kern_descrip.c revision 262458
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 262458 2014-02-24 21:03:38Z 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 and map. 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 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1534 M_ZERO | M_WAITOK); 1535 1536 /* copy the old data over and point at the new tables */ 1537 memcpy(ntable, otable, onfiles * sizeof(*otable)); 1538 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1539 1540 /* update the pointers and counters */ 1541 memcpy(ntable, otable, onfiles * sizeof(ntable[0])); 1542 fdp->fd_ofiles = ntable; 1543 fdp->fd_map = nmap; 1544 1545 /* 1546 * In order to have a valid pattern for fget_unlocked() 1547 * fdp->fd_nfiles must be the last member to be updated, otherwise 1548 * fget_unlocked() consumers may reference a new, higher value for 1549 * fdp->fd_nfiles before to access the fdp->fd_ofiles array, 1550 * resulting in OOB accesses. 1551 */ 1552 atomic_store_rel_int(&fdp->fd_nfiles, nnfiles); 1553 1554 /* 1555 * Do not free the old file table, as some threads may still 1556 * reference entries within it. Instead, place it on a freelist 1557 * which will be processed when the struct filedesc is released. 1558 * 1559 * Do, however, free the old map. 1560 * 1561 * Note that if onfiles == NDFILE, we're dealing with the original 1562 * static allocation contained within (struct filedesc0 *)fdp, 1563 * which must not be freed. 1564 */ 1565 if (onfiles > NDFILE) { 1566 ft = (struct freetable *)&otable[onfiles]; 1567 fdp0 = (struct filedesc0 *)fdp; 1568 ft->ft_table = otable; 1569 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1570 free(omap, M_FILEDESC); 1571 } 1572} 1573 1574/* 1575 * Allocate a file descriptor for the process. 1576 */ 1577int 1578fdalloc(struct thread *td, int minfd, int *result) 1579{ 1580 struct proc *p = td->td_proc; 1581 struct filedesc *fdp = p->p_fd; 1582 int fd = -1, maxfd, allocfd; 1583#ifdef RACCT 1584 int error; 1585#endif 1586 1587 FILEDESC_XLOCK_ASSERT(fdp); 1588 1589 if (fdp->fd_freefile > minfd) 1590 minfd = fdp->fd_freefile; 1591 1592 maxfd = getmaxfd(p); 1593 1594 /* 1595 * Search the bitmap for a free descriptor starting at minfd. 1596 * If none is found, grow the file table. 1597 */ 1598 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1599 if (fd >= maxfd) 1600 return (EMFILE); 1601 if (fd >= fdp->fd_nfiles) { 1602 allocfd = min(fd * 2, maxfd); 1603#ifdef RACCT 1604 PROC_LOCK(p); 1605 error = racct_set(p, RACCT_NOFILE, allocfd); 1606 PROC_UNLOCK(p); 1607 if (error != 0) 1608 return (EMFILE); 1609#endif 1610 /* 1611 * fd is already equal to first free descriptor >= minfd, so 1612 * we only need to grow the table and we are done. 1613 */ 1614 fdgrowtable_exp(fdp, allocfd); 1615 } 1616 1617 /* 1618 * Perform some sanity checks, then mark the file descriptor as 1619 * used and return it to the caller. 1620 */ 1621 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1622 ("invalid descriptor %d", fd)); 1623 KASSERT(!fdisused(fdp, fd), 1624 ("fd_first_free() returned non-free descriptor")); 1625 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1626 ("file descriptor isn't free")); 1627 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); 1628 fdused(fdp, fd); 1629 *result = fd; 1630 return (0); 1631} 1632 1633/* 1634 * Allocate n file descriptors for the process. 1635 */ 1636int 1637fdallocn(struct thread *td, int minfd, int *fds, int n) 1638{ 1639 struct proc *p = td->td_proc; 1640 struct filedesc *fdp = p->p_fd; 1641 int i; 1642 1643 FILEDESC_XLOCK_ASSERT(fdp); 1644 1645 if (!fdavail(td, n)) 1646 return (EMFILE); 1647 1648 for (i = 0; i < n; i++) 1649 if (fdalloc(td, 0, &fds[i]) != 0) 1650 break; 1651 1652 if (i < n) { 1653 for (i--; i >= 0; i--) 1654 fdunused(fdp, fds[i]); 1655 return (EMFILE); 1656 } 1657 1658 return (0); 1659} 1660 1661/* 1662 * Check to see whether n user file descriptors are available to the process 1663 * p. 1664 */ 1665int 1666fdavail(struct thread *td, int n) 1667{ 1668 struct proc *p = td->td_proc; 1669 struct filedesc *fdp = td->td_proc->p_fd; 1670 int i, lim, last; 1671 1672 FILEDESC_LOCK_ASSERT(fdp); 1673 1674 /* 1675 * XXX: This is only called from uipc_usrreq.c:unp_externalize(); 1676 * call racct_add() from there instead of dealing with containers 1677 * here. 1678 */ 1679 lim = getmaxfd(p); 1680 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1681 return (1); 1682 last = min(fdp->fd_nfiles, lim); 1683 for (i = fdp->fd_freefile; i < last; i++) { 1684 if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0) 1685 return (1); 1686 } 1687 return (0); 1688} 1689 1690/* 1691 * Create a new open file structure and allocate a file decriptor for the 1692 * process that refers to it. We add one reference to the file for the 1693 * descriptor table and one reference for resultfp. This is to prevent us 1694 * being preempted and the entry in the descriptor table closed after we 1695 * release the FILEDESC lock. 1696 */ 1697int 1698falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1699{ 1700 struct file *fp; 1701 int error, fd; 1702 1703 error = falloc_noinstall(td, &fp); 1704 if (error) 1705 return (error); /* no reference held on error */ 1706 1707 error = finstall(td, fp, &fd, flags, NULL); 1708 if (error) { 1709 fdrop(fp, td); /* one reference (fp only) */ 1710 return (error); 1711 } 1712 1713 if (resultfp != NULL) 1714 *resultfp = fp; /* copy out result */ 1715 else 1716 fdrop(fp, td); /* release local reference */ 1717 1718 if (resultfd != NULL) 1719 *resultfd = fd; 1720 1721 return (0); 1722} 1723 1724/* 1725 * Create a new open file structure without allocating a file descriptor. 1726 */ 1727int 1728falloc_noinstall(struct thread *td, struct file **resultfp) 1729{ 1730 struct file *fp; 1731 int maxuserfiles = maxfiles - (maxfiles / 20); 1732 static struct timeval lastfail; 1733 static int curfail; 1734 1735 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1736 1737 if ((openfiles >= maxuserfiles && 1738 priv_check(td, PRIV_MAXFILES) != 0) || 1739 openfiles >= maxfiles) { 1740 if (ppsratecheck(&lastfail, &curfail, 1)) { 1741 printf("kern.maxfiles limit exceeded by uid %i, " 1742 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1743 } 1744 return (ENFILE); 1745 } 1746 atomic_add_int(&openfiles, 1); 1747 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1748 refcount_init(&fp->f_count, 1); 1749 fp->f_cred = crhold(td->td_ucred); 1750 fp->f_ops = &badfileops; 1751 fp->f_data = NULL; 1752 fp->f_vnode = NULL; 1753 *resultfp = fp; 1754 return (0); 1755} 1756 1757/* 1758 * Install a file in a file descriptor table. 1759 */ 1760int 1761finstall(struct thread *td, struct file *fp, int *fd, int flags, 1762 struct filecaps *fcaps) 1763{ 1764 struct filedesc *fdp = td->td_proc->p_fd; 1765 struct filedescent *fde; 1766 int error; 1767 1768 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1769 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1770 if (fcaps != NULL) 1771 filecaps_validate(fcaps, __func__); 1772 1773 FILEDESC_XLOCK(fdp); 1774 if ((error = fdalloc(td, 0, fd))) { 1775 FILEDESC_XUNLOCK(fdp); 1776 return (error); 1777 } 1778 fhold(fp); 1779 fde = &fdp->fd_ofiles[*fd]; 1780 fde->fde_file = fp; 1781 if ((flags & O_CLOEXEC) != 0) 1782 fde->fde_flags |= UF_EXCLOSE; 1783 if (fcaps != NULL) 1784 filecaps_move(fcaps, &fde->fde_caps); 1785 else 1786 filecaps_fill(&fde->fde_caps); 1787 FILEDESC_XUNLOCK(fdp); 1788 return (0); 1789} 1790 1791/* 1792 * Build a new filedesc structure from another. 1793 * Copy the current, root, and jail root vnode references. 1794 */ 1795struct filedesc * 1796fdinit(struct filedesc *fdp) 1797{ 1798 struct filedesc0 *newfdp; 1799 1800 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1801 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1802 if (fdp != NULL) { 1803 FILEDESC_XLOCK(fdp); 1804 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1805 if (newfdp->fd_fd.fd_cdir) 1806 VREF(newfdp->fd_fd.fd_cdir); 1807 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1808 if (newfdp->fd_fd.fd_rdir) 1809 VREF(newfdp->fd_fd.fd_rdir); 1810 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1811 if (newfdp->fd_fd.fd_jdir) 1812 VREF(newfdp->fd_fd.fd_jdir); 1813 FILEDESC_XUNLOCK(fdp); 1814 } 1815 1816 /* Create the file descriptor table. */ 1817 newfdp->fd_fd.fd_refcnt = 1; 1818 newfdp->fd_fd.fd_holdcnt = 1; 1819 newfdp->fd_fd.fd_cmask = CMASK; 1820 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1821 newfdp->fd_fd.fd_nfiles = NDFILE; 1822 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1823 newfdp->fd_fd.fd_lastfile = -1; 1824 return (&newfdp->fd_fd); 1825} 1826 1827static struct filedesc * 1828fdhold(struct proc *p) 1829{ 1830 struct filedesc *fdp; 1831 1832 mtx_lock(&fdesc_mtx); 1833 fdp = p->p_fd; 1834 if (fdp != NULL) 1835 fdp->fd_holdcnt++; 1836 mtx_unlock(&fdesc_mtx); 1837 return (fdp); 1838} 1839 1840static void 1841fddrop(struct filedesc *fdp) 1842{ 1843 struct filedesc0 *fdp0; 1844 struct freetable *ft; 1845 int i; 1846 1847 mtx_lock(&fdesc_mtx); 1848 i = --fdp->fd_holdcnt; 1849 mtx_unlock(&fdesc_mtx); 1850 if (i > 0) 1851 return; 1852 1853 FILEDESC_LOCK_DESTROY(fdp); 1854 fdp0 = (struct filedesc0 *)fdp; 1855 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1856 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1857 free(ft->ft_table, M_FILEDESC); 1858 } 1859 free(fdp, M_FILEDESC); 1860} 1861 1862/* 1863 * Share a filedesc structure. 1864 */ 1865struct filedesc * 1866fdshare(struct filedesc *fdp) 1867{ 1868 1869 FILEDESC_XLOCK(fdp); 1870 fdp->fd_refcnt++; 1871 FILEDESC_XUNLOCK(fdp); 1872 return (fdp); 1873} 1874 1875/* 1876 * Unshare a filedesc structure, if necessary by making a copy 1877 */ 1878void 1879fdunshare(struct proc *p, struct thread *td) 1880{ 1881 1882 FILEDESC_XLOCK(p->p_fd); 1883 if (p->p_fd->fd_refcnt > 1) { 1884 struct filedesc *tmp; 1885 1886 FILEDESC_XUNLOCK(p->p_fd); 1887 tmp = fdcopy(p->p_fd); 1888 fdescfree(td); 1889 p->p_fd = tmp; 1890 } else 1891 FILEDESC_XUNLOCK(p->p_fd); 1892} 1893 1894/* 1895 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1896 * this is to ease callers, not catch errors. 1897 */ 1898struct filedesc * 1899fdcopy(struct filedesc *fdp) 1900{ 1901 struct filedesc *newfdp; 1902 struct filedescent *nfde, *ofde; 1903 int i; 1904 1905 /* Certain daemons might not have file descriptors. */ 1906 if (fdp == NULL) 1907 return (NULL); 1908 1909 newfdp = fdinit(fdp); 1910 FILEDESC_SLOCK(fdp); 1911 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1912 FILEDESC_SUNLOCK(fdp); 1913 FILEDESC_XLOCK(newfdp); 1914 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1915 FILEDESC_XUNLOCK(newfdp); 1916 FILEDESC_SLOCK(fdp); 1917 } 1918 /* copy all passable descriptors (i.e. not kqueue) */ 1919 newfdp->fd_freefile = -1; 1920 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1921 ofde = &fdp->fd_ofiles[i]; 1922 if (fdisused(fdp, i) && 1923 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) && 1924 ofde->fde_file->f_ops != &badfileops) { 1925 nfde = &newfdp->fd_ofiles[i]; 1926 *nfde = *ofde; 1927 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1928 fhold(nfde->fde_file); 1929 newfdp->fd_lastfile = i; 1930 } else { 1931 if (newfdp->fd_freefile == -1) 1932 newfdp->fd_freefile = i; 1933 } 1934 } 1935 newfdp->fd_cmask = fdp->fd_cmask; 1936 FILEDESC_SUNLOCK(fdp); 1937 FILEDESC_XLOCK(newfdp); 1938 for (i = 0; i <= newfdp->fd_lastfile; ++i) { 1939 if (newfdp->fd_ofiles[i].fde_file != NULL) 1940 fdused(newfdp, i); 1941 } 1942 if (newfdp->fd_freefile == -1) 1943 newfdp->fd_freefile = i; 1944 FILEDESC_XUNLOCK(newfdp); 1945 return (newfdp); 1946} 1947 1948/* 1949 * Release a filedesc structure. 1950 */ 1951void 1952fdescfree(struct thread *td) 1953{ 1954 struct filedesc *fdp; 1955 int i; 1956 struct filedesc_to_leader *fdtol; 1957 struct file *fp; 1958 struct vnode *cdir, *jdir, *rdir, *vp; 1959 struct flock lf; 1960 1961 /* Certain daemons might not have file descriptors. */ 1962 fdp = td->td_proc->p_fd; 1963 if (fdp == NULL) 1964 return; 1965 1966#ifdef RACCT 1967 PROC_LOCK(td->td_proc); 1968 racct_set(td->td_proc, RACCT_NOFILE, 0); 1969 PROC_UNLOCK(td->td_proc); 1970#endif 1971 1972 /* Check for special need to clear POSIX style locks */ 1973 fdtol = td->td_proc->p_fdtol; 1974 if (fdtol != NULL) { 1975 FILEDESC_XLOCK(fdp); 1976 KASSERT(fdtol->fdl_refcount > 0, 1977 ("filedesc_to_refcount botch: fdl_refcount=%d", 1978 fdtol->fdl_refcount)); 1979 if (fdtol->fdl_refcount == 1 && 1980 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1981 for (i = 0; i <= fdp->fd_lastfile; i++) { 1982 fp = fdp->fd_ofiles[i].fde_file; 1983 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1984 continue; 1985 fhold(fp); 1986 FILEDESC_XUNLOCK(fdp); 1987 lf.l_whence = SEEK_SET; 1988 lf.l_start = 0; 1989 lf.l_len = 0; 1990 lf.l_type = F_UNLCK; 1991 vp = fp->f_vnode; 1992 (void) VOP_ADVLOCK(vp, 1993 (caddr_t)td->td_proc->p_leader, F_UNLCK, 1994 &lf, F_POSIX); 1995 FILEDESC_XLOCK(fdp); 1996 fdrop(fp, td); 1997 } 1998 } 1999 retry: 2000 if (fdtol->fdl_refcount == 1) { 2001 if (fdp->fd_holdleaderscount > 0 && 2002 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2003 /* 2004 * close() or do_dup() has cleared a reference 2005 * in a shared file descriptor table. 2006 */ 2007 fdp->fd_holdleaderswakeup = 1; 2008 sx_sleep(&fdp->fd_holdleaderscount, 2009 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2010 goto retry; 2011 } 2012 if (fdtol->fdl_holdcount > 0) { 2013 /* 2014 * Ensure that fdtol->fdl_leader remains 2015 * valid in closef(). 2016 */ 2017 fdtol->fdl_wakeup = 1; 2018 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2019 "fdlhold", 0); 2020 goto retry; 2021 } 2022 } 2023 fdtol->fdl_refcount--; 2024 if (fdtol->fdl_refcount == 0 && 2025 fdtol->fdl_holdcount == 0) { 2026 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2027 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2028 } else 2029 fdtol = NULL; 2030 td->td_proc->p_fdtol = NULL; 2031 FILEDESC_XUNLOCK(fdp); 2032 if (fdtol != NULL) 2033 free(fdtol, M_FILEDESC_TO_LEADER); 2034 } 2035 FILEDESC_XLOCK(fdp); 2036 i = --fdp->fd_refcnt; 2037 FILEDESC_XUNLOCK(fdp); 2038 if (i > 0) 2039 return; 2040 2041 for (i = 0; i <= fdp->fd_lastfile; i++) { 2042 fp = fdp->fd_ofiles[i].fde_file; 2043 if (fp != NULL) { 2044 FILEDESC_XLOCK(fdp); 2045 fdfree(fdp, i); 2046 FILEDESC_XUNLOCK(fdp); 2047 (void) closef(fp, td); 2048 } 2049 } 2050 FILEDESC_XLOCK(fdp); 2051 2052 /* XXX This should happen earlier. */ 2053 mtx_lock(&fdesc_mtx); 2054 td->td_proc->p_fd = NULL; 2055 mtx_unlock(&fdesc_mtx); 2056 2057 if (fdp->fd_nfiles > NDFILE) 2058 free(fdp->fd_ofiles, M_FILEDESC); 2059 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2060 free(fdp->fd_map, M_FILEDESC); 2061 2062 fdp->fd_nfiles = 0; 2063 2064 cdir = fdp->fd_cdir; 2065 fdp->fd_cdir = NULL; 2066 rdir = fdp->fd_rdir; 2067 fdp->fd_rdir = NULL; 2068 jdir = fdp->fd_jdir; 2069 fdp->fd_jdir = NULL; 2070 FILEDESC_XUNLOCK(fdp); 2071 2072 if (cdir != NULL) 2073 vrele(cdir); 2074 if (rdir != NULL) 2075 vrele(rdir); 2076 if (jdir != NULL) 2077 vrele(jdir); 2078 2079 fddrop(fdp); 2080} 2081 2082/* 2083 * For setugid programs, we don't want to people to use that setugidness 2084 * to generate error messages which write to a file which otherwise would 2085 * otherwise be off-limits to the process. We check for filesystems where 2086 * the vnode can change out from under us after execve (like [lin]procfs). 2087 * 2088 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2089 * sufficient. We also don't check for setugidness since we know we are. 2090 */ 2091static int 2092is_unsafe(struct file *fp) 2093{ 2094 if (fp->f_type == DTYPE_VNODE) { 2095 struct vnode *vp = fp->f_vnode; 2096 2097 if ((vp->v_vflag & VV_PROCDEP) != 0) 2098 return (1); 2099 } 2100 return (0); 2101} 2102 2103/* 2104 * Make this setguid thing safe, if at all possible. 2105 */ 2106void 2107setugidsafety(struct thread *td) 2108{ 2109 struct filedesc *fdp; 2110 struct file *fp; 2111 int i; 2112 2113 /* Certain daemons might not have file descriptors. */ 2114 fdp = td->td_proc->p_fd; 2115 if (fdp == NULL) 2116 return; 2117 2118 /* 2119 * Note: fdp->fd_ofiles may be reallocated out from under us while 2120 * we are blocked in a close. Be careful! 2121 */ 2122 FILEDESC_XLOCK(fdp); 2123 for (i = 0; i <= fdp->fd_lastfile; i++) { 2124 if (i > 2) 2125 break; 2126 fp = fdp->fd_ofiles[i].fde_file; 2127 if (fp != NULL && is_unsafe(fp)) { 2128 knote_fdclose(td, i); 2129 /* 2130 * NULL-out descriptor prior to close to avoid 2131 * a race while close blocks. 2132 */ 2133 fdfree(fdp, i); 2134 FILEDESC_XUNLOCK(fdp); 2135 (void) closef(fp, td); 2136 FILEDESC_XLOCK(fdp); 2137 } 2138 } 2139 FILEDESC_XUNLOCK(fdp); 2140} 2141 2142/* 2143 * If a specific file object occupies a specific file descriptor, close the 2144 * file descriptor entry and drop a reference on the file object. This is a 2145 * convenience function to handle a subsequent error in a function that calls 2146 * falloc() that handles the race that another thread might have closed the 2147 * file descriptor out from under the thread creating the file object. 2148 */ 2149void 2150fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2151{ 2152 2153 FILEDESC_XLOCK(fdp); 2154 if (fdp->fd_ofiles[idx].fde_file == fp) { 2155 fdfree(fdp, idx); 2156 FILEDESC_XUNLOCK(fdp); 2157 fdrop(fp, td); 2158 } else 2159 FILEDESC_XUNLOCK(fdp); 2160} 2161 2162/* 2163 * Close any files on exec? 2164 */ 2165void 2166fdcloseexec(struct thread *td) 2167{ 2168 struct filedesc *fdp; 2169 struct filedescent *fde; 2170 struct file *fp; 2171 int i; 2172 2173 /* Certain daemons might not have file descriptors. */ 2174 fdp = td->td_proc->p_fd; 2175 if (fdp == NULL) 2176 return; 2177 2178 /* 2179 * We cannot cache fd_ofiles since operations 2180 * may block and rip them out from under us. 2181 */ 2182 FILEDESC_XLOCK(fdp); 2183 for (i = 0; i <= fdp->fd_lastfile; i++) { 2184 fde = &fdp->fd_ofiles[i]; 2185 fp = fde->fde_file; 2186 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2187 (fde->fde_flags & UF_EXCLOSE))) { 2188 fdfree(fdp, i); 2189 (void) closefp(fdp, i, fp, td, 0); 2190 /* closefp() drops the FILEDESC lock. */ 2191 FILEDESC_XLOCK(fdp); 2192 } 2193 } 2194 FILEDESC_XUNLOCK(fdp); 2195} 2196 2197/* 2198 * It is unsafe for set[ug]id processes to be started with file 2199 * descriptors 0..2 closed, as these descriptors are given implicit 2200 * significance in the Standard C library. fdcheckstd() will create a 2201 * descriptor referencing /dev/null for each of stdin, stdout, and 2202 * stderr that is not already open. 2203 */ 2204int 2205fdcheckstd(struct thread *td) 2206{ 2207 struct filedesc *fdp; 2208 register_t retval, save; 2209 int i, error, devnull; 2210 2211 fdp = td->td_proc->p_fd; 2212 if (fdp == NULL) 2213 return (0); 2214 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2215 devnull = -1; 2216 error = 0; 2217 for (i = 0; i < 3; i++) { 2218 if (fdp->fd_ofiles[i].fde_file != NULL) 2219 continue; 2220 if (devnull < 0) { 2221 save = td->td_retval[0]; 2222 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2223 O_RDWR, 0); 2224 devnull = td->td_retval[0]; 2225 td->td_retval[0] = save; 2226 if (error) 2227 break; 2228 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2229 } else { 2230 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2231 if (error != 0) 2232 break; 2233 } 2234 } 2235 return (error); 2236} 2237 2238/* 2239 * Internal form of close. Decrement reference count on file structure. 2240 * Note: td may be NULL when closing a file that was being passed in a 2241 * message. 2242 * 2243 * XXXRW: Giant is not required for the caller, but often will be held; this 2244 * makes it moderately likely the Giant will be recursed in the VFS case. 2245 */ 2246int 2247closef(struct file *fp, struct thread *td) 2248{ 2249 struct vnode *vp; 2250 struct flock lf; 2251 struct filedesc_to_leader *fdtol; 2252 struct filedesc *fdp; 2253 2254 /* 2255 * POSIX record locking dictates that any close releases ALL 2256 * locks owned by this process. This is handled by setting 2257 * a flag in the unlock to free ONLY locks obeying POSIX 2258 * semantics, and not to free BSD-style file locks. 2259 * If the descriptor was in a message, POSIX-style locks 2260 * aren't passed with the descriptor, and the thread pointer 2261 * will be NULL. Callers should be careful only to pass a 2262 * NULL thread pointer when there really is no owning 2263 * context that might have locks, or the locks will be 2264 * leaked. 2265 */ 2266 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2267 vp = fp->f_vnode; 2268 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2269 lf.l_whence = SEEK_SET; 2270 lf.l_start = 0; 2271 lf.l_len = 0; 2272 lf.l_type = F_UNLCK; 2273 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2274 F_UNLCK, &lf, F_POSIX); 2275 } 2276 fdtol = td->td_proc->p_fdtol; 2277 if (fdtol != NULL) { 2278 /* 2279 * Handle special case where file descriptor table is 2280 * shared between multiple process leaders. 2281 */ 2282 fdp = td->td_proc->p_fd; 2283 FILEDESC_XLOCK(fdp); 2284 for (fdtol = fdtol->fdl_next; 2285 fdtol != td->td_proc->p_fdtol; 2286 fdtol = fdtol->fdl_next) { 2287 if ((fdtol->fdl_leader->p_flag & 2288 P_ADVLOCK) == 0) 2289 continue; 2290 fdtol->fdl_holdcount++; 2291 FILEDESC_XUNLOCK(fdp); 2292 lf.l_whence = SEEK_SET; 2293 lf.l_start = 0; 2294 lf.l_len = 0; 2295 lf.l_type = F_UNLCK; 2296 vp = fp->f_vnode; 2297 (void) VOP_ADVLOCK(vp, 2298 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2299 F_POSIX); 2300 FILEDESC_XLOCK(fdp); 2301 fdtol->fdl_holdcount--; 2302 if (fdtol->fdl_holdcount == 0 && 2303 fdtol->fdl_wakeup != 0) { 2304 fdtol->fdl_wakeup = 0; 2305 wakeup(fdtol); 2306 } 2307 } 2308 FILEDESC_XUNLOCK(fdp); 2309 } 2310 } 2311 return (fdrop(fp, td)); 2312} 2313 2314/* 2315 * Initialize the file pointer with the specified properties. 2316 * 2317 * The ops are set with release semantics to be certain that the flags, type, 2318 * and data are visible when ops is. This is to prevent ops methods from being 2319 * called with bad data. 2320 */ 2321void 2322finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2323{ 2324 fp->f_data = data; 2325 fp->f_flag = flag; 2326 fp->f_type = type; 2327 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2328} 2329 2330int 2331fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2332 int needfcntl, struct file **fpp, cap_rights_t *haverightsp) 2333{ 2334 struct file *fp; 2335 u_int count; 2336#ifdef CAPABILITIES 2337 cap_rights_t haverights; 2338 int error; 2339#endif 2340 2341 /* 2342 * Avoid reads reordering and then a first access to the 2343 * fdp->fd_ofiles table which could result in OOB operation. 2344 */ 2345 if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles)) 2346 return (EBADF); 2347 /* 2348 * Fetch the descriptor locklessly. We avoid fdrop() races by 2349 * never raising a refcount above 0. To accomplish this we have 2350 * to use a cmpset loop rather than an atomic_add. The descriptor 2351 * must be re-verified once we acquire a reference to be certain 2352 * that the identity is still correct and we did not lose a race 2353 * due to preemption. 2354 */ 2355 for (;;) { 2356 fp = fdp->fd_ofiles[fd].fde_file; 2357 if (fp == NULL) 2358 return (EBADF); 2359#ifdef CAPABILITIES 2360 haverights = *cap_rights(fdp, fd); 2361 if (needrightsp != NULL) { 2362 error = cap_check(&haverights, needrightsp); 2363 if (error != 0) 2364 return (error); 2365 if (cap_rights_is_set(needrightsp, CAP_FCNTL)) { 2366 error = cap_fcntl_check(fdp, fd, needfcntl); 2367 if (error != 0) 2368 return (error); 2369 } 2370 } 2371#endif 2372 count = fp->f_count; 2373 if (count == 0) 2374 continue; 2375 /* 2376 * Use an acquire barrier to prevent caching of fd_ofiles 2377 * so it is refreshed for verification. 2378 */ 2379 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2380 continue; 2381 if (fp == fdp->fd_ofiles[fd].fde_file) 2382 break; 2383 fdrop(fp, curthread); 2384 } 2385 *fpp = fp; 2386 if (haverightsp != NULL) { 2387#ifdef CAPABILITIES 2388 *haverightsp = haverights; 2389#else 2390 CAP_ALL(haverightsp); 2391#endif 2392 } 2393 return (0); 2394} 2395 2396/* 2397 * Extract the file pointer associated with the specified descriptor for the 2398 * current user process. 2399 * 2400 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2401 * returned. 2402 * 2403 * File's rights will be checked against the capability rights mask. 2404 * 2405 * If an error occured the non-zero error is returned and *fpp is set to 2406 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2407 * responsible for fdrop(). 2408 */ 2409static __inline int 2410_fget(struct thread *td, int fd, struct file **fpp, int flags, 2411 cap_rights_t *needrightsp, u_char *maxprotp) 2412{ 2413 struct filedesc *fdp; 2414 struct file *fp; 2415 cap_rights_t haverights, needrights; 2416 int error; 2417 2418 *fpp = NULL; 2419 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2420 return (EBADF); 2421 if (needrightsp != NULL) 2422 needrights = *needrightsp; 2423 else 2424 cap_rights_init(&needrights); 2425 if (maxprotp != NULL) 2426 cap_rights_set(&needrights, CAP_MMAP); 2427 error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights); 2428 if (error != 0) 2429 return (error); 2430 if (fp->f_ops == &badfileops) { 2431 fdrop(fp, td); 2432 return (EBADF); 2433 } 2434 2435#ifdef CAPABILITIES 2436 /* 2437 * If requested, convert capability rights to access flags. 2438 */ 2439 if (maxprotp != NULL) 2440 *maxprotp = cap_rights_to_vmprot(&haverights); 2441#else /* !CAPABILITIES */ 2442 if (maxprotp != NULL) 2443 *maxprotp = VM_PROT_ALL; 2444#endif /* CAPABILITIES */ 2445 2446 /* 2447 * FREAD and FWRITE failure return EBADF as per POSIX. 2448 */ 2449 error = 0; 2450 switch (flags) { 2451 case FREAD: 2452 case FWRITE: 2453 if ((fp->f_flag & flags) == 0) 2454 error = EBADF; 2455 break; 2456 case FEXEC: 2457 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2458 ((fp->f_flag & FWRITE) != 0)) 2459 error = EBADF; 2460 break; 2461 case 0: 2462 break; 2463 default: 2464 KASSERT(0, ("wrong flags")); 2465 } 2466 2467 if (error != 0) { 2468 fdrop(fp, td); 2469 return (error); 2470 } 2471 2472 *fpp = fp; 2473 return (0); 2474} 2475 2476int 2477fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2478{ 2479 2480 return(_fget(td, fd, fpp, 0, rightsp, NULL)); 2481} 2482 2483int 2484fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2485 struct file **fpp) 2486{ 2487 2488 return (_fget(td, fd, fpp, 0, rightsp, maxprotp)); 2489} 2490 2491int 2492fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2493{ 2494 2495 return(_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2496} 2497 2498int 2499fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2500{ 2501 2502 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2503} 2504 2505/* 2506 * Like fget() but loads the underlying vnode, or returns an error if the 2507 * descriptor does not represent a vnode. Note that pipes use vnodes but 2508 * never have VM objects. The returned vnode will be vref()'d. 2509 * 2510 * XXX: what about the unused flags ? 2511 */ 2512static __inline int 2513_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2514 struct vnode **vpp) 2515{ 2516 struct file *fp; 2517 int error; 2518 2519 *vpp = NULL; 2520 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2521 if (error != 0) 2522 return (error); 2523 if (fp->f_vnode == NULL) { 2524 error = EINVAL; 2525 } else { 2526 *vpp = fp->f_vnode; 2527 vref(*vpp); 2528 } 2529 fdrop(fp, td); 2530 2531 return (error); 2532} 2533 2534int 2535fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2536{ 2537 2538 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2539} 2540 2541int 2542fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2543 struct filecaps *havecaps, struct vnode **vpp) 2544{ 2545 struct filedesc *fdp; 2546 struct file *fp; 2547#ifdef CAPABILITIES 2548 int error; 2549#endif 2550 2551 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2552 return (EBADF); 2553 2554 fp = fget_locked(fdp, fd); 2555 if (fp == NULL || fp->f_ops == &badfileops) 2556 return (EBADF); 2557 2558#ifdef CAPABILITIES 2559 if (needrightsp != NULL) { 2560 error = cap_check(cap_rights(fdp, fd), needrightsp); 2561 if (error != 0) 2562 return (error); 2563 } 2564#endif 2565 2566 if (fp->f_vnode == NULL) 2567 return (EINVAL); 2568 2569 *vpp = fp->f_vnode; 2570 vref(*vpp); 2571 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2572 2573 return (0); 2574} 2575 2576int 2577fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2578{ 2579 2580 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2581} 2582 2583int 2584fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2585{ 2586 2587 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2588} 2589 2590#ifdef notyet 2591int 2592fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2593 struct vnode **vpp) 2594{ 2595 2596 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2597} 2598#endif 2599 2600/* 2601 * Like fget() but loads the underlying socket, or returns an error if the 2602 * descriptor does not represent a socket. 2603 * 2604 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2605 * in the future. 2606 * 2607 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2608 * on their file descriptor reference to prevent the socket from being free'd 2609 * during use. 2610 */ 2611int 2612fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2613 u_int *fflagp) 2614{ 2615 struct file *fp; 2616 int error; 2617 2618 *spp = NULL; 2619 if (fflagp != NULL) 2620 *fflagp = 0; 2621 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2622 return (error); 2623 if (fp->f_type != DTYPE_SOCKET) { 2624 error = ENOTSOCK; 2625 } else { 2626 *spp = fp->f_data; 2627 if (fflagp) 2628 *fflagp = fp->f_flag; 2629 SOCK_LOCK(*spp); 2630 soref(*spp); 2631 SOCK_UNLOCK(*spp); 2632 } 2633 fdrop(fp, td); 2634 2635 return (error); 2636} 2637 2638/* 2639 * Drop the reference count on the socket and XXX release the SX lock in the 2640 * future. The last reference closes the socket. 2641 * 2642 * Note: fputsock() is deprecated, see comment for fgetsock(). 2643 */ 2644void 2645fputsock(struct socket *so) 2646{ 2647 2648 ACCEPT_LOCK(); 2649 SOCK_LOCK(so); 2650 CURVNET_SET(so->so_vnet); 2651 sorele(so); 2652 CURVNET_RESTORE(); 2653} 2654 2655/* 2656 * Handle the last reference to a file being closed. 2657 */ 2658int 2659_fdrop(struct file *fp, struct thread *td) 2660{ 2661 int error; 2662 2663 error = 0; 2664 if (fp->f_count != 0) 2665 panic("fdrop: count %d", fp->f_count); 2666 if (fp->f_ops != &badfileops) 2667 error = fo_close(fp, td); 2668 atomic_subtract_int(&openfiles, 1); 2669 crfree(fp->f_cred); 2670 free(fp->f_advice, M_FADVISE); 2671 uma_zfree(file_zone, fp); 2672 2673 return (error); 2674} 2675 2676/* 2677 * Apply an advisory lock on a file descriptor. 2678 * 2679 * Just attempt to get a record lock of the requested type on the entire file 2680 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2681 */ 2682#ifndef _SYS_SYSPROTO_H_ 2683struct flock_args { 2684 int fd; 2685 int how; 2686}; 2687#endif 2688/* ARGSUSED */ 2689int 2690sys_flock(struct thread *td, struct flock_args *uap) 2691{ 2692 struct file *fp; 2693 struct vnode *vp; 2694 struct flock lf; 2695 cap_rights_t rights; 2696 int error; 2697 2698 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2699 if (error != 0) 2700 return (error); 2701 if (fp->f_type != DTYPE_VNODE) { 2702 fdrop(fp, td); 2703 return (EOPNOTSUPP); 2704 } 2705 2706 vp = fp->f_vnode; 2707 lf.l_whence = SEEK_SET; 2708 lf.l_start = 0; 2709 lf.l_len = 0; 2710 if (uap->how & LOCK_UN) { 2711 lf.l_type = F_UNLCK; 2712 atomic_clear_int(&fp->f_flag, FHASLOCK); 2713 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2714 goto done2; 2715 } 2716 if (uap->how & LOCK_EX) 2717 lf.l_type = F_WRLCK; 2718 else if (uap->how & LOCK_SH) 2719 lf.l_type = F_RDLCK; 2720 else { 2721 error = EBADF; 2722 goto done2; 2723 } 2724 atomic_set_int(&fp->f_flag, FHASLOCK); 2725 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2726 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2727done2: 2728 fdrop(fp, td); 2729 return (error); 2730} 2731/* 2732 * Duplicate the specified descriptor to a free descriptor. 2733 */ 2734int 2735dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2736 int openerror, int *indxp) 2737{ 2738 struct file *fp; 2739 int error, indx; 2740 2741 KASSERT(openerror == ENODEV || openerror == ENXIO, 2742 ("unexpected error %d in %s", openerror, __func__)); 2743 2744 /* 2745 * If the to-be-dup'd fd number is greater than the allowed number 2746 * of file descriptors, or the fd to be dup'd has already been 2747 * closed, then reject. 2748 */ 2749 FILEDESC_XLOCK(fdp); 2750 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2751 FILEDESC_XUNLOCK(fdp); 2752 return (EBADF); 2753 } 2754 2755 error = fdalloc(td, 0, &indx); 2756 if (error != 0) { 2757 FILEDESC_XUNLOCK(fdp); 2758 return (error); 2759 } 2760 2761 /* 2762 * There are two cases of interest here. 2763 * 2764 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2765 * 2766 * For ENXIO steal away the file structure from (dfd) and store it in 2767 * (indx). (dfd) is effectively closed by this operation. 2768 */ 2769 switch (openerror) { 2770 case ENODEV: 2771 /* 2772 * Check that the mode the file is being opened for is a 2773 * subset of the mode of the existing descriptor. 2774 */ 2775 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2776 fdunused(fdp, indx); 2777 FILEDESC_XUNLOCK(fdp); 2778 return (EACCES); 2779 } 2780 fhold(fp); 2781 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2782 filecaps_copy(&fdp->fd_ofiles[dfd].fde_caps, 2783 &fdp->fd_ofiles[indx].fde_caps); 2784 break; 2785 case ENXIO: 2786 /* 2787 * Steal away the file pointer from dfd and stuff it into indx. 2788 */ 2789 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2790 bzero(&fdp->fd_ofiles[dfd], sizeof(fdp->fd_ofiles[dfd])); 2791 fdunused(fdp, dfd); 2792 break; 2793 } 2794 FILEDESC_XUNLOCK(fdp); 2795 *indxp = indx; 2796 return (0); 2797} 2798 2799/* 2800 * Scan all active processes and prisons to see if any of them have a current 2801 * or root directory of `olddp'. If so, replace them with the new mount point. 2802 */ 2803void 2804mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2805{ 2806 struct filedesc *fdp; 2807 struct prison *pr; 2808 struct proc *p; 2809 int nrele; 2810 2811 if (vrefcnt(olddp) == 1) 2812 return; 2813 nrele = 0; 2814 sx_slock(&allproc_lock); 2815 FOREACH_PROC_IN_SYSTEM(p) { 2816 fdp = fdhold(p); 2817 if (fdp == NULL) 2818 continue; 2819 FILEDESC_XLOCK(fdp); 2820 if (fdp->fd_cdir == olddp) { 2821 vref(newdp); 2822 fdp->fd_cdir = newdp; 2823 nrele++; 2824 } 2825 if (fdp->fd_rdir == olddp) { 2826 vref(newdp); 2827 fdp->fd_rdir = newdp; 2828 nrele++; 2829 } 2830 if (fdp->fd_jdir == olddp) { 2831 vref(newdp); 2832 fdp->fd_jdir = newdp; 2833 nrele++; 2834 } 2835 FILEDESC_XUNLOCK(fdp); 2836 fddrop(fdp); 2837 } 2838 sx_sunlock(&allproc_lock); 2839 if (rootvnode == olddp) { 2840 vref(newdp); 2841 rootvnode = newdp; 2842 nrele++; 2843 } 2844 mtx_lock(&prison0.pr_mtx); 2845 if (prison0.pr_root == olddp) { 2846 vref(newdp); 2847 prison0.pr_root = newdp; 2848 nrele++; 2849 } 2850 mtx_unlock(&prison0.pr_mtx); 2851 sx_slock(&allprison_lock); 2852 TAILQ_FOREACH(pr, &allprison, pr_list) { 2853 mtx_lock(&pr->pr_mtx); 2854 if (pr->pr_root == olddp) { 2855 vref(newdp); 2856 pr->pr_root = newdp; 2857 nrele++; 2858 } 2859 mtx_unlock(&pr->pr_mtx); 2860 } 2861 sx_sunlock(&allprison_lock); 2862 while (nrele--) 2863 vrele(olddp); 2864} 2865 2866struct filedesc_to_leader * 2867filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2868{ 2869 struct filedesc_to_leader *fdtol; 2870 2871 fdtol = malloc(sizeof(struct filedesc_to_leader), 2872 M_FILEDESC_TO_LEADER, 2873 M_WAITOK); 2874 fdtol->fdl_refcount = 1; 2875 fdtol->fdl_holdcount = 0; 2876 fdtol->fdl_wakeup = 0; 2877 fdtol->fdl_leader = leader; 2878 if (old != NULL) { 2879 FILEDESC_XLOCK(fdp); 2880 fdtol->fdl_next = old->fdl_next; 2881 fdtol->fdl_prev = old; 2882 old->fdl_next = fdtol; 2883 fdtol->fdl_next->fdl_prev = fdtol; 2884 FILEDESC_XUNLOCK(fdp); 2885 } else { 2886 fdtol->fdl_next = fdtol; 2887 fdtol->fdl_prev = fdtol; 2888 } 2889 return (fdtol); 2890} 2891 2892/* 2893 * Get file structures globally. 2894 */ 2895static int 2896sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2897{ 2898 struct xfile xf; 2899 struct filedesc *fdp; 2900 struct file *fp; 2901 struct proc *p; 2902 int error, n; 2903 2904 error = sysctl_wire_old_buffer(req, 0); 2905 if (error != 0) 2906 return (error); 2907 if (req->oldptr == NULL) { 2908 n = 0; 2909 sx_slock(&allproc_lock); 2910 FOREACH_PROC_IN_SYSTEM(p) { 2911 if (p->p_state == PRS_NEW) 2912 continue; 2913 fdp = fdhold(p); 2914 if (fdp == NULL) 2915 continue; 2916 /* overestimates sparse tables. */ 2917 if (fdp->fd_lastfile > 0) 2918 n += fdp->fd_lastfile; 2919 fddrop(fdp); 2920 } 2921 sx_sunlock(&allproc_lock); 2922 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2923 } 2924 error = 0; 2925 bzero(&xf, sizeof(xf)); 2926 xf.xf_size = sizeof(xf); 2927 sx_slock(&allproc_lock); 2928 FOREACH_PROC_IN_SYSTEM(p) { 2929 PROC_LOCK(p); 2930 if (p->p_state == PRS_NEW) { 2931 PROC_UNLOCK(p); 2932 continue; 2933 } 2934 if (p_cansee(req->td, p) != 0) { 2935 PROC_UNLOCK(p); 2936 continue; 2937 } 2938 xf.xf_pid = p->p_pid; 2939 xf.xf_uid = p->p_ucred->cr_uid; 2940 PROC_UNLOCK(p); 2941 fdp = fdhold(p); 2942 if (fdp == NULL) 2943 continue; 2944 FILEDESC_SLOCK(fdp); 2945 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2946 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 2947 continue; 2948 xf.xf_fd = n; 2949 xf.xf_file = fp; 2950 xf.xf_data = fp->f_data; 2951 xf.xf_vnode = fp->f_vnode; 2952 xf.xf_type = fp->f_type; 2953 xf.xf_count = fp->f_count; 2954 xf.xf_msgcount = 0; 2955 xf.xf_offset = foffset_get(fp); 2956 xf.xf_flag = fp->f_flag; 2957 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2958 if (error) 2959 break; 2960 } 2961 FILEDESC_SUNLOCK(fdp); 2962 fddrop(fdp); 2963 if (error) 2964 break; 2965 } 2966 sx_sunlock(&allproc_lock); 2967 return (error); 2968} 2969 2970SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, 2971 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2972 2973#ifdef KINFO_OFILE_SIZE 2974CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2975#endif 2976 2977#ifdef COMPAT_FREEBSD7 2978static int 2979export_vnode_for_osysctl(struct vnode *vp, int type, 2980 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2981{ 2982 int error; 2983 char *fullpath, *freepath; 2984 2985 bzero(kif, sizeof(*kif)); 2986 kif->kf_structsize = sizeof(*kif); 2987 2988 vref(vp); 2989 kif->kf_fd = type; 2990 kif->kf_type = KF_TYPE_VNODE; 2991 /* This function only handles directories. */ 2992 if (vp->v_type != VDIR) { 2993 vrele(vp); 2994 return (ENOTDIR); 2995 } 2996 kif->kf_vnode_type = KF_VTYPE_VDIR; 2997 2998 /* 2999 * This is not a true file descriptor, so we set a bogus refcount 3000 * and offset to indicate these fields should be ignored. 3001 */ 3002 kif->kf_ref_count = -1; 3003 kif->kf_offset = -1; 3004 3005 freepath = NULL; 3006 fullpath = "-"; 3007 FILEDESC_SUNLOCK(fdp); 3008 vn_fullpath(curthread, vp, &fullpath, &freepath); 3009 vrele(vp); 3010 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3011 if (freepath != NULL) 3012 free(freepath, M_TEMP); 3013 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3014 FILEDESC_SLOCK(fdp); 3015 return (error); 3016} 3017 3018/* 3019 * Get per-process file descriptors for use by procstat(1), et al. 3020 */ 3021static int 3022sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3023{ 3024 char *fullpath, *freepath; 3025 struct kinfo_ofile *kif; 3026 struct filedesc *fdp; 3027 int error, i, *name; 3028 struct shmfd *shmfd; 3029 struct socket *so; 3030 struct vnode *vp; 3031 struct ksem *ks; 3032 struct file *fp; 3033 struct proc *p; 3034 struct tty *tp; 3035 3036 name = (int *)arg1; 3037 error = pget((pid_t)name[0], PGET_CANDEBUG, &p); 3038 if (error != 0) 3039 return (error); 3040 fdp = fdhold(p); 3041 PROC_UNLOCK(p); 3042 if (fdp == NULL) 3043 return (ENOENT); 3044 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3045 FILEDESC_SLOCK(fdp); 3046 if (fdp->fd_cdir != NULL) 3047 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3048 fdp, req); 3049 if (fdp->fd_rdir != NULL) 3050 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3051 fdp, req); 3052 if (fdp->fd_jdir != NULL) 3053 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3054 fdp, req); 3055 for (i = 0; fdp->fd_refcnt > 0 && i < fdp->fd_nfiles; i++) { 3056 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3057 continue; 3058 bzero(kif, sizeof(*kif)); 3059 kif->kf_structsize = sizeof(*kif); 3060 ks = NULL; 3061 vp = NULL; 3062 so = NULL; 3063 tp = NULL; 3064 shmfd = NULL; 3065 kif->kf_fd = i; 3066 3067 switch (fp->f_type) { 3068 case DTYPE_VNODE: 3069 kif->kf_type = KF_TYPE_VNODE; 3070 vp = fp->f_vnode; 3071 break; 3072 3073 case DTYPE_SOCKET: 3074 kif->kf_type = KF_TYPE_SOCKET; 3075 so = fp->f_data; 3076 break; 3077 3078 case DTYPE_PIPE: 3079 kif->kf_type = KF_TYPE_PIPE; 3080 break; 3081 3082 case DTYPE_FIFO: 3083 kif->kf_type = KF_TYPE_FIFO; 3084 vp = fp->f_vnode; 3085 break; 3086 3087 case DTYPE_KQUEUE: 3088 kif->kf_type = KF_TYPE_KQUEUE; 3089 break; 3090 3091 case DTYPE_CRYPTO: 3092 kif->kf_type = KF_TYPE_CRYPTO; 3093 break; 3094 3095 case DTYPE_MQUEUE: 3096 kif->kf_type = KF_TYPE_MQUEUE; 3097 break; 3098 3099 case DTYPE_SHM: 3100 kif->kf_type = KF_TYPE_SHM; 3101 shmfd = fp->f_data; 3102 break; 3103 3104 case DTYPE_SEM: 3105 kif->kf_type = KF_TYPE_SEM; 3106 ks = fp->f_data; 3107 break; 3108 3109 case DTYPE_PTS: 3110 kif->kf_type = KF_TYPE_PTS; 3111 tp = fp->f_data; 3112 break; 3113 3114#ifdef PROCDESC 3115 case DTYPE_PROCDESC: 3116 kif->kf_type = KF_TYPE_PROCDESC; 3117 break; 3118#endif 3119 3120 default: 3121 kif->kf_type = KF_TYPE_UNKNOWN; 3122 break; 3123 } 3124 kif->kf_ref_count = fp->f_count; 3125 if (fp->f_flag & FREAD) 3126 kif->kf_flags |= KF_FLAG_READ; 3127 if (fp->f_flag & FWRITE) 3128 kif->kf_flags |= KF_FLAG_WRITE; 3129 if (fp->f_flag & FAPPEND) 3130 kif->kf_flags |= KF_FLAG_APPEND; 3131 if (fp->f_flag & FASYNC) 3132 kif->kf_flags |= KF_FLAG_ASYNC; 3133 if (fp->f_flag & FFSYNC) 3134 kif->kf_flags |= KF_FLAG_FSYNC; 3135 if (fp->f_flag & FNONBLOCK) 3136 kif->kf_flags |= KF_FLAG_NONBLOCK; 3137 if (fp->f_flag & O_DIRECT) 3138 kif->kf_flags |= KF_FLAG_DIRECT; 3139 if (fp->f_flag & FHASLOCK) 3140 kif->kf_flags |= KF_FLAG_HASLOCK; 3141 kif->kf_offset = foffset_get(fp); 3142 if (vp != NULL) { 3143 vref(vp); 3144 switch (vp->v_type) { 3145 case VNON: 3146 kif->kf_vnode_type = KF_VTYPE_VNON; 3147 break; 3148 case VREG: 3149 kif->kf_vnode_type = KF_VTYPE_VREG; 3150 break; 3151 case VDIR: 3152 kif->kf_vnode_type = KF_VTYPE_VDIR; 3153 break; 3154 case VBLK: 3155 kif->kf_vnode_type = KF_VTYPE_VBLK; 3156 break; 3157 case VCHR: 3158 kif->kf_vnode_type = KF_VTYPE_VCHR; 3159 break; 3160 case VLNK: 3161 kif->kf_vnode_type = KF_VTYPE_VLNK; 3162 break; 3163 case VSOCK: 3164 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3165 break; 3166 case VFIFO: 3167 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3168 break; 3169 case VBAD: 3170 kif->kf_vnode_type = KF_VTYPE_VBAD; 3171 break; 3172 default: 3173 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3174 break; 3175 } 3176 /* 3177 * It is OK to drop the filedesc lock here as we will 3178 * re-validate and re-evaluate its properties when 3179 * the loop continues. 3180 */ 3181 freepath = NULL; 3182 fullpath = "-"; 3183 FILEDESC_SUNLOCK(fdp); 3184 vn_fullpath(curthread, vp, &fullpath, &freepath); 3185 vrele(vp); 3186 strlcpy(kif->kf_path, fullpath, 3187 sizeof(kif->kf_path)); 3188 if (freepath != NULL) 3189 free(freepath, M_TEMP); 3190 FILEDESC_SLOCK(fdp); 3191 } 3192 if (so != NULL) { 3193 struct sockaddr *sa; 3194 3195 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3196 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3197 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3198 free(sa, M_SONAME); 3199 } 3200 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3201 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3202 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3203 free(sa, M_SONAME); 3204 } 3205 kif->kf_sock_domain = 3206 so->so_proto->pr_domain->dom_family; 3207 kif->kf_sock_type = so->so_type; 3208 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3209 } 3210 if (tp != NULL) { 3211 strlcpy(kif->kf_path, tty_devname(tp), 3212 sizeof(kif->kf_path)); 3213 } 3214 if (shmfd != NULL) 3215 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3216 if (ks != NULL && ksem_info != NULL) 3217 ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL); 3218 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3219 if (error) 3220 break; 3221 } 3222 FILEDESC_SUNLOCK(fdp); 3223 fddrop(fdp); 3224 free(kif, M_TEMP); 3225 return (0); 3226} 3227 3228static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD, 3229 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); 3230#endif /* COMPAT_FREEBSD7 */ 3231 3232#ifdef KINFO_FILE_SIZE 3233CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3234#endif 3235 3236struct export_fd_buf { 3237 struct filedesc *fdp; 3238 struct sbuf *sb; 3239 ssize_t remainder; 3240 struct kinfo_file kif; 3241}; 3242 3243static int 3244export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt, 3245 int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf) 3246{ 3247 struct { 3248 int fflag; 3249 int kf_fflag; 3250 } fflags_table[] = { 3251 { FAPPEND, KF_FLAG_APPEND }, 3252 { FASYNC, KF_FLAG_ASYNC }, 3253 { FFSYNC, KF_FLAG_FSYNC }, 3254 { FHASLOCK, KF_FLAG_HASLOCK }, 3255 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3256 { FREAD, KF_FLAG_READ }, 3257 { FWRITE, KF_FLAG_WRITE }, 3258 { O_CREAT, KF_FLAG_CREAT }, 3259 { O_DIRECT, KF_FLAG_DIRECT }, 3260 { O_EXCL, KF_FLAG_EXCL }, 3261 { O_EXEC, KF_FLAG_EXEC }, 3262 { O_EXLOCK, KF_FLAG_EXLOCK }, 3263 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3264 { O_SHLOCK, KF_FLAG_SHLOCK }, 3265 { O_TRUNC, KF_FLAG_TRUNC } 3266 }; 3267#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3268 struct kinfo_file *kif; 3269 struct vnode *vp; 3270 int error, locked; 3271 unsigned int i; 3272 3273 if (efbuf->remainder == 0) 3274 return (0); 3275 kif = &efbuf->kif; 3276 bzero(kif, sizeof(*kif)); 3277 locked = efbuf->fdp != NULL; 3278 switch (type) { 3279 case KF_TYPE_FIFO: 3280 case KF_TYPE_VNODE: 3281 if (locked) { 3282 FILEDESC_SUNLOCK(efbuf->fdp); 3283 locked = 0; 3284 } 3285 vp = (struct vnode *)data; 3286 error = fill_vnode_info(vp, kif); 3287 vrele(vp); 3288 break; 3289 case KF_TYPE_SOCKET: 3290 error = fill_socket_info((struct socket *)data, kif); 3291 break; 3292 case KF_TYPE_PIPE: 3293 error = fill_pipe_info((struct pipe *)data, kif); 3294 break; 3295 case KF_TYPE_PTS: 3296 error = fill_pts_info((struct tty *)data, kif); 3297 break; 3298 case KF_TYPE_PROCDESC: 3299 error = fill_procdesc_info((struct procdesc *)data, kif); 3300 break; 3301 case KF_TYPE_SEM: 3302 error = fill_sem_info((struct file *)data, kif); 3303 break; 3304 case KF_TYPE_SHM: 3305 error = fill_shm_info((struct file *)data, kif); 3306 break; 3307 default: 3308 error = 0; 3309 } 3310 if (error == 0) 3311 kif->kf_status |= KF_ATTR_VALID; 3312 3313 /* 3314 * Translate file access flags. 3315 */ 3316 for (i = 0; i < NFFLAGS; i++) 3317 if (fflags & fflags_table[i].fflag) 3318 kif->kf_flags |= fflags_table[i].kf_fflag; 3319 if (rightsp != NULL) 3320 kif->kf_cap_rights = *rightsp; 3321 else 3322 cap_rights_init(&kif->kf_cap_rights); 3323 kif->kf_fd = fd; 3324 kif->kf_type = type; 3325 kif->kf_ref_count = refcnt; 3326 kif->kf_offset = offset; 3327 /* Pack record size down */ 3328 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3329 strlen(kif->kf_path) + 1; 3330 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3331 if (efbuf->remainder != -1) { 3332 if (efbuf->remainder < kif->kf_structsize) { 3333 /* Terminate export. */ 3334 efbuf->remainder = 0; 3335 if (efbuf->fdp != NULL && !locked) 3336 FILEDESC_SLOCK(efbuf->fdp); 3337 return (0); 3338 } 3339 efbuf->remainder -= kif->kf_structsize; 3340 } 3341 if (locked) 3342 FILEDESC_SUNLOCK(efbuf->fdp); 3343 error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize); 3344 if (efbuf->fdp != NULL) 3345 FILEDESC_SLOCK(efbuf->fdp); 3346 return (error); 3347} 3348 3349/* 3350 * Store a process file descriptor information to sbuf. 3351 * 3352 * Takes a locked proc as argument, and returns with the proc unlocked. 3353 */ 3354int 3355kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3356{ 3357 struct file *fp; 3358 struct filedesc *fdp; 3359 struct export_fd_buf *efbuf; 3360 struct vnode *cttyvp, *textvp, *tracevp; 3361 int64_t offset; 3362 void *data; 3363 int error, i; 3364 int type, refcnt, fflags; 3365 cap_rights_t rights; 3366 3367 PROC_LOCK_ASSERT(p, MA_OWNED); 3368 3369 /* ktrace vnode */ 3370 tracevp = p->p_tracevp; 3371 if (tracevp != NULL) 3372 vref(tracevp); 3373 /* text vnode */ 3374 textvp = p->p_textvp; 3375 if (textvp != NULL) 3376 vref(textvp); 3377 /* Controlling tty. */ 3378 cttyvp = NULL; 3379 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3380 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3381 if (cttyvp != NULL) 3382 vref(cttyvp); 3383 } 3384 fdp = fdhold(p); 3385 PROC_UNLOCK(p); 3386 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3387 efbuf->fdp = NULL; 3388 efbuf->sb = sb; 3389 efbuf->remainder = maxlen; 3390 if (tracevp != NULL) 3391 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3392 FREAD | FWRITE, -1, -1, NULL, efbuf); 3393 if (textvp != NULL) 3394 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3395 FREAD, -1, -1, NULL, efbuf); 3396 if (cttyvp != NULL) 3397 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3398 FREAD | FWRITE, -1, -1, NULL, efbuf); 3399 error = 0; 3400 if (fdp == NULL) 3401 goto fail; 3402 efbuf->fdp = fdp; 3403 FILEDESC_SLOCK(fdp); 3404 /* working directory */ 3405 if (fdp->fd_cdir != NULL) { 3406 vref(fdp->fd_cdir); 3407 data = fdp->fd_cdir; 3408 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3409 FREAD, -1, -1, NULL, efbuf); 3410 } 3411 /* root directory */ 3412 if (fdp->fd_rdir != NULL) { 3413 vref(fdp->fd_rdir); 3414 data = fdp->fd_rdir; 3415 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3416 FREAD, -1, -1, NULL, efbuf); 3417 } 3418 /* jail directory */ 3419 if (fdp->fd_jdir != NULL) { 3420 vref(fdp->fd_jdir); 3421 data = fdp->fd_jdir; 3422 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3423 FREAD, -1, -1, NULL, efbuf); 3424 } 3425 for (i = 0; fdp->fd_refcnt > 0 && i < fdp->fd_nfiles; i++) { 3426 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3427 continue; 3428 data = NULL; 3429#ifdef CAPABILITIES 3430 rights = *cap_rights(fdp, i); 3431#else /* !CAPABILITIES */ 3432 cap_rights_init(&rights); 3433#endif 3434 switch (fp->f_type) { 3435 case DTYPE_VNODE: 3436 type = KF_TYPE_VNODE; 3437 vref(fp->f_vnode); 3438 data = fp->f_vnode; 3439 break; 3440 3441 case DTYPE_SOCKET: 3442 type = KF_TYPE_SOCKET; 3443 data = fp->f_data; 3444 break; 3445 3446 case DTYPE_PIPE: 3447 type = KF_TYPE_PIPE; 3448 data = fp->f_data; 3449 break; 3450 3451 case DTYPE_FIFO: 3452 type = KF_TYPE_FIFO; 3453 vref(fp->f_vnode); 3454 data = fp->f_vnode; 3455 break; 3456 3457 case DTYPE_KQUEUE: 3458 type = KF_TYPE_KQUEUE; 3459 break; 3460 3461 case DTYPE_CRYPTO: 3462 type = KF_TYPE_CRYPTO; 3463 break; 3464 3465 case DTYPE_MQUEUE: 3466 type = KF_TYPE_MQUEUE; 3467 break; 3468 3469 case DTYPE_SHM: 3470 type = KF_TYPE_SHM; 3471 data = fp; 3472 break; 3473 3474 case DTYPE_SEM: 3475 type = KF_TYPE_SEM; 3476 data = fp; 3477 break; 3478 3479 case DTYPE_PTS: 3480 type = KF_TYPE_PTS; 3481 data = fp->f_data; 3482 break; 3483 3484#ifdef PROCDESC 3485 case DTYPE_PROCDESC: 3486 type = KF_TYPE_PROCDESC; 3487 data = fp->f_data; 3488 break; 3489#endif 3490 3491 default: 3492 type = KF_TYPE_UNKNOWN; 3493 break; 3494 } 3495 refcnt = fp->f_count; 3496 fflags = fp->f_flag; 3497 offset = foffset_get(fp); 3498 3499 /* 3500 * Create sysctl entry. 3501 * It is OK to drop the filedesc lock here as we will 3502 * re-validate and re-evaluate its properties when 3503 * the loop continues. 3504 */ 3505 error = export_fd_to_sb(data, type, i, fflags, refcnt, 3506 offset, &rights, efbuf); 3507 if (error != 0) 3508 break; 3509 } 3510 FILEDESC_SUNLOCK(fdp); 3511 fddrop(fdp); 3512fail: 3513 free(efbuf, M_TEMP); 3514 return (error); 3515} 3516 3517#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3518 3519/* 3520 * Get per-process file descriptors for use by procstat(1), et al. 3521 */ 3522static int 3523sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3524{ 3525 struct sbuf sb; 3526 struct proc *p; 3527 ssize_t maxlen; 3528 int error, error2, *name; 3529 3530 name = (int *)arg1; 3531 3532 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3533 error = pget((pid_t)name[0], PGET_CANDEBUG, &p); 3534 if (error != 0) { 3535 sbuf_delete(&sb); 3536 return (error); 3537 } 3538 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3539 error = kern_proc_filedesc_out(p, &sb, maxlen); 3540 error2 = sbuf_finish(&sb); 3541 sbuf_delete(&sb); 3542 return (error != 0 ? error : error2); 3543} 3544 3545int 3546vntype_to_kinfo(int vtype) 3547{ 3548 struct { 3549 int vtype; 3550 int kf_vtype; 3551 } vtypes_table[] = { 3552 { VBAD, KF_VTYPE_VBAD }, 3553 { VBLK, KF_VTYPE_VBLK }, 3554 { VCHR, KF_VTYPE_VCHR }, 3555 { VDIR, KF_VTYPE_VDIR }, 3556 { VFIFO, KF_VTYPE_VFIFO }, 3557 { VLNK, KF_VTYPE_VLNK }, 3558 { VNON, KF_VTYPE_VNON }, 3559 { VREG, KF_VTYPE_VREG }, 3560 { VSOCK, KF_VTYPE_VSOCK } 3561 }; 3562#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3563 unsigned int i; 3564 3565 /* 3566 * Perform vtype translation. 3567 */ 3568 for (i = 0; i < NVTYPES; i++) 3569 if (vtypes_table[i].vtype == vtype) 3570 break; 3571 if (i < NVTYPES) 3572 return (vtypes_table[i].kf_vtype); 3573 3574 return (KF_VTYPE_UNKNOWN); 3575} 3576 3577static int 3578fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3579{ 3580 struct vattr va; 3581 char *fullpath, *freepath; 3582 int error; 3583 3584 if (vp == NULL) 3585 return (1); 3586 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3587 freepath = NULL; 3588 fullpath = "-"; 3589 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3590 if (error == 0) { 3591 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3592 } 3593 if (freepath != NULL) 3594 free(freepath, M_TEMP); 3595 3596 /* 3597 * Retrieve vnode attributes. 3598 */ 3599 va.va_fsid = VNOVAL; 3600 va.va_rdev = NODEV; 3601 vn_lock(vp, LK_SHARED | LK_RETRY); 3602 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3603 VOP_UNLOCK(vp, 0); 3604 if (error != 0) 3605 return (error); 3606 if (va.va_fsid != VNOVAL) 3607 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3608 else 3609 kif->kf_un.kf_file.kf_file_fsid = 3610 vp->v_mount->mnt_stat.f_fsid.val[0]; 3611 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3612 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3613 kif->kf_un.kf_file.kf_file_size = va.va_size; 3614 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3615 return (0); 3616} 3617 3618static int 3619fill_socket_info(struct socket *so, struct kinfo_file *kif) 3620{ 3621 struct sockaddr *sa; 3622 struct inpcb *inpcb; 3623 struct unpcb *unpcb; 3624 int error; 3625 3626 if (so == NULL) 3627 return (1); 3628 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3629 kif->kf_sock_type = so->so_type; 3630 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3631 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3632 switch(kif->kf_sock_domain) { 3633 case AF_INET: 3634 case AF_INET6: 3635 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3636 if (so->so_pcb != NULL) { 3637 inpcb = (struct inpcb *)(so->so_pcb); 3638 kif->kf_un.kf_sock.kf_sock_inpcb = 3639 (uintptr_t)inpcb->inp_ppcb; 3640 } 3641 } 3642 break; 3643 case AF_UNIX: 3644 if (so->so_pcb != NULL) { 3645 unpcb = (struct unpcb *)(so->so_pcb); 3646 if (unpcb->unp_conn) { 3647 kif->kf_un.kf_sock.kf_sock_unpconn = 3648 (uintptr_t)unpcb->unp_conn; 3649 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3650 so->so_rcv.sb_state; 3651 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3652 so->so_snd.sb_state; 3653 } 3654 } 3655 break; 3656 } 3657 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3658 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3659 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3660 free(sa, M_SONAME); 3661 } 3662 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3663 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3664 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3665 free(sa, M_SONAME); 3666 } 3667 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3668 sizeof(kif->kf_path)); 3669 return (0); 3670} 3671 3672static int 3673fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3674{ 3675 3676 if (tp == NULL) 3677 return (1); 3678 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3679 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3680 return (0); 3681} 3682 3683static int 3684fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3685{ 3686 3687 if (pi == NULL) 3688 return (1); 3689 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3690 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3691 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3692 return (0); 3693} 3694 3695static int 3696fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3697{ 3698 3699 if (pdp == NULL) 3700 return (1); 3701 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3702 return (0); 3703} 3704 3705static int 3706fill_sem_info(struct file *fp, struct kinfo_file *kif) 3707{ 3708 struct thread *td; 3709 struct stat sb; 3710 3711 td = curthread; 3712 if (fp->f_data == NULL) 3713 return (1); 3714 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3715 return (1); 3716 if (ksem_info == NULL) 3717 return (1); 3718 ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path), 3719 &kif->kf_un.kf_sem.kf_sem_value); 3720 kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode; 3721 return (0); 3722} 3723 3724static int 3725fill_shm_info(struct file *fp, struct kinfo_file *kif) 3726{ 3727 struct thread *td; 3728 struct stat sb; 3729 3730 td = curthread; 3731 if (fp->f_data == NULL) 3732 return (1); 3733 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3734 return (1); 3735 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3736 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3737 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3738 return (0); 3739} 3740 3741static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD, 3742 sysctl_kern_proc_filedesc, "Process filedesc entries"); 3743 3744#ifdef DDB 3745/* 3746 * For the purposes of debugging, generate a human-readable string for the 3747 * file type. 3748 */ 3749static const char * 3750file_type_to_name(short type) 3751{ 3752 3753 switch (type) { 3754 case 0: 3755 return ("zero"); 3756 case DTYPE_VNODE: 3757 return ("vnod"); 3758 case DTYPE_SOCKET: 3759 return ("sock"); 3760 case DTYPE_PIPE: 3761 return ("pipe"); 3762 case DTYPE_FIFO: 3763 return ("fifo"); 3764 case DTYPE_KQUEUE: 3765 return ("kque"); 3766 case DTYPE_CRYPTO: 3767 return ("crpt"); 3768 case DTYPE_MQUEUE: 3769 return ("mque"); 3770 case DTYPE_SHM: 3771 return ("shm"); 3772 case DTYPE_SEM: 3773 return ("ksem"); 3774 default: 3775 return ("unkn"); 3776 } 3777} 3778 3779/* 3780 * For the purposes of debugging, identify a process (if any, perhaps one of 3781 * many) that references the passed file in its file descriptor array. Return 3782 * NULL if none. 3783 */ 3784static struct proc * 3785file_to_first_proc(struct file *fp) 3786{ 3787 struct filedesc *fdp; 3788 struct proc *p; 3789 int n; 3790 3791 FOREACH_PROC_IN_SYSTEM(p) { 3792 if (p->p_state == PRS_NEW) 3793 continue; 3794 fdp = p->p_fd; 3795 if (fdp == NULL) 3796 continue; 3797 for (n = 0; n < fdp->fd_nfiles; n++) { 3798 if (fp == fdp->fd_ofiles[n].fde_file) 3799 return (p); 3800 } 3801 } 3802 return (NULL); 3803} 3804 3805static void 3806db_print_file(struct file *fp, int header) 3807{ 3808 struct proc *p; 3809 3810 if (header) 3811 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3812 "File", "Type", "Data", "Flag", "GCFl", "Count", 3813 "MCount", "Vnode", "FPID", "FCmd"); 3814 p = file_to_first_proc(fp); 3815 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3816 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3817 0, fp->f_count, 0, fp->f_vnode, 3818 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3819} 3820 3821DB_SHOW_COMMAND(file, db_show_file) 3822{ 3823 struct file *fp; 3824 3825 if (!have_addr) { 3826 db_printf("usage: show file <addr>\n"); 3827 return; 3828 } 3829 fp = (struct file *)addr; 3830 db_print_file(fp, 1); 3831} 3832 3833DB_SHOW_COMMAND(files, db_show_files) 3834{ 3835 struct filedesc *fdp; 3836 struct file *fp; 3837 struct proc *p; 3838 int header; 3839 int n; 3840 3841 header = 1; 3842 FOREACH_PROC_IN_SYSTEM(p) { 3843 if (p->p_state == PRS_NEW) 3844 continue; 3845 if ((fdp = p->p_fd) == NULL) 3846 continue; 3847 for (n = 0; n < fdp->fd_nfiles; ++n) { 3848 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3849 continue; 3850 db_print_file(fp, header); 3851 header = 0; 3852 } 3853 } 3854} 3855#endif 3856 3857SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3858 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3859 3860SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3861 &maxfiles, 0, "Maximum number of files"); 3862 3863SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3864 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3865 3866/* ARGSUSED*/ 3867static void 3868filelistinit(void *dummy) 3869{ 3870 3871 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3872 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3873 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3874 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3875} 3876SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3877 3878/*-------------------------------------------------------------------*/ 3879 3880static int 3881badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3882 int flags, struct thread *td) 3883{ 3884 3885 return (EBADF); 3886} 3887 3888static int 3889badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3890 struct thread *td) 3891{ 3892 3893 return (EINVAL); 3894} 3895 3896static int 3897badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3898 struct thread *td) 3899{ 3900 3901 return (EBADF); 3902} 3903 3904static int 3905badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3906 struct thread *td) 3907{ 3908 3909 return (0); 3910} 3911 3912static int 3913badfo_kqfilter(struct file *fp, struct knote *kn) 3914{ 3915 3916 return (EBADF); 3917} 3918 3919static int 3920badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3921 struct thread *td) 3922{ 3923 3924 return (EBADF); 3925} 3926 3927static int 3928badfo_close(struct file *fp, struct thread *td) 3929{ 3930 3931 return (EBADF); 3932} 3933 3934static int 3935badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3936 struct thread *td) 3937{ 3938 3939 return (EBADF); 3940} 3941 3942static int 3943badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3944 struct thread *td) 3945{ 3946 3947 return (EBADF); 3948} 3949 3950static int 3951badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3952 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3953 int kflags, struct thread *td) 3954{ 3955 3956 return (EBADF); 3957} 3958 3959struct fileops badfileops = { 3960 .fo_read = badfo_readwrite, 3961 .fo_write = badfo_readwrite, 3962 .fo_truncate = badfo_truncate, 3963 .fo_ioctl = badfo_ioctl, 3964 .fo_poll = badfo_poll, 3965 .fo_kqfilter = badfo_kqfilter, 3966 .fo_stat = badfo_stat, 3967 .fo_close = badfo_close, 3968 .fo_chmod = badfo_chmod, 3969 .fo_chown = badfo_chown, 3970 .fo_sendfile = badfo_sendfile, 3971}; 3972 3973int 3974invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3975 struct thread *td) 3976{ 3977 3978 return (EINVAL); 3979} 3980 3981int 3982invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3983 struct thread *td) 3984{ 3985 3986 return (EINVAL); 3987} 3988 3989int 3990invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3991 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3992 int kflags, struct thread *td) 3993{ 3994 3995 return (EINVAL); 3996} 3997 3998/*-------------------------------------------------------------------*/ 3999 4000/* 4001 * File Descriptor pseudo-device driver (/dev/fd/). 4002 * 4003 * Opening minor device N dup()s the file (if any) connected to file 4004 * descriptor N belonging to the calling process. Note that this driver 4005 * consists of only the ``open()'' routine, because all subsequent 4006 * references to this file will be direct to the other driver. 4007 * 4008 * XXX: we could give this one a cloning event handler if necessary. 4009 */ 4010 4011/* ARGSUSED */ 4012static int 4013fdopen(struct cdev *dev, int mode, int type, struct thread *td) 4014{ 4015 4016 /* 4017 * XXX Kludge: set curthread->td_dupfd to contain the value of the 4018 * the file descriptor being sought for duplication. The error 4019 * return ensures that the vnode for this device will be released 4020 * by vn_open. Open will detect this special error and take the 4021 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 4022 * will simply report the error. 4023 */ 4024 td->td_dupfd = dev2unit(dev); 4025 return (ENODEV); 4026} 4027 4028static struct cdevsw fildesc_cdevsw = { 4029 .d_version = D_VERSION, 4030 .d_open = fdopen, 4031 .d_name = "FD", 4032}; 4033 4034static void 4035fildesc_drvinit(void *unused) 4036{ 4037 struct cdev *dev; 4038 4039 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4040 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4041 make_dev_alias(dev, "stdin"); 4042 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4043 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4044 make_dev_alias(dev, "stdout"); 4045 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4046 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4047 make_dev_alias(dev, "stderr"); 4048} 4049 4050SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4051