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 176471 2008-02-23 01:01:49Z des $");
39
40#include "opt_compat.h"
41#include "opt_ddb.h"
42#include "opt_ktrace.h"
43
44#include <sys/param.h>
45#include <sys/systm.h>
46
47#include <sys/conf.h>
48#include <sys/domain.h>
49#include <sys/fcntl.h>
50#include <sys/file.h>
51#include <sys/filedesc.h>
52#include <sys/filio.h>
53#include <sys/jail.h>
54#include <sys/kernel.h>
55#include <sys/limits.h>
56#include <sys/lock.h>
57#include <sys/malloc.h>
58#include <sys/mount.h>
59#include <sys/mqueue.h>
60#include <sys/mutex.h>
61#include <sys/namei.h>
62#include <sys/priv.h>
63#include <sys/proc.h>
64#include <sys/protosw.h>
65#include <sys/resourcevar.h>
66#include <sys/signalvar.h>
67#include <sys/socketvar.h>
68#include <sys/stat.h>
69#include <sys/sx.h>
70#include <sys/syscallsubr.h>
71#include <sys/sysctl.h>
72#include <sys/sysproto.h>
73#include <sys/unistd.h>
74#include <sys/user.h>
75#include <sys/vnode.h>
76#ifdef KTRACE
77#include <sys/ktrace.h>
78#endif
79
80#include <security/audit/audit.h>
81
82#include <vm/uma.h>
83
84#include <ddb/ddb.h>
85
86static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
87static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
88 "file desc to leader structures");
89static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
90
91static uma_zone_t file_zone;
92
93
94/* How to treat 'new' parameter when allocating a fd for do_dup(). */
95enum dup_type { DUP_VARIABLE, DUP_FIXED };
96
97static int do_dup(struct thread *td, enum dup_type type, int old, int new,
98 register_t *retval);
99static int fd_first_free(struct filedesc *, int, int);
100static int fd_last_used(struct filedesc *, int, int);
101static void fdgrowtable(struct filedesc *, int);
102static void fdunused(struct filedesc *fdp, int fd);
103static void fdused(struct filedesc *fdp, int fd);
104
105/*
106 * A process is initially started out with NDFILE descriptors stored within
107 * this structure, selected to be enough for typical applications based on
108 * the historical limit of 20 open files (and the usage of descriptors by
109 * shells). If these descriptors are exhausted, a larger descriptor table
110 * may be allocated, up to a process' resource limit; the internal arrays
111 * are then unused.
112 */
113#define NDFILE 20
114#define NDSLOTSIZE sizeof(NDSLOTTYPE)
115#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
116#define NDSLOT(x) ((x) / NDENTRIES)
117#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
118#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
119
120/*
121 * Storage required per open file descriptor.
122 */
123#define OFILESIZE (sizeof(struct file *) + sizeof(char))
124
125/*
126 * Basic allocation of descriptors:
127 * one of the above, plus arrays for NDFILE descriptors.
128 */
129struct filedesc0 {
130 struct filedesc fd_fd;
131 /*
132 * These arrays are used when the number of open files is
133 * <= NDFILE, and are then pointed to by the pointers above.
134 */
135 struct file *fd_dfiles[NDFILE];
136 char fd_dfileflags[NDFILE];
137 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
138};
139
140/*
141 * Descriptor management.
142 */
143volatile int openfiles; /* actual number of open files */
144struct mtx sigio_lock; /* mtx to protect pointers to sigio */
145void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
146
147/* A mutex to protect the association between a proc and filedesc. */
148static struct mtx fdesc_mtx;
149
150/*
151 * Find the first zero bit in the given bitmap, starting at low and not
152 * exceeding size - 1.
153 */
154static int
155fd_first_free(struct filedesc *fdp, int low, int size)
156{
157 NDSLOTTYPE *map = fdp->fd_map;
158 NDSLOTTYPE mask;
159 int off, maxoff;
160
161 if (low >= size)
162 return (low);
163
164 off = NDSLOT(low);
165 if (low % NDENTRIES) {
166 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
167 if ((mask &= ~map[off]) != 0UL)
168 return (off * NDENTRIES + ffsl(mask) - 1);
169 ++off;
170 }
171 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
172 if (map[off] != ~0UL)
173 return (off * NDENTRIES + ffsl(~map[off]) - 1);
174 return (size);
175}
176
177/*
178 * Find the highest non-zero bit in the given bitmap, starting at low and
179 * not exceeding size - 1.
180 */
181static int
182fd_last_used(struct filedesc *fdp, int low, int size)
183{
184 NDSLOTTYPE *map = fdp->fd_map;
185 NDSLOTTYPE mask;
186 int off, minoff;
187
188 if (low >= size)
189 return (-1);
190
191 off = NDSLOT(size);
192 if (size % NDENTRIES) {
193 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
194 if ((mask &= map[off]) != 0)
195 return (off * NDENTRIES + flsl(mask) - 1);
196 --off;
197 }
198 for (minoff = NDSLOT(low); off >= minoff; --off)
199 if (map[off] != 0)
200 return (off * NDENTRIES + flsl(map[off]) - 1);
201 return (low - 1);
202}
203
204static int
205fdisused(struct filedesc *fdp, int fd)
206{
207 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
208 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
209 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
210}
211
212/*
213 * Mark a file descriptor as used.
214 */
215static void
216fdused(struct filedesc *fdp, int fd)
217{
218
219 FILEDESC_XLOCK_ASSERT(fdp);
220 KASSERT(!fdisused(fdp, fd),
221 ("fd already used"));
222
223 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
224 if (fd > fdp->fd_lastfile)
225 fdp->fd_lastfile = fd;
226 if (fd == fdp->fd_freefile)
227 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
228}
229
230/*
231 * Mark a file descriptor as unused.
232 */
233static void
234fdunused(struct filedesc *fdp, int fd)
235{
236
237 FILEDESC_XLOCK_ASSERT(fdp);
238 KASSERT(fdisused(fdp, fd),
239 ("fd is already unused"));
240 KASSERT(fdp->fd_ofiles[fd] == NULL,
241 ("fd is still in use"));
242
243 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
244 if (fd < fdp->fd_freefile)
245 fdp->fd_freefile = fd;
246 if (fd == fdp->fd_lastfile)
247 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
248}
249
250/*
251 * System calls on descriptors.
252 */
253#ifndef _SYS_SYSPROTO_H_
254struct getdtablesize_args {
255 int dummy;
256};
257#endif
258/* ARGSUSED */
259int
260getdtablesize(struct thread *td, struct getdtablesize_args *uap)
261{
262 struct proc *p = td->td_proc;
263
264 PROC_LOCK(p);
265 td->td_retval[0] =
266 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
267 PROC_UNLOCK(p);
268 return (0);
269}
270
271/*
272 * Duplicate a file descriptor to a particular value.
273 *
274 * Note: keep in mind that a potential race condition exists when closing
275 * descriptors from a shared descriptor table (via rfork).
276 */
277#ifndef _SYS_SYSPROTO_H_
278struct dup2_args {
279 u_int from;
280 u_int to;
281};
282#endif
283/* ARGSUSED */
284int
285dup2(struct thread *td, struct dup2_args *uap)
286{
287
288 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
289 td->td_retval));
290}
291
292/*
293 * Duplicate a file descriptor.
294 */
295#ifndef _SYS_SYSPROTO_H_
296struct dup_args {
297 u_int fd;
298};
299#endif
300/* ARGSUSED */
301int
302dup(struct thread *td, struct dup_args *uap)
303{
304
305 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
306}
307
308/*
309 * The file control system call.
310 */
311#ifndef _SYS_SYSPROTO_H_
312struct fcntl_args {
313 int fd;
314 int cmd;
315 long arg;
316};
317#endif
318/* ARGSUSED */
319int
320fcntl(struct thread *td, struct fcntl_args *uap)
321{
322 struct flock fl;
323 intptr_t arg;
324 int error;
325
326 error = 0;
327 switch (uap->cmd) {
328 case F_GETLK:
329 case F_SETLK:
330 case F_SETLKW:
331 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
332 arg = (intptr_t)&fl;
333 break;
334 default:
335 arg = uap->arg;
336 break;
337 }
338 if (error)
339 return (error);
340 error = kern_fcntl(td, uap->fd, uap->cmd, arg);
341 if (error)
342 return (error);
343 if (uap->cmd == F_GETLK)
344 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
345 return (error);
346}
347
348static inline struct file *
349fdtofp(int fd, struct filedesc *fdp)
350{
351 struct file *fp;
352
353 FILEDESC_LOCK_ASSERT(fdp);
354 if ((unsigned)fd >= fdp->fd_nfiles ||
355 (fp = fdp->fd_ofiles[fd]) == NULL)
356 return (NULL);
357 return (fp);
358}
359
360int
361kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
362{
363 struct filedesc *fdp;
364 struct flock *flp;
365 struct file *fp;
366 struct proc *p;
367 char *pop;
368 struct vnode *vp;
369 u_int newmin;
370 int error, flg, tmp;
371 int vfslocked;
372
373 vfslocked = 0;
374 error = 0;
375 flg = F_POSIX;
376 p = td->td_proc;
377 fdp = p->p_fd;
378
379 switch (cmd) {
380 case F_DUPFD:
381 FILEDESC_SLOCK(fdp);
382 if ((fp = fdtofp(fd, fdp)) == NULL) {
383 FILEDESC_SUNLOCK(fdp);
384 error = EBADF;
385 break;
386 }
387 FILEDESC_SUNLOCK(fdp);
388 newmin = arg;
389 PROC_LOCK(p);
390 if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
391 newmin >= maxfilesperproc) {
392 PROC_UNLOCK(p);
393 error = EINVAL;
394 break;
395 }
396 PROC_UNLOCK(p);
397 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
398 break;
399
400 case F_GETFD:
401 FILEDESC_SLOCK(fdp);
402 if ((fp = fdtofp(fd, fdp)) == NULL) {
403 FILEDESC_SUNLOCK(fdp);
404 error = EBADF;
405 break;
406 }
407 pop = &fdp->fd_ofileflags[fd];
408 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
409 FILEDESC_SUNLOCK(fdp);
410 break;
411
412 case F_SETFD:
413 FILEDESC_XLOCK(fdp);
414 if ((fp = fdtofp(fd, fdp)) == NULL) {
415 FILEDESC_XUNLOCK(fdp);
416 error = EBADF;
417 break;
418 }
419 pop = &fdp->fd_ofileflags[fd];
420 *pop = (*pop &~ UF_EXCLOSE) |
421 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
422 FILEDESC_XUNLOCK(fdp);
423 break;
424
425 case F_GETFL:
426 FILEDESC_SLOCK(fdp);
427 if ((fp = fdtofp(fd, fdp)) == NULL) {
428 FILEDESC_SUNLOCK(fdp);
429 error = EBADF;
430 break;
431 }
432 td->td_retval[0] = OFLAGS(fp->f_flag);
433 FILEDESC_SUNLOCK(fdp);
434 break;
435
436 case F_SETFL:
437 FILEDESC_SLOCK(fdp);
438 if ((fp = fdtofp(fd, fdp)) == NULL) {
439 FILEDESC_SUNLOCK(fdp);
440 error = EBADF;
441 break;
442 }
443 fhold(fp);
444 FILEDESC_SUNLOCK(fdp);
445 do {
446 tmp = flg = fp->f_flag;
447 tmp &= ~FCNTLFLAGS;
448 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
449 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
450 tmp = fp->f_flag & FNONBLOCK;
451 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
452 if (error) {
453 fdrop(fp, td);
454 break;
455 }
456 tmp = fp->f_flag & FASYNC;
457 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
458 if (error == 0) {
459 fdrop(fp, td);
460 break;
461 }
462 atomic_clear_int(&fp->f_flag, FNONBLOCK);
463 tmp = 0;
464 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
465 fdrop(fp, td);
466 break;
467
468 case F_GETOWN:
469 FILEDESC_SLOCK(fdp);
470 if ((fp = fdtofp(fd, fdp)) == NULL) {
471 FILEDESC_SUNLOCK(fdp);
472 error = EBADF;
473 break;
474 }
475 fhold(fp);
476 FILEDESC_SUNLOCK(fdp);
477 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
478 if (error == 0)
479 td->td_retval[0] = tmp;
480 fdrop(fp, td);
481 break;
482
483 case F_SETOWN:
484 FILEDESC_SLOCK(fdp);
485 if ((fp = fdtofp(fd, fdp)) == NULL) {
486 FILEDESC_SUNLOCK(fdp);
487 error = EBADF;
488 break;
489 }
490 fhold(fp);
491 FILEDESC_SUNLOCK(fdp);
492 tmp = arg;
493 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
494 fdrop(fp, td);
495 break;
496
497 case F_SETLKW:
498 flg |= F_WAIT;
499 /* FALLTHROUGH F_SETLK */
500
501 case F_SETLK:
502 FILEDESC_SLOCK(fdp);
503 if ((fp = fdtofp(fd, fdp)) == NULL) {
504 FILEDESC_SUNLOCK(fdp);
505 error = EBADF;
506 break;
507 }
508 if (fp->f_type != DTYPE_VNODE) {
509 FILEDESC_SUNLOCK(fdp);
510 error = EBADF;
511 break;
512 }
513 flp = (struct flock *)arg;
514 if (flp->l_whence == SEEK_CUR) {
515 if (fp->f_offset < 0 ||
516 (flp->l_start > 0 &&
517 fp->f_offset > OFF_MAX - flp->l_start)) {
518 FILEDESC_SUNLOCK(fdp);
519 error = EOVERFLOW;
520 break;
521 }
522 flp->l_start += fp->f_offset;
523 }
524
525 /*
526 * VOP_ADVLOCK() may block.
527 */
528 fhold(fp);
529 FILEDESC_SUNLOCK(fdp);
530 vp = fp->f_vnode;
531 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
532 switch (flp->l_type) {
533 case F_RDLCK:
534 if ((fp->f_flag & FREAD) == 0) {
535 error = EBADF;
536 break;
537 }
538 PROC_LOCK(p->p_leader);
539 p->p_leader->p_flag |= P_ADVLOCK;
540 PROC_UNLOCK(p->p_leader);
541 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
542 flp, flg);
543 break;
544 case F_WRLCK:
545 if ((fp->f_flag & FWRITE) == 0) {
546 error = EBADF;
547 break;
548 }
549 PROC_LOCK(p->p_leader);
550 p->p_leader->p_flag |= P_ADVLOCK;
551 PROC_UNLOCK(p->p_leader);
552 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
553 flp, flg);
554 break;
555 case F_UNLCK:
556 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
557 flp, F_POSIX);
558 break;
559 default:
560 error = EINVAL;
561 break;
562 }
563 VFS_UNLOCK_GIANT(vfslocked);
564 vfslocked = 0;
565 /* Check for race with close */
566 FILEDESC_SLOCK(fdp);
567 if ((unsigned) fd >= fdp->fd_nfiles ||
568 fp != fdp->fd_ofiles[fd]) {
569 FILEDESC_SUNLOCK(fdp);
570 flp->l_whence = SEEK_SET;
571 flp->l_start = 0;
572 flp->l_len = 0;
573 flp->l_type = F_UNLCK;
574 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
575 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
576 F_UNLCK, flp, F_POSIX);
577 VFS_UNLOCK_GIANT(vfslocked);
578 vfslocked = 0;
579 } else
580 FILEDESC_SUNLOCK(fdp);
581 fdrop(fp, td);
582 break;
583
584 case F_GETLK:
585 FILEDESC_SLOCK(fdp);
586 if ((fp = fdtofp(fd, fdp)) == NULL) {
587 FILEDESC_SUNLOCK(fdp);
588 error = EBADF;
589 break;
590 }
591 if (fp->f_type != DTYPE_VNODE) {
592 FILEDESC_SUNLOCK(fdp);
593 error = EBADF;
594 break;
595 }
596 flp = (struct flock *)arg;
597 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
598 flp->l_type != F_UNLCK) {
599 FILEDESC_SUNLOCK(fdp);
600 error = EINVAL;
601 break;
602 }
603 if (flp->l_whence == SEEK_CUR) {
604 if ((flp->l_start > 0 &&
605 fp->f_offset > OFF_MAX - flp->l_start) ||
606 (flp->l_start < 0 &&
607 fp->f_offset < OFF_MIN - flp->l_start)) {
608 FILEDESC_SUNLOCK(fdp);
609 error = EOVERFLOW;
610 break;
611 }
612 flp->l_start += fp->f_offset;
613 }
614 /*
615 * VOP_ADVLOCK() may block.
616 */
617 fhold(fp);
618 FILEDESC_SUNLOCK(fdp);
619 vp = fp->f_vnode;
620 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
621 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
622 F_POSIX);
623 VFS_UNLOCK_GIANT(vfslocked);
624 vfslocked = 0;
625 fdrop(fp, td);
626 break;
627 default:
628 error = EINVAL;
629 break;
630 }
631 VFS_UNLOCK_GIANT(vfslocked);
632 return (error);
633}
634
635/*
636 * Common code for dup, dup2, and fcntl(F_DUPFD).
637 */
638static int
639do_dup(struct thread *td, enum dup_type type, int old, int new,
640 register_t *retval)
641{
642 struct filedesc *fdp;
643 struct proc *p;
644 struct file *fp;
645 struct file *delfp;
646 int error, holdleaders, maxfd;
647
648 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
649 ("invalid dup type %d", type));
650
651 p = td->td_proc;
652 fdp = p->p_fd;
653
654 /*
655 * Verify we have a valid descriptor to dup from and possibly to
656 * dup to.
657 */
658 if (old < 0 || new < 0)
659 return (EBADF);
660 PROC_LOCK(p);
661 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
662 PROC_UNLOCK(p);
663 if (new >= maxfd)
664 return (EMFILE);
665
666 FILEDESC_XLOCK(fdp);
667 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
668 FILEDESC_XUNLOCK(fdp);
669 return (EBADF);
670 }
671 if (type == DUP_FIXED && old == new) {
672 *retval = new;
673 FILEDESC_XUNLOCK(fdp);
674 return (0);
675 }
676 fp = fdp->fd_ofiles[old];
677 fhold(fp);
678
679 /*
680 * If the caller specified a file descriptor, make sure the file
681 * table is large enough to hold it, and grab it. Otherwise, just
682 * allocate a new descriptor the usual way. Since the filedesc
683 * lock may be temporarily dropped in the process, we have to look
684 * out for a race.
685 */
686 if (type == DUP_FIXED) {
687 if (new >= fdp->fd_nfiles)
688 fdgrowtable(fdp, new + 1);
689 if (fdp->fd_ofiles[new] == NULL)
690 fdused(fdp, new);
691 } else {
692 if ((error = fdalloc(td, new, &new)) != 0) {
693 FILEDESC_XUNLOCK(fdp);
694 fdrop(fp, td);
695 return (error);
696 }
697 }
698
699 /*
700 * If the old file changed out from under us then treat it as a
701 * bad file descriptor. Userland should do its own locking to
702 * avoid this case.
703 */
704 if (fdp->fd_ofiles[old] != fp) {
705 /* we've allocated a descriptor which we won't use */
706 if (fdp->fd_ofiles[new] == NULL)
707 fdunused(fdp, new);
708 FILEDESC_XUNLOCK(fdp);
709 fdrop(fp, td);
710 return (EBADF);
711 }
712 KASSERT(old != new,
713 ("new fd is same as old"));
714
715 /*
716 * Save info on the descriptor being overwritten. We cannot close
717 * it without introducing an ownership race for the slot, since we
718 * need to drop the filedesc lock to call closef().
719 *
720 * XXX this duplicates parts of close().
721 */
722 delfp = fdp->fd_ofiles[new];
723 holdleaders = 0;
724 if (delfp != NULL) {
725 if (td->td_proc->p_fdtol != NULL) {
726 /*
727 * Ask fdfree() to sleep to ensure that all relevant
728 * process leaders can be traversed in closef().
729 */
730 fdp->fd_holdleaderscount++;
731 holdleaders = 1;
732 }
733 }
734
735 /*
736 * Duplicate the source descriptor
737 */
738 fdp->fd_ofiles[new] = fp;
739 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
740 if (new > fdp->fd_lastfile)
741 fdp->fd_lastfile = new;
742 *retval = new;
743
744 /*
745 * If we dup'd over a valid file, we now own the reference to it
746 * and must dispose of it using closef() semantics (as if a
747 * close() were performed on it).
748 *
749 * XXX this duplicates parts of close().
750 */
751 if (delfp != NULL) {
752 knote_fdclose(td, new);
753 if (delfp->f_type == DTYPE_MQUEUE)
754 mq_fdclose(td, new, delfp);
755 FILEDESC_XUNLOCK(fdp);
756 (void) closef(delfp, td);
757 if (holdleaders) {
758 FILEDESC_XLOCK(fdp);
759 fdp->fd_holdleaderscount--;
760 if (fdp->fd_holdleaderscount == 0 &&
761 fdp->fd_holdleaderswakeup != 0) {
762 fdp->fd_holdleaderswakeup = 0;
763 wakeup(&fdp->fd_holdleaderscount);
764 }
765 FILEDESC_XUNLOCK(fdp);
766 }
767 } else {
768 FILEDESC_XUNLOCK(fdp);
769 }
770 return (0);
771}
772
773/*
774 * If sigio is on the list associated with a process or process group,
775 * disable signalling from the device, remove sigio from the list and
776 * free sigio.
777 */
778void
779funsetown(struct sigio **sigiop)
780{
781 struct sigio *sigio;
782
783 SIGIO_LOCK();
784 sigio = *sigiop;
785 if (sigio == NULL) {
786 SIGIO_UNLOCK();
787 return;
788 }
789 *(sigio->sio_myref) = NULL;
790 if ((sigio)->sio_pgid < 0) {
791 struct pgrp *pg = (sigio)->sio_pgrp;
792 PGRP_LOCK(pg);
793 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
794 sigio, sio_pgsigio);
795 PGRP_UNLOCK(pg);
796 } else {
797 struct proc *p = (sigio)->sio_proc;
798 PROC_LOCK(p);
799 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
800 sigio, sio_pgsigio);
801 PROC_UNLOCK(p);
802 }
803 SIGIO_UNLOCK();
804 crfree(sigio->sio_ucred);
805 FREE(sigio, M_SIGIO);
806}
807
808/*
809 * Free a list of sigio structures.
810 * We only need to lock the SIGIO_LOCK because we have made ourselves
811 * inaccessible to callers of fsetown and therefore do not need to lock
812 * the proc or pgrp struct for the list manipulation.
813 */
814void
815funsetownlst(struct sigiolst *sigiolst)
816{
817 struct proc *p;
818 struct pgrp *pg;
819 struct sigio *sigio;
820
821 sigio = SLIST_FIRST(sigiolst);
822 if (sigio == NULL)
823 return;
824 p = NULL;
825 pg = NULL;
826
827 /*
828 * Every entry of the list should belong
829 * to a single proc or pgrp.
830 */
831 if (sigio->sio_pgid < 0) {
832 pg = sigio->sio_pgrp;
833 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
834 } else /* if (sigio->sio_pgid > 0) */ {
835 p = sigio->sio_proc;
836 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
837 }
838
839 SIGIO_LOCK();
840 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
841 *(sigio->sio_myref) = NULL;
842 if (pg != NULL) {
843 KASSERT(sigio->sio_pgid < 0,
844 ("Proc sigio in pgrp sigio list"));
845 KASSERT(sigio->sio_pgrp == pg,
846 ("Bogus pgrp in sigio list"));
847 PGRP_LOCK(pg);
848 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
849 sio_pgsigio);
850 PGRP_UNLOCK(pg);
851 } else /* if (p != NULL) */ {
852 KASSERT(sigio->sio_pgid > 0,
853 ("Pgrp sigio in proc sigio list"));
854 KASSERT(sigio->sio_proc == p,
855 ("Bogus proc in sigio list"));
856 PROC_LOCK(p);
857 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
858 sio_pgsigio);
859 PROC_UNLOCK(p);
860 }
861 SIGIO_UNLOCK();
862 crfree(sigio->sio_ucred);
863 FREE(sigio, M_SIGIO);
864 SIGIO_LOCK();
865 }
866 SIGIO_UNLOCK();
867}
868
869/*
870 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
871 *
872 * After permission checking, add a sigio structure to the sigio list for
873 * the process or process group.
874 */
875int
876fsetown(pid_t pgid, struct sigio **sigiop)
877{
878 struct proc *proc;
879 struct pgrp *pgrp;
880 struct sigio *sigio;
881 int ret;
882
883 if (pgid == 0) {
884 funsetown(sigiop);
885 return (0);
886 }
887
888 ret = 0;
889
890 /* Allocate and fill in the new sigio out of locks. */
891 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
892 sigio->sio_pgid = pgid;
893 sigio->sio_ucred = crhold(curthread->td_ucred);
894 sigio->sio_myref = sigiop;
895
896 sx_slock(&proctree_lock);
897 if (pgid > 0) {
898 proc = pfind(pgid);
899 if (proc == NULL) {
900 ret = ESRCH;
901 goto fail;
902 }
903
904 /*
905 * Policy - Don't allow a process to FSETOWN a process
906 * in another session.
907 *
908 * Remove this test to allow maximum flexibility or
909 * restrict FSETOWN to the current process or process
910 * group for maximum safety.
911 */
912 PROC_UNLOCK(proc);
913 if (proc->p_session != curthread->td_proc->p_session) {
914 ret = EPERM;
915 goto fail;
916 }
917
918 pgrp = NULL;
919 } else /* if (pgid < 0) */ {
920 pgrp = pgfind(-pgid);
921 if (pgrp == NULL) {
922 ret = ESRCH;
923 goto fail;
924 }
925 PGRP_UNLOCK(pgrp);
926
927 /*
928 * Policy - Don't allow a process to FSETOWN a process
929 * in another session.
930 *
931 * Remove this test to allow maximum flexibility or
932 * restrict FSETOWN to the current process or process
933 * group for maximum safety.
934 */
935 if (pgrp->pg_session != curthread->td_proc->p_session) {
936 ret = EPERM;
937 goto fail;
938 }
939
940 proc = NULL;
941 }
942 funsetown(sigiop);
943 if (pgid > 0) {
944 PROC_LOCK(proc);
945 /*
946 * Since funsetownlst() is called without the proctree
947 * locked, we need to check for P_WEXIT.
948 * XXX: is ESRCH correct?
949 */
950 if ((proc->p_flag & P_WEXIT) != 0) {
951 PROC_UNLOCK(proc);
952 ret = ESRCH;
953 goto fail;
954 }
955 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
956 sigio->sio_proc = proc;
957 PROC_UNLOCK(proc);
958 } else {
959 PGRP_LOCK(pgrp);
960 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
961 sigio->sio_pgrp = pgrp;
962 PGRP_UNLOCK(pgrp);
963 }
964 sx_sunlock(&proctree_lock);
965 SIGIO_LOCK();
966 *sigiop = sigio;
967 SIGIO_UNLOCK();
968 return (0);
969
970fail:
971 sx_sunlock(&proctree_lock);
972 crfree(sigio->sio_ucred);
973 FREE(sigio, M_SIGIO);
974 return (ret);
975}
976
977/*
978 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
979 */
980pid_t
981fgetown(sigiop)
982 struct sigio **sigiop;
983{
984 pid_t pgid;
985
986 SIGIO_LOCK();
987 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
988 SIGIO_UNLOCK();
989 return (pgid);
990}
991
992/*
993 * Close a file descriptor.
994 */
995#ifndef _SYS_SYSPROTO_H_
996struct close_args {
997 int fd;
998};
999#endif
1000/* ARGSUSED */
1001int
1002close(td, uap)
1003 struct thread *td;
1004 struct close_args *uap;
1005{
1006
1007 return (kern_close(td, uap->fd));
1008}
1009
1010int
1011kern_close(td, fd)
1012 struct thread *td;
1013 int fd;
1014{
1015 struct filedesc *fdp;
1016 struct file *fp;
1017 int error;
1018 int holdleaders;
1019
1020 error = 0;
1021 holdleaders = 0;
1022 fdp = td->td_proc->p_fd;
1023
1024 AUDIT_SYSCLOSE(td, fd);
1025
1026 FILEDESC_XLOCK(fdp);
1027 if ((unsigned)fd >= fdp->fd_nfiles ||
1028 (fp = fdp->fd_ofiles[fd]) == NULL) {
1029 FILEDESC_XUNLOCK(fdp);
1030 return (EBADF);
1031 }
1032 fdp->fd_ofiles[fd] = NULL;
1033 fdp->fd_ofileflags[fd] = 0;
1034 fdunused(fdp, fd);
1035 if (td->td_proc->p_fdtol != NULL) {
1036 /*
1037 * Ask fdfree() to sleep to ensure that all relevant
1038 * process leaders can be traversed in closef().
1039 */
1040 fdp->fd_holdleaderscount++;
1041 holdleaders = 1;
1042 }
1043
1044 /*
1045 * We now hold the fp reference that used to be owned by the
1046 * descriptor array. We have to unlock the FILEDESC *AFTER*
1047 * knote_fdclose to prevent a race of the fd getting opened, a knote
1048 * added, and deleteing a knote for the new fd.
1049 */
1050 knote_fdclose(td, fd);
1051 if (fp->f_type == DTYPE_MQUEUE)
1052 mq_fdclose(td, fd, fp);
1053 FILEDESC_XUNLOCK(fdp);
1054
1055 error = closef(fp, td);
1056 if (holdleaders) {
1057 FILEDESC_XLOCK(fdp);
1058 fdp->fd_holdleaderscount--;
1059 if (fdp->fd_holdleaderscount == 0 &&
1060 fdp->fd_holdleaderswakeup != 0) {
1061 fdp->fd_holdleaderswakeup = 0;
1062 wakeup(&fdp->fd_holdleaderscount);
1063 }
1064 FILEDESC_XUNLOCK(fdp);
1065 }
1066 return (error);
1067}
1068
1069#if defined(COMPAT_43)
1070/*
1071 * Return status information about a file descriptor.
1072 */
1073#ifndef _SYS_SYSPROTO_H_
1074struct ofstat_args {
1075 int fd;
1076 struct ostat *sb;
1077};
1078#endif
1079/* ARGSUSED */
1080int
1081ofstat(struct thread *td, struct ofstat_args *uap)
1082{
1083 struct ostat oub;
1084 struct stat ub;
1085 int error;
1086
1087 error = kern_fstat(td, uap->fd, &ub);
1088 if (error == 0) {
1089 cvtstat(&ub, &oub);
1090 error = copyout(&oub, uap->sb, sizeof(oub));
1091 }
1092 return (error);
1093}
1094#endif /* COMPAT_43 */
1095
1096/*
1097 * Return status information about a file descriptor.
1098 */
1099#ifndef _SYS_SYSPROTO_H_
1100struct fstat_args {
1101 int fd;
1102 struct stat *sb;
1103};
1104#endif
1105/* ARGSUSED */
1106int
1107fstat(struct thread *td, struct fstat_args *uap)
1108{
1109 struct stat ub;
1110 int error;
1111
1112 error = kern_fstat(td, uap->fd, &ub);
1113 if (error == 0)
1114 error = copyout(&ub, uap->sb, sizeof(ub));
1115 return (error);
1116}
1117
1118int
1119kern_fstat(struct thread *td, int fd, struct stat *sbp)
1120{
1121 struct file *fp;
1122 int error;
1123
1124 AUDIT_ARG(fd, fd);
1125
1126 if ((error = fget(td, fd, &fp)) != 0)
1127 return (error);
1128
1129 AUDIT_ARG(file, td->td_proc, fp);
1130
1131 error = fo_stat(fp, sbp, td->td_ucred, td);
1132 fdrop(fp, td);
1133#ifdef KTRACE
1134 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1135 ktrstat(sbp);
1136#endif
1137 return (error);
1138}
1139
1140/*
1141 * Return status information about a file descriptor.
1142 */
1143#ifndef _SYS_SYSPROTO_H_
1144struct nfstat_args {
1145 int fd;
1146 struct nstat *sb;
1147};
1148#endif
1149/* ARGSUSED */
1150int
1151nfstat(struct thread *td, struct nfstat_args *uap)
1152{
1153 struct nstat nub;
1154 struct stat ub;
1155 int error;
1156
1157 error = kern_fstat(td, uap->fd, &ub);
1158 if (error == 0) {
1159 cvtnstat(&ub, &nub);
1160 error = copyout(&nub, uap->sb, sizeof(nub));
1161 }
1162 return (error);
1163}
1164
1165/*
1166 * Return pathconf information about a file descriptor.
1167 */
1168#ifndef _SYS_SYSPROTO_H_
1169struct fpathconf_args {
1170 int fd;
1171 int name;
1172};
1173#endif
1174/* ARGSUSED */
1175int
1176fpathconf(struct thread *td, struct fpathconf_args *uap)
1177{
1178 struct file *fp;
1179 struct vnode *vp;
1180 int error;
1181
1182 if ((error = fget(td, uap->fd, &fp)) != 0)
1183 return (error);
1184
1185 /* If asynchronous I/O is available, it works for all descriptors. */
1186 if (uap->name == _PC_ASYNC_IO) {
1187 td->td_retval[0] = async_io_version;
1188 goto out;
1189 }
1190 vp = fp->f_vnode;
1191 if (vp != NULL) {
1192 int vfslocked;
1193 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1194 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1195 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1196 VOP_UNLOCK(vp, 0);
1197 VFS_UNLOCK_GIANT(vfslocked);
1198 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1199 if (uap->name != _PC_PIPE_BUF) {
1200 error = EINVAL;
1201 } else {
1202 td->td_retval[0] = PIPE_BUF;
1203 error = 0;
1204 }
1205 } else {
1206 error = EOPNOTSUPP;
1207 }
1208out:
1209 fdrop(fp, td);
1210 return (error);
1211}
1212
1213/*
1214 * Grow the file table to accomodate (at least) nfd descriptors. This may
1215 * block and drop the filedesc lock, but it will reacquire it before
1216 * returning.
1217 */
1218static void
1219fdgrowtable(struct filedesc *fdp, int nfd)
1220{
1221 struct file **ntable;
1222 char *nfileflags;
1223 int nnfiles, onfiles;
1224 NDSLOTTYPE *nmap;
1225
1226 FILEDESC_XLOCK_ASSERT(fdp);
1227
1228 KASSERT(fdp->fd_nfiles > 0,
1229 ("zero-length file table"));
1230
1231 /* compute the size of the new table */
1232 onfiles = fdp->fd_nfiles;
1233 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1234 if (nnfiles <= onfiles)
1235 /* the table is already large enough */
1236 return;
1237
1238 /* allocate a new table and (if required) new bitmaps */
1239 FILEDESC_XUNLOCK(fdp);
1240 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1241 M_FILEDESC, M_ZERO | M_WAITOK);
1242 nfileflags = (char *)&ntable[nnfiles];
1243 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1244 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1245 M_FILEDESC, M_ZERO | M_WAITOK);
1246 else
1247 nmap = NULL;
1248 FILEDESC_XLOCK(fdp);
1249
1250 /*
1251 * We now have new tables ready to go. Since we dropped the
1252 * filedesc lock to call malloc(), watch out for a race.
1253 */
1254 onfiles = fdp->fd_nfiles;
1255 if (onfiles >= nnfiles) {
1256 /* we lost the race, but that's OK */
1257 free(ntable, M_FILEDESC);
1258 if (nmap != NULL)
1259 free(nmap, M_FILEDESC);
1260 return;
1261 }
1262 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1263 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1264 if (onfiles > NDFILE)
1265 free(fdp->fd_ofiles, M_FILEDESC);
1266 fdp->fd_ofiles = ntable;
1267 fdp->fd_ofileflags = nfileflags;
1268 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1269 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1270 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1271 free(fdp->fd_map, M_FILEDESC);
1272 fdp->fd_map = nmap;
1273 }
1274 fdp->fd_nfiles = nnfiles;
1275}
1276
1277/*
1278 * Allocate a file descriptor for the process.
1279 */
1280int
1281fdalloc(struct thread *td, int minfd, int *result)
1282{
1283 struct proc *p = td->td_proc;
1284 struct filedesc *fdp = p->p_fd;
1285 int fd = -1, maxfd;
1286
1287 FILEDESC_XLOCK_ASSERT(fdp);
1288
1289 if (fdp->fd_freefile > minfd)
1290 minfd = fdp->fd_freefile;
1291
1292 PROC_LOCK(p);
1293 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1294 PROC_UNLOCK(p);
1295
1296 /*
1297 * Search the bitmap for a free descriptor. If none is found, try
1298 * to grow the file table. Keep at it until we either get a file
1299 * descriptor or run into process or system limits; fdgrowtable()
1300 * may drop the filedesc lock, so we're in a race.
1301 */
1302 for (;;) {
1303 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1304 if (fd >= maxfd)
1305 return (EMFILE);
1306 if (fd < fdp->fd_nfiles)
1307 break;
1308 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1309 }
1310
1311 /*
1312 * Perform some sanity checks, then mark the file descriptor as
1313 * used and return it to the caller.
1314 */
1315 KASSERT(!fdisused(fdp, fd),
1316 ("fd_first_free() returned non-free descriptor"));
1317 KASSERT(fdp->fd_ofiles[fd] == NULL,
1318 ("free descriptor isn't"));
1319 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1320 fdused(fdp, fd);
1321 *result = fd;
1322 return (0);
1323}
1324
1325/*
1326 * Check to see whether n user file descriptors are available to the process
1327 * p.
1328 */
1329int
1330fdavail(struct thread *td, int n)
1331{
1332 struct proc *p = td->td_proc;
1333 struct filedesc *fdp = td->td_proc->p_fd;
1334 struct file **fpp;
1335 int i, lim, last;
1336
1337 FILEDESC_LOCK_ASSERT(fdp);
1338
1339 PROC_LOCK(p);
1340 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1341 PROC_UNLOCK(p);
1342 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1343 return (1);
1344 last = min(fdp->fd_nfiles, lim);
1345 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1346 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1347 if (*fpp == NULL && --n <= 0)
1348 return (1);
1349 }
1350 return (0);
1351}
1352
1353/*
1354 * Create a new open file structure and allocate a file decriptor for the
1355 * process that refers to it. We add one reference to the file for the
1356 * descriptor table and one reference for resultfp. This is to prevent us
1357 * being preempted and the entry in the descriptor table closed after we
1358 * release the FILEDESC lock.
1359 */
1360int
1361falloc(struct thread *td, struct file **resultfp, int *resultfd)
1362{
1363 struct proc *p = td->td_proc;
1364 struct file *fp;
1365 int error, i;
1366 int maxuserfiles = maxfiles - (maxfiles / 20);
1367 static struct timeval lastfail;
1368 static int curfail;
1369
1370 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1371 if ((openfiles >= maxuserfiles &&
1372 priv_check(td, PRIV_MAXFILES) != 0) ||
1373 openfiles >= maxfiles) {
1374 if (ppsratecheck(&lastfail, &curfail, 1)) {
1375 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1376 td->td_ucred->cr_ruid);
1377 }
1378 uma_zfree(file_zone, fp);
1379 return (ENFILE);
1380 }
1381 atomic_add_int(&openfiles, 1);
1382
1383 /*
1384 * If the process has file descriptor zero open, add the new file
1385 * descriptor to the list of open files at that point, otherwise
1386 * put it at the front of the list of open files.
1387 */
1388 fp->f_count = 1;
1389 if (resultfp)
1390 fp->f_count++;
1391 fp->f_cred = crhold(td->td_ucred);
1392 fp->f_ops = &badfileops;
1393 fp->f_data = NULL;
1394 fp->f_vnode = NULL;
1395 FILEDESC_XLOCK(p->p_fd);
1396 if ((error = fdalloc(td, 0, &i))) {
1397 FILEDESC_XUNLOCK(p->p_fd);
1398 fdrop(fp, td);
1399 if (resultfp)
1400 fdrop(fp, td);
1401 return (error);
1402 }
1403 p->p_fd->fd_ofiles[i] = fp;
1404 FILEDESC_XUNLOCK(p->p_fd);
1405 if (resultfp)
1406 *resultfp = fp;
1407 if (resultfd)
1408 *resultfd = i;
1409 return (0);
1410}
1411
1412/*
1413 * Build a new filedesc structure from another.
1414 * Copy the current, root, and jail root vnode references.
1415 */
1416struct filedesc *
1417fdinit(struct filedesc *fdp)
1418{
1419 struct filedesc0 *newfdp;
1420
1421 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1422 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1423 if (fdp != NULL) {
1424 FILEDESC_XLOCK(fdp);
1425 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1426 if (newfdp->fd_fd.fd_cdir)
1427 VREF(newfdp->fd_fd.fd_cdir);
1428 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1429 if (newfdp->fd_fd.fd_rdir)
1430 VREF(newfdp->fd_fd.fd_rdir);
1431 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1432 if (newfdp->fd_fd.fd_jdir)
1433 VREF(newfdp->fd_fd.fd_jdir);
1434 FILEDESC_XUNLOCK(fdp);
1435 }
1436
1437 /* Create the file descriptor table. */
1438 newfdp->fd_fd.fd_refcnt = 1;
1439 newfdp->fd_fd.fd_holdcnt = 1;
1440 newfdp->fd_fd.fd_cmask = CMASK;
1441 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1442 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1443 newfdp->fd_fd.fd_nfiles = NDFILE;
1444 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1445 newfdp->fd_fd.fd_lastfile = -1;
1446 return (&newfdp->fd_fd);
1447}
1448
1449static struct filedesc *
1450fdhold(struct proc *p)
1451{
1452 struct filedesc *fdp;
1453
1454 mtx_lock(&fdesc_mtx);
1455 fdp = p->p_fd;
1456 if (fdp != NULL)
1457 fdp->fd_holdcnt++;
1458 mtx_unlock(&fdesc_mtx);
1459 return (fdp);
1460}
1461
1462static void
1463fddrop(struct filedesc *fdp)
1464{
1465 int i;
1466
1467 mtx_lock(&fdesc_mtx);
1468 i = --fdp->fd_holdcnt;
1469 mtx_unlock(&fdesc_mtx);
1470 if (i > 0)
1471 return;
1472
1473 FILEDESC_LOCK_DESTROY(fdp);
1474 FREE(fdp, M_FILEDESC);
1475}
1476
1477/*
1478 * Share a filedesc structure.
1479 */
1480struct filedesc *
1481fdshare(struct filedesc *fdp)
1482{
1483
1484 FILEDESC_XLOCK(fdp);
1485 fdp->fd_refcnt++;
1486 FILEDESC_XUNLOCK(fdp);
1487 return (fdp);
1488}
1489
1490/*
1491 * Unshare a filedesc structure, if necessary by making a copy
1492 */
1493void
1494fdunshare(struct proc *p, struct thread *td)
1495{
1496
1497 FILEDESC_XLOCK(p->p_fd);
1498 if (p->p_fd->fd_refcnt > 1) {
1499 struct filedesc *tmp;
1500
1501 FILEDESC_XUNLOCK(p->p_fd);
1502 tmp = fdcopy(p->p_fd);
1503 fdfree(td);
1504 p->p_fd = tmp;
1505 } else
1506 FILEDESC_XUNLOCK(p->p_fd);
1507}
1508
1509/*
1510 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1511 * this is to ease callers, not catch errors.
1512 */
1513struct filedesc *
1514fdcopy(struct filedesc *fdp)
1515{
1516 struct filedesc *newfdp;
1517 int i;
1518
1519 /* Certain daemons might not have file descriptors. */
1520 if (fdp == NULL)
1521 return (NULL);
1522
1523 newfdp = fdinit(fdp);
1524 FILEDESC_SLOCK(fdp);
1525 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1526 FILEDESC_SUNLOCK(fdp);
1527 FILEDESC_XLOCK(newfdp);
1528 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1529 FILEDESC_XUNLOCK(newfdp);
1530 FILEDESC_SLOCK(fdp);
1531 }
1532 /* copy everything except kqueue descriptors */
1533 newfdp->fd_freefile = -1;
1534 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1535 if (fdisused(fdp, i) &&
1536 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1537 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1538 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1539 fhold(newfdp->fd_ofiles[i]);
1540 newfdp->fd_lastfile = i;
1541 } else {
1542 if (newfdp->fd_freefile == -1)
1543 newfdp->fd_freefile = i;
1544 }
1545 }
1546 FILEDESC_SUNLOCK(fdp);
1547 FILEDESC_XLOCK(newfdp);
1548 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1549 if (newfdp->fd_ofiles[i] != NULL)
1550 fdused(newfdp, i);
1551 FILEDESC_XUNLOCK(newfdp);
1552 FILEDESC_SLOCK(fdp);
1553 if (newfdp->fd_freefile == -1)
1554 newfdp->fd_freefile = i;
1555 newfdp->fd_cmask = fdp->fd_cmask;
1556 FILEDESC_SUNLOCK(fdp);
1557 return (newfdp);
1558}
1559
1560/*
1561 * Release a filedesc structure.
1562 */
1563void
1564fdfree(struct thread *td)
1565{
1566 struct filedesc *fdp;
1567 struct file **fpp;
1568 int i, locked;
1569 struct filedesc_to_leader *fdtol;
1570 struct file *fp;
1571 struct vnode *cdir, *jdir, *rdir, *vp;
1572 struct flock lf;
1573
1574 /* Certain daemons might not have file descriptors. */
1575 fdp = td->td_proc->p_fd;
1576 if (fdp == NULL)
1577 return;
1578
1579 /* Check for special need to clear POSIX style locks */
1580 fdtol = td->td_proc->p_fdtol;
1581 if (fdtol != NULL) {
1582 FILEDESC_XLOCK(fdp);
1583 KASSERT(fdtol->fdl_refcount > 0,
1584 ("filedesc_to_refcount botch: fdl_refcount=%d",
1585 fdtol->fdl_refcount));
1586 if (fdtol->fdl_refcount == 1 &&
1587 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1588 for (i = 0, fpp = fdp->fd_ofiles;
1589 i <= fdp->fd_lastfile;
1590 i++, fpp++) {
1591 if (*fpp == NULL ||
1592 (*fpp)->f_type != DTYPE_VNODE)
1593 continue;
1594 fp = *fpp;
1595 fhold(fp);
1596 FILEDESC_XUNLOCK(fdp);
1597 lf.l_whence = SEEK_SET;
1598 lf.l_start = 0;
1599 lf.l_len = 0;
1600 lf.l_type = F_UNLCK;
1601 vp = fp->f_vnode;
1602 locked = VFS_LOCK_GIANT(vp->v_mount);
1603 (void) VOP_ADVLOCK(vp,
1604 (caddr_t)td->td_proc->
1605 p_leader,
1606 F_UNLCK,
1607 &lf,
1608 F_POSIX);
1609 VFS_UNLOCK_GIANT(locked);
1610 FILEDESC_XLOCK(fdp);
1611 fdrop(fp, td);
1612 fpp = fdp->fd_ofiles + i;
1613 }
1614 }
1615 retry:
1616 if (fdtol->fdl_refcount == 1) {
1617 if (fdp->fd_holdleaderscount > 0 &&
1618 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1619 /*
1620 * close() or do_dup() has cleared a reference
1621 * in a shared file descriptor table.
1622 */
1623 fdp->fd_holdleaderswakeup = 1;
1624 sx_sleep(&fdp->fd_holdleaderscount,
1625 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1626 goto retry;
1627 }
1628 if (fdtol->fdl_holdcount > 0) {
1629 /*
1630 * Ensure that fdtol->fdl_leader remains
1631 * valid in closef().
1632 */
1633 fdtol->fdl_wakeup = 1;
1634 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1635 "fdlhold", 0);
1636 goto retry;
1637 }
1638 }
1639 fdtol->fdl_refcount--;
1640 if (fdtol->fdl_refcount == 0 &&
1641 fdtol->fdl_holdcount == 0) {
1642 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1643 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1644 } else
1645 fdtol = NULL;
1646 td->td_proc->p_fdtol = NULL;
1647 FILEDESC_XUNLOCK(fdp);
1648 if (fdtol != NULL)
1649 FREE(fdtol, M_FILEDESC_TO_LEADER);
1650 }
1651 FILEDESC_XLOCK(fdp);
1652 i = --fdp->fd_refcnt;
1653 FILEDESC_XUNLOCK(fdp);
1654 if (i > 0)
1655 return;
1656 /*
1657 * We are the last reference to the structure, so we can
1658 * safely assume it will not change out from under us.
1659 */
1660 fpp = fdp->fd_ofiles;
1661 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1662 if (*fpp)
1663 (void) closef(*fpp, td);
1664 }
1665 FILEDESC_XLOCK(fdp);
1666
1667 /* XXX This should happen earlier. */
1668 mtx_lock(&fdesc_mtx);
1669 td->td_proc->p_fd = NULL;
1670 mtx_unlock(&fdesc_mtx);
1671
1672 if (fdp->fd_nfiles > NDFILE)
1673 FREE(fdp->fd_ofiles, M_FILEDESC);
1674 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1675 FREE(fdp->fd_map, M_FILEDESC);
1676
1677 fdp->fd_nfiles = 0;
1678
1679 cdir = fdp->fd_cdir;
1680 fdp->fd_cdir = NULL;
1681 rdir = fdp->fd_rdir;
1682 fdp->fd_rdir = NULL;
1683 jdir = fdp->fd_jdir;
1684 fdp->fd_jdir = NULL;
1685 FILEDESC_XUNLOCK(fdp);
1686
1687 if (cdir) {
1688 locked = VFS_LOCK_GIANT(cdir->v_mount);
1689 vrele(cdir);
1690 VFS_UNLOCK_GIANT(locked);
1691 }
1692 if (rdir) {
1693 locked = VFS_LOCK_GIANT(rdir->v_mount);
1694 vrele(rdir);
1695 VFS_UNLOCK_GIANT(locked);
1696 }
1697 if (jdir) {
1698 locked = VFS_LOCK_GIANT(jdir->v_mount);
1699 vrele(jdir);
1700 VFS_UNLOCK_GIANT(locked);
1701 }
1702
1703 fddrop(fdp);
1704}
1705
1706/*
1707 * For setugid programs, we don't want to people to use that setugidness
1708 * to generate error messages which write to a file which otherwise would
1709 * otherwise be off-limits to the process. We check for filesystems where
1710 * the vnode can change out from under us after execve (like [lin]procfs).
1711 *
1712 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1713 * sufficient. We also don't check for setugidness since we know we are.
1714 */
1715static int
1716is_unsafe(struct file *fp)
1717{
1718 if (fp->f_type == DTYPE_VNODE) {
1719 struct vnode *vp = fp->f_vnode;
1720
1721 if ((vp->v_vflag & VV_PROCDEP) != 0)
1722 return (1);
1723 }
1724 return (0);
1725}
1726
1727/*
1728 * Make this setguid thing safe, if at all possible.
1729 */
1730void
1731setugidsafety(struct thread *td)
1732{
1733 struct filedesc *fdp;
1734 int i;
1735
1736 /* Certain daemons might not have file descriptors. */
1737 fdp = td->td_proc->p_fd;
1738 if (fdp == NULL)
1739 return;
1740
1741 /*
1742 * Note: fdp->fd_ofiles may be reallocated out from under us while
1743 * we are blocked in a close. Be careful!
1744 */
1745 FILEDESC_XLOCK(fdp);
1746 for (i = 0; i <= fdp->fd_lastfile; i++) {
1747 if (i > 2)
1748 break;
1749 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1750 struct file *fp;
1751
1752 knote_fdclose(td, i);
1753 /*
1754 * NULL-out descriptor prior to close to avoid
1755 * a race while close blocks.
1756 */
1757 fp = fdp->fd_ofiles[i];
1758 fdp->fd_ofiles[i] = NULL;
1759 fdp->fd_ofileflags[i] = 0;
1760 fdunused(fdp, i);
1761 FILEDESC_XUNLOCK(fdp);
1762 (void) closef(fp, td);
1763 FILEDESC_XLOCK(fdp);
1764 }
1765 }
1766 FILEDESC_XUNLOCK(fdp);
1767}
1768
1769/*
1770 * If a specific file object occupies a specific file descriptor, close the
1771 * file descriptor entry and drop a reference on the file object. This is a
1772 * convenience function to handle a subsequent error in a function that calls
1773 * falloc() that handles the race that another thread might have closed the
1774 * file descriptor out from under the thread creating the file object.
1775 */
1776void
1777fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1778{
1779
1780 FILEDESC_XLOCK(fdp);
1781 if (fdp->fd_ofiles[idx] == fp) {
1782 fdp->fd_ofiles[idx] = NULL;
1783 fdunused(fdp, idx);
1784 FILEDESC_XUNLOCK(fdp);
1785 fdrop(fp, td);
1786 } else
1787 FILEDESC_XUNLOCK(fdp);
1788}
1789
1790/*
1791 * Close any files on exec?
1792 */
1793void
1794fdcloseexec(struct thread *td)
1795{
1796 struct filedesc *fdp;
1797 int i;
1798
1799 /* Certain daemons might not have file descriptors. */
1800 fdp = td->td_proc->p_fd;
1801 if (fdp == NULL)
1802 return;
1803
1804 FILEDESC_XLOCK(fdp);
1805
1806 /*
1807 * We cannot cache fd_ofiles or fd_ofileflags since operations
1808 * may block and rip them out from under us.
1809 */
1810 for (i = 0; i <= fdp->fd_lastfile; i++) {
1811 if (fdp->fd_ofiles[i] != NULL &&
1812 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1813 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1814 struct file *fp;
1815
1816 knote_fdclose(td, i);
1817 /*
1818 * NULL-out descriptor prior to close to avoid
1819 * a race while close blocks.
1820 */
1821 fp = fdp->fd_ofiles[i];
1822 fdp->fd_ofiles[i] = NULL;
1823 fdp->fd_ofileflags[i] = 0;
1824 fdunused(fdp, i);
1825 if (fp->f_type == DTYPE_MQUEUE)
1826 mq_fdclose(td, i, fp);
1827 FILEDESC_XUNLOCK(fdp);
1828 (void) closef(fp, td);
1829 FILEDESC_XLOCK(fdp);
1830 }
1831 }
1832 FILEDESC_XUNLOCK(fdp);
1833}
1834
1835/*
1836 * It is unsafe for set[ug]id processes to be started with file
1837 * descriptors 0..2 closed, as these descriptors are given implicit
1838 * significance in the Standard C library. fdcheckstd() will create a
1839 * descriptor referencing /dev/null for each of stdin, stdout, and
1840 * stderr that is not already open.
1841 */
1842int
1843fdcheckstd(struct thread *td)
1844{
1845 struct filedesc *fdp;
1846 register_t retval, save;
1847 int i, error, devnull;
1848
1849 fdp = td->td_proc->p_fd;
1850 if (fdp == NULL)
1851 return (0);
1852 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1853 devnull = -1;
1854 error = 0;
1855 for (i = 0; i < 3; i++) {
1856 if (fdp->fd_ofiles[i] != NULL)
1857 continue;
1858 if (devnull < 0) {
1859 save = td->td_retval[0];
1860 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
1861 O_RDWR, 0);
1862 devnull = td->td_retval[0];
1863 KASSERT(devnull == i, ("oof, we didn't get our fd"));
1864 td->td_retval[0] = save;
1865 if (error)
1866 break;
1867 } else {
1868 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1869 if (error != 0)
1870 break;
1871 }
1872 }
1873 return (error);
1874}
1875
1876/*
1877 * Internal form of close. Decrement reference count on file structure.
1878 * Note: td may be NULL when closing a file that was being passed in a
1879 * message.
1880 *
1881 * XXXRW: Giant is not required for the caller, but often will be held; this
1882 * makes it moderately likely the Giant will be recursed in the VFS case.
1883 */
1884int
1885closef(struct file *fp, struct thread *td)
1886{
1887 struct vnode *vp;
1888 struct flock lf;
1889 struct filedesc_to_leader *fdtol;
1890 struct filedesc *fdp;
1891
1892 /*
1893 * POSIX record locking dictates that any close releases ALL
1894 * locks owned by this process. This is handled by setting
1895 * a flag in the unlock to free ONLY locks obeying POSIX
1896 * semantics, and not to free BSD-style file locks.
1897 * If the descriptor was in a message, POSIX-style locks
1898 * aren't passed with the descriptor, and the thread pointer
1899 * will be NULL. Callers should be careful only to pass a
1900 * NULL thread pointer when there really is no owning
1901 * context that might have locks, or the locks will be
1902 * leaked.
1903 */
1904 if (fp->f_type == DTYPE_VNODE && td != NULL) {
1905 int vfslocked;
1906
1907 vp = fp->f_vnode;
1908 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1909 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1910 lf.l_whence = SEEK_SET;
1911 lf.l_start = 0;
1912 lf.l_len = 0;
1913 lf.l_type = F_UNLCK;
1914 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1915 F_UNLCK, &lf, F_POSIX);
1916 }
1917 fdtol = td->td_proc->p_fdtol;
1918 if (fdtol != NULL) {
1919 /*
1920 * Handle special case where file descriptor table is
1921 * shared between multiple process leaders.
1922 */
1923 fdp = td->td_proc->p_fd;
1924 FILEDESC_XLOCK(fdp);
1925 for (fdtol = fdtol->fdl_next;
1926 fdtol != td->td_proc->p_fdtol;
1927 fdtol = fdtol->fdl_next) {
1928 if ((fdtol->fdl_leader->p_flag &
1929 P_ADVLOCK) == 0)
1930 continue;
1931 fdtol->fdl_holdcount++;
1932 FILEDESC_XUNLOCK(fdp);
1933 lf.l_whence = SEEK_SET;
1934 lf.l_start = 0;
1935 lf.l_len = 0;
1936 lf.l_type = F_UNLCK;
1937 vp = fp->f_vnode;
1938 (void) VOP_ADVLOCK(vp,
1939 (caddr_t)fdtol->fdl_leader,
1940 F_UNLCK, &lf, F_POSIX);
1941 FILEDESC_XLOCK(fdp);
1942 fdtol->fdl_holdcount--;
1943 if (fdtol->fdl_holdcount == 0 &&
1944 fdtol->fdl_wakeup != 0) {
1945 fdtol->fdl_wakeup = 0;
1946 wakeup(fdtol);
1947 }
1948 }
1949 FILEDESC_XUNLOCK(fdp);
1950 }
1951 VFS_UNLOCK_GIANT(vfslocked);
1952 }
1953 return (fdrop(fp, td));
1954}
1955
1956/*
1957 * Initialize the file pointer with the specified properties.
1958 *
1959 * The ops are set with release semantics to be certain that the flags, type,
1960 * and data are visible when ops is. This is to prevent ops methods from being
1961 * called with bad data.
1962 */
1963void
1964finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
1965{
1966 fp->f_data = data;
1967 fp->f_flag = flag;
1968 fp->f_type = type;
1969 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
1970}
1971
1972
1973/*
1974 * Extract the file pointer associated with the specified descriptor for the
1975 * current user process.
1976 *
1977 * If the descriptor doesn't exist, EBADF is returned.
1978 *
1979 * If the descriptor exists but doesn't match 'flags' then return EBADF for
1980 * read attempts and EINVAL for write attempts.
1981 *
1982 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1983 * It should be dropped with fdrop(). If it is not set, then the refcount
1984 * will not be bumped however the thread's filedesc struct will be returned
1985 * locked (for fgetsock).
1986 *
1987 * If an error occured the non-zero error is returned and *fpp is set to
1988 * NULL. Otherwise *fpp is set and zero is returned.
1989 */
1990static __inline int
1991_fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1992{
1993 struct filedesc *fdp;
1994 struct file *fp;
1995
1996 *fpp = NULL;
1997 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1998 return (EBADF);
1999 FILEDESC_SLOCK(fdp);
2000 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
2001 FILEDESC_SUNLOCK(fdp);
2002 return (EBADF);
2003 }
2004
2005 /*
2006 * FREAD and FWRITE failure return EBADF as per POSIX.
2007 *
2008 * Only one flag, or 0, may be specified.
2009 */
2010 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2011 FILEDESC_SUNLOCK(fdp);
2012 return (EBADF);
2013 }
2014 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2015 FILEDESC_SUNLOCK(fdp);
2016 return (EBADF);
2017 }
2018 if (hold) {
2019 fhold(fp);
2020 FILEDESC_SUNLOCK(fdp);
2021 }
2022 *fpp = fp;
2023 return (0);
2024}
2025
2026int
2027fget(struct thread *td, int fd, struct file **fpp)
2028{
2029
2030 return(_fget(td, fd, fpp, 0, 1));
2031}
2032
2033int
2034fget_read(struct thread *td, int fd, struct file **fpp)
2035{
2036
2037 return(_fget(td, fd, fpp, FREAD, 1));
2038}
2039
2040int
2041fget_write(struct thread *td, int fd, struct file **fpp)
2042{
2043
2044 return(_fget(td, fd, fpp, FWRITE, 1));
2045}
2046
2047/*
2048 * Like fget() but loads the underlying vnode, or returns an error if the
2049 * descriptor does not represent a vnode. Note that pipes use vnodes but
2050 * never have VM objects. The returned vnode will be vref()'d.
2051 *
2052 * XXX: what about the unused flags ?
2053 */
2054static __inline int
2055_fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2056{
2057 struct file *fp;
2058 int error;
2059
2060 *vpp = NULL;
2061 if ((error = _fget(td, fd, &fp, flags, 0)) != 0)
2062 return (error);
2063 if (fp->f_vnode == NULL) {
2064 error = EINVAL;
2065 } else {
2066 *vpp = fp->f_vnode;
2067 vref(*vpp);
2068 }
2069 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2070 return (error);
2071}
2072
2073int
2074fgetvp(struct thread *td, int fd, struct vnode **vpp)
2075{
2076
2077 return (_fgetvp(td, fd, vpp, 0));
2078}
2079
2080int
2081fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2082{
2083
2084 return (_fgetvp(td, fd, vpp, FREAD));
2085}
2086
2087#ifdef notyet
2088int
2089fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2090{
2091
2092 return (_fgetvp(td, fd, vpp, FWRITE));
2093}
2094#endif
2095
2096/*
2097 * Like fget() but loads the underlying socket, or returns an error if the
2098 * descriptor does not represent a socket.
2099 *
2100 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2101 * in the future.
2102 *
2103 * XXXRW: fgetsock() and fputsock() are deprecated, as consumers should rely
2104 * on their file descriptor reference to prevent the socket from being free'd
2105 * during use.
2106 */
2107int
2108fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2109{
2110 struct file *fp;
2111 int error;
2112
2113 *spp = NULL;
2114 if (fflagp != NULL)
2115 *fflagp = 0;
2116 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2117 return (error);
2118 if (fp->f_type != DTYPE_SOCKET) {
2119 error = ENOTSOCK;
2120 } else {
2121 *spp = fp->f_data;
2122 if (fflagp)
2123 *fflagp = fp->f_flag;
2124 SOCK_LOCK(*spp);
2125 soref(*spp);
2126 SOCK_UNLOCK(*spp);
2127 }
2128 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2129 return (error);
2130}
2131
2132/*
2133 * Drop the reference count on the socket and XXX release the SX lock in the
2134 * future. The last reference closes the socket.
2135 *
2136 * XXXRW: fputsock() is deprecated, see comment for fgetsock().
2137 */
2138void
2139fputsock(struct socket *so)
2140{
2141
2142 ACCEPT_LOCK();
2143 SOCK_LOCK(so);
2144 sorele(so);
2145}
2146
2147/*
2148 * Handle the last reference to a file being closed.
2149 */
2150int
2151_fdrop(struct file *fp, struct thread *td)
2152{
2153 int error;
2154
2155 error = 0;
2156 if (fp->f_count != 0)
2157 panic("fdrop: count %d", fp->f_count);
2158 if (fp->f_ops != &badfileops)
2159 error = fo_close(fp, td);
2160 atomic_subtract_int(&openfiles, 1);
2161 crfree(fp->f_cred);
2162 uma_zfree(file_zone, fp);
2163
2164 return (error);
2165}
2166
2167/*
2168 * Apply an advisory lock on a file descriptor.
2169 *
2170 * Just attempt to get a record lock of the requested type on the entire file
2171 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2172 */
2173#ifndef _SYS_SYSPROTO_H_
2174struct flock_args {
2175 int fd;
2176 int how;
2177};
2178#endif
2179/* ARGSUSED */
2180int
2181flock(struct thread *td, struct flock_args *uap)
2182{
2183 struct file *fp;
2184 struct vnode *vp;
2185 struct flock lf;
2186 int vfslocked;
2187 int error;
2188
2189 if ((error = fget(td, uap->fd, &fp)) != 0)
2190 return (error);
2191 if (fp->f_type != DTYPE_VNODE) {
2192 fdrop(fp, td);
2193 return (EOPNOTSUPP);
2194 }
2195
2196 vp = fp->f_vnode;
2197 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2198 lf.l_whence = SEEK_SET;
2199 lf.l_start = 0;
2200 lf.l_len = 0;
2201 if (uap->how & LOCK_UN) {
2202 lf.l_type = F_UNLCK;
2203 atomic_clear_int(&fp->f_flag, FHASLOCK);
2204 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2205 goto done2;
2206 }
2207 if (uap->how & LOCK_EX)
2208 lf.l_type = F_WRLCK;
2209 else if (uap->how & LOCK_SH)
2210 lf.l_type = F_RDLCK;
2211 else {
2212 error = EBADF;
2213 goto done2;
2214 }
2215 atomic_set_int(&fp->f_flag, FHASLOCK);
2216 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2217 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2218done2:
2219 fdrop(fp, td);
2220 VFS_UNLOCK_GIANT(vfslocked);
2221 return (error);
2222}
2223/*
2224 * Duplicate the specified descriptor to a free descriptor.
2225 */
2226int
2227dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2228{
2229 struct file *wfp;
2230 struct file *fp;
2231
2232 /*
2233 * If the to-be-dup'd fd number is greater than the allowed number
2234 * of file descriptors, or the fd to be dup'd has already been
2235 * closed, then reject.
2236 */
2237 FILEDESC_XLOCK(fdp);
2238 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2239 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2240 FILEDESC_XUNLOCK(fdp);
2241 return (EBADF);
2242 }
2243
2244 /*
2245 * There are two cases of interest here.
2246 *
2247 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2248 *
2249 * For ENXIO steal away the file structure from (dfd) and store it in
2250 * (indx). (dfd) is effectively closed by this operation.
2251 *
2252 * Any other error code is just returned.
2253 */
2254 switch (error) {
2255 case ENODEV:
2256 /*
2257 * Check that the mode the file is being opened for is a
2258 * subset of the mode of the existing descriptor.
2259 */
2260 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2261 FILEDESC_XUNLOCK(fdp);
2262 return (EACCES);
2263 }
2264 fp = fdp->fd_ofiles[indx];
2265 fdp->fd_ofiles[indx] = wfp;
2266 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2267 if (fp == NULL)
2268 fdused(fdp, indx);
2269 fhold(wfp);
2270 FILEDESC_XUNLOCK(fdp);
2271 if (fp != NULL)
2272 /*
2273 * We now own the reference to fp that the ofiles[]
2274 * array used to own. Release it.
2275 */
2276 fdrop(fp, td);
2277 return (0);
2278
2279 case ENXIO:
2280 /*
2281 * Steal away the file pointer from dfd and stuff it into indx.
2282 */
2283 fp = fdp->fd_ofiles[indx];
2284 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2285 fdp->fd_ofiles[dfd] = NULL;
2286 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2287 fdp->fd_ofileflags[dfd] = 0;
2288 fdunused(fdp, dfd);
2289 if (fp == NULL)
2290 fdused(fdp, indx);
2291 FILEDESC_XUNLOCK(fdp);
2292
2293 /*
2294 * We now own the reference to fp that the ofiles[] array
2295 * used to own. Release it.
2296 */
2297 if (fp != NULL)
2298 fdrop(fp, td);
2299 return (0);
2300
2301 default:
2302 FILEDESC_XUNLOCK(fdp);
2303 return (error);
2304 }
2305 /* NOTREACHED */
2306}
2307
2308/*
2309 * Scan all active processes to see if any of them have a current or root
2310 * directory of `olddp'. If so, replace them with the new mount point.
2311 */
2312void
2313mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2314{
2315 struct filedesc *fdp;
2316 struct proc *p;
2317 int nrele;
2318
2319 if (vrefcnt(olddp) == 1)
2320 return;
2321 sx_slock(&allproc_lock);
2322 FOREACH_PROC_IN_SYSTEM(p) {
2323 fdp = fdhold(p);
2324 if (fdp == NULL)
2325 continue;
2326 nrele = 0;
2327 FILEDESC_XLOCK(fdp);
2328 if (fdp->fd_cdir == olddp) {
2329 vref(newdp);
2330 fdp->fd_cdir = newdp;
2331 nrele++;
2332 }
2333 if (fdp->fd_rdir == olddp) {
2334 vref(newdp);
2335 fdp->fd_rdir = newdp;
2336 nrele++;
2337 }
2338 FILEDESC_XUNLOCK(fdp);
2339 fddrop(fdp);
2340 while (nrele--)
2341 vrele(olddp);
2342 }
2343 sx_sunlock(&allproc_lock);
2344 if (rootvnode == olddp) {
2345 vrele(rootvnode);
2346 vref(newdp);
2347 rootvnode = newdp;
2348 }
2349}
2350
2351struct filedesc_to_leader *
2352filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2353{
2354 struct filedesc_to_leader *fdtol;
2355
2356 MALLOC(fdtol, struct filedesc_to_leader *,
2357 sizeof(struct filedesc_to_leader),
2358 M_FILEDESC_TO_LEADER,
2359 M_WAITOK);
2360 fdtol->fdl_refcount = 1;
2361 fdtol->fdl_holdcount = 0;
2362 fdtol->fdl_wakeup = 0;
2363 fdtol->fdl_leader = leader;
2364 if (old != NULL) {
2365 FILEDESC_XLOCK(fdp);
2366 fdtol->fdl_next = old->fdl_next;
2367 fdtol->fdl_prev = old;
2368 old->fdl_next = fdtol;
2369 fdtol->fdl_next->fdl_prev = fdtol;
2370 FILEDESC_XUNLOCK(fdp);
2371 } else {
2372 fdtol->fdl_next = fdtol;
2373 fdtol->fdl_prev = fdtol;
2374 }
2375 return (fdtol);
2376}
2377
2378/*
2379 * Get file structures globally.
2380 */
2381static int
2382sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2383{
2384 struct xfile xf;
2385 struct filedesc *fdp;
2386 struct file *fp;
2387 struct proc *p;
2388 int error, n;
2389
2390 error = sysctl_wire_old_buffer(req, 0);
2391 if (error != 0)
2392 return (error);
2393 if (req->oldptr == NULL) {
2394 n = 0;
2395 sx_slock(&allproc_lock);
2396 FOREACH_PROC_IN_SYSTEM(p) {
2397 if (p->p_state == PRS_NEW)
2398 continue;
2399 fdp = fdhold(p);
2400 if (fdp == NULL)
2401 continue;
2402 /* overestimates sparse tables. */
2403 if (fdp->fd_lastfile > 0)
2404 n += fdp->fd_lastfile;
2405 fddrop(fdp);
2406 }
2407 sx_sunlock(&allproc_lock);
2408 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2409 }
2410 error = 0;
2411 bzero(&xf, sizeof(xf));
2412 xf.xf_size = sizeof(xf);
2413 sx_slock(&allproc_lock);
2414 FOREACH_PROC_IN_SYSTEM(p) {
2415 if (p->p_state == PRS_NEW)
2416 continue;
2417 PROC_LOCK(p);
2418 if (p_cansee(req->td, p) != 0) {
2419 PROC_UNLOCK(p);
2420 continue;
2421 }
2422 xf.xf_pid = p->p_pid;
2423 xf.xf_uid = p->p_ucred->cr_uid;
2424 PROC_UNLOCK(p);
2425 fdp = fdhold(p);
2426 if (fdp == NULL)
2427 continue;
2428 FILEDESC_SLOCK(fdp);
2429 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2430 if ((fp = fdp->fd_ofiles[n]) == NULL)
2431 continue;
2432 xf.xf_fd = n;
2433 xf.xf_file = fp;
2434 xf.xf_data = fp->f_data;
2435 xf.xf_vnode = fp->f_vnode;
2436 xf.xf_type = fp->f_type;
2437 xf.xf_count = fp->f_count;
2438 xf.xf_msgcount = 0;
2439 xf.xf_offset = fp->f_offset;
2440 xf.xf_flag = fp->f_flag;
2441 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2442 if (error)
2443 break;
2444 }
2445 FILEDESC_SUNLOCK(fdp);
2446 fddrop(fdp);
2447 if (error)
2448 break;
2449 }
2450 sx_sunlock(&allproc_lock);
2451 return (error);
2452}
2453
2454SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2455 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2456
2457static int
2458export_vnode_for_sysctl(struct vnode *vp, int type,
2459 struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req)
2460{
2461 int error;
2462 char *fullpath, *freepath;
2463 int vfslocked;
2464
2465 bzero(kif, sizeof(*kif));
2466 kif->kf_structsize = sizeof(*kif);
2467
2468 vref(vp);
2469 kif->kf_fd = type;
2470 kif->kf_type = KF_TYPE_VNODE;
2471 /* This function only handles directories. */
2472 KASSERT(vp->v_type == VDIR, ("export_vnode_for_sysctl: vnode not directory"));
2473 kif->kf_vnode_type = KF_VTYPE_VDIR;
2474
2475 /*
2476 * This is not a true file descriptor, so we set a bogus refcount
2477 * and offset to indicate these fields should be ignored.
2478 */
2479 kif->kf_ref_count = -1;
2480 kif->kf_offset = -1;
2481
2482 freepath = NULL;
2483 fullpath = "-";
2484 FILEDESC_SUNLOCK(fdp);
2485 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2486 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2487 vn_fullpath(curthread, vp, &fullpath, &freepath);
2488 vput(vp);
2489 VFS_UNLOCK_GIANT(vfslocked);
2490 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2491 if (freepath != NULL)
2492 free(freepath, M_TEMP);
2493 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2494 FILEDESC_SLOCK(fdp);
2495 return (error);
2496}
2497
2498/*
2499 * Get per-process file descriptors for use by procstat(1), et al.
2500 */
2501static int
2502sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
2503{
2504 char *fullpath, *freepath;
2505 struct kinfo_file *kif;
2506 struct filedesc *fdp;
2507 int error, i, *name;
2508 struct socket *so;
2509 struct vnode *vp;
2510 struct file *fp;
2511 struct proc *p;
2512 int vfslocked;
2513
2514 name = (int *)arg1;
2515 if ((p = pfind((pid_t)name[0])) == NULL)
2516 return (ESRCH);
2517 if ((error = p_candebug(curthread, p))) {
2518 PROC_UNLOCK(p);
2519 return (error);
2520 }
2521 fdp = fdhold(p);
2522 PROC_UNLOCK(p);
2523 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2524 FILEDESC_SLOCK(fdp);
2525 if (fdp->fd_cdir != NULL)
2526 export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2527 fdp, req);
2528 if (fdp->fd_rdir != NULL)
2529 export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2530 fdp, req);
2531 if (fdp->fd_jdir != NULL)
2532 export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2533 fdp, req);
2534 for (i = 0; i < fdp->fd_nfiles; i++) {
2535 if ((fp = fdp->fd_ofiles[i]) == NULL)
2536 continue;
2537 bzero(kif, sizeof(*kif));
2538 kif->kf_structsize = sizeof(*kif);
2539 vp = NULL;
2540 so = NULL;
2541 kif->kf_fd = i;
2542 switch (fp->f_type) {
2543 case DTYPE_VNODE:
2544 kif->kf_type = KF_TYPE_VNODE;
2545 vp = fp->f_vnode;
2546 break;
2547
2548 case DTYPE_SOCKET:
2549 kif->kf_type = KF_TYPE_SOCKET;
2550 so = fp->f_data;
2551 break;
2552
2553 case DTYPE_PIPE:
2554 kif->kf_type = KF_TYPE_PIPE;
2555 break;
2556
2557 case DTYPE_FIFO:
2558 kif->kf_type = KF_TYPE_FIFO;
2559 vp = fp->f_vnode;
2560 vref(vp);
2561 break;
2562
2563 case DTYPE_KQUEUE:
2564 kif->kf_type = KF_TYPE_KQUEUE;
2565 break;
2566
2567 case DTYPE_CRYPTO:
2568 kif->kf_type = KF_TYPE_CRYPTO;
2569 break;
2570
2571 case DTYPE_MQUEUE:
2572 kif->kf_type = KF_TYPE_MQUEUE;
2573 break;
2574
2575 case DTYPE_SHM:
2576 kif->kf_type = KF_TYPE_SHM;
2577 break;
2578
2579 default:
2580 kif->kf_type = KF_TYPE_UNKNOWN;
2581 break;
2582 }
2583 kif->kf_ref_count = fp->f_count;
2584 if (fp->f_flag & FREAD)
2585 kif->kf_flags |= KF_FLAG_READ;
2586 if (fp->f_flag & FWRITE)
2587 kif->kf_flags |= KF_FLAG_WRITE;
2588 if (fp->f_flag & FAPPEND)
2589 kif->kf_flags |= KF_FLAG_APPEND;
2590 if (fp->f_flag & FASYNC)
2591 kif->kf_flags |= KF_FLAG_ASYNC;
2592 if (fp->f_flag & FFSYNC)
2593 kif->kf_flags |= KF_FLAG_FSYNC;
2594 if (fp->f_flag & FNONBLOCK)
2595 kif->kf_flags |= KF_FLAG_NONBLOCK;
2596 if (fp->f_flag & O_DIRECT)
2597 kif->kf_flags |= KF_FLAG_DIRECT;
2598 if (fp->f_flag & FHASLOCK)
2599 kif->kf_flags |= KF_FLAG_HASLOCK;
2600 kif->kf_offset = fp->f_offset;
2601 if (vp != NULL) {
2602 vref(vp);
2603 switch (vp->v_type) {
2604 case VNON:
2605 kif->kf_vnode_type = KF_VTYPE_VNON;
2606 break;
2607 case VREG:
2608 kif->kf_vnode_type = KF_VTYPE_VREG;
2609 break;
2610 case VDIR:
2611 kif->kf_vnode_type = KF_VTYPE_VDIR;
2612 break;
2613 case VBLK:
2614 kif->kf_vnode_type = KF_VTYPE_VBLK;
2615 break;
2616 case VCHR:
2617 kif->kf_vnode_type = KF_VTYPE_VCHR;
2618 break;
2619 case VLNK:
2620 kif->kf_vnode_type = KF_VTYPE_VLNK;
2621 break;
2622 case VSOCK:
2623 kif->kf_vnode_type = KF_VTYPE_VSOCK;
2624 break;
2625 case VFIFO:
2626 kif->kf_vnode_type = KF_VTYPE_VFIFO;
2627 break;
2628 case VBAD:
2629 kif->kf_vnode_type = KF_VTYPE_VBAD;
2630 break;
2631 default:
2632 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2633 break;
2634 }
2635 /*
2636 * It is OK to drop the filedesc lock here as we will
2637 * re-validate and re-evaluate its properties when
2638 * the loop continues.
2639 */
2640 freepath = NULL;
2641 fullpath = "-";
2642 FILEDESC_SUNLOCK(fdp);
2643 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2644 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2645 vn_fullpath(curthread, vp, &fullpath, &freepath);
2646 vput(vp);
2647 VFS_UNLOCK_GIANT(vfslocked);
2648 strlcpy(kif->kf_path, fullpath,
2649 sizeof(kif->kf_path));
2650 if (freepath != NULL)
2651 free(freepath, M_TEMP);
2652 FILEDESC_SLOCK(fdp);
2653 }
2654 if (so != NULL) {
2655 struct sockaddr *sa;
2656
2657 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
2658 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
2659 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
2660 free(sa, M_SONAME);
2661 }
2662 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
2663 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
2664 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
2665 free(sa, M_SONAME);
2666 }
2667 kif->kf_sock_domain =
2668 so->so_proto->pr_domain->dom_family;
2669 kif->kf_sock_type = so->so_type;
2670 kif->kf_sock_protocol = so->so_proto->pr_protocol;
2671 }
2672 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2673 if (error)
2674 break;
2675 }
2676 FILEDESC_SUNLOCK(fdp);
2677 fddrop(fdp);
2678 free(kif, M_TEMP);
2679 return (0);
2680}
2681
2682static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
2683 sysctl_kern_proc_filedesc, "Process filedesc entries");
2684
2685#ifdef DDB
2686/*
2687 * For the purposes of debugging, generate a human-readable string for the
2688 * file type.
2689 */
2690static const char *
2691file_type_to_name(short type)
2692{
2693
2694 switch (type) {
2695 case 0:
2696 return ("zero");
2697 case DTYPE_VNODE:
2698 return ("vnod");
2699 case DTYPE_SOCKET:
2700 return ("sock");
2701 case DTYPE_PIPE:
2702 return ("pipe");
2703 case DTYPE_FIFO:
2704 return ("fifo");
2705 case DTYPE_KQUEUE:
2706 return ("kque");
2707 case DTYPE_CRYPTO:
2708 return ("crpt");
2709 case DTYPE_MQUEUE:
2710 return ("mque");
2711 case DTYPE_SHM:
2712 return ("shm");
2713 default:
2714 return ("unkn");
2715 }
2716}
2717
2718/*
2719 * For the purposes of debugging, identify a process (if any, perhaps one of
2720 * many) that references the passed file in its file descriptor array. Return
2721 * NULL if none.
2722 */
2723static struct proc *
2724file_to_first_proc(struct file *fp)
2725{
2726 struct filedesc *fdp;
2727 struct proc *p;
2728 int n;
2729
2730 FOREACH_PROC_IN_SYSTEM(p) {
2731 if (p->p_state == PRS_NEW)
2732 continue;
2733 fdp = p->p_fd;
2734 if (fdp == NULL)
2735 continue;
2736 for (n = 0; n < fdp->fd_nfiles; n++) {
2737 if (fp == fdp->fd_ofiles[n])
2738 return (p);
2739 }
2740 }
2741 return (NULL);
2742}
2743
2744static void
2745db_print_file(struct file *fp, int header)
2746{
2747 struct proc *p;
2748
2749 if (header)
2750 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
2751 "File", "Type", "Data", "Flag", "GCFl", "Count",
2752 "MCount", "Vnode", "FPID", "FCmd");
2753 p = file_to_first_proc(fp);
2754 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2755 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2756 0, fp->f_count, 0, fp->f_vnode,
2757 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2758}
2759
2760DB_SHOW_COMMAND(file, db_show_file)
2761{
2762 struct file *fp;
2763
2764 if (!have_addr) {
2765 db_printf("usage: show file <addr>\n");
2766 return;
2767 }
2768 fp = (struct file *)addr;
2769 db_print_file(fp, 1);
2770}
2771
2772DB_SHOW_COMMAND(files, db_show_files)
2773{
2774 struct filedesc *fdp;
2775 struct file *fp;
2776 struct proc *p;
2777 int header;
2778 int n;
2779
2780 header = 1;
2781 FOREACH_PROC_IN_SYSTEM(p) {
2782 if (p->p_state == PRS_NEW)
2783 continue;
2784 if ((fdp = p->p_fd) == NULL)
2785 continue;
2786 for (n = 0; n < fdp->fd_nfiles; ++n) {
2787 if ((fp = fdp->fd_ofiles[n]) == NULL)
2788 continue;
2789 db_print_file(fp, header);
2790 header = 0;
2791 }
2792 }
2793}
2794#endif
2795
2796SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2797 &maxfilesperproc, 0, "Maximum files allowed open per process");
2798
2799SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2800 &maxfiles, 0, "Maximum number of files");
2801
2802SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2803 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
2804
2805/* ARGSUSED*/
2806static void
2807filelistinit(void *dummy)
2808{
2809
2810 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2811 NULL, NULL, UMA_ALIGN_PTR, 0);
2812 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2813 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2814}
2815SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2816
2817/*-------------------------------------------------------------------*/
2818
2819static int
2820badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2821{
2822
2823 return (EBADF);
2824}
2825
2826static int
2827badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td)
2828{
2829
2830 return (EINVAL);
2831}
2832
2833static int
2834badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2835{
2836
2837 return (EBADF);
2838}
2839
2840static int
2841badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2842{
2843
2844 return (0);
2845}
2846
2847static int
2848badfo_kqfilter(struct file *fp, struct knote *kn)
2849{
2850
2851 return (EBADF);
2852}
2853
2854static int
2855badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2856{
2857
2858 return (EBADF);
2859}
2860
2861static int
2862badfo_close(struct file *fp, struct thread *td)
2863{
2864
2865 return (EBADF);
2866}
2867
2868struct fileops badfileops = {
2869 .fo_read = badfo_readwrite,
2870 .fo_write = badfo_readwrite,
2871 .fo_truncate = badfo_truncate,
2872 .fo_ioctl = badfo_ioctl,
2873 .fo_poll = badfo_poll,
2874 .fo_kqfilter = badfo_kqfilter,
2875 .fo_stat = badfo_stat,
2876 .fo_close = badfo_close,
2877};
2878
2879
2880/*-------------------------------------------------------------------*/
2881
2882/*
2883 * File Descriptor pseudo-device driver (/dev/fd/).
2884 *
2885 * Opening minor device N dup()s the file (if any) connected to file
2886 * descriptor N belonging to the calling process. Note that this driver
2887 * consists of only the ``open()'' routine, because all subsequent
2888 * references to this file will be direct to the other driver.
2889 *
2890 * XXX: we could give this one a cloning event handler if necessary.
2891 */
2892
2893/* ARGSUSED */
2894static int
2895fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2896{
2897
2898 /*
2899 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2900 * the file descriptor being sought for duplication. The error
2901 * return ensures that the vnode for this device will be released
2902 * by vn_open. Open will detect this special error and take the
2903 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2904 * will simply report the error.
2905 */
2906 td->td_dupfd = dev2unit(dev);
2907 return (ENODEV);
2908}
2909
2910static struct cdevsw fildesc_cdevsw = {
2911 .d_version = D_VERSION,
2912 .d_flags = D_NEEDGIANT,
2913 .d_open = fdopen,
2914 .d_name = "FD",
2915};
2916
2917static void
2918fildesc_drvinit(void *unused)
2919{
2920 struct cdev *dev;
2921
2922 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2923 make_dev_alias(dev, "stdin");
2924 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2925 make_dev_alias(dev, "stdout");
2926 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2927 make_dev_alias(dev, "stderr");
2928}
2929
2930SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
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 176471 2008-02-23 01:01:49Z des $");
39
40#include "opt_compat.h"
41#include "opt_ddb.h"
42#include "opt_ktrace.h"
43
44#include <sys/param.h>
45#include <sys/systm.h>
46
47#include <sys/conf.h>
48#include <sys/domain.h>
49#include <sys/fcntl.h>
50#include <sys/file.h>
51#include <sys/filedesc.h>
52#include <sys/filio.h>
53#include <sys/jail.h>
54#include <sys/kernel.h>
55#include <sys/limits.h>
56#include <sys/lock.h>
57#include <sys/malloc.h>
58#include <sys/mount.h>
59#include <sys/mqueue.h>
60#include <sys/mutex.h>
61#include <sys/namei.h>
62#include <sys/priv.h>
63#include <sys/proc.h>
64#include <sys/protosw.h>
65#include <sys/resourcevar.h>
66#include <sys/signalvar.h>
67#include <sys/socketvar.h>
68#include <sys/stat.h>
69#include <sys/sx.h>
70#include <sys/syscallsubr.h>
71#include <sys/sysctl.h>
72#include <sys/sysproto.h>
73#include <sys/unistd.h>
74#include <sys/user.h>
75#include <sys/vnode.h>
76#ifdef KTRACE
77#include <sys/ktrace.h>
78#endif
79
80#include <security/audit/audit.h>
81
82#include <vm/uma.h>
83
84#include <ddb/ddb.h>
85
86static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
87static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
88 "file desc to leader structures");
89static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
90
91static uma_zone_t file_zone;
92
93
94/* How to treat 'new' parameter when allocating a fd for do_dup(). */
95enum dup_type { DUP_VARIABLE, DUP_FIXED };
96
97static int do_dup(struct thread *td, enum dup_type type, int old, int new,
98 register_t *retval);
99static int fd_first_free(struct filedesc *, int, int);
100static int fd_last_used(struct filedesc *, int, int);
101static void fdgrowtable(struct filedesc *, int);
102static void fdunused(struct filedesc *fdp, int fd);
103static void fdused(struct filedesc *fdp, int fd);
104
105/*
106 * A process is initially started out with NDFILE descriptors stored within
107 * this structure, selected to be enough for typical applications based on
108 * the historical limit of 20 open files (and the usage of descriptors by
109 * shells). If these descriptors are exhausted, a larger descriptor table
110 * may be allocated, up to a process' resource limit; the internal arrays
111 * are then unused.
112 */
113#define NDFILE 20
114#define NDSLOTSIZE sizeof(NDSLOTTYPE)
115#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
116#define NDSLOT(x) ((x) / NDENTRIES)
117#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
118#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
119
120/*
121 * Storage required per open file descriptor.
122 */
123#define OFILESIZE (sizeof(struct file *) + sizeof(char))
124
125/*
126 * Basic allocation of descriptors:
127 * one of the above, plus arrays for NDFILE descriptors.
128 */
129struct filedesc0 {
130 struct filedesc fd_fd;
131 /*
132 * These arrays are used when the number of open files is
133 * <= NDFILE, and are then pointed to by the pointers above.
134 */
135 struct file *fd_dfiles[NDFILE];
136 char fd_dfileflags[NDFILE];
137 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
138};
139
140/*
141 * Descriptor management.
142 */
143volatile int openfiles; /* actual number of open files */
144struct mtx sigio_lock; /* mtx to protect pointers to sigio */
145void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
146
147/* A mutex to protect the association between a proc and filedesc. */
148static struct mtx fdesc_mtx;
149
150/*
151 * Find the first zero bit in the given bitmap, starting at low and not
152 * exceeding size - 1.
153 */
154static int
155fd_first_free(struct filedesc *fdp, int low, int size)
156{
157 NDSLOTTYPE *map = fdp->fd_map;
158 NDSLOTTYPE mask;
159 int off, maxoff;
160
161 if (low >= size)
162 return (low);
163
164 off = NDSLOT(low);
165 if (low % NDENTRIES) {
166 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
167 if ((mask &= ~map[off]) != 0UL)
168 return (off * NDENTRIES + ffsl(mask) - 1);
169 ++off;
170 }
171 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
172 if (map[off] != ~0UL)
173 return (off * NDENTRIES + ffsl(~map[off]) - 1);
174 return (size);
175}
176
177/*
178 * Find the highest non-zero bit in the given bitmap, starting at low and
179 * not exceeding size - 1.
180 */
181static int
182fd_last_used(struct filedesc *fdp, int low, int size)
183{
184 NDSLOTTYPE *map = fdp->fd_map;
185 NDSLOTTYPE mask;
186 int off, minoff;
187
188 if (low >= size)
189 return (-1);
190
191 off = NDSLOT(size);
192 if (size % NDENTRIES) {
193 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
194 if ((mask &= map[off]) != 0)
195 return (off * NDENTRIES + flsl(mask) - 1);
196 --off;
197 }
198 for (minoff = NDSLOT(low); off >= minoff; --off)
199 if (map[off] != 0)
200 return (off * NDENTRIES + flsl(map[off]) - 1);
201 return (low - 1);
202}
203
204static int
205fdisused(struct filedesc *fdp, int fd)
206{
207 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
208 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
209 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
210}
211
212/*
213 * Mark a file descriptor as used.
214 */
215static void
216fdused(struct filedesc *fdp, int fd)
217{
218
219 FILEDESC_XLOCK_ASSERT(fdp);
220 KASSERT(!fdisused(fdp, fd),
221 ("fd already used"));
222
223 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
224 if (fd > fdp->fd_lastfile)
225 fdp->fd_lastfile = fd;
226 if (fd == fdp->fd_freefile)
227 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
228}
229
230/*
231 * Mark a file descriptor as unused.
232 */
233static void
234fdunused(struct filedesc *fdp, int fd)
235{
236
237 FILEDESC_XLOCK_ASSERT(fdp);
238 KASSERT(fdisused(fdp, fd),
239 ("fd is already unused"));
240 KASSERT(fdp->fd_ofiles[fd] == NULL,
241 ("fd is still in use"));
242
243 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
244 if (fd < fdp->fd_freefile)
245 fdp->fd_freefile = fd;
246 if (fd == fdp->fd_lastfile)
247 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
248}
249
250/*
251 * System calls on descriptors.
252 */
253#ifndef _SYS_SYSPROTO_H_
254struct getdtablesize_args {
255 int dummy;
256};
257#endif
258/* ARGSUSED */
259int
260getdtablesize(struct thread *td, struct getdtablesize_args *uap)
261{
262 struct proc *p = td->td_proc;
263
264 PROC_LOCK(p);
265 td->td_retval[0] =
266 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
267 PROC_UNLOCK(p);
268 return (0);
269}
270
271/*
272 * Duplicate a file descriptor to a particular value.
273 *
274 * Note: keep in mind that a potential race condition exists when closing
275 * descriptors from a shared descriptor table (via rfork).
276 */
277#ifndef _SYS_SYSPROTO_H_
278struct dup2_args {
279 u_int from;
280 u_int to;
281};
282#endif
283/* ARGSUSED */
284int
285dup2(struct thread *td, struct dup2_args *uap)
286{
287
288 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
289 td->td_retval));
290}
291
292/*
293 * Duplicate a file descriptor.
294 */
295#ifndef _SYS_SYSPROTO_H_
296struct dup_args {
297 u_int fd;
298};
299#endif
300/* ARGSUSED */
301int
302dup(struct thread *td, struct dup_args *uap)
303{
304
305 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
306}
307
308/*
309 * The file control system call.
310 */
311#ifndef _SYS_SYSPROTO_H_
312struct fcntl_args {
313 int fd;
314 int cmd;
315 long arg;
316};
317#endif
318/* ARGSUSED */
319int
320fcntl(struct thread *td, struct fcntl_args *uap)
321{
322 struct flock fl;
323 intptr_t arg;
324 int error;
325
326 error = 0;
327 switch (uap->cmd) {
328 case F_GETLK:
329 case F_SETLK:
330 case F_SETLKW:
331 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
332 arg = (intptr_t)&fl;
333 break;
334 default:
335 arg = uap->arg;
336 break;
337 }
338 if (error)
339 return (error);
340 error = kern_fcntl(td, uap->fd, uap->cmd, arg);
341 if (error)
342 return (error);
343 if (uap->cmd == F_GETLK)
344 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
345 return (error);
346}
347
348static inline struct file *
349fdtofp(int fd, struct filedesc *fdp)
350{
351 struct file *fp;
352
353 FILEDESC_LOCK_ASSERT(fdp);
354 if ((unsigned)fd >= fdp->fd_nfiles ||
355 (fp = fdp->fd_ofiles[fd]) == NULL)
356 return (NULL);
357 return (fp);
358}
359
360int
361kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
362{
363 struct filedesc *fdp;
364 struct flock *flp;
365 struct file *fp;
366 struct proc *p;
367 char *pop;
368 struct vnode *vp;
369 u_int newmin;
370 int error, flg, tmp;
371 int vfslocked;
372
373 vfslocked = 0;
374 error = 0;
375 flg = F_POSIX;
376 p = td->td_proc;
377 fdp = p->p_fd;
378
379 switch (cmd) {
380 case F_DUPFD:
381 FILEDESC_SLOCK(fdp);
382 if ((fp = fdtofp(fd, fdp)) == NULL) {
383 FILEDESC_SUNLOCK(fdp);
384 error = EBADF;
385 break;
386 }
387 FILEDESC_SUNLOCK(fdp);
388 newmin = arg;
389 PROC_LOCK(p);
390 if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
391 newmin >= maxfilesperproc) {
392 PROC_UNLOCK(p);
393 error = EINVAL;
394 break;
395 }
396 PROC_UNLOCK(p);
397 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
398 break;
399
400 case F_GETFD:
401 FILEDESC_SLOCK(fdp);
402 if ((fp = fdtofp(fd, fdp)) == NULL) {
403 FILEDESC_SUNLOCK(fdp);
404 error = EBADF;
405 break;
406 }
407 pop = &fdp->fd_ofileflags[fd];
408 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
409 FILEDESC_SUNLOCK(fdp);
410 break;
411
412 case F_SETFD:
413 FILEDESC_XLOCK(fdp);
414 if ((fp = fdtofp(fd, fdp)) == NULL) {
415 FILEDESC_XUNLOCK(fdp);
416 error = EBADF;
417 break;
418 }
419 pop = &fdp->fd_ofileflags[fd];
420 *pop = (*pop &~ UF_EXCLOSE) |
421 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
422 FILEDESC_XUNLOCK(fdp);
423 break;
424
425 case F_GETFL:
426 FILEDESC_SLOCK(fdp);
427 if ((fp = fdtofp(fd, fdp)) == NULL) {
428 FILEDESC_SUNLOCK(fdp);
429 error = EBADF;
430 break;
431 }
432 td->td_retval[0] = OFLAGS(fp->f_flag);
433 FILEDESC_SUNLOCK(fdp);
434 break;
435
436 case F_SETFL:
437 FILEDESC_SLOCK(fdp);
438 if ((fp = fdtofp(fd, fdp)) == NULL) {
439 FILEDESC_SUNLOCK(fdp);
440 error = EBADF;
441 break;
442 }
443 fhold(fp);
444 FILEDESC_SUNLOCK(fdp);
445 do {
446 tmp = flg = fp->f_flag;
447 tmp &= ~FCNTLFLAGS;
448 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
449 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
450 tmp = fp->f_flag & FNONBLOCK;
451 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
452 if (error) {
453 fdrop(fp, td);
454 break;
455 }
456 tmp = fp->f_flag & FASYNC;
457 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
458 if (error == 0) {
459 fdrop(fp, td);
460 break;
461 }
462 atomic_clear_int(&fp->f_flag, FNONBLOCK);
463 tmp = 0;
464 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
465 fdrop(fp, td);
466 break;
467
468 case F_GETOWN:
469 FILEDESC_SLOCK(fdp);
470 if ((fp = fdtofp(fd, fdp)) == NULL) {
471 FILEDESC_SUNLOCK(fdp);
472 error = EBADF;
473 break;
474 }
475 fhold(fp);
476 FILEDESC_SUNLOCK(fdp);
477 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
478 if (error == 0)
479 td->td_retval[0] = tmp;
480 fdrop(fp, td);
481 break;
482
483 case F_SETOWN:
484 FILEDESC_SLOCK(fdp);
485 if ((fp = fdtofp(fd, fdp)) == NULL) {
486 FILEDESC_SUNLOCK(fdp);
487 error = EBADF;
488 break;
489 }
490 fhold(fp);
491 FILEDESC_SUNLOCK(fdp);
492 tmp = arg;
493 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
494 fdrop(fp, td);
495 break;
496
497 case F_SETLKW:
498 flg |= F_WAIT;
499 /* FALLTHROUGH F_SETLK */
500
501 case F_SETLK:
502 FILEDESC_SLOCK(fdp);
503 if ((fp = fdtofp(fd, fdp)) == NULL) {
504 FILEDESC_SUNLOCK(fdp);
505 error = EBADF;
506 break;
507 }
508 if (fp->f_type != DTYPE_VNODE) {
509 FILEDESC_SUNLOCK(fdp);
510 error = EBADF;
511 break;
512 }
513 flp = (struct flock *)arg;
514 if (flp->l_whence == SEEK_CUR) {
515 if (fp->f_offset < 0 ||
516 (flp->l_start > 0 &&
517 fp->f_offset > OFF_MAX - flp->l_start)) {
518 FILEDESC_SUNLOCK(fdp);
519 error = EOVERFLOW;
520 break;
521 }
522 flp->l_start += fp->f_offset;
523 }
524
525 /*
526 * VOP_ADVLOCK() may block.
527 */
528 fhold(fp);
529 FILEDESC_SUNLOCK(fdp);
530 vp = fp->f_vnode;
531 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
532 switch (flp->l_type) {
533 case F_RDLCK:
534 if ((fp->f_flag & FREAD) == 0) {
535 error = EBADF;
536 break;
537 }
538 PROC_LOCK(p->p_leader);
539 p->p_leader->p_flag |= P_ADVLOCK;
540 PROC_UNLOCK(p->p_leader);
541 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
542 flp, flg);
543 break;
544 case F_WRLCK:
545 if ((fp->f_flag & FWRITE) == 0) {
546 error = EBADF;
547 break;
548 }
549 PROC_LOCK(p->p_leader);
550 p->p_leader->p_flag |= P_ADVLOCK;
551 PROC_UNLOCK(p->p_leader);
552 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
553 flp, flg);
554 break;
555 case F_UNLCK:
556 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
557 flp, F_POSIX);
558 break;
559 default:
560 error = EINVAL;
561 break;
562 }
563 VFS_UNLOCK_GIANT(vfslocked);
564 vfslocked = 0;
565 /* Check for race with close */
566 FILEDESC_SLOCK(fdp);
567 if ((unsigned) fd >= fdp->fd_nfiles ||
568 fp != fdp->fd_ofiles[fd]) {
569 FILEDESC_SUNLOCK(fdp);
570 flp->l_whence = SEEK_SET;
571 flp->l_start = 0;
572 flp->l_len = 0;
573 flp->l_type = F_UNLCK;
574 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
575 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
576 F_UNLCK, flp, F_POSIX);
577 VFS_UNLOCK_GIANT(vfslocked);
578 vfslocked = 0;
579 } else
580 FILEDESC_SUNLOCK(fdp);
581 fdrop(fp, td);
582 break;
583
584 case F_GETLK:
585 FILEDESC_SLOCK(fdp);
586 if ((fp = fdtofp(fd, fdp)) == NULL) {
587 FILEDESC_SUNLOCK(fdp);
588 error = EBADF;
589 break;
590 }
591 if (fp->f_type != DTYPE_VNODE) {
592 FILEDESC_SUNLOCK(fdp);
593 error = EBADF;
594 break;
595 }
596 flp = (struct flock *)arg;
597 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
598 flp->l_type != F_UNLCK) {
599 FILEDESC_SUNLOCK(fdp);
600 error = EINVAL;
601 break;
602 }
603 if (flp->l_whence == SEEK_CUR) {
604 if ((flp->l_start > 0 &&
605 fp->f_offset > OFF_MAX - flp->l_start) ||
606 (flp->l_start < 0 &&
607 fp->f_offset < OFF_MIN - flp->l_start)) {
608 FILEDESC_SUNLOCK(fdp);
609 error = EOVERFLOW;
610 break;
611 }
612 flp->l_start += fp->f_offset;
613 }
614 /*
615 * VOP_ADVLOCK() may block.
616 */
617 fhold(fp);
618 FILEDESC_SUNLOCK(fdp);
619 vp = fp->f_vnode;
620 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
621 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
622 F_POSIX);
623 VFS_UNLOCK_GIANT(vfslocked);
624 vfslocked = 0;
625 fdrop(fp, td);
626 break;
627 default:
628 error = EINVAL;
629 break;
630 }
631 VFS_UNLOCK_GIANT(vfslocked);
632 return (error);
633}
634
635/*
636 * Common code for dup, dup2, and fcntl(F_DUPFD).
637 */
638static int
639do_dup(struct thread *td, enum dup_type type, int old, int new,
640 register_t *retval)
641{
642 struct filedesc *fdp;
643 struct proc *p;
644 struct file *fp;
645 struct file *delfp;
646 int error, holdleaders, maxfd;
647
648 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
649 ("invalid dup type %d", type));
650
651 p = td->td_proc;
652 fdp = p->p_fd;
653
654 /*
655 * Verify we have a valid descriptor to dup from and possibly to
656 * dup to.
657 */
658 if (old < 0 || new < 0)
659 return (EBADF);
660 PROC_LOCK(p);
661 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
662 PROC_UNLOCK(p);
663 if (new >= maxfd)
664 return (EMFILE);
665
666 FILEDESC_XLOCK(fdp);
667 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
668 FILEDESC_XUNLOCK(fdp);
669 return (EBADF);
670 }
671 if (type == DUP_FIXED && old == new) {
672 *retval = new;
673 FILEDESC_XUNLOCK(fdp);
674 return (0);
675 }
676 fp = fdp->fd_ofiles[old];
677 fhold(fp);
678
679 /*
680 * If the caller specified a file descriptor, make sure the file
681 * table is large enough to hold it, and grab it. Otherwise, just
682 * allocate a new descriptor the usual way. Since the filedesc
683 * lock may be temporarily dropped in the process, we have to look
684 * out for a race.
685 */
686 if (type == DUP_FIXED) {
687 if (new >= fdp->fd_nfiles)
688 fdgrowtable(fdp, new + 1);
689 if (fdp->fd_ofiles[new] == NULL)
690 fdused(fdp, new);
691 } else {
692 if ((error = fdalloc(td, new, &new)) != 0) {
693 FILEDESC_XUNLOCK(fdp);
694 fdrop(fp, td);
695 return (error);
696 }
697 }
698
699 /*
700 * If the old file changed out from under us then treat it as a
701 * bad file descriptor. Userland should do its own locking to
702 * avoid this case.
703 */
704 if (fdp->fd_ofiles[old] != fp) {
705 /* we've allocated a descriptor which we won't use */
706 if (fdp->fd_ofiles[new] == NULL)
707 fdunused(fdp, new);
708 FILEDESC_XUNLOCK(fdp);
709 fdrop(fp, td);
710 return (EBADF);
711 }
712 KASSERT(old != new,
713 ("new fd is same as old"));
714
715 /*
716 * Save info on the descriptor being overwritten. We cannot close
717 * it without introducing an ownership race for the slot, since we
718 * need to drop the filedesc lock to call closef().
719 *
720 * XXX this duplicates parts of close().
721 */
722 delfp = fdp->fd_ofiles[new];
723 holdleaders = 0;
724 if (delfp != NULL) {
725 if (td->td_proc->p_fdtol != NULL) {
726 /*
727 * Ask fdfree() to sleep to ensure that all relevant
728 * process leaders can be traversed in closef().
729 */
730 fdp->fd_holdleaderscount++;
731 holdleaders = 1;
732 }
733 }
734
735 /*
736 * Duplicate the source descriptor
737 */
738 fdp->fd_ofiles[new] = fp;
739 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
740 if (new > fdp->fd_lastfile)
741 fdp->fd_lastfile = new;
742 *retval = new;
743
744 /*
745 * If we dup'd over a valid file, we now own the reference to it
746 * and must dispose of it using closef() semantics (as if a
747 * close() were performed on it).
748 *
749 * XXX this duplicates parts of close().
750 */
751 if (delfp != NULL) {
752 knote_fdclose(td, new);
753 if (delfp->f_type == DTYPE_MQUEUE)
754 mq_fdclose(td, new, delfp);
755 FILEDESC_XUNLOCK(fdp);
756 (void) closef(delfp, td);
757 if (holdleaders) {
758 FILEDESC_XLOCK(fdp);
759 fdp->fd_holdleaderscount--;
760 if (fdp->fd_holdleaderscount == 0 &&
761 fdp->fd_holdleaderswakeup != 0) {
762 fdp->fd_holdleaderswakeup = 0;
763 wakeup(&fdp->fd_holdleaderscount);
764 }
765 FILEDESC_XUNLOCK(fdp);
766 }
767 } else {
768 FILEDESC_XUNLOCK(fdp);
769 }
770 return (0);
771}
772
773/*
774 * If sigio is on the list associated with a process or process group,
775 * disable signalling from the device, remove sigio from the list and
776 * free sigio.
777 */
778void
779funsetown(struct sigio **sigiop)
780{
781 struct sigio *sigio;
782
783 SIGIO_LOCK();
784 sigio = *sigiop;
785 if (sigio == NULL) {
786 SIGIO_UNLOCK();
787 return;
788 }
789 *(sigio->sio_myref) = NULL;
790 if ((sigio)->sio_pgid < 0) {
791 struct pgrp *pg = (sigio)->sio_pgrp;
792 PGRP_LOCK(pg);
793 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
794 sigio, sio_pgsigio);
795 PGRP_UNLOCK(pg);
796 } else {
797 struct proc *p = (sigio)->sio_proc;
798 PROC_LOCK(p);
799 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
800 sigio, sio_pgsigio);
801 PROC_UNLOCK(p);
802 }
803 SIGIO_UNLOCK();
804 crfree(sigio->sio_ucred);
805 FREE(sigio, M_SIGIO);
806}
807
808/*
809 * Free a list of sigio structures.
810 * We only need to lock the SIGIO_LOCK because we have made ourselves
811 * inaccessible to callers of fsetown and therefore do not need to lock
812 * the proc or pgrp struct for the list manipulation.
813 */
814void
815funsetownlst(struct sigiolst *sigiolst)
816{
817 struct proc *p;
818 struct pgrp *pg;
819 struct sigio *sigio;
820
821 sigio = SLIST_FIRST(sigiolst);
822 if (sigio == NULL)
823 return;
824 p = NULL;
825 pg = NULL;
826
827 /*
828 * Every entry of the list should belong
829 * to a single proc or pgrp.
830 */
831 if (sigio->sio_pgid < 0) {
832 pg = sigio->sio_pgrp;
833 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
834 } else /* if (sigio->sio_pgid > 0) */ {
835 p = sigio->sio_proc;
836 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
837 }
838
839 SIGIO_LOCK();
840 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
841 *(sigio->sio_myref) = NULL;
842 if (pg != NULL) {
843 KASSERT(sigio->sio_pgid < 0,
844 ("Proc sigio in pgrp sigio list"));
845 KASSERT(sigio->sio_pgrp == pg,
846 ("Bogus pgrp in sigio list"));
847 PGRP_LOCK(pg);
848 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
849 sio_pgsigio);
850 PGRP_UNLOCK(pg);
851 } else /* if (p != NULL) */ {
852 KASSERT(sigio->sio_pgid > 0,
853 ("Pgrp sigio in proc sigio list"));
854 KASSERT(sigio->sio_proc == p,
855 ("Bogus proc in sigio list"));
856 PROC_LOCK(p);
857 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
858 sio_pgsigio);
859 PROC_UNLOCK(p);
860 }
861 SIGIO_UNLOCK();
862 crfree(sigio->sio_ucred);
863 FREE(sigio, M_SIGIO);
864 SIGIO_LOCK();
865 }
866 SIGIO_UNLOCK();
867}
868
869/*
870 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
871 *
872 * After permission checking, add a sigio structure to the sigio list for
873 * the process or process group.
874 */
875int
876fsetown(pid_t pgid, struct sigio **sigiop)
877{
878 struct proc *proc;
879 struct pgrp *pgrp;
880 struct sigio *sigio;
881 int ret;
882
883 if (pgid == 0) {
884 funsetown(sigiop);
885 return (0);
886 }
887
888 ret = 0;
889
890 /* Allocate and fill in the new sigio out of locks. */
891 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
892 sigio->sio_pgid = pgid;
893 sigio->sio_ucred = crhold(curthread->td_ucred);
894 sigio->sio_myref = sigiop;
895
896 sx_slock(&proctree_lock);
897 if (pgid > 0) {
898 proc = pfind(pgid);
899 if (proc == NULL) {
900 ret = ESRCH;
901 goto fail;
902 }
903
904 /*
905 * Policy - Don't allow a process to FSETOWN a process
906 * in another session.
907 *
908 * Remove this test to allow maximum flexibility or
909 * restrict FSETOWN to the current process or process
910 * group for maximum safety.
911 */
912 PROC_UNLOCK(proc);
913 if (proc->p_session != curthread->td_proc->p_session) {
914 ret = EPERM;
915 goto fail;
916 }
917
918 pgrp = NULL;
919 } else /* if (pgid < 0) */ {
920 pgrp = pgfind(-pgid);
921 if (pgrp == NULL) {
922 ret = ESRCH;
923 goto fail;
924 }
925 PGRP_UNLOCK(pgrp);
926
927 /*
928 * Policy - Don't allow a process to FSETOWN a process
929 * in another session.
930 *
931 * Remove this test to allow maximum flexibility or
932 * restrict FSETOWN to the current process or process
933 * group for maximum safety.
934 */
935 if (pgrp->pg_session != curthread->td_proc->p_session) {
936 ret = EPERM;
937 goto fail;
938 }
939
940 proc = NULL;
941 }
942 funsetown(sigiop);
943 if (pgid > 0) {
944 PROC_LOCK(proc);
945 /*
946 * Since funsetownlst() is called without the proctree
947 * locked, we need to check for P_WEXIT.
948 * XXX: is ESRCH correct?
949 */
950 if ((proc->p_flag & P_WEXIT) != 0) {
951 PROC_UNLOCK(proc);
952 ret = ESRCH;
953 goto fail;
954 }
955 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
956 sigio->sio_proc = proc;
957 PROC_UNLOCK(proc);
958 } else {
959 PGRP_LOCK(pgrp);
960 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
961 sigio->sio_pgrp = pgrp;
962 PGRP_UNLOCK(pgrp);
963 }
964 sx_sunlock(&proctree_lock);
965 SIGIO_LOCK();
966 *sigiop = sigio;
967 SIGIO_UNLOCK();
968 return (0);
969
970fail:
971 sx_sunlock(&proctree_lock);
972 crfree(sigio->sio_ucred);
973 FREE(sigio, M_SIGIO);
974 return (ret);
975}
976
977/*
978 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
979 */
980pid_t
981fgetown(sigiop)
982 struct sigio **sigiop;
983{
984 pid_t pgid;
985
986 SIGIO_LOCK();
987 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
988 SIGIO_UNLOCK();
989 return (pgid);
990}
991
992/*
993 * Close a file descriptor.
994 */
995#ifndef _SYS_SYSPROTO_H_
996struct close_args {
997 int fd;
998};
999#endif
1000/* ARGSUSED */
1001int
1002close(td, uap)
1003 struct thread *td;
1004 struct close_args *uap;
1005{
1006
1007 return (kern_close(td, uap->fd));
1008}
1009
1010int
1011kern_close(td, fd)
1012 struct thread *td;
1013 int fd;
1014{
1015 struct filedesc *fdp;
1016 struct file *fp;
1017 int error;
1018 int holdleaders;
1019
1020 error = 0;
1021 holdleaders = 0;
1022 fdp = td->td_proc->p_fd;
1023
1024 AUDIT_SYSCLOSE(td, fd);
1025
1026 FILEDESC_XLOCK(fdp);
1027 if ((unsigned)fd >= fdp->fd_nfiles ||
1028 (fp = fdp->fd_ofiles[fd]) == NULL) {
1029 FILEDESC_XUNLOCK(fdp);
1030 return (EBADF);
1031 }
1032 fdp->fd_ofiles[fd] = NULL;
1033 fdp->fd_ofileflags[fd] = 0;
1034 fdunused(fdp, fd);
1035 if (td->td_proc->p_fdtol != NULL) {
1036 /*
1037 * Ask fdfree() to sleep to ensure that all relevant
1038 * process leaders can be traversed in closef().
1039 */
1040 fdp->fd_holdleaderscount++;
1041 holdleaders = 1;
1042 }
1043
1044 /*
1045 * We now hold the fp reference that used to be owned by the
1046 * descriptor array. We have to unlock the FILEDESC *AFTER*
1047 * knote_fdclose to prevent a race of the fd getting opened, a knote
1048 * added, and deleteing a knote for the new fd.
1049 */
1050 knote_fdclose(td, fd);
1051 if (fp->f_type == DTYPE_MQUEUE)
1052 mq_fdclose(td, fd, fp);
1053 FILEDESC_XUNLOCK(fdp);
1054
1055 error = closef(fp, td);
1056 if (holdleaders) {
1057 FILEDESC_XLOCK(fdp);
1058 fdp->fd_holdleaderscount--;
1059 if (fdp->fd_holdleaderscount == 0 &&
1060 fdp->fd_holdleaderswakeup != 0) {
1061 fdp->fd_holdleaderswakeup = 0;
1062 wakeup(&fdp->fd_holdleaderscount);
1063 }
1064 FILEDESC_XUNLOCK(fdp);
1065 }
1066 return (error);
1067}
1068
1069#if defined(COMPAT_43)
1070/*
1071 * Return status information about a file descriptor.
1072 */
1073#ifndef _SYS_SYSPROTO_H_
1074struct ofstat_args {
1075 int fd;
1076 struct ostat *sb;
1077};
1078#endif
1079/* ARGSUSED */
1080int
1081ofstat(struct thread *td, struct ofstat_args *uap)
1082{
1083 struct ostat oub;
1084 struct stat ub;
1085 int error;
1086
1087 error = kern_fstat(td, uap->fd, &ub);
1088 if (error == 0) {
1089 cvtstat(&ub, &oub);
1090 error = copyout(&oub, uap->sb, sizeof(oub));
1091 }
1092 return (error);
1093}
1094#endif /* COMPAT_43 */
1095
1096/*
1097 * Return status information about a file descriptor.
1098 */
1099#ifndef _SYS_SYSPROTO_H_
1100struct fstat_args {
1101 int fd;
1102 struct stat *sb;
1103};
1104#endif
1105/* ARGSUSED */
1106int
1107fstat(struct thread *td, struct fstat_args *uap)
1108{
1109 struct stat ub;
1110 int error;
1111
1112 error = kern_fstat(td, uap->fd, &ub);
1113 if (error == 0)
1114 error = copyout(&ub, uap->sb, sizeof(ub));
1115 return (error);
1116}
1117
1118int
1119kern_fstat(struct thread *td, int fd, struct stat *sbp)
1120{
1121 struct file *fp;
1122 int error;
1123
1124 AUDIT_ARG(fd, fd);
1125
1126 if ((error = fget(td, fd, &fp)) != 0)
1127 return (error);
1128
1129 AUDIT_ARG(file, td->td_proc, fp);
1130
1131 error = fo_stat(fp, sbp, td->td_ucred, td);
1132 fdrop(fp, td);
1133#ifdef KTRACE
1134 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1135 ktrstat(sbp);
1136#endif
1137 return (error);
1138}
1139
1140/*
1141 * Return status information about a file descriptor.
1142 */
1143#ifndef _SYS_SYSPROTO_H_
1144struct nfstat_args {
1145 int fd;
1146 struct nstat *sb;
1147};
1148#endif
1149/* ARGSUSED */
1150int
1151nfstat(struct thread *td, struct nfstat_args *uap)
1152{
1153 struct nstat nub;
1154 struct stat ub;
1155 int error;
1156
1157 error = kern_fstat(td, uap->fd, &ub);
1158 if (error == 0) {
1159 cvtnstat(&ub, &nub);
1160 error = copyout(&nub, uap->sb, sizeof(nub));
1161 }
1162 return (error);
1163}
1164
1165/*
1166 * Return pathconf information about a file descriptor.
1167 */
1168#ifndef _SYS_SYSPROTO_H_
1169struct fpathconf_args {
1170 int fd;
1171 int name;
1172};
1173#endif
1174/* ARGSUSED */
1175int
1176fpathconf(struct thread *td, struct fpathconf_args *uap)
1177{
1178 struct file *fp;
1179 struct vnode *vp;
1180 int error;
1181
1182 if ((error = fget(td, uap->fd, &fp)) != 0)
1183 return (error);
1184
1185 /* If asynchronous I/O is available, it works for all descriptors. */
1186 if (uap->name == _PC_ASYNC_IO) {
1187 td->td_retval[0] = async_io_version;
1188 goto out;
1189 }
1190 vp = fp->f_vnode;
1191 if (vp != NULL) {
1192 int vfslocked;
1193 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1194 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1195 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1196 VOP_UNLOCK(vp, 0);
1197 VFS_UNLOCK_GIANT(vfslocked);
1198 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1199 if (uap->name != _PC_PIPE_BUF) {
1200 error = EINVAL;
1201 } else {
1202 td->td_retval[0] = PIPE_BUF;
1203 error = 0;
1204 }
1205 } else {
1206 error = EOPNOTSUPP;
1207 }
1208out:
1209 fdrop(fp, td);
1210 return (error);
1211}
1212
1213/*
1214 * Grow the file table to accomodate (at least) nfd descriptors. This may
1215 * block and drop the filedesc lock, but it will reacquire it before
1216 * returning.
1217 */
1218static void
1219fdgrowtable(struct filedesc *fdp, int nfd)
1220{
1221 struct file **ntable;
1222 char *nfileflags;
1223 int nnfiles, onfiles;
1224 NDSLOTTYPE *nmap;
1225
1226 FILEDESC_XLOCK_ASSERT(fdp);
1227
1228 KASSERT(fdp->fd_nfiles > 0,
1229 ("zero-length file table"));
1230
1231 /* compute the size of the new table */
1232 onfiles = fdp->fd_nfiles;
1233 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1234 if (nnfiles <= onfiles)
1235 /* the table is already large enough */
1236 return;
1237
1238 /* allocate a new table and (if required) new bitmaps */
1239 FILEDESC_XUNLOCK(fdp);
1240 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1241 M_FILEDESC, M_ZERO | M_WAITOK);
1242 nfileflags = (char *)&ntable[nnfiles];
1243 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1244 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1245 M_FILEDESC, M_ZERO | M_WAITOK);
1246 else
1247 nmap = NULL;
1248 FILEDESC_XLOCK(fdp);
1249
1250 /*
1251 * We now have new tables ready to go. Since we dropped the
1252 * filedesc lock to call malloc(), watch out for a race.
1253 */
1254 onfiles = fdp->fd_nfiles;
1255 if (onfiles >= nnfiles) {
1256 /* we lost the race, but that's OK */
1257 free(ntable, M_FILEDESC);
1258 if (nmap != NULL)
1259 free(nmap, M_FILEDESC);
1260 return;
1261 }
1262 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1263 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1264 if (onfiles > NDFILE)
1265 free(fdp->fd_ofiles, M_FILEDESC);
1266 fdp->fd_ofiles = ntable;
1267 fdp->fd_ofileflags = nfileflags;
1268 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1269 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1270 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1271 free(fdp->fd_map, M_FILEDESC);
1272 fdp->fd_map = nmap;
1273 }
1274 fdp->fd_nfiles = nnfiles;
1275}
1276
1277/*
1278 * Allocate a file descriptor for the process.
1279 */
1280int
1281fdalloc(struct thread *td, int minfd, int *result)
1282{
1283 struct proc *p = td->td_proc;
1284 struct filedesc *fdp = p->p_fd;
1285 int fd = -1, maxfd;
1286
1287 FILEDESC_XLOCK_ASSERT(fdp);
1288
1289 if (fdp->fd_freefile > minfd)
1290 minfd = fdp->fd_freefile;
1291
1292 PROC_LOCK(p);
1293 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1294 PROC_UNLOCK(p);
1295
1296 /*
1297 * Search the bitmap for a free descriptor. If none is found, try
1298 * to grow the file table. Keep at it until we either get a file
1299 * descriptor or run into process or system limits; fdgrowtable()
1300 * may drop the filedesc lock, so we're in a race.
1301 */
1302 for (;;) {
1303 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1304 if (fd >= maxfd)
1305 return (EMFILE);
1306 if (fd < fdp->fd_nfiles)
1307 break;
1308 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1309 }
1310
1311 /*
1312 * Perform some sanity checks, then mark the file descriptor as
1313 * used and return it to the caller.
1314 */
1315 KASSERT(!fdisused(fdp, fd),
1316 ("fd_first_free() returned non-free descriptor"));
1317 KASSERT(fdp->fd_ofiles[fd] == NULL,
1318 ("free descriptor isn't"));
1319 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1320 fdused(fdp, fd);
1321 *result = fd;
1322 return (0);
1323}
1324
1325/*
1326 * Check to see whether n user file descriptors are available to the process
1327 * p.
1328 */
1329int
1330fdavail(struct thread *td, int n)
1331{
1332 struct proc *p = td->td_proc;
1333 struct filedesc *fdp = td->td_proc->p_fd;
1334 struct file **fpp;
1335 int i, lim, last;
1336
1337 FILEDESC_LOCK_ASSERT(fdp);
1338
1339 PROC_LOCK(p);
1340 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1341 PROC_UNLOCK(p);
1342 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1343 return (1);
1344 last = min(fdp->fd_nfiles, lim);
1345 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1346 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1347 if (*fpp == NULL && --n <= 0)
1348 return (1);
1349 }
1350 return (0);
1351}
1352
1353/*
1354 * Create a new open file structure and allocate a file decriptor for the
1355 * process that refers to it. We add one reference to the file for the
1356 * descriptor table and one reference for resultfp. This is to prevent us
1357 * being preempted and the entry in the descriptor table closed after we
1358 * release the FILEDESC lock.
1359 */
1360int
1361falloc(struct thread *td, struct file **resultfp, int *resultfd)
1362{
1363 struct proc *p = td->td_proc;
1364 struct file *fp;
1365 int error, i;
1366 int maxuserfiles = maxfiles - (maxfiles / 20);
1367 static struct timeval lastfail;
1368 static int curfail;
1369
1370 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1371 if ((openfiles >= maxuserfiles &&
1372 priv_check(td, PRIV_MAXFILES) != 0) ||
1373 openfiles >= maxfiles) {
1374 if (ppsratecheck(&lastfail, &curfail, 1)) {
1375 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1376 td->td_ucred->cr_ruid);
1377 }
1378 uma_zfree(file_zone, fp);
1379 return (ENFILE);
1380 }
1381 atomic_add_int(&openfiles, 1);
1382
1383 /*
1384 * If the process has file descriptor zero open, add the new file
1385 * descriptor to the list of open files at that point, otherwise
1386 * put it at the front of the list of open files.
1387 */
1388 fp->f_count = 1;
1389 if (resultfp)
1390 fp->f_count++;
1391 fp->f_cred = crhold(td->td_ucred);
1392 fp->f_ops = &badfileops;
1393 fp->f_data = NULL;
1394 fp->f_vnode = NULL;
1395 FILEDESC_XLOCK(p->p_fd);
1396 if ((error = fdalloc(td, 0, &i))) {
1397 FILEDESC_XUNLOCK(p->p_fd);
1398 fdrop(fp, td);
1399 if (resultfp)
1400 fdrop(fp, td);
1401 return (error);
1402 }
1403 p->p_fd->fd_ofiles[i] = fp;
1404 FILEDESC_XUNLOCK(p->p_fd);
1405 if (resultfp)
1406 *resultfp = fp;
1407 if (resultfd)
1408 *resultfd = i;
1409 return (0);
1410}
1411
1412/*
1413 * Build a new filedesc structure from another.
1414 * Copy the current, root, and jail root vnode references.
1415 */
1416struct filedesc *
1417fdinit(struct filedesc *fdp)
1418{
1419 struct filedesc0 *newfdp;
1420
1421 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1422 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1423 if (fdp != NULL) {
1424 FILEDESC_XLOCK(fdp);
1425 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1426 if (newfdp->fd_fd.fd_cdir)
1427 VREF(newfdp->fd_fd.fd_cdir);
1428 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1429 if (newfdp->fd_fd.fd_rdir)
1430 VREF(newfdp->fd_fd.fd_rdir);
1431 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1432 if (newfdp->fd_fd.fd_jdir)
1433 VREF(newfdp->fd_fd.fd_jdir);
1434 FILEDESC_XUNLOCK(fdp);
1435 }
1436
1437 /* Create the file descriptor table. */
1438 newfdp->fd_fd.fd_refcnt = 1;
1439 newfdp->fd_fd.fd_holdcnt = 1;
1440 newfdp->fd_fd.fd_cmask = CMASK;
1441 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1442 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1443 newfdp->fd_fd.fd_nfiles = NDFILE;
1444 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1445 newfdp->fd_fd.fd_lastfile = -1;
1446 return (&newfdp->fd_fd);
1447}
1448
1449static struct filedesc *
1450fdhold(struct proc *p)
1451{
1452 struct filedesc *fdp;
1453
1454 mtx_lock(&fdesc_mtx);
1455 fdp = p->p_fd;
1456 if (fdp != NULL)
1457 fdp->fd_holdcnt++;
1458 mtx_unlock(&fdesc_mtx);
1459 return (fdp);
1460}
1461
1462static void
1463fddrop(struct filedesc *fdp)
1464{
1465 int i;
1466
1467 mtx_lock(&fdesc_mtx);
1468 i = --fdp->fd_holdcnt;
1469 mtx_unlock(&fdesc_mtx);
1470 if (i > 0)
1471 return;
1472
1473 FILEDESC_LOCK_DESTROY(fdp);
1474 FREE(fdp, M_FILEDESC);
1475}
1476
1477/*
1478 * Share a filedesc structure.
1479 */
1480struct filedesc *
1481fdshare(struct filedesc *fdp)
1482{
1483
1484 FILEDESC_XLOCK(fdp);
1485 fdp->fd_refcnt++;
1486 FILEDESC_XUNLOCK(fdp);
1487 return (fdp);
1488}
1489
1490/*
1491 * Unshare a filedesc structure, if necessary by making a copy
1492 */
1493void
1494fdunshare(struct proc *p, struct thread *td)
1495{
1496
1497 FILEDESC_XLOCK(p->p_fd);
1498 if (p->p_fd->fd_refcnt > 1) {
1499 struct filedesc *tmp;
1500
1501 FILEDESC_XUNLOCK(p->p_fd);
1502 tmp = fdcopy(p->p_fd);
1503 fdfree(td);
1504 p->p_fd = tmp;
1505 } else
1506 FILEDESC_XUNLOCK(p->p_fd);
1507}
1508
1509/*
1510 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1511 * this is to ease callers, not catch errors.
1512 */
1513struct filedesc *
1514fdcopy(struct filedesc *fdp)
1515{
1516 struct filedesc *newfdp;
1517 int i;
1518
1519 /* Certain daemons might not have file descriptors. */
1520 if (fdp == NULL)
1521 return (NULL);
1522
1523 newfdp = fdinit(fdp);
1524 FILEDESC_SLOCK(fdp);
1525 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1526 FILEDESC_SUNLOCK(fdp);
1527 FILEDESC_XLOCK(newfdp);
1528 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1529 FILEDESC_XUNLOCK(newfdp);
1530 FILEDESC_SLOCK(fdp);
1531 }
1532 /* copy everything except kqueue descriptors */
1533 newfdp->fd_freefile = -1;
1534 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1535 if (fdisused(fdp, i) &&
1536 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1537 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1538 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1539 fhold(newfdp->fd_ofiles[i]);
1540 newfdp->fd_lastfile = i;
1541 } else {
1542 if (newfdp->fd_freefile == -1)
1543 newfdp->fd_freefile = i;
1544 }
1545 }
1546 FILEDESC_SUNLOCK(fdp);
1547 FILEDESC_XLOCK(newfdp);
1548 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1549 if (newfdp->fd_ofiles[i] != NULL)
1550 fdused(newfdp, i);
1551 FILEDESC_XUNLOCK(newfdp);
1552 FILEDESC_SLOCK(fdp);
1553 if (newfdp->fd_freefile == -1)
1554 newfdp->fd_freefile = i;
1555 newfdp->fd_cmask = fdp->fd_cmask;
1556 FILEDESC_SUNLOCK(fdp);
1557 return (newfdp);
1558}
1559
1560/*
1561 * Release a filedesc structure.
1562 */
1563void
1564fdfree(struct thread *td)
1565{
1566 struct filedesc *fdp;
1567 struct file **fpp;
1568 int i, locked;
1569 struct filedesc_to_leader *fdtol;
1570 struct file *fp;
1571 struct vnode *cdir, *jdir, *rdir, *vp;
1572 struct flock lf;
1573
1574 /* Certain daemons might not have file descriptors. */
1575 fdp = td->td_proc->p_fd;
1576 if (fdp == NULL)
1577 return;
1578
1579 /* Check for special need to clear POSIX style locks */
1580 fdtol = td->td_proc->p_fdtol;
1581 if (fdtol != NULL) {
1582 FILEDESC_XLOCK(fdp);
1583 KASSERT(fdtol->fdl_refcount > 0,
1584 ("filedesc_to_refcount botch: fdl_refcount=%d",
1585 fdtol->fdl_refcount));
1586 if (fdtol->fdl_refcount == 1 &&
1587 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1588 for (i = 0, fpp = fdp->fd_ofiles;
1589 i <= fdp->fd_lastfile;
1590 i++, fpp++) {
1591 if (*fpp == NULL ||
1592 (*fpp)->f_type != DTYPE_VNODE)
1593 continue;
1594 fp = *fpp;
1595 fhold(fp);
1596 FILEDESC_XUNLOCK(fdp);
1597 lf.l_whence = SEEK_SET;
1598 lf.l_start = 0;
1599 lf.l_len = 0;
1600 lf.l_type = F_UNLCK;
1601 vp = fp->f_vnode;
1602 locked = VFS_LOCK_GIANT(vp->v_mount);
1603 (void) VOP_ADVLOCK(vp,
1604 (caddr_t)td->td_proc->
1605 p_leader,
1606 F_UNLCK,
1607 &lf,
1608 F_POSIX);
1609 VFS_UNLOCK_GIANT(locked);
1610 FILEDESC_XLOCK(fdp);
1611 fdrop(fp, td);
1612 fpp = fdp->fd_ofiles + i;
1613 }
1614 }
1615 retry:
1616 if (fdtol->fdl_refcount == 1) {
1617 if (fdp->fd_holdleaderscount > 0 &&
1618 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1619 /*
1620 * close() or do_dup() has cleared a reference
1621 * in a shared file descriptor table.
1622 */
1623 fdp->fd_holdleaderswakeup = 1;
1624 sx_sleep(&fdp->fd_holdleaderscount,
1625 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1626 goto retry;
1627 }
1628 if (fdtol->fdl_holdcount > 0) {
1629 /*
1630 * Ensure that fdtol->fdl_leader remains
1631 * valid in closef().
1632 */
1633 fdtol->fdl_wakeup = 1;
1634 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1635 "fdlhold", 0);
1636 goto retry;
1637 }
1638 }
1639 fdtol->fdl_refcount--;
1640 if (fdtol->fdl_refcount == 0 &&
1641 fdtol->fdl_holdcount == 0) {
1642 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1643 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1644 } else
1645 fdtol = NULL;
1646 td->td_proc->p_fdtol = NULL;
1647 FILEDESC_XUNLOCK(fdp);
1648 if (fdtol != NULL)
1649 FREE(fdtol, M_FILEDESC_TO_LEADER);
1650 }
1651 FILEDESC_XLOCK(fdp);
1652 i = --fdp->fd_refcnt;
1653 FILEDESC_XUNLOCK(fdp);
1654 if (i > 0)
1655 return;
1656 /*
1657 * We are the last reference to the structure, so we can
1658 * safely assume it will not change out from under us.
1659 */
1660 fpp = fdp->fd_ofiles;
1661 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1662 if (*fpp)
1663 (void) closef(*fpp, td);
1664 }
1665 FILEDESC_XLOCK(fdp);
1666
1667 /* XXX This should happen earlier. */
1668 mtx_lock(&fdesc_mtx);
1669 td->td_proc->p_fd = NULL;
1670 mtx_unlock(&fdesc_mtx);
1671
1672 if (fdp->fd_nfiles > NDFILE)
1673 FREE(fdp->fd_ofiles, M_FILEDESC);
1674 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1675 FREE(fdp->fd_map, M_FILEDESC);
1676
1677 fdp->fd_nfiles = 0;
1678
1679 cdir = fdp->fd_cdir;
1680 fdp->fd_cdir = NULL;
1681 rdir = fdp->fd_rdir;
1682 fdp->fd_rdir = NULL;
1683 jdir = fdp->fd_jdir;
1684 fdp->fd_jdir = NULL;
1685 FILEDESC_XUNLOCK(fdp);
1686
1687 if (cdir) {
1688 locked = VFS_LOCK_GIANT(cdir->v_mount);
1689 vrele(cdir);
1690 VFS_UNLOCK_GIANT(locked);
1691 }
1692 if (rdir) {
1693 locked = VFS_LOCK_GIANT(rdir->v_mount);
1694 vrele(rdir);
1695 VFS_UNLOCK_GIANT(locked);
1696 }
1697 if (jdir) {
1698 locked = VFS_LOCK_GIANT(jdir->v_mount);
1699 vrele(jdir);
1700 VFS_UNLOCK_GIANT(locked);
1701 }
1702
1703 fddrop(fdp);
1704}
1705
1706/*
1707 * For setugid programs, we don't want to people to use that setugidness
1708 * to generate error messages which write to a file which otherwise would
1709 * otherwise be off-limits to the process. We check for filesystems where
1710 * the vnode can change out from under us after execve (like [lin]procfs).
1711 *
1712 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1713 * sufficient. We also don't check for setugidness since we know we are.
1714 */
1715static int
1716is_unsafe(struct file *fp)
1717{
1718 if (fp->f_type == DTYPE_VNODE) {
1719 struct vnode *vp = fp->f_vnode;
1720
1721 if ((vp->v_vflag & VV_PROCDEP) != 0)
1722 return (1);
1723 }
1724 return (0);
1725}
1726
1727/*
1728 * Make this setguid thing safe, if at all possible.
1729 */
1730void
1731setugidsafety(struct thread *td)
1732{
1733 struct filedesc *fdp;
1734 int i;
1735
1736 /* Certain daemons might not have file descriptors. */
1737 fdp = td->td_proc->p_fd;
1738 if (fdp == NULL)
1739 return;
1740
1741 /*
1742 * Note: fdp->fd_ofiles may be reallocated out from under us while
1743 * we are blocked in a close. Be careful!
1744 */
1745 FILEDESC_XLOCK(fdp);
1746 for (i = 0; i <= fdp->fd_lastfile; i++) {
1747 if (i > 2)
1748 break;
1749 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1750 struct file *fp;
1751
1752 knote_fdclose(td, i);
1753 /*
1754 * NULL-out descriptor prior to close to avoid
1755 * a race while close blocks.
1756 */
1757 fp = fdp->fd_ofiles[i];
1758 fdp->fd_ofiles[i] = NULL;
1759 fdp->fd_ofileflags[i] = 0;
1760 fdunused(fdp, i);
1761 FILEDESC_XUNLOCK(fdp);
1762 (void) closef(fp, td);
1763 FILEDESC_XLOCK(fdp);
1764 }
1765 }
1766 FILEDESC_XUNLOCK(fdp);
1767}
1768
1769/*
1770 * If a specific file object occupies a specific file descriptor, close the
1771 * file descriptor entry and drop a reference on the file object. This is a
1772 * convenience function to handle a subsequent error in a function that calls
1773 * falloc() that handles the race that another thread might have closed the
1774 * file descriptor out from under the thread creating the file object.
1775 */
1776void
1777fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1778{
1779
1780 FILEDESC_XLOCK(fdp);
1781 if (fdp->fd_ofiles[idx] == fp) {
1782 fdp->fd_ofiles[idx] = NULL;
1783 fdunused(fdp, idx);
1784 FILEDESC_XUNLOCK(fdp);
1785 fdrop(fp, td);
1786 } else
1787 FILEDESC_XUNLOCK(fdp);
1788}
1789
1790/*
1791 * Close any files on exec?
1792 */
1793void
1794fdcloseexec(struct thread *td)
1795{
1796 struct filedesc *fdp;
1797 int i;
1798
1799 /* Certain daemons might not have file descriptors. */
1800 fdp = td->td_proc->p_fd;
1801 if (fdp == NULL)
1802 return;
1803
1804 FILEDESC_XLOCK(fdp);
1805
1806 /*
1807 * We cannot cache fd_ofiles or fd_ofileflags since operations
1808 * may block and rip them out from under us.
1809 */
1810 for (i = 0; i <= fdp->fd_lastfile; i++) {
1811 if (fdp->fd_ofiles[i] != NULL &&
1812 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1813 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1814 struct file *fp;
1815
1816 knote_fdclose(td, i);
1817 /*
1818 * NULL-out descriptor prior to close to avoid
1819 * a race while close blocks.
1820 */
1821 fp = fdp->fd_ofiles[i];
1822 fdp->fd_ofiles[i] = NULL;
1823 fdp->fd_ofileflags[i] = 0;
1824 fdunused(fdp, i);
1825 if (fp->f_type == DTYPE_MQUEUE)
1826 mq_fdclose(td, i, fp);
1827 FILEDESC_XUNLOCK(fdp);
1828 (void) closef(fp, td);
1829 FILEDESC_XLOCK(fdp);
1830 }
1831 }
1832 FILEDESC_XUNLOCK(fdp);
1833}
1834
1835/*
1836 * It is unsafe for set[ug]id processes to be started with file
1837 * descriptors 0..2 closed, as these descriptors are given implicit
1838 * significance in the Standard C library. fdcheckstd() will create a
1839 * descriptor referencing /dev/null for each of stdin, stdout, and
1840 * stderr that is not already open.
1841 */
1842int
1843fdcheckstd(struct thread *td)
1844{
1845 struct filedesc *fdp;
1846 register_t retval, save;
1847 int i, error, devnull;
1848
1849 fdp = td->td_proc->p_fd;
1850 if (fdp == NULL)
1851 return (0);
1852 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1853 devnull = -1;
1854 error = 0;
1855 for (i = 0; i < 3; i++) {
1856 if (fdp->fd_ofiles[i] != NULL)
1857 continue;
1858 if (devnull < 0) {
1859 save = td->td_retval[0];
1860 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
1861 O_RDWR, 0);
1862 devnull = td->td_retval[0];
1863 KASSERT(devnull == i, ("oof, we didn't get our fd"));
1864 td->td_retval[0] = save;
1865 if (error)
1866 break;
1867 } else {
1868 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1869 if (error != 0)
1870 break;
1871 }
1872 }
1873 return (error);
1874}
1875
1876/*
1877 * Internal form of close. Decrement reference count on file structure.
1878 * Note: td may be NULL when closing a file that was being passed in a
1879 * message.
1880 *
1881 * XXXRW: Giant is not required for the caller, but often will be held; this
1882 * makes it moderately likely the Giant will be recursed in the VFS case.
1883 */
1884int
1885closef(struct file *fp, struct thread *td)
1886{
1887 struct vnode *vp;
1888 struct flock lf;
1889 struct filedesc_to_leader *fdtol;
1890 struct filedesc *fdp;
1891
1892 /*
1893 * POSIX record locking dictates that any close releases ALL
1894 * locks owned by this process. This is handled by setting
1895 * a flag in the unlock to free ONLY locks obeying POSIX
1896 * semantics, and not to free BSD-style file locks.
1897 * If the descriptor was in a message, POSIX-style locks
1898 * aren't passed with the descriptor, and the thread pointer
1899 * will be NULL. Callers should be careful only to pass a
1900 * NULL thread pointer when there really is no owning
1901 * context that might have locks, or the locks will be
1902 * leaked.
1903 */
1904 if (fp->f_type == DTYPE_VNODE && td != NULL) {
1905 int vfslocked;
1906
1907 vp = fp->f_vnode;
1908 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1909 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1910 lf.l_whence = SEEK_SET;
1911 lf.l_start = 0;
1912 lf.l_len = 0;
1913 lf.l_type = F_UNLCK;
1914 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1915 F_UNLCK, &lf, F_POSIX);
1916 }
1917 fdtol = td->td_proc->p_fdtol;
1918 if (fdtol != NULL) {
1919 /*
1920 * Handle special case where file descriptor table is
1921 * shared between multiple process leaders.
1922 */
1923 fdp = td->td_proc->p_fd;
1924 FILEDESC_XLOCK(fdp);
1925 for (fdtol = fdtol->fdl_next;
1926 fdtol != td->td_proc->p_fdtol;
1927 fdtol = fdtol->fdl_next) {
1928 if ((fdtol->fdl_leader->p_flag &
1929 P_ADVLOCK) == 0)
1930 continue;
1931 fdtol->fdl_holdcount++;
1932 FILEDESC_XUNLOCK(fdp);
1933 lf.l_whence = SEEK_SET;
1934 lf.l_start = 0;
1935 lf.l_len = 0;
1936 lf.l_type = F_UNLCK;
1937 vp = fp->f_vnode;
1938 (void) VOP_ADVLOCK(vp,
1939 (caddr_t)fdtol->fdl_leader,
1940 F_UNLCK, &lf, F_POSIX);
1941 FILEDESC_XLOCK(fdp);
1942 fdtol->fdl_holdcount--;
1943 if (fdtol->fdl_holdcount == 0 &&
1944 fdtol->fdl_wakeup != 0) {
1945 fdtol->fdl_wakeup = 0;
1946 wakeup(fdtol);
1947 }
1948 }
1949 FILEDESC_XUNLOCK(fdp);
1950 }
1951 VFS_UNLOCK_GIANT(vfslocked);
1952 }
1953 return (fdrop(fp, td));
1954}
1955
1956/*
1957 * Initialize the file pointer with the specified properties.
1958 *
1959 * The ops are set with release semantics to be certain that the flags, type,
1960 * and data are visible when ops is. This is to prevent ops methods from being
1961 * called with bad data.
1962 */
1963void
1964finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
1965{
1966 fp->f_data = data;
1967 fp->f_flag = flag;
1968 fp->f_type = type;
1969 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
1970}
1971
1972
1973/*
1974 * Extract the file pointer associated with the specified descriptor for the
1975 * current user process.
1976 *
1977 * If the descriptor doesn't exist, EBADF is returned.
1978 *
1979 * If the descriptor exists but doesn't match 'flags' then return EBADF for
1980 * read attempts and EINVAL for write attempts.
1981 *
1982 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1983 * It should be dropped with fdrop(). If it is not set, then the refcount
1984 * will not be bumped however the thread's filedesc struct will be returned
1985 * locked (for fgetsock).
1986 *
1987 * If an error occured the non-zero error is returned and *fpp is set to
1988 * NULL. Otherwise *fpp is set and zero is returned.
1989 */
1990static __inline int
1991_fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1992{
1993 struct filedesc *fdp;
1994 struct file *fp;
1995
1996 *fpp = NULL;
1997 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1998 return (EBADF);
1999 FILEDESC_SLOCK(fdp);
2000 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
2001 FILEDESC_SUNLOCK(fdp);
2002 return (EBADF);
2003 }
2004
2005 /*
2006 * FREAD and FWRITE failure return EBADF as per POSIX.
2007 *
2008 * Only one flag, or 0, may be specified.
2009 */
2010 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2011 FILEDESC_SUNLOCK(fdp);
2012 return (EBADF);
2013 }
2014 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2015 FILEDESC_SUNLOCK(fdp);
2016 return (EBADF);
2017 }
2018 if (hold) {
2019 fhold(fp);
2020 FILEDESC_SUNLOCK(fdp);
2021 }
2022 *fpp = fp;
2023 return (0);
2024}
2025
2026int
2027fget(struct thread *td, int fd, struct file **fpp)
2028{
2029
2030 return(_fget(td, fd, fpp, 0, 1));
2031}
2032
2033int
2034fget_read(struct thread *td, int fd, struct file **fpp)
2035{
2036
2037 return(_fget(td, fd, fpp, FREAD, 1));
2038}
2039
2040int
2041fget_write(struct thread *td, int fd, struct file **fpp)
2042{
2043
2044 return(_fget(td, fd, fpp, FWRITE, 1));
2045}
2046
2047/*
2048 * Like fget() but loads the underlying vnode, or returns an error if the
2049 * descriptor does not represent a vnode. Note that pipes use vnodes but
2050 * never have VM objects. The returned vnode will be vref()'d.
2051 *
2052 * XXX: what about the unused flags ?
2053 */
2054static __inline int
2055_fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2056{
2057 struct file *fp;
2058 int error;
2059
2060 *vpp = NULL;
2061 if ((error = _fget(td, fd, &fp, flags, 0)) != 0)
2062 return (error);
2063 if (fp->f_vnode == NULL) {
2064 error = EINVAL;
2065 } else {
2066 *vpp = fp->f_vnode;
2067 vref(*vpp);
2068 }
2069 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2070 return (error);
2071}
2072
2073int
2074fgetvp(struct thread *td, int fd, struct vnode **vpp)
2075{
2076
2077 return (_fgetvp(td, fd, vpp, 0));
2078}
2079
2080int
2081fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2082{
2083
2084 return (_fgetvp(td, fd, vpp, FREAD));
2085}
2086
2087#ifdef notyet
2088int
2089fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2090{
2091
2092 return (_fgetvp(td, fd, vpp, FWRITE));
2093}
2094#endif
2095
2096/*
2097 * Like fget() but loads the underlying socket, or returns an error if the
2098 * descriptor does not represent a socket.
2099 *
2100 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2101 * in the future.
2102 *
2103 * XXXRW: fgetsock() and fputsock() are deprecated, as consumers should rely
2104 * on their file descriptor reference to prevent the socket from being free'd
2105 * during use.
2106 */
2107int
2108fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2109{
2110 struct file *fp;
2111 int error;
2112
2113 *spp = NULL;
2114 if (fflagp != NULL)
2115 *fflagp = 0;
2116 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2117 return (error);
2118 if (fp->f_type != DTYPE_SOCKET) {
2119 error = ENOTSOCK;
2120 } else {
2121 *spp = fp->f_data;
2122 if (fflagp)
2123 *fflagp = fp->f_flag;
2124 SOCK_LOCK(*spp);
2125 soref(*spp);
2126 SOCK_UNLOCK(*spp);
2127 }
2128 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2129 return (error);
2130}
2131
2132/*
2133 * Drop the reference count on the socket and XXX release the SX lock in the
2134 * future. The last reference closes the socket.
2135 *
2136 * XXXRW: fputsock() is deprecated, see comment for fgetsock().
2137 */
2138void
2139fputsock(struct socket *so)
2140{
2141
2142 ACCEPT_LOCK();
2143 SOCK_LOCK(so);
2144 sorele(so);
2145}
2146
2147/*
2148 * Handle the last reference to a file being closed.
2149 */
2150int
2151_fdrop(struct file *fp, struct thread *td)
2152{
2153 int error;
2154
2155 error = 0;
2156 if (fp->f_count != 0)
2157 panic("fdrop: count %d", fp->f_count);
2158 if (fp->f_ops != &badfileops)
2159 error = fo_close(fp, td);
2160 atomic_subtract_int(&openfiles, 1);
2161 crfree(fp->f_cred);
2162 uma_zfree(file_zone, fp);
2163
2164 return (error);
2165}
2166
2167/*
2168 * Apply an advisory lock on a file descriptor.
2169 *
2170 * Just attempt to get a record lock of the requested type on the entire file
2171 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2172 */
2173#ifndef _SYS_SYSPROTO_H_
2174struct flock_args {
2175 int fd;
2176 int how;
2177};
2178#endif
2179/* ARGSUSED */
2180int
2181flock(struct thread *td, struct flock_args *uap)
2182{
2183 struct file *fp;
2184 struct vnode *vp;
2185 struct flock lf;
2186 int vfslocked;
2187 int error;
2188
2189 if ((error = fget(td, uap->fd, &fp)) != 0)
2190 return (error);
2191 if (fp->f_type != DTYPE_VNODE) {
2192 fdrop(fp, td);
2193 return (EOPNOTSUPP);
2194 }
2195
2196 vp = fp->f_vnode;
2197 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2198 lf.l_whence = SEEK_SET;
2199 lf.l_start = 0;
2200 lf.l_len = 0;
2201 if (uap->how & LOCK_UN) {
2202 lf.l_type = F_UNLCK;
2203 atomic_clear_int(&fp->f_flag, FHASLOCK);
2204 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2205 goto done2;
2206 }
2207 if (uap->how & LOCK_EX)
2208 lf.l_type = F_WRLCK;
2209 else if (uap->how & LOCK_SH)
2210 lf.l_type = F_RDLCK;
2211 else {
2212 error = EBADF;
2213 goto done2;
2214 }
2215 atomic_set_int(&fp->f_flag, FHASLOCK);
2216 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2217 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2218done2:
2219 fdrop(fp, td);
2220 VFS_UNLOCK_GIANT(vfslocked);
2221 return (error);
2222}
2223/*
2224 * Duplicate the specified descriptor to a free descriptor.
2225 */
2226int
2227dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2228{
2229 struct file *wfp;
2230 struct file *fp;
2231
2232 /*
2233 * If the to-be-dup'd fd number is greater than the allowed number
2234 * of file descriptors, or the fd to be dup'd has already been
2235 * closed, then reject.
2236 */
2237 FILEDESC_XLOCK(fdp);
2238 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2239 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2240 FILEDESC_XUNLOCK(fdp);
2241 return (EBADF);
2242 }
2243
2244 /*
2245 * There are two cases of interest here.
2246 *
2247 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2248 *
2249 * For ENXIO steal away the file structure from (dfd) and store it in
2250 * (indx). (dfd) is effectively closed by this operation.
2251 *
2252 * Any other error code is just returned.
2253 */
2254 switch (error) {
2255 case ENODEV:
2256 /*
2257 * Check that the mode the file is being opened for is a
2258 * subset of the mode of the existing descriptor.
2259 */
2260 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2261 FILEDESC_XUNLOCK(fdp);
2262 return (EACCES);
2263 }
2264 fp = fdp->fd_ofiles[indx];
2265 fdp->fd_ofiles[indx] = wfp;
2266 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2267 if (fp == NULL)
2268 fdused(fdp, indx);
2269 fhold(wfp);
2270 FILEDESC_XUNLOCK(fdp);
2271 if (fp != NULL)
2272 /*
2273 * We now own the reference to fp that the ofiles[]
2274 * array used to own. Release it.
2275 */
2276 fdrop(fp, td);
2277 return (0);
2278
2279 case ENXIO:
2280 /*
2281 * Steal away the file pointer from dfd and stuff it into indx.
2282 */
2283 fp = fdp->fd_ofiles[indx];
2284 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2285 fdp->fd_ofiles[dfd] = NULL;
2286 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2287 fdp->fd_ofileflags[dfd] = 0;
2288 fdunused(fdp, dfd);
2289 if (fp == NULL)
2290 fdused(fdp, indx);
2291 FILEDESC_XUNLOCK(fdp);
2292
2293 /*
2294 * We now own the reference to fp that the ofiles[] array
2295 * used to own. Release it.
2296 */
2297 if (fp != NULL)
2298 fdrop(fp, td);
2299 return (0);
2300
2301 default:
2302 FILEDESC_XUNLOCK(fdp);
2303 return (error);
2304 }
2305 /* NOTREACHED */
2306}
2307
2308/*
2309 * Scan all active processes to see if any of them have a current or root
2310 * directory of `olddp'. If so, replace them with the new mount point.
2311 */
2312void
2313mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2314{
2315 struct filedesc *fdp;
2316 struct proc *p;
2317 int nrele;
2318
2319 if (vrefcnt(olddp) == 1)
2320 return;
2321 sx_slock(&allproc_lock);
2322 FOREACH_PROC_IN_SYSTEM(p) {
2323 fdp = fdhold(p);
2324 if (fdp == NULL)
2325 continue;
2326 nrele = 0;
2327 FILEDESC_XLOCK(fdp);
2328 if (fdp->fd_cdir == olddp) {
2329 vref(newdp);
2330 fdp->fd_cdir = newdp;
2331 nrele++;
2332 }
2333 if (fdp->fd_rdir == olddp) {
2334 vref(newdp);
2335 fdp->fd_rdir = newdp;
2336 nrele++;
2337 }
2338 FILEDESC_XUNLOCK(fdp);
2339 fddrop(fdp);
2340 while (nrele--)
2341 vrele(olddp);
2342 }
2343 sx_sunlock(&allproc_lock);
2344 if (rootvnode == olddp) {
2345 vrele(rootvnode);
2346 vref(newdp);
2347 rootvnode = newdp;
2348 }
2349}
2350
2351struct filedesc_to_leader *
2352filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2353{
2354 struct filedesc_to_leader *fdtol;
2355
2356 MALLOC(fdtol, struct filedesc_to_leader *,
2357 sizeof(struct filedesc_to_leader),
2358 M_FILEDESC_TO_LEADER,
2359 M_WAITOK);
2360 fdtol->fdl_refcount = 1;
2361 fdtol->fdl_holdcount = 0;
2362 fdtol->fdl_wakeup = 0;
2363 fdtol->fdl_leader = leader;
2364 if (old != NULL) {
2365 FILEDESC_XLOCK(fdp);
2366 fdtol->fdl_next = old->fdl_next;
2367 fdtol->fdl_prev = old;
2368 old->fdl_next = fdtol;
2369 fdtol->fdl_next->fdl_prev = fdtol;
2370 FILEDESC_XUNLOCK(fdp);
2371 } else {
2372 fdtol->fdl_next = fdtol;
2373 fdtol->fdl_prev = fdtol;
2374 }
2375 return (fdtol);
2376}
2377
2378/*
2379 * Get file structures globally.
2380 */
2381static int
2382sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2383{
2384 struct xfile xf;
2385 struct filedesc *fdp;
2386 struct file *fp;
2387 struct proc *p;
2388 int error, n;
2389
2390 error = sysctl_wire_old_buffer(req, 0);
2391 if (error != 0)
2392 return (error);
2393 if (req->oldptr == NULL) {
2394 n = 0;
2395 sx_slock(&allproc_lock);
2396 FOREACH_PROC_IN_SYSTEM(p) {
2397 if (p->p_state == PRS_NEW)
2398 continue;
2399 fdp = fdhold(p);
2400 if (fdp == NULL)
2401 continue;
2402 /* overestimates sparse tables. */
2403 if (fdp->fd_lastfile > 0)
2404 n += fdp->fd_lastfile;
2405 fddrop(fdp);
2406 }
2407 sx_sunlock(&allproc_lock);
2408 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2409 }
2410 error = 0;
2411 bzero(&xf, sizeof(xf));
2412 xf.xf_size = sizeof(xf);
2413 sx_slock(&allproc_lock);
2414 FOREACH_PROC_IN_SYSTEM(p) {
2415 if (p->p_state == PRS_NEW)
2416 continue;
2417 PROC_LOCK(p);
2418 if (p_cansee(req->td, p) != 0) {
2419 PROC_UNLOCK(p);
2420 continue;
2421 }
2422 xf.xf_pid = p->p_pid;
2423 xf.xf_uid = p->p_ucred->cr_uid;
2424 PROC_UNLOCK(p);
2425 fdp = fdhold(p);
2426 if (fdp == NULL)
2427 continue;
2428 FILEDESC_SLOCK(fdp);
2429 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2430 if ((fp = fdp->fd_ofiles[n]) == NULL)
2431 continue;
2432 xf.xf_fd = n;
2433 xf.xf_file = fp;
2434 xf.xf_data = fp->f_data;
2435 xf.xf_vnode = fp->f_vnode;
2436 xf.xf_type = fp->f_type;
2437 xf.xf_count = fp->f_count;
2438 xf.xf_msgcount = 0;
2439 xf.xf_offset = fp->f_offset;
2440 xf.xf_flag = fp->f_flag;
2441 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2442 if (error)
2443 break;
2444 }
2445 FILEDESC_SUNLOCK(fdp);
2446 fddrop(fdp);
2447 if (error)
2448 break;
2449 }
2450 sx_sunlock(&allproc_lock);
2451 return (error);
2452}
2453
2454SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2455 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2456
2457static int
2458export_vnode_for_sysctl(struct vnode *vp, int type,
2459 struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req)
2460{
2461 int error;
2462 char *fullpath, *freepath;
2463 int vfslocked;
2464
2465 bzero(kif, sizeof(*kif));
2466 kif->kf_structsize = sizeof(*kif);
2467
2468 vref(vp);
2469 kif->kf_fd = type;
2470 kif->kf_type = KF_TYPE_VNODE;
2471 /* This function only handles directories. */
2472 KASSERT(vp->v_type == VDIR, ("export_vnode_for_sysctl: vnode not directory"));
2473 kif->kf_vnode_type = KF_VTYPE_VDIR;
2474
2475 /*
2476 * This is not a true file descriptor, so we set a bogus refcount
2477 * and offset to indicate these fields should be ignored.
2478 */
2479 kif->kf_ref_count = -1;
2480 kif->kf_offset = -1;
2481
2482 freepath = NULL;
2483 fullpath = "-";
2484 FILEDESC_SUNLOCK(fdp);
2485 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2486 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2487 vn_fullpath(curthread, vp, &fullpath, &freepath);
2488 vput(vp);
2489 VFS_UNLOCK_GIANT(vfslocked);
2490 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2491 if (freepath != NULL)
2492 free(freepath, M_TEMP);
2493 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2494 FILEDESC_SLOCK(fdp);
2495 return (error);
2496}
2497
2498/*
2499 * Get per-process file descriptors for use by procstat(1), et al.
2500 */
2501static int
2502sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
2503{
2504 char *fullpath, *freepath;
2505 struct kinfo_file *kif;
2506 struct filedesc *fdp;
2507 int error, i, *name;
2508 struct socket *so;
2509 struct vnode *vp;
2510 struct file *fp;
2511 struct proc *p;
2512 int vfslocked;
2513
2514 name = (int *)arg1;
2515 if ((p = pfind((pid_t)name[0])) == NULL)
2516 return (ESRCH);
2517 if ((error = p_candebug(curthread, p))) {
2518 PROC_UNLOCK(p);
2519 return (error);
2520 }
2521 fdp = fdhold(p);
2522 PROC_UNLOCK(p);
2523 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2524 FILEDESC_SLOCK(fdp);
2525 if (fdp->fd_cdir != NULL)
2526 export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2527 fdp, req);
2528 if (fdp->fd_rdir != NULL)
2529 export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2530 fdp, req);
2531 if (fdp->fd_jdir != NULL)
2532 export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2533 fdp, req);
2534 for (i = 0; i < fdp->fd_nfiles; i++) {
2535 if ((fp = fdp->fd_ofiles[i]) == NULL)
2536 continue;
2537 bzero(kif, sizeof(*kif));
2538 kif->kf_structsize = sizeof(*kif);
2539 vp = NULL;
2540 so = NULL;
2541 kif->kf_fd = i;
2542 switch (fp->f_type) {
2543 case DTYPE_VNODE:
2544 kif->kf_type = KF_TYPE_VNODE;
2545 vp = fp->f_vnode;
2546 break;
2547
2548 case DTYPE_SOCKET:
2549 kif->kf_type = KF_TYPE_SOCKET;
2550 so = fp->f_data;
2551 break;
2552
2553 case DTYPE_PIPE:
2554 kif->kf_type = KF_TYPE_PIPE;
2555 break;
2556
2557 case DTYPE_FIFO:
2558 kif->kf_type = KF_TYPE_FIFO;
2559 vp = fp->f_vnode;
2560 vref(vp);
2561 break;
2562
2563 case DTYPE_KQUEUE:
2564 kif->kf_type = KF_TYPE_KQUEUE;
2565 break;
2566
2567 case DTYPE_CRYPTO:
2568 kif->kf_type = KF_TYPE_CRYPTO;
2569 break;
2570
2571 case DTYPE_MQUEUE:
2572 kif->kf_type = KF_TYPE_MQUEUE;
2573 break;
2574
2575 case DTYPE_SHM:
2576 kif->kf_type = KF_TYPE_SHM;
2577 break;
2578
2579 default:
2580 kif->kf_type = KF_TYPE_UNKNOWN;
2581 break;
2582 }
2583 kif->kf_ref_count = fp->f_count;
2584 if (fp->f_flag & FREAD)
2585 kif->kf_flags |= KF_FLAG_READ;
2586 if (fp->f_flag & FWRITE)
2587 kif->kf_flags |= KF_FLAG_WRITE;
2588 if (fp->f_flag & FAPPEND)
2589 kif->kf_flags |= KF_FLAG_APPEND;
2590 if (fp->f_flag & FASYNC)
2591 kif->kf_flags |= KF_FLAG_ASYNC;
2592 if (fp->f_flag & FFSYNC)
2593 kif->kf_flags |= KF_FLAG_FSYNC;
2594 if (fp->f_flag & FNONBLOCK)
2595 kif->kf_flags |= KF_FLAG_NONBLOCK;
2596 if (fp->f_flag & O_DIRECT)
2597 kif->kf_flags |= KF_FLAG_DIRECT;
2598 if (fp->f_flag & FHASLOCK)
2599 kif->kf_flags |= KF_FLAG_HASLOCK;
2600 kif->kf_offset = fp->f_offset;
2601 if (vp != NULL) {
2602 vref(vp);
2603 switch (vp->v_type) {
2604 case VNON:
2605 kif->kf_vnode_type = KF_VTYPE_VNON;
2606 break;
2607 case VREG:
2608 kif->kf_vnode_type = KF_VTYPE_VREG;
2609 break;
2610 case VDIR:
2611 kif->kf_vnode_type = KF_VTYPE_VDIR;
2612 break;
2613 case VBLK:
2614 kif->kf_vnode_type = KF_VTYPE_VBLK;
2615 break;
2616 case VCHR:
2617 kif->kf_vnode_type = KF_VTYPE_VCHR;
2618 break;
2619 case VLNK:
2620 kif->kf_vnode_type = KF_VTYPE_VLNK;
2621 break;
2622 case VSOCK:
2623 kif->kf_vnode_type = KF_VTYPE_VSOCK;
2624 break;
2625 case VFIFO:
2626 kif->kf_vnode_type = KF_VTYPE_VFIFO;
2627 break;
2628 case VBAD:
2629 kif->kf_vnode_type = KF_VTYPE_VBAD;
2630 break;
2631 default:
2632 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2633 break;
2634 }
2635 /*
2636 * It is OK to drop the filedesc lock here as we will
2637 * re-validate and re-evaluate its properties when
2638 * the loop continues.
2639 */
2640 freepath = NULL;
2641 fullpath = "-";
2642 FILEDESC_SUNLOCK(fdp);
2643 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2644 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2645 vn_fullpath(curthread, vp, &fullpath, &freepath);
2646 vput(vp);
2647 VFS_UNLOCK_GIANT(vfslocked);
2648 strlcpy(kif->kf_path, fullpath,
2649 sizeof(kif->kf_path));
2650 if (freepath != NULL)
2651 free(freepath, M_TEMP);
2652 FILEDESC_SLOCK(fdp);
2653 }
2654 if (so != NULL) {
2655 struct sockaddr *sa;
2656
2657 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
2658 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
2659 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
2660 free(sa, M_SONAME);
2661 }
2662 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
2663 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
2664 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
2665 free(sa, M_SONAME);
2666 }
2667 kif->kf_sock_domain =
2668 so->so_proto->pr_domain->dom_family;
2669 kif->kf_sock_type = so->so_type;
2670 kif->kf_sock_protocol = so->so_proto->pr_protocol;
2671 }
2672 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2673 if (error)
2674 break;
2675 }
2676 FILEDESC_SUNLOCK(fdp);
2677 fddrop(fdp);
2678 free(kif, M_TEMP);
2679 return (0);
2680}
2681
2682static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
2683 sysctl_kern_proc_filedesc, "Process filedesc entries");
2684
2685#ifdef DDB
2686/*
2687 * For the purposes of debugging, generate a human-readable string for the
2688 * file type.
2689 */
2690static const char *
2691file_type_to_name(short type)
2692{
2693
2694 switch (type) {
2695 case 0:
2696 return ("zero");
2697 case DTYPE_VNODE:
2698 return ("vnod");
2699 case DTYPE_SOCKET:
2700 return ("sock");
2701 case DTYPE_PIPE:
2702 return ("pipe");
2703 case DTYPE_FIFO:
2704 return ("fifo");
2705 case DTYPE_KQUEUE:
2706 return ("kque");
2707 case DTYPE_CRYPTO:
2708 return ("crpt");
2709 case DTYPE_MQUEUE:
2710 return ("mque");
2711 case DTYPE_SHM:
2712 return ("shm");
2713 default:
2714 return ("unkn");
2715 }
2716}
2717
2718/*
2719 * For the purposes of debugging, identify a process (if any, perhaps one of
2720 * many) that references the passed file in its file descriptor array. Return
2721 * NULL if none.
2722 */
2723static struct proc *
2724file_to_first_proc(struct file *fp)
2725{
2726 struct filedesc *fdp;
2727 struct proc *p;
2728 int n;
2729
2730 FOREACH_PROC_IN_SYSTEM(p) {
2731 if (p->p_state == PRS_NEW)
2732 continue;
2733 fdp = p->p_fd;
2734 if (fdp == NULL)
2735 continue;
2736 for (n = 0; n < fdp->fd_nfiles; n++) {
2737 if (fp == fdp->fd_ofiles[n])
2738 return (p);
2739 }
2740 }
2741 return (NULL);
2742}
2743
2744static void
2745db_print_file(struct file *fp, int header)
2746{
2747 struct proc *p;
2748
2749 if (header)
2750 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
2751 "File", "Type", "Data", "Flag", "GCFl", "Count",
2752 "MCount", "Vnode", "FPID", "FCmd");
2753 p = file_to_first_proc(fp);
2754 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2755 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2756 0, fp->f_count, 0, fp->f_vnode,
2757 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2758}
2759
2760DB_SHOW_COMMAND(file, db_show_file)
2761{
2762 struct file *fp;
2763
2764 if (!have_addr) {
2765 db_printf("usage: show file <addr>\n");
2766 return;
2767 }
2768 fp = (struct file *)addr;
2769 db_print_file(fp, 1);
2770}
2771
2772DB_SHOW_COMMAND(files, db_show_files)
2773{
2774 struct filedesc *fdp;
2775 struct file *fp;
2776 struct proc *p;
2777 int header;
2778 int n;
2779
2780 header = 1;
2781 FOREACH_PROC_IN_SYSTEM(p) {
2782 if (p->p_state == PRS_NEW)
2783 continue;
2784 if ((fdp = p->p_fd) == NULL)
2785 continue;
2786 for (n = 0; n < fdp->fd_nfiles; ++n) {
2787 if ((fp = fdp->fd_ofiles[n]) == NULL)
2788 continue;
2789 db_print_file(fp, header);
2790 header = 0;
2791 }
2792 }
2793}
2794#endif
2795
2796SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2797 &maxfilesperproc, 0, "Maximum files allowed open per process");
2798
2799SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2800 &maxfiles, 0, "Maximum number of files");
2801
2802SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2803 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
2804
2805/* ARGSUSED*/
2806static void
2807filelistinit(void *dummy)
2808{
2809
2810 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2811 NULL, NULL, UMA_ALIGN_PTR, 0);
2812 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2813 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2814}
2815SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2816
2817/*-------------------------------------------------------------------*/
2818
2819static int
2820badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2821{
2822
2823 return (EBADF);
2824}
2825
2826static int
2827badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td)
2828{
2829
2830 return (EINVAL);
2831}
2832
2833static int
2834badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2835{
2836
2837 return (EBADF);
2838}
2839
2840static int
2841badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2842{
2843
2844 return (0);
2845}
2846
2847static int
2848badfo_kqfilter(struct file *fp, struct knote *kn)
2849{
2850
2851 return (EBADF);
2852}
2853
2854static int
2855badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2856{
2857
2858 return (EBADF);
2859}
2860
2861static int
2862badfo_close(struct file *fp, struct thread *td)
2863{
2864
2865 return (EBADF);
2866}
2867
2868struct fileops badfileops = {
2869 .fo_read = badfo_readwrite,
2870 .fo_write = badfo_readwrite,
2871 .fo_truncate = badfo_truncate,
2872 .fo_ioctl = badfo_ioctl,
2873 .fo_poll = badfo_poll,
2874 .fo_kqfilter = badfo_kqfilter,
2875 .fo_stat = badfo_stat,
2876 .fo_close = badfo_close,
2877};
2878
2879
2880/*-------------------------------------------------------------------*/
2881
2882/*
2883 * File Descriptor pseudo-device driver (/dev/fd/).
2884 *
2885 * Opening minor device N dup()s the file (if any) connected to file
2886 * descriptor N belonging to the calling process. Note that this driver
2887 * consists of only the ``open()'' routine, because all subsequent
2888 * references to this file will be direct to the other driver.
2889 *
2890 * XXX: we could give this one a cloning event handler if necessary.
2891 */
2892
2893/* ARGSUSED */
2894static int
2895fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2896{
2897
2898 /*
2899 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2900 * the file descriptor being sought for duplication. The error
2901 * return ensures that the vnode for this device will be released
2902 * by vn_open. Open will detect this special error and take the
2903 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2904 * will simply report the error.
2905 */
2906 td->td_dupfd = dev2unit(dev);
2907 return (ENODEV);
2908}
2909
2910static struct cdevsw fildesc_cdevsw = {
2911 .d_version = D_VERSION,
2912 .d_flags = D_NEEDGIANT,
2913 .d_open = fdopen,
2914 .d_name = "FD",
2915};
2916
2917static void
2918fildesc_drvinit(void *unused)
2919{
2920 struct cdev *dev;
2921
2922 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2923 make_dev_alias(dev, "stdin");
2924 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2925 make_dev_alias(dev, "stdout");
2926 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2927 make_dev_alias(dev, "stderr");
2928}
2929
2930SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)