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: releng/10.2/sys/kern/kern_descrip.c 285964 2015-07-28 17:00:03Z kib $");
39
40 #include "opt_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ddb.h"
43 #include "opt_ktrace.h"
44 #include "opt_procdesc.h"
45
46 #include <sys/param.h>
47 #include <sys/systm.h>
48
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/domain.h>
52 #include <sys/fcntl.h>
53 #include <sys/file.h>
54 #include <sys/filedesc.h>
55 #include <sys/filio.h>
56 #include <sys/jail.h>
57 #include <sys/kernel.h>
58 #include <sys/ksem.h>
59 #include <sys/limits.h>
60 #include <sys/lock.h>
61 #include <sys/malloc.h>
62 #include <sys/mman.h>
63 #include <sys/mount.h>
64 #include <sys/mqueue.h>
65 #include <sys/mutex.h>
66 #include <sys/namei.h>
67 #include <sys/selinfo.h>
68 #include <sys/pipe.h>
69 #include <sys/priv.h>
70 #include <sys/proc.h>
71 #include <sys/procdesc.h>
72 #include <sys/protosw.h>
73 #include <sys/racct.h>
74 #include <sys/resourcevar.h>
75 #include <sys/sbuf.h>
76 #include <sys/signalvar.h>
77 #include <sys/socketvar.h>
78 #include <sys/stat.h>
79 #include <sys/sx.h>
80 #include <sys/syscallsubr.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysproto.h>
83 #include <sys/tty.h>
84 #include <sys/unistd.h>
85 #include <sys/un.h>
86 #include <sys/unpcb.h>
87 #include <sys/user.h>
88 #include <sys/vnode.h>
89 #ifdef KTRACE
90 #include <sys/ktrace.h>
91 #endif
92
93 #include <net/vnet.h>
94
95 #include <netinet/in.h>
96 #include <netinet/in_pcb.h>
97
98 #include <security/audit/audit.h>
99
100 #include <vm/uma.h>
101 #include <vm/vm.h>
102
103 #include <ddb/ddb.h>
104
105 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
106 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
107 "file desc to leader structures");
108 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
109 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
110
111 MALLOC_DECLARE(M_FADVISE);
112
113 static uma_zone_t file_zone;
114
115 void (*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value);
116
117 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
118 struct thread *td, int holdleaders);
119 static int fd_first_free(struct filedesc *fdp, int low, int size);
120 static int fd_last_used(struct filedesc *fdp, int size);
121 static void fdgrowtable(struct filedesc *fdp, int nfd);
122 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
123 static void fdunused(struct filedesc *fdp, int fd);
124 static void fdused(struct filedesc *fdp, int fd);
125 static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
126 static int fill_procdesc_info(struct procdesc *pdp,
127 struct kinfo_file *kif);
128 static int fill_pts_info(struct tty *tp, struct kinfo_file *kif);
129 static int fill_sem_info(struct file *fp, struct kinfo_file *kif);
130 static int fill_shm_info(struct file *fp, struct kinfo_file *kif);
131 static int fill_socket_info(struct socket *so, struct kinfo_file *kif);
132 static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
133 static int getmaxfd(struct proc *p);
134
135 /*
136 * Each process has:
137 *
138 * - An array of open file descriptors (fd_ofiles)
139 * - An array of file flags (fd_ofileflags)
140 * - A bitmap recording which descriptors are in use (fd_map)
141 *
142 * A process starts out with NDFILE descriptors. The value of NDFILE has
143 * been selected based the historical limit of 20 open files, and an
144 * assumption that the majority of processes, especially short-lived
145 * processes like shells, will never need more.
146 *
147 * If this initial allocation is exhausted, a larger descriptor table and
148 * map are allocated dynamically, and the pointers in the process's struct
149 * filedesc are updated to point to those. This is repeated every time
150 * the process runs out of file descriptors (provided it hasn't hit its
151 * resource limit).
152 *
153 * Since threads may hold references to individual descriptor table
154 * entries, the tables are never freed. Instead, they are placed on a
155 * linked list and freed only when the struct filedesc is released.
156 */
157 #define NDFILE 20
158 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
159 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
160 #define NDSLOT(x) ((x) / NDENTRIES)
161 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
162 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
163
164 /*
165 * SLIST entry used to keep track of ofiles which must be reclaimed when
166 * the process exits.
167 */
168 struct freetable {
169 struct filedescent *ft_table;
170 SLIST_ENTRY(freetable) ft_next;
171 };
172
173 /*
174 * Initial allocation: a filedesc structure + the head of SLIST used to
175 * keep track of old ofiles + enough space for NDFILE descriptors.
176 */
177 struct filedesc0 {
178 struct filedesc fd_fd;
179 SLIST_HEAD(, freetable) fd_free;
180 struct filedescent fd_dfiles[NDFILE];
181 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
182 };
183
184 /*
185 * Descriptor management.
186 */
187 volatile int openfiles; /* actual number of open files */
188 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
189 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
190
191 /* A mutex to protect the association between a proc and filedesc. */
192 static struct mtx fdesc_mtx;
193
194 /*
195 * If low >= size, just return low. Otherwise find the first zero bit in the
196 * given bitmap, starting at low and not exceeding size - 1. Return size if
197 * not found.
198 */
199 static int
200 fd_first_free(struct filedesc *fdp, int low, int size)
201 {
202 NDSLOTTYPE *map = fdp->fd_map;
203 NDSLOTTYPE mask;
204 int off, maxoff;
205
206 if (low >= size)
207 return (low);
208
209 off = NDSLOT(low);
210 if (low % NDENTRIES) {
211 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
212 if ((mask &= ~map[off]) != 0UL)
213 return (off * NDENTRIES + ffsl(mask) - 1);
214 ++off;
215 }
216 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
217 if (map[off] != ~0UL)
218 return (off * NDENTRIES + ffsl(~map[off]) - 1);
219 return (size);
220 }
221
222 /*
223 * Find the highest non-zero bit in the given bitmap, starting at 0 and
224 * not exceeding size - 1. Return -1 if not found.
225 */
226 static int
227 fd_last_used(struct filedesc *fdp, int size)
228 {
229 NDSLOTTYPE *map = fdp->fd_map;
230 NDSLOTTYPE mask;
231 int off, minoff;
232
233 off = NDSLOT(size);
234 if (size % NDENTRIES) {
235 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
236 if ((mask &= map[off]) != 0)
237 return (off * NDENTRIES + flsl(mask) - 1);
238 --off;
239 }
240 for (minoff = NDSLOT(0); off >= minoff; --off)
241 if (map[off] != 0)
242 return (off * NDENTRIES + flsl(map[off]) - 1);
243 return (-1);
244 }
245
246 static int
247 fdisused(struct filedesc *fdp, int fd)
248 {
249
250 FILEDESC_LOCK_ASSERT(fdp);
251
252 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
253 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
254
255 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
256 }
257
258 /*
259 * Mark a file descriptor as used.
260 */
261 static void
262 fdused(struct filedesc *fdp, int fd)
263 {
264
265 FILEDESC_XLOCK_ASSERT(fdp);
266
267 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
268
269 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
270 if (fd > fdp->fd_lastfile)
271 fdp->fd_lastfile = fd;
272 if (fd == fdp->fd_freefile)
273 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
274 }
275
276 /*
277 * Mark a file descriptor as unused.
278 */
279 static void
280 fdunused(struct filedesc *fdp, int fd)
281 {
282
283 FILEDESC_XLOCK_ASSERT(fdp);
284
285 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
286 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
287 ("fd=%d is still in use", fd));
288
289 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
290 if (fd < fdp->fd_freefile)
291 fdp->fd_freefile = fd;
292 if (fd == fdp->fd_lastfile)
293 fdp->fd_lastfile = fd_last_used(fdp, fd);
294 }
295
296 /*
297 * Free a file descriptor.
298 *
299 * Avoid some work if fdp is about to be destroyed.
300 */
301 static inline void
302 _fdfree(struct filedesc *fdp, int fd, int last)
303 {
304 struct filedescent *fde;
305
306 fde = &fdp->fd_ofiles[fd];
307 #ifdef CAPABILITIES
308 if (!last)
309 seq_write_begin(&fde->fde_seq);
310 #endif
311 filecaps_free(&fde->fde_caps);
312 if (last)
313 return;
314 bzero(fde, fde_change_size);
315 fdunused(fdp, fd);
316 #ifdef CAPABILITIES
317 seq_write_end(&fde->fde_seq);
318 #endif
319 }
320
321 static inline void
322 fdfree(struct filedesc *fdp, int fd)
323 {
324
325 _fdfree(fdp, fd, 0);
326 }
327
328 static inline void
329 fdfree_last(struct filedesc *fdp, int fd)
330 {
331
332 _fdfree(fdp, fd, 1);
333 }
334
335 /*
336 * System calls on descriptors.
337 */
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
340 int dummy;
341 };
342 #endif
343 /* ARGSUSED */
344 int
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
346 {
347 struct proc *p = td->td_proc;
348 uint64_t lim;
349
350 PROC_LOCK(p);
351 td->td_retval[0] =
352 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
353 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
354 PROC_UNLOCK(p);
355 if (lim < td->td_retval[0])
356 td->td_retval[0] = lim;
357 return (0);
358 }
359
360 /*
361 * Duplicate a file descriptor to a particular value.
362 *
363 * Note: keep in mind that a potential race condition exists when closing
364 * descriptors from a shared descriptor table (via rfork).
365 */
366 #ifndef _SYS_SYSPROTO_H_
367 struct dup2_args {
368 u_int from;
369 u_int to;
370 };
371 #endif
372 /* ARGSUSED */
373 int
374 sys_dup2(struct thread *td, struct dup2_args *uap)
375 {
376
377 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
378 td->td_retval));
379 }
380
381 /*
382 * Duplicate a file descriptor.
383 */
384 #ifndef _SYS_SYSPROTO_H_
385 struct dup_args {
386 u_int fd;
387 };
388 #endif
389 /* ARGSUSED */
390 int
391 sys_dup(struct thread *td, struct dup_args *uap)
392 {
393
394 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
395 }
396
397 /*
398 * The file control system call.
399 */
400 #ifndef _SYS_SYSPROTO_H_
401 struct fcntl_args {
402 int fd;
403 int cmd;
404 long arg;
405 };
406 #endif
407 /* ARGSUSED */
408 int
409 sys_fcntl(struct thread *td, struct fcntl_args *uap)
410 {
411
412 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
413 }
414
415 int
416 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
417 {
418 struct flock fl;
419 struct __oflock ofl;
420 intptr_t arg1;
421 int error, newcmd;
422
423 error = 0;
424 newcmd = cmd;
425 switch (cmd) {
426 case F_OGETLK:
427 case F_OSETLK:
428 case F_OSETLKW:
429 /*
430 * Convert old flock structure to new.
431 */
432 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
433 fl.l_start = ofl.l_start;
434 fl.l_len = ofl.l_len;
435 fl.l_pid = ofl.l_pid;
436 fl.l_type = ofl.l_type;
437 fl.l_whence = ofl.l_whence;
438 fl.l_sysid = 0;
439
440 switch (cmd) {
441 case F_OGETLK:
442 newcmd = F_GETLK;
443 break;
444 case F_OSETLK:
445 newcmd = F_SETLK;
446 break;
447 case F_OSETLKW:
448 newcmd = F_SETLKW;
449 break;
450 }
451 arg1 = (intptr_t)&fl;
452 break;
453 case F_GETLK:
454 case F_SETLK:
455 case F_SETLKW:
456 case F_SETLK_REMOTE:
457 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
458 arg1 = (intptr_t)&fl;
459 break;
460 default:
461 arg1 = arg;
462 break;
463 }
464 if (error)
465 return (error);
466 error = kern_fcntl(td, fd, newcmd, arg1);
467 if (error)
468 return (error);
469 if (cmd == F_OGETLK) {
470 ofl.l_start = fl.l_start;
471 ofl.l_len = fl.l_len;
472 ofl.l_pid = fl.l_pid;
473 ofl.l_type = fl.l_type;
474 ofl.l_whence = fl.l_whence;
475 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
476 } else if (cmd == F_GETLK) {
477 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
478 }
479 return (error);
480 }
481
482 int
483 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
484 {
485 struct filedesc *fdp;
486 struct flock *flp;
487 struct file *fp, *fp2;
488 struct filedescent *fde;
489 struct proc *p;
490 struct vnode *vp;
491 cap_rights_t rights;
492 int error, flg, tmp;
493 uint64_t bsize;
494 off_t foffset;
495
496 error = 0;
497 flg = F_POSIX;
498 p = td->td_proc;
499 fdp = p->p_fd;
500
501 switch (cmd) {
502 case F_DUPFD:
503 tmp = arg;
504 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
505 break;
506
507 case F_DUPFD_CLOEXEC:
508 tmp = arg;
509 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp,
510 td->td_retval);
511 break;
512
513 case F_DUP2FD:
514 tmp = arg;
515 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
516 break;
517
518 case F_DUP2FD_CLOEXEC:
519 tmp = arg;
520 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp,
521 td->td_retval);
522 break;
523
524 case F_GETFD:
525 FILEDESC_SLOCK(fdp);
526 if ((fp = fget_locked(fdp, fd)) == NULL) {
527 FILEDESC_SUNLOCK(fdp);
528 error = EBADF;
529 break;
530 }
531 fde = &fdp->fd_ofiles[fd];
532 td->td_retval[0] =
533 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
534 FILEDESC_SUNLOCK(fdp);
535 break;
536
537 case F_SETFD:
538 FILEDESC_XLOCK(fdp);
539 if ((fp = fget_locked(fdp, fd)) == NULL) {
540 FILEDESC_XUNLOCK(fdp);
541 error = EBADF;
542 break;
543 }
544 fde = &fdp->fd_ofiles[fd];
545 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
546 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
547 FILEDESC_XUNLOCK(fdp);
548 break;
549
550 case F_GETFL:
551 error = fget_unlocked(fdp, fd,
552 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL);
553 if (error != 0)
554 break;
555 td->td_retval[0] = OFLAGS(fp->f_flag);
556 fdrop(fp, td);
557 break;
558
559 case F_SETFL:
560 error = fget_unlocked(fdp, fd,
561 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL);
562 if (error != 0)
563 break;
564 do {
565 tmp = flg = fp->f_flag;
566 tmp &= ~FCNTLFLAGS;
567 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
568 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
569 tmp = fp->f_flag & FNONBLOCK;
570 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
571 if (error != 0) {
572 fdrop(fp, td);
573 break;
574 }
575 tmp = fp->f_flag & FASYNC;
576 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
577 if (error == 0) {
578 fdrop(fp, td);
579 break;
580 }
581 atomic_clear_int(&fp->f_flag, FNONBLOCK);
582 tmp = 0;
583 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
584 fdrop(fp, td);
585 break;
586
587 case F_GETOWN:
588 error = fget_unlocked(fdp, fd,
589 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL);
590 if (error != 0)
591 break;
592 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
593 if (error == 0)
594 td->td_retval[0] = tmp;
595 fdrop(fp, td);
596 break;
597
598 case F_SETOWN:
599 error = fget_unlocked(fdp, fd,
600 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL);
601 if (error != 0)
602 break;
603 tmp = arg;
604 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
605 fdrop(fp, td);
606 break;
607
608 case F_SETLK_REMOTE:
609 error = priv_check(td, PRIV_NFS_LOCKD);
610 if (error)
611 return (error);
612 flg = F_REMOTE;
613 goto do_setlk;
614
615 case F_SETLKW:
616 flg |= F_WAIT;
617 /* FALLTHROUGH F_SETLK */
618
619 case F_SETLK:
620 do_setlk:
621 cap_rights_init(&rights, CAP_FLOCK);
622 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL);
623 if (error != 0)
624 break;
625 if (fp->f_type != DTYPE_VNODE) {
626 error = EBADF;
627 fdrop(fp, td);
628 break;
629 }
630
631 flp = (struct flock *)arg;
632 if (flp->l_whence == SEEK_CUR) {
633 foffset = foffset_get(fp);
634 if (foffset < 0 ||
635 (flp->l_start > 0 &&
636 foffset > OFF_MAX - flp->l_start)) {
637 FILEDESC_SUNLOCK(fdp);
638 error = EOVERFLOW;
639 fdrop(fp, td);
640 break;
641 }
642 flp->l_start += foffset;
643 }
644
645 vp = fp->f_vnode;
646 switch (flp->l_type) {
647 case F_RDLCK:
648 if ((fp->f_flag & FREAD) == 0) {
649 error = EBADF;
650 break;
651 }
652 PROC_LOCK(p->p_leader);
653 p->p_leader->p_flag |= P_ADVLOCK;
654 PROC_UNLOCK(p->p_leader);
655 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
656 flp, flg);
657 break;
658 case F_WRLCK:
659 if ((fp->f_flag & FWRITE) == 0) {
660 error = EBADF;
661 break;
662 }
663 PROC_LOCK(p->p_leader);
664 p->p_leader->p_flag |= P_ADVLOCK;
665 PROC_UNLOCK(p->p_leader);
666 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
667 flp, flg);
668 break;
669 case F_UNLCK:
670 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
671 flp, flg);
672 break;
673 case F_UNLCKSYS:
674 /*
675 * Temporary api for testing remote lock
676 * infrastructure.
677 */
678 if (flg != F_REMOTE) {
679 error = EINVAL;
680 break;
681 }
682 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
683 F_UNLCKSYS, flp, flg);
684 break;
685 default:
686 error = EINVAL;
687 break;
688 }
689 if (error != 0 || flp->l_type == F_UNLCK ||
690 flp->l_type == F_UNLCKSYS) {
691 fdrop(fp, td);
692 break;
693 }
694
695 /*
696 * Check for a race with close.
697 *
698 * The vnode is now advisory locked (or unlocked, but this case
699 * is not really important) as the caller requested.
700 * We had to drop the filedesc lock, so we need to recheck if
701 * the descriptor is still valid, because if it was closed
702 * in the meantime we need to remove advisory lock from the
703 * vnode - close on any descriptor leading to an advisory
704 * locked vnode, removes that lock.
705 * We will return 0 on purpose in that case, as the result of
706 * successful advisory lock might have been externally visible
707 * already. This is fine - effectively we pretend to the caller
708 * that the closing thread was a bit slower and that the
709 * advisory lock succeeded before the close.
710 */
711 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL);
712 if (error != 0) {
713 fdrop(fp, td);
714 break;
715 }
716 if (fp != fp2) {
717 flp->l_whence = SEEK_SET;
718 flp->l_start = 0;
719 flp->l_len = 0;
720 flp->l_type = F_UNLCK;
721 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
722 F_UNLCK, flp, F_POSIX);
723 }
724 fdrop(fp, td);
725 fdrop(fp2, td);
726 break;
727
728 case F_GETLK:
729 error = fget_unlocked(fdp, fd,
730 cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL);
731 if (error != 0)
732 break;
733 if (fp->f_type != DTYPE_VNODE) {
734 error = EBADF;
735 fdrop(fp, td);
736 break;
737 }
738 flp = (struct flock *)arg;
739 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
740 flp->l_type != F_UNLCK) {
741 error = EINVAL;
742 fdrop(fp, td);
743 break;
744 }
745 if (flp->l_whence == SEEK_CUR) {
746 foffset = foffset_get(fp);
747 if ((flp->l_start > 0 &&
748 foffset > OFF_MAX - flp->l_start) ||
749 (flp->l_start < 0 &&
750 foffset < OFF_MIN - flp->l_start)) {
751 FILEDESC_SUNLOCK(fdp);
752 error = EOVERFLOW;
753 fdrop(fp, td);
754 break;
755 }
756 flp->l_start += foffset;
757 }
758 vp = fp->f_vnode;
759 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
760 F_POSIX);
761 fdrop(fp, td);
762 break;
763
764 case F_RDAHEAD:
765 arg = arg ? 128 * 1024: 0;
766 /* FALLTHROUGH */
767 case F_READAHEAD:
768 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL);
769 if (error != 0)
770 break;
771 if (fp->f_type != DTYPE_VNODE) {
772 fdrop(fp, td);
773 error = EBADF;
774 break;
775 }
776 vp = fp->f_vnode;
777 /*
778 * Exclusive lock synchronizes against f_seqcount reads and
779 * writes in sequential_heuristic().
780 */
781 error = vn_lock(vp, LK_EXCLUSIVE);
782 if (error != 0) {
783 fdrop(fp, td);
784 break;
785 }
786 if (arg >= 0) {
787 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
788 fp->f_seqcount = (arg + bsize - 1) / bsize;
789 atomic_set_int(&fp->f_flag, FRDAHEAD);
790 } else {
791 atomic_clear_int(&fp->f_flag, FRDAHEAD);
792 }
793 VOP_UNLOCK(vp, 0);
794 fdrop(fp, td);
795 break;
796
797 default:
798 error = EINVAL;
799 break;
800 }
801 return (error);
802 }
803
804 static int
805 getmaxfd(struct proc *p)
806 {
807 int maxfd;
808
809 PROC_LOCK(p);
810 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
811 PROC_UNLOCK(p);
812
813 return (maxfd);
814 }
815
816 /*
817 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
818 */
819 int
820 do_dup(struct thread *td, int flags, int old, int new,
821 register_t *retval)
822 {
823 struct filedesc *fdp;
824 struct filedescent *oldfde, *newfde;
825 struct proc *p;
826 struct file *fp;
827 struct file *delfp;
828 int error, maxfd;
829
830 p = td->td_proc;
831 fdp = p->p_fd;
832
833 /*
834 * Verify we have a valid descriptor to dup from and possibly to
835 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
836 * return EINVAL when the new descriptor is out of bounds.
837 */
838 if (old < 0)
839 return (EBADF);
840 if (new < 0)
841 return (flags & DUP_FCNTL ? EINVAL : EBADF);
842 maxfd = getmaxfd(p);
843 if (new >= maxfd)
844 return (flags & DUP_FCNTL ? EINVAL : EBADF);
845
846 FILEDESC_XLOCK(fdp);
847 if (fget_locked(fdp, old) == NULL) {
848 FILEDESC_XUNLOCK(fdp);
849 return (EBADF);
850 }
851 oldfde = &fdp->fd_ofiles[old];
852 if (flags & DUP_FIXED && old == new) {
853 *retval = new;
854 if (flags & DUP_CLOEXEC)
855 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
856 FILEDESC_XUNLOCK(fdp);
857 return (0);
858 }
859 fp = oldfde->fde_file;
860 fhold(fp);
861
862 /*
863 * If the caller specified a file descriptor, make sure the file
864 * table is large enough to hold it, and grab it. Otherwise, just
865 * allocate a new descriptor the usual way.
866 */
867 if (flags & DUP_FIXED) {
868 if (new >= fdp->fd_nfiles) {
869 /*
870 * The resource limits are here instead of e.g.
871 * fdalloc(), because the file descriptor table may be
872 * shared between processes, so we can't really use
873 * racct_add()/racct_sub(). Instead of counting the
874 * number of actually allocated descriptors, just put
875 * the limit on the size of the file descriptor table.
876 */
877 #ifdef RACCT
878 if (racct_enable) {
879 PROC_LOCK(p);
880 error = racct_set(p, RACCT_NOFILE, new + 1);
881 PROC_UNLOCK(p);
882 if (error != 0) {
883 FILEDESC_XUNLOCK(fdp);
884 fdrop(fp, td);
885 return (EMFILE);
886 }
887 }
888 #endif
889 fdgrowtable_exp(fdp, new + 1);
890 oldfde = &fdp->fd_ofiles[old];
891 }
892 newfde = &fdp->fd_ofiles[new];
893 if (newfde->fde_file == NULL)
894 fdused(fdp, new);
895 } else {
896 if ((error = fdalloc(td, new, &new)) != 0) {
897 FILEDESC_XUNLOCK(fdp);
898 fdrop(fp, td);
899 return (error);
900 }
901 newfde = &fdp->fd_ofiles[new];
902 }
903
904 KASSERT(fp == oldfde->fde_file, ("old fd has been modified"));
905 KASSERT(old != new, ("new fd is same as old"));
906
907 delfp = newfde->fde_file;
908
909 /*
910 * Duplicate the source descriptor.
911 */
912 #ifdef CAPABILITIES
913 seq_write_begin(&newfde->fde_seq);
914 #endif
915 filecaps_free(&newfde->fde_caps);
916 memcpy(newfde, oldfde, fde_change_size);
917 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
918 if ((flags & DUP_CLOEXEC) != 0)
919 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
920 else
921 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
922 #ifdef CAPABILITIES
923 seq_write_end(&newfde->fde_seq);
924 #endif
925 *retval = new;
926
927 if (delfp != NULL) {
928 (void) closefp(fdp, new, delfp, td, 1);
929 /* closefp() drops the FILEDESC lock for us. */
930 } else {
931 FILEDESC_XUNLOCK(fdp);
932 }
933
934 return (0);
935 }
936
937 /*
938 * If sigio is on the list associated with a process or process group,
939 * disable signalling from the device, remove sigio from the list and
940 * free sigio.
941 */
942 void
943 funsetown(struct sigio **sigiop)
944 {
945 struct sigio *sigio;
946
947 SIGIO_LOCK();
948 sigio = *sigiop;
949 if (sigio == NULL) {
950 SIGIO_UNLOCK();
951 return;
952 }
953 *(sigio->sio_myref) = NULL;
954 if ((sigio)->sio_pgid < 0) {
955 struct pgrp *pg = (sigio)->sio_pgrp;
956 PGRP_LOCK(pg);
957 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
958 sigio, sio_pgsigio);
959 PGRP_UNLOCK(pg);
960 } else {
961 struct proc *p = (sigio)->sio_proc;
962 PROC_LOCK(p);
963 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
964 sigio, sio_pgsigio);
965 PROC_UNLOCK(p);
966 }
967 SIGIO_UNLOCK();
968 crfree(sigio->sio_ucred);
969 free(sigio, M_SIGIO);
970 }
971
972 /*
973 * Free a list of sigio structures.
974 * We only need to lock the SIGIO_LOCK because we have made ourselves
975 * inaccessible to callers of fsetown and therefore do not need to lock
976 * the proc or pgrp struct for the list manipulation.
977 */
978 void
979 funsetownlst(struct sigiolst *sigiolst)
980 {
981 struct proc *p;
982 struct pgrp *pg;
983 struct sigio *sigio;
984
985 sigio = SLIST_FIRST(sigiolst);
986 if (sigio == NULL)
987 return;
988 p = NULL;
989 pg = NULL;
990
991 /*
992 * Every entry of the list should belong
993 * to a single proc or pgrp.
994 */
995 if (sigio->sio_pgid < 0) {
996 pg = sigio->sio_pgrp;
997 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
998 } else /* if (sigio->sio_pgid > 0) */ {
999 p = sigio->sio_proc;
1000 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1001 }
1002
1003 SIGIO_LOCK();
1004 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1005 *(sigio->sio_myref) = NULL;
1006 if (pg != NULL) {
1007 KASSERT(sigio->sio_pgid < 0,
1008 ("Proc sigio in pgrp sigio list"));
1009 KASSERT(sigio->sio_pgrp == pg,
1010 ("Bogus pgrp in sigio list"));
1011 PGRP_LOCK(pg);
1012 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1013 sio_pgsigio);
1014 PGRP_UNLOCK(pg);
1015 } else /* if (p != NULL) */ {
1016 KASSERT(sigio->sio_pgid > 0,
1017 ("Pgrp sigio in proc sigio list"));
1018 KASSERT(sigio->sio_proc == p,
1019 ("Bogus proc in sigio list"));
1020 PROC_LOCK(p);
1021 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1022 sio_pgsigio);
1023 PROC_UNLOCK(p);
1024 }
1025 SIGIO_UNLOCK();
1026 crfree(sigio->sio_ucred);
1027 free(sigio, M_SIGIO);
1028 SIGIO_LOCK();
1029 }
1030 SIGIO_UNLOCK();
1031 }
1032
1033 /*
1034 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1035 *
1036 * After permission checking, add a sigio structure to the sigio list for
1037 * the process or process group.
1038 */
1039 int
1040 fsetown(pid_t pgid, struct sigio **sigiop)
1041 {
1042 struct proc *proc;
1043 struct pgrp *pgrp;
1044 struct sigio *sigio;
1045 int ret;
1046
1047 if (pgid == 0) {
1048 funsetown(sigiop);
1049 return (0);
1050 }
1051
1052 ret = 0;
1053
1054 /* Allocate and fill in the new sigio out of locks. */
1055 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1056 sigio->sio_pgid = pgid;
1057 sigio->sio_ucred = crhold(curthread->td_ucred);
1058 sigio->sio_myref = sigiop;
1059
1060 sx_slock(&proctree_lock);
1061 if (pgid > 0) {
1062 proc = pfind(pgid);
1063 if (proc == NULL) {
1064 ret = ESRCH;
1065 goto fail;
1066 }
1067
1068 /*
1069 * Policy - Don't allow a process to FSETOWN a process
1070 * in another session.
1071 *
1072 * Remove this test to allow maximum flexibility or
1073 * restrict FSETOWN to the current process or process
1074 * group for maximum safety.
1075 */
1076 PROC_UNLOCK(proc);
1077 if (proc->p_session != curthread->td_proc->p_session) {
1078 ret = EPERM;
1079 goto fail;
1080 }
1081
1082 pgrp = NULL;
1083 } else /* if (pgid < 0) */ {
1084 pgrp = pgfind(-pgid);
1085 if (pgrp == NULL) {
1086 ret = ESRCH;
1087 goto fail;
1088 }
1089 PGRP_UNLOCK(pgrp);
1090
1091 /*
1092 * Policy - Don't allow a process to FSETOWN a process
1093 * in another session.
1094 *
1095 * Remove this test to allow maximum flexibility or
1096 * restrict FSETOWN to the current process or process
1097 * group for maximum safety.
1098 */
1099 if (pgrp->pg_session != curthread->td_proc->p_session) {
1100 ret = EPERM;
1101 goto fail;
1102 }
1103
1104 proc = NULL;
1105 }
1106 funsetown(sigiop);
1107 if (pgid > 0) {
1108 PROC_LOCK(proc);
1109 /*
1110 * Since funsetownlst() is called without the proctree
1111 * locked, we need to check for P_WEXIT.
1112 * XXX: is ESRCH correct?
1113 */
1114 if ((proc->p_flag & P_WEXIT) != 0) {
1115 PROC_UNLOCK(proc);
1116 ret = ESRCH;
1117 goto fail;
1118 }
1119 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1120 sigio->sio_proc = proc;
1121 PROC_UNLOCK(proc);
1122 } else {
1123 PGRP_LOCK(pgrp);
1124 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1125 sigio->sio_pgrp = pgrp;
1126 PGRP_UNLOCK(pgrp);
1127 }
1128 sx_sunlock(&proctree_lock);
1129 SIGIO_LOCK();
1130 *sigiop = sigio;
1131 SIGIO_UNLOCK();
1132 return (0);
1133
1134 fail:
1135 sx_sunlock(&proctree_lock);
1136 crfree(sigio->sio_ucred);
1137 free(sigio, M_SIGIO);
1138 return (ret);
1139 }
1140
1141 /*
1142 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1143 */
1144 pid_t
1145 fgetown(sigiop)
1146 struct sigio **sigiop;
1147 {
1148 pid_t pgid;
1149
1150 SIGIO_LOCK();
1151 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1152 SIGIO_UNLOCK();
1153 return (pgid);
1154 }
1155
1156 /*
1157 * Function drops the filedesc lock on return.
1158 */
1159 static int
1160 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1161 int holdleaders)
1162 {
1163 int error;
1164
1165 FILEDESC_XLOCK_ASSERT(fdp);
1166
1167 if (holdleaders) {
1168 if (td->td_proc->p_fdtol != NULL) {
1169 /*
1170 * Ask fdfree() to sleep to ensure that all relevant
1171 * process leaders can be traversed in closef().
1172 */
1173 fdp->fd_holdleaderscount++;
1174 } else {
1175 holdleaders = 0;
1176 }
1177 }
1178
1179 /*
1180 * We now hold the fp reference that used to be owned by the
1181 * descriptor array. We have to unlock the FILEDESC *AFTER*
1182 * knote_fdclose to prevent a race of the fd getting opened, a knote
1183 * added, and deleteing a knote for the new fd.
1184 */
1185 knote_fdclose(td, fd);
1186
1187 /*
1188 * We need to notify mqueue if the object is of type mqueue.
1189 */
1190 if (fp->f_type == DTYPE_MQUEUE)
1191 mq_fdclose(td, fd, fp);
1192 FILEDESC_XUNLOCK(fdp);
1193
1194 error = closef(fp, td);
1195 if (holdleaders) {
1196 FILEDESC_XLOCK(fdp);
1197 fdp->fd_holdleaderscount--;
1198 if (fdp->fd_holdleaderscount == 0 &&
1199 fdp->fd_holdleaderswakeup != 0) {
1200 fdp->fd_holdleaderswakeup = 0;
1201 wakeup(&fdp->fd_holdleaderscount);
1202 }
1203 FILEDESC_XUNLOCK(fdp);
1204 }
1205 return (error);
1206 }
1207
1208 /*
1209 * Close a file descriptor.
1210 */
1211 #ifndef _SYS_SYSPROTO_H_
1212 struct close_args {
1213 int fd;
1214 };
1215 #endif
1216 /* ARGSUSED */
1217 int
1218 sys_close(struct thread *td, struct close_args *uap)
1219 {
1220
1221 return (kern_close(td, uap->fd));
1222 }
1223
1224 int
1225 kern_close(struct thread *td, int fd)
1226 {
1227 struct filedesc *fdp;
1228 struct file *fp;
1229
1230 fdp = td->td_proc->p_fd;
1231
1232 AUDIT_SYSCLOSE(td, fd);
1233
1234 FILEDESC_XLOCK(fdp);
1235 if ((fp = fget_locked(fdp, fd)) == NULL) {
1236 FILEDESC_XUNLOCK(fdp);
1237 return (EBADF);
1238 }
1239 fdfree(fdp, fd);
1240
1241 /* closefp() drops the FILEDESC lock for us. */
1242 return (closefp(fdp, fd, fp, td, 1));
1243 }
1244
1245 /*
1246 * Close open file descriptors.
1247 */
1248 #ifndef _SYS_SYSPROTO_H_
1249 struct closefrom_args {
1250 int lowfd;
1251 };
1252 #endif
1253 /* ARGSUSED */
1254 int
1255 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1256 {
1257 struct filedesc *fdp;
1258 int fd;
1259
1260 fdp = td->td_proc->p_fd;
1261 AUDIT_ARG_FD(uap->lowfd);
1262
1263 /*
1264 * Treat negative starting file descriptor values identical to
1265 * closefrom(0) which closes all files.
1266 */
1267 if (uap->lowfd < 0)
1268 uap->lowfd = 0;
1269 FILEDESC_SLOCK(fdp);
1270 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1271 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1272 FILEDESC_SUNLOCK(fdp);
1273 (void)kern_close(td, fd);
1274 FILEDESC_SLOCK(fdp);
1275 }
1276 }
1277 FILEDESC_SUNLOCK(fdp);
1278 return (0);
1279 }
1280
1281 #if defined(COMPAT_43)
1282 /*
1283 * Return status information about a file descriptor.
1284 */
1285 #ifndef _SYS_SYSPROTO_H_
1286 struct ofstat_args {
1287 int fd;
1288 struct ostat *sb;
1289 };
1290 #endif
1291 /* ARGSUSED */
1292 int
1293 ofstat(struct thread *td, struct ofstat_args *uap)
1294 {
1295 struct ostat oub;
1296 struct stat ub;
1297 int error;
1298
1299 error = kern_fstat(td, uap->fd, &ub);
1300 if (error == 0) {
1301 cvtstat(&ub, &oub);
1302 error = copyout(&oub, uap->sb, sizeof(oub));
1303 }
1304 return (error);
1305 }
1306 #endif /* COMPAT_43 */
1307
1308 /*
1309 * Return status information about a file descriptor.
1310 */
1311 #ifndef _SYS_SYSPROTO_H_
1312 struct fstat_args {
1313 int fd;
1314 struct stat *sb;
1315 };
1316 #endif
1317 /* ARGSUSED */
1318 int
1319 sys_fstat(struct thread *td, struct fstat_args *uap)
1320 {
1321 struct stat ub;
1322 int error;
1323
1324 error = kern_fstat(td, uap->fd, &ub);
1325 if (error == 0)
1326 error = copyout(&ub, uap->sb, sizeof(ub));
1327 return (error);
1328 }
1329
1330 int
1331 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1332 {
1333 struct file *fp;
1334 cap_rights_t rights;
1335 int error;
1336
1337 AUDIT_ARG_FD(fd);
1338
1339 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1340 if (error != 0)
1341 return (error);
1342
1343 AUDIT_ARG_FILE(td->td_proc, fp);
1344
1345 error = fo_stat(fp, sbp, td->td_ucred, td);
1346 fdrop(fp, td);
1347 #ifdef KTRACE
1348 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1349 ktrstat(sbp);
1350 #endif
1351 return (error);
1352 }
1353
1354 /*
1355 * Return status information about a file descriptor.
1356 */
1357 #ifndef _SYS_SYSPROTO_H_
1358 struct nfstat_args {
1359 int fd;
1360 struct nstat *sb;
1361 };
1362 #endif
1363 /* ARGSUSED */
1364 int
1365 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1366 {
1367 struct nstat nub;
1368 struct stat ub;
1369 int error;
1370
1371 error = kern_fstat(td, uap->fd, &ub);
1372 if (error == 0) {
1373 cvtnstat(&ub, &nub);
1374 error = copyout(&nub, uap->sb, sizeof(nub));
1375 }
1376 return (error);
1377 }
1378
1379 /*
1380 * Return pathconf information about a file descriptor.
1381 */
1382 #ifndef _SYS_SYSPROTO_H_
1383 struct fpathconf_args {
1384 int fd;
1385 int name;
1386 };
1387 #endif
1388 /* ARGSUSED */
1389 int
1390 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1391 {
1392 struct file *fp;
1393 struct vnode *vp;
1394 cap_rights_t rights;
1395 int error;
1396
1397 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1398 if (error != 0)
1399 return (error);
1400
1401 /* If asynchronous I/O is available, it works for all descriptors. */
1402 if (uap->name == _PC_ASYNC_IO) {
1403 td->td_retval[0] = async_io_version;
1404 goto out;
1405 }
1406 vp = fp->f_vnode;
1407 if (vp != NULL) {
1408 vn_lock(vp, LK_SHARED | LK_RETRY);
1409 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1410 VOP_UNLOCK(vp, 0);
1411 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1412 if (uap->name != _PC_PIPE_BUF) {
1413 error = EINVAL;
1414 } else {
1415 td->td_retval[0] = PIPE_BUF;
1416 error = 0;
1417 }
1418 } else {
1419 error = EOPNOTSUPP;
1420 }
1421 out:
1422 fdrop(fp, td);
1423 return (error);
1424 }
1425
1426 /*
1427 * Initialize filecaps structure.
1428 */
1429 void
1430 filecaps_init(struct filecaps *fcaps)
1431 {
1432
1433 bzero(fcaps, sizeof(*fcaps));
1434 fcaps->fc_nioctls = -1;
1435 }
1436
1437 /*
1438 * Copy filecaps structure allocating memory for ioctls array if needed.
1439 */
1440 void
1441 filecaps_copy(const struct filecaps *src, struct filecaps *dst)
1442 {
1443 size_t size;
1444
1445 *dst = *src;
1446 if (src->fc_ioctls != NULL) {
1447 KASSERT(src->fc_nioctls > 0,
1448 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1449
1450 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1451 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1452 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1453 }
1454 }
1455
1456 /*
1457 * Move filecaps structure to the new place and clear the old place.
1458 */
1459 void
1460 filecaps_move(struct filecaps *src, struct filecaps *dst)
1461 {
1462
1463 *dst = *src;
1464 bzero(src, sizeof(*src));
1465 }
1466
1467 /*
1468 * Fill the given filecaps structure with full rights.
1469 */
1470 static void
1471 filecaps_fill(struct filecaps *fcaps)
1472 {
1473
1474 CAP_ALL(&fcaps->fc_rights);
1475 fcaps->fc_ioctls = NULL;
1476 fcaps->fc_nioctls = -1;
1477 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1478 }
1479
1480 /*
1481 * Free memory allocated within filecaps structure.
1482 */
1483 void
1484 filecaps_free(struct filecaps *fcaps)
1485 {
1486
1487 free(fcaps->fc_ioctls, M_FILECAPS);
1488 bzero(fcaps, sizeof(*fcaps));
1489 }
1490
1491 /*
1492 * Validate the given filecaps structure.
1493 */
1494 static void
1495 filecaps_validate(const struct filecaps *fcaps, const char *func)
1496 {
1497
1498 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1499 ("%s: invalid rights", func));
1500 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1501 ("%s: invalid fcntls", func));
1502 KASSERT(fcaps->fc_fcntls == 0 ||
1503 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1504 ("%s: fcntls without CAP_FCNTL", func));
1505 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1506 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1507 ("%s: invalid ioctls", func));
1508 KASSERT(fcaps->fc_nioctls == 0 ||
1509 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1510 ("%s: ioctls without CAP_IOCTL", func));
1511 }
1512
1513 static void
1514 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1515 {
1516 int nfd1;
1517
1518 FILEDESC_XLOCK_ASSERT(fdp);
1519
1520 nfd1 = fdp->fd_nfiles * 2;
1521 if (nfd1 < nfd)
1522 nfd1 = nfd;
1523 fdgrowtable(fdp, nfd1);
1524 }
1525
1526 /*
1527 * Grow the file table to accomodate (at least) nfd descriptors.
1528 */
1529 static void
1530 fdgrowtable(struct filedesc *fdp, int nfd)
1531 {
1532 struct filedesc0 *fdp0;
1533 struct freetable *ft;
1534 struct filedescent *ntable;
1535 struct filedescent *otable;
1536 int nnfiles, onfiles;
1537 NDSLOTTYPE *nmap, *omap;
1538
1539 FILEDESC_XLOCK_ASSERT(fdp);
1540
1541 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1542
1543 /* save old values */
1544 onfiles = fdp->fd_nfiles;
1545 otable = fdp->fd_ofiles;
1546 omap = fdp->fd_map;
1547
1548 /* compute the size of the new table */
1549 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1550 if (nnfiles <= onfiles)
1551 /* the table is already large enough */
1552 return;
1553
1554 /*
1555 * Allocate a new table. We need enough space for the
1556 * file entries themselves and the struct freetable we will use
1557 * when we decommission the table and place it on the freelist.
1558 * We place the struct freetable in the middle so we don't have
1559 * to worry about padding.
1560 */
1561 ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable),
1562 M_FILEDESC, M_ZERO | M_WAITOK);
1563 /* copy the old data over and point at the new tables */
1564 memcpy(ntable, otable, onfiles * sizeof(*otable));
1565 fdp->fd_ofiles = ntable;
1566
1567 /*
1568 * Allocate a new map only if the old is not large enough. It will
1569 * grow at a slower rate than the table as it can map more
1570 * entries than the table can hold.
1571 */
1572 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1573 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1574 M_ZERO | M_WAITOK);
1575 /* copy over the old data and update the pointer */
1576 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1577 fdp->fd_map = nmap;
1578 }
1579
1580 /*
1581 * In order to have a valid pattern for fget_unlocked()
1582 * fdp->fd_nfiles must be the last member to be updated, otherwise
1583 * fget_unlocked() consumers may reference a new, higher value for
1584 * fdp->fd_nfiles before to access the fdp->fd_ofiles array,
1585 * resulting in OOB accesses.
1586 */
1587 atomic_store_rel_int(&fdp->fd_nfiles, nnfiles);
1588
1589 /*
1590 * Do not free the old file table, as some threads may still
1591 * reference entries within it. Instead, place it on a freelist
1592 * which will be processed when the struct filedesc is released.
1593 *
1594 * Note that if onfiles == NDFILE, we're dealing with the original
1595 * static allocation contained within (struct filedesc0 *)fdp,
1596 * which must not be freed.
1597 */
1598 if (onfiles > NDFILE) {
1599 ft = (struct freetable *)&otable[onfiles];
1600 fdp0 = (struct filedesc0 *)fdp;
1601 ft->ft_table = otable;
1602 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1603 }
1604 /*
1605 * The map does not have the same possibility of threads still
1606 * holding references to it. So always free it as long as it
1607 * does not reference the original static allocation.
1608 */
1609 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1610 free(omap, M_FILEDESC);
1611 }
1612
1613 /*
1614 * Allocate a file descriptor for the process.
1615 */
1616 int
1617 fdalloc(struct thread *td, int minfd, int *result)
1618 {
1619 struct proc *p = td->td_proc;
1620 struct filedesc *fdp = p->p_fd;
1621 int fd = -1, maxfd, allocfd;
1622 #ifdef RACCT
1623 int error;
1624 #endif
1625
1626 FILEDESC_XLOCK_ASSERT(fdp);
1627
1628 if (fdp->fd_freefile > minfd)
1629 minfd = fdp->fd_freefile;
1630
1631 maxfd = getmaxfd(p);
1632
1633 /*
1634 * Search the bitmap for a free descriptor starting at minfd.
1635 * If none is found, grow the file table.
1636 */
1637 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1638 if (fd >= maxfd)
1639 return (EMFILE);
1640 if (fd >= fdp->fd_nfiles) {
1641 allocfd = min(fd * 2, maxfd);
1642 #ifdef RACCT
1643 if (racct_enable) {
1644 PROC_LOCK(p);
1645 error = racct_set(p, RACCT_NOFILE, allocfd);
1646 PROC_UNLOCK(p);
1647 if (error != 0)
1648 return (EMFILE);
1649 }
1650 #endif
1651 /*
1652 * fd is already equal to first free descriptor >= minfd, so
1653 * we only need to grow the table and we are done.
1654 */
1655 fdgrowtable_exp(fdp, allocfd);
1656 }
1657
1658 /*
1659 * Perform some sanity checks, then mark the file descriptor as
1660 * used and return it to the caller.
1661 */
1662 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1663 ("invalid descriptor %d", fd));
1664 KASSERT(!fdisused(fdp, fd),
1665 ("fd_first_free() returned non-free descriptor"));
1666 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1667 ("file descriptor isn't free"));
1668 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
1669 fdused(fdp, fd);
1670 *result = fd;
1671 return (0);
1672 }
1673
1674 /*
1675 * Allocate n file descriptors for the process.
1676 */
1677 int
1678 fdallocn(struct thread *td, int minfd, int *fds, int n)
1679 {
1680 struct proc *p = td->td_proc;
1681 struct filedesc *fdp = p->p_fd;
1682 int i;
1683
1684 FILEDESC_XLOCK_ASSERT(fdp);
1685
1686 if (!fdavail(td, n))
1687 return (EMFILE);
1688
1689 for (i = 0; i < n; i++)
1690 if (fdalloc(td, 0, &fds[i]) != 0)
1691 break;
1692
1693 if (i < n) {
1694 for (i--; i >= 0; i--)
1695 fdunused(fdp, fds[i]);
1696 return (EMFILE);
1697 }
1698
1699 return (0);
1700 }
1701
1702 /*
1703 * Check to see whether n user file descriptors are available to the process
1704 * p.
1705 */
1706 int
1707 fdavail(struct thread *td, int n)
1708 {
1709 struct proc *p = td->td_proc;
1710 struct filedesc *fdp = td->td_proc->p_fd;
1711 int i, lim, last;
1712
1713 FILEDESC_LOCK_ASSERT(fdp);
1714
1715 /*
1716 * XXX: This is only called from uipc_usrreq.c:unp_externalize();
1717 * call racct_add() from there instead of dealing with containers
1718 * here.
1719 */
1720 lim = getmaxfd(p);
1721 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1722 return (1);
1723 last = min(fdp->fd_nfiles, lim);
1724 for (i = fdp->fd_freefile; i < last; i++) {
1725 if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0)
1726 return (1);
1727 }
1728 return (0);
1729 }
1730
1731 /*
1732 * Create a new open file structure and allocate a file decriptor for the
1733 * process that refers to it. We add one reference to the file for the
1734 * descriptor table and one reference for resultfp. This is to prevent us
1735 * being preempted and the entry in the descriptor table closed after we
1736 * release the FILEDESC lock.
1737 */
1738 int
1739 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1740 {
1741 struct file *fp;
1742 int error, fd;
1743
1744 error = falloc_noinstall(td, &fp);
1745 if (error)
1746 return (error); /* no reference held on error */
1747
1748 error = finstall(td, fp, &fd, flags, NULL);
1749 if (error) {
1750 fdrop(fp, td); /* one reference (fp only) */
1751 return (error);
1752 }
1753
1754 if (resultfp != NULL)
1755 *resultfp = fp; /* copy out result */
1756 else
1757 fdrop(fp, td); /* release local reference */
1758
1759 if (resultfd != NULL)
1760 *resultfd = fd;
1761
1762 return (0);
1763 }
1764
1765 /*
1766 * Create a new open file structure without allocating a file descriptor.
1767 */
1768 int
1769 falloc_noinstall(struct thread *td, struct file **resultfp)
1770 {
1771 struct file *fp;
1772 int maxuserfiles = maxfiles - (maxfiles / 20);
1773 static struct timeval lastfail;
1774 static int curfail;
1775
1776 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1777
1778 if ((openfiles >= maxuserfiles &&
1779 priv_check(td, PRIV_MAXFILES) != 0) ||
1780 openfiles >= maxfiles) {
1781 if (ppsratecheck(&lastfail, &curfail, 1)) {
1782 printf("kern.maxfiles limit exceeded by uid %i, "
1783 "please see tuning(7).\n", td->td_ucred->cr_ruid);
1784 }
1785 return (ENFILE);
1786 }
1787 atomic_add_int(&openfiles, 1);
1788 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1789 refcount_init(&fp->f_count, 1);
1790 fp->f_cred = crhold(td->td_ucred);
1791 fp->f_ops = &badfileops;
1792 fp->f_data = NULL;
1793 fp->f_vnode = NULL;
1794 *resultfp = fp;
1795 return (0);
1796 }
1797
1798 /*
1799 * Install a file in a file descriptor table.
1800 */
1801 int
1802 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1803 struct filecaps *fcaps)
1804 {
1805 struct filedesc *fdp = td->td_proc->p_fd;
1806 struct filedescent *fde;
1807 int error;
1808
1809 KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1810 KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1811 if (fcaps != NULL)
1812 filecaps_validate(fcaps, __func__);
1813
1814 FILEDESC_XLOCK(fdp);
1815 if ((error = fdalloc(td, 0, fd))) {
1816 FILEDESC_XUNLOCK(fdp);
1817 return (error);
1818 }
1819 fhold(fp);
1820 fde = &fdp->fd_ofiles[*fd];
1821 #ifdef CAPABILITIES
1822 seq_write_begin(&fde->fde_seq);
1823 #endif
1824 fde->fde_file = fp;
1825 if ((flags & O_CLOEXEC) != 0)
1826 fde->fde_flags |= UF_EXCLOSE;
1827 if (fcaps != NULL)
1828 filecaps_move(fcaps, &fde->fde_caps);
1829 else
1830 filecaps_fill(&fde->fde_caps);
1831 #ifdef CAPABILITIES
1832 seq_write_end(&fde->fde_seq);
1833 #endif
1834 FILEDESC_XUNLOCK(fdp);
1835 return (0);
1836 }
1837
1838 /*
1839 * Build a new filedesc structure from another.
1840 * Copy the current, root, and jail root vnode references.
1841 */
1842 struct filedesc *
1843 fdinit(struct filedesc *fdp)
1844 {
1845 struct filedesc0 *newfdp;
1846
1847 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1848 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1849 if (fdp != NULL) {
1850 FILEDESC_SLOCK(fdp);
1851 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1852 if (newfdp->fd_fd.fd_cdir)
1853 VREF(newfdp->fd_fd.fd_cdir);
1854 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1855 if (newfdp->fd_fd.fd_rdir)
1856 VREF(newfdp->fd_fd.fd_rdir);
1857 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1858 if (newfdp->fd_fd.fd_jdir)
1859 VREF(newfdp->fd_fd.fd_jdir);
1860 FILEDESC_SUNLOCK(fdp);
1861 }
1862
1863 /* Create the file descriptor table. */
1864 newfdp->fd_fd.fd_refcnt = 1;
1865 newfdp->fd_fd.fd_holdcnt = 1;
1866 newfdp->fd_fd.fd_cmask = CMASK;
1867 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1868 newfdp->fd_fd.fd_nfiles = NDFILE;
1869 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1870 newfdp->fd_fd.fd_lastfile = -1;
1871 return (&newfdp->fd_fd);
1872 }
1873
1874 static struct filedesc *
1875 fdhold(struct proc *p)
1876 {
1877 struct filedesc *fdp;
1878
1879 mtx_lock(&fdesc_mtx);
1880 fdp = p->p_fd;
1881 if (fdp != NULL)
1882 fdp->fd_holdcnt++;
1883 mtx_unlock(&fdesc_mtx);
1884 return (fdp);
1885 }
1886
1887 static void
1888 fddrop(struct filedesc *fdp)
1889 {
1890 struct filedesc0 *fdp0;
1891 struct freetable *ft;
1892 int i;
1893
1894 mtx_lock(&fdesc_mtx);
1895 i = --fdp->fd_holdcnt;
1896 mtx_unlock(&fdesc_mtx);
1897 if (i > 0)
1898 return;
1899
1900 FILEDESC_LOCK_DESTROY(fdp);
1901 fdp0 = (struct filedesc0 *)fdp;
1902 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1903 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1904 free(ft->ft_table, M_FILEDESC);
1905 }
1906 free(fdp, M_FILEDESC);
1907 }
1908
1909 /*
1910 * Share a filedesc structure.
1911 */
1912 struct filedesc *
1913 fdshare(struct filedesc *fdp)
1914 {
1915
1916 FILEDESC_XLOCK(fdp);
1917 fdp->fd_refcnt++;
1918 FILEDESC_XUNLOCK(fdp);
1919 return (fdp);
1920 }
1921
1922 /*
1923 * Unshare a filedesc structure, if necessary by making a copy
1924 */
1925 void
1926 fdunshare(struct thread *td)
1927 {
1928 struct filedesc *tmp;
1929 struct proc *p = td->td_proc;
1930
1931 if (p->p_fd->fd_refcnt == 1)
1932 return;
1933
1934 tmp = fdcopy(p->p_fd);
1935 fdescfree(td);
1936 p->p_fd = tmp;
1937 }
1938
1939 /*
1940 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1941 * this is to ease callers, not catch errors.
1942 */
1943 struct filedesc *
1944 fdcopy(struct filedesc *fdp)
1945 {
1946 struct filedesc *newfdp;
1947 struct filedescent *nfde, *ofde;
1948 int i;
1949
1950 /* Certain daemons might not have file descriptors. */
1951 if (fdp == NULL)
1952 return (NULL);
1953
1954 newfdp = fdinit(fdp);
1955 FILEDESC_SLOCK(fdp);
1956 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1957 FILEDESC_SUNLOCK(fdp);
1958 FILEDESC_XLOCK(newfdp);
1959 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1960 FILEDESC_XUNLOCK(newfdp);
1961 FILEDESC_SLOCK(fdp);
1962 }
1963 /* copy all passable descriptors (i.e. not kqueue) */
1964 newfdp->fd_freefile = -1;
1965 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1966 ofde = &fdp->fd_ofiles[i];
1967 if (fdisused(fdp, i) &&
1968 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) &&
1969 ofde->fde_file->f_ops != &badfileops) {
1970 nfde = &newfdp->fd_ofiles[i];
1971 *nfde = *ofde;
1972 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
1973 fhold(nfde->fde_file);
1974 newfdp->fd_lastfile = i;
1975 } else {
1976 if (newfdp->fd_freefile == -1)
1977 newfdp->fd_freefile = i;
1978 }
1979 }
1980 newfdp->fd_cmask = fdp->fd_cmask;
1981 FILEDESC_SUNLOCK(fdp);
1982 FILEDESC_XLOCK(newfdp);
1983 for (i = 0; i <= newfdp->fd_lastfile; ++i) {
1984 if (newfdp->fd_ofiles[i].fde_file != NULL)
1985 fdused(newfdp, i);
1986 }
1987 if (newfdp->fd_freefile == -1)
1988 newfdp->fd_freefile = i;
1989 FILEDESC_XUNLOCK(newfdp);
1990 return (newfdp);
1991 }
1992
1993 /*
1994 * Release a filedesc structure.
1995 */
1996 void
1997 fdescfree(struct thread *td)
1998 {
1999 struct filedesc *fdp;
2000 int i;
2001 struct filedesc_to_leader *fdtol;
2002 struct file *fp;
2003 struct vnode *cdir, *jdir, *rdir, *vp;
2004 struct flock lf;
2005
2006 /* Certain daemons might not have file descriptors. */
2007 fdp = td->td_proc->p_fd;
2008 if (fdp == NULL)
2009 return;
2010
2011 #ifdef RACCT
2012 if (racct_enable) {
2013 PROC_LOCK(td->td_proc);
2014 racct_set(td->td_proc, RACCT_NOFILE, 0);
2015 PROC_UNLOCK(td->td_proc);
2016 }
2017 #endif
2018
2019 /* Check for special need to clear POSIX style locks */
2020 fdtol = td->td_proc->p_fdtol;
2021 if (fdtol != NULL) {
2022 FILEDESC_XLOCK(fdp);
2023 KASSERT(fdtol->fdl_refcount > 0,
2024 ("filedesc_to_refcount botch: fdl_refcount=%d",
2025 fdtol->fdl_refcount));
2026 if (fdtol->fdl_refcount == 1 &&
2027 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2028 for (i = 0; i <= fdp->fd_lastfile; i++) {
2029 fp = fdp->fd_ofiles[i].fde_file;
2030 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2031 continue;
2032 fhold(fp);
2033 FILEDESC_XUNLOCK(fdp);
2034 lf.l_whence = SEEK_SET;
2035 lf.l_start = 0;
2036 lf.l_len = 0;
2037 lf.l_type = F_UNLCK;
2038 vp = fp->f_vnode;
2039 (void) VOP_ADVLOCK(vp,
2040 (caddr_t)td->td_proc->p_leader, F_UNLCK,
2041 &lf, F_POSIX);
2042 FILEDESC_XLOCK(fdp);
2043 fdrop(fp, td);
2044 }
2045 }
2046 retry:
2047 if (fdtol->fdl_refcount == 1) {
2048 if (fdp->fd_holdleaderscount > 0 &&
2049 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2050 /*
2051 * close() or do_dup() has cleared a reference
2052 * in a shared file descriptor table.
2053 */
2054 fdp->fd_holdleaderswakeup = 1;
2055 sx_sleep(&fdp->fd_holdleaderscount,
2056 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2057 goto retry;
2058 }
2059 if (fdtol->fdl_holdcount > 0) {
2060 /*
2061 * Ensure that fdtol->fdl_leader remains
2062 * valid in closef().
2063 */
2064 fdtol->fdl_wakeup = 1;
2065 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2066 "fdlhold", 0);
2067 goto retry;
2068 }
2069 }
2070 fdtol->fdl_refcount--;
2071 if (fdtol->fdl_refcount == 0 &&
2072 fdtol->fdl_holdcount == 0) {
2073 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2074 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2075 } else
2076 fdtol = NULL;
2077 td->td_proc->p_fdtol = NULL;
2078 FILEDESC_XUNLOCK(fdp);
2079 if (fdtol != NULL)
2080 free(fdtol, M_FILEDESC_TO_LEADER);
2081 }
2082
2083 mtx_lock(&fdesc_mtx);
2084 td->td_proc->p_fd = NULL;
2085 mtx_unlock(&fdesc_mtx);
2086
2087 FILEDESC_XLOCK(fdp);
2088 i = --fdp->fd_refcnt;
2089 if (i > 0) {
2090 FILEDESC_XUNLOCK(fdp);
2091 return;
2092 }
2093
2094 cdir = fdp->fd_cdir;
2095 fdp->fd_cdir = NULL;
2096 rdir = fdp->fd_rdir;
2097 fdp->fd_rdir = NULL;
2098 jdir = fdp->fd_jdir;
2099 fdp->fd_jdir = NULL;
2100 FILEDESC_XUNLOCK(fdp);
2101
2102 for (i = 0; i <= fdp->fd_lastfile; i++) {
2103 fp = fdp->fd_ofiles[i].fde_file;
2104 if (fp != NULL) {
2105 fdfree_last(fdp, i);
2106 (void) closef(fp, td);
2107 }
2108 }
2109
2110 if (fdp->fd_nfiles > NDFILE)
2111 free(fdp->fd_ofiles, M_FILEDESC);
2112 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2113 free(fdp->fd_map, M_FILEDESC);
2114
2115 if (cdir != NULL)
2116 vrele(cdir);
2117 if (rdir != NULL)
2118 vrele(rdir);
2119 if (jdir != NULL)
2120 vrele(jdir);
2121
2122 fddrop(fdp);
2123 }
2124
2125 /*
2126 * For setugid programs, we don't want to people to use that setugidness
2127 * to generate error messages which write to a file which otherwise would
2128 * otherwise be off-limits to the process. We check for filesystems where
2129 * the vnode can change out from under us after execve (like [lin]procfs).
2130 *
2131 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2132 * sufficient. We also don't check for setugidness since we know we are.
2133 */
2134 static int
2135 is_unsafe(struct file *fp)
2136 {
2137 if (fp->f_type == DTYPE_VNODE) {
2138 struct vnode *vp = fp->f_vnode;
2139
2140 if ((vp->v_vflag & VV_PROCDEP) != 0)
2141 return (1);
2142 }
2143 return (0);
2144 }
2145
2146 /*
2147 * Make this setguid thing safe, if at all possible.
2148 */
2149 void
2150 setugidsafety(struct thread *td)
2151 {
2152 struct filedesc *fdp;
2153 struct file *fp;
2154 int i;
2155
2156 fdp = td->td_proc->p_fd;
2157 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2158 FILEDESC_XLOCK(fdp);
2159 for (i = 0; i <= fdp->fd_lastfile; i++) {
2160 if (i > 2)
2161 break;
2162 fp = fdp->fd_ofiles[i].fde_file;
2163 if (fp != NULL && is_unsafe(fp)) {
2164 knote_fdclose(td, i);
2165 /*
2166 * NULL-out descriptor prior to close to avoid
2167 * a race while close blocks.
2168 */
2169 fdfree(fdp, i);
2170 FILEDESC_XUNLOCK(fdp);
2171 (void) closef(fp, td);
2172 FILEDESC_XLOCK(fdp);
2173 }
2174 }
2175 FILEDESC_XUNLOCK(fdp);
2176 }
2177
2178 /*
2179 * If a specific file object occupies a specific file descriptor, close the
2180 * file descriptor entry and drop a reference on the file object. This is a
2181 * convenience function to handle a subsequent error in a function that calls
2182 * falloc() that handles the race that another thread might have closed the
2183 * file descriptor out from under the thread creating the file object.
2184 */
2185 void
2186 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2187 {
2188
2189 FILEDESC_XLOCK(fdp);
2190 if (fdp->fd_ofiles[idx].fde_file == fp) {
2191 fdfree(fdp, idx);
2192 FILEDESC_XUNLOCK(fdp);
2193 fdrop(fp, td);
2194 } else
2195 FILEDESC_XUNLOCK(fdp);
2196 }
2197
2198 /*
2199 * Close any files on exec?
2200 */
2201 void
2202 fdcloseexec(struct thread *td)
2203 {
2204 struct filedesc *fdp;
2205 struct filedescent *fde;
2206 struct file *fp;
2207 int i;
2208
2209 fdp = td->td_proc->p_fd;
2210 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2211 FILEDESC_XLOCK(fdp);
2212 for (i = 0; i <= fdp->fd_lastfile; i++) {
2213 fde = &fdp->fd_ofiles[i];
2214 fp = fde->fde_file;
2215 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2216 (fde->fde_flags & UF_EXCLOSE))) {
2217 fdfree(fdp, i);
2218 (void) closefp(fdp, i, fp, td, 0);
2219 /* closefp() drops the FILEDESC lock. */
2220 FILEDESC_XLOCK(fdp);
2221 }
2222 }
2223 FILEDESC_XUNLOCK(fdp);
2224 }
2225
2226 /*
2227 * It is unsafe for set[ug]id processes to be started with file
2228 * descriptors 0..2 closed, as these descriptors are given implicit
2229 * significance in the Standard C library. fdcheckstd() will create a
2230 * descriptor referencing /dev/null for each of stdin, stdout, and
2231 * stderr that is not already open.
2232 */
2233 int
2234 fdcheckstd(struct thread *td)
2235 {
2236 struct filedesc *fdp;
2237 register_t retval, save;
2238 int i, error, devnull;
2239
2240 fdp = td->td_proc->p_fd;
2241 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2242 devnull = -1;
2243 error = 0;
2244 for (i = 0; i < 3; i++) {
2245 if (fdp->fd_ofiles[i].fde_file != NULL)
2246 continue;
2247 if (devnull < 0) {
2248 save = td->td_retval[0];
2249 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2250 O_RDWR, 0);
2251 devnull = td->td_retval[0];
2252 td->td_retval[0] = save;
2253 if (error)
2254 break;
2255 KASSERT(devnull == i, ("oof, we didn't get our fd"));
2256 } else {
2257 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2258 if (error != 0)
2259 break;
2260 }
2261 }
2262 return (error);
2263 }
2264
2265 /*
2266 * Internal form of close. Decrement reference count on file structure.
2267 * Note: td may be NULL when closing a file that was being passed in a
2268 * message.
2269 *
2270 * XXXRW: Giant is not required for the caller, but often will be held; this
2271 * makes it moderately likely the Giant will be recursed in the VFS case.
2272 */
2273 int
2274 closef(struct file *fp, struct thread *td)
2275 {
2276 struct vnode *vp;
2277 struct flock lf;
2278 struct filedesc_to_leader *fdtol;
2279 struct filedesc *fdp;
2280
2281 /*
2282 * POSIX record locking dictates that any close releases ALL
2283 * locks owned by this process. This is handled by setting
2284 * a flag in the unlock to free ONLY locks obeying POSIX
2285 * semantics, and not to free BSD-style file locks.
2286 * If the descriptor was in a message, POSIX-style locks
2287 * aren't passed with the descriptor, and the thread pointer
2288 * will be NULL. Callers should be careful only to pass a
2289 * NULL thread pointer when there really is no owning
2290 * context that might have locks, or the locks will be
2291 * leaked.
2292 */
2293 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2294 vp = fp->f_vnode;
2295 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2296 lf.l_whence = SEEK_SET;
2297 lf.l_start = 0;
2298 lf.l_len = 0;
2299 lf.l_type = F_UNLCK;
2300 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2301 F_UNLCK, &lf, F_POSIX);
2302 }
2303 fdtol = td->td_proc->p_fdtol;
2304 if (fdtol != NULL) {
2305 /*
2306 * Handle special case where file descriptor table is
2307 * shared between multiple process leaders.
2308 */
2309 fdp = td->td_proc->p_fd;
2310 FILEDESC_XLOCK(fdp);
2311 for (fdtol = fdtol->fdl_next;
2312 fdtol != td->td_proc->p_fdtol;
2313 fdtol = fdtol->fdl_next) {
2314 if ((fdtol->fdl_leader->p_flag &
2315 P_ADVLOCK) == 0)
2316 continue;
2317 fdtol->fdl_holdcount++;
2318 FILEDESC_XUNLOCK(fdp);
2319 lf.l_whence = SEEK_SET;
2320 lf.l_start = 0;
2321 lf.l_len = 0;
2322 lf.l_type = F_UNLCK;
2323 vp = fp->f_vnode;
2324 (void) VOP_ADVLOCK(vp,
2325 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2326 F_POSIX);
2327 FILEDESC_XLOCK(fdp);
2328 fdtol->fdl_holdcount--;
2329 if (fdtol->fdl_holdcount == 0 &&
2330 fdtol->fdl_wakeup != 0) {
2331 fdtol->fdl_wakeup = 0;
2332 wakeup(fdtol);
2333 }
2334 }
2335 FILEDESC_XUNLOCK(fdp);
2336 }
2337 }
2338 return (fdrop(fp, td));
2339 }
2340
2341 /*
2342 * Initialize the file pointer with the specified properties.
2343 *
2344 * The ops are set with release semantics to be certain that the flags, type,
2345 * and data are visible when ops is. This is to prevent ops methods from being
2346 * called with bad data.
2347 */
2348 void
2349 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2350 {
2351 fp->f_data = data;
2352 fp->f_flag = flag;
2353 fp->f_type = type;
2354 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2355 }
2356
2357 int
2358 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2359 int needfcntl, struct file **fpp, cap_rights_t *haverightsp)
2360 {
2361 #ifdef CAPABILITIES
2362 struct filedescent fde;
2363 #endif
2364 struct file *fp;
2365 u_int count;
2366 #ifdef CAPABILITIES
2367 seq_t seq;
2368 cap_rights_t haverights;
2369 int error;
2370 #endif
2371
2372 /*
2373 * Avoid reads reordering and then a first access to the
2374 * fdp->fd_ofiles table which could result in OOB operation.
2375 */
2376 if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles))
2377 return (EBADF);
2378 /*
2379 * Fetch the descriptor locklessly. We avoid fdrop() races by
2380 * never raising a refcount above 0. To accomplish this we have
2381 * to use a cmpset loop rather than an atomic_add. The descriptor
2382 * must be re-verified once we acquire a reference to be certain
2383 * that the identity is still correct and we did not lose a race
2384 * due to preemption.
2385 */
2386 for (;;) {
2387 #ifdef CAPABILITIES
2388 seq = seq_read(fd_seq(fdp, fd));
2389 fde = fdp->fd_ofiles[fd];
2390 if (!seq_consistent(fd_seq(fdp, fd), seq)) {
2391 cpu_spinwait();
2392 continue;
2393 }
2394 fp = fde.fde_file;
2395 #else
2396 fp = fdp->fd_ofiles[fd].fde_file;
2397 #endif
2398 if (fp == NULL)
2399 return (EBADF);
2400 #ifdef CAPABILITIES
2401 haverights = *cap_rights_fde(&fde);
2402 if (needrightsp != NULL) {
2403 error = cap_check(&haverights, needrightsp);
2404 if (error != 0)
2405 return (error);
2406 if (cap_rights_is_set(needrightsp, CAP_FCNTL)) {
2407 error = cap_fcntl_check_fde(&fde, needfcntl);
2408 if (error != 0)
2409 return (error);
2410 }
2411 }
2412 #endif
2413 count = fp->f_count;
2414 if (count == 0)
2415 continue;
2416 /*
2417 * Use an acquire barrier to prevent caching of fd_ofiles
2418 * so it is refreshed for verification.
2419 */
2420 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2421 continue;
2422 #ifdef CAPABILITIES
2423 if (seq_consistent_nomb(fd_seq(fdp, fd), seq))
2424 #else
2425 if (fp == fdp->fd_ofiles[fd].fde_file)
2426 #endif
2427 break;
2428 fdrop(fp, curthread);
2429 }
2430 *fpp = fp;
2431 if (haverightsp != NULL) {
2432 #ifdef CAPABILITIES
2433 *haverightsp = haverights;
2434 #else
2435 CAP_ALL(haverightsp);
2436 #endif
2437 }
2438 return (0);
2439 }
2440
2441 /*
2442 * Extract the file pointer associated with the specified descriptor for the
2443 * current user process.
2444 *
2445 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2446 * returned.
2447 *
2448 * File's rights will be checked against the capability rights mask.
2449 *
2450 * If an error occured the non-zero error is returned and *fpp is set to
2451 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2452 * responsible for fdrop().
2453 */
2454 static __inline int
2455 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2456 cap_rights_t *needrightsp, u_char *maxprotp)
2457 {
2458 struct filedesc *fdp;
2459 struct file *fp;
2460 cap_rights_t haverights, needrights;
2461 int error;
2462
2463 *fpp = NULL;
2464 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2465 return (EBADF);
2466 if (needrightsp != NULL)
2467 needrights = *needrightsp;
2468 else
2469 cap_rights_init(&needrights);
2470 if (maxprotp != NULL)
2471 cap_rights_set(&needrights, CAP_MMAP);
2472 error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights);
2473 if (error != 0)
2474 return (error);
2475 if (fp->f_ops == &badfileops) {
2476 fdrop(fp, td);
2477 return (EBADF);
2478 }
2479
2480 #ifdef CAPABILITIES
2481 /*
2482 * If requested, convert capability rights to access flags.
2483 */
2484 if (maxprotp != NULL)
2485 *maxprotp = cap_rights_to_vmprot(&haverights);
2486 #else /* !CAPABILITIES */
2487 if (maxprotp != NULL)
2488 *maxprotp = VM_PROT_ALL;
2489 #endif /* CAPABILITIES */
2490
2491 /*
2492 * FREAD and FWRITE failure return EBADF as per POSIX.
2493 */
2494 error = 0;
2495 switch (flags) {
2496 case FREAD:
2497 case FWRITE:
2498 if ((fp->f_flag & flags) == 0)
2499 error = EBADF;
2500 break;
2501 case FEXEC:
2502 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2503 ((fp->f_flag & FWRITE) != 0))
2504 error = EBADF;
2505 break;
2506 case 0:
2507 break;
2508 default:
2509 KASSERT(0, ("wrong flags"));
2510 }
2511
2512 if (error != 0) {
2513 fdrop(fp, td);
2514 return (error);
2515 }
2516
2517 *fpp = fp;
2518 return (0);
2519 }
2520
2521 int
2522 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2523 {
2524
2525 return(_fget(td, fd, fpp, 0, rightsp, NULL));
2526 }
2527
2528 int
2529 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2530 struct file **fpp)
2531 {
2532
2533 return (_fget(td, fd, fpp, 0, rightsp, maxprotp));
2534 }
2535
2536 int
2537 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2538 {
2539
2540 return(_fget(td, fd, fpp, FREAD, rightsp, NULL));
2541 }
2542
2543 int
2544 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2545 {
2546
2547 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2548 }
2549
2550 /*
2551 * Like fget() but loads the underlying vnode, or returns an error if the
2552 * descriptor does not represent a vnode. Note that pipes use vnodes but
2553 * never have VM objects. The returned vnode will be vref()'d.
2554 *
2555 * XXX: what about the unused flags ?
2556 */
2557 static __inline int
2558 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2559 struct vnode **vpp)
2560 {
2561 struct file *fp;
2562 int error;
2563
2564 *vpp = NULL;
2565 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2566 if (error != 0)
2567 return (error);
2568 if (fp->f_vnode == NULL) {
2569 error = EINVAL;
2570 } else {
2571 *vpp = fp->f_vnode;
2572 vref(*vpp);
2573 }
2574 fdrop(fp, td);
2575
2576 return (error);
2577 }
2578
2579 int
2580 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2581 {
2582
2583 return (_fgetvp(td, fd, 0, rightsp, vpp));
2584 }
2585
2586 int
2587 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2588 struct filecaps *havecaps, struct vnode **vpp)
2589 {
2590 struct filedesc *fdp;
2591 struct file *fp;
2592 #ifdef CAPABILITIES
2593 int error;
2594 #endif
2595
2596 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2597 return (EBADF);
2598
2599 fp = fget_locked(fdp, fd);
2600 if (fp == NULL || fp->f_ops == &badfileops)
2601 return (EBADF);
2602
2603 #ifdef CAPABILITIES
2604 if (needrightsp != NULL) {
2605 error = cap_check(cap_rights(fdp, fd), needrightsp);
2606 if (error != 0)
2607 return (error);
2608 }
2609 #endif
2610
2611 if (fp->f_vnode == NULL)
2612 return (EINVAL);
2613
2614 *vpp = fp->f_vnode;
2615 vref(*vpp);
2616 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
2617
2618 return (0);
2619 }
2620
2621 int
2622 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2623 {
2624
2625 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2626 }
2627
2628 int
2629 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2630 {
2631
2632 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2633 }
2634
2635 #ifdef notyet
2636 int
2637 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2638 struct vnode **vpp)
2639 {
2640
2641 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2642 }
2643 #endif
2644
2645 /*
2646 * Like fget() but loads the underlying socket, or returns an error if the
2647 * descriptor does not represent a socket.
2648 *
2649 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2650 * in the future.
2651 *
2652 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2653 * on their file descriptor reference to prevent the socket from being free'd
2654 * during use.
2655 */
2656 int
2657 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp,
2658 u_int *fflagp)
2659 {
2660 struct file *fp;
2661 int error;
2662
2663 *spp = NULL;
2664 if (fflagp != NULL)
2665 *fflagp = 0;
2666 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0)
2667 return (error);
2668 if (fp->f_type != DTYPE_SOCKET) {
2669 error = ENOTSOCK;
2670 } else {
2671 *spp = fp->f_data;
2672 if (fflagp)
2673 *fflagp = fp->f_flag;
2674 SOCK_LOCK(*spp);
2675 soref(*spp);
2676 SOCK_UNLOCK(*spp);
2677 }
2678 fdrop(fp, td);
2679
2680 return (error);
2681 }
2682
2683 /*
2684 * Drop the reference count on the socket and XXX release the SX lock in the
2685 * future. The last reference closes the socket.
2686 *
2687 * Note: fputsock() is deprecated, see comment for fgetsock().
2688 */
2689 void
2690 fputsock(struct socket *so)
2691 {
2692
2693 ACCEPT_LOCK();
2694 SOCK_LOCK(so);
2695 CURVNET_SET(so->so_vnet);
2696 sorele(so);
2697 CURVNET_RESTORE();
2698 }
2699
2700 /*
2701 * Handle the last reference to a file being closed.
2702 */
2703 int
2704 _fdrop(struct file *fp, struct thread *td)
2705 {
2706 int error;
2707
2708 error = 0;
2709 if (fp->f_count != 0)
2710 panic("fdrop: count %d", fp->f_count);
2711 if (fp->f_ops != &badfileops)
2712 error = fo_close(fp, td);
2713 atomic_subtract_int(&openfiles, 1);
2714 crfree(fp->f_cred);
2715 free(fp->f_advice, M_FADVISE);
2716 uma_zfree(file_zone, fp);
2717
2718 return (error);
2719 }
2720
2721 /*
2722 * Apply an advisory lock on a file descriptor.
2723 *
2724 * Just attempt to get a record lock of the requested type on the entire file
2725 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2726 */
2727 #ifndef _SYS_SYSPROTO_H_
2728 struct flock_args {
2729 int fd;
2730 int how;
2731 };
2732 #endif
2733 /* ARGSUSED */
2734 int
2735 sys_flock(struct thread *td, struct flock_args *uap)
2736 {
2737 struct file *fp;
2738 struct vnode *vp;
2739 struct flock lf;
2740 cap_rights_t rights;
2741 int error;
2742
2743 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2744 if (error != 0)
2745 return (error);
2746 if (fp->f_type != DTYPE_VNODE) {
2747 fdrop(fp, td);
2748 return (EOPNOTSUPP);
2749 }
2750
2751 vp = fp->f_vnode;
2752 lf.l_whence = SEEK_SET;
2753 lf.l_start = 0;
2754 lf.l_len = 0;
2755 if (uap->how & LOCK_UN) {
2756 lf.l_type = F_UNLCK;
2757 atomic_clear_int(&fp->f_flag, FHASLOCK);
2758 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2759 goto done2;
2760 }
2761 if (uap->how & LOCK_EX)
2762 lf.l_type = F_WRLCK;
2763 else if (uap->how & LOCK_SH)
2764 lf.l_type = F_RDLCK;
2765 else {
2766 error = EBADF;
2767 goto done2;
2768 }
2769 atomic_set_int(&fp->f_flag, FHASLOCK);
2770 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2771 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2772 done2:
2773 fdrop(fp, td);
2774 return (error);
2775 }
2776 /*
2777 * Duplicate the specified descriptor to a free descriptor.
2778 */
2779 int
2780 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2781 int openerror, int *indxp)
2782 {
2783 struct filedescent *newfde, *oldfde;
2784 struct file *fp;
2785 int error, indx;
2786
2787 KASSERT(openerror == ENODEV || openerror == ENXIO,
2788 ("unexpected error %d in %s", openerror, __func__));
2789
2790 /*
2791 * If the to-be-dup'd fd number is greater than the allowed number
2792 * of file descriptors, or the fd to be dup'd has already been
2793 * closed, then reject.
2794 */
2795 FILEDESC_XLOCK(fdp);
2796 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2797 FILEDESC_XUNLOCK(fdp);
2798 return (EBADF);
2799 }
2800
2801 error = fdalloc(td, 0, &indx);
2802 if (error != 0) {
2803 FILEDESC_XUNLOCK(fdp);
2804 return (error);
2805 }
2806
2807 /*
2808 * There are two cases of interest here.
2809 *
2810 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2811 *
2812 * For ENXIO steal away the file structure from (dfd) and store it in
2813 * (indx). (dfd) is effectively closed by this operation.
2814 */
2815 switch (openerror) {
2816 case ENODEV:
2817 /*
2818 * Check that the mode the file is being opened for is a
2819 * subset of the mode of the existing descriptor.
2820 */
2821 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2822 fdunused(fdp, indx);
2823 FILEDESC_XUNLOCK(fdp);
2824 return (EACCES);
2825 }
2826 fhold(fp);
2827 newfde = &fdp->fd_ofiles[indx];
2828 oldfde = &fdp->fd_ofiles[dfd];
2829 #ifdef CAPABILITIES
2830 seq_write_begin(&newfde->fde_seq);
2831 #endif
2832 memcpy(newfde, oldfde, fde_change_size);
2833 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
2834 #ifdef CAPABILITIES
2835 seq_write_end(&newfde->fde_seq);
2836 #endif
2837 break;
2838 case ENXIO:
2839 /*
2840 * Steal away the file pointer from dfd and stuff it into indx.
2841 */
2842 newfde = &fdp->fd_ofiles[indx];
2843 oldfde = &fdp->fd_ofiles[dfd];
2844 #ifdef CAPABILITIES
2845 seq_write_begin(&newfde->fde_seq);
2846 #endif
2847 memcpy(newfde, oldfde, fde_change_size);
2848 bzero(oldfde, fde_change_size);
2849 fdunused(fdp, dfd);
2850 #ifdef CAPABILITIES
2851 seq_write_end(&newfde->fde_seq);
2852 #endif
2853 break;
2854 }
2855 FILEDESC_XUNLOCK(fdp);
2856 *indxp = indx;
2857 return (0);
2858 }
2859
2860 /*
2861 * Scan all active processes and prisons to see if any of them have a current
2862 * or root directory of `olddp'. If so, replace them with the new mount point.
2863 */
2864 void
2865 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2866 {
2867 struct filedesc *fdp;
2868 struct prison *pr;
2869 struct proc *p;
2870 int nrele;
2871
2872 if (vrefcnt(olddp) == 1)
2873 return;
2874 nrele = 0;
2875 sx_slock(&allproc_lock);
2876 FOREACH_PROC_IN_SYSTEM(p) {
2877 fdp = fdhold(p);
2878 if (fdp == NULL)
2879 continue;
2880 FILEDESC_XLOCK(fdp);
2881 if (fdp->fd_cdir == olddp) {
2882 vref(newdp);
2883 fdp->fd_cdir = newdp;
2884 nrele++;
2885 }
2886 if (fdp->fd_rdir == olddp) {
2887 vref(newdp);
2888 fdp->fd_rdir = newdp;
2889 nrele++;
2890 }
2891 if (fdp->fd_jdir == olddp) {
2892 vref(newdp);
2893 fdp->fd_jdir = newdp;
2894 nrele++;
2895 }
2896 FILEDESC_XUNLOCK(fdp);
2897 fddrop(fdp);
2898 }
2899 sx_sunlock(&allproc_lock);
2900 if (rootvnode == olddp) {
2901 vref(newdp);
2902 rootvnode = newdp;
2903 nrele++;
2904 }
2905 mtx_lock(&prison0.pr_mtx);
2906 if (prison0.pr_root == olddp) {
2907 vref(newdp);
2908 prison0.pr_root = newdp;
2909 nrele++;
2910 }
2911 mtx_unlock(&prison0.pr_mtx);
2912 sx_slock(&allprison_lock);
2913 TAILQ_FOREACH(pr, &allprison, pr_list) {
2914 mtx_lock(&pr->pr_mtx);
2915 if (pr->pr_root == olddp) {
2916 vref(newdp);
2917 pr->pr_root = newdp;
2918 nrele++;
2919 }
2920 mtx_unlock(&pr->pr_mtx);
2921 }
2922 sx_sunlock(&allprison_lock);
2923 while (nrele--)
2924 vrele(olddp);
2925 }
2926
2927 struct filedesc_to_leader *
2928 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2929 {
2930 struct filedesc_to_leader *fdtol;
2931
2932 fdtol = malloc(sizeof(struct filedesc_to_leader),
2933 M_FILEDESC_TO_LEADER, M_WAITOK);
2934 fdtol->fdl_refcount = 1;
2935 fdtol->fdl_holdcount = 0;
2936 fdtol->fdl_wakeup = 0;
2937 fdtol->fdl_leader = leader;
2938 if (old != NULL) {
2939 FILEDESC_XLOCK(fdp);
2940 fdtol->fdl_next = old->fdl_next;
2941 fdtol->fdl_prev = old;
2942 old->fdl_next = fdtol;
2943 fdtol->fdl_next->fdl_prev = fdtol;
2944 FILEDESC_XUNLOCK(fdp);
2945 } else {
2946 fdtol->fdl_next = fdtol;
2947 fdtol->fdl_prev = fdtol;
2948 }
2949 return (fdtol);
2950 }
2951
2952 /*
2953 * Get file structures globally.
2954 */
2955 static int
2956 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2957 {
2958 struct xfile xf;
2959 struct filedesc *fdp;
2960 struct file *fp;
2961 struct proc *p;
2962 int error, n;
2963
2964 error = sysctl_wire_old_buffer(req, 0);
2965 if (error != 0)
2966 return (error);
2967 if (req->oldptr == NULL) {
2968 n = 0;
2969 sx_slock(&allproc_lock);
2970 FOREACH_PROC_IN_SYSTEM(p) {
2971 if (p->p_state == PRS_NEW)
2972 continue;
2973 fdp = fdhold(p);
2974 if (fdp == NULL)
2975 continue;
2976 /* overestimates sparse tables. */
2977 if (fdp->fd_lastfile > 0)
2978 n += fdp->fd_lastfile;
2979 fddrop(fdp);
2980 }
2981 sx_sunlock(&allproc_lock);
2982 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2983 }
2984 error = 0;
2985 bzero(&xf, sizeof(xf));
2986 xf.xf_size = sizeof(xf);
2987 sx_slock(&allproc_lock);
2988 FOREACH_PROC_IN_SYSTEM(p) {
2989 PROC_LOCK(p);
2990 if (p->p_state == PRS_NEW) {
2991 PROC_UNLOCK(p);
2992 continue;
2993 }
2994 if (p_cansee(req->td, p) != 0) {
2995 PROC_UNLOCK(p);
2996 continue;
2997 }
2998 xf.xf_pid = p->p_pid;
2999 xf.xf_uid = p->p_ucred->cr_uid;
3000 PROC_UNLOCK(p);
3001 fdp = fdhold(p);
3002 if (fdp == NULL)
3003 continue;
3004 FILEDESC_SLOCK(fdp);
3005 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3006 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3007 continue;
3008 xf.xf_fd = n;
3009 xf.xf_file = fp;
3010 xf.xf_data = fp->f_data;
3011 xf.xf_vnode = fp->f_vnode;
3012 xf.xf_type = fp->f_type;
3013 xf.xf_count = fp->f_count;
3014 xf.xf_msgcount = 0;
3015 xf.xf_offset = foffset_get(fp);
3016 xf.xf_flag = fp->f_flag;
3017 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3018 if (error)
3019 break;
3020 }
3021 FILEDESC_SUNLOCK(fdp);
3022 fddrop(fdp);
3023 if (error)
3024 break;
3025 }
3026 sx_sunlock(&allproc_lock);
3027 return (error);
3028 }
3029
3030 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3031 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3032
3033 #ifdef KINFO_OFILE_SIZE
3034 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3035 #endif
3036
3037 #ifdef COMPAT_FREEBSD7
3038 static int
3039 export_vnode_for_osysctl(struct vnode *vp, int type,
3040 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
3041 {
3042 int error;
3043 char *fullpath, *freepath;
3044
3045 bzero(kif, sizeof(*kif));
3046 kif->kf_structsize = sizeof(*kif);
3047
3048 vref(vp);
3049 kif->kf_fd = type;
3050 kif->kf_type = KF_TYPE_VNODE;
3051 /* This function only handles directories. */
3052 if (vp->v_type != VDIR) {
3053 vrele(vp);
3054 return (ENOTDIR);
3055 }
3056 kif->kf_vnode_type = KF_VTYPE_VDIR;
3057
3058 /*
3059 * This is not a true file descriptor, so we set a bogus refcount
3060 * and offset to indicate these fields should be ignored.
3061 */
3062 kif->kf_ref_count = -1;
3063 kif->kf_offset = -1;
3064
3065 freepath = NULL;
3066 fullpath = "-";
3067 FILEDESC_SUNLOCK(fdp);
3068 vn_fullpath(curthread, vp, &fullpath, &freepath);
3069 vrele(vp);
3070 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3071 if (freepath != NULL)
3072 free(freepath, M_TEMP);
3073 error = SYSCTL_OUT(req, kif, sizeof(*kif));
3074 FILEDESC_SLOCK(fdp);
3075 return (error);
3076 }
3077
3078 /*
3079 * Get per-process file descriptors for use by procstat(1), et al.
3080 */
3081 static int
3082 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3083 {
3084 char *fullpath, *freepath;
3085 struct kinfo_ofile *kif;
3086 struct filedesc *fdp;
3087 int error, i, *name;
3088 struct shmfd *shmfd;
3089 struct socket *so;
3090 struct vnode *vp;
3091 struct ksem *ks;
3092 struct file *fp;
3093 struct proc *p;
3094 struct tty *tp;
3095
3096 name = (int *)arg1;
3097 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3098 if (error != 0)
3099 return (error);
3100 fdp = fdhold(p);
3101 PROC_UNLOCK(p);
3102 if (fdp == NULL)
3103 return (ENOENT);
3104 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3105 FILEDESC_SLOCK(fdp);
3106 if (fdp->fd_cdir != NULL)
3107 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3108 fdp, req);
3109 if (fdp->fd_rdir != NULL)
3110 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3111 fdp, req);
3112 if (fdp->fd_jdir != NULL)
3113 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3114 fdp, req);
3115 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3116 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3117 continue;
3118 bzero(kif, sizeof(*kif));
3119 kif->kf_structsize = sizeof(*kif);
3120 ks = NULL;
3121 vp = NULL;
3122 so = NULL;
3123 tp = NULL;
3124 shmfd = NULL;
3125 kif->kf_fd = i;
3126
3127 switch (fp->f_type) {
3128 case DTYPE_VNODE:
3129 kif->kf_type = KF_TYPE_VNODE;
3130 vp = fp->f_vnode;
3131 break;
3132
3133 case DTYPE_SOCKET:
3134 kif->kf_type = KF_TYPE_SOCKET;
3135 so = fp->f_data;
3136 break;
3137
3138 case DTYPE_PIPE:
3139 kif->kf_type = KF_TYPE_PIPE;
3140 break;
3141
3142 case DTYPE_FIFO:
3143 kif->kf_type = KF_TYPE_FIFO;
3144 vp = fp->f_vnode;
3145 break;
3146
3147 case DTYPE_KQUEUE:
3148 kif->kf_type = KF_TYPE_KQUEUE;
3149 break;
3150
3151 case DTYPE_CRYPTO:
3152 kif->kf_type = KF_TYPE_CRYPTO;
3153 break;
3154
3155 case DTYPE_MQUEUE:
3156 kif->kf_type = KF_TYPE_MQUEUE;
3157 break;
3158
3159 case DTYPE_SHM:
3160 kif->kf_type = KF_TYPE_SHM;
3161 shmfd = fp->f_data;
3162 break;
3163
3164 case DTYPE_SEM:
3165 kif->kf_type = KF_TYPE_SEM;
3166 ks = fp->f_data;
3167 break;
3168
3169 case DTYPE_PTS:
3170 kif->kf_type = KF_TYPE_PTS;
3171 tp = fp->f_data;
3172 break;
3173
3174 #ifdef PROCDESC
3175 case DTYPE_PROCDESC:
3176 kif->kf_type = KF_TYPE_PROCDESC;
3177 break;
3178 #endif
3179
3180 default:
3181 kif->kf_type = KF_TYPE_UNKNOWN;
3182 break;
3183 }
3184 kif->kf_ref_count = fp->f_count;
3185 if (fp->f_flag & FREAD)
3186 kif->kf_flags |= KF_FLAG_READ;
3187 if (fp->f_flag & FWRITE)
3188 kif->kf_flags |= KF_FLAG_WRITE;
3189 if (fp->f_flag & FAPPEND)
3190 kif->kf_flags |= KF_FLAG_APPEND;
3191 if (fp->f_flag & FASYNC)
3192 kif->kf_flags |= KF_FLAG_ASYNC;
3193 if (fp->f_flag & FFSYNC)
3194 kif->kf_flags |= KF_FLAG_FSYNC;
3195 if (fp->f_flag & FNONBLOCK)
3196 kif->kf_flags |= KF_FLAG_NONBLOCK;
3197 if (fp->f_flag & O_DIRECT)
3198 kif->kf_flags |= KF_FLAG_DIRECT;
3199 if (fp->f_flag & FHASLOCK)
3200 kif->kf_flags |= KF_FLAG_HASLOCK;
3201 kif->kf_offset = foffset_get(fp);
3202 if (vp != NULL) {
3203 vref(vp);
3204 switch (vp->v_type) {
3205 case VNON:
3206 kif->kf_vnode_type = KF_VTYPE_VNON;
3207 break;
3208 case VREG:
3209 kif->kf_vnode_type = KF_VTYPE_VREG;
3210 break;
3211 case VDIR:
3212 kif->kf_vnode_type = KF_VTYPE_VDIR;
3213 break;
3214 case VBLK:
3215 kif->kf_vnode_type = KF_VTYPE_VBLK;
3216 break;
3217 case VCHR:
3218 kif->kf_vnode_type = KF_VTYPE_VCHR;
3219 break;
3220 case VLNK:
3221 kif->kf_vnode_type = KF_VTYPE_VLNK;
3222 break;
3223 case VSOCK:
3224 kif->kf_vnode_type = KF_VTYPE_VSOCK;
3225 break;
3226 case VFIFO:
3227 kif->kf_vnode_type = KF_VTYPE_VFIFO;
3228 break;
3229 case VBAD:
3230 kif->kf_vnode_type = KF_VTYPE_VBAD;
3231 break;
3232 default:
3233 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3234 break;
3235 }
3236 /*
3237 * It is OK to drop the filedesc lock here as we will
3238 * re-validate and re-evaluate its properties when
3239 * the loop continues.
3240 */
3241 freepath = NULL;
3242 fullpath = "-";
3243 FILEDESC_SUNLOCK(fdp);
3244 vn_fullpath(curthread, vp, &fullpath, &freepath);
3245 vrele(vp);
3246 strlcpy(kif->kf_path, fullpath,
3247 sizeof(kif->kf_path));
3248 if (freepath != NULL)
3249 free(freepath, M_TEMP);
3250 FILEDESC_SLOCK(fdp);
3251 }
3252 if (so != NULL) {
3253 struct sockaddr *sa;
3254
3255 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3256 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3257 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3258 free(sa, M_SONAME);
3259 }
3260 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3261 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3262 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3263 free(sa, M_SONAME);
3264 }
3265 kif->kf_sock_domain =
3266 so->so_proto->pr_domain->dom_family;
3267 kif->kf_sock_type = so->so_type;
3268 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3269 }
3270 if (tp != NULL) {
3271 strlcpy(kif->kf_path, tty_devname(tp),
3272 sizeof(kif->kf_path));
3273 }
3274 if (shmfd != NULL)
3275 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
3276 if (ks != NULL && ksem_info != NULL)
3277 ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL);
3278 error = SYSCTL_OUT(req, kif, sizeof(*kif));
3279 if (error)
3280 break;
3281 }
3282 FILEDESC_SUNLOCK(fdp);
3283 fddrop(fdp);
3284 free(kif, M_TEMP);
3285 return (0);
3286 }
3287
3288 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3289 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3290 "Process ofiledesc entries");
3291 #endif /* COMPAT_FREEBSD7 */
3292
3293 #ifdef KINFO_FILE_SIZE
3294 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3295 #endif
3296
3297 struct export_fd_buf {
3298 struct filedesc *fdp;
3299 struct sbuf *sb;
3300 ssize_t remainder;
3301 struct kinfo_file kif;
3302 };
3303
3304 static int
3305 export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt,
3306 int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf)
3307 {
3308 struct {
3309 int fflag;
3310 int kf_fflag;
3311 } fflags_table[] = {
3312 { FAPPEND, KF_FLAG_APPEND },
3313 { FASYNC, KF_FLAG_ASYNC },
3314 { FFSYNC, KF_FLAG_FSYNC },
3315 { FHASLOCK, KF_FLAG_HASLOCK },
3316 { FNONBLOCK, KF_FLAG_NONBLOCK },
3317 { FREAD, KF_FLAG_READ },
3318 { FWRITE, KF_FLAG_WRITE },
3319 { O_CREAT, KF_FLAG_CREAT },
3320 { O_DIRECT, KF_FLAG_DIRECT },
3321 { O_EXCL, KF_FLAG_EXCL },
3322 { O_EXEC, KF_FLAG_EXEC },
3323 { O_EXLOCK, KF_FLAG_EXLOCK },
3324 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3325 { O_SHLOCK, KF_FLAG_SHLOCK },
3326 { O_TRUNC, KF_FLAG_TRUNC }
3327 };
3328 #define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table))
3329 struct kinfo_file *kif;
3330 struct vnode *vp;
3331 int error, locked;
3332 unsigned int i;
3333
3334 if (efbuf->remainder == 0)
3335 return (0);
3336 kif = &efbuf->kif;
3337 bzero(kif, sizeof(*kif));
3338 locked = efbuf->fdp != NULL;
3339 switch (type) {
3340 case KF_TYPE_FIFO:
3341 case KF_TYPE_VNODE:
3342 if (locked) {
3343 FILEDESC_SUNLOCK(efbuf->fdp);
3344 locked = 0;
3345 }
3346 vp = (struct vnode *)data;
3347 error = fill_vnode_info(vp, kif);
3348 vrele(vp);
3349 break;
3350 case KF_TYPE_SOCKET:
3351 error = fill_socket_info((struct socket *)data, kif);
3352 break;
3353 case KF_TYPE_PIPE:
3354 error = fill_pipe_info((struct pipe *)data, kif);
3355 break;
3356 case KF_TYPE_PTS:
3357 error = fill_pts_info((struct tty *)data, kif);
3358 break;
3359 case KF_TYPE_PROCDESC:
3360 error = fill_procdesc_info((struct procdesc *)data, kif);
3361 break;
3362 case KF_TYPE_SEM:
3363 error = fill_sem_info((struct file *)data, kif);
3364 break;
3365 case KF_TYPE_SHM:
3366 error = fill_shm_info((struct file *)data, kif);
3367 break;
3368 default:
3369 error = 0;
3370 }
3371 if (error == 0)
3372 kif->kf_status |= KF_ATTR_VALID;
3373
3374 /*
3375 * Translate file access flags.
3376 */
3377 for (i = 0; i < NFFLAGS; i++)
3378 if (fflags & fflags_table[i].fflag)
3379 kif->kf_flags |= fflags_table[i].kf_fflag;
3380 if (rightsp != NULL)
3381 kif->kf_cap_rights = *rightsp;
3382 else
3383 cap_rights_init(&kif->kf_cap_rights);
3384 kif->kf_fd = fd;
3385 kif->kf_type = type;
3386 kif->kf_ref_count = refcnt;
3387 kif->kf_offset = offset;
3388 /* Pack record size down */
3389 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3390 strlen(kif->kf_path) + 1;
3391 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3392 if (efbuf->remainder != -1) {
3393 if (efbuf->remainder < kif->kf_structsize) {
3394 /* Terminate export. */
3395 efbuf->remainder = 0;
3396 if (efbuf->fdp != NULL && !locked)
3397 FILEDESC_SLOCK(efbuf->fdp);
3398 return (0);
3399 }
3400 efbuf->remainder -= kif->kf_structsize;
3401 }
3402 if (locked)
3403 FILEDESC_SUNLOCK(efbuf->fdp);
3404 error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM;
3405 if (efbuf->fdp != NULL)
3406 FILEDESC_SLOCK(efbuf->fdp);
3407 return (error);
3408 }
3409
3410 /*
3411 * Store a process file descriptor information to sbuf.
3412 *
3413 * Takes a locked proc as argument, and returns with the proc unlocked.
3414 */
3415 int
3416 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3417 {
3418 struct file *fp;
3419 struct filedesc *fdp;
3420 struct export_fd_buf *efbuf;
3421 struct vnode *cttyvp, *textvp, *tracevp;
3422 int64_t offset;
3423 void *data;
3424 int error, i;
3425 int type, refcnt, fflags;
3426 cap_rights_t rights;
3427
3428 PROC_LOCK_ASSERT(p, MA_OWNED);
3429
3430 /* ktrace vnode */
3431 tracevp = p->p_tracevp;
3432 if (tracevp != NULL)
3433 vref(tracevp);
3434 /* text vnode */
3435 textvp = p->p_textvp;
3436 if (textvp != NULL)
3437 vref(textvp);
3438 /* Controlling tty. */
3439 cttyvp = NULL;
3440 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3441 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3442 if (cttyvp != NULL)
3443 vref(cttyvp);
3444 }
3445 fdp = fdhold(p);
3446 PROC_UNLOCK(p);
3447 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3448 efbuf->fdp = NULL;
3449 efbuf->sb = sb;
3450 efbuf->remainder = maxlen;
3451 if (tracevp != NULL)
3452 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
3453 FREAD | FWRITE, -1, -1, NULL, efbuf);
3454 if (textvp != NULL)
3455 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
3456 FREAD, -1, -1, NULL, efbuf);
3457 if (cttyvp != NULL)
3458 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
3459 FREAD | FWRITE, -1, -1, NULL, efbuf);
3460 error = 0;
3461 if (fdp == NULL)
3462 goto fail;
3463 efbuf->fdp = fdp;
3464 FILEDESC_SLOCK(fdp);
3465 /* working directory */
3466 if (fdp->fd_cdir != NULL) {
3467 vref(fdp->fd_cdir);
3468 data = fdp->fd_cdir;
3469 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
3470 FREAD, -1, -1, NULL, efbuf);
3471 }
3472 /* root directory */
3473 if (fdp->fd_rdir != NULL) {
3474 vref(fdp->fd_rdir);
3475 data = fdp->fd_rdir;
3476 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
3477 FREAD, -1, -1, NULL, efbuf);
3478 }
3479 /* jail directory */
3480 if (fdp->fd_jdir != NULL) {
3481 vref(fdp->fd_jdir);
3482 data = fdp->fd_jdir;
3483 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
3484 FREAD, -1, -1, NULL, efbuf);
3485 }
3486 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3487 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3488 continue;
3489 data = NULL;
3490 #ifdef CAPABILITIES
3491 rights = *cap_rights(fdp, i);
3492 #else /* !CAPABILITIES */
3493 cap_rights_init(&rights);
3494 #endif
3495 switch (fp->f_type) {
3496 case DTYPE_VNODE:
3497 type = KF_TYPE_VNODE;
3498 vref(fp->f_vnode);
3499 data = fp->f_vnode;
3500 break;
3501
3502 case DTYPE_SOCKET:
3503 type = KF_TYPE_SOCKET;
3504 data = fp->f_data;
3505 break;
3506
3507 case DTYPE_PIPE:
3508 type = KF_TYPE_PIPE;
3509 data = fp->f_data;
3510 break;
3511
3512 case DTYPE_FIFO:
3513 type = KF_TYPE_FIFO;
3514 vref(fp->f_vnode);
3515 data = fp->f_vnode;
3516 break;
3517
3518 case DTYPE_KQUEUE:
3519 type = KF_TYPE_KQUEUE;
3520 break;
3521
3522 case DTYPE_CRYPTO:
3523 type = KF_TYPE_CRYPTO;
3524 break;
3525
3526 case DTYPE_MQUEUE:
3527 type = KF_TYPE_MQUEUE;
3528 break;
3529
3530 case DTYPE_SHM:
3531 type = KF_TYPE_SHM;
3532 data = fp;
3533 break;
3534
3535 case DTYPE_SEM:
3536 type = KF_TYPE_SEM;
3537 data = fp;
3538 break;
3539
3540 case DTYPE_PTS:
3541 type = KF_TYPE_PTS;
3542 data = fp->f_data;
3543 break;
3544
3545 #ifdef PROCDESC
3546 case DTYPE_PROCDESC:
3547 type = KF_TYPE_PROCDESC;
3548 data = fp->f_data;
3549 break;
3550 #endif
3551
3552 default:
3553 type = KF_TYPE_UNKNOWN;
3554 break;
3555 }
3556 refcnt = fp->f_count;
3557 fflags = fp->f_flag;
3558 offset = foffset_get(fp);
3559
3560 /*
3561 * Create sysctl entry.
3562 * It is OK to drop the filedesc lock here as we will
3563 * re-validate and re-evaluate its properties when
3564 * the loop continues.
3565 */
3566 error = export_fd_to_sb(data, type, i, fflags, refcnt,
3567 offset, &rights, efbuf);
3568 if (error != 0)
3569 break;
3570 }
3571 FILEDESC_SUNLOCK(fdp);
3572 fddrop(fdp);
3573 fail:
3574 free(efbuf, M_TEMP);
3575 return (error);
3576 }
3577
3578 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3579
3580 /*
3581 * Get per-process file descriptors for use by procstat(1), et al.
3582 */
3583 static int
3584 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3585 {
3586 struct sbuf sb;
3587 struct proc *p;
3588 ssize_t maxlen;
3589 int error, error2, *name;
3590
3591 name = (int *)arg1;
3592
3593 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3594 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3595 if (error != 0) {
3596 sbuf_delete(&sb);
3597 return (error);
3598 }
3599 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3600 error = kern_proc_filedesc_out(p, &sb, maxlen);
3601 error2 = sbuf_finish(&sb);
3602 sbuf_delete(&sb);
3603 return (error != 0 ? error : error2);
3604 }
3605
3606 int
3607 vntype_to_kinfo(int vtype)
3608 {
3609 struct {
3610 int vtype;
3611 int kf_vtype;
3612 } vtypes_table[] = {
3613 { VBAD, KF_VTYPE_VBAD },
3614 { VBLK, KF_VTYPE_VBLK },
3615 { VCHR, KF_VTYPE_VCHR },
3616 { VDIR, KF_VTYPE_VDIR },
3617 { VFIFO, KF_VTYPE_VFIFO },
3618 { VLNK, KF_VTYPE_VLNK },
3619 { VNON, KF_VTYPE_VNON },
3620 { VREG, KF_VTYPE_VREG },
3621 { VSOCK, KF_VTYPE_VSOCK }
3622 };
3623 #define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table))
3624 unsigned int i;
3625
3626 /*
3627 * Perform vtype translation.
3628 */
3629 for (i = 0; i < NVTYPES; i++)
3630 if (vtypes_table[i].vtype == vtype)
3631 break;
3632 if (i < NVTYPES)
3633 return (vtypes_table[i].kf_vtype);
3634
3635 return (KF_VTYPE_UNKNOWN);
3636 }
3637
3638 static int
3639 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
3640 {
3641 struct vattr va;
3642 char *fullpath, *freepath;
3643 int error;
3644
3645 if (vp == NULL)
3646 return (1);
3647 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
3648 freepath = NULL;
3649 fullpath = "-";
3650 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
3651 if (error == 0) {
3652 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3653 }
3654 if (freepath != NULL)
3655 free(freepath, M_TEMP);
3656
3657 /*
3658 * Retrieve vnode attributes.
3659 */
3660 va.va_fsid = VNOVAL;
3661 va.va_rdev = NODEV;
3662 vn_lock(vp, LK_SHARED | LK_RETRY);
3663 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
3664 VOP_UNLOCK(vp, 0);
3665 if (error != 0)
3666 return (error);
3667 if (va.va_fsid != VNOVAL)
3668 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
3669 else
3670 kif->kf_un.kf_file.kf_file_fsid =
3671 vp->v_mount->mnt_stat.f_fsid.val[0];
3672 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
3673 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
3674 kif->kf_un.kf_file.kf_file_size = va.va_size;
3675 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
3676 return (0);
3677 }
3678
3679 static int
3680 fill_socket_info(struct socket *so, struct kinfo_file *kif)
3681 {
3682 struct sockaddr *sa;
3683 struct inpcb *inpcb;
3684 struct unpcb *unpcb;
3685 int error;
3686
3687 if (so == NULL)
3688 return (1);
3689 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
3690 kif->kf_sock_type = so->so_type;
3691 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3692 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
3693 switch(kif->kf_sock_domain) {
3694 case AF_INET:
3695 case AF_INET6:
3696 if (kif->kf_sock_protocol == IPPROTO_TCP) {
3697 if (so->so_pcb != NULL) {
3698 inpcb = (struct inpcb *)(so->so_pcb);
3699 kif->kf_un.kf_sock.kf_sock_inpcb =
3700 (uintptr_t)inpcb->inp_ppcb;
3701 }
3702 }
3703 break;
3704 case AF_UNIX:
3705 if (so->so_pcb != NULL) {
3706 unpcb = (struct unpcb *)(so->so_pcb);
3707 if (unpcb->unp_conn) {
3708 kif->kf_un.kf_sock.kf_sock_unpconn =
3709 (uintptr_t)unpcb->unp_conn;
3710 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
3711 so->so_rcv.sb_state;
3712 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
3713 so->so_snd.sb_state;
3714 }
3715 }
3716 break;
3717 }
3718 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
3719 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3720 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3721 free(sa, M_SONAME);
3722 }
3723 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
3724 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3725 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3726 free(sa, M_SONAME);
3727 }
3728 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
3729 sizeof(kif->kf_path));
3730 return (0);
3731 }
3732
3733 static int
3734 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
3735 {
3736
3737 if (tp == NULL)
3738 return (1);
3739 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
3740 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
3741 return (0);
3742 }
3743
3744 static int
3745 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
3746 {
3747
3748 if (pi == NULL)
3749 return (1);
3750 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
3751 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
3752 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
3753 return (0);
3754 }
3755
3756 static int
3757 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
3758 {
3759
3760 if (pdp == NULL)
3761 return (1);
3762 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
3763 return (0);
3764 }
3765
3766 static int
3767 fill_sem_info(struct file *fp, struct kinfo_file *kif)
3768 {
3769 struct thread *td;
3770 struct stat sb;
3771
3772 td = curthread;
3773 if (fp->f_data == NULL)
3774 return (1);
3775 if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3776 return (1);
3777 if (ksem_info == NULL)
3778 return (1);
3779 ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path),
3780 &kif->kf_un.kf_sem.kf_sem_value);
3781 kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode;
3782 return (0);
3783 }
3784
3785 static int
3786 fill_shm_info(struct file *fp, struct kinfo_file *kif)
3787 {
3788 struct thread *td;
3789 struct stat sb;
3790
3791 td = curthread;
3792 if (fp->f_data == NULL)
3793 return (1);
3794 if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3795 return (1);
3796 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
3797 kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
3798 kif->kf_un.kf_file.kf_file_size = sb.st_size;
3799 return (0);
3800 }
3801
3802 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3803 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3804 "Process filedesc entries");
3805
3806 #ifdef DDB
3807 /*
3808 * For the purposes of debugging, generate a human-readable string for the
3809 * file type.
3810 */
3811 static const char *
3812 file_type_to_name(short type)
3813 {
3814
3815 switch (type) {
3816 case 0:
3817 return ("zero");
3818 case DTYPE_VNODE:
3819 return ("vnod");
3820 case DTYPE_SOCKET:
3821 return ("sock");
3822 case DTYPE_PIPE:
3823 return ("pipe");
3824 case DTYPE_FIFO:
3825 return ("fifo");
3826 case DTYPE_KQUEUE:
3827 return ("kque");
3828 case DTYPE_CRYPTO:
3829 return ("crpt");
3830 case DTYPE_MQUEUE:
3831 return ("mque");
3832 case DTYPE_SHM:
3833 return ("shm");
3834 case DTYPE_SEM:
3835 return ("ksem");
3836 default:
3837 return ("unkn");
3838 }
3839 }
3840
3841 /*
3842 * For the purposes of debugging, identify a process (if any, perhaps one of
3843 * many) that references the passed file in its file descriptor array. Return
3844 * NULL if none.
3845 */
3846 static struct proc *
3847 file_to_first_proc(struct file *fp)
3848 {
3849 struct filedesc *fdp;
3850 struct proc *p;
3851 int n;
3852
3853 FOREACH_PROC_IN_SYSTEM(p) {
3854 if (p->p_state == PRS_NEW)
3855 continue;
3856 fdp = p->p_fd;
3857 if (fdp == NULL)
3858 continue;
3859 for (n = 0; n <= fdp->fd_lastfile; n++) {
3860 if (fp == fdp->fd_ofiles[n].fde_file)
3861 return (p);
3862 }
3863 }
3864 return (NULL);
3865 }
3866
3867 static void
3868 db_print_file(struct file *fp, int header)
3869 {
3870 struct proc *p;
3871
3872 if (header)
3873 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3874 "File", "Type", "Data", "Flag", "GCFl", "Count",
3875 "MCount", "Vnode", "FPID", "FCmd");
3876 p = file_to_first_proc(fp);
3877 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3878 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3879 0, fp->f_count, 0, fp->f_vnode,
3880 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3881 }
3882
3883 DB_SHOW_COMMAND(file, db_show_file)
3884 {
3885 struct file *fp;
3886
3887 if (!have_addr) {
3888 db_printf("usage: show file <addr>\n");
3889 return;
3890 }
3891 fp = (struct file *)addr;
3892 db_print_file(fp, 1);
3893 }
3894
3895 DB_SHOW_COMMAND(files, db_show_files)
3896 {
3897 struct filedesc *fdp;
3898 struct file *fp;
3899 struct proc *p;
3900 int header;
3901 int n;
3902
3903 header = 1;
3904 FOREACH_PROC_IN_SYSTEM(p) {
3905 if (p->p_state == PRS_NEW)
3906 continue;
3907 if ((fdp = p->p_fd) == NULL)
3908 continue;
3909 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3910 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3911 continue;
3912 db_print_file(fp, header);
3913 header = 0;
3914 }
3915 }
3916 }
3917 #endif
3918
3919 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3920 &maxfilesperproc, 0, "Maximum files allowed open per process");
3921
3922 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3923 &maxfiles, 0, "Maximum number of files");
3924
3925 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3926 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3927
3928 /* ARGSUSED*/
3929 static void
3930 filelistinit(void *dummy)
3931 {
3932
3933 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3934 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3935 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3936 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3937 }
3938 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3939
3940 /*-------------------------------------------------------------------*/
3941
3942 static int
3943 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3944 int flags, struct thread *td)
3945 {
3946
3947 return (EBADF);
3948 }
3949
3950 static int
3951 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3952 struct thread *td)
3953 {
3954
3955 return (EINVAL);
3956 }
3957
3958 static int
3959 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3960 struct thread *td)
3961 {
3962
3963 return (EBADF);
3964 }
3965
3966 static int
3967 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3968 struct thread *td)
3969 {
3970
3971 return (0);
3972 }
3973
3974 static int
3975 badfo_kqfilter(struct file *fp, struct knote *kn)
3976 {
3977
3978 return (EBADF);
3979 }
3980
3981 static int
3982 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3983 struct thread *td)
3984 {
3985
3986 return (EBADF);
3987 }
3988
3989 static int
3990 badfo_close(struct file *fp, struct thread *td)
3991 {
3992
3993 return (EBADF);
3994 }
3995
3996 static int
3997 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3998 struct thread *td)
3999 {
4000
4001 return (EBADF);
4002 }
4003
4004 static int
4005 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4006 struct thread *td)
4007 {
4008
4009 return (EBADF);
4010 }
4011
4012 static int
4013 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4014 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4015 int kflags, struct thread *td)
4016 {
4017
4018 return (EBADF);
4019 }
4020
4021 struct fileops badfileops = {
4022 .fo_read = badfo_readwrite,
4023 .fo_write = badfo_readwrite,
4024 .fo_truncate = badfo_truncate,
4025 .fo_ioctl = badfo_ioctl,
4026 .fo_poll = badfo_poll,
4027 .fo_kqfilter = badfo_kqfilter,
4028 .fo_stat = badfo_stat,
4029 .fo_close = badfo_close,
4030 .fo_chmod = badfo_chmod,
4031 .fo_chown = badfo_chown,
4032 .fo_sendfile = badfo_sendfile,
4033 };
4034
4035 int
4036 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4037 struct thread *td)
4038 {
4039
4040 return (EINVAL);
4041 }
4042
4043 int
4044 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4045 struct thread *td)
4046 {
4047
4048 return (EINVAL);
4049 }
4050
4051 int
4052 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4053 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4054 int kflags, struct thread *td)
4055 {
4056
4057 return (EINVAL);
4058 }
4059
4060 /*-------------------------------------------------------------------*/
4061
4062 /*
4063 * File Descriptor pseudo-device driver (/dev/fd/).
4064 *
4065 * Opening minor device N dup()s the file (if any) connected to file
4066 * descriptor N belonging to the calling process. Note that this driver
4067 * consists of only the ``open()'' routine, because all subsequent
4068 * references to this file will be direct to the other driver.
4069 *
4070 * XXX: we could give this one a cloning event handler if necessary.
4071 */
4072
4073 /* ARGSUSED */
4074 static int
4075 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4076 {
4077
4078 /*
4079 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4080 * the file descriptor being sought for duplication. The error
4081 * return ensures that the vnode for this device will be released
4082 * by vn_open. Open will detect this special error and take the
4083 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4084 * will simply report the error.
4085 */
4086 td->td_dupfd = dev2unit(dev);
4087 return (ENODEV);
4088 }
4089
4090 static struct cdevsw fildesc_cdevsw = {
4091 .d_version = D_VERSION,
4092 .d_open = fdopen,
4093 .d_name = "FD",
4094 };
4095
4096 static void
4097 fildesc_drvinit(void *unused)
4098 {
4099 struct cdev *dev;
4100
4101 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4102 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4103 make_dev_alias(dev, "stdin");
4104 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4105 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4106 make_dev_alias(dev, "stdout");
4107 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4108 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4109 make_dev_alias(dev, "stderr");
4110 }
4111
4112 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
Cache object: ccb5213878f35634dca348ed08cba8c5
|