kern_descrip.c revision 237016
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 237016 2012-06-13 19:00:29Z 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 closefp(struct filedesc *fdp, int fd, struct file *fp, 118 struct thread *td); 119static int do_dup(struct thread *td, int flags, int old, int new, 120 register_t *retval); 121static int fd_first_free(struct filedesc *, int, int); 122static int fd_last_used(struct filedesc *, int); 123static void fdgrowtable(struct filedesc *, int); 124static void fdunused(struct filedesc *fdp, int fd); 125static void fdused(struct filedesc *fdp, int fd); 126static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif); 127static int fill_socket_info(struct socket *so, struct kinfo_file *kif); 128static int fill_pts_info(struct tty *tp, struct kinfo_file *kif); 129static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif); 130static int fill_procdesc_info(struct procdesc *pdp, 131 struct kinfo_file *kif); 132static int fill_shm_info(struct file *fp, struct kinfo_file *kif); 133 134/* 135 * A process is initially started out with NDFILE descriptors stored within 136 * this structure, selected to be enough for typical applications based on 137 * the historical limit of 20 open files (and the usage of descriptors by 138 * shells). If these descriptors are exhausted, a larger descriptor table 139 * may be allocated, up to a process' resource limit; the internal arrays 140 * are then unused. 141 */ 142#define NDFILE 20 143#define NDSLOTSIZE sizeof(NDSLOTTYPE) 144#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 145#define NDSLOT(x) ((x) / NDENTRIES) 146#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 147#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 148 149/* 150 * Storage required per open file descriptor. 151 */ 152#define OFILESIZE (sizeof(struct file *) + sizeof(char)) 153 154/* 155 * Storage to hold unused ofiles that need to be reclaimed. 156 */ 157struct freetable { 158 struct file **ft_table; 159 SLIST_ENTRY(freetable) ft_next; 160}; 161 162/* 163 * Basic allocation of descriptors: 164 * one of the above, plus arrays for NDFILE descriptors. 165 */ 166struct filedesc0 { 167 struct filedesc fd_fd; 168 /* 169 * ofiles which need to be reclaimed on free. 170 */ 171 SLIST_HEAD(,freetable) fd_free; 172 /* 173 * These arrays are used when the number of open files is 174 * <= NDFILE, and are then pointed to by the pointers above. 175 */ 176 struct file *fd_dfiles[NDFILE]; 177 char fd_dfileflags[NDFILE]; 178 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 179}; 180 181/* 182 * Descriptor management. 183 */ 184volatile int openfiles; /* actual number of open files */ 185struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 186void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 187 188/* A mutex to protect the association between a proc and filedesc. */ 189static struct mtx fdesc_mtx; 190 191/* 192 * If low >= size, just return low. Otherwise find the first zero bit in the 193 * given bitmap, starting at low and not exceeding size - 1. Return size if 194 * not found. 195 */ 196static int 197fd_first_free(struct filedesc *fdp, int low, int size) 198{ 199 NDSLOTTYPE *map = fdp->fd_map; 200 NDSLOTTYPE mask; 201 int off, maxoff; 202 203 if (low >= size) 204 return (low); 205 206 off = NDSLOT(low); 207 if (low % NDENTRIES) { 208 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 209 if ((mask &= ~map[off]) != 0UL) 210 return (off * NDENTRIES + ffsl(mask) - 1); 211 ++off; 212 } 213 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 214 if (map[off] != ~0UL) 215 return (off * NDENTRIES + ffsl(~map[off]) - 1); 216 return (size); 217} 218 219/* 220 * Find the highest non-zero bit in the given bitmap, starting at 0 and 221 * not exceeding size - 1. Return -1 if not found. 222 */ 223static int 224fd_last_used(struct filedesc *fdp, int size) 225{ 226 NDSLOTTYPE *map = fdp->fd_map; 227 NDSLOTTYPE mask; 228 int off, minoff; 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(0); off >= minoff; --off) 238 if (map[off] != 0) 239 return (off * NDENTRIES + flsl(map[off]) - 1); 240 return (-1); 241} 242 243static int 244fdisused(struct filedesc *fdp, int fd) 245{ 246 KASSERT((unsigned int)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, 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, 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 * Duplicate the source descriptor. 880 */ 881 fdp->fd_ofiles[new] = fp; 882 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 883 if (new > fdp->fd_lastfile) 884 fdp->fd_lastfile = new; 885 *retval = new; 886 887 if (delfp != NULL) { 888 (void) closefp(fdp, new, delfp, td); 889 /* closefp() drops the FILEDESC lock for us. */ 890 } else { 891 FILEDESC_XUNLOCK(fdp); 892 } 893 894 return (0); 895} 896 897/* 898 * If sigio is on the list associated with a process or process group, 899 * disable signalling from the device, remove sigio from the list and 900 * free sigio. 901 */ 902void 903funsetown(struct sigio **sigiop) 904{ 905 struct sigio *sigio; 906 907 SIGIO_LOCK(); 908 sigio = *sigiop; 909 if (sigio == NULL) { 910 SIGIO_UNLOCK(); 911 return; 912 } 913 *(sigio->sio_myref) = NULL; 914 if ((sigio)->sio_pgid < 0) { 915 struct pgrp *pg = (sigio)->sio_pgrp; 916 PGRP_LOCK(pg); 917 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 918 sigio, sio_pgsigio); 919 PGRP_UNLOCK(pg); 920 } else { 921 struct proc *p = (sigio)->sio_proc; 922 PROC_LOCK(p); 923 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 924 sigio, sio_pgsigio); 925 PROC_UNLOCK(p); 926 } 927 SIGIO_UNLOCK(); 928 crfree(sigio->sio_ucred); 929 free(sigio, M_SIGIO); 930} 931 932/* 933 * Free a list of sigio structures. 934 * We only need to lock the SIGIO_LOCK because we have made ourselves 935 * inaccessible to callers of fsetown and therefore do not need to lock 936 * the proc or pgrp struct for the list manipulation. 937 */ 938void 939funsetownlst(struct sigiolst *sigiolst) 940{ 941 struct proc *p; 942 struct pgrp *pg; 943 struct sigio *sigio; 944 945 sigio = SLIST_FIRST(sigiolst); 946 if (sigio == NULL) 947 return; 948 p = NULL; 949 pg = NULL; 950 951 /* 952 * Every entry of the list should belong 953 * to a single proc or pgrp. 954 */ 955 if (sigio->sio_pgid < 0) { 956 pg = sigio->sio_pgrp; 957 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 958 } else /* if (sigio->sio_pgid > 0) */ { 959 p = sigio->sio_proc; 960 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 961 } 962 963 SIGIO_LOCK(); 964 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 965 *(sigio->sio_myref) = NULL; 966 if (pg != NULL) { 967 KASSERT(sigio->sio_pgid < 0, 968 ("Proc sigio in pgrp sigio list")); 969 KASSERT(sigio->sio_pgrp == pg, 970 ("Bogus pgrp in sigio list")); 971 PGRP_LOCK(pg); 972 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 973 sio_pgsigio); 974 PGRP_UNLOCK(pg); 975 } else /* if (p != NULL) */ { 976 KASSERT(sigio->sio_pgid > 0, 977 ("Pgrp sigio in proc sigio list")); 978 KASSERT(sigio->sio_proc == p, 979 ("Bogus proc in sigio list")); 980 PROC_LOCK(p); 981 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 982 sio_pgsigio); 983 PROC_UNLOCK(p); 984 } 985 SIGIO_UNLOCK(); 986 crfree(sigio->sio_ucred); 987 free(sigio, M_SIGIO); 988 SIGIO_LOCK(); 989 } 990 SIGIO_UNLOCK(); 991} 992 993/* 994 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 995 * 996 * After permission checking, add a sigio structure to the sigio list for 997 * the process or process group. 998 */ 999int 1000fsetown(pid_t pgid, struct sigio **sigiop) 1001{ 1002 struct proc *proc; 1003 struct pgrp *pgrp; 1004 struct sigio *sigio; 1005 int ret; 1006 1007 if (pgid == 0) { 1008 funsetown(sigiop); 1009 return (0); 1010 } 1011 1012 ret = 0; 1013 1014 /* Allocate and fill in the new sigio out of locks. */ 1015 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1016 sigio->sio_pgid = pgid; 1017 sigio->sio_ucred = crhold(curthread->td_ucred); 1018 sigio->sio_myref = sigiop; 1019 1020 sx_slock(&proctree_lock); 1021 if (pgid > 0) { 1022 proc = pfind(pgid); 1023 if (proc == NULL) { 1024 ret = ESRCH; 1025 goto fail; 1026 } 1027 1028 /* 1029 * Policy - Don't allow a process to FSETOWN a process 1030 * in another session. 1031 * 1032 * Remove this test to allow maximum flexibility or 1033 * restrict FSETOWN to the current process or process 1034 * group for maximum safety. 1035 */ 1036 PROC_UNLOCK(proc); 1037 if (proc->p_session != curthread->td_proc->p_session) { 1038 ret = EPERM; 1039 goto fail; 1040 } 1041 1042 pgrp = NULL; 1043 } else /* if (pgid < 0) */ { 1044 pgrp = pgfind(-pgid); 1045 if (pgrp == NULL) { 1046 ret = ESRCH; 1047 goto fail; 1048 } 1049 PGRP_UNLOCK(pgrp); 1050 1051 /* 1052 * Policy - Don't allow a process to FSETOWN a process 1053 * in another session. 1054 * 1055 * Remove this test to allow maximum flexibility or 1056 * restrict FSETOWN to the current process or process 1057 * group for maximum safety. 1058 */ 1059 if (pgrp->pg_session != curthread->td_proc->p_session) { 1060 ret = EPERM; 1061 goto fail; 1062 } 1063 1064 proc = NULL; 1065 } 1066 funsetown(sigiop); 1067 if (pgid > 0) { 1068 PROC_LOCK(proc); 1069 /* 1070 * Since funsetownlst() is called without the proctree 1071 * locked, we need to check for P_WEXIT. 1072 * XXX: is ESRCH correct? 1073 */ 1074 if ((proc->p_flag & P_WEXIT) != 0) { 1075 PROC_UNLOCK(proc); 1076 ret = ESRCH; 1077 goto fail; 1078 } 1079 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1080 sigio->sio_proc = proc; 1081 PROC_UNLOCK(proc); 1082 } else { 1083 PGRP_LOCK(pgrp); 1084 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1085 sigio->sio_pgrp = pgrp; 1086 PGRP_UNLOCK(pgrp); 1087 } 1088 sx_sunlock(&proctree_lock); 1089 SIGIO_LOCK(); 1090 *sigiop = sigio; 1091 SIGIO_UNLOCK(); 1092 return (0); 1093 1094fail: 1095 sx_sunlock(&proctree_lock); 1096 crfree(sigio->sio_ucred); 1097 free(sigio, M_SIGIO); 1098 return (ret); 1099} 1100 1101/* 1102 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1103 */ 1104pid_t 1105fgetown(sigiop) 1106 struct sigio **sigiop; 1107{ 1108 pid_t pgid; 1109 1110 SIGIO_LOCK(); 1111 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1112 SIGIO_UNLOCK(); 1113 return (pgid); 1114} 1115 1116/* 1117 * Function drops the filedesc lock on return. 1118 */ 1119static int 1120closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td) 1121{ 1122 struct file *fp_object; 1123 int error, holdleaders; 1124 1125 FILEDESC_XLOCK_ASSERT(fdp); 1126 1127 if (td->td_proc->p_fdtol != NULL) { 1128 /* 1129 * Ask fdfree() to sleep to ensure that all relevant 1130 * process leaders can be traversed in closef(). 1131 */ 1132 fdp->fd_holdleaderscount++; 1133 holdleaders = 1; 1134 } else { 1135 holdleaders = 0; 1136 } 1137 1138 /* 1139 * We now hold the fp reference that used to be owned by the 1140 * descriptor array. We have to unlock the FILEDESC *AFTER* 1141 * knote_fdclose to prevent a race of the fd getting opened, a knote 1142 * added, and deleteing a knote for the new fd. 1143 */ 1144 knote_fdclose(td, fd); 1145 1146 /* 1147 * When we're closing an fd with a capability, we need to notify 1148 * mqueue if the underlying object is of type mqueue. 1149 */ 1150 (void)cap_funwrap(fp, 0, &fp_object); 1151 if (fp_object->f_type == DTYPE_MQUEUE) 1152 mq_fdclose(td, fd, fp_object); 1153 FILEDESC_XUNLOCK(fdp); 1154 1155 error = closef(fp, td); 1156 if (holdleaders) { 1157 FILEDESC_XLOCK(fdp); 1158 fdp->fd_holdleaderscount--; 1159 if (fdp->fd_holdleaderscount == 0 && 1160 fdp->fd_holdleaderswakeup != 0) { 1161 fdp->fd_holdleaderswakeup = 0; 1162 wakeup(&fdp->fd_holdleaderscount); 1163 } 1164 FILEDESC_XUNLOCK(fdp); 1165 } 1166 return (error); 1167} 1168 1169/* 1170 * Close a file descriptor. 1171 */ 1172#ifndef _SYS_SYSPROTO_H_ 1173struct close_args { 1174 int fd; 1175}; 1176#endif 1177/* ARGSUSED */ 1178int 1179sys_close(td, uap) 1180 struct thread *td; 1181 struct close_args *uap; 1182{ 1183 1184 return (kern_close(td, uap->fd)); 1185} 1186 1187int 1188kern_close(td, fd) 1189 struct thread *td; 1190 int fd; 1191{ 1192 struct filedesc *fdp; 1193 struct file *fp; 1194 1195 fdp = td->td_proc->p_fd; 1196 1197 AUDIT_SYSCLOSE(td, fd); 1198 1199 FILEDESC_XLOCK(fdp); 1200 if ((unsigned)fd >= fdp->fd_nfiles || 1201 (fp = fdp->fd_ofiles[fd]) == NULL) { 1202 FILEDESC_XUNLOCK(fdp); 1203 return (EBADF); 1204 } 1205 fdp->fd_ofiles[fd] = NULL; 1206 fdp->fd_ofileflags[fd] = 0; 1207 fdunused(fdp, fd); 1208 1209 /* closefp() drops the FILEDESC lock for us. */ 1210 return (closefp(fdp, fd, fp, td)); 1211} 1212 1213/* 1214 * Close open file descriptors. 1215 */ 1216#ifndef _SYS_SYSPROTO_H_ 1217struct closefrom_args { 1218 int lowfd; 1219}; 1220#endif 1221/* ARGSUSED */ 1222int 1223sys_closefrom(struct thread *td, struct closefrom_args *uap) 1224{ 1225 struct filedesc *fdp; 1226 int fd; 1227 1228 fdp = td->td_proc->p_fd; 1229 AUDIT_ARG_FD(uap->lowfd); 1230 1231 /* 1232 * Treat negative starting file descriptor values identical to 1233 * closefrom(0) which closes all files. 1234 */ 1235 if (uap->lowfd < 0) 1236 uap->lowfd = 0; 1237 FILEDESC_SLOCK(fdp); 1238 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) { 1239 if (fdp->fd_ofiles[fd] != NULL) { 1240 FILEDESC_SUNLOCK(fdp); 1241 (void)kern_close(td, fd); 1242 FILEDESC_SLOCK(fdp); 1243 } 1244 } 1245 FILEDESC_SUNLOCK(fdp); 1246 return (0); 1247} 1248 1249#if defined(COMPAT_43) 1250/* 1251 * Return status information about a file descriptor. 1252 */ 1253#ifndef _SYS_SYSPROTO_H_ 1254struct ofstat_args { 1255 int fd; 1256 struct ostat *sb; 1257}; 1258#endif 1259/* ARGSUSED */ 1260int 1261ofstat(struct thread *td, struct ofstat_args *uap) 1262{ 1263 struct ostat oub; 1264 struct stat ub; 1265 int error; 1266 1267 error = kern_fstat(td, uap->fd, &ub); 1268 if (error == 0) { 1269 cvtstat(&ub, &oub); 1270 error = copyout(&oub, uap->sb, sizeof(oub)); 1271 } 1272 return (error); 1273} 1274#endif /* COMPAT_43 */ 1275 1276/* 1277 * Return status information about a file descriptor. 1278 */ 1279#ifndef _SYS_SYSPROTO_H_ 1280struct fstat_args { 1281 int fd; 1282 struct stat *sb; 1283}; 1284#endif 1285/* ARGSUSED */ 1286int 1287sys_fstat(struct thread *td, struct fstat_args *uap) 1288{ 1289 struct stat ub; 1290 int error; 1291 1292 error = kern_fstat(td, uap->fd, &ub); 1293 if (error == 0) 1294 error = copyout(&ub, uap->sb, sizeof(ub)); 1295 return (error); 1296} 1297 1298int 1299kern_fstat(struct thread *td, int fd, struct stat *sbp) 1300{ 1301 struct file *fp; 1302 int error; 1303 1304 AUDIT_ARG_FD(fd); 1305 1306 if ((error = fget(td, fd, CAP_FSTAT, &fp)) != 0) 1307 return (error); 1308 1309 AUDIT_ARG_FILE(td->td_proc, fp); 1310 1311 error = fo_stat(fp, sbp, td->td_ucred, td); 1312 fdrop(fp, td); 1313#ifdef KTRACE 1314 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1315 ktrstat(sbp); 1316#endif 1317 return (error); 1318} 1319 1320/* 1321 * Return status information about a file descriptor. 1322 */ 1323#ifndef _SYS_SYSPROTO_H_ 1324struct nfstat_args { 1325 int fd; 1326 struct nstat *sb; 1327}; 1328#endif 1329/* ARGSUSED */ 1330int 1331sys_nfstat(struct thread *td, struct nfstat_args *uap) 1332{ 1333 struct nstat nub; 1334 struct stat ub; 1335 int error; 1336 1337 error = kern_fstat(td, uap->fd, &ub); 1338 if (error == 0) { 1339 cvtnstat(&ub, &nub); 1340 error = copyout(&nub, uap->sb, sizeof(nub)); 1341 } 1342 return (error); 1343} 1344 1345/* 1346 * Return pathconf information about a file descriptor. 1347 */ 1348#ifndef _SYS_SYSPROTO_H_ 1349struct fpathconf_args { 1350 int fd; 1351 int name; 1352}; 1353#endif 1354/* ARGSUSED */ 1355int 1356sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1357{ 1358 struct file *fp; 1359 struct vnode *vp; 1360 int error; 1361 1362 if ((error = fget(td, uap->fd, CAP_FPATHCONF, &fp)) != 0) 1363 return (error); 1364 1365 /* If asynchronous I/O is available, it works for all descriptors. */ 1366 if (uap->name == _PC_ASYNC_IO) { 1367 td->td_retval[0] = async_io_version; 1368 goto out; 1369 } 1370 vp = fp->f_vnode; 1371 if (vp != NULL) { 1372 int vfslocked; 1373 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1374 vn_lock(vp, LK_SHARED | LK_RETRY); 1375 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1376 VOP_UNLOCK(vp, 0); 1377 VFS_UNLOCK_GIANT(vfslocked); 1378 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1379 if (uap->name != _PC_PIPE_BUF) { 1380 error = EINVAL; 1381 } else { 1382 td->td_retval[0] = PIPE_BUF; 1383 error = 0; 1384 } 1385 } else { 1386 error = EOPNOTSUPP; 1387 } 1388out: 1389 fdrop(fp, td); 1390 return (error); 1391} 1392 1393/* 1394 * Grow the file table to accomodate (at least) nfd descriptors. 1395 */ 1396static void 1397fdgrowtable(struct filedesc *fdp, int nfd) 1398{ 1399 struct filedesc0 *fdp0; 1400 struct freetable *fo; 1401 struct file **ntable; 1402 struct file **otable; 1403 char *nfileflags; 1404 int nnfiles, onfiles; 1405 NDSLOTTYPE *nmap; 1406 1407 FILEDESC_XLOCK_ASSERT(fdp); 1408 1409 KASSERT(fdp->fd_nfiles > 0, 1410 ("zero-length file table")); 1411 1412 /* compute the size of the new table */ 1413 onfiles = fdp->fd_nfiles; 1414 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1415 if (nnfiles <= onfiles) 1416 /* the table is already large enough */ 1417 return; 1418 1419 /* allocate a new table and (if required) new bitmaps */ 1420 ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable), 1421 M_FILEDESC, M_ZERO | M_WAITOK); 1422 nfileflags = (char *)&ntable[nnfiles]; 1423 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1424 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, 1425 M_FILEDESC, M_ZERO | M_WAITOK); 1426 else 1427 nmap = NULL; 1428 1429 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1430 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1431 otable = fdp->fd_ofiles; 1432 fdp->fd_ofileflags = nfileflags; 1433 fdp->fd_ofiles = ntable; 1434 /* 1435 * We must preserve ofiles until the process exits because we can't 1436 * be certain that no threads have references to the old table via 1437 * _fget(). 1438 */ 1439 if (onfiles > NDFILE) { 1440 fo = (struct freetable *)&otable[onfiles]; 1441 fdp0 = (struct filedesc0 *)fdp; 1442 fo->ft_table = otable; 1443 SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next); 1444 } 1445 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1446 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1447 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1448 free(fdp->fd_map, M_FILEDESC); 1449 fdp->fd_map = nmap; 1450 } 1451 fdp->fd_nfiles = nnfiles; 1452} 1453 1454/* 1455 * Allocate a file descriptor for the process. 1456 */ 1457int 1458fdalloc(struct thread *td, int minfd, int *result) 1459{ 1460 struct proc *p = td->td_proc; 1461 struct filedesc *fdp = p->p_fd; 1462 int fd = -1, maxfd, allocfd; 1463#ifdef RACCT 1464 int error; 1465#endif 1466 1467 FILEDESC_XLOCK_ASSERT(fdp); 1468 1469 if (fdp->fd_freefile > minfd) 1470 minfd = fdp->fd_freefile; 1471 1472 PROC_LOCK(p); 1473 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1474 PROC_UNLOCK(p); 1475 1476 /* 1477 * Search the bitmap for a free descriptor starting at minfd. 1478 * If none is found, grow the file table. 1479 */ 1480 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1481 if (fd >= maxfd) 1482 return (EMFILE); 1483 if (fd >= fdp->fd_nfiles) { 1484 allocfd = min(fd * 2, maxfd); 1485#ifdef RACCT 1486 PROC_LOCK(p); 1487 error = racct_set(p, RACCT_NOFILE, allocfd); 1488 PROC_UNLOCK(p); 1489 if (error != 0) 1490 return (EMFILE); 1491#endif 1492 /* 1493 * fd is already equal to first free descriptor >= minfd, so 1494 * we only need to grow the table and we are done. 1495 */ 1496 fdgrowtable(fdp, allocfd); 1497 } 1498 1499 /* 1500 * Perform some sanity checks, then mark the file descriptor as 1501 * used and return it to the caller. 1502 */ 1503 KASSERT((unsigned int)fd < min(maxfd, fdp->fd_nfiles), 1504 ("invalid descriptor %d", fd)); 1505 KASSERT(!fdisused(fdp, fd), 1506 ("fd_first_free() returned non-free descriptor")); 1507 KASSERT(fdp->fd_ofiles[fd] == NULL, ("file descriptor isn't free")); 1508 KASSERT(fdp->fd_ofileflags[fd] == 0, ("file flags are set")); 1509 fdused(fdp, fd); 1510 *result = fd; 1511 return (0); 1512} 1513 1514/* 1515 * Check to see whether n user file descriptors are available to the process 1516 * p. 1517 */ 1518int 1519fdavail(struct thread *td, int n) 1520{ 1521 struct proc *p = td->td_proc; 1522 struct filedesc *fdp = td->td_proc->p_fd; 1523 int i, lim, last; 1524 1525 FILEDESC_LOCK_ASSERT(fdp); 1526 1527 /* 1528 * XXX: This is only called from uipc_usrreq.c:unp_externalize(); 1529 * call racct_add() from there instead of dealing with containers 1530 * here. 1531 */ 1532 PROC_LOCK(p); 1533 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1534 PROC_UNLOCK(p); 1535 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1536 return (1); 1537 last = min(fdp->fd_nfiles, lim); 1538 for (i = fdp->fd_freefile; i < last; i++) { 1539 if (fdp->fd_ofiles[i] == NULL && --n <= 0) 1540 return (1); 1541 } 1542 return (0); 1543} 1544 1545/* 1546 * Create a new open file structure and allocate a file decriptor for the 1547 * process that refers to it. We add one reference to the file for the 1548 * descriptor table and one reference for resultfp. This is to prevent us 1549 * being preempted and the entry in the descriptor table closed after we 1550 * release the FILEDESC lock. 1551 */ 1552int 1553falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1554{ 1555 struct file *fp; 1556 int error, fd; 1557 1558 error = falloc_noinstall(td, &fp); 1559 if (error) 1560 return (error); /* no reference held on error */ 1561 1562 error = finstall(td, fp, &fd, flags); 1563 if (error) { 1564 fdrop(fp, td); /* one reference (fp only) */ 1565 return (error); 1566 } 1567 1568 if (resultfp != NULL) 1569 *resultfp = fp; /* copy out result */ 1570 else 1571 fdrop(fp, td); /* release local reference */ 1572 1573 if (resultfd != NULL) 1574 *resultfd = fd; 1575 1576 return (0); 1577} 1578 1579/* 1580 * Create a new open file structure without allocating a file descriptor. 1581 */ 1582int 1583falloc_noinstall(struct thread *td, struct file **resultfp) 1584{ 1585 struct file *fp; 1586 int maxuserfiles = maxfiles - (maxfiles / 20); 1587 static struct timeval lastfail; 1588 static int curfail; 1589 1590 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1591 1592 if ((openfiles >= maxuserfiles && 1593 priv_check(td, PRIV_MAXFILES) != 0) || 1594 openfiles >= maxfiles) { 1595 if (ppsratecheck(&lastfail, &curfail, 1)) { 1596 printf("kern.maxfiles limit exceeded by uid %i, " 1597 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1598 } 1599 return (ENFILE); 1600 } 1601 atomic_add_int(&openfiles, 1); 1602 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1603 refcount_init(&fp->f_count, 1); 1604 fp->f_cred = crhold(td->td_ucred); 1605 fp->f_ops = &badfileops; 1606 fp->f_data = NULL; 1607 fp->f_vnode = NULL; 1608 *resultfp = fp; 1609 return (0); 1610} 1611 1612/* 1613 * Install a file in a file descriptor table. 1614 */ 1615int 1616finstall(struct thread *td, struct file *fp, int *fd, int flags) 1617{ 1618 struct filedesc *fdp = td->td_proc->p_fd; 1619 int error; 1620 1621 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1622 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1623 1624 FILEDESC_XLOCK(fdp); 1625 if ((error = fdalloc(td, 0, fd))) { 1626 FILEDESC_XUNLOCK(fdp); 1627 return (error); 1628 } 1629 fhold(fp); 1630 fdp->fd_ofiles[*fd] = fp; 1631 if ((flags & O_CLOEXEC) != 0) 1632 fdp->fd_ofileflags[*fd] |= UF_EXCLOSE; 1633 FILEDESC_XUNLOCK(fdp); 1634 return (0); 1635} 1636 1637/* 1638 * Build a new filedesc structure from another. 1639 * Copy the current, root, and jail root vnode references. 1640 */ 1641struct filedesc * 1642fdinit(struct filedesc *fdp) 1643{ 1644 struct filedesc0 *newfdp; 1645 1646 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1647 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1648 if (fdp != NULL) { 1649 FILEDESC_XLOCK(fdp); 1650 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1651 if (newfdp->fd_fd.fd_cdir) 1652 VREF(newfdp->fd_fd.fd_cdir); 1653 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1654 if (newfdp->fd_fd.fd_rdir) 1655 VREF(newfdp->fd_fd.fd_rdir); 1656 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1657 if (newfdp->fd_fd.fd_jdir) 1658 VREF(newfdp->fd_fd.fd_jdir); 1659 FILEDESC_XUNLOCK(fdp); 1660 } 1661 1662 /* Create the file descriptor table. */ 1663 newfdp->fd_fd.fd_refcnt = 1; 1664 newfdp->fd_fd.fd_holdcnt = 1; 1665 newfdp->fd_fd.fd_cmask = CMASK; 1666 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1667 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1668 newfdp->fd_fd.fd_nfiles = NDFILE; 1669 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1670 newfdp->fd_fd.fd_lastfile = -1; 1671 return (&newfdp->fd_fd); 1672} 1673 1674static struct filedesc * 1675fdhold(struct proc *p) 1676{ 1677 struct filedesc *fdp; 1678 1679 mtx_lock(&fdesc_mtx); 1680 fdp = p->p_fd; 1681 if (fdp != NULL) 1682 fdp->fd_holdcnt++; 1683 mtx_unlock(&fdesc_mtx); 1684 return (fdp); 1685} 1686 1687static void 1688fddrop(struct filedesc *fdp) 1689{ 1690 struct filedesc0 *fdp0; 1691 struct freetable *ft; 1692 int i; 1693 1694 mtx_lock(&fdesc_mtx); 1695 i = --fdp->fd_holdcnt; 1696 mtx_unlock(&fdesc_mtx); 1697 if (i > 0) 1698 return; 1699 1700 FILEDESC_LOCK_DESTROY(fdp); 1701 fdp0 = (struct filedesc0 *)fdp; 1702 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1703 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1704 free(ft->ft_table, M_FILEDESC); 1705 } 1706 free(fdp, M_FILEDESC); 1707} 1708 1709/* 1710 * Share a filedesc structure. 1711 */ 1712struct filedesc * 1713fdshare(struct filedesc *fdp) 1714{ 1715 1716 FILEDESC_XLOCK(fdp); 1717 fdp->fd_refcnt++; 1718 FILEDESC_XUNLOCK(fdp); 1719 return (fdp); 1720} 1721 1722/* 1723 * Unshare a filedesc structure, if necessary by making a copy 1724 */ 1725void 1726fdunshare(struct proc *p, struct thread *td) 1727{ 1728 1729 FILEDESC_XLOCK(p->p_fd); 1730 if (p->p_fd->fd_refcnt > 1) { 1731 struct filedesc *tmp; 1732 1733 FILEDESC_XUNLOCK(p->p_fd); 1734 tmp = fdcopy(p->p_fd); 1735 fdfree(td); 1736 p->p_fd = tmp; 1737 } else 1738 FILEDESC_XUNLOCK(p->p_fd); 1739} 1740 1741/* 1742 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1743 * this is to ease callers, not catch errors. 1744 */ 1745struct filedesc * 1746fdcopy(struct filedesc *fdp) 1747{ 1748 struct filedesc *newfdp; 1749 int i; 1750 1751 /* Certain daemons might not have file descriptors. */ 1752 if (fdp == NULL) 1753 return (NULL); 1754 1755 newfdp = fdinit(fdp); 1756 FILEDESC_SLOCK(fdp); 1757 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1758 FILEDESC_SUNLOCK(fdp); 1759 FILEDESC_XLOCK(newfdp); 1760 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1761 FILEDESC_XUNLOCK(newfdp); 1762 FILEDESC_SLOCK(fdp); 1763 } 1764 /* copy all passable descriptors (i.e. not kqueue) */ 1765 newfdp->fd_freefile = -1; 1766 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1767 if (fdisused(fdp, i) && 1768 (fdp->fd_ofiles[i]->f_ops->fo_flags & DFLAG_PASSABLE) && 1769 fdp->fd_ofiles[i]->f_ops != &badfileops) { 1770 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1771 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1772 fhold(newfdp->fd_ofiles[i]); 1773 newfdp->fd_lastfile = i; 1774 } else { 1775 if (newfdp->fd_freefile == -1) 1776 newfdp->fd_freefile = i; 1777 } 1778 } 1779 newfdp->fd_cmask = fdp->fd_cmask; 1780 FILEDESC_SUNLOCK(fdp); 1781 FILEDESC_XLOCK(newfdp); 1782 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1783 if (newfdp->fd_ofiles[i] != NULL) 1784 fdused(newfdp, i); 1785 if (newfdp->fd_freefile == -1) 1786 newfdp->fd_freefile = i; 1787 FILEDESC_XUNLOCK(newfdp); 1788 return (newfdp); 1789} 1790 1791/* 1792 * Release a filedesc structure. 1793 */ 1794void 1795fdfree(struct thread *td) 1796{ 1797 struct filedesc *fdp; 1798 int i, locked; 1799 struct filedesc_to_leader *fdtol; 1800 struct file *fp; 1801 struct vnode *cdir, *jdir, *rdir, *vp; 1802 struct flock lf; 1803 1804 /* Certain daemons might not have file descriptors. */ 1805 fdp = td->td_proc->p_fd; 1806 if (fdp == NULL) 1807 return; 1808 1809#ifdef RACCT 1810 PROC_LOCK(td->td_proc); 1811 racct_set(td->td_proc, RACCT_NOFILE, 0); 1812 PROC_UNLOCK(td->td_proc); 1813#endif 1814 1815 /* Check for special need to clear POSIX style locks */ 1816 fdtol = td->td_proc->p_fdtol; 1817 if (fdtol != NULL) { 1818 FILEDESC_XLOCK(fdp); 1819 KASSERT(fdtol->fdl_refcount > 0, 1820 ("filedesc_to_refcount botch: fdl_refcount=%d", 1821 fdtol->fdl_refcount)); 1822 if (fdtol->fdl_refcount == 1 && 1823 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1824 for (i = 0; i <= fdp->fd_lastfile; i++) { 1825 fp = fdp->fd_ofiles[i]; 1826 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1827 continue; 1828 fhold(fp); 1829 FILEDESC_XUNLOCK(fdp); 1830 lf.l_whence = SEEK_SET; 1831 lf.l_start = 0; 1832 lf.l_len = 0; 1833 lf.l_type = F_UNLCK; 1834 vp = fp->f_vnode; 1835 locked = VFS_LOCK_GIANT(vp->v_mount); 1836 (void) VOP_ADVLOCK(vp, 1837 (caddr_t)td->td_proc-> 1838 p_leader, 1839 F_UNLCK, 1840 &lf, 1841 F_POSIX); 1842 VFS_UNLOCK_GIANT(locked); 1843 FILEDESC_XLOCK(fdp); 1844 fdrop(fp, td); 1845 } 1846 } 1847 retry: 1848 if (fdtol->fdl_refcount == 1) { 1849 if (fdp->fd_holdleaderscount > 0 && 1850 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1851 /* 1852 * close() or do_dup() has cleared a reference 1853 * in a shared file descriptor table. 1854 */ 1855 fdp->fd_holdleaderswakeup = 1; 1856 sx_sleep(&fdp->fd_holdleaderscount, 1857 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1858 goto retry; 1859 } 1860 if (fdtol->fdl_holdcount > 0) { 1861 /* 1862 * Ensure that fdtol->fdl_leader remains 1863 * valid in closef(). 1864 */ 1865 fdtol->fdl_wakeup = 1; 1866 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 1867 "fdlhold", 0); 1868 goto retry; 1869 } 1870 } 1871 fdtol->fdl_refcount--; 1872 if (fdtol->fdl_refcount == 0 && 1873 fdtol->fdl_holdcount == 0) { 1874 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1875 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1876 } else 1877 fdtol = NULL; 1878 td->td_proc->p_fdtol = NULL; 1879 FILEDESC_XUNLOCK(fdp); 1880 if (fdtol != NULL) 1881 free(fdtol, M_FILEDESC_TO_LEADER); 1882 } 1883 FILEDESC_XLOCK(fdp); 1884 i = --fdp->fd_refcnt; 1885 FILEDESC_XUNLOCK(fdp); 1886 if (i > 0) 1887 return; 1888 1889 for (i = 0; i <= fdp->fd_lastfile; i++) { 1890 fp = fdp->fd_ofiles[i]; 1891 if (fp != NULL) { 1892 FILEDESC_XLOCK(fdp); 1893 fdp->fd_ofiles[i] = NULL; 1894 FILEDESC_XUNLOCK(fdp); 1895 (void) closef(fp, td); 1896 } 1897 } 1898 FILEDESC_XLOCK(fdp); 1899 1900 /* XXX This should happen earlier. */ 1901 mtx_lock(&fdesc_mtx); 1902 td->td_proc->p_fd = NULL; 1903 mtx_unlock(&fdesc_mtx); 1904 1905 if (fdp->fd_nfiles > NDFILE) 1906 free(fdp->fd_ofiles, M_FILEDESC); 1907 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1908 free(fdp->fd_map, M_FILEDESC); 1909 1910 fdp->fd_nfiles = 0; 1911 1912 cdir = fdp->fd_cdir; 1913 fdp->fd_cdir = NULL; 1914 rdir = fdp->fd_rdir; 1915 fdp->fd_rdir = NULL; 1916 jdir = fdp->fd_jdir; 1917 fdp->fd_jdir = NULL; 1918 FILEDESC_XUNLOCK(fdp); 1919 1920 if (cdir) { 1921 locked = VFS_LOCK_GIANT(cdir->v_mount); 1922 vrele(cdir); 1923 VFS_UNLOCK_GIANT(locked); 1924 } 1925 if (rdir) { 1926 locked = VFS_LOCK_GIANT(rdir->v_mount); 1927 vrele(rdir); 1928 VFS_UNLOCK_GIANT(locked); 1929 } 1930 if (jdir) { 1931 locked = VFS_LOCK_GIANT(jdir->v_mount); 1932 vrele(jdir); 1933 VFS_UNLOCK_GIANT(locked); 1934 } 1935 1936 fddrop(fdp); 1937} 1938 1939/* 1940 * For setugid programs, we don't want to people to use that setugidness 1941 * to generate error messages which write to a file which otherwise would 1942 * otherwise be off-limits to the process. We check for filesystems where 1943 * the vnode can change out from under us after execve (like [lin]procfs). 1944 * 1945 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1946 * sufficient. We also don't check for setugidness since we know we are. 1947 */ 1948static int 1949is_unsafe(struct file *fp) 1950{ 1951 if (fp->f_type == DTYPE_VNODE) { 1952 struct vnode *vp = fp->f_vnode; 1953 1954 if ((vp->v_vflag & VV_PROCDEP) != 0) 1955 return (1); 1956 } 1957 return (0); 1958} 1959 1960/* 1961 * Make this setguid thing safe, if at all possible. 1962 */ 1963void 1964setugidsafety(struct thread *td) 1965{ 1966 struct filedesc *fdp; 1967 int i; 1968 1969 /* Certain daemons might not have file descriptors. */ 1970 fdp = td->td_proc->p_fd; 1971 if (fdp == NULL) 1972 return; 1973 1974 /* 1975 * Note: fdp->fd_ofiles may be reallocated out from under us while 1976 * we are blocked in a close. Be careful! 1977 */ 1978 FILEDESC_XLOCK(fdp); 1979 for (i = 0; i <= fdp->fd_lastfile; i++) { 1980 if (i > 2) 1981 break; 1982 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1983 struct file *fp; 1984 1985 knote_fdclose(td, i); 1986 /* 1987 * NULL-out descriptor prior to close to avoid 1988 * a race while close blocks. 1989 */ 1990 fp = fdp->fd_ofiles[i]; 1991 fdp->fd_ofiles[i] = NULL; 1992 fdp->fd_ofileflags[i] = 0; 1993 fdunused(fdp, i); 1994 FILEDESC_XUNLOCK(fdp); 1995 (void) closef(fp, td); 1996 FILEDESC_XLOCK(fdp); 1997 } 1998 } 1999 FILEDESC_XUNLOCK(fdp); 2000} 2001 2002/* 2003 * If a specific file object occupies a specific file descriptor, close the 2004 * file descriptor entry and drop a reference on the file object. This is a 2005 * convenience function to handle a subsequent error in a function that calls 2006 * falloc() that handles the race that another thread might have closed the 2007 * file descriptor out from under the thread creating the file object. 2008 */ 2009void 2010fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2011{ 2012 2013 FILEDESC_XLOCK(fdp); 2014 if (fdp->fd_ofiles[idx] == fp) { 2015 fdp->fd_ofiles[idx] = NULL; 2016 fdunused(fdp, idx); 2017 FILEDESC_XUNLOCK(fdp); 2018 fdrop(fp, td); 2019 } else 2020 FILEDESC_XUNLOCK(fdp); 2021} 2022 2023/* 2024 * Close any files on exec? 2025 */ 2026void 2027fdcloseexec(struct thread *td) 2028{ 2029 struct filedesc *fdp; 2030 int i; 2031 2032 /* Certain daemons might not have file descriptors. */ 2033 fdp = td->td_proc->p_fd; 2034 if (fdp == NULL) 2035 return; 2036 2037 FILEDESC_XLOCK(fdp); 2038 2039 /* 2040 * We cannot cache fd_ofiles or fd_ofileflags since operations 2041 * may block and rip them out from under us. 2042 */ 2043 for (i = 0; i <= fdp->fd_lastfile; i++) { 2044 if (fdp->fd_ofiles[i] != NULL && 2045 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE || 2046 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) { 2047 struct file *fp; 2048 2049 knote_fdclose(td, i); 2050 /* 2051 * NULL-out descriptor prior to close to avoid 2052 * a race while close blocks. 2053 */ 2054 fp = fdp->fd_ofiles[i]; 2055 fdp->fd_ofiles[i] = NULL; 2056 fdp->fd_ofileflags[i] = 0; 2057 fdunused(fdp, i); 2058 if (fp->f_type == DTYPE_MQUEUE) 2059 mq_fdclose(td, i, fp); 2060 FILEDESC_XUNLOCK(fdp); 2061 (void) closef(fp, td); 2062 FILEDESC_XLOCK(fdp); 2063 } 2064 } 2065 FILEDESC_XUNLOCK(fdp); 2066} 2067 2068/* 2069 * It is unsafe for set[ug]id processes to be started with file 2070 * descriptors 0..2 closed, as these descriptors are given implicit 2071 * significance in the Standard C library. fdcheckstd() will create a 2072 * descriptor referencing /dev/null for each of stdin, stdout, and 2073 * stderr that is not already open. 2074 */ 2075int 2076fdcheckstd(struct thread *td) 2077{ 2078 struct filedesc *fdp; 2079 register_t retval, save; 2080 int i, error, devnull; 2081 2082 fdp = td->td_proc->p_fd; 2083 if (fdp == NULL) 2084 return (0); 2085 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2086 devnull = -1; 2087 error = 0; 2088 for (i = 0; i < 3; i++) { 2089 if (fdp->fd_ofiles[i] != NULL) 2090 continue; 2091 if (devnull < 0) { 2092 save = td->td_retval[0]; 2093 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2094 O_RDWR, 0); 2095 devnull = td->td_retval[0]; 2096 td->td_retval[0] = save; 2097 if (error) 2098 break; 2099 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2100 } else { 2101 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2102 if (error != 0) 2103 break; 2104 } 2105 } 2106 return (error); 2107} 2108 2109/* 2110 * Internal form of close. Decrement reference count on file structure. 2111 * Note: td may be NULL when closing a file that was being passed in a 2112 * message. 2113 * 2114 * XXXRW: Giant is not required for the caller, but often will be held; this 2115 * makes it moderately likely the Giant will be recursed in the VFS case. 2116 */ 2117int 2118closef(struct file *fp, struct thread *td) 2119{ 2120 struct vnode *vp; 2121 struct flock lf; 2122 struct filedesc_to_leader *fdtol; 2123 struct filedesc *fdp; 2124 struct file *fp_object; 2125 2126 /* 2127 * POSIX record locking dictates that any close releases ALL 2128 * locks owned by this process. This is handled by setting 2129 * a flag in the unlock to free ONLY locks obeying POSIX 2130 * semantics, and not to free BSD-style file locks. 2131 * If the descriptor was in a message, POSIX-style locks 2132 * aren't passed with the descriptor, and the thread pointer 2133 * will be NULL. Callers should be careful only to pass a 2134 * NULL thread pointer when there really is no owning 2135 * context that might have locks, or the locks will be 2136 * leaked. 2137 * 2138 * If this is a capability, we do lock processing under the underlying 2139 * node, not the capability itself. 2140 */ 2141 (void)cap_funwrap(fp, 0, &fp_object); 2142 if ((fp_object->f_type == DTYPE_VNODE) && (td != NULL)) { 2143 int vfslocked; 2144 2145 vp = fp_object->f_vnode; 2146 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2147 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2148 lf.l_whence = SEEK_SET; 2149 lf.l_start = 0; 2150 lf.l_len = 0; 2151 lf.l_type = F_UNLCK; 2152 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2153 F_UNLCK, &lf, F_POSIX); 2154 } 2155 fdtol = td->td_proc->p_fdtol; 2156 if (fdtol != NULL) { 2157 /* 2158 * Handle special case where file descriptor table is 2159 * shared between multiple process leaders. 2160 */ 2161 fdp = td->td_proc->p_fd; 2162 FILEDESC_XLOCK(fdp); 2163 for (fdtol = fdtol->fdl_next; 2164 fdtol != td->td_proc->p_fdtol; 2165 fdtol = fdtol->fdl_next) { 2166 if ((fdtol->fdl_leader->p_flag & 2167 P_ADVLOCK) == 0) 2168 continue; 2169 fdtol->fdl_holdcount++; 2170 FILEDESC_XUNLOCK(fdp); 2171 lf.l_whence = SEEK_SET; 2172 lf.l_start = 0; 2173 lf.l_len = 0; 2174 lf.l_type = F_UNLCK; 2175 vp = fp_object->f_vnode; 2176 (void) VOP_ADVLOCK(vp, 2177 (caddr_t)fdtol->fdl_leader, 2178 F_UNLCK, &lf, F_POSIX); 2179 FILEDESC_XLOCK(fdp); 2180 fdtol->fdl_holdcount--; 2181 if (fdtol->fdl_holdcount == 0 && 2182 fdtol->fdl_wakeup != 0) { 2183 fdtol->fdl_wakeup = 0; 2184 wakeup(fdtol); 2185 } 2186 } 2187 FILEDESC_XUNLOCK(fdp); 2188 } 2189 VFS_UNLOCK_GIANT(vfslocked); 2190 } 2191 return (fdrop(fp, td)); 2192} 2193 2194/* 2195 * Initialize the file pointer with the specified properties. 2196 * 2197 * The ops are set with release semantics to be certain that the flags, type, 2198 * and data are visible when ops is. This is to prevent ops methods from being 2199 * called with bad data. 2200 */ 2201void 2202finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2203{ 2204 fp->f_data = data; 2205 fp->f_flag = flag; 2206 fp->f_type = type; 2207 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2208} 2209 2210struct file * 2211fget_unlocked(struct filedesc *fdp, int fd) 2212{ 2213 struct file *fp; 2214 u_int count; 2215 2216 if ((unsigned int)fd >= fdp->fd_nfiles) 2217 return (NULL); 2218 /* 2219 * Fetch the descriptor locklessly. We avoid fdrop() races by 2220 * never raising a refcount above 0. To accomplish this we have 2221 * to use a cmpset loop rather than an atomic_add. The descriptor 2222 * must be re-verified once we acquire a reference to be certain 2223 * that the identity is still correct and we did not lose a race 2224 * due to preemption. 2225 */ 2226 for (;;) { 2227 fp = fdp->fd_ofiles[fd]; 2228 if (fp == NULL) 2229 break; 2230 count = fp->f_count; 2231 if (count == 0) 2232 continue; 2233 /* 2234 * Use an acquire barrier to prevent caching of fd_ofiles 2235 * so it is refreshed for verification. 2236 */ 2237 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2238 continue; 2239 if (fp == fdp->fd_ofiles[fd]) 2240 break; 2241 fdrop(fp, curthread); 2242 } 2243 2244 return (fp); 2245} 2246 2247/* 2248 * Extract the file pointer associated with the specified descriptor for the 2249 * current user process. 2250 * 2251 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2252 * returned. 2253 * 2254 * If the FGET_GETCAP flag is set, the capability itself will be returned. 2255 * Calling _fget() with FGET_GETCAP on a non-capability will return EINVAL. 2256 * Otherwise, if the file is a capability, its rights will be checked against 2257 * the capability rights mask, and if successful, the object will be unwrapped. 2258 * 2259 * If an error occured the non-zero error is returned and *fpp is set to 2260 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2261 * responsible for fdrop(). 2262 */ 2263#define FGET_GETCAP 0x00000001 2264static __inline int 2265_fget(struct thread *td, int fd, struct file **fpp, int flags, 2266 cap_rights_t needrights, cap_rights_t *haverightsp, u_char *maxprotp, 2267 int fget_flags) 2268{ 2269 struct filedesc *fdp; 2270 struct file *fp; 2271#ifdef CAPABILITIES 2272 struct file *fp_fromcap; 2273 int error; 2274#endif 2275 2276 *fpp = NULL; 2277 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2278 return (EBADF); 2279 if ((fp = fget_unlocked(fdp, fd)) == NULL) 2280 return (EBADF); 2281 if (fp->f_ops == &badfileops) { 2282 fdrop(fp, td); 2283 return (EBADF); 2284 } 2285 2286#ifdef CAPABILITIES 2287 /* 2288 * If this is a capability, what rights does it have? 2289 */ 2290 if (haverightsp != NULL) { 2291 if (fp->f_type == DTYPE_CAPABILITY) 2292 *haverightsp = cap_rights(fp); 2293 else 2294 *haverightsp = CAP_MASK_VALID; 2295 } 2296 2297 /* 2298 * If a capability has been requested, return the capability directly. 2299 * Otherwise, check capability rights, extract the underlying object, 2300 * and check its access flags. 2301 */ 2302 if (fget_flags & FGET_GETCAP) { 2303 if (fp->f_type != DTYPE_CAPABILITY) { 2304 fdrop(fp, td); 2305 return (EINVAL); 2306 } 2307 } else { 2308 if (maxprotp == NULL) 2309 error = cap_funwrap(fp, needrights, &fp_fromcap); 2310 else 2311 error = cap_funwrap_mmap(fp, needrights, maxprotp, 2312 &fp_fromcap); 2313 if (error) { 2314 fdrop(fp, td); 2315 return (error); 2316 } 2317 2318 /* 2319 * If we've unwrapped a file, drop the original capability 2320 * and hold the new descriptor. fp after this point refers to 2321 * the actual (unwrapped) object, not the capability. 2322 */ 2323 if (fp != fp_fromcap) { 2324 fhold(fp_fromcap); 2325 fdrop(fp, td); 2326 fp = fp_fromcap; 2327 } 2328 } 2329#else /* !CAPABILITIES */ 2330 KASSERT(fp->f_type != DTYPE_CAPABILITY, 2331 ("%s: saw capability", __func__)); 2332 if (maxprotp != NULL) 2333 *maxprotp = VM_PROT_ALL; 2334#endif /* CAPABILITIES */ 2335 2336 /* 2337 * FREAD and FWRITE failure return EBADF as per POSIX. 2338 * 2339 * Only one flag, or 0, may be specified. 2340 */ 2341 if ((flags == FREAD && (fp->f_flag & FREAD) == 0) || 2342 (flags == FWRITE && (fp->f_flag & FWRITE) == 0)) { 2343 fdrop(fp, td); 2344 return (EBADF); 2345 } 2346 *fpp = fp; 2347 return (0); 2348} 2349 2350int 2351fget(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2352{ 2353 2354 return(_fget(td, fd, fpp, 0, rights, NULL, NULL, 0)); 2355} 2356 2357int 2358fget_mmap(struct thread *td, int fd, cap_rights_t rights, u_char *maxprotp, 2359 struct file **fpp) 2360{ 2361 2362 return (_fget(td, fd, fpp, 0, rights, NULL, maxprotp, 0)); 2363} 2364 2365int 2366fget_read(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2367{ 2368 2369 return(_fget(td, fd, fpp, FREAD, rights, NULL, NULL, 0)); 2370} 2371 2372int 2373fget_write(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2374{ 2375 2376 return (_fget(td, fd, fpp, FWRITE, rights, NULL, NULL, 0)); 2377} 2378 2379/* 2380 * Unlike the other fget() calls, which accept and check capability rights 2381 * but never return capabilities, fgetcap() returns the capability but doesn't 2382 * check capability rights. 2383 */ 2384int 2385fgetcap(struct thread *td, int fd, struct file **fpp) 2386{ 2387 2388 return (_fget(td, fd, fpp, 0, 0, NULL, NULL, FGET_GETCAP)); 2389} 2390 2391 2392/* 2393 * Like fget() but loads the underlying vnode, or returns an error if the 2394 * descriptor does not represent a vnode. Note that pipes use vnodes but 2395 * never have VM objects. The returned vnode will be vref()'d. 2396 * 2397 * XXX: what about the unused flags ? 2398 */ 2399static __inline int 2400_fgetvp(struct thread *td, int fd, int flags, cap_rights_t needrights, 2401 cap_rights_t *haverightsp, struct vnode **vpp) 2402{ 2403 struct file *fp; 2404 int error; 2405 2406 *vpp = NULL; 2407 if ((error = _fget(td, fd, &fp, flags, needrights, haverightsp, 2408 NULL, 0)) != 0) 2409 return (error); 2410 if (fp->f_vnode == NULL) { 2411 error = EINVAL; 2412 } else { 2413 *vpp = fp->f_vnode; 2414 vref(*vpp); 2415 } 2416 fdrop(fp, td); 2417 2418 return (error); 2419} 2420 2421int 2422fgetvp(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp) 2423{ 2424 2425 return (_fgetvp(td, fd, 0, rights, NULL, vpp)); 2426} 2427 2428int 2429fgetvp_rights(struct thread *td, int fd, cap_rights_t need, cap_rights_t *have, 2430 struct vnode **vpp) 2431{ 2432 return (_fgetvp(td, fd, 0, need, have, vpp)); 2433} 2434 2435int 2436fgetvp_read(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp) 2437{ 2438 2439 return (_fgetvp(td, fd, FREAD, rights, NULL, vpp)); 2440} 2441 2442#ifdef notyet 2443int 2444fgetvp_write(struct thread *td, int fd, cap_rights_t rights, 2445 struct vnode **vpp) 2446{ 2447 2448 return (_fgetvp(td, fd, FWRITE, rights, NULL, vpp)); 2449} 2450#endif 2451 2452/* 2453 * Like fget() but loads the underlying socket, or returns an error if the 2454 * descriptor does not represent a socket. 2455 * 2456 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2457 * in the future. 2458 * 2459 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2460 * on their file descriptor reference to prevent the socket from being free'd 2461 * during use. 2462 */ 2463int 2464fgetsock(struct thread *td, int fd, cap_rights_t rights, struct socket **spp, 2465 u_int *fflagp) 2466{ 2467 struct file *fp; 2468 int error; 2469 2470 *spp = NULL; 2471 if (fflagp != NULL) 2472 *fflagp = 0; 2473 if ((error = _fget(td, fd, &fp, 0, rights, NULL, NULL, 0)) != 0) 2474 return (error); 2475 if (fp->f_type != DTYPE_SOCKET) { 2476 error = ENOTSOCK; 2477 } else { 2478 *spp = fp->f_data; 2479 if (fflagp) 2480 *fflagp = fp->f_flag; 2481 SOCK_LOCK(*spp); 2482 soref(*spp); 2483 SOCK_UNLOCK(*spp); 2484 } 2485 fdrop(fp, td); 2486 2487 return (error); 2488} 2489 2490/* 2491 * Drop the reference count on the socket and XXX release the SX lock in the 2492 * future. The last reference closes the socket. 2493 * 2494 * Note: fputsock() is deprecated, see comment for fgetsock(). 2495 */ 2496void 2497fputsock(struct socket *so) 2498{ 2499 2500 ACCEPT_LOCK(); 2501 SOCK_LOCK(so); 2502 CURVNET_SET(so->so_vnet); 2503 sorele(so); 2504 CURVNET_RESTORE(); 2505} 2506 2507/* 2508 * Handle the last reference to a file being closed. 2509 * 2510 * No special capability handling here, as the capability's fo_close will run 2511 * instead of the object here, and perform any necessary drop on the object. 2512 */ 2513int 2514_fdrop(struct file *fp, struct thread *td) 2515{ 2516 int error; 2517 2518 error = 0; 2519 if (fp->f_count != 0) 2520 panic("fdrop: count %d", fp->f_count); 2521 if (fp->f_ops != &badfileops) 2522 error = fo_close(fp, td); 2523 atomic_subtract_int(&openfiles, 1); 2524 crfree(fp->f_cred); 2525 free(fp->f_advice, M_FADVISE); 2526 uma_zfree(file_zone, fp); 2527 2528 return (error); 2529} 2530 2531/* 2532 * Apply an advisory lock on a file descriptor. 2533 * 2534 * Just attempt to get a record lock of the requested type on the entire file 2535 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2536 */ 2537#ifndef _SYS_SYSPROTO_H_ 2538struct flock_args { 2539 int fd; 2540 int how; 2541}; 2542#endif 2543/* ARGSUSED */ 2544int 2545sys_flock(struct thread *td, struct flock_args *uap) 2546{ 2547 struct file *fp; 2548 struct vnode *vp; 2549 struct flock lf; 2550 int vfslocked; 2551 int error; 2552 2553 if ((error = fget(td, uap->fd, CAP_FLOCK, &fp)) != 0) 2554 return (error); 2555 if (fp->f_type != DTYPE_VNODE) { 2556 fdrop(fp, td); 2557 return (EOPNOTSUPP); 2558 } 2559 2560 vp = fp->f_vnode; 2561 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2562 lf.l_whence = SEEK_SET; 2563 lf.l_start = 0; 2564 lf.l_len = 0; 2565 if (uap->how & LOCK_UN) { 2566 lf.l_type = F_UNLCK; 2567 atomic_clear_int(&fp->f_flag, FHASLOCK); 2568 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2569 goto done2; 2570 } 2571 if (uap->how & LOCK_EX) 2572 lf.l_type = F_WRLCK; 2573 else if (uap->how & LOCK_SH) 2574 lf.l_type = F_RDLCK; 2575 else { 2576 error = EBADF; 2577 goto done2; 2578 } 2579 atomic_set_int(&fp->f_flag, FHASLOCK); 2580 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2581 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2582done2: 2583 fdrop(fp, td); 2584 VFS_UNLOCK_GIANT(vfslocked); 2585 return (error); 2586} 2587/* 2588 * Duplicate the specified descriptor to a free descriptor. 2589 */ 2590int 2591dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error) 2592{ 2593 struct file *wfp; 2594 struct file *fp; 2595 2596 KASSERT(error == ENODEV || error == ENXIO, 2597 ("unexpected error %d in %s", error, __func__)); 2598 2599 /* 2600 * If the to-be-dup'd fd number is greater than the allowed number 2601 * of file descriptors, or the fd to be dup'd has already been 2602 * closed, then reject. 2603 */ 2604 FILEDESC_XLOCK(fdp); 2605 if ((unsigned int)dfd >= fdp->fd_nfiles || 2606 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 2607 FILEDESC_XUNLOCK(fdp); 2608 return (EBADF); 2609 } 2610 2611 /* 2612 * There are two cases of interest here. 2613 * 2614 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2615 * 2616 * For ENXIO steal away the file structure from (dfd) and store it in 2617 * (indx). (dfd) is effectively closed by this operation. 2618 */ 2619 fp = fdp->fd_ofiles[indx]; 2620 switch (error) { 2621 case ENODEV: 2622 /* 2623 * Check that the mode the file is being opened for is a 2624 * subset of the mode of the existing descriptor. 2625 */ 2626 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2627 FILEDESC_XUNLOCK(fdp); 2628 return (EACCES); 2629 } 2630 fdp->fd_ofiles[indx] = wfp; 2631 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2632 fhold(wfp); 2633 break; 2634 case ENXIO: 2635 /* 2636 * Steal away the file pointer from dfd and stuff it into indx. 2637 */ 2638 fdp->fd_ofiles[indx] = wfp; 2639 fdp->fd_ofiles[dfd] = NULL; 2640 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2641 fdp->fd_ofileflags[dfd] = 0; 2642 fdunused(fdp, dfd); 2643 break; 2644 } 2645 FILEDESC_XUNLOCK(fdp); 2646 /* 2647 * We now own the reference to fp that the ofiles[] array used to own. 2648 * Release it. 2649 */ 2650 fdrop(fp, td); 2651 return (0); 2652} 2653 2654/* 2655 * Scan all active processes and prisons to see if any of them have a current 2656 * or root directory of `olddp'. If so, replace them with the new mount point. 2657 */ 2658void 2659mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2660{ 2661 struct filedesc *fdp; 2662 struct prison *pr; 2663 struct proc *p; 2664 int nrele; 2665 2666 if (vrefcnt(olddp) == 1) 2667 return; 2668 nrele = 0; 2669 sx_slock(&allproc_lock); 2670 FOREACH_PROC_IN_SYSTEM(p) { 2671 fdp = fdhold(p); 2672 if (fdp == NULL) 2673 continue; 2674 FILEDESC_XLOCK(fdp); 2675 if (fdp->fd_cdir == olddp) { 2676 vref(newdp); 2677 fdp->fd_cdir = newdp; 2678 nrele++; 2679 } 2680 if (fdp->fd_rdir == olddp) { 2681 vref(newdp); 2682 fdp->fd_rdir = newdp; 2683 nrele++; 2684 } 2685 if (fdp->fd_jdir == olddp) { 2686 vref(newdp); 2687 fdp->fd_jdir = newdp; 2688 nrele++; 2689 } 2690 FILEDESC_XUNLOCK(fdp); 2691 fddrop(fdp); 2692 } 2693 sx_sunlock(&allproc_lock); 2694 if (rootvnode == olddp) { 2695 vref(newdp); 2696 rootvnode = newdp; 2697 nrele++; 2698 } 2699 mtx_lock(&prison0.pr_mtx); 2700 if (prison0.pr_root == olddp) { 2701 vref(newdp); 2702 prison0.pr_root = newdp; 2703 nrele++; 2704 } 2705 mtx_unlock(&prison0.pr_mtx); 2706 sx_slock(&allprison_lock); 2707 TAILQ_FOREACH(pr, &allprison, pr_list) { 2708 mtx_lock(&pr->pr_mtx); 2709 if (pr->pr_root == olddp) { 2710 vref(newdp); 2711 pr->pr_root = newdp; 2712 nrele++; 2713 } 2714 mtx_unlock(&pr->pr_mtx); 2715 } 2716 sx_sunlock(&allprison_lock); 2717 while (nrele--) 2718 vrele(olddp); 2719} 2720 2721struct filedesc_to_leader * 2722filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2723{ 2724 struct filedesc_to_leader *fdtol; 2725 2726 fdtol = malloc(sizeof(struct filedesc_to_leader), 2727 M_FILEDESC_TO_LEADER, 2728 M_WAITOK); 2729 fdtol->fdl_refcount = 1; 2730 fdtol->fdl_holdcount = 0; 2731 fdtol->fdl_wakeup = 0; 2732 fdtol->fdl_leader = leader; 2733 if (old != NULL) { 2734 FILEDESC_XLOCK(fdp); 2735 fdtol->fdl_next = old->fdl_next; 2736 fdtol->fdl_prev = old; 2737 old->fdl_next = fdtol; 2738 fdtol->fdl_next->fdl_prev = fdtol; 2739 FILEDESC_XUNLOCK(fdp); 2740 } else { 2741 fdtol->fdl_next = fdtol; 2742 fdtol->fdl_prev = fdtol; 2743 } 2744 return (fdtol); 2745} 2746 2747/* 2748 * Get file structures globally. 2749 */ 2750static int 2751sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2752{ 2753 struct xfile xf; 2754 struct filedesc *fdp; 2755 struct file *fp; 2756 struct proc *p; 2757 int error, n; 2758 2759 error = sysctl_wire_old_buffer(req, 0); 2760 if (error != 0) 2761 return (error); 2762 if (req->oldptr == NULL) { 2763 n = 0; 2764 sx_slock(&allproc_lock); 2765 FOREACH_PROC_IN_SYSTEM(p) { 2766 if (p->p_state == PRS_NEW) 2767 continue; 2768 fdp = fdhold(p); 2769 if (fdp == NULL) 2770 continue; 2771 /* overestimates sparse tables. */ 2772 if (fdp->fd_lastfile > 0) 2773 n += fdp->fd_lastfile; 2774 fddrop(fdp); 2775 } 2776 sx_sunlock(&allproc_lock); 2777 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2778 } 2779 error = 0; 2780 bzero(&xf, sizeof(xf)); 2781 xf.xf_size = sizeof(xf); 2782 sx_slock(&allproc_lock); 2783 FOREACH_PROC_IN_SYSTEM(p) { 2784 PROC_LOCK(p); 2785 if (p->p_state == PRS_NEW) { 2786 PROC_UNLOCK(p); 2787 continue; 2788 } 2789 if (p_cansee(req->td, p) != 0) { 2790 PROC_UNLOCK(p); 2791 continue; 2792 } 2793 xf.xf_pid = p->p_pid; 2794 xf.xf_uid = p->p_ucred->cr_uid; 2795 PROC_UNLOCK(p); 2796 fdp = fdhold(p); 2797 if (fdp == NULL) 2798 continue; 2799 FILEDESC_SLOCK(fdp); 2800 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2801 if ((fp = fdp->fd_ofiles[n]) == NULL) 2802 continue; 2803 xf.xf_fd = n; 2804 xf.xf_file = fp; 2805 xf.xf_data = fp->f_data; 2806 xf.xf_vnode = fp->f_vnode; 2807 xf.xf_type = fp->f_type; 2808 xf.xf_count = fp->f_count; 2809 xf.xf_msgcount = 0; 2810 xf.xf_offset = fp->f_offset; 2811 xf.xf_flag = fp->f_flag; 2812 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2813 if (error) 2814 break; 2815 } 2816 FILEDESC_SUNLOCK(fdp); 2817 fddrop(fdp); 2818 if (error) 2819 break; 2820 } 2821 sx_sunlock(&allproc_lock); 2822 return (error); 2823} 2824 2825SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2826 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2827 2828#ifdef KINFO_OFILE_SIZE 2829CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2830#endif 2831 2832#ifdef COMPAT_FREEBSD7 2833static int 2834export_vnode_for_osysctl(struct vnode *vp, int type, 2835 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2836{ 2837 int error; 2838 char *fullpath, *freepath; 2839 int vfslocked; 2840 2841 bzero(kif, sizeof(*kif)); 2842 kif->kf_structsize = sizeof(*kif); 2843 2844 vref(vp); 2845 kif->kf_fd = type; 2846 kif->kf_type = KF_TYPE_VNODE; 2847 /* This function only handles directories. */ 2848 if (vp->v_type != VDIR) { 2849 vrele(vp); 2850 return (ENOTDIR); 2851 } 2852 kif->kf_vnode_type = KF_VTYPE_VDIR; 2853 2854 /* 2855 * This is not a true file descriptor, so we set a bogus refcount 2856 * and offset to indicate these fields should be ignored. 2857 */ 2858 kif->kf_ref_count = -1; 2859 kif->kf_offset = -1; 2860 2861 freepath = NULL; 2862 fullpath = "-"; 2863 FILEDESC_SUNLOCK(fdp); 2864 vn_fullpath(curthread, vp, &fullpath, &freepath); 2865 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2866 vrele(vp); 2867 VFS_UNLOCK_GIANT(vfslocked); 2868 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2869 if (freepath != NULL) 2870 free(freepath, M_TEMP); 2871 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2872 FILEDESC_SLOCK(fdp); 2873 return (error); 2874} 2875 2876/* 2877 * Get per-process file descriptors for use by procstat(1), et al. 2878 */ 2879static int 2880sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 2881{ 2882 char *fullpath, *freepath; 2883 struct kinfo_ofile *kif; 2884 struct filedesc *fdp; 2885 int error, i, *name; 2886 struct shmfd *shmfd; 2887 struct socket *so; 2888 struct vnode *vp; 2889 struct file *fp; 2890 struct proc *p; 2891 struct tty *tp; 2892 int vfslocked; 2893 2894 name = (int *)arg1; 2895 if ((p = pfind((pid_t)name[0])) == NULL) 2896 return (ESRCH); 2897 if ((error = p_candebug(curthread, p))) { 2898 PROC_UNLOCK(p); 2899 return (error); 2900 } 2901 fdp = fdhold(p); 2902 PROC_UNLOCK(p); 2903 if (fdp == NULL) 2904 return (ENOENT); 2905 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 2906 FILEDESC_SLOCK(fdp); 2907 if (fdp->fd_cdir != NULL) 2908 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 2909 fdp, req); 2910 if (fdp->fd_rdir != NULL) 2911 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 2912 fdp, req); 2913 if (fdp->fd_jdir != NULL) 2914 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 2915 fdp, req); 2916 for (i = 0; i < fdp->fd_nfiles; i++) { 2917 if ((fp = fdp->fd_ofiles[i]) == NULL) 2918 continue; 2919 bzero(kif, sizeof(*kif)); 2920 kif->kf_structsize = sizeof(*kif); 2921 vp = NULL; 2922 so = NULL; 2923 tp = NULL; 2924 shmfd = NULL; 2925 kif->kf_fd = i; 2926 2927#ifdef CAPABILITIES 2928 /* 2929 * When reporting a capability, most fields will be from the 2930 * underlying object, but do mark as a capability. With 2931 * ofiledesc, we don't have a field to export the cap_rights_t, 2932 * but we do with the new filedesc. 2933 */ 2934 if (fp->f_type == DTYPE_CAPABILITY) { 2935 kif->kf_flags |= KF_FLAG_CAPABILITY; 2936 (void)cap_funwrap(fp, 0, &fp); 2937 } 2938#else 2939 KASSERT(fp->f_type != DTYPE_CAPABILITY, 2940 ("sysctl_kern_proc_ofiledesc: saw capability")); 2941#endif 2942 switch (fp->f_type) { 2943 case DTYPE_VNODE: 2944 kif->kf_type = KF_TYPE_VNODE; 2945 vp = fp->f_vnode; 2946 break; 2947 2948 case DTYPE_SOCKET: 2949 kif->kf_type = KF_TYPE_SOCKET; 2950 so = fp->f_data; 2951 break; 2952 2953 case DTYPE_PIPE: 2954 kif->kf_type = KF_TYPE_PIPE; 2955 break; 2956 2957 case DTYPE_FIFO: 2958 kif->kf_type = KF_TYPE_FIFO; 2959 vp = fp->f_vnode; 2960 break; 2961 2962 case DTYPE_KQUEUE: 2963 kif->kf_type = KF_TYPE_KQUEUE; 2964 break; 2965 2966 case DTYPE_CRYPTO: 2967 kif->kf_type = KF_TYPE_CRYPTO; 2968 break; 2969 2970 case DTYPE_MQUEUE: 2971 kif->kf_type = KF_TYPE_MQUEUE; 2972 break; 2973 2974 case DTYPE_SHM: 2975 kif->kf_type = KF_TYPE_SHM; 2976 shmfd = fp->f_data; 2977 break; 2978 2979 case DTYPE_SEM: 2980 kif->kf_type = KF_TYPE_SEM; 2981 break; 2982 2983 case DTYPE_PTS: 2984 kif->kf_type = KF_TYPE_PTS; 2985 tp = fp->f_data; 2986 break; 2987 2988#ifdef PROCDESC 2989 case DTYPE_PROCDESC: 2990 kif->kf_type = KF_TYPE_PROCDESC; 2991 break; 2992#endif 2993 2994 default: 2995 kif->kf_type = KF_TYPE_UNKNOWN; 2996 break; 2997 } 2998 kif->kf_ref_count = fp->f_count; 2999 if (fp->f_flag & FREAD) 3000 kif->kf_flags |= KF_FLAG_READ; 3001 if (fp->f_flag & FWRITE) 3002 kif->kf_flags |= KF_FLAG_WRITE; 3003 if (fp->f_flag & FAPPEND) 3004 kif->kf_flags |= KF_FLAG_APPEND; 3005 if (fp->f_flag & FASYNC) 3006 kif->kf_flags |= KF_FLAG_ASYNC; 3007 if (fp->f_flag & FFSYNC) 3008 kif->kf_flags |= KF_FLAG_FSYNC; 3009 if (fp->f_flag & FNONBLOCK) 3010 kif->kf_flags |= KF_FLAG_NONBLOCK; 3011 if (fp->f_flag & O_DIRECT) 3012 kif->kf_flags |= KF_FLAG_DIRECT; 3013 if (fp->f_flag & FHASLOCK) 3014 kif->kf_flags |= KF_FLAG_HASLOCK; 3015 kif->kf_offset = fp->f_offset; 3016 if (vp != NULL) { 3017 vref(vp); 3018 switch (vp->v_type) { 3019 case VNON: 3020 kif->kf_vnode_type = KF_VTYPE_VNON; 3021 break; 3022 case VREG: 3023 kif->kf_vnode_type = KF_VTYPE_VREG; 3024 break; 3025 case VDIR: 3026 kif->kf_vnode_type = KF_VTYPE_VDIR; 3027 break; 3028 case VBLK: 3029 kif->kf_vnode_type = KF_VTYPE_VBLK; 3030 break; 3031 case VCHR: 3032 kif->kf_vnode_type = KF_VTYPE_VCHR; 3033 break; 3034 case VLNK: 3035 kif->kf_vnode_type = KF_VTYPE_VLNK; 3036 break; 3037 case VSOCK: 3038 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3039 break; 3040 case VFIFO: 3041 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3042 break; 3043 case VBAD: 3044 kif->kf_vnode_type = KF_VTYPE_VBAD; 3045 break; 3046 default: 3047 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3048 break; 3049 } 3050 /* 3051 * It is OK to drop the filedesc lock here as we will 3052 * re-validate and re-evaluate its properties when 3053 * the loop continues. 3054 */ 3055 freepath = NULL; 3056 fullpath = "-"; 3057 FILEDESC_SUNLOCK(fdp); 3058 vn_fullpath(curthread, vp, &fullpath, &freepath); 3059 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3060 vrele(vp); 3061 VFS_UNLOCK_GIANT(vfslocked); 3062 strlcpy(kif->kf_path, fullpath, 3063 sizeof(kif->kf_path)); 3064 if (freepath != NULL) 3065 free(freepath, M_TEMP); 3066 FILEDESC_SLOCK(fdp); 3067 } 3068 if (so != NULL) { 3069 struct sockaddr *sa; 3070 3071 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3072 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3073 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3074 free(sa, M_SONAME); 3075 } 3076 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3077 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3078 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3079 free(sa, M_SONAME); 3080 } 3081 kif->kf_sock_domain = 3082 so->so_proto->pr_domain->dom_family; 3083 kif->kf_sock_type = so->so_type; 3084 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3085 } 3086 if (tp != NULL) { 3087 strlcpy(kif->kf_path, tty_devname(tp), 3088 sizeof(kif->kf_path)); 3089 } 3090 if (shmfd != NULL) 3091 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3092 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3093 if (error) 3094 break; 3095 } 3096 FILEDESC_SUNLOCK(fdp); 3097 fddrop(fdp); 3098 free(kif, M_TEMP); 3099 return (0); 3100} 3101 3102static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD, 3103 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); 3104#endif /* COMPAT_FREEBSD7 */ 3105 3106#ifdef KINFO_FILE_SIZE 3107CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3108#endif 3109 3110static int 3111export_fd_for_sysctl(void *data, int type, int fd, int fflags, int refcnt, 3112 int64_t offset, int fd_is_cap, cap_rights_t fd_cap_rights, 3113 struct kinfo_file *kif, struct sysctl_req *req) 3114{ 3115 struct { 3116 int fflag; 3117 int kf_fflag; 3118 } fflags_table[] = { 3119 { FAPPEND, KF_FLAG_APPEND }, 3120 { FASYNC, KF_FLAG_ASYNC }, 3121 { FFSYNC, KF_FLAG_FSYNC }, 3122 { FHASLOCK, KF_FLAG_HASLOCK }, 3123 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3124 { FREAD, KF_FLAG_READ }, 3125 { FWRITE, KF_FLAG_WRITE }, 3126 { O_CREAT, KF_FLAG_CREAT }, 3127 { O_DIRECT, KF_FLAG_DIRECT }, 3128 { O_EXCL, KF_FLAG_EXCL }, 3129 { O_EXEC, KF_FLAG_EXEC }, 3130 { O_EXLOCK, KF_FLAG_EXLOCK }, 3131 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3132 { O_SHLOCK, KF_FLAG_SHLOCK }, 3133 { O_TRUNC, KF_FLAG_TRUNC } 3134 }; 3135#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3136 struct vnode *vp; 3137 int error, vfslocked; 3138 unsigned int i; 3139 3140 bzero(kif, sizeof(*kif)); 3141 switch (type) { 3142 case KF_TYPE_FIFO: 3143 case KF_TYPE_VNODE: 3144 vp = (struct vnode *)data; 3145 error = fill_vnode_info(vp, kif); 3146 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3147 vrele(vp); 3148 VFS_UNLOCK_GIANT(vfslocked); 3149 break; 3150 case KF_TYPE_SOCKET: 3151 error = fill_socket_info((struct socket *)data, kif); 3152 break; 3153 case KF_TYPE_PIPE: 3154 error = fill_pipe_info((struct pipe *)data, kif); 3155 break; 3156 case KF_TYPE_PTS: 3157 error = fill_pts_info((struct tty *)data, kif); 3158 break; 3159 case KF_TYPE_PROCDESC: 3160 error = fill_procdesc_info((struct procdesc *)data, kif); 3161 break; 3162 case KF_TYPE_SHM: 3163 error = fill_shm_info((struct file *)data, kif); 3164 break; 3165 default: 3166 error = 0; 3167 } 3168 if (error == 0) 3169 kif->kf_status |= KF_ATTR_VALID; 3170 3171 /* 3172 * Translate file access flags. 3173 */ 3174 for (i = 0; i < NFFLAGS; i++) 3175 if (fflags & fflags_table[i].fflag) 3176 kif->kf_flags |= fflags_table[i].kf_fflag; 3177 if (fd_is_cap) 3178 kif->kf_flags |= KF_FLAG_CAPABILITY; 3179 if (fd_is_cap) 3180 kif->kf_cap_rights = fd_cap_rights; 3181 kif->kf_fd = fd; 3182 kif->kf_type = type; 3183 kif->kf_ref_count = refcnt; 3184 kif->kf_offset = offset; 3185 /* Pack record size down */ 3186 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3187 strlen(kif->kf_path) + 1; 3188 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3189 error = SYSCTL_OUT(req, kif, kif->kf_structsize); 3190 return (error); 3191} 3192 3193/* 3194 * Get per-process file descriptors for use by procstat(1), et al. 3195 */ 3196static int 3197sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3198{ 3199 struct file *fp; 3200 struct filedesc *fdp; 3201 struct kinfo_file *kif; 3202 struct proc *p; 3203 struct vnode *cttyvp, *textvp, *tracevp; 3204 size_t oldidx; 3205 int64_t offset; 3206 void *data; 3207 int error, i, *name; 3208 int fd_is_cap, type, refcnt, fflags; 3209 cap_rights_t fd_cap_rights; 3210 3211 name = (int *)arg1; 3212 if ((p = pfind((pid_t)name[0])) == NULL) 3213 return (ESRCH); 3214 if ((error = p_candebug(curthread, p))) { 3215 PROC_UNLOCK(p); 3216 return (error); 3217 } 3218 /* ktrace vnode */ 3219 tracevp = p->p_tracevp; 3220 if (tracevp != NULL) 3221 vref(tracevp); 3222 /* text vnode */ 3223 textvp = p->p_textvp; 3224 if (textvp != NULL) 3225 vref(textvp); 3226 /* Controlling tty. */ 3227 cttyvp = NULL; 3228 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3229 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3230 if (cttyvp != NULL) 3231 vref(cttyvp); 3232 } 3233 fdp = fdhold(p); 3234 PROC_UNLOCK(p); 3235 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3236 if (tracevp != NULL) 3237 export_fd_for_sysctl(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3238 FREAD | FWRITE, -1, -1, 0, 0, kif, req); 3239 if (textvp != NULL) 3240 export_fd_for_sysctl(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3241 FREAD, -1, -1, 0, 0, kif, req); 3242 if (cttyvp != NULL) 3243 export_fd_for_sysctl(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3244 FREAD | FWRITE, -1, -1, 0, 0, kif, req); 3245 if (fdp == NULL) 3246 goto fail; 3247 FILEDESC_SLOCK(fdp); 3248 /* working directory */ 3249 if (fdp->fd_cdir != NULL) { 3250 vref(fdp->fd_cdir); 3251 data = fdp->fd_cdir; 3252 FILEDESC_SUNLOCK(fdp); 3253 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3254 FREAD, -1, -1, 0, 0, kif, req); 3255 FILEDESC_SLOCK(fdp); 3256 } 3257 /* root directory */ 3258 if (fdp->fd_rdir != NULL) { 3259 vref(fdp->fd_rdir); 3260 data = fdp->fd_rdir; 3261 FILEDESC_SUNLOCK(fdp); 3262 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3263 FREAD, -1, -1, 0, 0, kif, req); 3264 FILEDESC_SLOCK(fdp); 3265 } 3266 /* jail directory */ 3267 if (fdp->fd_jdir != NULL) { 3268 vref(fdp->fd_jdir); 3269 data = fdp->fd_jdir; 3270 FILEDESC_SUNLOCK(fdp); 3271 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3272 FREAD, -1, -1, 0, 0, kif, req); 3273 FILEDESC_SLOCK(fdp); 3274 } 3275 for (i = 0; i < fdp->fd_nfiles; i++) { 3276 if ((fp = fdp->fd_ofiles[i]) == NULL) 3277 continue; 3278 data = NULL; 3279 fd_is_cap = 0; 3280 fd_cap_rights = 0; 3281 3282#ifdef CAPABILITIES 3283 /* 3284 * When reporting a capability, most fields will be from the 3285 * underlying object, but do mark as a capability and export 3286 * the capability rights mask. 3287 */ 3288 if (fp->f_type == DTYPE_CAPABILITY) { 3289 fd_is_cap = 1; 3290 fd_cap_rights = cap_rights(fp); 3291 (void)cap_funwrap(fp, 0, &fp); 3292 } 3293#else /* !CAPABILITIES */ 3294 KASSERT(fp->f_type != DTYPE_CAPABILITY, 3295 ("sysctl_kern_proc_filedesc: saw capability")); 3296#endif 3297 switch (fp->f_type) { 3298 case DTYPE_VNODE: 3299 type = KF_TYPE_VNODE; 3300 vref(fp->f_vnode); 3301 data = fp->f_vnode; 3302 break; 3303 3304 case DTYPE_SOCKET: 3305 type = KF_TYPE_SOCKET; 3306 data = fp->f_data; 3307 break; 3308 3309 case DTYPE_PIPE: 3310 type = KF_TYPE_PIPE; 3311 data = fp->f_data; 3312 break; 3313 3314 case DTYPE_FIFO: 3315 type = KF_TYPE_FIFO; 3316 vref(fp->f_vnode); 3317 data = fp->f_vnode; 3318 break; 3319 3320 case DTYPE_KQUEUE: 3321 type = KF_TYPE_KQUEUE; 3322 break; 3323 3324 case DTYPE_CRYPTO: 3325 type = KF_TYPE_CRYPTO; 3326 break; 3327 3328 case DTYPE_MQUEUE: 3329 type = KF_TYPE_MQUEUE; 3330 break; 3331 3332 case DTYPE_SHM: 3333 type = KF_TYPE_SHM; 3334 data = fp; 3335 break; 3336 3337 case DTYPE_SEM: 3338 type = KF_TYPE_SEM; 3339 break; 3340 3341 case DTYPE_PTS: 3342 type = KF_TYPE_PTS; 3343 data = fp->f_data; 3344 break; 3345 3346#ifdef PROCDESC 3347 case DTYPE_PROCDESC: 3348 type = KF_TYPE_PROCDESC; 3349 data = fp->f_data; 3350 break; 3351#endif 3352 3353 default: 3354 type = KF_TYPE_UNKNOWN; 3355 break; 3356 } 3357 refcnt = fp->f_count; 3358 fflags = fp->f_flag; 3359 offset = fp->f_offset; 3360 3361 /* 3362 * Create sysctl entry. 3363 * It is OK to drop the filedesc lock here as we will 3364 * re-validate and re-evaluate its properties when 3365 * the loop continues. 3366 */ 3367 oldidx = req->oldidx; 3368 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO) 3369 FILEDESC_SUNLOCK(fdp); 3370 error = export_fd_for_sysctl(data, type, i, fflags, refcnt, 3371 offset, fd_is_cap, fd_cap_rights, kif, req); 3372 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO) 3373 FILEDESC_SLOCK(fdp); 3374 if (error) { 3375 if (error == ENOMEM) { 3376 /* 3377 * The hack to keep the ABI of sysctl 3378 * kern.proc.filedesc intact, but not 3379 * to account a partially copied 3380 * kinfo_file into the oldidx. 3381 */ 3382 req->oldidx = oldidx; 3383 error = 0; 3384 } 3385 break; 3386 } 3387 } 3388 FILEDESC_SUNLOCK(fdp); 3389fail: 3390 if (fdp != NULL) 3391 fddrop(fdp); 3392 free(kif, M_TEMP); 3393 return (error); 3394} 3395 3396int 3397vntype_to_kinfo(int vtype) 3398{ 3399 struct { 3400 int vtype; 3401 int kf_vtype; 3402 } vtypes_table[] = { 3403 { VBAD, KF_VTYPE_VBAD }, 3404 { VBLK, KF_VTYPE_VBLK }, 3405 { VCHR, KF_VTYPE_VCHR }, 3406 { VDIR, KF_VTYPE_VDIR }, 3407 { VFIFO, KF_VTYPE_VFIFO }, 3408 { VLNK, KF_VTYPE_VLNK }, 3409 { VNON, KF_VTYPE_VNON }, 3410 { VREG, KF_VTYPE_VREG }, 3411 { VSOCK, KF_VTYPE_VSOCK } 3412 }; 3413#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3414 unsigned int i; 3415 3416 /* 3417 * Perform vtype translation. 3418 */ 3419 for (i = 0; i < NVTYPES; i++) 3420 if (vtypes_table[i].vtype == vtype) 3421 break; 3422 if (i < NVTYPES) 3423 return (vtypes_table[i].kf_vtype); 3424 3425 return (KF_VTYPE_UNKNOWN); 3426} 3427 3428static int 3429fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3430{ 3431 struct vattr va; 3432 char *fullpath, *freepath; 3433 int error, vfslocked; 3434 3435 if (vp == NULL) 3436 return (1); 3437 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3438 freepath = NULL; 3439 fullpath = "-"; 3440 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3441 if (error == 0) { 3442 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3443 } 3444 if (freepath != NULL) 3445 free(freepath, M_TEMP); 3446 3447 /* 3448 * Retrieve vnode attributes. 3449 */ 3450 va.va_fsid = VNOVAL; 3451 va.va_rdev = NODEV; 3452 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3453 vn_lock(vp, LK_SHARED | LK_RETRY); 3454 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3455 VOP_UNLOCK(vp, 0); 3456 VFS_UNLOCK_GIANT(vfslocked); 3457 if (error != 0) 3458 return (error); 3459 if (va.va_fsid != VNOVAL) 3460 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3461 else 3462 kif->kf_un.kf_file.kf_file_fsid = 3463 vp->v_mount->mnt_stat.f_fsid.val[0]; 3464 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3465 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3466 kif->kf_un.kf_file.kf_file_size = va.va_size; 3467 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3468 return (0); 3469} 3470 3471static int 3472fill_socket_info(struct socket *so, struct kinfo_file *kif) 3473{ 3474 struct sockaddr *sa; 3475 struct inpcb *inpcb; 3476 struct unpcb *unpcb; 3477 int error; 3478 3479 if (so == NULL) 3480 return (1); 3481 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3482 kif->kf_sock_type = so->so_type; 3483 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3484 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3485 switch(kif->kf_sock_domain) { 3486 case AF_INET: 3487 case AF_INET6: 3488 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3489 if (so->so_pcb != NULL) { 3490 inpcb = (struct inpcb *)(so->so_pcb); 3491 kif->kf_un.kf_sock.kf_sock_inpcb = 3492 (uintptr_t)inpcb->inp_ppcb; 3493 } 3494 } 3495 break; 3496 case AF_UNIX: 3497 if (so->so_pcb != NULL) { 3498 unpcb = (struct unpcb *)(so->so_pcb); 3499 if (unpcb->unp_conn) { 3500 kif->kf_un.kf_sock.kf_sock_unpconn = 3501 (uintptr_t)unpcb->unp_conn; 3502 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3503 so->so_rcv.sb_state; 3504 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3505 so->so_snd.sb_state; 3506 } 3507 } 3508 break; 3509 } 3510 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3511 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3512 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3513 free(sa, M_SONAME); 3514 } 3515 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3516 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3517 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3518 free(sa, M_SONAME); 3519 } 3520 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3521 sizeof(kif->kf_path)); 3522 return (0); 3523} 3524 3525static int 3526fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3527{ 3528 3529 if (tp == NULL) 3530 return (1); 3531 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3532 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3533 return (0); 3534} 3535 3536static int 3537fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3538{ 3539 3540 if (pi == NULL) 3541 return (1); 3542 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3543 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3544 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3545 return (0); 3546} 3547 3548static int 3549fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3550{ 3551 3552 if (pdp == NULL) 3553 return (1); 3554 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3555 return (0); 3556} 3557 3558static int 3559fill_shm_info(struct file *fp, struct kinfo_file *kif) 3560{ 3561 struct thread *td; 3562 struct stat sb; 3563 3564 td = curthread; 3565 if (fp->f_data == NULL) 3566 return (1); 3567 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3568 return (1); 3569 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3570 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3571 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3572 return (0); 3573} 3574 3575static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD, 3576 sysctl_kern_proc_filedesc, "Process filedesc entries"); 3577 3578#ifdef DDB 3579/* 3580 * For the purposes of debugging, generate a human-readable string for the 3581 * file type. 3582 */ 3583static const char * 3584file_type_to_name(short type) 3585{ 3586 3587 switch (type) { 3588 case 0: 3589 return ("zero"); 3590 case DTYPE_VNODE: 3591 return ("vnod"); 3592 case DTYPE_SOCKET: 3593 return ("sock"); 3594 case DTYPE_PIPE: 3595 return ("pipe"); 3596 case DTYPE_FIFO: 3597 return ("fifo"); 3598 case DTYPE_KQUEUE: 3599 return ("kque"); 3600 case DTYPE_CRYPTO: 3601 return ("crpt"); 3602 case DTYPE_MQUEUE: 3603 return ("mque"); 3604 case DTYPE_SHM: 3605 return ("shm"); 3606 case DTYPE_SEM: 3607 return ("ksem"); 3608 default: 3609 return ("unkn"); 3610 } 3611} 3612 3613/* 3614 * For the purposes of debugging, identify a process (if any, perhaps one of 3615 * many) that references the passed file in its file descriptor array. Return 3616 * NULL if none. 3617 */ 3618static struct proc * 3619file_to_first_proc(struct file *fp) 3620{ 3621 struct filedesc *fdp; 3622 struct proc *p; 3623 int n; 3624 3625 FOREACH_PROC_IN_SYSTEM(p) { 3626 if (p->p_state == PRS_NEW) 3627 continue; 3628 fdp = p->p_fd; 3629 if (fdp == NULL) 3630 continue; 3631 for (n = 0; n < fdp->fd_nfiles; n++) { 3632 if (fp == fdp->fd_ofiles[n]) 3633 return (p); 3634 } 3635 } 3636 return (NULL); 3637} 3638 3639static void 3640db_print_file(struct file *fp, int header) 3641{ 3642 struct proc *p; 3643 3644 if (header) 3645 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3646 "File", "Type", "Data", "Flag", "GCFl", "Count", 3647 "MCount", "Vnode", "FPID", "FCmd"); 3648 p = file_to_first_proc(fp); 3649 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3650 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3651 0, fp->f_count, 0, fp->f_vnode, 3652 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3653} 3654 3655DB_SHOW_COMMAND(file, db_show_file) 3656{ 3657 struct file *fp; 3658 3659 if (!have_addr) { 3660 db_printf("usage: show file <addr>\n"); 3661 return; 3662 } 3663 fp = (struct file *)addr; 3664 db_print_file(fp, 1); 3665} 3666 3667DB_SHOW_COMMAND(files, db_show_files) 3668{ 3669 struct filedesc *fdp; 3670 struct file *fp; 3671 struct proc *p; 3672 int header; 3673 int n; 3674 3675 header = 1; 3676 FOREACH_PROC_IN_SYSTEM(p) { 3677 if (p->p_state == PRS_NEW) 3678 continue; 3679 if ((fdp = p->p_fd) == NULL) 3680 continue; 3681 for (n = 0; n < fdp->fd_nfiles; ++n) { 3682 if ((fp = fdp->fd_ofiles[n]) == NULL) 3683 continue; 3684 db_print_file(fp, header); 3685 header = 0; 3686 } 3687 } 3688} 3689#endif 3690 3691SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3692 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3693 3694SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3695 &maxfiles, 0, "Maximum number of files"); 3696 3697SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3698 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3699 3700/* ARGSUSED*/ 3701static void 3702filelistinit(void *dummy) 3703{ 3704 3705 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3706 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3707 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3708 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3709} 3710SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3711 3712/*-------------------------------------------------------------------*/ 3713 3714static int 3715badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3716 int flags, struct thread *td) 3717{ 3718 3719 return (EBADF); 3720} 3721 3722static int 3723badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3724 struct thread *td) 3725{ 3726 3727 return (EINVAL); 3728} 3729 3730static int 3731badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3732 struct thread *td) 3733{ 3734 3735 return (EBADF); 3736} 3737 3738static int 3739badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3740 struct thread *td) 3741{ 3742 3743 return (0); 3744} 3745 3746static int 3747badfo_kqfilter(struct file *fp, struct knote *kn) 3748{ 3749 3750 return (EBADF); 3751} 3752 3753static int 3754badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3755 struct thread *td) 3756{ 3757 3758 return (EBADF); 3759} 3760 3761static int 3762badfo_close(struct file *fp, struct thread *td) 3763{ 3764 3765 return (EBADF); 3766} 3767 3768static int 3769badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3770 struct thread *td) 3771{ 3772 3773 return (EBADF); 3774} 3775 3776static int 3777badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3778 struct thread *td) 3779{ 3780 3781 return (EBADF); 3782} 3783 3784struct fileops badfileops = { 3785 .fo_read = badfo_readwrite, 3786 .fo_write = badfo_readwrite, 3787 .fo_truncate = badfo_truncate, 3788 .fo_ioctl = badfo_ioctl, 3789 .fo_poll = badfo_poll, 3790 .fo_kqfilter = badfo_kqfilter, 3791 .fo_stat = badfo_stat, 3792 .fo_close = badfo_close, 3793 .fo_chmod = badfo_chmod, 3794 .fo_chown = badfo_chown, 3795}; 3796 3797int 3798invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3799 struct thread *td) 3800{ 3801 3802 return (EINVAL); 3803} 3804 3805int 3806invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3807 struct thread *td) 3808{ 3809 3810 return (EINVAL); 3811} 3812 3813/*-------------------------------------------------------------------*/ 3814 3815/* 3816 * File Descriptor pseudo-device driver (/dev/fd/). 3817 * 3818 * Opening minor device N dup()s the file (if any) connected to file 3819 * descriptor N belonging to the calling process. Note that this driver 3820 * consists of only the ``open()'' routine, because all subsequent 3821 * references to this file will be direct to the other driver. 3822 * 3823 * XXX: we could give this one a cloning event handler if necessary. 3824 */ 3825 3826/* ARGSUSED */ 3827static int 3828fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3829{ 3830 3831 /* 3832 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3833 * the file descriptor being sought for duplication. The error 3834 * return ensures that the vnode for this device will be released 3835 * by vn_open. Open will detect this special error and take the 3836 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3837 * will simply report the error. 3838 */ 3839 td->td_dupfd = dev2unit(dev); 3840 return (ENODEV); 3841} 3842 3843static struct cdevsw fildesc_cdevsw = { 3844 .d_version = D_VERSION, 3845 .d_open = fdopen, 3846 .d_name = "FD", 3847}; 3848 3849static void 3850fildesc_drvinit(void *unused) 3851{ 3852 struct cdev *dev; 3853 3854 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3855 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3856 make_dev_alias(dev, "stdin"); 3857 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3858 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3859 make_dev_alias(dev, "stdout"); 3860 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3861 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3862 make_dev_alias(dev, "stderr"); 3863} 3864 3865SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3866