kern_descrip.c revision 263958
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: stable/10/sys/kern/kern_descrip.c 263958 2014-03-31 02:17:58Z mjg $"); 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/ksem.h> 59#include <sys/limits.h> 60#include <sys/lock.h> 61#include <sys/malloc.h> 62#include <sys/mman.h> 63#include <sys/mount.h> 64#include <sys/mqueue.h> 65#include <sys/mutex.h> 66#include <sys/namei.h> 67#include <sys/selinfo.h> 68#include <sys/pipe.h> 69#include <sys/priv.h> 70#include <sys/proc.h> 71#include <sys/procdesc.h> 72#include <sys/protosw.h> 73#include <sys/racct.h> 74#include <sys/resourcevar.h> 75#include <sys/sbuf.h> 76#include <sys/signalvar.h> 77#include <sys/socketvar.h> 78#include <sys/stat.h> 79#include <sys/sx.h> 80#include <sys/syscallsubr.h> 81#include <sys/sysctl.h> 82#include <sys/sysproto.h> 83#include <sys/tty.h> 84#include <sys/unistd.h> 85#include <sys/un.h> 86#include <sys/unpcb.h> 87#include <sys/user.h> 88#include <sys/vnode.h> 89#ifdef KTRACE 90#include <sys/ktrace.h> 91#endif 92 93#include <net/vnet.h> 94 95#include <netinet/in.h> 96#include <netinet/in_pcb.h> 97 98#include <security/audit/audit.h> 99 100#include <vm/uma.h> 101#include <vm/vm.h> 102 103#include <ddb/ddb.h> 104 105static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 106static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 107 "file desc to leader structures"); 108static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 109MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 110 111MALLOC_DECLARE(M_FADVISE); 112 113static uma_zone_t file_zone; 114 115void (*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value); 116 117static int closefp(struct filedesc *fdp, int fd, struct file *fp, 118 struct thread *td, int holdleaders); 119static int fd_first_free(struct filedesc *fdp, int low, int size); 120static int fd_last_used(struct filedesc *fdp, int size); 121static void fdgrowtable(struct filedesc *fdp, int nfd); 122static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 123static void fdunused(struct filedesc *fdp, int fd); 124static void fdused(struct filedesc *fdp, int fd); 125static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif); 126static int fill_procdesc_info(struct procdesc *pdp, 127 struct kinfo_file *kif); 128static int fill_pts_info(struct tty *tp, struct kinfo_file *kif); 129static int fill_sem_info(struct file *fp, struct kinfo_file *kif); 130static int fill_shm_info(struct file *fp, struct kinfo_file *kif); 131static int fill_socket_info(struct socket *so, struct kinfo_file *kif); 132static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif); 133static int getmaxfd(struct proc *p); 134 135/* 136 * Each process has: 137 * 138 * - An array of open file descriptors (fd_ofiles) 139 * - An array of file flags (fd_ofileflags) 140 * - A bitmap recording which descriptors are in use (fd_map) 141 * 142 * A process starts out with NDFILE descriptors. The value of NDFILE has 143 * been selected based the historical limit of 20 open files, and an 144 * assumption that the majority of processes, especially short-lived 145 * processes like shells, will never need more. 146 * 147 * If this initial allocation is exhausted, a larger descriptor table and 148 * map are allocated dynamically, and the pointers in the process's struct 149 * filedesc are updated to point to those. This is repeated every time 150 * the process runs out of file descriptors (provided it hasn't hit its 151 * resource limit). 152 * 153 * Since threads may hold references to individual descriptor table 154 * entries, the tables are never freed. Instead, they are placed on a 155 * linked list and freed only when the struct filedesc is released. 156 */ 157#define NDFILE 20 158#define NDSLOTSIZE sizeof(NDSLOTTYPE) 159#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 160#define NDSLOT(x) ((x) / NDENTRIES) 161#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 162#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 163 164/* 165 * SLIST entry used to keep track of ofiles which must be reclaimed when 166 * the process exits. 167 */ 168struct freetable { 169 struct filedescent *ft_table; 170 SLIST_ENTRY(freetable) ft_next; 171}; 172 173/* 174 * Initial allocation: a filedesc structure + the head of SLIST used to 175 * keep track of old ofiles + enough space for NDFILE descriptors. 176 */ 177struct filedesc0 { 178 struct filedesc fd_fd; 179 SLIST_HEAD(, freetable) fd_free; 180 struct filedescent fd_dfiles[NDFILE]; 181 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 182}; 183 184/* 185 * Descriptor management. 186 */ 187volatile int openfiles; /* actual number of open files */ 188struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 189void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 190 191/* A mutex to protect the association between a proc and filedesc. */ 192static struct mtx fdesc_mtx; 193 194/* 195 * If low >= size, just return low. Otherwise find the first zero bit in the 196 * given bitmap, starting at low and not exceeding size - 1. Return size if 197 * not found. 198 */ 199static int 200fd_first_free(struct filedesc *fdp, int low, int size) 201{ 202 NDSLOTTYPE *map = fdp->fd_map; 203 NDSLOTTYPE mask; 204 int off, maxoff; 205 206 if (low >= size) 207 return (low); 208 209 off = NDSLOT(low); 210 if (low % NDENTRIES) { 211 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 212 if ((mask &= ~map[off]) != 0UL) 213 return (off * NDENTRIES + ffsl(mask) - 1); 214 ++off; 215 } 216 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 217 if (map[off] != ~0UL) 218 return (off * NDENTRIES + ffsl(~map[off]) - 1); 219 return (size); 220} 221 222/* 223 * Find the highest non-zero bit in the given bitmap, starting at 0 and 224 * not exceeding size - 1. Return -1 if not found. 225 */ 226static int 227fd_last_used(struct filedesc *fdp, int size) 228{ 229 NDSLOTTYPE *map = fdp->fd_map; 230 NDSLOTTYPE mask; 231 int off, minoff; 232 233 off = NDSLOT(size); 234 if (size % NDENTRIES) { 235 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 236 if ((mask &= map[off]) != 0) 237 return (off * NDENTRIES + flsl(mask) - 1); 238 --off; 239 } 240 for (minoff = NDSLOT(0); off >= minoff; --off) 241 if (map[off] != 0) 242 return (off * NDENTRIES + flsl(map[off]) - 1); 243 return (-1); 244} 245 246static int 247fdisused(struct filedesc *fdp, int fd) 248{ 249 250 FILEDESC_LOCK_ASSERT(fdp); 251 252 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 253 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 254 255 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 256} 257 258/* 259 * Mark a file descriptor as used. 260 */ 261static void 262fdused(struct filedesc *fdp, int fd) 263{ 264 265 FILEDESC_XLOCK_ASSERT(fdp); 266 267 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 268 269 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 270 if (fd > fdp->fd_lastfile) 271 fdp->fd_lastfile = fd; 272 if (fd == fdp->fd_freefile) 273 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 274} 275 276/* 277 * Mark a file descriptor as unused. 278 */ 279static void 280fdunused(struct filedesc *fdp, int fd) 281{ 282 283 FILEDESC_XLOCK_ASSERT(fdp); 284 285 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 286 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 287 ("fd=%d is still in use", fd)); 288 289 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 290 if (fd < fdp->fd_freefile) 291 fdp->fd_freefile = fd; 292 if (fd == fdp->fd_lastfile) 293 fdp->fd_lastfile = fd_last_used(fdp, fd); 294} 295 296/* 297 * Free a file descriptor. 298 */ 299static inline void 300fdfree(struct filedesc *fdp, int fd) 301{ 302 struct filedescent *fde; 303 304 fde = &fdp->fd_ofiles[fd]; 305 filecaps_free(&fde->fde_caps); 306 bzero(fde, sizeof(*fde)); 307 fdunused(fdp, fd); 308} 309 310/* 311 * System calls on descriptors. 312 */ 313#ifndef _SYS_SYSPROTO_H_ 314struct getdtablesize_args { 315 int dummy; 316}; 317#endif 318/* ARGSUSED */ 319int 320sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 321{ 322 struct proc *p = td->td_proc; 323 uint64_t lim; 324 325 PROC_LOCK(p); 326 td->td_retval[0] = 327 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 328 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 329 PROC_UNLOCK(p); 330 if (lim < td->td_retval[0]) 331 td->td_retval[0] = lim; 332 return (0); 333} 334 335/* 336 * Duplicate a file descriptor to a particular value. 337 * 338 * Note: keep in mind that a potential race condition exists when closing 339 * descriptors from a shared descriptor table (via rfork). 340 */ 341#ifndef _SYS_SYSPROTO_H_ 342struct dup2_args { 343 u_int from; 344 u_int to; 345}; 346#endif 347/* ARGSUSED */ 348int 349sys_dup2(struct thread *td, struct dup2_args *uap) 350{ 351 352 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 353 td->td_retval)); 354} 355 356/* 357 * Duplicate a file descriptor. 358 */ 359#ifndef _SYS_SYSPROTO_H_ 360struct dup_args { 361 u_int fd; 362}; 363#endif 364/* ARGSUSED */ 365int 366sys_dup(struct thread *td, struct dup_args *uap) 367{ 368 369 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 370} 371 372/* 373 * The file control system call. 374 */ 375#ifndef _SYS_SYSPROTO_H_ 376struct fcntl_args { 377 int fd; 378 int cmd; 379 long arg; 380}; 381#endif 382/* ARGSUSED */ 383int 384sys_fcntl(struct thread *td, struct fcntl_args *uap) 385{ 386 struct flock fl; 387 struct __oflock ofl; 388 intptr_t arg; 389 int error; 390 int cmd; 391 392 error = 0; 393 cmd = uap->cmd; 394 switch (uap->cmd) { 395 case F_OGETLK: 396 case F_OSETLK: 397 case F_OSETLKW: 398 /* 399 * Convert old flock structure to new. 400 */ 401 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 402 fl.l_start = ofl.l_start; 403 fl.l_len = ofl.l_len; 404 fl.l_pid = ofl.l_pid; 405 fl.l_type = ofl.l_type; 406 fl.l_whence = ofl.l_whence; 407 fl.l_sysid = 0; 408 409 switch (uap->cmd) { 410 case F_OGETLK: 411 cmd = F_GETLK; 412 break; 413 case F_OSETLK: 414 cmd = F_SETLK; 415 break; 416 case F_OSETLKW: 417 cmd = F_SETLKW; 418 break; 419 } 420 arg = (intptr_t)&fl; 421 break; 422 case F_GETLK: 423 case F_SETLK: 424 case F_SETLKW: 425 case F_SETLK_REMOTE: 426 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 427 arg = (intptr_t)&fl; 428 break; 429 default: 430 arg = uap->arg; 431 break; 432 } 433 if (error) 434 return (error); 435 error = kern_fcntl(td, uap->fd, cmd, arg); 436 if (error) 437 return (error); 438 if (uap->cmd == F_OGETLK) { 439 ofl.l_start = fl.l_start; 440 ofl.l_len = fl.l_len; 441 ofl.l_pid = fl.l_pid; 442 ofl.l_type = fl.l_type; 443 ofl.l_whence = fl.l_whence; 444 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 445 } else if (uap->cmd == F_GETLK) { 446 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 447 } 448 return (error); 449} 450 451int 452kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 453{ 454 struct filedesc *fdp; 455 struct flock *flp; 456 struct file *fp, *fp2; 457 struct filedescent *fde; 458 struct proc *p; 459 struct vnode *vp; 460 cap_rights_t rights; 461 int error, flg, tmp; 462 u_int old, new; 463 uint64_t bsize; 464 off_t foffset; 465 466 error = 0; 467 flg = F_POSIX; 468 p = td->td_proc; 469 fdp = p->p_fd; 470 471 switch (cmd) { 472 case F_DUPFD: 473 tmp = arg; 474 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 475 break; 476 477 case F_DUPFD_CLOEXEC: 478 tmp = arg; 479 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp, 480 td->td_retval); 481 break; 482 483 case F_DUP2FD: 484 tmp = arg; 485 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 486 break; 487 488 case F_DUP2FD_CLOEXEC: 489 tmp = arg; 490 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp, 491 td->td_retval); 492 break; 493 494 case F_GETFD: 495 FILEDESC_SLOCK(fdp); 496 if ((fp = fget_locked(fdp, fd)) == NULL) { 497 FILEDESC_SUNLOCK(fdp); 498 error = EBADF; 499 break; 500 } 501 fde = &fdp->fd_ofiles[fd]; 502 td->td_retval[0] = 503 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 504 FILEDESC_SUNLOCK(fdp); 505 break; 506 507 case F_SETFD: 508 FILEDESC_XLOCK(fdp); 509 if ((fp = fget_locked(fdp, fd)) == NULL) { 510 FILEDESC_XUNLOCK(fdp); 511 error = EBADF; 512 break; 513 } 514 fde = &fdp->fd_ofiles[fd]; 515 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 516 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 517 FILEDESC_XUNLOCK(fdp); 518 break; 519 520 case F_GETFL: 521 error = fget_unlocked(fdp, fd, 522 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL); 523 if (error != 0) 524 break; 525 td->td_retval[0] = OFLAGS(fp->f_flag); 526 fdrop(fp, td); 527 break; 528 529 case F_SETFL: 530 error = fget_unlocked(fdp, fd, 531 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL); 532 if (error != 0) 533 break; 534 do { 535 tmp = flg = fp->f_flag; 536 tmp &= ~FCNTLFLAGS; 537 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 538 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 539 tmp = fp->f_flag & FNONBLOCK; 540 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 541 if (error != 0) { 542 fdrop(fp, td); 543 break; 544 } 545 tmp = fp->f_flag & FASYNC; 546 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 547 if (error == 0) { 548 fdrop(fp, td); 549 break; 550 } 551 atomic_clear_int(&fp->f_flag, FNONBLOCK); 552 tmp = 0; 553 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 554 fdrop(fp, td); 555 break; 556 557 case F_GETOWN: 558 error = fget_unlocked(fdp, fd, 559 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL); 560 if (error != 0) 561 break; 562 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 563 if (error == 0) 564 td->td_retval[0] = tmp; 565 fdrop(fp, td); 566 break; 567 568 case F_SETOWN: 569 error = fget_unlocked(fdp, fd, 570 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL); 571 if (error != 0) 572 break; 573 tmp = arg; 574 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 575 fdrop(fp, td); 576 break; 577 578 case F_SETLK_REMOTE: 579 error = priv_check(td, PRIV_NFS_LOCKD); 580 if (error) 581 return (error); 582 flg = F_REMOTE; 583 goto do_setlk; 584 585 case F_SETLKW: 586 flg |= F_WAIT; 587 /* FALLTHROUGH F_SETLK */ 588 589 case F_SETLK: 590 do_setlk: 591 cap_rights_init(&rights, CAP_FLOCK); 592 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL); 593 if (error != 0) 594 break; 595 if (fp->f_type != DTYPE_VNODE) { 596 error = EBADF; 597 fdrop(fp, td); 598 break; 599 } 600 601 flp = (struct flock *)arg; 602 if (flp->l_whence == SEEK_CUR) { 603 foffset = foffset_get(fp); 604 if (foffset < 0 || 605 (flp->l_start > 0 && 606 foffset > OFF_MAX - flp->l_start)) { 607 FILEDESC_SUNLOCK(fdp); 608 error = EOVERFLOW; 609 fdrop(fp, td); 610 break; 611 } 612 flp->l_start += foffset; 613 } 614 615 vp = fp->f_vnode; 616 switch (flp->l_type) { 617 case F_RDLCK: 618 if ((fp->f_flag & FREAD) == 0) { 619 error = EBADF; 620 break; 621 } 622 PROC_LOCK(p->p_leader); 623 p->p_leader->p_flag |= P_ADVLOCK; 624 PROC_UNLOCK(p->p_leader); 625 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 626 flp, flg); 627 break; 628 case F_WRLCK: 629 if ((fp->f_flag & FWRITE) == 0) { 630 error = EBADF; 631 break; 632 } 633 PROC_LOCK(p->p_leader); 634 p->p_leader->p_flag |= P_ADVLOCK; 635 PROC_UNLOCK(p->p_leader); 636 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 637 flp, flg); 638 break; 639 case F_UNLCK: 640 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 641 flp, flg); 642 break; 643 case F_UNLCKSYS: 644 /* 645 * Temporary api for testing remote lock 646 * infrastructure. 647 */ 648 if (flg != F_REMOTE) { 649 error = EINVAL; 650 break; 651 } 652 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 653 F_UNLCKSYS, flp, flg); 654 break; 655 default: 656 error = EINVAL; 657 break; 658 } 659 if (error != 0 || flp->l_type == F_UNLCK || 660 flp->l_type == F_UNLCKSYS) { 661 fdrop(fp, td); 662 break; 663 } 664 665 /* 666 * Check for a race with close. 667 * 668 * The vnode is now advisory locked (or unlocked, but this case 669 * is not really important) as the caller requested. 670 * We had to drop the filedesc lock, so we need to recheck if 671 * the descriptor is still valid, because if it was closed 672 * in the meantime we need to remove advisory lock from the 673 * vnode - close on any descriptor leading to an advisory 674 * locked vnode, removes that lock. 675 * We will return 0 on purpose in that case, as the result of 676 * successful advisory lock might have been externally visible 677 * already. This is fine - effectively we pretend to the caller 678 * that the closing thread was a bit slower and that the 679 * advisory lock succeeded before the close. 680 */ 681 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL); 682 if (error != 0) { 683 fdrop(fp, td); 684 break; 685 } 686 if (fp != fp2) { 687 flp->l_whence = SEEK_SET; 688 flp->l_start = 0; 689 flp->l_len = 0; 690 flp->l_type = F_UNLCK; 691 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 692 F_UNLCK, flp, F_POSIX); 693 } 694 fdrop(fp, td); 695 fdrop(fp2, td); 696 break; 697 698 case F_GETLK: 699 error = fget_unlocked(fdp, fd, 700 cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL); 701 if (error != 0) 702 break; 703 if (fp->f_type != DTYPE_VNODE) { 704 error = EBADF; 705 fdrop(fp, td); 706 break; 707 } 708 flp = (struct flock *)arg; 709 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 710 flp->l_type != F_UNLCK) { 711 error = EINVAL; 712 fdrop(fp, td); 713 break; 714 } 715 if (flp->l_whence == SEEK_CUR) { 716 foffset = foffset_get(fp); 717 if ((flp->l_start > 0 && 718 foffset > OFF_MAX - flp->l_start) || 719 (flp->l_start < 0 && 720 foffset < OFF_MIN - flp->l_start)) { 721 FILEDESC_SUNLOCK(fdp); 722 error = EOVERFLOW; 723 fdrop(fp, td); 724 break; 725 } 726 flp->l_start += foffset; 727 } 728 vp = fp->f_vnode; 729 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 730 F_POSIX); 731 fdrop(fp, td); 732 break; 733 734 case F_RDAHEAD: 735 arg = arg ? 128 * 1024: 0; 736 /* FALLTHROUGH */ 737 case F_READAHEAD: 738 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL); 739 if (error != 0) 740 break; 741 if (fp->f_type != DTYPE_VNODE) { 742 fdrop(fp, td); 743 error = EBADF; 744 break; 745 } 746 if (arg >= 0) { 747 vp = fp->f_vnode; 748 error = vn_lock(vp, LK_SHARED); 749 if (error != 0) { 750 fdrop(fp, td); 751 break; 752 } 753 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 754 VOP_UNLOCK(vp, 0); 755 fp->f_seqcount = (arg + bsize - 1) / bsize; 756 do { 757 new = old = fp->f_flag; 758 new |= FRDAHEAD; 759 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 760 } else { 761 do { 762 new = old = fp->f_flag; 763 new &= ~FRDAHEAD; 764 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 765 } 766 fdrop(fp, td); 767 break; 768 769 default: 770 error = EINVAL; 771 break; 772 } 773 return (error); 774} 775 776static int 777getmaxfd(struct proc *p) 778{ 779 int maxfd; 780 781 PROC_LOCK(p); 782 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 783 PROC_UNLOCK(p); 784 785 return (maxfd); 786} 787 788/* 789 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 790 */ 791int 792do_dup(struct thread *td, int flags, int old, int new, 793 register_t *retval) 794{ 795 struct filedesc *fdp; 796 struct filedescent *oldfde, *newfde; 797 struct proc *p; 798 struct file *fp; 799 struct file *delfp; 800 int error, maxfd; 801 802 p = td->td_proc; 803 fdp = p->p_fd; 804 805 /* 806 * Verify we have a valid descriptor to dup from and possibly to 807 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 808 * return EINVAL when the new descriptor is out of bounds. 809 */ 810 if (old < 0) 811 return (EBADF); 812 if (new < 0) 813 return (flags & DUP_FCNTL ? EINVAL : EBADF); 814 maxfd = getmaxfd(p); 815 if (new >= maxfd) 816 return (flags & DUP_FCNTL ? EINVAL : EBADF); 817 818 FILEDESC_XLOCK(fdp); 819 if (fget_locked(fdp, old) == NULL) { 820 FILEDESC_XUNLOCK(fdp); 821 return (EBADF); 822 } 823 oldfde = &fdp->fd_ofiles[old]; 824 if (flags & DUP_FIXED && old == new) { 825 *retval = new; 826 if (flags & DUP_CLOEXEC) 827 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 828 FILEDESC_XUNLOCK(fdp); 829 return (0); 830 } 831 fp = oldfde->fde_file; 832 fhold(fp); 833 834 /* 835 * If the caller specified a file descriptor, make sure the file 836 * table is large enough to hold it, and grab it. Otherwise, just 837 * allocate a new descriptor the usual way. 838 */ 839 if (flags & DUP_FIXED) { 840 if (new >= fdp->fd_nfiles) { 841 /* 842 * The resource limits are here instead of e.g. 843 * fdalloc(), because the file descriptor table may be 844 * shared between processes, so we can't really use 845 * racct_add()/racct_sub(). Instead of counting the 846 * number of actually allocated descriptors, just put 847 * the limit on the size of the file descriptor table. 848 */ 849#ifdef RACCT 850 PROC_LOCK(p); 851 error = racct_set(p, RACCT_NOFILE, new + 1); 852 PROC_UNLOCK(p); 853 if (error != 0) { 854 FILEDESC_XUNLOCK(fdp); 855 fdrop(fp, td); 856 return (EMFILE); 857 } 858#endif 859 fdgrowtable_exp(fdp, new + 1); 860 oldfde = &fdp->fd_ofiles[old]; 861 } 862 newfde = &fdp->fd_ofiles[new]; 863 if (newfde->fde_file == NULL) 864 fdused(fdp, new); 865 } else { 866 if ((error = fdalloc(td, new, &new)) != 0) { 867 FILEDESC_XUNLOCK(fdp); 868 fdrop(fp, td); 869 return (error); 870 } 871 newfde = &fdp->fd_ofiles[new]; 872 } 873 874 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 875 KASSERT(old != new, ("new fd is same as old")); 876 877 delfp = newfde->fde_file; 878 879 /* 880 * Duplicate the source descriptor. 881 */ 882 filecaps_free(&newfde->fde_caps); 883 *newfde = *oldfde; 884 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 885 if ((flags & DUP_CLOEXEC) != 0) 886 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 887 else 888 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 889 if (new > fdp->fd_lastfile) 890 fdp->fd_lastfile = new; 891 *retval = new; 892 893 if (delfp != NULL) { 894 (void) closefp(fdp, new, delfp, td, 1); 895 /* closefp() drops the FILEDESC lock for us. */ 896 } else { 897 FILEDESC_XUNLOCK(fdp); 898 } 899 900 return (0); 901} 902 903/* 904 * If sigio is on the list associated with a process or process group, 905 * disable signalling from the device, remove sigio from the list and 906 * free sigio. 907 */ 908void 909funsetown(struct sigio **sigiop) 910{ 911 struct sigio *sigio; 912 913 SIGIO_LOCK(); 914 sigio = *sigiop; 915 if (sigio == NULL) { 916 SIGIO_UNLOCK(); 917 return; 918 } 919 *(sigio->sio_myref) = NULL; 920 if ((sigio)->sio_pgid < 0) { 921 struct pgrp *pg = (sigio)->sio_pgrp; 922 PGRP_LOCK(pg); 923 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 924 sigio, sio_pgsigio); 925 PGRP_UNLOCK(pg); 926 } else { 927 struct proc *p = (sigio)->sio_proc; 928 PROC_LOCK(p); 929 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 930 sigio, sio_pgsigio); 931 PROC_UNLOCK(p); 932 } 933 SIGIO_UNLOCK(); 934 crfree(sigio->sio_ucred); 935 free(sigio, M_SIGIO); 936} 937 938/* 939 * Free a list of sigio structures. 940 * We only need to lock the SIGIO_LOCK because we have made ourselves 941 * inaccessible to callers of fsetown and therefore do not need to lock 942 * the proc or pgrp struct for the list manipulation. 943 */ 944void 945funsetownlst(struct sigiolst *sigiolst) 946{ 947 struct proc *p; 948 struct pgrp *pg; 949 struct sigio *sigio; 950 951 sigio = SLIST_FIRST(sigiolst); 952 if (sigio == NULL) 953 return; 954 p = NULL; 955 pg = NULL; 956 957 /* 958 * Every entry of the list should belong 959 * to a single proc or pgrp. 960 */ 961 if (sigio->sio_pgid < 0) { 962 pg = sigio->sio_pgrp; 963 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 964 } else /* if (sigio->sio_pgid > 0) */ { 965 p = sigio->sio_proc; 966 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 967 } 968 969 SIGIO_LOCK(); 970 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 971 *(sigio->sio_myref) = NULL; 972 if (pg != NULL) { 973 KASSERT(sigio->sio_pgid < 0, 974 ("Proc sigio in pgrp sigio list")); 975 KASSERT(sigio->sio_pgrp == pg, 976 ("Bogus pgrp in sigio list")); 977 PGRP_LOCK(pg); 978 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 979 sio_pgsigio); 980 PGRP_UNLOCK(pg); 981 } else /* if (p != NULL) */ { 982 KASSERT(sigio->sio_pgid > 0, 983 ("Pgrp sigio in proc sigio list")); 984 KASSERT(sigio->sio_proc == p, 985 ("Bogus proc in sigio list")); 986 PROC_LOCK(p); 987 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 988 sio_pgsigio); 989 PROC_UNLOCK(p); 990 } 991 SIGIO_UNLOCK(); 992 crfree(sigio->sio_ucred); 993 free(sigio, M_SIGIO); 994 SIGIO_LOCK(); 995 } 996 SIGIO_UNLOCK(); 997} 998 999/* 1000 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1001 * 1002 * After permission checking, add a sigio structure to the sigio list for 1003 * the process or process group. 1004 */ 1005int 1006fsetown(pid_t pgid, struct sigio **sigiop) 1007{ 1008 struct proc *proc; 1009 struct pgrp *pgrp; 1010 struct sigio *sigio; 1011 int ret; 1012 1013 if (pgid == 0) { 1014 funsetown(sigiop); 1015 return (0); 1016 } 1017 1018 ret = 0; 1019 1020 /* Allocate and fill in the new sigio out of locks. */ 1021 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1022 sigio->sio_pgid = pgid; 1023 sigio->sio_ucred = crhold(curthread->td_ucred); 1024 sigio->sio_myref = sigiop; 1025 1026 sx_slock(&proctree_lock); 1027 if (pgid > 0) { 1028 proc = pfind(pgid); 1029 if (proc == NULL) { 1030 ret = ESRCH; 1031 goto fail; 1032 } 1033 1034 /* 1035 * Policy - Don't allow a process to FSETOWN a process 1036 * in another session. 1037 * 1038 * Remove this test to allow maximum flexibility or 1039 * restrict FSETOWN to the current process or process 1040 * group for maximum safety. 1041 */ 1042 PROC_UNLOCK(proc); 1043 if (proc->p_session != curthread->td_proc->p_session) { 1044 ret = EPERM; 1045 goto fail; 1046 } 1047 1048 pgrp = NULL; 1049 } else /* if (pgid < 0) */ { 1050 pgrp = pgfind(-pgid); 1051 if (pgrp == NULL) { 1052 ret = ESRCH; 1053 goto fail; 1054 } 1055 PGRP_UNLOCK(pgrp); 1056 1057 /* 1058 * Policy - Don't allow a process to FSETOWN a process 1059 * in another session. 1060 * 1061 * Remove this test to allow maximum flexibility or 1062 * restrict FSETOWN to the current process or process 1063 * group for maximum safety. 1064 */ 1065 if (pgrp->pg_session != curthread->td_proc->p_session) { 1066 ret = EPERM; 1067 goto fail; 1068 } 1069 1070 proc = NULL; 1071 } 1072 funsetown(sigiop); 1073 if (pgid > 0) { 1074 PROC_LOCK(proc); 1075 /* 1076 * Since funsetownlst() is called without the proctree 1077 * locked, we need to check for P_WEXIT. 1078 * XXX: is ESRCH correct? 1079 */ 1080 if ((proc->p_flag & P_WEXIT) != 0) { 1081 PROC_UNLOCK(proc); 1082 ret = ESRCH; 1083 goto fail; 1084 } 1085 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1086 sigio->sio_proc = proc; 1087 PROC_UNLOCK(proc); 1088 } else { 1089 PGRP_LOCK(pgrp); 1090 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1091 sigio->sio_pgrp = pgrp; 1092 PGRP_UNLOCK(pgrp); 1093 } 1094 sx_sunlock(&proctree_lock); 1095 SIGIO_LOCK(); 1096 *sigiop = sigio; 1097 SIGIO_UNLOCK(); 1098 return (0); 1099 1100fail: 1101 sx_sunlock(&proctree_lock); 1102 crfree(sigio->sio_ucred); 1103 free(sigio, M_SIGIO); 1104 return (ret); 1105} 1106 1107/* 1108 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1109 */ 1110pid_t 1111fgetown(sigiop) 1112 struct sigio **sigiop; 1113{ 1114 pid_t pgid; 1115 1116 SIGIO_LOCK(); 1117 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1118 SIGIO_UNLOCK(); 1119 return (pgid); 1120} 1121 1122/* 1123 * Function drops the filedesc lock on return. 1124 */ 1125static int 1126closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1127 int holdleaders) 1128{ 1129 int error; 1130 1131 FILEDESC_XLOCK_ASSERT(fdp); 1132 1133 if (holdleaders) { 1134 if (td->td_proc->p_fdtol != NULL) { 1135 /* 1136 * Ask fdfree() to sleep to ensure that all relevant 1137 * process leaders can be traversed in closef(). 1138 */ 1139 fdp->fd_holdleaderscount++; 1140 } else { 1141 holdleaders = 0; 1142 } 1143 } 1144 1145 /* 1146 * We now hold the fp reference that used to be owned by the 1147 * descriptor array. We have to unlock the FILEDESC *AFTER* 1148 * knote_fdclose to prevent a race of the fd getting opened, a knote 1149 * added, and deleteing a knote for the new fd. 1150 */ 1151 knote_fdclose(td, fd); 1152 1153 /* 1154 * We need to notify mqueue if the object is of type mqueue. 1155 */ 1156 if (fp->f_type == DTYPE_MQUEUE) 1157 mq_fdclose(td, fd, fp); 1158 FILEDESC_XUNLOCK(fdp); 1159 1160 error = closef(fp, td); 1161 if (holdleaders) { 1162 FILEDESC_XLOCK(fdp); 1163 fdp->fd_holdleaderscount--; 1164 if (fdp->fd_holdleaderscount == 0 && 1165 fdp->fd_holdleaderswakeup != 0) { 1166 fdp->fd_holdleaderswakeup = 0; 1167 wakeup(&fdp->fd_holdleaderscount); 1168 } 1169 FILEDESC_XUNLOCK(fdp); 1170 } 1171 return (error); 1172} 1173 1174/* 1175 * Close a file descriptor. 1176 */ 1177#ifndef _SYS_SYSPROTO_H_ 1178struct close_args { 1179 int fd; 1180}; 1181#endif 1182/* ARGSUSED */ 1183int 1184sys_close(td, uap) 1185 struct thread *td; 1186 struct close_args *uap; 1187{ 1188 1189 return (kern_close(td, uap->fd)); 1190} 1191 1192int 1193kern_close(td, fd) 1194 struct thread *td; 1195 int fd; 1196{ 1197 struct filedesc *fdp; 1198 struct file *fp; 1199 1200 fdp = td->td_proc->p_fd; 1201 1202 AUDIT_SYSCLOSE(td, fd); 1203 1204 FILEDESC_XLOCK(fdp); 1205 if ((fp = fget_locked(fdp, fd)) == NULL) { 1206 FILEDESC_XUNLOCK(fdp); 1207 return (EBADF); 1208 } 1209 fdfree(fdp, fd); 1210 1211 /* closefp() drops the FILEDESC lock for us. */ 1212 return (closefp(fdp, fd, fp, td, 1)); 1213} 1214 1215/* 1216 * Close open file descriptors. 1217 */ 1218#ifndef _SYS_SYSPROTO_H_ 1219struct closefrom_args { 1220 int lowfd; 1221}; 1222#endif 1223/* ARGSUSED */ 1224int 1225sys_closefrom(struct thread *td, struct closefrom_args *uap) 1226{ 1227 struct filedesc *fdp; 1228 int fd; 1229 1230 fdp = td->td_proc->p_fd; 1231 AUDIT_ARG_FD(uap->lowfd); 1232 1233 /* 1234 * Treat negative starting file descriptor values identical to 1235 * closefrom(0) which closes all files. 1236 */ 1237 if (uap->lowfd < 0) 1238 uap->lowfd = 0; 1239 FILEDESC_SLOCK(fdp); 1240 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) { 1241 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1242 FILEDESC_SUNLOCK(fdp); 1243 (void)kern_close(td, fd); 1244 FILEDESC_SLOCK(fdp); 1245 } 1246 } 1247 FILEDESC_SUNLOCK(fdp); 1248 return (0); 1249} 1250 1251#if defined(COMPAT_43) 1252/* 1253 * Return status information about a file descriptor. 1254 */ 1255#ifndef _SYS_SYSPROTO_H_ 1256struct ofstat_args { 1257 int fd; 1258 struct ostat *sb; 1259}; 1260#endif 1261/* ARGSUSED */ 1262int 1263ofstat(struct thread *td, struct ofstat_args *uap) 1264{ 1265 struct ostat oub; 1266 struct stat ub; 1267 int error; 1268 1269 error = kern_fstat(td, uap->fd, &ub); 1270 if (error == 0) { 1271 cvtstat(&ub, &oub); 1272 error = copyout(&oub, uap->sb, sizeof(oub)); 1273 } 1274 return (error); 1275} 1276#endif /* COMPAT_43 */ 1277 1278/* 1279 * Return status information about a file descriptor. 1280 */ 1281#ifndef _SYS_SYSPROTO_H_ 1282struct fstat_args { 1283 int fd; 1284 struct stat *sb; 1285}; 1286#endif 1287/* ARGSUSED */ 1288int 1289sys_fstat(struct thread *td, struct fstat_args *uap) 1290{ 1291 struct stat ub; 1292 int error; 1293 1294 error = kern_fstat(td, uap->fd, &ub); 1295 if (error == 0) 1296 error = copyout(&ub, uap->sb, sizeof(ub)); 1297 return (error); 1298} 1299 1300int 1301kern_fstat(struct thread *td, int fd, struct stat *sbp) 1302{ 1303 struct file *fp; 1304 cap_rights_t rights; 1305 int error; 1306 1307 AUDIT_ARG_FD(fd); 1308 1309 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1310 if (error != 0) 1311 return (error); 1312 1313 AUDIT_ARG_FILE(td->td_proc, fp); 1314 1315 error = fo_stat(fp, sbp, td->td_ucred, td); 1316 fdrop(fp, td); 1317#ifdef KTRACE 1318 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1319 ktrstat(sbp); 1320#endif 1321 return (error); 1322} 1323 1324/* 1325 * Return status information about a file descriptor. 1326 */ 1327#ifndef _SYS_SYSPROTO_H_ 1328struct nfstat_args { 1329 int fd; 1330 struct nstat *sb; 1331}; 1332#endif 1333/* ARGSUSED */ 1334int 1335sys_nfstat(struct thread *td, struct nfstat_args *uap) 1336{ 1337 struct nstat nub; 1338 struct stat ub; 1339 int error; 1340 1341 error = kern_fstat(td, uap->fd, &ub); 1342 if (error == 0) { 1343 cvtnstat(&ub, &nub); 1344 error = copyout(&nub, uap->sb, sizeof(nub)); 1345 } 1346 return (error); 1347} 1348 1349/* 1350 * Return pathconf information about a file descriptor. 1351 */ 1352#ifndef _SYS_SYSPROTO_H_ 1353struct fpathconf_args { 1354 int fd; 1355 int name; 1356}; 1357#endif 1358/* ARGSUSED */ 1359int 1360sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1361{ 1362 struct file *fp; 1363 struct vnode *vp; 1364 cap_rights_t rights; 1365 int error; 1366 1367 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1368 if (error != 0) 1369 return (error); 1370 1371 /* If asynchronous I/O is available, it works for all descriptors. */ 1372 if (uap->name == _PC_ASYNC_IO) { 1373 td->td_retval[0] = async_io_version; 1374 goto out; 1375 } 1376 vp = fp->f_vnode; 1377 if (vp != NULL) { 1378 vn_lock(vp, LK_SHARED | LK_RETRY); 1379 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1380 VOP_UNLOCK(vp, 0); 1381 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1382 if (uap->name != _PC_PIPE_BUF) { 1383 error = EINVAL; 1384 } else { 1385 td->td_retval[0] = PIPE_BUF; 1386 error = 0; 1387 } 1388 } else { 1389 error = EOPNOTSUPP; 1390 } 1391out: 1392 fdrop(fp, td); 1393 return (error); 1394} 1395 1396/* 1397 * Initialize filecaps structure. 1398 */ 1399void 1400filecaps_init(struct filecaps *fcaps) 1401{ 1402 1403 bzero(fcaps, sizeof(*fcaps)); 1404 fcaps->fc_nioctls = -1; 1405} 1406 1407/* 1408 * Copy filecaps structure allocating memory for ioctls array if needed. 1409 */ 1410void 1411filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1412{ 1413 size_t size; 1414 1415 *dst = *src; 1416 if (src->fc_ioctls != NULL) { 1417 KASSERT(src->fc_nioctls > 0, 1418 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1419 1420 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1421 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1422 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1423 } 1424} 1425 1426/* 1427 * Move filecaps structure to the new place and clear the old place. 1428 */ 1429void 1430filecaps_move(struct filecaps *src, struct filecaps *dst) 1431{ 1432 1433 *dst = *src; 1434 bzero(src, sizeof(*src)); 1435} 1436 1437/* 1438 * Fill the given filecaps structure with full rights. 1439 */ 1440static void 1441filecaps_fill(struct filecaps *fcaps) 1442{ 1443 1444 CAP_ALL(&fcaps->fc_rights); 1445 fcaps->fc_ioctls = NULL; 1446 fcaps->fc_nioctls = -1; 1447 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1448} 1449 1450/* 1451 * Free memory allocated within filecaps structure. 1452 */ 1453void 1454filecaps_free(struct filecaps *fcaps) 1455{ 1456 1457 free(fcaps->fc_ioctls, M_FILECAPS); 1458 bzero(fcaps, sizeof(*fcaps)); 1459} 1460 1461/* 1462 * Validate the given filecaps structure. 1463 */ 1464static void 1465filecaps_validate(const struct filecaps *fcaps, const char *func) 1466{ 1467 1468 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1469 ("%s: invalid rights", func)); 1470 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1471 ("%s: invalid fcntls", func)); 1472 KASSERT(fcaps->fc_fcntls == 0 || 1473 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1474 ("%s: fcntls without CAP_FCNTL", func)); 1475 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1476 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1477 ("%s: invalid ioctls", func)); 1478 KASSERT(fcaps->fc_nioctls == 0 || 1479 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1480 ("%s: ioctls without CAP_IOCTL", func)); 1481} 1482 1483static void 1484fdgrowtable_exp(struct filedesc *fdp, int nfd) 1485{ 1486 int nfd1; 1487 1488 FILEDESC_XLOCK_ASSERT(fdp); 1489 1490 nfd1 = fdp->fd_nfiles * 2; 1491 if (nfd1 < nfd) 1492 nfd1 = nfd; 1493 fdgrowtable(fdp, nfd1); 1494} 1495 1496/* 1497 * Grow the file table to accomodate (at least) nfd descriptors. 1498 */ 1499static void 1500fdgrowtable(struct filedesc *fdp, int nfd) 1501{ 1502 struct filedesc0 *fdp0; 1503 struct freetable *ft; 1504 struct filedescent *ntable; 1505 struct filedescent *otable; 1506 int nnfiles, onfiles; 1507 NDSLOTTYPE *nmap, *omap; 1508 1509 FILEDESC_XLOCK_ASSERT(fdp); 1510 1511 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1512 1513 /* save old values */ 1514 onfiles = fdp->fd_nfiles; 1515 otable = fdp->fd_ofiles; 1516 omap = fdp->fd_map; 1517 1518 /* compute the size of the new table */ 1519 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1520 if (nnfiles <= onfiles) 1521 /* the table is already large enough */ 1522 return; 1523 1524 /* 1525 * Allocate a new table. We need enough space for the 1526 * file entries themselves and the struct freetable we will use 1527 * when we decommission the table and place it on the freelist. 1528 * We place the struct freetable in the middle so we don't have 1529 * to worry about padding. 1530 */ 1531 ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable), 1532 M_FILEDESC, M_ZERO | M_WAITOK); 1533 /* copy the old data over and point at the new tables */ 1534 memcpy(ntable, otable, onfiles * sizeof(*otable)); 1535 fdp->fd_ofiles = ntable; 1536 1537 /* 1538 * Allocate a new map only if the old is not large enough. It will 1539 * grow at a slower rate than the table as it can map more 1540 * entries than the table can hold. 1541 */ 1542 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1543 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1544 M_ZERO | M_WAITOK); 1545 /* copy over the old data and update the pointer */ 1546 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1547 fdp->fd_map = nmap; 1548 } 1549 1550 /* 1551 * In order to have a valid pattern for fget_unlocked() 1552 * fdp->fd_nfiles must be the last member to be updated, otherwise 1553 * fget_unlocked() consumers may reference a new, higher value for 1554 * fdp->fd_nfiles before to access the fdp->fd_ofiles array, 1555 * resulting in OOB accesses. 1556 */ 1557 atomic_store_rel_int(&fdp->fd_nfiles, nnfiles); 1558 1559 /* 1560 * Do not free the old file table, as some threads may still 1561 * reference entries within it. Instead, place it on a freelist 1562 * which will be processed when the struct filedesc is released. 1563 * 1564 * Note that if onfiles == NDFILE, we're dealing with the original 1565 * static allocation contained within (struct filedesc0 *)fdp, 1566 * which must not be freed. 1567 */ 1568 if (onfiles > NDFILE) { 1569 ft = (struct freetable *)&otable[onfiles]; 1570 fdp0 = (struct filedesc0 *)fdp; 1571 ft->ft_table = otable; 1572 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1573 } 1574 /* 1575 * The map does not have the same possibility of threads still 1576 * holding references to it. So always free it as long as it 1577 * does not reference the original static allocation. 1578 */ 1579 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1580 free(omap, M_FILEDESC); 1581} 1582 1583/* 1584 * Allocate a file descriptor for the process. 1585 */ 1586int 1587fdalloc(struct thread *td, int minfd, int *result) 1588{ 1589 struct proc *p = td->td_proc; 1590 struct filedesc *fdp = p->p_fd; 1591 int fd = -1, maxfd, allocfd; 1592#ifdef RACCT 1593 int error; 1594#endif 1595 1596 FILEDESC_XLOCK_ASSERT(fdp); 1597 1598 if (fdp->fd_freefile > minfd) 1599 minfd = fdp->fd_freefile; 1600 1601 maxfd = getmaxfd(p); 1602 1603 /* 1604 * Search the bitmap for a free descriptor starting at minfd. 1605 * If none is found, grow the file table. 1606 */ 1607 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1608 if (fd >= maxfd) 1609 return (EMFILE); 1610 if (fd >= fdp->fd_nfiles) { 1611 allocfd = min(fd * 2, maxfd); 1612#ifdef RACCT 1613 PROC_LOCK(p); 1614 error = racct_set(p, RACCT_NOFILE, allocfd); 1615 PROC_UNLOCK(p); 1616 if (error != 0) 1617 return (EMFILE); 1618#endif 1619 /* 1620 * fd is already equal to first free descriptor >= minfd, so 1621 * we only need to grow the table and we are done. 1622 */ 1623 fdgrowtable_exp(fdp, allocfd); 1624 } 1625 1626 /* 1627 * Perform some sanity checks, then mark the file descriptor as 1628 * used and return it to the caller. 1629 */ 1630 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1631 ("invalid descriptor %d", fd)); 1632 KASSERT(!fdisused(fdp, fd), 1633 ("fd_first_free() returned non-free descriptor")); 1634 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1635 ("file descriptor isn't free")); 1636 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); 1637 fdused(fdp, fd); 1638 *result = fd; 1639 return (0); 1640} 1641 1642/* 1643 * Allocate n file descriptors for the process. 1644 */ 1645int 1646fdallocn(struct thread *td, int minfd, int *fds, int n) 1647{ 1648 struct proc *p = td->td_proc; 1649 struct filedesc *fdp = p->p_fd; 1650 int i; 1651 1652 FILEDESC_XLOCK_ASSERT(fdp); 1653 1654 if (!fdavail(td, n)) 1655 return (EMFILE); 1656 1657 for (i = 0; i < n; i++) 1658 if (fdalloc(td, 0, &fds[i]) != 0) 1659 break; 1660 1661 if (i < n) { 1662 for (i--; i >= 0; i--) 1663 fdunused(fdp, fds[i]); 1664 return (EMFILE); 1665 } 1666 1667 return (0); 1668} 1669 1670/* 1671 * Check to see whether n user file descriptors are available to the process 1672 * p. 1673 */ 1674int 1675fdavail(struct thread *td, int n) 1676{ 1677 struct proc *p = td->td_proc; 1678 struct filedesc *fdp = td->td_proc->p_fd; 1679 int i, lim, last; 1680 1681 FILEDESC_LOCK_ASSERT(fdp); 1682 1683 /* 1684 * XXX: This is only called from uipc_usrreq.c:unp_externalize(); 1685 * call racct_add() from there instead of dealing with containers 1686 * here. 1687 */ 1688 lim = getmaxfd(p); 1689 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1690 return (1); 1691 last = min(fdp->fd_nfiles, lim); 1692 for (i = fdp->fd_freefile; i < last; i++) { 1693 if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0) 1694 return (1); 1695 } 1696 return (0); 1697} 1698 1699/* 1700 * Create a new open file structure and allocate a file decriptor for the 1701 * process that refers to it. We add one reference to the file for the 1702 * descriptor table and one reference for resultfp. This is to prevent us 1703 * being preempted and the entry in the descriptor table closed after we 1704 * release the FILEDESC lock. 1705 */ 1706int 1707falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1708{ 1709 struct file *fp; 1710 int error, fd; 1711 1712 error = falloc_noinstall(td, &fp); 1713 if (error) 1714 return (error); /* no reference held on error */ 1715 1716 error = finstall(td, fp, &fd, flags, NULL); 1717 if (error) { 1718 fdrop(fp, td); /* one reference (fp only) */ 1719 return (error); 1720 } 1721 1722 if (resultfp != NULL) 1723 *resultfp = fp; /* copy out result */ 1724 else 1725 fdrop(fp, td); /* release local reference */ 1726 1727 if (resultfd != NULL) 1728 *resultfd = fd; 1729 1730 return (0); 1731} 1732 1733/* 1734 * Create a new open file structure without allocating a file descriptor. 1735 */ 1736int 1737falloc_noinstall(struct thread *td, struct file **resultfp) 1738{ 1739 struct file *fp; 1740 int maxuserfiles = maxfiles - (maxfiles / 20); 1741 static struct timeval lastfail; 1742 static int curfail; 1743 1744 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1745 1746 if ((openfiles >= maxuserfiles && 1747 priv_check(td, PRIV_MAXFILES) != 0) || 1748 openfiles >= maxfiles) { 1749 if (ppsratecheck(&lastfail, &curfail, 1)) { 1750 printf("kern.maxfiles limit exceeded by uid %i, " 1751 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1752 } 1753 return (ENFILE); 1754 } 1755 atomic_add_int(&openfiles, 1); 1756 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1757 refcount_init(&fp->f_count, 1); 1758 fp->f_cred = crhold(td->td_ucred); 1759 fp->f_ops = &badfileops; 1760 fp->f_data = NULL; 1761 fp->f_vnode = NULL; 1762 *resultfp = fp; 1763 return (0); 1764} 1765 1766/* 1767 * Install a file in a file descriptor table. 1768 */ 1769int 1770finstall(struct thread *td, struct file *fp, int *fd, int flags, 1771 struct filecaps *fcaps) 1772{ 1773 struct filedesc *fdp = td->td_proc->p_fd; 1774 struct filedescent *fde; 1775 int error; 1776 1777 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1778 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1779 if (fcaps != NULL) 1780 filecaps_validate(fcaps, __func__); 1781 1782 FILEDESC_XLOCK(fdp); 1783 if ((error = fdalloc(td, 0, fd))) { 1784 FILEDESC_XUNLOCK(fdp); 1785 return (error); 1786 } 1787 fhold(fp); 1788 fde = &fdp->fd_ofiles[*fd]; 1789 fde->fde_file = fp; 1790 if ((flags & O_CLOEXEC) != 0) 1791 fde->fde_flags |= UF_EXCLOSE; 1792 if (fcaps != NULL) 1793 filecaps_move(fcaps, &fde->fde_caps); 1794 else 1795 filecaps_fill(&fde->fde_caps); 1796 FILEDESC_XUNLOCK(fdp); 1797 return (0); 1798} 1799 1800/* 1801 * Build a new filedesc structure from another. 1802 * Copy the current, root, and jail root vnode references. 1803 */ 1804struct filedesc * 1805fdinit(struct filedesc *fdp) 1806{ 1807 struct filedesc0 *newfdp; 1808 1809 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1810 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1811 if (fdp != NULL) { 1812 FILEDESC_SLOCK(fdp); 1813 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1814 if (newfdp->fd_fd.fd_cdir) 1815 VREF(newfdp->fd_fd.fd_cdir); 1816 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1817 if (newfdp->fd_fd.fd_rdir) 1818 VREF(newfdp->fd_fd.fd_rdir); 1819 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1820 if (newfdp->fd_fd.fd_jdir) 1821 VREF(newfdp->fd_fd.fd_jdir); 1822 FILEDESC_SUNLOCK(fdp); 1823 } 1824 1825 /* Create the file descriptor table. */ 1826 newfdp->fd_fd.fd_refcnt = 1; 1827 newfdp->fd_fd.fd_holdcnt = 1; 1828 newfdp->fd_fd.fd_cmask = CMASK; 1829 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1830 newfdp->fd_fd.fd_nfiles = NDFILE; 1831 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1832 newfdp->fd_fd.fd_lastfile = -1; 1833 return (&newfdp->fd_fd); 1834} 1835 1836static struct filedesc * 1837fdhold(struct proc *p) 1838{ 1839 struct filedesc *fdp; 1840 1841 mtx_lock(&fdesc_mtx); 1842 fdp = p->p_fd; 1843 if (fdp != NULL) 1844 fdp->fd_holdcnt++; 1845 mtx_unlock(&fdesc_mtx); 1846 return (fdp); 1847} 1848 1849static void 1850fddrop(struct filedesc *fdp) 1851{ 1852 struct filedesc0 *fdp0; 1853 struct freetable *ft; 1854 int i; 1855 1856 mtx_lock(&fdesc_mtx); 1857 i = --fdp->fd_holdcnt; 1858 mtx_unlock(&fdesc_mtx); 1859 if (i > 0) 1860 return; 1861 1862 FILEDESC_LOCK_DESTROY(fdp); 1863 fdp0 = (struct filedesc0 *)fdp; 1864 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1865 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1866 free(ft->ft_table, M_FILEDESC); 1867 } 1868 free(fdp, M_FILEDESC); 1869} 1870 1871/* 1872 * Share a filedesc structure. 1873 */ 1874struct filedesc * 1875fdshare(struct filedesc *fdp) 1876{ 1877 1878 FILEDESC_XLOCK(fdp); 1879 fdp->fd_refcnt++; 1880 FILEDESC_XUNLOCK(fdp); 1881 return (fdp); 1882} 1883 1884/* 1885 * Unshare a filedesc structure, if necessary by making a copy 1886 */ 1887void 1888fdunshare(struct proc *p, struct thread *td) 1889{ 1890 1891 FILEDESC_XLOCK(p->p_fd); 1892 if (p->p_fd->fd_refcnt > 1) { 1893 struct filedesc *tmp; 1894 1895 FILEDESC_XUNLOCK(p->p_fd); 1896 tmp = fdcopy(p->p_fd); 1897 fdescfree(td); 1898 p->p_fd = tmp; 1899 } else 1900 FILEDESC_XUNLOCK(p->p_fd); 1901} 1902 1903/* 1904 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1905 * this is to ease callers, not catch errors. 1906 */ 1907struct filedesc * 1908fdcopy(struct filedesc *fdp) 1909{ 1910 struct filedesc *newfdp; 1911 struct filedescent *nfde, *ofde; 1912 int i; 1913 1914 /* Certain daemons might not have file descriptors. */ 1915 if (fdp == NULL) 1916 return (NULL); 1917 1918 newfdp = fdinit(fdp); 1919 FILEDESC_SLOCK(fdp); 1920 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1921 FILEDESC_SUNLOCK(fdp); 1922 FILEDESC_XLOCK(newfdp); 1923 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1924 FILEDESC_XUNLOCK(newfdp); 1925 FILEDESC_SLOCK(fdp); 1926 } 1927 /* copy all passable descriptors (i.e. not kqueue) */ 1928 newfdp->fd_freefile = -1; 1929 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1930 ofde = &fdp->fd_ofiles[i]; 1931 if (fdisused(fdp, i) && 1932 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) && 1933 ofde->fde_file->f_ops != &badfileops) { 1934 nfde = &newfdp->fd_ofiles[i]; 1935 *nfde = *ofde; 1936 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1937 fhold(nfde->fde_file); 1938 newfdp->fd_lastfile = i; 1939 } else { 1940 if (newfdp->fd_freefile == -1) 1941 newfdp->fd_freefile = i; 1942 } 1943 } 1944 newfdp->fd_cmask = fdp->fd_cmask; 1945 FILEDESC_SUNLOCK(fdp); 1946 FILEDESC_XLOCK(newfdp); 1947 for (i = 0; i <= newfdp->fd_lastfile; ++i) { 1948 if (newfdp->fd_ofiles[i].fde_file != NULL) 1949 fdused(newfdp, i); 1950 } 1951 if (newfdp->fd_freefile == -1) 1952 newfdp->fd_freefile = i; 1953 FILEDESC_XUNLOCK(newfdp); 1954 return (newfdp); 1955} 1956 1957/* 1958 * Release a filedesc structure. 1959 */ 1960void 1961fdescfree(struct thread *td) 1962{ 1963 struct filedesc *fdp; 1964 int i; 1965 struct filedesc_to_leader *fdtol; 1966 struct file *fp; 1967 struct vnode *cdir, *jdir, *rdir, *vp; 1968 struct flock lf; 1969 1970 /* Certain daemons might not have file descriptors. */ 1971 fdp = td->td_proc->p_fd; 1972 if (fdp == NULL) 1973 return; 1974 1975#ifdef RACCT 1976 PROC_LOCK(td->td_proc); 1977 racct_set(td->td_proc, RACCT_NOFILE, 0); 1978 PROC_UNLOCK(td->td_proc); 1979#endif 1980 1981 /* Check for special need to clear POSIX style locks */ 1982 fdtol = td->td_proc->p_fdtol; 1983 if (fdtol != NULL) { 1984 FILEDESC_XLOCK(fdp); 1985 KASSERT(fdtol->fdl_refcount > 0, 1986 ("filedesc_to_refcount botch: fdl_refcount=%d", 1987 fdtol->fdl_refcount)); 1988 if (fdtol->fdl_refcount == 1 && 1989 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1990 for (i = 0; i <= fdp->fd_lastfile; i++) { 1991 fp = fdp->fd_ofiles[i].fde_file; 1992 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1993 continue; 1994 fhold(fp); 1995 FILEDESC_XUNLOCK(fdp); 1996 lf.l_whence = SEEK_SET; 1997 lf.l_start = 0; 1998 lf.l_len = 0; 1999 lf.l_type = F_UNLCK; 2000 vp = fp->f_vnode; 2001 (void) VOP_ADVLOCK(vp, 2002 (caddr_t)td->td_proc->p_leader, F_UNLCK, 2003 &lf, F_POSIX); 2004 FILEDESC_XLOCK(fdp); 2005 fdrop(fp, td); 2006 } 2007 } 2008 retry: 2009 if (fdtol->fdl_refcount == 1) { 2010 if (fdp->fd_holdleaderscount > 0 && 2011 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2012 /* 2013 * close() or do_dup() has cleared a reference 2014 * in a shared file descriptor table. 2015 */ 2016 fdp->fd_holdleaderswakeup = 1; 2017 sx_sleep(&fdp->fd_holdleaderscount, 2018 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2019 goto retry; 2020 } 2021 if (fdtol->fdl_holdcount > 0) { 2022 /* 2023 * Ensure that fdtol->fdl_leader remains 2024 * valid in closef(). 2025 */ 2026 fdtol->fdl_wakeup = 1; 2027 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2028 "fdlhold", 0); 2029 goto retry; 2030 } 2031 } 2032 fdtol->fdl_refcount--; 2033 if (fdtol->fdl_refcount == 0 && 2034 fdtol->fdl_holdcount == 0) { 2035 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2036 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2037 } else 2038 fdtol = NULL; 2039 td->td_proc->p_fdtol = NULL; 2040 FILEDESC_XUNLOCK(fdp); 2041 if (fdtol != NULL) 2042 free(fdtol, M_FILEDESC_TO_LEADER); 2043 } 2044 FILEDESC_XLOCK(fdp); 2045 i = --fdp->fd_refcnt; 2046 FILEDESC_XUNLOCK(fdp); 2047 if (i > 0) 2048 return; 2049 2050 for (i = 0; i <= fdp->fd_lastfile; i++) { 2051 fp = fdp->fd_ofiles[i].fde_file; 2052 if (fp != NULL) { 2053 FILEDESC_XLOCK(fdp); 2054 fdfree(fdp, i); 2055 FILEDESC_XUNLOCK(fdp); 2056 (void) closef(fp, td); 2057 } 2058 } 2059 FILEDESC_XLOCK(fdp); 2060 2061 /* XXX This should happen earlier. */ 2062 mtx_lock(&fdesc_mtx); 2063 td->td_proc->p_fd = NULL; 2064 mtx_unlock(&fdesc_mtx); 2065 2066 if (fdp->fd_nfiles > NDFILE) 2067 free(fdp->fd_ofiles, M_FILEDESC); 2068 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2069 free(fdp->fd_map, M_FILEDESC); 2070 2071 fdp->fd_nfiles = 0; 2072 2073 cdir = fdp->fd_cdir; 2074 fdp->fd_cdir = NULL; 2075 rdir = fdp->fd_rdir; 2076 fdp->fd_rdir = NULL; 2077 jdir = fdp->fd_jdir; 2078 fdp->fd_jdir = NULL; 2079 FILEDESC_XUNLOCK(fdp); 2080 2081 if (cdir != NULL) 2082 vrele(cdir); 2083 if (rdir != NULL) 2084 vrele(rdir); 2085 if (jdir != NULL) 2086 vrele(jdir); 2087 2088 fddrop(fdp); 2089} 2090 2091/* 2092 * For setugid programs, we don't want to people to use that setugidness 2093 * to generate error messages which write to a file which otherwise would 2094 * otherwise be off-limits to the process. We check for filesystems where 2095 * the vnode can change out from under us after execve (like [lin]procfs). 2096 * 2097 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2098 * sufficient. We also don't check for setugidness since we know we are. 2099 */ 2100static int 2101is_unsafe(struct file *fp) 2102{ 2103 if (fp->f_type == DTYPE_VNODE) { 2104 struct vnode *vp = fp->f_vnode; 2105 2106 if ((vp->v_vflag & VV_PROCDEP) != 0) 2107 return (1); 2108 } 2109 return (0); 2110} 2111 2112/* 2113 * Make this setguid thing safe, if at all possible. 2114 */ 2115void 2116setugidsafety(struct thread *td) 2117{ 2118 struct filedesc *fdp; 2119 struct file *fp; 2120 int i; 2121 2122 /* Certain daemons might not have file descriptors. */ 2123 fdp = td->td_proc->p_fd; 2124 if (fdp == NULL) 2125 return; 2126 2127 /* 2128 * Note: fdp->fd_ofiles may be reallocated out from under us while 2129 * we are blocked in a close. Be careful! 2130 */ 2131 FILEDESC_XLOCK(fdp); 2132 for (i = 0; i <= fdp->fd_lastfile; i++) { 2133 if (i > 2) 2134 break; 2135 fp = fdp->fd_ofiles[i].fde_file; 2136 if (fp != NULL && is_unsafe(fp)) { 2137 knote_fdclose(td, i); 2138 /* 2139 * NULL-out descriptor prior to close to avoid 2140 * a race while close blocks. 2141 */ 2142 fdfree(fdp, i); 2143 FILEDESC_XUNLOCK(fdp); 2144 (void) closef(fp, td); 2145 FILEDESC_XLOCK(fdp); 2146 } 2147 } 2148 FILEDESC_XUNLOCK(fdp); 2149} 2150 2151/* 2152 * If a specific file object occupies a specific file descriptor, close the 2153 * file descriptor entry and drop a reference on the file object. This is a 2154 * convenience function to handle a subsequent error in a function that calls 2155 * falloc() that handles the race that another thread might have closed the 2156 * file descriptor out from under the thread creating the file object. 2157 */ 2158void 2159fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2160{ 2161 2162 FILEDESC_XLOCK(fdp); 2163 if (fdp->fd_ofiles[idx].fde_file == fp) { 2164 fdfree(fdp, idx); 2165 FILEDESC_XUNLOCK(fdp); 2166 fdrop(fp, td); 2167 } else 2168 FILEDESC_XUNLOCK(fdp); 2169} 2170 2171/* 2172 * Close any files on exec? 2173 */ 2174void 2175fdcloseexec(struct thread *td) 2176{ 2177 struct filedesc *fdp; 2178 struct filedescent *fde; 2179 struct file *fp; 2180 int i; 2181 2182 /* Certain daemons might not have file descriptors. */ 2183 fdp = td->td_proc->p_fd; 2184 if (fdp == NULL) 2185 return; 2186 2187 /* 2188 * We cannot cache fd_ofiles since operations 2189 * may block and rip them out from under us. 2190 */ 2191 FILEDESC_XLOCK(fdp); 2192 for (i = 0; i <= fdp->fd_lastfile; i++) { 2193 fde = &fdp->fd_ofiles[i]; 2194 fp = fde->fde_file; 2195 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2196 (fde->fde_flags & UF_EXCLOSE))) { 2197 fdfree(fdp, i); 2198 (void) closefp(fdp, i, fp, td, 0); 2199 /* closefp() drops the FILEDESC lock. */ 2200 FILEDESC_XLOCK(fdp); 2201 } 2202 } 2203 FILEDESC_XUNLOCK(fdp); 2204} 2205 2206/* 2207 * It is unsafe for set[ug]id processes to be started with file 2208 * descriptors 0..2 closed, as these descriptors are given implicit 2209 * significance in the Standard C library. fdcheckstd() will create a 2210 * descriptor referencing /dev/null for each of stdin, stdout, and 2211 * stderr that is not already open. 2212 */ 2213int 2214fdcheckstd(struct thread *td) 2215{ 2216 struct filedesc *fdp; 2217 register_t retval, save; 2218 int i, error, devnull; 2219 2220 fdp = td->td_proc->p_fd; 2221 if (fdp == NULL) 2222 return (0); 2223 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2224 devnull = -1; 2225 error = 0; 2226 for (i = 0; i < 3; i++) { 2227 if (fdp->fd_ofiles[i].fde_file != NULL) 2228 continue; 2229 if (devnull < 0) { 2230 save = td->td_retval[0]; 2231 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2232 O_RDWR, 0); 2233 devnull = td->td_retval[0]; 2234 td->td_retval[0] = save; 2235 if (error) 2236 break; 2237 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2238 } else { 2239 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2240 if (error != 0) 2241 break; 2242 } 2243 } 2244 return (error); 2245} 2246 2247/* 2248 * Internal form of close. Decrement reference count on file structure. 2249 * Note: td may be NULL when closing a file that was being passed in a 2250 * message. 2251 * 2252 * XXXRW: Giant is not required for the caller, but often will be held; this 2253 * makes it moderately likely the Giant will be recursed in the VFS case. 2254 */ 2255int 2256closef(struct file *fp, struct thread *td) 2257{ 2258 struct vnode *vp; 2259 struct flock lf; 2260 struct filedesc_to_leader *fdtol; 2261 struct filedesc *fdp; 2262 2263 /* 2264 * POSIX record locking dictates that any close releases ALL 2265 * locks owned by this process. This is handled by setting 2266 * a flag in the unlock to free ONLY locks obeying POSIX 2267 * semantics, and not to free BSD-style file locks. 2268 * If the descriptor was in a message, POSIX-style locks 2269 * aren't passed with the descriptor, and the thread pointer 2270 * will be NULL. Callers should be careful only to pass a 2271 * NULL thread pointer when there really is no owning 2272 * context that might have locks, or the locks will be 2273 * leaked. 2274 */ 2275 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2276 vp = fp->f_vnode; 2277 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2278 lf.l_whence = SEEK_SET; 2279 lf.l_start = 0; 2280 lf.l_len = 0; 2281 lf.l_type = F_UNLCK; 2282 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2283 F_UNLCK, &lf, F_POSIX); 2284 } 2285 fdtol = td->td_proc->p_fdtol; 2286 if (fdtol != NULL) { 2287 /* 2288 * Handle special case where file descriptor table is 2289 * shared between multiple process leaders. 2290 */ 2291 fdp = td->td_proc->p_fd; 2292 FILEDESC_XLOCK(fdp); 2293 for (fdtol = fdtol->fdl_next; 2294 fdtol != td->td_proc->p_fdtol; 2295 fdtol = fdtol->fdl_next) { 2296 if ((fdtol->fdl_leader->p_flag & 2297 P_ADVLOCK) == 0) 2298 continue; 2299 fdtol->fdl_holdcount++; 2300 FILEDESC_XUNLOCK(fdp); 2301 lf.l_whence = SEEK_SET; 2302 lf.l_start = 0; 2303 lf.l_len = 0; 2304 lf.l_type = F_UNLCK; 2305 vp = fp->f_vnode; 2306 (void) VOP_ADVLOCK(vp, 2307 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2308 F_POSIX); 2309 FILEDESC_XLOCK(fdp); 2310 fdtol->fdl_holdcount--; 2311 if (fdtol->fdl_holdcount == 0 && 2312 fdtol->fdl_wakeup != 0) { 2313 fdtol->fdl_wakeup = 0; 2314 wakeup(fdtol); 2315 } 2316 } 2317 FILEDESC_XUNLOCK(fdp); 2318 } 2319 } 2320 return (fdrop(fp, td)); 2321} 2322 2323/* 2324 * Initialize the file pointer with the specified properties. 2325 * 2326 * The ops are set with release semantics to be certain that the flags, type, 2327 * and data are visible when ops is. This is to prevent ops methods from being 2328 * called with bad data. 2329 */ 2330void 2331finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2332{ 2333 fp->f_data = data; 2334 fp->f_flag = flag; 2335 fp->f_type = type; 2336 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2337} 2338 2339int 2340fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2341 int needfcntl, struct file **fpp, cap_rights_t *haverightsp) 2342{ 2343 struct file *fp; 2344 u_int count; 2345#ifdef CAPABILITIES 2346 cap_rights_t haverights; 2347 int error; 2348#endif 2349 2350 /* 2351 * Avoid reads reordering and then a first access to the 2352 * fdp->fd_ofiles table which could result in OOB operation. 2353 */ 2354 if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles)) 2355 return (EBADF); 2356 /* 2357 * Fetch the descriptor locklessly. We avoid fdrop() races by 2358 * never raising a refcount above 0. To accomplish this we have 2359 * to use a cmpset loop rather than an atomic_add. The descriptor 2360 * must be re-verified once we acquire a reference to be certain 2361 * that the identity is still correct and we did not lose a race 2362 * due to preemption. 2363 */ 2364 for (;;) { 2365 fp = fdp->fd_ofiles[fd].fde_file; 2366 if (fp == NULL) 2367 return (EBADF); 2368#ifdef CAPABILITIES 2369 haverights = *cap_rights(fdp, fd); 2370 if (needrightsp != NULL) { 2371 error = cap_check(&haverights, needrightsp); 2372 if (error != 0) 2373 return (error); 2374 if (cap_rights_is_set(needrightsp, CAP_FCNTL)) { 2375 error = cap_fcntl_check(fdp, fd, needfcntl); 2376 if (error != 0) 2377 return (error); 2378 } 2379 } 2380#endif 2381 count = fp->f_count; 2382 if (count == 0) 2383 continue; 2384 /* 2385 * Use an acquire barrier to prevent caching of fd_ofiles 2386 * so it is refreshed for verification. 2387 */ 2388 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2389 continue; 2390 if (fp == fdp->fd_ofiles[fd].fde_file) 2391 break; 2392 fdrop(fp, curthread); 2393 } 2394 *fpp = fp; 2395 if (haverightsp != NULL) { 2396#ifdef CAPABILITIES 2397 *haverightsp = haverights; 2398#else 2399 CAP_ALL(haverightsp); 2400#endif 2401 } 2402 return (0); 2403} 2404 2405/* 2406 * Extract the file pointer associated with the specified descriptor for the 2407 * current user process. 2408 * 2409 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2410 * returned. 2411 * 2412 * File's rights will be checked against the capability rights mask. 2413 * 2414 * If an error occured the non-zero error is returned and *fpp is set to 2415 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2416 * responsible for fdrop(). 2417 */ 2418static __inline int 2419_fget(struct thread *td, int fd, struct file **fpp, int flags, 2420 cap_rights_t *needrightsp, u_char *maxprotp) 2421{ 2422 struct filedesc *fdp; 2423 struct file *fp; 2424 cap_rights_t haverights, needrights; 2425 int error; 2426 2427 *fpp = NULL; 2428 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2429 return (EBADF); 2430 if (needrightsp != NULL) 2431 needrights = *needrightsp; 2432 else 2433 cap_rights_init(&needrights); 2434 if (maxprotp != NULL) 2435 cap_rights_set(&needrights, CAP_MMAP); 2436 error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights); 2437 if (error != 0) 2438 return (error); 2439 if (fp->f_ops == &badfileops) { 2440 fdrop(fp, td); 2441 return (EBADF); 2442 } 2443 2444#ifdef CAPABILITIES 2445 /* 2446 * If requested, convert capability rights to access flags. 2447 */ 2448 if (maxprotp != NULL) 2449 *maxprotp = cap_rights_to_vmprot(&haverights); 2450#else /* !CAPABILITIES */ 2451 if (maxprotp != NULL) 2452 *maxprotp = VM_PROT_ALL; 2453#endif /* CAPABILITIES */ 2454 2455 /* 2456 * FREAD and FWRITE failure return EBADF as per POSIX. 2457 */ 2458 error = 0; 2459 switch (flags) { 2460 case FREAD: 2461 case FWRITE: 2462 if ((fp->f_flag & flags) == 0) 2463 error = EBADF; 2464 break; 2465 case FEXEC: 2466 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2467 ((fp->f_flag & FWRITE) != 0)) 2468 error = EBADF; 2469 break; 2470 case 0: 2471 break; 2472 default: 2473 KASSERT(0, ("wrong flags")); 2474 } 2475 2476 if (error != 0) { 2477 fdrop(fp, td); 2478 return (error); 2479 } 2480 2481 *fpp = fp; 2482 return (0); 2483} 2484 2485int 2486fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2487{ 2488 2489 return(_fget(td, fd, fpp, 0, rightsp, NULL)); 2490} 2491 2492int 2493fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2494 struct file **fpp) 2495{ 2496 2497 return (_fget(td, fd, fpp, 0, rightsp, maxprotp)); 2498} 2499 2500int 2501fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2502{ 2503 2504 return(_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2505} 2506 2507int 2508fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2509{ 2510 2511 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2512} 2513 2514/* 2515 * Like fget() but loads the underlying vnode, or returns an error if the 2516 * descriptor does not represent a vnode. Note that pipes use vnodes but 2517 * never have VM objects. The returned vnode will be vref()'d. 2518 * 2519 * XXX: what about the unused flags ? 2520 */ 2521static __inline int 2522_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2523 struct vnode **vpp) 2524{ 2525 struct file *fp; 2526 int error; 2527 2528 *vpp = NULL; 2529 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2530 if (error != 0) 2531 return (error); 2532 if (fp->f_vnode == NULL) { 2533 error = EINVAL; 2534 } else { 2535 *vpp = fp->f_vnode; 2536 vref(*vpp); 2537 } 2538 fdrop(fp, td); 2539 2540 return (error); 2541} 2542 2543int 2544fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2545{ 2546 2547 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2548} 2549 2550int 2551fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2552 struct filecaps *havecaps, struct vnode **vpp) 2553{ 2554 struct filedesc *fdp; 2555 struct file *fp; 2556#ifdef CAPABILITIES 2557 int error; 2558#endif 2559 2560 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2561 return (EBADF); 2562 2563 fp = fget_locked(fdp, fd); 2564 if (fp == NULL || fp->f_ops == &badfileops) 2565 return (EBADF); 2566 2567#ifdef CAPABILITIES 2568 if (needrightsp != NULL) { 2569 error = cap_check(cap_rights(fdp, fd), needrightsp); 2570 if (error != 0) 2571 return (error); 2572 } 2573#endif 2574 2575 if (fp->f_vnode == NULL) 2576 return (EINVAL); 2577 2578 *vpp = fp->f_vnode; 2579 vref(*vpp); 2580 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2581 2582 return (0); 2583} 2584 2585int 2586fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2587{ 2588 2589 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2590} 2591 2592int 2593fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2594{ 2595 2596 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2597} 2598 2599#ifdef notyet 2600int 2601fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2602 struct vnode **vpp) 2603{ 2604 2605 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2606} 2607#endif 2608 2609/* 2610 * Like fget() but loads the underlying socket, or returns an error if the 2611 * descriptor does not represent a socket. 2612 * 2613 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2614 * in the future. 2615 * 2616 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2617 * on their file descriptor reference to prevent the socket from being free'd 2618 * during use. 2619 */ 2620int 2621fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2622 u_int *fflagp) 2623{ 2624 struct file *fp; 2625 int error; 2626 2627 *spp = NULL; 2628 if (fflagp != NULL) 2629 *fflagp = 0; 2630 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2631 return (error); 2632 if (fp->f_type != DTYPE_SOCKET) { 2633 error = ENOTSOCK; 2634 } else { 2635 *spp = fp->f_data; 2636 if (fflagp) 2637 *fflagp = fp->f_flag; 2638 SOCK_LOCK(*spp); 2639 soref(*spp); 2640 SOCK_UNLOCK(*spp); 2641 } 2642 fdrop(fp, td); 2643 2644 return (error); 2645} 2646 2647/* 2648 * Drop the reference count on the socket and XXX release the SX lock in the 2649 * future. The last reference closes the socket. 2650 * 2651 * Note: fputsock() is deprecated, see comment for fgetsock(). 2652 */ 2653void 2654fputsock(struct socket *so) 2655{ 2656 2657 ACCEPT_LOCK(); 2658 SOCK_LOCK(so); 2659 CURVNET_SET(so->so_vnet); 2660 sorele(so); 2661 CURVNET_RESTORE(); 2662} 2663 2664/* 2665 * Handle the last reference to a file being closed. 2666 */ 2667int 2668_fdrop(struct file *fp, struct thread *td) 2669{ 2670 int error; 2671 2672 error = 0; 2673 if (fp->f_count != 0) 2674 panic("fdrop: count %d", fp->f_count); 2675 if (fp->f_ops != &badfileops) 2676 error = fo_close(fp, td); 2677 atomic_subtract_int(&openfiles, 1); 2678 crfree(fp->f_cred); 2679 free(fp->f_advice, M_FADVISE); 2680 uma_zfree(file_zone, fp); 2681 2682 return (error); 2683} 2684 2685/* 2686 * Apply an advisory lock on a file descriptor. 2687 * 2688 * Just attempt to get a record lock of the requested type on the entire file 2689 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2690 */ 2691#ifndef _SYS_SYSPROTO_H_ 2692struct flock_args { 2693 int fd; 2694 int how; 2695}; 2696#endif 2697/* ARGSUSED */ 2698int 2699sys_flock(struct thread *td, struct flock_args *uap) 2700{ 2701 struct file *fp; 2702 struct vnode *vp; 2703 struct flock lf; 2704 cap_rights_t rights; 2705 int error; 2706 2707 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2708 if (error != 0) 2709 return (error); 2710 if (fp->f_type != DTYPE_VNODE) { 2711 fdrop(fp, td); 2712 return (EOPNOTSUPP); 2713 } 2714 2715 vp = fp->f_vnode; 2716 lf.l_whence = SEEK_SET; 2717 lf.l_start = 0; 2718 lf.l_len = 0; 2719 if (uap->how & LOCK_UN) { 2720 lf.l_type = F_UNLCK; 2721 atomic_clear_int(&fp->f_flag, FHASLOCK); 2722 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2723 goto done2; 2724 } 2725 if (uap->how & LOCK_EX) 2726 lf.l_type = F_WRLCK; 2727 else if (uap->how & LOCK_SH) 2728 lf.l_type = F_RDLCK; 2729 else { 2730 error = EBADF; 2731 goto done2; 2732 } 2733 atomic_set_int(&fp->f_flag, FHASLOCK); 2734 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2735 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2736done2: 2737 fdrop(fp, td); 2738 return (error); 2739} 2740/* 2741 * Duplicate the specified descriptor to a free descriptor. 2742 */ 2743int 2744dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2745 int openerror, int *indxp) 2746{ 2747 struct file *fp; 2748 int error, indx; 2749 2750 KASSERT(openerror == ENODEV || openerror == ENXIO, 2751 ("unexpected error %d in %s", openerror, __func__)); 2752 2753 /* 2754 * If the to-be-dup'd fd number is greater than the allowed number 2755 * of file descriptors, or the fd to be dup'd has already been 2756 * closed, then reject. 2757 */ 2758 FILEDESC_XLOCK(fdp); 2759 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2760 FILEDESC_XUNLOCK(fdp); 2761 return (EBADF); 2762 } 2763 2764 error = fdalloc(td, 0, &indx); 2765 if (error != 0) { 2766 FILEDESC_XUNLOCK(fdp); 2767 return (error); 2768 } 2769 2770 /* 2771 * There are two cases of interest here. 2772 * 2773 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2774 * 2775 * For ENXIO steal away the file structure from (dfd) and store it in 2776 * (indx). (dfd) is effectively closed by this operation. 2777 */ 2778 switch (openerror) { 2779 case ENODEV: 2780 /* 2781 * Check that the mode the file is being opened for is a 2782 * subset of the mode of the existing descriptor. 2783 */ 2784 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2785 fdunused(fdp, indx); 2786 FILEDESC_XUNLOCK(fdp); 2787 return (EACCES); 2788 } 2789 fhold(fp); 2790 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2791 filecaps_copy(&fdp->fd_ofiles[dfd].fde_caps, 2792 &fdp->fd_ofiles[indx].fde_caps); 2793 break; 2794 case ENXIO: 2795 /* 2796 * Steal away the file pointer from dfd and stuff it into indx. 2797 */ 2798 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2799 bzero(&fdp->fd_ofiles[dfd], sizeof(fdp->fd_ofiles[dfd])); 2800 fdunused(fdp, dfd); 2801 break; 2802 } 2803 FILEDESC_XUNLOCK(fdp); 2804 *indxp = indx; 2805 return (0); 2806} 2807 2808/* 2809 * Scan all active processes and prisons to see if any of them have a current 2810 * or root directory of `olddp'. If so, replace them with the new mount point. 2811 */ 2812void 2813mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2814{ 2815 struct filedesc *fdp; 2816 struct prison *pr; 2817 struct proc *p; 2818 int nrele; 2819 2820 if (vrefcnt(olddp) == 1) 2821 return; 2822 nrele = 0; 2823 sx_slock(&allproc_lock); 2824 FOREACH_PROC_IN_SYSTEM(p) { 2825 fdp = fdhold(p); 2826 if (fdp == NULL) 2827 continue; 2828 FILEDESC_XLOCK(fdp); 2829 if (fdp->fd_cdir == olddp) { 2830 vref(newdp); 2831 fdp->fd_cdir = newdp; 2832 nrele++; 2833 } 2834 if (fdp->fd_rdir == olddp) { 2835 vref(newdp); 2836 fdp->fd_rdir = newdp; 2837 nrele++; 2838 } 2839 if (fdp->fd_jdir == olddp) { 2840 vref(newdp); 2841 fdp->fd_jdir = newdp; 2842 nrele++; 2843 } 2844 FILEDESC_XUNLOCK(fdp); 2845 fddrop(fdp); 2846 } 2847 sx_sunlock(&allproc_lock); 2848 if (rootvnode == olddp) { 2849 vref(newdp); 2850 rootvnode = newdp; 2851 nrele++; 2852 } 2853 mtx_lock(&prison0.pr_mtx); 2854 if (prison0.pr_root == olddp) { 2855 vref(newdp); 2856 prison0.pr_root = newdp; 2857 nrele++; 2858 } 2859 mtx_unlock(&prison0.pr_mtx); 2860 sx_slock(&allprison_lock); 2861 TAILQ_FOREACH(pr, &allprison, pr_list) { 2862 mtx_lock(&pr->pr_mtx); 2863 if (pr->pr_root == olddp) { 2864 vref(newdp); 2865 pr->pr_root = newdp; 2866 nrele++; 2867 } 2868 mtx_unlock(&pr->pr_mtx); 2869 } 2870 sx_sunlock(&allprison_lock); 2871 while (nrele--) 2872 vrele(olddp); 2873} 2874 2875struct filedesc_to_leader * 2876filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2877{ 2878 struct filedesc_to_leader *fdtol; 2879 2880 fdtol = malloc(sizeof(struct filedesc_to_leader), 2881 M_FILEDESC_TO_LEADER, 2882 M_WAITOK); 2883 fdtol->fdl_refcount = 1; 2884 fdtol->fdl_holdcount = 0; 2885 fdtol->fdl_wakeup = 0; 2886 fdtol->fdl_leader = leader; 2887 if (old != NULL) { 2888 FILEDESC_XLOCK(fdp); 2889 fdtol->fdl_next = old->fdl_next; 2890 fdtol->fdl_prev = old; 2891 old->fdl_next = fdtol; 2892 fdtol->fdl_next->fdl_prev = fdtol; 2893 FILEDESC_XUNLOCK(fdp); 2894 } else { 2895 fdtol->fdl_next = fdtol; 2896 fdtol->fdl_prev = fdtol; 2897 } 2898 return (fdtol); 2899} 2900 2901/* 2902 * Get file structures globally. 2903 */ 2904static int 2905sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2906{ 2907 struct xfile xf; 2908 struct filedesc *fdp; 2909 struct file *fp; 2910 struct proc *p; 2911 int error, n; 2912 2913 error = sysctl_wire_old_buffer(req, 0); 2914 if (error != 0) 2915 return (error); 2916 if (req->oldptr == NULL) { 2917 n = 0; 2918 sx_slock(&allproc_lock); 2919 FOREACH_PROC_IN_SYSTEM(p) { 2920 if (p->p_state == PRS_NEW) 2921 continue; 2922 fdp = fdhold(p); 2923 if (fdp == NULL) 2924 continue; 2925 /* overestimates sparse tables. */ 2926 if (fdp->fd_lastfile > 0) 2927 n += fdp->fd_lastfile; 2928 fddrop(fdp); 2929 } 2930 sx_sunlock(&allproc_lock); 2931 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2932 } 2933 error = 0; 2934 bzero(&xf, sizeof(xf)); 2935 xf.xf_size = sizeof(xf); 2936 sx_slock(&allproc_lock); 2937 FOREACH_PROC_IN_SYSTEM(p) { 2938 PROC_LOCK(p); 2939 if (p->p_state == PRS_NEW) { 2940 PROC_UNLOCK(p); 2941 continue; 2942 } 2943 if (p_cansee(req->td, p) != 0) { 2944 PROC_UNLOCK(p); 2945 continue; 2946 } 2947 xf.xf_pid = p->p_pid; 2948 xf.xf_uid = p->p_ucred->cr_uid; 2949 PROC_UNLOCK(p); 2950 fdp = fdhold(p); 2951 if (fdp == NULL) 2952 continue; 2953 FILEDESC_SLOCK(fdp); 2954 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2955 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 2956 continue; 2957 xf.xf_fd = n; 2958 xf.xf_file = fp; 2959 xf.xf_data = fp->f_data; 2960 xf.xf_vnode = fp->f_vnode; 2961 xf.xf_type = fp->f_type; 2962 xf.xf_count = fp->f_count; 2963 xf.xf_msgcount = 0; 2964 xf.xf_offset = foffset_get(fp); 2965 xf.xf_flag = fp->f_flag; 2966 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2967 if (error) 2968 break; 2969 } 2970 FILEDESC_SUNLOCK(fdp); 2971 fddrop(fdp); 2972 if (error) 2973 break; 2974 } 2975 sx_sunlock(&allproc_lock); 2976 return (error); 2977} 2978 2979SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2980 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2981 2982#ifdef KINFO_OFILE_SIZE 2983CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2984#endif 2985 2986#ifdef COMPAT_FREEBSD7 2987static int 2988export_vnode_for_osysctl(struct vnode *vp, int type, 2989 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2990{ 2991 int error; 2992 char *fullpath, *freepath; 2993 2994 bzero(kif, sizeof(*kif)); 2995 kif->kf_structsize = sizeof(*kif); 2996 2997 vref(vp); 2998 kif->kf_fd = type; 2999 kif->kf_type = KF_TYPE_VNODE; 3000 /* This function only handles directories. */ 3001 if (vp->v_type != VDIR) { 3002 vrele(vp); 3003 return (ENOTDIR); 3004 } 3005 kif->kf_vnode_type = KF_VTYPE_VDIR; 3006 3007 /* 3008 * This is not a true file descriptor, so we set a bogus refcount 3009 * and offset to indicate these fields should be ignored. 3010 */ 3011 kif->kf_ref_count = -1; 3012 kif->kf_offset = -1; 3013 3014 freepath = NULL; 3015 fullpath = "-"; 3016 FILEDESC_SUNLOCK(fdp); 3017 vn_fullpath(curthread, vp, &fullpath, &freepath); 3018 vrele(vp); 3019 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3020 if (freepath != NULL) 3021 free(freepath, M_TEMP); 3022 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3023 FILEDESC_SLOCK(fdp); 3024 return (error); 3025} 3026 3027/* 3028 * Get per-process file descriptors for use by procstat(1), et al. 3029 */ 3030static int 3031sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3032{ 3033 char *fullpath, *freepath; 3034 struct kinfo_ofile *kif; 3035 struct filedesc *fdp; 3036 int error, i, *name; 3037 struct shmfd *shmfd; 3038 struct socket *so; 3039 struct vnode *vp; 3040 struct ksem *ks; 3041 struct file *fp; 3042 struct proc *p; 3043 struct tty *tp; 3044 3045 name = (int *)arg1; 3046 error = pget((pid_t)name[0], PGET_CANDEBUG, &p); 3047 if (error != 0) 3048 return (error); 3049 fdp = fdhold(p); 3050 PROC_UNLOCK(p); 3051 if (fdp == NULL) 3052 return (ENOENT); 3053 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3054 FILEDESC_SLOCK(fdp); 3055 if (fdp->fd_cdir != NULL) 3056 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3057 fdp, req); 3058 if (fdp->fd_rdir != NULL) 3059 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3060 fdp, req); 3061 if (fdp->fd_jdir != NULL) 3062 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3063 fdp, req); 3064 for (i = 0; fdp->fd_refcnt > 0 && i < fdp->fd_nfiles; i++) { 3065 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3066 continue; 3067 bzero(kif, sizeof(*kif)); 3068 kif->kf_structsize = sizeof(*kif); 3069 ks = NULL; 3070 vp = NULL; 3071 so = NULL; 3072 tp = NULL; 3073 shmfd = NULL; 3074 kif->kf_fd = i; 3075 3076 switch (fp->f_type) { 3077 case DTYPE_VNODE: 3078 kif->kf_type = KF_TYPE_VNODE; 3079 vp = fp->f_vnode; 3080 break; 3081 3082 case DTYPE_SOCKET: 3083 kif->kf_type = KF_TYPE_SOCKET; 3084 so = fp->f_data; 3085 break; 3086 3087 case DTYPE_PIPE: 3088 kif->kf_type = KF_TYPE_PIPE; 3089 break; 3090 3091 case DTYPE_FIFO: 3092 kif->kf_type = KF_TYPE_FIFO; 3093 vp = fp->f_vnode; 3094 break; 3095 3096 case DTYPE_KQUEUE: 3097 kif->kf_type = KF_TYPE_KQUEUE; 3098 break; 3099 3100 case DTYPE_CRYPTO: 3101 kif->kf_type = KF_TYPE_CRYPTO; 3102 break; 3103 3104 case DTYPE_MQUEUE: 3105 kif->kf_type = KF_TYPE_MQUEUE; 3106 break; 3107 3108 case DTYPE_SHM: 3109 kif->kf_type = KF_TYPE_SHM; 3110 shmfd = fp->f_data; 3111 break; 3112 3113 case DTYPE_SEM: 3114 kif->kf_type = KF_TYPE_SEM; 3115 ks = fp->f_data; 3116 break; 3117 3118 case DTYPE_PTS: 3119 kif->kf_type = KF_TYPE_PTS; 3120 tp = fp->f_data; 3121 break; 3122 3123#ifdef PROCDESC 3124 case DTYPE_PROCDESC: 3125 kif->kf_type = KF_TYPE_PROCDESC; 3126 break; 3127#endif 3128 3129 default: 3130 kif->kf_type = KF_TYPE_UNKNOWN; 3131 break; 3132 } 3133 kif->kf_ref_count = fp->f_count; 3134 if (fp->f_flag & FREAD) 3135 kif->kf_flags |= KF_FLAG_READ; 3136 if (fp->f_flag & FWRITE) 3137 kif->kf_flags |= KF_FLAG_WRITE; 3138 if (fp->f_flag & FAPPEND) 3139 kif->kf_flags |= KF_FLAG_APPEND; 3140 if (fp->f_flag & FASYNC) 3141 kif->kf_flags |= KF_FLAG_ASYNC; 3142 if (fp->f_flag & FFSYNC) 3143 kif->kf_flags |= KF_FLAG_FSYNC; 3144 if (fp->f_flag & FNONBLOCK) 3145 kif->kf_flags |= KF_FLAG_NONBLOCK; 3146 if (fp->f_flag & O_DIRECT) 3147 kif->kf_flags |= KF_FLAG_DIRECT; 3148 if (fp->f_flag & FHASLOCK) 3149 kif->kf_flags |= KF_FLAG_HASLOCK; 3150 kif->kf_offset = foffset_get(fp); 3151 if (vp != NULL) { 3152 vref(vp); 3153 switch (vp->v_type) { 3154 case VNON: 3155 kif->kf_vnode_type = KF_VTYPE_VNON; 3156 break; 3157 case VREG: 3158 kif->kf_vnode_type = KF_VTYPE_VREG; 3159 break; 3160 case VDIR: 3161 kif->kf_vnode_type = KF_VTYPE_VDIR; 3162 break; 3163 case VBLK: 3164 kif->kf_vnode_type = KF_VTYPE_VBLK; 3165 break; 3166 case VCHR: 3167 kif->kf_vnode_type = KF_VTYPE_VCHR; 3168 break; 3169 case VLNK: 3170 kif->kf_vnode_type = KF_VTYPE_VLNK; 3171 break; 3172 case VSOCK: 3173 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3174 break; 3175 case VFIFO: 3176 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3177 break; 3178 case VBAD: 3179 kif->kf_vnode_type = KF_VTYPE_VBAD; 3180 break; 3181 default: 3182 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3183 break; 3184 } 3185 /* 3186 * It is OK to drop the filedesc lock here as we will 3187 * re-validate and re-evaluate its properties when 3188 * the loop continues. 3189 */ 3190 freepath = NULL; 3191 fullpath = "-"; 3192 FILEDESC_SUNLOCK(fdp); 3193 vn_fullpath(curthread, vp, &fullpath, &freepath); 3194 vrele(vp); 3195 strlcpy(kif->kf_path, fullpath, 3196 sizeof(kif->kf_path)); 3197 if (freepath != NULL) 3198 free(freepath, M_TEMP); 3199 FILEDESC_SLOCK(fdp); 3200 } 3201 if (so != NULL) { 3202 struct sockaddr *sa; 3203 3204 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3205 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3206 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3207 free(sa, M_SONAME); 3208 } 3209 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3210 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3211 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3212 free(sa, M_SONAME); 3213 } 3214 kif->kf_sock_domain = 3215 so->so_proto->pr_domain->dom_family; 3216 kif->kf_sock_type = so->so_type; 3217 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3218 } 3219 if (tp != NULL) { 3220 strlcpy(kif->kf_path, tty_devname(tp), 3221 sizeof(kif->kf_path)); 3222 } 3223 if (shmfd != NULL) 3224 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3225 if (ks != NULL && ksem_info != NULL) 3226 ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL); 3227 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3228 if (error) 3229 break; 3230 } 3231 FILEDESC_SUNLOCK(fdp); 3232 fddrop(fdp); 3233 free(kif, M_TEMP); 3234 return (0); 3235} 3236 3237static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD, 3238 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); 3239#endif /* COMPAT_FREEBSD7 */ 3240 3241#ifdef KINFO_FILE_SIZE 3242CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3243#endif 3244 3245struct export_fd_buf { 3246 struct filedesc *fdp; 3247 struct sbuf *sb; 3248 ssize_t remainder; 3249 struct kinfo_file kif; 3250}; 3251 3252static int 3253export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt, 3254 int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf) 3255{ 3256 struct { 3257 int fflag; 3258 int kf_fflag; 3259 } fflags_table[] = { 3260 { FAPPEND, KF_FLAG_APPEND }, 3261 { FASYNC, KF_FLAG_ASYNC }, 3262 { FFSYNC, KF_FLAG_FSYNC }, 3263 { FHASLOCK, KF_FLAG_HASLOCK }, 3264 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3265 { FREAD, KF_FLAG_READ }, 3266 { FWRITE, KF_FLAG_WRITE }, 3267 { O_CREAT, KF_FLAG_CREAT }, 3268 { O_DIRECT, KF_FLAG_DIRECT }, 3269 { O_EXCL, KF_FLAG_EXCL }, 3270 { O_EXEC, KF_FLAG_EXEC }, 3271 { O_EXLOCK, KF_FLAG_EXLOCK }, 3272 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3273 { O_SHLOCK, KF_FLAG_SHLOCK }, 3274 { O_TRUNC, KF_FLAG_TRUNC } 3275 }; 3276#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3277 struct kinfo_file *kif; 3278 struct vnode *vp; 3279 int error, locked; 3280 unsigned int i; 3281 3282 if (efbuf->remainder == 0) 3283 return (0); 3284 kif = &efbuf->kif; 3285 bzero(kif, sizeof(*kif)); 3286 locked = efbuf->fdp != NULL; 3287 switch (type) { 3288 case KF_TYPE_FIFO: 3289 case KF_TYPE_VNODE: 3290 if (locked) { 3291 FILEDESC_SUNLOCK(efbuf->fdp); 3292 locked = 0; 3293 } 3294 vp = (struct vnode *)data; 3295 error = fill_vnode_info(vp, kif); 3296 vrele(vp); 3297 break; 3298 case KF_TYPE_SOCKET: 3299 error = fill_socket_info((struct socket *)data, kif); 3300 break; 3301 case KF_TYPE_PIPE: 3302 error = fill_pipe_info((struct pipe *)data, kif); 3303 break; 3304 case KF_TYPE_PTS: 3305 error = fill_pts_info((struct tty *)data, kif); 3306 break; 3307 case KF_TYPE_PROCDESC: 3308 error = fill_procdesc_info((struct procdesc *)data, kif); 3309 break; 3310 case KF_TYPE_SEM: 3311 error = fill_sem_info((struct file *)data, kif); 3312 break; 3313 case KF_TYPE_SHM: 3314 error = fill_shm_info((struct file *)data, kif); 3315 break; 3316 default: 3317 error = 0; 3318 } 3319 if (error == 0) 3320 kif->kf_status |= KF_ATTR_VALID; 3321 3322 /* 3323 * Translate file access flags. 3324 */ 3325 for (i = 0; i < NFFLAGS; i++) 3326 if (fflags & fflags_table[i].fflag) 3327 kif->kf_flags |= fflags_table[i].kf_fflag; 3328 if (rightsp != NULL) 3329 kif->kf_cap_rights = *rightsp; 3330 else 3331 cap_rights_init(&kif->kf_cap_rights); 3332 kif->kf_fd = fd; 3333 kif->kf_type = type; 3334 kif->kf_ref_count = refcnt; 3335 kif->kf_offset = offset; 3336 /* Pack record size down */ 3337 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3338 strlen(kif->kf_path) + 1; 3339 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3340 if (efbuf->remainder != -1) { 3341 if (efbuf->remainder < kif->kf_structsize) { 3342 /* Terminate export. */ 3343 efbuf->remainder = 0; 3344 if (efbuf->fdp != NULL && !locked) 3345 FILEDESC_SLOCK(efbuf->fdp); 3346 return (0); 3347 } 3348 efbuf->remainder -= kif->kf_structsize; 3349 } 3350 if (locked) 3351 FILEDESC_SUNLOCK(efbuf->fdp); 3352 error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize); 3353 if (efbuf->fdp != NULL) 3354 FILEDESC_SLOCK(efbuf->fdp); 3355 return (error); 3356} 3357 3358/* 3359 * Store a process file descriptor information to sbuf. 3360 * 3361 * Takes a locked proc as argument, and returns with the proc unlocked. 3362 */ 3363int 3364kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3365{ 3366 struct file *fp; 3367 struct filedesc *fdp; 3368 struct export_fd_buf *efbuf; 3369 struct vnode *cttyvp, *textvp, *tracevp; 3370 int64_t offset; 3371 void *data; 3372 int error, i; 3373 int type, refcnt, fflags; 3374 cap_rights_t rights; 3375 3376 PROC_LOCK_ASSERT(p, MA_OWNED); 3377 3378 /* ktrace vnode */ 3379 tracevp = p->p_tracevp; 3380 if (tracevp != NULL) 3381 vref(tracevp); 3382 /* text vnode */ 3383 textvp = p->p_textvp; 3384 if (textvp != NULL) 3385 vref(textvp); 3386 /* Controlling tty. */ 3387 cttyvp = NULL; 3388 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3389 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3390 if (cttyvp != NULL) 3391 vref(cttyvp); 3392 } 3393 fdp = fdhold(p); 3394 PROC_UNLOCK(p); 3395 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3396 efbuf->fdp = NULL; 3397 efbuf->sb = sb; 3398 efbuf->remainder = maxlen; 3399 if (tracevp != NULL) 3400 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3401 FREAD | FWRITE, -1, -1, NULL, efbuf); 3402 if (textvp != NULL) 3403 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3404 FREAD, -1, -1, NULL, efbuf); 3405 if (cttyvp != NULL) 3406 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3407 FREAD | FWRITE, -1, -1, NULL, efbuf); 3408 error = 0; 3409 if (fdp == NULL) 3410 goto fail; 3411 efbuf->fdp = fdp; 3412 FILEDESC_SLOCK(fdp); 3413 /* working directory */ 3414 if (fdp->fd_cdir != NULL) { 3415 vref(fdp->fd_cdir); 3416 data = fdp->fd_cdir; 3417 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3418 FREAD, -1, -1, NULL, efbuf); 3419 } 3420 /* root directory */ 3421 if (fdp->fd_rdir != NULL) { 3422 vref(fdp->fd_rdir); 3423 data = fdp->fd_rdir; 3424 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3425 FREAD, -1, -1, NULL, efbuf); 3426 } 3427 /* jail directory */ 3428 if (fdp->fd_jdir != NULL) { 3429 vref(fdp->fd_jdir); 3430 data = fdp->fd_jdir; 3431 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3432 FREAD, -1, -1, NULL, efbuf); 3433 } 3434 for (i = 0; fdp->fd_refcnt > 0 && i < fdp->fd_nfiles; i++) { 3435 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3436 continue; 3437 data = NULL; 3438#ifdef CAPABILITIES 3439 rights = *cap_rights(fdp, i); 3440#else /* !CAPABILITIES */ 3441 cap_rights_init(&rights); 3442#endif 3443 switch (fp->f_type) { 3444 case DTYPE_VNODE: 3445 type = KF_TYPE_VNODE; 3446 vref(fp->f_vnode); 3447 data = fp->f_vnode; 3448 break; 3449 3450 case DTYPE_SOCKET: 3451 type = KF_TYPE_SOCKET; 3452 data = fp->f_data; 3453 break; 3454 3455 case DTYPE_PIPE: 3456 type = KF_TYPE_PIPE; 3457 data = fp->f_data; 3458 break; 3459 3460 case DTYPE_FIFO: 3461 type = KF_TYPE_FIFO; 3462 vref(fp->f_vnode); 3463 data = fp->f_vnode; 3464 break; 3465 3466 case DTYPE_KQUEUE: 3467 type = KF_TYPE_KQUEUE; 3468 break; 3469 3470 case DTYPE_CRYPTO: 3471 type = KF_TYPE_CRYPTO; 3472 break; 3473 3474 case DTYPE_MQUEUE: 3475 type = KF_TYPE_MQUEUE; 3476 break; 3477 3478 case DTYPE_SHM: 3479 type = KF_TYPE_SHM; 3480 data = fp; 3481 break; 3482 3483 case DTYPE_SEM: 3484 type = KF_TYPE_SEM; 3485 data = fp; 3486 break; 3487 3488 case DTYPE_PTS: 3489 type = KF_TYPE_PTS; 3490 data = fp->f_data; 3491 break; 3492 3493#ifdef PROCDESC 3494 case DTYPE_PROCDESC: 3495 type = KF_TYPE_PROCDESC; 3496 data = fp->f_data; 3497 break; 3498#endif 3499 3500 default: 3501 type = KF_TYPE_UNKNOWN; 3502 break; 3503 } 3504 refcnt = fp->f_count; 3505 fflags = fp->f_flag; 3506 offset = foffset_get(fp); 3507 3508 /* 3509 * Create sysctl entry. 3510 * It is OK to drop the filedesc lock here as we will 3511 * re-validate and re-evaluate its properties when 3512 * the loop continues. 3513 */ 3514 error = export_fd_to_sb(data, type, i, fflags, refcnt, 3515 offset, &rights, efbuf); 3516 if (error != 0) 3517 break; 3518 } 3519 FILEDESC_SUNLOCK(fdp); 3520 fddrop(fdp); 3521fail: 3522 free(efbuf, M_TEMP); 3523 return (error); 3524} 3525 3526#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3527 3528/* 3529 * Get per-process file descriptors for use by procstat(1), et al. 3530 */ 3531static int 3532sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3533{ 3534 struct sbuf sb; 3535 struct proc *p; 3536 ssize_t maxlen; 3537 int error, error2, *name; 3538 3539 name = (int *)arg1; 3540 3541 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3542 error = pget((pid_t)name[0], PGET_CANDEBUG, &p); 3543 if (error != 0) { 3544 sbuf_delete(&sb); 3545 return (error); 3546 } 3547 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3548 error = kern_proc_filedesc_out(p, &sb, maxlen); 3549 error2 = sbuf_finish(&sb); 3550 sbuf_delete(&sb); 3551 return (error != 0 ? error : error2); 3552} 3553 3554int 3555vntype_to_kinfo(int vtype) 3556{ 3557 struct { 3558 int vtype; 3559 int kf_vtype; 3560 } vtypes_table[] = { 3561 { VBAD, KF_VTYPE_VBAD }, 3562 { VBLK, KF_VTYPE_VBLK }, 3563 { VCHR, KF_VTYPE_VCHR }, 3564 { VDIR, KF_VTYPE_VDIR }, 3565 { VFIFO, KF_VTYPE_VFIFO }, 3566 { VLNK, KF_VTYPE_VLNK }, 3567 { VNON, KF_VTYPE_VNON }, 3568 { VREG, KF_VTYPE_VREG }, 3569 { VSOCK, KF_VTYPE_VSOCK } 3570 }; 3571#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3572 unsigned int i; 3573 3574 /* 3575 * Perform vtype translation. 3576 */ 3577 for (i = 0; i < NVTYPES; i++) 3578 if (vtypes_table[i].vtype == vtype) 3579 break; 3580 if (i < NVTYPES) 3581 return (vtypes_table[i].kf_vtype); 3582 3583 return (KF_VTYPE_UNKNOWN); 3584} 3585 3586static int 3587fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3588{ 3589 struct vattr va; 3590 char *fullpath, *freepath; 3591 int error; 3592 3593 if (vp == NULL) 3594 return (1); 3595 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3596 freepath = NULL; 3597 fullpath = "-"; 3598 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3599 if (error == 0) { 3600 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3601 } 3602 if (freepath != NULL) 3603 free(freepath, M_TEMP); 3604 3605 /* 3606 * Retrieve vnode attributes. 3607 */ 3608 va.va_fsid = VNOVAL; 3609 va.va_rdev = NODEV; 3610 vn_lock(vp, LK_SHARED | LK_RETRY); 3611 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3612 VOP_UNLOCK(vp, 0); 3613 if (error != 0) 3614 return (error); 3615 if (va.va_fsid != VNOVAL) 3616 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3617 else 3618 kif->kf_un.kf_file.kf_file_fsid = 3619 vp->v_mount->mnt_stat.f_fsid.val[0]; 3620 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3621 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3622 kif->kf_un.kf_file.kf_file_size = va.va_size; 3623 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3624 return (0); 3625} 3626 3627static int 3628fill_socket_info(struct socket *so, struct kinfo_file *kif) 3629{ 3630 struct sockaddr *sa; 3631 struct inpcb *inpcb; 3632 struct unpcb *unpcb; 3633 int error; 3634 3635 if (so == NULL) 3636 return (1); 3637 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3638 kif->kf_sock_type = so->so_type; 3639 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3640 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3641 switch(kif->kf_sock_domain) { 3642 case AF_INET: 3643 case AF_INET6: 3644 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3645 if (so->so_pcb != NULL) { 3646 inpcb = (struct inpcb *)(so->so_pcb); 3647 kif->kf_un.kf_sock.kf_sock_inpcb = 3648 (uintptr_t)inpcb->inp_ppcb; 3649 } 3650 } 3651 break; 3652 case AF_UNIX: 3653 if (so->so_pcb != NULL) { 3654 unpcb = (struct unpcb *)(so->so_pcb); 3655 if (unpcb->unp_conn) { 3656 kif->kf_un.kf_sock.kf_sock_unpconn = 3657 (uintptr_t)unpcb->unp_conn; 3658 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3659 so->so_rcv.sb_state; 3660 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3661 so->so_snd.sb_state; 3662 } 3663 } 3664 break; 3665 } 3666 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3667 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3668 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3669 free(sa, M_SONAME); 3670 } 3671 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3672 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3673 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3674 free(sa, M_SONAME); 3675 } 3676 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3677 sizeof(kif->kf_path)); 3678 return (0); 3679} 3680 3681static int 3682fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3683{ 3684 3685 if (tp == NULL) 3686 return (1); 3687 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3688 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3689 return (0); 3690} 3691 3692static int 3693fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3694{ 3695 3696 if (pi == NULL) 3697 return (1); 3698 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3699 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3700 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3701 return (0); 3702} 3703 3704static int 3705fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3706{ 3707 3708 if (pdp == NULL) 3709 return (1); 3710 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3711 return (0); 3712} 3713 3714static int 3715fill_sem_info(struct file *fp, struct kinfo_file *kif) 3716{ 3717 struct thread *td; 3718 struct stat sb; 3719 3720 td = curthread; 3721 if (fp->f_data == NULL) 3722 return (1); 3723 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3724 return (1); 3725 if (ksem_info == NULL) 3726 return (1); 3727 ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path), 3728 &kif->kf_un.kf_sem.kf_sem_value); 3729 kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode; 3730 return (0); 3731} 3732 3733static int 3734fill_shm_info(struct file *fp, struct kinfo_file *kif) 3735{ 3736 struct thread *td; 3737 struct stat sb; 3738 3739 td = curthread; 3740 if (fp->f_data == NULL) 3741 return (1); 3742 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3743 return (1); 3744 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3745 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3746 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3747 return (0); 3748} 3749 3750static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD, 3751 sysctl_kern_proc_filedesc, "Process filedesc entries"); 3752 3753#ifdef DDB 3754/* 3755 * For the purposes of debugging, generate a human-readable string for the 3756 * file type. 3757 */ 3758static const char * 3759file_type_to_name(short type) 3760{ 3761 3762 switch (type) { 3763 case 0: 3764 return ("zero"); 3765 case DTYPE_VNODE: 3766 return ("vnod"); 3767 case DTYPE_SOCKET: 3768 return ("sock"); 3769 case DTYPE_PIPE: 3770 return ("pipe"); 3771 case DTYPE_FIFO: 3772 return ("fifo"); 3773 case DTYPE_KQUEUE: 3774 return ("kque"); 3775 case DTYPE_CRYPTO: 3776 return ("crpt"); 3777 case DTYPE_MQUEUE: 3778 return ("mque"); 3779 case DTYPE_SHM: 3780 return ("shm"); 3781 case DTYPE_SEM: 3782 return ("ksem"); 3783 default: 3784 return ("unkn"); 3785 } 3786} 3787 3788/* 3789 * For the purposes of debugging, identify a process (if any, perhaps one of 3790 * many) that references the passed file in its file descriptor array. Return 3791 * NULL if none. 3792 */ 3793static struct proc * 3794file_to_first_proc(struct file *fp) 3795{ 3796 struct filedesc *fdp; 3797 struct proc *p; 3798 int n; 3799 3800 FOREACH_PROC_IN_SYSTEM(p) { 3801 if (p->p_state == PRS_NEW) 3802 continue; 3803 fdp = p->p_fd; 3804 if (fdp == NULL) 3805 continue; 3806 for (n = 0; n < fdp->fd_nfiles; n++) { 3807 if (fp == fdp->fd_ofiles[n].fde_file) 3808 return (p); 3809 } 3810 } 3811 return (NULL); 3812} 3813 3814static void 3815db_print_file(struct file *fp, int header) 3816{ 3817 struct proc *p; 3818 3819 if (header) 3820 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3821 "File", "Type", "Data", "Flag", "GCFl", "Count", 3822 "MCount", "Vnode", "FPID", "FCmd"); 3823 p = file_to_first_proc(fp); 3824 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3825 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3826 0, fp->f_count, 0, fp->f_vnode, 3827 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3828} 3829 3830DB_SHOW_COMMAND(file, db_show_file) 3831{ 3832 struct file *fp; 3833 3834 if (!have_addr) { 3835 db_printf("usage: show file <addr>\n"); 3836 return; 3837 } 3838 fp = (struct file *)addr; 3839 db_print_file(fp, 1); 3840} 3841 3842DB_SHOW_COMMAND(files, db_show_files) 3843{ 3844 struct filedesc *fdp; 3845 struct file *fp; 3846 struct proc *p; 3847 int header; 3848 int n; 3849 3850 header = 1; 3851 FOREACH_PROC_IN_SYSTEM(p) { 3852 if (p->p_state == PRS_NEW) 3853 continue; 3854 if ((fdp = p->p_fd) == NULL) 3855 continue; 3856 for (n = 0; n < fdp->fd_nfiles; ++n) { 3857 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3858 continue; 3859 db_print_file(fp, header); 3860 header = 0; 3861 } 3862 } 3863} 3864#endif 3865 3866SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3867 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3868 3869SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3870 &maxfiles, 0, "Maximum number of files"); 3871 3872SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3873 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3874 3875/* ARGSUSED*/ 3876static void 3877filelistinit(void *dummy) 3878{ 3879 3880 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3881 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3882 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3883 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3884} 3885SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3886 3887/*-------------------------------------------------------------------*/ 3888 3889static int 3890badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3891 int flags, struct thread *td) 3892{ 3893 3894 return (EBADF); 3895} 3896 3897static int 3898badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3899 struct thread *td) 3900{ 3901 3902 return (EINVAL); 3903} 3904 3905static int 3906badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3907 struct thread *td) 3908{ 3909 3910 return (EBADF); 3911} 3912 3913static int 3914badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3915 struct thread *td) 3916{ 3917 3918 return (0); 3919} 3920 3921static int 3922badfo_kqfilter(struct file *fp, struct knote *kn) 3923{ 3924 3925 return (EBADF); 3926} 3927 3928static int 3929badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3930 struct thread *td) 3931{ 3932 3933 return (EBADF); 3934} 3935 3936static int 3937badfo_close(struct file *fp, struct thread *td) 3938{ 3939 3940 return (EBADF); 3941} 3942 3943static int 3944badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3945 struct thread *td) 3946{ 3947 3948 return (EBADF); 3949} 3950 3951static int 3952badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3953 struct thread *td) 3954{ 3955 3956 return (EBADF); 3957} 3958 3959static int 3960badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3961 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3962 int kflags, struct thread *td) 3963{ 3964 3965 return (EBADF); 3966} 3967 3968struct fileops badfileops = { 3969 .fo_read = badfo_readwrite, 3970 .fo_write = badfo_readwrite, 3971 .fo_truncate = badfo_truncate, 3972 .fo_ioctl = badfo_ioctl, 3973 .fo_poll = badfo_poll, 3974 .fo_kqfilter = badfo_kqfilter, 3975 .fo_stat = badfo_stat, 3976 .fo_close = badfo_close, 3977 .fo_chmod = badfo_chmod, 3978 .fo_chown = badfo_chown, 3979 .fo_sendfile = badfo_sendfile, 3980}; 3981 3982int 3983invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3984 struct thread *td) 3985{ 3986 3987 return (EINVAL); 3988} 3989 3990int 3991invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3992 struct thread *td) 3993{ 3994 3995 return (EINVAL); 3996} 3997 3998int 3999invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 4000 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 4001 int kflags, struct thread *td) 4002{ 4003 4004 return (EINVAL); 4005} 4006 4007/*-------------------------------------------------------------------*/ 4008 4009/* 4010 * File Descriptor pseudo-device driver (/dev/fd/). 4011 * 4012 * Opening minor device N dup()s the file (if any) connected to file 4013 * descriptor N belonging to the calling process. Note that this driver 4014 * consists of only the ``open()'' routine, because all subsequent 4015 * references to this file will be direct to the other driver. 4016 * 4017 * XXX: we could give this one a cloning event handler if necessary. 4018 */ 4019 4020/* ARGSUSED */ 4021static int 4022fdopen(struct cdev *dev, int mode, int type, struct thread *td) 4023{ 4024 4025 /* 4026 * XXX Kludge: set curthread->td_dupfd to contain the value of the 4027 * the file descriptor being sought for duplication. The error 4028 * return ensures that the vnode for this device will be released 4029 * by vn_open. Open will detect this special error and take the 4030 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 4031 * will simply report the error. 4032 */ 4033 td->td_dupfd = dev2unit(dev); 4034 return (ENODEV); 4035} 4036 4037static struct cdevsw fildesc_cdevsw = { 4038 .d_version = D_VERSION, 4039 .d_open = fdopen, 4040 .d_name = "FD", 4041}; 4042 4043static void 4044fildesc_drvinit(void *unused) 4045{ 4046 struct cdev *dev; 4047 4048 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4049 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4050 make_dev_alias(dev, "stdin"); 4051 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4052 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4053 make_dev_alias(dev, "stdout"); 4054 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4055 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4056 make_dev_alias(dev, "stderr"); 4057} 4058 4059SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4060