kern_descrip.c revision 236821
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: head/sys/kern/kern_descrip.c 236821 2012-06-09 18:48:06Z pjd $"); 39 40#include "opt_capsicum.h" 41#include "opt_compat.h" 42#include "opt_ddb.h" 43#include "opt_ktrace.h" 44#include "opt_procdesc.h" 45 46#include <sys/param.h> 47#include <sys/systm.h> 48 49#include <sys/capability.h> 50#include <sys/conf.h> 51#include <sys/domain.h> 52#include <sys/fcntl.h> 53#include <sys/file.h> 54#include <sys/filedesc.h> 55#include <sys/filio.h> 56#include <sys/jail.h> 57#include <sys/kernel.h> 58#include <sys/limits.h> 59#include <sys/lock.h> 60#include <sys/malloc.h> 61#include <sys/mman.h> 62#include <sys/mount.h> 63#include <sys/mqueue.h> 64#include <sys/mutex.h> 65#include <sys/namei.h> 66#include <sys/selinfo.h> 67#include <sys/pipe.h> 68#include <sys/priv.h> 69#include <sys/proc.h> 70#include <sys/procdesc.h> 71#include <sys/protosw.h> 72#include <sys/racct.h> 73#include <sys/resourcevar.h> 74#include <sys/signalvar.h> 75#include <sys/socketvar.h> 76#include <sys/stat.h> 77#include <sys/sx.h> 78#include <sys/syscallsubr.h> 79#include <sys/sysctl.h> 80#include <sys/sysproto.h> 81#include <sys/tty.h> 82#include <sys/unistd.h> 83#include <sys/un.h> 84#include <sys/unpcb.h> 85#include <sys/user.h> 86#include <sys/vnode.h> 87#ifdef KTRACE 88#include <sys/ktrace.h> 89#endif 90 91#include <net/vnet.h> 92 93#include <netinet/in.h> 94#include <netinet/in_pcb.h> 95 96#include <security/audit/audit.h> 97 98#include <vm/uma.h> 99#include <vm/vm.h> 100 101#include <ddb/ddb.h> 102 103static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 104static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 105 "file desc to leader structures"); 106static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 107 108MALLOC_DECLARE(M_FADVISE); 109 110static uma_zone_t file_zone; 111 112 113/* Flags for do_dup() */ 114#define DUP_FIXED 0x1 /* Force fixed allocation */ 115#define DUP_FCNTL 0x2 /* fcntl()-style errors */ 116 117static int do_dup(struct thread *td, int flags, int old, int new, 118 register_t *retval); 119static int fd_first_free(struct filedesc *, int, int); 120static int fd_last_used(struct filedesc *, int, int); 121static void fdgrowtable(struct filedesc *, int); 122static void fdunused(struct filedesc *fdp, int fd); 123static void fdused(struct filedesc *fdp, int fd); 124static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif); 125static int fill_socket_info(struct socket *so, struct kinfo_file *kif); 126static int fill_pts_info(struct tty *tp, struct kinfo_file *kif); 127static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif); 128static int fill_procdesc_info(struct procdesc *pdp, 129 struct kinfo_file *kif); 130static int fill_shm_info(struct file *fp, struct kinfo_file *kif); 131 132/* 133 * A process is initially started out with NDFILE descriptors stored within 134 * this structure, selected to be enough for typical applications based on 135 * the historical limit of 20 open files (and the usage of descriptors by 136 * shells). If these descriptors are exhausted, a larger descriptor table 137 * may be allocated, up to a process' resource limit; the internal arrays 138 * are then unused. 139 */ 140#define NDFILE 20 141#define NDSLOTSIZE sizeof(NDSLOTTYPE) 142#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 143#define NDSLOT(x) ((x) / NDENTRIES) 144#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 145#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 146 147/* 148 * Storage required per open file descriptor. 149 */ 150#define OFILESIZE (sizeof(struct file *) + sizeof(char)) 151 152/* 153 * Storage to hold unused ofiles that need to be reclaimed. 154 */ 155struct freetable { 156 struct file **ft_table; 157 SLIST_ENTRY(freetable) ft_next; 158}; 159 160/* 161 * Basic allocation of descriptors: 162 * one of the above, plus arrays for NDFILE descriptors. 163 */ 164struct filedesc0 { 165 struct filedesc fd_fd; 166 /* 167 * ofiles which need to be reclaimed on free. 168 */ 169 SLIST_HEAD(,freetable) fd_free; 170 /* 171 * These arrays are used when the number of open files is 172 * <= NDFILE, and are then pointed to by the pointers above. 173 */ 174 struct file *fd_dfiles[NDFILE]; 175 char fd_dfileflags[NDFILE]; 176 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 177}; 178 179/* 180 * Descriptor management. 181 */ 182volatile int openfiles; /* actual number of open files */ 183struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 184void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 185 186/* A mutex to protect the association between a proc and filedesc. */ 187static struct mtx fdesc_mtx; 188 189/* 190 * Find the first zero bit in the given bitmap, starting at low and not 191 * exceeding size - 1. 192 */ 193static int 194fd_first_free(struct filedesc *fdp, int low, int size) 195{ 196 NDSLOTTYPE *map = fdp->fd_map; 197 NDSLOTTYPE mask; 198 int off, maxoff; 199 200 if (low >= size) 201 return (low); 202 203 off = NDSLOT(low); 204 if (low % NDENTRIES) { 205 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 206 if ((mask &= ~map[off]) != 0UL) 207 return (off * NDENTRIES + ffsl(mask) - 1); 208 ++off; 209 } 210 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 211 if (map[off] != ~0UL) 212 return (off * NDENTRIES + ffsl(~map[off]) - 1); 213 return (size); 214} 215 216/* 217 * Find the highest non-zero bit in the given bitmap, starting at low and 218 * not exceeding size - 1. 219 */ 220static int 221fd_last_used(struct filedesc *fdp, int low, int size) 222{ 223 NDSLOTTYPE *map = fdp->fd_map; 224 NDSLOTTYPE mask; 225 int off, minoff; 226 227 if (low >= size) 228 return (-1); 229 230 off = NDSLOT(size); 231 if (size % NDENTRIES) { 232 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 233 if ((mask &= map[off]) != 0) 234 return (off * NDENTRIES + flsl(mask) - 1); 235 --off; 236 } 237 for (minoff = NDSLOT(low); off >= minoff; --off) 238 if (map[off] != 0) 239 return (off * NDENTRIES + flsl(map[off]) - 1); 240 return (low - 1); 241} 242 243static int 244fdisused(struct filedesc *fdp, int fd) 245{ 246 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 247 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 248 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 249} 250 251/* 252 * Mark a file descriptor as used. 253 */ 254static void 255fdused(struct filedesc *fdp, int fd) 256{ 257 258 FILEDESC_XLOCK_ASSERT(fdp); 259 KASSERT(!fdisused(fdp, fd), 260 ("fd already used")); 261 262 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 263 if (fd > fdp->fd_lastfile) 264 fdp->fd_lastfile = fd; 265 if (fd == fdp->fd_freefile) 266 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 267} 268 269/* 270 * Mark a file descriptor as unused. 271 */ 272static void 273fdunused(struct filedesc *fdp, int fd) 274{ 275 276 FILEDESC_XLOCK_ASSERT(fdp); 277 KASSERT(fdisused(fdp, fd), 278 ("fd is already unused")); 279 KASSERT(fdp->fd_ofiles[fd] == NULL, 280 ("fd is still in use")); 281 282 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 283 if (fd < fdp->fd_freefile) 284 fdp->fd_freefile = fd; 285 if (fd == fdp->fd_lastfile) 286 fdp->fd_lastfile = fd_last_used(fdp, 0, fd); 287} 288 289/* 290 * System calls on descriptors. 291 */ 292#ifndef _SYS_SYSPROTO_H_ 293struct getdtablesize_args { 294 int dummy; 295}; 296#endif 297/* ARGSUSED */ 298int 299sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 300{ 301 struct proc *p = td->td_proc; 302 uint64_t lim; 303 304 PROC_LOCK(p); 305 td->td_retval[0] = 306 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 307 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 308 PROC_UNLOCK(p); 309 if (lim < td->td_retval[0]) 310 td->td_retval[0] = lim; 311 return (0); 312} 313 314/* 315 * Duplicate a file descriptor to a particular value. 316 * 317 * Note: keep in mind that a potential race condition exists when closing 318 * descriptors from a shared descriptor table (via rfork). 319 */ 320#ifndef _SYS_SYSPROTO_H_ 321struct dup2_args { 322 u_int from; 323 u_int to; 324}; 325#endif 326/* ARGSUSED */ 327int 328sys_dup2(struct thread *td, struct dup2_args *uap) 329{ 330 331 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 332 td->td_retval)); 333} 334 335/* 336 * Duplicate a file descriptor. 337 */ 338#ifndef _SYS_SYSPROTO_H_ 339struct dup_args { 340 u_int fd; 341}; 342#endif 343/* ARGSUSED */ 344int 345sys_dup(struct thread *td, struct dup_args *uap) 346{ 347 348 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 349} 350 351/* 352 * The file control system call. 353 */ 354#ifndef _SYS_SYSPROTO_H_ 355struct fcntl_args { 356 int fd; 357 int cmd; 358 long arg; 359}; 360#endif 361/* ARGSUSED */ 362int 363sys_fcntl(struct thread *td, struct fcntl_args *uap) 364{ 365 struct flock fl; 366 struct oflock ofl; 367 intptr_t arg; 368 int error; 369 int cmd; 370 371 error = 0; 372 cmd = uap->cmd; 373 switch (uap->cmd) { 374 case F_OGETLK: 375 case F_OSETLK: 376 case F_OSETLKW: 377 /* 378 * Convert old flock structure to new. 379 */ 380 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 381 fl.l_start = ofl.l_start; 382 fl.l_len = ofl.l_len; 383 fl.l_pid = ofl.l_pid; 384 fl.l_type = ofl.l_type; 385 fl.l_whence = ofl.l_whence; 386 fl.l_sysid = 0; 387 388 switch (uap->cmd) { 389 case F_OGETLK: 390 cmd = F_GETLK; 391 break; 392 case F_OSETLK: 393 cmd = F_SETLK; 394 break; 395 case F_OSETLKW: 396 cmd = F_SETLKW; 397 break; 398 } 399 arg = (intptr_t)&fl; 400 break; 401 case F_GETLK: 402 case F_SETLK: 403 case F_SETLKW: 404 case F_SETLK_REMOTE: 405 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 406 arg = (intptr_t)&fl; 407 break; 408 default: 409 arg = uap->arg; 410 break; 411 } 412 if (error) 413 return (error); 414 error = kern_fcntl(td, uap->fd, cmd, arg); 415 if (error) 416 return (error); 417 if (uap->cmd == F_OGETLK) { 418 ofl.l_start = fl.l_start; 419 ofl.l_len = fl.l_len; 420 ofl.l_pid = fl.l_pid; 421 ofl.l_type = fl.l_type; 422 ofl.l_whence = fl.l_whence; 423 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 424 } else if (uap->cmd == F_GETLK) { 425 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 426 } 427 return (error); 428} 429 430static inline struct file * 431fdtofp(int fd, struct filedesc *fdp) 432{ 433 struct file *fp; 434 435 FILEDESC_LOCK_ASSERT(fdp); 436 if ((unsigned)fd >= fdp->fd_nfiles || 437 (fp = fdp->fd_ofiles[fd]) == NULL) 438 return (NULL); 439 return (fp); 440} 441 442static inline int 443fdunwrap(int fd, cap_rights_t rights, struct filedesc *fdp, struct file **fpp) 444{ 445 446 *fpp = fdtofp(fd, fdp); 447 if (*fpp == NULL) 448 return (EBADF); 449 450#ifdef CAPABILITIES 451 if ((*fpp)->f_type == DTYPE_CAPABILITY) { 452 int err = cap_funwrap(*fpp, rights, fpp); 453 if (err != 0) { 454 *fpp = NULL; 455 return (err); 456 } 457 } 458#endif /* CAPABILITIES */ 459 return (0); 460} 461 462int 463kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 464{ 465 struct filedesc *fdp; 466 struct flock *flp; 467 struct file *fp; 468 struct proc *p; 469 char *pop; 470 struct vnode *vp; 471 int error, flg, tmp; 472 int vfslocked; 473 u_int old, new; 474 uint64_t bsize; 475 476 vfslocked = 0; 477 error = 0; 478 flg = F_POSIX; 479 p = td->td_proc; 480 fdp = p->p_fd; 481 482 switch (cmd) { 483 case F_DUPFD: 484 tmp = arg; 485 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 486 break; 487 488 case F_DUP2FD: 489 tmp = arg; 490 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 491 break; 492 493 case F_GETFD: 494 FILEDESC_SLOCK(fdp); 495 if ((fp = fdtofp(fd, fdp)) == NULL) { 496 FILEDESC_SUNLOCK(fdp); 497 error = EBADF; 498 break; 499 } 500 pop = &fdp->fd_ofileflags[fd]; 501 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0; 502 FILEDESC_SUNLOCK(fdp); 503 break; 504 505 case F_SETFD: 506 FILEDESC_XLOCK(fdp); 507 if ((fp = fdtofp(fd, fdp)) == NULL) { 508 FILEDESC_XUNLOCK(fdp); 509 error = EBADF; 510 break; 511 } 512 pop = &fdp->fd_ofileflags[fd]; 513 *pop = (*pop &~ UF_EXCLOSE) | 514 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 515 FILEDESC_XUNLOCK(fdp); 516 break; 517 518 case F_GETFL: 519 FILEDESC_SLOCK(fdp); 520 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 521 if (error != 0) { 522 FILEDESC_SUNLOCK(fdp); 523 break; 524 } 525 td->td_retval[0] = OFLAGS(fp->f_flag); 526 FILEDESC_SUNLOCK(fdp); 527 break; 528 529 case F_SETFL: 530 FILEDESC_SLOCK(fdp); 531 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 532 if (error != 0) { 533 FILEDESC_SUNLOCK(fdp); 534 break; 535 } 536 fhold(fp); 537 FILEDESC_SUNLOCK(fdp); 538 do { 539 tmp = flg = fp->f_flag; 540 tmp &= ~FCNTLFLAGS; 541 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 542 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 543 tmp = fp->f_flag & FNONBLOCK; 544 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 545 if (error) { 546 fdrop(fp, td); 547 break; 548 } 549 tmp = fp->f_flag & FASYNC; 550 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 551 if (error == 0) { 552 fdrop(fp, td); 553 break; 554 } 555 atomic_clear_int(&fp->f_flag, FNONBLOCK); 556 tmp = 0; 557 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 558 fdrop(fp, td); 559 break; 560 561 case F_GETOWN: 562 FILEDESC_SLOCK(fdp); 563 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 564 if (error != 0) { 565 FILEDESC_SUNLOCK(fdp); 566 break; 567 } 568 fhold(fp); 569 FILEDESC_SUNLOCK(fdp); 570 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 571 if (error == 0) 572 td->td_retval[0] = tmp; 573 fdrop(fp, td); 574 break; 575 576 case F_SETOWN: 577 FILEDESC_SLOCK(fdp); 578 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp); 579 if (error != 0) { 580 FILEDESC_SUNLOCK(fdp); 581 break; 582 } 583 fhold(fp); 584 FILEDESC_SUNLOCK(fdp); 585 tmp = arg; 586 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 587 fdrop(fp, td); 588 break; 589 590 case F_SETLK_REMOTE: 591 error = priv_check(td, PRIV_NFS_LOCKD); 592 if (error) 593 return (error); 594 flg = F_REMOTE; 595 goto do_setlk; 596 597 case F_SETLKW: 598 flg |= F_WAIT; 599 /* FALLTHROUGH F_SETLK */ 600 601 case F_SETLK: 602 do_setlk: 603 FILEDESC_SLOCK(fdp); 604 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp); 605 if (error != 0) { 606 FILEDESC_SUNLOCK(fdp); 607 break; 608 } 609 if (fp->f_type != DTYPE_VNODE) { 610 FILEDESC_SUNLOCK(fdp); 611 error = EBADF; 612 break; 613 } 614 flp = (struct flock *)arg; 615 if (flp->l_whence == SEEK_CUR) { 616 if (fp->f_offset < 0 || 617 (flp->l_start > 0 && 618 fp->f_offset > OFF_MAX - flp->l_start)) { 619 FILEDESC_SUNLOCK(fdp); 620 error = EOVERFLOW; 621 break; 622 } 623 flp->l_start += fp->f_offset; 624 } 625 626 /* 627 * VOP_ADVLOCK() may block. 628 */ 629 fhold(fp); 630 FILEDESC_SUNLOCK(fdp); 631 vp = fp->f_vnode; 632 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 633 switch (flp->l_type) { 634 case F_RDLCK: 635 if ((fp->f_flag & FREAD) == 0) { 636 error = EBADF; 637 break; 638 } 639 PROC_LOCK(p->p_leader); 640 p->p_leader->p_flag |= P_ADVLOCK; 641 PROC_UNLOCK(p->p_leader); 642 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 643 flp, flg); 644 break; 645 case F_WRLCK: 646 if ((fp->f_flag & FWRITE) == 0) { 647 error = EBADF; 648 break; 649 } 650 PROC_LOCK(p->p_leader); 651 p->p_leader->p_flag |= P_ADVLOCK; 652 PROC_UNLOCK(p->p_leader); 653 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 654 flp, flg); 655 break; 656 case F_UNLCK: 657 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 658 flp, flg); 659 break; 660 case F_UNLCKSYS: 661 /* 662 * Temporary api for testing remote lock 663 * infrastructure. 664 */ 665 if (flg != F_REMOTE) { 666 error = EINVAL; 667 break; 668 } 669 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 670 F_UNLCKSYS, flp, flg); 671 break; 672 default: 673 error = EINVAL; 674 break; 675 } 676 VFS_UNLOCK_GIANT(vfslocked); 677 vfslocked = 0; 678 /* Check for race with close */ 679 FILEDESC_SLOCK(fdp); 680 if ((unsigned) fd >= fdp->fd_nfiles || 681 fp != fdp->fd_ofiles[fd]) { 682 FILEDESC_SUNLOCK(fdp); 683 flp->l_whence = SEEK_SET; 684 flp->l_start = 0; 685 flp->l_len = 0; 686 flp->l_type = F_UNLCK; 687 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 688 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 689 F_UNLCK, flp, F_POSIX); 690 VFS_UNLOCK_GIANT(vfslocked); 691 vfslocked = 0; 692 } else 693 FILEDESC_SUNLOCK(fdp); 694 fdrop(fp, td); 695 break; 696 697 case F_GETLK: 698 FILEDESC_SLOCK(fdp); 699 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp); 700 if (error != 0) { 701 FILEDESC_SUNLOCK(fdp); 702 break; 703 } 704 if (fp->f_type != DTYPE_VNODE) { 705 FILEDESC_SUNLOCK(fdp); 706 error = EBADF; 707 break; 708 } 709 flp = (struct flock *)arg; 710 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 711 flp->l_type != F_UNLCK) { 712 FILEDESC_SUNLOCK(fdp); 713 error = EINVAL; 714 break; 715 } 716 if (flp->l_whence == SEEK_CUR) { 717 if ((flp->l_start > 0 && 718 fp->f_offset > OFF_MAX - flp->l_start) || 719 (flp->l_start < 0 && 720 fp->f_offset < OFF_MIN - flp->l_start)) { 721 FILEDESC_SUNLOCK(fdp); 722 error = EOVERFLOW; 723 break; 724 } 725 flp->l_start += fp->f_offset; 726 } 727 /* 728 * VOP_ADVLOCK() may block. 729 */ 730 fhold(fp); 731 FILEDESC_SUNLOCK(fdp); 732 vp = fp->f_vnode; 733 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 734 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 735 F_POSIX); 736 VFS_UNLOCK_GIANT(vfslocked); 737 vfslocked = 0; 738 fdrop(fp, td); 739 break; 740 741 case F_RDAHEAD: 742 arg = arg ? 128 * 1024: 0; 743 /* FALLTHROUGH */ 744 case F_READAHEAD: 745 FILEDESC_SLOCK(fdp); 746 if ((fp = fdtofp(fd, fdp)) == NULL) { 747 FILEDESC_SUNLOCK(fdp); 748 error = EBADF; 749 break; 750 } 751 if (fp->f_type != DTYPE_VNODE) { 752 FILEDESC_SUNLOCK(fdp); 753 error = EBADF; 754 break; 755 } 756 fhold(fp); 757 FILEDESC_SUNLOCK(fdp); 758 if (arg != 0) { 759 vp = fp->f_vnode; 760 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 761 error = vn_lock(vp, LK_SHARED); 762 if (error != 0) 763 goto readahead_vnlock_fail; 764 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 765 VOP_UNLOCK(vp, 0); 766 fp->f_seqcount = (arg + bsize - 1) / bsize; 767 do { 768 new = old = fp->f_flag; 769 new |= FRDAHEAD; 770 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 771readahead_vnlock_fail: 772 VFS_UNLOCK_GIANT(vfslocked); 773 vfslocked = 0; 774 } else { 775 do { 776 new = old = fp->f_flag; 777 new &= ~FRDAHEAD; 778 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 779 } 780 fdrop(fp, td); 781 break; 782 783 default: 784 error = EINVAL; 785 break; 786 } 787 VFS_UNLOCK_GIANT(vfslocked); 788 return (error); 789} 790 791/* 792 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 793 */ 794static int 795do_dup(struct thread *td, int flags, int old, int new, 796 register_t *retval) 797{ 798 struct filedesc *fdp; 799 struct proc *p; 800 struct file *fp; 801 struct file *delfp; 802 int error, holdleaders, maxfd; 803 804 p = td->td_proc; 805 fdp = p->p_fd; 806 807 /* 808 * Verify we have a valid descriptor to dup from and possibly to 809 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 810 * return EINVAL when the new descriptor is out of bounds. 811 */ 812 if (old < 0) 813 return (EBADF); 814 if (new < 0) 815 return (flags & DUP_FCNTL ? EINVAL : EBADF); 816 PROC_LOCK(p); 817 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 818 PROC_UNLOCK(p); 819 if (new >= maxfd) 820 return (flags & DUP_FCNTL ? EINVAL : EBADF); 821 822 FILEDESC_XLOCK(fdp); 823 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) { 824 FILEDESC_XUNLOCK(fdp); 825 return (EBADF); 826 } 827 if (flags & DUP_FIXED && old == new) { 828 *retval = new; 829 FILEDESC_XUNLOCK(fdp); 830 return (0); 831 } 832 fp = fdp->fd_ofiles[old]; 833 fhold(fp); 834 835 /* 836 * If the caller specified a file descriptor, make sure the file 837 * table is large enough to hold it, and grab it. Otherwise, just 838 * allocate a new descriptor the usual way. Since the filedesc 839 * lock may be temporarily dropped in the process, we have to look 840 * out for a race. 841 */ 842 if (flags & DUP_FIXED) { 843 if (new >= fdp->fd_nfiles) { 844 /* 845 * The resource limits are here instead of e.g. 846 * fdalloc(), because the file descriptor table may be 847 * shared between processes, so we can't really use 848 * racct_add()/racct_sub(). Instead of counting the 849 * number of actually allocated descriptors, just put 850 * the limit on the size of the file descriptor table. 851 */ 852#ifdef RACCT 853 PROC_LOCK(p); 854 error = racct_set(p, RACCT_NOFILE, new + 1); 855 PROC_UNLOCK(p); 856 if (error != 0) { 857 FILEDESC_XUNLOCK(fdp); 858 fdrop(fp, td); 859 return (EMFILE); 860 } 861#endif 862 fdgrowtable(fdp, new + 1); 863 } 864 if (fdp->fd_ofiles[new] == NULL) 865 fdused(fdp, new); 866 } else { 867 if ((error = fdalloc(td, new, &new)) != 0) { 868 FILEDESC_XUNLOCK(fdp); 869 fdrop(fp, td); 870 return (error); 871 } 872 } 873 874 /* 875 * If the old file changed out from under us then treat it as a 876 * bad file descriptor. Userland should do its own locking to 877 * avoid this case. 878 */ 879 if (fdp->fd_ofiles[old] != fp) { 880 /* we've allocated a descriptor which we won't use */ 881 if (fdp->fd_ofiles[new] == NULL) 882 fdunused(fdp, new); 883 FILEDESC_XUNLOCK(fdp); 884 fdrop(fp, td); 885 return (EBADF); 886 } 887 KASSERT(old != new, 888 ("new fd is same as old")); 889 890 /* 891 * Save info on the descriptor being overwritten. We cannot close 892 * it without introducing an ownership race for the slot, since we 893 * need to drop the filedesc lock to call closef(). 894 * 895 * XXX this duplicates parts of close(). 896 */ 897 delfp = fdp->fd_ofiles[new]; 898 holdleaders = 0; 899 if (delfp != NULL) { 900 if (td->td_proc->p_fdtol != NULL) { 901 /* 902 * Ask fdfree() to sleep to ensure that all relevant 903 * process leaders can be traversed in closef(). 904 */ 905 fdp->fd_holdleaderscount++; 906 holdleaders = 1; 907 } 908 } 909 910 /* 911 * Duplicate the source descriptor 912 */ 913 fdp->fd_ofiles[new] = fp; 914 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 915 if (new > fdp->fd_lastfile) 916 fdp->fd_lastfile = new; 917 *retval = new; 918 919 /* 920 * If we dup'd over a valid file, we now own the reference to it 921 * and must dispose of it using closef() semantics (as if a 922 * close() were performed on it). 923 * 924 * XXX this duplicates parts of close(). 925 */ 926 if (delfp != NULL) { 927 knote_fdclose(td, new); 928 if (delfp->f_type == DTYPE_MQUEUE) 929 mq_fdclose(td, new, delfp); 930 FILEDESC_XUNLOCK(fdp); 931 (void) closef(delfp, td); 932 if (holdleaders) { 933 FILEDESC_XLOCK(fdp); 934 fdp->fd_holdleaderscount--; 935 if (fdp->fd_holdleaderscount == 0 && 936 fdp->fd_holdleaderswakeup != 0) { 937 fdp->fd_holdleaderswakeup = 0; 938 wakeup(&fdp->fd_holdleaderscount); 939 } 940 FILEDESC_XUNLOCK(fdp); 941 } 942 } else { 943 FILEDESC_XUNLOCK(fdp); 944 } 945 return (0); 946} 947 948/* 949 * If sigio is on the list associated with a process or process group, 950 * disable signalling from the device, remove sigio from the list and 951 * free sigio. 952 */ 953void 954funsetown(struct sigio **sigiop) 955{ 956 struct sigio *sigio; 957 958 SIGIO_LOCK(); 959 sigio = *sigiop; 960 if (sigio == NULL) { 961 SIGIO_UNLOCK(); 962 return; 963 } 964 *(sigio->sio_myref) = NULL; 965 if ((sigio)->sio_pgid < 0) { 966 struct pgrp *pg = (sigio)->sio_pgrp; 967 PGRP_LOCK(pg); 968 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 969 sigio, sio_pgsigio); 970 PGRP_UNLOCK(pg); 971 } else { 972 struct proc *p = (sigio)->sio_proc; 973 PROC_LOCK(p); 974 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 975 sigio, sio_pgsigio); 976 PROC_UNLOCK(p); 977 } 978 SIGIO_UNLOCK(); 979 crfree(sigio->sio_ucred); 980 free(sigio, M_SIGIO); 981} 982 983/* 984 * Free a list of sigio structures. 985 * We only need to lock the SIGIO_LOCK because we have made ourselves 986 * inaccessible to callers of fsetown and therefore do not need to lock 987 * the proc or pgrp struct for the list manipulation. 988 */ 989void 990funsetownlst(struct sigiolst *sigiolst) 991{ 992 struct proc *p; 993 struct pgrp *pg; 994 struct sigio *sigio; 995 996 sigio = SLIST_FIRST(sigiolst); 997 if (sigio == NULL) 998 return; 999 p = NULL; 1000 pg = NULL; 1001 1002 /* 1003 * Every entry of the list should belong 1004 * to a single proc or pgrp. 1005 */ 1006 if (sigio->sio_pgid < 0) { 1007 pg = sigio->sio_pgrp; 1008 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 1009 } else /* if (sigio->sio_pgid > 0) */ { 1010 p = sigio->sio_proc; 1011 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 1012 } 1013 1014 SIGIO_LOCK(); 1015 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 1016 *(sigio->sio_myref) = NULL; 1017 if (pg != NULL) { 1018 KASSERT(sigio->sio_pgid < 0, 1019 ("Proc sigio in pgrp sigio list")); 1020 KASSERT(sigio->sio_pgrp == pg, 1021 ("Bogus pgrp in sigio list")); 1022 PGRP_LOCK(pg); 1023 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 1024 sio_pgsigio); 1025 PGRP_UNLOCK(pg); 1026 } else /* if (p != NULL) */ { 1027 KASSERT(sigio->sio_pgid > 0, 1028 ("Pgrp sigio in proc sigio list")); 1029 KASSERT(sigio->sio_proc == p, 1030 ("Bogus proc in sigio list")); 1031 PROC_LOCK(p); 1032 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1033 sio_pgsigio); 1034 PROC_UNLOCK(p); 1035 } 1036 SIGIO_UNLOCK(); 1037 crfree(sigio->sio_ucred); 1038 free(sigio, M_SIGIO); 1039 SIGIO_LOCK(); 1040 } 1041 SIGIO_UNLOCK(); 1042} 1043 1044/* 1045 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1046 * 1047 * After permission checking, add a sigio structure to the sigio list for 1048 * the process or process group. 1049 */ 1050int 1051fsetown(pid_t pgid, struct sigio **sigiop) 1052{ 1053 struct proc *proc; 1054 struct pgrp *pgrp; 1055 struct sigio *sigio; 1056 int ret; 1057 1058 if (pgid == 0) { 1059 funsetown(sigiop); 1060 return (0); 1061 } 1062 1063 ret = 0; 1064 1065 /* Allocate and fill in the new sigio out of locks. */ 1066 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1067 sigio->sio_pgid = pgid; 1068 sigio->sio_ucred = crhold(curthread->td_ucred); 1069 sigio->sio_myref = sigiop; 1070 1071 sx_slock(&proctree_lock); 1072 if (pgid > 0) { 1073 proc = pfind(pgid); 1074 if (proc == NULL) { 1075 ret = ESRCH; 1076 goto fail; 1077 } 1078 1079 /* 1080 * Policy - Don't allow a process to FSETOWN a process 1081 * in another session. 1082 * 1083 * Remove this test to allow maximum flexibility or 1084 * restrict FSETOWN to the current process or process 1085 * group for maximum safety. 1086 */ 1087 PROC_UNLOCK(proc); 1088 if (proc->p_session != curthread->td_proc->p_session) { 1089 ret = EPERM; 1090 goto fail; 1091 } 1092 1093 pgrp = NULL; 1094 } else /* if (pgid < 0) */ { 1095 pgrp = pgfind(-pgid); 1096 if (pgrp == NULL) { 1097 ret = ESRCH; 1098 goto fail; 1099 } 1100 PGRP_UNLOCK(pgrp); 1101 1102 /* 1103 * Policy - Don't allow a process to FSETOWN a process 1104 * in another session. 1105 * 1106 * Remove this test to allow maximum flexibility or 1107 * restrict FSETOWN to the current process or process 1108 * group for maximum safety. 1109 */ 1110 if (pgrp->pg_session != curthread->td_proc->p_session) { 1111 ret = EPERM; 1112 goto fail; 1113 } 1114 1115 proc = NULL; 1116 } 1117 funsetown(sigiop); 1118 if (pgid > 0) { 1119 PROC_LOCK(proc); 1120 /* 1121 * Since funsetownlst() is called without the proctree 1122 * locked, we need to check for P_WEXIT. 1123 * XXX: is ESRCH correct? 1124 */ 1125 if ((proc->p_flag & P_WEXIT) != 0) { 1126 PROC_UNLOCK(proc); 1127 ret = ESRCH; 1128 goto fail; 1129 } 1130 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1131 sigio->sio_proc = proc; 1132 PROC_UNLOCK(proc); 1133 } else { 1134 PGRP_LOCK(pgrp); 1135 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1136 sigio->sio_pgrp = pgrp; 1137 PGRP_UNLOCK(pgrp); 1138 } 1139 sx_sunlock(&proctree_lock); 1140 SIGIO_LOCK(); 1141 *sigiop = sigio; 1142 SIGIO_UNLOCK(); 1143 return (0); 1144 1145fail: 1146 sx_sunlock(&proctree_lock); 1147 crfree(sigio->sio_ucred); 1148 free(sigio, M_SIGIO); 1149 return (ret); 1150} 1151 1152/* 1153 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1154 */ 1155pid_t 1156fgetown(sigiop) 1157 struct sigio **sigiop; 1158{ 1159 pid_t pgid; 1160 1161 SIGIO_LOCK(); 1162 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1163 SIGIO_UNLOCK(); 1164 return (pgid); 1165} 1166 1167/* 1168 * Close a file descriptor. 1169 */ 1170#ifndef _SYS_SYSPROTO_H_ 1171struct close_args { 1172 int fd; 1173}; 1174#endif 1175/* ARGSUSED */ 1176int 1177sys_close(td, uap) 1178 struct thread *td; 1179 struct close_args *uap; 1180{ 1181 1182 return (kern_close(td, uap->fd)); 1183} 1184 1185int 1186kern_close(td, fd) 1187 struct thread *td; 1188 int fd; 1189{ 1190 struct filedesc *fdp; 1191 struct file *fp, *fp_object; 1192 int error; 1193 int holdleaders; 1194 1195 error = 0; 1196 holdleaders = 0; 1197 fdp = td->td_proc->p_fd; 1198 1199 AUDIT_SYSCLOSE(td, fd); 1200 1201 FILEDESC_XLOCK(fdp); 1202 if ((unsigned)fd >= fdp->fd_nfiles || 1203 (fp = fdp->fd_ofiles[fd]) == NULL) { 1204 FILEDESC_XUNLOCK(fdp); 1205 return (EBADF); 1206 } 1207 fdp->fd_ofiles[fd] = NULL; 1208 fdp->fd_ofileflags[fd] = 0; 1209 fdunused(fdp, fd); 1210 if (td->td_proc->p_fdtol != NULL) { 1211 /* 1212 * Ask fdfree() to sleep to ensure that all relevant 1213 * process leaders can be traversed in closef(). 1214 */ 1215 fdp->fd_holdleaderscount++; 1216 holdleaders = 1; 1217 } 1218 1219 /* 1220 * We now hold the fp reference that used to be owned by the 1221 * descriptor array. We have to unlock the FILEDESC *AFTER* 1222 * knote_fdclose to prevent a race of the fd getting opened, a knote 1223 * added, and deleteing a knote for the new fd. 1224 */ 1225 knote_fdclose(td, fd); 1226 1227 /* 1228 * When we're closing an fd with a capability, we need to notify 1229 * mqueue if the underlying object is of type mqueue. 1230 */ 1231 (void)cap_funwrap(fp, 0, &fp_object); 1232 if (fp_object->f_type == DTYPE_MQUEUE) 1233 mq_fdclose(td, fd, fp_object); 1234 FILEDESC_XUNLOCK(fdp); 1235 1236 error = closef(fp, td); 1237 if (holdleaders) { 1238 FILEDESC_XLOCK(fdp); 1239 fdp->fd_holdleaderscount--; 1240 if (fdp->fd_holdleaderscount == 0 && 1241 fdp->fd_holdleaderswakeup != 0) { 1242 fdp->fd_holdleaderswakeup = 0; 1243 wakeup(&fdp->fd_holdleaderscount); 1244 } 1245 FILEDESC_XUNLOCK(fdp); 1246 } 1247 return (error); 1248} 1249 1250/* 1251 * Close open file descriptors. 1252 */ 1253#ifndef _SYS_SYSPROTO_H_ 1254struct closefrom_args { 1255 int lowfd; 1256}; 1257#endif 1258/* ARGSUSED */ 1259int 1260sys_closefrom(struct thread *td, struct closefrom_args *uap) 1261{ 1262 struct filedesc *fdp; 1263 int fd; 1264 1265 fdp = td->td_proc->p_fd; 1266 AUDIT_ARG_FD(uap->lowfd); 1267 1268 /* 1269 * Treat negative starting file descriptor values identical to 1270 * closefrom(0) which closes all files. 1271 */ 1272 if (uap->lowfd < 0) 1273 uap->lowfd = 0; 1274 FILEDESC_SLOCK(fdp); 1275 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) { 1276 if (fdp->fd_ofiles[fd] != NULL) { 1277 FILEDESC_SUNLOCK(fdp); 1278 (void)kern_close(td, fd); 1279 FILEDESC_SLOCK(fdp); 1280 } 1281 } 1282 FILEDESC_SUNLOCK(fdp); 1283 return (0); 1284} 1285 1286#if defined(COMPAT_43) 1287/* 1288 * Return status information about a file descriptor. 1289 */ 1290#ifndef _SYS_SYSPROTO_H_ 1291struct ofstat_args { 1292 int fd; 1293 struct ostat *sb; 1294}; 1295#endif 1296/* ARGSUSED */ 1297int 1298ofstat(struct thread *td, struct ofstat_args *uap) 1299{ 1300 struct ostat oub; 1301 struct stat ub; 1302 int error; 1303 1304 error = kern_fstat(td, uap->fd, &ub); 1305 if (error == 0) { 1306 cvtstat(&ub, &oub); 1307 error = copyout(&oub, uap->sb, sizeof(oub)); 1308 } 1309 return (error); 1310} 1311#endif /* COMPAT_43 */ 1312 1313/* 1314 * Return status information about a file descriptor. 1315 */ 1316#ifndef _SYS_SYSPROTO_H_ 1317struct fstat_args { 1318 int fd; 1319 struct stat *sb; 1320}; 1321#endif 1322/* ARGSUSED */ 1323int 1324sys_fstat(struct thread *td, struct fstat_args *uap) 1325{ 1326 struct stat ub; 1327 int error; 1328 1329 error = kern_fstat(td, uap->fd, &ub); 1330 if (error == 0) 1331 error = copyout(&ub, uap->sb, sizeof(ub)); 1332 return (error); 1333} 1334 1335int 1336kern_fstat(struct thread *td, int fd, struct stat *sbp) 1337{ 1338 struct file *fp; 1339 int error; 1340 1341 AUDIT_ARG_FD(fd); 1342 1343 if ((error = fget(td, fd, CAP_FSTAT, &fp)) != 0) 1344 return (error); 1345 1346 AUDIT_ARG_FILE(td->td_proc, fp); 1347 1348 error = fo_stat(fp, sbp, td->td_ucred, td); 1349 fdrop(fp, td); 1350#ifdef KTRACE 1351 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1352 ktrstat(sbp); 1353#endif 1354 return (error); 1355} 1356 1357/* 1358 * Return status information about a file descriptor. 1359 */ 1360#ifndef _SYS_SYSPROTO_H_ 1361struct nfstat_args { 1362 int fd; 1363 struct nstat *sb; 1364}; 1365#endif 1366/* ARGSUSED */ 1367int 1368sys_nfstat(struct thread *td, struct nfstat_args *uap) 1369{ 1370 struct nstat nub; 1371 struct stat ub; 1372 int error; 1373 1374 error = kern_fstat(td, uap->fd, &ub); 1375 if (error == 0) { 1376 cvtnstat(&ub, &nub); 1377 error = copyout(&nub, uap->sb, sizeof(nub)); 1378 } 1379 return (error); 1380} 1381 1382/* 1383 * Return pathconf information about a file descriptor. 1384 */ 1385#ifndef _SYS_SYSPROTO_H_ 1386struct fpathconf_args { 1387 int fd; 1388 int name; 1389}; 1390#endif 1391/* ARGSUSED */ 1392int 1393sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1394{ 1395 struct file *fp; 1396 struct vnode *vp; 1397 int error; 1398 1399 if ((error = fget(td, uap->fd, CAP_FPATHCONF, &fp)) != 0) 1400 return (error); 1401 1402 /* If asynchronous I/O is available, it works for all descriptors. */ 1403 if (uap->name == _PC_ASYNC_IO) { 1404 td->td_retval[0] = async_io_version; 1405 goto out; 1406 } 1407 vp = fp->f_vnode; 1408 if (vp != NULL) { 1409 int vfslocked; 1410 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1411 vn_lock(vp, LK_SHARED | LK_RETRY); 1412 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1413 VOP_UNLOCK(vp, 0); 1414 VFS_UNLOCK_GIANT(vfslocked); 1415 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1416 if (uap->name != _PC_PIPE_BUF) { 1417 error = EINVAL; 1418 } else { 1419 td->td_retval[0] = PIPE_BUF; 1420 error = 0; 1421 } 1422 } else { 1423 error = EOPNOTSUPP; 1424 } 1425out: 1426 fdrop(fp, td); 1427 return (error); 1428} 1429 1430/* 1431 * Grow the file table to accomodate (at least) nfd descriptors. This may 1432 * block and drop the filedesc lock, but it will reacquire it before 1433 * returning. 1434 */ 1435static void 1436fdgrowtable(struct filedesc *fdp, int nfd) 1437{ 1438 struct filedesc0 *fdp0; 1439 struct freetable *fo; 1440 struct file **ntable; 1441 struct file **otable; 1442 char *nfileflags; 1443 int nnfiles, onfiles; 1444 NDSLOTTYPE *nmap; 1445 1446 FILEDESC_XLOCK_ASSERT(fdp); 1447 1448 KASSERT(fdp->fd_nfiles > 0, 1449 ("zero-length file table")); 1450 1451 /* compute the size of the new table */ 1452 onfiles = fdp->fd_nfiles; 1453 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1454 if (nnfiles <= onfiles) 1455 /* the table is already large enough */ 1456 return; 1457 1458 /* allocate a new table and (if required) new bitmaps */ 1459 FILEDESC_XUNLOCK(fdp); 1460 ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable), 1461 M_FILEDESC, M_ZERO | M_WAITOK); 1462 nfileflags = (char *)&ntable[nnfiles]; 1463 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) 1464 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, 1465 M_FILEDESC, M_ZERO | M_WAITOK); 1466 else 1467 nmap = NULL; 1468 FILEDESC_XLOCK(fdp); 1469 1470 /* 1471 * We now have new tables ready to go. Since we dropped the 1472 * filedesc lock to call malloc(), watch out for a race. 1473 */ 1474 onfiles = fdp->fd_nfiles; 1475 if (onfiles >= nnfiles) { 1476 /* we lost the race, but that's OK */ 1477 free(ntable, M_FILEDESC); 1478 if (nmap != NULL) 1479 free(nmap, M_FILEDESC); 1480 return; 1481 } 1482 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable)); 1483 bcopy(fdp->fd_ofileflags, nfileflags, onfiles); 1484 otable = fdp->fd_ofiles; 1485 fdp->fd_ofileflags = nfileflags; 1486 fdp->fd_ofiles = ntable; 1487 /* 1488 * We must preserve ofiles until the process exits because we can't 1489 * be certain that no threads have references to the old table via 1490 * _fget(). 1491 */ 1492 if (onfiles > NDFILE) { 1493 fo = (struct freetable *)&otable[onfiles]; 1494 fdp0 = (struct filedesc0 *)fdp; 1495 fo->ft_table = otable; 1496 SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next); 1497 } 1498 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1499 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap)); 1500 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1501 free(fdp->fd_map, M_FILEDESC); 1502 fdp->fd_map = nmap; 1503 } 1504 fdp->fd_nfiles = nnfiles; 1505} 1506 1507/* 1508 * Allocate a file descriptor for the process. 1509 */ 1510int 1511fdalloc(struct thread *td, int minfd, int *result) 1512{ 1513 struct proc *p = td->td_proc; 1514 struct filedesc *fdp = p->p_fd; 1515 int fd = -1, maxfd; 1516#ifdef RACCT 1517 int error; 1518#endif 1519 1520 FILEDESC_XLOCK_ASSERT(fdp); 1521 1522 if (fdp->fd_freefile > minfd) 1523 minfd = fdp->fd_freefile; 1524 1525 PROC_LOCK(p); 1526 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1527 PROC_UNLOCK(p); 1528 1529 /* 1530 * Search the bitmap for a free descriptor. If none is found, try 1531 * to grow the file table. Keep at it until we either get a file 1532 * descriptor or run into process or system limits; fdgrowtable() 1533 * may drop the filedesc lock, so we're in a race. 1534 */ 1535 for (;;) { 1536 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1537 if (fd >= maxfd) 1538 return (EMFILE); 1539 if (fd < fdp->fd_nfiles) 1540 break; 1541#ifdef RACCT 1542 PROC_LOCK(p); 1543 error = racct_set(p, RACCT_NOFILE, min(fdp->fd_nfiles * 2, maxfd)); 1544 PROC_UNLOCK(p); 1545 if (error != 0) 1546 return (EMFILE); 1547#endif 1548 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd)); 1549 } 1550 1551 /* 1552 * Perform some sanity checks, then mark the file descriptor as 1553 * used and return it to the caller. 1554 */ 1555 KASSERT(!fdisused(fdp, fd), 1556 ("fd_first_free() returned non-free descriptor")); 1557 KASSERT(fdp->fd_ofiles[fd] == NULL, ("file descriptor isn't free")); 1558 KASSERT(fdp->fd_ofileflags[fd] == 0, ("file flags are set")); 1559 fdused(fdp, fd); 1560 *result = fd; 1561 return (0); 1562} 1563 1564/* 1565 * Check to see whether n user file descriptors are available to the process 1566 * p. 1567 */ 1568int 1569fdavail(struct thread *td, int n) 1570{ 1571 struct proc *p = td->td_proc; 1572 struct filedesc *fdp = td->td_proc->p_fd; 1573 int i, lim, last; 1574 1575 FILEDESC_LOCK_ASSERT(fdp); 1576 1577 /* 1578 * XXX: This is only called from uipc_usrreq.c:unp_externalize(); 1579 * call racct_add() from there instead of dealing with containers 1580 * here. 1581 */ 1582 PROC_LOCK(p); 1583 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 1584 PROC_UNLOCK(p); 1585 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1586 return (1); 1587 last = min(fdp->fd_nfiles, lim); 1588 for (i = fdp->fd_freefile; i < last; i++) { 1589 if (fdp->fd_ofiles[i] == NULL && --n <= 0) 1590 return (1); 1591 } 1592 return (0); 1593} 1594 1595/* 1596 * Create a new open file structure and allocate a file decriptor for the 1597 * process that refers to it. We add one reference to the file for the 1598 * descriptor table and one reference for resultfp. This is to prevent us 1599 * being preempted and the entry in the descriptor table closed after we 1600 * release the FILEDESC lock. 1601 */ 1602int 1603falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1604{ 1605 struct file *fp; 1606 int error, fd; 1607 1608 error = falloc_noinstall(td, &fp); 1609 if (error) 1610 return (error); /* no reference held on error */ 1611 1612 error = finstall(td, fp, &fd, flags); 1613 if (error) { 1614 fdrop(fp, td); /* one reference (fp only) */ 1615 return (error); 1616 } 1617 1618 if (resultfp != NULL) 1619 *resultfp = fp; /* copy out result */ 1620 else 1621 fdrop(fp, td); /* release local reference */ 1622 1623 if (resultfd != NULL) 1624 *resultfd = fd; 1625 1626 return (0); 1627} 1628 1629/* 1630 * Create a new open file structure without allocating a file descriptor. 1631 */ 1632int 1633falloc_noinstall(struct thread *td, struct file **resultfp) 1634{ 1635 struct file *fp; 1636 int maxuserfiles = maxfiles - (maxfiles / 20); 1637 static struct timeval lastfail; 1638 static int curfail; 1639 1640 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1641 1642 if ((openfiles >= maxuserfiles && 1643 priv_check(td, PRIV_MAXFILES) != 0) || 1644 openfiles >= maxfiles) { 1645 if (ppsratecheck(&lastfail, &curfail, 1)) { 1646 printf("kern.maxfiles limit exceeded by uid %i, " 1647 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1648 } 1649 return (ENFILE); 1650 } 1651 atomic_add_int(&openfiles, 1); 1652 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1653 refcount_init(&fp->f_count, 1); 1654 fp->f_cred = crhold(td->td_ucred); 1655 fp->f_ops = &badfileops; 1656 fp->f_data = NULL; 1657 fp->f_vnode = NULL; 1658 *resultfp = fp; 1659 return (0); 1660} 1661 1662/* 1663 * Install a file in a file descriptor table. 1664 */ 1665int 1666finstall(struct thread *td, struct file *fp, int *fd, int flags) 1667{ 1668 struct filedesc *fdp = td->td_proc->p_fd; 1669 int error; 1670 1671 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1672 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1673 1674 FILEDESC_XLOCK(fdp); 1675 if ((error = fdalloc(td, 0, fd))) { 1676 FILEDESC_XUNLOCK(fdp); 1677 return (error); 1678 } 1679 fhold(fp); 1680 fdp->fd_ofiles[*fd] = fp; 1681 if ((flags & O_CLOEXEC) != 0) 1682 fdp->fd_ofileflags[*fd] |= UF_EXCLOSE; 1683 FILEDESC_XUNLOCK(fdp); 1684 return (0); 1685} 1686 1687/* 1688 * Build a new filedesc structure from another. 1689 * Copy the current, root, and jail root vnode references. 1690 */ 1691struct filedesc * 1692fdinit(struct filedesc *fdp) 1693{ 1694 struct filedesc0 *newfdp; 1695 1696 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1697 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1698 if (fdp != NULL) { 1699 FILEDESC_XLOCK(fdp); 1700 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1701 if (newfdp->fd_fd.fd_cdir) 1702 VREF(newfdp->fd_fd.fd_cdir); 1703 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1704 if (newfdp->fd_fd.fd_rdir) 1705 VREF(newfdp->fd_fd.fd_rdir); 1706 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1707 if (newfdp->fd_fd.fd_jdir) 1708 VREF(newfdp->fd_fd.fd_jdir); 1709 FILEDESC_XUNLOCK(fdp); 1710 } 1711 1712 /* Create the file descriptor table. */ 1713 newfdp->fd_fd.fd_refcnt = 1; 1714 newfdp->fd_fd.fd_holdcnt = 1; 1715 newfdp->fd_fd.fd_cmask = CMASK; 1716 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1717 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags; 1718 newfdp->fd_fd.fd_nfiles = NDFILE; 1719 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1720 newfdp->fd_fd.fd_lastfile = -1; 1721 return (&newfdp->fd_fd); 1722} 1723 1724static struct filedesc * 1725fdhold(struct proc *p) 1726{ 1727 struct filedesc *fdp; 1728 1729 mtx_lock(&fdesc_mtx); 1730 fdp = p->p_fd; 1731 if (fdp != NULL) 1732 fdp->fd_holdcnt++; 1733 mtx_unlock(&fdesc_mtx); 1734 return (fdp); 1735} 1736 1737static void 1738fddrop(struct filedesc *fdp) 1739{ 1740 struct filedesc0 *fdp0; 1741 struct freetable *ft; 1742 int i; 1743 1744 mtx_lock(&fdesc_mtx); 1745 i = --fdp->fd_holdcnt; 1746 mtx_unlock(&fdesc_mtx); 1747 if (i > 0) 1748 return; 1749 1750 FILEDESC_LOCK_DESTROY(fdp); 1751 fdp0 = (struct filedesc0 *)fdp; 1752 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1753 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1754 free(ft->ft_table, M_FILEDESC); 1755 } 1756 free(fdp, M_FILEDESC); 1757} 1758 1759/* 1760 * Share a filedesc structure. 1761 */ 1762struct filedesc * 1763fdshare(struct filedesc *fdp) 1764{ 1765 1766 FILEDESC_XLOCK(fdp); 1767 fdp->fd_refcnt++; 1768 FILEDESC_XUNLOCK(fdp); 1769 return (fdp); 1770} 1771 1772/* 1773 * Unshare a filedesc structure, if necessary by making a copy 1774 */ 1775void 1776fdunshare(struct proc *p, struct thread *td) 1777{ 1778 1779 FILEDESC_XLOCK(p->p_fd); 1780 if (p->p_fd->fd_refcnt > 1) { 1781 struct filedesc *tmp; 1782 1783 FILEDESC_XUNLOCK(p->p_fd); 1784 tmp = fdcopy(p->p_fd); 1785 fdfree(td); 1786 p->p_fd = tmp; 1787 } else 1788 FILEDESC_XUNLOCK(p->p_fd); 1789} 1790 1791/* 1792 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1793 * this is to ease callers, not catch errors. 1794 */ 1795struct filedesc * 1796fdcopy(struct filedesc *fdp) 1797{ 1798 struct filedesc *newfdp; 1799 int i; 1800 1801 /* Certain daemons might not have file descriptors. */ 1802 if (fdp == NULL) 1803 return (NULL); 1804 1805 newfdp = fdinit(fdp); 1806 FILEDESC_SLOCK(fdp); 1807 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1808 FILEDESC_SUNLOCK(fdp); 1809 FILEDESC_XLOCK(newfdp); 1810 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1811 FILEDESC_XUNLOCK(newfdp); 1812 FILEDESC_SLOCK(fdp); 1813 } 1814 /* copy all passable descriptors (i.e. not kqueue) */ 1815 newfdp->fd_freefile = -1; 1816 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1817 if (fdisused(fdp, i) && 1818 (fdp->fd_ofiles[i]->f_ops->fo_flags & DFLAG_PASSABLE) && 1819 fdp->fd_ofiles[i]->f_ops != &badfileops) { 1820 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i]; 1821 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i]; 1822 fhold(newfdp->fd_ofiles[i]); 1823 newfdp->fd_lastfile = i; 1824 } else { 1825 if (newfdp->fd_freefile == -1) 1826 newfdp->fd_freefile = i; 1827 } 1828 } 1829 newfdp->fd_cmask = fdp->fd_cmask; 1830 FILEDESC_SUNLOCK(fdp); 1831 FILEDESC_XLOCK(newfdp); 1832 for (i = 0; i <= newfdp->fd_lastfile; ++i) 1833 if (newfdp->fd_ofiles[i] != NULL) 1834 fdused(newfdp, i); 1835 if (newfdp->fd_freefile == -1) 1836 newfdp->fd_freefile = i; 1837 FILEDESC_XUNLOCK(newfdp); 1838 return (newfdp); 1839} 1840 1841/* 1842 * Release a filedesc structure. 1843 */ 1844void 1845fdfree(struct thread *td) 1846{ 1847 struct filedesc *fdp; 1848 int i, locked; 1849 struct filedesc_to_leader *fdtol; 1850 struct file *fp; 1851 struct vnode *cdir, *jdir, *rdir, *vp; 1852 struct flock lf; 1853 1854 /* Certain daemons might not have file descriptors. */ 1855 fdp = td->td_proc->p_fd; 1856 if (fdp == NULL) 1857 return; 1858 1859#ifdef RACCT 1860 PROC_LOCK(td->td_proc); 1861 racct_set(td->td_proc, RACCT_NOFILE, 0); 1862 PROC_UNLOCK(td->td_proc); 1863#endif 1864 1865 /* Check for special need to clear POSIX style locks */ 1866 fdtol = td->td_proc->p_fdtol; 1867 if (fdtol != NULL) { 1868 FILEDESC_XLOCK(fdp); 1869 KASSERT(fdtol->fdl_refcount > 0, 1870 ("filedesc_to_refcount botch: fdl_refcount=%d", 1871 fdtol->fdl_refcount)); 1872 if (fdtol->fdl_refcount == 1 && 1873 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1874 for (i = 0; i <= fdp->fd_lastfile; i++) { 1875 fp = fdp->fd_ofiles[i]; 1876 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1877 continue; 1878 fhold(fp); 1879 FILEDESC_XUNLOCK(fdp); 1880 lf.l_whence = SEEK_SET; 1881 lf.l_start = 0; 1882 lf.l_len = 0; 1883 lf.l_type = F_UNLCK; 1884 vp = fp->f_vnode; 1885 locked = VFS_LOCK_GIANT(vp->v_mount); 1886 (void) VOP_ADVLOCK(vp, 1887 (caddr_t)td->td_proc-> 1888 p_leader, 1889 F_UNLCK, 1890 &lf, 1891 F_POSIX); 1892 VFS_UNLOCK_GIANT(locked); 1893 FILEDESC_XLOCK(fdp); 1894 fdrop(fp, td); 1895 } 1896 } 1897 retry: 1898 if (fdtol->fdl_refcount == 1) { 1899 if (fdp->fd_holdleaderscount > 0 && 1900 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1901 /* 1902 * close() or do_dup() has cleared a reference 1903 * in a shared file descriptor table. 1904 */ 1905 fdp->fd_holdleaderswakeup = 1; 1906 sx_sleep(&fdp->fd_holdleaderscount, 1907 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1908 goto retry; 1909 } 1910 if (fdtol->fdl_holdcount > 0) { 1911 /* 1912 * Ensure that fdtol->fdl_leader remains 1913 * valid in closef(). 1914 */ 1915 fdtol->fdl_wakeup = 1; 1916 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 1917 "fdlhold", 0); 1918 goto retry; 1919 } 1920 } 1921 fdtol->fdl_refcount--; 1922 if (fdtol->fdl_refcount == 0 && 1923 fdtol->fdl_holdcount == 0) { 1924 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1925 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 1926 } else 1927 fdtol = NULL; 1928 td->td_proc->p_fdtol = NULL; 1929 FILEDESC_XUNLOCK(fdp); 1930 if (fdtol != NULL) 1931 free(fdtol, M_FILEDESC_TO_LEADER); 1932 } 1933 FILEDESC_XLOCK(fdp); 1934 i = --fdp->fd_refcnt; 1935 FILEDESC_XUNLOCK(fdp); 1936 if (i > 0) 1937 return; 1938 1939 for (i = 0; i <= fdp->fd_lastfile; i++) { 1940 fp = fdp->fd_ofiles[i]; 1941 if (fp != NULL) { 1942 FILEDESC_XLOCK(fdp); 1943 fdp->fd_ofiles[i] = NULL; 1944 FILEDESC_XUNLOCK(fdp); 1945 (void) closef(fp, td); 1946 } 1947 } 1948 FILEDESC_XLOCK(fdp); 1949 1950 /* XXX This should happen earlier. */ 1951 mtx_lock(&fdesc_mtx); 1952 td->td_proc->p_fd = NULL; 1953 mtx_unlock(&fdesc_mtx); 1954 1955 if (fdp->fd_nfiles > NDFILE) 1956 free(fdp->fd_ofiles, M_FILEDESC); 1957 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 1958 free(fdp->fd_map, M_FILEDESC); 1959 1960 fdp->fd_nfiles = 0; 1961 1962 cdir = fdp->fd_cdir; 1963 fdp->fd_cdir = NULL; 1964 rdir = fdp->fd_rdir; 1965 fdp->fd_rdir = NULL; 1966 jdir = fdp->fd_jdir; 1967 fdp->fd_jdir = NULL; 1968 FILEDESC_XUNLOCK(fdp); 1969 1970 if (cdir) { 1971 locked = VFS_LOCK_GIANT(cdir->v_mount); 1972 vrele(cdir); 1973 VFS_UNLOCK_GIANT(locked); 1974 } 1975 if (rdir) { 1976 locked = VFS_LOCK_GIANT(rdir->v_mount); 1977 vrele(rdir); 1978 VFS_UNLOCK_GIANT(locked); 1979 } 1980 if (jdir) { 1981 locked = VFS_LOCK_GIANT(jdir->v_mount); 1982 vrele(jdir); 1983 VFS_UNLOCK_GIANT(locked); 1984 } 1985 1986 fddrop(fdp); 1987} 1988 1989/* 1990 * For setugid programs, we don't want to people to use that setugidness 1991 * to generate error messages which write to a file which otherwise would 1992 * otherwise be off-limits to the process. We check for filesystems where 1993 * the vnode can change out from under us after execve (like [lin]procfs). 1994 * 1995 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 1996 * sufficient. We also don't check for setugidness since we know we are. 1997 */ 1998static int 1999is_unsafe(struct file *fp) 2000{ 2001 if (fp->f_type == DTYPE_VNODE) { 2002 struct vnode *vp = fp->f_vnode; 2003 2004 if ((vp->v_vflag & VV_PROCDEP) != 0) 2005 return (1); 2006 } 2007 return (0); 2008} 2009 2010/* 2011 * Make this setguid thing safe, if at all possible. 2012 */ 2013void 2014setugidsafety(struct thread *td) 2015{ 2016 struct filedesc *fdp; 2017 int i; 2018 2019 /* Certain daemons might not have file descriptors. */ 2020 fdp = td->td_proc->p_fd; 2021 if (fdp == NULL) 2022 return; 2023 2024 /* 2025 * Note: fdp->fd_ofiles may be reallocated out from under us while 2026 * we are blocked in a close. Be careful! 2027 */ 2028 FILEDESC_XLOCK(fdp); 2029 for (i = 0; i <= fdp->fd_lastfile; i++) { 2030 if (i > 2) 2031 break; 2032 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) { 2033 struct file *fp; 2034 2035 knote_fdclose(td, i); 2036 /* 2037 * NULL-out descriptor prior to close to avoid 2038 * a race while close blocks. 2039 */ 2040 fp = fdp->fd_ofiles[i]; 2041 fdp->fd_ofiles[i] = NULL; 2042 fdp->fd_ofileflags[i] = 0; 2043 fdunused(fdp, i); 2044 FILEDESC_XUNLOCK(fdp); 2045 (void) closef(fp, td); 2046 FILEDESC_XLOCK(fdp); 2047 } 2048 } 2049 FILEDESC_XUNLOCK(fdp); 2050} 2051 2052/* 2053 * If a specific file object occupies a specific file descriptor, close the 2054 * file descriptor entry and drop a reference on the file object. This is a 2055 * convenience function to handle a subsequent error in a function that calls 2056 * falloc() that handles the race that another thread might have closed the 2057 * file descriptor out from under the thread creating the file object. 2058 */ 2059void 2060fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2061{ 2062 2063 FILEDESC_XLOCK(fdp); 2064 if (fdp->fd_ofiles[idx] == fp) { 2065 fdp->fd_ofiles[idx] = NULL; 2066 fdunused(fdp, idx); 2067 FILEDESC_XUNLOCK(fdp); 2068 fdrop(fp, td); 2069 } else 2070 FILEDESC_XUNLOCK(fdp); 2071} 2072 2073/* 2074 * Close any files on exec? 2075 */ 2076void 2077fdcloseexec(struct thread *td) 2078{ 2079 struct filedesc *fdp; 2080 int i; 2081 2082 /* Certain daemons might not have file descriptors. */ 2083 fdp = td->td_proc->p_fd; 2084 if (fdp == NULL) 2085 return; 2086 2087 FILEDESC_XLOCK(fdp); 2088 2089 /* 2090 * We cannot cache fd_ofiles or fd_ofileflags since operations 2091 * may block and rip them out from under us. 2092 */ 2093 for (i = 0; i <= fdp->fd_lastfile; i++) { 2094 if (fdp->fd_ofiles[i] != NULL && 2095 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE || 2096 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) { 2097 struct file *fp; 2098 2099 knote_fdclose(td, i); 2100 /* 2101 * NULL-out descriptor prior to close to avoid 2102 * a race while close blocks. 2103 */ 2104 fp = fdp->fd_ofiles[i]; 2105 fdp->fd_ofiles[i] = NULL; 2106 fdp->fd_ofileflags[i] = 0; 2107 fdunused(fdp, i); 2108 if (fp->f_type == DTYPE_MQUEUE) 2109 mq_fdclose(td, i, fp); 2110 FILEDESC_XUNLOCK(fdp); 2111 (void) closef(fp, td); 2112 FILEDESC_XLOCK(fdp); 2113 } 2114 } 2115 FILEDESC_XUNLOCK(fdp); 2116} 2117 2118/* 2119 * It is unsafe for set[ug]id processes to be started with file 2120 * descriptors 0..2 closed, as these descriptors are given implicit 2121 * significance in the Standard C library. fdcheckstd() will create a 2122 * descriptor referencing /dev/null for each of stdin, stdout, and 2123 * stderr that is not already open. 2124 */ 2125int 2126fdcheckstd(struct thread *td) 2127{ 2128 struct filedesc *fdp; 2129 register_t retval, save; 2130 int i, error, devnull; 2131 2132 fdp = td->td_proc->p_fd; 2133 if (fdp == NULL) 2134 return (0); 2135 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2136 devnull = -1; 2137 error = 0; 2138 for (i = 0; i < 3; i++) { 2139 if (fdp->fd_ofiles[i] != NULL) 2140 continue; 2141 if (devnull < 0) { 2142 save = td->td_retval[0]; 2143 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2144 O_RDWR, 0); 2145 devnull = td->td_retval[0]; 2146 td->td_retval[0] = save; 2147 if (error) 2148 break; 2149 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2150 } else { 2151 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2152 if (error != 0) 2153 break; 2154 } 2155 } 2156 return (error); 2157} 2158 2159/* 2160 * Internal form of close. Decrement reference count on file structure. 2161 * Note: td may be NULL when closing a file that was being passed in a 2162 * message. 2163 * 2164 * XXXRW: Giant is not required for the caller, but often will be held; this 2165 * makes it moderately likely the Giant will be recursed in the VFS case. 2166 */ 2167int 2168closef(struct file *fp, struct thread *td) 2169{ 2170 struct vnode *vp; 2171 struct flock lf; 2172 struct filedesc_to_leader *fdtol; 2173 struct filedesc *fdp; 2174 struct file *fp_object; 2175 2176 /* 2177 * POSIX record locking dictates that any close releases ALL 2178 * locks owned by this process. This is handled by setting 2179 * a flag in the unlock to free ONLY locks obeying POSIX 2180 * semantics, and not to free BSD-style file locks. 2181 * If the descriptor was in a message, POSIX-style locks 2182 * aren't passed with the descriptor, and the thread pointer 2183 * will be NULL. Callers should be careful only to pass a 2184 * NULL thread pointer when there really is no owning 2185 * context that might have locks, or the locks will be 2186 * leaked. 2187 * 2188 * If this is a capability, we do lock processing under the underlying 2189 * node, not the capability itself. 2190 */ 2191 (void)cap_funwrap(fp, 0, &fp_object); 2192 if ((fp_object->f_type == DTYPE_VNODE) && (td != NULL)) { 2193 int vfslocked; 2194 2195 vp = fp_object->f_vnode; 2196 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2197 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2198 lf.l_whence = SEEK_SET; 2199 lf.l_start = 0; 2200 lf.l_len = 0; 2201 lf.l_type = F_UNLCK; 2202 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2203 F_UNLCK, &lf, F_POSIX); 2204 } 2205 fdtol = td->td_proc->p_fdtol; 2206 if (fdtol != NULL) { 2207 /* 2208 * Handle special case where file descriptor table is 2209 * shared between multiple process leaders. 2210 */ 2211 fdp = td->td_proc->p_fd; 2212 FILEDESC_XLOCK(fdp); 2213 for (fdtol = fdtol->fdl_next; 2214 fdtol != td->td_proc->p_fdtol; 2215 fdtol = fdtol->fdl_next) { 2216 if ((fdtol->fdl_leader->p_flag & 2217 P_ADVLOCK) == 0) 2218 continue; 2219 fdtol->fdl_holdcount++; 2220 FILEDESC_XUNLOCK(fdp); 2221 lf.l_whence = SEEK_SET; 2222 lf.l_start = 0; 2223 lf.l_len = 0; 2224 lf.l_type = F_UNLCK; 2225 vp = fp_object->f_vnode; 2226 (void) VOP_ADVLOCK(vp, 2227 (caddr_t)fdtol->fdl_leader, 2228 F_UNLCK, &lf, F_POSIX); 2229 FILEDESC_XLOCK(fdp); 2230 fdtol->fdl_holdcount--; 2231 if (fdtol->fdl_holdcount == 0 && 2232 fdtol->fdl_wakeup != 0) { 2233 fdtol->fdl_wakeup = 0; 2234 wakeup(fdtol); 2235 } 2236 } 2237 FILEDESC_XUNLOCK(fdp); 2238 } 2239 VFS_UNLOCK_GIANT(vfslocked); 2240 } 2241 return (fdrop(fp, td)); 2242} 2243 2244/* 2245 * Initialize the file pointer with the specified properties. 2246 * 2247 * The ops are set with release semantics to be certain that the flags, type, 2248 * and data are visible when ops is. This is to prevent ops methods from being 2249 * called with bad data. 2250 */ 2251void 2252finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2253{ 2254 fp->f_data = data; 2255 fp->f_flag = flag; 2256 fp->f_type = type; 2257 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2258} 2259 2260struct file * 2261fget_unlocked(struct filedesc *fdp, int fd) 2262{ 2263 struct file *fp; 2264 u_int count; 2265 2266 if (fd < 0 || fd >= fdp->fd_nfiles) 2267 return (NULL); 2268 /* 2269 * Fetch the descriptor locklessly. We avoid fdrop() races by 2270 * never raising a refcount above 0. To accomplish this we have 2271 * to use a cmpset loop rather than an atomic_add. The descriptor 2272 * must be re-verified once we acquire a reference to be certain 2273 * that the identity is still correct and we did not lose a race 2274 * due to preemption. 2275 */ 2276 for (;;) { 2277 fp = fdp->fd_ofiles[fd]; 2278 if (fp == NULL) 2279 break; 2280 count = fp->f_count; 2281 if (count == 0) 2282 continue; 2283 /* 2284 * Use an acquire barrier to prevent caching of fd_ofiles 2285 * so it is refreshed for verification. 2286 */ 2287 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2288 continue; 2289 if (fp == fdp->fd_ofiles[fd]) 2290 break; 2291 fdrop(fp, curthread); 2292 } 2293 2294 return (fp); 2295} 2296 2297/* 2298 * Extract the file pointer associated with the specified descriptor for the 2299 * current user process. 2300 * 2301 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2302 * returned. 2303 * 2304 * If the FGET_GETCAP flag is set, the capability itself will be returned. 2305 * Calling _fget() with FGET_GETCAP on a non-capability will return EINVAL. 2306 * Otherwise, if the file is a capability, its rights will be checked against 2307 * the capability rights mask, and if successful, the object will be unwrapped. 2308 * 2309 * If an error occured the non-zero error is returned and *fpp is set to 2310 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2311 * responsible for fdrop(). 2312 */ 2313#define FGET_GETCAP 0x00000001 2314static __inline int 2315_fget(struct thread *td, int fd, struct file **fpp, int flags, 2316 cap_rights_t needrights, cap_rights_t *haverightsp, u_char *maxprotp, 2317 int fget_flags) 2318{ 2319 struct filedesc *fdp; 2320 struct file *fp; 2321#ifdef CAPABILITIES 2322 struct file *fp_fromcap; 2323 int error; 2324#endif 2325 2326 *fpp = NULL; 2327 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2328 return (EBADF); 2329 if ((fp = fget_unlocked(fdp, fd)) == NULL) 2330 return (EBADF); 2331 if (fp->f_ops == &badfileops) { 2332 fdrop(fp, td); 2333 return (EBADF); 2334 } 2335 2336#ifdef CAPABILITIES 2337 /* 2338 * If this is a capability, what rights does it have? 2339 */ 2340 if (haverightsp != NULL) { 2341 if (fp->f_type == DTYPE_CAPABILITY) 2342 *haverightsp = cap_rights(fp); 2343 else 2344 *haverightsp = CAP_MASK_VALID; 2345 } 2346 2347 /* 2348 * If a capability has been requested, return the capability directly. 2349 * Otherwise, check capability rights, extract the underlying object, 2350 * and check its access flags. 2351 */ 2352 if (fget_flags & FGET_GETCAP) { 2353 if (fp->f_type != DTYPE_CAPABILITY) { 2354 fdrop(fp, td); 2355 return (EINVAL); 2356 } 2357 } else { 2358 if (maxprotp == NULL) 2359 error = cap_funwrap(fp, needrights, &fp_fromcap); 2360 else 2361 error = cap_funwrap_mmap(fp, needrights, maxprotp, 2362 &fp_fromcap); 2363 if (error) { 2364 fdrop(fp, td); 2365 return (error); 2366 } 2367 2368 /* 2369 * If we've unwrapped a file, drop the original capability 2370 * and hold the new descriptor. fp after this point refers to 2371 * the actual (unwrapped) object, not the capability. 2372 */ 2373 if (fp != fp_fromcap) { 2374 fhold(fp_fromcap); 2375 fdrop(fp, td); 2376 fp = fp_fromcap; 2377 } 2378 } 2379#else /* !CAPABILITIES */ 2380 KASSERT(fp->f_type != DTYPE_CAPABILITY, 2381 ("%s: saw capability", __func__)); 2382 if (maxprotp != NULL) 2383 *maxprotp = VM_PROT_ALL; 2384#endif /* CAPABILITIES */ 2385 2386 /* 2387 * FREAD and FWRITE failure return EBADF as per POSIX. 2388 * 2389 * Only one flag, or 0, may be specified. 2390 */ 2391 if ((flags == FREAD && (fp->f_flag & FREAD) == 0) || 2392 (flags == FWRITE && (fp->f_flag & FWRITE) == 0)) { 2393 fdrop(fp, td); 2394 return (EBADF); 2395 } 2396 *fpp = fp; 2397 return (0); 2398} 2399 2400int 2401fget(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2402{ 2403 2404 return(_fget(td, fd, fpp, 0, rights, NULL, NULL, 0)); 2405} 2406 2407int 2408fget_mmap(struct thread *td, int fd, cap_rights_t rights, u_char *maxprotp, 2409 struct file **fpp) 2410{ 2411 2412 return (_fget(td, fd, fpp, 0, rights, NULL, maxprotp, 0)); 2413} 2414 2415int 2416fget_read(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2417{ 2418 2419 return(_fget(td, fd, fpp, FREAD, rights, NULL, NULL, 0)); 2420} 2421 2422int 2423fget_write(struct thread *td, int fd, cap_rights_t rights, struct file **fpp) 2424{ 2425 2426 return (_fget(td, fd, fpp, FWRITE, rights, NULL, NULL, 0)); 2427} 2428 2429/* 2430 * Unlike the other fget() calls, which accept and check capability rights 2431 * but never return capabilities, fgetcap() returns the capability but doesn't 2432 * check capability rights. 2433 */ 2434int 2435fgetcap(struct thread *td, int fd, struct file **fpp) 2436{ 2437 2438 return (_fget(td, fd, fpp, 0, 0, NULL, NULL, FGET_GETCAP)); 2439} 2440 2441 2442/* 2443 * Like fget() but loads the underlying vnode, or returns an error if the 2444 * descriptor does not represent a vnode. Note that pipes use vnodes but 2445 * never have VM objects. The returned vnode will be vref()'d. 2446 * 2447 * XXX: what about the unused flags ? 2448 */ 2449static __inline int 2450_fgetvp(struct thread *td, int fd, int flags, cap_rights_t needrights, 2451 cap_rights_t *haverightsp, struct vnode **vpp) 2452{ 2453 struct file *fp; 2454 int error; 2455 2456 *vpp = NULL; 2457 if ((error = _fget(td, fd, &fp, flags, needrights, haverightsp, 2458 NULL, 0)) != 0) 2459 return (error); 2460 if (fp->f_vnode == NULL) { 2461 error = EINVAL; 2462 } else { 2463 *vpp = fp->f_vnode; 2464 vref(*vpp); 2465 } 2466 fdrop(fp, td); 2467 2468 return (error); 2469} 2470 2471int 2472fgetvp(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp) 2473{ 2474 2475 return (_fgetvp(td, fd, 0, rights, NULL, vpp)); 2476} 2477 2478int 2479fgetvp_rights(struct thread *td, int fd, cap_rights_t need, cap_rights_t *have, 2480 struct vnode **vpp) 2481{ 2482 return (_fgetvp(td, fd, 0, need, have, vpp)); 2483} 2484 2485int 2486fgetvp_read(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp) 2487{ 2488 2489 return (_fgetvp(td, fd, FREAD, rights, NULL, vpp)); 2490} 2491 2492#ifdef notyet 2493int 2494fgetvp_write(struct thread *td, int fd, cap_rights_t rights, 2495 struct vnode **vpp) 2496{ 2497 2498 return (_fgetvp(td, fd, FWRITE, rights, NULL, vpp)); 2499} 2500#endif 2501 2502/* 2503 * Like fget() but loads the underlying socket, or returns an error if the 2504 * descriptor does not represent a socket. 2505 * 2506 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2507 * in the future. 2508 * 2509 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2510 * on their file descriptor reference to prevent the socket from being free'd 2511 * during use. 2512 */ 2513int 2514fgetsock(struct thread *td, int fd, cap_rights_t rights, struct socket **spp, 2515 u_int *fflagp) 2516{ 2517 struct file *fp; 2518 int error; 2519 2520 *spp = NULL; 2521 if (fflagp != NULL) 2522 *fflagp = 0; 2523 if ((error = _fget(td, fd, &fp, 0, rights, NULL, NULL, 0)) != 0) 2524 return (error); 2525 if (fp->f_type != DTYPE_SOCKET) { 2526 error = ENOTSOCK; 2527 } else { 2528 *spp = fp->f_data; 2529 if (fflagp) 2530 *fflagp = fp->f_flag; 2531 SOCK_LOCK(*spp); 2532 soref(*spp); 2533 SOCK_UNLOCK(*spp); 2534 } 2535 fdrop(fp, td); 2536 2537 return (error); 2538} 2539 2540/* 2541 * Drop the reference count on the socket and XXX release the SX lock in the 2542 * future. The last reference closes the socket. 2543 * 2544 * Note: fputsock() is deprecated, see comment for fgetsock(). 2545 */ 2546void 2547fputsock(struct socket *so) 2548{ 2549 2550 ACCEPT_LOCK(); 2551 SOCK_LOCK(so); 2552 CURVNET_SET(so->so_vnet); 2553 sorele(so); 2554 CURVNET_RESTORE(); 2555} 2556 2557/* 2558 * Handle the last reference to a file being closed. 2559 * 2560 * No special capability handling here, as the capability's fo_close will run 2561 * instead of the object here, and perform any necessary drop on the object. 2562 */ 2563int 2564_fdrop(struct file *fp, struct thread *td) 2565{ 2566 int error; 2567 2568 error = 0; 2569 if (fp->f_count != 0) 2570 panic("fdrop: count %d", fp->f_count); 2571 if (fp->f_ops != &badfileops) 2572 error = fo_close(fp, td); 2573 atomic_subtract_int(&openfiles, 1); 2574 crfree(fp->f_cred); 2575 free(fp->f_advice, M_FADVISE); 2576 uma_zfree(file_zone, fp); 2577 2578 return (error); 2579} 2580 2581/* 2582 * Apply an advisory lock on a file descriptor. 2583 * 2584 * Just attempt to get a record lock of the requested type on the entire file 2585 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2586 */ 2587#ifndef _SYS_SYSPROTO_H_ 2588struct flock_args { 2589 int fd; 2590 int how; 2591}; 2592#endif 2593/* ARGSUSED */ 2594int 2595sys_flock(struct thread *td, struct flock_args *uap) 2596{ 2597 struct file *fp; 2598 struct vnode *vp; 2599 struct flock lf; 2600 int vfslocked; 2601 int error; 2602 2603 if ((error = fget(td, uap->fd, CAP_FLOCK, &fp)) != 0) 2604 return (error); 2605 if (fp->f_type != DTYPE_VNODE) { 2606 fdrop(fp, td); 2607 return (EOPNOTSUPP); 2608 } 2609 2610 vp = fp->f_vnode; 2611 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2612 lf.l_whence = SEEK_SET; 2613 lf.l_start = 0; 2614 lf.l_len = 0; 2615 if (uap->how & LOCK_UN) { 2616 lf.l_type = F_UNLCK; 2617 atomic_clear_int(&fp->f_flag, FHASLOCK); 2618 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2619 goto done2; 2620 } 2621 if (uap->how & LOCK_EX) 2622 lf.l_type = F_WRLCK; 2623 else if (uap->how & LOCK_SH) 2624 lf.l_type = F_RDLCK; 2625 else { 2626 error = EBADF; 2627 goto done2; 2628 } 2629 atomic_set_int(&fp->f_flag, FHASLOCK); 2630 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2631 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2632done2: 2633 fdrop(fp, td); 2634 VFS_UNLOCK_GIANT(vfslocked); 2635 return (error); 2636} 2637/* 2638 * Duplicate the specified descriptor to a free descriptor. 2639 */ 2640int 2641dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error) 2642{ 2643 struct file *wfp; 2644 struct file *fp; 2645 2646 /* 2647 * If the to-be-dup'd fd number is greater than the allowed number 2648 * of file descriptors, or the fd to be dup'd has already been 2649 * closed, then reject. 2650 */ 2651 FILEDESC_XLOCK(fdp); 2652 if (dfd < 0 || dfd >= fdp->fd_nfiles || 2653 (wfp = fdp->fd_ofiles[dfd]) == NULL) { 2654 FILEDESC_XUNLOCK(fdp); 2655 return (EBADF); 2656 } 2657 2658 /* 2659 * There are two cases of interest here. 2660 * 2661 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2662 * 2663 * For ENXIO steal away the file structure from (dfd) and store it in 2664 * (indx). (dfd) is effectively closed by this operation. 2665 * 2666 * Any other error code is just returned. 2667 */ 2668 switch (error) { 2669 case ENODEV: 2670 /* 2671 * Check that the mode the file is being opened for is a 2672 * subset of the mode of the existing descriptor. 2673 */ 2674 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { 2675 FILEDESC_XUNLOCK(fdp); 2676 return (EACCES); 2677 } 2678 fp = fdp->fd_ofiles[indx]; 2679 fdp->fd_ofiles[indx] = wfp; 2680 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2681 if (fp == NULL) 2682 fdused(fdp, indx); 2683 fhold(wfp); 2684 FILEDESC_XUNLOCK(fdp); 2685 if (fp != NULL) 2686 /* 2687 * We now own the reference to fp that the ofiles[] 2688 * array used to own. Release it. 2689 */ 2690 fdrop(fp, td); 2691 return (0); 2692 2693 case ENXIO: 2694 /* 2695 * Steal away the file pointer from dfd and stuff it into indx. 2696 */ 2697 fp = fdp->fd_ofiles[indx]; 2698 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2699 fdp->fd_ofiles[dfd] = NULL; 2700 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 2701 fdp->fd_ofileflags[dfd] = 0; 2702 fdunused(fdp, dfd); 2703 if (fp == NULL) 2704 fdused(fdp, indx); 2705 FILEDESC_XUNLOCK(fdp); 2706 2707 /* 2708 * We now own the reference to fp that the ofiles[] array 2709 * used to own. Release it. 2710 */ 2711 if (fp != NULL) 2712 fdrop(fp, td); 2713 return (0); 2714 2715 default: 2716 FILEDESC_XUNLOCK(fdp); 2717 return (error); 2718 } 2719 /* NOTREACHED */ 2720} 2721 2722/* 2723 * Scan all active processes and prisons to see if any of them have a current 2724 * or root directory of `olddp'. If so, replace them with the new mount point. 2725 */ 2726void 2727mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2728{ 2729 struct filedesc *fdp; 2730 struct prison *pr; 2731 struct proc *p; 2732 int nrele; 2733 2734 if (vrefcnt(olddp) == 1) 2735 return; 2736 nrele = 0; 2737 sx_slock(&allproc_lock); 2738 FOREACH_PROC_IN_SYSTEM(p) { 2739 fdp = fdhold(p); 2740 if (fdp == NULL) 2741 continue; 2742 FILEDESC_XLOCK(fdp); 2743 if (fdp->fd_cdir == olddp) { 2744 vref(newdp); 2745 fdp->fd_cdir = newdp; 2746 nrele++; 2747 } 2748 if (fdp->fd_rdir == olddp) { 2749 vref(newdp); 2750 fdp->fd_rdir = newdp; 2751 nrele++; 2752 } 2753 if (fdp->fd_jdir == olddp) { 2754 vref(newdp); 2755 fdp->fd_jdir = newdp; 2756 nrele++; 2757 } 2758 FILEDESC_XUNLOCK(fdp); 2759 fddrop(fdp); 2760 } 2761 sx_sunlock(&allproc_lock); 2762 if (rootvnode == olddp) { 2763 vref(newdp); 2764 rootvnode = newdp; 2765 nrele++; 2766 } 2767 mtx_lock(&prison0.pr_mtx); 2768 if (prison0.pr_root == olddp) { 2769 vref(newdp); 2770 prison0.pr_root = newdp; 2771 nrele++; 2772 } 2773 mtx_unlock(&prison0.pr_mtx); 2774 sx_slock(&allprison_lock); 2775 TAILQ_FOREACH(pr, &allprison, pr_list) { 2776 mtx_lock(&pr->pr_mtx); 2777 if (pr->pr_root == olddp) { 2778 vref(newdp); 2779 pr->pr_root = newdp; 2780 nrele++; 2781 } 2782 mtx_unlock(&pr->pr_mtx); 2783 } 2784 sx_sunlock(&allprison_lock); 2785 while (nrele--) 2786 vrele(olddp); 2787} 2788 2789struct filedesc_to_leader * 2790filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2791{ 2792 struct filedesc_to_leader *fdtol; 2793 2794 fdtol = malloc(sizeof(struct filedesc_to_leader), 2795 M_FILEDESC_TO_LEADER, 2796 M_WAITOK); 2797 fdtol->fdl_refcount = 1; 2798 fdtol->fdl_holdcount = 0; 2799 fdtol->fdl_wakeup = 0; 2800 fdtol->fdl_leader = leader; 2801 if (old != NULL) { 2802 FILEDESC_XLOCK(fdp); 2803 fdtol->fdl_next = old->fdl_next; 2804 fdtol->fdl_prev = old; 2805 old->fdl_next = fdtol; 2806 fdtol->fdl_next->fdl_prev = fdtol; 2807 FILEDESC_XUNLOCK(fdp); 2808 } else { 2809 fdtol->fdl_next = fdtol; 2810 fdtol->fdl_prev = fdtol; 2811 } 2812 return (fdtol); 2813} 2814 2815/* 2816 * Get file structures globally. 2817 */ 2818static int 2819sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2820{ 2821 struct xfile xf; 2822 struct filedesc *fdp; 2823 struct file *fp; 2824 struct proc *p; 2825 int error, n; 2826 2827 error = sysctl_wire_old_buffer(req, 0); 2828 if (error != 0) 2829 return (error); 2830 if (req->oldptr == NULL) { 2831 n = 0; 2832 sx_slock(&allproc_lock); 2833 FOREACH_PROC_IN_SYSTEM(p) { 2834 if (p->p_state == PRS_NEW) 2835 continue; 2836 fdp = fdhold(p); 2837 if (fdp == NULL) 2838 continue; 2839 /* overestimates sparse tables. */ 2840 if (fdp->fd_lastfile > 0) 2841 n += fdp->fd_lastfile; 2842 fddrop(fdp); 2843 } 2844 sx_sunlock(&allproc_lock); 2845 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2846 } 2847 error = 0; 2848 bzero(&xf, sizeof(xf)); 2849 xf.xf_size = sizeof(xf); 2850 sx_slock(&allproc_lock); 2851 FOREACH_PROC_IN_SYSTEM(p) { 2852 PROC_LOCK(p); 2853 if (p->p_state == PRS_NEW) { 2854 PROC_UNLOCK(p); 2855 continue; 2856 } 2857 if (p_cansee(req->td, p) != 0) { 2858 PROC_UNLOCK(p); 2859 continue; 2860 } 2861 xf.xf_pid = p->p_pid; 2862 xf.xf_uid = p->p_ucred->cr_uid; 2863 PROC_UNLOCK(p); 2864 fdp = fdhold(p); 2865 if (fdp == NULL) 2866 continue; 2867 FILEDESC_SLOCK(fdp); 2868 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2869 if ((fp = fdp->fd_ofiles[n]) == NULL) 2870 continue; 2871 xf.xf_fd = n; 2872 xf.xf_file = fp; 2873 xf.xf_data = fp->f_data; 2874 xf.xf_vnode = fp->f_vnode; 2875 xf.xf_type = fp->f_type; 2876 xf.xf_count = fp->f_count; 2877 xf.xf_msgcount = 0; 2878 xf.xf_offset = fp->f_offset; 2879 xf.xf_flag = fp->f_flag; 2880 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2881 if (error) 2882 break; 2883 } 2884 FILEDESC_SUNLOCK(fdp); 2885 fddrop(fdp); 2886 if (error) 2887 break; 2888 } 2889 sx_sunlock(&allproc_lock); 2890 return (error); 2891} 2892 2893SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2894 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2895 2896#ifdef KINFO_OFILE_SIZE 2897CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2898#endif 2899 2900#ifdef COMPAT_FREEBSD7 2901static int 2902export_vnode_for_osysctl(struct vnode *vp, int type, 2903 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2904{ 2905 int error; 2906 char *fullpath, *freepath; 2907 int vfslocked; 2908 2909 bzero(kif, sizeof(*kif)); 2910 kif->kf_structsize = sizeof(*kif); 2911 2912 vref(vp); 2913 kif->kf_fd = type; 2914 kif->kf_type = KF_TYPE_VNODE; 2915 /* This function only handles directories. */ 2916 if (vp->v_type != VDIR) { 2917 vrele(vp); 2918 return (ENOTDIR); 2919 } 2920 kif->kf_vnode_type = KF_VTYPE_VDIR; 2921 2922 /* 2923 * This is not a true file descriptor, so we set a bogus refcount 2924 * and offset to indicate these fields should be ignored. 2925 */ 2926 kif->kf_ref_count = -1; 2927 kif->kf_offset = -1; 2928 2929 freepath = NULL; 2930 fullpath = "-"; 2931 FILEDESC_SUNLOCK(fdp); 2932 vn_fullpath(curthread, vp, &fullpath, &freepath); 2933 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2934 vrele(vp); 2935 VFS_UNLOCK_GIANT(vfslocked); 2936 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2937 if (freepath != NULL) 2938 free(freepath, M_TEMP); 2939 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2940 FILEDESC_SLOCK(fdp); 2941 return (error); 2942} 2943 2944/* 2945 * Get per-process file descriptors for use by procstat(1), et al. 2946 */ 2947static int 2948sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 2949{ 2950 char *fullpath, *freepath; 2951 struct kinfo_ofile *kif; 2952 struct filedesc *fdp; 2953 int error, i, *name; 2954 struct shmfd *shmfd; 2955 struct socket *so; 2956 struct vnode *vp; 2957 struct file *fp; 2958 struct proc *p; 2959 struct tty *tp; 2960 int vfslocked; 2961 2962 name = (int *)arg1; 2963 if ((p = pfind((pid_t)name[0])) == NULL) 2964 return (ESRCH); 2965 if ((error = p_candebug(curthread, p))) { 2966 PROC_UNLOCK(p); 2967 return (error); 2968 } 2969 fdp = fdhold(p); 2970 PROC_UNLOCK(p); 2971 if (fdp == NULL) 2972 return (ENOENT); 2973 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 2974 FILEDESC_SLOCK(fdp); 2975 if (fdp->fd_cdir != NULL) 2976 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 2977 fdp, req); 2978 if (fdp->fd_rdir != NULL) 2979 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 2980 fdp, req); 2981 if (fdp->fd_jdir != NULL) 2982 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 2983 fdp, req); 2984 for (i = 0; i < fdp->fd_nfiles; i++) { 2985 if ((fp = fdp->fd_ofiles[i]) == NULL) 2986 continue; 2987 bzero(kif, sizeof(*kif)); 2988 kif->kf_structsize = sizeof(*kif); 2989 vp = NULL; 2990 so = NULL; 2991 tp = NULL; 2992 shmfd = NULL; 2993 kif->kf_fd = i; 2994 2995#ifdef CAPABILITIES 2996 /* 2997 * When reporting a capability, most fields will be from the 2998 * underlying object, but do mark as a capability. With 2999 * ofiledesc, we don't have a field to export the cap_rights_t, 3000 * but we do with the new filedesc. 3001 */ 3002 if (fp->f_type == DTYPE_CAPABILITY) { 3003 kif->kf_flags |= KF_FLAG_CAPABILITY; 3004 (void)cap_funwrap(fp, 0, &fp); 3005 } 3006#else 3007 KASSERT(fp->f_type != DTYPE_CAPABILITY, 3008 ("sysctl_kern_proc_ofiledesc: saw capability")); 3009#endif 3010 switch (fp->f_type) { 3011 case DTYPE_VNODE: 3012 kif->kf_type = KF_TYPE_VNODE; 3013 vp = fp->f_vnode; 3014 break; 3015 3016 case DTYPE_SOCKET: 3017 kif->kf_type = KF_TYPE_SOCKET; 3018 so = fp->f_data; 3019 break; 3020 3021 case DTYPE_PIPE: 3022 kif->kf_type = KF_TYPE_PIPE; 3023 break; 3024 3025 case DTYPE_FIFO: 3026 kif->kf_type = KF_TYPE_FIFO; 3027 vp = fp->f_vnode; 3028 break; 3029 3030 case DTYPE_KQUEUE: 3031 kif->kf_type = KF_TYPE_KQUEUE; 3032 break; 3033 3034 case DTYPE_CRYPTO: 3035 kif->kf_type = KF_TYPE_CRYPTO; 3036 break; 3037 3038 case DTYPE_MQUEUE: 3039 kif->kf_type = KF_TYPE_MQUEUE; 3040 break; 3041 3042 case DTYPE_SHM: 3043 kif->kf_type = KF_TYPE_SHM; 3044 shmfd = fp->f_data; 3045 break; 3046 3047 case DTYPE_SEM: 3048 kif->kf_type = KF_TYPE_SEM; 3049 break; 3050 3051 case DTYPE_PTS: 3052 kif->kf_type = KF_TYPE_PTS; 3053 tp = fp->f_data; 3054 break; 3055 3056#ifdef PROCDESC 3057 case DTYPE_PROCDESC: 3058 kif->kf_type = KF_TYPE_PROCDESC; 3059 break; 3060#endif 3061 3062 default: 3063 kif->kf_type = KF_TYPE_UNKNOWN; 3064 break; 3065 } 3066 kif->kf_ref_count = fp->f_count; 3067 if (fp->f_flag & FREAD) 3068 kif->kf_flags |= KF_FLAG_READ; 3069 if (fp->f_flag & FWRITE) 3070 kif->kf_flags |= KF_FLAG_WRITE; 3071 if (fp->f_flag & FAPPEND) 3072 kif->kf_flags |= KF_FLAG_APPEND; 3073 if (fp->f_flag & FASYNC) 3074 kif->kf_flags |= KF_FLAG_ASYNC; 3075 if (fp->f_flag & FFSYNC) 3076 kif->kf_flags |= KF_FLAG_FSYNC; 3077 if (fp->f_flag & FNONBLOCK) 3078 kif->kf_flags |= KF_FLAG_NONBLOCK; 3079 if (fp->f_flag & O_DIRECT) 3080 kif->kf_flags |= KF_FLAG_DIRECT; 3081 if (fp->f_flag & FHASLOCK) 3082 kif->kf_flags |= KF_FLAG_HASLOCK; 3083 kif->kf_offset = fp->f_offset; 3084 if (vp != NULL) { 3085 vref(vp); 3086 switch (vp->v_type) { 3087 case VNON: 3088 kif->kf_vnode_type = KF_VTYPE_VNON; 3089 break; 3090 case VREG: 3091 kif->kf_vnode_type = KF_VTYPE_VREG; 3092 break; 3093 case VDIR: 3094 kif->kf_vnode_type = KF_VTYPE_VDIR; 3095 break; 3096 case VBLK: 3097 kif->kf_vnode_type = KF_VTYPE_VBLK; 3098 break; 3099 case VCHR: 3100 kif->kf_vnode_type = KF_VTYPE_VCHR; 3101 break; 3102 case VLNK: 3103 kif->kf_vnode_type = KF_VTYPE_VLNK; 3104 break; 3105 case VSOCK: 3106 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3107 break; 3108 case VFIFO: 3109 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3110 break; 3111 case VBAD: 3112 kif->kf_vnode_type = KF_VTYPE_VBAD; 3113 break; 3114 default: 3115 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3116 break; 3117 } 3118 /* 3119 * It is OK to drop the filedesc lock here as we will 3120 * re-validate and re-evaluate its properties when 3121 * the loop continues. 3122 */ 3123 freepath = NULL; 3124 fullpath = "-"; 3125 FILEDESC_SUNLOCK(fdp); 3126 vn_fullpath(curthread, vp, &fullpath, &freepath); 3127 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3128 vrele(vp); 3129 VFS_UNLOCK_GIANT(vfslocked); 3130 strlcpy(kif->kf_path, fullpath, 3131 sizeof(kif->kf_path)); 3132 if (freepath != NULL) 3133 free(freepath, M_TEMP); 3134 FILEDESC_SLOCK(fdp); 3135 } 3136 if (so != NULL) { 3137 struct sockaddr *sa; 3138 3139 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3140 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3141 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3142 free(sa, M_SONAME); 3143 } 3144 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3145 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3146 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3147 free(sa, M_SONAME); 3148 } 3149 kif->kf_sock_domain = 3150 so->so_proto->pr_domain->dom_family; 3151 kif->kf_sock_type = so->so_type; 3152 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3153 } 3154 if (tp != NULL) { 3155 strlcpy(kif->kf_path, tty_devname(tp), 3156 sizeof(kif->kf_path)); 3157 } 3158 if (shmfd != NULL) 3159 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3160 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3161 if (error) 3162 break; 3163 } 3164 FILEDESC_SUNLOCK(fdp); 3165 fddrop(fdp); 3166 free(kif, M_TEMP); 3167 return (0); 3168} 3169 3170static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD, 3171 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); 3172#endif /* COMPAT_FREEBSD7 */ 3173 3174#ifdef KINFO_FILE_SIZE 3175CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3176#endif 3177 3178static int 3179export_fd_for_sysctl(void *data, int type, int fd, int fflags, int refcnt, 3180 int64_t offset, int fd_is_cap, cap_rights_t fd_cap_rights, 3181 struct kinfo_file *kif, struct sysctl_req *req) 3182{ 3183 struct { 3184 int fflag; 3185 int kf_fflag; 3186 } fflags_table[] = { 3187 { FAPPEND, KF_FLAG_APPEND }, 3188 { FASYNC, KF_FLAG_ASYNC }, 3189 { FFSYNC, KF_FLAG_FSYNC }, 3190 { FHASLOCK, KF_FLAG_HASLOCK }, 3191 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3192 { FREAD, KF_FLAG_READ }, 3193 { FWRITE, KF_FLAG_WRITE }, 3194 { O_CREAT, KF_FLAG_CREAT }, 3195 { O_DIRECT, KF_FLAG_DIRECT }, 3196 { O_EXCL, KF_FLAG_EXCL }, 3197 { O_EXEC, KF_FLAG_EXEC }, 3198 { O_EXLOCK, KF_FLAG_EXLOCK }, 3199 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3200 { O_SHLOCK, KF_FLAG_SHLOCK }, 3201 { O_TRUNC, KF_FLAG_TRUNC } 3202 }; 3203#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3204 struct vnode *vp; 3205 int error, vfslocked; 3206 unsigned int i; 3207 3208 bzero(kif, sizeof(*kif)); 3209 switch (type) { 3210 case KF_TYPE_FIFO: 3211 case KF_TYPE_VNODE: 3212 vp = (struct vnode *)data; 3213 error = fill_vnode_info(vp, kif); 3214 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3215 vrele(vp); 3216 VFS_UNLOCK_GIANT(vfslocked); 3217 break; 3218 case KF_TYPE_SOCKET: 3219 error = fill_socket_info((struct socket *)data, kif); 3220 break; 3221 case KF_TYPE_PIPE: 3222 error = fill_pipe_info((struct pipe *)data, kif); 3223 break; 3224 case KF_TYPE_PTS: 3225 error = fill_pts_info((struct tty *)data, kif); 3226 break; 3227 case KF_TYPE_PROCDESC: 3228 error = fill_procdesc_info((struct procdesc *)data, kif); 3229 break; 3230 case KF_TYPE_SHM: 3231 error = fill_shm_info((struct file *)data, kif); 3232 break; 3233 default: 3234 error = 0; 3235 } 3236 if (error == 0) 3237 kif->kf_status |= KF_ATTR_VALID; 3238 3239 /* 3240 * Translate file access flags. 3241 */ 3242 for (i = 0; i < NFFLAGS; i++) 3243 if (fflags & fflags_table[i].fflag) 3244 kif->kf_flags |= fflags_table[i].kf_fflag; 3245 if (fd_is_cap) 3246 kif->kf_flags |= KF_FLAG_CAPABILITY; 3247 if (fd_is_cap) 3248 kif->kf_cap_rights = fd_cap_rights; 3249 kif->kf_fd = fd; 3250 kif->kf_type = type; 3251 kif->kf_ref_count = refcnt; 3252 kif->kf_offset = offset; 3253 /* Pack record size down */ 3254 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3255 strlen(kif->kf_path) + 1; 3256 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3257 error = SYSCTL_OUT(req, kif, kif->kf_structsize); 3258 return (error); 3259} 3260 3261/* 3262 * Get per-process file descriptors for use by procstat(1), et al. 3263 */ 3264static int 3265sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3266{ 3267 struct file *fp; 3268 struct filedesc *fdp; 3269 struct kinfo_file *kif; 3270 struct proc *p; 3271 struct vnode *cttyvp, *textvp, *tracevp; 3272 size_t oldidx; 3273 int64_t offset; 3274 void *data; 3275 int error, i, *name; 3276 int fd_is_cap, type, refcnt, fflags; 3277 cap_rights_t fd_cap_rights; 3278 3279 name = (int *)arg1; 3280 if ((p = pfind((pid_t)name[0])) == NULL) 3281 return (ESRCH); 3282 if ((error = p_candebug(curthread, p))) { 3283 PROC_UNLOCK(p); 3284 return (error); 3285 } 3286 /* ktrace vnode */ 3287 tracevp = p->p_tracevp; 3288 if (tracevp != NULL) 3289 vref(tracevp); 3290 /* text vnode */ 3291 textvp = p->p_textvp; 3292 if (textvp != NULL) 3293 vref(textvp); 3294 /* Controlling tty. */ 3295 cttyvp = NULL; 3296 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3297 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3298 if (cttyvp != NULL) 3299 vref(cttyvp); 3300 } 3301 fdp = fdhold(p); 3302 PROC_UNLOCK(p); 3303 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3304 if (tracevp != NULL) 3305 export_fd_for_sysctl(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3306 FREAD | FWRITE, -1, -1, 0, 0, kif, req); 3307 if (textvp != NULL) 3308 export_fd_for_sysctl(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3309 FREAD, -1, -1, 0, 0, kif, req); 3310 if (cttyvp != NULL) 3311 export_fd_for_sysctl(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3312 FREAD | FWRITE, -1, -1, 0, 0, kif, req); 3313 if (fdp == NULL) 3314 goto fail; 3315 FILEDESC_SLOCK(fdp); 3316 /* working directory */ 3317 if (fdp->fd_cdir != NULL) { 3318 vref(fdp->fd_cdir); 3319 data = fdp->fd_cdir; 3320 FILEDESC_SUNLOCK(fdp); 3321 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3322 FREAD, -1, -1, 0, 0, kif, req); 3323 FILEDESC_SLOCK(fdp); 3324 } 3325 /* root directory */ 3326 if (fdp->fd_rdir != NULL) { 3327 vref(fdp->fd_rdir); 3328 data = fdp->fd_rdir; 3329 FILEDESC_SUNLOCK(fdp); 3330 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3331 FREAD, -1, -1, 0, 0, kif, req); 3332 FILEDESC_SLOCK(fdp); 3333 } 3334 /* jail directory */ 3335 if (fdp->fd_jdir != NULL) { 3336 vref(fdp->fd_jdir); 3337 data = fdp->fd_jdir; 3338 FILEDESC_SUNLOCK(fdp); 3339 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3340 FREAD, -1, -1, 0, 0, kif, req); 3341 FILEDESC_SLOCK(fdp); 3342 } 3343 for (i = 0; i < fdp->fd_nfiles; i++) { 3344 if ((fp = fdp->fd_ofiles[i]) == NULL) 3345 continue; 3346 data = NULL; 3347 fd_is_cap = 0; 3348 fd_cap_rights = 0; 3349 3350#ifdef CAPABILITIES 3351 /* 3352 * When reporting a capability, most fields will be from the 3353 * underlying object, but do mark as a capability and export 3354 * the capability rights mask. 3355 */ 3356 if (fp->f_type == DTYPE_CAPABILITY) { 3357 fd_is_cap = 1; 3358 fd_cap_rights = cap_rights(fp); 3359 (void)cap_funwrap(fp, 0, &fp); 3360 } 3361#else /* !CAPABILITIES */ 3362 KASSERT(fp->f_type != DTYPE_CAPABILITY, 3363 ("sysctl_kern_proc_filedesc: saw capability")); 3364#endif 3365 switch (fp->f_type) { 3366 case DTYPE_VNODE: 3367 type = KF_TYPE_VNODE; 3368 vref(fp->f_vnode); 3369 data = fp->f_vnode; 3370 break; 3371 3372 case DTYPE_SOCKET: 3373 type = KF_TYPE_SOCKET; 3374 data = fp->f_data; 3375 break; 3376 3377 case DTYPE_PIPE: 3378 type = KF_TYPE_PIPE; 3379 data = fp->f_data; 3380 break; 3381 3382 case DTYPE_FIFO: 3383 type = KF_TYPE_FIFO; 3384 vref(fp->f_vnode); 3385 data = fp->f_vnode; 3386 break; 3387 3388 case DTYPE_KQUEUE: 3389 type = KF_TYPE_KQUEUE; 3390 break; 3391 3392 case DTYPE_CRYPTO: 3393 type = KF_TYPE_CRYPTO; 3394 break; 3395 3396 case DTYPE_MQUEUE: 3397 type = KF_TYPE_MQUEUE; 3398 break; 3399 3400 case DTYPE_SHM: 3401 type = KF_TYPE_SHM; 3402 data = fp; 3403 break; 3404 3405 case DTYPE_SEM: 3406 type = KF_TYPE_SEM; 3407 break; 3408 3409 case DTYPE_PTS: 3410 type = KF_TYPE_PTS; 3411 data = fp->f_data; 3412 break; 3413 3414#ifdef PROCDESC 3415 case DTYPE_PROCDESC: 3416 type = KF_TYPE_PROCDESC; 3417 data = fp->f_data; 3418 break; 3419#endif 3420 3421 default: 3422 type = KF_TYPE_UNKNOWN; 3423 break; 3424 } 3425 refcnt = fp->f_count; 3426 fflags = fp->f_flag; 3427 offset = fp->f_offset; 3428 3429 /* 3430 * Create sysctl entry. 3431 * It is OK to drop the filedesc lock here as we will 3432 * re-validate and re-evaluate its properties when 3433 * the loop continues. 3434 */ 3435 oldidx = req->oldidx; 3436 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO) 3437 FILEDESC_SUNLOCK(fdp); 3438 error = export_fd_for_sysctl(data, type, i, fflags, refcnt, 3439 offset, fd_is_cap, fd_cap_rights, kif, req); 3440 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO) 3441 FILEDESC_SLOCK(fdp); 3442 if (error) { 3443 if (error == ENOMEM) { 3444 /* 3445 * The hack to keep the ABI of sysctl 3446 * kern.proc.filedesc intact, but not 3447 * to account a partially copied 3448 * kinfo_file into the oldidx. 3449 */ 3450 req->oldidx = oldidx; 3451 error = 0; 3452 } 3453 break; 3454 } 3455 } 3456 FILEDESC_SUNLOCK(fdp); 3457fail: 3458 if (fdp != NULL) 3459 fddrop(fdp); 3460 free(kif, M_TEMP); 3461 return (error); 3462} 3463 3464int 3465vntype_to_kinfo(int vtype) 3466{ 3467 struct { 3468 int vtype; 3469 int kf_vtype; 3470 } vtypes_table[] = { 3471 { VBAD, KF_VTYPE_VBAD }, 3472 { VBLK, KF_VTYPE_VBLK }, 3473 { VCHR, KF_VTYPE_VCHR }, 3474 { VDIR, KF_VTYPE_VDIR }, 3475 { VFIFO, KF_VTYPE_VFIFO }, 3476 { VLNK, KF_VTYPE_VLNK }, 3477 { VNON, KF_VTYPE_VNON }, 3478 { VREG, KF_VTYPE_VREG }, 3479 { VSOCK, KF_VTYPE_VSOCK } 3480 }; 3481#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3482 unsigned int i; 3483 3484 /* 3485 * Perform vtype translation. 3486 */ 3487 for (i = 0; i < NVTYPES; i++) 3488 if (vtypes_table[i].vtype == vtype) 3489 break; 3490 if (i < NVTYPES) 3491 return (vtypes_table[i].kf_vtype); 3492 3493 return (KF_VTYPE_UNKNOWN); 3494} 3495 3496static int 3497fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3498{ 3499 struct vattr va; 3500 char *fullpath, *freepath; 3501 int error, vfslocked; 3502 3503 if (vp == NULL) 3504 return (1); 3505 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3506 freepath = NULL; 3507 fullpath = "-"; 3508 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3509 if (error == 0) { 3510 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3511 } 3512 if (freepath != NULL) 3513 free(freepath, M_TEMP); 3514 3515 /* 3516 * Retrieve vnode attributes. 3517 */ 3518 va.va_fsid = VNOVAL; 3519 va.va_rdev = NODEV; 3520 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 3521 vn_lock(vp, LK_SHARED | LK_RETRY); 3522 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3523 VOP_UNLOCK(vp, 0); 3524 VFS_UNLOCK_GIANT(vfslocked); 3525 if (error != 0) 3526 return (error); 3527 if (va.va_fsid != VNOVAL) 3528 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3529 else 3530 kif->kf_un.kf_file.kf_file_fsid = 3531 vp->v_mount->mnt_stat.f_fsid.val[0]; 3532 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3533 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3534 kif->kf_un.kf_file.kf_file_size = va.va_size; 3535 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3536 return (0); 3537} 3538 3539static int 3540fill_socket_info(struct socket *so, struct kinfo_file *kif) 3541{ 3542 struct sockaddr *sa; 3543 struct inpcb *inpcb; 3544 struct unpcb *unpcb; 3545 int error; 3546 3547 if (so == NULL) 3548 return (1); 3549 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3550 kif->kf_sock_type = so->so_type; 3551 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3552 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3553 switch(kif->kf_sock_domain) { 3554 case AF_INET: 3555 case AF_INET6: 3556 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3557 if (so->so_pcb != NULL) { 3558 inpcb = (struct inpcb *)(so->so_pcb); 3559 kif->kf_un.kf_sock.kf_sock_inpcb = 3560 (uintptr_t)inpcb->inp_ppcb; 3561 } 3562 } 3563 break; 3564 case AF_UNIX: 3565 if (so->so_pcb != NULL) { 3566 unpcb = (struct unpcb *)(so->so_pcb); 3567 if (unpcb->unp_conn) { 3568 kif->kf_un.kf_sock.kf_sock_unpconn = 3569 (uintptr_t)unpcb->unp_conn; 3570 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3571 so->so_rcv.sb_state; 3572 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3573 so->so_snd.sb_state; 3574 } 3575 } 3576 break; 3577 } 3578 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3579 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3580 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3581 free(sa, M_SONAME); 3582 } 3583 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3584 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3585 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3586 free(sa, M_SONAME); 3587 } 3588 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3589 sizeof(kif->kf_path)); 3590 return (0); 3591} 3592 3593static int 3594fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3595{ 3596 3597 if (tp == NULL) 3598 return (1); 3599 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3600 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3601 return (0); 3602} 3603 3604static int 3605fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3606{ 3607 3608 if (pi == NULL) 3609 return (1); 3610 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3611 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3612 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3613 return (0); 3614} 3615 3616static int 3617fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3618{ 3619 3620 if (pdp == NULL) 3621 return (1); 3622 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3623 return (0); 3624} 3625 3626static int 3627fill_shm_info(struct file *fp, struct kinfo_file *kif) 3628{ 3629 struct thread *td; 3630 struct stat sb; 3631 3632 td = curthread; 3633 if (fp->f_data == NULL) 3634 return (1); 3635 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3636 return (1); 3637 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3638 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3639 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3640 return (0); 3641} 3642 3643static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD, 3644 sysctl_kern_proc_filedesc, "Process filedesc entries"); 3645 3646#ifdef DDB 3647/* 3648 * For the purposes of debugging, generate a human-readable string for the 3649 * file type. 3650 */ 3651static const char * 3652file_type_to_name(short type) 3653{ 3654 3655 switch (type) { 3656 case 0: 3657 return ("zero"); 3658 case DTYPE_VNODE: 3659 return ("vnod"); 3660 case DTYPE_SOCKET: 3661 return ("sock"); 3662 case DTYPE_PIPE: 3663 return ("pipe"); 3664 case DTYPE_FIFO: 3665 return ("fifo"); 3666 case DTYPE_KQUEUE: 3667 return ("kque"); 3668 case DTYPE_CRYPTO: 3669 return ("crpt"); 3670 case DTYPE_MQUEUE: 3671 return ("mque"); 3672 case DTYPE_SHM: 3673 return ("shm"); 3674 case DTYPE_SEM: 3675 return ("ksem"); 3676 default: 3677 return ("unkn"); 3678 } 3679} 3680 3681/* 3682 * For the purposes of debugging, identify a process (if any, perhaps one of 3683 * many) that references the passed file in its file descriptor array. Return 3684 * NULL if none. 3685 */ 3686static struct proc * 3687file_to_first_proc(struct file *fp) 3688{ 3689 struct filedesc *fdp; 3690 struct proc *p; 3691 int n; 3692 3693 FOREACH_PROC_IN_SYSTEM(p) { 3694 if (p->p_state == PRS_NEW) 3695 continue; 3696 fdp = p->p_fd; 3697 if (fdp == NULL) 3698 continue; 3699 for (n = 0; n < fdp->fd_nfiles; n++) { 3700 if (fp == fdp->fd_ofiles[n]) 3701 return (p); 3702 } 3703 } 3704 return (NULL); 3705} 3706 3707static void 3708db_print_file(struct file *fp, int header) 3709{ 3710 struct proc *p; 3711 3712 if (header) 3713 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3714 "File", "Type", "Data", "Flag", "GCFl", "Count", 3715 "MCount", "Vnode", "FPID", "FCmd"); 3716 p = file_to_first_proc(fp); 3717 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3718 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3719 0, fp->f_count, 0, fp->f_vnode, 3720 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3721} 3722 3723DB_SHOW_COMMAND(file, db_show_file) 3724{ 3725 struct file *fp; 3726 3727 if (!have_addr) { 3728 db_printf("usage: show file <addr>\n"); 3729 return; 3730 } 3731 fp = (struct file *)addr; 3732 db_print_file(fp, 1); 3733} 3734 3735DB_SHOW_COMMAND(files, db_show_files) 3736{ 3737 struct filedesc *fdp; 3738 struct file *fp; 3739 struct proc *p; 3740 int header; 3741 int n; 3742 3743 header = 1; 3744 FOREACH_PROC_IN_SYSTEM(p) { 3745 if (p->p_state == PRS_NEW) 3746 continue; 3747 if ((fdp = p->p_fd) == NULL) 3748 continue; 3749 for (n = 0; n < fdp->fd_nfiles; ++n) { 3750 if ((fp = fdp->fd_ofiles[n]) == NULL) 3751 continue; 3752 db_print_file(fp, header); 3753 header = 0; 3754 } 3755 } 3756} 3757#endif 3758 3759SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3760 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3761 3762SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3763 &maxfiles, 0, "Maximum number of files"); 3764 3765SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3766 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3767 3768/* ARGSUSED*/ 3769static void 3770filelistinit(void *dummy) 3771{ 3772 3773 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3774 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3775 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3776 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3777} 3778SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3779 3780/*-------------------------------------------------------------------*/ 3781 3782static int 3783badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3784 int flags, struct thread *td) 3785{ 3786 3787 return (EBADF); 3788} 3789 3790static int 3791badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3792 struct thread *td) 3793{ 3794 3795 return (EINVAL); 3796} 3797 3798static int 3799badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3800 struct thread *td) 3801{ 3802 3803 return (EBADF); 3804} 3805 3806static int 3807badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3808 struct thread *td) 3809{ 3810 3811 return (0); 3812} 3813 3814static int 3815badfo_kqfilter(struct file *fp, struct knote *kn) 3816{ 3817 3818 return (EBADF); 3819} 3820 3821static int 3822badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3823 struct thread *td) 3824{ 3825 3826 return (EBADF); 3827} 3828 3829static int 3830badfo_close(struct file *fp, struct thread *td) 3831{ 3832 3833 return (EBADF); 3834} 3835 3836static int 3837badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3838 struct thread *td) 3839{ 3840 3841 return (EBADF); 3842} 3843 3844static int 3845badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3846 struct thread *td) 3847{ 3848 3849 return (EBADF); 3850} 3851 3852struct fileops badfileops = { 3853 .fo_read = badfo_readwrite, 3854 .fo_write = badfo_readwrite, 3855 .fo_truncate = badfo_truncate, 3856 .fo_ioctl = badfo_ioctl, 3857 .fo_poll = badfo_poll, 3858 .fo_kqfilter = badfo_kqfilter, 3859 .fo_stat = badfo_stat, 3860 .fo_close = badfo_close, 3861 .fo_chmod = badfo_chmod, 3862 .fo_chown = badfo_chown, 3863}; 3864 3865int 3866invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3867 struct thread *td) 3868{ 3869 3870 return (EINVAL); 3871} 3872 3873int 3874invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3875 struct thread *td) 3876{ 3877 3878 return (EINVAL); 3879} 3880 3881/*-------------------------------------------------------------------*/ 3882 3883/* 3884 * File Descriptor pseudo-device driver (/dev/fd/). 3885 * 3886 * Opening minor device N dup()s the file (if any) connected to file 3887 * descriptor N belonging to the calling process. Note that this driver 3888 * consists of only the ``open()'' routine, because all subsequent 3889 * references to this file will be direct to the other driver. 3890 * 3891 * XXX: we could give this one a cloning event handler if necessary. 3892 */ 3893 3894/* ARGSUSED */ 3895static int 3896fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3897{ 3898 3899 /* 3900 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3901 * the file descriptor being sought for duplication. The error 3902 * return ensures that the vnode for this device will be released 3903 * by vn_open. Open will detect this special error and take the 3904 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3905 * will simply report the error. 3906 */ 3907 td->td_dupfd = dev2unit(dev); 3908 return (ENODEV); 3909} 3910 3911static struct cdevsw fildesc_cdevsw = { 3912 .d_version = D_VERSION, 3913 .d_open = fdopen, 3914 .d_name = "FD", 3915}; 3916 3917static void 3918fildesc_drvinit(void *unused) 3919{ 3920 struct cdev *dev; 3921 3922 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3923 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3924 make_dev_alias(dev, "stdin"); 3925 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3926 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3927 make_dev_alias(dev, "stdout"); 3928 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3929 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3930 make_dev_alias(dev, "stderr"); 3931} 3932 3933SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3934