kern_descrip.c revision 283059
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 283059 2015-05-18 14:27:04Z 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 do_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 proc *p); 113 114/* Flags for do_dup() */ 115#define DUP_FIXED 0x1 /* Force fixed allocation. */ 116#define DUP_FCNTL 0x2 /* fcntl()-style errors. */ 117#define DUP_CLOEXEC 0x4 /* Atomically set FD_CLOEXEC. */ 118 119/* 120 * Each process has: 121 * 122 * - An array of open file descriptors (fd_ofiles) 123 * - An array of file flags (fd_ofileflags) 124 * - A bitmap recording which descriptors are in use (fd_map) 125 * 126 * A process starts out with NDFILE descriptors. The value of NDFILE has 127 * been selected based the historical limit of 20 open files, and an 128 * assumption that the majority of processes, especially short-lived 129 * processes like shells, will never need more. 130 * 131 * If this initial allocation is exhausted, a larger descriptor table and 132 * map are allocated dynamically, and the pointers in the process's struct 133 * filedesc are updated to point to those. This is repeated every time 134 * the process runs out of file descriptors (provided it hasn't hit its 135 * resource limit). 136 * 137 * Since threads may hold references to individual descriptor table 138 * entries, the tables are never freed. Instead, they are placed on a 139 * linked list and freed only when the struct filedesc is released. 140 */ 141#define NDFILE 20 142#define NDSLOTSIZE sizeof(NDSLOTTYPE) 143#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 144#define NDSLOT(x) ((x) / NDENTRIES) 145#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 146#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 147 148/* 149 * SLIST entry used to keep track of ofiles which must be reclaimed when 150 * the process exits. 151 */ 152struct freetable { 153 struct fdescenttbl *ft_table; 154 SLIST_ENTRY(freetable) ft_next; 155}; 156 157/* 158 * Initial allocation: a filedesc structure + the head of SLIST used to 159 * keep track of old ofiles + enough space for NDFILE descriptors. 160 */ 161 162struct fdescenttbl0 { 163 int fdt_nfiles; 164 struct filedescent fdt_ofiles[NDFILE]; 165}; 166 167struct filedesc0 { 168 struct filedesc fd_fd; 169 SLIST_HEAD(, freetable) fd_free; 170 struct fdescenttbl0 fd_dfiles; 171 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 172}; 173 174/* 175 * Descriptor management. 176 */ 177volatile int openfiles; /* actual number of open files */ 178struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 179void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 180 181/* A mutex to protect the association between a proc and filedesc. */ 182static struct mtx fdesc_mtx; 183 184/* 185 * If low >= size, just return low. Otherwise find the first zero bit in the 186 * given bitmap, starting at low and not exceeding size - 1. Return size if 187 * not found. 188 */ 189static int 190fd_first_free(struct filedesc *fdp, int low, int size) 191{ 192 NDSLOTTYPE *map = fdp->fd_map; 193 NDSLOTTYPE mask; 194 int off, maxoff; 195 196 if (low >= size) 197 return (low); 198 199 off = NDSLOT(low); 200 if (low % NDENTRIES) { 201 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 202 if ((mask &= ~map[off]) != 0UL) 203 return (off * NDENTRIES + ffsl(mask) - 1); 204 ++off; 205 } 206 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 207 if (map[off] != ~0UL) 208 return (off * NDENTRIES + ffsl(~map[off]) - 1); 209 return (size); 210} 211 212/* 213 * Find the highest non-zero bit in the given bitmap, starting at 0 and 214 * not exceeding size - 1. Return -1 if not found. 215 */ 216static int 217fd_last_used(struct filedesc *fdp, int size) 218{ 219 NDSLOTTYPE *map = fdp->fd_map; 220 NDSLOTTYPE mask; 221 int off, minoff; 222 223 off = NDSLOT(size); 224 if (size % NDENTRIES) { 225 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 226 if ((mask &= map[off]) != 0) 227 return (off * NDENTRIES + flsl(mask) - 1); 228 --off; 229 } 230 for (minoff = NDSLOT(0); off >= minoff; --off) 231 if (map[off] != 0) 232 return (off * NDENTRIES + flsl(map[off]) - 1); 233 return (-1); 234} 235 236#ifdef INVARIANTS 237static int 238fdisused(struct filedesc *fdp, int fd) 239{ 240 241 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 242 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 243 244 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 245} 246#endif 247 248/* 249 * Mark a file descriptor as used. 250 */ 251static void 252fdused_init(struct filedesc *fdp, int fd) 253{ 254 255 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 256 257 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 258} 259 260static void 261fdused(struct filedesc *fdp, int fd) 262{ 263 264 FILEDESC_XLOCK_ASSERT(fdp); 265 266 fdused_init(fdp, fd); 267 if (fd > fdp->fd_lastfile) 268 fdp->fd_lastfile = fd; 269 if (fd == fdp->fd_freefile) 270 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 271} 272 273/* 274 * Mark a file descriptor as unused. 275 */ 276static void 277fdunused(struct filedesc *fdp, int fd) 278{ 279 280 FILEDESC_XLOCK_ASSERT(fdp); 281 282 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 283 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 284 ("fd=%d is still in use", fd)); 285 286 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 287 if (fd < fdp->fd_freefile) 288 fdp->fd_freefile = fd; 289 if (fd == fdp->fd_lastfile) 290 fdp->fd_lastfile = fd_last_used(fdp, fd); 291} 292 293/* 294 * Free a file descriptor. 295 * 296 * Avoid some work if fdp is about to be destroyed. 297 */ 298static inline void 299fdefree_last(struct filedescent *fde) 300{ 301 302 filecaps_free(&fde->fde_caps); 303} 304 305static inline void 306fdfree(struct filedesc *fdp, int fd) 307{ 308 struct filedescent *fde; 309 310 fde = &fdp->fd_ofiles[fd]; 311#ifdef CAPABILITIES 312 seq_write_begin(&fde->fde_seq); 313#endif 314 fdefree_last(fde); 315 bzero(fde, fde_change_size); 316 fdunused(fdp, fd); 317#ifdef CAPABILITIES 318 seq_write_end(&fde->fde_seq); 319#endif 320} 321 322/* 323 * System calls on descriptors. 324 */ 325#ifndef _SYS_SYSPROTO_H_ 326struct getdtablesize_args { 327 int dummy; 328}; 329#endif 330/* ARGSUSED */ 331int 332sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 333{ 334 struct proc *p = td->td_proc; 335 uint64_t lim; 336 337 PROC_LOCK(p); 338 td->td_retval[0] = 339 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 340 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 341 PROC_UNLOCK(p); 342 if (lim < td->td_retval[0]) 343 td->td_retval[0] = lim; 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 (do_dup(td, DUP_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 (do_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; 408 409 error = 0; 410 switch (cmd) { 411 case F_OGETLK: 412 case F_OSETLK: 413 case F_OSETLKW: 414 /* 415 * Convert old flock structure to new. 416 */ 417 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 418 fl.l_start = ofl.l_start; 419 fl.l_len = ofl.l_len; 420 fl.l_pid = ofl.l_pid; 421 fl.l_type = ofl.l_type; 422 fl.l_whence = ofl.l_whence; 423 fl.l_sysid = 0; 424 425 switch (cmd) { 426 case F_OGETLK: 427 cmd = F_GETLK; 428 break; 429 case F_OSETLK: 430 cmd = F_SETLK; 431 break; 432 case F_OSETLKW: 433 cmd = F_SETLKW; 434 break; 435 } 436 arg1 = (intptr_t)&fl; 437 break; 438 case F_GETLK: 439 case F_SETLK: 440 case F_SETLKW: 441 case F_SETLK_REMOTE: 442 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 443 arg1 = (intptr_t)&fl; 444 break; 445 default: 446 arg1 = arg; 447 break; 448 } 449 if (error) 450 return (error); 451 error = kern_fcntl(td, fd, cmd, arg1); 452 if (error) 453 return (error); 454 if (cmd == F_OGETLK) { 455 ofl.l_start = fl.l_start; 456 ofl.l_len = fl.l_len; 457 ofl.l_pid = fl.l_pid; 458 ofl.l_type = fl.l_type; 459 ofl.l_whence = fl.l_whence; 460 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 461 } else if (cmd == F_GETLK) { 462 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 463 } 464 return (error); 465} 466 467int 468kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 469{ 470 struct filedesc *fdp; 471 struct flock *flp; 472 struct file *fp, *fp2; 473 struct filedescent *fde; 474 struct proc *p; 475 struct vnode *vp; 476 cap_rights_t rights; 477 int error, flg, tmp; 478 uint64_t bsize; 479 off_t foffset; 480 481 error = 0; 482 flg = F_POSIX; 483 p = td->td_proc; 484 fdp = p->p_fd; 485 486 switch (cmd) { 487 case F_DUPFD: 488 tmp = arg; 489 error = do_dup(td, DUP_FCNTL, fd, tmp); 490 break; 491 492 case F_DUPFD_CLOEXEC: 493 tmp = arg; 494 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp); 495 break; 496 497 case F_DUP2FD: 498 tmp = arg; 499 error = do_dup(td, DUP_FIXED, fd, tmp); 500 break; 501 502 case F_DUP2FD_CLOEXEC: 503 tmp = arg; 504 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp); 505 break; 506 507 case F_GETFD: 508 FILEDESC_SLOCK(fdp); 509 if (fget_locked(fdp, fd) == NULL) { 510 FILEDESC_SUNLOCK(fdp); 511 error = EBADF; 512 break; 513 } 514 fde = &fdp->fd_ofiles[fd]; 515 td->td_retval[0] = 516 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 517 FILEDESC_SUNLOCK(fdp); 518 break; 519 520 case F_SETFD: 521 FILEDESC_XLOCK(fdp); 522 if (fget_locked(fdp, fd) == NULL) { 523 FILEDESC_XUNLOCK(fdp); 524 error = EBADF; 525 break; 526 } 527 fde = &fdp->fd_ofiles[fd]; 528 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 529 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 530 FILEDESC_XUNLOCK(fdp); 531 break; 532 533 case F_GETFL: 534 error = fget_fcntl(td, fd, 535 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp); 536 if (error != 0) 537 break; 538 td->td_retval[0] = OFLAGS(fp->f_flag); 539 fdrop(fp, td); 540 break; 541 542 case F_SETFL: 543 error = fget_fcntl(td, fd, 544 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp); 545 if (error != 0) 546 break; 547 do { 548 tmp = flg = fp->f_flag; 549 tmp &= ~FCNTLFLAGS; 550 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 551 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 552 tmp = fp->f_flag & FNONBLOCK; 553 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 554 if (error != 0) { 555 fdrop(fp, td); 556 break; 557 } 558 tmp = fp->f_flag & FASYNC; 559 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 560 if (error == 0) { 561 fdrop(fp, td); 562 break; 563 } 564 atomic_clear_int(&fp->f_flag, FNONBLOCK); 565 tmp = 0; 566 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 567 fdrop(fp, td); 568 break; 569 570 case F_GETOWN: 571 error = fget_fcntl(td, fd, 572 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp); 573 if (error != 0) 574 break; 575 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 576 if (error == 0) 577 td->td_retval[0] = tmp; 578 fdrop(fp, td); 579 break; 580 581 case F_SETOWN: 582 error = fget_fcntl(td, fd, 583 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp); 584 if (error != 0) 585 break; 586 tmp = arg; 587 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 588 fdrop(fp, td); 589 break; 590 591 case F_SETLK_REMOTE: 592 error = priv_check(td, PRIV_NFS_LOCKD); 593 if (error) 594 return (error); 595 flg = F_REMOTE; 596 goto do_setlk; 597 598 case F_SETLKW: 599 flg |= F_WAIT; 600 /* FALLTHROUGH F_SETLK */ 601 602 case F_SETLK: 603 do_setlk: 604 cap_rights_init(&rights, CAP_FLOCK); 605 error = fget_unlocked(fdp, fd, &rights, &fp, NULL); 606 if (error != 0) 607 break; 608 if (fp->f_type != DTYPE_VNODE) { 609 error = EBADF; 610 fdrop(fp, td); 611 break; 612 } 613 614 flp = (struct flock *)arg; 615 if (flp->l_whence == SEEK_CUR) { 616 foffset = foffset_get(fp); 617 if (foffset < 0 || 618 (flp->l_start > 0 && 619 foffset > OFF_MAX - flp->l_start)) { 620 error = EOVERFLOW; 621 fdrop(fp, td); 622 break; 623 } 624 flp->l_start += foffset; 625 } 626 627 vp = fp->f_vnode; 628 switch (flp->l_type) { 629 case F_RDLCK: 630 if ((fp->f_flag & FREAD) == 0) { 631 error = EBADF; 632 break; 633 } 634 PROC_LOCK(p->p_leader); 635 p->p_leader->p_flag |= P_ADVLOCK; 636 PROC_UNLOCK(p->p_leader); 637 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 638 flp, flg); 639 break; 640 case F_WRLCK: 641 if ((fp->f_flag & FWRITE) == 0) { 642 error = EBADF; 643 break; 644 } 645 PROC_LOCK(p->p_leader); 646 p->p_leader->p_flag |= P_ADVLOCK; 647 PROC_UNLOCK(p->p_leader); 648 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 649 flp, flg); 650 break; 651 case F_UNLCK: 652 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 653 flp, flg); 654 break; 655 case F_UNLCKSYS: 656 /* 657 * Temporary api for testing remote lock 658 * infrastructure. 659 */ 660 if (flg != F_REMOTE) { 661 error = EINVAL; 662 break; 663 } 664 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 665 F_UNLCKSYS, flp, flg); 666 break; 667 default: 668 error = EINVAL; 669 break; 670 } 671 if (error != 0 || flp->l_type == F_UNLCK || 672 flp->l_type == F_UNLCKSYS) { 673 fdrop(fp, td); 674 break; 675 } 676 677 /* 678 * Check for a race with close. 679 * 680 * The vnode is now advisory locked (or unlocked, but this case 681 * is not really important) as the caller requested. 682 * We had to drop the filedesc lock, so we need to recheck if 683 * the descriptor is still valid, because if it was closed 684 * in the meantime we need to remove advisory lock from the 685 * vnode - close on any descriptor leading to an advisory 686 * locked vnode, removes that lock. 687 * We will return 0 on purpose in that case, as the result of 688 * successful advisory lock might have been externally visible 689 * already. This is fine - effectively we pretend to the caller 690 * that the closing thread was a bit slower and that the 691 * advisory lock succeeded before the close. 692 */ 693 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL); 694 if (error != 0) { 695 fdrop(fp, td); 696 break; 697 } 698 if (fp != fp2) { 699 flp->l_whence = SEEK_SET; 700 flp->l_start = 0; 701 flp->l_len = 0; 702 flp->l_type = F_UNLCK; 703 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 704 F_UNLCK, flp, F_POSIX); 705 } 706 fdrop(fp, td); 707 fdrop(fp2, td); 708 break; 709 710 case F_GETLK: 711 error = fget_unlocked(fdp, fd, 712 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL); 713 if (error != 0) 714 break; 715 if (fp->f_type != DTYPE_VNODE) { 716 error = EBADF; 717 fdrop(fp, td); 718 break; 719 } 720 flp = (struct flock *)arg; 721 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 722 flp->l_type != F_UNLCK) { 723 error = EINVAL; 724 fdrop(fp, td); 725 break; 726 } 727 if (flp->l_whence == SEEK_CUR) { 728 foffset = foffset_get(fp); 729 if ((flp->l_start > 0 && 730 foffset > OFF_MAX - flp->l_start) || 731 (flp->l_start < 0 && 732 foffset < OFF_MIN - flp->l_start)) { 733 error = EOVERFLOW; 734 fdrop(fp, td); 735 break; 736 } 737 flp->l_start += foffset; 738 } 739 vp = fp->f_vnode; 740 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 741 F_POSIX); 742 fdrop(fp, td); 743 break; 744 745 case F_RDAHEAD: 746 arg = arg ? 128 * 1024: 0; 747 /* FALLTHROUGH */ 748 case F_READAHEAD: 749 error = fget_unlocked(fdp, fd, NULL, &fp, NULL); 750 if (error != 0) 751 break; 752 if (fp->f_type != DTYPE_VNODE) { 753 fdrop(fp, td); 754 error = EBADF; 755 break; 756 } 757 vp = fp->f_vnode; 758 /* 759 * Exclusive lock synchronizes against f_seqcount reads and 760 * writes in sequential_heuristic(). 761 */ 762 error = vn_lock(vp, LK_EXCLUSIVE); 763 if (error != 0) { 764 fdrop(fp, td); 765 break; 766 } 767 if (arg >= 0) { 768 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 769 fp->f_seqcount = (arg + bsize - 1) / bsize; 770 atomic_set_int(&fp->f_flag, FRDAHEAD); 771 } else { 772 atomic_clear_int(&fp->f_flag, FRDAHEAD); 773 } 774 VOP_UNLOCK(vp, 0); 775 fdrop(fp, td); 776 break; 777 778 default: 779 error = EINVAL; 780 break; 781 } 782 return (error); 783} 784 785static int 786getmaxfd(struct proc *p) 787{ 788 int maxfd; 789 790 PROC_LOCK(p); 791 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 792 PROC_UNLOCK(p); 793 794 return (maxfd); 795} 796 797/* 798 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 799 */ 800static int 801do_dup(struct thread *td, int flags, int old, int new) 802{ 803 struct filedesc *fdp; 804 struct filedescent *oldfde, *newfde; 805 struct proc *p; 806 struct file *fp; 807 struct file *delfp; 808 int error, maxfd; 809 810 p = td->td_proc; 811 fdp = p->p_fd; 812 813 /* 814 * Verify we have a valid descriptor to dup from and possibly to 815 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 816 * return EINVAL when the new descriptor is out of bounds. 817 */ 818 if (old < 0) 819 return (EBADF); 820 if (new < 0) 821 return (flags & DUP_FCNTL ? EINVAL : EBADF); 822 maxfd = getmaxfd(p); 823 if (new >= maxfd) 824 return (flags & DUP_FCNTL ? EINVAL : EBADF); 825 826 FILEDESC_XLOCK(fdp); 827 if (fget_locked(fdp, old) == NULL) { 828 FILEDESC_XUNLOCK(fdp); 829 return (EBADF); 830 } 831 oldfde = &fdp->fd_ofiles[old]; 832 if (flags & DUP_FIXED && old == new) { 833 td->td_retval[0] = new; 834 if (flags & DUP_CLOEXEC) 835 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 836 FILEDESC_XUNLOCK(fdp); 837 return (0); 838 } 839 fp = oldfde->fde_file; 840 fhold(fp); 841 842 /* 843 * If the caller specified a file descriptor, make sure the file 844 * table is large enough to hold it, and grab it. Otherwise, just 845 * allocate a new descriptor the usual way. 846 */ 847 if (flags & DUP_FIXED) { 848 if (new >= fdp->fd_nfiles) { 849 /* 850 * The resource limits are here instead of e.g. 851 * fdalloc(), because the file descriptor table may be 852 * shared between processes, so we can't really use 853 * racct_add()/racct_sub(). Instead of counting the 854 * number of actually allocated descriptors, just put 855 * the limit on the size of the file descriptor table. 856 */ 857#ifdef RACCT 858 if (racct_enable) { 859 PROC_LOCK(p); 860 error = racct_set(p, RACCT_NOFILE, new + 1); 861 PROC_UNLOCK(p); 862 if (error != 0) { 863 FILEDESC_XUNLOCK(fdp); 864 fdrop(fp, td); 865 return (EMFILE); 866 } 867 } 868#endif 869 fdgrowtable_exp(fdp, new + 1); 870 oldfde = &fdp->fd_ofiles[old]; 871 } 872 newfde = &fdp->fd_ofiles[new]; 873 if (newfde->fde_file == NULL) 874 fdused(fdp, new); 875 } else { 876 if ((error = fdalloc(td, new, &new)) != 0) { 877 FILEDESC_XUNLOCK(fdp); 878 fdrop(fp, td); 879 return (error); 880 } 881 newfde = &fdp->fd_ofiles[new]; 882 } 883 884 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 885 KASSERT(old != new, ("new fd is same as old")); 886 887 delfp = newfde->fde_file; 888 889 /* 890 * Duplicate the source descriptor. 891 */ 892#ifdef CAPABILITIES 893 seq_write_begin(&newfde->fde_seq); 894#endif 895 filecaps_free(&newfde->fde_caps); 896 memcpy(newfde, oldfde, fde_change_size); 897 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 898 if ((flags & DUP_CLOEXEC) != 0) 899 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 900 else 901 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 902#ifdef CAPABILITIES 903 seq_write_end(&newfde->fde_seq); 904#endif 905 td->td_retval[0] = new; 906 907 if (delfp != NULL) { 908 (void) closefp(fdp, new, delfp, td, 1); 909 /* closefp() drops the FILEDESC lock for us. */ 910 } else { 911 FILEDESC_XUNLOCK(fdp); 912 } 913 914 return (0); 915} 916 917/* 918 * If sigio is on the list associated with a process or process group, 919 * disable signalling from the device, remove sigio from the list and 920 * free sigio. 921 */ 922void 923funsetown(struct sigio **sigiop) 924{ 925 struct sigio *sigio; 926 927 SIGIO_LOCK(); 928 sigio = *sigiop; 929 if (sigio == NULL) { 930 SIGIO_UNLOCK(); 931 return; 932 } 933 *(sigio->sio_myref) = NULL; 934 if ((sigio)->sio_pgid < 0) { 935 struct pgrp *pg = (sigio)->sio_pgrp; 936 PGRP_LOCK(pg); 937 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 938 sigio, sio_pgsigio); 939 PGRP_UNLOCK(pg); 940 } else { 941 struct proc *p = (sigio)->sio_proc; 942 PROC_LOCK(p); 943 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 944 sigio, sio_pgsigio); 945 PROC_UNLOCK(p); 946 } 947 SIGIO_UNLOCK(); 948 crfree(sigio->sio_ucred); 949 free(sigio, M_SIGIO); 950} 951 952/* 953 * Free a list of sigio structures. 954 * We only need to lock the SIGIO_LOCK because we have made ourselves 955 * inaccessible to callers of fsetown and therefore do not need to lock 956 * the proc or pgrp struct for the list manipulation. 957 */ 958void 959funsetownlst(struct sigiolst *sigiolst) 960{ 961 struct proc *p; 962 struct pgrp *pg; 963 struct sigio *sigio; 964 965 sigio = SLIST_FIRST(sigiolst); 966 if (sigio == NULL) 967 return; 968 p = NULL; 969 pg = NULL; 970 971 /* 972 * Every entry of the list should belong 973 * to a single proc or pgrp. 974 */ 975 if (sigio->sio_pgid < 0) { 976 pg = sigio->sio_pgrp; 977 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 978 } else /* if (sigio->sio_pgid > 0) */ { 979 p = sigio->sio_proc; 980 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 981 } 982 983 SIGIO_LOCK(); 984 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 985 *(sigio->sio_myref) = NULL; 986 if (pg != NULL) { 987 KASSERT(sigio->sio_pgid < 0, 988 ("Proc sigio in pgrp sigio list")); 989 KASSERT(sigio->sio_pgrp == pg, 990 ("Bogus pgrp in sigio list")); 991 PGRP_LOCK(pg); 992 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 993 sio_pgsigio); 994 PGRP_UNLOCK(pg); 995 } else /* if (p != NULL) */ { 996 KASSERT(sigio->sio_pgid > 0, 997 ("Pgrp sigio in proc sigio list")); 998 KASSERT(sigio->sio_proc == p, 999 ("Bogus proc in sigio list")); 1000 PROC_LOCK(p); 1001 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 1002 sio_pgsigio); 1003 PROC_UNLOCK(p); 1004 } 1005 SIGIO_UNLOCK(); 1006 crfree(sigio->sio_ucred); 1007 free(sigio, M_SIGIO); 1008 SIGIO_LOCK(); 1009 } 1010 SIGIO_UNLOCK(); 1011} 1012 1013/* 1014 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1015 * 1016 * After permission checking, add a sigio structure to the sigio list for 1017 * the process or process group. 1018 */ 1019int 1020fsetown(pid_t pgid, struct sigio **sigiop) 1021{ 1022 struct proc *proc; 1023 struct pgrp *pgrp; 1024 struct sigio *sigio; 1025 int ret; 1026 1027 if (pgid == 0) { 1028 funsetown(sigiop); 1029 return (0); 1030 } 1031 1032 ret = 0; 1033 1034 /* Allocate and fill in the new sigio out of locks. */ 1035 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1036 sigio->sio_pgid = pgid; 1037 sigio->sio_ucred = crhold(curthread->td_ucred); 1038 sigio->sio_myref = sigiop; 1039 1040 sx_slock(&proctree_lock); 1041 if (pgid > 0) { 1042 proc = pfind(pgid); 1043 if (proc == NULL) { 1044 ret = ESRCH; 1045 goto fail; 1046 } 1047 1048 /* 1049 * Policy - Don't allow a process to FSETOWN a process 1050 * in another session. 1051 * 1052 * Remove this test to allow maximum flexibility or 1053 * restrict FSETOWN to the current process or process 1054 * group for maximum safety. 1055 */ 1056 PROC_UNLOCK(proc); 1057 if (proc->p_session != curthread->td_proc->p_session) { 1058 ret = EPERM; 1059 goto fail; 1060 } 1061 1062 pgrp = NULL; 1063 } else /* if (pgid < 0) */ { 1064 pgrp = pgfind(-pgid); 1065 if (pgrp == NULL) { 1066 ret = ESRCH; 1067 goto fail; 1068 } 1069 PGRP_UNLOCK(pgrp); 1070 1071 /* 1072 * Policy - Don't allow a process to FSETOWN a process 1073 * in another session. 1074 * 1075 * Remove this test to allow maximum flexibility or 1076 * restrict FSETOWN to the current process or process 1077 * group for maximum safety. 1078 */ 1079 if (pgrp->pg_session != curthread->td_proc->p_session) { 1080 ret = EPERM; 1081 goto fail; 1082 } 1083 1084 proc = NULL; 1085 } 1086 funsetown(sigiop); 1087 if (pgid > 0) { 1088 PROC_LOCK(proc); 1089 /* 1090 * Since funsetownlst() is called without the proctree 1091 * locked, we need to check for P_WEXIT. 1092 * XXX: is ESRCH correct? 1093 */ 1094 if ((proc->p_flag & P_WEXIT) != 0) { 1095 PROC_UNLOCK(proc); 1096 ret = ESRCH; 1097 goto fail; 1098 } 1099 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1100 sigio->sio_proc = proc; 1101 PROC_UNLOCK(proc); 1102 } else { 1103 PGRP_LOCK(pgrp); 1104 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1105 sigio->sio_pgrp = pgrp; 1106 PGRP_UNLOCK(pgrp); 1107 } 1108 sx_sunlock(&proctree_lock); 1109 SIGIO_LOCK(); 1110 *sigiop = sigio; 1111 SIGIO_UNLOCK(); 1112 return (0); 1113 1114fail: 1115 sx_sunlock(&proctree_lock); 1116 crfree(sigio->sio_ucred); 1117 free(sigio, M_SIGIO); 1118 return (ret); 1119} 1120 1121/* 1122 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1123 */ 1124pid_t 1125fgetown(sigiop) 1126 struct sigio **sigiop; 1127{ 1128 pid_t pgid; 1129 1130 SIGIO_LOCK(); 1131 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1132 SIGIO_UNLOCK(); 1133 return (pgid); 1134} 1135 1136/* 1137 * Function drops the filedesc lock on return. 1138 */ 1139static int 1140closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1141 int holdleaders) 1142{ 1143 int error; 1144 1145 FILEDESC_XLOCK_ASSERT(fdp); 1146 1147 if (holdleaders) { 1148 if (td->td_proc->p_fdtol != NULL) { 1149 /* 1150 * Ask fdfree() to sleep to ensure that all relevant 1151 * process leaders can be traversed in closef(). 1152 */ 1153 fdp->fd_holdleaderscount++; 1154 } else { 1155 holdleaders = 0; 1156 } 1157 } 1158 1159 /* 1160 * We now hold the fp reference that used to be owned by the 1161 * descriptor array. We have to unlock the FILEDESC *AFTER* 1162 * knote_fdclose to prevent a race of the fd getting opened, a knote 1163 * added, and deleteing a knote for the new fd. 1164 */ 1165 knote_fdclose(td, fd); 1166 1167 /* 1168 * We need to notify mqueue if the object is of type mqueue. 1169 */ 1170 if (fp->f_type == DTYPE_MQUEUE) 1171 mq_fdclose(td, fd, fp); 1172 FILEDESC_XUNLOCK(fdp); 1173 1174 error = closef(fp, td); 1175 if (holdleaders) { 1176 FILEDESC_XLOCK(fdp); 1177 fdp->fd_holdleaderscount--; 1178 if (fdp->fd_holdleaderscount == 0 && 1179 fdp->fd_holdleaderswakeup != 0) { 1180 fdp->fd_holdleaderswakeup = 0; 1181 wakeup(&fdp->fd_holdleaderscount); 1182 } 1183 FILEDESC_XUNLOCK(fdp); 1184 } 1185 return (error); 1186} 1187 1188/* 1189 * Close a file descriptor. 1190 */ 1191#ifndef _SYS_SYSPROTO_H_ 1192struct close_args { 1193 int fd; 1194}; 1195#endif 1196/* ARGSUSED */ 1197int 1198sys_close(td, uap) 1199 struct thread *td; 1200 struct close_args *uap; 1201{ 1202 1203 return (kern_close(td, uap->fd)); 1204} 1205 1206int 1207kern_close(td, fd) 1208 struct thread *td; 1209 int fd; 1210{ 1211 struct filedesc *fdp; 1212 struct file *fp; 1213 1214 fdp = td->td_proc->p_fd; 1215 1216 AUDIT_SYSCLOSE(td, fd); 1217 1218 FILEDESC_XLOCK(fdp); 1219 if ((fp = fget_locked(fdp, fd)) == NULL) { 1220 FILEDESC_XUNLOCK(fdp); 1221 return (EBADF); 1222 } 1223 fdfree(fdp, fd); 1224 1225 /* closefp() drops the FILEDESC lock for us. */ 1226 return (closefp(fdp, fd, fp, td, 1)); 1227} 1228 1229/* 1230 * Close open file descriptors. 1231 */ 1232#ifndef _SYS_SYSPROTO_H_ 1233struct closefrom_args { 1234 int lowfd; 1235}; 1236#endif 1237/* ARGSUSED */ 1238int 1239sys_closefrom(struct thread *td, struct closefrom_args *uap) 1240{ 1241 struct filedesc *fdp; 1242 int fd; 1243 1244 fdp = td->td_proc->p_fd; 1245 AUDIT_ARG_FD(uap->lowfd); 1246 1247 /* 1248 * Treat negative starting file descriptor values identical to 1249 * closefrom(0) which closes all files. 1250 */ 1251 if (uap->lowfd < 0) 1252 uap->lowfd = 0; 1253 FILEDESC_SLOCK(fdp); 1254 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { 1255 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1256 FILEDESC_SUNLOCK(fdp); 1257 (void)kern_close(td, fd); 1258 FILEDESC_SLOCK(fdp); 1259 } 1260 } 1261 FILEDESC_SUNLOCK(fdp); 1262 return (0); 1263} 1264 1265#if defined(COMPAT_43) 1266/* 1267 * Return status information about a file descriptor. 1268 */ 1269#ifndef _SYS_SYSPROTO_H_ 1270struct ofstat_args { 1271 int fd; 1272 struct ostat *sb; 1273}; 1274#endif 1275/* ARGSUSED */ 1276int 1277ofstat(struct thread *td, struct ofstat_args *uap) 1278{ 1279 struct ostat oub; 1280 struct stat ub; 1281 int error; 1282 1283 error = kern_fstat(td, uap->fd, &ub); 1284 if (error == 0) { 1285 cvtstat(&ub, &oub); 1286 error = copyout(&oub, uap->sb, sizeof(oub)); 1287 } 1288 return (error); 1289} 1290#endif /* COMPAT_43 */ 1291 1292/* 1293 * Return status information about a file descriptor. 1294 */ 1295#ifndef _SYS_SYSPROTO_H_ 1296struct fstat_args { 1297 int fd; 1298 struct stat *sb; 1299}; 1300#endif 1301/* ARGSUSED */ 1302int 1303sys_fstat(struct thread *td, struct fstat_args *uap) 1304{ 1305 struct stat ub; 1306 int error; 1307 1308 error = kern_fstat(td, uap->fd, &ub); 1309 if (error == 0) 1310 error = copyout(&ub, uap->sb, sizeof(ub)); 1311 return (error); 1312} 1313 1314int 1315kern_fstat(struct thread *td, int fd, struct stat *sbp) 1316{ 1317 struct file *fp; 1318 cap_rights_t rights; 1319 int error; 1320 1321 AUDIT_ARG_FD(fd); 1322 1323 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1324 if (error != 0) 1325 return (error); 1326 1327 AUDIT_ARG_FILE(td->td_proc, fp); 1328 1329 error = fo_stat(fp, sbp, td->td_ucred, td); 1330 fdrop(fp, td); 1331#ifdef KTRACE 1332 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1333 ktrstat(sbp); 1334#endif 1335 return (error); 1336} 1337 1338/* 1339 * Return status information about a file descriptor. 1340 */ 1341#ifndef _SYS_SYSPROTO_H_ 1342struct nfstat_args { 1343 int fd; 1344 struct nstat *sb; 1345}; 1346#endif 1347/* ARGSUSED */ 1348int 1349sys_nfstat(struct thread *td, struct nfstat_args *uap) 1350{ 1351 struct nstat nub; 1352 struct stat ub; 1353 int error; 1354 1355 error = kern_fstat(td, uap->fd, &ub); 1356 if (error == 0) { 1357 cvtnstat(&ub, &nub); 1358 error = copyout(&nub, uap->sb, sizeof(nub)); 1359 } 1360 return (error); 1361} 1362 1363/* 1364 * Return pathconf information about a file descriptor. 1365 */ 1366#ifndef _SYS_SYSPROTO_H_ 1367struct fpathconf_args { 1368 int fd; 1369 int name; 1370}; 1371#endif 1372/* ARGSUSED */ 1373int 1374sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1375{ 1376 struct file *fp; 1377 struct vnode *vp; 1378 cap_rights_t rights; 1379 int error; 1380 1381 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1382 if (error != 0) 1383 return (error); 1384 1385 /* If asynchronous I/O is available, it works for all descriptors. */ 1386 if (uap->name == _PC_ASYNC_IO) { 1387 td->td_retval[0] = async_io_version; 1388 goto out; 1389 } 1390 vp = fp->f_vnode; 1391 if (vp != NULL) { 1392 vn_lock(vp, LK_SHARED | LK_RETRY); 1393 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1394 VOP_UNLOCK(vp, 0); 1395 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1396 if (uap->name != _PC_PIPE_BUF) { 1397 error = EINVAL; 1398 } else { 1399 td->td_retval[0] = PIPE_BUF; 1400 error = 0; 1401 } 1402 } else { 1403 error = EOPNOTSUPP; 1404 } 1405out: 1406 fdrop(fp, td); 1407 return (error); 1408} 1409 1410/* 1411 * Initialize filecaps structure. 1412 */ 1413void 1414filecaps_init(struct filecaps *fcaps) 1415{ 1416 1417 bzero(fcaps, sizeof(*fcaps)); 1418 fcaps->fc_nioctls = -1; 1419} 1420 1421/* 1422 * Copy filecaps structure allocating memory for ioctls array if needed. 1423 */ 1424void 1425filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1426{ 1427 size_t size; 1428 1429 *dst = *src; 1430 if (src->fc_ioctls != NULL) { 1431 KASSERT(src->fc_nioctls > 0, 1432 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1433 1434 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1435 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1436 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1437 } 1438} 1439 1440/* 1441 * Move filecaps structure to the new place and clear the old place. 1442 */ 1443void 1444filecaps_move(struct filecaps *src, struct filecaps *dst) 1445{ 1446 1447 *dst = *src; 1448 bzero(src, sizeof(*src)); 1449} 1450 1451/* 1452 * Fill the given filecaps structure with full rights. 1453 */ 1454static void 1455filecaps_fill(struct filecaps *fcaps) 1456{ 1457 1458 CAP_ALL(&fcaps->fc_rights); 1459 fcaps->fc_ioctls = NULL; 1460 fcaps->fc_nioctls = -1; 1461 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1462} 1463 1464/* 1465 * Free memory allocated within filecaps structure. 1466 */ 1467void 1468filecaps_free(struct filecaps *fcaps) 1469{ 1470 1471 free(fcaps->fc_ioctls, M_FILECAPS); 1472 bzero(fcaps, sizeof(*fcaps)); 1473} 1474 1475/* 1476 * Validate the given filecaps structure. 1477 */ 1478static void 1479filecaps_validate(const struct filecaps *fcaps, const char *func) 1480{ 1481 1482 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1483 ("%s: invalid rights", func)); 1484 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1485 ("%s: invalid fcntls", func)); 1486 KASSERT(fcaps->fc_fcntls == 0 || 1487 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1488 ("%s: fcntls without CAP_FCNTL", func)); 1489 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1490 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1491 ("%s: invalid ioctls", func)); 1492 KASSERT(fcaps->fc_nioctls == 0 || 1493 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1494 ("%s: ioctls without CAP_IOCTL", func)); 1495} 1496 1497static void 1498fdgrowtable_exp(struct filedesc *fdp, int nfd) 1499{ 1500 int nfd1; 1501 1502 FILEDESC_XLOCK_ASSERT(fdp); 1503 1504 nfd1 = fdp->fd_nfiles * 2; 1505 if (nfd1 < nfd) 1506 nfd1 = nfd; 1507 fdgrowtable(fdp, nfd1); 1508} 1509 1510/* 1511 * Grow the file table to accomodate (at least) nfd descriptors. 1512 */ 1513static void 1514fdgrowtable(struct filedesc *fdp, int nfd) 1515{ 1516 struct filedesc0 *fdp0; 1517 struct freetable *ft; 1518 struct fdescenttbl *ntable; 1519 struct fdescenttbl *otable; 1520 int nnfiles, onfiles; 1521 NDSLOTTYPE *nmap, *omap; 1522 1523 /* 1524 * If lastfile is -1 this struct filedesc was just allocated and we are 1525 * growing it to accomodate for the one we are going to copy from. There 1526 * is no need to have a lock on this one as it's not visible to anyone. 1527 */ 1528 if (fdp->fd_lastfile != -1) 1529 FILEDESC_XLOCK_ASSERT(fdp); 1530 1531 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1532 1533 /* save old values */ 1534 onfiles = fdp->fd_nfiles; 1535 otable = fdp->fd_files; 1536 omap = fdp->fd_map; 1537 1538 /* compute the size of the new table */ 1539 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1540 if (nnfiles <= onfiles) 1541 /* the table is already large enough */ 1542 return; 1543 1544 /* 1545 * Allocate a new table. We need enough space for the number of 1546 * entries, file entries themselves and the struct freetable we will use 1547 * when we decommission the table and place it on the freelist. 1548 * We place the struct freetable in the middle so we don't have 1549 * to worry about padding. 1550 */ 1551 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1552 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1553 sizeof(struct freetable), 1554 M_FILEDESC, M_ZERO | M_WAITOK); 1555 /* copy the old data */ 1556 ntable->fdt_nfiles = nnfiles; 1557 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1558 onfiles * sizeof(ntable->fdt_ofiles[0])); 1559 1560 /* 1561 * Allocate a new map only if the old is not large enough. It will 1562 * grow at a slower rate than the table as it can map more 1563 * entries than the table can hold. 1564 */ 1565 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1566 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1567 M_ZERO | M_WAITOK); 1568 /* copy over the old data and update the pointer */ 1569 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1570 fdp->fd_map = nmap; 1571 } 1572 1573 /* 1574 * Make sure that ntable is correctly initialized before we replace 1575 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1576 * data. 1577 */ 1578 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1579 1580 /* 1581 * Do not free the old file table, as some threads may still 1582 * reference entries within it. Instead, place it on a freelist 1583 * which will be processed when the struct filedesc is released. 1584 * 1585 * Note that if onfiles == NDFILE, we're dealing with the original 1586 * static allocation contained within (struct filedesc0 *)fdp, 1587 * which must not be freed. 1588 */ 1589 if (onfiles > NDFILE) { 1590 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1591 fdp0 = (struct filedesc0 *)fdp; 1592 ft->ft_table = otable; 1593 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1594 } 1595 /* 1596 * The map does not have the same possibility of threads still 1597 * holding references to it. So always free it as long as it 1598 * does not reference the original static allocation. 1599 */ 1600 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1601 free(omap, M_FILEDESC); 1602} 1603 1604/* 1605 * Allocate a file descriptor for the process. 1606 */ 1607int 1608fdalloc(struct thread *td, int minfd, int *result) 1609{ 1610 struct proc *p = td->td_proc; 1611 struct filedesc *fdp = p->p_fd; 1612 int fd, maxfd, allocfd; 1613#ifdef RACCT 1614 int error; 1615#endif 1616 1617 FILEDESC_XLOCK_ASSERT(fdp); 1618 1619 if (fdp->fd_freefile > minfd) 1620 minfd = fdp->fd_freefile; 1621 1622 maxfd = getmaxfd(p); 1623 1624 /* 1625 * Search the bitmap for a free descriptor starting at minfd. 1626 * If none is found, grow the file table. 1627 */ 1628 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1629 if (fd >= maxfd) 1630 return (EMFILE); 1631 if (fd >= fdp->fd_nfiles) { 1632 allocfd = min(fd * 2, maxfd); 1633#ifdef RACCT 1634 if (racct_enable) { 1635 PROC_LOCK(p); 1636 error = racct_set(p, RACCT_NOFILE, allocfd); 1637 PROC_UNLOCK(p); 1638 if (error != 0) 1639 return (EMFILE); 1640 } 1641#endif 1642 /* 1643 * fd is already equal to first free descriptor >= minfd, so 1644 * we only need to grow the table and we are done. 1645 */ 1646 fdgrowtable_exp(fdp, allocfd); 1647 } 1648 1649 /* 1650 * Perform some sanity checks, then mark the file descriptor as 1651 * used and return it to the caller. 1652 */ 1653 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1654 ("invalid descriptor %d", fd)); 1655 KASSERT(!fdisused(fdp, fd), 1656 ("fd_first_free() returned non-free descriptor")); 1657 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1658 ("file descriptor isn't free")); 1659 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); 1660 fdused(fdp, fd); 1661 *result = fd; 1662 return (0); 1663} 1664 1665/* 1666 * Allocate n file descriptors for the process. 1667 */ 1668int 1669fdallocn(struct thread *td, int minfd, int *fds, int n) 1670{ 1671 struct proc *p = td->td_proc; 1672 struct filedesc *fdp = p->p_fd; 1673 int i; 1674 1675 FILEDESC_XLOCK_ASSERT(fdp); 1676 1677 for (i = 0; i < n; i++) 1678 if (fdalloc(td, 0, &fds[i]) != 0) 1679 break; 1680 1681 if (i < n) { 1682 for (i--; i >= 0; i--) 1683 fdunused(fdp, fds[i]); 1684 return (EMFILE); 1685 } 1686 1687 return (0); 1688} 1689 1690/* 1691 * Create a new open file structure and allocate a file decriptor for the 1692 * process that refers to it. We add one reference to the file for the 1693 * descriptor table and one reference for resultfp. This is to prevent us 1694 * being preempted and the entry in the descriptor table closed after we 1695 * release the FILEDESC lock. 1696 */ 1697int 1698falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1699{ 1700 struct file *fp; 1701 int error, fd; 1702 1703 error = falloc_noinstall(td, &fp); 1704 if (error) 1705 return (error); /* no reference held on error */ 1706 1707 error = finstall(td, fp, &fd, flags, NULL); 1708 if (error) { 1709 fdrop(fp, td); /* one reference (fp only) */ 1710 return (error); 1711 } 1712 1713 if (resultfp != NULL) 1714 *resultfp = fp; /* copy out result */ 1715 else 1716 fdrop(fp, td); /* release local reference */ 1717 1718 if (resultfd != NULL) 1719 *resultfd = fd; 1720 1721 return (0); 1722} 1723 1724/* 1725 * Create a new open file structure without allocating a file descriptor. 1726 */ 1727int 1728falloc_noinstall(struct thread *td, struct file **resultfp) 1729{ 1730 struct file *fp; 1731 int maxuserfiles = maxfiles - (maxfiles / 20); 1732 static struct timeval lastfail; 1733 static int curfail; 1734 1735 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1736 1737 if ((openfiles >= maxuserfiles && 1738 priv_check(td, PRIV_MAXFILES) != 0) || 1739 openfiles >= maxfiles) { 1740 if (ppsratecheck(&lastfail, &curfail, 1)) { 1741 printf("kern.maxfiles limit exceeded by uid %i, " 1742 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1743 } 1744 return (ENFILE); 1745 } 1746 atomic_add_int(&openfiles, 1); 1747 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1748 refcount_init(&fp->f_count, 1); 1749 fp->f_cred = crhold(td->td_ucred); 1750 fp->f_ops = &badfileops; 1751 *resultfp = fp; 1752 return (0); 1753} 1754 1755/* 1756 * Install a file in a file descriptor table. 1757 */ 1758int 1759finstall(struct thread *td, struct file *fp, int *fd, int flags, 1760 struct filecaps *fcaps) 1761{ 1762 struct filedesc *fdp = td->td_proc->p_fd; 1763 struct filedescent *fde; 1764 int error; 1765 1766 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1767 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1768 if (fcaps != NULL) 1769 filecaps_validate(fcaps, __func__); 1770 1771 FILEDESC_XLOCK(fdp); 1772 if ((error = fdalloc(td, 0, fd))) { 1773 FILEDESC_XUNLOCK(fdp); 1774 return (error); 1775 } 1776 fhold(fp); 1777 fde = &fdp->fd_ofiles[*fd]; 1778#ifdef CAPABILITIES 1779 seq_write_begin(&fde->fde_seq); 1780#endif 1781 fde->fde_file = fp; 1782 if ((flags & O_CLOEXEC) != 0) 1783 fde->fde_flags |= UF_EXCLOSE; 1784 if (fcaps != NULL) 1785 filecaps_move(fcaps, &fde->fde_caps); 1786 else 1787 filecaps_fill(&fde->fde_caps); 1788#ifdef CAPABILITIES 1789 seq_write_end(&fde->fde_seq); 1790#endif 1791 FILEDESC_XUNLOCK(fdp); 1792 return (0); 1793} 1794 1795/* 1796 * Build a new filedesc structure from another. 1797 * Copy the current, root, and jail root vnode references. 1798 * 1799 * If fdp is not NULL, return with it shared locked. 1800 */ 1801struct filedesc * 1802fdinit(struct filedesc *fdp, bool prepfiles) 1803{ 1804 struct filedesc0 *newfdp0; 1805 struct filedesc *newfdp; 1806 1807 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 1808 newfdp = &newfdp0->fd_fd; 1809 1810 /* Create the file descriptor table. */ 1811 FILEDESC_LOCK_INIT(newfdp); 1812 newfdp->fd_refcnt = 1; 1813 newfdp->fd_holdcnt = 1; 1814 newfdp->fd_cmask = CMASK; 1815 newfdp->fd_map = newfdp0->fd_dmap; 1816 newfdp->fd_lastfile = -1; 1817 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 1818 newfdp->fd_files->fdt_nfiles = NDFILE; 1819 1820 if (fdp == NULL) 1821 return (newfdp); 1822 1823 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) 1824 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1825 1826 FILEDESC_SLOCK(fdp); 1827 newfdp->fd_cdir = fdp->fd_cdir; 1828 if (newfdp->fd_cdir) 1829 VREF(newfdp->fd_cdir); 1830 newfdp->fd_rdir = fdp->fd_rdir; 1831 if (newfdp->fd_rdir) 1832 VREF(newfdp->fd_rdir); 1833 newfdp->fd_jdir = fdp->fd_jdir; 1834 if (newfdp->fd_jdir) 1835 VREF(newfdp->fd_jdir); 1836 1837 if (!prepfiles) { 1838 FILEDESC_SUNLOCK(fdp); 1839 } else { 1840 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1841 FILEDESC_SUNLOCK(fdp); 1842 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1843 FILEDESC_SLOCK(fdp); 1844 } 1845 } 1846 1847 return (newfdp); 1848} 1849 1850static struct filedesc * 1851fdhold(struct proc *p) 1852{ 1853 struct filedesc *fdp; 1854 1855 mtx_lock(&fdesc_mtx); 1856 fdp = p->p_fd; 1857 if (fdp != NULL) 1858 fdp->fd_holdcnt++; 1859 mtx_unlock(&fdesc_mtx); 1860 return (fdp); 1861} 1862 1863static void 1864fddrop(struct filedesc *fdp) 1865{ 1866 int i; 1867 1868 if (fdp->fd_holdcnt > 1) { 1869 mtx_lock(&fdesc_mtx); 1870 i = --fdp->fd_holdcnt; 1871 mtx_unlock(&fdesc_mtx); 1872 if (i > 0) 1873 return; 1874 } 1875 1876 FILEDESC_LOCK_DESTROY(fdp); 1877 uma_zfree(filedesc0_zone, fdp); 1878} 1879 1880/* 1881 * Share a filedesc structure. 1882 */ 1883struct filedesc * 1884fdshare(struct filedesc *fdp) 1885{ 1886 1887 FILEDESC_XLOCK(fdp); 1888 fdp->fd_refcnt++; 1889 FILEDESC_XUNLOCK(fdp); 1890 return (fdp); 1891} 1892 1893/* 1894 * Unshare a filedesc structure, if necessary by making a copy 1895 */ 1896void 1897fdunshare(struct thread *td) 1898{ 1899 struct filedesc *tmp; 1900 struct proc *p = td->td_proc; 1901 1902 if (p->p_fd->fd_refcnt == 1) 1903 return; 1904 1905 tmp = fdcopy(p->p_fd); 1906 fdescfree(td); 1907 p->p_fd = tmp; 1908} 1909 1910/* 1911 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1912 * this is to ease callers, not catch errors. 1913 */ 1914struct filedesc * 1915fdcopy(struct filedesc *fdp) 1916{ 1917 struct filedesc *newfdp; 1918 struct filedescent *nfde, *ofde; 1919 int i; 1920 1921 MPASS(fdp != NULL); 1922 1923 newfdp = fdinit(fdp, true); 1924 /* copy all passable descriptors (i.e. not kqueue) */ 1925 newfdp->fd_freefile = -1; 1926 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1927 ofde = &fdp->fd_ofiles[i]; 1928 if (ofde->fde_file == NULL || 1929 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { 1930 if (newfdp->fd_freefile == -1) 1931 newfdp->fd_freefile = i; 1932 continue; 1933 } 1934 nfde = &newfdp->fd_ofiles[i]; 1935 *nfde = *ofde; 1936 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1937 fhold(nfde->fde_file); 1938 fdused_init(newfdp, i); 1939 newfdp->fd_lastfile = i; 1940 } 1941 if (newfdp->fd_freefile == -1) 1942 newfdp->fd_freefile = i; 1943 newfdp->fd_cmask = fdp->fd_cmask; 1944 FILEDESC_SUNLOCK(fdp); 1945 return (newfdp); 1946} 1947 1948/* 1949 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 1950 * one of processes using it exits) and the table used to be shared. 1951 */ 1952static void 1953fdclearlocks(struct thread *td) 1954{ 1955 struct filedesc *fdp; 1956 struct filedesc_to_leader *fdtol; 1957 struct flock lf; 1958 struct file *fp; 1959 struct proc *p; 1960 struct vnode *vp; 1961 int i; 1962 1963 p = td->td_proc; 1964 fdp = p->p_fd; 1965 fdtol = p->p_fdtol; 1966 MPASS(fdtol != NULL); 1967 1968 FILEDESC_XLOCK(fdp); 1969 KASSERT(fdtol->fdl_refcount > 0, 1970 ("filedesc_to_refcount botch: fdl_refcount=%d", 1971 fdtol->fdl_refcount)); 1972 if (fdtol->fdl_refcount == 1 && 1973 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1974 for (i = 0; i <= fdp->fd_lastfile; i++) { 1975 fp = fdp->fd_ofiles[i].fde_file; 1976 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1977 continue; 1978 fhold(fp); 1979 FILEDESC_XUNLOCK(fdp); 1980 lf.l_whence = SEEK_SET; 1981 lf.l_start = 0; 1982 lf.l_len = 0; 1983 lf.l_type = F_UNLCK; 1984 vp = fp->f_vnode; 1985 (void) VOP_ADVLOCK(vp, 1986 (caddr_t)p->p_leader, F_UNLCK, 1987 &lf, F_POSIX); 1988 FILEDESC_XLOCK(fdp); 1989 fdrop(fp, td); 1990 } 1991 } 1992retry: 1993 if (fdtol->fdl_refcount == 1) { 1994 if (fdp->fd_holdleaderscount > 0 && 1995 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 1996 /* 1997 * close() or do_dup() has cleared a reference 1998 * in a shared file descriptor table. 1999 */ 2000 fdp->fd_holdleaderswakeup = 1; 2001 sx_sleep(&fdp->fd_holdleaderscount, 2002 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2003 goto retry; 2004 } 2005 if (fdtol->fdl_holdcount > 0) { 2006 /* 2007 * Ensure that fdtol->fdl_leader remains 2008 * valid in closef(). 2009 */ 2010 fdtol->fdl_wakeup = 1; 2011 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2012 "fdlhold", 0); 2013 goto retry; 2014 } 2015 } 2016 fdtol->fdl_refcount--; 2017 if (fdtol->fdl_refcount == 0 && 2018 fdtol->fdl_holdcount == 0) { 2019 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2020 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2021 } else 2022 fdtol = NULL; 2023 p->p_fdtol = NULL; 2024 FILEDESC_XUNLOCK(fdp); 2025 if (fdtol != NULL) 2026 free(fdtol, M_FILEDESC_TO_LEADER); 2027} 2028 2029/* 2030 * Release a filedesc structure. 2031 */ 2032void 2033fdescfree(struct thread *td) 2034{ 2035 struct filedesc0 *fdp0; 2036 struct filedesc *fdp; 2037 struct freetable *ft, *tft; 2038 struct filedescent *fde; 2039 struct file *fp; 2040 struct vnode *cdir, *jdir, *rdir; 2041 int i; 2042 2043 fdp = td->td_proc->p_fd; 2044 MPASS(fdp != NULL); 2045 2046#ifdef RACCT 2047 if (racct_enable) { 2048 PROC_LOCK(td->td_proc); 2049 racct_set(td->td_proc, RACCT_NOFILE, 0); 2050 PROC_UNLOCK(td->td_proc); 2051 } 2052#endif 2053 2054 if (td->td_proc->p_fdtol != NULL) 2055 fdclearlocks(td); 2056 2057 mtx_lock(&fdesc_mtx); 2058 td->td_proc->p_fd = NULL; 2059 mtx_unlock(&fdesc_mtx); 2060 2061 FILEDESC_XLOCK(fdp); 2062 i = --fdp->fd_refcnt; 2063 if (i > 0) { 2064 FILEDESC_XUNLOCK(fdp); 2065 return; 2066 } 2067 2068 cdir = fdp->fd_cdir; 2069 fdp->fd_cdir = NULL; 2070 rdir = fdp->fd_rdir; 2071 fdp->fd_rdir = NULL; 2072 jdir = fdp->fd_jdir; 2073 fdp->fd_jdir = NULL; 2074 FILEDESC_XUNLOCK(fdp); 2075 2076 for (i = 0; i <= fdp->fd_lastfile; i++) { 2077 fde = &fdp->fd_ofiles[i]; 2078 fp = fde->fde_file; 2079 if (fp != NULL) { 2080 fdefree_last(fde); 2081 (void) closef(fp, td); 2082 } 2083 } 2084 2085 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2086 free(fdp->fd_map, M_FILEDESC); 2087 if (fdp->fd_nfiles > NDFILE) 2088 free(fdp->fd_files, M_FILEDESC); 2089 2090 fdp0 = (struct filedesc0 *)fdp; 2091 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2092 free(ft->ft_table, M_FILEDESC); 2093 2094 if (cdir != NULL) 2095 vrele(cdir); 2096 if (rdir != NULL) 2097 vrele(rdir); 2098 if (jdir != NULL) 2099 vrele(jdir); 2100 2101 fddrop(fdp); 2102} 2103 2104/* 2105 * For setugid programs, we don't want to people to use that setugidness 2106 * to generate error messages which write to a file which otherwise would 2107 * otherwise be off-limits to the process. We check for filesystems where 2108 * the vnode can change out from under us after execve (like [lin]procfs). 2109 * 2110 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2111 * sufficient. We also don't check for setugidness since we know we are. 2112 */ 2113static bool 2114is_unsafe(struct file *fp) 2115{ 2116 struct vnode *vp; 2117 2118 if (fp->f_type != DTYPE_VNODE) 2119 return (false); 2120 2121 vp = fp->f_vnode; 2122 return ((vp->v_vflag & VV_PROCDEP) != 0); 2123} 2124 2125/* 2126 * Make this setguid thing safe, if at all possible. 2127 */ 2128void 2129fdsetugidsafety(struct thread *td) 2130{ 2131 struct filedesc *fdp; 2132 struct file *fp; 2133 int i; 2134 2135 fdp = td->td_proc->p_fd; 2136 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2137 MPASS(fdp->fd_nfiles >= 3); 2138 for (i = 0; i <= 2; i++) { 2139 fp = fdp->fd_ofiles[i].fde_file; 2140 if (fp != NULL && is_unsafe(fp)) { 2141 FILEDESC_XLOCK(fdp); 2142 knote_fdclose(td, i); 2143 /* 2144 * NULL-out descriptor prior to close to avoid 2145 * a race while close blocks. 2146 */ 2147 fdfree(fdp, i); 2148 FILEDESC_XUNLOCK(fdp); 2149 (void) closef(fp, td); 2150 } 2151 } 2152} 2153 2154/* 2155 * If a specific file object occupies a specific file descriptor, close the 2156 * file descriptor entry and drop a reference on the file object. This is a 2157 * convenience function to handle a subsequent error in a function that calls 2158 * falloc() that handles the race that another thread might have closed the 2159 * file descriptor out from under the thread creating the file object. 2160 */ 2161void 2162fdclose(struct thread *td, struct file *fp, int idx) 2163{ 2164 struct filedesc *fdp = td->td_proc->p_fd; 2165 2166 FILEDESC_XLOCK(fdp); 2167 if (fdp->fd_ofiles[idx].fde_file == fp) { 2168 fdfree(fdp, idx); 2169 FILEDESC_XUNLOCK(fdp); 2170 fdrop(fp, td); 2171 } else 2172 FILEDESC_XUNLOCK(fdp); 2173} 2174 2175/* 2176 * Close any files on exec? 2177 */ 2178void 2179fdcloseexec(struct thread *td) 2180{ 2181 struct filedesc *fdp; 2182 struct filedescent *fde; 2183 struct file *fp; 2184 int i; 2185 2186 fdp = td->td_proc->p_fd; 2187 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2188 for (i = 0; i <= fdp->fd_lastfile; i++) { 2189 fde = &fdp->fd_ofiles[i]; 2190 fp = fde->fde_file; 2191 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2192 (fde->fde_flags & UF_EXCLOSE))) { 2193 FILEDESC_XLOCK(fdp); 2194 fdfree(fdp, i); 2195 (void) closefp(fdp, i, fp, td, 0); 2196 /* closefp() drops the FILEDESC lock. */ 2197 } 2198 } 2199} 2200 2201/* 2202 * It is unsafe for set[ug]id processes to be started with file 2203 * descriptors 0..2 closed, as these descriptors are given implicit 2204 * significance in the Standard C library. fdcheckstd() will create a 2205 * descriptor referencing /dev/null for each of stdin, stdout, and 2206 * stderr that is not already open. 2207 */ 2208int 2209fdcheckstd(struct thread *td) 2210{ 2211 struct filedesc *fdp; 2212 register_t save; 2213 int i, error, devnull; 2214 2215 fdp = td->td_proc->p_fd; 2216 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2217 MPASS(fdp->fd_nfiles >= 3); 2218 devnull = -1; 2219 for (i = 0; i <= 2; i++) { 2220 if (fdp->fd_ofiles[i].fde_file != NULL) 2221 continue; 2222 2223 save = td->td_retval[0]; 2224 if (devnull != -1) { 2225 error = do_dup(td, DUP_FIXED, devnull, i); 2226 } else { 2227 error = kern_openat(td, AT_FDCWD, "/dev/null", 2228 UIO_SYSSPACE, O_RDWR, 0); 2229 if (error == 0) { 2230 devnull = td->td_retval[0]; 2231 KASSERT(devnull == i, ("we didn't get our fd")); 2232 } 2233 } 2234 td->td_retval[0] = save; 2235 if (error != 0) 2236 return (error); 2237 } 2238 return (0); 2239} 2240 2241/* 2242 * Internal form of close. Decrement reference count on file structure. 2243 * Note: td may be NULL when closing a file that was being passed in a 2244 * message. 2245 * 2246 * XXXRW: Giant is not required for the caller, but often will be held; this 2247 * makes it moderately likely the Giant will be recursed in the VFS case. 2248 */ 2249int 2250closef(struct file *fp, struct thread *td) 2251{ 2252 struct vnode *vp; 2253 struct flock lf; 2254 struct filedesc_to_leader *fdtol; 2255 struct filedesc *fdp; 2256 2257 /* 2258 * POSIX record locking dictates that any close releases ALL 2259 * locks owned by this process. This is handled by setting 2260 * a flag in the unlock to free ONLY locks obeying POSIX 2261 * semantics, and not to free BSD-style file locks. 2262 * If the descriptor was in a message, POSIX-style locks 2263 * aren't passed with the descriptor, and the thread pointer 2264 * will be NULL. Callers should be careful only to pass a 2265 * NULL thread pointer when there really is no owning 2266 * context that might have locks, or the locks will be 2267 * leaked. 2268 */ 2269 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2270 vp = fp->f_vnode; 2271 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2272 lf.l_whence = SEEK_SET; 2273 lf.l_start = 0; 2274 lf.l_len = 0; 2275 lf.l_type = F_UNLCK; 2276 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2277 F_UNLCK, &lf, F_POSIX); 2278 } 2279 fdtol = td->td_proc->p_fdtol; 2280 if (fdtol != NULL) { 2281 /* 2282 * Handle special case where file descriptor table is 2283 * shared between multiple process leaders. 2284 */ 2285 fdp = td->td_proc->p_fd; 2286 FILEDESC_XLOCK(fdp); 2287 for (fdtol = fdtol->fdl_next; 2288 fdtol != td->td_proc->p_fdtol; 2289 fdtol = fdtol->fdl_next) { 2290 if ((fdtol->fdl_leader->p_flag & 2291 P_ADVLOCK) == 0) 2292 continue; 2293 fdtol->fdl_holdcount++; 2294 FILEDESC_XUNLOCK(fdp); 2295 lf.l_whence = SEEK_SET; 2296 lf.l_start = 0; 2297 lf.l_len = 0; 2298 lf.l_type = F_UNLCK; 2299 vp = fp->f_vnode; 2300 (void) VOP_ADVLOCK(vp, 2301 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2302 F_POSIX); 2303 FILEDESC_XLOCK(fdp); 2304 fdtol->fdl_holdcount--; 2305 if (fdtol->fdl_holdcount == 0 && 2306 fdtol->fdl_wakeup != 0) { 2307 fdtol->fdl_wakeup = 0; 2308 wakeup(fdtol); 2309 } 2310 } 2311 FILEDESC_XUNLOCK(fdp); 2312 } 2313 } 2314 return (fdrop(fp, td)); 2315} 2316 2317/* 2318 * Initialize the file pointer with the specified properties. 2319 * 2320 * The ops are set with release semantics to be certain that the flags, type, 2321 * and data are visible when ops is. This is to prevent ops methods from being 2322 * called with bad data. 2323 */ 2324void 2325finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2326{ 2327 fp->f_data = data; 2328 fp->f_flag = flag; 2329 fp->f_type = type; 2330 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2331} 2332 2333int 2334fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2335 struct file **fpp, seq_t *seqp) 2336{ 2337#ifdef CAPABILITIES 2338 struct filedescent *fde; 2339#endif 2340 struct fdescenttbl *fdt; 2341 struct file *fp; 2342 u_int count; 2343#ifdef CAPABILITIES 2344 seq_t seq; 2345 cap_rights_t haverights; 2346 int error; 2347#endif 2348 2349 fdt = fdp->fd_files; 2350 if ((u_int)fd >= fdt->fdt_nfiles) 2351 return (EBADF); 2352 /* 2353 * Fetch the descriptor locklessly. We avoid fdrop() races by 2354 * never raising a refcount above 0. To accomplish this we have 2355 * to use a cmpset loop rather than an atomic_add. The descriptor 2356 * must be re-verified once we acquire a reference to be certain 2357 * that the identity is still correct and we did not lose a race 2358 * due to preemption. 2359 */ 2360 for (;;) { 2361#ifdef CAPABILITIES 2362 seq = seq_read(fd_seq(fdt, fd)); 2363 fde = &fdt->fdt_ofiles[fd]; 2364 haverights = *cap_rights_fde(fde); 2365 fp = fde->fde_file; 2366 if (!seq_consistent(fd_seq(fdt, fd), seq)) { 2367 cpu_spinwait(); 2368 continue; 2369 } 2370#else 2371 fp = fdt->fdt_ofiles[fd].fde_file; 2372#endif 2373 if (fp == NULL) 2374 return (EBADF); 2375#ifdef CAPABILITIES 2376 if (needrightsp != NULL) { 2377 error = cap_check(&haverights, needrightsp); 2378 if (error != 0) 2379 return (error); 2380 } 2381#endif 2382 retry: 2383 count = fp->f_count; 2384 if (count == 0) { 2385 /* 2386 * Force a reload. Other thread could reallocate the 2387 * table before this fd was closed, so it possible that 2388 * there is a stale fp pointer in cached version. 2389 */ 2390 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); 2391 continue; 2392 } 2393 /* 2394 * Use an acquire barrier to force re-reading of fdt so it is 2395 * refreshed for verification. 2396 */ 2397 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) 2398 goto retry; 2399 fdt = fdp->fd_files; 2400#ifdef CAPABILITIES 2401 if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) 2402#else 2403 if (fp == fdt->fdt_ofiles[fd].fde_file) 2404#endif 2405 break; 2406 fdrop(fp, curthread); 2407 } 2408 *fpp = fp; 2409 if (seqp != NULL) { 2410#ifdef CAPABILITIES 2411 *seqp = seq; 2412#endif 2413 } 2414 return (0); 2415} 2416 2417/* 2418 * Extract the file pointer associated with the specified descriptor for the 2419 * current user process. 2420 * 2421 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2422 * returned. 2423 * 2424 * File's rights will be checked against the capability rights mask. 2425 * 2426 * If an error occured the non-zero error is returned and *fpp is set to 2427 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2428 * responsible for fdrop(). 2429 */ 2430static __inline int 2431_fget(struct thread *td, int fd, struct file **fpp, int flags, 2432 cap_rights_t *needrightsp, seq_t *seqp) 2433{ 2434 struct filedesc *fdp; 2435 struct file *fp; 2436 cap_rights_t needrights; 2437 int error; 2438 2439 *fpp = NULL; 2440 fdp = td->td_proc->p_fd; 2441 if (needrightsp != NULL) 2442 needrights = *needrightsp; 2443 else 2444 cap_rights_init(&needrights); 2445 error = fget_unlocked(fdp, fd, &needrights, &fp, seqp); 2446 if (error != 0) 2447 return (error); 2448 if (fp->f_ops == &badfileops) { 2449 fdrop(fp, td); 2450 return (EBADF); 2451 } 2452 2453 /* 2454 * FREAD and FWRITE failure return EBADF as per POSIX. 2455 */ 2456 error = 0; 2457 switch (flags) { 2458 case FREAD: 2459 case FWRITE: 2460 if ((fp->f_flag & flags) == 0) 2461 error = EBADF; 2462 break; 2463 case FEXEC: 2464 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2465 ((fp->f_flag & FWRITE) != 0)) 2466 error = EBADF; 2467 break; 2468 case 0: 2469 break; 2470 default: 2471 KASSERT(0, ("wrong flags")); 2472 } 2473 2474 if (error != 0) { 2475 fdrop(fp, td); 2476 return (error); 2477 } 2478 2479 *fpp = fp; 2480 return (0); 2481} 2482 2483int 2484fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2485{ 2486 2487 return (_fget(td, fd, fpp, 0, rightsp, NULL)); 2488} 2489 2490int 2491fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2492 struct file **fpp) 2493{ 2494 int error; 2495#ifndef CAPABILITIES 2496 error = _fget(td, fd, fpp, 0, rightsp, NULL); 2497 if (maxprotp != NULL) 2498 *maxprotp = VM_PROT_ALL; 2499#else 2500 struct filedesc *fdp = td->td_proc->p_fd; 2501 seq_t seq; 2502 2503 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 2504 for (;;) { 2505 error = _fget(td, fd, fpp, 0, rightsp, &seq); 2506 if (error != 0) 2507 return (error); 2508 /* 2509 * If requested, convert capability rights to access flags. 2510 */ 2511 if (maxprotp != NULL) 2512 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); 2513 if (!fd_modified(fdp, fd, seq)) 2514 break; 2515 fdrop(*fpp, td); 2516 } 2517#endif 2518 return (error); 2519} 2520 2521int 2522fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2523{ 2524 2525 return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2526} 2527 2528int 2529fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2530{ 2531 2532 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2533} 2534 2535int 2536fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 2537 struct file **fpp) 2538{ 2539 struct filedesc *fdp = td->td_proc->p_fd; 2540#ifndef CAPABILITIES 2541 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); 2542#else 2543 int error; 2544 seq_t seq; 2545 2546 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 2547 for (;;) { 2548 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); 2549 if (error != 0) 2550 return (error); 2551 error = cap_fcntl_check(fdp, fd, needfcntl); 2552 if (!fd_modified(fdp, fd, seq)) 2553 break; 2554 fdrop(*fpp, td); 2555 } 2556 if (error != 0) { 2557 fdrop(*fpp, td); 2558 *fpp = NULL; 2559 } 2560 return (error); 2561#endif 2562} 2563 2564/* 2565 * Like fget() but loads the underlying vnode, or returns an error if the 2566 * descriptor does not represent a vnode. Note that pipes use vnodes but 2567 * never have VM objects. The returned vnode will be vref()'d. 2568 * 2569 * XXX: what about the unused flags ? 2570 */ 2571static __inline int 2572_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2573 struct vnode **vpp) 2574{ 2575 struct file *fp; 2576 int error; 2577 2578 *vpp = NULL; 2579 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2580 if (error != 0) 2581 return (error); 2582 if (fp->f_vnode == NULL) { 2583 error = EINVAL; 2584 } else { 2585 *vpp = fp->f_vnode; 2586 vref(*vpp); 2587 } 2588 fdrop(fp, td); 2589 2590 return (error); 2591} 2592 2593int 2594fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2595{ 2596 2597 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2598} 2599 2600int 2601fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2602 struct filecaps *havecaps, struct vnode **vpp) 2603{ 2604 struct filedesc *fdp; 2605 struct file *fp; 2606#ifdef CAPABILITIES 2607 int error; 2608#endif 2609 2610 fdp = td->td_proc->p_fd; 2611 fp = fget_locked(fdp, fd); 2612 if (fp == NULL || fp->f_ops == &badfileops) 2613 return (EBADF); 2614 2615#ifdef CAPABILITIES 2616 if (needrightsp != NULL) { 2617 error = cap_check(cap_rights(fdp, fd), needrightsp); 2618 if (error != 0) 2619 return (error); 2620 } 2621#endif 2622 2623 if (fp->f_vnode == NULL) 2624 return (EINVAL); 2625 2626 *vpp = fp->f_vnode; 2627 vref(*vpp); 2628 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2629 2630 return (0); 2631} 2632 2633int 2634fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2635{ 2636 2637 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2638} 2639 2640int 2641fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2642{ 2643 2644 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2645} 2646 2647#ifdef notyet 2648int 2649fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2650 struct vnode **vpp) 2651{ 2652 2653 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2654} 2655#endif 2656 2657/* 2658 * Like fget() but loads the underlying socket, or returns an error if the 2659 * descriptor does not represent a socket. 2660 * 2661 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2662 * in the future. 2663 * 2664 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2665 * on their file descriptor reference to prevent the socket from being free'd 2666 * during use. 2667 */ 2668int 2669fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2670 u_int *fflagp) 2671{ 2672 struct file *fp; 2673 int error; 2674 2675 *spp = NULL; 2676 if (fflagp != NULL) 2677 *fflagp = 0; 2678 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2679 return (error); 2680 if (fp->f_type != DTYPE_SOCKET) { 2681 error = ENOTSOCK; 2682 } else { 2683 *spp = fp->f_data; 2684 if (fflagp) 2685 *fflagp = fp->f_flag; 2686 SOCK_LOCK(*spp); 2687 soref(*spp); 2688 SOCK_UNLOCK(*spp); 2689 } 2690 fdrop(fp, td); 2691 2692 return (error); 2693} 2694 2695/* 2696 * Drop the reference count on the socket and XXX release the SX lock in the 2697 * future. The last reference closes the socket. 2698 * 2699 * Note: fputsock() is deprecated, see comment for fgetsock(). 2700 */ 2701void 2702fputsock(struct socket *so) 2703{ 2704 2705 ACCEPT_LOCK(); 2706 SOCK_LOCK(so); 2707 CURVNET_SET(so->so_vnet); 2708 sorele(so); 2709 CURVNET_RESTORE(); 2710} 2711 2712/* 2713 * Handle the last reference to a file being closed. 2714 */ 2715int 2716_fdrop(struct file *fp, struct thread *td) 2717{ 2718 int error; 2719 2720 if (fp->f_count != 0) 2721 panic("fdrop: count %d", fp->f_count); 2722 error = fo_close(fp, td); 2723 atomic_subtract_int(&openfiles, 1); 2724 crfree(fp->f_cred); 2725 free(fp->f_advice, M_FADVISE); 2726 uma_zfree(file_zone, fp); 2727 2728 return (error); 2729} 2730 2731/* 2732 * Apply an advisory lock on a file descriptor. 2733 * 2734 * Just attempt to get a record lock of the requested type on the entire file 2735 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2736 */ 2737#ifndef _SYS_SYSPROTO_H_ 2738struct flock_args { 2739 int fd; 2740 int how; 2741}; 2742#endif 2743/* ARGSUSED */ 2744int 2745sys_flock(struct thread *td, struct flock_args *uap) 2746{ 2747 struct file *fp; 2748 struct vnode *vp; 2749 struct flock lf; 2750 cap_rights_t rights; 2751 int error; 2752 2753 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2754 if (error != 0) 2755 return (error); 2756 if (fp->f_type != DTYPE_VNODE) { 2757 fdrop(fp, td); 2758 return (EOPNOTSUPP); 2759 } 2760 2761 vp = fp->f_vnode; 2762 lf.l_whence = SEEK_SET; 2763 lf.l_start = 0; 2764 lf.l_len = 0; 2765 if (uap->how & LOCK_UN) { 2766 lf.l_type = F_UNLCK; 2767 atomic_clear_int(&fp->f_flag, FHASLOCK); 2768 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2769 goto done2; 2770 } 2771 if (uap->how & LOCK_EX) 2772 lf.l_type = F_WRLCK; 2773 else if (uap->how & LOCK_SH) 2774 lf.l_type = F_RDLCK; 2775 else { 2776 error = EBADF; 2777 goto done2; 2778 } 2779 atomic_set_int(&fp->f_flag, FHASLOCK); 2780 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2781 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2782done2: 2783 fdrop(fp, td); 2784 return (error); 2785} 2786/* 2787 * Duplicate the specified descriptor to a free descriptor. 2788 */ 2789int 2790dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2791 int openerror, int *indxp) 2792{ 2793 struct filedescent *newfde, *oldfde; 2794 struct file *fp; 2795 int error, indx; 2796 2797 KASSERT(openerror == ENODEV || openerror == ENXIO, 2798 ("unexpected error %d in %s", openerror, __func__)); 2799 2800 /* 2801 * If the to-be-dup'd fd number is greater than the allowed number 2802 * of file descriptors, or the fd to be dup'd has already been 2803 * closed, then reject. 2804 */ 2805 FILEDESC_XLOCK(fdp); 2806 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2807 FILEDESC_XUNLOCK(fdp); 2808 return (EBADF); 2809 } 2810 2811 error = fdalloc(td, 0, &indx); 2812 if (error != 0) { 2813 FILEDESC_XUNLOCK(fdp); 2814 return (error); 2815 } 2816 2817 /* 2818 * There are two cases of interest here. 2819 * 2820 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2821 * 2822 * For ENXIO steal away the file structure from (dfd) and store it in 2823 * (indx). (dfd) is effectively closed by this operation. 2824 */ 2825 switch (openerror) { 2826 case ENODEV: 2827 /* 2828 * Check that the mode the file is being opened for is a 2829 * subset of the mode of the existing descriptor. 2830 */ 2831 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2832 fdunused(fdp, indx); 2833 FILEDESC_XUNLOCK(fdp); 2834 return (EACCES); 2835 } 2836 fhold(fp); 2837 newfde = &fdp->fd_ofiles[indx]; 2838 oldfde = &fdp->fd_ofiles[dfd]; 2839#ifdef CAPABILITIES 2840 seq_write_begin(&newfde->fde_seq); 2841#endif 2842 memcpy(newfde, oldfde, fde_change_size); 2843 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 2844#ifdef CAPABILITIES 2845 seq_write_end(&newfde->fde_seq); 2846#endif 2847 break; 2848 case ENXIO: 2849 /* 2850 * Steal away the file pointer from dfd and stuff it into indx. 2851 */ 2852 newfde = &fdp->fd_ofiles[indx]; 2853 oldfde = &fdp->fd_ofiles[dfd]; 2854#ifdef CAPABILITIES 2855 seq_write_begin(&newfde->fde_seq); 2856#endif 2857 memcpy(newfde, oldfde, fde_change_size); 2858 bzero(oldfde, fde_change_size); 2859 fdunused(fdp, dfd); 2860#ifdef CAPABILITIES 2861 seq_write_end(&newfde->fde_seq); 2862#endif 2863 break; 2864 } 2865 FILEDESC_XUNLOCK(fdp); 2866 *indxp = indx; 2867 return (0); 2868} 2869 2870/* 2871 * Scan all active processes and prisons to see if any of them have a current 2872 * or root directory of `olddp'. If so, replace them with the new mount point. 2873 */ 2874void 2875mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2876{ 2877 struct filedesc *fdp; 2878 struct prison *pr; 2879 struct proc *p; 2880 int nrele; 2881 2882 if (vrefcnt(olddp) == 1) 2883 return; 2884 nrele = 0; 2885 sx_slock(&allproc_lock); 2886 FOREACH_PROC_IN_SYSTEM(p) { 2887 fdp = fdhold(p); 2888 if (fdp == NULL) 2889 continue; 2890 FILEDESC_XLOCK(fdp); 2891 if (fdp->fd_cdir == olddp) { 2892 vref(newdp); 2893 fdp->fd_cdir = newdp; 2894 nrele++; 2895 } 2896 if (fdp->fd_rdir == olddp) { 2897 vref(newdp); 2898 fdp->fd_rdir = newdp; 2899 nrele++; 2900 } 2901 if (fdp->fd_jdir == olddp) { 2902 vref(newdp); 2903 fdp->fd_jdir = newdp; 2904 nrele++; 2905 } 2906 FILEDESC_XUNLOCK(fdp); 2907 fddrop(fdp); 2908 } 2909 sx_sunlock(&allproc_lock); 2910 if (rootvnode == olddp) { 2911 vref(newdp); 2912 rootvnode = newdp; 2913 nrele++; 2914 } 2915 mtx_lock(&prison0.pr_mtx); 2916 if (prison0.pr_root == olddp) { 2917 vref(newdp); 2918 prison0.pr_root = newdp; 2919 nrele++; 2920 } 2921 mtx_unlock(&prison0.pr_mtx); 2922 sx_slock(&allprison_lock); 2923 TAILQ_FOREACH(pr, &allprison, pr_list) { 2924 mtx_lock(&pr->pr_mtx); 2925 if (pr->pr_root == olddp) { 2926 vref(newdp); 2927 pr->pr_root = newdp; 2928 nrele++; 2929 } 2930 mtx_unlock(&pr->pr_mtx); 2931 } 2932 sx_sunlock(&allprison_lock); 2933 while (nrele--) 2934 vrele(olddp); 2935} 2936 2937struct filedesc_to_leader * 2938filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2939{ 2940 struct filedesc_to_leader *fdtol; 2941 2942 fdtol = malloc(sizeof(struct filedesc_to_leader), 2943 M_FILEDESC_TO_LEADER, 2944 M_WAITOK); 2945 fdtol->fdl_refcount = 1; 2946 fdtol->fdl_holdcount = 0; 2947 fdtol->fdl_wakeup = 0; 2948 fdtol->fdl_leader = leader; 2949 if (old != NULL) { 2950 FILEDESC_XLOCK(fdp); 2951 fdtol->fdl_next = old->fdl_next; 2952 fdtol->fdl_prev = old; 2953 old->fdl_next = fdtol; 2954 fdtol->fdl_next->fdl_prev = fdtol; 2955 FILEDESC_XUNLOCK(fdp); 2956 } else { 2957 fdtol->fdl_next = fdtol; 2958 fdtol->fdl_prev = fdtol; 2959 } 2960 return (fdtol); 2961} 2962 2963/* 2964 * Get file structures globally. 2965 */ 2966static int 2967sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2968{ 2969 struct xfile xf; 2970 struct filedesc *fdp; 2971 struct file *fp; 2972 struct proc *p; 2973 int error, n; 2974 2975 error = sysctl_wire_old_buffer(req, 0); 2976 if (error != 0) 2977 return (error); 2978 if (req->oldptr == NULL) { 2979 n = 0; 2980 sx_slock(&allproc_lock); 2981 FOREACH_PROC_IN_SYSTEM(p) { 2982 if (p->p_state == PRS_NEW) 2983 continue; 2984 fdp = fdhold(p); 2985 if (fdp == NULL) 2986 continue; 2987 /* overestimates sparse tables. */ 2988 if (fdp->fd_lastfile > 0) 2989 n += fdp->fd_lastfile; 2990 fddrop(fdp); 2991 } 2992 sx_sunlock(&allproc_lock); 2993 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2994 } 2995 error = 0; 2996 bzero(&xf, sizeof(xf)); 2997 xf.xf_size = sizeof(xf); 2998 sx_slock(&allproc_lock); 2999 FOREACH_PROC_IN_SYSTEM(p) { 3000 PROC_LOCK(p); 3001 if (p->p_state == PRS_NEW) { 3002 PROC_UNLOCK(p); 3003 continue; 3004 } 3005 if (p_cansee(req->td, p) != 0) { 3006 PROC_UNLOCK(p); 3007 continue; 3008 } 3009 xf.xf_pid = p->p_pid; 3010 xf.xf_uid = p->p_ucred->cr_uid; 3011 PROC_UNLOCK(p); 3012 fdp = fdhold(p); 3013 if (fdp == NULL) 3014 continue; 3015 FILEDESC_SLOCK(fdp); 3016 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { 3017 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3018 continue; 3019 xf.xf_fd = n; 3020 xf.xf_file = fp; 3021 xf.xf_data = fp->f_data; 3022 xf.xf_vnode = fp->f_vnode; 3023 xf.xf_type = fp->f_type; 3024 xf.xf_count = fp->f_count; 3025 xf.xf_msgcount = 0; 3026 xf.xf_offset = foffset_get(fp); 3027 xf.xf_flag = fp->f_flag; 3028 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 3029 if (error) 3030 break; 3031 } 3032 FILEDESC_SUNLOCK(fdp); 3033 fddrop(fdp); 3034 if (error) 3035 break; 3036 } 3037 sx_sunlock(&allproc_lock); 3038 return (error); 3039} 3040 3041SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 3042 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 3043 3044#ifdef KINFO_FILE_SIZE 3045CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3046#endif 3047 3048static int 3049xlate_fflags(int fflags) 3050{ 3051 static const struct { 3052 int fflag; 3053 int kf_fflag; 3054 } fflags_table[] = { 3055 { FAPPEND, KF_FLAG_APPEND }, 3056 { FASYNC, KF_FLAG_ASYNC }, 3057 { FFSYNC, KF_FLAG_FSYNC }, 3058 { FHASLOCK, KF_FLAG_HASLOCK }, 3059 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3060 { FREAD, KF_FLAG_READ }, 3061 { FWRITE, KF_FLAG_WRITE }, 3062 { O_CREAT, KF_FLAG_CREAT }, 3063 { O_DIRECT, KF_FLAG_DIRECT }, 3064 { O_EXCL, KF_FLAG_EXCL }, 3065 { O_EXEC, KF_FLAG_EXEC }, 3066 { O_EXLOCK, KF_FLAG_EXLOCK }, 3067 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3068 { O_SHLOCK, KF_FLAG_SHLOCK }, 3069 { O_TRUNC, KF_FLAG_TRUNC } 3070 }; 3071 unsigned int i; 3072 int kflags; 3073 3074 kflags = 0; 3075 for (i = 0; i < nitems(fflags_table); i++) 3076 if (fflags & fflags_table[i].fflag) 3077 kflags |= fflags_table[i].kf_fflag; 3078 return (kflags); 3079} 3080 3081/* Trim unused data from kf_path by truncating the structure size. */ 3082static void 3083pack_kinfo(struct kinfo_file *kif) 3084{ 3085 3086 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3087 strlen(kif->kf_path) + 1; 3088 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3089} 3090 3091static void 3092export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 3093 struct kinfo_file *kif, struct filedesc *fdp) 3094{ 3095 int error; 3096 3097 bzero(kif, sizeof(*kif)); 3098 3099 /* Set a default type to allow for empty fill_kinfo() methods. */ 3100 kif->kf_type = KF_TYPE_UNKNOWN; 3101 kif->kf_flags = xlate_fflags(fp->f_flag); 3102 if (rightsp != NULL) 3103 kif->kf_cap_rights = *rightsp; 3104 else 3105 cap_rights_init(&kif->kf_cap_rights); 3106 kif->kf_fd = fd; 3107 kif->kf_ref_count = fp->f_count; 3108 kif->kf_offset = foffset_get(fp); 3109 3110 /* 3111 * This may drop the filedesc lock, so the 'fp' cannot be 3112 * accessed after this call. 3113 */ 3114 error = fo_fill_kinfo(fp, kif, fdp); 3115 if (error == 0) 3116 kif->kf_status |= KF_ATTR_VALID; 3117 pack_kinfo(kif); 3118} 3119 3120static void 3121export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 3122 struct kinfo_file *kif) 3123{ 3124 int error; 3125 3126 bzero(kif, sizeof(*kif)); 3127 3128 kif->kf_type = KF_TYPE_VNODE; 3129 error = vn_fill_kinfo_vnode(vp, kif); 3130 if (error == 0) 3131 kif->kf_status |= KF_ATTR_VALID; 3132 kif->kf_flags = xlate_fflags(fflags); 3133 cap_rights_init(&kif->kf_cap_rights); 3134 kif->kf_fd = fd; 3135 kif->kf_ref_count = -1; 3136 kif->kf_offset = -1; 3137 pack_kinfo(kif); 3138 vrele(vp); 3139} 3140 3141struct export_fd_buf { 3142 struct filedesc *fdp; 3143 struct sbuf *sb; 3144 ssize_t remainder; 3145 struct kinfo_file kif; 3146}; 3147 3148static int 3149export_kinfo_to_sb(struct export_fd_buf *efbuf) 3150{ 3151 struct kinfo_file *kif; 3152 3153 kif = &efbuf->kif; 3154 if (efbuf->remainder != -1) { 3155 if (efbuf->remainder < kif->kf_structsize) { 3156 /* Terminate export. */ 3157 efbuf->remainder = 0; 3158 return (0); 3159 } 3160 efbuf->remainder -= kif->kf_structsize; 3161 } 3162 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); 3163} 3164 3165static int 3166export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 3167 struct export_fd_buf *efbuf) 3168{ 3169 int error; 3170 3171 if (efbuf->remainder == 0) 3172 return (0); 3173 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp); 3174 FILEDESC_SUNLOCK(efbuf->fdp); 3175 error = export_kinfo_to_sb(efbuf); 3176 FILEDESC_SLOCK(efbuf->fdp); 3177 return (error); 3178} 3179 3180static int 3181export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 3182 struct export_fd_buf *efbuf) 3183{ 3184 int error; 3185 3186 if (efbuf->remainder == 0) 3187 return (0); 3188 if (efbuf->fdp != NULL) 3189 FILEDESC_SUNLOCK(efbuf->fdp); 3190 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif); 3191 error = export_kinfo_to_sb(efbuf); 3192 if (efbuf->fdp != NULL) 3193 FILEDESC_SLOCK(efbuf->fdp); 3194 return (error); 3195} 3196 3197/* 3198 * Store a process file descriptor information to sbuf. 3199 * 3200 * Takes a locked proc as argument, and returns with the proc unlocked. 3201 */ 3202int 3203kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3204{ 3205 struct file *fp; 3206 struct filedesc *fdp; 3207 struct export_fd_buf *efbuf; 3208 struct vnode *cttyvp, *textvp, *tracevp; 3209 int error, i; 3210 cap_rights_t rights; 3211 3212 PROC_LOCK_ASSERT(p, MA_OWNED); 3213 3214 /* ktrace vnode */ 3215 tracevp = p->p_tracevp; 3216 if (tracevp != NULL) 3217 vref(tracevp); 3218 /* text vnode */ 3219 textvp = p->p_textvp; 3220 if (textvp != NULL) 3221 vref(textvp); 3222 /* Controlling tty. */ 3223 cttyvp = NULL; 3224 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3225 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3226 if (cttyvp != NULL) 3227 vref(cttyvp); 3228 } 3229 fdp = fdhold(p); 3230 PROC_UNLOCK(p); 3231 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3232 efbuf->fdp = NULL; 3233 efbuf->sb = sb; 3234 efbuf->remainder = maxlen; 3235 if (tracevp != NULL) 3236 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, 3237 efbuf); 3238 if (textvp != NULL) 3239 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); 3240 if (cttyvp != NULL) 3241 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, 3242 efbuf); 3243 error = 0; 3244 if (fdp == NULL) 3245 goto fail; 3246 efbuf->fdp = fdp; 3247 FILEDESC_SLOCK(fdp); 3248 /* working directory */ 3249 if (fdp->fd_cdir != NULL) { 3250 vref(fdp->fd_cdir); 3251 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 3252 } 3253 /* root directory */ 3254 if (fdp->fd_rdir != NULL) { 3255 vref(fdp->fd_rdir); 3256 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); 3257 } 3258 /* jail directory */ 3259 if (fdp->fd_jdir != NULL) { 3260 vref(fdp->fd_jdir); 3261 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); 3262 } 3263 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3264 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3265 continue; 3266#ifdef CAPABILITIES 3267 rights = *cap_rights(fdp, i); 3268#else /* !CAPABILITIES */ 3269 cap_rights_init(&rights); 3270#endif 3271 /* 3272 * Create sysctl entry. It is OK to drop the filedesc 3273 * lock inside of export_file_to_sb() as we will 3274 * re-validate and re-evaluate its properties when the 3275 * loop continues. 3276 */ 3277 error = export_file_to_sb(fp, i, &rights, efbuf); 3278 if (error != 0 || efbuf->remainder == 0) 3279 break; 3280 } 3281 FILEDESC_SUNLOCK(fdp); 3282 fddrop(fdp); 3283fail: 3284 free(efbuf, M_TEMP); 3285 return (error); 3286} 3287 3288#define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3289 3290/* 3291 * Get per-process file descriptors for use by procstat(1), et al. 3292 */ 3293static int 3294sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3295{ 3296 struct sbuf sb; 3297 struct proc *p; 3298 ssize_t maxlen; 3299 int error, error2, *name; 3300 3301 name = (int *)arg1; 3302 3303 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3304 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3305 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3306 if (error != 0) { 3307 sbuf_delete(&sb); 3308 return (error); 3309 } 3310 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3311 error = kern_proc_filedesc_out(p, &sb, maxlen); 3312 error2 = sbuf_finish(&sb); 3313 sbuf_delete(&sb); 3314 return (error != 0 ? error : error2); 3315} 3316 3317#ifdef KINFO_OFILE_SIZE 3318CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 3319#endif 3320 3321#ifdef COMPAT_FREEBSD7 3322static void 3323kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 3324{ 3325 3326 okif->kf_structsize = sizeof(*okif); 3327 okif->kf_type = kif->kf_type; 3328 okif->kf_fd = kif->kf_fd; 3329 okif->kf_ref_count = kif->kf_ref_count; 3330 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 3331 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 3332 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 3333 okif->kf_offset = kif->kf_offset; 3334 okif->kf_vnode_type = kif->kf_vnode_type; 3335 okif->kf_sock_domain = kif->kf_sock_domain; 3336 okif->kf_sock_type = kif->kf_sock_type; 3337 okif->kf_sock_protocol = kif->kf_sock_protocol; 3338 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 3339 okif->kf_sa_local = kif->kf_sa_local; 3340 okif->kf_sa_peer = kif->kf_sa_peer; 3341} 3342 3343static int 3344export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 3345 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) 3346{ 3347 int error; 3348 3349 vref(vp); 3350 FILEDESC_SUNLOCK(fdp); 3351 export_vnode_to_kinfo(vp, type, 0, kif); 3352 kinfo_to_okinfo(kif, okif); 3353 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3354 FILEDESC_SLOCK(fdp); 3355 return (error); 3356} 3357 3358/* 3359 * Get per-process file descriptors for use by procstat(1), et al. 3360 */ 3361static int 3362sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 3363{ 3364 struct kinfo_ofile *okif; 3365 struct kinfo_file *kif; 3366 struct filedesc *fdp; 3367 int error, i, *name; 3368 struct file *fp; 3369 struct proc *p; 3370 3371 name = (int *)arg1; 3372 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3373 if (error != 0) 3374 return (error); 3375 fdp = fdhold(p); 3376 PROC_UNLOCK(p); 3377 if (fdp == NULL) 3378 return (ENOENT); 3379 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3380 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 3381 FILEDESC_SLOCK(fdp); 3382 if (fdp->fd_cdir != NULL) 3383 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3384 okif, fdp, req); 3385 if (fdp->fd_rdir != NULL) 3386 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3387 okif, fdp, req); 3388 if (fdp->fd_jdir != NULL) 3389 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3390 okif, fdp, req); 3391 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { 3392 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3393 continue; 3394 export_file_to_kinfo(fp, i, NULL, kif, fdp); 3395 FILEDESC_SUNLOCK(fdp); 3396 kinfo_to_okinfo(kif, okif); 3397 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 3398 FILEDESC_SLOCK(fdp); 3399 if (error) 3400 break; 3401 } 3402 FILEDESC_SUNLOCK(fdp); 3403 fddrop(fdp); 3404 free(kif, M_TEMP); 3405 free(okif, M_TEMP); 3406 return (0); 3407} 3408 3409static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3410 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3411 "Process ofiledesc entries"); 3412#endif /* COMPAT_FREEBSD7 */ 3413 3414int 3415vntype_to_kinfo(int vtype) 3416{ 3417 struct { 3418 int vtype; 3419 int kf_vtype; 3420 } vtypes_table[] = { 3421 { VBAD, KF_VTYPE_VBAD }, 3422 { VBLK, KF_VTYPE_VBLK }, 3423 { VCHR, KF_VTYPE_VCHR }, 3424 { VDIR, KF_VTYPE_VDIR }, 3425 { VFIFO, KF_VTYPE_VFIFO }, 3426 { VLNK, KF_VTYPE_VLNK }, 3427 { VNON, KF_VTYPE_VNON }, 3428 { VREG, KF_VTYPE_VREG }, 3429 { VSOCK, KF_VTYPE_VSOCK } 3430 }; 3431 unsigned int i; 3432 3433 /* 3434 * Perform vtype translation. 3435 */ 3436 for (i = 0; i < nitems(vtypes_table); i++) 3437 if (vtypes_table[i].vtype == vtype) 3438 return (vtypes_table[i].kf_vtype); 3439 3440 return (KF_VTYPE_UNKNOWN); 3441} 3442 3443static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3444 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3445 "Process filedesc entries"); 3446 3447/* 3448 * Store a process current working directory information to sbuf. 3449 * 3450 * Takes a locked proc as argument, and returns with the proc unlocked. 3451 */ 3452int 3453kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3454{ 3455 struct filedesc *fdp; 3456 struct export_fd_buf *efbuf; 3457 int error; 3458 3459 PROC_LOCK_ASSERT(p, MA_OWNED); 3460 3461 fdp = fdhold(p); 3462 PROC_UNLOCK(p); 3463 if (fdp == NULL) 3464 return (EINVAL); 3465 3466 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3467 efbuf->fdp = fdp; 3468 efbuf->sb = sb; 3469 efbuf->remainder = maxlen; 3470 3471 FILEDESC_SLOCK(fdp); 3472 if (fdp->fd_cdir == NULL) 3473 error = EINVAL; 3474 else { 3475 vref(fdp->fd_cdir); 3476 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, 3477 FREAD, efbuf); 3478 } 3479 FILEDESC_SUNLOCK(fdp); 3480 fddrop(fdp); 3481 free(efbuf, M_TEMP); 3482 return (error); 3483} 3484 3485/* 3486 * Get per-process current working directory. 3487 */ 3488static int 3489sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 3490{ 3491 struct sbuf sb; 3492 struct proc *p; 3493 ssize_t maxlen; 3494 int error, error2, *name; 3495 3496 name = (int *)arg1; 3497 3498 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 3499 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 3500 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 3501 if (error != 0) { 3502 sbuf_delete(&sb); 3503 return (error); 3504 } 3505 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3506 error = kern_proc_cwd_out(p, &sb, maxlen); 3507 error2 = sbuf_finish(&sb); 3508 sbuf_delete(&sb); 3509 return (error != 0 ? error : error2); 3510} 3511 3512static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 3513 sysctl_kern_proc_cwd, "Process current working directory"); 3514 3515#ifdef DDB 3516/* 3517 * For the purposes of debugging, generate a human-readable string for the 3518 * file type. 3519 */ 3520static const char * 3521file_type_to_name(short type) 3522{ 3523 3524 switch (type) { 3525 case 0: 3526 return ("zero"); 3527 case DTYPE_VNODE: 3528 return ("vnod"); 3529 case DTYPE_SOCKET: 3530 return ("sock"); 3531 case DTYPE_PIPE: 3532 return ("pipe"); 3533 case DTYPE_FIFO: 3534 return ("fifo"); 3535 case DTYPE_KQUEUE: 3536 return ("kque"); 3537 case DTYPE_CRYPTO: 3538 return ("crpt"); 3539 case DTYPE_MQUEUE: 3540 return ("mque"); 3541 case DTYPE_SHM: 3542 return ("shm"); 3543 case DTYPE_SEM: 3544 return ("ksem"); 3545 default: 3546 return ("unkn"); 3547 } 3548} 3549 3550/* 3551 * For the purposes of debugging, identify a process (if any, perhaps one of 3552 * many) that references the passed file in its file descriptor array. Return 3553 * NULL if none. 3554 */ 3555static struct proc * 3556file_to_first_proc(struct file *fp) 3557{ 3558 struct filedesc *fdp; 3559 struct proc *p; 3560 int n; 3561 3562 FOREACH_PROC_IN_SYSTEM(p) { 3563 if (p->p_state == PRS_NEW) 3564 continue; 3565 fdp = p->p_fd; 3566 if (fdp == NULL) 3567 continue; 3568 for (n = 0; n <= fdp->fd_lastfile; n++) { 3569 if (fp == fdp->fd_ofiles[n].fde_file) 3570 return (p); 3571 } 3572 } 3573 return (NULL); 3574} 3575 3576static void 3577db_print_file(struct file *fp, int header) 3578{ 3579 struct proc *p; 3580 3581 if (header) 3582 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3583 "File", "Type", "Data", "Flag", "GCFl", "Count", 3584 "MCount", "Vnode", "FPID", "FCmd"); 3585 p = file_to_first_proc(fp); 3586 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3587 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3588 0, fp->f_count, 0, fp->f_vnode, 3589 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3590} 3591 3592DB_SHOW_COMMAND(file, db_show_file) 3593{ 3594 struct file *fp; 3595 3596 if (!have_addr) { 3597 db_printf("usage: show file <addr>\n"); 3598 return; 3599 } 3600 fp = (struct file *)addr; 3601 db_print_file(fp, 1); 3602} 3603 3604DB_SHOW_COMMAND(files, db_show_files) 3605{ 3606 struct filedesc *fdp; 3607 struct file *fp; 3608 struct proc *p; 3609 int header; 3610 int n; 3611 3612 header = 1; 3613 FOREACH_PROC_IN_SYSTEM(p) { 3614 if (p->p_state == PRS_NEW) 3615 continue; 3616 if ((fdp = p->p_fd) == NULL) 3617 continue; 3618 for (n = 0; n <= fdp->fd_lastfile; ++n) { 3619 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3620 continue; 3621 db_print_file(fp, header); 3622 header = 0; 3623 } 3624 } 3625} 3626#endif 3627 3628SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3629 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3630 3631SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3632 &maxfiles, 0, "Maximum number of files"); 3633 3634SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3635 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3636 3637/* ARGSUSED*/ 3638static void 3639filelistinit(void *dummy) 3640{ 3641 3642 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3643 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3644 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 3645 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 3646 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3647 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3648} 3649SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3650 3651/*-------------------------------------------------------------------*/ 3652 3653static int 3654badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3655 int flags, struct thread *td) 3656{ 3657 3658 return (EBADF); 3659} 3660 3661static int 3662badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3663 struct thread *td) 3664{ 3665 3666 return (EINVAL); 3667} 3668 3669static int 3670badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3671 struct thread *td) 3672{ 3673 3674 return (EBADF); 3675} 3676 3677static int 3678badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3679 struct thread *td) 3680{ 3681 3682 return (0); 3683} 3684 3685static int 3686badfo_kqfilter(struct file *fp, struct knote *kn) 3687{ 3688 3689 return (EBADF); 3690} 3691 3692static int 3693badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3694 struct thread *td) 3695{ 3696 3697 return (EBADF); 3698} 3699 3700static int 3701badfo_close(struct file *fp, struct thread *td) 3702{ 3703 3704 return (0); 3705} 3706 3707static int 3708badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3709 struct thread *td) 3710{ 3711 3712 return (EBADF); 3713} 3714 3715static int 3716badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3717 struct thread *td) 3718{ 3719 3720 return (EBADF); 3721} 3722 3723static int 3724badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3725 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3726 int kflags, struct thread *td) 3727{ 3728 3729 return (EBADF); 3730} 3731 3732static int 3733badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 3734{ 3735 3736 return (0); 3737} 3738 3739struct fileops badfileops = { 3740 .fo_read = badfo_readwrite, 3741 .fo_write = badfo_readwrite, 3742 .fo_truncate = badfo_truncate, 3743 .fo_ioctl = badfo_ioctl, 3744 .fo_poll = badfo_poll, 3745 .fo_kqfilter = badfo_kqfilter, 3746 .fo_stat = badfo_stat, 3747 .fo_close = badfo_close, 3748 .fo_chmod = badfo_chmod, 3749 .fo_chown = badfo_chown, 3750 .fo_sendfile = badfo_sendfile, 3751 .fo_fill_kinfo = badfo_fill_kinfo, 3752}; 3753 3754int 3755invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 3756 int flags, struct thread *td) 3757{ 3758 3759 return (EOPNOTSUPP); 3760} 3761 3762int 3763invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3764 struct thread *td) 3765{ 3766 3767 return (EINVAL); 3768} 3769 3770int 3771invfo_ioctl(struct file *fp, u_long com, void *data, 3772 struct ucred *active_cred, struct thread *td) 3773{ 3774 3775 return (ENOTTY); 3776} 3777 3778int 3779invfo_poll(struct file *fp, int events, struct ucred *active_cred, 3780 struct thread *td) 3781{ 3782 3783 return (poll_no_poll(events)); 3784} 3785 3786int 3787invfo_kqfilter(struct file *fp, struct knote *kn) 3788{ 3789 3790 return (EINVAL); 3791} 3792 3793int 3794invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3795 struct thread *td) 3796{ 3797 3798 return (EINVAL); 3799} 3800 3801int 3802invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3803 struct thread *td) 3804{ 3805 3806 return (EINVAL); 3807} 3808 3809int 3810invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3811 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3812 int kflags, struct thread *td) 3813{ 3814 3815 return (EINVAL); 3816} 3817 3818/*-------------------------------------------------------------------*/ 3819 3820/* 3821 * File Descriptor pseudo-device driver (/dev/fd/). 3822 * 3823 * Opening minor device N dup()s the file (if any) connected to file 3824 * descriptor N belonging to the calling process. Note that this driver 3825 * consists of only the ``open()'' routine, because all subsequent 3826 * references to this file will be direct to the other driver. 3827 * 3828 * XXX: we could give this one a cloning event handler if necessary. 3829 */ 3830 3831/* ARGSUSED */ 3832static int 3833fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3834{ 3835 3836 /* 3837 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3838 * the file descriptor being sought for duplication. The error 3839 * return ensures that the vnode for this device will be released 3840 * by vn_open. Open will detect this special error and take the 3841 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3842 * will simply report the error. 3843 */ 3844 td->td_dupfd = dev2unit(dev); 3845 return (ENODEV); 3846} 3847 3848static struct cdevsw fildesc_cdevsw = { 3849 .d_version = D_VERSION, 3850 .d_open = fdopen, 3851 .d_name = "FD", 3852}; 3853 3854static void 3855fildesc_drvinit(void *unused) 3856{ 3857 struct cdev *dev; 3858 3859 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 3860 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 3861 make_dev_alias(dev, "stdin"); 3862 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 3863 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 3864 make_dev_alias(dev, "stdout"); 3865 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 3866 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 3867 make_dev_alias(dev, "stderr"); 3868} 3869 3870SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 3871