kern_descrip.c revision 285321
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 285321 2015-07-09 15:19:45Z 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; 101static uma_zone_t filedesc0_zone; 102 103static int closefp(struct filedesc *fdp, int fd, struct file *fp, 104 struct thread *td, int holdleaders); 105static int kern_dup(struct thread *td, int flags, int old, int new); 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 thread *td); 113 114/* Flags for kern_dup() */ 115#define FDDUP_FIXED 0x1 /* Force fixed allocation. */ 116#define FDDUP_FCNTL 0x2 /* fcntl()-style errors. */ 117#define FDDUP_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/* 182 * If low >= size, just return low. Otherwise find the first zero bit in the 183 * given bitmap, starting at low and not exceeding size - 1. Return size if 184 * not found. 185 */ 186static int 187fd_first_free(struct filedesc *fdp, int low, int size) 188{ 189 NDSLOTTYPE *map = fdp->fd_map; 190 NDSLOTTYPE mask; 191 int off, maxoff; 192 193 if (low >= size) 194 return (low); 195 196 off = NDSLOT(low); 197 if (low % NDENTRIES) { 198 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 199 if ((mask &= ~map[off]) != 0UL) 200 return (off * NDENTRIES + ffsl(mask) - 1); 201 ++off; 202 } 203 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 204 if (map[off] != ~0UL) 205 return (off * NDENTRIES + ffsl(~map[off]) - 1); 206 return (size); 207} 208 209/* 210 * Find the highest non-zero bit in the given bitmap, starting at 0 and 211 * not exceeding size - 1. Return -1 if not found. 212 */ 213static int 214fd_last_used(struct filedesc *fdp, int size) 215{ 216 NDSLOTTYPE *map = fdp->fd_map; 217 NDSLOTTYPE mask; 218 int off, minoff; 219 220 off = NDSLOT(size); 221 if (size % NDENTRIES) { 222 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 223 if ((mask &= map[off]) != 0) 224 return (off * NDENTRIES + flsl(mask) - 1); 225 --off; 226 } 227 for (minoff = NDSLOT(0); off >= minoff; --off) 228 if (map[off] != 0) 229 return (off * NDENTRIES + flsl(map[off]) - 1); 230 return (-1); 231} 232 233#ifdef INVARIANTS 234static int 235fdisused(struct filedesc *fdp, int fd) 236{ 237 238 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 239 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 240 241 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 242} 243#endif 244 245/* 246 * Mark a file descriptor as used. 247 */ 248static void 249fdused_init(struct filedesc *fdp, int fd) 250{ 251 252 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 253 254 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 255} 256 257static void 258fdused(struct filedesc *fdp, int fd) 259{ 260 261 FILEDESC_XLOCK_ASSERT(fdp); 262 263 fdused_init(fdp, fd); 264 if (fd > fdp->fd_lastfile) 265 fdp->fd_lastfile = fd; 266 if (fd == fdp->fd_freefile) 267 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 268} 269 270/* 271 * Mark a file descriptor as unused. 272 */ 273static void 274fdunused(struct filedesc *fdp, int fd) 275{ 276 277 FILEDESC_XLOCK_ASSERT(fdp); 278 279 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 280 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 281 ("fd=%d is still in use", fd)); 282 283 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 284 if (fd < fdp->fd_freefile) 285 fdp->fd_freefile = fd; 286 if (fd == fdp->fd_lastfile) 287 fdp->fd_lastfile = fd_last_used(fdp, fd); 288} 289 290/* 291 * Free a file descriptor. 292 * 293 * Avoid some work if fdp is about to be destroyed. 294 */ 295static inline void 296fdefree_last(struct filedescent *fde) 297{ 298 299 filecaps_free(&fde->fde_caps); 300} 301 302static inline void 303fdfree(struct filedesc *fdp, int fd) 304{ 305 struct filedescent *fde; 306 307 fde = &fdp->fd_ofiles[fd]; 308#ifdef CAPABILITIES 309 seq_write_begin(&fde->fde_seq); 310#endif 311 fdefree_last(fde); 312 fde->fde_file = NULL; 313 fdunused(fdp, fd); 314#ifdef CAPABILITIES 315 seq_write_end(&fde->fde_seq); 316#endif 317} 318 319/* 320 * System calls on descriptors. 321 */ 322#ifndef _SYS_SYSPROTO_H_ 323struct getdtablesize_args { 324 int dummy; 325}; 326#endif 327/* ARGSUSED */ 328int 329sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 330{ 331#ifdef RACCT 332 uint64_t lim; 333#endif 334 335 td->td_retval[0] = 336 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc); 337#ifdef RACCT 338 PROC_LOCK(td->td_proc); 339 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 340 PROC_UNLOCK(td->td_proc); 341 if (lim < td->td_retval[0]) 342 td->td_retval[0] = lim; 343#endif 344 return (0); 345} 346 347/* 348 * Duplicate a file descriptor to a particular value. 349 * 350 * Note: keep in mind that a potential race condition exists when closing 351 * descriptors from a shared descriptor table (via rfork). 352 */ 353#ifndef _SYS_SYSPROTO_H_ 354struct dup2_args { 355 u_int from; 356 u_int to; 357}; 358#endif 359/* ARGSUSED */ 360int 361sys_dup2(struct thread *td, struct dup2_args *uap) 362{ 363 364 return (kern_dup(td, FDDUP_FIXED, (int)uap->from, (int)uap->to)); 365} 366 367/* 368 * Duplicate a file descriptor. 369 */ 370#ifndef _SYS_SYSPROTO_H_ 371struct dup_args { 372 u_int fd; 373}; 374#endif 375/* ARGSUSED */ 376int 377sys_dup(struct thread *td, struct dup_args *uap) 378{ 379 380 return (kern_dup(td, 0, (int)uap->fd, 0)); 381} 382 383/* 384 * The file control system call. 385 */ 386#ifndef _SYS_SYSPROTO_H_ 387struct fcntl_args { 388 int fd; 389 int cmd; 390 long arg; 391}; 392#endif 393/* ARGSUSED */ 394int 395sys_fcntl(struct thread *td, struct fcntl_args *uap) 396{ 397 398 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 399} 400 401int 402kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 403{ 404 struct flock fl; 405 struct __oflock ofl; 406 intptr_t arg1; 407 int error, newcmd; 408 409 error = 0; 410 newcmd = cmd; 411 switch (cmd) { 412 case F_OGETLK: 413 case F_OSETLK: 414 case F_OSETLKW: 415 /* 416 * Convert old flock structure to new. 417 */ 418 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 419 fl.l_start = ofl.l_start; 420 fl.l_len = ofl.l_len; 421 fl.l_pid = ofl.l_pid; 422 fl.l_type = ofl.l_type; 423 fl.l_whence = ofl.l_whence; 424 fl.l_sysid = 0; 425 426 switch (cmd) { 427 case F_OGETLK: 428 newcmd = F_GETLK; 429 break; 430 case F_OSETLK: 431 newcmd = F_SETLK; 432 break; 433 case F_OSETLKW: 434 newcmd = F_SETLKW; 435 break; 436 } 437 arg1 = (intptr_t)&fl; 438 break; 439 case F_GETLK: 440 case F_SETLK: 441 case F_SETLKW: 442 case F_SETLK_REMOTE: 443 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 444 arg1 = (intptr_t)&fl; 445 break; 446 default: 447 arg1 = arg; 448 break; 449 } 450 if (error) 451 return (error); 452 error = kern_fcntl(td, fd, newcmd, arg1); 453 if (error) 454 return (error); 455 if (cmd == F_OGETLK) { 456 ofl.l_start = fl.l_start; 457 ofl.l_len = fl.l_len; 458 ofl.l_pid = fl.l_pid; 459 ofl.l_type = fl.l_type; 460 ofl.l_whence = fl.l_whence; 461 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 462 } else if (cmd == F_GETLK) { 463 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 464 } 465 return (error); 466} 467 468int 469kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 470{ 471 struct filedesc *fdp; 472 struct flock *flp; 473 struct file *fp, *fp2; 474 struct filedescent *fde; 475 struct proc *p; 476 struct vnode *vp; 477 cap_rights_t rights; 478 int error, flg, tmp; 479 uint64_t bsize; 480 off_t foffset; 481 482 error = 0; 483 flg = F_POSIX; 484 p = td->td_proc; 485 fdp = p->p_fd; 486 487 switch (cmd) { 488 case F_DUPFD: 489 tmp = arg; 490 error = kern_dup(td, FDDUP_FCNTL, fd, tmp); 491 break; 492 493 case F_DUPFD_CLOEXEC: 494 tmp = arg; 495 error = kern_dup(td, FDDUP_FCNTL | FDDUP_CLOEXEC, fd, tmp); 496 break; 497 498 case F_DUP2FD: 499 tmp = arg; 500 error = kern_dup(td, FDDUP_FIXED, fd, tmp); 501 break; 502 503 case F_DUP2FD_CLOEXEC: 504 tmp = arg; 505 error = kern_dup(td, FDDUP_FIXED | FDDUP_CLOEXEC, fd, tmp); 506 break; 507 508 case F_GETFD: 509 FILEDESC_SLOCK(fdp); 510 if (fget_locked(fdp, fd) == NULL) { 511 FILEDESC_SUNLOCK(fdp); 512 error = EBADF; 513 break; 514 } 515 fde = &fdp->fd_ofiles[fd]; 516 td->td_retval[0] = 517 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 518 FILEDESC_SUNLOCK(fdp); 519 break; 520 521 case F_SETFD: 522 FILEDESC_XLOCK(fdp); 523 if (fget_locked(fdp, fd) == NULL) { 524 FILEDESC_XUNLOCK(fdp); 525 error = EBADF; 526 break; 527 } 528 fde = &fdp->fd_ofiles[fd]; 529 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 530 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 531 FILEDESC_XUNLOCK(fdp); 532 break; 533 534 case F_GETFL: 535 error = fget_fcntl(td, fd, 536 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp); 537 if (error != 0) 538 break; 539 td->td_retval[0] = OFLAGS(fp->f_flag); 540 fdrop(fp, td); 541 break; 542 543 case F_SETFL: 544 error = fget_fcntl(td, fd, 545 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp); 546 if (error != 0) 547 break; 548 do { 549 tmp = flg = fp->f_flag; 550 tmp &= ~FCNTLFLAGS; 551 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 552 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 553 tmp = fp->f_flag & FNONBLOCK; 554 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 555 if (error != 0) { 556 fdrop(fp, td); 557 break; 558 } 559 tmp = fp->f_flag & FASYNC; 560 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 561 if (error == 0) { 562 fdrop(fp, td); 563 break; 564 } 565 atomic_clear_int(&fp->f_flag, FNONBLOCK); 566 tmp = 0; 567 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 568 fdrop(fp, td); 569 break; 570 571 case F_GETOWN: 572 error = fget_fcntl(td, fd, 573 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp); 574 if (error != 0) 575 break; 576 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 577 if (error == 0) 578 td->td_retval[0] = tmp; 579 fdrop(fp, td); 580 break; 581 582 case F_SETOWN: 583 error = fget_fcntl(td, fd, 584 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp); 585 if (error != 0) 586 break; 587 tmp = arg; 588 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 589 fdrop(fp, td); 590 break; 591 592 case F_SETLK_REMOTE: 593 error = priv_check(td, PRIV_NFS_LOCKD); 594 if (error) 595 return (error); 596 flg = F_REMOTE; 597 goto do_setlk; 598 599 case F_SETLKW: 600 flg |= F_WAIT; 601 /* FALLTHROUGH F_SETLK */ 602 603 case F_SETLK: 604 do_setlk: 605 cap_rights_init(&rights, CAP_FLOCK); 606 error = fget_unlocked(fdp, fd, &rights, &fp, NULL); 607 if (error != 0) 608 break; 609 if (fp->f_type != DTYPE_VNODE) { 610 error = EBADF; 611 fdrop(fp, td); 612 break; 613 } 614 615 flp = (struct flock *)arg; 616 if (flp->l_whence == SEEK_CUR) { 617 foffset = foffset_get(fp); 618 if (foffset < 0 || 619 (flp->l_start > 0 && 620 foffset > OFF_MAX - flp->l_start)) { 621 error = EOVERFLOW; 622 fdrop(fp, td); 623 break; 624 } 625 flp->l_start += foffset; 626 } 627 628 vp = fp->f_vnode; 629 switch (flp->l_type) { 630 case F_RDLCK: 631 if ((fp->f_flag & FREAD) == 0) { 632 error = EBADF; 633 break; 634 } 635 PROC_LOCK(p->p_leader); 636 p->p_leader->p_flag |= P_ADVLOCK; 637 PROC_UNLOCK(p->p_leader); 638 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 639 flp, flg); 640 break; 641 case F_WRLCK: 642 if ((fp->f_flag & FWRITE) == 0) { 643 error = EBADF; 644 break; 645 } 646 PROC_LOCK(p->p_leader); 647 p->p_leader->p_flag |= P_ADVLOCK; 648 PROC_UNLOCK(p->p_leader); 649 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 650 flp, flg); 651 break; 652 case F_UNLCK: 653 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 654 flp, flg); 655 break; 656 case F_UNLCKSYS: 657 /* 658 * Temporary api for testing remote lock 659 * infrastructure. 660 */ 661 if (flg != F_REMOTE) { 662 error = EINVAL; 663 break; 664 } 665 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 666 F_UNLCKSYS, flp, flg); 667 break; 668 default: 669 error = EINVAL; 670 break; 671 } 672 if (error != 0 || flp->l_type == F_UNLCK || 673 flp->l_type == F_UNLCKSYS) { 674 fdrop(fp, td); 675 break; 676 } 677 678 /* 679 * Check for a race with close. 680 * 681 * The vnode is now advisory locked (or unlocked, but this case 682 * is not really important) as the caller requested. 683 * We had to drop the filedesc lock, so we need to recheck if 684 * the descriptor is still valid, because if it was closed 685 * in the meantime we need to remove advisory lock from the 686 * vnode - close on any descriptor leading to an advisory 687 * locked vnode, removes that lock. 688 * We will return 0 on purpose in that case, as the result of 689 * successful advisory lock might have been externally visible 690 * already. This is fine - effectively we pretend to the caller 691 * that the closing thread was a bit slower and that the 692 * advisory lock succeeded before the close. 693 */ 694 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL); 695 if (error != 0) { 696 fdrop(fp, td); 697 break; 698 } 699 if (fp != fp2) { 700 flp->l_whence = SEEK_SET; 701 flp->l_start = 0; 702 flp->l_len = 0; 703 flp->l_type = F_UNLCK; 704 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 705 F_UNLCK, flp, F_POSIX); 706 } 707 fdrop(fp, td); 708 fdrop(fp2, td); 709 break; 710 711 case F_GETLK: 712 error = fget_unlocked(fdp, fd, 713 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL); 714 if (error != 0) 715 break; 716 if (fp->f_type != DTYPE_VNODE) { 717 error = EBADF; 718 fdrop(fp, td); 719 break; 720 } 721 flp = (struct flock *)arg; 722 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 723 flp->l_type != F_UNLCK) { 724 error = EINVAL; 725 fdrop(fp, td); 726 break; 727 } 728 if (flp->l_whence == SEEK_CUR) { 729 foffset = foffset_get(fp); 730 if ((flp->l_start > 0 && 731 foffset > OFF_MAX - flp->l_start) || 732 (flp->l_start < 0 && 733 foffset < OFF_MIN - flp->l_start)) { 734 error = EOVERFLOW; 735 fdrop(fp, td); 736 break; 737 } 738 flp->l_start += foffset; 739 } 740 vp = fp->f_vnode; 741 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 742 F_POSIX); 743 fdrop(fp, td); 744 break; 745 746 case F_RDAHEAD: 747 arg = arg ? 128 * 1024: 0; 748 /* FALLTHROUGH */ 749 case F_READAHEAD: 750 error = fget_unlocked(fdp, fd, 751 cap_rights_init(&rights), &fp, NULL); 752 if (error != 0) 753 break; 754 if (fp->f_type != DTYPE_VNODE) { 755 fdrop(fp, td); 756 error = EBADF; 757 break; 758 } 759 vp = fp->f_vnode; 760 /* 761 * Exclusive lock synchronizes against f_seqcount reads and 762 * writes in sequential_heuristic(). 763 */ 764 error = vn_lock(vp, LK_EXCLUSIVE); 765 if (error != 0) { 766 fdrop(fp, td); 767 break; 768 } 769 if (arg >= 0) { 770 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 771 fp->f_seqcount = (arg + bsize - 1) / bsize; 772 atomic_set_int(&fp->f_flag, FRDAHEAD); 773 } else { 774 atomic_clear_int(&fp->f_flag, FRDAHEAD); 775 } 776 VOP_UNLOCK(vp, 0); 777 fdrop(fp, td); 778 break; 779 780 default: 781 error = EINVAL; 782 break; 783 } 784 return (error); 785} 786 787static int 788getmaxfd(struct thread *td) 789{ 790 791 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 792} 793 794/* 795 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 796 */ 797static int 798kern_dup(struct thread *td, int flags, int old, int new) 799{ 800 struct filedesc *fdp; 801 struct filedescent *oldfde, *newfde; 802 struct proc *p; 803 struct file *fp; 804 struct file *delfp; 805 int error, maxfd; 806 807 p = td->td_proc; 808 fdp = p->p_fd; 809 810 MPASS((flags & ~(FDDUP_FIXED | FDDUP_FCNTL | FDDUP_CLOEXEC)) == 0); 811 812 /* 813 * Verify we have a valid descriptor to dup from and possibly to 814 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 815 * return EINVAL when the new descriptor is out of bounds. 816 */ 817 if (old < 0) 818 return (EBADF); 819 if (new < 0) 820 return (flags & FDDUP_FCNTL ? EINVAL : EBADF); 821 maxfd = getmaxfd(td); 822 if (new >= maxfd) 823 return (flags & FDDUP_FCNTL ? EINVAL : EBADF); 824 825 FILEDESC_XLOCK(fdp); 826 if (fget_locked(fdp, old) == NULL) { 827 FILEDESC_XUNLOCK(fdp); 828 return (EBADF); 829 } 830 oldfde = &fdp->fd_ofiles[old]; 831 if (flags & FDDUP_FIXED && old == new) { 832 td->td_retval[0] = new; 833 if (flags & FDDUP_CLOEXEC) 834 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 835 FILEDESC_XUNLOCK(fdp); 836 return (0); 837 } 838 fp = oldfde->fde_file; 839 fhold(fp); 840 841 /* 842 * If the caller specified a file descriptor, make sure the file 843 * table is large enough to hold it, and grab it. Otherwise, just 844 * allocate a new descriptor the usual way. 845 */ 846 if (flags & FDDUP_FIXED) { 847 if (new >= fdp->fd_nfiles) { 848 /* 849 * The resource limits are here instead of e.g. 850 * fdalloc(), because the file descriptor table may be 851 * shared between processes, so we can't really use 852 * racct_add()/racct_sub(). Instead of counting the 853 * number of actually allocated descriptors, just put 854 * the limit on the size of the file descriptor table. 855 */ 856#ifdef RACCT 857 if (racct_enable) { 858 PROC_LOCK(p); 859 error = racct_set(p, RACCT_NOFILE, new + 1); 860 PROC_UNLOCK(p); 861 if (error != 0) { 862 FILEDESC_XUNLOCK(fdp); 863 fdrop(fp, td); 864 return (EMFILE); 865 } 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 & FDDUP_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 td->td_retval[0] = 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(struct thread *td, struct close_args *uap) 1198{ 1199 1200 return (kern_close(td, uap->fd)); 1201} 1202 1203int 1204kern_close(struct thread *td, int fd) 1205{ 1206 struct filedesc *fdp; 1207 struct file *fp; 1208 1209 fdp = td->td_proc->p_fd; 1210 1211 AUDIT_SYSCLOSE(td, fd); 1212 1213 FILEDESC_XLOCK(fdp); 1214 if ((fp = fget_locked(fdp, fd)) == NULL) { 1215 FILEDESC_XUNLOCK(fdp); 1216 return (EBADF); 1217 } 1218 fdfree(fdp, fd); 1219 1220 /* closefp() drops the FILEDESC lock for us. */ 1221 return (closefp(fdp, fd, fp, td, 1)); 1222} 1223 1224/* 1225 * Close open file descriptors. 1226 */ 1227#ifndef _SYS_SYSPROTO_H_ 1228struct closefrom_args { 1229 int lowfd; 1230}; 1231#endif 1232/* ARGSUSED */ 1233int 1234sys_closefrom(struct thread *td, struct closefrom_args *uap) 1235{ 1236 struct filedesc *fdp; 1237 int fd; 1238 1239 fdp = td->td_proc->p_fd; 1240 AUDIT_ARG_FD(uap->lowfd); 1241 1242 /* 1243 * Treat negative starting file descriptor values identical to 1244 * closefrom(0) which closes all files. 1245 */ 1246 if (uap->lowfd < 0) 1247 uap->lowfd = 0; 1248 FILEDESC_SLOCK(fdp); 1249 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1250 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1251 FILEDESC_SUNLOCK(fdp); 1252 (void)kern_close(td, fd); 1253 FILEDESC_SLOCK(fdp); 1254 } 1255 } 1256 FILEDESC_SUNLOCK(fdp); 1257 return (0); 1258} 1259 1260#if defined(COMPAT_43) 1261/* 1262 * Return status information about a file descriptor. 1263 */ 1264#ifndef _SYS_SYSPROTO_H_ 1265struct ofstat_args { 1266 int fd; 1267 struct ostat *sb; 1268}; 1269#endif 1270/* ARGSUSED */ 1271int 1272ofstat(struct thread *td, struct ofstat_args *uap) 1273{ 1274 struct ostat oub; 1275 struct stat ub; 1276 int error; 1277 1278 error = kern_fstat(td, uap->fd, &ub); 1279 if (error == 0) { 1280 cvtstat(&ub, &oub); 1281 error = copyout(&oub, uap->sb, sizeof(oub)); 1282 } 1283 return (error); 1284} 1285#endif /* COMPAT_43 */ 1286 1287/* 1288 * Return status information about a file descriptor. 1289 */ 1290#ifndef _SYS_SYSPROTO_H_ 1291struct fstat_args { 1292 int fd; 1293 struct stat *sb; 1294}; 1295#endif 1296/* ARGSUSED */ 1297int 1298sys_fstat(struct thread *td, struct fstat_args *uap) 1299{ 1300 struct stat ub; 1301 int error; 1302 1303 error = kern_fstat(td, uap->fd, &ub); 1304 if (error == 0) 1305 error = copyout(&ub, uap->sb, sizeof(ub)); 1306 return (error); 1307} 1308 1309int 1310kern_fstat(struct thread *td, int fd, struct stat *sbp) 1311{ 1312 struct file *fp; 1313 cap_rights_t rights; 1314 int error; 1315 1316 AUDIT_ARG_FD(fd); 1317 1318 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1319 if (error != 0) 1320 return (error); 1321 1322 AUDIT_ARG_FILE(td->td_proc, fp); 1323 1324 error = fo_stat(fp, sbp, td->td_ucred, td); 1325 fdrop(fp, td); 1326#ifdef KTRACE 1327 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1328 ktrstat(sbp); 1329#endif 1330 return (error); 1331} 1332 1333/* 1334 * Return status information about a file descriptor. 1335 */ 1336#ifndef _SYS_SYSPROTO_H_ 1337struct nfstat_args { 1338 int fd; 1339 struct nstat *sb; 1340}; 1341#endif 1342/* ARGSUSED */ 1343int 1344sys_nfstat(struct thread *td, struct nfstat_args *uap) 1345{ 1346 struct nstat nub; 1347 struct stat ub; 1348 int error; 1349 1350 error = kern_fstat(td, uap->fd, &ub); 1351 if (error == 0) { 1352 cvtnstat(&ub, &nub); 1353 error = copyout(&nub, uap->sb, sizeof(nub)); 1354 } 1355 return (error); 1356} 1357 1358/* 1359 * Return pathconf information about a file descriptor. 1360 */ 1361#ifndef _SYS_SYSPROTO_H_ 1362struct fpathconf_args { 1363 int fd; 1364 int name; 1365}; 1366#endif 1367/* ARGSUSED */ 1368int 1369sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1370{ 1371 struct file *fp; 1372 struct vnode *vp; 1373 cap_rights_t rights; 1374 int error; 1375 1376 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1377 if (error != 0) 1378 return (error); 1379 1380 /* If asynchronous I/O is available, it works for all descriptors. */ 1381 if (uap->name == _PC_ASYNC_IO) { 1382 td->td_retval[0] = async_io_version; 1383 goto out; 1384 } 1385 vp = fp->f_vnode; 1386 if (vp != NULL) { 1387 vn_lock(vp, LK_SHARED | LK_RETRY); 1388 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1389 VOP_UNLOCK(vp, 0); 1390 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1391 if (uap->name != _PC_PIPE_BUF) { 1392 error = EINVAL; 1393 } else { 1394 td->td_retval[0] = PIPE_BUF; 1395 error = 0; 1396 } 1397 } else { 1398 error = EOPNOTSUPP; 1399 } 1400out: 1401 fdrop(fp, td); 1402 return (error); 1403} 1404 1405/* 1406 * Initialize filecaps structure. 1407 */ 1408void 1409filecaps_init(struct filecaps *fcaps) 1410{ 1411 1412 bzero(fcaps, sizeof(*fcaps)); 1413 fcaps->fc_nioctls = -1; 1414} 1415 1416/* 1417 * Copy filecaps structure allocating memory for ioctls array if needed. 1418 */ 1419void 1420filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1421{ 1422 size_t size; 1423 1424 *dst = *src; 1425 if (src->fc_ioctls != NULL) { 1426 KASSERT(src->fc_nioctls > 0, 1427 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1428 1429 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1430 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1431 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1432 } 1433} 1434 1435/* 1436 * Move filecaps structure to the new place and clear the old place. 1437 */ 1438void 1439filecaps_move(struct filecaps *src, struct filecaps *dst) 1440{ 1441 1442 *dst = *src; 1443 bzero(src, sizeof(*src)); 1444} 1445 1446/* 1447 * Fill the given filecaps structure with full rights. 1448 */ 1449static void 1450filecaps_fill(struct filecaps *fcaps) 1451{ 1452 1453 CAP_ALL(&fcaps->fc_rights); 1454 fcaps->fc_ioctls = NULL; 1455 fcaps->fc_nioctls = -1; 1456 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1457} 1458 1459/* 1460 * Free memory allocated within filecaps structure. 1461 */ 1462void 1463filecaps_free(struct filecaps *fcaps) 1464{ 1465 1466 free(fcaps->fc_ioctls, M_FILECAPS); 1467 bzero(fcaps, sizeof(*fcaps)); 1468} 1469 1470/* 1471 * Validate the given filecaps structure. 1472 */ 1473static void 1474filecaps_validate(const struct filecaps *fcaps, const char *func) 1475{ 1476 1477 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1478 ("%s: invalid rights", func)); 1479 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1480 ("%s: invalid fcntls", func)); 1481 KASSERT(fcaps->fc_fcntls == 0 || 1482 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1483 ("%s: fcntls without CAP_FCNTL", func)); 1484 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1485 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1486 ("%s: invalid ioctls", func)); 1487 KASSERT(fcaps->fc_nioctls == 0 || 1488 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1489 ("%s: ioctls without CAP_IOCTL", func)); 1490} 1491 1492static void 1493fdgrowtable_exp(struct filedesc *fdp, int nfd) 1494{ 1495 int nfd1; 1496 1497 FILEDESC_XLOCK_ASSERT(fdp); 1498 1499 nfd1 = fdp->fd_nfiles * 2; 1500 if (nfd1 < nfd) 1501 nfd1 = nfd; 1502 fdgrowtable(fdp, nfd1); 1503} 1504 1505/* 1506 * Grow the file table to accomodate (at least) nfd descriptors. 1507 */ 1508static void 1509fdgrowtable(struct filedesc *fdp, int nfd) 1510{ 1511 struct filedesc0 *fdp0; 1512 struct freetable *ft; 1513 struct fdescenttbl *ntable; 1514 struct fdescenttbl *otable; 1515 int nnfiles, onfiles; 1516 NDSLOTTYPE *nmap, *omap; 1517 1518 /* 1519 * If lastfile is -1 this struct filedesc was just allocated and we are 1520 * growing it to accomodate for the one we are going to copy from. There 1521 * is no need to have a lock on this one as it's not visible to anyone. 1522 */ 1523 if (fdp->fd_lastfile != -1) 1524 FILEDESC_XLOCK_ASSERT(fdp); 1525 1526 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1527 1528 /* save old values */ 1529 onfiles = fdp->fd_nfiles; 1530 otable = fdp->fd_files; 1531 omap = fdp->fd_map; 1532 1533 /* compute the size of the new table */ 1534 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1535 if (nnfiles <= onfiles) 1536 /* the table is already large enough */ 1537 return; 1538 1539 /* 1540 * Allocate a new table. We need enough space for the number of 1541 * entries, file entries themselves and the struct freetable we will use 1542 * when we decommission the table and place it on the freelist. 1543 * We place the struct freetable in the middle so we don't have 1544 * to worry about padding. 1545 */ 1546 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1547 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1548 sizeof(struct freetable), 1549 M_FILEDESC, M_ZERO | M_WAITOK); 1550 /* copy the old data */ 1551 ntable->fdt_nfiles = nnfiles; 1552 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1553 onfiles * sizeof(ntable->fdt_ofiles[0])); 1554 1555 /* 1556 * Allocate a new map only if the old is not large enough. It will 1557 * grow at a slower rate than the table as it can map more 1558 * entries than the table can hold. 1559 */ 1560 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1561 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1562 M_ZERO | M_WAITOK); 1563 /* copy over the old data and update the pointer */ 1564 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1565 fdp->fd_map = nmap; 1566 } 1567 1568 /* 1569 * Make sure that ntable is correctly initialized before we replace 1570 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1571 * data. 1572 */ 1573 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1574 1575 /* 1576 * Do not free the old file table, as some threads may still 1577 * reference entries within it. Instead, place it on a freelist 1578 * which will be processed when the struct filedesc is released. 1579 * 1580 * Note that if onfiles == NDFILE, we're dealing with the original 1581 * static allocation contained within (struct filedesc0 *)fdp, 1582 * which must not be freed. 1583 */ 1584 if (onfiles > NDFILE) { 1585 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1586 fdp0 = (struct filedesc0 *)fdp; 1587 ft->ft_table = otable; 1588 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1589 } 1590 /* 1591 * The map does not have the same possibility of threads still 1592 * holding references to it. So always free it as long as it 1593 * does not reference the original static allocation. 1594 */ 1595 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1596 free(omap, M_FILEDESC); 1597} 1598 1599/* 1600 * Allocate a file descriptor for the process. 1601 */ 1602int 1603fdalloc(struct thread *td, int minfd, int *result) 1604{ 1605 struct proc *p = td->td_proc; 1606 struct filedesc *fdp = p->p_fd; 1607 int fd, maxfd, allocfd; 1608#ifdef RACCT 1609 int error; 1610#endif 1611 1612 FILEDESC_XLOCK_ASSERT(fdp); 1613 1614 if (fdp->fd_freefile > minfd) 1615 minfd = fdp->fd_freefile; 1616 1617 maxfd = getmaxfd(td); 1618 1619 /* 1620 * Search the bitmap for a free descriptor starting at minfd. 1621 * If none is found, grow the file table. 1622 */ 1623 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1624 if (fd >= maxfd) 1625 return (EMFILE); 1626 if (fd >= fdp->fd_nfiles) { 1627 allocfd = min(fd * 2, maxfd); 1628#ifdef RACCT 1629 if (racct_enable) { 1630 PROC_LOCK(p); 1631 error = racct_set(p, RACCT_NOFILE, allocfd); 1632 PROC_UNLOCK(p); 1633 if (error != 0) 1634 return (EMFILE); 1635 } 1636#endif 1637 /* 1638 * fd is already equal to first free descriptor >= minfd, so 1639 * we only need to grow the table and we are done. 1640 */ 1641 fdgrowtable_exp(fdp, allocfd); 1642 } 1643 1644 /* 1645 * Perform some sanity checks, then mark the file descriptor as 1646 * used and return it to the caller. 1647 */ 1648 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1649 ("invalid descriptor %d", fd)); 1650 KASSERT(!fdisused(fdp, fd), 1651 ("fd_first_free() returned non-free descriptor")); 1652 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1653 ("file descriptor isn't free")); 1654 fdused(fdp, fd); 1655 *result = fd; 1656 return (0); 1657} 1658 1659/* 1660 * Allocate n file descriptors for the process. 1661 */ 1662int 1663fdallocn(struct thread *td, int minfd, int *fds, int n) 1664{ 1665 struct proc *p = td->td_proc; 1666 struct filedesc *fdp = p->p_fd; 1667 int i; 1668 1669 FILEDESC_XLOCK_ASSERT(fdp); 1670 1671 for (i = 0; i < n; i++) 1672 if (fdalloc(td, 0, &fds[i]) != 0) 1673 break; 1674 1675 if (i < n) { 1676 for (i--; i >= 0; i--) 1677 fdunused(fdp, fds[i]); 1678 return (EMFILE); 1679 } 1680 1681 return (0); 1682} 1683 1684/* 1685 * Create a new open file structure and allocate a file decriptor for the 1686 * process that refers to it. We add one reference to the file for the 1687 * descriptor table and one reference for resultfp. This is to prevent us 1688 * being preempted and the entry in the descriptor table closed after we 1689 * release the FILEDESC lock. 1690 */ 1691int 1692falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1693{ 1694 struct file *fp; 1695 int error, fd; 1696 1697 error = falloc_noinstall(td, &fp); 1698 if (error) 1699 return (error); /* no reference held on error */ 1700 1701 error = finstall(td, fp, &fd, flags, NULL); 1702 if (error) { 1703 fdrop(fp, td); /* one reference (fp only) */ 1704 return (error); 1705 } 1706 1707 if (resultfp != NULL) 1708 *resultfp = fp; /* copy out result */ 1709 else 1710 fdrop(fp, td); /* release local reference */ 1711 1712 if (resultfd != NULL) 1713 *resultfd = fd; 1714 1715 return (0); 1716} 1717 1718/* 1719 * Create a new open file structure without allocating a file descriptor. 1720 */ 1721int 1722falloc_noinstall(struct thread *td, struct file **resultfp) 1723{ 1724 struct file *fp; 1725 int maxuserfiles = maxfiles - (maxfiles / 20); 1726 static struct timeval lastfail; 1727 static int curfail; 1728 1729 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1730 1731 if ((openfiles >= maxuserfiles && 1732 priv_check(td, PRIV_MAXFILES) != 0) || 1733 openfiles >= maxfiles) { 1734 if (ppsratecheck(&lastfail, &curfail, 1)) { 1735 printf("kern.maxfiles limit exceeded by uid %i, " 1736 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1737 } 1738 return (ENFILE); 1739 } 1740 atomic_add_int(&openfiles, 1); 1741 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1742 refcount_init(&fp->f_count, 1); 1743 fp->f_cred = crhold(td->td_ucred); 1744 fp->f_ops = &badfileops; 1745 *resultfp = fp; 1746 return (0); 1747} 1748 1749/* 1750 * Install a file in a file descriptor table. 1751 */ 1752void 1753_finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 1754 struct filecaps *fcaps) 1755{ 1756 struct filedescent *fde; 1757 1758 MPASS(fp != NULL); 1759 if (fcaps != NULL) 1760 filecaps_validate(fcaps, __func__); 1761 FILEDESC_XLOCK_ASSERT(fdp); 1762 1763 fde = &fdp->fd_ofiles[fd]; 1764#ifdef CAPABILITIES 1765 seq_write_begin(&fde->fde_seq); 1766#endif 1767 fde->fde_file = fp; 1768 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 1769 if (fcaps != NULL) 1770 filecaps_move(fcaps, &fde->fde_caps); 1771 else 1772 filecaps_fill(&fde->fde_caps); 1773#ifdef CAPABILITIES 1774 seq_write_end(&fde->fde_seq); 1775#endif 1776} 1777 1778int 1779finstall(struct thread *td, struct file *fp, int *fd, int flags, 1780 struct filecaps *fcaps) 1781{ 1782 struct filedesc *fdp = td->td_proc->p_fd; 1783 int error; 1784 1785 MPASS(fd != NULL); 1786 1787 FILEDESC_XLOCK(fdp); 1788 if ((error = fdalloc(td, 0, fd))) { 1789 FILEDESC_XUNLOCK(fdp); 1790 return (error); 1791 } 1792 fhold(fp); 1793 _finstall(fdp, fp, *fd, flags, fcaps); 1794 FILEDESC_XUNLOCK(fdp); 1795 return (0); 1796} 1797 1798/* 1799 * Build a new filedesc structure from another. 1800 * Copy the current, root, and jail root vnode references. 1801 * 1802 * If fdp is not NULL, return with it shared locked. 1803 */ 1804struct filedesc * 1805fdinit(struct filedesc *fdp, bool prepfiles) 1806{ 1807 struct filedesc0 *newfdp0; 1808 struct filedesc *newfdp; 1809 1810 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1811 newfdp = &newfdp0->fd_fd; 1812 1813 /* Create the file descriptor table. */ 1814 FILEDESC_LOCK_INIT(newfdp); 1815 refcount_init(&newfdp->fd_refcnt, 1); 1816 refcount_init(&newfdp->fd_holdcnt, 1); 1817 newfdp->fd_cmask = CMASK; 1818 newfdp->fd_map = newfdp0->fd_dmap; 1819 newfdp->fd_lastfile = -1; 1820 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1821 newfdp->fd_files->fdt_nfiles = NDFILE; 1822 1823 if (fdp == NULL) 1824 return (newfdp); 1825 1826 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1827 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1828 1829 FILEDESC_SLOCK(fdp); 1830 newfdp->fd_cdir = fdp->fd_cdir; 1831 if (newfdp->fd_cdir) 1832 VREF(newfdp->fd_cdir); 1833 newfdp->fd_rdir = fdp->fd_rdir; 1834 if (newfdp->fd_rdir) 1835 VREF(newfdp->fd_rdir); 1836 newfdp->fd_jdir = fdp->fd_jdir; 1837 if (newfdp->fd_jdir) 1838 VREF(newfdp->fd_jdir); 1839 1840 if (!prepfiles) { 1841 FILEDESC_SUNLOCK(fdp); 1842 } else { 1843 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1844 FILEDESC_SUNLOCK(fdp); 1845 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1846 FILEDESC_SLOCK(fdp); 1847 } 1848 } 1849 1850 return (newfdp); 1851} 1852 1853static struct filedesc * 1854fdhold(struct proc *p) 1855{ 1856 struct filedesc *fdp; 1857 1858 PROC_LOCK_ASSERT(p, MA_OWNED); 1859 fdp = p->p_fd; 1860 if (fdp != NULL) 1861 refcount_acquire(&fdp->fd_holdcnt); 1862 return (fdp); 1863} 1864 1865static void 1866fddrop(struct filedesc *fdp) 1867{ 1868 1869 if (fdp->fd_holdcnt > 1) { 1870 if (refcount_release(&fdp->fd_holdcnt) == 0) 1871 return; 1872 } 1873 1874 FILEDESC_LOCK_DESTROY(fdp); 1875 uma_zfree(filedesc0_zone, fdp); 1876} 1877 1878/* 1879 * Share a filedesc structure. 1880 */ 1881struct filedesc * 1882fdshare(struct filedesc *fdp) 1883{ 1884 1885 refcount_acquire(&fdp->fd_refcnt); 1886 return (fdp); 1887} 1888 1889/* 1890 * Unshare a filedesc structure, if necessary by making a copy 1891 */ 1892void 1893fdunshare(struct thread *td) 1894{ 1895 struct filedesc *tmp; 1896 struct proc *p = td->td_proc; 1897 1898 if (p->p_fd->fd_refcnt == 1) 1899 return; 1900 1901 tmp = fdcopy(p->p_fd); 1902 fdescfree(td); 1903 p->p_fd = tmp; 1904} 1905 1906/* 1907 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1908 * this is to ease callers, not catch errors. 1909 */ 1910struct filedesc * 1911fdcopy(struct filedesc *fdp) 1912{ 1913 struct filedesc *newfdp; 1914 struct filedescent *nfde, *ofde; 1915 int i; 1916 1917 MPASS(fdp != NULL); 1918 1919 newfdp = fdinit(fdp, true); 1920 /* copy all passable descriptors (i.e. not kqueue) */ 1921 newfdp->fd_freefile = -1; 1922 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1923 ofde = &fdp->fd_ofiles[i]; 1924 if (ofde->fde_file == NULL || 1925 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 1926 if (newfdp->fd_freefile == -1) 1927 newfdp->fd_freefile = i; 1928 continue; 1929 } 1930 nfde = &newfdp->fd_ofiles[i]; 1931 *nfde = *ofde; 1932 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1933 fhold(nfde->fde_file); 1934 fdused_init(newfdp, i); 1935 newfdp->fd_lastfile = i; 1936 } 1937 if (newfdp->fd_freefile == -1) 1938 newfdp->fd_freefile = i; 1939 newfdp->fd_cmask = fdp->fd_cmask; 1940 FILEDESC_SUNLOCK(fdp); 1941 return (newfdp); 1942} 1943 1944/* 1945 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 1946 * one of processes using it exits) and the table used to be shared. 1947 */ 1948static void 1949fdclearlocks(struct thread *td) 1950{ 1951 struct filedesc *fdp; 1952 struct filedesc_to_leader *fdtol; 1953 struct flock lf; 1954 struct file *fp; 1955 struct proc *p; 1956 struct vnode *vp; 1957 int i; 1958 1959 p = td->td_proc; 1960 fdp = p->p_fd; 1961 fdtol = p->p_fdtol; 1962 MPASS(fdtol != NULL); 1963 1964 FILEDESC_XLOCK(fdp); 1965 KASSERT(fdtol->fdl_refcount > 0, 1966 ("filedesc_to_refcount botch: fdl_refcount=%d", 1967 fdtol->fdl_refcount)); 1968 if (fdtol->fdl_refcount == 1 && 1969 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1970 for (i = 0; i <= fdp->fd_lastfile; i++) { 1971 fp = fdp->fd_ofiles[i].fde_file; 1972 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1973 continue; 1974 fhold(fp); 1975 FILEDESC_XUNLOCK(fdp); 1976 lf.l_whence = SEEK_SET; 1977 lf.l_start = 0; 1978 lf.l_len = 0; 1979 lf.l_type = F_UNLCK; 1980 vp = fp->f_vnode; 1981 (void) VOP_ADVLOCK(vp, 1982 (caddr_t)p->p_leader, F_UNLCK, 1983 &lf, F_POSIX); 1984 FILEDESC_XLOCK(fdp); 1985 fdrop(fp, td); 1986 } 1987 } 1988retry: 1989 if (fdtol->fdl_refcount == 1) { 1990 if (fdp->fd_holdleaderscount > 0 && 1991 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1992 /* 1993 * close() or kern_dup() has cleared a reference 1994 * in a shared file descriptor table. 1995 */ 1996 fdp->fd_holdleaderswakeup = 1; 1997 sx_sleep(&fdp->fd_holdleaderscount, 1998 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1999 goto retry; 2000 } 2001 if (fdtol->fdl_holdcount > 0) { 2002 /* 2003 * Ensure that fdtol->fdl_leader remains 2004 * valid in closef(). 2005 */ 2006 fdtol->fdl_wakeup = 1; 2007 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2008 "fdlhold", 0); 2009 goto retry; 2010 } 2011 } 2012 fdtol->fdl_refcount--; 2013 if (fdtol->fdl_refcount == 0 && 2014 fdtol->fdl_holdcount == 0) { 2015 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2016 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2017 } else 2018 fdtol = NULL; 2019 p->p_fdtol = NULL; 2020 FILEDESC_XUNLOCK(fdp); 2021 if (fdtol != NULL) 2022 free(fdtol, M_FILEDESC_TO_LEADER); 2023} 2024 2025/* 2026 * Release a filedesc structure. 2027 */ 2028void 2029fdescfree(struct thread *td) 2030{ 2031 struct proc *p; 2032 struct filedesc0 *fdp0; 2033 struct filedesc *fdp; 2034 struct freetable *ft, *tft; 2035 struct filedescent *fde; 2036 struct file *fp; 2037 struct vnode *cdir, *jdir, *rdir; 2038 int i; 2039 2040 p = td->td_proc; 2041 fdp = p->p_fd; 2042 MPASS(fdp != NULL); 2043 2044#ifdef RACCT 2045 if (racct_enable) { 2046 PROC_LOCK(p); 2047 racct_set(p, RACCT_NOFILE, 0); 2048 PROC_UNLOCK(p); 2049 } 2050#endif 2051 2052 if (td->td_proc->p_fdtol != NULL) 2053 fdclearlocks(td); 2054 2055 PROC_LOCK(p); 2056 p->p_fd = NULL; 2057 PROC_UNLOCK(p); 2058 2059 if (refcount_release(&fdp->fd_refcnt) == 0) 2060 return; 2061 2062 FILEDESC_XLOCK(fdp); 2063 cdir = fdp->fd_cdir; 2064 fdp->fd_cdir = NULL; 2065 rdir = fdp->fd_rdir; 2066 fdp->fd_rdir = NULL; 2067 jdir = fdp->fd_jdir; 2068 fdp->fd_jdir = NULL; 2069 FILEDESC_XUNLOCK(fdp); 2070 2071 for (i = 0; i <= fdp->fd_lastfile; i++) { 2072 fde = &fdp->fd_ofiles[i]; 2073 fp = fde->fde_file; 2074 if (fp != NULL) { 2075 fdefree_last(fde); 2076 (void) closef(fp, td); 2077 } 2078 } 2079 2080 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2081 free(fdp->fd_map, M_FILEDESC); 2082 if (fdp->fd_nfiles > NDFILE) 2083 free(fdp->fd_files, M_FILEDESC); 2084 2085 fdp0 = (struct filedesc0 *)fdp; 2086 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2087 free(ft->ft_table, M_FILEDESC); 2088 2089 if (cdir != NULL) 2090 vrele(cdir); 2091 if (rdir != NULL) 2092 vrele(rdir); 2093 if (jdir != NULL) 2094 vrele(jdir); 2095 2096 fddrop(fdp); 2097} 2098 2099/* 2100 * For setugid programs, we don't want to people to use that setugidness 2101 * to generate error messages which write to a file which otherwise would 2102 * otherwise be off-limits to the process. We check for filesystems where 2103 * the vnode can change out from under us after execve (like [lin]procfs). 2104 * 2105 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2106 * sufficient. We also don't check for setugidness since we know we are. 2107 */ 2108static bool 2109is_unsafe(struct file *fp) 2110{ 2111 struct vnode *vp; 2112 2113 if (fp->f_type != DTYPE_VNODE) 2114 return (false); 2115 2116 vp = fp->f_vnode; 2117 return ((vp->v_vflag & VV_PROCDEP) != 0); 2118} 2119 2120/* 2121 * Make this setguid thing safe, if at all possible. 2122 */ 2123void 2124fdsetugidsafety(struct thread *td) 2125{ 2126 struct filedesc *fdp; 2127 struct file *fp; 2128 int i; 2129 2130 fdp = td->td_proc->p_fd; 2131 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2132 MPASS(fdp->fd_nfiles >= 3); 2133 for (i = 0; i <= 2; i++) { 2134 fp = fdp->fd_ofiles[i].fde_file; 2135 if (fp != NULL && is_unsafe(fp)) { 2136 FILEDESC_XLOCK(fdp); 2137 knote_fdclose(td, i); 2138 /* 2139 * NULL-out descriptor prior to close to avoid 2140 * a race while close blocks. 2141 */ 2142 fdfree(fdp, i); 2143 FILEDESC_XUNLOCK(fdp); 2144 (void) closef(fp, td); 2145 } 2146 } 2147} 2148 2149/* 2150 * If a specific file object occupies a specific file descriptor, close the 2151 * file descriptor entry and drop a reference on the file object. This is a 2152 * convenience function to handle a subsequent error in a function that calls 2153 * falloc() that handles the race that another thread might have closed the 2154 * file descriptor out from under the thread creating the file object. 2155 */ 2156void 2157fdclose(struct thread *td, struct file *fp, int idx) 2158{ 2159 struct filedesc *fdp = td->td_proc->p_fd; 2160 2161 FILEDESC_XLOCK(fdp); 2162 if (fdp->fd_ofiles[idx].fde_file == fp) { 2163 fdfree(fdp, idx); 2164 FILEDESC_XUNLOCK(fdp); 2165 fdrop(fp, td); 2166 } else 2167 FILEDESC_XUNLOCK(fdp); 2168} 2169 2170/* 2171 * Close any files on exec? 2172 */ 2173void 2174fdcloseexec(struct thread *td) 2175{ 2176 struct filedesc *fdp; 2177 struct filedescent *fde; 2178 struct file *fp; 2179 int i; 2180 2181 fdp = td->td_proc->p_fd; 2182 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2183 for (i = 0; i <= fdp->fd_lastfile; i++) { 2184 fde = &fdp->fd_ofiles[i]; 2185 fp = fde->fde_file; 2186 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2187 (fde->fde_flags & UF_EXCLOSE))) { 2188 FILEDESC_XLOCK(fdp); 2189 fdfree(fdp, i); 2190 (void) closefp(fdp, i, fp, td, 0); 2191 /* closefp() drops the FILEDESC lock. */ 2192 } 2193 } 2194} 2195 2196/* 2197 * It is unsafe for set[ug]id processes to be started with file 2198 * descriptors 0..2 closed, as these descriptors are given implicit 2199 * significance in the Standard C library. fdcheckstd() will create a 2200 * descriptor referencing /dev/null for each of stdin, stdout, and 2201 * stderr that is not already open. 2202 */ 2203int 2204fdcheckstd(struct thread *td) 2205{ 2206 struct filedesc *fdp; 2207 register_t save; 2208 int i, error, devnull; 2209 2210 fdp = td->td_proc->p_fd; 2211 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2212 MPASS(fdp->fd_nfiles >= 3); 2213 devnull = -1; 2214 for (i = 0; i <= 2; i++) { 2215 if (fdp->fd_ofiles[i].fde_file != NULL) 2216 continue; 2217 2218 save = td->td_retval[0]; 2219 if (devnull != -1) { 2220 error = kern_dup(td, FDDUP_FIXED, devnull, i); 2221 } else { 2222 error = kern_openat(td, AT_FDCWD, "/dev/null", 2223 UIO_SYSSPACE, O_RDWR, 0); 2224 if (error == 0) { 2225 devnull = td->td_retval[0]; 2226 KASSERT(devnull == i, ("we didn't get our fd")); 2227 } 2228 } 2229 td->td_retval[0] = save; 2230 if (error != 0) 2231 return (error); 2232 } 2233 return (0); 2234} 2235 2236/* 2237 * Internal form of close. Decrement reference count on file structure. 2238 * Note: td may be NULL when closing a file that was being passed in a 2239 * message. 2240 * 2241 * XXXRW: Giant is not required for the caller, but often will be held; this 2242 * makes it moderately likely the Giant will be recursed in the VFS case. 2243 */ 2244int 2245closef(struct file *fp, struct thread *td) 2246{ 2247 struct vnode *vp; 2248 struct flock lf; 2249 struct filedesc_to_leader *fdtol; 2250 struct filedesc *fdp; 2251 2252 /* 2253 * POSIX record locking dictates that any close releases ALL 2254 * locks owned by this process. This is handled by setting 2255 * a flag in the unlock to free ONLY locks obeying POSIX 2256 * semantics, and not to free BSD-style file locks. 2257 * If the descriptor was in a message, POSIX-style locks 2258 * aren't passed with the descriptor, and the thread pointer 2259 * will be NULL. Callers should be careful only to pass a 2260 * NULL thread pointer when there really is no owning 2261 * context that might have locks, or the locks will be 2262 * leaked. 2263 */ 2264 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2265 vp = fp->f_vnode; 2266 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2267 lf.l_whence = SEEK_SET; 2268 lf.l_start = 0; 2269 lf.l_len = 0; 2270 lf.l_type = F_UNLCK; 2271 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2272 F_UNLCK, &lf, F_POSIX); 2273 } 2274 fdtol = td->td_proc->p_fdtol; 2275 if (fdtol != NULL) { 2276 /* 2277 * Handle special case where file descriptor table is 2278 * shared between multiple process leaders. 2279 */ 2280 fdp = td->td_proc->p_fd; 2281 FILEDESC_XLOCK(fdp); 2282 for (fdtol = fdtol->fdl_next; 2283 fdtol != td->td_proc->p_fdtol; 2284 fdtol = fdtol->fdl_next) { 2285 if ((fdtol->fdl_leader->p_flag & 2286 P_ADVLOCK) == 0) 2287 continue; 2288 fdtol->fdl_holdcount++; 2289 FILEDESC_XUNLOCK(fdp); 2290 lf.l_whence = SEEK_SET; 2291 lf.l_start = 0; 2292 lf.l_len = 0; 2293 lf.l_type = F_UNLCK; 2294 vp = fp->f_vnode; 2295 (void) VOP_ADVLOCK(vp, 2296 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2297 F_POSIX); 2298 FILEDESC_XLOCK(fdp); 2299 fdtol->fdl_holdcount--; 2300 if (fdtol->fdl_holdcount == 0 && 2301 fdtol->fdl_wakeup != 0) { 2302 fdtol->fdl_wakeup = 0; 2303 wakeup(fdtol); 2304 } 2305 } 2306 FILEDESC_XUNLOCK(fdp); 2307 } 2308 } 2309 return (fdrop(fp, td)); 2310} 2311 2312/* 2313 * Initialize the file pointer with the specified properties. 2314 * 2315 * The ops are set with release semantics to be certain that the flags, type, 2316 * and data are visible when ops is. This is to prevent ops methods from being 2317 * called with bad data. 2318 */ 2319void 2320finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2321{ 2322 fp->f_data = data; 2323 fp->f_flag = flag; 2324 fp->f_type = type; 2325 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2326} 2327 2328int 2329fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2330 struct file **fpp, seq_t *seqp) 2331{ 2332#ifdef CAPABILITIES 2333 struct filedescent *fde; 2334#endif 2335 struct fdescenttbl *fdt; 2336 struct file *fp; 2337 u_int count; 2338#ifdef CAPABILITIES 2339 seq_t seq; 2340 cap_rights_t haverights; 2341 int error; 2342#endif 2343 2344 fdt = fdp->fd_files; 2345 if ((u_int)fd >= fdt->fdt_nfiles) 2346 return (EBADF); 2347 /* 2348 * Fetch the descriptor locklessly. We avoid fdrop() races by 2349 * never raising a refcount above 0. To accomplish this we have 2350 * to use a cmpset loop rather than an atomic_add. The descriptor 2351 * must be re-verified once we acquire a reference to be certain 2352 * that the identity is still correct and we did not lose a race 2353 * due to preemption. 2354 */ 2355 for (;;) { 2356#ifdef CAPABILITIES 2357 seq = seq_read(fd_seq(fdt, fd)); 2358 fde = &fdt->fdt_ofiles[fd]; 2359 haverights = *cap_rights_fde(fde); 2360 fp = fde->fde_file; 2361 if (!seq_consistent(fd_seq(fdt, fd), seq)) { 2362 cpu_spinwait(); 2363 continue; 2364 } 2365#else 2366 fp = fdt->fdt_ofiles[fd].fde_file; 2367#endif 2368 if (fp == NULL) 2369 return (EBADF); 2370#ifdef CAPABILITIES 2371 error = cap_check(&haverights, needrightsp); 2372 if (error != 0) 2373 return (error); 2374#endif 2375 retry: 2376 count = fp->f_count; 2377 if (count == 0) { 2378 /* 2379 * Force a reload. Other thread could reallocate the 2380 * table before this fd was closed, so it possible that 2381 * there is a stale fp pointer in cached version. 2382 */ 2383 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); 2384 continue; 2385 } 2386 /* 2387 * Use an acquire barrier to force re-reading of fdt so it is 2388 * refreshed for verification. 2389 */ 2390 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) 2391 goto retry; 2392 fdt = fdp->fd_files; 2393#ifdef CAPABILITIES 2394 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2395#else 2396 if (fp == fdt->fdt_ofiles[fd].fde_file) 2397#endif 2398 break; 2399 fdrop(fp, curthread); 2400 } 2401 *fpp = fp; 2402 if (seqp != NULL) { 2403#ifdef CAPABILITIES 2404 *seqp = seq; 2405#endif 2406 } 2407 return (0); 2408} 2409 2410/* 2411 * Extract the file pointer associated with the specified descriptor for the 2412 * current user process. 2413 * 2414 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2415 * returned. 2416 * 2417 * File's rights will be checked against the capability rights mask. 2418 * 2419 * If an error occured the non-zero error is returned and *fpp is set to 2420 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2421 * responsible for fdrop(). 2422 */ 2423static __inline int 2424_fget(struct thread *td, int fd, struct file **fpp, int flags, 2425 cap_rights_t *needrightsp, seq_t *seqp) 2426{ 2427 struct filedesc *fdp; 2428 struct file *fp; 2429 int error; 2430 2431 *fpp = NULL; 2432 fdp = td->td_proc->p_fd; 2433 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); 2434 if (error != 0) 2435 return (error); 2436 if (fp->f_ops == &badfileops) { 2437 fdrop(fp, td); 2438 return (EBADF); 2439 } 2440 2441 /* 2442 * FREAD and FWRITE failure return EBADF as per POSIX. 2443 */ 2444 error = 0; 2445 switch (flags) { 2446 case FREAD: 2447 case FWRITE: 2448 if ((fp->f_flag & flags) == 0) 2449 error = EBADF; 2450 break; 2451 case FEXEC: 2452 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2453 ((fp->f_flag & FWRITE) != 0)) 2454 error = EBADF; 2455 break; 2456 case 0: 2457 break; 2458 default: 2459 KASSERT(0, ("wrong flags")); 2460 } 2461 2462 if (error != 0) { 2463 fdrop(fp, td); 2464 return (error); 2465 } 2466 2467 *fpp = fp; 2468 return (0); 2469} 2470 2471int 2472fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2473{ 2474 2475 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2476} 2477 2478int 2479fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2480 struct file **fpp) 2481{ 2482 int error; 2483#ifndef CAPABILITIES 2484 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2485 if (maxprotp != NULL) 2486 *maxprotp = VM_PROT_ALL; 2487#else 2488 struct filedesc *fdp = td->td_proc->p_fd; 2489 seq_t seq; 2490 2491 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2492 for (;;) { 2493 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2494 if (error != 0) 2495 return (error); 2496 /* 2497 * If requested, convert capability rights to access flags. 2498 */ 2499 if (maxprotp != NULL) 2500 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2501 if (!fd_modified(fdp, fd, seq)) 2502 break; 2503 fdrop(*fpp, td); 2504 } 2505#endif 2506 return (error); 2507} 2508 2509int 2510fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2511{ 2512 2513 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2514} 2515 2516int 2517fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2518{ 2519 2520 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2521} 2522 2523int 2524fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2525 struct file **fpp) 2526{ 2527 struct filedesc *fdp = td->td_proc->p_fd; 2528#ifndef CAPABILITIES 2529 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2530#else 2531 int error; 2532 seq_t seq; 2533 2534 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2535 for (;;) { 2536 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2537 if (error != 0) 2538 return (error); 2539 error = cap_fcntl_check(fdp, fd, needfcntl); 2540 if (!fd_modified(fdp, fd, seq)) 2541 break; 2542 fdrop(*fpp, td); 2543 } 2544 if (error != 0) { 2545 fdrop(*fpp, td); 2546 *fpp = NULL; 2547 } 2548 return (error); 2549#endif 2550} 2551 2552/* 2553 * Like fget() but loads the underlying vnode, or returns an error if the 2554 * descriptor does not represent a vnode. Note that pipes use vnodes but 2555 * never have VM objects. The returned vnode will be vref()'d. 2556 * 2557 * XXX: what about the unused flags ? 2558 */ 2559static __inline int 2560_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2561 struct vnode **vpp) 2562{ 2563 struct file *fp; 2564 int error; 2565 2566 *vpp = NULL; 2567 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2568 if (error != 0) 2569 return (error); 2570 if (fp->f_vnode == NULL) { 2571 error = EINVAL; 2572 } else { 2573 *vpp = fp->f_vnode; 2574 vref(*vpp); 2575 } 2576 fdrop(fp, td); 2577 2578 return (error); 2579} 2580 2581int 2582fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2583{ 2584 2585 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2586} 2587 2588int 2589fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2590 struct filecaps *havecaps, struct vnode **vpp) 2591{ 2592 struct filedesc *fdp; 2593 struct file *fp; 2594#ifdef CAPABILITIES 2595 int error; 2596#endif 2597 2598 fdp = td->td_proc->p_fd; 2599 fp = fget_locked(fdp, fd); 2600 if (fp == NULL || fp->f_ops == &badfileops) 2601 return (EBADF); 2602 2603#ifdef CAPABILITIES 2604 if (needrightsp != NULL) { 2605 error = cap_check(cap_rights(fdp, fd), needrightsp); 2606 if (error != 0) 2607 return (error); 2608 } 2609#endif 2610 2611 if (fp->f_vnode == NULL) 2612 return (EINVAL); 2613 2614 *vpp = fp->f_vnode; 2615 vref(*vpp); 2616 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2617 2618 return (0); 2619} 2620 2621int 2622fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2623{ 2624 2625 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2626} 2627 2628int 2629fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2630{ 2631 2632 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2633} 2634 2635#ifdef notyet 2636int 2637fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2638 struct vnode **vpp) 2639{ 2640 2641 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2642} 2643#endif 2644 2645/* 2646 * Like fget() but loads the underlying socket, or returns an error if the 2647 * descriptor does not represent a socket. 2648 * 2649 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2650 * in the future. 2651 * 2652 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2653 * on their file descriptor reference to prevent the socket from being free'd 2654 * during use. 2655 */ 2656int 2657fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2658 u_int *fflagp) 2659{ 2660 struct file *fp; 2661 int error; 2662 2663 *spp = NULL; 2664 if (fflagp != NULL) 2665 *fflagp = 0; 2666 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2667 return (error); 2668 if (fp->f_type != DTYPE_SOCKET) { 2669 error = ENOTSOCK; 2670 } else { 2671 *spp = fp->f_data; 2672 if (fflagp) 2673 *fflagp = fp->f_flag; 2674 SOCK_LOCK(*spp); 2675 soref(*spp); 2676 SOCK_UNLOCK(*spp); 2677 } 2678 fdrop(fp, td); 2679 2680 return (error); 2681} 2682 2683/* 2684 * Drop the reference count on the socket and XXX release the SX lock in the 2685 * future. The last reference closes the socket. 2686 * 2687 * Note: fputsock() is deprecated, see comment for fgetsock(). 2688 */ 2689void 2690fputsock(struct socket *so) 2691{ 2692 2693 ACCEPT_LOCK(); 2694 SOCK_LOCK(so); 2695 CURVNET_SET(so->so_vnet); 2696 sorele(so); 2697 CURVNET_RESTORE(); 2698} 2699 2700/* 2701 * Handle the last reference to a file being closed. 2702 */ 2703int 2704_fdrop(struct file *fp, struct thread *td) 2705{ 2706 int error; 2707 2708 if (fp->f_count != 0) 2709 panic("fdrop: count %d", fp->f_count); 2710 error = fo_close(fp, td); 2711 atomic_subtract_int(&openfiles, 1); 2712 crfree(fp->f_cred); 2713 free(fp->f_advice, M_FADVISE); 2714 uma_zfree(file_zone, fp); 2715 2716 return (error); 2717} 2718 2719/* 2720 * Apply an advisory lock on a file descriptor. 2721 * 2722 * Just attempt to get a record lock of the requested type on the entire file 2723 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2724 */ 2725#ifndef _SYS_SYSPROTO_H_ 2726struct flock_args { 2727 int fd; 2728 int how; 2729}; 2730#endif 2731/* ARGSUSED */ 2732int 2733sys_flock(struct thread *td, struct flock_args *uap) 2734{ 2735 struct file *fp; 2736 struct vnode *vp; 2737 struct flock lf; 2738 cap_rights_t rights; 2739 int error; 2740 2741 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2742 if (error != 0) 2743 return (error); 2744 if (fp->f_type != DTYPE_VNODE) { 2745 fdrop(fp, td); 2746 return (EOPNOTSUPP); 2747 } 2748 2749 vp = fp->f_vnode; 2750 lf.l_whence = SEEK_SET; 2751 lf.l_start = 0; 2752 lf.l_len = 0; 2753 if (uap->how & LOCK_UN) { 2754 lf.l_type = F_UNLCK; 2755 atomic_clear_int(&fp->f_flag, FHASLOCK); 2756 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2757 goto done2; 2758 } 2759 if (uap->how & LOCK_EX) 2760 lf.l_type = F_WRLCK; 2761 else if (uap->how & LOCK_SH) 2762 lf.l_type = F_RDLCK; 2763 else { 2764 error = EBADF; 2765 goto done2; 2766 } 2767 atomic_set_int(&fp->f_flag, FHASLOCK); 2768 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2769 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2770done2: 2771 fdrop(fp, td); 2772 return (error); 2773} 2774/* 2775 * Duplicate the specified descriptor to a free descriptor. 2776 */ 2777int 2778dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2779 int openerror, int *indxp) 2780{ 2781 struct filedescent *newfde, *oldfde; 2782 struct file *fp; 2783 int error, indx; 2784 2785 KASSERT(openerror == ENODEV || openerror == ENXIO, 2786 ("unexpected error %d in %s", openerror, __func__)); 2787 2788 /* 2789 * If the to-be-dup'd fd number is greater than the allowed number 2790 * of file descriptors, or the fd to be dup'd has already been 2791 * closed, then reject. 2792 */ 2793 FILEDESC_XLOCK(fdp); 2794 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2795 FILEDESC_XUNLOCK(fdp); 2796 return (EBADF); 2797 } 2798 2799 error = fdalloc(td, 0, &indx); 2800 if (error != 0) { 2801 FILEDESC_XUNLOCK(fdp); 2802 return (error); 2803 } 2804 2805 /* 2806 * There are two cases of interest here. 2807 * 2808 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2809 * 2810 * For ENXIO steal away the file structure from (dfd) and store it in 2811 * (indx). (dfd) is effectively closed by this operation. 2812 */ 2813 switch (openerror) { 2814 case ENODEV: 2815 /* 2816 * Check that the mode the file is being opened for is a 2817 * subset of the mode of the existing descriptor. 2818 */ 2819 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2820 fdunused(fdp, indx); 2821 FILEDESC_XUNLOCK(fdp); 2822 return (EACCES); 2823 } 2824 fhold(fp); 2825 newfde = &fdp->fd_ofiles[indx]; 2826 oldfde = &fdp->fd_ofiles[dfd]; 2827#ifdef CAPABILITIES 2828 seq_write_begin(&newfde->fde_seq); 2829#endif 2830 memcpy(newfde, oldfde, fde_change_size); 2831 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 2832#ifdef CAPABILITIES 2833 seq_write_end(&newfde->fde_seq); 2834#endif 2835 break; 2836 case ENXIO: 2837 /* 2838 * Steal away the file pointer from dfd and stuff it into indx. 2839 */ 2840 newfde = &fdp->fd_ofiles[indx]; 2841 oldfde = &fdp->fd_ofiles[dfd]; 2842#ifdef CAPABILITIES 2843 seq_write_begin(&newfde->fde_seq); 2844#endif 2845 memcpy(newfde, oldfde, fde_change_size); 2846 oldfde->fde_file = NULL; 2847 fdunused(fdp, dfd); 2848#ifdef CAPABILITIES 2849 seq_write_end(&newfde->fde_seq); 2850#endif 2851 break; 2852 } 2853 FILEDESC_XUNLOCK(fdp); 2854 *indxp = indx; 2855 return (0); 2856} 2857 2858/* 2859 * Scan all active processes and prisons to see if any of them have a current 2860 * or root directory of `olddp'. If so, replace them with the new mount point. 2861 */ 2862void 2863mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2864{ 2865 struct filedesc *fdp; 2866 struct prison *pr; 2867 struct proc *p; 2868 int nrele; 2869 2870 if (vrefcnt(olddp) == 1) 2871 return; 2872 nrele = 0; 2873 sx_slock(&allproc_lock); 2874 FOREACH_PROC_IN_SYSTEM(p) { 2875 PROC_LOCK(p); 2876 fdp = fdhold(p); 2877 PROC_UNLOCK(p); 2878 if (fdp == NULL) 2879 continue; 2880 FILEDESC_XLOCK(fdp); 2881 if (fdp->fd_cdir == olddp) { 2882 vref(newdp); 2883 fdp->fd_cdir = newdp; 2884 nrele++; 2885 } 2886 if (fdp->fd_rdir == olddp) { 2887 vref(newdp); 2888 fdp->fd_rdir = newdp; 2889 nrele++; 2890 } 2891 if (fdp->fd_jdir == olddp) { 2892 vref(newdp); 2893 fdp->fd_jdir = newdp; 2894 nrele++; 2895 } 2896 FILEDESC_XUNLOCK(fdp); 2897 fddrop(fdp); 2898 } 2899 sx_sunlock(&allproc_lock); 2900 if (rootvnode == olddp) { 2901 vref(newdp); 2902 rootvnode = newdp; 2903 nrele++; 2904 } 2905 mtx_lock(&prison0.pr_mtx); 2906 if (prison0.pr_root == olddp) { 2907 vref(newdp); 2908 prison0.pr_root = newdp; 2909 nrele++; 2910 } 2911 mtx_unlock(&prison0.pr_mtx); 2912 sx_slock(&allprison_lock); 2913 TAILQ_FOREACH(pr, &allprison, pr_list) { 2914 mtx_lock(&pr->pr_mtx); 2915 if (pr->pr_root == olddp) { 2916 vref(newdp); 2917 pr->pr_root = newdp; 2918 nrele++; 2919 } 2920 mtx_unlock(&pr->pr_mtx); 2921 } 2922 sx_sunlock(&allprison_lock); 2923 while (nrele--) 2924 vrele(olddp); 2925} 2926 2927struct filedesc_to_leader * 2928filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2929{ 2930 struct filedesc_to_leader *fdtol; 2931 2932 fdtol = malloc(sizeof(struct filedesc_to_leader), 2933 M_FILEDESC_TO_LEADER, M_WAITOK); 2934 fdtol->fdl_refcount = 1; 2935 fdtol->fdl_holdcount = 0; 2936 fdtol->fdl_wakeup = 0; 2937 fdtol->fdl_leader = leader; 2938 if (old != NULL) { 2939 FILEDESC_XLOCK(fdp); 2940 fdtol->fdl_next = old->fdl_next; 2941 fdtol->fdl_prev = old; 2942 old->fdl_next = fdtol; 2943 fdtol->fdl_next->fdl_prev = fdtol; 2944 FILEDESC_XUNLOCK(fdp); 2945 } else { 2946 fdtol->fdl_next = fdtol; 2947 fdtol->fdl_prev = fdtol; 2948 } 2949 return (fdtol); 2950} 2951 2952/* 2953 * Get file structures globally. 2954 */ 2955static int 2956sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2957{ 2958 struct xfile xf; 2959 struct filedesc *fdp; 2960 struct file *fp; 2961 struct proc *p; 2962 int error, n; 2963 2964 error = sysctl_wire_old_buffer(req, 0); 2965 if (error != 0) 2966 return (error); 2967 if (req->oldptr == NULL) { 2968 n = 0; 2969 sx_slock(&allproc_lock); 2970 FOREACH_PROC_IN_SYSTEM(p) { 2971 PROC_LOCK(p); 2972 if (p->p_state == PRS_NEW) { 2973 PROC_UNLOCK(p); 2974 continue; 2975 } 2976 fdp = fdhold(p); 2977 PROC_UNLOCK(p); 2978 if (fdp == NULL) 2979 continue; 2980 /* overestimates sparse tables. */ 2981 if (fdp->fd_lastfile > 0) 2982 n += fdp->fd_lastfile; 2983 fddrop(fdp); 2984 } 2985 sx_sunlock(&allproc_lock); 2986 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2987 } 2988 error = 0; 2989 bzero(&xf, sizeof(xf)); 2990 xf.xf_size = sizeof(xf); 2991 sx_slock(&allproc_lock); 2992 FOREACH_PROC_IN_SYSTEM(p) { 2993 PROC_LOCK(p); 2994 if (p->p_state == PRS_NEW) { 2995 PROC_UNLOCK(p); 2996 continue; 2997 } 2998 if (p_cansee(req->td, p) != 0) { 2999 PROC_UNLOCK(p); 3000 continue; 3001 } 3002 xf.xf_pid = p->p_pid; 3003 xf.xf_uid = p->p_ucred->cr_uid; 3004 fdp = fdhold(p); 3005 PROC_UNLOCK(p); 3006 if (fdp == NULL) 3007 continue; 3008 FILEDESC_SLOCK(fdp); 3009 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3010 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3011 continue; 3012 xf.xf_fd = n; 3013 xf.xf_file = fp; 3014 xf.xf_data = fp->f_data; 3015 xf.xf_vnode = fp->f_vnode; 3016 xf.xf_type = fp->f_type; 3017 xf.xf_count = fp->f_count; 3018 xf.xf_msgcount = 0; 3019 xf.xf_offset = foffset_get(fp); 3020 xf.xf_flag = fp->f_flag; 3021 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3022 if (error) 3023 break; 3024 } 3025 FILEDESC_SUNLOCK(fdp); 3026 fddrop(fdp); 3027 if (error) 3028 break; 3029 } 3030 sx_sunlock(&allproc_lock); 3031 return (error); 3032} 3033 3034SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3035 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3036 3037#ifdef KINFO_FILE_SIZE 3038CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3039#endif 3040 3041static int 3042xlate_fflags(int fflags) 3043{ 3044 static const struct { 3045 int fflag; 3046 int kf_fflag; 3047 } fflags_table[] = { 3048 { FAPPEND, KF_FLAG_APPEND }, 3049 { FASYNC, KF_FLAG_ASYNC }, 3050 { FFSYNC, KF_FLAG_FSYNC }, 3051 { FHASLOCK, KF_FLAG_HASLOCK }, 3052 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3053 { FREAD, KF_FLAG_READ }, 3054 { FWRITE, KF_FLAG_WRITE }, 3055 { O_CREAT, KF_FLAG_CREAT }, 3056 { O_DIRECT, KF_FLAG_DIRECT }, 3057 { O_EXCL, KF_FLAG_EXCL }, 3058 { O_EXEC, KF_FLAG_EXEC }, 3059 { O_EXLOCK, KF_FLAG_EXLOCK }, 3060 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3061 { O_SHLOCK, KF_FLAG_SHLOCK }, 3062 { O_TRUNC, KF_FLAG_TRUNC } 3063 }; 3064 unsigned int i; 3065 int kflags; 3066 3067 kflags = 0; 3068 for (i = 0; i < nitems(fflags_table); i++) 3069 if (fflags & fflags_table[i].fflag) 3070 kflags |= fflags_table[i].kf_fflag; 3071 return (kflags); 3072} 3073 3074/* Trim unused data from kf_path by truncating the structure size. */ 3075static void 3076pack_kinfo(struct kinfo_file *kif) 3077{ 3078 3079 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3080 strlen(kif->kf_path) + 1; 3081 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3082} 3083 3084static void 3085export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3086 struct kinfo_file *kif, struct filedesc *fdp) 3087{ 3088 int error; 3089 3090 bzero(kif, sizeof(*kif)); 3091 3092 /* Set a default type to allow for empty fill_kinfo() methods. */ 3093 kif->kf_type = KF_TYPE_UNKNOWN; 3094 kif->kf_flags = xlate_fflags(fp->f_flag); 3095 if (rightsp != NULL) 3096 kif->kf_cap_rights = *rightsp; 3097 else 3098 cap_rights_init(&kif->kf_cap_rights); 3099 kif->kf_fd = fd; 3100 kif->kf_ref_count = fp->f_count; 3101 kif->kf_offset = foffset_get(fp); 3102 3103 /* 3104 * This may drop the filedesc lock, so the 'fp' cannot be 3105 * accessed after this call. 3106 */ 3107 error = fo_fill_kinfo(fp, kif, fdp); 3108 if (error == 0) 3109 kif->kf_status |= KF_ATTR_VALID; 3110 pack_kinfo(kif); 3111} 3112 3113static void 3114export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3115 struct kinfo_file *kif) 3116{ 3117 int error; 3118 3119 bzero(kif, sizeof(*kif)); 3120 3121 kif->kf_type = KF_TYPE_VNODE; 3122 error = vn_fill_kinfo_vnode(vp, kif); 3123 if (error == 0) 3124 kif->kf_status |= KF_ATTR_VALID; 3125 kif->kf_flags = xlate_fflags(fflags); 3126 cap_rights_init(&kif->kf_cap_rights); 3127 kif->kf_fd = fd; 3128 kif->kf_ref_count = -1; 3129 kif->kf_offset = -1; 3130 pack_kinfo(kif); 3131 vrele(vp); 3132} 3133 3134struct export_fd_buf { 3135 struct filedesc *fdp; 3136 struct sbuf *sb; 3137 ssize_t remainder; 3138 struct kinfo_file kif; 3139}; 3140 3141static int 3142export_kinfo_to_sb(struct export_fd_buf *efbuf) 3143{ 3144 struct kinfo_file *kif; 3145 3146 kif = &efbuf->kif; 3147 if (efbuf->remainder != -1) { 3148 if (efbuf->remainder < kif->kf_structsize) { 3149 /* Terminate export. */ 3150 efbuf->remainder = 0; 3151 return (0); 3152 } 3153 efbuf->remainder -= kif->kf_structsize; 3154 } 3155 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3156} 3157 3158static int 3159export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3160 struct export_fd_buf *efbuf) 3161{ 3162 int error; 3163 3164 if (efbuf->remainder == 0) 3165 return (0); 3166 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp); 3167 FILEDESC_SUNLOCK(efbuf->fdp); 3168 error = export_kinfo_to_sb(efbuf); 3169 FILEDESC_SLOCK(efbuf->fdp); 3170 return (error); 3171} 3172 3173static int 3174export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3175 struct export_fd_buf *efbuf) 3176{ 3177 int error; 3178 3179 if (efbuf->remainder == 0) 3180 return (0); 3181 if (efbuf->fdp != NULL) 3182 FILEDESC_SUNLOCK(efbuf->fdp); 3183 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif); 3184 error = export_kinfo_to_sb(efbuf); 3185 if (efbuf->fdp != NULL) 3186 FILEDESC_SLOCK(efbuf->fdp); 3187 return (error); 3188} 3189 3190/* 3191 * Store a process file descriptor information to sbuf. 3192 * 3193 * Takes a locked proc as argument, and returns with the proc unlocked. 3194 */ 3195int 3196kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3197{ 3198 struct file *fp; 3199 struct filedesc *fdp; 3200 struct export_fd_buf *efbuf; 3201 struct vnode *cttyvp, *textvp, *tracevp; 3202 int error, i; 3203 cap_rights_t rights; 3204 3205 PROC_LOCK_ASSERT(p, MA_OWNED); 3206 3207 /* ktrace vnode */ 3208 tracevp = p->p_tracevp; 3209 if (tracevp != NULL) 3210 vref(tracevp); 3211 /* text vnode */ 3212 textvp = p->p_textvp; 3213 if (textvp != NULL) 3214 vref(textvp); 3215 /* Controlling tty. */ 3216 cttyvp = NULL; 3217 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3218 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3219 if (cttyvp != NULL) 3220 vref(cttyvp); 3221 } 3222 fdp = fdhold(p); 3223 PROC_UNLOCK(p); 3224 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3225 efbuf->fdp = NULL; 3226 efbuf->sb = sb; 3227 efbuf->remainder = maxlen; 3228 if (tracevp != NULL) 3229 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3230 efbuf); 3231 if (textvp != NULL) 3232 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3233 if (cttyvp != NULL) 3234 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3235 efbuf); 3236 error = 0; 3237 if (fdp == NULL) 3238 goto fail; 3239 efbuf->fdp = fdp; 3240 FILEDESC_SLOCK(fdp); 3241 /* working directory */ 3242 if (fdp->fd_cdir != NULL) { 3243 vref(fdp->fd_cdir); 3244 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3245 } 3246 /* root directory */ 3247 if (fdp->fd_rdir != NULL) { 3248 vref(fdp->fd_rdir); 3249 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3250 } 3251 /* jail directory */ 3252 if (fdp->fd_jdir != NULL) { 3253 vref(fdp->fd_jdir); 3254 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3255 } 3256 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3257 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3258 continue; 3259#ifdef CAPABILITIES 3260 rights = *cap_rights(fdp, i); 3261#else /* !CAPABILITIES */ 3262 cap_rights_init(&rights); 3263#endif 3264 /* 3265 * Create sysctl entry. It is OK to drop the filedesc 3266 * lock inside of export_file_to_sb() as we will 3267 * re-validate and re-evaluate its properties when the 3268 * loop continues. 3269 */ 3270 error = export_file_to_sb(fp, i, &rights, efbuf); 3271 if (error != 0 || efbuf->remainder == 0) 3272 break; 3273 } 3274 FILEDESC_SUNLOCK(fdp); 3275 fddrop(fdp); 3276fail: 3277 free(efbuf, M_TEMP); 3278 return (error); 3279} 3280 3281#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3282 3283/* 3284 * Get per-process file descriptors for use by procstat(1), et al. 3285 */ 3286static int 3287sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3288{ 3289 struct sbuf sb; 3290 struct proc *p; 3291 ssize_t maxlen; 3292 int error, error2, *name; 3293 3294 name = (int *)arg1; 3295 3296 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3297 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3298 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3299 if (error != 0) { 3300 sbuf_delete(&sb); 3301 return (error); 3302 } 3303 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3304 error = kern_proc_filedesc_out(p, &sb, maxlen); 3305 error2 = sbuf_finish(&sb); 3306 sbuf_delete(&sb); 3307 return (error != 0 ? error : error2); 3308} 3309 3310#ifdef KINFO_OFILE_SIZE 3311CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3312#endif 3313 3314#ifdef COMPAT_FREEBSD7 3315static void 3316kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3317{ 3318 3319 okif->kf_structsize = sizeof(*okif); 3320 okif->kf_type = kif->kf_type; 3321 okif->kf_fd = kif->kf_fd; 3322 okif->kf_ref_count = kif->kf_ref_count; 3323 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3324 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3325 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3326 okif->kf_offset = kif->kf_offset; 3327 okif->kf_vnode_type = kif->kf_vnode_type; 3328 okif->kf_sock_domain = kif->kf_sock_domain; 3329 okif->kf_sock_type = kif->kf_sock_type; 3330 okif->kf_sock_protocol = kif->kf_sock_protocol; 3331 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3332 okif->kf_sa_local = kif->kf_sa_local; 3333 okif->kf_sa_peer = kif->kf_sa_peer; 3334} 3335 3336static int 3337export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3338 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3339{ 3340 int error; 3341 3342 vref(vp); 3343 FILEDESC_SUNLOCK(fdp); 3344 export_vnode_to_kinfo(vp, type, 0, kif); 3345 kinfo_to_okinfo(kif, okif); 3346 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3347 FILEDESC_SLOCK(fdp); 3348 return (error); 3349} 3350 3351/* 3352 * Get per-process file descriptors for use by procstat(1), et al. 3353 */ 3354static int 3355sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3356{ 3357 struct kinfo_ofile *okif; 3358 struct kinfo_file *kif; 3359 struct filedesc *fdp; 3360 int error, i, *name; 3361 struct file *fp; 3362 struct proc *p; 3363 3364 name = (int *)arg1; 3365 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3366 if (error != 0) 3367 return (error); 3368 fdp = fdhold(p); 3369 PROC_UNLOCK(p); 3370 if (fdp == NULL) 3371 return (ENOENT); 3372 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3373 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3374 FILEDESC_SLOCK(fdp); 3375 if (fdp->fd_cdir != NULL) 3376 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3377 okif, fdp, req); 3378 if (fdp->fd_rdir != NULL) 3379 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3380 okif, fdp, req); 3381 if (fdp->fd_jdir != NULL) 3382 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3383 okif, fdp, req); 3384 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3385 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3386 continue; 3387 export_file_to_kinfo(fp, i, NULL, kif, fdp); 3388 FILEDESC_SUNLOCK(fdp); 3389 kinfo_to_okinfo(kif, okif); 3390 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3391 FILEDESC_SLOCK(fdp); 3392 if (error) 3393 break; 3394 } 3395 FILEDESC_SUNLOCK(fdp); 3396 fddrop(fdp); 3397 free(kif, M_TEMP); 3398 free(okif, M_TEMP); 3399 return (0); 3400} 3401 3402static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3403 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3404 "Process ofiledesc entries"); 3405#endif /* COMPAT_FREEBSD7 */ 3406 3407int 3408vntype_to_kinfo(int vtype) 3409{ 3410 struct { 3411 int vtype; 3412 int kf_vtype; 3413 } vtypes_table[] = { 3414 { VBAD, KF_VTYPE_VBAD }, 3415 { VBLK, KF_VTYPE_VBLK }, 3416 { VCHR, KF_VTYPE_VCHR }, 3417 { VDIR, KF_VTYPE_VDIR }, 3418 { VFIFO, KF_VTYPE_VFIFO }, 3419 { VLNK, KF_VTYPE_VLNK }, 3420 { VNON, KF_VTYPE_VNON }, 3421 { VREG, KF_VTYPE_VREG }, 3422 { VSOCK, KF_VTYPE_VSOCK } 3423 }; 3424 unsigned int i; 3425 3426 /* 3427 * Perform vtype translation. 3428 */ 3429 for (i = 0; i < nitems(vtypes_table); i++) 3430 if (vtypes_table[i].vtype == vtype) 3431 return (vtypes_table[i].kf_vtype); 3432 3433 return (KF_VTYPE_UNKNOWN); 3434} 3435 3436static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3437 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3438 "Process filedesc entries"); 3439 3440/* 3441 * Store a process current working directory information to sbuf. 3442 * 3443 * Takes a locked proc as argument, and returns with the proc unlocked. 3444 */ 3445int 3446kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3447{ 3448 struct filedesc *fdp; 3449 struct export_fd_buf *efbuf; 3450 int error; 3451 3452 PROC_LOCK_ASSERT(p, MA_OWNED); 3453 3454 fdp = fdhold(p); 3455 PROC_UNLOCK(p); 3456 if (fdp == NULL) 3457 return (EINVAL); 3458 3459 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3460 efbuf->fdp = fdp; 3461 efbuf->sb = sb; 3462 efbuf->remainder = maxlen; 3463 3464 FILEDESC_SLOCK(fdp); 3465 if (fdp->fd_cdir == NULL) 3466 error = EINVAL; 3467 else { 3468 vref(fdp->fd_cdir); 3469 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3470 FREAD, efbuf); 3471 } 3472 FILEDESC_SUNLOCK(fdp); 3473 fddrop(fdp); 3474 free(efbuf, M_TEMP); 3475 return (error); 3476} 3477 3478/* 3479 * Get per-process current working directory. 3480 */ 3481static int 3482sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3483{ 3484 struct sbuf sb; 3485 struct proc *p; 3486 ssize_t maxlen; 3487 int error, error2, *name; 3488 3489 name = (int *)arg1; 3490 3491 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3492 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3493 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3494 if (error != 0) { 3495 sbuf_delete(&sb); 3496 return (error); 3497 } 3498 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3499 error = kern_proc_cwd_out(p, &sb, maxlen); 3500 error2 = sbuf_finish(&sb); 3501 sbuf_delete(&sb); 3502 return (error != 0 ? error : error2); 3503} 3504 3505static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3506 sysctl_kern_proc_cwd, "Process current working directory"); 3507 3508#ifdef DDB 3509/* 3510 * For the purposes of debugging, generate a human-readable string for the 3511 * file type. 3512 */ 3513static const char * 3514file_type_to_name(short type) 3515{ 3516 3517 switch (type) { 3518 case 0: 3519 return ("zero"); 3520 case DTYPE_VNODE: 3521 return ("vnod"); 3522 case DTYPE_SOCKET: 3523 return ("sock"); 3524 case DTYPE_PIPE: 3525 return ("pipe"); 3526 case DTYPE_FIFO: 3527 return ("fifo"); 3528 case DTYPE_KQUEUE: 3529 return ("kque"); 3530 case DTYPE_CRYPTO: 3531 return ("crpt"); 3532 case DTYPE_MQUEUE: 3533 return ("mque"); 3534 case DTYPE_SHM: 3535 return ("shm"); 3536 case DTYPE_SEM: 3537 return ("ksem"); 3538 default: 3539 return ("unkn"); 3540 } 3541} 3542 3543/* 3544 * For the purposes of debugging, identify a process (if any, perhaps one of 3545 * many) that references the passed file in its file descriptor array. Return 3546 * NULL if none. 3547 */ 3548static struct proc * 3549file_to_first_proc(struct file *fp) 3550{ 3551 struct filedesc *fdp; 3552 struct proc *p; 3553 int n; 3554 3555 FOREACH_PROC_IN_SYSTEM(p) { 3556 if (p->p_state == PRS_NEW) 3557 continue; 3558 fdp = p->p_fd; 3559 if (fdp == NULL) 3560 continue; 3561 for (n = 0; n <= fdp->fd_lastfile; n++) { 3562 if (fp == fdp->fd_ofiles[n].fde_file) 3563 return (p); 3564 } 3565 } 3566 return (NULL); 3567} 3568 3569static void 3570db_print_file(struct file *fp, int header) 3571{ 3572 struct proc *p; 3573 3574 if (header) 3575 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3576 "File", "Type", "Data", "Flag", "GCFl", "Count", 3577 "MCount", "Vnode", "FPID", "FCmd"); 3578 p = file_to_first_proc(fp); 3579 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3580 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3581 0, fp->f_count, 0, fp->f_vnode, 3582 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3583} 3584 3585DB_SHOW_COMMAND(file, db_show_file) 3586{ 3587 struct file *fp; 3588 3589 if (!have_addr) { 3590 db_printf("usage: show file <addr>\n"); 3591 return; 3592 } 3593 fp = (struct file *)addr; 3594 db_print_file(fp, 1); 3595} 3596 3597DB_SHOW_COMMAND(files, db_show_files) 3598{ 3599 struct filedesc *fdp; 3600 struct file *fp; 3601 struct proc *p; 3602 int header; 3603 int n; 3604 3605 header = 1; 3606 FOREACH_PROC_IN_SYSTEM(p) { 3607 if (p->p_state == PRS_NEW) 3608 continue; 3609 if ((fdp = p->p_fd) == NULL) 3610 continue; 3611 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3612 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3613 continue; 3614 db_print_file(fp, header); 3615 header = 0; 3616 } 3617 } 3618} 3619#endif 3620 3621SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3622 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3623 3624SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3625 &maxfiles, 0, "Maximum number of files"); 3626 3627SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3628 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3629 3630/* ARGSUSED*/ 3631static void 3632filelistinit(void *dummy) 3633{ 3634 3635 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3636 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3637 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 3638 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 3639 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3640} 3641SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3642 3643/*-------------------------------------------------------------------*/ 3644 3645static int 3646badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3647 int flags, struct thread *td) 3648{ 3649 3650 return (EBADF); 3651} 3652 3653static int 3654badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3655 struct thread *td) 3656{ 3657 3658 return (EINVAL); 3659} 3660 3661static int 3662badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3663 struct thread *td) 3664{ 3665 3666 return (EBADF); 3667} 3668 3669static int 3670badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3671 struct thread *td) 3672{ 3673 3674 return (0); 3675} 3676 3677static int 3678badfo_kqfilter(struct file *fp, struct knote *kn) 3679{ 3680 3681 return (EBADF); 3682} 3683 3684static int 3685badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3686 struct thread *td) 3687{ 3688 3689 return (EBADF); 3690} 3691 3692static int 3693badfo_close(struct file *fp, struct thread *td) 3694{ 3695 3696 return (0); 3697} 3698 3699static int 3700badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3701 struct thread *td) 3702{ 3703 3704 return (EBADF); 3705} 3706 3707static int 3708badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3709 struct thread *td) 3710{ 3711 3712 return (EBADF); 3713} 3714 3715static int 3716badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3717 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3718 int kflags, struct thread *td) 3719{ 3720 3721 return (EBADF); 3722} 3723 3724static int 3725badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 3726{ 3727 3728 return (0); 3729} 3730 3731struct fileops badfileops = { 3732 .fo_read = badfo_readwrite, 3733 .fo_write = badfo_readwrite, 3734 .fo_truncate = badfo_truncate, 3735 .fo_ioctl = badfo_ioctl, 3736 .fo_poll = badfo_poll, 3737 .fo_kqfilter = badfo_kqfilter, 3738 .fo_stat = badfo_stat, 3739 .fo_close = badfo_close, 3740 .fo_chmod = badfo_chmod, 3741 .fo_chown = badfo_chown, 3742 .fo_sendfile = badfo_sendfile, 3743 .fo_fill_kinfo = badfo_fill_kinfo, 3744}; 3745 3746int 3747invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 3748 int flags, struct thread *td) 3749{ 3750 3751 return (EOPNOTSUPP); 3752} 3753 3754int 3755invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3756 struct thread *td) 3757{ 3758 3759 return (EINVAL); 3760} 3761 3762int 3763invfo_ioctl(struct file *fp, u_long com, void *data, 3764 struct ucred *active_cred, struct thread *td) 3765{ 3766 3767 return (ENOTTY); 3768} 3769 3770int 3771invfo_poll(struct file *fp, int events, struct ucred *active_cred, 3772 struct thread *td) 3773{ 3774 3775 return (poll_no_poll(events)); 3776} 3777 3778int 3779invfo_kqfilter(struct file *fp, struct knote *kn) 3780{ 3781 3782 return (EINVAL); 3783} 3784 3785int 3786invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3787 struct thread *td) 3788{ 3789 3790 return (EINVAL); 3791} 3792 3793int 3794invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3795 struct thread *td) 3796{ 3797 3798 return (EINVAL); 3799} 3800 3801int 3802invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3803 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3804 int kflags, struct thread *td) 3805{ 3806 3807 return (EINVAL); 3808} 3809 3810/*-------------------------------------------------------------------*/ 3811 3812/* 3813 * File Descriptor pseudo-device driver (/dev/fd/). 3814 * 3815 * Opening minor device N dup()s the file (if any) connected to file 3816 * descriptor N belonging to the calling process. Note that this driver 3817 * consists of only the ``open()'' routine, because all subsequent 3818 * references to this file will be direct to the other driver. 3819 * 3820 * XXX: we could give this one a cloning event handler if necessary. 3821 */ 3822 3823/* ARGSUSED */ 3824static int 3825fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3826{ 3827 3828 /* 3829 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3830 * the file descriptor being sought for duplication. The error 3831 * return ensures that the vnode for this device will be released 3832 * by vn_open. Open will detect this special error and take the 3833 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3834 * will simply report the error. 3835 */ 3836 td->td_dupfd = dev2unit(dev); 3837 return (ENODEV); 3838} 3839 3840static struct cdevsw fildesc_cdevsw = { 3841 .d_version = D_VERSION, 3842 .d_open = fdopen, 3843 .d_name = "FD", 3844}; 3845 3846static void 3847fildesc_drvinit(void *unused) 3848{ 3849 struct cdev *dev; 3850 3851 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3852 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3853 make_dev_alias(dev, "stdin"); 3854 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3855 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3856 make_dev_alias(dev, "stdout"); 3857 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3858 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3859 make_dev_alias(dev, "stderr"); 3860} 3861 3862SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3863