kern_descrip.c revision 268505
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 268505 2014-07-10 20:59:54Z 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 * 298 * Avoid some work if fdp is about to be destroyed. 299 */ 300static inline void 301_fdfree(struct filedesc *fdp, int fd, int last) 302{ 303 struct filedescent *fde; 304 305 fde = &fdp->fd_ofiles[fd]; 306 filecaps_free(&fde->fde_caps); 307 if (last) 308 return; 309 bzero(fde, sizeof(*fde)); 310 fdunused(fdp, fd); 311} 312 313static inline void 314fdfree(struct filedesc *fdp, int fd) 315{ 316 317 _fdfree(fdp, fd, 0); 318} 319 320static inline void 321fdfree_last(struct filedesc *fdp, int fd) 322{ 323 324 _fdfree(fdp, fd, 1); 325} 326 327/* 328 * System calls on descriptors. 329 */ 330#ifndef _SYS_SYSPROTO_H_ 331struct getdtablesize_args { 332 int dummy; 333}; 334#endif 335/* ARGSUSED */ 336int 337sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 338{ 339 struct proc *p = td->td_proc; 340 uint64_t lim; 341 342 PROC_LOCK(p); 343 td->td_retval[0] = 344 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 345 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 346 PROC_UNLOCK(p); 347 if (lim < td->td_retval[0]) 348 td->td_retval[0] = lim; 349 return (0); 350} 351 352/* 353 * Duplicate a file descriptor to a particular value. 354 * 355 * Note: keep in mind that a potential race condition exists when closing 356 * descriptors from a shared descriptor table (via rfork). 357 */ 358#ifndef _SYS_SYSPROTO_H_ 359struct dup2_args { 360 u_int from; 361 u_int to; 362}; 363#endif 364/* ARGSUSED */ 365int 366sys_dup2(struct thread *td, struct dup2_args *uap) 367{ 368 369 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 370 td->td_retval)); 371} 372 373/* 374 * Duplicate a file descriptor. 375 */ 376#ifndef _SYS_SYSPROTO_H_ 377struct dup_args { 378 u_int fd; 379}; 380#endif 381/* ARGSUSED */ 382int 383sys_dup(struct thread *td, struct dup_args *uap) 384{ 385 386 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 387} 388 389/* 390 * The file control system call. 391 */ 392#ifndef _SYS_SYSPROTO_H_ 393struct fcntl_args { 394 int fd; 395 int cmd; 396 long arg; 397}; 398#endif 399/* ARGSUSED */ 400int 401sys_fcntl(struct thread *td, struct fcntl_args *uap) 402{ 403 struct flock fl; 404 struct __oflock ofl; 405 intptr_t arg; 406 int error; 407 int cmd; 408 409 error = 0; 410 cmd = uap->cmd; 411 switch (uap->cmd) { 412 case F_OGETLK: 413 case F_OSETLK: 414 case F_OSETLKW: 415 /* 416 * Convert old flock structure to new. 417 */ 418 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 419 fl.l_start = ofl.l_start; 420 fl.l_len = ofl.l_len; 421 fl.l_pid = ofl.l_pid; 422 fl.l_type = ofl.l_type; 423 fl.l_whence = ofl.l_whence; 424 fl.l_sysid = 0; 425 426 switch (uap->cmd) { 427 case F_OGETLK: 428 cmd = F_GETLK; 429 break; 430 case F_OSETLK: 431 cmd = F_SETLK; 432 break; 433 case F_OSETLKW: 434 cmd = F_SETLKW; 435 break; 436 } 437 arg = (intptr_t)&fl; 438 break; 439 case F_GETLK: 440 case F_SETLK: 441 case F_SETLKW: 442 case F_SETLK_REMOTE: 443 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 444 arg = (intptr_t)&fl; 445 break; 446 default: 447 arg = uap->arg; 448 break; 449 } 450 if (error) 451 return (error); 452 error = kern_fcntl(td, uap->fd, cmd, arg); 453 if (error) 454 return (error); 455 if (uap->cmd == F_OGETLK) { 456 ofl.l_start = fl.l_start; 457 ofl.l_len = fl.l_len; 458 ofl.l_pid = fl.l_pid; 459 ofl.l_type = fl.l_type; 460 ofl.l_whence = fl.l_whence; 461 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 462 } else if (uap->cmd == F_GETLK) { 463 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 464 } 465 return (error); 466} 467 468int 469kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 470{ 471 struct filedesc *fdp; 472 struct flock *flp; 473 struct file *fp, *fp2; 474 struct filedescent *fde; 475 struct proc *p; 476 struct vnode *vp; 477 cap_rights_t rights; 478 int error, flg, tmp; 479 u_int old, new; 480 uint64_t bsize; 481 off_t foffset; 482 483 error = 0; 484 flg = F_POSIX; 485 p = td->td_proc; 486 fdp = p->p_fd; 487 488 switch (cmd) { 489 case F_DUPFD: 490 tmp = arg; 491 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 492 break; 493 494 case F_DUPFD_CLOEXEC: 495 tmp = arg; 496 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp, 497 td->td_retval); 498 break; 499 500 case F_DUP2FD: 501 tmp = arg; 502 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 503 break; 504 505 case F_DUP2FD_CLOEXEC: 506 tmp = arg; 507 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp, 508 td->td_retval); 509 break; 510 511 case F_GETFD: 512 FILEDESC_SLOCK(fdp); 513 if ((fp = fget_locked(fdp, fd)) == NULL) { 514 FILEDESC_SUNLOCK(fdp); 515 error = EBADF; 516 break; 517 } 518 fde = &fdp->fd_ofiles[fd]; 519 td->td_retval[0] = 520 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 521 FILEDESC_SUNLOCK(fdp); 522 break; 523 524 case F_SETFD: 525 FILEDESC_XLOCK(fdp); 526 if ((fp = fget_locked(fdp, fd)) == NULL) { 527 FILEDESC_XUNLOCK(fdp); 528 error = EBADF; 529 break; 530 } 531 fde = &fdp->fd_ofiles[fd]; 532 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 533 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 534 FILEDESC_XUNLOCK(fdp); 535 break; 536 537 case F_GETFL: 538 error = fget_unlocked(fdp, fd, 539 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL); 540 if (error != 0) 541 break; 542 td->td_retval[0] = OFLAGS(fp->f_flag); 543 fdrop(fp, td); 544 break; 545 546 case F_SETFL: 547 error = fget_unlocked(fdp, fd, 548 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL); 549 if (error != 0) 550 break; 551 do { 552 tmp = flg = fp->f_flag; 553 tmp &= ~FCNTLFLAGS; 554 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 555 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 556 tmp = fp->f_flag & FNONBLOCK; 557 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 558 if (error != 0) { 559 fdrop(fp, td); 560 break; 561 } 562 tmp = fp->f_flag & FASYNC; 563 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 564 if (error == 0) { 565 fdrop(fp, td); 566 break; 567 } 568 atomic_clear_int(&fp->f_flag, FNONBLOCK); 569 tmp = 0; 570 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 571 fdrop(fp, td); 572 break; 573 574 case F_GETOWN: 575 error = fget_unlocked(fdp, fd, 576 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL); 577 if (error != 0) 578 break; 579 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 580 if (error == 0) 581 td->td_retval[0] = tmp; 582 fdrop(fp, td); 583 break; 584 585 case F_SETOWN: 586 error = fget_unlocked(fdp, fd, 587 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL); 588 if (error != 0) 589 break; 590 tmp = arg; 591 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 592 fdrop(fp, td); 593 break; 594 595 case F_SETLK_REMOTE: 596 error = priv_check(td, PRIV_NFS_LOCKD); 597 if (error) 598 return (error); 599 flg = F_REMOTE; 600 goto do_setlk; 601 602 case F_SETLKW: 603 flg |= F_WAIT; 604 /* FALLTHROUGH F_SETLK */ 605 606 case F_SETLK: 607 do_setlk: 608 cap_rights_init(&rights, CAP_FLOCK); 609 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL); 610 if (error != 0) 611 break; 612 if (fp->f_type != DTYPE_VNODE) { 613 error = EBADF; 614 fdrop(fp, td); 615 break; 616 } 617 618 flp = (struct flock *)arg; 619 if (flp->l_whence == SEEK_CUR) { 620 foffset = foffset_get(fp); 621 if (foffset < 0 || 622 (flp->l_start > 0 && 623 foffset > OFF_MAX - flp->l_start)) { 624 FILEDESC_SUNLOCK(fdp); 625 error = EOVERFLOW; 626 fdrop(fp, td); 627 break; 628 } 629 flp->l_start += foffset; 630 } 631 632 vp = fp->f_vnode; 633 switch (flp->l_type) { 634 case F_RDLCK: 635 if ((fp->f_flag & FREAD) == 0) { 636 error = EBADF; 637 break; 638 } 639 PROC_LOCK(p->p_leader); 640 p->p_leader->p_flag |= P_ADVLOCK; 641 PROC_UNLOCK(p->p_leader); 642 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 643 flp, flg); 644 break; 645 case F_WRLCK: 646 if ((fp->f_flag & FWRITE) == 0) { 647 error = EBADF; 648 break; 649 } 650 PROC_LOCK(p->p_leader); 651 p->p_leader->p_flag |= P_ADVLOCK; 652 PROC_UNLOCK(p->p_leader); 653 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 654 flp, flg); 655 break; 656 case F_UNLCK: 657 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 658 flp, flg); 659 break; 660 case F_UNLCKSYS: 661 /* 662 * Temporary api for testing remote lock 663 * infrastructure. 664 */ 665 if (flg != F_REMOTE) { 666 error = EINVAL; 667 break; 668 } 669 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 670 F_UNLCKSYS, flp, flg); 671 break; 672 default: 673 error = EINVAL; 674 break; 675 } 676 if (error != 0 || flp->l_type == F_UNLCK || 677 flp->l_type == F_UNLCKSYS) { 678 fdrop(fp, td); 679 break; 680 } 681 682 /* 683 * Check for a race with close. 684 * 685 * The vnode is now advisory locked (or unlocked, but this case 686 * is not really important) as the caller requested. 687 * We had to drop the filedesc lock, so we need to recheck if 688 * the descriptor is still valid, because if it was closed 689 * in the meantime we need to remove advisory lock from the 690 * vnode - close on any descriptor leading to an advisory 691 * locked vnode, removes that lock. 692 * We will return 0 on purpose in that case, as the result of 693 * successful advisory lock might have been externally visible 694 * already. This is fine - effectively we pretend to the caller 695 * that the closing thread was a bit slower and that the 696 * advisory lock succeeded before the close. 697 */ 698 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL); 699 if (error != 0) { 700 fdrop(fp, td); 701 break; 702 } 703 if (fp != fp2) { 704 flp->l_whence = SEEK_SET; 705 flp->l_start = 0; 706 flp->l_len = 0; 707 flp->l_type = F_UNLCK; 708 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 709 F_UNLCK, flp, F_POSIX); 710 } 711 fdrop(fp, td); 712 fdrop(fp2, td); 713 break; 714 715 case F_GETLK: 716 error = fget_unlocked(fdp, fd, 717 cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL); 718 if (error != 0) 719 break; 720 if (fp->f_type != DTYPE_VNODE) { 721 error = EBADF; 722 fdrop(fp, td); 723 break; 724 } 725 flp = (struct flock *)arg; 726 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 727 flp->l_type != F_UNLCK) { 728 error = EINVAL; 729 fdrop(fp, td); 730 break; 731 } 732 if (flp->l_whence == SEEK_CUR) { 733 foffset = foffset_get(fp); 734 if ((flp->l_start > 0 && 735 foffset > OFF_MAX - flp->l_start) || 736 (flp->l_start < 0 && 737 foffset < OFF_MIN - flp->l_start)) { 738 FILEDESC_SUNLOCK(fdp); 739 error = EOVERFLOW; 740 fdrop(fp, td); 741 break; 742 } 743 flp->l_start += foffset; 744 } 745 vp = fp->f_vnode; 746 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 747 F_POSIX); 748 fdrop(fp, td); 749 break; 750 751 case F_RDAHEAD: 752 arg = arg ? 128 * 1024: 0; 753 /* FALLTHROUGH */ 754 case F_READAHEAD: 755 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL); 756 if (error != 0) 757 break; 758 if (fp->f_type != DTYPE_VNODE) { 759 fdrop(fp, td); 760 error = EBADF; 761 break; 762 } 763 if (arg >= 0) { 764 vp = fp->f_vnode; 765 error = vn_lock(vp, LK_SHARED); 766 if (error != 0) { 767 fdrop(fp, td); 768 break; 769 } 770 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 771 VOP_UNLOCK(vp, 0); 772 fp->f_seqcount = (arg + bsize - 1) / bsize; 773 do { 774 new = old = fp->f_flag; 775 new |= FRDAHEAD; 776 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 777 } else { 778 do { 779 new = old = fp->f_flag; 780 new &= ~FRDAHEAD; 781 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 782 } 783 fdrop(fp, td); 784 break; 785 786 default: 787 error = EINVAL; 788 break; 789 } 790 return (error); 791} 792 793static int 794getmaxfd(struct proc *p) 795{ 796 int maxfd; 797 798 PROC_LOCK(p); 799 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 800 PROC_UNLOCK(p); 801 802 return (maxfd); 803} 804 805/* 806 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 807 */ 808int 809do_dup(struct thread *td, int flags, int old, int new, 810 register_t *retval) 811{ 812 struct filedesc *fdp; 813 struct filedescent *oldfde, *newfde; 814 struct proc *p; 815 struct file *fp; 816 struct file *delfp; 817 int error, maxfd; 818 819 p = td->td_proc; 820 fdp = p->p_fd; 821 822 /* 823 * Verify we have a valid descriptor to dup from and possibly to 824 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 825 * return EINVAL when the new descriptor is out of bounds. 826 */ 827 if (old < 0) 828 return (EBADF); 829 if (new < 0) 830 return (flags & DUP_FCNTL ? EINVAL : EBADF); 831 maxfd = getmaxfd(p); 832 if (new >= maxfd) 833 return (flags & DUP_FCNTL ? EINVAL : EBADF); 834 835 FILEDESC_XLOCK(fdp); 836 if (fget_locked(fdp, old) == NULL) { 837 FILEDESC_XUNLOCK(fdp); 838 return (EBADF); 839 } 840 oldfde = &fdp->fd_ofiles[old]; 841 if (flags & DUP_FIXED && old == new) { 842 *retval = new; 843 if (flags & DUP_CLOEXEC) 844 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 845 FILEDESC_XUNLOCK(fdp); 846 return (0); 847 } 848 fp = oldfde->fde_file; 849 fhold(fp); 850 851 /* 852 * If the caller specified a file descriptor, make sure the file 853 * table is large enough to hold it, and grab it. Otherwise, just 854 * allocate a new descriptor the usual way. 855 */ 856 if (flags & DUP_FIXED) { 857 if (new >= fdp->fd_nfiles) { 858 /* 859 * The resource limits are here instead of e.g. 860 * fdalloc(), because the file descriptor table may be 861 * shared between processes, so we can't really use 862 * racct_add()/racct_sub(). Instead of counting the 863 * number of actually allocated descriptors, just put 864 * the limit on the size of the file descriptor table. 865 */ 866#ifdef RACCT 867 PROC_LOCK(p); 868 error = racct_set(p, RACCT_NOFILE, new + 1); 869 PROC_UNLOCK(p); 870 if (error != 0) { 871 FILEDESC_XUNLOCK(fdp); 872 fdrop(fp, td); 873 return (EMFILE); 874 } 875#endif 876 fdgrowtable_exp(fdp, new + 1); 877 oldfde = &fdp->fd_ofiles[old]; 878 } 879 newfde = &fdp->fd_ofiles[new]; 880 if (newfde->fde_file == NULL) 881 fdused(fdp, new); 882 } else { 883 if ((error = fdalloc(td, new, &new)) != 0) { 884 FILEDESC_XUNLOCK(fdp); 885 fdrop(fp, td); 886 return (error); 887 } 888 newfde = &fdp->fd_ofiles[new]; 889 } 890 891 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 892 KASSERT(old != new, ("new fd is same as old")); 893 894 delfp = newfde->fde_file; 895 896 /* 897 * Duplicate the source descriptor. 898 */ 899 filecaps_free(&newfde->fde_caps); 900 *newfde = *oldfde; 901 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 902 if ((flags & DUP_CLOEXEC) != 0) 903 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 904 else 905 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 906 *retval = new; 907 908 if (delfp != NULL) { 909 (void) closefp(fdp, new, delfp, td, 1); 910 /* closefp() drops the FILEDESC lock for us. */ 911 } else { 912 FILEDESC_XUNLOCK(fdp); 913 } 914 915 return (0); 916} 917 918/* 919 * If sigio is on the list associated with a process or process group, 920 * disable signalling from the device, remove sigio from the list and 921 * free sigio. 922 */ 923void 924funsetown(struct sigio **sigiop) 925{ 926 struct sigio *sigio; 927 928 SIGIO_LOCK(); 929 sigio = *sigiop; 930 if (sigio == NULL) { 931 SIGIO_UNLOCK(); 932 return; 933 } 934 *(sigio->sio_myref) = NULL; 935 if ((sigio)->sio_pgid < 0) { 936 struct pgrp *pg = (sigio)->sio_pgrp; 937 PGRP_LOCK(pg); 938 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 939 sigio, sio_pgsigio); 940 PGRP_UNLOCK(pg); 941 } else { 942 struct proc *p = (sigio)->sio_proc; 943 PROC_LOCK(p); 944 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 945 sigio, sio_pgsigio); 946 PROC_UNLOCK(p); 947 } 948 SIGIO_UNLOCK(); 949 crfree(sigio->sio_ucred); 950 free(sigio, M_SIGIO); 951} 952 953/* 954 * Free a list of sigio structures. 955 * We only need to lock the SIGIO_LOCK because we have made ourselves 956 * inaccessible to callers of fsetown and therefore do not need to lock 957 * the proc or pgrp struct for the list manipulation. 958 */ 959void 960funsetownlst(struct sigiolst *sigiolst) 961{ 962 struct proc *p; 963 struct pgrp *pg; 964 struct sigio *sigio; 965 966 sigio = SLIST_FIRST(sigiolst); 967 if (sigio == NULL) 968 return; 969 p = NULL; 970 pg = NULL; 971 972 /* 973 * Every entry of the list should belong 974 * to a single proc or pgrp. 975 */ 976 if (sigio->sio_pgid < 0) { 977 pg = sigio->sio_pgrp; 978 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 979 } else /* if (sigio->sio_pgid > 0) */ { 980 p = sigio->sio_proc; 981 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 982 } 983 984 SIGIO_LOCK(); 985 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 986 *(sigio->sio_myref) = NULL; 987 if (pg != NULL) { 988 KASSERT(sigio->sio_pgid < 0, 989 ("Proc sigio in pgrp sigio list")); 990 KASSERT(sigio->sio_pgrp == pg, 991 ("Bogus pgrp in sigio list")); 992 PGRP_LOCK(pg); 993 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 994 sio_pgsigio); 995 PGRP_UNLOCK(pg); 996 } else /* if (p != NULL) */ { 997 KASSERT(sigio->sio_pgid > 0, 998 ("Pgrp sigio in proc sigio list")); 999 KASSERT(sigio->sio_proc == p, 1000 ("Bogus proc in sigio list")); 1001 PROC_LOCK(p); 1002 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1003 sio_pgsigio); 1004 PROC_UNLOCK(p); 1005 } 1006 SIGIO_UNLOCK(); 1007 crfree(sigio->sio_ucred); 1008 free(sigio, M_SIGIO); 1009 SIGIO_LOCK(); 1010 } 1011 SIGIO_UNLOCK(); 1012} 1013 1014/* 1015 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1016 * 1017 * After permission checking, add a sigio structure to the sigio list for 1018 * the process or process group. 1019 */ 1020int 1021fsetown(pid_t pgid, struct sigio **sigiop) 1022{ 1023 struct proc *proc; 1024 struct pgrp *pgrp; 1025 struct sigio *sigio; 1026 int ret; 1027 1028 if (pgid == 0) { 1029 funsetown(sigiop); 1030 return (0); 1031 } 1032 1033 ret = 0; 1034 1035 /* Allocate and fill in the new sigio out of locks. */ 1036 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1037 sigio->sio_pgid = pgid; 1038 sigio->sio_ucred = crhold(curthread->td_ucred); 1039 sigio->sio_myref = sigiop; 1040 1041 sx_slock(&proctree_lock); 1042 if (pgid > 0) { 1043 proc = pfind(pgid); 1044 if (proc == NULL) { 1045 ret = ESRCH; 1046 goto fail; 1047 } 1048 1049 /* 1050 * Policy - Don't allow a process to FSETOWN a process 1051 * in another session. 1052 * 1053 * Remove this test to allow maximum flexibility or 1054 * restrict FSETOWN to the current process or process 1055 * group for maximum safety. 1056 */ 1057 PROC_UNLOCK(proc); 1058 if (proc->p_session != curthread->td_proc->p_session) { 1059 ret = EPERM; 1060 goto fail; 1061 } 1062 1063 pgrp = NULL; 1064 } else /* if (pgid < 0) */ { 1065 pgrp = pgfind(-pgid); 1066 if (pgrp == NULL) { 1067 ret = ESRCH; 1068 goto fail; 1069 } 1070 PGRP_UNLOCK(pgrp); 1071 1072 /* 1073 * Policy - Don't allow a process to FSETOWN a process 1074 * in another session. 1075 * 1076 * Remove this test to allow maximum flexibility or 1077 * restrict FSETOWN to the current process or process 1078 * group for maximum safety. 1079 */ 1080 if (pgrp->pg_session != curthread->td_proc->p_session) { 1081 ret = EPERM; 1082 goto fail; 1083 } 1084 1085 proc = NULL; 1086 } 1087 funsetown(sigiop); 1088 if (pgid > 0) { 1089 PROC_LOCK(proc); 1090 /* 1091 * Since funsetownlst() is called without the proctree 1092 * locked, we need to check for P_WEXIT. 1093 * XXX: is ESRCH correct? 1094 */ 1095 if ((proc->p_flag & P_WEXIT) != 0) { 1096 PROC_UNLOCK(proc); 1097 ret = ESRCH; 1098 goto fail; 1099 } 1100 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1101 sigio->sio_proc = proc; 1102 PROC_UNLOCK(proc); 1103 } else { 1104 PGRP_LOCK(pgrp); 1105 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1106 sigio->sio_pgrp = pgrp; 1107 PGRP_UNLOCK(pgrp); 1108 } 1109 sx_sunlock(&proctree_lock); 1110 SIGIO_LOCK(); 1111 *sigiop = sigio; 1112 SIGIO_UNLOCK(); 1113 return (0); 1114 1115fail: 1116 sx_sunlock(&proctree_lock); 1117 crfree(sigio->sio_ucred); 1118 free(sigio, M_SIGIO); 1119 return (ret); 1120} 1121 1122/* 1123 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1124 */ 1125pid_t 1126fgetown(sigiop) 1127 struct sigio **sigiop; 1128{ 1129 pid_t pgid; 1130 1131 SIGIO_LOCK(); 1132 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1133 SIGIO_UNLOCK(); 1134 return (pgid); 1135} 1136 1137/* 1138 * Function drops the filedesc lock on return. 1139 */ 1140static int 1141closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1142 int holdleaders) 1143{ 1144 int error; 1145 1146 FILEDESC_XLOCK_ASSERT(fdp); 1147 1148 if (holdleaders) { 1149 if (td->td_proc->p_fdtol != NULL) { 1150 /* 1151 * Ask fdfree() to sleep to ensure that all relevant 1152 * process leaders can be traversed in closef(). 1153 */ 1154 fdp->fd_holdleaderscount++; 1155 } else { 1156 holdleaders = 0; 1157 } 1158 } 1159 1160 /* 1161 * We now hold the fp reference that used to be owned by the 1162 * descriptor array. We have to unlock the FILEDESC *AFTER* 1163 * knote_fdclose to prevent a race of the fd getting opened, a knote 1164 * added, and deleteing a knote for the new fd. 1165 */ 1166 knote_fdclose(td, fd); 1167 1168 /* 1169 * We need to notify mqueue if the object is of type mqueue. 1170 */ 1171 if (fp->f_type == DTYPE_MQUEUE) 1172 mq_fdclose(td, fd, fp); 1173 FILEDESC_XUNLOCK(fdp); 1174 1175 error = closef(fp, td); 1176 if (holdleaders) { 1177 FILEDESC_XLOCK(fdp); 1178 fdp->fd_holdleaderscount--; 1179 if (fdp->fd_holdleaderscount == 0 && 1180 fdp->fd_holdleaderswakeup != 0) { 1181 fdp->fd_holdleaderswakeup = 0; 1182 wakeup(&fdp->fd_holdleaderscount); 1183 } 1184 FILEDESC_XUNLOCK(fdp); 1185 } 1186 return (error); 1187} 1188 1189/* 1190 * Close a file descriptor. 1191 */ 1192#ifndef _SYS_SYSPROTO_H_ 1193struct close_args { 1194 int fd; 1195}; 1196#endif 1197/* ARGSUSED */ 1198int 1199sys_close(td, uap) 1200 struct thread *td; 1201 struct close_args *uap; 1202{ 1203 1204 return (kern_close(td, uap->fd)); 1205} 1206 1207int 1208kern_close(td, fd) 1209 struct thread *td; 1210 int fd; 1211{ 1212 struct filedesc *fdp; 1213 struct file *fp; 1214 1215 fdp = td->td_proc->p_fd; 1216 1217 AUDIT_SYSCLOSE(td, fd); 1218 1219 FILEDESC_XLOCK(fdp); 1220 if ((fp = fget_locked(fdp, fd)) == NULL) { 1221 FILEDESC_XUNLOCK(fdp); 1222 return (EBADF); 1223 } 1224 fdfree(fdp, fd); 1225 1226 /* closefp() drops the FILEDESC lock for us. */ 1227 return (closefp(fdp, fd, fp, td, 1)); 1228} 1229 1230/* 1231 * Close open file descriptors. 1232 */ 1233#ifndef _SYS_SYSPROTO_H_ 1234struct closefrom_args { 1235 int lowfd; 1236}; 1237#endif 1238/* ARGSUSED */ 1239int 1240sys_closefrom(struct thread *td, struct closefrom_args *uap) 1241{ 1242 struct filedesc *fdp; 1243 int fd; 1244 1245 fdp = td->td_proc->p_fd; 1246 AUDIT_ARG_FD(uap->lowfd); 1247 1248 /* 1249 * Treat negative starting file descriptor values identical to 1250 * closefrom(0) which closes all files. 1251 */ 1252 if (uap->lowfd < 0) 1253 uap->lowfd = 0; 1254 FILEDESC_SLOCK(fdp); 1255 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1256 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1257 FILEDESC_SUNLOCK(fdp); 1258 (void)kern_close(td, fd); 1259 FILEDESC_SLOCK(fdp); 1260 } 1261 } 1262 FILEDESC_SUNLOCK(fdp); 1263 return (0); 1264} 1265 1266#if defined(COMPAT_43) 1267/* 1268 * Return status information about a file descriptor. 1269 */ 1270#ifndef _SYS_SYSPROTO_H_ 1271struct ofstat_args { 1272 int fd; 1273 struct ostat *sb; 1274}; 1275#endif 1276/* ARGSUSED */ 1277int 1278ofstat(struct thread *td, struct ofstat_args *uap) 1279{ 1280 struct ostat oub; 1281 struct stat ub; 1282 int error; 1283 1284 error = kern_fstat(td, uap->fd, &ub); 1285 if (error == 0) { 1286 cvtstat(&ub, &oub); 1287 error = copyout(&oub, uap->sb, sizeof(oub)); 1288 } 1289 return (error); 1290} 1291#endif /* COMPAT_43 */ 1292 1293/* 1294 * Return status information about a file descriptor. 1295 */ 1296#ifndef _SYS_SYSPROTO_H_ 1297struct fstat_args { 1298 int fd; 1299 struct stat *sb; 1300}; 1301#endif 1302/* ARGSUSED */ 1303int 1304sys_fstat(struct thread *td, struct fstat_args *uap) 1305{ 1306 struct stat ub; 1307 int error; 1308 1309 error = kern_fstat(td, uap->fd, &ub); 1310 if (error == 0) 1311 error = copyout(&ub, uap->sb, sizeof(ub)); 1312 return (error); 1313} 1314 1315int 1316kern_fstat(struct thread *td, int fd, struct stat *sbp) 1317{ 1318 struct file *fp; 1319 cap_rights_t rights; 1320 int error; 1321 1322 AUDIT_ARG_FD(fd); 1323 1324 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1325 if (error != 0) 1326 return (error); 1327 1328 AUDIT_ARG_FILE(td->td_proc, fp); 1329 1330 error = fo_stat(fp, sbp, td->td_ucred, td); 1331 fdrop(fp, td); 1332#ifdef KTRACE 1333 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1334 ktrstat(sbp); 1335#endif 1336 return (error); 1337} 1338 1339/* 1340 * Return status information about a file descriptor. 1341 */ 1342#ifndef _SYS_SYSPROTO_H_ 1343struct nfstat_args { 1344 int fd; 1345 struct nstat *sb; 1346}; 1347#endif 1348/* ARGSUSED */ 1349int 1350sys_nfstat(struct thread *td, struct nfstat_args *uap) 1351{ 1352 struct nstat nub; 1353 struct stat ub; 1354 int error; 1355 1356 error = kern_fstat(td, uap->fd, &ub); 1357 if (error == 0) { 1358 cvtnstat(&ub, &nub); 1359 error = copyout(&nub, uap->sb, sizeof(nub)); 1360 } 1361 return (error); 1362} 1363 1364/* 1365 * Return pathconf information about a file descriptor. 1366 */ 1367#ifndef _SYS_SYSPROTO_H_ 1368struct fpathconf_args { 1369 int fd; 1370 int name; 1371}; 1372#endif 1373/* ARGSUSED */ 1374int 1375sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1376{ 1377 struct file *fp; 1378 struct vnode *vp; 1379 cap_rights_t rights; 1380 int error; 1381 1382 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1383 if (error != 0) 1384 return (error); 1385 1386 /* If asynchronous I/O is available, it works for all descriptors. */ 1387 if (uap->name == _PC_ASYNC_IO) { 1388 td->td_retval[0] = async_io_version; 1389 goto out; 1390 } 1391 vp = fp->f_vnode; 1392 if (vp != NULL) { 1393 vn_lock(vp, LK_SHARED | LK_RETRY); 1394 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1395 VOP_UNLOCK(vp, 0); 1396 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1397 if (uap->name != _PC_PIPE_BUF) { 1398 error = EINVAL; 1399 } else { 1400 td->td_retval[0] = PIPE_BUF; 1401 error = 0; 1402 } 1403 } else { 1404 error = EOPNOTSUPP; 1405 } 1406out: 1407 fdrop(fp, td); 1408 return (error); 1409} 1410 1411/* 1412 * Initialize filecaps structure. 1413 */ 1414void 1415filecaps_init(struct filecaps *fcaps) 1416{ 1417 1418 bzero(fcaps, sizeof(*fcaps)); 1419 fcaps->fc_nioctls = -1; 1420} 1421 1422/* 1423 * Copy filecaps structure allocating memory for ioctls array if needed. 1424 */ 1425void 1426filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1427{ 1428 size_t size; 1429 1430 *dst = *src; 1431 if (src->fc_ioctls != NULL) { 1432 KASSERT(src->fc_nioctls > 0, 1433 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1434 1435 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1436 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1437 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1438 } 1439} 1440 1441/* 1442 * Move filecaps structure to the new place and clear the old place. 1443 */ 1444void 1445filecaps_move(struct filecaps *src, struct filecaps *dst) 1446{ 1447 1448 *dst = *src; 1449 bzero(src, sizeof(*src)); 1450} 1451 1452/* 1453 * Fill the given filecaps structure with full rights. 1454 */ 1455static void 1456filecaps_fill(struct filecaps *fcaps) 1457{ 1458 1459 CAP_ALL(&fcaps->fc_rights); 1460 fcaps->fc_ioctls = NULL; 1461 fcaps->fc_nioctls = -1; 1462 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1463} 1464 1465/* 1466 * Free memory allocated within filecaps structure. 1467 */ 1468void 1469filecaps_free(struct filecaps *fcaps) 1470{ 1471 1472 free(fcaps->fc_ioctls, M_FILECAPS); 1473 bzero(fcaps, sizeof(*fcaps)); 1474} 1475 1476/* 1477 * Validate the given filecaps structure. 1478 */ 1479static void 1480filecaps_validate(const struct filecaps *fcaps, const char *func) 1481{ 1482 1483 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1484 ("%s: invalid rights", func)); 1485 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1486 ("%s: invalid fcntls", func)); 1487 KASSERT(fcaps->fc_fcntls == 0 || 1488 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1489 ("%s: fcntls without CAP_FCNTL", func)); 1490 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1491 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1492 ("%s: invalid ioctls", func)); 1493 KASSERT(fcaps->fc_nioctls == 0 || 1494 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1495 ("%s: ioctls without CAP_IOCTL", func)); 1496} 1497 1498static void 1499fdgrowtable_exp(struct filedesc *fdp, int nfd) 1500{ 1501 int nfd1; 1502 1503 FILEDESC_XLOCK_ASSERT(fdp); 1504 1505 nfd1 = fdp->fd_nfiles * 2; 1506 if (nfd1 < nfd) 1507 nfd1 = nfd; 1508 fdgrowtable(fdp, nfd1); 1509} 1510 1511/* 1512 * Grow the file table to accomodate (at least) nfd descriptors. 1513 */ 1514static void 1515fdgrowtable(struct filedesc *fdp, int nfd) 1516{ 1517 struct filedesc0 *fdp0; 1518 struct freetable *ft; 1519 struct filedescent *ntable; 1520 struct filedescent *otable; 1521 int nnfiles, onfiles; 1522 NDSLOTTYPE *nmap, *omap; 1523 1524 FILEDESC_XLOCK_ASSERT(fdp); 1525 1526 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1527 1528 /* save old values */ 1529 onfiles = fdp->fd_nfiles; 1530 otable = fdp->fd_ofiles; 1531 omap = fdp->fd_map; 1532 1533 /* compute the size of the new table */ 1534 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1535 if (nnfiles <= onfiles) 1536 /* the table is already large enough */ 1537 return; 1538 1539 /* 1540 * Allocate a new table. We need enough space for the 1541 * file entries themselves and the struct freetable we will use 1542 * when we decommission the table and place it on the freelist. 1543 * We place the struct freetable in the middle so we don't have 1544 * to worry about padding. 1545 */ 1546 ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable), 1547 M_FILEDESC, M_ZERO | M_WAITOK); 1548 /* copy the old data over and point at the new tables */ 1549 memcpy(ntable, otable, onfiles * sizeof(*otable)); 1550 fdp->fd_ofiles = ntable; 1551 1552 /* 1553 * Allocate a new map only if the old is not large enough. It will 1554 * grow at a slower rate than the table as it can map more 1555 * entries than the table can hold. 1556 */ 1557 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1558 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1559 M_ZERO | M_WAITOK); 1560 /* copy over the old data and update the pointer */ 1561 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1562 fdp->fd_map = nmap; 1563 } 1564 1565 /* 1566 * In order to have a valid pattern for fget_unlocked() 1567 * fdp->fd_nfiles must be the last member to be updated, otherwise 1568 * fget_unlocked() consumers may reference a new, higher value for 1569 * fdp->fd_nfiles before to access the fdp->fd_ofiles array, 1570 * resulting in OOB accesses. 1571 */ 1572 atomic_store_rel_int(&fdp->fd_nfiles, nnfiles); 1573 1574 /* 1575 * Do not free the old file table, as some threads may still 1576 * reference entries within it. Instead, place it on a freelist 1577 * which will be processed when the struct filedesc is released. 1578 * 1579 * Note that if onfiles == NDFILE, we're dealing with the original 1580 * static allocation contained within (struct filedesc0 *)fdp, 1581 * which must not be freed. 1582 */ 1583 if (onfiles > NDFILE) { 1584 ft = (struct freetable *)&otable[onfiles]; 1585 fdp0 = (struct filedesc0 *)fdp; 1586 ft->ft_table = otable; 1587 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1588 } 1589 /* 1590 * The map does not have the same possibility of threads still 1591 * holding references to it. So always free it as long as it 1592 * does not reference the original static allocation. 1593 */ 1594 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1595 free(omap, M_FILEDESC); 1596} 1597 1598/* 1599 * Allocate a file descriptor for the process. 1600 */ 1601int 1602fdalloc(struct thread *td, int minfd, int *result) 1603{ 1604 struct proc *p = td->td_proc; 1605 struct filedesc *fdp = p->p_fd; 1606 int fd = -1, maxfd, allocfd; 1607#ifdef RACCT 1608 int error; 1609#endif 1610 1611 FILEDESC_XLOCK_ASSERT(fdp); 1612 1613 if (fdp->fd_freefile > minfd) 1614 minfd = fdp->fd_freefile; 1615 1616 maxfd = getmaxfd(p); 1617 1618 /* 1619 * Search the bitmap for a free descriptor starting at minfd. 1620 * If none is found, grow the file table. 1621 */ 1622 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1623 if (fd >= maxfd) 1624 return (EMFILE); 1625 if (fd >= fdp->fd_nfiles) { 1626 allocfd = min(fd * 2, maxfd); 1627#ifdef RACCT 1628 PROC_LOCK(p); 1629 error = racct_set(p, RACCT_NOFILE, allocfd); 1630 PROC_UNLOCK(p); 1631 if (error != 0) 1632 return (EMFILE); 1633#endif 1634 /* 1635 * fd is already equal to first free descriptor >= minfd, so 1636 * we only need to grow the table and we are done. 1637 */ 1638 fdgrowtable_exp(fdp, allocfd); 1639 } 1640 1641 /* 1642 * Perform some sanity checks, then mark the file descriptor as 1643 * used and return it to the caller. 1644 */ 1645 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1646 ("invalid descriptor %d", fd)); 1647 KASSERT(!fdisused(fdp, fd), 1648 ("fd_first_free() returned non-free descriptor")); 1649 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1650 ("file descriptor isn't free")); 1651 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); 1652 fdused(fdp, fd); 1653 *result = fd; 1654 return (0); 1655} 1656 1657/* 1658 * Allocate n file descriptors for the process. 1659 */ 1660int 1661fdallocn(struct thread *td, int minfd, int *fds, int n) 1662{ 1663 struct proc *p = td->td_proc; 1664 struct filedesc *fdp = p->p_fd; 1665 int i; 1666 1667 FILEDESC_XLOCK_ASSERT(fdp); 1668 1669 for (i = 0; i < n; i++) 1670 if (fdalloc(td, 0, &fds[i]) != 0) 1671 break; 1672 1673 if (i < n) { 1674 for (i--; i >= 0; i--) 1675 fdunused(fdp, fds[i]); 1676 return (EMFILE); 1677 } 1678 1679 return (0); 1680} 1681 1682/* 1683 * Create a new open file structure and allocate a file decriptor for the 1684 * process that refers to it. We add one reference to the file for the 1685 * descriptor table and one reference for resultfp. This is to prevent us 1686 * being preempted and the entry in the descriptor table closed after we 1687 * release the FILEDESC lock. 1688 */ 1689int 1690falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1691{ 1692 struct file *fp; 1693 int error, fd; 1694 1695 error = falloc_noinstall(td, &fp); 1696 if (error) 1697 return (error); /* no reference held on error */ 1698 1699 error = finstall(td, fp, &fd, flags, NULL); 1700 if (error) { 1701 fdrop(fp, td); /* one reference (fp only) */ 1702 return (error); 1703 } 1704 1705 if (resultfp != NULL) 1706 *resultfp = fp; /* copy out result */ 1707 else 1708 fdrop(fp, td); /* release local reference */ 1709 1710 if (resultfd != NULL) 1711 *resultfd = fd; 1712 1713 return (0); 1714} 1715 1716/* 1717 * Create a new open file structure without allocating a file descriptor. 1718 */ 1719int 1720falloc_noinstall(struct thread *td, struct file **resultfp) 1721{ 1722 struct file *fp; 1723 int maxuserfiles = maxfiles - (maxfiles / 20); 1724 static struct timeval lastfail; 1725 static int curfail; 1726 1727 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1728 1729 if ((openfiles >= maxuserfiles && 1730 priv_check(td, PRIV_MAXFILES) != 0) || 1731 openfiles >= maxfiles) { 1732 if (ppsratecheck(&lastfail, &curfail, 1)) { 1733 printf("kern.maxfiles limit exceeded by uid %i, " 1734 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1735 } 1736 return (ENFILE); 1737 } 1738 atomic_add_int(&openfiles, 1); 1739 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1740 refcount_init(&fp->f_count, 1); 1741 fp->f_cred = crhold(td->td_ucred); 1742 fp->f_ops = &badfileops; 1743 fp->f_data = NULL; 1744 fp->f_vnode = NULL; 1745 *resultfp = fp; 1746 return (0); 1747} 1748 1749/* 1750 * Install a file in a file descriptor table. 1751 */ 1752int 1753finstall(struct thread *td, struct file *fp, int *fd, int flags, 1754 struct filecaps *fcaps) 1755{ 1756 struct filedesc *fdp = td->td_proc->p_fd; 1757 struct filedescent *fde; 1758 int error; 1759 1760 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1761 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1762 if (fcaps != NULL) 1763 filecaps_validate(fcaps, __func__); 1764 1765 FILEDESC_XLOCK(fdp); 1766 if ((error = fdalloc(td, 0, fd))) { 1767 FILEDESC_XUNLOCK(fdp); 1768 return (error); 1769 } 1770 fhold(fp); 1771 fde = &fdp->fd_ofiles[*fd]; 1772 fde->fde_file = fp; 1773 if ((flags & O_CLOEXEC) != 0) 1774 fde->fde_flags |= UF_EXCLOSE; 1775 if (fcaps != NULL) 1776 filecaps_move(fcaps, &fde->fde_caps); 1777 else 1778 filecaps_fill(&fde->fde_caps); 1779 FILEDESC_XUNLOCK(fdp); 1780 return (0); 1781} 1782 1783/* 1784 * Build a new filedesc structure from another. 1785 * Copy the current, root, and jail root vnode references. 1786 */ 1787struct filedesc * 1788fdinit(struct filedesc *fdp) 1789{ 1790 struct filedesc0 *newfdp; 1791 1792 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1793 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1794 if (fdp != NULL) { 1795 FILEDESC_SLOCK(fdp); 1796 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1797 if (newfdp->fd_fd.fd_cdir) 1798 VREF(newfdp->fd_fd.fd_cdir); 1799 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1800 if (newfdp->fd_fd.fd_rdir) 1801 VREF(newfdp->fd_fd.fd_rdir); 1802 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1803 if (newfdp->fd_fd.fd_jdir) 1804 VREF(newfdp->fd_fd.fd_jdir); 1805 FILEDESC_SUNLOCK(fdp); 1806 } 1807 1808 /* Create the file descriptor table. */ 1809 newfdp->fd_fd.fd_refcnt = 1; 1810 newfdp->fd_fd.fd_holdcnt = 1; 1811 newfdp->fd_fd.fd_cmask = CMASK; 1812 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1813 newfdp->fd_fd.fd_nfiles = NDFILE; 1814 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1815 newfdp->fd_fd.fd_lastfile = -1; 1816 return (&newfdp->fd_fd); 1817} 1818 1819static struct filedesc * 1820fdhold(struct proc *p) 1821{ 1822 struct filedesc *fdp; 1823 1824 mtx_lock(&fdesc_mtx); 1825 fdp = p->p_fd; 1826 if (fdp != NULL) 1827 fdp->fd_holdcnt++; 1828 mtx_unlock(&fdesc_mtx); 1829 return (fdp); 1830} 1831 1832static void 1833fddrop(struct filedesc *fdp) 1834{ 1835 struct filedesc0 *fdp0; 1836 struct freetable *ft; 1837 int i; 1838 1839 mtx_lock(&fdesc_mtx); 1840 i = --fdp->fd_holdcnt; 1841 mtx_unlock(&fdesc_mtx); 1842 if (i > 0) 1843 return; 1844 1845 FILEDESC_LOCK_DESTROY(fdp); 1846 fdp0 = (struct filedesc0 *)fdp; 1847 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1848 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1849 free(ft->ft_table, M_FILEDESC); 1850 } 1851 free(fdp, M_FILEDESC); 1852} 1853 1854/* 1855 * Share a filedesc structure. 1856 */ 1857struct filedesc * 1858fdshare(struct filedesc *fdp) 1859{ 1860 1861 FILEDESC_XLOCK(fdp); 1862 fdp->fd_refcnt++; 1863 FILEDESC_XUNLOCK(fdp); 1864 return (fdp); 1865} 1866 1867/* 1868 * Unshare a filedesc structure, if necessary by making a copy 1869 */ 1870void 1871fdunshare(struct thread *td) 1872{ 1873 struct filedesc *tmp; 1874 struct proc *p = td->td_proc; 1875 1876 if (p->p_fd->fd_refcnt == 1) 1877 return; 1878 1879 tmp = fdcopy(p->p_fd); 1880 fdescfree(td); 1881 p->p_fd = tmp; 1882} 1883 1884/* 1885 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1886 * this is to ease callers, not catch errors. 1887 */ 1888struct filedesc * 1889fdcopy(struct filedesc *fdp) 1890{ 1891 struct filedesc *newfdp; 1892 struct filedescent *nfde, *ofde; 1893 int i; 1894 1895 /* Certain daemons might not have file descriptors. */ 1896 if (fdp == NULL) 1897 return (NULL); 1898 1899 newfdp = fdinit(fdp); 1900 FILEDESC_SLOCK(fdp); 1901 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1902 FILEDESC_SUNLOCK(fdp); 1903 FILEDESC_XLOCK(newfdp); 1904 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1905 FILEDESC_XUNLOCK(newfdp); 1906 FILEDESC_SLOCK(fdp); 1907 } 1908 /* copy all passable descriptors (i.e. not kqueue) */ 1909 newfdp->fd_freefile = -1; 1910 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1911 ofde = &fdp->fd_ofiles[i]; 1912 if (fdisused(fdp, i) && 1913 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) && 1914 ofde->fde_file->f_ops != &badfileops) { 1915 nfde = &newfdp->fd_ofiles[i]; 1916 *nfde = *ofde; 1917 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1918 fhold(nfde->fde_file); 1919 newfdp->fd_lastfile = i; 1920 } else { 1921 if (newfdp->fd_freefile == -1) 1922 newfdp->fd_freefile = i; 1923 } 1924 } 1925 newfdp->fd_cmask = fdp->fd_cmask; 1926 FILEDESC_SUNLOCK(fdp); 1927 FILEDESC_XLOCK(newfdp); 1928 for (i = 0; i <= newfdp->fd_lastfile; ++i) { 1929 if (newfdp->fd_ofiles[i].fde_file != NULL) 1930 fdused(newfdp, i); 1931 } 1932 if (newfdp->fd_freefile == -1) 1933 newfdp->fd_freefile = i; 1934 FILEDESC_XUNLOCK(newfdp); 1935 return (newfdp); 1936} 1937 1938/* 1939 * Release a filedesc structure. 1940 */ 1941void 1942fdescfree(struct thread *td) 1943{ 1944 struct filedesc *fdp; 1945 int i; 1946 struct filedesc_to_leader *fdtol; 1947 struct file *fp; 1948 struct vnode *cdir, *jdir, *rdir, *vp; 1949 struct flock lf; 1950 1951 /* Certain daemons might not have file descriptors. */ 1952 fdp = td->td_proc->p_fd; 1953 if (fdp == NULL) 1954 return; 1955 1956#ifdef RACCT 1957 PROC_LOCK(td->td_proc); 1958 racct_set(td->td_proc, RACCT_NOFILE, 0); 1959 PROC_UNLOCK(td->td_proc); 1960#endif 1961 1962 /* Check for special need to clear POSIX style locks */ 1963 fdtol = td->td_proc->p_fdtol; 1964 if (fdtol != NULL) { 1965 FILEDESC_XLOCK(fdp); 1966 KASSERT(fdtol->fdl_refcount > 0, 1967 ("filedesc_to_refcount botch: fdl_refcount=%d", 1968 fdtol->fdl_refcount)); 1969 if (fdtol->fdl_refcount == 1 && 1970 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1971 for (i = 0; i <= fdp->fd_lastfile; i++) { 1972 fp = fdp->fd_ofiles[i].fde_file; 1973 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1974 continue; 1975 fhold(fp); 1976 FILEDESC_XUNLOCK(fdp); 1977 lf.l_whence = SEEK_SET; 1978 lf.l_start = 0; 1979 lf.l_len = 0; 1980 lf.l_type = F_UNLCK; 1981 vp = fp->f_vnode; 1982 (void) VOP_ADVLOCK(vp, 1983 (caddr_t)td->td_proc->p_leader, F_UNLCK, 1984 &lf, F_POSIX); 1985 FILEDESC_XLOCK(fdp); 1986 fdrop(fp, td); 1987 } 1988 } 1989 retry: 1990 if (fdtol->fdl_refcount == 1) { 1991 if (fdp->fd_holdleaderscount > 0 && 1992 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1993 /* 1994 * close() or do_dup() has cleared a reference 1995 * in a shared file descriptor table. 1996 */ 1997 fdp->fd_holdleaderswakeup = 1; 1998 sx_sleep(&fdp->fd_holdleaderscount, 1999 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2000 goto retry; 2001 } 2002 if (fdtol->fdl_holdcount > 0) { 2003 /* 2004 * Ensure that fdtol->fdl_leader remains 2005 * valid in closef(). 2006 */ 2007 fdtol->fdl_wakeup = 1; 2008 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2009 "fdlhold", 0); 2010 goto retry; 2011 } 2012 } 2013 fdtol->fdl_refcount--; 2014 if (fdtol->fdl_refcount == 0 && 2015 fdtol->fdl_holdcount == 0) { 2016 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2017 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2018 } else 2019 fdtol = NULL; 2020 td->td_proc->p_fdtol = NULL; 2021 FILEDESC_XUNLOCK(fdp); 2022 if (fdtol != NULL) 2023 free(fdtol, M_FILEDESC_TO_LEADER); 2024 } 2025 2026 mtx_lock(&fdesc_mtx); 2027 td->td_proc->p_fd = NULL; 2028 mtx_unlock(&fdesc_mtx); 2029 2030 FILEDESC_XLOCK(fdp); 2031 i = --fdp->fd_refcnt; 2032 if (i > 0) { 2033 FILEDESC_XUNLOCK(fdp); 2034 return; 2035 } 2036 2037 fdp->fd_nfiles = 0; 2038 2039 cdir = fdp->fd_cdir; 2040 fdp->fd_cdir = NULL; 2041 rdir = fdp->fd_rdir; 2042 fdp->fd_rdir = NULL; 2043 jdir = fdp->fd_jdir; 2044 fdp->fd_jdir = NULL; 2045 FILEDESC_XUNLOCK(fdp); 2046 2047 for (i = 0; i <= fdp->fd_lastfile; i++) { 2048 fp = fdp->fd_ofiles[i].fde_file; 2049 if (fp != NULL) { 2050 fdfree_last(fdp, i); 2051 (void) closef(fp, td); 2052 } 2053 } 2054 2055 if (fdp->fd_nfiles > NDFILE) 2056 free(fdp->fd_ofiles, M_FILEDESC); 2057 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2058 free(fdp->fd_map, M_FILEDESC); 2059 2060 if (cdir != NULL) 2061 vrele(cdir); 2062 if (rdir != NULL) 2063 vrele(rdir); 2064 if (jdir != NULL) 2065 vrele(jdir); 2066 2067 fddrop(fdp); 2068} 2069 2070/* 2071 * For setugid programs, we don't want to people to use that setugidness 2072 * to generate error messages which write to a file which otherwise would 2073 * otherwise be off-limits to the process. We check for filesystems where 2074 * the vnode can change out from under us after execve (like [lin]procfs). 2075 * 2076 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2077 * sufficient. We also don't check for setugidness since we know we are. 2078 */ 2079static int 2080is_unsafe(struct file *fp) 2081{ 2082 if (fp->f_type == DTYPE_VNODE) { 2083 struct vnode *vp = fp->f_vnode; 2084 2085 if ((vp->v_vflag & VV_PROCDEP) != 0) 2086 return (1); 2087 } 2088 return (0); 2089} 2090 2091/* 2092 * Make this setguid thing safe, if at all possible. 2093 */ 2094void 2095setugidsafety(struct thread *td) 2096{ 2097 struct filedesc *fdp; 2098 struct file *fp; 2099 int i; 2100 2101 fdp = td->td_proc->p_fd; 2102 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2103 FILEDESC_XLOCK(fdp); 2104 for (i = 0; i <= fdp->fd_lastfile; i++) { 2105 if (i > 2) 2106 break; 2107 fp = fdp->fd_ofiles[i].fde_file; 2108 if (fp != NULL && is_unsafe(fp)) { 2109 knote_fdclose(td, i); 2110 /* 2111 * NULL-out descriptor prior to close to avoid 2112 * a race while close blocks. 2113 */ 2114 fdfree(fdp, i); 2115 FILEDESC_XUNLOCK(fdp); 2116 (void) closef(fp, td); 2117 FILEDESC_XLOCK(fdp); 2118 } 2119 } 2120 FILEDESC_XUNLOCK(fdp); 2121} 2122 2123/* 2124 * If a specific file object occupies a specific file descriptor, close the 2125 * file descriptor entry and drop a reference on the file object. This is a 2126 * convenience function to handle a subsequent error in a function that calls 2127 * falloc() that handles the race that another thread might have closed the 2128 * file descriptor out from under the thread creating the file object. 2129 */ 2130void 2131fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2132{ 2133 2134 FILEDESC_XLOCK(fdp); 2135 if (fdp->fd_ofiles[idx].fde_file == fp) { 2136 fdfree(fdp, idx); 2137 FILEDESC_XUNLOCK(fdp); 2138 fdrop(fp, td); 2139 } else 2140 FILEDESC_XUNLOCK(fdp); 2141} 2142 2143/* 2144 * Close any files on exec? 2145 */ 2146void 2147fdcloseexec(struct thread *td) 2148{ 2149 struct filedesc *fdp; 2150 struct filedescent *fde; 2151 struct file *fp; 2152 int i; 2153 2154 fdp = td->td_proc->p_fd; 2155 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2156 FILEDESC_XLOCK(fdp); 2157 for (i = 0; i <= fdp->fd_lastfile; i++) { 2158 fde = &fdp->fd_ofiles[i]; 2159 fp = fde->fde_file; 2160 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2161 (fde->fde_flags & UF_EXCLOSE))) { 2162 fdfree(fdp, i); 2163 (void) closefp(fdp, i, fp, td, 0); 2164 /* closefp() drops the FILEDESC lock. */ 2165 FILEDESC_XLOCK(fdp); 2166 } 2167 } 2168 FILEDESC_XUNLOCK(fdp); 2169} 2170 2171/* 2172 * It is unsafe for set[ug]id processes to be started with file 2173 * descriptors 0..2 closed, as these descriptors are given implicit 2174 * significance in the Standard C library. fdcheckstd() will create a 2175 * descriptor referencing /dev/null for each of stdin, stdout, and 2176 * stderr that is not already open. 2177 */ 2178int 2179fdcheckstd(struct thread *td) 2180{ 2181 struct filedesc *fdp; 2182 register_t retval, save; 2183 int i, error, devnull; 2184 2185 fdp = td->td_proc->p_fd; 2186 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2187 devnull = -1; 2188 error = 0; 2189 for (i = 0; i < 3; i++) { 2190 if (fdp->fd_ofiles[i].fde_file != NULL) 2191 continue; 2192 if (devnull < 0) { 2193 save = td->td_retval[0]; 2194 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2195 O_RDWR, 0); 2196 devnull = td->td_retval[0]; 2197 td->td_retval[0] = save; 2198 if (error) 2199 break; 2200 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2201 } else { 2202 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2203 if (error != 0) 2204 break; 2205 } 2206 } 2207 return (error); 2208} 2209 2210/* 2211 * Internal form of close. Decrement reference count on file structure. 2212 * Note: td may be NULL when closing a file that was being passed in a 2213 * message. 2214 * 2215 * XXXRW: Giant is not required for the caller, but often will be held; this 2216 * makes it moderately likely the Giant will be recursed in the VFS case. 2217 */ 2218int 2219closef(struct file *fp, struct thread *td) 2220{ 2221 struct vnode *vp; 2222 struct flock lf; 2223 struct filedesc_to_leader *fdtol; 2224 struct filedesc *fdp; 2225 2226 /* 2227 * POSIX record locking dictates that any close releases ALL 2228 * locks owned by this process. This is handled by setting 2229 * a flag in the unlock to free ONLY locks obeying POSIX 2230 * semantics, and not to free BSD-style file locks. 2231 * If the descriptor was in a message, POSIX-style locks 2232 * aren't passed with the descriptor, and the thread pointer 2233 * will be NULL. Callers should be careful only to pass a 2234 * NULL thread pointer when there really is no owning 2235 * context that might have locks, or the locks will be 2236 * leaked. 2237 */ 2238 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2239 vp = fp->f_vnode; 2240 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2241 lf.l_whence = SEEK_SET; 2242 lf.l_start = 0; 2243 lf.l_len = 0; 2244 lf.l_type = F_UNLCK; 2245 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2246 F_UNLCK, &lf, F_POSIX); 2247 } 2248 fdtol = td->td_proc->p_fdtol; 2249 if (fdtol != NULL) { 2250 /* 2251 * Handle special case where file descriptor table is 2252 * shared between multiple process leaders. 2253 */ 2254 fdp = td->td_proc->p_fd; 2255 FILEDESC_XLOCK(fdp); 2256 for (fdtol = fdtol->fdl_next; 2257 fdtol != td->td_proc->p_fdtol; 2258 fdtol = fdtol->fdl_next) { 2259 if ((fdtol->fdl_leader->p_flag & 2260 P_ADVLOCK) == 0) 2261 continue; 2262 fdtol->fdl_holdcount++; 2263 FILEDESC_XUNLOCK(fdp); 2264 lf.l_whence = SEEK_SET; 2265 lf.l_start = 0; 2266 lf.l_len = 0; 2267 lf.l_type = F_UNLCK; 2268 vp = fp->f_vnode; 2269 (void) VOP_ADVLOCK(vp, 2270 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2271 F_POSIX); 2272 FILEDESC_XLOCK(fdp); 2273 fdtol->fdl_holdcount--; 2274 if (fdtol->fdl_holdcount == 0 && 2275 fdtol->fdl_wakeup != 0) { 2276 fdtol->fdl_wakeup = 0; 2277 wakeup(fdtol); 2278 } 2279 } 2280 FILEDESC_XUNLOCK(fdp); 2281 } 2282 } 2283 return (fdrop(fp, td)); 2284} 2285 2286/* 2287 * Initialize the file pointer with the specified properties. 2288 * 2289 * The ops are set with release semantics to be certain that the flags, type, 2290 * and data are visible when ops is. This is to prevent ops methods from being 2291 * called with bad data. 2292 */ 2293void 2294finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2295{ 2296 fp->f_data = data; 2297 fp->f_flag = flag; 2298 fp->f_type = type; 2299 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2300} 2301 2302int 2303fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2304 int needfcntl, struct file **fpp, cap_rights_t *haverightsp) 2305{ 2306 struct file *fp; 2307 u_int count; 2308#ifdef CAPABILITIES 2309 cap_rights_t haverights; 2310 int error; 2311#endif 2312 2313 /* 2314 * Avoid reads reordering and then a first access to the 2315 * fdp->fd_ofiles table which could result in OOB operation. 2316 */ 2317 if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles)) 2318 return (EBADF); 2319 /* 2320 * Fetch the descriptor locklessly. We avoid fdrop() races by 2321 * never raising a refcount above 0. To accomplish this we have 2322 * to use a cmpset loop rather than an atomic_add. The descriptor 2323 * must be re-verified once we acquire a reference to be certain 2324 * that the identity is still correct and we did not lose a race 2325 * due to preemption. 2326 */ 2327 for (;;) { 2328 fp = fdp->fd_ofiles[fd].fde_file; 2329 if (fp == NULL) 2330 return (EBADF); 2331#ifdef CAPABILITIES 2332 haverights = *cap_rights(fdp, fd); 2333 if (needrightsp != NULL) { 2334 error = cap_check(&haverights, needrightsp); 2335 if (error != 0) 2336 return (error); 2337 if (cap_rights_is_set(needrightsp, CAP_FCNTL)) { 2338 error = cap_fcntl_check(fdp, fd, needfcntl); 2339 if (error != 0) 2340 return (error); 2341 } 2342 } 2343#endif 2344 count = fp->f_count; 2345 if (count == 0) 2346 continue; 2347 /* 2348 * Use an acquire barrier to prevent caching of fd_ofiles 2349 * so it is refreshed for verification. 2350 */ 2351 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2352 continue; 2353 if (fp == fdp->fd_ofiles[fd].fde_file) 2354 break; 2355 fdrop(fp, curthread); 2356 } 2357 *fpp = fp; 2358 if (haverightsp != NULL) { 2359#ifdef CAPABILITIES 2360 *haverightsp = haverights; 2361#else 2362 CAP_ALL(haverightsp); 2363#endif 2364 } 2365 return (0); 2366} 2367 2368/* 2369 * Extract the file pointer associated with the specified descriptor for the 2370 * current user process. 2371 * 2372 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2373 * returned. 2374 * 2375 * File's rights will be checked against the capability rights mask. 2376 * 2377 * If an error occured the non-zero error is returned and *fpp is set to 2378 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2379 * responsible for fdrop(). 2380 */ 2381static __inline int 2382_fget(struct thread *td, int fd, struct file **fpp, int flags, 2383 cap_rights_t *needrightsp, u_char *maxprotp) 2384{ 2385 struct filedesc *fdp; 2386 struct file *fp; 2387 cap_rights_t haverights, needrights; 2388 int error; 2389 2390 *fpp = NULL; 2391 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2392 return (EBADF); 2393 if (needrightsp != NULL) 2394 needrights = *needrightsp; 2395 else 2396 cap_rights_init(&needrights); 2397 if (maxprotp != NULL) 2398 cap_rights_set(&needrights, CAP_MMAP); 2399 error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights); 2400 if (error != 0) 2401 return (error); 2402 if (fp->f_ops == &badfileops) { 2403 fdrop(fp, td); 2404 return (EBADF); 2405 } 2406 2407#ifdef CAPABILITIES 2408 /* 2409 * If requested, convert capability rights to access flags. 2410 */ 2411 if (maxprotp != NULL) 2412 *maxprotp = cap_rights_to_vmprot(&haverights); 2413#else /* !CAPABILITIES */ 2414 if (maxprotp != NULL) 2415 *maxprotp = VM_PROT_ALL; 2416#endif /* CAPABILITIES */ 2417 2418 /* 2419 * FREAD and FWRITE failure return EBADF as per POSIX. 2420 */ 2421 error = 0; 2422 switch (flags) { 2423 case FREAD: 2424 case FWRITE: 2425 if ((fp->f_flag & flags) == 0) 2426 error = EBADF; 2427 break; 2428 case FEXEC: 2429 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2430 ((fp->f_flag & FWRITE) != 0)) 2431 error = EBADF; 2432 break; 2433 case 0: 2434 break; 2435 default: 2436 KASSERT(0, ("wrong flags")); 2437 } 2438 2439 if (error != 0) { 2440 fdrop(fp, td); 2441 return (error); 2442 } 2443 2444 *fpp = fp; 2445 return (0); 2446} 2447 2448int 2449fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2450{ 2451 2452 return(_fget(td, fd, fpp, 0, rightsp, NULL)); 2453} 2454 2455int 2456fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2457 struct file **fpp) 2458{ 2459 2460 return (_fget(td, fd, fpp, 0, rightsp, maxprotp)); 2461} 2462 2463int 2464fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2465{ 2466 2467 return(_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2468} 2469 2470int 2471fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2472{ 2473 2474 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2475} 2476 2477/* 2478 * Like fget() but loads the underlying vnode, or returns an error if the 2479 * descriptor does not represent a vnode. Note that pipes use vnodes but 2480 * never have VM objects. The returned vnode will be vref()'d. 2481 * 2482 * XXX: what about the unused flags ? 2483 */ 2484static __inline int 2485_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2486 struct vnode **vpp) 2487{ 2488 struct file *fp; 2489 int error; 2490 2491 *vpp = NULL; 2492 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2493 if (error != 0) 2494 return (error); 2495 if (fp->f_vnode == NULL) { 2496 error = EINVAL; 2497 } else { 2498 *vpp = fp->f_vnode; 2499 vref(*vpp); 2500 } 2501 fdrop(fp, td); 2502 2503 return (error); 2504} 2505 2506int 2507fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2508{ 2509 2510 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2511} 2512 2513int 2514fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2515 struct filecaps *havecaps, struct vnode **vpp) 2516{ 2517 struct filedesc *fdp; 2518 struct file *fp; 2519#ifdef CAPABILITIES 2520 int error; 2521#endif 2522 2523 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2524 return (EBADF); 2525 2526 fp = fget_locked(fdp, fd); 2527 if (fp == NULL || fp->f_ops == &badfileops) 2528 return (EBADF); 2529 2530#ifdef CAPABILITIES 2531 if (needrightsp != NULL) { 2532 error = cap_check(cap_rights(fdp, fd), needrightsp); 2533 if (error != 0) 2534 return (error); 2535 } 2536#endif 2537 2538 if (fp->f_vnode == NULL) 2539 return (EINVAL); 2540 2541 *vpp = fp->f_vnode; 2542 vref(*vpp); 2543 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2544 2545 return (0); 2546} 2547 2548int 2549fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2550{ 2551 2552 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2553} 2554 2555int 2556fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2557{ 2558 2559 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2560} 2561 2562#ifdef notyet 2563int 2564fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2565 struct vnode **vpp) 2566{ 2567 2568 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2569} 2570#endif 2571 2572/* 2573 * Like fget() but loads the underlying socket, or returns an error if the 2574 * descriptor does not represent a socket. 2575 * 2576 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2577 * in the future. 2578 * 2579 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2580 * on their file descriptor reference to prevent the socket from being free'd 2581 * during use. 2582 */ 2583int 2584fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2585 u_int *fflagp) 2586{ 2587 struct file *fp; 2588 int error; 2589 2590 *spp = NULL; 2591 if (fflagp != NULL) 2592 *fflagp = 0; 2593 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2594 return (error); 2595 if (fp->f_type != DTYPE_SOCKET) { 2596 error = ENOTSOCK; 2597 } else { 2598 *spp = fp->f_data; 2599 if (fflagp) 2600 *fflagp = fp->f_flag; 2601 SOCK_LOCK(*spp); 2602 soref(*spp); 2603 SOCK_UNLOCK(*spp); 2604 } 2605 fdrop(fp, td); 2606 2607 return (error); 2608} 2609 2610/* 2611 * Drop the reference count on the socket and XXX release the SX lock in the 2612 * future. The last reference closes the socket. 2613 * 2614 * Note: fputsock() is deprecated, see comment for fgetsock(). 2615 */ 2616void 2617fputsock(struct socket *so) 2618{ 2619 2620 ACCEPT_LOCK(); 2621 SOCK_LOCK(so); 2622 CURVNET_SET(so->so_vnet); 2623 sorele(so); 2624 CURVNET_RESTORE(); 2625} 2626 2627/* 2628 * Handle the last reference to a file being closed. 2629 */ 2630int 2631_fdrop(struct file *fp, struct thread *td) 2632{ 2633 int error; 2634 2635 error = 0; 2636 if (fp->f_count != 0) 2637 panic("fdrop: count %d", fp->f_count); 2638 if (fp->f_ops != &badfileops) 2639 error = fo_close(fp, td); 2640 atomic_subtract_int(&openfiles, 1); 2641 crfree(fp->f_cred); 2642 free(fp->f_advice, M_FADVISE); 2643 uma_zfree(file_zone, fp); 2644 2645 return (error); 2646} 2647 2648/* 2649 * Apply an advisory lock on a file descriptor. 2650 * 2651 * Just attempt to get a record lock of the requested type on the entire file 2652 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2653 */ 2654#ifndef _SYS_SYSPROTO_H_ 2655struct flock_args { 2656 int fd; 2657 int how; 2658}; 2659#endif 2660/* ARGSUSED */ 2661int 2662sys_flock(struct thread *td, struct flock_args *uap) 2663{ 2664 struct file *fp; 2665 struct vnode *vp; 2666 struct flock lf; 2667 cap_rights_t rights; 2668 int error; 2669 2670 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2671 if (error != 0) 2672 return (error); 2673 if (fp->f_type != DTYPE_VNODE) { 2674 fdrop(fp, td); 2675 return (EOPNOTSUPP); 2676 } 2677 2678 vp = fp->f_vnode; 2679 lf.l_whence = SEEK_SET; 2680 lf.l_start = 0; 2681 lf.l_len = 0; 2682 if (uap->how & LOCK_UN) { 2683 lf.l_type = F_UNLCK; 2684 atomic_clear_int(&fp->f_flag, FHASLOCK); 2685 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2686 goto done2; 2687 } 2688 if (uap->how & LOCK_EX) 2689 lf.l_type = F_WRLCK; 2690 else if (uap->how & LOCK_SH) 2691 lf.l_type = F_RDLCK; 2692 else { 2693 error = EBADF; 2694 goto done2; 2695 } 2696 atomic_set_int(&fp->f_flag, FHASLOCK); 2697 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2698 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2699done2: 2700 fdrop(fp, td); 2701 return (error); 2702} 2703/* 2704 * Duplicate the specified descriptor to a free descriptor. 2705 */ 2706int 2707dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2708 int openerror, int *indxp) 2709{ 2710 struct file *fp; 2711 int error, indx; 2712 2713 KASSERT(openerror == ENODEV || openerror == ENXIO, 2714 ("unexpected error %d in %s", openerror, __func__)); 2715 2716 /* 2717 * If the to-be-dup'd fd number is greater than the allowed number 2718 * of file descriptors, or the fd to be dup'd has already been 2719 * closed, then reject. 2720 */ 2721 FILEDESC_XLOCK(fdp); 2722 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2723 FILEDESC_XUNLOCK(fdp); 2724 return (EBADF); 2725 } 2726 2727 error = fdalloc(td, 0, &indx); 2728 if (error != 0) { 2729 FILEDESC_XUNLOCK(fdp); 2730 return (error); 2731 } 2732 2733 /* 2734 * There are two cases of interest here. 2735 * 2736 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2737 * 2738 * For ENXIO steal away the file structure from (dfd) and store it in 2739 * (indx). (dfd) is effectively closed by this operation. 2740 */ 2741 switch (openerror) { 2742 case ENODEV: 2743 /* 2744 * Check that the mode the file is being opened for is a 2745 * subset of the mode of the existing descriptor. 2746 */ 2747 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2748 fdunused(fdp, indx); 2749 FILEDESC_XUNLOCK(fdp); 2750 return (EACCES); 2751 } 2752 fhold(fp); 2753 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2754 filecaps_copy(&fdp->fd_ofiles[dfd].fde_caps, 2755 &fdp->fd_ofiles[indx].fde_caps); 2756 break; 2757 case ENXIO: 2758 /* 2759 * Steal away the file pointer from dfd and stuff it into indx. 2760 */ 2761 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2762 bzero(&fdp->fd_ofiles[dfd], sizeof(fdp->fd_ofiles[dfd])); 2763 fdunused(fdp, dfd); 2764 break; 2765 } 2766 FILEDESC_XUNLOCK(fdp); 2767 *indxp = indx; 2768 return (0); 2769} 2770 2771/* 2772 * Scan all active processes and prisons to see if any of them have a current 2773 * or root directory of `olddp'. If so, replace them with the new mount point. 2774 */ 2775void 2776mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2777{ 2778 struct filedesc *fdp; 2779 struct prison *pr; 2780 struct proc *p; 2781 int nrele; 2782 2783 if (vrefcnt(olddp) == 1) 2784 return; 2785 nrele = 0; 2786 sx_slock(&allproc_lock); 2787 FOREACH_PROC_IN_SYSTEM(p) { 2788 fdp = fdhold(p); 2789 if (fdp == NULL) 2790 continue; 2791 FILEDESC_XLOCK(fdp); 2792 if (fdp->fd_cdir == olddp) { 2793 vref(newdp); 2794 fdp->fd_cdir = newdp; 2795 nrele++; 2796 } 2797 if (fdp->fd_rdir == olddp) { 2798 vref(newdp); 2799 fdp->fd_rdir = newdp; 2800 nrele++; 2801 } 2802 if (fdp->fd_jdir == olddp) { 2803 vref(newdp); 2804 fdp->fd_jdir = newdp; 2805 nrele++; 2806 } 2807 FILEDESC_XUNLOCK(fdp); 2808 fddrop(fdp); 2809 } 2810 sx_sunlock(&allproc_lock); 2811 if (rootvnode == olddp) { 2812 vref(newdp); 2813 rootvnode = newdp; 2814 nrele++; 2815 } 2816 mtx_lock(&prison0.pr_mtx); 2817 if (prison0.pr_root == olddp) { 2818 vref(newdp); 2819 prison0.pr_root = newdp; 2820 nrele++; 2821 } 2822 mtx_unlock(&prison0.pr_mtx); 2823 sx_slock(&allprison_lock); 2824 TAILQ_FOREACH(pr, &allprison, pr_list) { 2825 mtx_lock(&pr->pr_mtx); 2826 if (pr->pr_root == olddp) { 2827 vref(newdp); 2828 pr->pr_root = newdp; 2829 nrele++; 2830 } 2831 mtx_unlock(&pr->pr_mtx); 2832 } 2833 sx_sunlock(&allprison_lock); 2834 while (nrele--) 2835 vrele(olddp); 2836} 2837 2838struct filedesc_to_leader * 2839filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2840{ 2841 struct filedesc_to_leader *fdtol; 2842 2843 fdtol = malloc(sizeof(struct filedesc_to_leader), 2844 M_FILEDESC_TO_LEADER, 2845 M_WAITOK); 2846 fdtol->fdl_refcount = 1; 2847 fdtol->fdl_holdcount = 0; 2848 fdtol->fdl_wakeup = 0; 2849 fdtol->fdl_leader = leader; 2850 if (old != NULL) { 2851 FILEDESC_XLOCK(fdp); 2852 fdtol->fdl_next = old->fdl_next; 2853 fdtol->fdl_prev = old; 2854 old->fdl_next = fdtol; 2855 fdtol->fdl_next->fdl_prev = fdtol; 2856 FILEDESC_XUNLOCK(fdp); 2857 } else { 2858 fdtol->fdl_next = fdtol; 2859 fdtol->fdl_prev = fdtol; 2860 } 2861 return (fdtol); 2862} 2863 2864/* 2865 * Get file structures globally. 2866 */ 2867static int 2868sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2869{ 2870 struct xfile xf; 2871 struct filedesc *fdp; 2872 struct file *fp; 2873 struct proc *p; 2874 int error, n; 2875 2876 error = sysctl_wire_old_buffer(req, 0); 2877 if (error != 0) 2878 return (error); 2879 if (req->oldptr == NULL) { 2880 n = 0; 2881 sx_slock(&allproc_lock); 2882 FOREACH_PROC_IN_SYSTEM(p) { 2883 if (p->p_state == PRS_NEW) 2884 continue; 2885 fdp = fdhold(p); 2886 if (fdp == NULL) 2887 continue; 2888 /* overestimates sparse tables. */ 2889 if (fdp->fd_lastfile > 0) 2890 n += fdp->fd_lastfile; 2891 fddrop(fdp); 2892 } 2893 sx_sunlock(&allproc_lock); 2894 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2895 } 2896 error = 0; 2897 bzero(&xf, sizeof(xf)); 2898 xf.xf_size = sizeof(xf); 2899 sx_slock(&allproc_lock); 2900 FOREACH_PROC_IN_SYSTEM(p) { 2901 PROC_LOCK(p); 2902 if (p->p_state == PRS_NEW) { 2903 PROC_UNLOCK(p); 2904 continue; 2905 } 2906 if (p_cansee(req->td, p) != 0) { 2907 PROC_UNLOCK(p); 2908 continue; 2909 } 2910 xf.xf_pid = p->p_pid; 2911 xf.xf_uid = p->p_ucred->cr_uid; 2912 PROC_UNLOCK(p); 2913 fdp = fdhold(p); 2914 if (fdp == NULL) 2915 continue; 2916 FILEDESC_SLOCK(fdp); 2917 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 2918 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 2919 continue; 2920 xf.xf_fd = n; 2921 xf.xf_file = fp; 2922 xf.xf_data = fp->f_data; 2923 xf.xf_vnode = fp->f_vnode; 2924 xf.xf_type = fp->f_type; 2925 xf.xf_count = fp->f_count; 2926 xf.xf_msgcount = 0; 2927 xf.xf_offset = foffset_get(fp); 2928 xf.xf_flag = fp->f_flag; 2929 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2930 if (error) 2931 break; 2932 } 2933 FILEDESC_SUNLOCK(fdp); 2934 fddrop(fdp); 2935 if (error) 2936 break; 2937 } 2938 sx_sunlock(&allproc_lock); 2939 return (error); 2940} 2941 2942SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 2943 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2944 2945#ifdef KINFO_OFILE_SIZE 2946CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2947#endif 2948 2949#ifdef COMPAT_FREEBSD7 2950static int 2951export_vnode_for_osysctl(struct vnode *vp, int type, 2952 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2953{ 2954 int error; 2955 char *fullpath, *freepath; 2956 2957 bzero(kif, sizeof(*kif)); 2958 kif->kf_structsize = sizeof(*kif); 2959 2960 vref(vp); 2961 kif->kf_fd = type; 2962 kif->kf_type = KF_TYPE_VNODE; 2963 /* This function only handles directories. */ 2964 if (vp->v_type != VDIR) { 2965 vrele(vp); 2966 return (ENOTDIR); 2967 } 2968 kif->kf_vnode_type = KF_VTYPE_VDIR; 2969 2970 /* 2971 * This is not a true file descriptor, so we set a bogus refcount 2972 * and offset to indicate these fields should be ignored. 2973 */ 2974 kif->kf_ref_count = -1; 2975 kif->kf_offset = -1; 2976 2977 freepath = NULL; 2978 fullpath = "-"; 2979 FILEDESC_SUNLOCK(fdp); 2980 vn_fullpath(curthread, vp, &fullpath, &freepath); 2981 vrele(vp); 2982 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2983 if (freepath != NULL) 2984 free(freepath, M_TEMP); 2985 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2986 FILEDESC_SLOCK(fdp); 2987 return (error); 2988} 2989 2990/* 2991 * Get per-process file descriptors for use by procstat(1), et al. 2992 */ 2993static int 2994sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 2995{ 2996 char *fullpath, *freepath; 2997 struct kinfo_ofile *kif; 2998 struct filedesc *fdp; 2999 int error, i, *name; 3000 struct shmfd *shmfd; 3001 struct socket *so; 3002 struct vnode *vp; 3003 struct ksem *ks; 3004 struct file *fp; 3005 struct proc *p; 3006 struct tty *tp; 3007 3008 name = (int *)arg1; 3009 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3010 if (error != 0) 3011 return (error); 3012 fdp = fdhold(p); 3013 PROC_UNLOCK(p); 3014 if (fdp == NULL) 3015 return (ENOENT); 3016 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3017 FILEDESC_SLOCK(fdp); 3018 if (fdp->fd_cdir != NULL) 3019 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3020 fdp, req); 3021 if (fdp->fd_rdir != NULL) 3022 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3023 fdp, req); 3024 if (fdp->fd_jdir != NULL) 3025 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3026 fdp, req); 3027 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3028 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3029 continue; 3030 bzero(kif, sizeof(*kif)); 3031 kif->kf_structsize = sizeof(*kif); 3032 ks = NULL; 3033 vp = NULL; 3034 so = NULL; 3035 tp = NULL; 3036 shmfd = NULL; 3037 kif->kf_fd = i; 3038 3039 switch (fp->f_type) { 3040 case DTYPE_VNODE: 3041 kif->kf_type = KF_TYPE_VNODE; 3042 vp = fp->f_vnode; 3043 break; 3044 3045 case DTYPE_SOCKET: 3046 kif->kf_type = KF_TYPE_SOCKET; 3047 so = fp->f_data; 3048 break; 3049 3050 case DTYPE_PIPE: 3051 kif->kf_type = KF_TYPE_PIPE; 3052 break; 3053 3054 case DTYPE_FIFO: 3055 kif->kf_type = KF_TYPE_FIFO; 3056 vp = fp->f_vnode; 3057 break; 3058 3059 case DTYPE_KQUEUE: 3060 kif->kf_type = KF_TYPE_KQUEUE; 3061 break; 3062 3063 case DTYPE_CRYPTO: 3064 kif->kf_type = KF_TYPE_CRYPTO; 3065 break; 3066 3067 case DTYPE_MQUEUE: 3068 kif->kf_type = KF_TYPE_MQUEUE; 3069 break; 3070 3071 case DTYPE_SHM: 3072 kif->kf_type = KF_TYPE_SHM; 3073 shmfd = fp->f_data; 3074 break; 3075 3076 case DTYPE_SEM: 3077 kif->kf_type = KF_TYPE_SEM; 3078 ks = fp->f_data; 3079 break; 3080 3081 case DTYPE_PTS: 3082 kif->kf_type = KF_TYPE_PTS; 3083 tp = fp->f_data; 3084 break; 3085 3086 case DTYPE_PROCDESC: 3087 kif->kf_type = KF_TYPE_PROCDESC; 3088 break; 3089 3090 default: 3091 kif->kf_type = KF_TYPE_UNKNOWN; 3092 break; 3093 } 3094 kif->kf_ref_count = fp->f_count; 3095 if (fp->f_flag & FREAD) 3096 kif->kf_flags |= KF_FLAG_READ; 3097 if (fp->f_flag & FWRITE) 3098 kif->kf_flags |= KF_FLAG_WRITE; 3099 if (fp->f_flag & FAPPEND) 3100 kif->kf_flags |= KF_FLAG_APPEND; 3101 if (fp->f_flag & FASYNC) 3102 kif->kf_flags |= KF_FLAG_ASYNC; 3103 if (fp->f_flag & FFSYNC) 3104 kif->kf_flags |= KF_FLAG_FSYNC; 3105 if (fp->f_flag & FNONBLOCK) 3106 kif->kf_flags |= KF_FLAG_NONBLOCK; 3107 if (fp->f_flag & O_DIRECT) 3108 kif->kf_flags |= KF_FLAG_DIRECT; 3109 if (fp->f_flag & FHASLOCK) 3110 kif->kf_flags |= KF_FLAG_HASLOCK; 3111 kif->kf_offset = foffset_get(fp); 3112 if (vp != NULL) { 3113 vref(vp); 3114 switch (vp->v_type) { 3115 case VNON: 3116 kif->kf_vnode_type = KF_VTYPE_VNON; 3117 break; 3118 case VREG: 3119 kif->kf_vnode_type = KF_VTYPE_VREG; 3120 break; 3121 case VDIR: 3122 kif->kf_vnode_type = KF_VTYPE_VDIR; 3123 break; 3124 case VBLK: 3125 kif->kf_vnode_type = KF_VTYPE_VBLK; 3126 break; 3127 case VCHR: 3128 kif->kf_vnode_type = KF_VTYPE_VCHR; 3129 break; 3130 case VLNK: 3131 kif->kf_vnode_type = KF_VTYPE_VLNK; 3132 break; 3133 case VSOCK: 3134 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3135 break; 3136 case VFIFO: 3137 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3138 break; 3139 case VBAD: 3140 kif->kf_vnode_type = KF_VTYPE_VBAD; 3141 break; 3142 default: 3143 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3144 break; 3145 } 3146 /* 3147 * It is OK to drop the filedesc lock here as we will 3148 * re-validate and re-evaluate its properties when 3149 * the loop continues. 3150 */ 3151 freepath = NULL; 3152 fullpath = "-"; 3153 FILEDESC_SUNLOCK(fdp); 3154 vn_fullpath(curthread, vp, &fullpath, &freepath); 3155 vrele(vp); 3156 strlcpy(kif->kf_path, fullpath, 3157 sizeof(kif->kf_path)); 3158 if (freepath != NULL) 3159 free(freepath, M_TEMP); 3160 FILEDESC_SLOCK(fdp); 3161 } 3162 if (so != NULL) { 3163 struct sockaddr *sa; 3164 3165 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3166 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3167 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3168 free(sa, M_SONAME); 3169 } 3170 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3171 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3172 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3173 free(sa, M_SONAME); 3174 } 3175 kif->kf_sock_domain = 3176 so->so_proto->pr_domain->dom_family; 3177 kif->kf_sock_type = so->so_type; 3178 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3179 } 3180 if (tp != NULL) { 3181 strlcpy(kif->kf_path, tty_devname(tp), 3182 sizeof(kif->kf_path)); 3183 } 3184 if (shmfd != NULL) 3185 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3186 if (ks != NULL && ksem_info != NULL) 3187 ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL); 3188 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3189 if (error) 3190 break; 3191 } 3192 FILEDESC_SUNLOCK(fdp); 3193 fddrop(fdp); 3194 free(kif, M_TEMP); 3195 return (0); 3196} 3197 3198static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3199 CTLFLAG_RD||CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3200 "Process ofiledesc entries"); 3201#endif /* COMPAT_FREEBSD7 */ 3202 3203#ifdef KINFO_FILE_SIZE 3204CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3205#endif 3206 3207struct export_fd_buf { 3208 struct filedesc *fdp; 3209 struct sbuf *sb; 3210 ssize_t remainder; 3211 struct kinfo_file kif; 3212}; 3213 3214static int 3215export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt, 3216 int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf) 3217{ 3218 struct { 3219 int fflag; 3220 int kf_fflag; 3221 } fflags_table[] = { 3222 { FAPPEND, KF_FLAG_APPEND }, 3223 { FASYNC, KF_FLAG_ASYNC }, 3224 { FFSYNC, KF_FLAG_FSYNC }, 3225 { FHASLOCK, KF_FLAG_HASLOCK }, 3226 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3227 { FREAD, KF_FLAG_READ }, 3228 { FWRITE, KF_FLAG_WRITE }, 3229 { O_CREAT, KF_FLAG_CREAT }, 3230 { O_DIRECT, KF_FLAG_DIRECT }, 3231 { O_EXCL, KF_FLAG_EXCL }, 3232 { O_EXEC, KF_FLAG_EXEC }, 3233 { O_EXLOCK, KF_FLAG_EXLOCK }, 3234 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3235 { O_SHLOCK, KF_FLAG_SHLOCK }, 3236 { O_TRUNC, KF_FLAG_TRUNC } 3237 }; 3238#define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3239 struct kinfo_file *kif; 3240 struct vnode *vp; 3241 int error, locked; 3242 unsigned int i; 3243 3244 if (efbuf->remainder == 0) 3245 return (0); 3246 kif = &efbuf->kif; 3247 bzero(kif, sizeof(*kif)); 3248 locked = efbuf->fdp != NULL; 3249 switch (type) { 3250 case KF_TYPE_FIFO: 3251 case KF_TYPE_VNODE: 3252 if (locked) { 3253 FILEDESC_SUNLOCK(efbuf->fdp); 3254 locked = 0; 3255 } 3256 vp = (struct vnode *)data; 3257 error = fill_vnode_info(vp, kif); 3258 vrele(vp); 3259 break; 3260 case KF_TYPE_SOCKET: 3261 error = fill_socket_info((struct socket *)data, kif); 3262 break; 3263 case KF_TYPE_PIPE: 3264 error = fill_pipe_info((struct pipe *)data, kif); 3265 break; 3266 case KF_TYPE_PTS: 3267 error = fill_pts_info((struct tty *)data, kif); 3268 break; 3269 case KF_TYPE_PROCDESC: 3270 error = fill_procdesc_info((struct procdesc *)data, kif); 3271 break; 3272 case KF_TYPE_SEM: 3273 error = fill_sem_info((struct file *)data, kif); 3274 break; 3275 case KF_TYPE_SHM: 3276 error = fill_shm_info((struct file *)data, kif); 3277 break; 3278 default: 3279 error = 0; 3280 } 3281 if (error == 0) 3282 kif->kf_status |= KF_ATTR_VALID; 3283 3284 /* 3285 * Translate file access flags. 3286 */ 3287 for (i = 0; i < NFFLAGS; i++) 3288 if (fflags & fflags_table[i].fflag) 3289 kif->kf_flags |= fflags_table[i].kf_fflag; 3290 if (rightsp != NULL) 3291 kif->kf_cap_rights = *rightsp; 3292 else 3293 cap_rights_init(&kif->kf_cap_rights); 3294 kif->kf_fd = fd; 3295 kif->kf_type = type; 3296 kif->kf_ref_count = refcnt; 3297 kif->kf_offset = offset; 3298 /* Pack record size down */ 3299 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3300 strlen(kif->kf_path) + 1; 3301 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3302 if (efbuf->remainder != -1) { 3303 if (efbuf->remainder < kif->kf_structsize) { 3304 /* Terminate export. */ 3305 efbuf->remainder = 0; 3306 if (efbuf->fdp != NULL && !locked) 3307 FILEDESC_SLOCK(efbuf->fdp); 3308 return (0); 3309 } 3310 efbuf->remainder -= kif->kf_structsize; 3311 } 3312 if (locked) 3313 FILEDESC_SUNLOCK(efbuf->fdp); 3314 error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize); 3315 if (efbuf->fdp != NULL) 3316 FILEDESC_SLOCK(efbuf->fdp); 3317 return (error); 3318} 3319 3320/* 3321 * Store a process file descriptor information to sbuf. 3322 * 3323 * Takes a locked proc as argument, and returns with the proc unlocked. 3324 */ 3325int 3326kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3327{ 3328 struct file *fp; 3329 struct filedesc *fdp; 3330 struct export_fd_buf *efbuf; 3331 struct vnode *cttyvp, *textvp, *tracevp; 3332 int64_t offset; 3333 void *data; 3334 int error, i; 3335 int type, refcnt, fflags; 3336 cap_rights_t rights; 3337 3338 PROC_LOCK_ASSERT(p, MA_OWNED); 3339 3340 /* ktrace vnode */ 3341 tracevp = p->p_tracevp; 3342 if (tracevp != NULL) 3343 vref(tracevp); 3344 /* text vnode */ 3345 textvp = p->p_textvp; 3346 if (textvp != NULL) 3347 vref(textvp); 3348 /* Controlling tty. */ 3349 cttyvp = NULL; 3350 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3351 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3352 if (cttyvp != NULL) 3353 vref(cttyvp); 3354 } 3355 fdp = fdhold(p); 3356 PROC_UNLOCK(p); 3357 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3358 efbuf->fdp = NULL; 3359 efbuf->sb = sb; 3360 efbuf->remainder = maxlen; 3361 if (tracevp != NULL) 3362 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3363 FREAD | FWRITE, -1, -1, NULL, efbuf); 3364 if (textvp != NULL) 3365 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3366 FREAD, -1, -1, NULL, efbuf); 3367 if (cttyvp != NULL) 3368 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3369 FREAD | FWRITE, -1, -1, NULL, efbuf); 3370 error = 0; 3371 if (fdp == NULL) 3372 goto fail; 3373 efbuf->fdp = fdp; 3374 FILEDESC_SLOCK(fdp); 3375 /* working directory */ 3376 if (fdp->fd_cdir != NULL) { 3377 vref(fdp->fd_cdir); 3378 data = fdp->fd_cdir; 3379 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3380 FREAD, -1, -1, NULL, efbuf); 3381 } 3382 /* root directory */ 3383 if (fdp->fd_rdir != NULL) { 3384 vref(fdp->fd_rdir); 3385 data = fdp->fd_rdir; 3386 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3387 FREAD, -1, -1, NULL, efbuf); 3388 } 3389 /* jail directory */ 3390 if (fdp->fd_jdir != NULL) { 3391 vref(fdp->fd_jdir); 3392 data = fdp->fd_jdir; 3393 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3394 FREAD, -1, -1, NULL, efbuf); 3395 } 3396 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3397 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3398 continue; 3399 data = NULL; 3400#ifdef CAPABILITIES 3401 rights = *cap_rights(fdp, i); 3402#else /* !CAPABILITIES */ 3403 cap_rights_init(&rights); 3404#endif 3405 switch (fp->f_type) { 3406 case DTYPE_VNODE: 3407 type = KF_TYPE_VNODE; 3408 vref(fp->f_vnode); 3409 data = fp->f_vnode; 3410 break; 3411 3412 case DTYPE_SOCKET: 3413 type = KF_TYPE_SOCKET; 3414 data = fp->f_data; 3415 break; 3416 3417 case DTYPE_PIPE: 3418 type = KF_TYPE_PIPE; 3419 data = fp->f_data; 3420 break; 3421 3422 case DTYPE_FIFO: 3423 type = KF_TYPE_FIFO; 3424 vref(fp->f_vnode); 3425 data = fp->f_vnode; 3426 break; 3427 3428 case DTYPE_KQUEUE: 3429 type = KF_TYPE_KQUEUE; 3430 break; 3431 3432 case DTYPE_CRYPTO: 3433 type = KF_TYPE_CRYPTO; 3434 break; 3435 3436 case DTYPE_MQUEUE: 3437 type = KF_TYPE_MQUEUE; 3438 break; 3439 3440 case DTYPE_SHM: 3441 type = KF_TYPE_SHM; 3442 data = fp; 3443 break; 3444 3445 case DTYPE_SEM: 3446 type = KF_TYPE_SEM; 3447 data = fp; 3448 break; 3449 3450 case DTYPE_PTS: 3451 type = KF_TYPE_PTS; 3452 data = fp->f_data; 3453 break; 3454 3455 case DTYPE_PROCDESC: 3456 type = KF_TYPE_PROCDESC; 3457 data = fp->f_data; 3458 break; 3459 3460 default: 3461 type = KF_TYPE_UNKNOWN; 3462 break; 3463 } 3464 refcnt = fp->f_count; 3465 fflags = fp->f_flag; 3466 offset = foffset_get(fp); 3467 3468 /* 3469 * Create sysctl entry. 3470 * It is OK to drop the filedesc lock here as we will 3471 * re-validate and re-evaluate its properties when 3472 * the loop continues. 3473 */ 3474 error = export_fd_to_sb(data, type, i, fflags, refcnt, 3475 offset, &rights, efbuf); 3476 if (error != 0) 3477 break; 3478 } 3479 FILEDESC_SUNLOCK(fdp); 3480 fddrop(fdp); 3481fail: 3482 free(efbuf, M_TEMP); 3483 return (error); 3484} 3485 3486#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3487 3488/* 3489 * Get per-process file descriptors for use by procstat(1), et al. 3490 */ 3491static int 3492sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3493{ 3494 struct sbuf sb; 3495 struct proc *p; 3496 ssize_t maxlen; 3497 int error, error2, *name; 3498 3499 name = (int *)arg1; 3500 3501 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3502 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3503 if (error != 0) { 3504 sbuf_delete(&sb); 3505 return (error); 3506 } 3507 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3508 error = kern_proc_filedesc_out(p, &sb, maxlen); 3509 error2 = sbuf_finish(&sb); 3510 sbuf_delete(&sb); 3511 return (error != 0 ? error : error2); 3512} 3513 3514int 3515vntype_to_kinfo(int vtype) 3516{ 3517 struct { 3518 int vtype; 3519 int kf_vtype; 3520 } vtypes_table[] = { 3521 { VBAD, KF_VTYPE_VBAD }, 3522 { VBLK, KF_VTYPE_VBLK }, 3523 { VCHR, KF_VTYPE_VCHR }, 3524 { VDIR, KF_VTYPE_VDIR }, 3525 { VFIFO, KF_VTYPE_VFIFO }, 3526 { VLNK, KF_VTYPE_VLNK }, 3527 { VNON, KF_VTYPE_VNON }, 3528 { VREG, KF_VTYPE_VREG }, 3529 { VSOCK, KF_VTYPE_VSOCK } 3530 }; 3531#define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3532 unsigned int i; 3533 3534 /* 3535 * Perform vtype translation. 3536 */ 3537 for (i = 0; i < NVTYPES; i++) 3538 if (vtypes_table[i].vtype == vtype) 3539 break; 3540 if (i < NVTYPES) 3541 return (vtypes_table[i].kf_vtype); 3542 3543 return (KF_VTYPE_UNKNOWN); 3544} 3545 3546static int 3547fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3548{ 3549 struct vattr va; 3550 char *fullpath, *freepath; 3551 int error; 3552 3553 if (vp == NULL) 3554 return (1); 3555 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3556 freepath = NULL; 3557 fullpath = "-"; 3558 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3559 if (error == 0) { 3560 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3561 } 3562 if (freepath != NULL) 3563 free(freepath, M_TEMP); 3564 3565 /* 3566 * Retrieve vnode attributes. 3567 */ 3568 va.va_fsid = VNOVAL; 3569 va.va_rdev = NODEV; 3570 vn_lock(vp, LK_SHARED | LK_RETRY); 3571 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3572 VOP_UNLOCK(vp, 0); 3573 if (error != 0) 3574 return (error); 3575 if (va.va_fsid != VNOVAL) 3576 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3577 else 3578 kif->kf_un.kf_file.kf_file_fsid = 3579 vp->v_mount->mnt_stat.f_fsid.val[0]; 3580 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3581 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3582 kif->kf_un.kf_file.kf_file_size = va.va_size; 3583 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3584 return (0); 3585} 3586 3587static int 3588fill_socket_info(struct socket *so, struct kinfo_file *kif) 3589{ 3590 struct sockaddr *sa; 3591 struct inpcb *inpcb; 3592 struct unpcb *unpcb; 3593 int error; 3594 3595 if (so == NULL) 3596 return (1); 3597 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3598 kif->kf_sock_type = so->so_type; 3599 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3600 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3601 switch(kif->kf_sock_domain) { 3602 case AF_INET: 3603 case AF_INET6: 3604 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3605 if (so->so_pcb != NULL) { 3606 inpcb = (struct inpcb *)(so->so_pcb); 3607 kif->kf_un.kf_sock.kf_sock_inpcb = 3608 (uintptr_t)inpcb->inp_ppcb; 3609 } 3610 } 3611 break; 3612 case AF_UNIX: 3613 if (so->so_pcb != NULL) { 3614 unpcb = (struct unpcb *)(so->so_pcb); 3615 if (unpcb->unp_conn) { 3616 kif->kf_un.kf_sock.kf_sock_unpconn = 3617 (uintptr_t)unpcb->unp_conn; 3618 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3619 so->so_rcv.sb_state; 3620 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3621 so->so_snd.sb_state; 3622 } 3623 } 3624 break; 3625 } 3626 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3627 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3628 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3629 free(sa, M_SONAME); 3630 } 3631 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3632 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3633 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3634 free(sa, M_SONAME); 3635 } 3636 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3637 sizeof(kif->kf_path)); 3638 return (0); 3639} 3640 3641static int 3642fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3643{ 3644 3645 if (tp == NULL) 3646 return (1); 3647 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3648 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3649 return (0); 3650} 3651 3652static int 3653fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3654{ 3655 3656 if (pi == NULL) 3657 return (1); 3658 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3659 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3660 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3661 return (0); 3662} 3663 3664static int 3665fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3666{ 3667 3668 if (pdp == NULL) 3669 return (1); 3670 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3671 return (0); 3672} 3673 3674static int 3675fill_sem_info(struct file *fp, struct kinfo_file *kif) 3676{ 3677 struct thread *td; 3678 struct stat sb; 3679 3680 td = curthread; 3681 if (fp->f_data == NULL) 3682 return (1); 3683 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3684 return (1); 3685 if (ksem_info == NULL) 3686 return (1); 3687 ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path), 3688 &kif->kf_un.kf_sem.kf_sem_value); 3689 kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode; 3690 return (0); 3691} 3692 3693static int 3694fill_shm_info(struct file *fp, struct kinfo_file *kif) 3695{ 3696 struct thread *td; 3697 struct stat sb; 3698 3699 td = curthread; 3700 if (fp->f_data == NULL) 3701 return (1); 3702 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3703 return (1); 3704 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3705 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3706 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3707 return (0); 3708} 3709 3710static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3711 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3712 "Process filedesc entries"); 3713 3714#ifdef DDB 3715/* 3716 * For the purposes of debugging, generate a human-readable string for the 3717 * file type. 3718 */ 3719static const char * 3720file_type_to_name(short type) 3721{ 3722 3723 switch (type) { 3724 case 0: 3725 return ("zero"); 3726 case DTYPE_VNODE: 3727 return ("vnod"); 3728 case DTYPE_SOCKET: 3729 return ("sock"); 3730 case DTYPE_PIPE: 3731 return ("pipe"); 3732 case DTYPE_FIFO: 3733 return ("fifo"); 3734 case DTYPE_KQUEUE: 3735 return ("kque"); 3736 case DTYPE_CRYPTO: 3737 return ("crpt"); 3738 case DTYPE_MQUEUE: 3739 return ("mque"); 3740 case DTYPE_SHM: 3741 return ("shm"); 3742 case DTYPE_SEM: 3743 return ("ksem"); 3744 default: 3745 return ("unkn"); 3746 } 3747} 3748 3749/* 3750 * For the purposes of debugging, identify a process (if any, perhaps one of 3751 * many) that references the passed file in its file descriptor array. Return 3752 * NULL if none. 3753 */ 3754static struct proc * 3755file_to_first_proc(struct file *fp) 3756{ 3757 struct filedesc *fdp; 3758 struct proc *p; 3759 int n; 3760 3761 FOREACH_PROC_IN_SYSTEM(p) { 3762 if (p->p_state == PRS_NEW) 3763 continue; 3764 fdp = p->p_fd; 3765 if (fdp == NULL) 3766 continue; 3767 for (n = 0; n <= fdp->fd_lastfile; n++) { 3768 if (fp == fdp->fd_ofiles[n].fde_file) 3769 return (p); 3770 } 3771 } 3772 return (NULL); 3773} 3774 3775static void 3776db_print_file(struct file *fp, int header) 3777{ 3778 struct proc *p; 3779 3780 if (header) 3781 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3782 "File", "Type", "Data", "Flag", "GCFl", "Count", 3783 "MCount", "Vnode", "FPID", "FCmd"); 3784 p = file_to_first_proc(fp); 3785 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3786 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3787 0, fp->f_count, 0, fp->f_vnode, 3788 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3789} 3790 3791DB_SHOW_COMMAND(file, db_show_file) 3792{ 3793 struct file *fp; 3794 3795 if (!have_addr) { 3796 db_printf("usage: show file <addr>\n"); 3797 return; 3798 } 3799 fp = (struct file *)addr; 3800 db_print_file(fp, 1); 3801} 3802 3803DB_SHOW_COMMAND(files, db_show_files) 3804{ 3805 struct filedesc *fdp; 3806 struct file *fp; 3807 struct proc *p; 3808 int header; 3809 int n; 3810 3811 header = 1; 3812 FOREACH_PROC_IN_SYSTEM(p) { 3813 if (p->p_state == PRS_NEW) 3814 continue; 3815 if ((fdp = p->p_fd) == NULL) 3816 continue; 3817 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3818 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3819 continue; 3820 db_print_file(fp, header); 3821 header = 0; 3822 } 3823 } 3824} 3825#endif 3826 3827SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3828 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3829 3830SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3831 &maxfiles, 0, "Maximum number of files"); 3832 3833SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3834 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3835 3836/* ARGSUSED*/ 3837static void 3838filelistinit(void *dummy) 3839{ 3840 3841 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3842 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3843 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3844 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3845} 3846SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3847 3848/*-------------------------------------------------------------------*/ 3849 3850static int 3851badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3852 int flags, struct thread *td) 3853{ 3854 3855 return (EBADF); 3856} 3857 3858static int 3859badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3860 struct thread *td) 3861{ 3862 3863 return (EINVAL); 3864} 3865 3866static int 3867badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3868 struct thread *td) 3869{ 3870 3871 return (EBADF); 3872} 3873 3874static int 3875badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3876 struct thread *td) 3877{ 3878 3879 return (0); 3880} 3881 3882static int 3883badfo_kqfilter(struct file *fp, struct knote *kn) 3884{ 3885 3886 return (EBADF); 3887} 3888 3889static int 3890badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3891 struct thread *td) 3892{ 3893 3894 return (EBADF); 3895} 3896 3897static int 3898badfo_close(struct file *fp, struct thread *td) 3899{ 3900 3901 return (EBADF); 3902} 3903 3904static int 3905badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3906 struct thread *td) 3907{ 3908 3909 return (EBADF); 3910} 3911 3912static int 3913badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3914 struct thread *td) 3915{ 3916 3917 return (EBADF); 3918} 3919 3920static int 3921badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3922 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3923 int kflags, struct sendfile_sync *sfs, struct thread *td) 3924{ 3925 3926 return (EBADF); 3927} 3928 3929struct fileops badfileops = { 3930 .fo_read = badfo_readwrite, 3931 .fo_write = badfo_readwrite, 3932 .fo_truncate = badfo_truncate, 3933 .fo_ioctl = badfo_ioctl, 3934 .fo_poll = badfo_poll, 3935 .fo_kqfilter = badfo_kqfilter, 3936 .fo_stat = badfo_stat, 3937 .fo_close = badfo_close, 3938 .fo_chmod = badfo_chmod, 3939 .fo_chown = badfo_chown, 3940 .fo_sendfile = badfo_sendfile, 3941}; 3942 3943int 3944invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3945 struct thread *td) 3946{ 3947 3948 return (EINVAL); 3949} 3950 3951int 3952invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3953 struct thread *td) 3954{ 3955 3956 return (EINVAL); 3957} 3958 3959int 3960invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3961 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3962 int kflags, struct sendfile_sync *sfs, struct thread *td) 3963{ 3964 3965 return (EINVAL); 3966} 3967 3968/*-------------------------------------------------------------------*/ 3969 3970/* 3971 * File Descriptor pseudo-device driver (/dev/fd/). 3972 * 3973 * Opening minor device N dup()s the file (if any) connected to file 3974 * descriptor N belonging to the calling process. Note that this driver 3975 * consists of only the ``open()'' routine, because all subsequent 3976 * references to this file will be direct to the other driver. 3977 * 3978 * XXX: we could give this one a cloning event handler if necessary. 3979 */ 3980 3981/* ARGSUSED */ 3982static int 3983fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3984{ 3985 3986 /* 3987 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3988 * the file descriptor being sought for duplication. The error 3989 * return ensures that the vnode for this device will be released 3990 * by vn_open. Open will detect this special error and take the 3991 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3992 * will simply report the error. 3993 */ 3994 td->td_dupfd = dev2unit(dev); 3995 return (ENODEV); 3996} 3997 3998static struct cdevsw fildesc_cdevsw = { 3999 .d_version = D_VERSION, 4000 .d_open = fdopen, 4001 .d_name = "FD", 4002}; 4003 4004static void 4005fildesc_drvinit(void *unused) 4006{ 4007 struct cdev *dev; 4008 4009 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4010 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4011 make_dev_alias(dev, "stdin"); 4012 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4013 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4014 make_dev_alias(dev, "stdout"); 4015 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4016 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4017 make_dev_alias(dev, "stderr"); 4018} 4019 4020SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4021