kern_descrip.c revision 237073
1212904Sdim/*- 2212904Sdim * 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 237073 2012-06-14 15:26:23Z 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, int holdleaders); 119static int do_dup(struct thread *td, int flags, int old, int new, 120 register_t *retval); 121static int fd_first_free(struct filedesc *fdp, int low, int size); 122static int fd_last_used(struct filedesc *fdp, int size); 123static void fdgrowtable(struct filedesc *fdp, int nfd); 124static void fdunused(struct filedesc *fdp, int fd); 125static void fdused(struct filedesc *fdp, int fd); 126static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif); 127static int fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif); 128static int fill_pts_info(struct tty *tp, struct kinfo_file *kif); 129static int fill_shm_info(struct file *fp, struct kinfo_file *kif); 130static int fill_socket_info(struct socket *so, struct kinfo_file *kif); 131static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif); 132 133/* 134 * A process is initially started out with NDFILE descriptors stored within 135 * this structure, selected to be enough for typical applications based on 136 * the historical limit of 20 open files (and the usage of descriptors by 137 * shells). If these descriptors are exhausted, a larger descriptor table 138 * may be allocated, up to a process' resource limit; the internal arrays 139 * are then unused. 140 */ 141#define NDFILE 20 142#define NDSLOTSIZE sizeof(NDSLOTTYPE) 143#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 144#define NDSLOT(x) ((x) / NDENTRIES) 145#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 146#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 147 148/* 149 * Storage required per open file descriptor. 150 */ 151#define OFILESIZE (sizeof(struct file *) + sizeof(char)) 152 153/* 154 * Storage to hold unused ofiles that need to be reclaimed. 155 */ 156struct freetable { 157 struct file **ft_table; 158 SLIST_ENTRY(freetable) ft_next; 159}; 160 161/* 162 * Basic allocation of descriptors: 163 * one of the above, plus arrays for NDFILE descriptors. 164 */ 165struct filedesc0 { 166 struct filedesc fd_fd; 167 /* 168 * ofiles which need to be reclaimed on free. 169 */ 170 SLIST_HEAD(,freetable) fd_free; 171 /* 172 * These arrays are used when the number of open files is 173 * <= NDFILE, and are then pointed to by the pointers above. 174 */ 175 struct file *fd_dfiles[NDFILE]; 176 char fd_dfileflags[NDFILE]; 177 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 178}; 179 180/* 181 * Descriptor management. 182 */ 183volatile int openfiles; /* actual number of open files */ 184struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 185void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 186 187/* A mutex to protect the association between a proc and filedesc. */ 188static struct mtx fdesc_mtx; 189 190/* 191 * If low >= size, just return low. Otherwise find the first zero bit in the 192 * given bitmap, starting at low and not exceeding size - 1. Return size if 193 * not found. 194 */ 195static int 196fd_first_free(struct filedesc *fdp, int low, int size) 197{ 198 NDSLOTTYPE *map = fdp->fd_map; 199 NDSLOTTYPE mask; 200 int off, maxoff; 201 202 if (low >= size) 203 return (low); 204 205 off = NDSLOT(low); 206 if (low % NDENTRIES) { 207 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 208 if ((mask &= ~map[off]) != 0UL) 209 return (off * NDENTRIES + ffsl(mask) - 1); 210 ++off; 211 } 212 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 213 if (map[off] != ~0UL) 214 return (off * NDENTRIES + ffsl(~map[off]) - 1); 215 return (size); 216} 217 218/* 219 * Find the highest non-zero bit in the given bitmap, starting at 0 and 220 * not exceeding size - 1. Return -1 if not found. 221 */ 222static int 223fd_last_used(struct filedesc *fdp, int size) 224{ 225 NDSLOTTYPE *map = fdp->fd_map; 226 NDSLOTTYPE mask; 227 int off, minoff; 228 229 off = NDSLOT(size); 230 if (size % NDENTRIES) { 231 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 232 if ((mask &= map[off]) != 0) 233 return (off * NDENTRIES + flsl(mask) - 1); 234 --off; 235 } 236 for (minoff = NDSLOT(0); off >= minoff; --off) 237 if (map[off] != 0) 238 return (off * NDENTRIES + flsl(map[off]) - 1); 239 return (-1); 240} 241 242static int 243fdisused(struct filedesc *fdp, int fd) 244{ 245 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 246 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 247 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 248} 249 250/* 251 * Mark a file descriptor as used. 252 */ 253static void 254fdused(struct filedesc *fdp, int fd) 255{ 256 257 FILEDESC_XLOCK_ASSERT(fdp); 258 KASSERT(!fdisused(fdp, fd), 259 ("fd already used")); 260 261 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 262 if (fd > fdp->fd_lastfile) 263 fdp->fd_lastfile = fd; 264 if (fd == fdp->fd_freefile) 265 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 266} 267 268/* 269 * Mark a file descriptor as unused. 270 */ 271static void 272fdunused(struct filedesc *fdp, int fd) 273{ 274 275 FILEDESC_XLOCK_ASSERT(fdp); 276 KASSERT(fdisused(fdp, fd), 277 ("fd is already unused")); 278 KASSERT(fdp->fd_ofiles[fd] == NULL, 279 ("fd is still in use")); 280 281 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 282 if (fd < fdp->fd_freefile) 283 fdp->fd_freefile = fd; 284 if (fd == fdp->fd_lastfile) 285 fdp->fd_lastfile = fd_last_used(fdp, fd); 286} 287 288/* 289 * System calls on descriptors. 290 */ 291#ifndef _SYS_SYSPROTO_H_ 292struct getdtablesize_args { 293 int dummy; 294}; 295#endif 296/* ARGSUSED */ 297int 298sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 299{ 300 struct proc *p = td->td_proc; 301 uint64_t lim; 302 303 PROC_LOCK(p); 304 td->td_retval[0] = 305 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 306 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 307 PROC_UNLOCK(p); 308 if (lim < td->td_retval[0]) 309 td->td_retval[0] = lim; 310 return (0); 311} 312 313/* 314 * Duplicate a file descriptor to a particular value. 315 * 316 * Note: keep in mind that a potential race condition exists when closing 317 * descriptors from a shared descriptor table (via rfork). 318 */ 319#ifndef _SYS_SYSPROTO_H_ 320struct dup2_args { 321 u_int from; 322 u_int to; 323}; 324#endif 325/* ARGSUSED */ 326int 327sys_dup2(struct thread *td, struct dup2_args *uap) 328{ 329 330 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 331 td->td_retval)); 332} 333 334/* 335 * Duplicate a file descriptor. 336 */ 337#ifndef _SYS_SYSPROTO_H_ 338struct dup_args { 339 u_int fd; 340}; 341#endif 342/* ARGSUSED */ 343int 344sys_dup(struct thread *td, struct dup_args *uap) 345{ 346 347 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 348} 349 350/* 351 * The file control system call. 352 */ 353#ifndef _SYS_SYSPROTO_H_ 354struct fcntl_args { 355 int fd; 356 int cmd; 357 long arg; 358}; 359#endif 360/* ARGSUSED */ 361int 362sys_fcntl(struct thread *td, struct fcntl_args *uap) 363{ 364 struct flock fl; 365 struct oflock ofl; 366 intptr_t arg; 367 int error; 368 int cmd; 369 370 error = 0; 371 cmd = uap->cmd; 372 switch (uap->cmd) { 373 case F_OGETLK: 374 case F_OSETLK: 375 case F_OSETLKW: 376 /* 377 * Convert old flock structure to new. 378 */ 379 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 380 fl.l_start = ofl.l_start; 381 fl.l_len = ofl.l_len; 382 fl.l_pid = ofl.l_pid; 383 fl.l_type = ofl.l_type; 384 fl.l_whence = ofl.l_whence; 385 fl.l_sysid = 0; 386 387 switch (uap->cmd) { 388 case F_OGETLK: 389 cmd = F_GETLK; 390 break; 391 case F_OSETLK: 392 cmd = F_SETLK; 393 break; 394 case F_OSETLKW: 395 cmd = F_SETLKW; 396 break; 397 } 398 arg = (intptr_t)&fl; 399 break; 400 case F_GETLK: 401 case F_SETLK: 402 case F_SETLKW: 403 case F_SETLK_REMOTE: 404 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 405 arg = (intptr_t)&fl; 406 break; 407 default: 408 arg = uap->arg; 409 break; 410 } 411 if (error) 412 return (error); 413 error = kern_fcntl(td, uap->fd, cmd, arg); 414 if (error) 415 return (error); 416 if (uap->cmd == F_OGETLK) { 417 ofl.l_start = fl.l_start; 418 ofl.l_len = fl.l_len; 419 ofl.l_pid = fl.l_pid; 420 ofl.l_type = fl.l_type; 421 ofl.l_whence = fl.l_whence; 422 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 423 } else if (uap->cmd == F_GETLK) { 424 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 425 } 426 return (error); 427} 428 429static inline struct file * 430fdtofp(int fd, struct filedesc *fdp) 431{ 432 433 FILEDESC_LOCK_ASSERT(fdp); 434 435 if (fd < 0 || fd >= fdp->fd_nfiles) 436 return (NULL); 437 438 return (fdp->fd_ofiles[fd]); 439} 440 441static inline int 442fdunwrap(int fd, cap_rights_t rights, struct filedesc *fdp, struct file **fpp) 443{ 444 445 *fpp = fdtofp(fd, fdp); 446 if (*fpp == NULL) 447 return (EBADF); 448 449#ifdef CAPABILITIES 450 if ((*fpp)->f_type == DTYPE_CAPABILITY) { 451 int err = cap_funwrap(*fpp, rights, fpp); 452 if (err != 0) { 453 *fpp = NULL; 454 return (err); 455 } 456 } 457#endif /* CAPABILITIES */ 458 return (0); 459} 460 461int 462kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 463{ 464 struct filedesc *fdp; 465 struct flock *flp; 466 struct file *fp; 467 struct proc *p; 468 char *pop; 469 struct vnode *vp; 470 int error, flg, tmp; 471 int vfslocked; 472 u_int old, new; 473 uint64_t bsize; 474 475 vfslocked = 0; 476 error = 0; 477 flg = F_POSIX; 478 p = td->td_proc; 479 fdp = p->p_fd; 480 481 switch (cmd) { 482 case F_DUPFD: 483 tmp = arg; 484 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 485 break; 486 487 case F_DUP2FD: 488 tmp = arg; 489 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 490 break; 491 492 case F_GETFD: 493 FILEDESC_SLOCK(fdp); 494 if ((fp = fdtofp(fd, fdp)) == NULL) { 495 FILEDESC_SUNLOCK(fdp); 496 error = EBADF; 497 break; 498 } 499 pop = &fdp->fd_ofileflags[fd]; 500 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 501 FILEDESC_SUNLOCK(fdp); 502 break; 503 504 case F_SETFD: 505 FILEDESC_XLOCK(fdp); 506 if ((fp = fdtofp(fd, fdp)) == NULL) { 507 FILEDESC_XUNLOCK(fdp); 508 error = EBADF; 509 break; 510 } 511 pop = &fdp->fd_ofileflags[fd]; 512 *pop = (*pop &~ UF_EXCLOSE) | 513 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 514 FILEDESC_XUNLOCK(fdp); 515 break; 516 517 case F_GETFL: 518 FILEDESC_SLOCK(fdp); 519 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 520 if (error != 0) { 521 FILEDESC_SUNLOCK(fdp); 522 break; 523 } 524 td->td_retval[0] = OFLAGS(fp->f_flag); 525 FILEDESC_SUNLOCK(fdp); 526 break; 527 528 case F_SETFL: 529 FILEDESC_SLOCK(fdp); 530 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 531 if (error != 0) { 532 FILEDESC_SUNLOCK(fdp); 533 break; 534 } 535 fhold(fp); 536 FILEDESC_SUNLOCK(fdp); 537 do { 538 tmp = flg = fp->f_flag; 539 tmp &= ~FCNTLFLAGS; 540 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 541 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 542 tmp = fp->f_flag & FNONBLOCK; 543 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 544 if (error) { 545 fdrop(fp, td); 546 break; 547 } 548 tmp = fp->f_flag & FASYNC; 549 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 550 if (error == 0) { 551 fdrop(fp, td); 552 break; 553 } 554 atomic_clear_int(&fp->f_flag, FNONBLOCK); 555 tmp = 0; 556 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 557 fdrop(fp, td); 558 break; 559 560 case F_GETOWN: 561 FILEDESC_SLOCK(fdp); 562 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 563 if (error != 0) { 564 FILEDESC_SUNLOCK(fdp); 565 break; 566 } 567 fhold(fp); 568 FILEDESC_SUNLOCK(fdp); 569 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 570 if (error == 0) 571 td->td_retval[0] = tmp; 572 fdrop(fp, td); 573 break; 574 575 case F_SETOWN: 576 FILEDESC_SLOCK(fdp); 577 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 578 if (error != 0) { 579 FILEDESC_SUNLOCK(fdp); 580 break; 581 } 582 fhold(fp); 583 FILEDESC_SUNLOCK(fdp); 584 tmp = arg; 585 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 586 fdrop(fp, td); 587 break; 588 589 case F_SETLK_REMOTE: 590 error = priv_check(td, PRIV_NFS_LOCKD); 591 if (error) 592 return (error); 593 flg = F_REMOTE; 594 goto do_setlk; 595 596 case F_SETLKW: 597 flg |= F_WAIT; 598 /* FALLTHROUGH F_SETLK */ 599 600 case F_SETLK: 601 do_setlk: 602 FILEDESC_SLOCK(fdp); 603 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp); 604 if (error != 0) { 605 FILEDESC_SUNLOCK(fdp); 606 break; 607 } 608 if (fp->f_type != DTYPE_VNODE) { 609 FILEDESC_SUNLOCK(fdp); 610 error = EBADF; 611 break; 612 } 613 flp = (struct flock *)arg; 614 if (flp->l_whence == SEEK_CUR) { 615 if (fp->f_offset < 0 || 616 (flp->l_start > 0 && 617 fp->f_offset > OFF_MAX - flp->l_start)) { 618 FILEDESC_SUNLOCK(fdp); 619 error = EOVERFLOW; 620 break; 621 } 622 flp->l_start += fp->f_offset; 623 } 624 625 /* 626 * VOP_ADVLOCK() may block. 627 */ 628 fhold(fp); 629 FILEDESC_SUNLOCK(fdp); 630 vp = fp->f_vnode; 631 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 632 switch (flp->l_type) { 633 case F_RDLCK: 634 if ((fp->f_flag & FREAD) == 0) { 635 error = EBADF; 636 break; 637 } 638 PROC_LOCK(p->p_leader); 639 p->p_leader->p_flag |= P_ADVLOCK; 640 PROC_UNLOCK(p->p_leader); 641 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 642 flp, flg); 643 break; 644 case F_WRLCK: 645 if ((fp->f_flag & FWRITE) == 0) { 646 error = EBADF; 647 break; 648 } 649 PROC_LOCK(p->p_leader); 650 p->p_leader->p_flag |= P_ADVLOCK; 651 PROC_UNLOCK(p->p_leader); 652 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 653 flp, flg); 654 break; 655 case F_UNLCK: 656 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 657 flp, flg); 658 break; 659 case F_UNLCKSYS: 660 /* 661 * Temporary api for testing remote lock 662 * infrastructure. 663 */ 664 if (flg != F_REMOTE) { 665 error = EINVAL; 666 break; 667 } 668 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 669 F_UNLCKSYS, flp, flg); 670 break; 671 default: 672 error = EINVAL; 673 break; 674 } 675 VFS_UNLOCK_GIANT(vfslocked); 676 vfslocked = 0; 677 /* Check for race with close */ 678 FILEDESC_SLOCK(fdp); 679 if (fd < 0 || fd >= fdp->fd_nfiles || 680 fp != fdp->fd_ofiles[fd]) { 681 FILEDESC_SUNLOCK(fdp); 682 flp->l_whence = SEEK_SET; 683 flp->l_start = 0; 684 flp->l_len = 0; 685 flp->l_type = F_UNLCK; 686 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 687 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 688 F_UNLCK, flp, F_POSIX); 689 VFS_UNLOCK_GIANT(vfslocked); 690 vfslocked = 0; 691 } else 692 FILEDESC_SUNLOCK(fdp); 693 fdrop(fp, td); 694 break; 695 696 case F_GETLK: 697 FILEDESC_SLOCK(fdp); 698 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp); 699 if (error != 0) { 700 FILEDESC_SUNLOCK(fdp); 701 break; 702 } 703 if (fp->f_type != DTYPE_VNODE) { 704 FILEDESC_SUNLOCK(fdp); 705 error = EBADF; 706 break; 707 } 708 flp = (struct flock *)arg; 709 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 710 flp->l_type != F_UNLCK) { 711 FILEDESC_SUNLOCK(fdp); 712 error = EINVAL; 713 break; 714 } 715 if (flp->l_whence == SEEK_CUR) { 716 if ((flp->l_start > 0 && 717 fp->f_offset > OFF_MAX - flp->l_start) || 718 (flp->l_start < 0 && 719 fp->f_offset < OFF_MIN - flp->l_start)) { 720 FILEDESC_SUNLOCK(fdp); 721 error = EOVERFLOW; 722 break; 723 } 724 flp->l_start += fp->f_offset; 725 } 726 /* 727 * VOP_ADVLOCK() may block. 728 */ 729 fhold(fp); 730 FILEDESC_SUNLOCK(fdp); 731 vp = fp->f_vnode; 732 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 733 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 734 F_POSIX); 735 VFS_UNLOCK_GIANT(vfslocked); 736 vfslocked = 0; 737 fdrop(fp, td); 738 break; 739 740 case F_RDAHEAD: 741 arg = arg ? 128 * 1024: 0; 742 /* FALLTHROUGH */ 743 case F_READAHEAD: 744 FILEDESC_SLOCK(fdp); 745 if ((fp = fdtofp(fd, fdp)) == NULL) { 746 FILEDESC_SUNLOCK(fdp); 747 error = EBADF; 748 break; 749 } 750 if (fp->f_type != DTYPE_VNODE) { 751 FILEDESC_SUNLOCK(fdp); 752 error = EBADF; 753 break; 754 } 755 fhold(fp); 756 FILEDESC_SUNLOCK(fdp); 757 if (arg != 0) { 758 vp = fp->f_vnode; 759 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 760 error = vn_lock(vp, LK_SHARED); 761 if (error != 0) 762 goto readahead_vnlock_fail; 763 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 764 VOP_UNLOCK(vp, 0); 765 fp->f_seqcount = (arg + bsize - 1) / bsize; 766 do { 767 new = old = fp->f_flag; 768 new |= FRDAHEAD; 769 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 770readahead_vnlock_fail: 771 VFS_UNLOCK_GIANT(vfslocked); 772 vfslocked = 0; 773 } else { 774 do { 775 new = old = fp->f_flag; 776 new &= ~FRDAHEAD; 777 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 778 } 779 fdrop(fp, td); 780 break; 781 782 default: 783 error = EINVAL; 784 break; 785 } 786 VFS_UNLOCK_GIANT(vfslocked); 787 return (error); 788} 789 790/* 791 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 792 */ 793static int 794do_dup(struct thread *td, int flags, int old, int new, 795 register_t *retval) 796{ 797 struct filedesc *fdp; 798 struct proc *p; 799 struct file *fp; 800 struct file *delfp; 801 int error, maxfd; 802 803 p = td->td_proc; 804 fdp = p->p_fd; 805 806 /* 807 * Verify we have a valid descriptor to dup from and possibly to 808 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 809 * return EINVAL when the new descriptor is out of bounds. 810 */ 811 if (old < 0) 812 return (EBADF); 813 if (new < 0) 814 return (flags & DUP_FCNTL ? EINVAL : EBADF); 815 PROC_LOCK(p); 816 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 817 PROC_UNLOCK(p); 818 if (new >= maxfd) 819 return (flags & DUP_FCNTL ? EINVAL : EBADF); 820 821 FILEDESC_XLOCK(fdp); 822 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) { 823 FILEDESC_XUNLOCK(fdp); 824 return (EBADF); 825 } 826 if (flags & DUP_FIXED && old == new) { 827 *retval = new; 828 FILEDESC_XUNLOCK(fdp); 829 return (0); 830 } 831 fp = fdp->fd_ofiles[old]; 832 fhold(fp); 833 834 /* 835 * If the caller specified a file descriptor, make sure the file 836 * table is large enough to hold it, and grab it. Otherwise, just 837 * allocate a new descriptor the usual way. Since the filedesc 838 * lock may be temporarily dropped in the process, we have to look 839 * out for a race. 840 */ 841 if (flags & DUP_FIXED) { 842 if (new >= fdp->fd_nfiles) { 843 /* 844 * The resource limits are here instead of e.g. 845 * fdalloc(), because the file descriptor table may be 846 * shared between processes, so we can't really use 847 * racct_add()/racct_sub(). Instead of counting the 848 * number of actually allocated descriptors, just put 849 * the limit on the size of the file descriptor table. 850 */ 851#ifdef RACCT 852 PROC_LOCK(p); 853 error = racct_set(p, RACCT_NOFILE, new + 1); 854 PROC_UNLOCK(p); 855 if (error != 0) { 856 FILEDESC_XUNLOCK(fdp); 857 fdrop(fp, td); 858 return (EMFILE); 859 } 860#endif 861 fdgrowtable(fdp, new + 1); 862 } 863 if (fdp->fd_ofiles[new] == NULL) 864 fdused(fdp, new); 865 } else { 866 if ((error = fdalloc(td, new, &new)) != 0) { 867 FILEDESC_XUNLOCK(fdp); 868 fdrop(fp, td); 869 return (error); 870 } 871 } 872 873 KASSERT(fp == fdp->fd_ofiles[old], ("old fd has been modified")); 874 KASSERT(old != new, ("new fd is same as old")); 875 876 delfp = fdp->fd_ofiles[new]; 877 /* 878 * Duplicate the source descriptor. 879 */ 880 fdp->fd_ofiles[new] = fp; 881 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 882 if (new > fdp->fd_lastfile) 883 fdp->fd_lastfile = new; 884 *retval = new; 885 886 if (delfp != NULL) { 887 (void) closefp(fdp, new, delfp, td, 1); 888 /* closefp() drops the FILEDESC lock for us. */ 889 } else { 890 FILEDESC_XUNLOCK(fdp); 891 } 892 893 return (0); 894} 895 896/* 897 * If sigio is on the list associated with a process or process group, 898 * disable signalling from the device, remove sigio from the list and 899 * free sigio. 900 */ 901void 902funsetown(struct sigio **sigiop) 903{ 904 struct sigio *sigio; 905 906 SIGIO_LOCK(); 907 sigio = *sigiop; 908 if (sigio == NULL) { 909 SIGIO_UNLOCK(); 910 return; 911 } 912 *(sigio->sio_myref) = NULL; 913 if ((sigio)->sio_pgid < 0) { 914 struct pgrp *pg = (sigio)->sio_pgrp; 915 PGRP_LOCK(pg); 916 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 917 sigio, sio_pgsigio); 918 PGRP_UNLOCK(pg); 919 } else { 920 struct proc *p = (sigio)->sio_proc; 921 PROC_LOCK(p); 922 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 923 sigio, sio_pgsigio); 924 PROC_UNLOCK(p); 925 } 926 SIGIO_UNLOCK(); 927 crfree(sigio->sio_ucred); 928 free(sigio, M_SIGIO); 929} 930 931/* 932 * Free a list of sigio structures. 933 * We only need to lock the SIGIO_LOCK because we have made ourselves 934 * inaccessible to callers of fsetown and therefore do not need to lock 935 * the proc or pgrp struct for the list manipulation. 936 */ 937void 938funsetownlst(struct sigiolst *sigiolst) 939{ 940 struct proc *p; 941 struct pgrp *pg; 942 struct sigio *sigio; 943 944 sigio = SLIST_FIRST(sigiolst); 945 if (sigio == NULL) 946 return; 947 p = NULL; 948 pg = NULL; 949 950 /* 951 * Every entry of the list should belong 952 * to a single proc or pgrp. 953 */ 954 if (sigio->sio_pgid < 0) { 955 pg = sigio->sio_pgrp; 956 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 957 } else /* if (sigio->sio_pgid > 0) */ { 958 p = sigio->sio_proc; 959 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 960 } 961 962 SIGIO_LOCK(); 963 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 964 *(sigio->sio_myref) = NULL; 965 if (pg != NULL) { 966 KASSERT(sigio->sio_pgid < 0, 967 ("Proc sigio in pgrp sigio list")); 968 KASSERT(sigio->sio_pgrp == pg, 969 ("Bogus pgrp in sigio list")); 970 PGRP_LOCK(pg); 971 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 972 sio_pgsigio); 973 PGRP_UNLOCK(pg); 974 } else /* if (p != NULL) */ { 975 KASSERT(sigio->sio_pgid > 0, 976 ("Pgrp sigio in proc sigio list")); 977 KASSERT(sigio->sio_proc == p, 978 ("Bogus proc in sigio list")); 979 PROC_LOCK(p); 980 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 981 sio_pgsigio); 982 PROC_UNLOCK(p); 983 } 984 SIGIO_UNLOCK(); 985 crfree(sigio->sio_ucred); 986 free(sigio, M_SIGIO); 987 SIGIO_LOCK(); 988 } 989 SIGIO_UNLOCK(); 990} 991 992/* 993 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 994 * 995 * After permission checking, add a sigio structure to the sigio list for 996 * the process or process group. 997 */ 998int 999fsetown(pid_t pgid, struct sigio **sigiop) 1000{ 1001 struct proc *proc; 1002 struct pgrp *pgrp; 1003 struct sigio *sigio; 1004 int ret; 1005 1006 if (pgid == 0) { 1007 funsetown(sigiop); 1008 return (0); 1009 } 1010 1011 ret = 0; 1012 1013 /* Allocate and fill in the new sigio out of locks. */ 1014 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1015 sigio->sio_pgid = pgid; 1016 sigio->sio_ucred = crhold(curthread->td_ucred); 1017 sigio->sio_myref = sigiop; 1018 1019 sx_slock(&proctree_lock); 1020 if (pgid > 0) { 1021 proc = pfind(pgid); 1022 if (proc == NULL) { 1023 ret = ESRCH; 1024 goto fail; 1025 } 1026 1027 /* 1028 * Policy - Don't allow a process to FSETOWN a process 1029 * in another session. 1030 * 1031 * Remove this test to allow maximum flexibility or 1032 * restrict FSETOWN to the current process or process 1033 * group for maximum safety. 1034 */ 1035 PROC_UNLOCK(proc); 1036 if (proc->p_session != curthread->td_proc->p_session) { 1037 ret = EPERM; 1038 goto fail; 1039 } 1040 1041 pgrp = NULL; 1042 } else /* if (pgid < 0) */ { 1043 pgrp = pgfind(-pgid); 1044 if (pgrp == NULL) { 1045 ret = ESRCH; 1046 goto fail; 1047 } 1048 PGRP_UNLOCK(pgrp); 1049 1050 /* 1051 * Policy - Don't allow a process to FSETOWN a process 1052 * in another session. 1053 * 1054 * Remove this test to allow maximum flexibility or 1055 * restrict FSETOWN to the current process or process 1056 * group for maximum safety. 1057 */ 1058 if (pgrp->pg_session != curthread->td_proc->p_session) { 1059 ret = EPERM; 1060 goto fail; 1061 } 1062 1063 proc = NULL; 1064 } 1065 funsetown(sigiop); 1066 if (pgid > 0) { 1067 PROC_LOCK(proc); 1068 /* 1069 * Since funsetownlst() is called without the proctree 1070 * locked, we need to check for P_WEXIT. 1071 * XXX: is ESRCH correct? 1072 */ 1073 if ((proc->p_flag & P_WEXIT) != 0) { 1074 PROC_UNLOCK(proc); 1075 ret = ESRCH; 1076 goto fail; 1077 } 1078 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1079 sigio->sio_proc = proc; 1080 PROC_UNLOCK(proc); 1081 } else { 1082 PGRP_LOCK(pgrp); 1083 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1084 sigio->sio_pgrp = pgrp; 1085 PGRP_UNLOCK(pgrp); 1086 } 1087 sx_sunlock(&proctree_lock); 1088 SIGIO_LOCK(); 1089 *sigiop = sigio; 1090 SIGIO_UNLOCK(); 1091 return (0); 1092 1093fail: 1094 sx_sunlock(&proctree_lock); 1095 crfree(sigio->sio_ucred); 1096 free(sigio, M_SIGIO); 1097 return (ret); 1098} 1099 1100/* 1101 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1102 */ 1103pid_t 1104fgetown(sigiop) 1105 struct sigio **sigiop; 1106{ 1107 pid_t pgid; 1108 1109 SIGIO_LOCK(); 1110 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1111 SIGIO_UNLOCK(); 1112 return (pgid); 1113} 1114 1115/* 1116 * Function drops the filedesc lock on return. 1117 */ 1118static int 1119closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1120 int holdleaders) 1121{ 1122 struct file *fp_object; 1123 int error; 1124 1125 FILEDESC_XLOCK_ASSERT(fdp); 1126 1127 if (holdleaders) { 1128 if (td->td_proc->p_fdtol != NULL) { 1129 /* 1130 * Ask fdfree() to sleep to ensure that all relevant 1131 * process leaders can be traversed in closef(). 1132 */ 1133 fdp->fd_holdleaderscount++; 1134 } else { 1135 holdleaders = 0; 1136 } 1137 } 1138 1139 /* 1140 * We now hold the fp reference that used to be owned by the 1141 * descriptor array. We have to unlock the FILEDESC *AFTER* 1142 * knote_fdclose to prevent a race of the fd getting opened, a knote 1143 * added, and deleteing a knote for the new fd. 1144 */ 1145 knote_fdclose(td, fd); 1146 1147 /* 1148 * When we're closing an fd with a capability, we need to notify 1149 * mqueue if the underlying object is of type mqueue. 1150 */ 1151 (void)cap_funwrap(fp, 0, &fp_object); 1152 if (fp_object->f_type == DTYPE_MQUEUE) 1153 mq_fdclose(td, fd, fp_object); 1154 FILEDESC_XUNLOCK(fdp); 1155 1156 error = closef(fp, td); 1157 if (holdleaders) { 1158 FILEDESC_XLOCK(fdp); 1159 fdp->fd_holdleaderscount--; 1160 if (fdp->fd_holdleaderscount == 0 && 1161 fdp->fd_holdleaderswakeup != 0) { 1162 fdp->fd_holdleaderswakeup = 0; 1163 wakeup(&fdp->fd_holdleaderscount); 1164 } 1165 FILEDESC_XUNLOCK(fdp); 1166 } 1167 return (error); 1168} 1169 1170/* 1171 * Close a file descriptor. 1172 */ 1173#ifndef _SYS_SYSPROTO_H_ 1174struct close_args { 1175 int fd; 1176}; 1177#endif 1178/* ARGSUSED */ 1179int 1180sys_close(td, uap) 1181 struct thread *td; 1182 struct close_args *uap; 1183{ 1184 1185 return (kern_close(td, uap->fd)); 1186} 1187 1188int 1189kern_close(td, fd) 1190 struct thread *td; 1191 int fd; 1192{ 1193 struct filedesc *fdp; 1194 struct file *fp; 1195 1196 fdp = td->td_proc->p_fd; 1197 1198 AUDIT_SYSCLOSE(td, fd); 1199 1200 FILEDESC_XLOCK(fdp); 1201 if (fd < 0 || fd >= fdp->fd_nfiles || 1202 (fp = fdp->fd_ofiles[fd]) == NULL) { 1203 FILEDESC_XUNLOCK(fdp); 1204 return (EBADF); 1205 } 1206 fdp->fd_ofiles[fd] = NULL; 1207 fdp->fd_ofileflags[fd] = 0; 1208 fdunused(fdp, fd); 1209 1210 /* closefp() drops the FILEDESC lock for us. */ 1211 return (closefp(fdp, fd, fp, td, 1)); 1212} 1213 1214/* 1215 * Close open file descriptors. 1216 */ 1217#ifndef _SYS_SYSPROTO_H_ 1218struct closefrom_args { 1219 int lowfd; 1220}; 1221#endif 1222/* ARGSUSED */ 1223int 1224sys_closefrom(struct thread *td, struct closefrom_args *uap) 1225{ 1226 struct filedesc *fdp; 1227 int fd; 1228 1229 fdp = td->td_proc->p_fd; 1230 AUDIT_ARG_FD(uap->lowfd); 1231 1232 /* 1233 * Treat negative starting file descriptor values identical to 1234 * closefrom(0) which closes all files. 1235 */ 1236 if (uap->lowfd < 0) 1237 uap->lowfd = 0; 1238 FILEDESC_SLOCK(fdp); 1239 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) { 1240 if (fdp->fd_ofiles[fd] != NULL) { 1241 FILEDESC_SUNLOCK(fdp); 1242 (void)kern_close(td, fd); 1243 FILEDESC_SLOCK(fdp); 1244 } 1245 } 1246 FILEDESC_SUNLOCK(fdp); 1247 return (0); 1248} 1249 1250#if defined(COMPAT_43) 1251/* 1252 * Return status information about a file descriptor. 1253 */ 1254#ifndef _SYS_SYSPROTO_H_ 1255struct ofstat_args { 1256 int fd; 1257 struct ostat *sb; 1258}; 1259#endif 1260/* ARGSUSED */ 1261int 1262ofstat(struct thread *td, struct ofstat_args *uap) 1263{ 1264 struct ostat oub; 1265 struct stat ub; 1266 int error; 1267 1268 error = kern_fstat(td, uap->fd, &ub); 1269 if (error == 0) { 1270 cvtstat(&ub, &oub); 1271 error = copyout(&oub, uap->sb, sizeof(oub)); 1272 } 1273 return (error); 1274} 1275#endif /* COMPAT_43 */ 1276 1277/* 1278 * Return status information about a file descriptor. 1279 */ 1280#ifndef _SYS_SYSPROTO_H_ 1281struct fstat_args { 1282 int fd; 1283 struct stat *sb; 1284}; 1285#endif 1286/* ARGSUSED */ 1287int 1288sys_fstat(struct thread *td, struct fstat_args *uap) 1289{ 1290 struct stat ub; 1291 int error; 1292 1293 error = kern_fstat(td, uap->fd, &ub); 1294 if (error == 0) 1295 error = copyout(&ub, uap->sb, sizeof(ub)); 1296 return (error); 1297} 1298 1299int 1300kern_fstat(struct thread *td, int fd, struct stat *sbp) 1301{ 1302 struct file *fp; 1303 int error; 1304 1305 AUDIT_ARG_FD(fd); 1306 1307 if ((error = fget(td, fd, CAP_FSTAT, &fp)) != 0) 1308 return (error); 1309 1310 AUDIT_ARG_FILE(td->td_proc, fp); 1311 1312 error = fo_stat(fp, sbp, td->td_ucred, td); 1313 fdrop(fp, td); 1314#ifdef KTRACE 1315 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1316 ktrstat(sbp); 1317#endif 1318 return (error); 1319} 1320 1321/* 1322 * Return status information about a file descriptor. 1323 */ 1324#ifndef _SYS_SYSPROTO_H_ 1325struct nfstat_args { 1326 int fd; 1327 struct nstat *sb; 1328}; 1329#endif 1330/* ARGSUSED */ 1331int 1332sys_nfstat(struct thread *td, struct nfstat_args *uap) 1333{ 1334 struct nstat nub; 1335 struct stat ub; 1336 int error; 1337 1338 error = kern_fstat(td, uap->fd, &ub); 1339 if (error == 0) { 1340 cvtnstat(&ub, &nub); 1341 error = copyout(&nub, uap->sb, sizeof(nub)); 1342 } 1343 return (error); 1344} 1345 1346/* 1347 * Return pathconf information about a file descriptor. 1348 */ 1349#ifndef _SYS_SYSPROTO_H_ 1350struct fpathconf_args { 1351 int fd; 1352 int name; 1353}; 1354#endif 1355/* ARGSUSED */ 1356int 1357sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1358{ 1359 struct file *fp; 1360 struct vnode *vp; 1361 int error; 1362 1363 if ((error = fget(td, uap->fd, CAP_FPATHCONF, &fp)) != 0) 1364 return (error); 1365 1366 /* If asynchronous I/O is available, it works for all descriptors. */ 1367 if (uap->name == _PC_ASYNC_IO) { 1368 td->td_retval[0] = async_io_version; 1369 goto out; 1370 } 1371 vp = fp->f_vnode; 1372 if (vp != NULL) { 1373 int vfslocked; 1374 1375 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1376 vn_lock(vp, LK_SHARED | LK_RETRY); 1377 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1378 VOP_UNLOCK(vp, 0); 1379 VFS_UNLOCK_GIANT(vfslocked); 1380 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1381 if (uap->name != _PC_PIPE_BUF) { 1382 error = EINVAL; 1383 } else { 1384 td->td_retval[0] = PIPE_BUF; 1385 error = 0; 1386 } 1387 } else { 1388 error = EOPNOTSUPP; 1389 } 1390out: 1391 fdrop(fp, td); 1392 return (error); 1393} 1394 1395/* 1396 * Grow the file table to accomodate (at least) nfd descriptors. 1397 */ 1398static void 1399fdgrowtable(struct filedesc *fdp, int nfd) 1400{ 1401 struct filedesc0 *fdp0; 1402 struct freetable *fo; 1403 struct file **ntable; 1404 struct file **otable; 1405 char *nfileflags; 1406 int nnfiles, onfiles; 1407 NDSLOTTYPE *nmap; 1408 1409 FILEDESC_XLOCK_ASSERT(fdp); 1410 1411 KASSERT(fdp->fd_nfiles > 0, 1412 ("zero-length file table")); 1413 1414 /* compute the size of the new table */ 1415 onfiles = fdp->fd_nfiles; 1416 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1417 if (nnfiles <= onfiles) 1418 /* the table is already large enough */ 1419 return; 1420 1421 /* allocate a new table and (if required) new bitmaps */ 1422 ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable), 1423 M_FILEDESC, M_ZERO | M_WAITOK); 1424 nfileflags = (char *)&ntable[nnfiles]; 1425 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1426 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, 1427 M_FILEDESC, M_ZERO | M_WAITOK); 1428 else 1429 nmap = NULL; 1430 1431 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1432 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1433 otable = fdp->fd_ofiles; 1434 fdp->fd_ofileflags = nfileflags; 1435 fdp->fd_ofiles = ntable; 1436 /* 1437 * We must preserve ofiles until the process exits because we can't 1438 * be certain that no threads have references to the old table via 1439 * _fget(). 1440 */ 1441 if (onfiles > NDFILE) { 1442 fo = (struct freetable *)&otable[onfiles]; 1443 fdp0 = (struct filedesc0 *)fdp; 1444 fo->ft_table = otable; 1445 SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next); 1446 } 1447 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1448 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1449 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1450 free(fdp->fd_map, M_FILEDESC); 1451 fdp->fd_map = nmap; 1452 } 1453 fdp->fd_nfiles = nnfiles; 1454} 1455 1456/* 1457 * Allocate a file descriptor for the process. 1458 */ 1459int 1460fdalloc(struct thread *td, int minfd, int *result) 1461{ 1462 struct proc *p = td->td_proc; 1463 struct filedesc *fdp = p->p_fd; 1464 int fd = -1, maxfd, allocfd; 1465#ifdef RACCT 1466 int error; 1467#endif 1468 1469 FILEDESC_XLOCK_ASSERT(fdp); 1470 1471 if (fdp->fd_freefile > minfd) 1472 minfd = fdp->fd_freefile; 1473 1474 PROC_LOCK(p); 1475 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1476 PROC_UNLOCK(p); 1477 1478 /* 1479 * Search the bitmap for a free descriptor starting at minfd. 1480 * If none is found, grow the file table. 1481 */ 1482 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1483 if (fd >= maxfd) 1484 return (EMFILE); 1485 if (fd >= fdp->fd_nfiles) { 1486 allocfd = min(fd * 2, maxfd); 1487#ifdef RACCT 1488 PROC_LOCK(p); 1489 error = racct_set(p, RACCT_NOFILE, allocfd); 1490 PROC_UNLOCK(p); 1491 if (error != 0) 1492 return (EMFILE); 1493#endif 1494 /* 1495 * fd is already equal to first free descriptor >= minfd, so 1496 * we only need to grow the table and we are done. 1497 */ 1498 fdgrowtable(fdp, allocfd); 1499 } 1500 1501 /* 1502 * Perform some sanity checks, then mark the file descriptor as 1503 * used and return it to the caller. 1504 */ 1505 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1506 ("invalid descriptor %d", fd)); 1507 KASSERT(!fdisused(fdp, fd), 1508 ("fd_first_free() returned non-free descriptor")); 1509 KASSERT(fdp->fd_ofiles[fd] == NULL, ("file descriptor isn't free")); 1510 KASSERT(fdp->fd_ofileflags[fd] == 0, ("file flags are set")); 1511 fdused(fdp, fd); 1512 *result = fd; 1513 return (0); 1514} 1515 1516/* 1517 * Check to see whether n user file descriptors are available to the process 1518 * p. 1519 */ 1520int 1521fdavail(struct thread *td, int n) 1522{ 1523 struct proc *p = td->td_proc; 1524 struct filedesc *fdp = td->td_proc->p_fd; 1525 int i, lim, last; 1526 1527 FILEDESC_LOCK_ASSERT(fdp); 1528 1529 /* 1530 * XXX: This is only called from uipc_usrreq.c:unp_externalize(); 1531 * call racct_add() from there instead of dealing with containers 1532 * here. 1533 */ 1534 PROC_LOCK(p); 1535 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1536 PROC_UNLOCK(p); 1537 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1538 return (1); 1539 last = min(fdp->fd_nfiles, lim); 1540 for (i = fdp->fd_freefile; i < last; i++) { 1541 if (fdp->fd_ofiles[i] == NULL && --n <= 0) 1542 return (1); 1543 } 1544 return (0); 1545} 1546 1547/* 1548 * Create a new open file structure and allocate a file decriptor for the 1549 * process that refers to it. We add one reference to the file for the 1550 * descriptor table and one reference for resultfp. This is to prevent us 1551 * being preempted and the entry in the descriptor table closed after we 1552 * release the FILEDESC lock. 1553 */ 1554int 1555falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1556{ 1557 struct file *fp; 1558 int error, fd; 1559 1560 error = falloc_noinstall(td, &fp); 1561 if (error) 1562 return (error); /* no reference held on error */ 1563 1564 error = finstall(td, fp, &fd, flags); 1565 if (error) { 1566 fdrop(fp, td); /* one reference (fp only) */ 1567 return (error); 1568 } 1569 1570 if (resultfp != NULL) 1571 *resultfp = fp; /* copy out result */ 1572 else 1573 fdrop(fp, td); /* release local reference */ 1574 1575 if (resultfd != NULL) 1576 *resultfd = fd; 1577 1578 return (0); 1579} 1580 1581/* 1582 * Create a new open file structure without allocating a file descriptor. 1583 */ 1584int 1585falloc_noinstall(struct thread *td, struct file **resultfp) 1586{ 1587 struct file *fp; 1588 int maxuserfiles = maxfiles - (maxfiles / 20); 1589 static struct timeval lastfail; 1590 static int curfail; 1591 1592 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1593 1594 if ((openfiles >= maxuserfiles && 1595 priv_check(td, PRIV_MAXFILES) != 0) || 1596 openfiles >= maxfiles) { 1597 if (ppsratecheck(&lastfail, &curfail, 1)) { 1598 printf("kern.maxfiles limit exceeded by uid %i, " 1599 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1600 } 1601 return (ENFILE); 1602 } 1603 atomic_add_int(&openfiles, 1); 1604 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1605 refcount_init(&fp->f_count, 1); 1606 fp->f_cred = crhold(td->td_ucred); 1607 fp->f_ops = &badfileops; 1608 fp->f_data = NULL; 1609 fp->f_vnode = NULL; 1610 *resultfp = fp; 1611 return (0); 1612} 1613 1614/* 1615 * Install a file in a file descriptor table. 1616 */ 1617int 1618finstall(struct thread *td, struct file *fp, int *fd, int flags) 1619{ 1620 struct filedesc *fdp = td->td_proc->p_fd; 1621 int error; 1622 1623 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1624 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1625 1626 FILEDESC_XLOCK(fdp); 1627 if ((error = fdalloc(td, 0, fd))) { 1628 FILEDESC_XUNLOCK(fdp); 1629 return (error); 1630 } 1631 fhold(fp); 1632 fdp->fd_ofiles[*fd] = fp; 1633 if ((flags & O_CLOEXEC) != 0) 1634 fdp->fd_ofileflags[*fd] |= UF_EXCLOSE; 1635 FILEDESC_XUNLOCK(fdp); 1636 return (0); 1637} 1638 1639/* 1640 * Build a new filedesc structure from another. 1641 * Copy the current, root, and jail root vnode references. 1642 */ 1643struct filedesc * 1644fdinit(struct filedesc *fdp) 1645{ 1646 struct filedesc0 *newfdp; 1647 1648 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1649 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1650 if (fdp != NULL) { 1651 FILEDESC_XLOCK(fdp); 1652 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1653 if (newfdp->fd_fd.fd_cdir) 1654 VREF(newfdp->fd_fd.fd_cdir); 1655 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1656 if (newfdp->fd_fd.fd_rdir) 1657 VREF(newfdp->fd_fd.fd_rdir); 1658 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1659 if (newfdp->fd_fd.fd_jdir) 1660 VREF(newfdp->fd_fd.fd_jdir); 1661 FILEDESC_XUNLOCK(fdp); 1662 } 1663 1664 /* Create the file descriptor table. */ 1665 newfdp->fd_fd.fd_refcnt = 1; 1666 newfdp->fd_fd.fd_holdcnt = 1; 1667 newfdp->fd_fd.fd_cmask = CMASK; 1668 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1669 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1670 newfdp->fd_fd.fd_nfiles = NDFILE; 1671 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1672 newfdp->fd_fd.fd_lastfile = -1; 1673 return (&newfdp->fd_fd); 1674} 1675 1676static struct filedesc * 1677fdhold(struct proc *p) 1678{ 1679 struct filedesc *fdp; 1680 1681 mtx_lock(&fdesc_mtx); 1682 fdp = p->p_fd; 1683 if (fdp != NULL) 1684 fdp->fd_holdcnt++; 1685 mtx_unlock(&fdesc_mtx); 1686 return (fdp); 1687} 1688 1689static void 1690fddrop(struct filedesc *fdp) 1691{ 1692 struct filedesc0 *fdp0; 1693 struct freetable *ft; 1694 int i; 1695 1696 mtx_lock(&fdesc_mtx); 1697 i = --fdp->fd_holdcnt; 1698 mtx_unlock(&fdesc_mtx); 1699 if (i > 0) 1700 return; 1701 1702 FILEDESC_LOCK_DESTROY(fdp); 1703 fdp0 = (struct filedesc0 *)fdp; 1704 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1705 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1706 free(ft->ft_table, M_FILEDESC); 1707 } 1708 free(fdp, M_FILEDESC); 1709} 1710 1711/* 1712 * Share a filedesc structure. 1713 */ 1714struct filedesc * 1715fdshare(struct filedesc *fdp) 1716{ 1717 1718 FILEDESC_XLOCK(fdp); 1719 fdp->fd_refcnt++; 1720 FILEDESC_XUNLOCK(fdp); 1721 return (fdp); 1722} 1723 1724/* 1725 * Unshare a filedesc structure, if necessary by making a copy 1726 */ 1727void 1728fdunshare(struct proc *p, struct thread *td) 1729{ 1730 1731 FILEDESC_XLOCK(p->p_fd); 1732 if (p->p_fd->fd_refcnt > 1) { 1733 struct filedesc *tmp; 1734 1735 FILEDESC_XUNLOCK(p->p_fd); 1736 tmp = fdcopy(p->p_fd); 1737 fdfree(td); 1738 p->p_fd = tmp; 1739 } else 1740 FILEDESC_XUNLOCK(p->p_fd); 1741} 1742 1743/* 1744 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1745 * this is to ease callers, not catch errors. 1746 */ 1747struct filedesc * 1748fdcopy(struct filedesc *fdp) 1749{ 1750 struct filedesc *newfdp; 1751 int i; 1752 1753 /* Certain daemons might not have file descriptors. */ 1754 if (fdp == NULL) 1755 return (NULL); 1756 1757 newfdp = fdinit(fdp); 1758 FILEDESC_SLOCK(fdp); 1759 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1760 FILEDESC_SUNLOCK(fdp); 1761 FILEDESC_XLOCK(newfdp); 1762 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1763 FILEDESC_XUNLOCK(newfdp); 1764 FILEDESC_SLOCK(fdp); 1765 } 1766 /* copy all passable descriptors (i.e. not kqueue) */ 1767 newfdp->fd_freefile = -1; 1768 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1769 if (fdisused(fdp, i) && 1770 (fdp->fd_ofiles[i]->f_ops->fo_flags & DFLAG_PASSABLE) && 1771 fdp->fd_ofiles[i]->f_ops != &badfileops) { 1772 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1773 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1774 fhold(newfdp->fd_ofiles[i]); 1775 newfdp->fd_lastfile = i; 1776 } else { 1777 if (newfdp->fd_freefile == -1) 1778 newfdp->fd_freefile = i; 1779 } 1780 } 1781 newfdp->fd_cmask = fdp->fd_cmask; 1782 FILEDESC_SUNLOCK(fdp); 1783 FILEDESC_XLOCK(newfdp); 1784 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1785 if (newfdp->fd_ofiles[i] != NULL) 1786 fdused(newfdp, i); 1787 if (newfdp->fd_freefile == -1) 1788 newfdp->fd_freefile = i; 1789 FILEDESC_XUNLOCK(newfdp); 1790 return (newfdp); 1791} 1792 1793/* 1794 * Release a filedesc structure. 1795 */ 1796void 1797fdfree(struct thread *td) 1798{ 1799 struct filedesc *fdp; 1800 int i, locked; 1801 struct filedesc_to_leader *fdtol; 1802 struct file *fp; 1803 struct vnode *cdir, *jdir, *rdir, *vp; 1804 struct flock lf; 1805 1806 /* Certain daemons might not have file descriptors. */ 1807 fdp = td->td_proc->p_fd; 1808 if (fdp == NULL) 1809 return; 1810 1811#ifdef RACCT 1812 PROC_LOCK(td->td_proc); 1813 racct_set(td->td_proc, RACCT_NOFILE, 0); 1814 PROC_UNLOCK(td->td_proc); 1815#endif 1816 1817 /* Check for special need to clear POSIX style locks */ 1818 fdtol = td->td_proc->p_fdtol; 1819 if (fdtol != NULL) { 1820 FILEDESC_XLOCK(fdp); 1821 KASSERT(fdtol->fdl_refcount > 0, 1822 ("filedesc_to_refcount botch: fdl_refcount=%d", 1823 fdtol->fdl_refcount)); 1824 if (fdtol->fdl_refcount == 1 && 1825 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1826 for (i = 0; i <= fdp->fd_lastfile; i++) { 1827 fp = fdp->fd_ofiles[i]; 1828 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1829 continue; 1830 fhold(fp); 1831 FILEDESC_XUNLOCK(fdp); 1832 lf.l_whence = SEEK_SET; 1833 lf.l_start = 0; 1834 lf.l_len = 0; 1835 lf.l_type = F_UNLCK; 1836 vp = fp->f_vnode; 1837 locked = VFS_LOCK_GIANT(vp->v_mount); 1838 (void) VOP_ADVLOCK(vp, 1839 (caddr_t)td->td_proc->p_leader, F_UNLCK, 1840 &lf, F_POSIX); 1841 VFS_UNLOCK_GIANT(locked); 1842 FILEDESC_XLOCK(fdp); 1843 fdrop(fp, td); 1844 } 1845 } 1846 retry: 1847 if (fdtol->fdl_refcount == 1) { 1848 if (fdp->fd_holdleaderscount > 0 && 1849 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1850 /* 1851 * close() or do_dup() has cleared a reference 1852 * in a shared file descriptor table. 1853 */ 1854 fdp->fd_holdleaderswakeup = 1; 1855 sx_sleep(&fdp->fd_holdleaderscount, 1856 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1857 goto retry; 1858 } 1859 if (fdtol->fdl_holdcount > 0) { 1860 /* 1861 * Ensure that fdtol->fdl_leader remains 1862 * valid in closef(). 1863 */ 1864 fdtol->fdl_wakeup = 1; 1865 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 1866 "fdlhold", 0); 1867 goto retry; 1868 } 1869 } 1870 fdtol->fdl_refcount--; 1871 if (fdtol->fdl_refcount == 0 && 1872 fdtol->fdl_holdcount == 0) { 1873 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1874 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1875 } else 1876 fdtol = NULL; 1877 td->td_proc->p_fdtol = NULL; 1878 FILEDESC_XUNLOCK(fdp); 1879 if (fdtol != NULL) 1880 free(fdtol, M_FILEDESC_TO_LEADER); 1881 } 1882 FILEDESC_XLOCK(fdp); 1883 i = --fdp->fd_refcnt; 1884 FILEDESC_XUNLOCK(fdp); 1885 if (i > 0) 1886 return; 1887 1888 for (i = 0; i <= fdp->fd_lastfile; i++) { 1889 fp = fdp->fd_ofiles[i]; 1890 if (fp != NULL) { 1891 FILEDESC_XLOCK(fdp); 1892 fdp->fd_ofiles[i] = NULL; 1893 FILEDESC_XUNLOCK(fdp); 1894 (void) closef(fp, td); 1895 } 1896 } 1897 FILEDESC_XLOCK(fdp); 1898 1899 /* XXX This should happen earlier. */ 1900 mtx_lock(&fdesc_mtx); 1901 td->td_proc->p_fd = NULL; 1902 mtx_unlock(&fdesc_mtx); 1903 1904 if (fdp->fd_nfiles > NDFILE) 1905 free(fdp->fd_ofiles, M_FILEDESC); 1906 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1907 free(fdp->fd_map, M_FILEDESC); 1908 1909 fdp->fd_nfiles = 0; 1910 1911 cdir = fdp->fd_cdir; 1912 fdp->fd_cdir = NULL; 1913 rdir = fdp->fd_rdir; 1914 fdp->fd_rdir = NULL; 1915 jdir = fdp->fd_jdir; 1916 fdp->fd_jdir = NULL; 1917 FILEDESC_XUNLOCK(fdp); 1918 1919 if (cdir) { 1920 locked = VFS_LOCK_GIANT(cdir->v_mount); 1921 vrele(cdir); 1922 VFS_UNLOCK_GIANT(locked); 1923 } 1924 if (rdir) { 1925 locked = VFS_LOCK_GIANT(rdir->v_mount); 1926 vrele(rdir); 1927 VFS_UNLOCK_GIANT(locked); 1928 } 1929 if (jdir) { 1930 locked = VFS_LOCK_GIANT(jdir->v_mount); 1931 vrele(jdir); 1932 VFS_UNLOCK_GIANT(locked); 1933 } 1934 1935 fddrop(fdp); 1936} 1937 1938/* 1939 * For setugid programs, we don't want to people to use that setugidness 1940 * to generate error messages which write to a file which otherwise would 1941 * otherwise be off-limits to the process. We check for filesystems where 1942 * the vnode can change out from under us after execve (like [lin]procfs). 1943 * 1944 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1945 * sufficient. We also don't check for setugidness since we know we are. 1946 */ 1947static int 1948is_unsafe(struct file *fp) 1949{ 1950 if (fp->f_type == DTYPE_VNODE) { 1951 struct vnode *vp = fp->f_vnode; 1952 1953 if ((vp->v_vflag & VV_PROCDEP) != 0) 1954 return (1); 1955 } 1956 return (0); 1957} 1958 1959/* 1960 * Make this setguid thing safe, if at all possible. 1961 */ 1962void 1963setugidsafety(struct thread *td) 1964{ 1965 struct filedesc *fdp; 1966 int i; 1967 1968 /* Certain daemons might not have file descriptors. */ 1969 fdp = td->td_proc->p_fd; 1970 if (fdp == NULL) 1971 return; 1972 1973 /* 1974 * Note: fdp->fd_ofiles may be reallocated out from under us while 1975 * we are blocked in a close. Be careful! 1976 */ 1977 FILEDESC_XLOCK(fdp); 1978 for (i = 0; i <= fdp->fd_lastfile; i++) { 1979 if (i > 2) 1980 break; 1981 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 1982 struct file *fp; 1983 1984 knote_fdclose(td, i); 1985 /* 1986 * NULL-out descriptor prior to close to avoid 1987 * a race while close blocks. 1988 */ 1989 fp = fdp->fd_ofiles[i]; 1990 fdp->fd_ofiles[i] = NULL; 1991 fdp->fd_ofileflags[i] = 0; 1992 fdunused(fdp, i); 1993 FILEDESC_XUNLOCK(fdp); 1994 (void) closef(fp, td); 1995 FILEDESC_XLOCK(fdp); 1996 } 1997 } 1998 FILEDESC_XUNLOCK(fdp); 1999} 2000 2001/* 2002 * If a specific file object occupies a specific file descriptor, close the 2003 * file descriptor entry and drop a reference on the file object. This is a 2004 * convenience function to handle a subsequent error in a function that calls 2005 * falloc() that handles the race that another thread might have closed the 2006 * file descriptor out from under the thread creating the file object. 2007 */ 2008void 2009fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2010{ 2011 2012 FILEDESC_XLOCK(fdp); 2013 if (fdp->fd_ofiles[idx] == fp) { 2014 fdp->fd_ofiles[idx] = NULL; 2015 fdunused(fdp, idx); 2016 FILEDESC_XUNLOCK(fdp); 2017 fdrop(fp, td); 2018 } else 2019 FILEDESC_XUNLOCK(fdp); 2020} 2021 2022/* 2023 * Close any files on exec? 2024 */ 2025void 2026fdcloseexec(struct thread *td) 2027{ 2028 struct filedesc *fdp; 2029 struct file *fp; 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 /* 2038 * We cannot cache fd_ofiles or fd_ofileflags since operations 2039 * may block and rip them out from under us. 2040 */ 2041 FILEDESC_XLOCK(fdp); 2042 for (i = 0; i <= fdp->fd_lastfile; i++) { 2043 fp = fdp->fd_ofiles[i]; 2044 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2045 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) { 2046 fdp->fd_ofiles[i] = NULL; 2047 fdp->fd_ofileflags[i] = 0; 2048 fdunused(fdp, i); 2049 (void) closefp(fdp, i, fp, td, 0); 2050 /* closefp() drops the FILEDESC lock. */ 2051 FILEDESC_XLOCK(fdp); 2052 } 2053 } 2054 FILEDESC_XUNLOCK(fdp); 2055} 2056 2057/* 2058 * It is unsafe for set[ug]id processes to be started with file 2059 * descriptors 0..2 closed, as these descriptors are given implicit 2060 * significance in the Standard C library. fdcheckstd() will create a 2061 * descriptor referencing /dev/null for each of stdin, stdout, and 2062 * stderr that is not already open. 2063 */ 2064int 2065fdcheckstd(struct thread *td) 2066{ 2067 struct filedesc *fdp; 2068 register_t retval, save; 2069 int i, error, devnull; 2070 2071 fdp = td->td_proc->p_fd; 2072 if (fdp == NULL) 2073 return (0); 2074 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2075 devnull = -1; 2076 error = 0; 2077 for (i = 0; i < 3; i++) { 2078 if (fdp->fd_ofiles[i] != NULL) 2079 continue; 2080 if (devnull < 0) { 2081 save = td->td_retval[0]; 2082 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2083 O_RDWR, 0); 2084 devnull = td->td_retval[0]; 2085 td->td_retval[0] = save; 2086 if (error) 2087 break; 2088 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2089 } else { 2090 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2091 if (error != 0) 2092 break; 2093 } 2094 } 2095 return (error); 2096} 2097 2098/* 2099 * Internal form of close. Decrement reference count on file structure. 2100 * Note: td may be NULL when closing a file that was being passed in a 2101 * message. 2102 * 2103 * XXXRW: Giant is not required for the caller, but often will be held; this 2104 * makes it moderately likely the Giant will be recursed in the VFS case. 2105 */ 2106int 2107closef(struct file *fp, struct thread *td) 2108{ 2109 struct vnode *vp; 2110 struct flock lf; 2111 struct filedesc_to_leader *fdtol; 2112 struct filedesc *fdp; 2113 struct file *fp_object; 2114 2115 fdp = td->td_proc->p_fd; 2116 FILEDESC_UNLOCK_ASSERT(fdp); 2117 2118 /* 2119 * POSIX record locking dictates that any close releases ALL 2120 * locks owned by this process. This is handled by setting 2121 * a flag in the unlock to free ONLY locks obeying POSIX 2122 * semantics, and not to free BSD-style file locks. 2123 * If the descriptor was in a message, POSIX-style locks 2124 * aren't passed with the descriptor, and the thread pointer 2125 * will be NULL. Callers should be careful only to pass a 2126 * NULL thread pointer when there really is no owning 2127 * context that might have locks, or the locks will be 2128 * leaked. 2129 * 2130 * If this is a capability, we do lock processing under the underlying 2131 * node, not the capability itself. 2132 */ 2133 (void)cap_funwrap(fp, 0, &fp_object); 2134 if (fp_object->f_type == DTYPE_VNODE && td != NULL) { 2135 int vfslocked; 2136 2137 vp = fp_object->f_vnode; 2138 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2139 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2140 lf.l_whence = SEEK_SET; 2141 lf.l_start = 0; 2142 lf.l_len = 0; 2143 lf.l_type = F_UNLCK; 2144 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2145 F_UNLCK, &lf, F_POSIX); 2146 } 2147 fdtol = td->td_proc->p_fdtol; 2148 if (fdtol != NULL) { 2149 /* 2150 * Handle special case where file descriptor table is 2151 * shared between multiple process leaders. 2152 */ 2153 FILEDESC_XLOCK(fdp); 2154 for (fdtol = fdtol->fdl_next; 2155 fdtol != td->td_proc->p_fdtol; 2156 fdtol = fdtol->fdl_next) { 2157 if ((fdtol->fdl_leader->p_flag & 2158 P_ADVLOCK) == 0) 2159 continue; 2160 fdtol->fdl_holdcount++; 2161 FILEDESC_XUNLOCK(fdp); 2162 lf.l_whence = SEEK_SET; 2163 lf.l_start = 0; 2164 lf.l_len = 0; 2165 lf.l_type = F_UNLCK; 2166 vp = fp_object->f_vnode; 2167 (void) VOP_ADVLOCK(vp, 2168 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2169 F_POSIX); 2170 FILEDESC_XLOCK(fdp); 2171 fdtol->fdl_holdcount--; 2172 if (fdtol->fdl_holdcount == 0 && 2173 fdtol->fdl_wakeup != 0) { 2174 fdtol->fdl_wakeup = 0; 2175 wakeup(fdtol); 2176 } 2177 } 2178 FILEDESC_XUNLOCK(fdp); 2179 } 2180 VFS_UNLOCK_GIANT(vfslocked); 2181 } 2182 return (fdrop(fp, td)); 2183} 2184 2185/* 2186 * Initialize the file pointer with the specified properties. 2187 * 2188 * The ops are set with release semantics to be certain that the flags, type, 2189 * and data are visible when ops is. This is to prevent ops methods from being 2190 * called with bad data. 2191 */ 2192void 2193finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2194{ 2195 fp->f_data = data; 2196 fp->f_flag = flag; 2197 fp->f_type = type; 2198 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2199} 2200 2201struct file * 2202fget_unlocked(struct filedesc *fdp, int fd) 2203{ 2204 struct file *fp; 2205 u_int count; 2206 2207 if (fd < 0 || fd >= fdp->fd_nfiles) 2208 return (NULL); 2209 /* 2210 * Fetch the descriptor locklessly. We avoid fdrop() races by 2211 * never raising a refcount above 0. To accomplish this we have 2212 * to use a cmpset loop rather than an atomic_add. The descriptor 2213 * must be re-verified once we acquire a reference to be certain 2214 * that the identity is still correct and we did not lose a race 2215 * due to preemption. 2216 */ 2217 for (;;) { 2218 fp = fdp->fd_ofiles[fd]; 2219 if (fp == NULL) 2220 break; 2221 count = fp->f_count; 2222 if (count == 0) 2223 continue; 2224 /* 2225 * Use an acquire barrier to prevent caching of fd_ofiles 2226 * so it is refreshed for verification. 2227 */ 2228 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2229 continue; 2230 if (fp == fdp->fd_ofiles[fd]) 2231 break; 2232 fdrop(fp, curthread); 2233 } 2234 2235 return (fp); 2236} 2237 2238/* 2239 * Extract the file pointer associated with the specified descriptor for the 2240 * current user process. 2241 * 2242 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2243 * returned. 2244 * 2245 * If the FGET_GETCAP flag is set, the capability itself will be returned. 2246 * Calling _fget() with FGET_GETCAP on a non-capability will return EINVAL. 2247 * Otherwise, if the file is a capability, its rights will be checked against 2248 * the capability rights mask, and if successful, the object will be unwrapped. 2249 * 2250 * If an error occured the non-zero error is returned and *fpp is set to 2251 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2252 * responsible for fdrop(). 2253 */ 2254#define FGET_GETCAP 0x00000001 2255static __inline int 2256_fget(struct thread *td, int fd, struct file **fpp, int flags, 2257 cap_rights_t needrights, cap_rights_t *haverightsp, u_char *maxprotp, 2258 int fget_flags) 2259{ 2260 struct filedesc *fdp; 2261 struct file *fp; 2262#ifdef CAPABILITIES 2263 struct file *fp_fromcap; 2264 int error; 2265#endif 2266 2267 *fpp = NULL; 2268 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2269 return (EBADF); 2270 if ((fp = fget_unlocked(fdp, fd)) == NULL) 2271 return (EBADF); 2272 if (fp->f_ops == &badfileops) { 2273 fdrop(fp, td); 2274 return (EBADF); 2275 } 2276 2277#ifdef CAPABILITIES 2278 /* 2279 * If this is a capability, what rights does it have? 2280 */ 2281 if (haverightsp != NULL) { 2282 if (fp->f_type == DTYPE_CAPABILITY) 2283 *haverightsp = cap_rights(fp); 2284 else 2285 *haverightsp = CAP_MASK_VALID; 2286 } 2287 2288 /* 2289 * If a capability has been requested, return the capability directly. 2290 * Otherwise, check capability rights, extract the underlying object, 2291 * and check its access flags. 2292 */ 2293 if (fget_flags & FGET_GETCAP) { 2294 if (fp->f_type != DTYPE_CAPABILITY) { 2295 fdrop(fp, td); 2296 return (EINVAL); 2297 } 2298 } else { 2299 if (maxprotp == NULL) 2300 error = cap_funwrap(fp, needrights, &fp_fromcap); 2301 else 2302 error = cap_funwrap_mmap(fp, needrights, maxprotp, 2303 &fp_fromcap); 2304 if (error) { 2305 fdrop(fp, td); 2306 return (error); 2307 } 2308 2309 /* 2310 * If we've unwrapped a file, drop the original capability 2311 * and hold the new descriptor. fp after this point refers to 2312 * the actual (unwrapped) object, not the capability. 2313 */ 2314 if (fp != fp_fromcap) { 2315 fhold(fp_fromcap); 2316 fdrop(fp, td); 2317 fp = fp_fromcap; 2318 } 2319 } 2320#else /* !CAPABILITIES */ 2321 KASSERT(fp->f_type != DTYPE_CAPABILITY, 2322 ("%s: saw capability", __func__)); 2323 if (maxprotp != NULL) 2324 *maxprotp = VM_PROT_ALL; 2325#endif /* CAPABILITIES */ 2326 2327 /* 2328 * FREAD and FWRITE failure return EBADF as per POSIX. 2329 * 2330 * Only one flag, or 0, may be specified. 2331 */ 2332 if ((flags == FREAD && (fp->f_flag & FREAD) == 0) || 2333 (flags == FWRITE && (fp->f_flag & FWRITE) == 0)) { 2334 fdrop(fp, td); 2335 return (EBADF); 2336 } 2337 *fpp = fp; 2338 return (0); 2339} 2340 2341int 2342fget(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2343{ 2344 2345 return(_fget(td, fd, fpp, 0, rights, NULL, NULL, 0)); 2346} 2347 2348int 2349fget_mmap(struct thread *td, int fd, cap_rights_t rights, u_char *maxprotp, 2350 struct file **fpp) 2351{ 2352 2353 return (_fget(td, fd, fpp, 0, rights, NULL, maxprotp, 0)); 2354} 2355 2356int 2357fget_read(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2358{ 2359 2360 return(_fget(td, fd, fpp, FREAD, rights, NULL, NULL, 0)); 2361} 2362 2363int 2364fget_write(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2365{ 2366 2367 return (_fget(td, fd, fpp, FWRITE, rights, NULL, NULL, 0)); 2368} 2369 2370/* 2371 * Unlike the other fget() calls, which accept and check capability rights 2372 * but never return capabilities, fgetcap() returns the capability but doesn't 2373 * check capability rights. 2374 */ 2375int 2376fgetcap(struct thread *td, int fd, struct file **fpp) 2377{ 2378 2379 return (_fget(td, fd, fpp, 0, 0, NULL, NULL, FGET_GETCAP)); 2380} 2381 2382 2383/* 2384 * Like fget() but loads the underlying vnode, or returns an error if the 2385 * descriptor does not represent a vnode. Note that pipes use vnodes but 2386 * never have VM objects. The returned vnode will be vref()'d. 2387 * 2388 * XXX: what about the unused flags ? 2389 */ 2390static __inline int 2391_fgetvp(struct thread *td, int fd, int flags, cap_rights_t needrights, 2392 cap_rights_t *haverightsp, struct vnode **vpp) 2393{ 2394 struct file *fp; 2395 int error; 2396 2397 *vpp = NULL; 2398 if ((error = _fget(td, fd, &fp, flags, needrights, haverightsp, 2399 NULL, 0)) != 0) 2400 return (error); 2401 if (fp->f_vnode == NULL) { 2402 error = EINVAL; 2403 } else { 2404 *vpp = fp->f_vnode; 2405 vref(*vpp); 2406 } 2407 fdrop(fp, td); 2408 2409 return (error); 2410} 2411 2412int 2413fgetvp(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp) 2414{ 2415 2416 return (_fgetvp(td, fd, 0, rights, NULL, vpp)); 2417} 2418 2419int 2420fgetvp_rights(struct thread *td, int fd, cap_rights_t need, cap_rights_t *have, 2421 struct vnode **vpp) 2422{ 2423 return (_fgetvp(td, fd, 0, need, have, vpp)); 2424} 2425 2426int 2427fgetvp_read(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp) 2428{ 2429 2430 return (_fgetvp(td, fd, FREAD, rights, NULL, vpp)); 2431} 2432 2433#ifdef notyet 2434int 2435fgetvp_write(struct thread *td, int fd, cap_rights_t rights, 2436 struct vnode **vpp) 2437{ 2438 2439 return (_fgetvp(td, fd, FWRITE, rights, NULL, vpp)); 2440} 2441#endif 2442 2443/* 2444 * Like fget() but loads the underlying socket, or returns an error if the 2445 * descriptor does not represent a socket. 2446 * 2447 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2448 * in the future. 2449 * 2450 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2451 * on their file descriptor reference to prevent the socket from being free'd 2452 * during use. 2453 */ 2454int 2455fgetsock(struct thread *td, int fd, cap_rights_t rights, struct socket **spp, 2456 u_int *fflagp) 2457{ 2458 struct file *fp; 2459 int error; 2460 2461 *spp = NULL; 2462 if (fflagp != NULL) 2463 *fflagp = 0; 2464 if ((error = _fget(td, fd, &fp, 0, rights, NULL, NULL, 0)) != 0) 2465 return (error); 2466 if (fp->f_type != DTYPE_SOCKET) { 2467 error = ENOTSOCK; 2468 } else { 2469 *spp = fp->f_data; 2470 if (fflagp) 2471 *fflagp = fp->f_flag; 2472 SOCK_LOCK(*spp); 2473 soref(*spp); 2474 SOCK_UNLOCK(*spp); 2475 } 2476 fdrop(fp, td); 2477 2478 return (error); 2479} 2480 2481/* 2482 * Drop the reference count on the socket and XXX release the SX lock in the 2483 * future. The last reference closes the socket. 2484 * 2485 * Note: fputsock() is deprecated, see comment for fgetsock(). 2486 */ 2487void 2488fputsock(struct socket *so) 2489{ 2490 2491 ACCEPT_LOCK(); 2492 SOCK_LOCK(so); 2493 CURVNET_SET(so->so_vnet); 2494 sorele(so); 2495 CURVNET_RESTORE(); 2496} 2497 2498/* 2499 * Handle the last reference to a file being closed. 2500 * 2501 * No special capability handling here, as the capability's fo_close will run 2502 * instead of the object here, and perform any necessary drop on the object. 2503 */ 2504int 2505_fdrop(struct file *fp, struct thread *td) 2506{ 2507 int error; 2508 2509 error = 0; 2510 if (fp->f_count != 0) 2511 panic("fdrop: count %d", fp->f_count); 2512 if (fp->f_ops != &badfileops) 2513 error = fo_close(fp, td); 2514 atomic_subtract_int(&openfiles, 1); 2515 crfree(fp->f_cred); 2516 free(fp->f_advice, M_FADVISE); 2517 uma_zfree(file_zone, fp); 2518 2519 return (error); 2520} 2521 2522/* 2523 * Apply an advisory lock on a file descriptor. 2524 * 2525 * Just attempt to get a record lock of the requested type on the entire file 2526 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2527 */ 2528#ifndef _SYS_SYSPROTO_H_ 2529struct flock_args { 2530 int fd; 2531 int how; 2532}; 2533#endif 2534/* ARGSUSED */ 2535int 2536sys_flock(struct thread *td, struct flock_args *uap) 2537{ 2538 struct file *fp; 2539 struct vnode *vp; 2540 struct flock lf; 2541 int vfslocked; 2542 int error; 2543 2544 if ((error = fget(td, uap->fd, CAP_FLOCK, &fp)) != 0) 2545 return (error); 2546 if (fp->f_type != DTYPE_VNODE) { 2547 fdrop(fp, td); 2548 return (EOPNOTSUPP); 2549 } 2550 2551 vp = fp->f_vnode; 2552 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2553 lf.l_whence = SEEK_SET; 2554 lf.l_start = 0; 2555 lf.l_len = 0; 2556 if (uap->how & LOCK_UN) { 2557 lf.l_type = F_UNLCK; 2558 atomic_clear_int(&fp->f_flag, FHASLOCK); 2559 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2560 goto done2; 2561 } 2562 if (uap->how & LOCK_EX) 2563 lf.l_type = F_WRLCK; 2564 else if (uap->how & LOCK_SH) 2565 lf.l_type = F_RDLCK; 2566 else { 2567 error = EBADF; 2568 goto done2; 2569 } 2570 atomic_set_int(&fp->f_flag, FHASLOCK); 2571 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2572 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2573done2: 2574 fdrop(fp, td); 2575 VFS_UNLOCK_GIANT(vfslocked); 2576 return (error); 2577} 2578/* 2579 * Duplicate the specified descriptor to a free descriptor. 2580 */ 2581int 2582dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, int openerror, int *indxp) 2583{ 2584 struct file *fp; 2585 int error, indx; 2586 2587 KASSERT(openerror == ENODEV || openerror == ENXIO, 2588 ("unexpected error %d in %s", openerror, __func__)); 2589 2590 /* 2591 * If the to-be-dup'd fd number is greater than the allowed number 2592 * of file descriptors, or the fd to be dup'd has already been 2593 * closed, then reject. 2594 */ 2595 FILEDESC_XLOCK(fdp); 2596 if (dfd < 0 || dfd >= fdp->fd_nfiles || 2597 (fp = fdp->fd_ofiles[dfd]) == NULL) { 2598 FILEDESC_XUNLOCK(fdp); 2599 return (EBADF); 2600 } 2601 2602 error = fdalloc(td, 0, &indx); 2603 if (error != 0) { 2604 FILEDESC_XUNLOCK(fdp); 2605 return (error); 2606 } 2607 2608 /* 2609 * There are two cases of interest here. 2610 * 2611 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2612 * 2613 * For ENXIO steal away the file structure from (dfd) and store it in 2614 * (indx). (dfd) is effectively closed by this operation. 2615 */ 2616 switch (openerror) { 2617 case ENODEV: 2618 /* 2619 * Check that the mode the file is being opened for is a 2620 * subset of the mode of the existing descriptor. 2621 */ 2622 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2623 fdunused(fdp, indx); 2624 FILEDESC_XUNLOCK(fdp); 2625 return (EACCES); 2626 } 2627 fdp->fd_ofiles[indx] = fp; 2628 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2629 fhold(fp); 2630 break; 2631 case ENXIO: 2632 /* 2633 * Steal away the file pointer from dfd and stuff it into indx. 2634 */ 2635 fdp->fd_ofiles[indx] = fp; 2636 fdp->fd_ofiles[dfd] = NULL; 2637 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2638 fdp->fd_ofileflags[dfd] = 0; 2639 fdunused(fdp, dfd); 2640 break; 2641 } 2642 FILEDESC_XUNLOCK(fdp); 2643 *indxp = indx; 2644 return (0); 2645} 2646 2647/* 2648 * Scan all active processes and prisons to see if any of them have a current 2649 * or root directory of `olddp'. If so, replace them with the new mount point. 2650 */ 2651void 2652mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2653{ 2654 struct filedesc *fdp; 2655 struct prison *pr; 2656 struct proc *p; 2657 int nrele; 2658 2659 if (vrefcnt(olddp) == 1) 2660 return; 2661 nrele = 0; 2662 sx_slock(&allproc_lock); 2663 FOREACH_PROC_IN_SYSTEM(p) { 2664 fdp = fdhold(p); 2665 if (fdp == NULL) 2666 continue; 2667 FILEDESC_XLOCK(fdp); 2668 if (fdp->fd_cdir == olddp) { 2669 vref(newdp); 2670 fdp->fd_cdir = newdp; 2671 nrele++; 2672 } 2673 if (fdp->fd_rdir == olddp) { 2674 vref(newdp); 2675 fdp->fd_rdir = newdp; 2676 nrele++; 2677 } 2678 if (fdp->fd_jdir == olddp) { 2679 vref(newdp); 2680 fdp->fd_jdir = newdp; 2681 nrele++; 2682 } 2683 FILEDESC_XUNLOCK(fdp); 2684 fddrop(fdp); 2685 } 2686 sx_sunlock(&allproc_lock); 2687 if (rootvnode == olddp) { 2688 vref(newdp); 2689 rootvnode = newdp; 2690 nrele++; 2691 } 2692 mtx_lock(&prison0.pr_mtx); 2693 if (prison0.pr_root == olddp) { 2694 vref(newdp); 2695 prison0.pr_root = newdp; 2696 nrele++; 2697 } 2698 mtx_unlock(&prison0.pr_mtx); 2699 sx_slock(&allprison_lock); 2700 TAILQ_FOREACH(pr, &allprison, pr_list) { 2701 mtx_lock(&pr->pr_mtx); 2702 if (pr->pr_root == olddp) { 2703 vref(newdp); 2704 pr->pr_root = newdp; 2705 nrele++; 2706 } 2707 mtx_unlock(&pr->pr_mtx); 2708 } 2709 sx_sunlock(&allprison_lock); 2710 while (nrele--) 2711 vrele(olddp); 2712} 2713 2714struct filedesc_to_leader * 2715filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2716{ 2717 struct filedesc_to_leader *fdtol; 2718 2719 fdtol = malloc(sizeof(struct filedesc_to_leader), 2720 M_FILEDESC_TO_LEADER, 2721 M_WAITOK); 2722 fdtol->fdl_refcount = 1; 2723 fdtol->fdl_holdcount = 0; 2724 fdtol->fdl_wakeup = 0; 2725 fdtol->fdl_leader = leader; 2726 if (old != NULL) { 2727 FILEDESC_XLOCK(fdp); 2728 fdtol->fdl_next = old->fdl_next; 2729 fdtol->fdl_prev = old; 2730 old->fdl_next = fdtol; 2731 fdtol->fdl_next->fdl_prev = fdtol; 2732 FILEDESC_XUNLOCK(fdp); 2733 } else { 2734 fdtol->fdl_next = fdtol; 2735 fdtol->fdl_prev = fdtol; 2736 } 2737 return (fdtol); 2738} 2739 2740/* 2741 * Get file structures globally. 2742 */ 2743static int 2744sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2745{ 2746 struct xfile xf; 2747 struct filedesc *fdp; 2748 struct file *fp; 2749 struct proc *p; 2750 int error, n; 2751 2752 error = sysctl_wire_old_buffer(req, 0); 2753 if (error != 0) 2754 return (error); 2755 if (req->oldptr == NULL) { 2756 n = 0; 2757 sx_slock(&allproc_lock); 2758 FOREACH_PROC_IN_SYSTEM(p) { 2759 if (p->p_state == PRS_NEW) 2760 continue; 2761 fdp = fdhold(p); 2762 if (fdp == NULL) 2763 continue; 2764 /* overestimates sparse tables. */ 2765 if (fdp->fd_lastfile > 0) 2766 n += fdp->fd_lastfile; 2767 fddrop(fdp); 2768 } 2769 sx_sunlock(&allproc_lock); 2770 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2771 } 2772 error = 0; 2773 bzero(&xf, sizeof(xf)); 2774 xf.xf_size = sizeof(xf); 2775 sx_slock(&allproc_lock); 2776 FOREACH_PROC_IN_SYSTEM(p) { 2777 PROC_LOCK(p); 2778 if (p->p_state == PRS_NEW) { 2779 PROC_UNLOCK(p); 2780 continue; 2781 } 2782 if (p_cansee(req->td, p) != 0) { 2783 PROC_UNLOCK(p); 2784 continue; 2785 } 2786 xf.xf_pid = p->p_pid; 2787 xf.xf_uid = p->p_ucred->cr_uid; 2788 PROC_UNLOCK(p); 2789 fdp = fdhold(p); 2790 if (fdp == NULL) 2791 continue; 2792 FILEDESC_SLOCK(fdp); 2793 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2794 if ((fp = fdp->fd_ofiles[n]) == NULL) 2795 continue; 2796 xf.xf_fd = n; 2797 xf.xf_file = fp; 2798 xf.xf_data = fp->f_data; 2799 xf.xf_vnode = fp->f_vnode; 2800 xf.xf_type = fp->f_type; 2801 xf.xf_count = fp->f_count; 2802 xf.xf_msgcount = 0; 2803 xf.xf_offset = fp->f_offset; 2804 xf.xf_flag = fp->f_flag; 2805 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2806 if (error) 2807 break; 2808 } 2809 FILEDESC_SUNLOCK(fdp); 2810 fddrop(fdp); 2811 if (error) 2812 break; 2813 } 2814 sx_sunlock(&allproc_lock); 2815 return (error); 2816} 2817 2818SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2819 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2820 2821#ifdef KINFO_OFILE_SIZE 2822CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2823#endif 2824 2825#ifdef COMPAT_FREEBSD7 2826static int 2827export_vnode_for_osysctl(struct vnode *vp, int type, 2828 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2829{ 2830 int error; 2831 char *fullpath, *freepath; 2832 int vfslocked; 2833 2834 bzero(kif, sizeof(*kif)); 2835 kif->kf_structsize = sizeof(*kif); 2836 2837 vref(vp); 2838 kif->kf_fd = type; 2839 kif->kf_type = KF_TYPE_VNODE; 2840 /* This function only handles directories. */ 2841 if (vp->v_type != VDIR) { 2842 vrele(vp); 2843 return (ENOTDIR); 2844 } 2845 kif->kf_vnode_type = KF_VTYPE_VDIR; 2846 2847 /* 2848 * This is not a true file descriptor, so we set a bogus refcount 2849 * and offset to indicate these fields should be ignored. 2850 */ 2851 kif->kf_ref_count = -1; 2852 kif->kf_offset = -1; 2853 2854 freepath = NULL; 2855 fullpath = "-"; 2856 FILEDESC_SUNLOCK(fdp); 2857 vn_fullpath(curthread, vp, &fullpath, &freepath); 2858 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2859 vrele(vp); 2860 VFS_UNLOCK_GIANT(vfslocked); 2861 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2862 if (freepath != NULL) 2863 free(freepath, M_TEMP); 2864 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2865 FILEDESC_SLOCK(fdp); 2866 return (error); 2867} 2868 2869/* 2870 * Get per-process file descriptors for use by procstat(1), et al. 2871 */ 2872static int 2873sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 2874{ 2875 char *fullpath, *freepath; 2876 struct kinfo_ofile *kif; 2877 struct filedesc *fdp; 2878 int error, i, *name; 2879 struct shmfd *shmfd; 2880 struct socket *so; 2881 struct vnode *vp; 2882 struct file *fp; 2883 struct proc *p; 2884 struct tty *tp; 2885 int vfslocked; 2886 2887 name = (int *)arg1; 2888 if ((p = pfind((pid_t)name[0])) == NULL) 2889 return (ESRCH); 2890 if ((error = p_candebug(curthread, p))) { 2891 PROC_UNLOCK(p); 2892 return (error); 2893 } 2894 fdp = fdhold(p); 2895 PROC_UNLOCK(p); 2896 if (fdp == NULL) 2897 return (ENOENT); 2898 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 2899 FILEDESC_SLOCK(fdp); 2900 if (fdp->fd_cdir != NULL) 2901 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 2902 fdp, req); 2903 if (fdp->fd_rdir != NULL) 2904 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 2905 fdp, req); 2906 if (fdp->fd_jdir != NULL) 2907 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 2908 fdp, req); 2909 for (i = 0; i < fdp->fd_nfiles; i++) { 2910 if ((fp = fdp->fd_ofiles[i]) == NULL) 2911 continue; 2912 bzero(kif, sizeof(*kif)); 2913 kif->kf_structsize = sizeof(*kif); 2914 vp = NULL; 2915 so = NULL; 2916 tp = NULL; 2917 shmfd = NULL; 2918 kif->kf_fd = i; 2919 2920#ifdef CAPABILITIES 2921 /* 2922 * When reporting a capability, most fields will be from the 2923 * underlying object, but do mark as a capability. With 2924 * ofiledesc, we don't have a field to export the cap_rights_t, 2925 * but we do with the new filedesc. 2926 */ 2927 if (fp->f_type == DTYPE_CAPABILITY) { 2928 kif->kf_flags |= KF_FLAG_CAPABILITY; 2929 (void)cap_funwrap(fp, 0, &fp); 2930 } 2931#else 2932 KASSERT(fp->f_type != DTYPE_CAPABILITY, 2933 ("sysctl_kern_proc_ofiledesc: saw capability")); 2934#endif 2935 switch (fp->f_type) { 2936 case DTYPE_VNODE: 2937 kif->kf_type = KF_TYPE_VNODE; 2938 vp = fp->f_vnode; 2939 break; 2940 2941 case DTYPE_SOCKET: 2942 kif->kf_type = KF_TYPE_SOCKET; 2943 so = fp->f_data; 2944 break; 2945 2946 case DTYPE_PIPE: 2947 kif->kf_type = KF_TYPE_PIPE; 2948 break; 2949 2950 case DTYPE_FIFO: 2951 kif->kf_type = KF_TYPE_FIFO; 2952 vp = fp->f_vnode; 2953 break; 2954 2955 case DTYPE_KQUEUE: 2956 kif->kf_type = KF_TYPE_KQUEUE; 2957 break; 2958 2959 case DTYPE_CRYPTO: 2960 kif->kf_type = KF_TYPE_CRYPTO; 2961 break; 2962 2963 case DTYPE_MQUEUE: 2964 kif->kf_type = KF_TYPE_MQUEUE; 2965 break; 2966 2967 case DTYPE_SHM: 2968 kif->kf_type = KF_TYPE_SHM; 2969 shmfd = fp->f_data; 2970 break; 2971 2972 case DTYPE_SEM: 2973 kif->kf_type = KF_TYPE_SEM; 2974 break; 2975 2976 case DTYPE_PTS: 2977 kif->kf_type = KF_TYPE_PTS; 2978 tp = fp->f_data; 2979 break; 2980 2981#ifdef PROCDESC 2982 case DTYPE_PROCDESC: 2983 kif->kf_type = KF_TYPE_PROCDESC; 2984 break; 2985#endif 2986 2987 default: 2988 kif->kf_type = KF_TYPE_UNKNOWN; 2989 break; 2990 } 2991 kif->kf_ref_count = fp->f_count; 2992 if (fp->f_flag & FREAD) 2993 kif->kf_flags |= KF_FLAG_READ; 2994 if (fp->f_flag & FWRITE) 2995 kif->kf_flags |= KF_FLAG_WRITE; 2996 if (fp->f_flag & FAPPEND) 2997 kif->kf_flags |= KF_FLAG_APPEND; 2998 if (fp->f_flag & FASYNC) 2999 kif->kf_flags |= KF_FLAG_ASYNC; 3000 if (fp->f_flag & FFSYNC) 3001 kif->kf_flags |= KF_FLAG_FSYNC; 3002 if (fp->f_flag & FNONBLOCK) 3003 kif->kf_flags |= KF_FLAG_NONBLOCK; 3004 if (fp->f_flag & O_DIRECT) 3005 kif->kf_flags |= KF_FLAG_DIRECT; 3006 if (fp->f_flag & FHASLOCK) 3007 kif->kf_flags |= KF_FLAG_HASLOCK; 3008 kif->kf_offset = fp->f_offset; 3009 if (vp != NULL) { 3010 vref(vp); 3011 switch (vp->v_type) { 3012 case VNON: 3013 kif->kf_vnode_type = KF_VTYPE_VNON; 3014 break; 3015 case VREG: 3016 kif->kf_vnode_type = KF_VTYPE_VREG; 3017 break; 3018 case VDIR: 3019 kif->kf_vnode_type = KF_VTYPE_VDIR; 3020 break; 3021 case VBLK: 3022 kif->kf_vnode_type = KF_VTYPE_VBLK; 3023 break; 3024 case VCHR: 3025 kif->kf_vnode_type = KF_VTYPE_VCHR; 3026 break; 3027 case VLNK: 3028 kif->kf_vnode_type = KF_VTYPE_VLNK; 3029 break; 3030 case VSOCK: 3031 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3032 break; 3033 case VFIFO: 3034 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3035 break; 3036 case VBAD: 3037 kif->kf_vnode_type = KF_VTYPE_VBAD; 3038 break; 3039 default: 3040 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3041 break; 3042 } 3043 /* 3044 * It is OK to drop the filedesc lock here as we will 3045 * re-validate and re-evaluate its properties when 3046 * the loop continues. 3047 */ 3048 freepath = NULL; 3049 fullpath = "-"; 3050 FILEDESC_SUNLOCK(fdp); 3051 vn_fullpath(curthread, vp, &fullpath, &freepath); 3052 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3053 vrele(vp); 3054 VFS_UNLOCK_GIANT(vfslocked); 3055 strlcpy(kif->kf_path, fullpath, 3056 sizeof(kif->kf_path)); 3057 if (freepath != NULL) 3058 free(freepath, M_TEMP); 3059 FILEDESC_SLOCK(fdp); 3060 } 3061 if (so != NULL) { 3062 struct sockaddr *sa; 3063 3064 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3065 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3066 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3067 free(sa, M_SONAME); 3068 } 3069 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3070 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3071 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3072 free(sa, M_SONAME); 3073 } 3074 kif->kf_sock_domain = 3075 so->so_proto->pr_domain->dom_family; 3076 kif->kf_sock_type = so->so_type; 3077 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3078 } 3079 if (tp != NULL) { 3080 strlcpy(kif->kf_path, tty_devname(tp), 3081 sizeof(kif->kf_path)); 3082 } 3083 if (shmfd != NULL) 3084 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3085 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3086 if (error) 3087 break; 3088 } 3089 FILEDESC_SUNLOCK(fdp); 3090 fddrop(fdp); 3091 free(kif, M_TEMP); 3092 return (0); 3093} 3094 3095static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD, 3096 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); 3097#endif /* COMPAT_FREEBSD7 */ 3098 3099#ifdef KINFO_FILE_SIZE 3100CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3101#endif 3102 3103static int 3104export_fd_for_sysctl(void *data, int type, int fd, int fflags, int refcnt, 3105 int64_t offset, int fd_is_cap, cap_rights_t fd_cap_rights, 3106 struct kinfo_file *kif, struct sysctl_req *req) 3107{ 3108 struct { 3109 int fflag; 3110 int kf_fflag; 3111 } fflags_table[] = { 3112 { FAPPEND, KF_FLAG_APPEND }, 3113 { FASYNC, KF_FLAG_ASYNC }, 3114 { FFSYNC, KF_FLAG_FSYNC }, 3115 { FHASLOCK, KF_FLAG_HASLOCK }, 3116 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3117 { FREAD, KF_FLAG_READ }, 3118 { FWRITE, KF_FLAG_WRITE }, 3119 { O_CREAT, KF_FLAG_CREAT }, 3120 { O_DIRECT, KF_FLAG_DIRECT }, 3121 { O_EXCL, KF_FLAG_EXCL }, 3122 { O_EXEC, KF_FLAG_EXEC }, 3123 { O_EXLOCK, KF_FLAG_EXLOCK }, 3124 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3125 { O_SHLOCK, KF_FLAG_SHLOCK }, 3126 { O_TRUNC, KF_FLAG_TRUNC } 3127 }; 3128#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3129 struct vnode *vp; 3130 int error, vfslocked; 3131 unsigned int i; 3132 3133 bzero(kif, sizeof(*kif)); 3134 switch (type) { 3135 case KF_TYPE_FIFO: 3136 case KF_TYPE_VNODE: 3137 vp = (struct vnode *)data; 3138 error = fill_vnode_info(vp, kif); 3139 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3140 vrele(vp); 3141 VFS_UNLOCK_GIANT(vfslocked); 3142 break; 3143 case KF_TYPE_SOCKET: 3144 error = fill_socket_info((struct socket *)data, kif); 3145 break; 3146 case KF_TYPE_PIPE: 3147 error = fill_pipe_info((struct pipe *)data, kif); 3148 break; 3149 case KF_TYPE_PTS: 3150 error = fill_pts_info((struct tty *)data, kif); 3151 break; 3152 case KF_TYPE_PROCDESC: 3153 error = fill_procdesc_info((struct procdesc *)data, kif); 3154 break; 3155 case KF_TYPE_SHM: 3156 error = fill_shm_info((struct file *)data, kif); 3157 break; 3158 default: 3159 error = 0; 3160 } 3161 if (error == 0) 3162 kif->kf_status |= KF_ATTR_VALID; 3163 3164 /* 3165 * Translate file access flags. 3166 */ 3167 for (i = 0; i < NFFLAGS; i++) 3168 if (fflags & fflags_table[i].fflag) 3169 kif->kf_flags |= fflags_table[i].kf_fflag; 3170 if (fd_is_cap) 3171 kif->kf_flags |= KF_FLAG_CAPABILITY; 3172 if (fd_is_cap) 3173 kif->kf_cap_rights = fd_cap_rights; 3174 kif->kf_fd = fd; 3175 kif->kf_type = type; 3176 kif->kf_ref_count = refcnt; 3177 kif->kf_offset = offset; 3178 /* Pack record size down */ 3179 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3180 strlen(kif->kf_path) + 1; 3181 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3182 error = SYSCTL_OUT(req, kif, kif->kf_structsize); 3183 return (error); 3184} 3185 3186/* 3187 * Get per-process file descriptors for use by procstat(1), et al. 3188 */ 3189static int 3190sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3191{ 3192 struct file *fp; 3193 struct filedesc *fdp; 3194 struct kinfo_file *kif; 3195 struct proc *p; 3196 struct vnode *cttyvp, *textvp, *tracevp; 3197 size_t oldidx; 3198 int64_t offset; 3199 void *data; 3200 int error, i, *name; 3201 int fd_is_cap, type, refcnt, fflags; 3202 cap_rights_t fd_cap_rights; 3203 3204 name = (int *)arg1; 3205 if ((p = pfind((pid_t)name[0])) == NULL) 3206 return (ESRCH); 3207 if ((error = p_candebug(curthread, p))) { 3208 PROC_UNLOCK(p); 3209 return (error); 3210 } 3211 /* ktrace vnode */ 3212 tracevp = p->p_tracevp; 3213 if (tracevp != NULL) 3214 vref(tracevp); 3215 /* text vnode */ 3216 textvp = p->p_textvp; 3217 if (textvp != NULL) 3218 vref(textvp); 3219 /* Controlling tty. */ 3220 cttyvp = NULL; 3221 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3222 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3223 if (cttyvp != NULL) 3224 vref(cttyvp); 3225 } 3226 fdp = fdhold(p); 3227 PROC_UNLOCK(p); 3228 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3229 if (tracevp != NULL) 3230 export_fd_for_sysctl(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3231 FREAD | FWRITE, -1, -1, 0, 0, kif, req); 3232 if (textvp != NULL) 3233 export_fd_for_sysctl(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3234 FREAD, -1, -1, 0, 0, kif, req); 3235 if (cttyvp != NULL) 3236 export_fd_for_sysctl(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3237 FREAD | FWRITE, -1, -1, 0, 0, kif, req); 3238 if (fdp == NULL) 3239 goto fail; 3240 FILEDESC_SLOCK(fdp); 3241 /* working directory */ 3242 if (fdp->fd_cdir != NULL) { 3243 vref(fdp->fd_cdir); 3244 data = fdp->fd_cdir; 3245 FILEDESC_SUNLOCK(fdp); 3246 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3247 FREAD, -1, -1, 0, 0, kif, req); 3248 FILEDESC_SLOCK(fdp); 3249 } 3250 /* root directory */ 3251 if (fdp->fd_rdir != NULL) { 3252 vref(fdp->fd_rdir); 3253 data = fdp->fd_rdir; 3254 FILEDESC_SUNLOCK(fdp); 3255 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3256 FREAD, -1, -1, 0, 0, kif, req); 3257 FILEDESC_SLOCK(fdp); 3258 } 3259 /* jail directory */ 3260 if (fdp->fd_jdir != NULL) { 3261 vref(fdp->fd_jdir); 3262 data = fdp->fd_jdir; 3263 FILEDESC_SUNLOCK(fdp); 3264 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3265 FREAD, -1, -1, 0, 0, kif, req); 3266 FILEDESC_SLOCK(fdp); 3267 } 3268 for (i = 0; i < fdp->fd_nfiles; i++) { 3269 if ((fp = fdp->fd_ofiles[i]) == NULL) 3270 continue; 3271 data = NULL; 3272 fd_is_cap = 0; 3273 fd_cap_rights = 0; 3274 3275#ifdef CAPABILITIES 3276 /* 3277 * When reporting a capability, most fields will be from the 3278 * underlying object, but do mark as a capability and export 3279 * the capability rights mask. 3280 */ 3281 if (fp->f_type == DTYPE_CAPABILITY) { 3282 fd_is_cap = 1; 3283 fd_cap_rights = cap_rights(fp); 3284 (void)cap_funwrap(fp, 0, &fp); 3285 } 3286#else /* !CAPABILITIES */ 3287 KASSERT(fp->f_type != DTYPE_CAPABILITY, 3288 ("sysctl_kern_proc_filedesc: saw capability")); 3289#endif 3290 switch (fp->f_type) { 3291 case DTYPE_VNODE: 3292 type = KF_TYPE_VNODE; 3293 vref(fp->f_vnode); 3294 data = fp->f_vnode; 3295 break; 3296 3297 case DTYPE_SOCKET: 3298 type = KF_TYPE_SOCKET; 3299 data = fp->f_data; 3300 break; 3301 3302 case DTYPE_PIPE: 3303 type = KF_TYPE_PIPE; 3304 data = fp->f_data; 3305 break; 3306 3307 case DTYPE_FIFO: 3308 type = KF_TYPE_FIFO; 3309 vref(fp->f_vnode); 3310 data = fp->f_vnode; 3311 break; 3312 3313 case DTYPE_KQUEUE: 3314 type = KF_TYPE_KQUEUE; 3315 break; 3316 3317 case DTYPE_CRYPTO: 3318 type = KF_TYPE_CRYPTO; 3319 break; 3320 3321 case DTYPE_MQUEUE: 3322 type = KF_TYPE_MQUEUE; 3323 break; 3324 3325 case DTYPE_SHM: 3326 type = KF_TYPE_SHM; 3327 data = fp; 3328 break; 3329 3330 case DTYPE_SEM: 3331 type = KF_TYPE_SEM; 3332 break; 3333 3334 case DTYPE_PTS: 3335 type = KF_TYPE_PTS; 3336 data = fp->f_data; 3337 break; 3338 3339#ifdef PROCDESC 3340 case DTYPE_PROCDESC: 3341 type = KF_TYPE_PROCDESC; 3342 data = fp->f_data; 3343 break; 3344#endif 3345 3346 default: 3347 type = KF_TYPE_UNKNOWN; 3348 break; 3349 } 3350 refcnt = fp->f_count; 3351 fflags = fp->f_flag; 3352 offset = fp->f_offset; 3353 3354 /* 3355 * Create sysctl entry. 3356 * It is OK to drop the filedesc lock here as we will 3357 * re-validate and re-evaluate its properties when 3358 * the loop continues. 3359 */ 3360 oldidx = req->oldidx; 3361 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO) 3362 FILEDESC_SUNLOCK(fdp); 3363 error = export_fd_for_sysctl(data, type, i, fflags, refcnt, 3364 offset, fd_is_cap, fd_cap_rights, kif, req); 3365 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO) 3366 FILEDESC_SLOCK(fdp); 3367 if (error) { 3368 if (error == ENOMEM) { 3369 /* 3370 * The hack to keep the ABI of sysctl 3371 * kern.proc.filedesc intact, but not 3372 * to account a partially copied 3373 * kinfo_file into the oldidx. 3374 */ 3375 req->oldidx = oldidx; 3376 error = 0; 3377 } 3378 break; 3379 } 3380 } 3381 FILEDESC_SUNLOCK(fdp); 3382fail: 3383 if (fdp != NULL) 3384 fddrop(fdp); 3385 free(kif, M_TEMP); 3386 return (error); 3387} 3388 3389int 3390vntype_to_kinfo(int vtype) 3391{ 3392 struct { 3393 int vtype; 3394 int kf_vtype; 3395 } vtypes_table[] = { 3396 { VBAD, KF_VTYPE_VBAD }, 3397 { VBLK, KF_VTYPE_VBLK }, 3398 { VCHR, KF_VTYPE_VCHR }, 3399 { VDIR, KF_VTYPE_VDIR }, 3400 { VFIFO, KF_VTYPE_VFIFO }, 3401 { VLNK, KF_VTYPE_VLNK }, 3402 { VNON, KF_VTYPE_VNON }, 3403 { VREG, KF_VTYPE_VREG }, 3404 { VSOCK, KF_VTYPE_VSOCK } 3405 }; 3406#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3407 unsigned int i; 3408 3409 /* 3410 * Perform vtype translation. 3411 */ 3412 for (i = 0; i < NVTYPES; i++) 3413 if (vtypes_table[i].vtype == vtype) 3414 break; 3415 if (i < NVTYPES) 3416 return (vtypes_table[i].kf_vtype); 3417 3418 return (KF_VTYPE_UNKNOWN); 3419} 3420 3421static int 3422fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3423{ 3424 struct vattr va; 3425 char *fullpath, *freepath; 3426 int error, vfslocked; 3427 3428 if (vp == NULL) 3429 return (1); 3430 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3431 freepath = NULL; 3432 fullpath = "-"; 3433 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3434 if (error == 0) { 3435 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3436 } 3437 if (freepath != NULL) 3438 free(freepath, M_TEMP); 3439 3440 /* 3441 * Retrieve vnode attributes. 3442 */ 3443 va.va_fsid = VNOVAL; 3444 va.va_rdev = NODEV; 3445 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3446 vn_lock(vp, LK_SHARED | LK_RETRY); 3447 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3448 VOP_UNLOCK(vp, 0); 3449 VFS_UNLOCK_GIANT(vfslocked); 3450 if (error != 0) 3451 return (error); 3452 if (va.va_fsid != VNOVAL) 3453 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3454 else 3455 kif->kf_un.kf_file.kf_file_fsid = 3456 vp->v_mount->mnt_stat.f_fsid.val[0]; 3457 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3458 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3459 kif->kf_un.kf_file.kf_file_size = va.va_size; 3460 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3461 return (0); 3462} 3463 3464static int 3465fill_socket_info(struct socket *so, struct kinfo_file *kif) 3466{ 3467 struct sockaddr *sa; 3468 struct inpcb *inpcb; 3469 struct unpcb *unpcb; 3470 int error; 3471 3472 if (so == NULL) 3473 return (1); 3474 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3475 kif->kf_sock_type = so->so_type; 3476 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3477 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3478 switch(kif->kf_sock_domain) { 3479 case AF_INET: 3480 case AF_INET6: 3481 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3482 if (so->so_pcb != NULL) { 3483 inpcb = (struct inpcb *)(so->so_pcb); 3484 kif->kf_un.kf_sock.kf_sock_inpcb = 3485 (uintptr_t)inpcb->inp_ppcb; 3486 } 3487 } 3488 break; 3489 case AF_UNIX: 3490 if (so->so_pcb != NULL) { 3491 unpcb = (struct unpcb *)(so->so_pcb); 3492 if (unpcb->unp_conn) { 3493 kif->kf_un.kf_sock.kf_sock_unpconn = 3494 (uintptr_t)unpcb->unp_conn; 3495 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3496 so->so_rcv.sb_state; 3497 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3498 so->so_snd.sb_state; 3499 } 3500 } 3501 break; 3502 } 3503 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3504 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3505 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3506 free(sa, M_SONAME); 3507 } 3508 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3509 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3510 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3511 free(sa, M_SONAME); 3512 } 3513 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3514 sizeof(kif->kf_path)); 3515 return (0); 3516} 3517 3518static int 3519fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3520{ 3521 3522 if (tp == NULL) 3523 return (1); 3524 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3525 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3526 return (0); 3527} 3528 3529static int 3530fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3531{ 3532 3533 if (pi == NULL) 3534 return (1); 3535 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3536 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3537 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3538 return (0); 3539} 3540 3541static int 3542fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3543{ 3544 3545 if (pdp == NULL) 3546 return (1); 3547 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3548 return (0); 3549} 3550 3551static int 3552fill_shm_info(struct file *fp, struct kinfo_file *kif) 3553{ 3554 struct thread *td; 3555 struct stat sb; 3556 3557 td = curthread; 3558 if (fp->f_data == NULL) 3559 return (1); 3560 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3561 return (1); 3562 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3563 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3564 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3565 return (0); 3566} 3567 3568static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD, 3569 sysctl_kern_proc_filedesc, "Process filedesc entries"); 3570 3571#ifdef DDB 3572/* 3573 * For the purposes of debugging, generate a human-readable string for the 3574 * file type. 3575 */ 3576static const char * 3577file_type_to_name(short type) 3578{ 3579 3580 switch (type) { 3581 case 0: 3582 return ("zero"); 3583 case DTYPE_VNODE: 3584 return ("vnod"); 3585 case DTYPE_SOCKET: 3586 return ("sock"); 3587 case DTYPE_PIPE: 3588 return ("pipe"); 3589 case DTYPE_FIFO: 3590 return ("fifo"); 3591 case DTYPE_KQUEUE: 3592 return ("kque"); 3593 case DTYPE_CRYPTO: 3594 return ("crpt"); 3595 case DTYPE_MQUEUE: 3596 return ("mque"); 3597 case DTYPE_SHM: 3598 return ("shm"); 3599 case DTYPE_SEM: 3600 return ("ksem"); 3601 default: 3602 return ("unkn"); 3603 } 3604} 3605 3606/* 3607 * For the purposes of debugging, identify a process (if any, perhaps one of 3608 * many) that references the passed file in its file descriptor array. Return 3609 * NULL if none. 3610 */ 3611static struct proc * 3612file_to_first_proc(struct file *fp) 3613{ 3614 struct filedesc *fdp; 3615 struct proc *p; 3616 int n; 3617 3618 FOREACH_PROC_IN_SYSTEM(p) { 3619 if (p->p_state == PRS_NEW) 3620 continue; 3621 fdp = p->p_fd; 3622 if (fdp == NULL) 3623 continue; 3624 for (n = 0; n < fdp->fd_nfiles; n++) { 3625 if (fp == fdp->fd_ofiles[n]) 3626 return (p); 3627 } 3628 } 3629 return (NULL); 3630} 3631 3632static void 3633db_print_file(struct file *fp, int header) 3634{ 3635 struct proc *p; 3636 3637 if (header) 3638 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3639 "File", "Type", "Data", "Flag", "GCFl", "Count", 3640 "MCount", "Vnode", "FPID", "FCmd"); 3641 p = file_to_first_proc(fp); 3642 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3643 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3644 0, fp->f_count, 0, fp->f_vnode, 3645 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3646} 3647 3648DB_SHOW_COMMAND(file, db_show_file) 3649{ 3650 struct file *fp; 3651 3652 if (!have_addr) { 3653 db_printf("usage: show file <addr>\n"); 3654 return; 3655 } 3656 fp = (struct file *)addr; 3657 db_print_file(fp, 1); 3658} 3659 3660DB_SHOW_COMMAND(files, db_show_files) 3661{ 3662 struct filedesc *fdp; 3663 struct file *fp; 3664 struct proc *p; 3665 int header; 3666 int n; 3667 3668 header = 1; 3669 FOREACH_PROC_IN_SYSTEM(p) { 3670 if (p->p_state == PRS_NEW) 3671 continue; 3672 if ((fdp = p->p_fd) == NULL) 3673 continue; 3674 for (n = 0; n < fdp->fd_nfiles; ++n) { 3675 if ((fp = fdp->fd_ofiles[n]) == NULL) 3676 continue; 3677 db_print_file(fp, header); 3678 header = 0; 3679 } 3680 } 3681} 3682#endif 3683 3684SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3685 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3686 3687SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3688 &maxfiles, 0, "Maximum number of files"); 3689 3690SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3691 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3692 3693/* ARGSUSED*/ 3694static void 3695filelistinit(void *dummy) 3696{ 3697 3698 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3699 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3700 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3701 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3702} 3703SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3704 3705/*-------------------------------------------------------------------*/ 3706 3707static int 3708badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3709 int flags, struct thread *td) 3710{ 3711 3712 return (EBADF); 3713} 3714 3715static int 3716badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3717 struct thread *td) 3718{ 3719 3720 return (EINVAL); 3721} 3722 3723static int 3724badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3725 struct thread *td) 3726{ 3727 3728 return (EBADF); 3729} 3730 3731static int 3732badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3733 struct thread *td) 3734{ 3735 3736 return (0); 3737} 3738 3739static int 3740badfo_kqfilter(struct file *fp, struct knote *kn) 3741{ 3742 3743 return (EBADF); 3744} 3745 3746static int 3747badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3748 struct thread *td) 3749{ 3750 3751 return (EBADF); 3752} 3753 3754static int 3755badfo_close(struct file *fp, struct thread *td) 3756{ 3757 3758 return (EBADF); 3759} 3760 3761static int 3762badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3763 struct thread *td) 3764{ 3765 3766 return (EBADF); 3767} 3768 3769static int 3770badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3771 struct thread *td) 3772{ 3773 3774 return (EBADF); 3775} 3776 3777struct fileops badfileops = { 3778 .fo_read = badfo_readwrite, 3779 .fo_write = badfo_readwrite, 3780 .fo_truncate = badfo_truncate, 3781 .fo_ioctl = badfo_ioctl, 3782 .fo_poll = badfo_poll, 3783 .fo_kqfilter = badfo_kqfilter, 3784 .fo_stat = badfo_stat, 3785 .fo_close = badfo_close, 3786 .fo_chmod = badfo_chmod, 3787 .fo_chown = badfo_chown, 3788}; 3789 3790int 3791invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3792 struct thread *td) 3793{ 3794 3795 return (EINVAL); 3796} 3797 3798int 3799invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3800 struct thread *td) 3801{ 3802 3803 return (EINVAL); 3804} 3805 3806/*-------------------------------------------------------------------*/ 3807 3808/* 3809 * File Descriptor pseudo-device driver (/dev/fd/). 3810 * 3811 * Opening minor device N dup()s the file (if any) connected to file 3812 * descriptor N belonging to the calling process. Note that this driver 3813 * consists of only the ``open()'' routine, because all subsequent 3814 * references to this file will be direct to the other driver. 3815 * 3816 * XXX: we could give this one a cloning event handler if necessary. 3817 */ 3818 3819/* ARGSUSED */ 3820static int 3821fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3822{ 3823 3824 /* 3825 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3826 * the file descriptor being sought for duplication. The error 3827 * return ensures that the vnode for this device will be released 3828 * by vn_open. Open will detect this special error and take the 3829 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3830 * will simply report the error. 3831 */ 3832 td->td_dupfd = dev2unit(dev); 3833 return (ENODEV); 3834} 3835 3836static struct cdevsw fildesc_cdevsw = { 3837 .d_version = D_VERSION, 3838 .d_open = fdopen, 3839 .d_name = "FD", 3840}; 3841 3842static void 3843fildesc_drvinit(void *unused) 3844{ 3845 struct cdev *dev; 3846 3847 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3848 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3849 make_dev_alias(dev, "stdin"); 3850 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3851 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3852 make_dev_alias(dev, "stdout"); 3853 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3854 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3855 make_dev_alias(dev, "stderr"); 3856} 3857 3858SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3859