kern_descrip.c revision 273893
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 273893 2014-10-31 09:15:59Z 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/fcntl.h> 51#include <sys/file.h> 52#include <sys/filedesc.h> 53#include <sys/filio.h> 54#include <sys/jail.h> 55#include <sys/kernel.h> 56#include <sys/limits.h> 57#include <sys/lock.h> 58#include <sys/malloc.h> 59#include <sys/mount.h> 60#include <sys/mutex.h> 61#include <sys/namei.h> 62#include <sys/selinfo.h> 63#include <sys/priv.h> 64#include <sys/proc.h> 65#include <sys/protosw.h> 66#include <sys/racct.h> 67#include <sys/resourcevar.h> 68#include <sys/sbuf.h> 69#include <sys/signalvar.h> 70#include <sys/socketvar.h> 71#include <sys/stat.h> 72#include <sys/sx.h> 73#include <sys/syscallsubr.h> 74#include <sys/sysctl.h> 75#include <sys/sysproto.h> 76#include <sys/unistd.h> 77#include <sys/user.h> 78#include <sys/vnode.h> 79#ifdef KTRACE 80#include <sys/ktrace.h> 81#endif 82 83#include <net/vnet.h> 84 85#include <security/audit/audit.h> 86 87#include <vm/uma.h> 88#include <vm/vm.h> 89 90#include <ddb/ddb.h> 91 92static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 93static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 94 "file desc to leader structures"); 95static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 96MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 97 98MALLOC_DECLARE(M_FADVISE); 99 100static uma_zone_t file_zone; 101 102static int closefp(struct filedesc *fdp, int fd, struct file *fp, 103 struct thread *td, int holdleaders); 104static int do_dup(struct thread *td, int flags, int old, int new, 105 register_t *retval); 106static int fd_first_free(struct filedesc *fdp, int low, int size); 107static int fd_last_used(struct filedesc *fdp, int size); 108static void fdgrowtable(struct filedesc *fdp, int nfd); 109static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 110static void fdunused(struct filedesc *fdp, int fd); 111static void fdused(struct filedesc *fdp, int fd); 112static int getmaxfd(struct proc *p); 113 114/* Flags for do_dup() */ 115#define DUP_FIXED 0x1 /* Force fixed allocation. */ 116#define DUP_FCNTL 0x2 /* fcntl()-style errors. */ 117#define DUP_CLOEXEC 0x4 /* Atomically set FD_CLOEXEC. */ 118 119/* 120 * Each process has: 121 * 122 * - An array of open file descriptors (fd_ofiles) 123 * - An array of file flags (fd_ofileflags) 124 * - A bitmap recording which descriptors are in use (fd_map) 125 * 126 * A process starts out with NDFILE descriptors. The value of NDFILE has 127 * been selected based the historical limit of 20 open files, and an 128 * assumption that the majority of processes, especially short-lived 129 * processes like shells, will never need more. 130 * 131 * If this initial allocation is exhausted, a larger descriptor table and 132 * map are allocated dynamically, and the pointers in the process's struct 133 * filedesc are updated to point to those. This is repeated every time 134 * the process runs out of file descriptors (provided it hasn't hit its 135 * resource limit). 136 * 137 * Since threads may hold references to individual descriptor table 138 * entries, the tables are never freed. Instead, they are placed on a 139 * linked list and freed only when the struct filedesc is released. 140 */ 141#define NDFILE 20 142#define NDSLOTSIZE sizeof(NDSLOTTYPE) 143#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 144#define NDSLOT(x) ((x) / NDENTRIES) 145#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 146#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 147 148/* 149 * SLIST entry used to keep track of ofiles which must be reclaimed when 150 * the process exits. 151 */ 152struct freetable { 153 struct fdescenttbl *ft_table; 154 SLIST_ENTRY(freetable) ft_next; 155}; 156 157/* 158 * Initial allocation: a filedesc structure + the head of SLIST used to 159 * keep track of old ofiles + enough space for NDFILE descriptors. 160 */ 161 162struct fdescenttbl0 { 163 int fdt_nfiles; 164 struct filedescent fdt_ofiles[NDFILE]; 165}; 166 167struct filedesc0 { 168 struct filedesc fd_fd; 169 SLIST_HEAD(, freetable) fd_free; 170 struct fdescenttbl0 fd_dfiles; 171 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 172}; 173 174/* 175 * Descriptor management. 176 */ 177volatile int openfiles; /* actual number of open files */ 178struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 179void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 180 181/* A mutex to protect the association between a proc and filedesc. */ 182static struct mtx fdesc_mtx; 183 184/* 185 * If low >= size, just return low. Otherwise find the first zero bit in the 186 * given bitmap, starting at low and not exceeding size - 1. Return size if 187 * not found. 188 */ 189static int 190fd_first_free(struct filedesc *fdp, int low, int size) 191{ 192 NDSLOTTYPE *map = fdp->fd_map; 193 NDSLOTTYPE mask; 194 int off, maxoff; 195 196 if (low >= size) 197 return (low); 198 199 off = NDSLOT(low); 200 if (low % NDENTRIES) { 201 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 202 if ((mask &= ~map[off]) != 0UL) 203 return (off * NDENTRIES + ffsl(mask) - 1); 204 ++off; 205 } 206 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 207 if (map[off] != ~0UL) 208 return (off * NDENTRIES + ffsl(~map[off]) - 1); 209 return (size); 210} 211 212/* 213 * Find the highest non-zero bit in the given bitmap, starting at 0 and 214 * not exceeding size - 1. Return -1 if not found. 215 */ 216static int 217fd_last_used(struct filedesc *fdp, int size) 218{ 219 NDSLOTTYPE *map = fdp->fd_map; 220 NDSLOTTYPE mask; 221 int off, minoff; 222 223 off = NDSLOT(size); 224 if (size % NDENTRIES) { 225 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 226 if ((mask &= map[off]) != 0) 227 return (off * NDENTRIES + flsl(mask) - 1); 228 --off; 229 } 230 for (minoff = NDSLOT(0); off >= minoff; --off) 231 if (map[off] != 0) 232 return (off * NDENTRIES + flsl(map[off]) - 1); 233 return (-1); 234} 235 236#ifdef INVARIANTS 237static int 238fdisused(struct filedesc *fdp, int fd) 239{ 240 241 FILEDESC_LOCK_ASSERT(fdp); 242 243 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 244 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 245 246 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 247} 248#endif 249 250/* 251 * Mark a file descriptor as used. 252 */ 253static void 254fdused(struct filedesc *fdp, int fd) 255{ 256 257 FILEDESC_XLOCK_ASSERT(fdp); 258 259 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 260 261 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 262 if (fd > fdp->fd_lastfile) 263 fdp->fd_lastfile = fd; 264 if (fd == fdp->fd_freefile) 265 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 266} 267 268/* 269 * Mark a file descriptor as unused. 270 */ 271static void 272fdunused(struct filedesc *fdp, int fd) 273{ 274 275 FILEDESC_XLOCK_ASSERT(fdp); 276 277 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 278 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 279 ("fd=%d is still in use", fd)); 280 281 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 282 if (fd < fdp->fd_freefile) 283 fdp->fd_freefile = fd; 284 if (fd == fdp->fd_lastfile) 285 fdp->fd_lastfile = fd_last_used(fdp, fd); 286} 287 288/* 289 * Free a file descriptor. 290 * 291 * Avoid some work if fdp is about to be destroyed. 292 */ 293static inline void 294fdefree_last(struct filedescent *fde) 295{ 296 297 filecaps_free(&fde->fde_caps); 298} 299 300static inline void 301fdfree(struct filedesc *fdp, int fd) 302{ 303 struct filedescent *fde; 304 305 fde = &fdp->fd_ofiles[fd]; 306#ifdef CAPABILITIES 307 seq_write_begin(&fde->fde_seq); 308#endif 309 fdefree_last(fde); 310 bzero(fde, fde_change_size); 311 fdunused(fdp, fd); 312#ifdef CAPABILITIES 313 seq_write_end(&fde->fde_seq); 314#endif 315} 316 317/* 318 * System calls on descriptors. 319 */ 320#ifndef _SYS_SYSPROTO_H_ 321struct getdtablesize_args { 322 int dummy; 323}; 324#endif 325/* ARGSUSED */ 326int 327sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 328{ 329 struct proc *p = td->td_proc; 330 uint64_t lim; 331 332 PROC_LOCK(p); 333 td->td_retval[0] = 334 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 335 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 336 PROC_UNLOCK(p); 337 if (lim < td->td_retval[0]) 338 td->td_retval[0] = lim; 339 return (0); 340} 341 342/* 343 * Duplicate a file descriptor to a particular value. 344 * 345 * Note: keep in mind that a potential race condition exists when closing 346 * descriptors from a shared descriptor table (via rfork). 347 */ 348#ifndef _SYS_SYSPROTO_H_ 349struct dup2_args { 350 u_int from; 351 u_int to; 352}; 353#endif 354/* ARGSUSED */ 355int 356sys_dup2(struct thread *td, struct dup2_args *uap) 357{ 358 359 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 360 td->td_retval)); 361} 362 363/* 364 * Duplicate a file descriptor. 365 */ 366#ifndef _SYS_SYSPROTO_H_ 367struct dup_args { 368 u_int fd; 369}; 370#endif 371/* ARGSUSED */ 372int 373sys_dup(struct thread *td, struct dup_args *uap) 374{ 375 376 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 377} 378 379/* 380 * The file control system call. 381 */ 382#ifndef _SYS_SYSPROTO_H_ 383struct fcntl_args { 384 int fd; 385 int cmd; 386 long arg; 387}; 388#endif 389/* ARGSUSED */ 390int 391sys_fcntl(struct thread *td, struct fcntl_args *uap) 392{ 393 394 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 395} 396 397int 398kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 399{ 400 struct flock fl; 401 struct __oflock ofl; 402 intptr_t arg1; 403 int error; 404 405 error = 0; 406 switch (cmd) { 407 case F_OGETLK: 408 case F_OSETLK: 409 case F_OSETLKW: 410 /* 411 * Convert old flock structure to new. 412 */ 413 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 414 fl.l_start = ofl.l_start; 415 fl.l_len = ofl.l_len; 416 fl.l_pid = ofl.l_pid; 417 fl.l_type = ofl.l_type; 418 fl.l_whence = ofl.l_whence; 419 fl.l_sysid = 0; 420 421 switch (cmd) { 422 case F_OGETLK: 423 cmd = F_GETLK; 424 break; 425 case F_OSETLK: 426 cmd = F_SETLK; 427 break; 428 case F_OSETLKW: 429 cmd = F_SETLKW; 430 break; 431 } 432 arg1 = (intptr_t)&fl; 433 break; 434 case F_GETLK: 435 case F_SETLK: 436 case F_SETLKW: 437 case F_SETLK_REMOTE: 438 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 439 arg1 = (intptr_t)&fl; 440 break; 441 default: 442 arg1 = arg; 443 break; 444 } 445 if (error) 446 return (error); 447 error = kern_fcntl(td, fd, cmd, arg1); 448 if (error) 449 return (error); 450 if (cmd == F_OGETLK) { 451 ofl.l_start = fl.l_start; 452 ofl.l_len = fl.l_len; 453 ofl.l_pid = fl.l_pid; 454 ofl.l_type = fl.l_type; 455 ofl.l_whence = fl.l_whence; 456 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 457 } else if (cmd == F_GETLK) { 458 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 459 } 460 return (error); 461} 462 463int 464kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 465{ 466 struct filedesc *fdp; 467 struct flock *flp; 468 struct file *fp, *fp2; 469 struct filedescent *fde; 470 struct proc *p; 471 struct vnode *vp; 472 cap_rights_t rights; 473 int error, flg, tmp; 474 uint64_t bsize; 475 off_t foffset; 476 477 error = 0; 478 flg = F_POSIX; 479 p = td->td_proc; 480 fdp = p->p_fd; 481 482 switch (cmd) { 483 case F_DUPFD: 484 tmp = arg; 485 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 486 break; 487 488 case F_DUPFD_CLOEXEC: 489 tmp = arg; 490 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp, 491 td->td_retval); 492 break; 493 494 case F_DUP2FD: 495 tmp = arg; 496 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 497 break; 498 499 case F_DUP2FD_CLOEXEC: 500 tmp = arg; 501 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp, 502 td->td_retval); 503 break; 504 505 case F_GETFD: 506 FILEDESC_SLOCK(fdp); 507 if (fget_locked(fdp, fd) == NULL) { 508 FILEDESC_SUNLOCK(fdp); 509 error = EBADF; 510 break; 511 } 512 fde = &fdp->fd_ofiles[fd]; 513 td->td_retval[0] = 514 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 515 FILEDESC_SUNLOCK(fdp); 516 break; 517 518 case F_SETFD: 519 FILEDESC_XLOCK(fdp); 520 if (fget_locked(fdp, fd) == NULL) { 521 FILEDESC_XUNLOCK(fdp); 522 error = EBADF; 523 break; 524 } 525 fde = &fdp->fd_ofiles[fd]; 526 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 527 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 528 FILEDESC_XUNLOCK(fdp); 529 break; 530 531 case F_GETFL: 532 error = fget_unlocked(fdp, fd, 533 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL); 534 if (error != 0) 535 break; 536 td->td_retval[0] = OFLAGS(fp->f_flag); 537 fdrop(fp, td); 538 break; 539 540 case F_SETFL: 541 error = fget_unlocked(fdp, fd, 542 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL); 543 if (error != 0) 544 break; 545 do { 546 tmp = flg = fp->f_flag; 547 tmp &= ~FCNTLFLAGS; 548 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 549 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 550 tmp = fp->f_flag & FNONBLOCK; 551 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 552 if (error != 0) { 553 fdrop(fp, td); 554 break; 555 } 556 tmp = fp->f_flag & FASYNC; 557 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 558 if (error == 0) { 559 fdrop(fp, td); 560 break; 561 } 562 atomic_clear_int(&fp->f_flag, FNONBLOCK); 563 tmp = 0; 564 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 565 fdrop(fp, td); 566 break; 567 568 case F_GETOWN: 569 error = fget_unlocked(fdp, fd, 570 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL); 571 if (error != 0) 572 break; 573 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 574 if (error == 0) 575 td->td_retval[0] = tmp; 576 fdrop(fp, td); 577 break; 578 579 case F_SETOWN: 580 error = fget_unlocked(fdp, fd, 581 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL); 582 if (error != 0) 583 break; 584 tmp = arg; 585 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 586 fdrop(fp, td); 587 break; 588 589 case F_SETLK_REMOTE: 590 error = priv_check(td, PRIV_NFS_LOCKD); 591 if (error) 592 return (error); 593 flg = F_REMOTE; 594 goto do_setlk; 595 596 case F_SETLKW: 597 flg |= F_WAIT; 598 /* FALLTHROUGH F_SETLK */ 599 600 case F_SETLK: 601 do_setlk: 602 cap_rights_init(&rights, CAP_FLOCK); 603 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL); 604 if (error != 0) 605 break; 606 if (fp->f_type != DTYPE_VNODE) { 607 error = EBADF; 608 fdrop(fp, td); 609 break; 610 } 611 612 flp = (struct flock *)arg; 613 if (flp->l_whence == SEEK_CUR) { 614 foffset = foffset_get(fp); 615 if (foffset < 0 || 616 (flp->l_start > 0 && 617 foffset > OFF_MAX - flp->l_start)) { 618 FILEDESC_SUNLOCK(fdp); 619 error = EOVERFLOW; 620 fdrop(fp, td); 621 break; 622 } 623 flp->l_start += foffset; 624 } 625 626 vp = fp->f_vnode; 627 switch (flp->l_type) { 628 case F_RDLCK: 629 if ((fp->f_flag & FREAD) == 0) { 630 error = EBADF; 631 break; 632 } 633 PROC_LOCK(p->p_leader); 634 p->p_leader->p_flag |= P_ADVLOCK; 635 PROC_UNLOCK(p->p_leader); 636 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 637 flp, flg); 638 break; 639 case F_WRLCK: 640 if ((fp->f_flag & FWRITE) == 0) { 641 error = EBADF; 642 break; 643 } 644 PROC_LOCK(p->p_leader); 645 p->p_leader->p_flag |= P_ADVLOCK; 646 PROC_UNLOCK(p->p_leader); 647 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 648 flp, flg); 649 break; 650 case F_UNLCK: 651 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 652 flp, flg); 653 break; 654 case F_UNLCKSYS: 655 /* 656 * Temporary api for testing remote lock 657 * infrastructure. 658 */ 659 if (flg != F_REMOTE) { 660 error = EINVAL; 661 break; 662 } 663 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 664 F_UNLCKSYS, flp, flg); 665 break; 666 default: 667 error = EINVAL; 668 break; 669 } 670 if (error != 0 || flp->l_type == F_UNLCK || 671 flp->l_type == F_UNLCKSYS) { 672 fdrop(fp, td); 673 break; 674 } 675 676 /* 677 * Check for a race with close. 678 * 679 * The vnode is now advisory locked (or unlocked, but this case 680 * is not really important) as the caller requested. 681 * We had to drop the filedesc lock, so we need to recheck if 682 * the descriptor is still valid, because if it was closed 683 * in the meantime we need to remove advisory lock from the 684 * vnode - close on any descriptor leading to an advisory 685 * locked vnode, removes that lock. 686 * We will return 0 on purpose in that case, as the result of 687 * successful advisory lock might have been externally visible 688 * already. This is fine - effectively we pretend to the caller 689 * that the closing thread was a bit slower and that the 690 * advisory lock succeeded before the close. 691 */ 692 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL); 693 if (error != 0) { 694 fdrop(fp, td); 695 break; 696 } 697 if (fp != fp2) { 698 flp->l_whence = SEEK_SET; 699 flp->l_start = 0; 700 flp->l_len = 0; 701 flp->l_type = F_UNLCK; 702 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 703 F_UNLCK, flp, F_POSIX); 704 } 705 fdrop(fp, td); 706 fdrop(fp2, td); 707 break; 708 709 case F_GETLK: 710 error = fget_unlocked(fdp, fd, 711 cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL); 712 if (error != 0) 713 break; 714 if (fp->f_type != DTYPE_VNODE) { 715 error = EBADF; 716 fdrop(fp, td); 717 break; 718 } 719 flp = (struct flock *)arg; 720 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 721 flp->l_type != F_UNLCK) { 722 error = EINVAL; 723 fdrop(fp, td); 724 break; 725 } 726 if (flp->l_whence == SEEK_CUR) { 727 foffset = foffset_get(fp); 728 if ((flp->l_start > 0 && 729 foffset > OFF_MAX - flp->l_start) || 730 (flp->l_start < 0 && 731 foffset < OFF_MIN - flp->l_start)) { 732 FILEDESC_SUNLOCK(fdp); 733 error = EOVERFLOW; 734 fdrop(fp, td); 735 break; 736 } 737 flp->l_start += foffset; 738 } 739 vp = fp->f_vnode; 740 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 741 F_POSIX); 742 fdrop(fp, td); 743 break; 744 745 case F_RDAHEAD: 746 arg = arg ? 128 * 1024: 0; 747 /* FALLTHROUGH */ 748 case F_READAHEAD: 749 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL); 750 if (error != 0) 751 break; 752 if (fp->f_type != DTYPE_VNODE) { 753 fdrop(fp, td); 754 error = EBADF; 755 break; 756 } 757 vp = fp->f_vnode; 758 /* 759 * Exclusive lock synchronizes against f_seqcount reads and 760 * writes in sequential_heuristic(). 761 */ 762 error = vn_lock(vp, LK_EXCLUSIVE); 763 if (error != 0) { 764 fdrop(fp, td); 765 break; 766 } 767 if (arg >= 0) { 768 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 769 fp->f_seqcount = (arg + bsize - 1) / bsize; 770 atomic_set_int(&fp->f_flag, FRDAHEAD); 771 } else { 772 atomic_clear_int(&fp->f_flag, FRDAHEAD); 773 } 774 VOP_UNLOCK(vp, 0); 775 fdrop(fp, td); 776 break; 777 778 default: 779 error = EINVAL; 780 break; 781 } 782 return (error); 783} 784 785static int 786getmaxfd(struct proc *p) 787{ 788 int maxfd; 789 790 PROC_LOCK(p); 791 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 792 PROC_UNLOCK(p); 793 794 return (maxfd); 795} 796 797/* 798 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 799 */ 800static int 801do_dup(struct thread *td, int flags, int old, int new, 802 register_t *retval) 803{ 804 struct filedesc *fdp; 805 struct filedescent *oldfde, *newfde; 806 struct proc *p; 807 struct file *fp; 808 struct file *delfp; 809 int error, maxfd; 810 811 p = td->td_proc; 812 fdp = p->p_fd; 813 814 /* 815 * Verify we have a valid descriptor to dup from and possibly to 816 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 817 * return EINVAL when the new descriptor is out of bounds. 818 */ 819 if (old < 0) 820 return (EBADF); 821 if (new < 0) 822 return (flags & DUP_FCNTL ? EINVAL : EBADF); 823 maxfd = getmaxfd(p); 824 if (new >= maxfd) 825 return (flags & DUP_FCNTL ? EINVAL : EBADF); 826 827 FILEDESC_XLOCK(fdp); 828 if (fget_locked(fdp, old) == NULL) { 829 FILEDESC_XUNLOCK(fdp); 830 return (EBADF); 831 } 832 oldfde = &fdp->fd_ofiles[old]; 833 if (flags & DUP_FIXED && old == new) { 834 *retval = new; 835 if (flags & DUP_CLOEXEC) 836 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 837 FILEDESC_XUNLOCK(fdp); 838 return (0); 839 } 840 fp = oldfde->fde_file; 841 fhold(fp); 842 843 /* 844 * If the caller specified a file descriptor, make sure the file 845 * table is large enough to hold it, and grab it. Otherwise, just 846 * allocate a new descriptor the usual way. 847 */ 848 if (flags & DUP_FIXED) { 849 if (new >= fdp->fd_nfiles) { 850 /* 851 * The resource limits are here instead of e.g. 852 * fdalloc(), because the file descriptor table may be 853 * shared between processes, so we can't really use 854 * racct_add()/racct_sub(). Instead of counting the 855 * number of actually allocated descriptors, just put 856 * the limit on the size of the file descriptor table. 857 */ 858#ifdef RACCT 859 PROC_LOCK(p); 860 error = racct_set(p, RACCT_NOFILE, new + 1); 861 PROC_UNLOCK(p); 862 if (error != 0) { 863 FILEDESC_XUNLOCK(fdp); 864 fdrop(fp, td); 865 return (EMFILE); 866 } 867#endif 868 fdgrowtable_exp(fdp, new + 1); 869 oldfde = &fdp->fd_ofiles[old]; 870 } 871 newfde = &fdp->fd_ofiles[new]; 872 if (newfde->fde_file == NULL) 873 fdused(fdp, new); 874 } else { 875 if ((error = fdalloc(td, new, &new)) != 0) { 876 FILEDESC_XUNLOCK(fdp); 877 fdrop(fp, td); 878 return (error); 879 } 880 newfde = &fdp->fd_ofiles[new]; 881 } 882 883 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 884 KASSERT(old != new, ("new fd is same as old")); 885 886 delfp = newfde->fde_file; 887 888 /* 889 * Duplicate the source descriptor. 890 */ 891#ifdef CAPABILITIES 892 seq_write_begin(&newfde->fde_seq); 893#endif 894 filecaps_free(&newfde->fde_caps); 895 memcpy(newfde, oldfde, fde_change_size); 896 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 897 if ((flags & DUP_CLOEXEC) != 0) 898 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 899 else 900 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 901#ifdef CAPABILITIES 902 seq_write_end(&newfde->fde_seq); 903#endif 904 *retval = new; 905 906 if (delfp != NULL) { 907 (void) closefp(fdp, new, delfp, td, 1); 908 /* closefp() drops the FILEDESC lock for us. */ 909 } else { 910 FILEDESC_XUNLOCK(fdp); 911 } 912 913 return (0); 914} 915 916/* 917 * If sigio is on the list associated with a process or process group, 918 * disable signalling from the device, remove sigio from the list and 919 * free sigio. 920 */ 921void 922funsetown(struct sigio **sigiop) 923{ 924 struct sigio *sigio; 925 926 SIGIO_LOCK(); 927 sigio = *sigiop; 928 if (sigio == NULL) { 929 SIGIO_UNLOCK(); 930 return; 931 } 932 *(sigio->sio_myref) = NULL; 933 if ((sigio)->sio_pgid < 0) { 934 struct pgrp *pg = (sigio)->sio_pgrp; 935 PGRP_LOCK(pg); 936 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 937 sigio, sio_pgsigio); 938 PGRP_UNLOCK(pg); 939 } else { 940 struct proc *p = (sigio)->sio_proc; 941 PROC_LOCK(p); 942 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 943 sigio, sio_pgsigio); 944 PROC_UNLOCK(p); 945 } 946 SIGIO_UNLOCK(); 947 crfree(sigio->sio_ucred); 948 free(sigio, M_SIGIO); 949} 950 951/* 952 * Free a list of sigio structures. 953 * We only need to lock the SIGIO_LOCK because we have made ourselves 954 * inaccessible to callers of fsetown and therefore do not need to lock 955 * the proc or pgrp struct for the list manipulation. 956 */ 957void 958funsetownlst(struct sigiolst *sigiolst) 959{ 960 struct proc *p; 961 struct pgrp *pg; 962 struct sigio *sigio; 963 964 sigio = SLIST_FIRST(sigiolst); 965 if (sigio == NULL) 966 return; 967 p = NULL; 968 pg = NULL; 969 970 /* 971 * Every entry of the list should belong 972 * to a single proc or pgrp. 973 */ 974 if (sigio->sio_pgid < 0) { 975 pg = sigio->sio_pgrp; 976 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 977 } else /* if (sigio->sio_pgid > 0) */ { 978 p = sigio->sio_proc; 979 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 980 } 981 982 SIGIO_LOCK(); 983 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 984 *(sigio->sio_myref) = NULL; 985 if (pg != NULL) { 986 KASSERT(sigio->sio_pgid < 0, 987 ("Proc sigio in pgrp sigio list")); 988 KASSERT(sigio->sio_pgrp == pg, 989 ("Bogus pgrp in sigio list")); 990 PGRP_LOCK(pg); 991 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 992 sio_pgsigio); 993 PGRP_UNLOCK(pg); 994 } else /* if (p != NULL) */ { 995 KASSERT(sigio->sio_pgid > 0, 996 ("Pgrp sigio in proc sigio list")); 997 KASSERT(sigio->sio_proc == p, 998 ("Bogus proc in sigio list")); 999 PROC_LOCK(p); 1000 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1001 sio_pgsigio); 1002 PROC_UNLOCK(p); 1003 } 1004 SIGIO_UNLOCK(); 1005 crfree(sigio->sio_ucred); 1006 free(sigio, M_SIGIO); 1007 SIGIO_LOCK(); 1008 } 1009 SIGIO_UNLOCK(); 1010} 1011 1012/* 1013 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1014 * 1015 * After permission checking, add a sigio structure to the sigio list for 1016 * the process or process group. 1017 */ 1018int 1019fsetown(pid_t pgid, struct sigio **sigiop) 1020{ 1021 struct proc *proc; 1022 struct pgrp *pgrp; 1023 struct sigio *sigio; 1024 int ret; 1025 1026 if (pgid == 0) { 1027 funsetown(sigiop); 1028 return (0); 1029 } 1030 1031 ret = 0; 1032 1033 /* Allocate and fill in the new sigio out of locks. */ 1034 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1035 sigio->sio_pgid = pgid; 1036 sigio->sio_ucred = crhold(curthread->td_ucred); 1037 sigio->sio_myref = sigiop; 1038 1039 sx_slock(&proctree_lock); 1040 if (pgid > 0) { 1041 proc = pfind(pgid); 1042 if (proc == NULL) { 1043 ret = ESRCH; 1044 goto fail; 1045 } 1046 1047 /* 1048 * Policy - Don't allow a process to FSETOWN a process 1049 * in another session. 1050 * 1051 * Remove this test to allow maximum flexibility or 1052 * restrict FSETOWN to the current process or process 1053 * group for maximum safety. 1054 */ 1055 PROC_UNLOCK(proc); 1056 if (proc->p_session != curthread->td_proc->p_session) { 1057 ret = EPERM; 1058 goto fail; 1059 } 1060 1061 pgrp = NULL; 1062 } else /* if (pgid < 0) */ { 1063 pgrp = pgfind(-pgid); 1064 if (pgrp == NULL) { 1065 ret = ESRCH; 1066 goto fail; 1067 } 1068 PGRP_UNLOCK(pgrp); 1069 1070 /* 1071 * Policy - Don't allow a process to FSETOWN a process 1072 * in another session. 1073 * 1074 * Remove this test to allow maximum flexibility or 1075 * restrict FSETOWN to the current process or process 1076 * group for maximum safety. 1077 */ 1078 if (pgrp->pg_session != curthread->td_proc->p_session) { 1079 ret = EPERM; 1080 goto fail; 1081 } 1082 1083 proc = NULL; 1084 } 1085 funsetown(sigiop); 1086 if (pgid > 0) { 1087 PROC_LOCK(proc); 1088 /* 1089 * Since funsetownlst() is called without the proctree 1090 * locked, we need to check for P_WEXIT. 1091 * XXX: is ESRCH correct? 1092 */ 1093 if ((proc->p_flag & P_WEXIT) != 0) { 1094 PROC_UNLOCK(proc); 1095 ret = ESRCH; 1096 goto fail; 1097 } 1098 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1099 sigio->sio_proc = proc; 1100 PROC_UNLOCK(proc); 1101 } else { 1102 PGRP_LOCK(pgrp); 1103 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1104 sigio->sio_pgrp = pgrp; 1105 PGRP_UNLOCK(pgrp); 1106 } 1107 sx_sunlock(&proctree_lock); 1108 SIGIO_LOCK(); 1109 *sigiop = sigio; 1110 SIGIO_UNLOCK(); 1111 return (0); 1112 1113fail: 1114 sx_sunlock(&proctree_lock); 1115 crfree(sigio->sio_ucred); 1116 free(sigio, M_SIGIO); 1117 return (ret); 1118} 1119 1120/* 1121 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1122 */ 1123pid_t 1124fgetown(sigiop) 1125 struct sigio **sigiop; 1126{ 1127 pid_t pgid; 1128 1129 SIGIO_LOCK(); 1130 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1131 SIGIO_UNLOCK(); 1132 return (pgid); 1133} 1134 1135/* 1136 * Function drops the filedesc lock on return. 1137 */ 1138static int 1139closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1140 int holdleaders) 1141{ 1142 int error; 1143 1144 FILEDESC_XLOCK_ASSERT(fdp); 1145 1146 if (holdleaders) { 1147 if (td->td_proc->p_fdtol != NULL) { 1148 /* 1149 * Ask fdfree() to sleep to ensure that all relevant 1150 * process leaders can be traversed in closef(). 1151 */ 1152 fdp->fd_holdleaderscount++; 1153 } else { 1154 holdleaders = 0; 1155 } 1156 } 1157 1158 /* 1159 * We now hold the fp reference that used to be owned by the 1160 * descriptor array. We have to unlock the FILEDESC *AFTER* 1161 * knote_fdclose to prevent a race of the fd getting opened, a knote 1162 * added, and deleteing a knote for the new fd. 1163 */ 1164 knote_fdclose(td, fd); 1165 1166 /* 1167 * We need to notify mqueue if the object is of type mqueue. 1168 */ 1169 if (fp->f_type == DTYPE_MQUEUE) 1170 mq_fdclose(td, fd, fp); 1171 FILEDESC_XUNLOCK(fdp); 1172 1173 error = closef(fp, td); 1174 if (holdleaders) { 1175 FILEDESC_XLOCK(fdp); 1176 fdp->fd_holdleaderscount--; 1177 if (fdp->fd_holdleaderscount == 0 && 1178 fdp->fd_holdleaderswakeup != 0) { 1179 fdp->fd_holdleaderswakeup = 0; 1180 wakeup(&fdp->fd_holdleaderscount); 1181 } 1182 FILEDESC_XUNLOCK(fdp); 1183 } 1184 return (error); 1185} 1186 1187/* 1188 * Close a file descriptor. 1189 */ 1190#ifndef _SYS_SYSPROTO_H_ 1191struct close_args { 1192 int fd; 1193}; 1194#endif 1195/* ARGSUSED */ 1196int 1197sys_close(td, uap) 1198 struct thread *td; 1199 struct close_args *uap; 1200{ 1201 1202 return (kern_close(td, uap->fd)); 1203} 1204 1205int 1206kern_close(td, fd) 1207 struct thread *td; 1208 int fd; 1209{ 1210 struct filedesc *fdp; 1211 struct file *fp; 1212 1213 fdp = td->td_proc->p_fd; 1214 1215 AUDIT_SYSCLOSE(td, fd); 1216 1217 FILEDESC_XLOCK(fdp); 1218 if ((fp = fget_locked(fdp, fd)) == NULL) { 1219 FILEDESC_XUNLOCK(fdp); 1220 return (EBADF); 1221 } 1222 fdfree(fdp, fd); 1223 1224 /* closefp() drops the FILEDESC lock for us. */ 1225 return (closefp(fdp, fd, fp, td, 1)); 1226} 1227 1228/* 1229 * Close open file descriptors. 1230 */ 1231#ifndef _SYS_SYSPROTO_H_ 1232struct closefrom_args { 1233 int lowfd; 1234}; 1235#endif 1236/* ARGSUSED */ 1237int 1238sys_closefrom(struct thread *td, struct closefrom_args *uap) 1239{ 1240 struct filedesc *fdp; 1241 int fd; 1242 1243 fdp = td->td_proc->p_fd; 1244 AUDIT_ARG_FD(uap->lowfd); 1245 1246 /* 1247 * Treat negative starting file descriptor values identical to 1248 * closefrom(0) which closes all files. 1249 */ 1250 if (uap->lowfd < 0) 1251 uap->lowfd = 0; 1252 FILEDESC_SLOCK(fdp); 1253 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1254 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1255 FILEDESC_SUNLOCK(fdp); 1256 (void)kern_close(td, fd); 1257 FILEDESC_SLOCK(fdp); 1258 } 1259 } 1260 FILEDESC_SUNLOCK(fdp); 1261 return (0); 1262} 1263 1264#if defined(COMPAT_43) 1265/* 1266 * Return status information about a file descriptor. 1267 */ 1268#ifndef _SYS_SYSPROTO_H_ 1269struct ofstat_args { 1270 int fd; 1271 struct ostat *sb; 1272}; 1273#endif 1274/* ARGSUSED */ 1275int 1276ofstat(struct thread *td, struct ofstat_args *uap) 1277{ 1278 struct ostat oub; 1279 struct stat ub; 1280 int error; 1281 1282 error = kern_fstat(td, uap->fd, &ub); 1283 if (error == 0) { 1284 cvtstat(&ub, &oub); 1285 error = copyout(&oub, uap->sb, sizeof(oub)); 1286 } 1287 return (error); 1288} 1289#endif /* COMPAT_43 */ 1290 1291/* 1292 * Return status information about a file descriptor. 1293 */ 1294#ifndef _SYS_SYSPROTO_H_ 1295struct fstat_args { 1296 int fd; 1297 struct stat *sb; 1298}; 1299#endif 1300/* ARGSUSED */ 1301int 1302sys_fstat(struct thread *td, struct fstat_args *uap) 1303{ 1304 struct stat ub; 1305 int error; 1306 1307 error = kern_fstat(td, uap->fd, &ub); 1308 if (error == 0) 1309 error = copyout(&ub, uap->sb, sizeof(ub)); 1310 return (error); 1311} 1312 1313int 1314kern_fstat(struct thread *td, int fd, struct stat *sbp) 1315{ 1316 struct file *fp; 1317 cap_rights_t rights; 1318 int error; 1319 1320 AUDIT_ARG_FD(fd); 1321 1322 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1323 if (error != 0) 1324 return (error); 1325 1326 AUDIT_ARG_FILE(td->td_proc, fp); 1327 1328 error = fo_stat(fp, sbp, td->td_ucred, td); 1329 fdrop(fp, td); 1330#ifdef KTRACE 1331 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1332 ktrstat(sbp); 1333#endif 1334 return (error); 1335} 1336 1337/* 1338 * Return status information about a file descriptor. 1339 */ 1340#ifndef _SYS_SYSPROTO_H_ 1341struct nfstat_args { 1342 int fd; 1343 struct nstat *sb; 1344}; 1345#endif 1346/* ARGSUSED */ 1347int 1348sys_nfstat(struct thread *td, struct nfstat_args *uap) 1349{ 1350 struct nstat nub; 1351 struct stat ub; 1352 int error; 1353 1354 error = kern_fstat(td, uap->fd, &ub); 1355 if (error == 0) { 1356 cvtnstat(&ub, &nub); 1357 error = copyout(&nub, uap->sb, sizeof(nub)); 1358 } 1359 return (error); 1360} 1361 1362/* 1363 * Return pathconf information about a file descriptor. 1364 */ 1365#ifndef _SYS_SYSPROTO_H_ 1366struct fpathconf_args { 1367 int fd; 1368 int name; 1369}; 1370#endif 1371/* ARGSUSED */ 1372int 1373sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1374{ 1375 struct file *fp; 1376 struct vnode *vp; 1377 cap_rights_t rights; 1378 int error; 1379 1380 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1381 if (error != 0) 1382 return (error); 1383 1384 /* If asynchronous I/O is available, it works for all descriptors. */ 1385 if (uap->name == _PC_ASYNC_IO) { 1386 td->td_retval[0] = async_io_version; 1387 goto out; 1388 } 1389 vp = fp->f_vnode; 1390 if (vp != NULL) { 1391 vn_lock(vp, LK_SHARED | LK_RETRY); 1392 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1393 VOP_UNLOCK(vp, 0); 1394 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1395 if (uap->name != _PC_PIPE_BUF) { 1396 error = EINVAL; 1397 } else { 1398 td->td_retval[0] = PIPE_BUF; 1399 error = 0; 1400 } 1401 } else { 1402 error = EOPNOTSUPP; 1403 } 1404out: 1405 fdrop(fp, td); 1406 return (error); 1407} 1408 1409/* 1410 * Initialize filecaps structure. 1411 */ 1412void 1413filecaps_init(struct filecaps *fcaps) 1414{ 1415 1416 bzero(fcaps, sizeof(*fcaps)); 1417 fcaps->fc_nioctls = -1; 1418} 1419 1420/* 1421 * Copy filecaps structure allocating memory for ioctls array if needed. 1422 */ 1423void 1424filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1425{ 1426 size_t size; 1427 1428 *dst = *src; 1429 if (src->fc_ioctls != NULL) { 1430 KASSERT(src->fc_nioctls > 0, 1431 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1432 1433 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1434 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1435 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1436 } 1437} 1438 1439/* 1440 * Move filecaps structure to the new place and clear the old place. 1441 */ 1442void 1443filecaps_move(struct filecaps *src, struct filecaps *dst) 1444{ 1445 1446 *dst = *src; 1447 bzero(src, sizeof(*src)); 1448} 1449 1450/* 1451 * Fill the given filecaps structure with full rights. 1452 */ 1453static void 1454filecaps_fill(struct filecaps *fcaps) 1455{ 1456 1457 CAP_ALL(&fcaps->fc_rights); 1458 fcaps->fc_ioctls = NULL; 1459 fcaps->fc_nioctls = -1; 1460 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1461} 1462 1463/* 1464 * Free memory allocated within filecaps structure. 1465 */ 1466void 1467filecaps_free(struct filecaps *fcaps) 1468{ 1469 1470 free(fcaps->fc_ioctls, M_FILECAPS); 1471 bzero(fcaps, sizeof(*fcaps)); 1472} 1473 1474/* 1475 * Validate the given filecaps structure. 1476 */ 1477static void 1478filecaps_validate(const struct filecaps *fcaps, const char *func) 1479{ 1480 1481 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1482 ("%s: invalid rights", func)); 1483 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1484 ("%s: invalid fcntls", func)); 1485 KASSERT(fcaps->fc_fcntls == 0 || 1486 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1487 ("%s: fcntls without CAP_FCNTL", func)); 1488 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1489 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1490 ("%s: invalid ioctls", func)); 1491 KASSERT(fcaps->fc_nioctls == 0 || 1492 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1493 ("%s: ioctls without CAP_IOCTL", func)); 1494} 1495 1496static void 1497fdgrowtable_exp(struct filedesc *fdp, int nfd) 1498{ 1499 int nfd1; 1500 1501 FILEDESC_XLOCK_ASSERT(fdp); 1502 1503 nfd1 = fdp->fd_nfiles * 2; 1504 if (nfd1 < nfd) 1505 nfd1 = nfd; 1506 fdgrowtable(fdp, nfd1); 1507} 1508 1509/* 1510 * Grow the file table to accomodate (at least) nfd descriptors. 1511 */ 1512static void 1513fdgrowtable(struct filedesc *fdp, int nfd) 1514{ 1515 struct filedesc0 *fdp0; 1516 struct freetable *ft; 1517 struct fdescenttbl *ntable; 1518 struct fdescenttbl *otable; 1519 int nnfiles, onfiles; 1520 NDSLOTTYPE *nmap, *omap; 1521 1522 FILEDESC_XLOCK_ASSERT(fdp); 1523 1524 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1525 1526 /* save old values */ 1527 onfiles = fdp->fd_nfiles; 1528 otable = fdp->fd_files; 1529 omap = fdp->fd_map; 1530 1531 /* compute the size of the new table */ 1532 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1533 if (nnfiles <= onfiles) 1534 /* the table is already large enough */ 1535 return; 1536 1537 /* 1538 * Allocate a new table. We need enough space for the number of 1539 * entries, file entries themselves and the struct freetable we will use 1540 * when we decommission the table and place it on the freelist. 1541 * We place the struct freetable in the middle so we don't have 1542 * to worry about padding. 1543 */ 1544 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1545 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1546 sizeof(struct freetable), 1547 M_FILEDESC, M_ZERO | M_WAITOK); 1548 /* copy the old data */ 1549 ntable->fdt_nfiles = nnfiles; 1550 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1551 onfiles * sizeof(ntable->fdt_ofiles[0])); 1552 1553 /* 1554 * Allocate a new map only if the old is not large enough. It will 1555 * grow at a slower rate than the table as it can map more 1556 * entries than the table can hold. 1557 */ 1558 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1559 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1560 M_ZERO | M_WAITOK); 1561 /* copy over the old data and update the pointer */ 1562 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1563 fdp->fd_map = nmap; 1564 } 1565 1566 /* 1567 * Make sure that ntable is correctly initialized before we replace 1568 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1569 * data. 1570 */ 1571 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1572 1573 /* 1574 * Do not free the old file table, as some threads may still 1575 * reference entries within it. Instead, place it on a freelist 1576 * which will be processed when the struct filedesc is released. 1577 * 1578 * Note that if onfiles == NDFILE, we're dealing with the original 1579 * static allocation contained within (struct filedesc0 *)fdp, 1580 * which must not be freed. 1581 */ 1582 if (onfiles > NDFILE) { 1583 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1584 fdp0 = (struct filedesc0 *)fdp; 1585 ft->ft_table = otable; 1586 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1587 } 1588 /* 1589 * The map does not have the same possibility of threads still 1590 * holding references to it. So always free it as long as it 1591 * does not reference the original static allocation. 1592 */ 1593 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1594 free(omap, M_FILEDESC); 1595} 1596 1597/* 1598 * Allocate a file descriptor for the process. 1599 */ 1600int 1601fdalloc(struct thread *td, int minfd, int *result) 1602{ 1603 struct proc *p = td->td_proc; 1604 struct filedesc *fdp = p->p_fd; 1605 int fd = -1, maxfd, allocfd; 1606#ifdef RACCT 1607 int error; 1608#endif 1609 1610 FILEDESC_XLOCK_ASSERT(fdp); 1611 1612 if (fdp->fd_freefile > minfd) 1613 minfd = fdp->fd_freefile; 1614 1615 maxfd = getmaxfd(p); 1616 1617 /* 1618 * Search the bitmap for a free descriptor starting at minfd. 1619 * If none is found, grow the file table. 1620 */ 1621 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1622 if (fd >= maxfd) 1623 return (EMFILE); 1624 if (fd >= fdp->fd_nfiles) { 1625 allocfd = min(fd * 2, maxfd); 1626#ifdef RACCT 1627 PROC_LOCK(p); 1628 error = racct_set(p, RACCT_NOFILE, allocfd); 1629 PROC_UNLOCK(p); 1630 if (error != 0) 1631 return (EMFILE); 1632#endif 1633 /* 1634 * fd is already equal to first free descriptor >= minfd, so 1635 * we only need to grow the table and we are done. 1636 */ 1637 fdgrowtable_exp(fdp, allocfd); 1638 } 1639 1640 /* 1641 * Perform some sanity checks, then mark the file descriptor as 1642 * used and return it to the caller. 1643 */ 1644 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1645 ("invalid descriptor %d", fd)); 1646 KASSERT(!fdisused(fdp, fd), 1647 ("fd_first_free() returned non-free descriptor")); 1648 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1649 ("file descriptor isn't free")); 1650 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); 1651 fdused(fdp, fd); 1652 *result = fd; 1653 return (0); 1654} 1655 1656/* 1657 * Allocate n file descriptors for the process. 1658 */ 1659int 1660fdallocn(struct thread *td, int minfd, int *fds, int n) 1661{ 1662 struct proc *p = td->td_proc; 1663 struct filedesc *fdp = p->p_fd; 1664 int i; 1665 1666 FILEDESC_XLOCK_ASSERT(fdp); 1667 1668 for (i = 0; i < n; i++) 1669 if (fdalloc(td, 0, &fds[i]) != 0) 1670 break; 1671 1672 if (i < n) { 1673 for (i--; i >= 0; i--) 1674 fdunused(fdp, fds[i]); 1675 return (EMFILE); 1676 } 1677 1678 return (0); 1679} 1680 1681/* 1682 * Create a new open file structure and allocate a file decriptor for the 1683 * process that refers to it. We add one reference to the file for the 1684 * descriptor table and one reference for resultfp. This is to prevent us 1685 * being preempted and the entry in the descriptor table closed after we 1686 * release the FILEDESC lock. 1687 */ 1688int 1689falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1690{ 1691 struct file *fp; 1692 int error, fd; 1693 1694 error = falloc_noinstall(td, &fp); 1695 if (error) 1696 return (error); /* no reference held on error */ 1697 1698 error = finstall(td, fp, &fd, flags, NULL); 1699 if (error) { 1700 fdrop(fp, td); /* one reference (fp only) */ 1701 return (error); 1702 } 1703 1704 if (resultfp != NULL) 1705 *resultfp = fp; /* copy out result */ 1706 else 1707 fdrop(fp, td); /* release local reference */ 1708 1709 if (resultfd != NULL) 1710 *resultfd = fd; 1711 1712 return (0); 1713} 1714 1715/* 1716 * Create a new open file structure without allocating a file descriptor. 1717 */ 1718int 1719falloc_noinstall(struct thread *td, struct file **resultfp) 1720{ 1721 struct file *fp; 1722 int maxuserfiles = maxfiles - (maxfiles / 20); 1723 static struct timeval lastfail; 1724 static int curfail; 1725 1726 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1727 1728 if ((openfiles >= maxuserfiles && 1729 priv_check(td, PRIV_MAXFILES) != 0) || 1730 openfiles >= maxfiles) { 1731 if (ppsratecheck(&lastfail, &curfail, 1)) { 1732 printf("kern.maxfiles limit exceeded by uid %i, " 1733 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1734 } 1735 return (ENFILE); 1736 } 1737 atomic_add_int(&openfiles, 1); 1738 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1739 refcount_init(&fp->f_count, 1); 1740 fp->f_cred = crhold(td->td_ucred); 1741 fp->f_ops = &badfileops; 1742 *resultfp = fp; 1743 return (0); 1744} 1745 1746/* 1747 * Install a file in a file descriptor table. 1748 */ 1749int 1750finstall(struct thread *td, struct file *fp, int *fd, int flags, 1751 struct filecaps *fcaps) 1752{ 1753 struct filedesc *fdp = td->td_proc->p_fd; 1754 struct filedescent *fde; 1755 int error; 1756 1757 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1758 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1759 if (fcaps != NULL) 1760 filecaps_validate(fcaps, __func__); 1761 1762 FILEDESC_XLOCK(fdp); 1763 if ((error = fdalloc(td, 0, fd))) { 1764 FILEDESC_XUNLOCK(fdp); 1765 return (error); 1766 } 1767 fhold(fp); 1768 fde = &fdp->fd_ofiles[*fd]; 1769#ifdef CAPABILITIES 1770 seq_write_begin(&fde->fde_seq); 1771#endif 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#ifdef CAPABILITIES 1780 seq_write_end(&fde->fde_seq); 1781#endif 1782 FILEDESC_XUNLOCK(fdp); 1783 return (0); 1784} 1785 1786/* 1787 * Build a new filedesc structure from another. 1788 * Copy the current, root, and jail root vnode references. 1789 */ 1790struct filedesc * 1791fdinit(struct filedesc *fdp) 1792{ 1793 struct filedesc0 *newfdp; 1794 1795 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1796 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1797 if (fdp != NULL) { 1798 FILEDESC_SLOCK(fdp); 1799 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1800 if (newfdp->fd_fd.fd_cdir) 1801 VREF(newfdp->fd_fd.fd_cdir); 1802 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1803 if (newfdp->fd_fd.fd_rdir) 1804 VREF(newfdp->fd_fd.fd_rdir); 1805 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1806 if (newfdp->fd_fd.fd_jdir) 1807 VREF(newfdp->fd_fd.fd_jdir); 1808 FILEDESC_SUNLOCK(fdp); 1809 } 1810 1811 /* Create the file descriptor table. */ 1812 newfdp->fd_fd.fd_refcnt = 1; 1813 newfdp->fd_fd.fd_holdcnt = 1; 1814 newfdp->fd_fd.fd_cmask = CMASK; 1815 newfdp->fd_dfiles.fdt_nfiles = NDFILE; 1816 newfdp->fd_fd.fd_files = (struct fdescenttbl *)&newfdp->fd_dfiles; 1817 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1818 newfdp->fd_fd.fd_lastfile = -1; 1819 return (&newfdp->fd_fd); 1820} 1821 1822static struct filedesc * 1823fdhold(struct proc *p) 1824{ 1825 struct filedesc *fdp; 1826 1827 mtx_lock(&fdesc_mtx); 1828 fdp = p->p_fd; 1829 if (fdp != NULL) 1830 fdp->fd_holdcnt++; 1831 mtx_unlock(&fdesc_mtx); 1832 return (fdp); 1833} 1834 1835static void 1836fddrop(struct filedesc *fdp) 1837{ 1838 struct filedesc0 *fdp0; 1839 struct freetable *ft; 1840 int i; 1841 1842 mtx_lock(&fdesc_mtx); 1843 i = --fdp->fd_holdcnt; 1844 mtx_unlock(&fdesc_mtx); 1845 if (i > 0) 1846 return; 1847 1848 FILEDESC_LOCK_DESTROY(fdp); 1849 fdp0 = (struct filedesc0 *)fdp; 1850 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1851 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1852 free(ft->ft_table, M_FILEDESC); 1853 } 1854 free(fdp, M_FILEDESC); 1855} 1856 1857/* 1858 * Share a filedesc structure. 1859 */ 1860struct filedesc * 1861fdshare(struct filedesc *fdp) 1862{ 1863 1864 FILEDESC_XLOCK(fdp); 1865 fdp->fd_refcnt++; 1866 FILEDESC_XUNLOCK(fdp); 1867 return (fdp); 1868} 1869 1870/* 1871 * Unshare a filedesc structure, if necessary by making a copy 1872 */ 1873void 1874fdunshare(struct thread *td) 1875{ 1876 struct filedesc *tmp; 1877 struct proc *p = td->td_proc; 1878 1879 if (p->p_fd->fd_refcnt == 1) 1880 return; 1881 1882 tmp = fdcopy(p->p_fd); 1883 fdescfree(td); 1884 p->p_fd = tmp; 1885} 1886 1887/* 1888 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1889 * this is to ease callers, not catch errors. 1890 */ 1891struct filedesc * 1892fdcopy(struct filedesc *fdp) 1893{ 1894 struct filedesc *newfdp; 1895 struct filedescent *nfde, *ofde; 1896 int i; 1897 1898 /* Certain daemons might not have file descriptors. */ 1899 if (fdp == NULL) 1900 return (NULL); 1901 1902 newfdp = fdinit(fdp); 1903 FILEDESC_SLOCK(fdp); 1904 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1905 FILEDESC_SUNLOCK(fdp); 1906 FILEDESC_XLOCK(newfdp); 1907 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1908 FILEDESC_XUNLOCK(newfdp); 1909 FILEDESC_SLOCK(fdp); 1910 } 1911 /* copy all passable descriptors (i.e. not kqueue) */ 1912 newfdp->fd_freefile = -1; 1913 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1914 ofde = &fdp->fd_ofiles[i]; 1915 if (ofde->fde_file != NULL && 1916 ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) { 1917 nfde = &newfdp->fd_ofiles[i]; 1918 *nfde = *ofde; 1919 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1920 fhold(nfde->fde_file); 1921 newfdp->fd_lastfile = i; 1922 } else { 1923 if (newfdp->fd_freefile == -1) 1924 newfdp->fd_freefile = i; 1925 } 1926 } 1927 newfdp->fd_cmask = fdp->fd_cmask; 1928 FILEDESC_SUNLOCK(fdp); 1929 FILEDESC_XLOCK(newfdp); 1930 for (i = 0; i <= newfdp->fd_lastfile; ++i) { 1931 if (newfdp->fd_ofiles[i].fde_file != NULL) 1932 fdused(newfdp, i); 1933 } 1934 if (newfdp->fd_freefile == -1) 1935 newfdp->fd_freefile = i; 1936 FILEDESC_XUNLOCK(newfdp); 1937 return (newfdp); 1938} 1939 1940/* 1941 * Release a filedesc structure. 1942 */ 1943void 1944fdescfree(struct thread *td) 1945{ 1946 struct filedesc *fdp; 1947 int i; 1948 struct filedesc_to_leader *fdtol; 1949 struct filedescent *fde; 1950 struct file *fp; 1951 struct vnode *cdir, *jdir, *rdir, *vp; 1952 struct flock lf; 1953 1954 /* Certain daemons might not have file descriptors. */ 1955 fdp = td->td_proc->p_fd; 1956 if (fdp == NULL) 1957 return; 1958 1959#ifdef RACCT 1960 PROC_LOCK(td->td_proc); 1961 racct_set(td->td_proc, RACCT_NOFILE, 0); 1962 PROC_UNLOCK(td->td_proc); 1963#endif 1964 1965 /* Check for special need to clear POSIX style locks */ 1966 fdtol = td->td_proc->p_fdtol; 1967 if (fdtol != NULL) { 1968 FILEDESC_XLOCK(fdp); 1969 KASSERT(fdtol->fdl_refcount > 0, 1970 ("filedesc_to_refcount botch: fdl_refcount=%d", 1971 fdtol->fdl_refcount)); 1972 if (fdtol->fdl_refcount == 1 && 1973 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1974 for (i = 0; i <= fdp->fd_lastfile; i++) { 1975 fp = fdp->fd_ofiles[i].fde_file; 1976 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1977 continue; 1978 fhold(fp); 1979 FILEDESC_XUNLOCK(fdp); 1980 lf.l_whence = SEEK_SET; 1981 lf.l_start = 0; 1982 lf.l_len = 0; 1983 lf.l_type = F_UNLCK; 1984 vp = fp->f_vnode; 1985 (void) VOP_ADVLOCK(vp, 1986 (caddr_t)td->td_proc->p_leader, F_UNLCK, 1987 &lf, F_POSIX); 1988 FILEDESC_XLOCK(fdp); 1989 fdrop(fp, td); 1990 } 1991 } 1992 retry: 1993 if (fdtol->fdl_refcount == 1) { 1994 if (fdp->fd_holdleaderscount > 0 && 1995 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1996 /* 1997 * close() or do_dup() has cleared a reference 1998 * in a shared file descriptor table. 1999 */ 2000 fdp->fd_holdleaderswakeup = 1; 2001 sx_sleep(&fdp->fd_holdleaderscount, 2002 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2003 goto retry; 2004 } 2005 if (fdtol->fdl_holdcount > 0) { 2006 /* 2007 * Ensure that fdtol->fdl_leader remains 2008 * valid in closef(). 2009 */ 2010 fdtol->fdl_wakeup = 1; 2011 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2012 "fdlhold", 0); 2013 goto retry; 2014 } 2015 } 2016 fdtol->fdl_refcount--; 2017 if (fdtol->fdl_refcount == 0 && 2018 fdtol->fdl_holdcount == 0) { 2019 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2020 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2021 } else 2022 fdtol = NULL; 2023 td->td_proc->p_fdtol = NULL; 2024 FILEDESC_XUNLOCK(fdp); 2025 if (fdtol != NULL) 2026 free(fdtol, M_FILEDESC_TO_LEADER); 2027 } 2028 2029 mtx_lock(&fdesc_mtx); 2030 td->td_proc->p_fd = NULL; 2031 mtx_unlock(&fdesc_mtx); 2032 2033 FILEDESC_XLOCK(fdp); 2034 i = --fdp->fd_refcnt; 2035 if (i > 0) { 2036 FILEDESC_XUNLOCK(fdp); 2037 return; 2038 } 2039 2040 cdir = fdp->fd_cdir; 2041 fdp->fd_cdir = NULL; 2042 rdir = fdp->fd_rdir; 2043 fdp->fd_rdir = NULL; 2044 jdir = fdp->fd_jdir; 2045 fdp->fd_jdir = NULL; 2046 FILEDESC_XUNLOCK(fdp); 2047 2048 for (i = 0; i <= fdp->fd_lastfile; i++) { 2049 fde = &fdp->fd_ofiles[i]; 2050 fp = fde->fde_file; 2051 if (fp != NULL) { 2052 fdefree_last(fde); 2053 (void) closef(fp, td); 2054 } 2055 } 2056 2057 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2058 free(fdp->fd_map, M_FILEDESC); 2059 if (fdp->fd_nfiles > NDFILE) 2060 free(fdp->fd_files, M_FILEDESC); 2061 2062 if (cdir != NULL) 2063 vrele(cdir); 2064 if (rdir != NULL) 2065 vrele(rdir); 2066 if (jdir != NULL) 2067 vrele(jdir); 2068 2069 fddrop(fdp); 2070} 2071 2072/* 2073 * For setugid programs, we don't want to people to use that setugidness 2074 * to generate error messages which write to a file which otherwise would 2075 * otherwise be off-limits to the process. We check for filesystems where 2076 * the vnode can change out from under us after execve (like [lin]procfs). 2077 * 2078 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2079 * sufficient. We also don't check for setugidness since we know we are. 2080 */ 2081static bool 2082is_unsafe(struct file *fp) 2083{ 2084 struct vnode *vp; 2085 2086 if (fp->f_type != DTYPE_VNODE) 2087 return (false); 2088 2089 vp = fp->f_vnode; 2090 return ((vp->v_vflag & VV_PROCDEP) != 0); 2091} 2092 2093/* 2094 * Make this setguid thing safe, if at all possible. 2095 */ 2096void 2097fdsetugidsafety(struct thread *td) 2098{ 2099 struct filedesc *fdp; 2100 struct file *fp; 2101 int i; 2102 2103 fdp = td->td_proc->p_fd; 2104 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2105 MPASS(fdp->fd_nfiles >= 3); 2106 for (i = 0; i <= 2; i++) { 2107 fp = fdp->fd_ofiles[i].fde_file; 2108 if (fp != NULL && is_unsafe(fp)) { 2109 FILEDESC_XLOCK(fdp); 2110 knote_fdclose(td, i); 2111 /* 2112 * NULL-out descriptor prior to close to avoid 2113 * a race while close blocks. 2114 */ 2115 fdfree(fdp, i); 2116 FILEDESC_XUNLOCK(fdp); 2117 (void) closef(fp, td); 2118 } 2119 } 2120} 2121 2122/* 2123 * If a specific file object occupies a specific file descriptor, close the 2124 * file descriptor entry and drop a reference on the file object. This is a 2125 * convenience function to handle a subsequent error in a function that calls 2126 * falloc() that handles the race that another thread might have closed the 2127 * file descriptor out from under the thread creating the file object. 2128 */ 2129void 2130fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2131{ 2132 2133 FILEDESC_XLOCK(fdp); 2134 if (fdp->fd_ofiles[idx].fde_file == fp) { 2135 fdfree(fdp, idx); 2136 FILEDESC_XUNLOCK(fdp); 2137 fdrop(fp, td); 2138 } else 2139 FILEDESC_XUNLOCK(fdp); 2140} 2141 2142/* 2143 * Close any files on exec? 2144 */ 2145void 2146fdcloseexec(struct thread *td) 2147{ 2148 struct filedesc *fdp; 2149 struct filedescent *fde; 2150 struct file *fp; 2151 int i; 2152 2153 fdp = td->td_proc->p_fd; 2154 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2155 FILEDESC_XLOCK(fdp); 2156 for (i = 0; i <= fdp->fd_lastfile; i++) { 2157 fde = &fdp->fd_ofiles[i]; 2158 fp = fde->fde_file; 2159 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2160 (fde->fde_flags & UF_EXCLOSE))) { 2161 fdfree(fdp, i); 2162 (void) closefp(fdp, i, fp, td, 0); 2163 /* closefp() drops the FILEDESC lock. */ 2164 FILEDESC_XLOCK(fdp); 2165 } 2166 } 2167 FILEDESC_XUNLOCK(fdp); 2168} 2169 2170/* 2171 * It is unsafe for set[ug]id processes to be started with file 2172 * descriptors 0..2 closed, as these descriptors are given implicit 2173 * significance in the Standard C library. fdcheckstd() will create a 2174 * descriptor referencing /dev/null for each of stdin, stdout, and 2175 * stderr that is not already open. 2176 */ 2177int 2178fdcheckstd(struct thread *td) 2179{ 2180 struct filedesc *fdp; 2181 register_t retval, save; 2182 int i, error, devnull; 2183 2184 fdp = td->td_proc->p_fd; 2185 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2186 devnull = -1; 2187 error = 0; 2188 for (i = 0; i < 3; i++) { 2189 if (fdp->fd_ofiles[i].fde_file != NULL) 2190 continue; 2191 if (devnull < 0) { 2192 save = td->td_retval[0]; 2193 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2194 O_RDWR, 0); 2195 devnull = td->td_retval[0]; 2196 td->td_retval[0] = save; 2197 if (error) 2198 break; 2199 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2200 } else { 2201 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2202 if (error != 0) 2203 break; 2204 } 2205 } 2206 return (error); 2207} 2208 2209/* 2210 * Internal form of close. Decrement reference count on file structure. 2211 * Note: td may be NULL when closing a file that was being passed in a 2212 * message. 2213 * 2214 * XXXRW: Giant is not required for the caller, but often will be held; this 2215 * makes it moderately likely the Giant will be recursed in the VFS case. 2216 */ 2217int 2218closef(struct file *fp, struct thread *td) 2219{ 2220 struct vnode *vp; 2221 struct flock lf; 2222 struct filedesc_to_leader *fdtol; 2223 struct filedesc *fdp; 2224 2225 /* 2226 * POSIX record locking dictates that any close releases ALL 2227 * locks owned by this process. This is handled by setting 2228 * a flag in the unlock to free ONLY locks obeying POSIX 2229 * semantics, and not to free BSD-style file locks. 2230 * If the descriptor was in a message, POSIX-style locks 2231 * aren't passed with the descriptor, and the thread pointer 2232 * will be NULL. Callers should be careful only to pass a 2233 * NULL thread pointer when there really is no owning 2234 * context that might have locks, or the locks will be 2235 * leaked. 2236 */ 2237 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2238 vp = fp->f_vnode; 2239 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2240 lf.l_whence = SEEK_SET; 2241 lf.l_start = 0; 2242 lf.l_len = 0; 2243 lf.l_type = F_UNLCK; 2244 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2245 F_UNLCK, &lf, F_POSIX); 2246 } 2247 fdtol = td->td_proc->p_fdtol; 2248 if (fdtol != NULL) { 2249 /* 2250 * Handle special case where file descriptor table is 2251 * shared between multiple process leaders. 2252 */ 2253 fdp = td->td_proc->p_fd; 2254 FILEDESC_XLOCK(fdp); 2255 for (fdtol = fdtol->fdl_next; 2256 fdtol != td->td_proc->p_fdtol; 2257 fdtol = fdtol->fdl_next) { 2258 if ((fdtol->fdl_leader->p_flag & 2259 P_ADVLOCK) == 0) 2260 continue; 2261 fdtol->fdl_holdcount++; 2262 FILEDESC_XUNLOCK(fdp); 2263 lf.l_whence = SEEK_SET; 2264 lf.l_start = 0; 2265 lf.l_len = 0; 2266 lf.l_type = F_UNLCK; 2267 vp = fp->f_vnode; 2268 (void) VOP_ADVLOCK(vp, 2269 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2270 F_POSIX); 2271 FILEDESC_XLOCK(fdp); 2272 fdtol->fdl_holdcount--; 2273 if (fdtol->fdl_holdcount == 0 && 2274 fdtol->fdl_wakeup != 0) { 2275 fdtol->fdl_wakeup = 0; 2276 wakeup(fdtol); 2277 } 2278 } 2279 FILEDESC_XUNLOCK(fdp); 2280 } 2281 } 2282 return (fdrop(fp, td)); 2283} 2284 2285/* 2286 * Initialize the file pointer with the specified properties. 2287 * 2288 * The ops are set with release semantics to be certain that the flags, type, 2289 * and data are visible when ops is. This is to prevent ops methods from being 2290 * called with bad data. 2291 */ 2292void 2293finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2294{ 2295 fp->f_data = data; 2296 fp->f_flag = flag; 2297 fp->f_type = type; 2298 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2299} 2300 2301int 2302fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2303 int needfcntl, struct file **fpp, cap_rights_t *haverightsp) 2304{ 2305#ifdef CAPABILITIES 2306 struct filedescent fde; 2307#endif 2308 struct fdescenttbl *fdt; 2309 struct file *fp; 2310 u_int count; 2311#ifdef CAPABILITIES 2312 seq_t seq; 2313 cap_rights_t haverights; 2314 int error; 2315#endif 2316 2317 fdt = fdp->fd_files; 2318 if (fd < 0 || fd >= fdt->fdt_nfiles) 2319 return (EBADF); 2320 /* 2321 * Fetch the descriptor locklessly. We avoid fdrop() races by 2322 * never raising a refcount above 0. To accomplish this we have 2323 * to use a cmpset loop rather than an atomic_add. The descriptor 2324 * must be re-verified once we acquire a reference to be certain 2325 * that the identity is still correct and we did not lose a race 2326 * due to preemption. 2327 */ 2328 for (;;) { 2329#ifdef CAPABILITIES 2330 seq = seq_read(fd_seq(fdt, fd)); 2331 fde = fdt->fdt_ofiles[fd]; 2332 if (!seq_consistent(fd_seq(fdt, fd), seq)) { 2333 cpu_spinwait(); 2334 continue; 2335 } 2336 fp = fde.fde_file; 2337#else 2338 fp = fdt->fdt_ofiles[fd].fde_file; 2339#endif 2340 if (fp == NULL) 2341 return (EBADF); 2342#ifdef CAPABILITIES 2343 haverights = *cap_rights_fde(&fde); 2344 if (needrightsp != NULL) { 2345 error = cap_check(&haverights, needrightsp); 2346 if (error != 0) 2347 return (error); 2348 if (cap_rights_is_set(needrightsp, CAP_FCNTL)) { 2349 error = cap_fcntl_check_fde(&fde, needfcntl); 2350 if (error != 0) 2351 return (error); 2352 } 2353 } 2354#endif 2355 retry: 2356 count = fp->f_count; 2357 if (count == 0) { 2358 /* 2359 * Force a reload. Other thread could reallocate the 2360 * table before this fd was closed, so it possible that 2361 * there is a stale fp pointer in cached version. 2362 */ 2363 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); 2364 continue; 2365 } 2366 /* 2367 * Use an acquire barrier to force re-reading of fdt so it is 2368 * refreshed for verification. 2369 */ 2370 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) 2371 goto retry; 2372 fdt = fdp->fd_files; 2373#ifdef CAPABILITIES 2374 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2375#else 2376 if (fp == fdt->fdt_ofiles[fd].fde_file) 2377#endif 2378 break; 2379 fdrop(fp, curthread); 2380 } 2381 *fpp = fp; 2382 if (haverightsp != NULL) { 2383#ifdef CAPABILITIES 2384 *haverightsp = haverights; 2385#else 2386 CAP_ALL(haverightsp); 2387#endif 2388 } 2389 return (0); 2390} 2391 2392/* 2393 * Extract the file pointer associated with the specified descriptor for the 2394 * current user process. 2395 * 2396 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2397 * returned. 2398 * 2399 * File's rights will be checked against the capability rights mask. 2400 * 2401 * If an error occured the non-zero error is returned and *fpp is set to 2402 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2403 * responsible for fdrop(). 2404 */ 2405static __inline int 2406_fget(struct thread *td, int fd, struct file **fpp, int flags, 2407 cap_rights_t *needrightsp, u_char *maxprotp) 2408{ 2409 struct filedesc *fdp; 2410 struct file *fp; 2411 cap_rights_t haverights, needrights; 2412 int error; 2413 2414 *fpp = NULL; 2415 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2416 return (EBADF); 2417 if (needrightsp != NULL) 2418 needrights = *needrightsp; 2419 else 2420 cap_rights_init(&needrights); 2421 if (maxprotp != NULL) 2422 cap_rights_set(&needrights, CAP_MMAP); 2423 error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights); 2424 if (error != 0) 2425 return (error); 2426 if (fp->f_ops == &badfileops) { 2427 fdrop(fp, td); 2428 return (EBADF); 2429 } 2430 2431#ifdef CAPABILITIES 2432 /* 2433 * If requested, convert capability rights to access flags. 2434 */ 2435 if (maxprotp != NULL) 2436 *maxprotp = cap_rights_to_vmprot(&haverights); 2437#else /* !CAPABILITIES */ 2438 if (maxprotp != NULL) 2439 *maxprotp = VM_PROT_ALL; 2440#endif /* CAPABILITIES */ 2441 2442 /* 2443 * FREAD and FWRITE failure return EBADF as per POSIX. 2444 */ 2445 error = 0; 2446 switch (flags) { 2447 case FREAD: 2448 case FWRITE: 2449 if ((fp->f_flag & flags) == 0) 2450 error = EBADF; 2451 break; 2452 case FEXEC: 2453 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2454 ((fp->f_flag & FWRITE) != 0)) 2455 error = EBADF; 2456 break; 2457 case 0: 2458 break; 2459 default: 2460 KASSERT(0, ("wrong flags")); 2461 } 2462 2463 if (error != 0) { 2464 fdrop(fp, td); 2465 return (error); 2466 } 2467 2468 *fpp = fp; 2469 return (0); 2470} 2471 2472int 2473fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2474{ 2475 2476 return(_fget(td, fd, fpp, 0, rightsp, NULL)); 2477} 2478 2479int 2480fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2481 struct file **fpp) 2482{ 2483 2484 return (_fget(td, fd, fpp, 0, rightsp, maxprotp)); 2485} 2486 2487int 2488fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2489{ 2490 2491 return(_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2492} 2493 2494int 2495fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2496{ 2497 2498 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2499} 2500 2501/* 2502 * Like fget() but loads the underlying vnode, or returns an error if the 2503 * descriptor does not represent a vnode. Note that pipes use vnodes but 2504 * never have VM objects. The returned vnode will be vref()'d. 2505 * 2506 * XXX: what about the unused flags ? 2507 */ 2508static __inline int 2509_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2510 struct vnode **vpp) 2511{ 2512 struct file *fp; 2513 int error; 2514 2515 *vpp = NULL; 2516 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2517 if (error != 0) 2518 return (error); 2519 if (fp->f_vnode == NULL) { 2520 error = EINVAL; 2521 } else { 2522 *vpp = fp->f_vnode; 2523 vref(*vpp); 2524 } 2525 fdrop(fp, td); 2526 2527 return (error); 2528} 2529 2530int 2531fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2532{ 2533 2534 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2535} 2536 2537int 2538fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2539 struct filecaps *havecaps, struct vnode **vpp) 2540{ 2541 struct filedesc *fdp; 2542 struct file *fp; 2543#ifdef CAPABILITIES 2544 int error; 2545#endif 2546 2547 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2548 return (EBADF); 2549 2550 fp = fget_locked(fdp, fd); 2551 if (fp == NULL || fp->f_ops == &badfileops) 2552 return (EBADF); 2553 2554#ifdef CAPABILITIES 2555 if (needrightsp != NULL) { 2556 error = cap_check(cap_rights(fdp, fd), needrightsp); 2557 if (error != 0) 2558 return (error); 2559 } 2560#endif 2561 2562 if (fp->f_vnode == NULL) 2563 return (EINVAL); 2564 2565 *vpp = fp->f_vnode; 2566 vref(*vpp); 2567 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2568 2569 return (0); 2570} 2571 2572int 2573fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2574{ 2575 2576 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2577} 2578 2579int 2580fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2581{ 2582 2583 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2584} 2585 2586#ifdef notyet 2587int 2588fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2589 struct vnode **vpp) 2590{ 2591 2592 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2593} 2594#endif 2595 2596/* 2597 * Like fget() but loads the underlying socket, or returns an error if the 2598 * descriptor does not represent a socket. 2599 * 2600 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2601 * in the future. 2602 * 2603 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2604 * on their file descriptor reference to prevent the socket from being free'd 2605 * during use. 2606 */ 2607int 2608fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2609 u_int *fflagp) 2610{ 2611 struct file *fp; 2612 int error; 2613 2614 *spp = NULL; 2615 if (fflagp != NULL) 2616 *fflagp = 0; 2617 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2618 return (error); 2619 if (fp->f_type != DTYPE_SOCKET) { 2620 error = ENOTSOCK; 2621 } else { 2622 *spp = fp->f_data; 2623 if (fflagp) 2624 *fflagp = fp->f_flag; 2625 SOCK_LOCK(*spp); 2626 soref(*spp); 2627 SOCK_UNLOCK(*spp); 2628 } 2629 fdrop(fp, td); 2630 2631 return (error); 2632} 2633 2634/* 2635 * Drop the reference count on the socket and XXX release the SX lock in the 2636 * future. The last reference closes the socket. 2637 * 2638 * Note: fputsock() is deprecated, see comment for fgetsock(). 2639 */ 2640void 2641fputsock(struct socket *so) 2642{ 2643 2644 ACCEPT_LOCK(); 2645 SOCK_LOCK(so); 2646 CURVNET_SET(so->so_vnet); 2647 sorele(so); 2648 CURVNET_RESTORE(); 2649} 2650 2651/* 2652 * Handle the last reference to a file being closed. 2653 */ 2654int 2655_fdrop(struct file *fp, struct thread *td) 2656{ 2657 int error; 2658 2659 error = 0; 2660 if (fp->f_count != 0) 2661 panic("fdrop: count %d", fp->f_count); 2662 if (fp->f_ops != &badfileops) 2663 error = fo_close(fp, td); 2664 atomic_subtract_int(&openfiles, 1); 2665 crfree(fp->f_cred); 2666 free(fp->f_advice, M_FADVISE); 2667 uma_zfree(file_zone, fp); 2668 2669 return (error); 2670} 2671 2672/* 2673 * Apply an advisory lock on a file descriptor. 2674 * 2675 * Just attempt to get a record lock of the requested type on the entire file 2676 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2677 */ 2678#ifndef _SYS_SYSPROTO_H_ 2679struct flock_args { 2680 int fd; 2681 int how; 2682}; 2683#endif 2684/* ARGSUSED */ 2685int 2686sys_flock(struct thread *td, struct flock_args *uap) 2687{ 2688 struct file *fp; 2689 struct vnode *vp; 2690 struct flock lf; 2691 cap_rights_t rights; 2692 int error; 2693 2694 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2695 if (error != 0) 2696 return (error); 2697 if (fp->f_type != DTYPE_VNODE) { 2698 fdrop(fp, td); 2699 return (EOPNOTSUPP); 2700 } 2701 2702 vp = fp->f_vnode; 2703 lf.l_whence = SEEK_SET; 2704 lf.l_start = 0; 2705 lf.l_len = 0; 2706 if (uap->how & LOCK_UN) { 2707 lf.l_type = F_UNLCK; 2708 atomic_clear_int(&fp->f_flag, FHASLOCK); 2709 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2710 goto done2; 2711 } 2712 if (uap->how & LOCK_EX) 2713 lf.l_type = F_WRLCK; 2714 else if (uap->how & LOCK_SH) 2715 lf.l_type = F_RDLCK; 2716 else { 2717 error = EBADF; 2718 goto done2; 2719 } 2720 atomic_set_int(&fp->f_flag, FHASLOCK); 2721 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2722 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2723done2: 2724 fdrop(fp, td); 2725 return (error); 2726} 2727/* 2728 * Duplicate the specified descriptor to a free descriptor. 2729 */ 2730int 2731dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2732 int openerror, int *indxp) 2733{ 2734 struct filedescent *newfde, *oldfde; 2735 struct file *fp; 2736 int error, indx; 2737 2738 KASSERT(openerror == ENODEV || openerror == ENXIO, 2739 ("unexpected error %d in %s", openerror, __func__)); 2740 2741 /* 2742 * If the to-be-dup'd fd number is greater than the allowed number 2743 * of file descriptors, or the fd to be dup'd has already been 2744 * closed, then reject. 2745 */ 2746 FILEDESC_XLOCK(fdp); 2747 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2748 FILEDESC_XUNLOCK(fdp); 2749 return (EBADF); 2750 } 2751 2752 error = fdalloc(td, 0, &indx); 2753 if (error != 0) { 2754 FILEDESC_XUNLOCK(fdp); 2755 return (error); 2756 } 2757 2758 /* 2759 * There are two cases of interest here. 2760 * 2761 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2762 * 2763 * For ENXIO steal away the file structure from (dfd) and store it in 2764 * (indx). (dfd) is effectively closed by this operation. 2765 */ 2766 switch (openerror) { 2767 case ENODEV: 2768 /* 2769 * Check that the mode the file is being opened for is a 2770 * subset of the mode of the existing descriptor. 2771 */ 2772 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2773 fdunused(fdp, indx); 2774 FILEDESC_XUNLOCK(fdp); 2775 return (EACCES); 2776 } 2777 fhold(fp); 2778 newfde = &fdp->fd_ofiles[indx]; 2779 oldfde = &fdp->fd_ofiles[dfd]; 2780#ifdef CAPABILITIES 2781 seq_write_begin(&newfde->fde_seq); 2782#endif 2783 memcpy(newfde, oldfde, fde_change_size); 2784 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 2785#ifdef CAPABILITIES 2786 seq_write_end(&newfde->fde_seq); 2787#endif 2788 break; 2789 case ENXIO: 2790 /* 2791 * Steal away the file pointer from dfd and stuff it into indx. 2792 */ 2793 newfde = &fdp->fd_ofiles[indx]; 2794 oldfde = &fdp->fd_ofiles[dfd]; 2795#ifdef CAPABILITIES 2796 seq_write_begin(&newfde->fde_seq); 2797#endif 2798 memcpy(newfde, oldfde, fde_change_size); 2799 bzero(oldfde, fde_change_size); 2800 fdunused(fdp, dfd); 2801#ifdef CAPABILITIES 2802 seq_write_end(&newfde->fde_seq); 2803#endif 2804 break; 2805 } 2806 FILEDESC_XUNLOCK(fdp); 2807 *indxp = indx; 2808 return (0); 2809} 2810 2811/* 2812 * Scan all active processes and prisons to see if any of them have a current 2813 * or root directory of `olddp'. If so, replace them with the new mount point. 2814 */ 2815void 2816mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2817{ 2818 struct filedesc *fdp; 2819 struct prison *pr; 2820 struct proc *p; 2821 int nrele; 2822 2823 if (vrefcnt(olddp) == 1) 2824 return; 2825 nrele = 0; 2826 sx_slock(&allproc_lock); 2827 FOREACH_PROC_IN_SYSTEM(p) { 2828 fdp = fdhold(p); 2829 if (fdp == NULL) 2830 continue; 2831 FILEDESC_XLOCK(fdp); 2832 if (fdp->fd_cdir == olddp) { 2833 vref(newdp); 2834 fdp->fd_cdir = newdp; 2835 nrele++; 2836 } 2837 if (fdp->fd_rdir == olddp) { 2838 vref(newdp); 2839 fdp->fd_rdir = newdp; 2840 nrele++; 2841 } 2842 if (fdp->fd_jdir == olddp) { 2843 vref(newdp); 2844 fdp->fd_jdir = newdp; 2845 nrele++; 2846 } 2847 FILEDESC_XUNLOCK(fdp); 2848 fddrop(fdp); 2849 } 2850 sx_sunlock(&allproc_lock); 2851 if (rootvnode == olddp) { 2852 vref(newdp); 2853 rootvnode = newdp; 2854 nrele++; 2855 } 2856 mtx_lock(&prison0.pr_mtx); 2857 if (prison0.pr_root == olddp) { 2858 vref(newdp); 2859 prison0.pr_root = newdp; 2860 nrele++; 2861 } 2862 mtx_unlock(&prison0.pr_mtx); 2863 sx_slock(&allprison_lock); 2864 TAILQ_FOREACH(pr, &allprison, pr_list) { 2865 mtx_lock(&pr->pr_mtx); 2866 if (pr->pr_root == olddp) { 2867 vref(newdp); 2868 pr->pr_root = newdp; 2869 nrele++; 2870 } 2871 mtx_unlock(&pr->pr_mtx); 2872 } 2873 sx_sunlock(&allprison_lock); 2874 while (nrele--) 2875 vrele(olddp); 2876} 2877 2878struct filedesc_to_leader * 2879filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2880{ 2881 struct filedesc_to_leader *fdtol; 2882 2883 fdtol = malloc(sizeof(struct filedesc_to_leader), 2884 M_FILEDESC_TO_LEADER, 2885 M_WAITOK); 2886 fdtol->fdl_refcount = 1; 2887 fdtol->fdl_holdcount = 0; 2888 fdtol->fdl_wakeup = 0; 2889 fdtol->fdl_leader = leader; 2890 if (old != NULL) { 2891 FILEDESC_XLOCK(fdp); 2892 fdtol->fdl_next = old->fdl_next; 2893 fdtol->fdl_prev = old; 2894 old->fdl_next = fdtol; 2895 fdtol->fdl_next->fdl_prev = fdtol; 2896 FILEDESC_XUNLOCK(fdp); 2897 } else { 2898 fdtol->fdl_next = fdtol; 2899 fdtol->fdl_prev = fdtol; 2900 } 2901 return (fdtol); 2902} 2903 2904/* 2905 * Get file structures globally. 2906 */ 2907static int 2908sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2909{ 2910 struct xfile xf; 2911 struct filedesc *fdp; 2912 struct file *fp; 2913 struct proc *p; 2914 int error, n; 2915 2916 error = sysctl_wire_old_buffer(req, 0); 2917 if (error != 0) 2918 return (error); 2919 if (req->oldptr == NULL) { 2920 n = 0; 2921 sx_slock(&allproc_lock); 2922 FOREACH_PROC_IN_SYSTEM(p) { 2923 if (p->p_state == PRS_NEW) 2924 continue; 2925 fdp = fdhold(p); 2926 if (fdp == NULL) 2927 continue; 2928 /* overestimates sparse tables. */ 2929 if (fdp->fd_lastfile > 0) 2930 n += fdp->fd_lastfile; 2931 fddrop(fdp); 2932 } 2933 sx_sunlock(&allproc_lock); 2934 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2935 } 2936 error = 0; 2937 bzero(&xf, sizeof(xf)); 2938 xf.xf_size = sizeof(xf); 2939 sx_slock(&allproc_lock); 2940 FOREACH_PROC_IN_SYSTEM(p) { 2941 PROC_LOCK(p); 2942 if (p->p_state == PRS_NEW) { 2943 PROC_UNLOCK(p); 2944 continue; 2945 } 2946 if (p_cansee(req->td, p) != 0) { 2947 PROC_UNLOCK(p); 2948 continue; 2949 } 2950 xf.xf_pid = p->p_pid; 2951 xf.xf_uid = p->p_ucred->cr_uid; 2952 PROC_UNLOCK(p); 2953 fdp = fdhold(p); 2954 if (fdp == NULL) 2955 continue; 2956 FILEDESC_SLOCK(fdp); 2957 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 2958 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 2959 continue; 2960 xf.xf_fd = n; 2961 xf.xf_file = fp; 2962 xf.xf_data = fp->f_data; 2963 xf.xf_vnode = fp->f_vnode; 2964 xf.xf_type = fp->f_type; 2965 xf.xf_count = fp->f_count; 2966 xf.xf_msgcount = 0; 2967 xf.xf_offset = foffset_get(fp); 2968 xf.xf_flag = fp->f_flag; 2969 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2970 if (error) 2971 break; 2972 } 2973 FILEDESC_SUNLOCK(fdp); 2974 fddrop(fdp); 2975 if (error) 2976 break; 2977 } 2978 sx_sunlock(&allproc_lock); 2979 return (error); 2980} 2981 2982SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 2983 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2984 2985#ifdef KINFO_FILE_SIZE 2986CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2987#endif 2988 2989static int 2990xlate_fflags(int fflags) 2991{ 2992 static const struct { 2993 int fflag; 2994 int kf_fflag; 2995 } fflags_table[] = { 2996 { FAPPEND, KF_FLAG_APPEND }, 2997 { FASYNC, KF_FLAG_ASYNC }, 2998 { FFSYNC, KF_FLAG_FSYNC }, 2999 { FHASLOCK, KF_FLAG_HASLOCK }, 3000 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3001 { FREAD, KF_FLAG_READ }, 3002 { FWRITE, KF_FLAG_WRITE }, 3003 { O_CREAT, KF_FLAG_CREAT }, 3004 { O_DIRECT, KF_FLAG_DIRECT }, 3005 { O_EXCL, KF_FLAG_EXCL }, 3006 { O_EXEC, KF_FLAG_EXEC }, 3007 { O_EXLOCK, KF_FLAG_EXLOCK }, 3008 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3009 { O_SHLOCK, KF_FLAG_SHLOCK }, 3010 { O_TRUNC, KF_FLAG_TRUNC } 3011 }; 3012 unsigned int i; 3013 int kflags; 3014 3015 kflags = 0; 3016 for (i = 0; i < nitems(fflags_table); i++) 3017 if (fflags & fflags_table[i].fflag) 3018 kflags |= fflags_table[i].kf_fflag; 3019 return (kflags); 3020} 3021 3022/* Trim unused data from kf_path by truncating the structure size. */ 3023static void 3024pack_kinfo(struct kinfo_file *kif) 3025{ 3026 3027 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3028 strlen(kif->kf_path) + 1; 3029 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3030} 3031 3032static void 3033export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3034 struct kinfo_file *kif, struct filedesc *fdp) 3035{ 3036 int error; 3037 3038 bzero(kif, sizeof(*kif)); 3039 3040 /* Set a default type to allow for empty fill_kinfo() methods. */ 3041 kif->kf_type = KF_TYPE_UNKNOWN; 3042 kif->kf_flags = xlate_fflags(fp->f_flag); 3043 if (rightsp != NULL) 3044 kif->kf_cap_rights = *rightsp; 3045 else 3046 cap_rights_init(&kif->kf_cap_rights); 3047 kif->kf_fd = fd; 3048 kif->kf_ref_count = fp->f_count; 3049 kif->kf_offset = foffset_get(fp); 3050 3051 /* 3052 * This may drop the filedesc lock, so the 'fp' cannot be 3053 * accessed after this call. 3054 */ 3055 error = fo_fill_kinfo(fp, kif, fdp); 3056 if (error == 0) 3057 kif->kf_status |= KF_ATTR_VALID; 3058 pack_kinfo(kif); 3059} 3060 3061static void 3062export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3063 struct kinfo_file *kif) 3064{ 3065 int error; 3066 3067 bzero(kif, sizeof(*kif)); 3068 3069 kif->kf_type = KF_TYPE_VNODE; 3070 error = vn_fill_kinfo_vnode(vp, kif); 3071 if (error == 0) 3072 kif->kf_status |= KF_ATTR_VALID; 3073 kif->kf_flags = xlate_fflags(fflags); 3074 kif->kf_fd = fd; 3075 kif->kf_ref_count = -1; 3076 kif->kf_offset = -1; 3077 pack_kinfo(kif); 3078 vrele(vp); 3079} 3080 3081struct export_fd_buf { 3082 struct filedesc *fdp; 3083 struct sbuf *sb; 3084 ssize_t remainder; 3085 struct kinfo_file kif; 3086}; 3087 3088static int 3089export_kinfo_to_sb(struct export_fd_buf *efbuf) 3090{ 3091 struct kinfo_file *kif; 3092 3093 kif = &efbuf->kif; 3094 if (efbuf->remainder != -1) { 3095 if (efbuf->remainder < kif->kf_structsize) { 3096 /* Terminate export. */ 3097 efbuf->remainder = 0; 3098 return (0); 3099 } 3100 efbuf->remainder -= kif->kf_structsize; 3101 } 3102 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3103} 3104 3105static int 3106export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3107 struct export_fd_buf *efbuf) 3108{ 3109 int error; 3110 3111 if (efbuf->remainder == 0) 3112 return (0); 3113 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp); 3114 FILEDESC_SUNLOCK(efbuf->fdp); 3115 error = export_kinfo_to_sb(efbuf); 3116 FILEDESC_SLOCK(efbuf->fdp); 3117 return (error); 3118} 3119 3120static int 3121export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3122 struct export_fd_buf *efbuf) 3123{ 3124 int error; 3125 3126 if (efbuf->remainder == 0) 3127 return (0); 3128 if (efbuf->fdp != NULL) 3129 FILEDESC_SUNLOCK(efbuf->fdp); 3130 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif); 3131 error = export_kinfo_to_sb(efbuf); 3132 if (efbuf->fdp != NULL) 3133 FILEDESC_SLOCK(efbuf->fdp); 3134 return (error); 3135} 3136 3137/* 3138 * Store a process file descriptor information to sbuf. 3139 * 3140 * Takes a locked proc as argument, and returns with the proc unlocked. 3141 */ 3142int 3143kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3144{ 3145 struct file *fp; 3146 struct filedesc *fdp; 3147 struct export_fd_buf *efbuf; 3148 struct vnode *cttyvp, *textvp, *tracevp; 3149 int error, i; 3150 cap_rights_t rights; 3151 3152 PROC_LOCK_ASSERT(p, MA_OWNED); 3153 3154 /* ktrace vnode */ 3155 tracevp = p->p_tracevp; 3156 if (tracevp != NULL) 3157 vref(tracevp); 3158 /* text vnode */ 3159 textvp = p->p_textvp; 3160 if (textvp != NULL) 3161 vref(textvp); 3162 /* Controlling tty. */ 3163 cttyvp = NULL; 3164 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3165 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3166 if (cttyvp != NULL) 3167 vref(cttyvp); 3168 } 3169 fdp = fdhold(p); 3170 PROC_UNLOCK(p); 3171 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3172 efbuf->fdp = NULL; 3173 efbuf->sb = sb; 3174 efbuf->remainder = maxlen; 3175 if (tracevp != NULL) 3176 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3177 efbuf); 3178 if (textvp != NULL) 3179 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3180 if (cttyvp != NULL) 3181 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3182 efbuf); 3183 error = 0; 3184 if (fdp == NULL) 3185 goto fail; 3186 efbuf->fdp = fdp; 3187 FILEDESC_SLOCK(fdp); 3188 /* working directory */ 3189 if (fdp->fd_cdir != NULL) { 3190 vref(fdp->fd_cdir); 3191 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3192 } 3193 /* root directory */ 3194 if (fdp->fd_rdir != NULL) { 3195 vref(fdp->fd_rdir); 3196 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3197 } 3198 /* jail directory */ 3199 if (fdp->fd_jdir != NULL) { 3200 vref(fdp->fd_jdir); 3201 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3202 } 3203 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3204 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3205 continue; 3206#ifdef CAPABILITIES 3207 rights = *cap_rights(fdp, i); 3208#else /* !CAPABILITIES */ 3209 cap_rights_init(&rights); 3210#endif 3211 /* 3212 * Create sysctl entry. It is OK to drop the filedesc 3213 * lock inside of export_file_to_sb() as we will 3214 * re-validate and re-evaluate its properties when the 3215 * loop continues. 3216 */ 3217 error = export_file_to_sb(fp, i, &rights, efbuf); 3218 if (error != 0 || efbuf->remainder == 0) 3219 break; 3220 } 3221 FILEDESC_SUNLOCK(fdp); 3222 fddrop(fdp); 3223fail: 3224 free(efbuf, M_TEMP); 3225 return (error); 3226} 3227 3228#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3229 3230/* 3231 * Get per-process file descriptors for use by procstat(1), et al. 3232 */ 3233static int 3234sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3235{ 3236 struct sbuf sb; 3237 struct proc *p; 3238 ssize_t maxlen; 3239 int error, error2, *name; 3240 3241 name = (int *)arg1; 3242 3243 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3244 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3245 if (error != 0) { 3246 sbuf_delete(&sb); 3247 return (error); 3248 } 3249 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3250 error = kern_proc_filedesc_out(p, &sb, maxlen); 3251 error2 = sbuf_finish(&sb); 3252 sbuf_delete(&sb); 3253 return (error != 0 ? error : error2); 3254} 3255 3256#ifdef KINFO_OFILE_SIZE 3257CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3258#endif 3259 3260#ifdef COMPAT_FREEBSD7 3261static void 3262kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3263{ 3264 3265 okif->kf_structsize = sizeof(*okif); 3266 okif->kf_type = kif->kf_type; 3267 okif->kf_fd = kif->kf_fd; 3268 okif->kf_ref_count = kif->kf_ref_count; 3269 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3270 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3271 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3272 okif->kf_offset = kif->kf_offset; 3273 okif->kf_vnode_type = kif->kf_vnode_type; 3274 okif->kf_sock_domain = kif->kf_sock_domain; 3275 okif->kf_sock_type = kif->kf_sock_type; 3276 okif->kf_sock_protocol = kif->kf_sock_protocol; 3277 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3278 okif->kf_sa_local = kif->kf_sa_local; 3279 okif->kf_sa_peer = kif->kf_sa_peer; 3280} 3281 3282static int 3283export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3284 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3285{ 3286 int error; 3287 3288 vref(vp); 3289 FILEDESC_SUNLOCK(fdp); 3290 export_vnode_to_kinfo(vp, type, 0, kif); 3291 kinfo_to_okinfo(kif, okif); 3292 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3293 FILEDESC_SLOCK(fdp); 3294 return (error); 3295} 3296 3297/* 3298 * Get per-process file descriptors for use by procstat(1), et al. 3299 */ 3300static int 3301sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3302{ 3303 struct kinfo_ofile *okif; 3304 struct kinfo_file *kif; 3305 struct filedesc *fdp; 3306 int error, i, *name; 3307 struct file *fp; 3308 struct proc *p; 3309 3310 name = (int *)arg1; 3311 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3312 if (error != 0) 3313 return (error); 3314 fdp = fdhold(p); 3315 PROC_UNLOCK(p); 3316 if (fdp == NULL) 3317 return (ENOENT); 3318 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3319 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3320 FILEDESC_SLOCK(fdp); 3321 if (fdp->fd_cdir != NULL) 3322 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3323 okif, fdp, req); 3324 if (fdp->fd_rdir != NULL) 3325 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3326 okif, fdp, req); 3327 if (fdp->fd_jdir != NULL) 3328 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3329 okif, fdp, req); 3330 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3331 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3332 continue; 3333 export_file_to_kinfo(fp, i, NULL, kif, fdp); 3334 FILEDESC_SUNLOCK(fdp); 3335 kinfo_to_okinfo(kif, okif); 3336 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3337 FILEDESC_SLOCK(fdp); 3338 if (error) 3339 break; 3340 } 3341 FILEDESC_SUNLOCK(fdp); 3342 fddrop(fdp); 3343 free(kif, M_TEMP); 3344 free(okif, M_TEMP); 3345 return (0); 3346} 3347 3348static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3349 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3350 "Process ofiledesc entries"); 3351#endif /* COMPAT_FREEBSD7 */ 3352 3353int 3354vntype_to_kinfo(int vtype) 3355{ 3356 struct { 3357 int vtype; 3358 int kf_vtype; 3359 } vtypes_table[] = { 3360 { VBAD, KF_VTYPE_VBAD }, 3361 { VBLK, KF_VTYPE_VBLK }, 3362 { VCHR, KF_VTYPE_VCHR }, 3363 { VDIR, KF_VTYPE_VDIR }, 3364 { VFIFO, KF_VTYPE_VFIFO }, 3365 { VLNK, KF_VTYPE_VLNK }, 3366 { VNON, KF_VTYPE_VNON }, 3367 { VREG, KF_VTYPE_VREG }, 3368 { VSOCK, KF_VTYPE_VSOCK } 3369 }; 3370 unsigned int i; 3371 3372 /* 3373 * Perform vtype translation. 3374 */ 3375 for (i = 0; i < nitems(vtypes_table); i++) 3376 if (vtypes_table[i].vtype == vtype) 3377 return (vtypes_table[i].kf_vtype); 3378 3379 return (KF_VTYPE_UNKNOWN); 3380} 3381 3382static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3383 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3384 "Process filedesc entries"); 3385 3386#ifdef DDB 3387/* 3388 * For the purposes of debugging, generate a human-readable string for the 3389 * file type. 3390 */ 3391static const char * 3392file_type_to_name(short type) 3393{ 3394 3395 switch (type) { 3396 case 0: 3397 return ("zero"); 3398 case DTYPE_VNODE: 3399 return ("vnod"); 3400 case DTYPE_SOCKET: 3401 return ("sock"); 3402 case DTYPE_PIPE: 3403 return ("pipe"); 3404 case DTYPE_FIFO: 3405 return ("fifo"); 3406 case DTYPE_KQUEUE: 3407 return ("kque"); 3408 case DTYPE_CRYPTO: 3409 return ("crpt"); 3410 case DTYPE_MQUEUE: 3411 return ("mque"); 3412 case DTYPE_SHM: 3413 return ("shm"); 3414 case DTYPE_SEM: 3415 return ("ksem"); 3416 default: 3417 return ("unkn"); 3418 } 3419} 3420 3421/* 3422 * For the purposes of debugging, identify a process (if any, perhaps one of 3423 * many) that references the passed file in its file descriptor array. Return 3424 * NULL if none. 3425 */ 3426static struct proc * 3427file_to_first_proc(struct file *fp) 3428{ 3429 struct filedesc *fdp; 3430 struct proc *p; 3431 int n; 3432 3433 FOREACH_PROC_IN_SYSTEM(p) { 3434 if (p->p_state == PRS_NEW) 3435 continue; 3436 fdp = p->p_fd; 3437 if (fdp == NULL) 3438 continue; 3439 for (n = 0; n <= fdp->fd_lastfile; n++) { 3440 if (fp == fdp->fd_ofiles[n].fde_file) 3441 return (p); 3442 } 3443 } 3444 return (NULL); 3445} 3446 3447static void 3448db_print_file(struct file *fp, int header) 3449{ 3450 struct proc *p; 3451 3452 if (header) 3453 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3454 "File", "Type", "Data", "Flag", "GCFl", "Count", 3455 "MCount", "Vnode", "FPID", "FCmd"); 3456 p = file_to_first_proc(fp); 3457 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3458 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3459 0, fp->f_count, 0, fp->f_vnode, 3460 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3461} 3462 3463DB_SHOW_COMMAND(file, db_show_file) 3464{ 3465 struct file *fp; 3466 3467 if (!have_addr) { 3468 db_printf("usage: show file <addr>\n"); 3469 return; 3470 } 3471 fp = (struct file *)addr; 3472 db_print_file(fp, 1); 3473} 3474 3475DB_SHOW_COMMAND(files, db_show_files) 3476{ 3477 struct filedesc *fdp; 3478 struct file *fp; 3479 struct proc *p; 3480 int header; 3481 int n; 3482 3483 header = 1; 3484 FOREACH_PROC_IN_SYSTEM(p) { 3485 if (p->p_state == PRS_NEW) 3486 continue; 3487 if ((fdp = p->p_fd) == NULL) 3488 continue; 3489 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3490 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3491 continue; 3492 db_print_file(fp, header); 3493 header = 0; 3494 } 3495 } 3496} 3497#endif 3498 3499SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3500 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3501 3502SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3503 &maxfiles, 0, "Maximum number of files"); 3504 3505SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3506 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3507 3508/* ARGSUSED*/ 3509static void 3510filelistinit(void *dummy) 3511{ 3512 3513 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3514 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3515 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3516 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3517} 3518SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3519 3520/*-------------------------------------------------------------------*/ 3521 3522static int 3523badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3524 int flags, struct thread *td) 3525{ 3526 3527 return (EBADF); 3528} 3529 3530static int 3531badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3532 struct thread *td) 3533{ 3534 3535 return (EINVAL); 3536} 3537 3538static int 3539badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3540 struct thread *td) 3541{ 3542 3543 return (EBADF); 3544} 3545 3546static int 3547badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3548 struct thread *td) 3549{ 3550 3551 return (0); 3552} 3553 3554static int 3555badfo_kqfilter(struct file *fp, struct knote *kn) 3556{ 3557 3558 return (EBADF); 3559} 3560 3561static int 3562badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3563 struct thread *td) 3564{ 3565 3566 return (EBADF); 3567} 3568 3569static int 3570badfo_close(struct file *fp, struct thread *td) 3571{ 3572 3573 return (EBADF); 3574} 3575 3576static int 3577badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3578 struct thread *td) 3579{ 3580 3581 return (EBADF); 3582} 3583 3584static int 3585badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3586 struct thread *td) 3587{ 3588 3589 return (EBADF); 3590} 3591 3592static int 3593badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3594 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3595 int kflags, struct sendfile_sync *sfs, struct thread *td) 3596{ 3597 3598 return (EBADF); 3599} 3600 3601static int 3602badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 3603{ 3604 3605 return (0); 3606} 3607 3608struct fileops badfileops = { 3609 .fo_read = badfo_readwrite, 3610 .fo_write = badfo_readwrite, 3611 .fo_truncate = badfo_truncate, 3612 .fo_ioctl = badfo_ioctl, 3613 .fo_poll = badfo_poll, 3614 .fo_kqfilter = badfo_kqfilter, 3615 .fo_stat = badfo_stat, 3616 .fo_close = badfo_close, 3617 .fo_chmod = badfo_chmod, 3618 .fo_chown = badfo_chown, 3619 .fo_sendfile = badfo_sendfile, 3620 .fo_fill_kinfo = badfo_fill_kinfo, 3621}; 3622 3623int 3624invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 3625 int flags, struct thread *td) 3626{ 3627 3628 return (EOPNOTSUPP); 3629} 3630 3631int 3632invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3633 struct thread *td) 3634{ 3635 3636 return (EINVAL); 3637} 3638 3639int 3640invfo_ioctl(struct file *fp, u_long com, void *data, 3641 struct ucred *active_cred, struct thread *td) 3642{ 3643 3644 return (ENOTTY); 3645} 3646 3647int 3648invfo_poll(struct file *fp, int events, struct ucred *active_cred, 3649 struct thread *td) 3650{ 3651 3652 return (poll_no_poll(events)); 3653} 3654 3655int 3656invfo_kqfilter(struct file *fp, struct knote *kn) 3657{ 3658 3659 return (EINVAL); 3660} 3661 3662int 3663invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3664 struct thread *td) 3665{ 3666 3667 return (EINVAL); 3668} 3669 3670int 3671invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3672 struct thread *td) 3673{ 3674 3675 return (EINVAL); 3676} 3677 3678int 3679invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3680 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3681 int kflags, struct sendfile_sync *sfs, struct thread *td) 3682{ 3683 3684 return (EINVAL); 3685} 3686 3687/*-------------------------------------------------------------------*/ 3688 3689/* 3690 * File Descriptor pseudo-device driver (/dev/fd/). 3691 * 3692 * Opening minor device N dup()s the file (if any) connected to file 3693 * descriptor N belonging to the calling process. Note that this driver 3694 * consists of only the ``open()'' routine, because all subsequent 3695 * references to this file will be direct to the other driver. 3696 * 3697 * XXX: we could give this one a cloning event handler if necessary. 3698 */ 3699 3700/* ARGSUSED */ 3701static int 3702fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3703{ 3704 3705 /* 3706 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3707 * the file descriptor being sought for duplication. The error 3708 * return ensures that the vnode for this device will be released 3709 * by vn_open. Open will detect this special error and take the 3710 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3711 * will simply report the error. 3712 */ 3713 td->td_dupfd = dev2unit(dev); 3714 return (ENODEV); 3715} 3716 3717static struct cdevsw fildesc_cdevsw = { 3718 .d_version = D_VERSION, 3719 .d_open = fdopen, 3720 .d_name = "FD", 3721}; 3722 3723static void 3724fildesc_drvinit(void *unused) 3725{ 3726 struct cdev *dev; 3727 3728 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3729 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3730 make_dev_alias(dev, "stdin"); 3731 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3732 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3733 make_dev_alias(dev, "stdout"); 3734 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3735 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3736 make_dev_alias(dev, "stderr"); 3737} 3738 3739SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3740