 struct file **otable; 1431 char *nfileflags; 1432 int nnfiles, onfiles; 1433 NDSLOTTYPE *nmap; 1434 1435 FILEDESC_XLOCK_ASSERT(fdp); 1436 1437 KASSERT(fdp->fd_nfiles > 0, 1438 ("zero-length file table")); 1439 1440 /* compute the size of the new table */ 1441 onfiles = fdp->fd_nfiles; 1442 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1443 if (nnfiles <= onfiles) 1444 /* the table is already large enough */ 1445 return; 1446 1447 /* allocate a new table and (if required) new bitmaps */ 1448 ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable), 1449 M_FILEDESC, M_ZERO | M_WAITOK); 1450 nfileflags = (char *)&ntable[nnfiles]; 1451 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1452 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, 1453 M_FILEDESC, M_ZERO | M_WAITOK); 1454 else 1455 nmap = NULL; 1456 1457 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1458 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1459 otable = fdp->fd_ofiles; 1460 fdp->fd_ofileflags = nfileflags; 1461 fdp->fd_ofiles = ntable; 1462 /* 1463 * We must preserve ofiles until the process exits because we can't 1464 * be certain that no threads have references to the old table via 1465 * _fget(). 1466 */ 1467 if (onfiles > NDFILE) { 1468 fo = (struct freetable *)&otable[onfiles]; 1469 fdp0 = (struct filedesc0 *)fdp; 1470 fo->ft_table = otable; 1471 SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next); 1472 } 1473 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1474 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1475 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1476 free(fdp->fd_map, M_FILEDESC); 1477 fdp->fd_map = nmap; 1478 } 1479 fdp->fd_nfiles = nnfiles; 1480} 1481 1482/* 1483 * Allocate a file descriptor for the process. 1484 */ 1485int 1486fdalloc(struct thread *td, int minfd, int *result) 1487{ 1488 struct proc *p = td->td_proc; 1489 struct filedesc *fdp = p->p_fd; 1490 int fd = -1, maxfd; 1491#ifdef RACCT 1492 int error; 1493#endif 1494 1495 FILEDESC_XLOCK_ASSERT(fdp); 1496 1497 if (fdp->fd_freefile > minfd) 1498 minfd = fdp->fd_freefile; 1499 1500 PROC_LOCK(p); 1501 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1502 PROC_UNLOCK(p); 1503 1504 /* 1505 * Search the bitmap for a free descriptor. If none is found, try 1506 * to grow the file table. Keep at it until we either get a file 1507 * descriptor or run into process or system limits; fdgrowtable() 1508 * may drop the filedesc lock, so we're in a race. 1509 */ 1510 for (;;) { 1511 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1512 if (fd >= maxfd) 1513 return (EMFILE); 1514 if (fd < fdp->fd_nfiles) 1515 break; 1516#ifdef RACCT 1517 PROC_LOCK(p); 1518 error = racct_set(p, RACCT_NOFILE, min(fdp->fd_nfiles * 2, maxfd)); 1519 PROC_UNLOCK(p); 1520 if (error != 0) 1521 return (EMFILE); 1522#endif 1523 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd)); 1524 } 1525 1526 /* 1527 * Perform some sanity checks, then mark the file descriptor as 1528 * used and return it to the caller. 1529 */ 1530 KASSERT(!fdisused(fdp, fd), 1531 ("fd_first_free() returned non-free descriptor")); 1532 KASSERT(fdp->fd_ofiles[fd] == NULL, ("file descriptor isn't free")); 1533 KASSERT(fdp->fd_ofileflags[fd] == 0, ("file flags are set")); 1534 fdused(fdp, fd); 1535 *result = fd; 1536 return (0); 1537} 1538 1539/* 1540 * Check to see whether n user file descriptors are available to the process 1541 * p. 1542 */ 1543int 1544fdavail(struct thread *td, int n) 1545{ 1546 struct proc *p = td->td_proc; 1547 struct filedesc *fdp = td->td_proc->p_fd; 1548 int i, lim, last; 1549 1550 FILEDESC_LOCK_ASSERT(fdp); 1551 1552 /* 1553 * XXX: This is only called from uipc_usrreq.c:unp_externalize(); 1554 * call racct_add() from there instead of dealing with containers 1555 * here. 1556 */ 1557 PROC_LOCK(p); 1558 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1559 PROC_UNLOCK(p); 1560 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1561 return (1); 1562 last = min(fdp->fd_nfiles, lim); 1563 for (i = fdp->fd_freefile; i < last; i++) { 1564 if (fdp->fd_ofiles[i] == NULL && --n <= 0) 1565 return (1); 1566 } 1567 return (0); 1568} 1569 1570/* 1571 * Create a new open file structure and allocate a file decriptor for the 1572 * process that refers to it. We add one reference to the file for the 1573 * descriptor table and one reference for resultfp. This is to prevent us 1574 * being preempted and the entry in the descriptor table closed after we 1575 * release the FILEDESC lock. 1576 */ 1577int 1578falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1579{ 1580 struct file *fp; 1581 int error, fd; 1582 1583 error = falloc_noinstall(td, &fp); 1584 if (error) 1585 return (error); /* no reference held on error */ 1586 1587 error = finstall(td, fp, &fd, flags); 1588 if (error) { 1589 fdrop(fp, td); /* one reference (fp only) */ 1590 return (error); 1591 } 1592 1593 if (resultfp != NULL) 1594 *resultfp = fp; /* copy out result */ 1595 else 1596 fdrop(fp, td); /* release local reference */ 1597 1598 if (resultfd != NULL) 1599 *resultfd = fd; 1600 1601 return (0); 1602} 1603 1604/* 1605 * Create a new open file structure without allocating a file descriptor. 1606 */ 1607int 1608falloc_noinstall(struct thread *td, struct file **resultfp) 1609{ 1610 struct file *fp; 1611 int maxuserfiles = maxfiles - (maxfiles / 20); 1612 static struct timeval lastfail; 1613 static int curfail; 1614 1615 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1616 1617 if ((openfiles >= maxuserfiles && 1618 priv_check(td, PRIV_MAXFILES) != 0) || 1619 openfiles >= maxfiles) { 1620 if (ppsratecheck(&lastfail, &curfail, 1)) { 1621 printf("kern.maxfiles limit exceeded by uid %i, " 1622 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1623 } 1624 return (ENFILE); 1625 } 1626 atomic_add_int(&openfiles, 1); 1627 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1628 refcount_init(&fp->f_count, 1); 1629 fp->f_cred = crhold(td->td_ucred); 1630 fp->f_ops = &badfileops; 1631 fp->f_data = NULL; 1632 fp->f_vnode = NULL; 1633 *resultfp = fp; 1634 return (0); 1635} 1636 1637/* 1638 * Install a file in a file descriptor table. 1639 */ 1640int 1641finstall(struct thread *td, struct file *fp, int *fd, int flags) 1642{ 1643 struct filedesc *fdp = td->td_proc->p_fd; 1644 int error; 1645 1646 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1647 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1648 1649 FILEDESC_XLOCK(fdp); 1650 if ((error = fdalloc(td, 0, fd))) { 1651 FILEDESC_XUNLOCK(fdp); 1652 return (error); 1653 } 1654 fhold(fp); 1655 fdp->fd_ofiles[*fd] = fp; 1656 if ((flags & O_CLOEXEC) != 0) 1657 fdp->fd_ofileflags[*fd] |= UF_EXCLOSE; 1658 FILEDESC_XUNLOCK(fdp); 1659 return (0); 1660} 1661 1662/* 1663 * Build a new filedesc structure from another. 1664 * Copy the current, root, and jail root vnode references. 1665 */ 1666struct filedesc * 1667fdinit(struct filedesc *fdp) 1668{ 1669 struct filedesc0 *newfdp; 1670 1671 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1672 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1673 if (fdp != NULL) { 1674 FILEDESC_XLOCK(fdp); 1675 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1676 if (newfdp->fd_fd.fd_cdir) 1677 VREF(newfdp->fd_fd.fd_cdir); 1678 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1679 if (newfdp->fd_fd.fd_rdir) 1680 VREF(newfdp->fd_fd.fd_rdir); 1681 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1682 if (newfdp->fd_fd.fd_jdir) 1683 VREF(newfdp->fd_fd.fd_jdir); 1684 FILEDESC_XUNLOCK(fdp); 1685 } 1686 1687 /* Create the file descriptor table. */ 1688 newfdp->fd_fd.fd_refcnt = 1; 1689 newfdp->fd_fd.fd_holdcnt = 1; 1690 newfdp->fd_fd.fd_cmask = CMASK; 1691 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1692 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1693 newfdp->fd_fd.fd_nfiles = NDFILE; 1694 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1695 newfdp->fd_fd.fd_lastfile = -1; 1696 return (&newfdp->fd_fd); 1697} 1698 1699static struct filedesc * 1700fdhold(struct proc *p) 1701{ 1702 struct filedesc *fdp; 1703 1704 mtx_lock(&fdesc_mtx); 1705 fdp = p->p_fd; 1706 if (fdp != NULL) 1707 fdp->fd_holdcnt++; 1708 mtx_unlock(&fdesc_mtx); 1709 return (fdp); 1710} 1711 1712static void 1713fddrop(struct filedesc *fdp) 1714{ 1715 struct filedesc0 *fdp0; 1716 struct freetable *ft; 1717 int i; 1718 1719 mtx_lock(&fdesc_mtx); 1720 i = --fdp->fd_holdcnt; 1721 mtx_unlock(&fdesc_mtx); 1722 if (i > 0) 1723 return; 1724 1725 FILEDESC_LOCK_DESTROY(fdp); 1726 fdp0 = (struct filedesc0 *)fdp; 1727 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1728 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1729 free(ft->ft_table, M_FILEDESC); 1730 } 1731 free(fdp, M_FILEDESC); 1732} 1733 1734/* 1735 * Share a filedesc structure. 1736 */ 1737struct filedesc * 1738fdshare(struct filedesc *fdp) 1739{ 1740 1741 FILEDESC_XLOCK(fdp); 1742 fdp->fd_refcnt++; 1743 FILEDESC_XUNLOCK(fdp); 1744 return (fdp); 1745} 1746 1747/* 1748 * Unshare a filedesc structure, if necessary by making a copy 1749 */ 1750void 1751fdunshare(struct proc *p, struct thread *td) 1752{ 1753 1754 FILEDESC_XLOCK(p->p_fd); 1755 if (p->p_fd->fd_refcnt > 1) { 1756 struct filedesc *tmp; 1757 1758 FILEDESC_XUNLOCK(p->p_fd); 1759 tmp = fdcopy(p->p_fd); 1760 fdfree(td); 1761 p->p_fd = tmp; 1762 } else 1763 FILEDESC_XUNLOCK(p->p_fd); 1764} 1765 1766/* 1767 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1768 * this is to ease callers, not catch errors. 1769 */ 1770struct filedesc * 1771fdcopy(struct filedesc *fdp) 1772{ 1773 struct filedesc *newfdp; 1774 int i; 1775 1776 /* Certain daemons might not have file descriptors. */ 1777 if (fdp == NULL) 1778 return (NULL); 1779 1780 newfdp = fdinit(fdp); 1781 FILEDESC_SLOCK(fdp); 1782 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1783 FILEDESC_SUNLOCK(fdp); 1784 FILEDESC_XLOCK(newfdp); 1785 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1786 FILEDESC_XUNLOCK(newfdp); 1787 FILEDESC_SLOCK(fdp); 1788 } 1789 /* copy all passable descriptors (i.e. not kqueue) */ 1790 newfdp->fd_freefile = -1; 1791 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1792 if (fdisused(fdp, i) && 1793 (fdp->fd_ofiles[i]->f_ops->fo_flags & DFLAG_PASSABLE) && 1794 fdp->fd_ofiles[i]->f_ops != &badfileops) { 1795 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1796 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1797 fhold(newfdp->fd_ofiles[i]); 1798 newfdp->fd_lastfile = i; 1799 } else { 1800 if (newfdp->fd_freefile == -1) 1801 newfdp->fd_freefile = i; 1802 } 1803 } 1804 newfdp->fd_cmask = fdp->fd_cmask; 1805 FILEDESC_SUNLOCK(fdp); 1806 FILEDESC_XLOCK(newfdp); 1807 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1808 if (newfdp->fd_ofiles[i] != NULL) 1809 fdused(newfdp, i); 1810 if (newfdp->fd_freefile == -1) 1811 newfdp->fd_freefile = i; 1812 FILEDESC_XUNLOCK(newfdp); 1813 return (newfdp); 1814} 1815 1816/* 1817 * Release a filedesc structure. 1818 */ 1819void 1820fdfree(struct thread *td) 1821{ 1822 struct filedesc *fdp; 1823 int i, locked; 1824 struct filedesc_to_leader *fdtol; 1825 struct file *fp; 1826 struct vnode *cdir, *jdir, *rdir, *vp; 1827 struct flock lf; 1828 1829 /* Certain daemons might not have file descriptors. */ 1830 fdp = td->td_proc->p_fd; 1831 if (fdp == NULL) 1832 return; 1833 1834#ifdef RACCT 1835 PROC_LOCK(td->td_proc); 1836 racct_set(td->td_proc, RACCT_NOFILE, 0); 1837 PROC_UNLOCK(td->td_proc); 1838#endif 1839 1840 /* Check for special need to clear POSIX style locks */ 1841 fdtol = td->td_proc->p_fdtol; 1842 if (fdtol != NULL) { 1843 FILEDESC_XLOCK(fdp); 1844 KASSERT(fdtol->fdl_refcount > 0, 1845 ("filedesc_to_refcount botch: fdl_refcount=%d", 1846 fdtol->fdl_refcount)); 1847 if (fdtol->fdl_refcount == 1 && 1848 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1849 for (i = 0; i <= fdp->fd_lastfile; i++) { 1850 fp = fdp->fd_ofiles[i]; 1851 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1852 continue; 1853 fhold(fp); 1854 FILEDESC_XUNLOCK(fdp); 1855 lf.l_whence = SEEK_SET; 1856 lf.l_start = 0; 1857 lf.l_len = 0; 1858 lf.l_type = F_UNLCK; 1859 vp = fp->f_vnode; 1860 locked = VFS_LOCK_GIANT(vp->v_mount); 1861 (void) VOP_ADVLOCK(vp, 1862 (caddr_t)td->td_proc-> 1863 p_leader, 1864 F_UNLCK, 1865 &lf, 1866 F_POSIX); 1867 VFS_UNLOCK_GIANT(locked); 1868 FILEDESC_XLOCK(fdp); 1869 fdrop(fp, td); 1870 } 1871 } 1872 retry: 1873 if (fdtol->fdl_refcount == 1) { 1874 if (fdp->fd_holdleaderscount > 0 && 1875 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1876 /* 1877 * close() or do_dup() has cleared a reference 1878 * in a shared file descriptor table. 1879 */ 1880 fdp->fd_holdleaderswakeup = 1; 1881 sx_sleep(&fdp->fd_holdleaderscount, 1882 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1883 goto retry; 1884 } 1885 if (fdtol->fdl_holdcount > 0) { 1886 /* 1887 * Ensure that fdtol->fdl_leader remains 1888 * valid in closef(). 1889 */ 1890 fdtol->fdl_wakeup = 1; 1891 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 1892 "fdlhold", 0); 1893 goto retry; 1894 } 1895 } 1896 fdtol->fdl_refcount--; 1897 if (fdtol->fdl_refcount == 0 && 1898 fdtol->fdl_holdcount == 0) { 1899 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1900 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1901 } else 1902 fdtol = NULL; 1903 td->td_proc->p_fdtol = NULL; 1904 FILEDESC_XUNLOCK(fdp); 1905 if (fdtol != NULL) 1906 free(fdtol, M_FILEDESC_TO_LEADER); 1907 } 1908 FILEDESC_XLOCK(fdp); 1909 i = --fdp->fd_refcnt; 1910 FILEDESC_XUNLOCK(fdp); 1911 if (i > 0) 1912 return; 1913 1914 for (i = 0; i <= fdp->fd_lastfile; i++) { 1915 fp = fdp->fd_ofiles[i]; 1916 if (fp != NULL) { 1917 FILEDESC_XLOCK(fdp); 1918 fdp->fd_ofiles[i] = NULL; 1919 FILEDESC_XUNLOCK(fdp); 1920 (void) closef(fp, td); 1921 } 1922 } 1923 FILEDESC_XLOCK(fdp); 1924 1925 /* XXX This should happen earlier. */ 1926 mtx_lock(&fdesc_mtx); 1927 td->td_proc->p_fd = NULL; 1928 mtx_unlock(&fdesc_mtx); 1929 1930 if (fdp->fd_nfiles > NDFILE) 1931 free(fdp->fd_ofiles, M_FILEDESC); 1932 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1933 free(fdp->fd_map, M_FILEDESC); 1934 1935 fdp->fd_nfiles = 0; 1936 1937 cdir = fdp->fd_cdir; 1938 fdp->fd_cdir = NULL; 1939 rdir = fdp->fd_rdir; 1940 fdp->fd_rdir = NULL; 1941 jdir = fdp->fd_jdir; 1942 fdp->fd_jdir = NULL; 1943 FILEDESC_XUNLOCK(fdp); 1944 1945 if (cdir) { 1946 locked = VFS_LOCK_GIANT(cdir->v_mount); 1947 vrele(cdir); 1948 VFS_UNLOCK_GIANT(locked); 1949 } 1950 if (rdir) { 1951 locked = VFS_LOCK_GIANT(rdir->v_mount); 1952 vrele(rdir); 1953 VFS_UNLOCK_GIANT(locked); 1954 } 1955 if (jdir) { 1956 locked = VFS_LOCK_GIANT(jdir->v_mount); 1957 vrele(jdir); 1958 VFS_UNLOCK_GIANT(locked); 1959 } 1960 1961 fddrop(fdp); 1962} 1963 1964/* 1965 * For setugid programs, we don't want to people to use that setugidness 1966 * to generate error messages which write to a file which otherwise would 1967 * otherwise be off-limits to the process. We check for filesystems where 1968 * the vnode can change out from under us after execve (like [lin]procfs). 1969 * 1970 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1971 * sufficient. We also don't check for setugidness since we know we are. 1972 */ 1973static int 1974is_unsafe(struct file *fp) 1975{ 1976 if (fp->f_type == DTYPE_VNODE) { 1977 struct vnode *vp = fp->f_vnode; 1978 1979 if ((vp->v_vflag & VV_PROCDEP) != 0) 1980 return (1); 1981 } 1982 return (0); 1983} 1984 1985/* 1986 * Make this setguid thing safe, if at all possible. 1987 */ 1988void 1989setugidsafety(struct thread *td) 1990{ 1991 struct filedesc *fdp; 1992 int i; 1993 1994 /* Certain daemons might not have file descriptors. */ 1995 fdp = td->td_proc->p_fd; 1996 if (fdp == NULL) 1997 return; 1998 1999 /* 2000 * Note: fdp->fd_ofiles may be reallocated out from under us while 2001 * we are blocked in a close. Be careful! 2002 */ 2003 FILEDESC_XLOCK(fdp); 2004 for (i = 0; i <= fdp->fd_lastfile; i++) { 2005 if (i > 2) 2006 break; 2007 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 2008 struct file *fp; 2009 2010 knote_fdclose(td, i); 2011 /* 2012 * NULL-out descriptor prior to close to avoid 2013 * a race while close blocks. 2014 */ 2015 fp = fdp->fd_ofiles[i]; 2016 fdp->fd_ofiles[i] = NULL; 2017 fdp->fd_ofileflags[i] = 0; 2018 fdunused(fdp, i); 2019 FILEDESC_XUNLOCK(fdp); 2020 (void) closef(fp, td); 2021 FILEDESC_XLOCK(fdp); 2022 } 2023 } 2024 FILEDESC_XUNLOCK(fdp); 2025} 2026 2027/* 2028 * If a specific file object occupies a specific file descriptor, close the 2029 * file descriptor entry and drop a reference on the file object. This is a 2030 * convenience function to handle a subsequent error in a function that calls 2031 * falloc() that handles the race that another thread might have closed the 2032 * file descriptor out from under the thread creating the file object. 2033 */ 2034void 2035fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2036{ 2037 2038 FILEDESC_XLOCK(fdp); 2039 if (fdp->fd_ofiles[idx] == fp) { 2040 fdp->fd_ofiles[idx] = NULL; 2041 fdunused(fdp, idx); 2042 FILEDESC_XUNLOCK(fdp); 2043 fdrop(fp, td); 2044 } else 2045 FILEDESC_XUNLOCK(fdp); 2046} 2047 2048/* 2049 * Close any files on exec? 2050 */ 2051void 2052fdcloseexec(struct thread *td) 2053{ 2054 struct filedesc *fdp; 2055 int i; 2056 2057 /* Certain daemons might not have file descriptors. */ 2058 fdp = td->td_proc->p_fd; 2059 if (fdp == NULL) 2060 return; 2061 2062 FILEDESC_XLOCK(fdp); 2063 2064 /* 2065 * We cannot cache fd_ofiles or fd_ofileflags since operations 2066 * may block and rip them out from under us. 2067 */ 2068 for (i = 0; i <= fdp->fd_lastfile; i++) { 2069 if (fdp->fd_ofiles[i] != NULL && 2070 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE || 2071 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) { 2072 struct file *fp; 2073 2074 knote_fdclose(td, i); 2075 /* 2076 * NULL-out descriptor prior to close to avoid 2077 * a race while close blocks. 2078 */ 2079 fp = fdp->fd_ofiles[i]; 2080 fdp->fd_ofiles[i] = NULL; 2081 fdp->fd_ofileflags[i] = 0; 2082 fdunused(fdp, i); 2083 if (fp->f_type == DTYPE_MQUEUE) 2084 mq_fdclose(td, i, fp); 2085 FILEDESC_XUNLOCK(fdp); 2086 (void) closef(fp, td); 2087 FILEDESC_XLOCK(fdp); 2088 } 2089 } 2090 FILEDESC_XUNLOCK(fdp); 2091} 2092 2093/* 2094 * It is unsafe for set[ug]id processes to be started with file 2095 * descriptors 0..2 closed, as these descriptors are given implicit 2096 * significance in the Standard C library. fdcheckstd() will create a 2097 * descriptor referencing /dev/null for each of stdin, stdout, and 2098 * stderr that is not already open. 2099 */ 2100int 2101fdcheckstd(struct thread *td) 2102{ 2103 struct filedesc *fdp; 2104 register_t retval, save; 2105 int i, error, devnull; 2106 2107 fdp = td->td_proc->p_fd; 2108 if (fdp == NULL) 2109 return (0); 2110 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2111 devnull = -1; 2112 error = 0; 2113 for (i = 0; i < 3; i++) { 2114 if (fdp->fd_ofiles[i] != NULL) 2115 continue; 2116 if (devnull < 0) { 2117 save = td->td_retval[0]; 2118 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2119 O_RDWR, 0); 2120 devnull = td->td_retval[0]; 2121 td->td_retval[0] = save; 2122 if (error) 2123 break; 2124 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2125 } else { 2126 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2127 if (error != 0) 2128 break; 2129 } 2130 } 2131 return (error); 2132} 2133 2134/* 2135 * Internal form of close. Decrement reference count on file structure. 2136 * Note: td may be NULL when closing a file that was being passed in a 2137 * message. 2138 * 2139 * XXXRW: Giant is not required for the caller, but often will be held; this 2140 * makes it moderately likely the Giant will be recursed in the VFS case. 2141 */ 2142int 2143closef(struct file *fp, struct thread *td) 2144{ 2145 struct vnode *vp; 2146 struct flock lf; 2147 struct filedesc_to_leader *fdtol; 2148 struct filedesc *fdp; 2149 struct file *fp_object; 2150 2151 /* 2152 * POSIX record locking dictates that any close releases ALL 2153 * locks owned by this process. This is handled by setting 2154 * a flag in the unlock to free ONLY locks obeying POSIX 2155 * semantics, and not to free BSD-style file locks. 2156 * If the descriptor was in a message, POSIX-style locks 2157 * aren't passed with the descriptor, and the thread pointer 2158 * will be NULL. Callers should be careful only to pass a 2159 * NULL thread pointer when there really is no owning 2160 * context that might have locks, or the locks will be 2161 * leaked. 2162 * 2163 * If this is a capability, we do lock processing under the underlying 2164 * node, not the capability itself. 2165 */ 2166 (void)cap_funwrap(fp, 0, &fp_object); 2167 if ((fp_object->f_type == DTYPE_VNODE) && (td != NULL)) { 2168 int vfslocked; 2169 2170 vp = fp_object->f_vnode; 2171 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2172 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2173 lf.l_whence = SEEK_SET; 2174 lf.l_start = 0; 2175 lf.l_len = 0; 2176 lf.l_type = F_UNLCK; 2177 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2178 F_UNLCK, &lf, F_POSIX); 2179 } 2180 fdtol = td->td_proc->p_fdtol; 2181 if (fdtol != NULL) { 2182 /* 2183 * Handle special case where file descriptor table is 2184 * shared between multiple process leaders. 2185 */ 2186 fdp = td->td_proc->p_fd; 2187 FILEDESC_XLOCK(fdp); 2188 for (fdtol = fdtol->fdl_next; 2189 fdtol != td->td_proc->p_fdtol; 2190 fdtol = fdtol->fdl_next) { 2191 if ((fdtol->fdl_leader->p_flag & 2192 P_ADVLOCK) == 0) 2193 continue; 2194 fdtol->fdl_holdcount++; 2195 FILEDESC_XUNLOCK(fdp); 2196 lf.l_whence = SEEK_SET; 2197 lf.l_start = 0; 2198 lf.l_len = 0; 2199 lf.l_type = F_UNLCK; 2200 vp = fp_object->f_vnode; 2201 (void) VOP_ADVLOCK(vp, 2202 (caddr_t)fdtol->fdl_leader, 2203 F_UNLCK, &lf, F_POSIX); 2204 FILEDESC_XLOCK(fdp); 2205 fdtol->fdl_holdcount--; 2206 if (fdtol->fdl_holdcount == 0 && 2207 fdtol->fdl_wakeup != 0) { 2208 fdtol->fdl_wakeup = 0; 2209 wakeup(fdtol); 2210 } 2211 } 2212 FILEDESC_XUNLOCK(fdp); 2213 } 2214 VFS_UNLOCK_GIANT(vfslocked); 2215 } 2216 return (fdrop(fp, td)); 2217} 2218 2219/* 2220 * Initialize the file pointer with the specified properties. 2221 * 2222 * The ops are set with release semantics to be certain that the flags, type, 2223 * and data are visible when ops is. This is to prevent ops methods from being 2224 * called with bad data. 2225 */ 2226void 2227finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2228{ 2229 fp->f_data = data; 2230 fp->f_flag = flag; 2231 fp->f_type = type; 2232 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2233} 2234 2235struct file * 2236fget_unlocked(struct filedesc *fdp, int fd) 2237{ 2238 struct file *fp; 2239 u_int count; 2240 2241 if (fd < 0 || fd >= fdp->fd_nfiles) 2242 return (NULL); 2243 /* 2244 * Fetch the descriptor locklessly. We avoid fdrop() races by 2245 * never raising a refcount above 0. To accomplish this we have 2246 * to use a cmpset loop rather than an atomic_add. The descriptor 2247 * must be re-verified once we acquire a reference to be certain 2248 * that the identity is still correct and we did not lose a race 2249 * due to preemption. 2250 */ 2251 for (;;) { 2252 fp = fdp->fd_ofiles[fd]; 2253 if (fp == NULL) 2254 break; 2255 count = fp->f_count; 2256 if (count == 0) 2257 continue; 2258 /* 2259 * Use an acquire barrier to prevent caching of fd_ofiles 2260 * so it is refreshed for verification. 2261 */ 2262 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2263 continue; 2264 if (fp == fdp->fd_ofiles[fd]) 2265 break; 2266 fdrop(fp, curthread); 2267 } 2268 2269 return (fp); 2270} 2271 2272/* 2273 * Extract the file pointer associated with the specified descriptor for the 2274 * current user process. 2275 * 2276 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2277 * returned. 2278 * 2279 * If the FGET_GETCAP flag is set, the capability itself will be returned. 2280 * Calling _fget() with FGET_GETCAP on a non-capability will return EINVAL. 2281 * Otherwise, if the file is a capability, its rights will be checked against 2282 * the capability rights mask, and if successful, the object will be unwrapped. 2283 * 2284 * If an error occured the non-zero error is returned and *fpp is set to 2285 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2286 * responsible for fdrop(). 2287 */ 2288#define FGET_GETCAP 0x00000001 2289static __inline int 2290_fget(struct thread *td, int fd, struct file **fpp, int flags, 2291 cap_rights_t needrights, cap_rights_t *haverightsp, u_char *maxprotp, 2292 int fget_flags) 2293{ 2294 struct filedesc *fdp; 2295 struct file *fp; 2296#ifdef CAPABILITIES 2297 struct file *fp_fromcap; 2298 int error; 2299#endif 2300 2301 *fpp = NULL; 2302 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2303 return (EBADF); 2304 if ((fp = fget_unlocked(fdp, fd)) == NULL) 2305 return (EBADF); 2306 if (fp->f_ops == &badfileops) { 2307 fdrop(fp, td); 2308 return (EBADF); 2309 } 2310 2311#ifdef CAPABILITIES 2312 /* 2313 * If this is a capability, what rights does it have? 2314 */ 2315 if (haverightsp != NULL) { 2316 if (fp->f_type == DTYPE_CAPABILITY) 2317 *haverightsp = cap_rights(fp); 2318 else 2319 *haverightsp = CAP_MASK_VALID; 2320 } 2321 2322 /* 2323 * If a capability has been requested, return the capability directly. 2324 * Otherwise, check capability rights, extract the underlying object, 2325 * and check its access flags. 2326 */ 2327 if (fget_flags & FGET_GETCAP) { 2328 if (fp->f_type != DTYPE_CAPABILITY) { 2329 fdrop(fp, td); 2330 return (EINVAL); 2331 } 2332 } else { 2333 if (maxprotp == NULL) 2334 error = cap_funwrap(fp, needrights, &fp_fromcap); 2335 else 2336 error = cap_funwrap_mmap(fp, needrights, maxprotp, 2337 &fp_fromcap); 2338 if (error) { 2339 fdrop(fp, td); 2340 return (error); 2341 } 2342 2343 /* 2344 * If we've unwrapped a file, drop the original capability 2345 * and hold the new descriptor. fp after this point refers to 2346 * the actual (unwrapped) object, not the capability. 2347 */ 2348 if (fp != fp_fromcap) { 2349 fhold(fp_fromcap); 2350 fdrop(fp, td); 2351 fp = fp_fromcap; 2352 } 2353 } 2354#else /* !CAPABILITIES */ 2355 KASSERT(fp->f_type != DTYPE_CAPABILITY, 2356 ("%s: saw capability", __func__)); 2357 if (maxprotp != NULL) 2358 *maxprotp = VM_PROT_ALL; 2359#endif /* CAPABILITIES */ 2360 2361 /* 2362 * FREAD and FWRITE failure return EBADF as per POSIX. 2363 * 2364 * Only one flag, or 0, may be specified. 2365 */ 2366 if ((flags == FREAD && (fp->f_flag & FREAD) == 0) || 2367 (flags == FWRITE && (fp->f_flag & FWRITE) == 0)) { 2368 fdrop(fp, td); 2369 return (EBADF); 2370 } 2371 *fpp = fp; 2372 return (0); 2373} 2374 2375int 2376fget(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2377{ 2378 2379 return(_fget(td, fd, fpp, 0, rights, NULL, NULL, 0)); 2380} 2381 2382int 2383fget_mmap(struct thread *td, int fd, cap_rights_t rights, u_char *maxprotp, 2384 struct file **fpp) 2385{ 2386 2387 return (_fget(td, fd, fpp, 0, rights, NULL, maxprotp, 0)); 2388} 2389 2390int 2391fget_read(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2392{ 2393 2394 return(_fget(td, fd, fpp, FREAD, rights, NULL, NULL, 0)); 2395} 2396 2397int 2398fget_write(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2399{ 2400 2401 return (_fget(td, fd, fpp, FWRITE, rights, NULL, NULL, 0)); 2402} 2403 2404/* 2405 * Unlike the other fget() calls, which accept and check capability rights 2406 * but never return capabilities, fgetcap() returns the capability but doesn't 2407 * check capability rights. 2408 */ 2409int 2410fgetcap(struct thread *td, int fd, struct file **fpp) 2411{ 2412 2413 return (_fget(td, fd, fpp, 0, 0, NULL, NULL, FGET_GETCAP)); 2414} 2415 2416 2417/* 2418 * Like fget() but loads the underlying vnode, or returns an error if the 2419 * descriptor does not represent a vnode. Note that pipes use vnodes but 2420 * never have VM objects. The returned vnode will be vref()'d. 2421 * 2422 * XXX: what about the unused flags ? 2423 */ 2424static __inline int 2425_fgetvp(struct thread *td, int fd, int flags, cap_rights_t needrights, 2426 cap_rights_t *haverightsp, struct vnode **vpp) 2427{ 2428 struct file *fp; 2429 int error; 2430 2431 *vpp = NULL; 2432 if ((error = _fget(td, fd, &fp, flags, needrights, haverightsp, 2433 NULL, 0)) != 0) 2434 return (error); 2435 if (fp->f_vnode == NULL) { 2436 error = EINVAL; 2437 } else { 2438 *vpp = fp->f_vnode; 2439 vref(*vpp); 2440 } 2441 fdrop(fp, td); 2442 2443 return (error); 2444} 2445 2446int 2447fgetvp(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp) 2448{ 2449 2450 return (_fgetvp(td, fd, 0, rights, NULL, vpp)); 2451} 2452 2453int 2454fgetvp_rights(struct thread *td, int fd, cap_rights_t need, cap_rights_t *have, 2455 struct vnode **vpp) 2456{ 2457 return (_fgetvp(td, fd, 0, need, have, vpp)); 2458} 2459 2460int 2461fgetvp_read(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp) 2462{ 2463 2464 return (_fgetvp(td, fd, FREAD, rights, NULL, vpp)); 2465} 2466 2467#ifdef notyet 2468int 2469fgetvp_write(struct thread *td, int fd, cap_rights_t rights, 2470 struct vnode **vpp) 2471{ 2472 2473 return (_fgetvp(td, fd, FWRITE, rights, NULL, vpp)); 2474} 2475#endif 2476 2477/* 2478 * Like fget() but loads the underlying socket, or returns an error if the 2479 * descriptor does not represent a socket. 2480 * 2481 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2482 * in the future. 2483 * 2484 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2485 * on their file descriptor reference to prevent the socket from being free'd 2486 * during use. 2487 */ 2488int 2489fgetsock(struct thread *td, int fd, cap_rights_t rights, struct socket **spp, 2490 u_int *fflagp) 2491{ 2492 struct file *fp; 2493 int error; 2494 2495 *spp = NULL; 2496 if (fflagp != NULL) 2497 *fflagp = 0; 2498 if ((error = _fget(td, fd, &fp, 0, rights, NULL, NULL, 0)) != 0) 2499 return (error); 2500 if (fp->f_type != DTYPE_SOCKET) { 2501 error = ENOTSOCK; 2502 } else { 2503 *spp = fp->f_data; 2504 if (fflagp) 2505 *fflagp = fp->f_flag; 2506 SOCK_LOCK(*spp); 2507 soref(*spp); 2508 SOCK_UNLOCK(*spp); 2509 } 2510 fdrop(fp, td); 2511 2512 return (error); 2513} 2514 2515/* 2516 * Drop the reference count on the socket and XXX release the SX lock in the 2517 * future. The last reference closes the socket. 2518 * 2519 * Note: fputsock() is deprecated, see comment for fgetsock(). 2520 */ 2521void 2522fputsock(struct socket *so) 2523{ 2524 2525 ACCEPT_LOCK(); 2526 SOCK_LOCK(so); 2527 CURVNET_SET(so->so_vnet); 2528 sorele(so); 2529 CURVNET_RESTORE(); 2530} 2531 2532/* 2533 * Handle the last reference to a file being closed. 2534 * 2535 * No special capability handling here, as the capability's fo_close will run 2536 * instead of the object here, and perform any necessary drop on the object. 2537 */ 2538int 2539_fdrop(struct file *fp, struct thread *td) 2540{ 2541 int error; 2542 2543 error = 0; 2544 if (fp->f_count != 0) 2545 panic("fdrop: count %d", fp->f_count); 2546 if (fp->f_ops != &badfileops) 2547 error = fo_close(fp, td); 2548 atomic_subtract_int(&openfiles, 1); 2549 crfree(fp->f_cred); 2550 free(fp->f_advice, M_FADVISE); 2551 uma_zfree(file_zone, fp); 2552 2553 return (error); 2554} 2555 2556/* 2557 * Apply an advisory lock on a file descriptor. 2558 * 2559 * Just attempt to get a record lock of the requested type on the entire file 2560 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2561 */ 2562#ifndef _SYS_SYSPROTO_H_ 2563struct flock_args { 2564 int fd; 2565 int how; 2566}; 2567#endif 2568/* ARGSUSED */ 2569int 2570sys_flock(struct thread *td, struct flock_args *uap) 2571{ 2572 struct file *fp; 2573 struct vnode *vp; 2574 struct flock lf; 2575 int vfslocked; 2576 int error; 2577 2578 if ((error = fget(td, uap->fd, CAP_FLOCK, &fp)) != 0) 2579 return (error); 2580 if (fp->f_type != DTYPE_VNODE) { 2581 fdrop(fp, td); 2582 return (EOPNOTSUPP); 2583 } 2584 2585 vp = fp->f_vnode; 2586 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2587 lf.l_whence = SEEK_SET; 2588 lf.l_start = 0; 2589 lf.l_len = 0; 2590 if (uap->how & LOCK_UN) { 2591 lf.l_type = F_UNLCK; 2592 atomic_clear_int(&fp->f_flag, FHASLOCK); 2593 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2594 goto done2; 2595 } 2596 if (uap->how & LOCK_EX) 2597 lf.l_type = F_WRLCK; 2598 else if (uap->how & LOCK_SH) 2599 lf.l_type = F_RDLCK; 2600 else { 2601 error = EBADF; 2602 goto done2; 2603 } 2604 atomic_set_int(&fp->f_flag, FHASLOCK); 2605 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2606 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2607done2: 2608 fdrop(fp, td); 2609 VFS_UNLOCK_GIANT(vfslocked); 2610 return (error); 2611} 2612/* 2613 * Duplicate the specified descriptor to a free descriptor. 2614 */ 2615int 2616dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error) 2617{ 2618 struct file *wfp; 2619 struct file *fp; 2620 2621 /* 2622 * If the to-be-dup'd fd number is greater than the allowed number 2623 * of file descriptors, or the fd to be dup'd has already been 2624 * closed, then reject. 2625 */ 2626 FILEDESC_XLOCK(fdp); 2627 if (dfd < 0 || dfd >= fdp->fd_nfiles || 2628 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 2629 FILEDESC_XUNLOCK(fdp); 2630 return (EBADF); 2631 } 2632 2633 /* 2634 * There are two cases of interest here. 2635 * 2636 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2637 * 2638 * For ENXIO steal away the file structure from (dfd) and store it in 2639 * (indx). (dfd) is effectively closed by this operation. 2640 * 2641 * Any other error code is just returned. 2642 */ 2643 switch (error) { 2644 case ENODEV: 2645 /* 2646 * Check that the mode the file is being opened for is a 2647 * subset of the mode of the existing descriptor. 2648 */ 2649 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2650 FILEDESC_XUNLOCK(fdp); 2651 return (EACCES); 2652 } 2653 fp = fdp->fd_ofiles[indx]; 2654 fdp->fd_ofiles[indx] = wfp; 2655 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2656 if (fp == NULL) 2657 fdused(fdp, indx); 2658 fhold(wfp); 2659 FILEDESC_XUNLOCK(fdp); 2660 if (fp != NULL) 2661 /* 2662 * We now own the reference to fp that the ofiles[] 2663 * array used to own. Release it. 2664 */ 2665 fdrop(fp, td); 2666 return (0); 2667 2668 case ENXIO: 2669 /* 2670 * Steal away the file pointer from dfd and stuff it into indx. 2671 */ 2672 fp = fdp->fd_ofiles[indx]; 2673 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2674 fdp->fd_ofiles[dfd] = NULL; 2675 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2676 fdp->fd_ofileflags[dfd] = 0; 2677 fdunused(fdp, dfd); 2678 if (fp == NULL) 2679 fdused(fdp, indx); 2680 FILEDESC_XUNLOCK(fdp); 2681 2682 /* 2683 * We now own the reference to fp that the ofiles[] array 2684 * used to own. Release it. 2685 */ 2686 if (fp != NULL) 2687 fdrop(fp, td); 2688 return (0); 2689 2690 default: 2691 FILEDESC_XUNLOCK(fdp); 2692 return (error); 2693 } 2694 /* NOTREACHED */ 2695} 2696 2697/* 2698 * Scan all active processes and prisons to see if any of them have a current 2699 * or root directory of `olddp'. If so, replace them with the new mount point. 2700 */ 2701void 2702mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2703{ 2704 struct filedesc *fdp; 2705 struct prison *pr; 2706 struct proc *p; 2707 int nrele; 2708 2709 if (vrefcnt(olddp) == 1) 2710 return; 2711 nrele = 0; 2712 sx_slock(&allproc_lock); 2713 FOREACH_PROC_IN_SYSTEM(p) { 2714 fdp = fdhold(p); 2715 if (fdp == NULL) 2716 continue; 2717 FILEDESC_XLOCK(fdp); 2718 if (fdp->fd_cdir == olddp) { 2719 vref(newdp); 2720 fdp->fd_cdir = newdp; 2721 nrele++; 2722 } 2723 if (fdp->fd_rdir == olddp) { 2724 vref(newdp); 2725 fdp->fd_rdir = newdp; 2726 nrele++; 2727 } 2728 if (fdp->fd_jdir == olddp) { 2729 vref(newdp); 2730 fdp->fd_jdir = newdp; 2731 nrele++; 2732 } 2733 FILEDESC_XUNLOCK(fdp); 2734 fddrop(fdp); 2735 } 2736 sx_sunlock(&allproc_lock); 2737 if (rootvnode == olddp) { 2738 vref(newdp); 2739 rootvnode = newdp; 2740 nrele++; 2741 } 2742 mtx_lock(&prison0.pr_mtx); 2743 if (prison0.pr_root == olddp) { 2744 vref(newdp); 2745 prison0.pr_root = newdp; 2746 nrele++; 2747 } 2748 mtx_unlock(&prison0.pr_mtx); 2749 sx_slock(&allprison_lock); 2750 TAILQ_FOREACH(pr, &allprison, pr_list) { 2751 mtx_lock(&pr->pr_mtx); 2752 if (pr->pr_root == olddp) { 2753 vref(newdp); 2754 pr->pr_root = newdp; 2755 nrele++; 2756 } 2757 mtx_unlock(&pr->pr_mtx); 2758 } 2759 sx_sunlock(&allprison_lock); 2760 while (nrele--) 2761 vrele(olddp); 2762} 2763 2764struct filedesc_to_leader * 2765filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2766{ 2767 struct filedesc_to_leader *fdtol; 2768 2769 fdtol = malloc(sizeof(struct filedesc_to_leader), 2770 M_FILEDESC_TO_LEADER, 2771 M_WAITOK); 2772 fdtol->fdl_refcount = 1; 2773 fdtol->fdl_holdcount = 0; 2774 fdtol->fdl_wakeup = 0; 2775 fdtol->fdl_leader = leader; 2776 if (old != NULL) { 2777 FILEDESC_XLOCK(fdp); 2778 fdtol->fdl_next = old->fdl_next; 2779 fdtol->fdl_prev = old; 2780 old->fdl_next = fdtol; 2781 fdtol->fdl_next->fdl_prev = fdtol; 2782 FILEDESC_XUNLOCK(fdp); 2783 } else { 2784 fdtol->fdl_next = fdtol; 2785 fdtol->fdl_prev = fdtol; 2786 } 2787 return (fdtol); 2788} 2789 2790/* 2791 * Get file structures globally. 2792 */ 2793static int 2794sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2795{ 2796 struct xfile xf; 2797 struct filedesc *fdp; 2798 struct file *fp; 2799 struct proc *p; 2800 int error, n; 2801 2802 error = sysctl_wire_old_buffer(req, 0); 2803 if (error != 0) 2804 return (error); 2805 if (req->oldptr == NULL) { 2806 n = 0; 2807 sx_slock(&allproc_lock); 2808 FOREACH_PROC_IN_SYSTEM(p) { 2809 if (p->p_state == PRS_NEW) 2810 continue; 2811 fdp = fdhold(p); 2812 if (fdp == NULL) 2813 continue; 2814 /* overestimates sparse tables. */ 2815 if (fdp->fd_lastfile > 0) 2816 n += fdp->fd_lastfile; 2817 fddrop(fdp); 2818 } 2819 sx_sunlock(&allproc_lock); 2820 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2821 } 2822 error = 0; 2823 bzero(&xf, sizeof(xf)); 2824 xf.xf_size = sizeof(xf); 2825 sx_slock(&allproc_lock); 2826 FOREACH_PROC_IN_SYSTEM(p) { 2827 PROC_LOCK(p); 2828 if (p->p_state == PRS_NEW) { 2829 PROC_UNLOCK(p); 2830 continue; 2831 } 2832 if (p_cansee(req->td, p) != 0) { 2833 PROC_UNLOCK(p); 2834 continue; 2835 } 2836 xf.xf_pid = p->p_pid; 2837 xf.xf_uid = p->p_ucred->cr_uid; 2838 PROC_UNLOCK(p); 2839 fdp = fdhold(p); 2840 if (fdp == NULL) 2841 continue; 2842 FILEDESC_SLOCK(fdp); 2843 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2844 if ((fp = fdp->fd_ofiles[n]) == NULL) 2845 continue; 2846 xf.xf_fd = n; 2847 xf.xf_file = fp; 2848 xf.xf_data = fp->f_data; 2849 xf.xf_vnode = fp->f_vnode; 2850 xf.xf_type = fp->f_type; 2851 xf.xf_count = fp->f_count; 2852 xf.xf_msgcount = 0; 2853 xf.xf_offset = fp->f_offset; 2854 xf.xf_flag = fp->f_flag; 2855 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2856 if (error) 2857 break; 2858 } 2859 FILEDESC_SUNLOCK(fdp); 2860 fddrop(fdp); 2861 if (error) 2862 break; 2863 } 2864 sx_sunlock(&allproc_lock); 2865 return (error); 2866} 2867 2868SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2869 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2870 2871#ifdef KINFO_OFILE_SIZE 2872CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2873#endif 2874 2875#ifdef COMPAT_FREEBSD7 2876static int 2877export_vnode_for_osysctl(struct vnode *vp, int type, 2878 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2879{ 2880 int error; 2881 char *fullpath, *freepath; 2882 int vfslocked; 2883 2884 bzero(kif, sizeof(*kif)); 2885 kif->kf_structsize = sizeof(*kif); 2886 2887 vref(vp); 2888 kif->kf_fd = type; 2889 kif->kf_type = KF_TYPE_VNODE; 2890 /* This function only handles directories. */ 2891 if (vp->v_type != VDIR) { 2892 vrele(vp); 2893 return (ENOTDIR); 2894 } 2895 kif->kf_vnode_type = KF_VTYPE_VDIR; 2896 2897 /* 2898 * This is not a true file descriptor, so we set a bogus refcount 2899 * and offset to indicate these fields should be ignored. 2900 */ 2901 kif->kf_ref_count = -1; 2902 kif->kf_offset = -1; 2903 2904 freepath = NULL; 2905 fullpath = "-"; 2906 FILEDESC_SUNLOCK(fdp); 2907 vn_fullpath(curthread, vp, &fullpath, &freepath); 2908 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2909 vrele(vp); 2910 VFS_UNLOCK_GIANT(vfslocked); 2911 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2912 if (freepath != NULL) 2913 free(freepath, M_TEMP); 2914 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2915 FILEDESC_SLOCK(fdp); 2916 return (error); 2917} 2918 2919/* 2920 * Get per-process file descriptors for use by procstat(1), et al. 2921 */ 2922static int 2923sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 2924{ 2925 char *fullpath, *freepath; 2926 struct kinfo_ofile *kif; 2927 struct filedesc *fdp; 2928 int error, i, *name; 2929 struct shmfd *shmfd; 2930 struct socket *so; 2931 struct vnode *vp; 2932 struct file *fp; 2933 struct proc *p; 2934 struct tty *tp; 2935 int vfslocked; 2936 2937 name = (int *)arg1; 2938 if ((p = pfind((pid_t)name[0])) == NULL) 2939 return (ESRCH); 2940 if ((error = p_candebug(curthread, p))) { 2941 PROC_UNLOCK(p); 2942 return (error); 2943 } 2944 fdp = fdhold(p); 2945 PROC_UNLOCK(p); 2946 if (fdp == NULL) 2947 return (ENOENT); 2948 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 2949 FILEDESC_SLOCK(fdp); 2950 if (fdp->fd_cdir != NULL) 2951 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 2952 fdp, req); 2953 if (fdp->fd_rdir != NULL) 2954 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 2955 fdp, req); 2956 if (fdp->fd_jdir != NULL) 2957 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 2958 fdp, req); 2959 for (i = 0; i < fdp->fd_nfiles; i++) { 2960 if ((fp = fdp->fd_ofiles[i]) == NULL) 2961 continue; 2962 bzero(kif, sizeof(*kif)); 2963 kif->kf_structsize = sizeof(*kif); 2964 vp = NULL; 2965 so = NULL; 2966 tp = NULL; 2967 shmfd = NULL; 2968 kif->kf_fd = i; 2969 2970#ifdef CAPABILITIES 2971 /* 2972 * When reporting a capability, most fields will be from the 2973 * underlying object, but do mark as a capability. With 2974 * ofiledesc, we don't have a field to export the cap_rights_t, 2975 * but we do with the new filedesc. 2976 */ 2977 if (fp->f_type == DTYPE_CAPABILITY) { 2978 kif->kf_flags |= KF_FLAG_CAPABILITY; 2979 (void)cap_funwrap(fp, 0, &fp); 2980 } 2981#else 2982 KASSERT(fp->f_type != DTYPE_CAPABILITY, 2983 ("sysctl_kern_proc_ofiledesc: saw capability")); 2984#endif 2985 switch (fp->f_type) { 2986 case DTYPE_VNODE: 2987 kif->kf_type = KF_TYPE_VNODE; 2988 vp = fp->f_vnode; 2989 break; 2990 2991 case DTYPE_SOCKET: 2992 kif->kf_type = KF_TYPE_SOCKET; 2993 so = fp->f_data; 2994 break; 2995 2996 case DTYPE_PIPE: 2997 kif->kf_type = KF_TYPE_PIPE; 2998 break; 2999 3000 case DTYPE_FIFO: 3001 kif->kf_type = KF_TYPE_FIFO; 3002 vp = fp->f_vnode; 3003 break; 3004 3005 case DTYPE_KQUEUE: 3006 kif->kf_type = KF_TYPE_KQUEUE; 3007 break; 3008 3009 case DTYPE_CRYPTO: 3010 kif->kf_type = KF_TYPE_CRYPTO; 3011 break; 3012 3013 case DTYPE_MQUEUE: 3014 kif->kf_type = KF_TYPE_MQUEUE; 3015 break; 3016 3017 case DTYPE_SHM: 3018 kif->kf_type = KF_TYPE_SHM; 3019 shmfd = fp->f_data; 3020 break; 3021 3022 case DTYPE_SEM: 3023 kif->kf_type = KF_TYPE_SEM; 3024 break; 3025 3026 case DTYPE_PTS: 3027 kif->kf_type = KF_TYPE_PTS; 3028 tp = fp->f_data; 3029 break; 3030 3031#ifdef PROCDESC 3032 case DTYPE_PROCDESC: 3033 kif->kf_type = KF_TYPE_PROCDESC; 3034 break; 3035#endif 3036 3037 default: 3038 kif->kf_type = KF_TYPE_UNKNOWN; 3039 break; 3040 } 3041 kif->kf_ref_count = fp->f_count; 3042 if (fp->f_flag & FREAD) 3043 kif->kf_flags |= KF_FLAG_READ; 3044 if (fp->f_flag & FWRITE) 3045 kif->kf_flags |= KF_FLAG_WRITE; 3046 if (fp->f_flag & FAPPEND) 3047 kif->kf_flags |= KF_FLAG_APPEND; 3048 if (fp->f_flag & FASYNC) 3049 kif->kf_flags |= KF_FLAG_ASYNC; 3050 if (fp->f_flag & FFSYNC) 3051 kif->kf_flags |= KF_FLAG_FSYNC; 3052 if (fp->f_flag & FNONBLOCK) 3053 kif->kf_flags |= KF_FLAG_NONBLOCK; 3054 if (fp->f_flag & O_DIRECT) 3055 kif->kf_flags |= KF_FLAG_DIRECT; 3056 if (fp->f_flag & FHASLOCK) 3057 kif->kf_flags |= KF_FLAG_HASLOCK; 3058 kif->kf_offset = fp->f_offset; 3059 if (vp != NULL) { 3060 vref(vp); 3061 switch (vp->v_type) { 3062 case VNON: 3063 kif->kf_vnode_type = KF_VTYPE_VNON; 3064 break; 3065 case VREG: 3066 kif->kf_vnode_type = KF_VTYPE_VREG; 3067 break; 3068 case VDIR: 3069 kif->kf_vnode_type = KF_VTYPE_VDIR; 3070 break; 3071 case VBLK: 3072 kif->kf_vnode_type = KF_VTYPE_VBLK; 3073 break; 3074 case VCHR: 3075 kif->kf_vnode_type = KF_VTYPE_VCHR; 3076 break; 3077 case VLNK: 3078 kif->kf_vnode_type = KF_VTYPE_VLNK; 3079 break; 3080 case VSOCK: 3081 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3082 break; 3083 case VFIFO: 3084 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3085 break; 3086 case VBAD: 3087 kif->kf_vnode_type = KF_VTYPE_VBAD; 3088 break; 3089 default: 3090 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3091 break; 3092 } 3093 /* 3094 * It is OK to drop the filedesc lock here as we will 3095 * re-validate and re-evaluate its properties when 3096 * the loop continues. 3097 */ 3098 freepath = NULL; 3099 fullpath = "-"; 3100 FILEDESC_SUNLOCK(fdp); 3101 vn_fullpath(curthread, vp, &fullpath, &freepath); 3102 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3103 vrele(vp); 3104 VFS_UNLOCK_GIANT(vfslocked); 3105 strlcpy(kif->kf_path, fullpath, 3106 sizeof(kif->kf_path)); 3107 if (freepath != NULL) 3108 free(freepath, M_TEMP); 3109 FILEDESC_SLOCK(fdp); 3110 } 3111 if (so != NULL) { 3112 struct sockaddr *sa; 3113 3114 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3115 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3116 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3117 free(sa, M_SONAME); 3118 } 3119 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3120 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3121 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3122 free(sa, M_SONAME); 3123 } 3124 kif->kf_sock_domain = 3125 so->so_proto->pr_domain->dom_family; 3126 kif->kf_sock_type = so->so_type; 3127 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3128 } 3129 if (tp != NULL) { 3130 strlcpy(kif->kf_path, tty_devname(tp), 3131 sizeof(kif->kf_path)); 3132 } 3133 if (shmfd != NULL) 3134 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3135 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3136 if (error) 3137 break; 3138 } 3139 FILEDESC_SUNLOCK(fdp); 3140 fddrop(fdp); 3141 free(kif, M_TEMP); 3142 return (0); 3143} 3144 3145static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD, 3146 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); 3147#endif /* COMPAT_FREEBSD7 */ 3148 3149#ifdef KINFO_FILE_SIZE 3150CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3151#endif 3152 3153static int 3154export_fd_for_sysctl(void *data, int type, int fd, int fflags, int refcnt, 3155 int64_t offset, int fd_is_cap, cap_rights_t fd_cap_rights, 3156 struct kinfo_file *kif, struct sysctl_req *req) 3157{ 3158 struct { 3159 int fflag; 3160 int kf_fflag; 3161 } fflags_table[] = { 3162 { FAPPEND, KF_FLAG_APPEND }, 3163 { FASYNC, KF_FLAG_ASYNC }, 3164 { FFSYNC, KF_FLAG_FSYNC }, 3165 { FHASLOCK, KF_FLAG_HASLOCK }, 3166 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3167 { FREAD, KF_FLAG_READ }, 3168 { FWRITE, KF_FLAG_WRITE }, 3169 { O_CREAT, KF_FLAG_CREAT }, 3170 { O_DIRECT, KF_FLAG_DIRECT }, 3171 { O_EXCL, KF_FLAG_EXCL }, 3172 { O_EXEC, KF_FLAG_EXEC }, 3173 { O_EXLOCK, KF_FLAG_EXLOCK }, 3174 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3175 { O_SHLOCK, KF_FLAG_SHLOCK }, 3176 { O_TRUNC, KF_FLAG_TRUNC } 3177 }; 3178#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3179 struct vnode *vp; 3180 int error, vfslocked; 3181 unsigned int i; 3182 3183 bzero(kif, sizeof(*kif)); 3184 switch (type) { 3185 case KF_TYPE_FIFO: 3186 case KF_TYPE_VNODE: 3187 vp = (struct vnode *)data; 3188 error = fill_vnode_info(vp, kif); 3189 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3190 vrele(vp); 3191 VFS_UNLOCK_GIANT(vfslocked); 3192 break; 3193 case KF_TYPE_SOCKET: 3194 error = fill_socket_info((struct socket *)data, kif); 3195 break; 3196 case KF_TYPE_PIPE: 3197 error = fill_pipe_info((struct pipe *)data, kif); 3198 break; 3199 case KF_TYPE_PTS: 3200 error = fill_pts_info((struct tty *)data, kif); 3201 break; 3202 case KF_TYPE_PROCDESC: 3203 error = fill_procdesc_info((struct procdesc *)data, kif); 3204 break; 3205 case KF_TYPE_SHM: 3206 error = fill_shm_info((struct file *)data, kif); 3207 break; 3208 default: 3209 error = 0; 3210 } 3211 if (error == 0) 3212 kif->kf_status |= KF_ATTR_VALID; 3213 3214 /* 3215 * Translate file access flags. 3216 */ 3217 for (i = 0; i < NFFLAGS; i++) 3218 if (fflags & fflags_table[i].fflag) 3219 kif->kf_flags |= fflags_table[i].kf_fflag; 3220 if (fd_is_cap) 3221 kif->kf_flags |= KF_FLAG_CAPABILITY; 3222 if (fd_is_cap) 3223 kif->kf_cap_rights = fd_cap_rights; 3224 kif->kf_fd = fd; 3225 kif->kf_type = type; 3226 kif->kf_ref_count = refcnt; 3227 kif->kf_offset = offset; 3228 /* Pack record size down */ 3229 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3230 strlen(kif->kf_path) + 1; 3231 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3232 error = SYSCTL_OUT(req, kif, kif->kf_structsize); 3233 return (error); 3234} 3235 3236/* 3237 * Get per-process file descriptors for use by procstat(1), et al. 3238 */ 3239static int 3240sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3241{ 3242 struct file *fp; 3243 struct filedesc *fdp; 3244 struct kinfo_file *kif; 3245 struct proc *p; 3246 struct vnode *cttyvp, *textvp, *tracevp; 3247 size_t oldidx; 3248 int64_t offset; 3249 void *data; 3250 int error, i, *name; 3251 int fd_is_cap, type, refcnt, fflags; 3252 cap_rights_t fd_cap_rights; 3253 3254 name = (int *)arg1; 3255 if ((p = pfind((pid_t)name[0])) == NULL) 3256 return (ESRCH); 3257 if ((error = p_candebug(curthread, p))) { 3258 PROC_UNLOCK(p); 3259 return (error); 3260 } 3261 /* ktrace vnode */ 3262 tracevp = p->p_tracevp; 3263 if (tracevp != NULL) 3264 vref(tracevp); 3265 /* text vnode */ 3266 textvp = p->p_textvp; 3267 if (textvp != NULL) 3268 vref(textvp); 3269 /* Controlling tty. */ 3270 cttyvp = NULL; 3271 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3272 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3273 if (cttyvp != NULL) 3274 vref(cttyvp); 3275 } 3276 fdp = fdhold(p); 3277 PROC_UNLOCK(p); 3278 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3279 if (tracevp != NULL) 3280 export_fd_for_sysctl(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3281 FREAD | FWRITE, -1, -1, 0, 0, kif, req); 3282 if (textvp != NULL) 3283 export_fd_for_sysctl(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3284 FREAD, -1, -1, 0, 0, kif, req); 3285 if (cttyvp != NULL) 3286 export_fd_for_sysctl(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3287 FREAD | FWRITE, -1, -1, 0, 0, kif, req); 3288 if (fdp == NULL) 3289 goto fail; 3290 FILEDESC_SLOCK(fdp); 3291 /* working directory */ 3292 if (fdp->fd_cdir != NULL) { 3293 vref(fdp->fd_cdir); 3294 data = fdp->fd_cdir; 3295 FILEDESC_SUNLOCK(fdp); 3296 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3297 FREAD, -1, -1, 0, 0, kif, req); 3298 FILEDESC_SLOCK(fdp); 3299 } 3300 /* root directory */ 3301 if (fdp->fd_rdir != NULL) { 3302 vref(fdp->fd_rdir); 3303 data = fdp->fd_rdir; 3304 FILEDESC_SUNLOCK(fdp); 3305 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3306 FREAD, -1, -1, 0, 0, kif, req); 3307 FILEDESC_SLOCK(fdp); 3308 } 3309 /* jail directory */ 3310 if (fdp->fd_jdir != NULL) { 3311 vref(fdp->fd_jdir); 3312 data = fdp->fd_jdir; 3313 FILEDESC_SUNLOCK(fdp); 3314 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3315 FREAD, -1, -1, 0, 0, kif, req); 3316 FILEDESC_SLOCK(fdp); 3317 } 3318 for (i = 0; i < fdp->fd_nfiles; i++) { 3319 if ((fp = fdp->fd_ofiles[i]) == NULL) 3320 continue; 3321 data = NULL; 3322 fd_is_cap = 0; 3323 fd_cap_rights = 0; 3324 3325#ifdef CAPABILITIES 3326 /* 3327 * When reporting a capability, most fields will be from the 3328 * underlying object, but do mark as a capability and export 3329 * the capability rights mask. 3330 */ 3331 if (fp->f_type == DTYPE_CAPABILITY) { 3332 fd_is_cap = 1; 3333 fd_cap_rights = cap_rights(fp); 3334 (void)cap_funwrap(fp, 0, &fp); 3335 } 3336#else /* !CAPABILITIES */ 3337 KASSERT(fp->f_type != DTYPE_CAPABILITY, 3338 ("sysctl_kern_proc_filedesc: saw capability")); 3339#endif 3340 switch (fp->f_type) { 3341 case DTYPE_VNODE: 3342 type = KF_TYPE_VNODE; 3343 vref(fp->f_vnode); 3344 data = fp->f_vnode; 3345 break; 3346 3347 case DTYPE_SOCKET: 3348 type = KF_TYPE_SOCKET; 3349 data = fp->f_data; 3350 break; 3351 3352 case DTYPE_PIPE: 3353 type = KF_TYPE_PIPE; 3354 data = fp->f_data; 3355 break; 3356 3357 case DTYPE_FIFO: 3358 type = KF_TYPE_FIFO; 3359 vref(fp->f_vnode); 3360 data = fp->f_vnode; 3361 break; 3362 3363 case DTYPE_KQUEUE: 3364 type = KF_TYPE_KQUEUE; 3365 break; 3366 3367 case DTYPE_CRYPTO: 3368 type = KF_TYPE_CRYPTO; 3369 break; 3370 3371 case DTYPE_MQUEUE: 3372 type = KF_TYPE_MQUEUE; 3373 break; 3374 3375 case DTYPE_SHM: 3376 type = KF_TYPE_SHM; 3377 data = fp; 3378 break; 3379 3380 case DTYPE_SEM: 3381 type = KF_TYPE_SEM; 3382 break; 3383 3384 case DTYPE_PTS: 3385 type = KF_TYPE_PTS; 3386 data = fp->f_data; 3387 break; 3388 3389#ifdef PROCDESC 3390 case DTYPE_PROCDESC: 3391 type = KF_TYPE_PROCDESC; 3392 data = fp->f_data; 3393 break; 3394#endif 3395 3396 default: 3397 type = KF_TYPE_UNKNOWN; 3398 break; 3399 } 3400 refcnt = fp->f_count; 3401 fflags = fp->f_flag; 3402 offset = fp->f_offset; 3403 3404 /* 3405 * Create sysctl entry. 3406 * It is OK to drop the filedesc lock here as we will 3407 * re-validate and re-evaluate its properties when 3408 * the loop continues. 3409 */ 3410 oldidx = req->oldidx; 3411 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO) 3412 FILEDESC_SUNLOCK(fdp); 3413 error = export_fd_for_sysctl(data, type, i, fflags, refcnt, 3414 offset, fd_is_cap, fd_cap_rights, kif, req); 3415 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO) 3416 FILEDESC_SLOCK(fdp); 3417 if (error) { 3418 if (error == ENOMEM) { 3419 /* 3420 * The hack to keep the ABI of sysctl 3421 * kern.proc.filedesc intact, but not 3422 * to account a partially copied 3423 * kinfo_file into the oldidx. 3424 */ 3425 req->oldidx = oldidx; 3426 error = 0; 3427 } 3428 break; 3429 } 3430 } 3431 FILEDESC_SUNLOCK(fdp); 3432fail: 3433 if (fdp != NULL) 3434 fddrop(fdp); 3435 free(kif, M_TEMP); 3436 return (error); 3437} 3438 3439int 3440vntype_to_kinfo(int vtype) 3441{ 3442 struct { 3443 int vtype; 3444 int kf_vtype; 3445 } vtypes_table[] = { 3446 { VBAD, KF_VTYPE_VBAD }, 3447 { VBLK, KF_VTYPE_VBLK }, 3448 { VCHR, KF_VTYPE_VCHR }, 3449 { VDIR, KF_VTYPE_VDIR }, 3450 { VFIFO, KF_VTYPE_VFIFO }, 3451 { VLNK, KF_VTYPE_VLNK }, 3452 { VNON, KF_VTYPE_VNON }, 3453 { VREG, KF_VTYPE_VREG }, 3454 { VSOCK, KF_VTYPE_VSOCK } 3455 }; 3456#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3457 unsigned int i; 3458 3459 /* 3460 * Perform vtype translation. 3461 */ 3462 for (i = 0; i < NVTYPES; i++) 3463 if (vtypes_table[i].vtype == vtype) 3464 break; 3465 if (i < NVTYPES) 3466 return (vtypes_table[i].kf_vtype); 3467 3468 return (KF_VTYPE_UNKNOWN); 3469} 3470 3471static int 3472fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3473{ 3474 struct vattr va; 3475 char *fullpath, *freepath; 3476 int error, vfslocked; 3477 3478 if (vp == NULL) 3479 return (1); 3480 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3481 freepath = NULL; 3482 fullpath = "-"; 3483 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3484 if (error == 0) { 3485 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3486 } 3487 if (freepath != NULL) 3488 free(freepath, M_TEMP); 3489 3490 /* 3491 * Retrieve vnode attributes. 3492 */ 3493 va.va_fsid = VNOVAL; 3494 va.va_rdev = NODEV; 3495 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3496 vn_lock(vp, LK_SHARED | LK_RETRY); 3497 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3498 VOP_UNLOCK(vp, 0); 3499 VFS_UNLOCK_GIANT(vfslocked); 3500 if (error != 0) 3501 return (error); 3502 if (va.va_fsid != VNOVAL) 3503 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3504 else 3505 kif->kf_un.kf_file.kf_file_fsid = 3506 vp->v_mount->mnt_stat.f_fsid.val[0]; 3507 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3508 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3509 kif->kf_un.kf_file.kf_file_size = va.va_size; 3510 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3511 return (0); 3512} 3513 3514static int 3515fill_socket_info(struct socket *so, struct kinfo_file *kif) 3516{ 3517 struct sockaddr *sa; 3518 struct inpcb *inpcb; 3519 struct unpcb *unpcb; 3520 int error; 3521 3522 if (so == NULL) 3523 return (1); 3524 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3525 kif->kf_sock_type = so->so_type; 3526 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3527 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3528 switch(kif->kf_sock_domain) { 3529 case AF_INET: 3530 case AF_INET6: 3531 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3532 if (so->so_pcb != NULL) { 3533 inpcb = (struct inpcb *)(so->so_pcb); 3534 kif->kf_un.kf_sock.kf_sock_inpcb = 3535 (uintptr_t)inpcb->inp_ppcb; 3536 } 3537 } 3538 break; 3539 case AF_UNIX: 3540 if (so->so_pcb != NULL) { 3541 unpcb = (struct unpcb *)(so->so_pcb); 3542 if (unpcb->unp_conn) { 3543 kif->kf_un.kf_sock.kf_sock_unpconn = 3544 (uintptr_t)unpcb->unp_conn; 3545 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3546 so->so_rcv.sb_state; 3547 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3548 so->so_snd.sb_state; 3549 } 3550 } 3551 break; 3552 } 3553 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3554 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3555 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3556 free(sa, M_SONAME); 3557 } 3558 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3559 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3560 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3561 free(sa, M_SONAME); 3562 } 3563 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3564 sizeof(kif->kf_path)); 3565 return (0); 3566} 3567 3568static int 3569fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3570{ 3571 3572 if (tp == NULL) 3573 return (1); 3574 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3575 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3576 return (0); 3577} 3578 3579static int 3580fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3581{ 3582 3583 if (pi == NULL) 3584 return (1); 3585 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3586 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3587 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3588 return (0); 3589} 3590 3591static int 3592fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3593{ 3594 3595 if (pdp == NULL) 3596 return (1); 3597 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3598 return (0); 3599} 3600 3601static int 3602fill_shm_info(struct file *fp, struct kinfo_file *kif) 3603{ 3604 struct thread *td; 3605 struct stat sb; 3606 3607 td = curthread; 3608 if (fp->f_data == NULL) 3609 return (1); 3610 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3611 return (1); 3612 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3613 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3614 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3615 return (0); 3616} 3617 3618static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD, 3619 sysctl_kern_proc_filedesc, "Process filedesc entries"); 3620 3621#ifdef DDB 3622/* 3623 * For the purposes of debugging, generate a human-readable string for the 3624 * file type. 3625 */ 3626static const char * 3627file_type_to_name(short type) 3628{ 3629 3630 switch (type) { 3631 case 0: 3632 return ("zero"); 3633 case DTYPE_VNODE: 3634 return ("vnod"); 3635 case DTYPE_SOCKET: 3636 return ("sock"); 3637 case DTYPE_PIPE: 3638 return ("pipe"); 3639 case DTYPE_FIFO: 3640 return ("fifo"); 3641 case DTYPE_KQUEUE: 3642 return ("kque"); 3643 case DTYPE_CRYPTO: 3644 return ("crpt"); 3645 case DTYPE_MQUEUE: 3646 return ("mque"); 3647 case DTYPE_SHM: 3648 return ("shm"); 3649 case DTYPE_SEM: 3650 return ("ksem"); 3651 default: 3652 return ("unkn"); 3653 } 3654} 3655 3656/* 3657 * For the purposes of debugging, identify a process (if any, perhaps one of 3658 * many) that references the passed file in its file descriptor array. Return 3659 * NULL if none. 3660 */ 3661static struct proc * 3662file_to_first_proc(struct file *fp) 3663{ 3664 struct filedesc *fdp; 3665 struct proc *p; 3666 int n; 3667 3668 FOREACH_PROC_IN_SYSTEM(p) { 3669 if (p->p_state == PRS_NEW) 3670 continue; 3671 fdp = p->p_fd; 3672 if (fdp == NULL) 3673 continue; 3674 for (n = 0; n < fdp->fd_nfiles; n++) { 3675 if (fp == fdp->fd_ofiles[n]) 3676 return (p); 3677 } 3678 } 3679 return (NULL); 3680} 3681 3682static void 3683db_print_file(struct file *fp, int header) 3684{ 3685 struct proc *p; 3686 3687 if (header) 3688 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3689 "File", "Type", "Data", "Flag", "GCFl", "Count", 3690 "MCount", "Vnode", "FPID", "FCmd"); 3691 p = file_to_first_proc(fp); 3692 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3693 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3694 0, fp->f_count, 0, fp->f_vnode, 3695 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3696} 3697 3698DB_SHOW_COMMAND(file, db_show_file) 3699{ 3700 struct file *fp; 3701 3702 if (!have_addr) { 3703 db_printf("usage: show file <addr>\n"); 3704 return; 3705 } 3706 fp = (struct file *)addr; 3707 db_print_file(fp, 1); 3708} 3709 3710DB_SHOW_COMMAND(files, db_show_files) 3711{ 3712 struct filedesc *fdp; 3713 struct file *fp; 3714 struct proc *p; 3715 int header; 3716 int n; 3717 3718 header = 1; 3719 FOREACH_PROC_IN_SYSTEM(p) { 3720 if (p->p_state == PRS_NEW) 3721 continue; 3722 if ((fdp = p->p_fd) == NULL) 3723 continue; 3724 for (n = 0; n < fdp->fd_nfiles; ++n) { 3725 if ((fp = fdp->fd_ofiles[n]) == NULL) 3726 continue; 3727 db_print_file(fp, header); 3728 header = 0; 3729 } 3730 } 3731} 3732#endif 3733 3734SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3735 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3736 3737SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3738 &maxfiles, 0, "Maximum number of files"); 3739 3740SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3741 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3742 3743/* ARGSUSED*/ 3744static void 3745filelistinit(void *dummy) 3746{ 3747 3748 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3749 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3750 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3751 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3752} 3753SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3754 3755/*-------------------------------------------------------------------*/ 3756 3757static int 3758badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3759 int flags, struct thread *td) 3760{ 3761 3762 return (EBADF); 3763} 3764 3765static int 3766badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3767 struct thread *td) 3768{ 3769 3770 return (EINVAL); 3771} 3772 3773static int 3774badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3775 struct thread *td) 3776{ 3777 3778 return (EBADF); 3779} 3780 3781static int 3782badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3783 struct thread *td) 3784{ 3785 3786 return (0); 3787} 3788 3789static int 3790badfo_kqfilter(struct file *fp, struct knote *kn) 3791{ 3792 3793 return (EBADF); 3794} 3795 3796static int 3797badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3798 struct thread *td) 3799{ 3800 3801 return (EBADF); 3802} 3803 3804static int 3805badfo_close(struct file *fp, struct thread *td) 3806{ 3807 3808 return (EBADF); 3809} 3810 3811static int 3812badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3813 struct thread *td) 3814{ 3815 3816 return (EBADF); 3817} 3818 3819static int 3820badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3821 struct thread *td) 3822{ 3823 3824 return (EBADF); 3825} 3826 3827struct fileops badfileops = { 3828 .fo_read = badfo_readwrite, 3829 .fo_write = badfo_readwrite, 3830 .fo_truncate = badfo_truncate, 3831 .fo_ioctl = badfo_ioctl, 3832 .fo_poll = badfo_poll, 3833 .fo_kqfilter = badfo_kqfilter, 3834 .fo_stat = badfo_stat, 3835 .fo_close = badfo_close, 3836 .fo_chmod = badfo_chmod, 3837 .fo_chown = badfo_chown, 3838}; 3839 3840int 3841invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3842 struct thread *td) 3843{ 3844 3845 return (EINVAL); 3846} 3847 3848int 3849invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3850 struct thread *td) 3851{ 3852 3853 return (EINVAL); 3854} 3855 3856/*-------------------------------------------------------------------*/ 3857 3858/* 3859 * File Descriptor pseudo-device driver (/dev/fd/). 3860 * 3861 * Opening minor device N dup()s the file (if any) connected to file 3862 * descriptor N belonging to the calling process. Note that this driver 3863 * consists of only the ``open()'' routine, because all subsequent 3864 * references to this file will be direct to the other driver. 3865 * 3866 * XXX: we could give this one a cloning event handler if necessary. 3867 */ 3868 3869/* ARGSUSED */ 3870static int 3871fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3872{ 3873 3874 /* 3875 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3876 * the file descriptor being sought for duplication. The error 3877 * return ensures that the vnode for this device will be released 3878 * by vn_open. Open will detect this special error and take the 3879 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3880 * will simply report the error. 3881 */ 3882 td->td_dupfd = dev2unit(dev); 3883 return (ENODEV); 3884} 3885 3886static struct cdevsw fildesc_cdevsw = { 3887 .d_version = D_VERSION, 3888 .d_open = fdopen, 3889 .d_name = "FD", 3890}; 3891 3892static void 3893fildesc_drvinit(void *unused) 3894{ 3895 struct cdev *dev; 3896 3897 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3898 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3899 make_dev_alias(dev, "stdin"); 3900 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3901 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3902 make_dev_alias(dev, "stdout"); 3903 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3904 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3905 make_dev_alias(dev, "stderr"); 3906} 3907 3908SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3909