kern_descrip.c revision 268001
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: head/sys/kern/kern_descrip.c 268001 2014-06-28 05:41:53Z mjg $"); 39 40#include "opt_capsicum.h" 41#include "opt_compat.h" 42#include "opt_ddb.h" 43#include "opt_ktrace.h" 44 45#include <sys/param.h> 46#include <sys/systm.h> 47 48#include <sys/capsicum.h> 49#include <sys/conf.h> 50#include <sys/domain.h> 51#include <sys/fcntl.h> 52#include <sys/file.h> 53#include <sys/filedesc.h> 54#include <sys/filio.h> 55#include <sys/jail.h> 56#include <sys/kernel.h> 57#include <sys/ksem.h> 58#include <sys/limits.h> 59#include <sys/lock.h> 60#include <sys/malloc.h> 61#include <sys/mman.h> 62#include <sys/mount.h> 63#include <sys/mqueue.h> 64#include <sys/mutex.h> 65#include <sys/namei.h> 66#include <sys/selinfo.h> 67#include <sys/pipe.h> 68#include <sys/priv.h> 69#include <sys/proc.h> 70#include <sys/procdesc.h> 71#include <sys/protosw.h> 72#include <sys/racct.h> 73#include <sys/resourcevar.h> 74#include <sys/sbuf.h> 75#include <sys/signalvar.h> 76#include <sys/socketvar.h> 77#include <sys/stat.h> 78#include <sys/sx.h> 79#include <sys/syscallsubr.h> 80#include <sys/sysctl.h> 81#include <sys/sysproto.h> 82#include <sys/tty.h> 83#include <sys/unistd.h> 84#include <sys/un.h> 85#include <sys/unpcb.h> 86#include <sys/user.h> 87#include <sys/vnode.h> 88#ifdef KTRACE 89#include <sys/ktrace.h> 90#endif 91 92#include <net/vnet.h> 93 94#include <netinet/in.h> 95#include <netinet/in_pcb.h> 96 97#include <security/audit/audit.h> 98 99#include <vm/uma.h> 100#include <vm/vm.h> 101 102#include <ddb/ddb.h> 103 104static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 105static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 106 "file desc to leader structures"); 107static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 108MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 109 110MALLOC_DECLARE(M_FADVISE); 111 112static uma_zone_t file_zone; 113 114void (*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value); 115 116static int closefp(struct filedesc *fdp, int fd, struct file *fp, 117 struct thread *td, int holdleaders); 118static int fd_first_free(struct filedesc *fdp, int low, int size); 119static int fd_last_used(struct filedesc *fdp, int size); 120static void fdgrowtable(struct filedesc *fdp, int nfd); 121static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 122static void fdunused(struct filedesc *fdp, int fd); 123static void fdused(struct filedesc *fdp, int fd); 124static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif); 125static int fill_procdesc_info(struct procdesc *pdp, 126 struct kinfo_file *kif); 127static int fill_pts_info(struct tty *tp, struct kinfo_file *kif); 128static int fill_sem_info(struct file *fp, struct kinfo_file *kif); 129static int fill_shm_info(struct file *fp, struct kinfo_file *kif); 130static int fill_socket_info(struct socket *so, struct kinfo_file *kif); 131static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif); 132static int getmaxfd(struct proc *p); 133 134/* 135 * Each process has: 136 * 137 * - An array of open file descriptors (fd_ofiles) 138 * - An array of file flags (fd_ofileflags) 139 * - A bitmap recording which descriptors are in use (fd_map) 140 * 141 * A process starts out with NDFILE descriptors. The value of NDFILE has 142 * been selected based the historical limit of 20 open files, and an 143 * assumption that the majority of processes, especially short-lived 144 * processes like shells, will never need more. 145 * 146 * If this initial allocation is exhausted, a larger descriptor table and 147 * map are allocated dynamically, and the pointers in the process's struct 148 * filedesc are updated to point to those. This is repeated every time 149 * the process runs out of file descriptors (provided it hasn't hit its 150 * resource limit). 151 * 152 * Since threads may hold references to individual descriptor table 153 * entries, the tables are never freed. Instead, they are placed on a 154 * linked list and freed only when the struct filedesc is released. 155 */ 156#define NDFILE 20 157#define NDSLOTSIZE sizeof(NDSLOTTYPE) 158#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 159#define NDSLOT(x) ((x) / NDENTRIES) 160#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 161#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 162 163/* 164 * SLIST entry used to keep track of ofiles which must be reclaimed when 165 * the process exits. 166 */ 167struct freetable { 168 struct filedescent *ft_table; 169 SLIST_ENTRY(freetable) ft_next; 170}; 171 172/* 173 * Initial allocation: a filedesc structure + the head of SLIST used to 174 * keep track of old ofiles + enough space for NDFILE descriptors. 175 */ 176struct filedesc0 { 177 struct filedesc fd_fd; 178 SLIST_HEAD(, freetable) fd_free; 179 struct filedescent fd_dfiles[NDFILE]; 180 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 181}; 182 183/* 184 * Descriptor management. 185 */ 186volatile int openfiles; /* actual number of open files */ 187struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 188void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 189 190/* A mutex to protect the association between a proc and filedesc. */ 191static struct mtx fdesc_mtx; 192 193/* 194 * If low >= size, just return low. Otherwise find the first zero bit in the 195 * given bitmap, starting at low and not exceeding size - 1. Return size if 196 * not found. 197 */ 198static int 199fd_first_free(struct filedesc *fdp, int low, int size) 200{ 201 NDSLOTTYPE *map = fdp->fd_map; 202 NDSLOTTYPE mask; 203 int off, maxoff; 204 205 if (low >= size) 206 return (low); 207 208 off = NDSLOT(low); 209 if (low % NDENTRIES) { 210 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 211 if ((mask &= ~map[off]) != 0UL) 212 return (off * NDENTRIES + ffsl(mask) - 1); 213 ++off; 214 } 215 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 216 if (map[off] != ~0UL) 217 return (off * NDENTRIES + ffsl(~map[off]) - 1); 218 return (size); 219} 220 221/* 222 * Find the highest non-zero bit in the given bitmap, starting at 0 and 223 * not exceeding size - 1. Return -1 if not found. 224 */ 225static int 226fd_last_used(struct filedesc *fdp, int size) 227{ 228 NDSLOTTYPE *map = fdp->fd_map; 229 NDSLOTTYPE mask; 230 int off, minoff; 231 232 off = NDSLOT(size); 233 if (size % NDENTRIES) { 234 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 235 if ((mask &= map[off]) != 0) 236 return (off * NDENTRIES + flsl(mask) - 1); 237 --off; 238 } 239 for (minoff = NDSLOT(0); off >= minoff; --off) 240 if (map[off] != 0) 241 return (off * NDENTRIES + flsl(map[off]) - 1); 242 return (-1); 243} 244 245static int 246fdisused(struct filedesc *fdp, int fd) 247{ 248 249 FILEDESC_LOCK_ASSERT(fdp); 250 251 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 252 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 253 254 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 255} 256 257/* 258 * Mark a file descriptor as used. 259 */ 260static void 261fdused(struct filedesc *fdp, int fd) 262{ 263 264 FILEDESC_XLOCK_ASSERT(fdp); 265 266 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 267 268 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 269 if (fd > fdp->fd_lastfile) 270 fdp->fd_lastfile = fd; 271 if (fd == fdp->fd_freefile) 272 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 273} 274 275/* 276 * Mark a file descriptor as unused. 277 */ 278static void 279fdunused(struct filedesc *fdp, int fd) 280{ 281 282 FILEDESC_XLOCK_ASSERT(fdp); 283 284 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 285 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 286 ("fd=%d is still in use", fd)); 287 288 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 289 if (fd < fdp->fd_freefile) 290 fdp->fd_freefile = fd; 291 if (fd == fdp->fd_lastfile) 292 fdp->fd_lastfile = fd_last_used(fdp, fd); 293} 294 295/* 296 * Free a file descriptor. 297 */ 298static inline void 299fdfree(struct filedesc *fdp, int fd) 300{ 301 struct filedescent *fde; 302 303 fde = &fdp->fd_ofiles[fd]; 304 filecaps_free(&fde->fde_caps); 305 bzero(fde, sizeof(*fde)); 306 fdunused(fdp, fd); 307} 308 309/* 310 * System calls on descriptors. 311 */ 312#ifndef _SYS_SYSPROTO_H_ 313struct getdtablesize_args { 314 int dummy; 315}; 316#endif 317/* ARGSUSED */ 318int 319sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 320{ 321 struct proc *p = td->td_proc; 322 uint64_t lim; 323 324 PROC_LOCK(p); 325 td->td_retval[0] = 326 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 327 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 328 PROC_UNLOCK(p); 329 if (lim < td->td_retval[0]) 330 td->td_retval[0] = lim; 331 return (0); 332} 333 334/* 335 * Duplicate a file descriptor to a particular value. 336 * 337 * Note: keep in mind that a potential race condition exists when closing 338 * descriptors from a shared descriptor table (via rfork). 339 */ 340#ifndef _SYS_SYSPROTO_H_ 341struct dup2_args { 342 u_int from; 343 u_int to; 344}; 345#endif 346/* ARGSUSED */ 347int 348sys_dup2(struct thread *td, struct dup2_args *uap) 349{ 350 351 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 352 td->td_retval)); 353} 354 355/* 356 * Duplicate a file descriptor. 357 */ 358#ifndef _SYS_SYSPROTO_H_ 359struct dup_args { 360 u_int fd; 361}; 362#endif 363/* ARGSUSED */ 364int 365sys_dup(struct thread *td, struct dup_args *uap) 366{ 367 368 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 369} 370 371/* 372 * The file control system call. 373 */ 374#ifndef _SYS_SYSPROTO_H_ 375struct fcntl_args { 376 int fd; 377 int cmd; 378 long arg; 379}; 380#endif 381/* ARGSUSED */ 382int 383sys_fcntl(struct thread *td, struct fcntl_args *uap) 384{ 385 struct flock fl; 386 struct __oflock ofl; 387 intptr_t arg; 388 int error; 389 int cmd; 390 391 error = 0; 392 cmd = uap->cmd; 393 switch (uap->cmd) { 394 case F_OGETLK: 395 case F_OSETLK: 396 case F_OSETLKW: 397 /* 398 * Convert old flock structure to new. 399 */ 400 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 401 fl.l_start = ofl.l_start; 402 fl.l_len = ofl.l_len; 403 fl.l_pid = ofl.l_pid; 404 fl.l_type = ofl.l_type; 405 fl.l_whence = ofl.l_whence; 406 fl.l_sysid = 0; 407 408 switch (uap->cmd) { 409 case F_OGETLK: 410 cmd = F_GETLK; 411 break; 412 case F_OSETLK: 413 cmd = F_SETLK; 414 break; 415 case F_OSETLKW: 416 cmd = F_SETLKW; 417 break; 418 } 419 arg = (intptr_t)&fl; 420 break; 421 case F_GETLK: 422 case F_SETLK: 423 case F_SETLKW: 424 case F_SETLK_REMOTE: 425 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 426 arg = (intptr_t)&fl; 427 break; 428 default: 429 arg = uap->arg; 430 break; 431 } 432 if (error) 433 return (error); 434 error = kern_fcntl(td, uap->fd, cmd, arg); 435 if (error) 436 return (error); 437 if (uap->cmd == F_OGETLK) { 438 ofl.l_start = fl.l_start; 439 ofl.l_len = fl.l_len; 440 ofl.l_pid = fl.l_pid; 441 ofl.l_type = fl.l_type; 442 ofl.l_whence = fl.l_whence; 443 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 444 } else if (uap->cmd == F_GETLK) { 445 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 446 } 447 return (error); 448} 449 450int 451kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 452{ 453 struct filedesc *fdp; 454 struct flock *flp; 455 struct file *fp, *fp2; 456 struct filedescent *fde; 457 struct proc *p; 458 struct vnode *vp; 459 cap_rights_t rights; 460 int error, flg, tmp; 461 u_int old, new; 462 uint64_t bsize; 463 off_t foffset; 464 465 error = 0; 466 flg = F_POSIX; 467 p = td->td_proc; 468 fdp = p->p_fd; 469 470 switch (cmd) { 471 case F_DUPFD: 472 tmp = arg; 473 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 474 break; 475 476 case F_DUPFD_CLOEXEC: 477 tmp = arg; 478 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp, 479 td->td_retval); 480 break; 481 482 case F_DUP2FD: 483 tmp = arg; 484 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 485 break; 486 487 case F_DUP2FD_CLOEXEC: 488 tmp = arg; 489 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp, 490 td->td_retval); 491 break; 492 493 case F_GETFD: 494 FILEDESC_SLOCK(fdp); 495 if ((fp = fget_locked(fdp, fd)) == NULL) { 496 FILEDESC_SUNLOCK(fdp); 497 error = EBADF; 498 break; 499 } 500 fde = &fdp->fd_ofiles[fd]; 501 td->td_retval[0] = 502 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 503 FILEDESC_SUNLOCK(fdp); 504 break; 505 506 case F_SETFD: 507 FILEDESC_XLOCK(fdp); 508 if ((fp = fget_locked(fdp, fd)) == NULL) { 509 FILEDESC_XUNLOCK(fdp); 510 error = EBADF; 511 break; 512 } 513 fde = &fdp->fd_ofiles[fd]; 514 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 515 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 516 FILEDESC_XUNLOCK(fdp); 517 break; 518 519 case F_GETFL: 520 error = fget_unlocked(fdp, fd, 521 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL); 522 if (error != 0) 523 break; 524 td->td_retval[0] = OFLAGS(fp->f_flag); 525 fdrop(fp, td); 526 break; 527 528 case F_SETFL: 529 error = fget_unlocked(fdp, fd, 530 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL); 531 if (error != 0) 532 break; 533 do { 534 tmp = flg = fp->f_flag; 535 tmp &= ~FCNTLFLAGS; 536 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 537 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 538 tmp = fp->f_flag & FNONBLOCK; 539 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 540 if (error != 0) { 541 fdrop(fp, td); 542 break; 543 } 544 tmp = fp->f_flag & FASYNC; 545 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 546 if (error == 0) { 547 fdrop(fp, td); 548 break; 549 } 550 atomic_clear_int(&fp->f_flag, FNONBLOCK); 551 tmp = 0; 552 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 553 fdrop(fp, td); 554 break; 555 556 case F_GETOWN: 557 error = fget_unlocked(fdp, fd, 558 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL); 559 if (error != 0) 560 break; 561 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 562 if (error == 0) 563 td->td_retval[0] = tmp; 564 fdrop(fp, td); 565 break; 566 567 case F_SETOWN: 568 error = fget_unlocked(fdp, fd, 569 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL); 570 if (error != 0) 571 break; 572 tmp = arg; 573 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 574 fdrop(fp, td); 575 break; 576 577 case F_SETLK_REMOTE: 578 error = priv_check(td, PRIV_NFS_LOCKD); 579 if (error) 580 return (error); 581 flg = F_REMOTE; 582 goto do_setlk; 583 584 case F_SETLKW: 585 flg |= F_WAIT; 586 /* FALLTHROUGH F_SETLK */ 587 588 case F_SETLK: 589 do_setlk: 590 cap_rights_init(&rights, CAP_FLOCK); 591 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL); 592 if (error != 0) 593 break; 594 if (fp->f_type != DTYPE_VNODE) { 595 error = EBADF; 596 fdrop(fp, td); 597 break; 598 } 599 600 flp = (struct flock *)arg; 601 if (flp->l_whence == SEEK_CUR) { 602 foffset = foffset_get(fp); 603 if (foffset < 0 || 604 (flp->l_start > 0 && 605 foffset > OFF_MAX - flp->l_start)) { 606 FILEDESC_SUNLOCK(fdp); 607 error = EOVERFLOW; 608 fdrop(fp, td); 609 break; 610 } 611 flp->l_start += foffset; 612 } 613 614 vp = fp->f_vnode; 615 switch (flp->l_type) { 616 case F_RDLCK: 617 if ((fp->f_flag & FREAD) == 0) { 618 error = EBADF; 619 break; 620 } 621 PROC_LOCK(p->p_leader); 622 p->p_leader->p_flag |= P_ADVLOCK; 623 PROC_UNLOCK(p->p_leader); 624 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 625 flp, flg); 626 break; 627 case F_WRLCK: 628 if ((fp->f_flag & FWRITE) == 0) { 629 error = EBADF; 630 break; 631 } 632 PROC_LOCK(p->p_leader); 633 p->p_leader->p_flag |= P_ADVLOCK; 634 PROC_UNLOCK(p->p_leader); 635 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 636 flp, flg); 637 break; 638 case F_UNLCK: 639 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 640 flp, flg); 641 break; 642 case F_UNLCKSYS: 643 /* 644 * Temporary api for testing remote lock 645 * infrastructure. 646 */ 647 if (flg != F_REMOTE) { 648 error = EINVAL; 649 break; 650 } 651 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 652 F_UNLCKSYS, flp, flg); 653 break; 654 default: 655 error = EINVAL; 656 break; 657 } 658 if (error != 0 || flp->l_type == F_UNLCK || 659 flp->l_type == F_UNLCKSYS) { 660 fdrop(fp, td); 661 break; 662 } 663 664 /* 665 * Check for a race with close. 666 * 667 * The vnode is now advisory locked (or unlocked, but this case 668 * is not really important) as the caller requested. 669 * We had to drop the filedesc lock, so we need to recheck if 670 * the descriptor is still valid, because if it was closed 671 * in the meantime we need to remove advisory lock from the 672 * vnode - close on any descriptor leading to an advisory 673 * locked vnode, removes that lock. 674 * We will return 0 on purpose in that case, as the result of 675 * successful advisory lock might have been externally visible 676 * already. This is fine - effectively we pretend to the caller 677 * that the closing thread was a bit slower and that the 678 * advisory lock succeeded before the close. 679 */ 680 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL); 681 if (error != 0) { 682 fdrop(fp, td); 683 break; 684 } 685 if (fp != fp2) { 686 flp->l_whence = SEEK_SET; 687 flp->l_start = 0; 688 flp->l_len = 0; 689 flp->l_type = F_UNLCK; 690 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 691 F_UNLCK, flp, F_POSIX); 692 } 693 fdrop(fp, td); 694 fdrop(fp2, td); 695 break; 696 697 case F_GETLK: 698 error = fget_unlocked(fdp, fd, 699 cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL); 700 if (error != 0) 701 break; 702 if (fp->f_type != DTYPE_VNODE) { 703 error = EBADF; 704 fdrop(fp, td); 705 break; 706 } 707 flp = (struct flock *)arg; 708 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 709 flp->l_type != F_UNLCK) { 710 error = EINVAL; 711 fdrop(fp, td); 712 break; 713 } 714 if (flp->l_whence == SEEK_CUR) { 715 foffset = foffset_get(fp); 716 if ((flp->l_start > 0 && 717 foffset > OFF_MAX - flp->l_start) || 718 (flp->l_start < 0 && 719 foffset < OFF_MIN - flp->l_start)) { 720 FILEDESC_SUNLOCK(fdp); 721 error = EOVERFLOW; 722 fdrop(fp, td); 723 break; 724 } 725 flp->l_start += foffset; 726 } 727 vp = fp->f_vnode; 728 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 729 F_POSIX); 730 fdrop(fp, td); 731 break; 732 733 case F_RDAHEAD: 734 arg = arg ? 128 * 1024: 0; 735 /* FALLTHROUGH */ 736 case F_READAHEAD: 737 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL); 738 if (error != 0) 739 break; 740 if (fp->f_type != DTYPE_VNODE) { 741 fdrop(fp, td); 742 error = EBADF; 743 break; 744 } 745 if (arg >= 0) { 746 vp = fp->f_vnode; 747 error = vn_lock(vp, LK_SHARED); 748 if (error != 0) { 749 fdrop(fp, td); 750 break; 751 } 752 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 753 VOP_UNLOCK(vp, 0); 754 fp->f_seqcount = (arg + bsize - 1) / bsize; 755 do { 756 new = old = fp->f_flag; 757 new |= FRDAHEAD; 758 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 759 } else { 760 do { 761 new = old = fp->f_flag; 762 new &= ~FRDAHEAD; 763 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 764 } 765 fdrop(fp, td); 766 break; 767 768 default: 769 error = EINVAL; 770 break; 771 } 772 return (error); 773} 774 775static int 776getmaxfd(struct proc *p) 777{ 778 int maxfd; 779 780 PROC_LOCK(p); 781 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 782 PROC_UNLOCK(p); 783 784 return (maxfd); 785} 786 787/* 788 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 789 */ 790int 791do_dup(struct thread *td, int flags, int old, int new, 792 register_t *retval) 793{ 794 struct filedesc *fdp; 795 struct filedescent *oldfde, *newfde; 796 struct proc *p; 797 struct file *fp; 798 struct file *delfp; 799 int error, maxfd; 800 801 p = td->td_proc; 802 fdp = p->p_fd; 803 804 /* 805 * Verify we have a valid descriptor to dup from and possibly to 806 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 807 * return EINVAL when the new descriptor is out of bounds. 808 */ 809 if (old < 0) 810 return (EBADF); 811 if (new < 0) 812 return (flags & DUP_FCNTL ? EINVAL : EBADF); 813 maxfd = getmaxfd(p); 814 if (new >= maxfd) 815 return (flags & DUP_FCNTL ? EINVAL : EBADF); 816 817 FILEDESC_XLOCK(fdp); 818 if (fget_locked(fdp, old) == NULL) { 819 FILEDESC_XUNLOCK(fdp); 820 return (EBADF); 821 } 822 oldfde = &fdp->fd_ofiles[old]; 823 if (flags & DUP_FIXED && old == new) { 824 *retval = new; 825 if (flags & DUP_CLOEXEC) 826 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 827 FILEDESC_XUNLOCK(fdp); 828 return (0); 829 } 830 fp = oldfde->fde_file; 831 fhold(fp); 832 833 /* 834 * If the caller specified a file descriptor, make sure the file 835 * table is large enough to hold it, and grab it. Otherwise, just 836 * allocate a new descriptor the usual way. 837 */ 838 if (flags & DUP_FIXED) { 839 if (new >= fdp->fd_nfiles) { 840 /* 841 * The resource limits are here instead of e.g. 842 * fdalloc(), because the file descriptor table may be 843 * shared between processes, so we can't really use 844 * racct_add()/racct_sub(). Instead of counting the 845 * number of actually allocated descriptors, just put 846 * the limit on the size of the file descriptor table. 847 */ 848#ifdef RACCT 849 PROC_LOCK(p); 850 error = racct_set(p, RACCT_NOFILE, new + 1); 851 PROC_UNLOCK(p); 852 if (error != 0) { 853 FILEDESC_XUNLOCK(fdp); 854 fdrop(fp, td); 855 return (EMFILE); 856 } 857#endif 858 fdgrowtable_exp(fdp, new + 1); 859 oldfde = &fdp->fd_ofiles[old]; 860 } 861 newfde = &fdp->fd_ofiles[new]; 862 if (newfde->fde_file == NULL) 863 fdused(fdp, new); 864 } else { 865 if ((error = fdalloc(td, new, &new)) != 0) { 866 FILEDESC_XUNLOCK(fdp); 867 fdrop(fp, td); 868 return (error); 869 } 870 newfde = &fdp->fd_ofiles[new]; 871 } 872 873 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 874 KASSERT(old != new, ("new fd is same as old")); 875 876 delfp = newfde->fde_file; 877 878 /* 879 * Duplicate the source descriptor. 880 */ 881 filecaps_free(&newfde->fde_caps); 882 *newfde = *oldfde; 883 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 884 if ((flags & DUP_CLOEXEC) != 0) 885 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 886 else 887 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 888 *retval = new; 889 890 if (delfp != NULL) { 891 (void) closefp(fdp, new, delfp, td, 1); 892 /* closefp() drops the FILEDESC lock for us. */ 893 } else { 894 FILEDESC_XUNLOCK(fdp); 895 } 896 897 return (0); 898} 899 900/* 901 * If sigio is on the list associated with a process or process group, 902 * disable signalling from the device, remove sigio from the list and 903 * free sigio. 904 */ 905void 906funsetown(struct sigio **sigiop) 907{ 908 struct sigio *sigio; 909 910 SIGIO_LOCK(); 911 sigio = *sigiop; 912 if (sigio == NULL) { 913 SIGIO_UNLOCK(); 914 return; 915 } 916 *(sigio->sio_myref) = NULL; 917 if ((sigio)->sio_pgid < 0) { 918 struct pgrp *pg = (sigio)->sio_pgrp; 919 PGRP_LOCK(pg); 920 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 921 sigio, sio_pgsigio); 922 PGRP_UNLOCK(pg); 923 } else { 924 struct proc *p = (sigio)->sio_proc; 925 PROC_LOCK(p); 926 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 927 sigio, sio_pgsigio); 928 PROC_UNLOCK(p); 929 } 930 SIGIO_UNLOCK(); 931 crfree(sigio->sio_ucred); 932 free(sigio, M_SIGIO); 933} 934 935/* 936 * Free a list of sigio structures. 937 * We only need to lock the SIGIO_LOCK because we have made ourselves 938 * inaccessible to callers of fsetown and therefore do not need to lock 939 * the proc or pgrp struct for the list manipulation. 940 */ 941void 942funsetownlst(struct sigiolst *sigiolst) 943{ 944 struct proc *p; 945 struct pgrp *pg; 946 struct sigio *sigio; 947 948 sigio = SLIST_FIRST(sigiolst); 949 if (sigio == NULL) 950 return; 951 p = NULL; 952 pg = NULL; 953 954 /* 955 * Every entry of the list should belong 956 * to a single proc or pgrp. 957 */ 958 if (sigio->sio_pgid < 0) { 959 pg = sigio->sio_pgrp; 960 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 961 } else /* if (sigio->sio_pgid > 0) */ { 962 p = sigio->sio_proc; 963 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 964 } 965 966 SIGIO_LOCK(); 967 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 968 *(sigio->sio_myref) = NULL; 969 if (pg != NULL) { 970 KASSERT(sigio->sio_pgid < 0, 971 ("Proc sigio in pgrp sigio list")); 972 KASSERT(sigio->sio_pgrp == pg, 973 ("Bogus pgrp in sigio list")); 974 PGRP_LOCK(pg); 975 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 976 sio_pgsigio); 977 PGRP_UNLOCK(pg); 978 } else /* if (p != NULL) */ { 979 KASSERT(sigio->sio_pgid > 0, 980 ("Pgrp sigio in proc sigio list")); 981 KASSERT(sigio->sio_proc == p, 982 ("Bogus proc in sigio list")); 983 PROC_LOCK(p); 984 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 985 sio_pgsigio); 986 PROC_UNLOCK(p); 987 } 988 SIGIO_UNLOCK(); 989 crfree(sigio->sio_ucred); 990 free(sigio, M_SIGIO); 991 SIGIO_LOCK(); 992 } 993 SIGIO_UNLOCK(); 994} 995 996/* 997 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 998 * 999 * After permission checking, add a sigio structure to the sigio list for 1000 * the process or process group. 1001 */ 1002int 1003fsetown(pid_t pgid, struct sigio **sigiop) 1004{ 1005 struct proc *proc; 1006 struct pgrp *pgrp; 1007 struct sigio *sigio; 1008 int ret; 1009 1010 if (pgid == 0) { 1011 funsetown(sigiop); 1012 return (0); 1013 } 1014 1015 ret = 0; 1016 1017 /* Allocate and fill in the new sigio out of locks. */ 1018 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1019 sigio->sio_pgid = pgid; 1020 sigio->sio_ucred = crhold(curthread->td_ucred); 1021 sigio->sio_myref = sigiop; 1022 1023 sx_slock(&proctree_lock); 1024 if (pgid > 0) { 1025 proc = pfind(pgid); 1026 if (proc == NULL) { 1027 ret = ESRCH; 1028 goto fail; 1029 } 1030 1031 /* 1032 * Policy - Don't allow a process to FSETOWN a process 1033 * in another session. 1034 * 1035 * Remove this test to allow maximum flexibility or 1036 * restrict FSETOWN to the current process or process 1037 * group for maximum safety. 1038 */ 1039 PROC_UNLOCK(proc); 1040 if (proc->p_session != curthread->td_proc->p_session) { 1041 ret = EPERM; 1042 goto fail; 1043 } 1044 1045 pgrp = NULL; 1046 } else /* if (pgid < 0) */ { 1047 pgrp = pgfind(-pgid); 1048 if (pgrp == NULL) { 1049 ret = ESRCH; 1050 goto fail; 1051 } 1052 PGRP_UNLOCK(pgrp); 1053 1054 /* 1055 * Policy - Don't allow a process to FSETOWN a process 1056 * in another session. 1057 * 1058 * Remove this test to allow maximum flexibility or 1059 * restrict FSETOWN to the current process or process 1060 * group for maximum safety. 1061 */ 1062 if (pgrp->pg_session != curthread->td_proc->p_session) { 1063 ret = EPERM; 1064 goto fail; 1065 } 1066 1067 proc = NULL; 1068 } 1069 funsetown(sigiop); 1070 if (pgid > 0) { 1071 PROC_LOCK(proc); 1072 /* 1073 * Since funsetownlst() is called without the proctree 1074 * locked, we need to check for P_WEXIT. 1075 * XXX: is ESRCH correct? 1076 */ 1077 if ((proc->p_flag & P_WEXIT) != 0) { 1078 PROC_UNLOCK(proc); 1079 ret = ESRCH; 1080 goto fail; 1081 } 1082 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1083 sigio->sio_proc = proc; 1084 PROC_UNLOCK(proc); 1085 } else { 1086 PGRP_LOCK(pgrp); 1087 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1088 sigio->sio_pgrp = pgrp; 1089 PGRP_UNLOCK(pgrp); 1090 } 1091 sx_sunlock(&proctree_lock); 1092 SIGIO_LOCK(); 1093 *sigiop = sigio; 1094 SIGIO_UNLOCK(); 1095 return (0); 1096 1097fail: 1098 sx_sunlock(&proctree_lock); 1099 crfree(sigio->sio_ucred); 1100 free(sigio, M_SIGIO); 1101 return (ret); 1102} 1103 1104/* 1105 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1106 */ 1107pid_t 1108fgetown(sigiop) 1109 struct sigio **sigiop; 1110{ 1111 pid_t pgid; 1112 1113 SIGIO_LOCK(); 1114 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1115 SIGIO_UNLOCK(); 1116 return (pgid); 1117} 1118 1119/* 1120 * Function drops the filedesc lock on return. 1121 */ 1122static int 1123closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1124 int holdleaders) 1125{ 1126 int error; 1127 1128 FILEDESC_XLOCK_ASSERT(fdp); 1129 1130 if (holdleaders) { 1131 if (td->td_proc->p_fdtol != NULL) { 1132 /* 1133 * Ask fdfree() to sleep to ensure that all relevant 1134 * process leaders can be traversed in closef(). 1135 */ 1136 fdp->fd_holdleaderscount++; 1137 } else { 1138 holdleaders = 0; 1139 } 1140 } 1141 1142 /* 1143 * We now hold the fp reference that used to be owned by the 1144 * descriptor array. We have to unlock the FILEDESC *AFTER* 1145 * knote_fdclose to prevent a race of the fd getting opened, a knote 1146 * added, and deleteing a knote for the new fd. 1147 */ 1148 knote_fdclose(td, fd); 1149 1150 /* 1151 * We need to notify mqueue if the object is of type mqueue. 1152 */ 1153 if (fp->f_type == DTYPE_MQUEUE) 1154 mq_fdclose(td, fd, fp); 1155 FILEDESC_XUNLOCK(fdp); 1156 1157 error = closef(fp, td); 1158 if (holdleaders) { 1159 FILEDESC_XLOCK(fdp); 1160 fdp->fd_holdleaderscount--; 1161 if (fdp->fd_holdleaderscount == 0 && 1162 fdp->fd_holdleaderswakeup != 0) { 1163 fdp->fd_holdleaderswakeup = 0; 1164 wakeup(&fdp->fd_holdleaderscount); 1165 } 1166 FILEDESC_XUNLOCK(fdp); 1167 } 1168 return (error); 1169} 1170 1171/* 1172 * Close a file descriptor. 1173 */ 1174#ifndef _SYS_SYSPROTO_H_ 1175struct close_args { 1176 int fd; 1177}; 1178#endif 1179/* ARGSUSED */ 1180int 1181sys_close(td, uap) 1182 struct thread *td; 1183 struct close_args *uap; 1184{ 1185 1186 return (kern_close(td, uap->fd)); 1187} 1188 1189int 1190kern_close(td, fd) 1191 struct thread *td; 1192 int fd; 1193{ 1194 struct filedesc *fdp; 1195 struct file *fp; 1196 1197 fdp = td->td_proc->p_fd; 1198 1199 AUDIT_SYSCLOSE(td, fd); 1200 1201 FILEDESC_XLOCK(fdp); 1202 if ((fp = fget_locked(fdp, fd)) == NULL) { 1203 FILEDESC_XUNLOCK(fdp); 1204 return (EBADF); 1205 } 1206 fdfree(fdp, fd); 1207 1208 /* closefp() drops the FILEDESC lock for us. */ 1209 return (closefp(fdp, fd, fp, td, 1)); 1210} 1211 1212/* 1213 * Close open file descriptors. 1214 */ 1215#ifndef _SYS_SYSPROTO_H_ 1216struct closefrom_args { 1217 int lowfd; 1218}; 1219#endif 1220/* ARGSUSED */ 1221int 1222sys_closefrom(struct thread *td, struct closefrom_args *uap) 1223{ 1224 struct filedesc *fdp; 1225 int fd; 1226 1227 fdp = td->td_proc->p_fd; 1228 AUDIT_ARG_FD(uap->lowfd); 1229 1230 /* 1231 * Treat negative starting file descriptor values identical to 1232 * closefrom(0) which closes all files. 1233 */ 1234 if (uap->lowfd < 0) 1235 uap->lowfd = 0; 1236 FILEDESC_SLOCK(fdp); 1237 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1238 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1239 FILEDESC_SUNLOCK(fdp); 1240 (void)kern_close(td, fd); 1241 FILEDESC_SLOCK(fdp); 1242 } 1243 } 1244 FILEDESC_SUNLOCK(fdp); 1245 return (0); 1246} 1247 1248#if defined(COMPAT_43) 1249/* 1250 * Return status information about a file descriptor. 1251 */ 1252#ifndef _SYS_SYSPROTO_H_ 1253struct ofstat_args { 1254 int fd; 1255 struct ostat *sb; 1256}; 1257#endif 1258/* ARGSUSED */ 1259int 1260ofstat(struct thread *td, struct ofstat_args *uap) 1261{ 1262 struct ostat oub; 1263 struct stat ub; 1264 int error; 1265 1266 error = kern_fstat(td, uap->fd, &ub); 1267 if (error == 0) { 1268 cvtstat(&ub, &oub); 1269 error = copyout(&oub, uap->sb, sizeof(oub)); 1270 } 1271 return (error); 1272} 1273#endif /* COMPAT_43 */ 1274 1275/* 1276 * Return status information about a file descriptor. 1277 */ 1278#ifndef _SYS_SYSPROTO_H_ 1279struct fstat_args { 1280 int fd; 1281 struct stat *sb; 1282}; 1283#endif 1284/* ARGSUSED */ 1285int 1286sys_fstat(struct thread *td, struct fstat_args *uap) 1287{ 1288 struct stat ub; 1289 int error; 1290 1291 error = kern_fstat(td, uap->fd, &ub); 1292 if (error == 0) 1293 error = copyout(&ub, uap->sb, sizeof(ub)); 1294 return (error); 1295} 1296 1297int 1298kern_fstat(struct thread *td, int fd, struct stat *sbp) 1299{ 1300 struct file *fp; 1301 cap_rights_t rights; 1302 int error; 1303 1304 AUDIT_ARG_FD(fd); 1305 1306 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1307 if (error != 0) 1308 return (error); 1309 1310 AUDIT_ARG_FILE(td->td_proc, fp); 1311 1312 error = fo_stat(fp, sbp, td->td_ucred, td); 1313 fdrop(fp, td); 1314#ifdef KTRACE 1315 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1316 ktrstat(sbp); 1317#endif 1318 return (error); 1319} 1320 1321/* 1322 * Return status information about a file descriptor. 1323 */ 1324#ifndef _SYS_SYSPROTO_H_ 1325struct nfstat_args { 1326 int fd; 1327 struct nstat *sb; 1328}; 1329#endif 1330/* ARGSUSED */ 1331int 1332sys_nfstat(struct thread *td, struct nfstat_args *uap) 1333{ 1334 struct nstat nub; 1335 struct stat ub; 1336 int error; 1337 1338 error = kern_fstat(td, uap->fd, &ub); 1339 if (error == 0) { 1340 cvtnstat(&ub, &nub); 1341 error = copyout(&nub, uap->sb, sizeof(nub)); 1342 } 1343 return (error); 1344} 1345 1346/* 1347 * Return pathconf information about a file descriptor. 1348 */ 1349#ifndef _SYS_SYSPROTO_H_ 1350struct fpathconf_args { 1351 int fd; 1352 int name; 1353}; 1354#endif 1355/* ARGSUSED */ 1356int 1357sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1358{ 1359 struct file *fp; 1360 struct vnode *vp; 1361 cap_rights_t rights; 1362 int error; 1363 1364 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1365 if (error != 0) 1366 return (error); 1367 1368 /* If asynchronous I/O is available, it works for all descriptors. */ 1369 if (uap->name == _PC_ASYNC_IO) { 1370 td->td_retval[0] = async_io_version; 1371 goto out; 1372 } 1373 vp = fp->f_vnode; 1374 if (vp != NULL) { 1375 vn_lock(vp, LK_SHARED | LK_RETRY); 1376 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1377 VOP_UNLOCK(vp, 0); 1378 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1379 if (uap->name != _PC_PIPE_BUF) { 1380 error = EINVAL; 1381 } else { 1382 td->td_retval[0] = PIPE_BUF; 1383 error = 0; 1384 } 1385 } else { 1386 error = EOPNOTSUPP; 1387 } 1388out: 1389 fdrop(fp, td); 1390 return (error); 1391} 1392 1393/* 1394 * Initialize filecaps structure. 1395 */ 1396void 1397filecaps_init(struct filecaps *fcaps) 1398{ 1399 1400 bzero(fcaps, sizeof(*fcaps)); 1401 fcaps->fc_nioctls = -1; 1402} 1403 1404/* 1405 * Copy filecaps structure allocating memory for ioctls array if needed. 1406 */ 1407void 1408filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1409{ 1410 size_t size; 1411 1412 *dst = *src; 1413 if (src->fc_ioctls != NULL) { 1414 KASSERT(src->fc_nioctls > 0, 1415 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1416 1417 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1418 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1419 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1420 } 1421} 1422 1423/* 1424 * Move filecaps structure to the new place and clear the old place. 1425 */ 1426void 1427filecaps_move(struct filecaps *src, struct filecaps *dst) 1428{ 1429 1430 *dst = *src; 1431 bzero(src, sizeof(*src)); 1432} 1433 1434/* 1435 * Fill the given filecaps structure with full rights. 1436 */ 1437static void 1438filecaps_fill(struct filecaps *fcaps) 1439{ 1440 1441 CAP_ALL(&fcaps->fc_rights); 1442 fcaps->fc_ioctls = NULL; 1443 fcaps->fc_nioctls = -1; 1444 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1445} 1446 1447/* 1448 * Free memory allocated within filecaps structure. 1449 */ 1450void 1451filecaps_free(struct filecaps *fcaps) 1452{ 1453 1454 free(fcaps->fc_ioctls, M_FILECAPS); 1455 bzero(fcaps, sizeof(*fcaps)); 1456} 1457 1458/* 1459 * Validate the given filecaps structure. 1460 */ 1461static void 1462filecaps_validate(const struct filecaps *fcaps, const char *func) 1463{ 1464 1465 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1466 ("%s: invalid rights", func)); 1467 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1468 ("%s: invalid fcntls", func)); 1469 KASSERT(fcaps->fc_fcntls == 0 || 1470 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1471 ("%s: fcntls without CAP_FCNTL", func)); 1472 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1473 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1474 ("%s: invalid ioctls", func)); 1475 KASSERT(fcaps->fc_nioctls == 0 || 1476 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1477 ("%s: ioctls without CAP_IOCTL", func)); 1478} 1479 1480static void 1481fdgrowtable_exp(struct filedesc *fdp, int nfd) 1482{ 1483 int nfd1; 1484 1485 FILEDESC_XLOCK_ASSERT(fdp); 1486 1487 nfd1 = fdp->fd_nfiles * 2; 1488 if (nfd1 < nfd) 1489 nfd1 = nfd; 1490 fdgrowtable(fdp, nfd1); 1491} 1492 1493/* 1494 * Grow the file table to accomodate (at least) nfd descriptors. 1495 */ 1496static void 1497fdgrowtable(struct filedesc *fdp, int nfd) 1498{ 1499 struct filedesc0 *fdp0; 1500 struct freetable *ft; 1501 struct filedescent *ntable; 1502 struct filedescent *otable; 1503 int nnfiles, onfiles; 1504 NDSLOTTYPE *nmap, *omap; 1505 1506 FILEDESC_XLOCK_ASSERT(fdp); 1507 1508 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1509 1510 /* save old values */ 1511 onfiles = fdp->fd_nfiles; 1512 otable = fdp->fd_ofiles; 1513 omap = fdp->fd_map; 1514 1515 /* compute the size of the new table */ 1516 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1517 if (nnfiles <= onfiles) 1518 /* the table is already large enough */ 1519 return; 1520 1521 /* 1522 * Allocate a new table. We need enough space for the 1523 * file entries themselves and the struct freetable we will use 1524 * when we decommission the table and place it on the freelist. 1525 * We place the struct freetable in the middle so we don't have 1526 * to worry about padding. 1527 */ 1528 ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable), 1529 M_FILEDESC, M_ZERO | M_WAITOK); 1530 /* copy the old data over and point at the new tables */ 1531 memcpy(ntable, otable, onfiles * sizeof(*otable)); 1532 fdp->fd_ofiles = ntable; 1533 1534 /* 1535 * Allocate a new map only if the old is not large enough. It will 1536 * grow at a slower rate than the table as it can map more 1537 * entries than the table can hold. 1538 */ 1539 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1540 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1541 M_ZERO | M_WAITOK); 1542 /* copy over the old data and update the pointer */ 1543 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1544 fdp->fd_map = nmap; 1545 } 1546 1547 /* 1548 * In order to have a valid pattern for fget_unlocked() 1549 * fdp->fd_nfiles must be the last member to be updated, otherwise 1550 * fget_unlocked() consumers may reference a new, higher value for 1551 * fdp->fd_nfiles before to access the fdp->fd_ofiles array, 1552 * resulting in OOB accesses. 1553 */ 1554 atomic_store_rel_int(&fdp->fd_nfiles, nnfiles); 1555 1556 /* 1557 * Do not free the old file table, as some threads may still 1558 * reference entries within it. Instead, place it on a freelist 1559 * which will be processed when the struct filedesc is released. 1560 * 1561 * Note that if onfiles == NDFILE, we're dealing with the original 1562 * static allocation contained within (struct filedesc0 *)fdp, 1563 * which must not be freed. 1564 */ 1565 if (onfiles > NDFILE) { 1566 ft = (struct freetable *)&otable[onfiles]; 1567 fdp0 = (struct filedesc0 *)fdp; 1568 ft->ft_table = otable; 1569 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1570 } 1571 /* 1572 * The map does not have the same possibility of threads still 1573 * holding references to it. So always free it as long as it 1574 * does not reference the original static allocation. 1575 */ 1576 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1577 free(omap, M_FILEDESC); 1578} 1579 1580/* 1581 * Allocate a file descriptor for the process. 1582 */ 1583int 1584fdalloc(struct thread *td, int minfd, int *result) 1585{ 1586 struct proc *p = td->td_proc; 1587 struct filedesc *fdp = p->p_fd; 1588 int fd = -1, maxfd, allocfd; 1589#ifdef RACCT 1590 int error; 1591#endif 1592 1593 FILEDESC_XLOCK_ASSERT(fdp); 1594 1595 if (fdp->fd_freefile > minfd) 1596 minfd = fdp->fd_freefile; 1597 1598 maxfd = getmaxfd(p); 1599 1600 /* 1601 * Search the bitmap for a free descriptor starting at minfd. 1602 * If none is found, grow the file table. 1603 */ 1604 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1605 if (fd >= maxfd) 1606 return (EMFILE); 1607 if (fd >= fdp->fd_nfiles) { 1608 allocfd = min(fd * 2, maxfd); 1609#ifdef RACCT 1610 PROC_LOCK(p); 1611 error = racct_set(p, RACCT_NOFILE, allocfd); 1612 PROC_UNLOCK(p); 1613 if (error != 0) 1614 return (EMFILE); 1615#endif 1616 /* 1617 * fd is already equal to first free descriptor >= minfd, so 1618 * we only need to grow the table and we are done. 1619 */ 1620 fdgrowtable_exp(fdp, allocfd); 1621 } 1622 1623 /* 1624 * Perform some sanity checks, then mark the file descriptor as 1625 * used and return it to the caller. 1626 */ 1627 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1628 ("invalid descriptor %d", fd)); 1629 KASSERT(!fdisused(fdp, fd), 1630 ("fd_first_free() returned non-free descriptor")); 1631 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1632 ("file descriptor isn't free")); 1633 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); 1634 fdused(fdp, fd); 1635 *result = fd; 1636 return (0); 1637} 1638 1639/* 1640 * Allocate n file descriptors for the process. 1641 */ 1642int 1643fdallocn(struct thread *td, int minfd, int *fds, int n) 1644{ 1645 struct proc *p = td->td_proc; 1646 struct filedesc *fdp = p->p_fd; 1647 int i; 1648 1649 FILEDESC_XLOCK_ASSERT(fdp); 1650 1651 for (i = 0; i < n; i++) 1652 if (fdalloc(td, 0, &fds[i]) != 0) 1653 break; 1654 1655 if (i < n) { 1656 for (i--; i >= 0; i--) 1657 fdunused(fdp, fds[i]); 1658 return (EMFILE); 1659 } 1660 1661 return (0); 1662} 1663 1664/* 1665 * Create a new open file structure and allocate a file decriptor for the 1666 * process that refers to it. We add one reference to the file for the 1667 * descriptor table and one reference for resultfp. This is to prevent us 1668 * being preempted and the entry in the descriptor table closed after we 1669 * release the FILEDESC lock. 1670 */ 1671int 1672falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1673{ 1674 struct file *fp; 1675 int error, fd; 1676 1677 error = falloc_noinstall(td, &fp); 1678 if (error) 1679 return (error); /* no reference held on error */ 1680 1681 error = finstall(td, fp, &fd, flags, NULL); 1682 if (error) { 1683 fdrop(fp, td); /* one reference (fp only) */ 1684 return (error); 1685 } 1686 1687 if (resultfp != NULL) 1688 *resultfp = fp; /* copy out result */ 1689 else 1690 fdrop(fp, td); /* release local reference */ 1691 1692 if (resultfd != NULL) 1693 *resultfd = fd; 1694 1695 return (0); 1696} 1697 1698/* 1699 * Create a new open file structure without allocating a file descriptor. 1700 */ 1701int 1702falloc_noinstall(struct thread *td, struct file **resultfp) 1703{ 1704 struct file *fp; 1705 int maxuserfiles = maxfiles - (maxfiles / 20); 1706 static struct timeval lastfail; 1707 static int curfail; 1708 1709 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1710 1711 if ((openfiles >= maxuserfiles && 1712 priv_check(td, PRIV_MAXFILES) != 0) || 1713 openfiles >= maxfiles) { 1714 if (ppsratecheck(&lastfail, &curfail, 1)) { 1715 printf("kern.maxfiles limit exceeded by uid %i, " 1716 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1717 } 1718 return (ENFILE); 1719 } 1720 atomic_add_int(&openfiles, 1); 1721 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1722 refcount_init(&fp->f_count, 1); 1723 fp->f_cred = crhold(td->td_ucred); 1724 fp->f_ops = &badfileops; 1725 fp->f_data = NULL; 1726 fp->f_vnode = NULL; 1727 *resultfp = fp; 1728 return (0); 1729} 1730 1731/* 1732 * Install a file in a file descriptor table. 1733 */ 1734int 1735finstall(struct thread *td, struct file *fp, int *fd, int flags, 1736 struct filecaps *fcaps) 1737{ 1738 struct filedesc *fdp = td->td_proc->p_fd; 1739 struct filedescent *fde; 1740 int error; 1741 1742 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1743 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1744 if (fcaps != NULL) 1745 filecaps_validate(fcaps, __func__); 1746 1747 FILEDESC_XLOCK(fdp); 1748 if ((error = fdalloc(td, 0, fd))) { 1749 FILEDESC_XUNLOCK(fdp); 1750 return (error); 1751 } 1752 fhold(fp); 1753 fde = &fdp->fd_ofiles[*fd]; 1754 fde->fde_file = fp; 1755 if ((flags & O_CLOEXEC) != 0) 1756 fde->fde_flags |= UF_EXCLOSE; 1757 if (fcaps != NULL) 1758 filecaps_move(fcaps, &fde->fde_caps); 1759 else 1760 filecaps_fill(&fde->fde_caps); 1761 FILEDESC_XUNLOCK(fdp); 1762 return (0); 1763} 1764 1765/* 1766 * Build a new filedesc structure from another. 1767 * Copy the current, root, and jail root vnode references. 1768 */ 1769struct filedesc * 1770fdinit(struct filedesc *fdp) 1771{ 1772 struct filedesc0 *newfdp; 1773 1774 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1775 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1776 if (fdp != NULL) { 1777 FILEDESC_SLOCK(fdp); 1778 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1779 if (newfdp->fd_fd.fd_cdir) 1780 VREF(newfdp->fd_fd.fd_cdir); 1781 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1782 if (newfdp->fd_fd.fd_rdir) 1783 VREF(newfdp->fd_fd.fd_rdir); 1784 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1785 if (newfdp->fd_fd.fd_jdir) 1786 VREF(newfdp->fd_fd.fd_jdir); 1787 FILEDESC_SUNLOCK(fdp); 1788 } 1789 1790 /* Create the file descriptor table. */ 1791 newfdp->fd_fd.fd_refcnt = 1; 1792 newfdp->fd_fd.fd_holdcnt = 1; 1793 newfdp->fd_fd.fd_cmask = CMASK; 1794 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1795 newfdp->fd_fd.fd_nfiles = NDFILE; 1796 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1797 newfdp->fd_fd.fd_lastfile = -1; 1798 return (&newfdp->fd_fd); 1799} 1800 1801static struct filedesc * 1802fdhold(struct proc *p) 1803{ 1804 struct filedesc *fdp; 1805 1806 mtx_lock(&fdesc_mtx); 1807 fdp = p->p_fd; 1808 if (fdp != NULL) 1809 fdp->fd_holdcnt++; 1810 mtx_unlock(&fdesc_mtx); 1811 return (fdp); 1812} 1813 1814static void 1815fddrop(struct filedesc *fdp) 1816{ 1817 struct filedesc0 *fdp0; 1818 struct freetable *ft; 1819 int i; 1820 1821 mtx_lock(&fdesc_mtx); 1822 i = --fdp->fd_holdcnt; 1823 mtx_unlock(&fdesc_mtx); 1824 if (i > 0) 1825 return; 1826 1827 FILEDESC_LOCK_DESTROY(fdp); 1828 fdp0 = (struct filedesc0 *)fdp; 1829 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1830 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1831 free(ft->ft_table, M_FILEDESC); 1832 } 1833 free(fdp, M_FILEDESC); 1834} 1835 1836/* 1837 * Share a filedesc structure. 1838 */ 1839struct filedesc * 1840fdshare(struct filedesc *fdp) 1841{ 1842 1843 FILEDESC_XLOCK(fdp); 1844 fdp->fd_refcnt++; 1845 FILEDESC_XUNLOCK(fdp); 1846 return (fdp); 1847} 1848 1849/* 1850 * Unshare a filedesc structure, if necessary by making a copy 1851 */ 1852void 1853fdunshare(struct thread *td) 1854{ 1855 struct filedesc *tmp; 1856 struct proc *p = td->td_proc; 1857 1858 if (p->p_fd->fd_refcnt == 1) 1859 return; 1860 1861 tmp = fdcopy(p->p_fd); 1862 fdescfree(td); 1863 p->p_fd = tmp; 1864} 1865 1866/* 1867 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1868 * this is to ease callers, not catch errors. 1869 */ 1870struct filedesc * 1871fdcopy(struct filedesc *fdp) 1872{ 1873 struct filedesc *newfdp; 1874 struct filedescent *nfde, *ofde; 1875 int i; 1876 1877 /* Certain daemons might not have file descriptors. */ 1878 if (fdp == NULL) 1879 return (NULL); 1880 1881 newfdp = fdinit(fdp); 1882 FILEDESC_SLOCK(fdp); 1883 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1884 FILEDESC_SUNLOCK(fdp); 1885 FILEDESC_XLOCK(newfdp); 1886 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1887 FILEDESC_XUNLOCK(newfdp); 1888 FILEDESC_SLOCK(fdp); 1889 } 1890 /* copy all passable descriptors (i.e. not kqueue) */ 1891 newfdp->fd_freefile = -1; 1892 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1893 ofde = &fdp->fd_ofiles[i]; 1894 if (fdisused(fdp, i) && 1895 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) && 1896 ofde->fde_file->f_ops != &badfileops) { 1897 nfde = &newfdp->fd_ofiles[i]; 1898 *nfde = *ofde; 1899 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1900 fhold(nfde->fde_file); 1901 newfdp->fd_lastfile = i; 1902 } else { 1903 if (newfdp->fd_freefile == -1) 1904 newfdp->fd_freefile = i; 1905 } 1906 } 1907 newfdp->fd_cmask = fdp->fd_cmask; 1908 FILEDESC_SUNLOCK(fdp); 1909 FILEDESC_XLOCK(newfdp); 1910 for (i = 0; i <= newfdp->fd_lastfile; ++i) { 1911 if (newfdp->fd_ofiles[i].fde_file != NULL) 1912 fdused(newfdp, i); 1913 } 1914 if (newfdp->fd_freefile == -1) 1915 newfdp->fd_freefile = i; 1916 FILEDESC_XUNLOCK(newfdp); 1917 return (newfdp); 1918} 1919 1920/* 1921 * Release a filedesc structure. 1922 */ 1923void 1924fdescfree(struct thread *td) 1925{ 1926 struct filedesc *fdp; 1927 int i; 1928 struct filedesc_to_leader *fdtol; 1929 struct file *fp; 1930 struct vnode *cdir, *jdir, *rdir, *vp; 1931 struct flock lf; 1932 1933 /* Certain daemons might not have file descriptors. */ 1934 fdp = td->td_proc->p_fd; 1935 if (fdp == NULL) 1936 return; 1937 1938#ifdef RACCT 1939 PROC_LOCK(td->td_proc); 1940 racct_set(td->td_proc, RACCT_NOFILE, 0); 1941 PROC_UNLOCK(td->td_proc); 1942#endif 1943 1944 /* Check for special need to clear POSIX style locks */ 1945 fdtol = td->td_proc->p_fdtol; 1946 if (fdtol != NULL) { 1947 FILEDESC_XLOCK(fdp); 1948 KASSERT(fdtol->fdl_refcount > 0, 1949 ("filedesc_to_refcount botch: fdl_refcount=%d", 1950 fdtol->fdl_refcount)); 1951 if (fdtol->fdl_refcount == 1 && 1952 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1953 for (i = 0; i <= fdp->fd_lastfile; i++) { 1954 fp = fdp->fd_ofiles[i].fde_file; 1955 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1956 continue; 1957 fhold(fp); 1958 FILEDESC_XUNLOCK(fdp); 1959 lf.l_whence = SEEK_SET; 1960 lf.l_start = 0; 1961 lf.l_len = 0; 1962 lf.l_type = F_UNLCK; 1963 vp = fp->f_vnode; 1964 (void) VOP_ADVLOCK(vp, 1965 (caddr_t)td->td_proc->p_leader, F_UNLCK, 1966 &lf, F_POSIX); 1967 FILEDESC_XLOCK(fdp); 1968 fdrop(fp, td); 1969 } 1970 } 1971 retry: 1972 if (fdtol->fdl_refcount == 1) { 1973 if (fdp->fd_holdleaderscount > 0 && 1974 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1975 /* 1976 * close() or do_dup() has cleared a reference 1977 * in a shared file descriptor table. 1978 */ 1979 fdp->fd_holdleaderswakeup = 1; 1980 sx_sleep(&fdp->fd_holdleaderscount, 1981 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1982 goto retry; 1983 } 1984 if (fdtol->fdl_holdcount > 0) { 1985 /* 1986 * Ensure that fdtol->fdl_leader remains 1987 * valid in closef(). 1988 */ 1989 fdtol->fdl_wakeup = 1; 1990 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 1991 "fdlhold", 0); 1992 goto retry; 1993 } 1994 } 1995 fdtol->fdl_refcount--; 1996 if (fdtol->fdl_refcount == 0 && 1997 fdtol->fdl_holdcount == 0) { 1998 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 1999 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2000 } else 2001 fdtol = NULL; 2002 td->td_proc->p_fdtol = NULL; 2003 FILEDESC_XUNLOCK(fdp); 2004 if (fdtol != NULL) 2005 free(fdtol, M_FILEDESC_TO_LEADER); 2006 } 2007 2008 mtx_lock(&fdesc_mtx); 2009 td->td_proc->p_fd = NULL; 2010 mtx_unlock(&fdesc_mtx); 2011 2012 FILEDESC_XLOCK(fdp); 2013 i = --fdp->fd_refcnt; 2014 FILEDESC_XUNLOCK(fdp); 2015 if (i > 0) 2016 return; 2017 2018 for (i = 0; i <= fdp->fd_lastfile; i++) { 2019 fp = fdp->fd_ofiles[i].fde_file; 2020 if (fp != NULL) { 2021 FILEDESC_XLOCK(fdp); 2022 fdfree(fdp, i); 2023 FILEDESC_XUNLOCK(fdp); 2024 (void) closef(fp, td); 2025 } 2026 } 2027 FILEDESC_XLOCK(fdp); 2028 2029 if (fdp->fd_nfiles > NDFILE) 2030 free(fdp->fd_ofiles, M_FILEDESC); 2031 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2032 free(fdp->fd_map, M_FILEDESC); 2033 2034 fdp->fd_nfiles = 0; 2035 2036 cdir = fdp->fd_cdir; 2037 fdp->fd_cdir = NULL; 2038 rdir = fdp->fd_rdir; 2039 fdp->fd_rdir = NULL; 2040 jdir = fdp->fd_jdir; 2041 fdp->fd_jdir = NULL; 2042 FILEDESC_XUNLOCK(fdp); 2043 2044 if (cdir != NULL) 2045 vrele(cdir); 2046 if (rdir != NULL) 2047 vrele(rdir); 2048 if (jdir != NULL) 2049 vrele(jdir); 2050 2051 fddrop(fdp); 2052} 2053 2054/* 2055 * For setugid programs, we don't want to people to use that setugidness 2056 * to generate error messages which write to a file which otherwise would 2057 * otherwise be off-limits to the process. We check for filesystems where 2058 * the vnode can change out from under us after execve (like [lin]procfs). 2059 * 2060 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2061 * sufficient. We also don't check for setugidness since we know we are. 2062 */ 2063static int 2064is_unsafe(struct file *fp) 2065{ 2066 if (fp->f_type == DTYPE_VNODE) { 2067 struct vnode *vp = fp->f_vnode; 2068 2069 if ((vp->v_vflag & VV_PROCDEP) != 0) 2070 return (1); 2071 } 2072 return (0); 2073} 2074 2075/* 2076 * Make this setguid thing safe, if at all possible. 2077 */ 2078void 2079setugidsafety(struct thread *td) 2080{ 2081 struct filedesc *fdp; 2082 struct file *fp; 2083 int i; 2084 2085 fdp = td->td_proc->p_fd; 2086 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2087 FILEDESC_XLOCK(fdp); 2088 for (i = 0; i <= fdp->fd_lastfile; i++) { 2089 if (i > 2) 2090 break; 2091 fp = fdp->fd_ofiles[i].fde_file; 2092 if (fp != NULL && is_unsafe(fp)) { 2093 knote_fdclose(td, i); 2094 /* 2095 * NULL-out descriptor prior to close to avoid 2096 * a race while close blocks. 2097 */ 2098 fdfree(fdp, i); 2099 FILEDESC_XUNLOCK(fdp); 2100 (void) closef(fp, td); 2101 FILEDESC_XLOCK(fdp); 2102 } 2103 } 2104 FILEDESC_XUNLOCK(fdp); 2105} 2106 2107/* 2108 * If a specific file object occupies a specific file descriptor, close the 2109 * file descriptor entry and drop a reference on the file object. This is a 2110 * convenience function to handle a subsequent error in a function that calls 2111 * falloc() that handles the race that another thread might have closed the 2112 * file descriptor out from under the thread creating the file object. 2113 */ 2114void 2115fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2116{ 2117 2118 FILEDESC_XLOCK(fdp); 2119 if (fdp->fd_ofiles[idx].fde_file == fp) { 2120 fdfree(fdp, idx); 2121 FILEDESC_XUNLOCK(fdp); 2122 fdrop(fp, td); 2123 } else 2124 FILEDESC_XUNLOCK(fdp); 2125} 2126 2127/* 2128 * Close any files on exec? 2129 */ 2130void 2131fdcloseexec(struct thread *td) 2132{ 2133 struct filedesc *fdp; 2134 struct filedescent *fde; 2135 struct file *fp; 2136 int i; 2137 2138 fdp = td->td_proc->p_fd; 2139 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2140 FILEDESC_XLOCK(fdp); 2141 for (i = 0; i <= fdp->fd_lastfile; i++) { 2142 fde = &fdp->fd_ofiles[i]; 2143 fp = fde->fde_file; 2144 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2145 (fde->fde_flags & UF_EXCLOSE))) { 2146 fdfree(fdp, i); 2147 (void) closefp(fdp, i, fp, td, 0); 2148 /* closefp() drops the FILEDESC lock. */ 2149 FILEDESC_XLOCK(fdp); 2150 } 2151 } 2152 FILEDESC_XUNLOCK(fdp); 2153} 2154 2155/* 2156 * It is unsafe for set[ug]id processes to be started with file 2157 * descriptors 0..2 closed, as these descriptors are given implicit 2158 * significance in the Standard C library. fdcheckstd() will create a 2159 * descriptor referencing /dev/null for each of stdin, stdout, and 2160 * stderr that is not already open. 2161 */ 2162int 2163fdcheckstd(struct thread *td) 2164{ 2165 struct filedesc *fdp; 2166 register_t retval, save; 2167 int i, error, devnull; 2168 2169 fdp = td->td_proc->p_fd; 2170 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2171 devnull = -1; 2172 error = 0; 2173 for (i = 0; i < 3; i++) { 2174 if (fdp->fd_ofiles[i].fde_file != NULL) 2175 continue; 2176 if (devnull < 0) { 2177 save = td->td_retval[0]; 2178 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2179 O_RDWR, 0); 2180 devnull = td->td_retval[0]; 2181 td->td_retval[0] = save; 2182 if (error) 2183 break; 2184 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2185 } else { 2186 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2187 if (error != 0) 2188 break; 2189 } 2190 } 2191 return (error); 2192} 2193 2194/* 2195 * Internal form of close. Decrement reference count on file structure. 2196 * Note: td may be NULL when closing a file that was being passed in a 2197 * message. 2198 * 2199 * XXXRW: Giant is not required for the caller, but often will be held; this 2200 * makes it moderately likely the Giant will be recursed in the VFS case. 2201 */ 2202int 2203closef(struct file *fp, struct thread *td) 2204{ 2205 struct vnode *vp; 2206 struct flock lf; 2207 struct filedesc_to_leader *fdtol; 2208 struct filedesc *fdp; 2209 2210 /* 2211 * POSIX record locking dictates that any close releases ALL 2212 * locks owned by this process. This is handled by setting 2213 * a flag in the unlock to free ONLY locks obeying POSIX 2214 * semantics, and not to free BSD-style file locks. 2215 * If the descriptor was in a message, POSIX-style locks 2216 * aren't passed with the descriptor, and the thread pointer 2217 * will be NULL. Callers should be careful only to pass a 2218 * NULL thread pointer when there really is no owning 2219 * context that might have locks, or the locks will be 2220 * leaked. 2221 */ 2222 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2223 vp = fp->f_vnode; 2224 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2225 lf.l_whence = SEEK_SET; 2226 lf.l_start = 0; 2227 lf.l_len = 0; 2228 lf.l_type = F_UNLCK; 2229 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2230 F_UNLCK, &lf, F_POSIX); 2231 } 2232 fdtol = td->td_proc->p_fdtol; 2233 if (fdtol != NULL) { 2234 /* 2235 * Handle special case where file descriptor table is 2236 * shared between multiple process leaders. 2237 */ 2238 fdp = td->td_proc->p_fd; 2239 FILEDESC_XLOCK(fdp); 2240 for (fdtol = fdtol->fdl_next; 2241 fdtol != td->td_proc->p_fdtol; 2242 fdtol = fdtol->fdl_next) { 2243 if ((fdtol->fdl_leader->p_flag & 2244 P_ADVLOCK) == 0) 2245 continue; 2246 fdtol->fdl_holdcount++; 2247 FILEDESC_XUNLOCK(fdp); 2248 lf.l_whence = SEEK_SET; 2249 lf.l_start = 0; 2250 lf.l_len = 0; 2251 lf.l_type = F_UNLCK; 2252 vp = fp->f_vnode; 2253 (void) VOP_ADVLOCK(vp, 2254 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2255 F_POSIX); 2256 FILEDESC_XLOCK(fdp); 2257 fdtol->fdl_holdcount--; 2258 if (fdtol->fdl_holdcount == 0 && 2259 fdtol->fdl_wakeup != 0) { 2260 fdtol->fdl_wakeup = 0; 2261 wakeup(fdtol); 2262 } 2263 } 2264 FILEDESC_XUNLOCK(fdp); 2265 } 2266 } 2267 return (fdrop(fp, td)); 2268} 2269 2270/* 2271 * Initialize the file pointer with the specified properties. 2272 * 2273 * The ops are set with release semantics to be certain that the flags, type, 2274 * and data are visible when ops is. This is to prevent ops methods from being 2275 * called with bad data. 2276 */ 2277void 2278finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2279{ 2280 fp->f_data = data; 2281 fp->f_flag = flag; 2282 fp->f_type = type; 2283 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2284} 2285 2286int 2287fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2288 int needfcntl, struct file **fpp, cap_rights_t *haverightsp) 2289{ 2290 struct file *fp; 2291 u_int count; 2292#ifdef CAPABILITIES 2293 cap_rights_t haverights; 2294 int error; 2295#endif 2296 2297 /* 2298 * Avoid reads reordering and then a first access to the 2299 * fdp->fd_ofiles table which could result in OOB operation. 2300 */ 2301 if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles)) 2302 return (EBADF); 2303 /* 2304 * Fetch the descriptor locklessly. We avoid fdrop() races by 2305 * never raising a refcount above 0. To accomplish this we have 2306 * to use a cmpset loop rather than an atomic_add. The descriptor 2307 * must be re-verified once we acquire a reference to be certain 2308 * that the identity is still correct and we did not lose a race 2309 * due to preemption. 2310 */ 2311 for (;;) { 2312 fp = fdp->fd_ofiles[fd].fde_file; 2313 if (fp == NULL) 2314 return (EBADF); 2315#ifdef CAPABILITIES 2316 haverights = *cap_rights(fdp, fd); 2317 if (needrightsp != NULL) { 2318 error = cap_check(&haverights, needrightsp); 2319 if (error != 0) 2320 return (error); 2321 if (cap_rights_is_set(needrightsp, CAP_FCNTL)) { 2322 error = cap_fcntl_check(fdp, fd, needfcntl); 2323 if (error != 0) 2324 return (error); 2325 } 2326 } 2327#endif 2328 count = fp->f_count; 2329 if (count == 0) 2330 continue; 2331 /* 2332 * Use an acquire barrier to prevent caching of fd_ofiles 2333 * so it is refreshed for verification. 2334 */ 2335 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2336 continue; 2337 if (fp == fdp->fd_ofiles[fd].fde_file) 2338 break; 2339 fdrop(fp, curthread); 2340 } 2341 *fpp = fp; 2342 if (haverightsp != NULL) { 2343#ifdef CAPABILITIES 2344 *haverightsp = haverights; 2345#else 2346 CAP_ALL(haverightsp); 2347#endif 2348 } 2349 return (0); 2350} 2351 2352/* 2353 * Extract the file pointer associated with the specified descriptor for the 2354 * current user process. 2355 * 2356 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2357 * returned. 2358 * 2359 * File's rights will be checked against the capability rights mask. 2360 * 2361 * If an error occured the non-zero error is returned and *fpp is set to 2362 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2363 * responsible for fdrop(). 2364 */ 2365static __inline int 2366_fget(struct thread *td, int fd, struct file **fpp, int flags, 2367 cap_rights_t *needrightsp, u_char *maxprotp) 2368{ 2369 struct filedesc *fdp; 2370 struct file *fp; 2371 cap_rights_t haverights, needrights; 2372 int error; 2373 2374 *fpp = NULL; 2375 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2376 return (EBADF); 2377 if (needrightsp != NULL) 2378 needrights = *needrightsp; 2379 else 2380 cap_rights_init(&needrights); 2381 if (maxprotp != NULL) 2382 cap_rights_set(&needrights, CAP_MMAP); 2383 error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights); 2384 if (error != 0) 2385 return (error); 2386 if (fp->f_ops == &badfileops) { 2387 fdrop(fp, td); 2388 return (EBADF); 2389 } 2390 2391#ifdef CAPABILITIES 2392 /* 2393 * If requested, convert capability rights to access flags. 2394 */ 2395 if (maxprotp != NULL) 2396 *maxprotp = cap_rights_to_vmprot(&haverights); 2397#else /* !CAPABILITIES */ 2398 if (maxprotp != NULL) 2399 *maxprotp = VM_PROT_ALL; 2400#endif /* CAPABILITIES */ 2401 2402 /* 2403 * FREAD and FWRITE failure return EBADF as per POSIX. 2404 */ 2405 error = 0; 2406 switch (flags) { 2407 case FREAD: 2408 case FWRITE: 2409 if ((fp->f_flag & flags) == 0) 2410 error = EBADF; 2411 break; 2412 case FEXEC: 2413 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2414 ((fp->f_flag & FWRITE) != 0)) 2415 error = EBADF; 2416 break; 2417 case 0: 2418 break; 2419 default: 2420 KASSERT(0, ("wrong flags")); 2421 } 2422 2423 if (error != 0) { 2424 fdrop(fp, td); 2425 return (error); 2426 } 2427 2428 *fpp = fp; 2429 return (0); 2430} 2431 2432int 2433fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2434{ 2435 2436 return(_fget(td, fd, fpp, 0, rightsp, NULL)); 2437} 2438 2439int 2440fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2441 struct file **fpp) 2442{ 2443 2444 return (_fget(td, fd, fpp, 0, rightsp, maxprotp)); 2445} 2446 2447int 2448fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2449{ 2450 2451 return(_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2452} 2453 2454int 2455fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2456{ 2457 2458 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2459} 2460 2461/* 2462 * Like fget() but loads the underlying vnode, or returns an error if the 2463 * descriptor does not represent a vnode. Note that pipes use vnodes but 2464 * never have VM objects. The returned vnode will be vref()'d. 2465 * 2466 * XXX: what about the unused flags ? 2467 */ 2468static __inline int 2469_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2470 struct vnode **vpp) 2471{ 2472 struct file *fp; 2473 int error; 2474 2475 *vpp = NULL; 2476 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2477 if (error != 0) 2478 return (error); 2479 if (fp->f_vnode == NULL) { 2480 error = EINVAL; 2481 } else { 2482 *vpp = fp->f_vnode; 2483 vref(*vpp); 2484 } 2485 fdrop(fp, td); 2486 2487 return (error); 2488} 2489 2490int 2491fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2492{ 2493 2494 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2495} 2496 2497int 2498fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2499 struct filecaps *havecaps, struct vnode **vpp) 2500{ 2501 struct filedesc *fdp; 2502 struct file *fp; 2503#ifdef CAPABILITIES 2504 int error; 2505#endif 2506 2507 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2508 return (EBADF); 2509 2510 fp = fget_locked(fdp, fd); 2511 if (fp == NULL || fp->f_ops == &badfileops) 2512 return (EBADF); 2513 2514#ifdef CAPABILITIES 2515 if (needrightsp != NULL) { 2516 error = cap_check(cap_rights(fdp, fd), needrightsp); 2517 if (error != 0) 2518 return (error); 2519 } 2520#endif 2521 2522 if (fp->f_vnode == NULL) 2523 return (EINVAL); 2524 2525 *vpp = fp->f_vnode; 2526 vref(*vpp); 2527 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2528 2529 return (0); 2530} 2531 2532int 2533fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2534{ 2535 2536 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2537} 2538 2539int 2540fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2541{ 2542 2543 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2544} 2545 2546#ifdef notyet 2547int 2548fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2549 struct vnode **vpp) 2550{ 2551 2552 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2553} 2554#endif 2555 2556/* 2557 * Like fget() but loads the underlying socket, or returns an error if the 2558 * descriptor does not represent a socket. 2559 * 2560 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2561 * in the future. 2562 * 2563 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2564 * on their file descriptor reference to prevent the socket from being free'd 2565 * during use. 2566 */ 2567int 2568fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2569 u_int *fflagp) 2570{ 2571 struct file *fp; 2572 int error; 2573 2574 *spp = NULL; 2575 if (fflagp != NULL) 2576 *fflagp = 0; 2577 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2578 return (error); 2579 if (fp->f_type != DTYPE_SOCKET) { 2580 error = ENOTSOCK; 2581 } else { 2582 *spp = fp->f_data; 2583 if (fflagp) 2584 *fflagp = fp->f_flag; 2585 SOCK_LOCK(*spp); 2586 soref(*spp); 2587 SOCK_UNLOCK(*spp); 2588 } 2589 fdrop(fp, td); 2590 2591 return (error); 2592} 2593 2594/* 2595 * Drop the reference count on the socket and XXX release the SX lock in the 2596 * future. The last reference closes the socket. 2597 * 2598 * Note: fputsock() is deprecated, see comment for fgetsock(). 2599 */ 2600void 2601fputsock(struct socket *so) 2602{ 2603 2604 ACCEPT_LOCK(); 2605 SOCK_LOCK(so); 2606 CURVNET_SET(so->so_vnet); 2607 sorele(so); 2608 CURVNET_RESTORE(); 2609} 2610 2611/* 2612 * Handle the last reference to a file being closed. 2613 */ 2614int 2615_fdrop(struct file *fp, struct thread *td) 2616{ 2617 int error; 2618 2619 error = 0; 2620 if (fp->f_count != 0) 2621 panic("fdrop: count %d", fp->f_count); 2622 if (fp->f_ops != &badfileops) 2623 error = fo_close(fp, td); 2624 atomic_subtract_int(&openfiles, 1); 2625 crfree(fp->f_cred); 2626 free(fp->f_advice, M_FADVISE); 2627 uma_zfree(file_zone, fp); 2628 2629 return (error); 2630} 2631 2632/* 2633 * Apply an advisory lock on a file descriptor. 2634 * 2635 * Just attempt to get a record lock of the requested type on the entire file 2636 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2637 */ 2638#ifndef _SYS_SYSPROTO_H_ 2639struct flock_args { 2640 int fd; 2641 int how; 2642}; 2643#endif 2644/* ARGSUSED */ 2645int 2646sys_flock(struct thread *td, struct flock_args *uap) 2647{ 2648 struct file *fp; 2649 struct vnode *vp; 2650 struct flock lf; 2651 cap_rights_t rights; 2652 int error; 2653 2654 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2655 if (error != 0) 2656 return (error); 2657 if (fp->f_type != DTYPE_VNODE) { 2658 fdrop(fp, td); 2659 return (EOPNOTSUPP); 2660 } 2661 2662 vp = fp->f_vnode; 2663 lf.l_whence = SEEK_SET; 2664 lf.l_start = 0; 2665 lf.l_len = 0; 2666 if (uap->how & LOCK_UN) { 2667 lf.l_type = F_UNLCK; 2668 atomic_clear_int(&fp->f_flag, FHASLOCK); 2669 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2670 goto done2; 2671 } 2672 if (uap->how & LOCK_EX) 2673 lf.l_type = F_WRLCK; 2674 else if (uap->how & LOCK_SH) 2675 lf.l_type = F_RDLCK; 2676 else { 2677 error = EBADF; 2678 goto done2; 2679 } 2680 atomic_set_int(&fp->f_flag, FHASLOCK); 2681 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2682 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2683done2: 2684 fdrop(fp, td); 2685 return (error); 2686} 2687/* 2688 * Duplicate the specified descriptor to a free descriptor. 2689 */ 2690int 2691dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2692 int openerror, int *indxp) 2693{ 2694 struct file *fp; 2695 int error, indx; 2696 2697 KASSERT(openerror == ENODEV || openerror == ENXIO, 2698 ("unexpected error %d in %s", openerror, __func__)); 2699 2700 /* 2701 * If the to-be-dup'd fd number is greater than the allowed number 2702 * of file descriptors, or the fd to be dup'd has already been 2703 * closed, then reject. 2704 */ 2705 FILEDESC_XLOCK(fdp); 2706 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2707 FILEDESC_XUNLOCK(fdp); 2708 return (EBADF); 2709 } 2710 2711 error = fdalloc(td, 0, &indx); 2712 if (error != 0) { 2713 FILEDESC_XUNLOCK(fdp); 2714 return (error); 2715 } 2716 2717 /* 2718 * There are two cases of interest here. 2719 * 2720 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2721 * 2722 * For ENXIO steal away the file structure from (dfd) and store it in 2723 * (indx). (dfd) is effectively closed by this operation. 2724 */ 2725 switch (openerror) { 2726 case ENODEV: 2727 /* 2728 * Check that the mode the file is being opened for is a 2729 * subset of the mode of the existing descriptor. 2730 */ 2731 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2732 fdunused(fdp, indx); 2733 FILEDESC_XUNLOCK(fdp); 2734 return (EACCES); 2735 } 2736 fhold(fp); 2737 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2738 filecaps_copy(&fdp->fd_ofiles[dfd].fde_caps, 2739 &fdp->fd_ofiles[indx].fde_caps); 2740 break; 2741 case ENXIO: 2742 /* 2743 * Steal away the file pointer from dfd and stuff it into indx. 2744 */ 2745 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2746 bzero(&fdp->fd_ofiles[dfd], sizeof(fdp->fd_ofiles[dfd])); 2747 fdunused(fdp, dfd); 2748 break; 2749 } 2750 FILEDESC_XUNLOCK(fdp); 2751 *indxp = indx; 2752 return (0); 2753} 2754 2755/* 2756 * Scan all active processes and prisons to see if any of them have a current 2757 * or root directory of `olddp'. If so, replace them with the new mount point. 2758 */ 2759void 2760mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2761{ 2762 struct filedesc *fdp; 2763 struct prison *pr; 2764 struct proc *p; 2765 int nrele; 2766 2767 if (vrefcnt(olddp) == 1) 2768 return; 2769 nrele = 0; 2770 sx_slock(&allproc_lock); 2771 FOREACH_PROC_IN_SYSTEM(p) { 2772 fdp = fdhold(p); 2773 if (fdp == NULL) 2774 continue; 2775 FILEDESC_XLOCK(fdp); 2776 if (fdp->fd_cdir == olddp) { 2777 vref(newdp); 2778 fdp->fd_cdir = newdp; 2779 nrele++; 2780 } 2781 if (fdp->fd_rdir == olddp) { 2782 vref(newdp); 2783 fdp->fd_rdir = newdp; 2784 nrele++; 2785 } 2786 if (fdp->fd_jdir == olddp) { 2787 vref(newdp); 2788 fdp->fd_jdir = newdp; 2789 nrele++; 2790 } 2791 FILEDESC_XUNLOCK(fdp); 2792 fddrop(fdp); 2793 } 2794 sx_sunlock(&allproc_lock); 2795 if (rootvnode == olddp) { 2796 vref(newdp); 2797 rootvnode = newdp; 2798 nrele++; 2799 } 2800 mtx_lock(&prison0.pr_mtx); 2801 if (prison0.pr_root == olddp) { 2802 vref(newdp); 2803 prison0.pr_root = newdp; 2804 nrele++; 2805 } 2806 mtx_unlock(&prison0.pr_mtx); 2807 sx_slock(&allprison_lock); 2808 TAILQ_FOREACH(pr, &allprison, pr_list) { 2809 mtx_lock(&pr->pr_mtx); 2810 if (pr->pr_root == olddp) { 2811 vref(newdp); 2812 pr->pr_root = newdp; 2813 nrele++; 2814 } 2815 mtx_unlock(&pr->pr_mtx); 2816 } 2817 sx_sunlock(&allprison_lock); 2818 while (nrele--) 2819 vrele(olddp); 2820} 2821 2822struct filedesc_to_leader * 2823filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2824{ 2825 struct filedesc_to_leader *fdtol; 2826 2827 fdtol = malloc(sizeof(struct filedesc_to_leader), 2828 M_FILEDESC_TO_LEADER, 2829 M_WAITOK); 2830 fdtol->fdl_refcount = 1; 2831 fdtol->fdl_holdcount = 0; 2832 fdtol->fdl_wakeup = 0; 2833 fdtol->fdl_leader = leader; 2834 if (old != NULL) { 2835 FILEDESC_XLOCK(fdp); 2836 fdtol->fdl_next = old->fdl_next; 2837 fdtol->fdl_prev = old; 2838 old->fdl_next = fdtol; 2839 fdtol->fdl_next->fdl_prev = fdtol; 2840 FILEDESC_XUNLOCK(fdp); 2841 } else { 2842 fdtol->fdl_next = fdtol; 2843 fdtol->fdl_prev = fdtol; 2844 } 2845 return (fdtol); 2846} 2847 2848/* 2849 * Get file structures globally. 2850 */ 2851static int 2852sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2853{ 2854 struct xfile xf; 2855 struct filedesc *fdp; 2856 struct file *fp; 2857 struct proc *p; 2858 int error, n; 2859 2860 error = sysctl_wire_old_buffer(req, 0); 2861 if (error != 0) 2862 return (error); 2863 if (req->oldptr == NULL) { 2864 n = 0; 2865 sx_slock(&allproc_lock); 2866 FOREACH_PROC_IN_SYSTEM(p) { 2867 if (p->p_state == PRS_NEW) 2868 continue; 2869 fdp = fdhold(p); 2870 if (fdp == NULL) 2871 continue; 2872 /* overestimates sparse tables. */ 2873 if (fdp->fd_lastfile > 0) 2874 n += fdp->fd_lastfile; 2875 fddrop(fdp); 2876 } 2877 sx_sunlock(&allproc_lock); 2878 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2879 } 2880 error = 0; 2881 bzero(&xf, sizeof(xf)); 2882 xf.xf_size = sizeof(xf); 2883 sx_slock(&allproc_lock); 2884 FOREACH_PROC_IN_SYSTEM(p) { 2885 PROC_LOCK(p); 2886 if (p->p_state == PRS_NEW) { 2887 PROC_UNLOCK(p); 2888 continue; 2889 } 2890 if (p_cansee(req->td, p) != 0) { 2891 PROC_UNLOCK(p); 2892 continue; 2893 } 2894 xf.xf_pid = p->p_pid; 2895 xf.xf_uid = p->p_ucred->cr_uid; 2896 PROC_UNLOCK(p); 2897 fdp = fdhold(p); 2898 if (fdp == NULL) 2899 continue; 2900 FILEDESC_SLOCK(fdp); 2901 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 2902 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 2903 continue; 2904 xf.xf_fd = n; 2905 xf.xf_file = fp; 2906 xf.xf_data = fp->f_data; 2907 xf.xf_vnode = fp->f_vnode; 2908 xf.xf_type = fp->f_type; 2909 xf.xf_count = fp->f_count; 2910 xf.xf_msgcount = 0; 2911 xf.xf_offset = foffset_get(fp); 2912 xf.xf_flag = fp->f_flag; 2913 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2914 if (error) 2915 break; 2916 } 2917 FILEDESC_SUNLOCK(fdp); 2918 fddrop(fdp); 2919 if (error) 2920 break; 2921 } 2922 sx_sunlock(&allproc_lock); 2923 return (error); 2924} 2925 2926SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 2927 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2928 2929#ifdef KINFO_OFILE_SIZE 2930CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2931#endif 2932 2933#ifdef COMPAT_FREEBSD7 2934static int 2935export_vnode_for_osysctl(struct vnode *vp, int type, 2936 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2937{ 2938 int error; 2939 char *fullpath, *freepath; 2940 2941 bzero(kif, sizeof(*kif)); 2942 kif->kf_structsize = sizeof(*kif); 2943 2944 vref(vp); 2945 kif->kf_fd = type; 2946 kif->kf_type = KF_TYPE_VNODE; 2947 /* This function only handles directories. */ 2948 if (vp->v_type != VDIR) { 2949 vrele(vp); 2950 return (ENOTDIR); 2951 } 2952 kif->kf_vnode_type = KF_VTYPE_VDIR; 2953 2954 /* 2955 * This is not a true file descriptor, so we set a bogus refcount 2956 * and offset to indicate these fields should be ignored. 2957 */ 2958 kif->kf_ref_count = -1; 2959 kif->kf_offset = -1; 2960 2961 freepath = NULL; 2962 fullpath = "-"; 2963 FILEDESC_SUNLOCK(fdp); 2964 vn_fullpath(curthread, vp, &fullpath, &freepath); 2965 vrele(vp); 2966 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2967 if (freepath != NULL) 2968 free(freepath, M_TEMP); 2969 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2970 FILEDESC_SLOCK(fdp); 2971 return (error); 2972} 2973 2974/* 2975 * Get per-process file descriptors for use by procstat(1), et al. 2976 */ 2977static int 2978sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 2979{ 2980 char *fullpath, *freepath; 2981 struct kinfo_ofile *kif; 2982 struct filedesc *fdp; 2983 int error, i, *name; 2984 struct shmfd *shmfd; 2985 struct socket *so; 2986 struct vnode *vp; 2987 struct ksem *ks; 2988 struct file *fp; 2989 struct proc *p; 2990 struct tty *tp; 2991 2992 name = (int *)arg1; 2993 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 2994 if (error != 0) 2995 return (error); 2996 fdp = fdhold(p); 2997 PROC_UNLOCK(p); 2998 if (fdp == NULL) 2999 return (ENOENT); 3000 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3001 FILEDESC_SLOCK(fdp); 3002 if (fdp->fd_cdir != NULL) 3003 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3004 fdp, req); 3005 if (fdp->fd_rdir != NULL) 3006 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3007 fdp, req); 3008 if (fdp->fd_jdir != NULL) 3009 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3010 fdp, req); 3011 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3012 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3013 continue; 3014 bzero(kif, sizeof(*kif)); 3015 kif->kf_structsize = sizeof(*kif); 3016 ks = NULL; 3017 vp = NULL; 3018 so = NULL; 3019 tp = NULL; 3020 shmfd = NULL; 3021 kif->kf_fd = i; 3022 3023 switch (fp->f_type) { 3024 case DTYPE_VNODE: 3025 kif->kf_type = KF_TYPE_VNODE; 3026 vp = fp->f_vnode; 3027 break; 3028 3029 case DTYPE_SOCKET: 3030 kif->kf_type = KF_TYPE_SOCKET; 3031 so = fp->f_data; 3032 break; 3033 3034 case DTYPE_PIPE: 3035 kif->kf_type = KF_TYPE_PIPE; 3036 break; 3037 3038 case DTYPE_FIFO: 3039 kif->kf_type = KF_TYPE_FIFO; 3040 vp = fp->f_vnode; 3041 break; 3042 3043 case DTYPE_KQUEUE: 3044 kif->kf_type = KF_TYPE_KQUEUE; 3045 break; 3046 3047 case DTYPE_CRYPTO: 3048 kif->kf_type = KF_TYPE_CRYPTO; 3049 break; 3050 3051 case DTYPE_MQUEUE: 3052 kif->kf_type = KF_TYPE_MQUEUE; 3053 break; 3054 3055 case DTYPE_SHM: 3056 kif->kf_type = KF_TYPE_SHM; 3057 shmfd = fp->f_data; 3058 break; 3059 3060 case DTYPE_SEM: 3061 kif->kf_type = KF_TYPE_SEM; 3062 ks = fp->f_data; 3063 break; 3064 3065 case DTYPE_PTS: 3066 kif->kf_type = KF_TYPE_PTS; 3067 tp = fp->f_data; 3068 break; 3069 3070 case DTYPE_PROCDESC: 3071 kif->kf_type = KF_TYPE_PROCDESC; 3072 break; 3073 3074 default: 3075 kif->kf_type = KF_TYPE_UNKNOWN; 3076 break; 3077 } 3078 kif->kf_ref_count = fp->f_count; 3079 if (fp->f_flag & FREAD) 3080 kif->kf_flags |= KF_FLAG_READ; 3081 if (fp->f_flag & FWRITE) 3082 kif->kf_flags |= KF_FLAG_WRITE; 3083 if (fp->f_flag & FAPPEND) 3084 kif->kf_flags |= KF_FLAG_APPEND; 3085 if (fp->f_flag & FASYNC) 3086 kif->kf_flags |= KF_FLAG_ASYNC; 3087 if (fp->f_flag & FFSYNC) 3088 kif->kf_flags |= KF_FLAG_FSYNC; 3089 if (fp->f_flag & FNONBLOCK) 3090 kif->kf_flags |= KF_FLAG_NONBLOCK; 3091 if (fp->f_flag & O_DIRECT) 3092 kif->kf_flags |= KF_FLAG_DIRECT; 3093 if (fp->f_flag & FHASLOCK) 3094 kif->kf_flags |= KF_FLAG_HASLOCK; 3095 kif->kf_offset = foffset_get(fp); 3096 if (vp != NULL) { 3097 vref(vp); 3098 switch (vp->v_type) { 3099 case VNON: 3100 kif->kf_vnode_type = KF_VTYPE_VNON; 3101 break; 3102 case VREG: 3103 kif->kf_vnode_type = KF_VTYPE_VREG; 3104 break; 3105 case VDIR: 3106 kif->kf_vnode_type = KF_VTYPE_VDIR; 3107 break; 3108 case VBLK: 3109 kif->kf_vnode_type = KF_VTYPE_VBLK; 3110 break; 3111 case VCHR: 3112 kif->kf_vnode_type = KF_VTYPE_VCHR; 3113 break; 3114 case VLNK: 3115 kif->kf_vnode_type = KF_VTYPE_VLNK; 3116 break; 3117 case VSOCK: 3118 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3119 break; 3120 case VFIFO: 3121 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3122 break; 3123 case VBAD: 3124 kif->kf_vnode_type = KF_VTYPE_VBAD; 3125 break; 3126 default: 3127 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3128 break; 3129 } 3130 /* 3131 * It is OK to drop the filedesc lock here as we will 3132 * re-validate and re-evaluate its properties when 3133 * the loop continues. 3134 */ 3135 freepath = NULL; 3136 fullpath = "-"; 3137 FILEDESC_SUNLOCK(fdp); 3138 vn_fullpath(curthread, vp, &fullpath, &freepath); 3139 vrele(vp); 3140 strlcpy(kif->kf_path, fullpath, 3141 sizeof(kif->kf_path)); 3142 if (freepath != NULL) 3143 free(freepath, M_TEMP); 3144 FILEDESC_SLOCK(fdp); 3145 } 3146 if (so != NULL) { 3147 struct sockaddr *sa; 3148 3149 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3150 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3151 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3152 free(sa, M_SONAME); 3153 } 3154 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3155 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3156 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3157 free(sa, M_SONAME); 3158 } 3159 kif->kf_sock_domain = 3160 so->so_proto->pr_domain->dom_family; 3161 kif->kf_sock_type = so->so_type; 3162 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3163 } 3164 if (tp != NULL) { 3165 strlcpy(kif->kf_path, tty_devname(tp), 3166 sizeof(kif->kf_path)); 3167 } 3168 if (shmfd != NULL) 3169 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3170 if (ks != NULL && ksem_info != NULL) 3171 ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL); 3172 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3173 if (error) 3174 break; 3175 } 3176 FILEDESC_SUNLOCK(fdp); 3177 fddrop(fdp); 3178 free(kif, M_TEMP); 3179 return (0); 3180} 3181 3182static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3183 CTLFLAG_RD||CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3184 "Process ofiledesc entries"); 3185#endif /* COMPAT_FREEBSD7 */ 3186 3187#ifdef KINFO_FILE_SIZE 3188CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3189#endif 3190 3191struct export_fd_buf { 3192 struct filedesc *fdp; 3193 struct sbuf *sb; 3194 ssize_t remainder; 3195 struct kinfo_file kif; 3196}; 3197 3198static int 3199export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt, 3200 int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf) 3201{ 3202 struct { 3203 int fflag; 3204 int kf_fflag; 3205 } fflags_table[] = { 3206 { FAPPEND, KF_FLAG_APPEND }, 3207 { FASYNC, KF_FLAG_ASYNC }, 3208 { FFSYNC, KF_FLAG_FSYNC }, 3209 { FHASLOCK, KF_FLAG_HASLOCK }, 3210 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3211 { FREAD, KF_FLAG_READ }, 3212 { FWRITE, KF_FLAG_WRITE }, 3213 { O_CREAT, KF_FLAG_CREAT }, 3214 { O_DIRECT, KF_FLAG_DIRECT }, 3215 { O_EXCL, KF_FLAG_EXCL }, 3216 { O_EXEC, KF_FLAG_EXEC }, 3217 { O_EXLOCK, KF_FLAG_EXLOCK }, 3218 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3219 { O_SHLOCK, KF_FLAG_SHLOCK }, 3220 { O_TRUNC, KF_FLAG_TRUNC } 3221 }; 3222#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3223 struct kinfo_file *kif; 3224 struct vnode *vp; 3225 int error, locked; 3226 unsigned int i; 3227 3228 if (efbuf->remainder == 0) 3229 return (0); 3230 kif = &efbuf->kif; 3231 bzero(kif, sizeof(*kif)); 3232 locked = efbuf->fdp != NULL; 3233 switch (type) { 3234 case KF_TYPE_FIFO: 3235 case KF_TYPE_VNODE: 3236 if (locked) { 3237 FILEDESC_SUNLOCK(efbuf->fdp); 3238 locked = 0; 3239 } 3240 vp = (struct vnode *)data; 3241 error = fill_vnode_info(vp, kif); 3242 vrele(vp); 3243 break; 3244 case KF_TYPE_SOCKET: 3245 error = fill_socket_info((struct socket *)data, kif); 3246 break; 3247 case KF_TYPE_PIPE: 3248 error = fill_pipe_info((struct pipe *)data, kif); 3249 break; 3250 case KF_TYPE_PTS: 3251 error = fill_pts_info((struct tty *)data, kif); 3252 break; 3253 case KF_TYPE_PROCDESC: 3254 error = fill_procdesc_info((struct procdesc *)data, kif); 3255 break; 3256 case KF_TYPE_SEM: 3257 error = fill_sem_info((struct file *)data, kif); 3258 break; 3259 case KF_TYPE_SHM: 3260 error = fill_shm_info((struct file *)data, kif); 3261 break; 3262 default: 3263 error = 0; 3264 } 3265 if (error == 0) 3266 kif->kf_status |= KF_ATTR_VALID; 3267 3268 /* 3269 * Translate file access flags. 3270 */ 3271 for (i = 0; i < NFFLAGS; i++) 3272 if (fflags & fflags_table[i].fflag) 3273 kif->kf_flags |= fflags_table[i].kf_fflag; 3274 if (rightsp != NULL) 3275 kif->kf_cap_rights = *rightsp; 3276 else 3277 cap_rights_init(&kif->kf_cap_rights); 3278 kif->kf_fd = fd; 3279 kif->kf_type = type; 3280 kif->kf_ref_count = refcnt; 3281 kif->kf_offset = offset; 3282 /* Pack record size down */ 3283 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3284 strlen(kif->kf_path) + 1; 3285 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3286 if (efbuf->remainder != -1) { 3287 if (efbuf->remainder < kif->kf_structsize) { 3288 /* Terminate export. */ 3289 efbuf->remainder = 0; 3290 if (efbuf->fdp != NULL && !locked) 3291 FILEDESC_SLOCK(efbuf->fdp); 3292 return (0); 3293 } 3294 efbuf->remainder -= kif->kf_structsize; 3295 } 3296 if (locked) 3297 FILEDESC_SUNLOCK(efbuf->fdp); 3298 error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize); 3299 if (efbuf->fdp != NULL) 3300 FILEDESC_SLOCK(efbuf->fdp); 3301 return (error); 3302} 3303 3304/* 3305 * Store a process file descriptor information to sbuf. 3306 * 3307 * Takes a locked proc as argument, and returns with the proc unlocked. 3308 */ 3309int 3310kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3311{ 3312 struct file *fp; 3313 struct filedesc *fdp; 3314 struct export_fd_buf *efbuf; 3315 struct vnode *cttyvp, *textvp, *tracevp; 3316 int64_t offset; 3317 void *data; 3318 int error, i; 3319 int type, refcnt, fflags; 3320 cap_rights_t rights; 3321 3322 PROC_LOCK_ASSERT(p, MA_OWNED); 3323 3324 /* ktrace vnode */ 3325 tracevp = p->p_tracevp; 3326 if (tracevp != NULL) 3327 vref(tracevp); 3328 /* text vnode */ 3329 textvp = p->p_textvp; 3330 if (textvp != NULL) 3331 vref(textvp); 3332 /* Controlling tty. */ 3333 cttyvp = NULL; 3334 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3335 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3336 if (cttyvp != NULL) 3337 vref(cttyvp); 3338 } 3339 fdp = fdhold(p); 3340 PROC_UNLOCK(p); 3341 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3342 efbuf->fdp = NULL; 3343 efbuf->sb = sb; 3344 efbuf->remainder = maxlen; 3345 if (tracevp != NULL) 3346 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3347 FREAD | FWRITE, -1, -1, NULL, efbuf); 3348 if (textvp != NULL) 3349 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3350 FREAD, -1, -1, NULL, efbuf); 3351 if (cttyvp != NULL) 3352 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3353 FREAD | FWRITE, -1, -1, NULL, efbuf); 3354 error = 0; 3355 if (fdp == NULL) 3356 goto fail; 3357 efbuf->fdp = fdp; 3358 FILEDESC_SLOCK(fdp); 3359 /* working directory */ 3360 if (fdp->fd_cdir != NULL) { 3361 vref(fdp->fd_cdir); 3362 data = fdp->fd_cdir; 3363 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3364 FREAD, -1, -1, NULL, efbuf); 3365 } 3366 /* root directory */ 3367 if (fdp->fd_rdir != NULL) { 3368 vref(fdp->fd_rdir); 3369 data = fdp->fd_rdir; 3370 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3371 FREAD, -1, -1, NULL, efbuf); 3372 } 3373 /* jail directory */ 3374 if (fdp->fd_jdir != NULL) { 3375 vref(fdp->fd_jdir); 3376 data = fdp->fd_jdir; 3377 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3378 FREAD, -1, -1, NULL, efbuf); 3379 } 3380 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3381 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3382 continue; 3383 data = NULL; 3384#ifdef CAPABILITIES 3385 rights = *cap_rights(fdp, i); 3386#else /* !CAPABILITIES */ 3387 cap_rights_init(&rights); 3388#endif 3389 switch (fp->f_type) { 3390 case DTYPE_VNODE: 3391 type = KF_TYPE_VNODE; 3392 vref(fp->f_vnode); 3393 data = fp->f_vnode; 3394 break; 3395 3396 case DTYPE_SOCKET: 3397 type = KF_TYPE_SOCKET; 3398 data = fp->f_data; 3399 break; 3400 3401 case DTYPE_PIPE: 3402 type = KF_TYPE_PIPE; 3403 data = fp->f_data; 3404 break; 3405 3406 case DTYPE_FIFO: 3407 type = KF_TYPE_FIFO; 3408 vref(fp->f_vnode); 3409 data = fp->f_vnode; 3410 break; 3411 3412 case DTYPE_KQUEUE: 3413 type = KF_TYPE_KQUEUE; 3414 break; 3415 3416 case DTYPE_CRYPTO: 3417 type = KF_TYPE_CRYPTO; 3418 break; 3419 3420 case DTYPE_MQUEUE: 3421 type = KF_TYPE_MQUEUE; 3422 break; 3423 3424 case DTYPE_SHM: 3425 type = KF_TYPE_SHM; 3426 data = fp; 3427 break; 3428 3429 case DTYPE_SEM: 3430 type = KF_TYPE_SEM; 3431 data = fp; 3432 break; 3433 3434 case DTYPE_PTS: 3435 type = KF_TYPE_PTS; 3436 data = fp->f_data; 3437 break; 3438 3439 case DTYPE_PROCDESC: 3440 type = KF_TYPE_PROCDESC; 3441 data = fp->f_data; 3442 break; 3443 3444 default: 3445 type = KF_TYPE_UNKNOWN; 3446 break; 3447 } 3448 refcnt = fp->f_count; 3449 fflags = fp->f_flag; 3450 offset = foffset_get(fp); 3451 3452 /* 3453 * Create sysctl entry. 3454 * It is OK to drop the filedesc lock here as we will 3455 * re-validate and re-evaluate its properties when 3456 * the loop continues. 3457 */ 3458 error = export_fd_to_sb(data, type, i, fflags, refcnt, 3459 offset, &rights, efbuf); 3460 if (error != 0) 3461 break; 3462 } 3463 FILEDESC_SUNLOCK(fdp); 3464 fddrop(fdp); 3465fail: 3466 free(efbuf, M_TEMP); 3467 return (error); 3468} 3469 3470#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3471 3472/* 3473 * Get per-process file descriptors for use by procstat(1), et al. 3474 */ 3475static int 3476sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3477{ 3478 struct sbuf sb; 3479 struct proc *p; 3480 ssize_t maxlen; 3481 int error, error2, *name; 3482 3483 name = (int *)arg1; 3484 3485 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3486 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3487 if (error != 0) { 3488 sbuf_delete(&sb); 3489 return (error); 3490 } 3491 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3492 error = kern_proc_filedesc_out(p, &sb, maxlen); 3493 error2 = sbuf_finish(&sb); 3494 sbuf_delete(&sb); 3495 return (error != 0 ? error : error2); 3496} 3497 3498int 3499vntype_to_kinfo(int vtype) 3500{ 3501 struct { 3502 int vtype; 3503 int kf_vtype; 3504 } vtypes_table[] = { 3505 { VBAD, KF_VTYPE_VBAD }, 3506 { VBLK, KF_VTYPE_VBLK }, 3507 { VCHR, KF_VTYPE_VCHR }, 3508 { VDIR, KF_VTYPE_VDIR }, 3509 { VFIFO, KF_VTYPE_VFIFO }, 3510 { VLNK, KF_VTYPE_VLNK }, 3511 { VNON, KF_VTYPE_VNON }, 3512 { VREG, KF_VTYPE_VREG }, 3513 { VSOCK, KF_VTYPE_VSOCK } 3514 }; 3515#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3516 unsigned int i; 3517 3518 /* 3519 * Perform vtype translation. 3520 */ 3521 for (i = 0; i < NVTYPES; i++) 3522 if (vtypes_table[i].vtype == vtype) 3523 break; 3524 if (i < NVTYPES) 3525 return (vtypes_table[i].kf_vtype); 3526 3527 return (KF_VTYPE_UNKNOWN); 3528} 3529 3530static int 3531fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3532{ 3533 struct vattr va; 3534 char *fullpath, *freepath; 3535 int error; 3536 3537 if (vp == NULL) 3538 return (1); 3539 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3540 freepath = NULL; 3541 fullpath = "-"; 3542 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3543 if (error == 0) { 3544 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3545 } 3546 if (freepath != NULL) 3547 free(freepath, M_TEMP); 3548 3549 /* 3550 * Retrieve vnode attributes. 3551 */ 3552 va.va_fsid = VNOVAL; 3553 va.va_rdev = NODEV; 3554 vn_lock(vp, LK_SHARED | LK_RETRY); 3555 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3556 VOP_UNLOCK(vp, 0); 3557 if (error != 0) 3558 return (error); 3559 if (va.va_fsid != VNOVAL) 3560 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3561 else 3562 kif->kf_un.kf_file.kf_file_fsid = 3563 vp->v_mount->mnt_stat.f_fsid.val[0]; 3564 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3565 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3566 kif->kf_un.kf_file.kf_file_size = va.va_size; 3567 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3568 return (0); 3569} 3570 3571static int 3572fill_socket_info(struct socket *so, struct kinfo_file *kif) 3573{ 3574 struct sockaddr *sa; 3575 struct inpcb *inpcb; 3576 struct unpcb *unpcb; 3577 int error; 3578 3579 if (so == NULL) 3580 return (1); 3581 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3582 kif->kf_sock_type = so->so_type; 3583 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3584 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3585 switch(kif->kf_sock_domain) { 3586 case AF_INET: 3587 case AF_INET6: 3588 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3589 if (so->so_pcb != NULL) { 3590 inpcb = (struct inpcb *)(so->so_pcb); 3591 kif->kf_un.kf_sock.kf_sock_inpcb = 3592 (uintptr_t)inpcb->inp_ppcb; 3593 } 3594 } 3595 break; 3596 case AF_UNIX: 3597 if (so->so_pcb != NULL) { 3598 unpcb = (struct unpcb *)(so->so_pcb); 3599 if (unpcb->unp_conn) { 3600 kif->kf_un.kf_sock.kf_sock_unpconn = 3601 (uintptr_t)unpcb->unp_conn; 3602 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3603 so->so_rcv.sb_state; 3604 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3605 so->so_snd.sb_state; 3606 } 3607 } 3608 break; 3609 } 3610 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3611 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3612 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3613 free(sa, M_SONAME); 3614 } 3615 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3616 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3617 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3618 free(sa, M_SONAME); 3619 } 3620 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3621 sizeof(kif->kf_path)); 3622 return (0); 3623} 3624 3625static int 3626fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3627{ 3628 3629 if (tp == NULL) 3630 return (1); 3631 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3632 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3633 return (0); 3634} 3635 3636static int 3637fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3638{ 3639 3640 if (pi == NULL) 3641 return (1); 3642 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3643 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3644 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3645 return (0); 3646} 3647 3648static int 3649fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3650{ 3651 3652 if (pdp == NULL) 3653 return (1); 3654 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3655 return (0); 3656} 3657 3658static int 3659fill_sem_info(struct file *fp, struct kinfo_file *kif) 3660{ 3661 struct thread *td; 3662 struct stat sb; 3663 3664 td = curthread; 3665 if (fp->f_data == NULL) 3666 return (1); 3667 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3668 return (1); 3669 if (ksem_info == NULL) 3670 return (1); 3671 ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path), 3672 &kif->kf_un.kf_sem.kf_sem_value); 3673 kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode; 3674 return (0); 3675} 3676 3677static int 3678fill_shm_info(struct file *fp, struct kinfo_file *kif) 3679{ 3680 struct thread *td; 3681 struct stat sb; 3682 3683 td = curthread; 3684 if (fp->f_data == NULL) 3685 return (1); 3686 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3687 return (1); 3688 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3689 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3690 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3691 return (0); 3692} 3693 3694static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3695 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3696 "Process filedesc entries"); 3697 3698#ifdef DDB 3699/* 3700 * For the purposes of debugging, generate a human-readable string for the 3701 * file type. 3702 */ 3703static const char * 3704file_type_to_name(short type) 3705{ 3706 3707 switch (type) { 3708 case 0: 3709 return ("zero"); 3710 case DTYPE_VNODE: 3711 return ("vnod"); 3712 case DTYPE_SOCKET: 3713 return ("sock"); 3714 case DTYPE_PIPE: 3715 return ("pipe"); 3716 case DTYPE_FIFO: 3717 return ("fifo"); 3718 case DTYPE_KQUEUE: 3719 return ("kque"); 3720 case DTYPE_CRYPTO: 3721 return ("crpt"); 3722 case DTYPE_MQUEUE: 3723 return ("mque"); 3724 case DTYPE_SHM: 3725 return ("shm"); 3726 case DTYPE_SEM: 3727 return ("ksem"); 3728 default: 3729 return ("unkn"); 3730 } 3731} 3732 3733/* 3734 * For the purposes of debugging, identify a process (if any, perhaps one of 3735 * many) that references the passed file in its file descriptor array. Return 3736 * NULL if none. 3737 */ 3738static struct proc * 3739file_to_first_proc(struct file *fp) 3740{ 3741 struct filedesc *fdp; 3742 struct proc *p; 3743 int n; 3744 3745 FOREACH_PROC_IN_SYSTEM(p) { 3746 if (p->p_state == PRS_NEW) 3747 continue; 3748 fdp = p->p_fd; 3749 if (fdp == NULL) 3750 continue; 3751 for (n = 0; n <= fdp->fd_lastfile; n++) { 3752 if (fp == fdp->fd_ofiles[n].fde_file) 3753 return (p); 3754 } 3755 } 3756 return (NULL); 3757} 3758 3759static void 3760db_print_file(struct file *fp, int header) 3761{ 3762 struct proc *p; 3763 3764 if (header) 3765 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3766 "File", "Type", "Data", "Flag", "GCFl", "Count", 3767 "MCount", "Vnode", "FPID", "FCmd"); 3768 p = file_to_first_proc(fp); 3769 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3770 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3771 0, fp->f_count, 0, fp->f_vnode, 3772 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3773} 3774 3775DB_SHOW_COMMAND(file, db_show_file) 3776{ 3777 struct file *fp; 3778 3779 if (!have_addr) { 3780 db_printf("usage: show file <addr>\n"); 3781 return; 3782 } 3783 fp = (struct file *)addr; 3784 db_print_file(fp, 1); 3785} 3786 3787DB_SHOW_COMMAND(files, db_show_files) 3788{ 3789 struct filedesc *fdp; 3790 struct file *fp; 3791 struct proc *p; 3792 int header; 3793 int n; 3794 3795 header = 1; 3796 FOREACH_PROC_IN_SYSTEM(p) { 3797 if (p->p_state == PRS_NEW) 3798 continue; 3799 if ((fdp = p->p_fd) == NULL) 3800 continue; 3801 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3802 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3803 continue; 3804 db_print_file(fp, header); 3805 header = 0; 3806 } 3807 } 3808} 3809#endif 3810 3811SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3812 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3813 3814SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3815 &maxfiles, 0, "Maximum number of files"); 3816 3817SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3818 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3819 3820/* ARGSUSED*/ 3821static void 3822filelistinit(void *dummy) 3823{ 3824 3825 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3826 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3827 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3828 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3829} 3830SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3831 3832/*-------------------------------------------------------------------*/ 3833 3834static int 3835badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3836 int flags, struct thread *td) 3837{ 3838 3839 return (EBADF); 3840} 3841 3842static int 3843badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3844 struct thread *td) 3845{ 3846 3847 return (EINVAL); 3848} 3849 3850static int 3851badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3852 struct thread *td) 3853{ 3854 3855 return (EBADF); 3856} 3857 3858static int 3859badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3860 struct thread *td) 3861{ 3862 3863 return (0); 3864} 3865 3866static int 3867badfo_kqfilter(struct file *fp, struct knote *kn) 3868{ 3869 3870 return (EBADF); 3871} 3872 3873static int 3874badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3875 struct thread *td) 3876{ 3877 3878 return (EBADF); 3879} 3880 3881static int 3882badfo_close(struct file *fp, struct thread *td) 3883{ 3884 3885 return (EBADF); 3886} 3887 3888static int 3889badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3890 struct thread *td) 3891{ 3892 3893 return (EBADF); 3894} 3895 3896static int 3897badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3898 struct thread *td) 3899{ 3900 3901 return (EBADF); 3902} 3903 3904static int 3905badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3906 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3907 int kflags, struct sendfile_sync *sfs, struct thread *td) 3908{ 3909 3910 return (EBADF); 3911} 3912 3913struct fileops badfileops = { 3914 .fo_read = badfo_readwrite, 3915 .fo_write = badfo_readwrite, 3916 .fo_truncate = badfo_truncate, 3917 .fo_ioctl = badfo_ioctl, 3918 .fo_poll = badfo_poll, 3919 .fo_kqfilter = badfo_kqfilter, 3920 .fo_stat = badfo_stat, 3921 .fo_close = badfo_close, 3922 .fo_chmod = badfo_chmod, 3923 .fo_chown = badfo_chown, 3924 .fo_sendfile = badfo_sendfile, 3925}; 3926 3927int 3928invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3929 struct thread *td) 3930{ 3931 3932 return (EINVAL); 3933} 3934 3935int 3936invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3937 struct thread *td) 3938{ 3939 3940 return (EINVAL); 3941} 3942 3943int 3944invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3945 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3946 int kflags, struct sendfile_sync *sfs, struct thread *td) 3947{ 3948 3949 return (EINVAL); 3950} 3951 3952/*-------------------------------------------------------------------*/ 3953 3954/* 3955 * File Descriptor pseudo-device driver (/dev/fd/). 3956 * 3957 * Opening minor device N dup()s the file (if any) connected to file 3958 * descriptor N belonging to the calling process. Note that this driver 3959 * consists of only the ``open()'' routine, because all subsequent 3960 * references to this file will be direct to the other driver. 3961 * 3962 * XXX: we could give this one a cloning event handler if necessary. 3963 */ 3964 3965/* ARGSUSED */ 3966static int 3967fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3968{ 3969 3970 /* 3971 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3972 * the file descriptor being sought for duplication. The error 3973 * return ensures that the vnode for this device will be released 3974 * by vn_open. Open will detect this special error and take the 3975 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3976 * will simply report the error. 3977 */ 3978 td->td_dupfd = dev2unit(dev); 3979 return (ENODEV); 3980} 3981 3982static struct cdevsw fildesc_cdevsw = { 3983 .d_version = D_VERSION, 3984 .d_open = fdopen, 3985 .d_name = "FD", 3986}; 3987 3988static void 3989fildesc_drvinit(void *unused) 3990{ 3991 struct cdev *dev; 3992 3993 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3994 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3995 make_dev_alias(dev, "stdin"); 3996 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3997 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3998 make_dev_alias(dev, "stdout"); 3999 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4000 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4001 make_dev_alias(dev, "stderr"); 4002} 4003 4004SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4005