kern_descrip.c revision 272187
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 272187 2014-09-26 20:05:28Z 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 * 299 * Avoid some work if fdp is about to be destroyed. 300 */ 301static inline void 302_fdfree(struct filedesc *fdp, int fd, int last) 303{ 304 struct filedescent *fde; 305 306 fde = &fdp->fd_ofiles[fd]; 307 filecaps_free(&fde->fde_caps); 308 if (last) 309 return; 310 bzero(fde, sizeof(*fde)); 311 fdunused(fdp, fd); 312} 313 314static inline void 315fdfree(struct filedesc *fdp, int fd) 316{ 317 318 _fdfree(fdp, fd, 0); 319} 320 321static inline void 322fdfree_last(struct filedesc *fdp, int fd) 323{ 324 325 _fdfree(fdp, fd, 1); 326} 327 328/* 329 * System calls on descriptors. 330 */ 331#ifndef _SYS_SYSPROTO_H_ 332struct getdtablesize_args { 333 int dummy; 334}; 335#endif 336/* ARGSUSED */ 337int 338sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 339{ 340 struct proc *p = td->td_proc; 341 uint64_t lim; 342 343 PROC_LOCK(p); 344 td->td_retval[0] = 345 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 346 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 347 PROC_UNLOCK(p); 348 if (lim < td->td_retval[0]) 349 td->td_retval[0] = lim; 350 return (0); 351} 352 353/* 354 * Duplicate a file descriptor to a particular value. 355 * 356 * Note: keep in mind that a potential race condition exists when closing 357 * descriptors from a shared descriptor table (via rfork). 358 */ 359#ifndef _SYS_SYSPROTO_H_ 360struct dup2_args { 361 u_int from; 362 u_int to; 363}; 364#endif 365/* ARGSUSED */ 366int 367sys_dup2(struct thread *td, struct dup2_args *uap) 368{ 369 370 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 371 td->td_retval)); 372} 373 374/* 375 * Duplicate a file descriptor. 376 */ 377#ifndef _SYS_SYSPROTO_H_ 378struct dup_args { 379 u_int fd; 380}; 381#endif 382/* ARGSUSED */ 383int 384sys_dup(struct thread *td, struct dup_args *uap) 385{ 386 387 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 388} 389 390/* 391 * The file control system call. 392 */ 393#ifndef _SYS_SYSPROTO_H_ 394struct fcntl_args { 395 int fd; 396 int cmd; 397 long arg; 398}; 399#endif 400/* ARGSUSED */ 401int 402sys_fcntl(struct thread *td, struct fcntl_args *uap) 403{ 404 struct flock fl; 405 struct __oflock ofl; 406 intptr_t arg; 407 int error; 408 int cmd; 409 410 error = 0; 411 cmd = uap->cmd; 412 switch (uap->cmd) { 413 case F_OGETLK: 414 case F_OSETLK: 415 case F_OSETLKW: 416 /* 417 * Convert old flock structure to new. 418 */ 419 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 420 fl.l_start = ofl.l_start; 421 fl.l_len = ofl.l_len; 422 fl.l_pid = ofl.l_pid; 423 fl.l_type = ofl.l_type; 424 fl.l_whence = ofl.l_whence; 425 fl.l_sysid = 0; 426 427 switch (uap->cmd) { 428 case F_OGETLK: 429 cmd = F_GETLK; 430 break; 431 case F_OSETLK: 432 cmd = F_SETLK; 433 break; 434 case F_OSETLKW: 435 cmd = F_SETLKW; 436 break; 437 } 438 arg = (intptr_t)&fl; 439 break; 440 case F_GETLK: 441 case F_SETLK: 442 case F_SETLKW: 443 case F_SETLK_REMOTE: 444 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 445 arg = (intptr_t)&fl; 446 break; 447 default: 448 arg = uap->arg; 449 break; 450 } 451 if (error) 452 return (error); 453 error = kern_fcntl(td, uap->fd, cmd, arg); 454 if (error) 455 return (error); 456 if (uap->cmd == F_OGETLK) { 457 ofl.l_start = fl.l_start; 458 ofl.l_len = fl.l_len; 459 ofl.l_pid = fl.l_pid; 460 ofl.l_type = fl.l_type; 461 ofl.l_whence = fl.l_whence; 462 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 463 } else if (uap->cmd == F_GETLK) { 464 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 465 } 466 return (error); 467} 468 469int 470kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 471{ 472 struct filedesc *fdp; 473 struct flock *flp; 474 struct file *fp, *fp2; 475 struct filedescent *fde; 476 struct proc *p; 477 struct vnode *vp; 478 cap_rights_t rights; 479 int error, flg, tmp; 480 uint64_t bsize; 481 off_t foffset; 482 483 error = 0; 484 flg = F_POSIX; 485 p = td->td_proc; 486 fdp = p->p_fd; 487 488 switch (cmd) { 489 case F_DUPFD: 490 tmp = arg; 491 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 492 break; 493 494 case F_DUPFD_CLOEXEC: 495 tmp = arg; 496 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp, 497 td->td_retval); 498 break; 499 500 case F_DUP2FD: 501 tmp = arg; 502 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 503 break; 504 505 case F_DUP2FD_CLOEXEC: 506 tmp = arg; 507 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp, 508 td->td_retval); 509 break; 510 511 case F_GETFD: 512 FILEDESC_SLOCK(fdp); 513 if ((fp = fget_locked(fdp, fd)) == NULL) { 514 FILEDESC_SUNLOCK(fdp); 515 error = EBADF; 516 break; 517 } 518 fde = &fdp->fd_ofiles[fd]; 519 td->td_retval[0] = 520 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 521 FILEDESC_SUNLOCK(fdp); 522 break; 523 524 case F_SETFD: 525 FILEDESC_XLOCK(fdp); 526 if ((fp = fget_locked(fdp, fd)) == NULL) { 527 FILEDESC_XUNLOCK(fdp); 528 error = EBADF; 529 break; 530 } 531 fde = &fdp->fd_ofiles[fd]; 532 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 533 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 534 FILEDESC_XUNLOCK(fdp); 535 break; 536 537 case F_GETFL: 538 error = fget_unlocked(fdp, fd, 539 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL); 540 if (error != 0) 541 break; 542 td->td_retval[0] = OFLAGS(fp->f_flag); 543 fdrop(fp, td); 544 break; 545 546 case F_SETFL: 547 error = fget_unlocked(fdp, fd, 548 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL); 549 if (error != 0) 550 break; 551 do { 552 tmp = flg = fp->f_flag; 553 tmp &= ~FCNTLFLAGS; 554 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 555 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 556 tmp = fp->f_flag & FNONBLOCK; 557 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 558 if (error != 0) { 559 fdrop(fp, td); 560 break; 561 } 562 tmp = fp->f_flag & FASYNC; 563 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 564 if (error == 0) { 565 fdrop(fp, td); 566 break; 567 } 568 atomic_clear_int(&fp->f_flag, FNONBLOCK); 569 tmp = 0; 570 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 571 fdrop(fp, td); 572 break; 573 574 case F_GETOWN: 575 error = fget_unlocked(fdp, fd, 576 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL); 577 if (error != 0) 578 break; 579 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 580 if (error == 0) 581 td->td_retval[0] = tmp; 582 fdrop(fp, td); 583 break; 584 585 case F_SETOWN: 586 error = fget_unlocked(fdp, fd, 587 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL); 588 if (error != 0) 589 break; 590 tmp = arg; 591 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 592 fdrop(fp, td); 593 break; 594 595 case F_SETLK_REMOTE: 596 error = priv_check(td, PRIV_NFS_LOCKD); 597 if (error) 598 return (error); 599 flg = F_REMOTE; 600 goto do_setlk; 601 602 case F_SETLKW: 603 flg |= F_WAIT; 604 /* FALLTHROUGH F_SETLK */ 605 606 case F_SETLK: 607 do_setlk: 608 cap_rights_init(&rights, CAP_FLOCK); 609 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL); 610 if (error != 0) 611 break; 612 if (fp->f_type != DTYPE_VNODE) { 613 error = EBADF; 614 fdrop(fp, td); 615 break; 616 } 617 618 flp = (struct flock *)arg; 619 if (flp->l_whence == SEEK_CUR) { 620 foffset = foffset_get(fp); 621 if (foffset < 0 || 622 (flp->l_start > 0 && 623 foffset > OFF_MAX - flp->l_start)) { 624 FILEDESC_SUNLOCK(fdp); 625 error = EOVERFLOW; 626 fdrop(fp, td); 627 break; 628 } 629 flp->l_start += foffset; 630 } 631 632 vp = fp->f_vnode; 633 switch (flp->l_type) { 634 case F_RDLCK: 635 if ((fp->f_flag & FREAD) == 0) { 636 error = EBADF; 637 break; 638 } 639 PROC_LOCK(p->p_leader); 640 p->p_leader->p_flag |= P_ADVLOCK; 641 PROC_UNLOCK(p->p_leader); 642 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 643 flp, flg); 644 break; 645 case F_WRLCK: 646 if ((fp->f_flag & FWRITE) == 0) { 647 error = EBADF; 648 break; 649 } 650 PROC_LOCK(p->p_leader); 651 p->p_leader->p_flag |= P_ADVLOCK; 652 PROC_UNLOCK(p->p_leader); 653 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 654 flp, flg); 655 break; 656 case F_UNLCK: 657 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 658 flp, flg); 659 break; 660 case F_UNLCKSYS: 661 /* 662 * Temporary api for testing remote lock 663 * infrastructure. 664 */ 665 if (flg != F_REMOTE) { 666 error = EINVAL; 667 break; 668 } 669 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 670 F_UNLCKSYS, flp, flg); 671 break; 672 default: 673 error = EINVAL; 674 break; 675 } 676 if (error != 0 || flp->l_type == F_UNLCK || 677 flp->l_type == F_UNLCKSYS) { 678 fdrop(fp, td); 679 break; 680 } 681 682 /* 683 * Check for a race with close. 684 * 685 * The vnode is now advisory locked (or unlocked, but this case 686 * is not really important) as the caller requested. 687 * We had to drop the filedesc lock, so we need to recheck if 688 * the descriptor is still valid, because if it was closed 689 * in the meantime we need to remove advisory lock from the 690 * vnode - close on any descriptor leading to an advisory 691 * locked vnode, removes that lock. 692 * We will return 0 on purpose in that case, as the result of 693 * successful advisory lock might have been externally visible 694 * already. This is fine - effectively we pretend to the caller 695 * that the closing thread was a bit slower and that the 696 * advisory lock succeeded before the close. 697 */ 698 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL); 699 if (error != 0) { 700 fdrop(fp, td); 701 break; 702 } 703 if (fp != fp2) { 704 flp->l_whence = SEEK_SET; 705 flp->l_start = 0; 706 flp->l_len = 0; 707 flp->l_type = F_UNLCK; 708 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 709 F_UNLCK, flp, F_POSIX); 710 } 711 fdrop(fp, td); 712 fdrop(fp2, td); 713 break; 714 715 case F_GETLK: 716 error = fget_unlocked(fdp, fd, 717 cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL); 718 if (error != 0) 719 break; 720 if (fp->f_type != DTYPE_VNODE) { 721 error = EBADF; 722 fdrop(fp, td); 723 break; 724 } 725 flp = (struct flock *)arg; 726 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 727 flp->l_type != F_UNLCK) { 728 error = EINVAL; 729 fdrop(fp, td); 730 break; 731 } 732 if (flp->l_whence == SEEK_CUR) { 733 foffset = foffset_get(fp); 734 if ((flp->l_start > 0 && 735 foffset > OFF_MAX - flp->l_start) || 736 (flp->l_start < 0 && 737 foffset < OFF_MIN - flp->l_start)) { 738 FILEDESC_SUNLOCK(fdp); 739 error = EOVERFLOW; 740 fdrop(fp, td); 741 break; 742 } 743 flp->l_start += foffset; 744 } 745 vp = fp->f_vnode; 746 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 747 F_POSIX); 748 fdrop(fp, td); 749 break; 750 751 case F_RDAHEAD: 752 arg = arg ? 128 * 1024: 0; 753 /* FALLTHROUGH */ 754 case F_READAHEAD: 755 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL); 756 if (error != 0) 757 break; 758 if (fp->f_type != DTYPE_VNODE) { 759 fdrop(fp, td); 760 error = EBADF; 761 break; 762 } 763 vp = fp->f_vnode; 764 /* 765 * Exclusive lock synchronizes against f_seqcount reads and 766 * writes in sequential_heuristic(). 767 */ 768 error = vn_lock(vp, LK_EXCLUSIVE); 769 if (error != 0) { 770 fdrop(fp, td); 771 break; 772 } 773 if (arg >= 0) { 774 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 775 fp->f_seqcount = (arg + bsize - 1) / bsize; 776 atomic_set_int(&fp->f_flag, FRDAHEAD); 777 } else { 778 atomic_clear_int(&fp->f_flag, FRDAHEAD); 779 } 780 VOP_UNLOCK(vp, 0); 781 fdrop(fp, td); 782 break; 783 784 default: 785 error = EINVAL; 786 break; 787 } 788 return (error); 789} 790 791static int 792getmaxfd(struct proc *p) 793{ 794 int maxfd; 795 796 PROC_LOCK(p); 797 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 798 PROC_UNLOCK(p); 799 800 return (maxfd); 801} 802 803/* 804 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 805 */ 806int 807do_dup(struct thread *td, int flags, int old, int new, 808 register_t *retval) 809{ 810 struct filedesc *fdp; 811 struct filedescent *oldfde, *newfde; 812 struct proc *p; 813 struct file *fp; 814 struct file *delfp; 815 int error, maxfd; 816 817 p = td->td_proc; 818 fdp = p->p_fd; 819 820 /* 821 * Verify we have a valid descriptor to dup from and possibly to 822 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 823 * return EINVAL when the new descriptor is out of bounds. 824 */ 825 if (old < 0) 826 return (EBADF); 827 if (new < 0) 828 return (flags & DUP_FCNTL ? EINVAL : EBADF); 829 maxfd = getmaxfd(p); 830 if (new >= maxfd) 831 return (flags & DUP_FCNTL ? EINVAL : EBADF); 832 833 FILEDESC_XLOCK(fdp); 834 if (fget_locked(fdp, old) == NULL) { 835 FILEDESC_XUNLOCK(fdp); 836 return (EBADF); 837 } 838 oldfde = &fdp->fd_ofiles[old]; 839 if (flags & DUP_FIXED && old == new) { 840 *retval = new; 841 if (flags & DUP_CLOEXEC) 842 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 843 FILEDESC_XUNLOCK(fdp); 844 return (0); 845 } 846 fp = oldfde->fde_file; 847 fhold(fp); 848 849 /* 850 * If the caller specified a file descriptor, make sure the file 851 * table is large enough to hold it, and grab it. Otherwise, just 852 * allocate a new descriptor the usual way. 853 */ 854 if (flags & DUP_FIXED) { 855 if (new >= fdp->fd_nfiles) { 856 /* 857 * The resource limits are here instead of e.g. 858 * fdalloc(), because the file descriptor table may be 859 * shared between processes, so we can't really use 860 * racct_add()/racct_sub(). Instead of counting the 861 * number of actually allocated descriptors, just put 862 * the limit on the size of the file descriptor table. 863 */ 864#ifdef RACCT 865 PROC_LOCK(p); 866 error = racct_set(p, RACCT_NOFILE, new + 1); 867 PROC_UNLOCK(p); 868 if (error != 0) { 869 FILEDESC_XUNLOCK(fdp); 870 fdrop(fp, td); 871 return (EMFILE); 872 } 873#endif 874 fdgrowtable_exp(fdp, new + 1); 875 oldfde = &fdp->fd_ofiles[old]; 876 } 877 newfde = &fdp->fd_ofiles[new]; 878 if (newfde->fde_file == NULL) 879 fdused(fdp, new); 880 } else { 881 if ((error = fdalloc(td, new, &new)) != 0) { 882 FILEDESC_XUNLOCK(fdp); 883 fdrop(fp, td); 884 return (error); 885 } 886 newfde = &fdp->fd_ofiles[new]; 887 } 888 889 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 890 KASSERT(old != new, ("new fd is same as old")); 891 892 delfp = newfde->fde_file; 893 894 /* 895 * Duplicate the source descriptor. 896 */ 897 filecaps_free(&newfde->fde_caps); 898 *newfde = *oldfde; 899 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 900 if ((flags & DUP_CLOEXEC) != 0) 901 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 902 else 903 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 904 *retval = new; 905 906 if (delfp != NULL) { 907 (void) closefp(fdp, new, delfp, td, 1); 908 /* closefp() drops the FILEDESC lock for us. */ 909 } else { 910 FILEDESC_XUNLOCK(fdp); 911 } 912 913 return (0); 914} 915 916/* 917 * If sigio is on the list associated with a process or process group, 918 * disable signalling from the device, remove sigio from the list and 919 * free sigio. 920 */ 921void 922funsetown(struct sigio **sigiop) 923{ 924 struct sigio *sigio; 925 926 SIGIO_LOCK(); 927 sigio = *sigiop; 928 if (sigio == NULL) { 929 SIGIO_UNLOCK(); 930 return; 931 } 932 *(sigio->sio_myref) = NULL; 933 if ((sigio)->sio_pgid < 0) { 934 struct pgrp *pg = (sigio)->sio_pgrp; 935 PGRP_LOCK(pg); 936 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 937 sigio, sio_pgsigio); 938 PGRP_UNLOCK(pg); 939 } else { 940 struct proc *p = (sigio)->sio_proc; 941 PROC_LOCK(p); 942 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 943 sigio, sio_pgsigio); 944 PROC_UNLOCK(p); 945 } 946 SIGIO_UNLOCK(); 947 crfree(sigio->sio_ucred); 948 free(sigio, M_SIGIO); 949} 950 951/* 952 * Free a list of sigio structures. 953 * We only need to lock the SIGIO_LOCK because we have made ourselves 954 * inaccessible to callers of fsetown and therefore do not need to lock 955 * the proc or pgrp struct for the list manipulation. 956 */ 957void 958funsetownlst(struct sigiolst *sigiolst) 959{ 960 struct proc *p; 961 struct pgrp *pg; 962 struct sigio *sigio; 963 964 sigio = SLIST_FIRST(sigiolst); 965 if (sigio == NULL) 966 return; 967 p = NULL; 968 pg = NULL; 969 970 /* 971 * Every entry of the list should belong 972 * to a single proc or pgrp. 973 */ 974 if (sigio->sio_pgid < 0) { 975 pg = sigio->sio_pgrp; 976 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 977 } else /* if (sigio->sio_pgid > 0) */ { 978 p = sigio->sio_proc; 979 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 980 } 981 982 SIGIO_LOCK(); 983 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 984 *(sigio->sio_myref) = NULL; 985 if (pg != NULL) { 986 KASSERT(sigio->sio_pgid < 0, 987 ("Proc sigio in pgrp sigio list")); 988 KASSERT(sigio->sio_pgrp == pg, 989 ("Bogus pgrp in sigio list")); 990 PGRP_LOCK(pg); 991 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 992 sio_pgsigio); 993 PGRP_UNLOCK(pg); 994 } else /* if (p != NULL) */ { 995 KASSERT(sigio->sio_pgid > 0, 996 ("Pgrp sigio in proc sigio list")); 997 KASSERT(sigio->sio_proc == p, 998 ("Bogus proc in sigio list")); 999 PROC_LOCK(p); 1000 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1001 sio_pgsigio); 1002 PROC_UNLOCK(p); 1003 } 1004 SIGIO_UNLOCK(); 1005 crfree(sigio->sio_ucred); 1006 free(sigio, M_SIGIO); 1007 SIGIO_LOCK(); 1008 } 1009 SIGIO_UNLOCK(); 1010} 1011 1012/* 1013 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1014 * 1015 * After permission checking, add a sigio structure to the sigio list for 1016 * the process or process group. 1017 */ 1018int 1019fsetown(pid_t pgid, struct sigio **sigiop) 1020{ 1021 struct proc *proc; 1022 struct pgrp *pgrp; 1023 struct sigio *sigio; 1024 int ret; 1025 1026 if (pgid == 0) { 1027 funsetown(sigiop); 1028 return (0); 1029 } 1030 1031 ret = 0; 1032 1033 /* Allocate and fill in the new sigio out of locks. */ 1034 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1035 sigio->sio_pgid = pgid; 1036 sigio->sio_ucred = crhold(curthread->td_ucred); 1037 sigio->sio_myref = sigiop; 1038 1039 sx_slock(&proctree_lock); 1040 if (pgid > 0) { 1041 proc = pfind(pgid); 1042 if (proc == NULL) { 1043 ret = ESRCH; 1044 goto fail; 1045 } 1046 1047 /* 1048 * Policy - Don't allow a process to FSETOWN a process 1049 * in another session. 1050 * 1051 * Remove this test to allow maximum flexibility or 1052 * restrict FSETOWN to the current process or process 1053 * group for maximum safety. 1054 */ 1055 PROC_UNLOCK(proc); 1056 if (proc->p_session != curthread->td_proc->p_session) { 1057 ret = EPERM; 1058 goto fail; 1059 } 1060 1061 pgrp = NULL; 1062 } else /* if (pgid < 0) */ { 1063 pgrp = pgfind(-pgid); 1064 if (pgrp == NULL) { 1065 ret = ESRCH; 1066 goto fail; 1067 } 1068 PGRP_UNLOCK(pgrp); 1069 1070 /* 1071 * Policy - Don't allow a process to FSETOWN a process 1072 * in another session. 1073 * 1074 * Remove this test to allow maximum flexibility or 1075 * restrict FSETOWN to the current process or process 1076 * group for maximum safety. 1077 */ 1078 if (pgrp->pg_session != curthread->td_proc->p_session) { 1079 ret = EPERM; 1080 goto fail; 1081 } 1082 1083 proc = NULL; 1084 } 1085 funsetown(sigiop); 1086 if (pgid > 0) { 1087 PROC_LOCK(proc); 1088 /* 1089 * Since funsetownlst() is called without the proctree 1090 * locked, we need to check for P_WEXIT. 1091 * XXX: is ESRCH correct? 1092 */ 1093 if ((proc->p_flag & P_WEXIT) != 0) { 1094 PROC_UNLOCK(proc); 1095 ret = ESRCH; 1096 goto fail; 1097 } 1098 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1099 sigio->sio_proc = proc; 1100 PROC_UNLOCK(proc); 1101 } else { 1102 PGRP_LOCK(pgrp); 1103 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1104 sigio->sio_pgrp = pgrp; 1105 PGRP_UNLOCK(pgrp); 1106 } 1107 sx_sunlock(&proctree_lock); 1108 SIGIO_LOCK(); 1109 *sigiop = sigio; 1110 SIGIO_UNLOCK(); 1111 return (0); 1112 1113fail: 1114 sx_sunlock(&proctree_lock); 1115 crfree(sigio->sio_ucred); 1116 free(sigio, M_SIGIO); 1117 return (ret); 1118} 1119 1120/* 1121 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1122 */ 1123pid_t 1124fgetown(sigiop) 1125 struct sigio **sigiop; 1126{ 1127 pid_t pgid; 1128 1129 SIGIO_LOCK(); 1130 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1131 SIGIO_UNLOCK(); 1132 return (pgid); 1133} 1134 1135/* 1136 * Function drops the filedesc lock on return. 1137 */ 1138static int 1139closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1140 int holdleaders) 1141{ 1142 int error; 1143 1144 FILEDESC_XLOCK_ASSERT(fdp); 1145 1146 if (holdleaders) { 1147 if (td->td_proc->p_fdtol != NULL) { 1148 /* 1149 * Ask fdfree() to sleep to ensure that all relevant 1150 * process leaders can be traversed in closef(). 1151 */ 1152 fdp->fd_holdleaderscount++; 1153 } else { 1154 holdleaders = 0; 1155 } 1156 } 1157 1158 /* 1159 * We now hold the fp reference that used to be owned by the 1160 * descriptor array. We have to unlock the FILEDESC *AFTER* 1161 * knote_fdclose to prevent a race of the fd getting opened, a knote 1162 * added, and deleteing a knote for the new fd. 1163 */ 1164 knote_fdclose(td, fd); 1165 1166 /* 1167 * We need to notify mqueue if the object is of type mqueue. 1168 */ 1169 if (fp->f_type == DTYPE_MQUEUE) 1170 mq_fdclose(td, fd, fp); 1171 FILEDESC_XUNLOCK(fdp); 1172 1173 error = closef(fp, td); 1174 if (holdleaders) { 1175 FILEDESC_XLOCK(fdp); 1176 fdp->fd_holdleaderscount--; 1177 if (fdp->fd_holdleaderscount == 0 && 1178 fdp->fd_holdleaderswakeup != 0) { 1179 fdp->fd_holdleaderswakeup = 0; 1180 wakeup(&fdp->fd_holdleaderscount); 1181 } 1182 FILEDESC_XUNLOCK(fdp); 1183 } 1184 return (error); 1185} 1186 1187/* 1188 * Close a file descriptor. 1189 */ 1190#ifndef _SYS_SYSPROTO_H_ 1191struct close_args { 1192 int fd; 1193}; 1194#endif 1195/* ARGSUSED */ 1196int 1197sys_close(td, uap) 1198 struct thread *td; 1199 struct close_args *uap; 1200{ 1201 1202 return (kern_close(td, uap->fd)); 1203} 1204 1205int 1206kern_close(td, fd) 1207 struct thread *td; 1208 int fd; 1209{ 1210 struct filedesc *fdp; 1211 struct file *fp; 1212 1213 fdp = td->td_proc->p_fd; 1214 1215 AUDIT_SYSCLOSE(td, fd); 1216 1217 FILEDESC_XLOCK(fdp); 1218 if ((fp = fget_locked(fdp, fd)) == NULL) { 1219 FILEDESC_XUNLOCK(fdp); 1220 return (EBADF); 1221 } 1222 fdfree(fdp, fd); 1223 1224 /* closefp() drops the FILEDESC lock for us. */ 1225 return (closefp(fdp, fd, fp, td, 1)); 1226} 1227 1228/* 1229 * Close open file descriptors. 1230 */ 1231#ifndef _SYS_SYSPROTO_H_ 1232struct closefrom_args { 1233 int lowfd; 1234}; 1235#endif 1236/* ARGSUSED */ 1237int 1238sys_closefrom(struct thread *td, struct closefrom_args *uap) 1239{ 1240 struct filedesc *fdp; 1241 int fd; 1242 1243 fdp = td->td_proc->p_fd; 1244 AUDIT_ARG_FD(uap->lowfd); 1245 1246 /* 1247 * Treat negative starting file descriptor values identical to 1248 * closefrom(0) which closes all files. 1249 */ 1250 if (uap->lowfd < 0) 1251 uap->lowfd = 0; 1252 FILEDESC_SLOCK(fdp); 1253 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1254 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1255 FILEDESC_SUNLOCK(fdp); 1256 (void)kern_close(td, fd); 1257 FILEDESC_SLOCK(fdp); 1258 } 1259 } 1260 FILEDESC_SUNLOCK(fdp); 1261 return (0); 1262} 1263 1264#if defined(COMPAT_43) 1265/* 1266 * Return status information about a file descriptor. 1267 */ 1268#ifndef _SYS_SYSPROTO_H_ 1269struct ofstat_args { 1270 int fd; 1271 struct ostat *sb; 1272}; 1273#endif 1274/* ARGSUSED */ 1275int 1276ofstat(struct thread *td, struct ofstat_args *uap) 1277{ 1278 struct ostat oub; 1279 struct stat ub; 1280 int error; 1281 1282 error = kern_fstat(td, uap->fd, &ub); 1283 if (error == 0) { 1284 cvtstat(&ub, &oub); 1285 error = copyout(&oub, uap->sb, sizeof(oub)); 1286 } 1287 return (error); 1288} 1289#endif /* COMPAT_43 */ 1290 1291/* 1292 * Return status information about a file descriptor. 1293 */ 1294#ifndef _SYS_SYSPROTO_H_ 1295struct fstat_args { 1296 int fd; 1297 struct stat *sb; 1298}; 1299#endif 1300/* ARGSUSED */ 1301int 1302sys_fstat(struct thread *td, struct fstat_args *uap) 1303{ 1304 struct stat ub; 1305 int error; 1306 1307 error = kern_fstat(td, uap->fd, &ub); 1308 if (error == 0) 1309 error = copyout(&ub, uap->sb, sizeof(ub)); 1310 return (error); 1311} 1312 1313int 1314kern_fstat(struct thread *td, int fd, struct stat *sbp) 1315{ 1316 struct file *fp; 1317 cap_rights_t rights; 1318 int error; 1319 1320 AUDIT_ARG_FD(fd); 1321 1322 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1323 if (error != 0) 1324 return (error); 1325 1326 AUDIT_ARG_FILE(td->td_proc, fp); 1327 1328 error = fo_stat(fp, sbp, td->td_ucred, td); 1329 fdrop(fp, td); 1330#ifdef KTRACE 1331 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1332 ktrstat(sbp); 1333#endif 1334 return (error); 1335} 1336 1337/* 1338 * Return status information about a file descriptor. 1339 */ 1340#ifndef _SYS_SYSPROTO_H_ 1341struct nfstat_args { 1342 int fd; 1343 struct nstat *sb; 1344}; 1345#endif 1346/* ARGSUSED */ 1347int 1348sys_nfstat(struct thread *td, struct nfstat_args *uap) 1349{ 1350 struct nstat nub; 1351 struct stat ub; 1352 int error; 1353 1354 error = kern_fstat(td, uap->fd, &ub); 1355 if (error == 0) { 1356 cvtnstat(&ub, &nub); 1357 error = copyout(&nub, uap->sb, sizeof(nub)); 1358 } 1359 return (error); 1360} 1361 1362/* 1363 * Return pathconf information about a file descriptor. 1364 */ 1365#ifndef _SYS_SYSPROTO_H_ 1366struct fpathconf_args { 1367 int fd; 1368 int name; 1369}; 1370#endif 1371/* ARGSUSED */ 1372int 1373sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1374{ 1375 struct file *fp; 1376 struct vnode *vp; 1377 cap_rights_t rights; 1378 int error; 1379 1380 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1381 if (error != 0) 1382 return (error); 1383 1384 /* If asynchronous I/O is available, it works for all descriptors. */ 1385 if (uap->name == _PC_ASYNC_IO) { 1386 td->td_retval[0] = async_io_version; 1387 goto out; 1388 } 1389 vp = fp->f_vnode; 1390 if (vp != NULL) { 1391 vn_lock(vp, LK_SHARED | LK_RETRY); 1392 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1393 VOP_UNLOCK(vp, 0); 1394 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1395 if (uap->name != _PC_PIPE_BUF) { 1396 error = EINVAL; 1397 } else { 1398 td->td_retval[0] = PIPE_BUF; 1399 error = 0; 1400 } 1401 } else { 1402 error = EOPNOTSUPP; 1403 } 1404out: 1405 fdrop(fp, td); 1406 return (error); 1407} 1408 1409/* 1410 * Initialize filecaps structure. 1411 */ 1412void 1413filecaps_init(struct filecaps *fcaps) 1414{ 1415 1416 bzero(fcaps, sizeof(*fcaps)); 1417 fcaps->fc_nioctls = -1; 1418} 1419 1420/* 1421 * Copy filecaps structure allocating memory for ioctls array if needed. 1422 */ 1423void 1424filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1425{ 1426 size_t size; 1427 1428 *dst = *src; 1429 if (src->fc_ioctls != NULL) { 1430 KASSERT(src->fc_nioctls > 0, 1431 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1432 1433 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1434 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1435 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1436 } 1437} 1438 1439/* 1440 * Move filecaps structure to the new place and clear the old place. 1441 */ 1442void 1443filecaps_move(struct filecaps *src, struct filecaps *dst) 1444{ 1445 1446 *dst = *src; 1447 bzero(src, sizeof(*src)); 1448} 1449 1450/* 1451 * Fill the given filecaps structure with full rights. 1452 */ 1453static void 1454filecaps_fill(struct filecaps *fcaps) 1455{ 1456 1457 CAP_ALL(&fcaps->fc_rights); 1458 fcaps->fc_ioctls = NULL; 1459 fcaps->fc_nioctls = -1; 1460 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1461} 1462 1463/* 1464 * Free memory allocated within filecaps structure. 1465 */ 1466void 1467filecaps_free(struct filecaps *fcaps) 1468{ 1469 1470 free(fcaps->fc_ioctls, M_FILECAPS); 1471 bzero(fcaps, sizeof(*fcaps)); 1472} 1473 1474/* 1475 * Validate the given filecaps structure. 1476 */ 1477static void 1478filecaps_validate(const struct filecaps *fcaps, const char *func) 1479{ 1480 1481 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1482 ("%s: invalid rights", func)); 1483 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1484 ("%s: invalid fcntls", func)); 1485 KASSERT(fcaps->fc_fcntls == 0 || 1486 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1487 ("%s: fcntls without CAP_FCNTL", func)); 1488 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1489 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1490 ("%s: invalid ioctls", func)); 1491 KASSERT(fcaps->fc_nioctls == 0 || 1492 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1493 ("%s: ioctls without CAP_IOCTL", func)); 1494} 1495 1496static void 1497fdgrowtable_exp(struct filedesc *fdp, int nfd) 1498{ 1499 int nfd1; 1500 1501 FILEDESC_XLOCK_ASSERT(fdp); 1502 1503 nfd1 = fdp->fd_nfiles * 2; 1504 if (nfd1 < nfd) 1505 nfd1 = nfd; 1506 fdgrowtable(fdp, nfd1); 1507} 1508 1509/* 1510 * Grow the file table to accomodate (at least) nfd descriptors. 1511 */ 1512static void 1513fdgrowtable(struct filedesc *fdp, int nfd) 1514{ 1515 struct filedesc0 *fdp0; 1516 struct freetable *ft; 1517 struct filedescent *ntable; 1518 struct filedescent *otable; 1519 int nnfiles, onfiles; 1520 NDSLOTTYPE *nmap, *omap; 1521 1522 FILEDESC_XLOCK_ASSERT(fdp); 1523 1524 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1525 1526 /* save old values */ 1527 onfiles = fdp->fd_nfiles; 1528 otable = fdp->fd_ofiles; 1529 omap = fdp->fd_map; 1530 1531 /* compute the size of the new table */ 1532 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1533 if (nnfiles <= onfiles) 1534 /* the table is already large enough */ 1535 return; 1536 1537 /* 1538 * Allocate a new table. We need enough space for the 1539 * file entries themselves and the struct freetable we will use 1540 * when we decommission the table and place it on the freelist. 1541 * We place the struct freetable in the middle so we don't have 1542 * to worry about padding. 1543 */ 1544 ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable), 1545 M_FILEDESC, M_ZERO | M_WAITOK); 1546 /* copy the old data over and point at the new tables */ 1547 memcpy(ntable, otable, onfiles * sizeof(*otable)); 1548 fdp->fd_ofiles = ntable; 1549 1550 /* 1551 * Allocate a new map only if the old is not large enough. It will 1552 * grow at a slower rate than the table as it can map more 1553 * entries than the table can hold. 1554 */ 1555 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1556 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1557 M_ZERO | M_WAITOK); 1558 /* copy over the old data and update the pointer */ 1559 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1560 fdp->fd_map = nmap; 1561 } 1562 1563 /* 1564 * In order to have a valid pattern for fget_unlocked() 1565 * fdp->fd_nfiles must be the last member to be updated, otherwise 1566 * fget_unlocked() consumers may reference a new, higher value for 1567 * fdp->fd_nfiles before to access the fdp->fd_ofiles array, 1568 * resulting in OOB accesses. 1569 */ 1570 atomic_store_rel_int(&fdp->fd_nfiles, nnfiles); 1571 1572 /* 1573 * Do not free the old file table, as some threads may still 1574 * reference entries within it. Instead, place it on a freelist 1575 * which will be processed when the struct filedesc is released. 1576 * 1577 * Note that if onfiles == NDFILE, we're dealing with the original 1578 * static allocation contained within (struct filedesc0 *)fdp, 1579 * which must not be freed. 1580 */ 1581 if (onfiles > NDFILE) { 1582 ft = (struct freetable *)&otable[onfiles]; 1583 fdp0 = (struct filedesc0 *)fdp; 1584 ft->ft_table = otable; 1585 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1586 } 1587 /* 1588 * The map does not have the same possibility of threads still 1589 * holding references to it. So always free it as long as it 1590 * does not reference the original static allocation. 1591 */ 1592 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1593 free(omap, M_FILEDESC); 1594} 1595 1596/* 1597 * Allocate a file descriptor for the process. 1598 */ 1599int 1600fdalloc(struct thread *td, int minfd, int *result) 1601{ 1602 struct proc *p = td->td_proc; 1603 struct filedesc *fdp = p->p_fd; 1604 int fd = -1, maxfd, allocfd; 1605#ifdef RACCT 1606 int error; 1607#endif 1608 1609 FILEDESC_XLOCK_ASSERT(fdp); 1610 1611 if (fdp->fd_freefile > minfd) 1612 minfd = fdp->fd_freefile; 1613 1614 maxfd = getmaxfd(p); 1615 1616 /* 1617 * Search the bitmap for a free descriptor starting at minfd. 1618 * If none is found, grow the file table. 1619 */ 1620 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1621 if (fd >= maxfd) 1622 return (EMFILE); 1623 if (fd >= fdp->fd_nfiles) { 1624 allocfd = min(fd * 2, maxfd); 1625#ifdef RACCT 1626 PROC_LOCK(p); 1627 error = racct_set(p, RACCT_NOFILE, allocfd); 1628 PROC_UNLOCK(p); 1629 if (error != 0) 1630 return (EMFILE); 1631#endif 1632 /* 1633 * fd is already equal to first free descriptor >= minfd, so 1634 * we only need to grow the table and we are done. 1635 */ 1636 fdgrowtable_exp(fdp, allocfd); 1637 } 1638 1639 /* 1640 * Perform some sanity checks, then mark the file descriptor as 1641 * used and return it to the caller. 1642 */ 1643 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1644 ("invalid descriptor %d", fd)); 1645 KASSERT(!fdisused(fdp, fd), 1646 ("fd_first_free() returned non-free descriptor")); 1647 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1648 ("file descriptor isn't free")); 1649 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); 1650 fdused(fdp, fd); 1651 *result = fd; 1652 return (0); 1653} 1654 1655/* 1656 * Allocate n file descriptors for the process. 1657 */ 1658int 1659fdallocn(struct thread *td, int minfd, int *fds, int n) 1660{ 1661 struct proc *p = td->td_proc; 1662 struct filedesc *fdp = p->p_fd; 1663 int i; 1664 1665 FILEDESC_XLOCK_ASSERT(fdp); 1666 1667 if (!fdavail(td, n)) 1668 return (EMFILE); 1669 1670 for (i = 0; i < n; i++) 1671 if (fdalloc(td, 0, &fds[i]) != 0) 1672 break; 1673 1674 if (i < n) { 1675 for (i--; i >= 0; i--) 1676 fdunused(fdp, fds[i]); 1677 return (EMFILE); 1678 } 1679 1680 return (0); 1681} 1682 1683/* 1684 * Check to see whether n user file descriptors are available to the process 1685 * p. 1686 */ 1687int 1688fdavail(struct thread *td, int n) 1689{ 1690 struct proc *p = td->td_proc; 1691 struct filedesc *fdp = td->td_proc->p_fd; 1692 int i, lim, last; 1693 1694 FILEDESC_LOCK_ASSERT(fdp); 1695 1696 /* 1697 * XXX: This is only called from uipc_usrreq.c:unp_externalize(); 1698 * call racct_add() from there instead of dealing with containers 1699 * here. 1700 */ 1701 lim = getmaxfd(p); 1702 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) 1703 return (1); 1704 last = min(fdp->fd_nfiles, lim); 1705 for (i = fdp->fd_freefile; i < last; i++) { 1706 if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0) 1707 return (1); 1708 } 1709 return (0); 1710} 1711 1712/* 1713 * Create a new open file structure and allocate a file decriptor for the 1714 * process that refers to it. We add one reference to the file for the 1715 * descriptor table and one reference for resultfp. This is to prevent us 1716 * being preempted and the entry in the descriptor table closed after we 1717 * release the FILEDESC lock. 1718 */ 1719int 1720falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1721{ 1722 struct file *fp; 1723 int error, fd; 1724 1725 error = falloc_noinstall(td, &fp); 1726 if (error) 1727 return (error); /* no reference held on error */ 1728 1729 error = finstall(td, fp, &fd, flags, NULL); 1730 if (error) { 1731 fdrop(fp, td); /* one reference (fp only) */ 1732 return (error); 1733 } 1734 1735 if (resultfp != NULL) 1736 *resultfp = fp; /* copy out result */ 1737 else 1738 fdrop(fp, td); /* release local reference */ 1739 1740 if (resultfd != NULL) 1741 *resultfd = fd; 1742 1743 return (0); 1744} 1745 1746/* 1747 * Create a new open file structure without allocating a file descriptor. 1748 */ 1749int 1750falloc_noinstall(struct thread *td, struct file **resultfp) 1751{ 1752 struct file *fp; 1753 int maxuserfiles = maxfiles - (maxfiles / 20); 1754 static struct timeval lastfail; 1755 static int curfail; 1756 1757 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1758 1759 if ((openfiles >= maxuserfiles && 1760 priv_check(td, PRIV_MAXFILES) != 0) || 1761 openfiles >= maxfiles) { 1762 if (ppsratecheck(&lastfail, &curfail, 1)) { 1763 printf("kern.maxfiles limit exceeded by uid %i, " 1764 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1765 } 1766 return (ENFILE); 1767 } 1768 atomic_add_int(&openfiles, 1); 1769 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1770 refcount_init(&fp->f_count, 1); 1771 fp->f_cred = crhold(td->td_ucred); 1772 fp->f_ops = &badfileops; 1773 fp->f_data = NULL; 1774 fp->f_vnode = NULL; 1775 *resultfp = fp; 1776 return (0); 1777} 1778 1779/* 1780 * Install a file in a file descriptor table. 1781 */ 1782int 1783finstall(struct thread *td, struct file *fp, int *fd, int flags, 1784 struct filecaps *fcaps) 1785{ 1786 struct filedesc *fdp = td->td_proc->p_fd; 1787 struct filedescent *fde; 1788 int error; 1789 1790 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1791 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1792 if (fcaps != NULL) 1793 filecaps_validate(fcaps, __func__); 1794 1795 FILEDESC_XLOCK(fdp); 1796 if ((error = fdalloc(td, 0, fd))) { 1797 FILEDESC_XUNLOCK(fdp); 1798 return (error); 1799 } 1800 fhold(fp); 1801 fde = &fdp->fd_ofiles[*fd]; 1802 fde->fde_file = fp; 1803 if ((flags & O_CLOEXEC) != 0) 1804 fde->fde_flags |= UF_EXCLOSE; 1805 if (fcaps != NULL) 1806 filecaps_move(fcaps, &fde->fde_caps); 1807 else 1808 filecaps_fill(&fde->fde_caps); 1809 FILEDESC_XUNLOCK(fdp); 1810 return (0); 1811} 1812 1813/* 1814 * Build a new filedesc structure from another. 1815 * Copy the current, root, and jail root vnode references. 1816 */ 1817struct filedesc * 1818fdinit(struct filedesc *fdp) 1819{ 1820 struct filedesc0 *newfdp; 1821 1822 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1823 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1824 if (fdp != NULL) { 1825 FILEDESC_SLOCK(fdp); 1826 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1827 if (newfdp->fd_fd.fd_cdir) 1828 VREF(newfdp->fd_fd.fd_cdir); 1829 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1830 if (newfdp->fd_fd.fd_rdir) 1831 VREF(newfdp->fd_fd.fd_rdir); 1832 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1833 if (newfdp->fd_fd.fd_jdir) 1834 VREF(newfdp->fd_fd.fd_jdir); 1835 FILEDESC_SUNLOCK(fdp); 1836 } 1837 1838 /* Create the file descriptor table. */ 1839 newfdp->fd_fd.fd_refcnt = 1; 1840 newfdp->fd_fd.fd_holdcnt = 1; 1841 newfdp->fd_fd.fd_cmask = CMASK; 1842 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1843 newfdp->fd_fd.fd_nfiles = NDFILE; 1844 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1845 newfdp->fd_fd.fd_lastfile = -1; 1846 return (&newfdp->fd_fd); 1847} 1848 1849static struct filedesc * 1850fdhold(struct proc *p) 1851{ 1852 struct filedesc *fdp; 1853 1854 mtx_lock(&fdesc_mtx); 1855 fdp = p->p_fd; 1856 if (fdp != NULL) 1857 fdp->fd_holdcnt++; 1858 mtx_unlock(&fdesc_mtx); 1859 return (fdp); 1860} 1861 1862static void 1863fddrop(struct filedesc *fdp) 1864{ 1865 struct filedesc0 *fdp0; 1866 struct freetable *ft; 1867 int i; 1868 1869 mtx_lock(&fdesc_mtx); 1870 i = --fdp->fd_holdcnt; 1871 mtx_unlock(&fdesc_mtx); 1872 if (i > 0) 1873 return; 1874 1875 FILEDESC_LOCK_DESTROY(fdp); 1876 fdp0 = (struct filedesc0 *)fdp; 1877 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1878 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1879 free(ft->ft_table, M_FILEDESC); 1880 } 1881 free(fdp, M_FILEDESC); 1882} 1883 1884/* 1885 * Share a filedesc structure. 1886 */ 1887struct filedesc * 1888fdshare(struct filedesc *fdp) 1889{ 1890 1891 FILEDESC_XLOCK(fdp); 1892 fdp->fd_refcnt++; 1893 FILEDESC_XUNLOCK(fdp); 1894 return (fdp); 1895} 1896 1897/* 1898 * Unshare a filedesc structure, if necessary by making a copy 1899 */ 1900void 1901fdunshare(struct thread *td) 1902{ 1903 struct filedesc *tmp; 1904 struct proc *p = td->td_proc; 1905 1906 if (p->p_fd->fd_refcnt == 1) 1907 return; 1908 1909 tmp = fdcopy(p->p_fd); 1910 fdescfree(td); 1911 p->p_fd = tmp; 1912} 1913 1914/* 1915 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1916 * this is to ease callers, not catch errors. 1917 */ 1918struct filedesc * 1919fdcopy(struct filedesc *fdp) 1920{ 1921 struct filedesc *newfdp; 1922 struct filedescent *nfde, *ofde; 1923 int i; 1924 1925 /* Certain daemons might not have file descriptors. */ 1926 if (fdp == NULL) 1927 return (NULL); 1928 1929 newfdp = fdinit(fdp); 1930 FILEDESC_SLOCK(fdp); 1931 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1932 FILEDESC_SUNLOCK(fdp); 1933 FILEDESC_XLOCK(newfdp); 1934 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1935 FILEDESC_XUNLOCK(newfdp); 1936 FILEDESC_SLOCK(fdp); 1937 } 1938 /* copy all passable descriptors (i.e. not kqueue) */ 1939 newfdp->fd_freefile = -1; 1940 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1941 ofde = &fdp->fd_ofiles[i]; 1942 if (fdisused(fdp, i) && 1943 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) && 1944 ofde->fde_file->f_ops != &badfileops) { 1945 nfde = &newfdp->fd_ofiles[i]; 1946 *nfde = *ofde; 1947 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1948 fhold(nfde->fde_file); 1949 newfdp->fd_lastfile = i; 1950 } else { 1951 if (newfdp->fd_freefile == -1) 1952 newfdp->fd_freefile = i; 1953 } 1954 } 1955 newfdp->fd_cmask = fdp->fd_cmask; 1956 FILEDESC_SUNLOCK(fdp); 1957 FILEDESC_XLOCK(newfdp); 1958 for (i = 0; i <= newfdp->fd_lastfile; ++i) { 1959 if (newfdp->fd_ofiles[i].fde_file != NULL) 1960 fdused(newfdp, i); 1961 } 1962 if (newfdp->fd_freefile == -1) 1963 newfdp->fd_freefile = i; 1964 FILEDESC_XUNLOCK(newfdp); 1965 return (newfdp); 1966} 1967 1968/* 1969 * Release a filedesc structure. 1970 */ 1971void 1972fdescfree(struct thread *td) 1973{ 1974 struct filedesc *fdp; 1975 int i; 1976 struct filedesc_to_leader *fdtol; 1977 struct file *fp; 1978 struct vnode *cdir, *jdir, *rdir, *vp; 1979 struct flock lf; 1980 1981 /* Certain daemons might not have file descriptors. */ 1982 fdp = td->td_proc->p_fd; 1983 if (fdp == NULL) 1984 return; 1985 1986#ifdef RACCT 1987 PROC_LOCK(td->td_proc); 1988 racct_set(td->td_proc, RACCT_NOFILE, 0); 1989 PROC_UNLOCK(td->td_proc); 1990#endif 1991 1992 /* Check for special need to clear POSIX style locks */ 1993 fdtol = td->td_proc->p_fdtol; 1994 if (fdtol != NULL) { 1995 FILEDESC_XLOCK(fdp); 1996 KASSERT(fdtol->fdl_refcount > 0, 1997 ("filedesc_to_refcount botch: fdl_refcount=%d", 1998 fdtol->fdl_refcount)); 1999 if (fdtol->fdl_refcount == 1 && 2000 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2001 for (i = 0; i <= fdp->fd_lastfile; i++) { 2002 fp = fdp->fd_ofiles[i].fde_file; 2003 if (fp == NULL || fp->f_type != DTYPE_VNODE) 2004 continue; 2005 fhold(fp); 2006 FILEDESC_XUNLOCK(fdp); 2007 lf.l_whence = SEEK_SET; 2008 lf.l_start = 0; 2009 lf.l_len = 0; 2010 lf.l_type = F_UNLCK; 2011 vp = fp->f_vnode; 2012 (void) VOP_ADVLOCK(vp, 2013 (caddr_t)td->td_proc->p_leader, F_UNLCK, 2014 &lf, F_POSIX); 2015 FILEDESC_XLOCK(fdp); 2016 fdrop(fp, td); 2017 } 2018 } 2019 retry: 2020 if (fdtol->fdl_refcount == 1) { 2021 if (fdp->fd_holdleaderscount > 0 && 2022 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2023 /* 2024 * close() or do_dup() has cleared a reference 2025 * in a shared file descriptor table. 2026 */ 2027 fdp->fd_holdleaderswakeup = 1; 2028 sx_sleep(&fdp->fd_holdleaderscount, 2029 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2030 goto retry; 2031 } 2032 if (fdtol->fdl_holdcount > 0) { 2033 /* 2034 * Ensure that fdtol->fdl_leader remains 2035 * valid in closef(). 2036 */ 2037 fdtol->fdl_wakeup = 1; 2038 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2039 "fdlhold", 0); 2040 goto retry; 2041 } 2042 } 2043 fdtol->fdl_refcount--; 2044 if (fdtol->fdl_refcount == 0 && 2045 fdtol->fdl_holdcount == 0) { 2046 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2047 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2048 } else 2049 fdtol = NULL; 2050 td->td_proc->p_fdtol = NULL; 2051 FILEDESC_XUNLOCK(fdp); 2052 if (fdtol != NULL) 2053 free(fdtol, M_FILEDESC_TO_LEADER); 2054 } 2055 2056 mtx_lock(&fdesc_mtx); 2057 td->td_proc->p_fd = NULL; 2058 mtx_unlock(&fdesc_mtx); 2059 2060 FILEDESC_XLOCK(fdp); 2061 i = --fdp->fd_refcnt; 2062 if (i > 0) { 2063 FILEDESC_XUNLOCK(fdp); 2064 return; 2065 } 2066 2067 cdir = fdp->fd_cdir; 2068 fdp->fd_cdir = NULL; 2069 rdir = fdp->fd_rdir; 2070 fdp->fd_rdir = NULL; 2071 jdir = fdp->fd_jdir; 2072 fdp->fd_jdir = NULL; 2073 FILEDESC_XUNLOCK(fdp); 2074 2075 for (i = 0; i <= fdp->fd_lastfile; i++) { 2076 fp = fdp->fd_ofiles[i].fde_file; 2077 if (fp != NULL) { 2078 fdfree_last(fdp, i); 2079 (void) closef(fp, td); 2080 } 2081 } 2082 2083 if (fdp->fd_nfiles > NDFILE) 2084 free(fdp->fd_ofiles, M_FILEDESC); 2085 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2086 free(fdp->fd_map, M_FILEDESC); 2087 2088 if (cdir != NULL) 2089 vrele(cdir); 2090 if (rdir != NULL) 2091 vrele(rdir); 2092 if (jdir != NULL) 2093 vrele(jdir); 2094 2095 fddrop(fdp); 2096} 2097 2098/* 2099 * For setugid programs, we don't want to people to use that setugidness 2100 * to generate error messages which write to a file which otherwise would 2101 * otherwise be off-limits to the process. We check for filesystems where 2102 * the vnode can change out from under us after execve (like [lin]procfs). 2103 * 2104 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2105 * sufficient. We also don't check for setugidness since we know we are. 2106 */ 2107static int 2108is_unsafe(struct file *fp) 2109{ 2110 if (fp->f_type == DTYPE_VNODE) { 2111 struct vnode *vp = fp->f_vnode; 2112 2113 if ((vp->v_vflag & VV_PROCDEP) != 0) 2114 return (1); 2115 } 2116 return (0); 2117} 2118 2119/* 2120 * Make this setguid thing safe, if at all possible. 2121 */ 2122void 2123setugidsafety(struct thread *td) 2124{ 2125 struct filedesc *fdp; 2126 struct file *fp; 2127 int i; 2128 2129 fdp = td->td_proc->p_fd; 2130 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2131 FILEDESC_XLOCK(fdp); 2132 for (i = 0; i <= fdp->fd_lastfile; i++) { 2133 if (i > 2) 2134 break; 2135 fp = fdp->fd_ofiles[i].fde_file; 2136 if (fp != NULL && is_unsafe(fp)) { 2137 knote_fdclose(td, i); 2138 /* 2139 * NULL-out descriptor prior to close to avoid 2140 * a race while close blocks. 2141 */ 2142 fdfree(fdp, i); 2143 FILEDESC_XUNLOCK(fdp); 2144 (void) closef(fp, td); 2145 FILEDESC_XLOCK(fdp); 2146 } 2147 } 2148 FILEDESC_XUNLOCK(fdp); 2149} 2150 2151/* 2152 * If a specific file object occupies a specific file descriptor, close the 2153 * file descriptor entry and drop a reference on the file object. This is a 2154 * convenience function to handle a subsequent error in a function that calls 2155 * falloc() that handles the race that another thread might have closed the 2156 * file descriptor out from under the thread creating the file object. 2157 */ 2158void 2159fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2160{ 2161 2162 FILEDESC_XLOCK(fdp); 2163 if (fdp->fd_ofiles[idx].fde_file == fp) { 2164 fdfree(fdp, idx); 2165 FILEDESC_XUNLOCK(fdp); 2166 fdrop(fp, td); 2167 } else 2168 FILEDESC_XUNLOCK(fdp); 2169} 2170 2171/* 2172 * Close any files on exec? 2173 */ 2174void 2175fdcloseexec(struct thread *td) 2176{ 2177 struct filedesc *fdp; 2178 struct filedescent *fde; 2179 struct file *fp; 2180 int i; 2181 2182 fdp = td->td_proc->p_fd; 2183 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2184 FILEDESC_XLOCK(fdp); 2185 for (i = 0; i <= fdp->fd_lastfile; i++) { 2186 fde = &fdp->fd_ofiles[i]; 2187 fp = fde->fde_file; 2188 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2189 (fde->fde_flags & UF_EXCLOSE))) { 2190 fdfree(fdp, i); 2191 (void) closefp(fdp, i, fp, td, 0); 2192 /* closefp() drops the FILEDESC lock. */ 2193 FILEDESC_XLOCK(fdp); 2194 } 2195 } 2196 FILEDESC_XUNLOCK(fdp); 2197} 2198 2199/* 2200 * It is unsafe for set[ug]id processes to be started with file 2201 * descriptors 0..2 closed, as these descriptors are given implicit 2202 * significance in the Standard C library. fdcheckstd() will create a 2203 * descriptor referencing /dev/null for each of stdin, stdout, and 2204 * stderr that is not already open. 2205 */ 2206int 2207fdcheckstd(struct thread *td) 2208{ 2209 struct filedesc *fdp; 2210 register_t retval, save; 2211 int i, error, devnull; 2212 2213 fdp = td->td_proc->p_fd; 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_lastfile; ++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|CTLFLAG_MPSAFE, 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 | PGET_NOTWEXIT, &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_lastfile; 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, 3229 CTLFLAG_RD||CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3230 "Process ofiledesc entries"); 3231#endif /* COMPAT_FREEBSD7 */ 3232 3233#ifdef KINFO_FILE_SIZE 3234CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3235#endif 3236 3237struct export_fd_buf { 3238 struct filedesc *fdp; 3239 struct sbuf *sb; 3240 ssize_t remainder; 3241 struct kinfo_file kif; 3242}; 3243 3244static int 3245export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt, 3246 int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf) 3247{ 3248 struct { 3249 int fflag; 3250 int kf_fflag; 3251 } fflags_table[] = { 3252 { FAPPEND, KF_FLAG_APPEND }, 3253 { FASYNC, KF_FLAG_ASYNC }, 3254 { FFSYNC, KF_FLAG_FSYNC }, 3255 { FHASLOCK, KF_FLAG_HASLOCK }, 3256 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3257 { FREAD, KF_FLAG_READ }, 3258 { FWRITE, KF_FLAG_WRITE }, 3259 { O_CREAT, KF_FLAG_CREAT }, 3260 { O_DIRECT, KF_FLAG_DIRECT }, 3261 { O_EXCL, KF_FLAG_EXCL }, 3262 { O_EXEC, KF_FLAG_EXEC }, 3263 { O_EXLOCK, KF_FLAG_EXLOCK }, 3264 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3265 { O_SHLOCK, KF_FLAG_SHLOCK }, 3266 { O_TRUNC, KF_FLAG_TRUNC } 3267 }; 3268#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3269 struct kinfo_file *kif; 3270 struct vnode *vp; 3271 int error, locked; 3272 unsigned int i; 3273 3274 if (efbuf->remainder == 0) 3275 return (0); 3276 kif = &efbuf->kif; 3277 bzero(kif, sizeof(*kif)); 3278 locked = efbuf->fdp != NULL; 3279 switch (type) { 3280 case KF_TYPE_FIFO: 3281 case KF_TYPE_VNODE: 3282 if (locked) { 3283 FILEDESC_SUNLOCK(efbuf->fdp); 3284 locked = 0; 3285 } 3286 vp = (struct vnode *)data; 3287 error = fill_vnode_info(vp, kif); 3288 vrele(vp); 3289 break; 3290 case KF_TYPE_SOCKET: 3291 error = fill_socket_info((struct socket *)data, kif); 3292 break; 3293 case KF_TYPE_PIPE: 3294 error = fill_pipe_info((struct pipe *)data, kif); 3295 break; 3296 case KF_TYPE_PTS: 3297 error = fill_pts_info((struct tty *)data, kif); 3298 break; 3299 case KF_TYPE_PROCDESC: 3300 error = fill_procdesc_info((struct procdesc *)data, kif); 3301 break; 3302 case KF_TYPE_SEM: 3303 error = fill_sem_info((struct file *)data, kif); 3304 break; 3305 case KF_TYPE_SHM: 3306 error = fill_shm_info((struct file *)data, kif); 3307 break; 3308 default: 3309 error = 0; 3310 } 3311 if (error == 0) 3312 kif->kf_status |= KF_ATTR_VALID; 3313 3314 /* 3315 * Translate file access flags. 3316 */ 3317 for (i = 0; i < NFFLAGS; i++) 3318 if (fflags & fflags_table[i].fflag) 3319 kif->kf_flags |= fflags_table[i].kf_fflag; 3320 if (rightsp != NULL) 3321 kif->kf_cap_rights = *rightsp; 3322 else 3323 cap_rights_init(&kif->kf_cap_rights); 3324 kif->kf_fd = fd; 3325 kif->kf_type = type; 3326 kif->kf_ref_count = refcnt; 3327 kif->kf_offset = offset; 3328 /* Pack record size down */ 3329 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3330 strlen(kif->kf_path) + 1; 3331 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3332 if (efbuf->remainder != -1) { 3333 if (efbuf->remainder < kif->kf_structsize) { 3334 /* Terminate export. */ 3335 efbuf->remainder = 0; 3336 if (efbuf->fdp != NULL && !locked) 3337 FILEDESC_SLOCK(efbuf->fdp); 3338 return (0); 3339 } 3340 efbuf->remainder -= kif->kf_structsize; 3341 } 3342 if (locked) 3343 FILEDESC_SUNLOCK(efbuf->fdp); 3344 error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize); 3345 if (efbuf->fdp != NULL) 3346 FILEDESC_SLOCK(efbuf->fdp); 3347 return (error); 3348} 3349 3350/* 3351 * Store a process file descriptor information to sbuf. 3352 * 3353 * Takes a locked proc as argument, and returns with the proc unlocked. 3354 */ 3355int 3356kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3357{ 3358 struct file *fp; 3359 struct filedesc *fdp; 3360 struct export_fd_buf *efbuf; 3361 struct vnode *cttyvp, *textvp, *tracevp; 3362 int64_t offset; 3363 void *data; 3364 int error, i; 3365 int type, refcnt, fflags; 3366 cap_rights_t rights; 3367 3368 PROC_LOCK_ASSERT(p, MA_OWNED); 3369 3370 /* ktrace vnode */ 3371 tracevp = p->p_tracevp; 3372 if (tracevp != NULL) 3373 vref(tracevp); 3374 /* text vnode */ 3375 textvp = p->p_textvp; 3376 if (textvp != NULL) 3377 vref(textvp); 3378 /* Controlling tty. */ 3379 cttyvp = NULL; 3380 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3381 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3382 if (cttyvp != NULL) 3383 vref(cttyvp); 3384 } 3385 fdp = fdhold(p); 3386 PROC_UNLOCK(p); 3387 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3388 efbuf->fdp = NULL; 3389 efbuf->sb = sb; 3390 efbuf->remainder = maxlen; 3391 if (tracevp != NULL) 3392 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3393 FREAD | FWRITE, -1, -1, NULL, efbuf); 3394 if (textvp != NULL) 3395 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3396 FREAD, -1, -1, NULL, efbuf); 3397 if (cttyvp != NULL) 3398 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3399 FREAD | FWRITE, -1, -1, NULL, efbuf); 3400 error = 0; 3401 if (fdp == NULL) 3402 goto fail; 3403 efbuf->fdp = fdp; 3404 FILEDESC_SLOCK(fdp); 3405 /* working directory */ 3406 if (fdp->fd_cdir != NULL) { 3407 vref(fdp->fd_cdir); 3408 data = fdp->fd_cdir; 3409 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3410 FREAD, -1, -1, NULL, efbuf); 3411 } 3412 /* root directory */ 3413 if (fdp->fd_rdir != NULL) { 3414 vref(fdp->fd_rdir); 3415 data = fdp->fd_rdir; 3416 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3417 FREAD, -1, -1, NULL, efbuf); 3418 } 3419 /* jail directory */ 3420 if (fdp->fd_jdir != NULL) { 3421 vref(fdp->fd_jdir); 3422 data = fdp->fd_jdir; 3423 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3424 FREAD, -1, -1, NULL, efbuf); 3425 } 3426 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3427 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3428 continue; 3429 data = NULL; 3430#ifdef CAPABILITIES 3431 rights = *cap_rights(fdp, i); 3432#else /* !CAPABILITIES */ 3433 cap_rights_init(&rights); 3434#endif 3435 switch (fp->f_type) { 3436 case DTYPE_VNODE: 3437 type = KF_TYPE_VNODE; 3438 vref(fp->f_vnode); 3439 data = fp->f_vnode; 3440 break; 3441 3442 case DTYPE_SOCKET: 3443 type = KF_TYPE_SOCKET; 3444 data = fp->f_data; 3445 break; 3446 3447 case DTYPE_PIPE: 3448 type = KF_TYPE_PIPE; 3449 data = fp->f_data; 3450 break; 3451 3452 case DTYPE_FIFO: 3453 type = KF_TYPE_FIFO; 3454 vref(fp->f_vnode); 3455 data = fp->f_vnode; 3456 break; 3457 3458 case DTYPE_KQUEUE: 3459 type = KF_TYPE_KQUEUE; 3460 break; 3461 3462 case DTYPE_CRYPTO: 3463 type = KF_TYPE_CRYPTO; 3464 break; 3465 3466 case DTYPE_MQUEUE: 3467 type = KF_TYPE_MQUEUE; 3468 break; 3469 3470 case DTYPE_SHM: 3471 type = KF_TYPE_SHM; 3472 data = fp; 3473 break; 3474 3475 case DTYPE_SEM: 3476 type = KF_TYPE_SEM; 3477 data = fp; 3478 break; 3479 3480 case DTYPE_PTS: 3481 type = KF_TYPE_PTS; 3482 data = fp->f_data; 3483 break; 3484 3485#ifdef PROCDESC 3486 case DTYPE_PROCDESC: 3487 type = KF_TYPE_PROCDESC; 3488 data = fp->f_data; 3489 break; 3490#endif 3491 3492 default: 3493 type = KF_TYPE_UNKNOWN; 3494 break; 3495 } 3496 refcnt = fp->f_count; 3497 fflags = fp->f_flag; 3498 offset = foffset_get(fp); 3499 3500 /* 3501 * Create sysctl entry. 3502 * It is OK to drop the filedesc lock here as we will 3503 * re-validate and re-evaluate its properties when 3504 * the loop continues. 3505 */ 3506 error = export_fd_to_sb(data, type, i, fflags, refcnt, 3507 offset, &rights, efbuf); 3508 if (error != 0) 3509 break; 3510 } 3511 FILEDESC_SUNLOCK(fdp); 3512 fddrop(fdp); 3513fail: 3514 free(efbuf, M_TEMP); 3515 return (error); 3516} 3517 3518#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3519 3520/* 3521 * Get per-process file descriptors for use by procstat(1), et al. 3522 */ 3523static int 3524sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3525{ 3526 struct sbuf sb; 3527 struct proc *p; 3528 ssize_t maxlen; 3529 int error, error2, *name; 3530 3531 name = (int *)arg1; 3532 3533 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3534 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3535 if (error != 0) { 3536 sbuf_delete(&sb); 3537 return (error); 3538 } 3539 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3540 error = kern_proc_filedesc_out(p, &sb, maxlen); 3541 error2 = sbuf_finish(&sb); 3542 sbuf_delete(&sb); 3543 return (error != 0 ? error : error2); 3544} 3545 3546int 3547vntype_to_kinfo(int vtype) 3548{ 3549 struct { 3550 int vtype; 3551 int kf_vtype; 3552 } vtypes_table[] = { 3553 { VBAD, KF_VTYPE_VBAD }, 3554 { VBLK, KF_VTYPE_VBLK }, 3555 { VCHR, KF_VTYPE_VCHR }, 3556 { VDIR, KF_VTYPE_VDIR }, 3557 { VFIFO, KF_VTYPE_VFIFO }, 3558 { VLNK, KF_VTYPE_VLNK }, 3559 { VNON, KF_VTYPE_VNON }, 3560 { VREG, KF_VTYPE_VREG }, 3561 { VSOCK, KF_VTYPE_VSOCK } 3562 }; 3563#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3564 unsigned int i; 3565 3566 /* 3567 * Perform vtype translation. 3568 */ 3569 for (i = 0; i < NVTYPES; i++) 3570 if (vtypes_table[i].vtype == vtype) 3571 break; 3572 if (i < NVTYPES) 3573 return (vtypes_table[i].kf_vtype); 3574 3575 return (KF_VTYPE_UNKNOWN); 3576} 3577 3578static int 3579fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3580{ 3581 struct vattr va; 3582 char *fullpath, *freepath; 3583 int error; 3584 3585 if (vp == NULL) 3586 return (1); 3587 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3588 freepath = NULL; 3589 fullpath = "-"; 3590 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3591 if (error == 0) { 3592 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3593 } 3594 if (freepath != NULL) 3595 free(freepath, M_TEMP); 3596 3597 /* 3598 * Retrieve vnode attributes. 3599 */ 3600 va.va_fsid = VNOVAL; 3601 va.va_rdev = NODEV; 3602 vn_lock(vp, LK_SHARED | LK_RETRY); 3603 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3604 VOP_UNLOCK(vp, 0); 3605 if (error != 0) 3606 return (error); 3607 if (va.va_fsid != VNOVAL) 3608 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3609 else 3610 kif->kf_un.kf_file.kf_file_fsid = 3611 vp->v_mount->mnt_stat.f_fsid.val[0]; 3612 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3613 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3614 kif->kf_un.kf_file.kf_file_size = va.va_size; 3615 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3616 return (0); 3617} 3618 3619static int 3620fill_socket_info(struct socket *so, struct kinfo_file *kif) 3621{ 3622 struct sockaddr *sa; 3623 struct inpcb *inpcb; 3624 struct unpcb *unpcb; 3625 int error; 3626 3627 if (so == NULL) 3628 return (1); 3629 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3630 kif->kf_sock_type = so->so_type; 3631 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3632 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3633 switch(kif->kf_sock_domain) { 3634 case AF_INET: 3635 case AF_INET6: 3636 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3637 if (so->so_pcb != NULL) { 3638 inpcb = (struct inpcb *)(so->so_pcb); 3639 kif->kf_un.kf_sock.kf_sock_inpcb = 3640 (uintptr_t)inpcb->inp_ppcb; 3641 } 3642 } 3643 break; 3644 case AF_UNIX: 3645 if (so->so_pcb != NULL) { 3646 unpcb = (struct unpcb *)(so->so_pcb); 3647 if (unpcb->unp_conn) { 3648 kif->kf_un.kf_sock.kf_sock_unpconn = 3649 (uintptr_t)unpcb->unp_conn; 3650 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3651 so->so_rcv.sb_state; 3652 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3653 so->so_snd.sb_state; 3654 } 3655 } 3656 break; 3657 } 3658 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3659 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3660 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3661 free(sa, M_SONAME); 3662 } 3663 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3664 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3665 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3666 free(sa, M_SONAME); 3667 } 3668 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3669 sizeof(kif->kf_path)); 3670 return (0); 3671} 3672 3673static int 3674fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3675{ 3676 3677 if (tp == NULL) 3678 return (1); 3679 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3680 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3681 return (0); 3682} 3683 3684static int 3685fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3686{ 3687 3688 if (pi == NULL) 3689 return (1); 3690 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3691 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3692 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3693 return (0); 3694} 3695 3696static int 3697fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3698{ 3699 3700 if (pdp == NULL) 3701 return (1); 3702 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3703 return (0); 3704} 3705 3706static int 3707fill_sem_info(struct file *fp, struct kinfo_file *kif) 3708{ 3709 struct thread *td; 3710 struct stat sb; 3711 3712 td = curthread; 3713 if (fp->f_data == NULL) 3714 return (1); 3715 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3716 return (1); 3717 if (ksem_info == NULL) 3718 return (1); 3719 ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path), 3720 &kif->kf_un.kf_sem.kf_sem_value); 3721 kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode; 3722 return (0); 3723} 3724 3725static int 3726fill_shm_info(struct file *fp, struct kinfo_file *kif) 3727{ 3728 struct thread *td; 3729 struct stat sb; 3730 3731 td = curthread; 3732 if (fp->f_data == NULL) 3733 return (1); 3734 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3735 return (1); 3736 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3737 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3738 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3739 return (0); 3740} 3741 3742static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3743 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3744 "Process filedesc entries"); 3745 3746#ifdef DDB 3747/* 3748 * For the purposes of debugging, generate a human-readable string for the 3749 * file type. 3750 */ 3751static const char * 3752file_type_to_name(short type) 3753{ 3754 3755 switch (type) { 3756 case 0: 3757 return ("zero"); 3758 case DTYPE_VNODE: 3759 return ("vnod"); 3760 case DTYPE_SOCKET: 3761 return ("sock"); 3762 case DTYPE_PIPE: 3763 return ("pipe"); 3764 case DTYPE_FIFO: 3765 return ("fifo"); 3766 case DTYPE_KQUEUE: 3767 return ("kque"); 3768 case DTYPE_CRYPTO: 3769 return ("crpt"); 3770 case DTYPE_MQUEUE: 3771 return ("mque"); 3772 case DTYPE_SHM: 3773 return ("shm"); 3774 case DTYPE_SEM: 3775 return ("ksem"); 3776 default: 3777 return ("unkn"); 3778 } 3779} 3780 3781/* 3782 * For the purposes of debugging, identify a process (if any, perhaps one of 3783 * many) that references the passed file in its file descriptor array. Return 3784 * NULL if none. 3785 */ 3786static struct proc * 3787file_to_first_proc(struct file *fp) 3788{ 3789 struct filedesc *fdp; 3790 struct proc *p; 3791 int n; 3792 3793 FOREACH_PROC_IN_SYSTEM(p) { 3794 if (p->p_state == PRS_NEW) 3795 continue; 3796 fdp = p->p_fd; 3797 if (fdp == NULL) 3798 continue; 3799 for (n = 0; n <= fdp->fd_lastfile; n++) { 3800 if (fp == fdp->fd_ofiles[n].fde_file) 3801 return (p); 3802 } 3803 } 3804 return (NULL); 3805} 3806 3807static void 3808db_print_file(struct file *fp, int header) 3809{ 3810 struct proc *p; 3811 3812 if (header) 3813 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3814 "File", "Type", "Data", "Flag", "GCFl", "Count", 3815 "MCount", "Vnode", "FPID", "FCmd"); 3816 p = file_to_first_proc(fp); 3817 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3818 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3819 0, fp->f_count, 0, fp->f_vnode, 3820 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3821} 3822 3823DB_SHOW_COMMAND(file, db_show_file) 3824{ 3825 struct file *fp; 3826 3827 if (!have_addr) { 3828 db_printf("usage: show file <addr>\n"); 3829 return; 3830 } 3831 fp = (struct file *)addr; 3832 db_print_file(fp, 1); 3833} 3834 3835DB_SHOW_COMMAND(files, db_show_files) 3836{ 3837 struct filedesc *fdp; 3838 struct file *fp; 3839 struct proc *p; 3840 int header; 3841 int n; 3842 3843 header = 1; 3844 FOREACH_PROC_IN_SYSTEM(p) { 3845 if (p->p_state == PRS_NEW) 3846 continue; 3847 if ((fdp = p->p_fd) == NULL) 3848 continue; 3849 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3850 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3851 continue; 3852 db_print_file(fp, header); 3853 header = 0; 3854 } 3855 } 3856} 3857#endif 3858 3859SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3860 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3861 3862SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3863 &maxfiles, 0, "Maximum number of files"); 3864 3865SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3866 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3867 3868/* ARGSUSED*/ 3869static void 3870filelistinit(void *dummy) 3871{ 3872 3873 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3874 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3875 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3876 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3877} 3878SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3879 3880/*-------------------------------------------------------------------*/ 3881 3882static int 3883badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3884 int flags, struct thread *td) 3885{ 3886 3887 return (EBADF); 3888} 3889 3890static int 3891badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3892 struct thread *td) 3893{ 3894 3895 return (EINVAL); 3896} 3897 3898static int 3899badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3900 struct thread *td) 3901{ 3902 3903 return (EBADF); 3904} 3905 3906static int 3907badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3908 struct thread *td) 3909{ 3910 3911 return (0); 3912} 3913 3914static int 3915badfo_kqfilter(struct file *fp, struct knote *kn) 3916{ 3917 3918 return (EBADF); 3919} 3920 3921static int 3922badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3923 struct thread *td) 3924{ 3925 3926 return (EBADF); 3927} 3928 3929static int 3930badfo_close(struct file *fp, struct thread *td) 3931{ 3932 3933 return (EBADF); 3934} 3935 3936static int 3937badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3938 struct thread *td) 3939{ 3940 3941 return (EBADF); 3942} 3943 3944static int 3945badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3946 struct thread *td) 3947{ 3948 3949 return (EBADF); 3950} 3951 3952static int 3953badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3954 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3955 int kflags, struct thread *td) 3956{ 3957 3958 return (EBADF); 3959} 3960 3961struct fileops badfileops = { 3962 .fo_read = badfo_readwrite, 3963 .fo_write = badfo_readwrite, 3964 .fo_truncate = badfo_truncate, 3965 .fo_ioctl = badfo_ioctl, 3966 .fo_poll = badfo_poll, 3967 .fo_kqfilter = badfo_kqfilter, 3968 .fo_stat = badfo_stat, 3969 .fo_close = badfo_close, 3970 .fo_chmod = badfo_chmod, 3971 .fo_chown = badfo_chown, 3972 .fo_sendfile = badfo_sendfile, 3973}; 3974 3975int 3976invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3977 struct thread *td) 3978{ 3979 3980 return (EINVAL); 3981} 3982 3983int 3984invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3985 struct thread *td) 3986{ 3987 3988 return (EINVAL); 3989} 3990 3991int 3992invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3993 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3994 int kflags, struct thread *td) 3995{ 3996 3997 return (EINVAL); 3998} 3999 4000/*-------------------------------------------------------------------*/ 4001 4002/* 4003 * File Descriptor pseudo-device driver (/dev/fd/). 4004 * 4005 * Opening minor device N dup()s the file (if any) connected to file 4006 * descriptor N belonging to the calling process. Note that this driver 4007 * consists of only the ``open()'' routine, because all subsequent 4008 * references to this file will be direct to the other driver. 4009 * 4010 * XXX: we could give this one a cloning event handler if necessary. 4011 */ 4012 4013/* ARGSUSED */ 4014static int 4015fdopen(struct cdev *dev, int mode, int type, struct thread *td) 4016{ 4017 4018 /* 4019 * XXX Kludge: set curthread->td_dupfd to contain the value of the 4020 * the file descriptor being sought for duplication. The error 4021 * return ensures that the vnode for this device will be released 4022 * by vn_open. Open will detect this special error and take the 4023 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 4024 * will simply report the error. 4025 */ 4026 td->td_dupfd = dev2unit(dev); 4027 return (ENODEV); 4028} 4029 4030static struct cdevsw fildesc_cdevsw = { 4031 .d_version = D_VERSION, 4032 .d_open = fdopen, 4033 .d_name = "FD", 4034}; 4035 4036static void 4037fildesc_drvinit(void *unused) 4038{ 4039 struct cdev *dev; 4040 4041 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4042 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4043 make_dev_alias(dev, "stdin"); 4044 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4045 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4046 make_dev_alias(dev, "stdout"); 4047 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4048 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4049 make_dev_alias(dev, "stderr"); 4050} 4051 4052SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4053