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/8.0/sys/kern/kern_descrip.c 195104 2009-06-27 13:58:44Z rwatson $");
39
40 #include "opt_compat.h"
41 #include "opt_ddb.h"
42 #include "opt_ktrace.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46
47 #include <sys/conf.h>
48 #include <sys/domain.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/filedesc.h>
52 #include <sys/filio.h>
53 #include <sys/jail.h>
54 #include <sys/kernel.h>
55 #include <sys/limits.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/mqueue.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/priv.h>
63 #include <sys/proc.h>
64 #include <sys/protosw.h>
65 #include <sys/resourcevar.h>
66 #include <sys/signalvar.h>
67 #include <sys/socketvar.h>
68 #include <sys/stat.h>
69 #include <sys/sx.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/sysctl.h>
72 #include <sys/sysproto.h>
73 #include <sys/tty.h>
74 #include <sys/unistd.h>
75 #include <sys/user.h>
76 #include <sys/vnode.h>
77 #ifdef KTRACE
78 #include <sys/ktrace.h>
79 #endif
80
81 #include <security/audit/audit.h>
82
83 #include <vm/uma.h>
84
85 #include <ddb/ddb.h>
86
87 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
88 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
89 "file desc to leader structures");
90 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
91
92 static uma_zone_t file_zone;
93
94
95 /* Flags for do_dup() */
96 #define DUP_FIXED 0x1 /* Force fixed allocation */
97 #define DUP_FCNTL 0x2 /* fcntl()-style errors */
98
99 static int do_dup(struct thread *td, int flags, int old, int new,
100 register_t *retval);
101 static int fd_first_free(struct filedesc *, int, int);
102 static int fd_last_used(struct filedesc *, int, int);
103 static void fdgrowtable(struct filedesc *, int);
104 static void fdunused(struct filedesc *fdp, int fd);
105 static void fdused(struct filedesc *fdp, int fd);
106
107 /*
108 * A process is initially started out with NDFILE descriptors stored within
109 * this structure, selected to be enough for typical applications based on
110 * the historical limit of 20 open files (and the usage of descriptors by
111 * shells). If these descriptors are exhausted, a larger descriptor table
112 * may be allocated, up to a process' resource limit; the internal arrays
113 * are then unused.
114 */
115 #define NDFILE 20
116 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
117 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
118 #define NDSLOT(x) ((x) / NDENTRIES)
119 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
120 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
121
122 /*
123 * Storage required per open file descriptor.
124 */
125 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
126
127 /*
128 * Storage to hold unused ofiles that need to be reclaimed.
129 */
130 struct freetable {
131 struct file **ft_table;
132 SLIST_ENTRY(freetable) ft_next;
133 };
134
135 /*
136 * Basic allocation of descriptors:
137 * one of the above, plus arrays for NDFILE descriptors.
138 */
139 struct filedesc0 {
140 struct filedesc fd_fd;
141 /*
142 * ofiles which need to be reclaimed on free.
143 */
144 SLIST_HEAD(,freetable) fd_free;
145 /*
146 * These arrays are used when the number of open files is
147 * <= NDFILE, and are then pointed to by the pointers above.
148 */
149 struct file *fd_dfiles[NDFILE];
150 char fd_dfileflags[NDFILE];
151 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
152 };
153
154 /*
155 * Descriptor management.
156 */
157 volatile int openfiles; /* actual number of open files */
158 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
159 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
160
161 /* A mutex to protect the association between a proc and filedesc. */
162 static struct mtx fdesc_mtx;
163
164 /*
165 * Find the first zero bit in the given bitmap, starting at low and not
166 * exceeding size - 1.
167 */
168 static int
169 fd_first_free(struct filedesc *fdp, int low, int size)
170 {
171 NDSLOTTYPE *map = fdp->fd_map;
172 NDSLOTTYPE mask;
173 int off, maxoff;
174
175 if (low >= size)
176 return (low);
177
178 off = NDSLOT(low);
179 if (low % NDENTRIES) {
180 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
181 if ((mask &= ~map[off]) != 0UL)
182 return (off * NDENTRIES + ffsl(mask) - 1);
183 ++off;
184 }
185 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
186 if (map[off] != ~0UL)
187 return (off * NDENTRIES + ffsl(~map[off]) - 1);
188 return (size);
189 }
190
191 /*
192 * Find the highest non-zero bit in the given bitmap, starting at low and
193 * not exceeding size - 1.
194 */
195 static int
196 fd_last_used(struct filedesc *fdp, int low, int size)
197 {
198 NDSLOTTYPE *map = fdp->fd_map;
199 NDSLOTTYPE mask;
200 int off, minoff;
201
202 if (low >= size)
203 return (-1);
204
205 off = NDSLOT(size);
206 if (size % NDENTRIES) {
207 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
208 if ((mask &= map[off]) != 0)
209 return (off * NDENTRIES + flsl(mask) - 1);
210 --off;
211 }
212 for (minoff = NDSLOT(low); off >= minoff; --off)
213 if (map[off] != 0)
214 return (off * NDENTRIES + flsl(map[off]) - 1);
215 return (low - 1);
216 }
217
218 static int
219 fdisused(struct filedesc *fdp, int fd)
220 {
221 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
222 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
223 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
224 }
225
226 /*
227 * Mark a file descriptor as used.
228 */
229 static void
230 fdused(struct filedesc *fdp, int fd)
231 {
232
233 FILEDESC_XLOCK_ASSERT(fdp);
234 KASSERT(!fdisused(fdp, fd),
235 ("fd already used"));
236
237 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
238 if (fd > fdp->fd_lastfile)
239 fdp->fd_lastfile = fd;
240 if (fd == fdp->fd_freefile)
241 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
242 }
243
244 /*
245 * Mark a file descriptor as unused.
246 */
247 static void
248 fdunused(struct filedesc *fdp, int fd)
249 {
250
251 FILEDESC_XLOCK_ASSERT(fdp);
252 KASSERT(fdisused(fdp, fd),
253 ("fd is already unused"));
254 KASSERT(fdp->fd_ofiles[fd] == NULL,
255 ("fd is still in use"));
256
257 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
258 if (fd < fdp->fd_freefile)
259 fdp->fd_freefile = fd;
260 if (fd == fdp->fd_lastfile)
261 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
262 }
263
264 /*
265 * System calls on descriptors.
266 */
267 #ifndef _SYS_SYSPROTO_H_
268 struct getdtablesize_args {
269 int dummy;
270 };
271 #endif
272 /* ARGSUSED */
273 int
274 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
275 {
276 struct proc *p = td->td_proc;
277
278 PROC_LOCK(p);
279 td->td_retval[0] =
280 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
281 PROC_UNLOCK(p);
282 return (0);
283 }
284
285 /*
286 * Duplicate a file descriptor to a particular value.
287 *
288 * Note: keep in mind that a potential race condition exists when closing
289 * descriptors from a shared descriptor table (via rfork).
290 */
291 #ifndef _SYS_SYSPROTO_H_
292 struct dup2_args {
293 u_int from;
294 u_int to;
295 };
296 #endif
297 /* ARGSUSED */
298 int
299 dup2(struct thread *td, struct dup2_args *uap)
300 {
301
302 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
303 td->td_retval));
304 }
305
306 /*
307 * Duplicate a file descriptor.
308 */
309 #ifndef _SYS_SYSPROTO_H_
310 struct dup_args {
311 u_int fd;
312 };
313 #endif
314 /* ARGSUSED */
315 int
316 dup(struct thread *td, struct dup_args *uap)
317 {
318
319 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
320 }
321
322 /*
323 * The file control system call.
324 */
325 #ifndef _SYS_SYSPROTO_H_
326 struct fcntl_args {
327 int fd;
328 int cmd;
329 long arg;
330 };
331 #endif
332 /* ARGSUSED */
333 int
334 fcntl(struct thread *td, struct fcntl_args *uap)
335 {
336 struct flock fl;
337 struct oflock ofl;
338 intptr_t arg;
339 int error;
340 int cmd;
341
342 error = 0;
343 cmd = uap->cmd;
344 switch (uap->cmd) {
345 case F_OGETLK:
346 case F_OSETLK:
347 case F_OSETLKW:
348 /*
349 * Convert old flock structure to new.
350 */
351 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
352 fl.l_start = ofl.l_start;
353 fl.l_len = ofl.l_len;
354 fl.l_pid = ofl.l_pid;
355 fl.l_type = ofl.l_type;
356 fl.l_whence = ofl.l_whence;
357 fl.l_sysid = 0;
358
359 switch (uap->cmd) {
360 case F_OGETLK:
361 cmd = F_GETLK;
362 break;
363 case F_OSETLK:
364 cmd = F_SETLK;
365 break;
366 case F_OSETLKW:
367 cmd = F_SETLKW;
368 break;
369 }
370 arg = (intptr_t)&fl;
371 break;
372 case F_GETLK:
373 case F_SETLK:
374 case F_SETLKW:
375 case F_SETLK_REMOTE:
376 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
377 arg = (intptr_t)&fl;
378 break;
379 default:
380 arg = uap->arg;
381 break;
382 }
383 if (error)
384 return (error);
385 error = kern_fcntl(td, uap->fd, cmd, arg);
386 if (error)
387 return (error);
388 if (uap->cmd == F_OGETLK) {
389 ofl.l_start = fl.l_start;
390 ofl.l_len = fl.l_len;
391 ofl.l_pid = fl.l_pid;
392 ofl.l_type = fl.l_type;
393 ofl.l_whence = fl.l_whence;
394 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
395 } else if (uap->cmd == F_GETLK) {
396 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
397 }
398 return (error);
399 }
400
401 static inline struct file *
402 fdtofp(int fd, struct filedesc *fdp)
403 {
404 struct file *fp;
405
406 FILEDESC_LOCK_ASSERT(fdp);
407 if ((unsigned)fd >= fdp->fd_nfiles ||
408 (fp = fdp->fd_ofiles[fd]) == NULL)
409 return (NULL);
410 return (fp);
411 }
412
413 int
414 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
415 {
416 struct filedesc *fdp;
417 struct flock *flp;
418 struct file *fp;
419 struct proc *p;
420 char *pop;
421 struct vnode *vp;
422 int error, flg, tmp;
423 int vfslocked;
424
425 vfslocked = 0;
426 error = 0;
427 flg = F_POSIX;
428 p = td->td_proc;
429 fdp = p->p_fd;
430
431 switch (cmd) {
432 case F_DUPFD:
433 tmp = arg;
434 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
435 break;
436
437 case F_DUP2FD:
438 tmp = arg;
439 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
440 break;
441
442 case F_GETFD:
443 FILEDESC_SLOCK(fdp);
444 if ((fp = fdtofp(fd, fdp)) == NULL) {
445 FILEDESC_SUNLOCK(fdp);
446 error = EBADF;
447 break;
448 }
449 pop = &fdp->fd_ofileflags[fd];
450 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
451 FILEDESC_SUNLOCK(fdp);
452 break;
453
454 case F_SETFD:
455 FILEDESC_XLOCK(fdp);
456 if ((fp = fdtofp(fd, fdp)) == NULL) {
457 FILEDESC_XUNLOCK(fdp);
458 error = EBADF;
459 break;
460 }
461 pop = &fdp->fd_ofileflags[fd];
462 *pop = (*pop &~ UF_EXCLOSE) |
463 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
464 FILEDESC_XUNLOCK(fdp);
465 break;
466
467 case F_GETFL:
468 FILEDESC_SLOCK(fdp);
469 if ((fp = fdtofp(fd, fdp)) == NULL) {
470 FILEDESC_SUNLOCK(fdp);
471 error = EBADF;
472 break;
473 }
474 td->td_retval[0] = OFLAGS(fp->f_flag);
475 FILEDESC_SUNLOCK(fdp);
476 break;
477
478 case F_SETFL:
479 FILEDESC_SLOCK(fdp);
480 if ((fp = fdtofp(fd, fdp)) == NULL) {
481 FILEDESC_SUNLOCK(fdp);
482 error = EBADF;
483 break;
484 }
485 fhold(fp);
486 FILEDESC_SUNLOCK(fdp);
487 do {
488 tmp = flg = fp->f_flag;
489 tmp &= ~FCNTLFLAGS;
490 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
491 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
492 tmp = fp->f_flag & FNONBLOCK;
493 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
494 if (error) {
495 fdrop(fp, td);
496 break;
497 }
498 tmp = fp->f_flag & FASYNC;
499 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
500 if (error == 0) {
501 fdrop(fp, td);
502 break;
503 }
504 atomic_clear_int(&fp->f_flag, FNONBLOCK);
505 tmp = 0;
506 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
507 fdrop(fp, td);
508 break;
509
510 case F_GETOWN:
511 FILEDESC_SLOCK(fdp);
512 if ((fp = fdtofp(fd, fdp)) == NULL) {
513 FILEDESC_SUNLOCK(fdp);
514 error = EBADF;
515 break;
516 }
517 fhold(fp);
518 FILEDESC_SUNLOCK(fdp);
519 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
520 if (error == 0)
521 td->td_retval[0] = tmp;
522 fdrop(fp, td);
523 break;
524
525 case F_SETOWN:
526 FILEDESC_SLOCK(fdp);
527 if ((fp = fdtofp(fd, fdp)) == NULL) {
528 FILEDESC_SUNLOCK(fdp);
529 error = EBADF;
530 break;
531 }
532 fhold(fp);
533 FILEDESC_SUNLOCK(fdp);
534 tmp = arg;
535 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
536 fdrop(fp, td);
537 break;
538
539 case F_SETLK_REMOTE:
540 error = priv_check(td, PRIV_NFS_LOCKD);
541 if (error)
542 return (error);
543 flg = F_REMOTE;
544 goto do_setlk;
545
546 case F_SETLKW:
547 flg |= F_WAIT;
548 /* FALLTHROUGH F_SETLK */
549
550 case F_SETLK:
551 do_setlk:
552 FILEDESC_SLOCK(fdp);
553 if ((fp = fdtofp(fd, fdp)) == NULL) {
554 FILEDESC_SUNLOCK(fdp);
555 error = EBADF;
556 break;
557 }
558 if (fp->f_type != DTYPE_VNODE) {
559 FILEDESC_SUNLOCK(fdp);
560 error = EBADF;
561 break;
562 }
563 flp = (struct flock *)arg;
564 if (flp->l_whence == SEEK_CUR) {
565 if (fp->f_offset < 0 ||
566 (flp->l_start > 0 &&
567 fp->f_offset > OFF_MAX - flp->l_start)) {
568 FILEDESC_SUNLOCK(fdp);
569 error = EOVERFLOW;
570 break;
571 }
572 flp->l_start += fp->f_offset;
573 }
574
575 /*
576 * VOP_ADVLOCK() may block.
577 */
578 fhold(fp);
579 FILEDESC_SUNLOCK(fdp);
580 vp = fp->f_vnode;
581 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
582 switch (flp->l_type) {
583 case F_RDLCK:
584 if ((fp->f_flag & FREAD) == 0) {
585 error = EBADF;
586 break;
587 }
588 PROC_LOCK(p->p_leader);
589 p->p_leader->p_flag |= P_ADVLOCK;
590 PROC_UNLOCK(p->p_leader);
591 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
592 flp, flg);
593 break;
594 case F_WRLCK:
595 if ((fp->f_flag & FWRITE) == 0) {
596 error = EBADF;
597 break;
598 }
599 PROC_LOCK(p->p_leader);
600 p->p_leader->p_flag |= P_ADVLOCK;
601 PROC_UNLOCK(p->p_leader);
602 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
603 flp, flg);
604 break;
605 case F_UNLCK:
606 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
607 flp, flg);
608 break;
609 case F_UNLCKSYS:
610 /*
611 * Temporary api for testing remote lock
612 * infrastructure.
613 */
614 if (flg != F_REMOTE) {
615 error = EINVAL;
616 break;
617 }
618 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
619 F_UNLCKSYS, flp, flg);
620 break;
621 default:
622 error = EINVAL;
623 break;
624 }
625 VFS_UNLOCK_GIANT(vfslocked);
626 vfslocked = 0;
627 /* Check for race with close */
628 FILEDESC_SLOCK(fdp);
629 if ((unsigned) fd >= fdp->fd_nfiles ||
630 fp != fdp->fd_ofiles[fd]) {
631 FILEDESC_SUNLOCK(fdp);
632 flp->l_whence = SEEK_SET;
633 flp->l_start = 0;
634 flp->l_len = 0;
635 flp->l_type = F_UNLCK;
636 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
637 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
638 F_UNLCK, flp, F_POSIX);
639 VFS_UNLOCK_GIANT(vfslocked);
640 vfslocked = 0;
641 } else
642 FILEDESC_SUNLOCK(fdp);
643 fdrop(fp, td);
644 break;
645
646 case F_GETLK:
647 FILEDESC_SLOCK(fdp);
648 if ((fp = fdtofp(fd, fdp)) == NULL) {
649 FILEDESC_SUNLOCK(fdp);
650 error = EBADF;
651 break;
652 }
653 if (fp->f_type != DTYPE_VNODE) {
654 FILEDESC_SUNLOCK(fdp);
655 error = EBADF;
656 break;
657 }
658 flp = (struct flock *)arg;
659 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
660 flp->l_type != F_UNLCK) {
661 FILEDESC_SUNLOCK(fdp);
662 error = EINVAL;
663 break;
664 }
665 if (flp->l_whence == SEEK_CUR) {
666 if ((flp->l_start > 0 &&
667 fp->f_offset > OFF_MAX - flp->l_start) ||
668 (flp->l_start < 0 &&
669 fp->f_offset < OFF_MIN - flp->l_start)) {
670 FILEDESC_SUNLOCK(fdp);
671 error = EOVERFLOW;
672 break;
673 }
674 flp->l_start += fp->f_offset;
675 }
676 /*
677 * VOP_ADVLOCK() may block.
678 */
679 fhold(fp);
680 FILEDESC_SUNLOCK(fdp);
681 vp = fp->f_vnode;
682 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
683 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
684 F_POSIX);
685 VFS_UNLOCK_GIANT(vfslocked);
686 vfslocked = 0;
687 fdrop(fp, td);
688 break;
689 default:
690 error = EINVAL;
691 break;
692 }
693 VFS_UNLOCK_GIANT(vfslocked);
694 return (error);
695 }
696
697 /*
698 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
699 */
700 static int
701 do_dup(struct thread *td, int flags, int old, int new,
702 register_t *retval)
703 {
704 struct filedesc *fdp;
705 struct proc *p;
706 struct file *fp;
707 struct file *delfp;
708 int error, holdleaders, maxfd;
709
710 p = td->td_proc;
711 fdp = p->p_fd;
712
713 /*
714 * Verify we have a valid descriptor to dup from and possibly to
715 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
716 * return EINVAL when the new descriptor is out of bounds.
717 */
718 if (old < 0)
719 return (EBADF);
720 if (new < 0)
721 return (flags & DUP_FCNTL ? EINVAL : EBADF);
722 PROC_LOCK(p);
723 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
724 PROC_UNLOCK(p);
725 if (new >= maxfd)
726 return (flags & DUP_FCNTL ? EINVAL : EMFILE);
727
728 FILEDESC_XLOCK(fdp);
729 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
730 FILEDESC_XUNLOCK(fdp);
731 return (EBADF);
732 }
733 if (flags & DUP_FIXED && old == new) {
734 *retval = new;
735 FILEDESC_XUNLOCK(fdp);
736 return (0);
737 }
738 fp = fdp->fd_ofiles[old];
739 fhold(fp);
740
741 /*
742 * If the caller specified a file descriptor, make sure the file
743 * table is large enough to hold it, and grab it. Otherwise, just
744 * allocate a new descriptor the usual way. Since the filedesc
745 * lock may be temporarily dropped in the process, we have to look
746 * out for a race.
747 */
748 if (flags & DUP_FIXED) {
749 if (new >= fdp->fd_nfiles)
750 fdgrowtable(fdp, new + 1);
751 if (fdp->fd_ofiles[new] == NULL)
752 fdused(fdp, new);
753 } else {
754 if ((error = fdalloc(td, new, &new)) != 0) {
755 FILEDESC_XUNLOCK(fdp);
756 fdrop(fp, td);
757 return (error);
758 }
759 }
760
761 /*
762 * If the old file changed out from under us then treat it as a
763 * bad file descriptor. Userland should do its own locking to
764 * avoid this case.
765 */
766 if (fdp->fd_ofiles[old] != fp) {
767 /* we've allocated a descriptor which we won't use */
768 if (fdp->fd_ofiles[new] == NULL)
769 fdunused(fdp, new);
770 FILEDESC_XUNLOCK(fdp);
771 fdrop(fp, td);
772 return (EBADF);
773 }
774 KASSERT(old != new,
775 ("new fd is same as old"));
776
777 /*
778 * Save info on the descriptor being overwritten. We cannot close
779 * it without introducing an ownership race for the slot, since we
780 * need to drop the filedesc lock to call closef().
781 *
782 * XXX this duplicates parts of close().
783 */
784 delfp = fdp->fd_ofiles[new];
785 holdleaders = 0;
786 if (delfp != NULL) {
787 if (td->td_proc->p_fdtol != NULL) {
788 /*
789 * Ask fdfree() to sleep to ensure that all relevant
790 * process leaders can be traversed in closef().
791 */
792 fdp->fd_holdleaderscount++;
793 holdleaders = 1;
794 }
795 }
796
797 /*
798 * Duplicate the source descriptor
799 */
800 fdp->fd_ofiles[new] = fp;
801 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
802 if (new > fdp->fd_lastfile)
803 fdp->fd_lastfile = new;
804 *retval = new;
805
806 /*
807 * If we dup'd over a valid file, we now own the reference to it
808 * and must dispose of it using closef() semantics (as if a
809 * close() were performed on it).
810 *
811 * XXX this duplicates parts of close().
812 */
813 if (delfp != NULL) {
814 knote_fdclose(td, new);
815 if (delfp->f_type == DTYPE_MQUEUE)
816 mq_fdclose(td, new, delfp);
817 FILEDESC_XUNLOCK(fdp);
818 (void) closef(delfp, td);
819 if (holdleaders) {
820 FILEDESC_XLOCK(fdp);
821 fdp->fd_holdleaderscount--;
822 if (fdp->fd_holdleaderscount == 0 &&
823 fdp->fd_holdleaderswakeup != 0) {
824 fdp->fd_holdleaderswakeup = 0;
825 wakeup(&fdp->fd_holdleaderscount);
826 }
827 FILEDESC_XUNLOCK(fdp);
828 }
829 } else {
830 FILEDESC_XUNLOCK(fdp);
831 }
832 return (0);
833 }
834
835 /*
836 * If sigio is on the list associated with a process or process group,
837 * disable signalling from the device, remove sigio from the list and
838 * free sigio.
839 */
840 void
841 funsetown(struct sigio **sigiop)
842 {
843 struct sigio *sigio;
844
845 SIGIO_LOCK();
846 sigio = *sigiop;
847 if (sigio == NULL) {
848 SIGIO_UNLOCK();
849 return;
850 }
851 *(sigio->sio_myref) = NULL;
852 if ((sigio)->sio_pgid < 0) {
853 struct pgrp *pg = (sigio)->sio_pgrp;
854 PGRP_LOCK(pg);
855 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
856 sigio, sio_pgsigio);
857 PGRP_UNLOCK(pg);
858 } else {
859 struct proc *p = (sigio)->sio_proc;
860 PROC_LOCK(p);
861 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
862 sigio, sio_pgsigio);
863 PROC_UNLOCK(p);
864 }
865 SIGIO_UNLOCK();
866 crfree(sigio->sio_ucred);
867 free(sigio, M_SIGIO);
868 }
869
870 /*
871 * Free a list of sigio structures.
872 * We only need to lock the SIGIO_LOCK because we have made ourselves
873 * inaccessible to callers of fsetown and therefore do not need to lock
874 * the proc or pgrp struct for the list manipulation.
875 */
876 void
877 funsetownlst(struct sigiolst *sigiolst)
878 {
879 struct proc *p;
880 struct pgrp *pg;
881 struct sigio *sigio;
882
883 sigio = SLIST_FIRST(sigiolst);
884 if (sigio == NULL)
885 return;
886 p = NULL;
887 pg = NULL;
888
889 /*
890 * Every entry of the list should belong
891 * to a single proc or pgrp.
892 */
893 if (sigio->sio_pgid < 0) {
894 pg = sigio->sio_pgrp;
895 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
896 } else /* if (sigio->sio_pgid > 0) */ {
897 p = sigio->sio_proc;
898 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
899 }
900
901 SIGIO_LOCK();
902 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
903 *(sigio->sio_myref) = NULL;
904 if (pg != NULL) {
905 KASSERT(sigio->sio_pgid < 0,
906 ("Proc sigio in pgrp sigio list"));
907 KASSERT(sigio->sio_pgrp == pg,
908 ("Bogus pgrp in sigio list"));
909 PGRP_LOCK(pg);
910 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
911 sio_pgsigio);
912 PGRP_UNLOCK(pg);
913 } else /* if (p != NULL) */ {
914 KASSERT(sigio->sio_pgid > 0,
915 ("Pgrp sigio in proc sigio list"));
916 KASSERT(sigio->sio_proc == p,
917 ("Bogus proc in sigio list"));
918 PROC_LOCK(p);
919 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
920 sio_pgsigio);
921 PROC_UNLOCK(p);
922 }
923 SIGIO_UNLOCK();
924 crfree(sigio->sio_ucred);
925 free(sigio, M_SIGIO);
926 SIGIO_LOCK();
927 }
928 SIGIO_UNLOCK();
929 }
930
931 /*
932 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
933 *
934 * After permission checking, add a sigio structure to the sigio list for
935 * the process or process group.
936 */
937 int
938 fsetown(pid_t pgid, struct sigio **sigiop)
939 {
940 struct proc *proc;
941 struct pgrp *pgrp;
942 struct sigio *sigio;
943 int ret;
944
945 if (pgid == 0) {
946 funsetown(sigiop);
947 return (0);
948 }
949
950 ret = 0;
951
952 /* Allocate and fill in the new sigio out of locks. */
953 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
954 sigio->sio_pgid = pgid;
955 sigio->sio_ucred = crhold(curthread->td_ucred);
956 sigio->sio_myref = sigiop;
957
958 sx_slock(&proctree_lock);
959 if (pgid > 0) {
960 proc = pfind(pgid);
961 if (proc == NULL) {
962 ret = ESRCH;
963 goto fail;
964 }
965
966 /*
967 * Policy - Don't allow a process to FSETOWN a process
968 * in another session.
969 *
970 * Remove this test to allow maximum flexibility or
971 * restrict FSETOWN to the current process or process
972 * group for maximum safety.
973 */
974 PROC_UNLOCK(proc);
975 if (proc->p_session != curthread->td_proc->p_session) {
976 ret = EPERM;
977 goto fail;
978 }
979
980 pgrp = NULL;
981 } else /* if (pgid < 0) */ {
982 pgrp = pgfind(-pgid);
983 if (pgrp == NULL) {
984 ret = ESRCH;
985 goto fail;
986 }
987 PGRP_UNLOCK(pgrp);
988
989 /*
990 * Policy - Don't allow a process to FSETOWN a process
991 * in another session.
992 *
993 * Remove this test to allow maximum flexibility or
994 * restrict FSETOWN to the current process or process
995 * group for maximum safety.
996 */
997 if (pgrp->pg_session != curthread->td_proc->p_session) {
998 ret = EPERM;
999 goto fail;
1000 }
1001
1002 proc = NULL;
1003 }
1004 funsetown(sigiop);
1005 if (pgid > 0) {
1006 PROC_LOCK(proc);
1007 /*
1008 * Since funsetownlst() is called without the proctree
1009 * locked, we need to check for P_WEXIT.
1010 * XXX: is ESRCH correct?
1011 */
1012 if ((proc->p_flag & P_WEXIT) != 0) {
1013 PROC_UNLOCK(proc);
1014 ret = ESRCH;
1015 goto fail;
1016 }
1017 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1018 sigio->sio_proc = proc;
1019 PROC_UNLOCK(proc);
1020 } else {
1021 PGRP_LOCK(pgrp);
1022 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1023 sigio->sio_pgrp = pgrp;
1024 PGRP_UNLOCK(pgrp);
1025 }
1026 sx_sunlock(&proctree_lock);
1027 SIGIO_LOCK();
1028 *sigiop = sigio;
1029 SIGIO_UNLOCK();
1030 return (0);
1031
1032 fail:
1033 sx_sunlock(&proctree_lock);
1034 crfree(sigio->sio_ucred);
1035 free(sigio, M_SIGIO);
1036 return (ret);
1037 }
1038
1039 /*
1040 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1041 */
1042 pid_t
1043 fgetown(sigiop)
1044 struct sigio **sigiop;
1045 {
1046 pid_t pgid;
1047
1048 SIGIO_LOCK();
1049 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1050 SIGIO_UNLOCK();
1051 return (pgid);
1052 }
1053
1054 /*
1055 * Close a file descriptor.
1056 */
1057 #ifndef _SYS_SYSPROTO_H_
1058 struct close_args {
1059 int fd;
1060 };
1061 #endif
1062 /* ARGSUSED */
1063 int
1064 close(td, uap)
1065 struct thread *td;
1066 struct close_args *uap;
1067 {
1068
1069 return (kern_close(td, uap->fd));
1070 }
1071
1072 int
1073 kern_close(td, fd)
1074 struct thread *td;
1075 int fd;
1076 {
1077 struct filedesc *fdp;
1078 struct file *fp;
1079 int error;
1080 int holdleaders;
1081
1082 error = 0;
1083 holdleaders = 0;
1084 fdp = td->td_proc->p_fd;
1085
1086 AUDIT_SYSCLOSE(td, fd);
1087
1088 FILEDESC_XLOCK(fdp);
1089 if ((unsigned)fd >= fdp->fd_nfiles ||
1090 (fp = fdp->fd_ofiles[fd]) == NULL) {
1091 FILEDESC_XUNLOCK(fdp);
1092 return (EBADF);
1093 }
1094 fdp->fd_ofiles[fd] = NULL;
1095 fdp->fd_ofileflags[fd] = 0;
1096 fdunused(fdp, fd);
1097 if (td->td_proc->p_fdtol != NULL) {
1098 /*
1099 * Ask fdfree() to sleep to ensure that all relevant
1100 * process leaders can be traversed in closef().
1101 */
1102 fdp->fd_holdleaderscount++;
1103 holdleaders = 1;
1104 }
1105
1106 /*
1107 * We now hold the fp reference that used to be owned by the
1108 * descriptor array. We have to unlock the FILEDESC *AFTER*
1109 * knote_fdclose to prevent a race of the fd getting opened, a knote
1110 * added, and deleteing a knote for the new fd.
1111 */
1112 knote_fdclose(td, fd);
1113 if (fp->f_type == DTYPE_MQUEUE)
1114 mq_fdclose(td, fd, fp);
1115 FILEDESC_XUNLOCK(fdp);
1116
1117 error = closef(fp, td);
1118 if (holdleaders) {
1119 FILEDESC_XLOCK(fdp);
1120 fdp->fd_holdleaderscount--;
1121 if (fdp->fd_holdleaderscount == 0 &&
1122 fdp->fd_holdleaderswakeup != 0) {
1123 fdp->fd_holdleaderswakeup = 0;
1124 wakeup(&fdp->fd_holdleaderscount);
1125 }
1126 FILEDESC_XUNLOCK(fdp);
1127 }
1128 return (error);
1129 }
1130
1131 /*
1132 * Close open file descriptors.
1133 */
1134 #ifndef _SYS_SYSPROTO_H_
1135 struct closefrom_args {
1136 int lowfd;
1137 };
1138 #endif
1139 /* ARGSUSED */
1140 int
1141 closefrom(struct thread *td, struct closefrom_args *uap)
1142 {
1143 struct filedesc *fdp;
1144 int fd;
1145
1146 fdp = td->td_proc->p_fd;
1147 AUDIT_ARG_FD(uap->lowfd);
1148
1149 /*
1150 * Treat negative starting file descriptor values identical to
1151 * closefrom(0) which closes all files.
1152 */
1153 if (uap->lowfd < 0)
1154 uap->lowfd = 0;
1155 FILEDESC_SLOCK(fdp);
1156 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
1157 if (fdp->fd_ofiles[fd] != NULL) {
1158 FILEDESC_SUNLOCK(fdp);
1159 (void)kern_close(td, fd);
1160 FILEDESC_SLOCK(fdp);
1161 }
1162 }
1163 FILEDESC_SUNLOCK(fdp);
1164 return (0);
1165 }
1166
1167 #if defined(COMPAT_43)
1168 /*
1169 * Return status information about a file descriptor.
1170 */
1171 #ifndef _SYS_SYSPROTO_H_
1172 struct ofstat_args {
1173 int fd;
1174 struct ostat *sb;
1175 };
1176 #endif
1177 /* ARGSUSED */
1178 int
1179 ofstat(struct thread *td, struct ofstat_args *uap)
1180 {
1181 struct ostat oub;
1182 struct stat ub;
1183 int error;
1184
1185 error = kern_fstat(td, uap->fd, &ub);
1186 if (error == 0) {
1187 cvtstat(&ub, &oub);
1188 error = copyout(&oub, uap->sb, sizeof(oub));
1189 }
1190 return (error);
1191 }
1192 #endif /* COMPAT_43 */
1193
1194 /*
1195 * Return status information about a file descriptor.
1196 */
1197 #ifndef _SYS_SYSPROTO_H_
1198 struct fstat_args {
1199 int fd;
1200 struct stat *sb;
1201 };
1202 #endif
1203 /* ARGSUSED */
1204 int
1205 fstat(struct thread *td, struct fstat_args *uap)
1206 {
1207 struct stat ub;
1208 int error;
1209
1210 error = kern_fstat(td, uap->fd, &ub);
1211 if (error == 0)
1212 error = copyout(&ub, uap->sb, sizeof(ub));
1213 return (error);
1214 }
1215
1216 int
1217 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1218 {
1219 struct file *fp;
1220 int error;
1221
1222 AUDIT_ARG_FD(fd);
1223
1224 if ((error = fget(td, fd, &fp)) != 0)
1225 return (error);
1226
1227 AUDIT_ARG_FILE(td->td_proc, fp);
1228
1229 error = fo_stat(fp, sbp, td->td_ucred, td);
1230 fdrop(fp, td);
1231 #ifdef KTRACE
1232 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1233 ktrstat(sbp);
1234 #endif
1235 return (error);
1236 }
1237
1238 /*
1239 * Return status information about a file descriptor.
1240 */
1241 #ifndef _SYS_SYSPROTO_H_
1242 struct nfstat_args {
1243 int fd;
1244 struct nstat *sb;
1245 };
1246 #endif
1247 /* ARGSUSED */
1248 int
1249 nfstat(struct thread *td, struct nfstat_args *uap)
1250 {
1251 struct nstat nub;
1252 struct stat ub;
1253 int error;
1254
1255 error = kern_fstat(td, uap->fd, &ub);
1256 if (error == 0) {
1257 cvtnstat(&ub, &nub);
1258 error = copyout(&nub, uap->sb, sizeof(nub));
1259 }
1260 return (error);
1261 }
1262
1263 /*
1264 * Return pathconf information about a file descriptor.
1265 */
1266 #ifndef _SYS_SYSPROTO_H_
1267 struct fpathconf_args {
1268 int fd;
1269 int name;
1270 };
1271 #endif
1272 /* ARGSUSED */
1273 int
1274 fpathconf(struct thread *td, struct fpathconf_args *uap)
1275 {
1276 struct file *fp;
1277 struct vnode *vp;
1278 int error;
1279
1280 if ((error = fget(td, uap->fd, &fp)) != 0)
1281 return (error);
1282
1283 /* If asynchronous I/O is available, it works for all descriptors. */
1284 if (uap->name == _PC_ASYNC_IO) {
1285 td->td_retval[0] = async_io_version;
1286 goto out;
1287 }
1288 vp = fp->f_vnode;
1289 if (vp != NULL) {
1290 int vfslocked;
1291 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1292 vn_lock(vp, LK_SHARED | LK_RETRY);
1293 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1294 VOP_UNLOCK(vp, 0);
1295 VFS_UNLOCK_GIANT(vfslocked);
1296 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1297 if (uap->name != _PC_PIPE_BUF) {
1298 error = EINVAL;
1299 } else {
1300 td->td_retval[0] = PIPE_BUF;
1301 error = 0;
1302 }
1303 } else {
1304 error = EOPNOTSUPP;
1305 }
1306 out:
1307 fdrop(fp, td);
1308 return (error);
1309 }
1310
1311 /*
1312 * Grow the file table to accomodate (at least) nfd descriptors. This may
1313 * block and drop the filedesc lock, but it will reacquire it before
1314 * returning.
1315 */
1316 static void
1317 fdgrowtable(struct filedesc *fdp, int nfd)
1318 {
1319 struct filedesc0 *fdp0;
1320 struct freetable *fo;
1321 struct file **ntable;
1322 struct file **otable;
1323 char *nfileflags;
1324 int nnfiles, onfiles;
1325 NDSLOTTYPE *nmap;
1326
1327 FILEDESC_XLOCK_ASSERT(fdp);
1328
1329 KASSERT(fdp->fd_nfiles > 0,
1330 ("zero-length file table"));
1331
1332 /* compute the size of the new table */
1333 onfiles = fdp->fd_nfiles;
1334 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1335 if (nnfiles <= onfiles)
1336 /* the table is already large enough */
1337 return;
1338
1339 /* allocate a new table and (if required) new bitmaps */
1340 FILEDESC_XUNLOCK(fdp);
1341 ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable),
1342 M_FILEDESC, M_ZERO | M_WAITOK);
1343 nfileflags = (char *)&ntable[nnfiles];
1344 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1345 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE,
1346 M_FILEDESC, M_ZERO | M_WAITOK);
1347 else
1348 nmap = NULL;
1349 FILEDESC_XLOCK(fdp);
1350
1351 /*
1352 * We now have new tables ready to go. Since we dropped the
1353 * filedesc lock to call malloc(), watch out for a race.
1354 */
1355 onfiles = fdp->fd_nfiles;
1356 if (onfiles >= nnfiles) {
1357 /* we lost the race, but that's OK */
1358 free(ntable, M_FILEDESC);
1359 if (nmap != NULL)
1360 free(nmap, M_FILEDESC);
1361 return;
1362 }
1363 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1364 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1365 otable = fdp->fd_ofiles;
1366 fdp->fd_ofileflags = nfileflags;
1367 fdp->fd_ofiles = ntable;
1368 /*
1369 * We must preserve ofiles until the process exits because we can't
1370 * be certain that no threads have references to the old table via
1371 * _fget().
1372 */
1373 if (onfiles > NDFILE) {
1374 fo = (struct freetable *)&otable[onfiles];
1375 fdp0 = (struct filedesc0 *)fdp;
1376 fo->ft_table = otable;
1377 SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next);
1378 }
1379 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1380 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1381 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1382 free(fdp->fd_map, M_FILEDESC);
1383 fdp->fd_map = nmap;
1384 }
1385 fdp->fd_nfiles = nnfiles;
1386 }
1387
1388 /*
1389 * Allocate a file descriptor for the process.
1390 */
1391 int
1392 fdalloc(struct thread *td, int minfd, int *result)
1393 {
1394 struct proc *p = td->td_proc;
1395 struct filedesc *fdp = p->p_fd;
1396 int fd = -1, maxfd;
1397
1398 FILEDESC_XLOCK_ASSERT(fdp);
1399
1400 if (fdp->fd_freefile > minfd)
1401 minfd = fdp->fd_freefile;
1402
1403 PROC_LOCK(p);
1404 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1405 PROC_UNLOCK(p);
1406
1407 /*
1408 * Search the bitmap for a free descriptor. If none is found, try
1409 * to grow the file table. Keep at it until we either get a file
1410 * descriptor or run into process or system limits; fdgrowtable()
1411 * may drop the filedesc lock, so we're in a race.
1412 */
1413 for (;;) {
1414 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1415 if (fd >= maxfd)
1416 return (EMFILE);
1417 if (fd < fdp->fd_nfiles)
1418 break;
1419 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1420 }
1421
1422 /*
1423 * Perform some sanity checks, then mark the file descriptor as
1424 * used and return it to the caller.
1425 */
1426 KASSERT(!fdisused(fdp, fd),
1427 ("fd_first_free() returned non-free descriptor"));
1428 KASSERT(fdp->fd_ofiles[fd] == NULL,
1429 ("free descriptor isn't"));
1430 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1431 fdused(fdp, fd);
1432 *result = fd;
1433 return (0);
1434 }
1435
1436 /*
1437 * Check to see whether n user file descriptors are available to the process
1438 * p.
1439 */
1440 int
1441 fdavail(struct thread *td, int n)
1442 {
1443 struct proc *p = td->td_proc;
1444 struct filedesc *fdp = td->td_proc->p_fd;
1445 struct file **fpp;
1446 int i, lim, last;
1447
1448 FILEDESC_LOCK_ASSERT(fdp);
1449
1450 PROC_LOCK(p);
1451 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1452 PROC_UNLOCK(p);
1453 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1454 return (1);
1455 last = min(fdp->fd_nfiles, lim);
1456 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1457 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1458 if (*fpp == NULL && --n <= 0)
1459 return (1);
1460 }
1461 return (0);
1462 }
1463
1464 /*
1465 * Create a new open file structure and allocate a file decriptor for the
1466 * process that refers to it. We add one reference to the file for the
1467 * descriptor table and one reference for resultfp. This is to prevent us
1468 * being preempted and the entry in the descriptor table closed after we
1469 * release the FILEDESC lock.
1470 */
1471 int
1472 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1473 {
1474 struct proc *p = td->td_proc;
1475 struct file *fp;
1476 int error, i;
1477 int maxuserfiles = maxfiles - (maxfiles / 20);
1478 static struct timeval lastfail;
1479 static int curfail;
1480
1481 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1482 if ((openfiles >= maxuserfiles &&
1483 priv_check(td, PRIV_MAXFILES) != 0) ||
1484 openfiles >= maxfiles) {
1485 if (ppsratecheck(&lastfail, &curfail, 1)) {
1486 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1487 td->td_ucred->cr_ruid);
1488 }
1489 uma_zfree(file_zone, fp);
1490 return (ENFILE);
1491 }
1492 atomic_add_int(&openfiles, 1);
1493
1494 /*
1495 * If the process has file descriptor zero open, add the new file
1496 * descriptor to the list of open files at that point, otherwise
1497 * put it at the front of the list of open files.
1498 */
1499 refcount_init(&fp->f_count, 1);
1500 if (resultfp)
1501 fhold(fp);
1502 fp->f_cred = crhold(td->td_ucred);
1503 fp->f_ops = &badfileops;
1504 fp->f_data = NULL;
1505 fp->f_vnode = NULL;
1506 FILEDESC_XLOCK(p->p_fd);
1507 if ((error = fdalloc(td, 0, &i))) {
1508 FILEDESC_XUNLOCK(p->p_fd);
1509 fdrop(fp, td);
1510 if (resultfp)
1511 fdrop(fp, td);
1512 return (error);
1513 }
1514 p->p_fd->fd_ofiles[i] = fp;
1515 FILEDESC_XUNLOCK(p->p_fd);
1516 if (resultfp)
1517 *resultfp = fp;
1518 if (resultfd)
1519 *resultfd = i;
1520 return (0);
1521 }
1522
1523 /*
1524 * Build a new filedesc structure from another.
1525 * Copy the current, root, and jail root vnode references.
1526 */
1527 struct filedesc *
1528 fdinit(struct filedesc *fdp)
1529 {
1530 struct filedesc0 *newfdp;
1531
1532 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1533 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1534 if (fdp != NULL) {
1535 FILEDESC_XLOCK(fdp);
1536 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1537 if (newfdp->fd_fd.fd_cdir)
1538 VREF(newfdp->fd_fd.fd_cdir);
1539 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1540 if (newfdp->fd_fd.fd_rdir)
1541 VREF(newfdp->fd_fd.fd_rdir);
1542 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1543 if (newfdp->fd_fd.fd_jdir)
1544 VREF(newfdp->fd_fd.fd_jdir);
1545 FILEDESC_XUNLOCK(fdp);
1546 }
1547
1548 /* Create the file descriptor table. */
1549 newfdp->fd_fd.fd_refcnt = 1;
1550 newfdp->fd_fd.fd_holdcnt = 1;
1551 newfdp->fd_fd.fd_cmask = CMASK;
1552 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1553 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1554 newfdp->fd_fd.fd_nfiles = NDFILE;
1555 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1556 newfdp->fd_fd.fd_lastfile = -1;
1557 return (&newfdp->fd_fd);
1558 }
1559
1560 static struct filedesc *
1561 fdhold(struct proc *p)
1562 {
1563 struct filedesc *fdp;
1564
1565 mtx_lock(&fdesc_mtx);
1566 fdp = p->p_fd;
1567 if (fdp != NULL)
1568 fdp->fd_holdcnt++;
1569 mtx_unlock(&fdesc_mtx);
1570 return (fdp);
1571 }
1572
1573 static void
1574 fddrop(struct filedesc *fdp)
1575 {
1576 struct filedesc0 *fdp0;
1577 struct freetable *ft;
1578 int i;
1579
1580 mtx_lock(&fdesc_mtx);
1581 i = --fdp->fd_holdcnt;
1582 mtx_unlock(&fdesc_mtx);
1583 if (i > 0)
1584 return;
1585
1586 FILEDESC_LOCK_DESTROY(fdp);
1587 fdp0 = (struct filedesc0 *)fdp;
1588 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1589 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1590 free(ft->ft_table, M_FILEDESC);
1591 }
1592 free(fdp, M_FILEDESC);
1593 }
1594
1595 /*
1596 * Share a filedesc structure.
1597 */
1598 struct filedesc *
1599 fdshare(struct filedesc *fdp)
1600 {
1601
1602 FILEDESC_XLOCK(fdp);
1603 fdp->fd_refcnt++;
1604 FILEDESC_XUNLOCK(fdp);
1605 return (fdp);
1606 }
1607
1608 /*
1609 * Unshare a filedesc structure, if necessary by making a copy
1610 */
1611 void
1612 fdunshare(struct proc *p, struct thread *td)
1613 {
1614
1615 FILEDESC_XLOCK(p->p_fd);
1616 if (p->p_fd->fd_refcnt > 1) {
1617 struct filedesc *tmp;
1618
1619 FILEDESC_XUNLOCK(p->p_fd);
1620 tmp = fdcopy(p->p_fd);
1621 fdfree(td);
1622 p->p_fd = tmp;
1623 } else
1624 FILEDESC_XUNLOCK(p->p_fd);
1625 }
1626
1627 /*
1628 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1629 * this is to ease callers, not catch errors.
1630 */
1631 struct filedesc *
1632 fdcopy(struct filedesc *fdp)
1633 {
1634 struct filedesc *newfdp;
1635 int i;
1636
1637 /* Certain daemons might not have file descriptors. */
1638 if (fdp == NULL)
1639 return (NULL);
1640
1641 newfdp = fdinit(fdp);
1642 FILEDESC_SLOCK(fdp);
1643 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1644 FILEDESC_SUNLOCK(fdp);
1645 FILEDESC_XLOCK(newfdp);
1646 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1647 FILEDESC_XUNLOCK(newfdp);
1648 FILEDESC_SLOCK(fdp);
1649 }
1650 /* copy everything except kqueue descriptors */
1651 newfdp->fd_freefile = -1;
1652 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1653 if (fdisused(fdp, i) &&
1654 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE &&
1655 fdp->fd_ofiles[i]->f_ops != &badfileops) {
1656 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1657 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1658 fhold(newfdp->fd_ofiles[i]);
1659 newfdp->fd_lastfile = i;
1660 } else {
1661 if (newfdp->fd_freefile == -1)
1662 newfdp->fd_freefile = i;
1663 }
1664 }
1665 newfdp->fd_cmask = fdp->fd_cmask;
1666 FILEDESC_SUNLOCK(fdp);
1667 FILEDESC_XLOCK(newfdp);
1668 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1669 if (newfdp->fd_ofiles[i] != NULL)
1670 fdused(newfdp, i);
1671 if (newfdp->fd_freefile == -1)
1672 newfdp->fd_freefile = i;
1673 FILEDESC_XUNLOCK(newfdp);
1674 return (newfdp);
1675 }
1676
1677 /*
1678 * Release a filedesc structure.
1679 */
1680 void
1681 fdfree(struct thread *td)
1682 {
1683 struct filedesc *fdp;
1684 struct file **fpp;
1685 int i, locked;
1686 struct filedesc_to_leader *fdtol;
1687 struct file *fp;
1688 struct vnode *cdir, *jdir, *rdir, *vp;
1689 struct flock lf;
1690
1691 /* Certain daemons might not have file descriptors. */
1692 fdp = td->td_proc->p_fd;
1693 if (fdp == NULL)
1694 return;
1695
1696 /* Check for special need to clear POSIX style locks */
1697 fdtol = td->td_proc->p_fdtol;
1698 if (fdtol != NULL) {
1699 FILEDESC_XLOCK(fdp);
1700 KASSERT(fdtol->fdl_refcount > 0,
1701 ("filedesc_to_refcount botch: fdl_refcount=%d",
1702 fdtol->fdl_refcount));
1703 if (fdtol->fdl_refcount == 1 &&
1704 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1705 for (i = 0, fpp = fdp->fd_ofiles;
1706 i <= fdp->fd_lastfile;
1707 i++, fpp++) {
1708 if (*fpp == NULL ||
1709 (*fpp)->f_type != DTYPE_VNODE)
1710 continue;
1711 fp = *fpp;
1712 fhold(fp);
1713 FILEDESC_XUNLOCK(fdp);
1714 lf.l_whence = SEEK_SET;
1715 lf.l_start = 0;
1716 lf.l_len = 0;
1717 lf.l_type = F_UNLCK;
1718 vp = fp->f_vnode;
1719 locked = VFS_LOCK_GIANT(vp->v_mount);
1720 (void) VOP_ADVLOCK(vp,
1721 (caddr_t)td->td_proc->
1722 p_leader,
1723 F_UNLCK,
1724 &lf,
1725 F_POSIX);
1726 VFS_UNLOCK_GIANT(locked);
1727 FILEDESC_XLOCK(fdp);
1728 fdrop(fp, td);
1729 fpp = fdp->fd_ofiles + i;
1730 }
1731 }
1732 retry:
1733 if (fdtol->fdl_refcount == 1) {
1734 if (fdp->fd_holdleaderscount > 0 &&
1735 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1736 /*
1737 * close() or do_dup() has cleared a reference
1738 * in a shared file descriptor table.
1739 */
1740 fdp->fd_holdleaderswakeup = 1;
1741 sx_sleep(&fdp->fd_holdleaderscount,
1742 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1743 goto retry;
1744 }
1745 if (fdtol->fdl_holdcount > 0) {
1746 /*
1747 * Ensure that fdtol->fdl_leader remains
1748 * valid in closef().
1749 */
1750 fdtol->fdl_wakeup = 1;
1751 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1752 "fdlhold", 0);
1753 goto retry;
1754 }
1755 }
1756 fdtol->fdl_refcount--;
1757 if (fdtol->fdl_refcount == 0 &&
1758 fdtol->fdl_holdcount == 0) {
1759 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1760 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1761 } else
1762 fdtol = NULL;
1763 td->td_proc->p_fdtol = NULL;
1764 FILEDESC_XUNLOCK(fdp);
1765 if (fdtol != NULL)
1766 free(fdtol, M_FILEDESC_TO_LEADER);
1767 }
1768 FILEDESC_XLOCK(fdp);
1769 i = --fdp->fd_refcnt;
1770 FILEDESC_XUNLOCK(fdp);
1771 if (i > 0)
1772 return;
1773
1774 fpp = fdp->fd_ofiles;
1775 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1776 if (*fpp) {
1777 FILEDESC_XLOCK(fdp);
1778 fp = *fpp;
1779 *fpp = NULL;
1780 FILEDESC_XUNLOCK(fdp);
1781 (void) closef(fp, td);
1782 }
1783 }
1784 FILEDESC_XLOCK(fdp);
1785
1786 /* XXX This should happen earlier. */
1787 mtx_lock(&fdesc_mtx);
1788 td->td_proc->p_fd = NULL;
1789 mtx_unlock(&fdesc_mtx);
1790
1791 if (fdp->fd_nfiles > NDFILE)
1792 free(fdp->fd_ofiles, M_FILEDESC);
1793 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1794 free(fdp->fd_map, M_FILEDESC);
1795
1796 fdp->fd_nfiles = 0;
1797
1798 cdir = fdp->fd_cdir;
1799 fdp->fd_cdir = NULL;
1800 rdir = fdp->fd_rdir;
1801 fdp->fd_rdir = NULL;
1802 jdir = fdp->fd_jdir;
1803 fdp->fd_jdir = NULL;
1804 FILEDESC_XUNLOCK(fdp);
1805
1806 if (cdir) {
1807 locked = VFS_LOCK_GIANT(cdir->v_mount);
1808 vrele(cdir);
1809 VFS_UNLOCK_GIANT(locked);
1810 }
1811 if (rdir) {
1812 locked = VFS_LOCK_GIANT(rdir->v_mount);
1813 vrele(rdir);
1814 VFS_UNLOCK_GIANT(locked);
1815 }
1816 if (jdir) {
1817 locked = VFS_LOCK_GIANT(jdir->v_mount);
1818 vrele(jdir);
1819 VFS_UNLOCK_GIANT(locked);
1820 }
1821
1822 fddrop(fdp);
1823 }
1824
1825 /*
1826 * For setugid programs, we don't want to people to use that setugidness
1827 * to generate error messages which write to a file which otherwise would
1828 * otherwise be off-limits to the process. We check for filesystems where
1829 * the vnode can change out from under us after execve (like [lin]procfs).
1830 *
1831 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1832 * sufficient. We also don't check for setugidness since we know we are.
1833 */
1834 static int
1835 is_unsafe(struct file *fp)
1836 {
1837 if (fp->f_type == DTYPE_VNODE) {
1838 struct vnode *vp = fp->f_vnode;
1839
1840 if ((vp->v_vflag & VV_PROCDEP) != 0)
1841 return (1);
1842 }
1843 return (0);
1844 }
1845
1846 /*
1847 * Make this setguid thing safe, if at all possible.
1848 */
1849 void
1850 setugidsafety(struct thread *td)
1851 {
1852 struct filedesc *fdp;
1853 int i;
1854
1855 /* Certain daemons might not have file descriptors. */
1856 fdp = td->td_proc->p_fd;
1857 if (fdp == NULL)
1858 return;
1859
1860 /*
1861 * Note: fdp->fd_ofiles may be reallocated out from under us while
1862 * we are blocked in a close. Be careful!
1863 */
1864 FILEDESC_XLOCK(fdp);
1865 for (i = 0; i <= fdp->fd_lastfile; i++) {
1866 if (i > 2)
1867 break;
1868 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1869 struct file *fp;
1870
1871 knote_fdclose(td, i);
1872 /*
1873 * NULL-out descriptor prior to close to avoid
1874 * a race while close blocks.
1875 */
1876 fp = fdp->fd_ofiles[i];
1877 fdp->fd_ofiles[i] = NULL;
1878 fdp->fd_ofileflags[i] = 0;
1879 fdunused(fdp, i);
1880 FILEDESC_XUNLOCK(fdp);
1881 (void) closef(fp, td);
1882 FILEDESC_XLOCK(fdp);
1883 }
1884 }
1885 FILEDESC_XUNLOCK(fdp);
1886 }
1887
1888 /*
1889 * If a specific file object occupies a specific file descriptor, close the
1890 * file descriptor entry and drop a reference on the file object. This is a
1891 * convenience function to handle a subsequent error in a function that calls
1892 * falloc() that handles the race that another thread might have closed the
1893 * file descriptor out from under the thread creating the file object.
1894 */
1895 void
1896 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1897 {
1898
1899 FILEDESC_XLOCK(fdp);
1900 if (fdp->fd_ofiles[idx] == fp) {
1901 fdp->fd_ofiles[idx] = NULL;
1902 fdunused(fdp, idx);
1903 FILEDESC_XUNLOCK(fdp);
1904 fdrop(fp, td);
1905 } else
1906 FILEDESC_XUNLOCK(fdp);
1907 }
1908
1909 /*
1910 * Close any files on exec?
1911 */
1912 void
1913 fdcloseexec(struct thread *td)
1914 {
1915 struct filedesc *fdp;
1916 int i;
1917
1918 /* Certain daemons might not have file descriptors. */
1919 fdp = td->td_proc->p_fd;
1920 if (fdp == NULL)
1921 return;
1922
1923 FILEDESC_XLOCK(fdp);
1924
1925 /*
1926 * We cannot cache fd_ofiles or fd_ofileflags since operations
1927 * may block and rip them out from under us.
1928 */
1929 for (i = 0; i <= fdp->fd_lastfile; i++) {
1930 if (fdp->fd_ofiles[i] != NULL &&
1931 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1932 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1933 struct file *fp;
1934
1935 knote_fdclose(td, i);
1936 /*
1937 * NULL-out descriptor prior to close to avoid
1938 * a race while close blocks.
1939 */
1940 fp = fdp->fd_ofiles[i];
1941 fdp->fd_ofiles[i] = NULL;
1942 fdp->fd_ofileflags[i] = 0;
1943 fdunused(fdp, i);
1944 if (fp->f_type == DTYPE_MQUEUE)
1945 mq_fdclose(td, i, fp);
1946 FILEDESC_XUNLOCK(fdp);
1947 (void) closef(fp, td);
1948 FILEDESC_XLOCK(fdp);
1949 }
1950 }
1951 FILEDESC_XUNLOCK(fdp);
1952 }
1953
1954 /*
1955 * It is unsafe for set[ug]id processes to be started with file
1956 * descriptors 0..2 closed, as these descriptors are given implicit
1957 * significance in the Standard C library. fdcheckstd() will create a
1958 * descriptor referencing /dev/null for each of stdin, stdout, and
1959 * stderr that is not already open.
1960 */
1961 int
1962 fdcheckstd(struct thread *td)
1963 {
1964 struct filedesc *fdp;
1965 register_t retval, save;
1966 int i, error, devnull;
1967
1968 fdp = td->td_proc->p_fd;
1969 if (fdp == NULL)
1970 return (0);
1971 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1972 devnull = -1;
1973 error = 0;
1974 for (i = 0; i < 3; i++) {
1975 if (fdp->fd_ofiles[i] != NULL)
1976 continue;
1977 if (devnull < 0) {
1978 save = td->td_retval[0];
1979 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
1980 O_RDWR, 0);
1981 devnull = td->td_retval[0];
1982 KASSERT(devnull == i, ("oof, we didn't get our fd"));
1983 td->td_retval[0] = save;
1984 if (error)
1985 break;
1986 } else {
1987 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1988 if (error != 0)
1989 break;
1990 }
1991 }
1992 return (error);
1993 }
1994
1995 /*
1996 * Internal form of close. Decrement reference count on file structure.
1997 * Note: td may be NULL when closing a file that was being passed in a
1998 * message.
1999 *
2000 * XXXRW: Giant is not required for the caller, but often will be held; this
2001 * makes it moderately likely the Giant will be recursed in the VFS case.
2002 */
2003 int
2004 closef(struct file *fp, struct thread *td)
2005 {
2006 struct vnode *vp;
2007 struct flock lf;
2008 struct filedesc_to_leader *fdtol;
2009 struct filedesc *fdp;
2010
2011 /*
2012 * POSIX record locking dictates that any close releases ALL
2013 * locks owned by this process. This is handled by setting
2014 * a flag in the unlock to free ONLY locks obeying POSIX
2015 * semantics, and not to free BSD-style file locks.
2016 * If the descriptor was in a message, POSIX-style locks
2017 * aren't passed with the descriptor, and the thread pointer
2018 * will be NULL. Callers should be careful only to pass a
2019 * NULL thread pointer when there really is no owning
2020 * context that might have locks, or the locks will be
2021 * leaked.
2022 */
2023 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2024 int vfslocked;
2025
2026 vp = fp->f_vnode;
2027 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2028 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2029 lf.l_whence = SEEK_SET;
2030 lf.l_start = 0;
2031 lf.l_len = 0;
2032 lf.l_type = F_UNLCK;
2033 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2034 F_UNLCK, &lf, F_POSIX);
2035 }
2036 fdtol = td->td_proc->p_fdtol;
2037 if (fdtol != NULL) {
2038 /*
2039 * Handle special case where file descriptor table is
2040 * shared between multiple process leaders.
2041 */
2042 fdp = td->td_proc->p_fd;
2043 FILEDESC_XLOCK(fdp);
2044 for (fdtol = fdtol->fdl_next;
2045 fdtol != td->td_proc->p_fdtol;
2046 fdtol = fdtol->fdl_next) {
2047 if ((fdtol->fdl_leader->p_flag &
2048 P_ADVLOCK) == 0)
2049 continue;
2050 fdtol->fdl_holdcount++;
2051 FILEDESC_XUNLOCK(fdp);
2052 lf.l_whence = SEEK_SET;
2053 lf.l_start = 0;
2054 lf.l_len = 0;
2055 lf.l_type = F_UNLCK;
2056 vp = fp->f_vnode;
2057 (void) VOP_ADVLOCK(vp,
2058 (caddr_t)fdtol->fdl_leader,
2059 F_UNLCK, &lf, F_POSIX);
2060 FILEDESC_XLOCK(fdp);
2061 fdtol->fdl_holdcount--;
2062 if (fdtol->fdl_holdcount == 0 &&
2063 fdtol->fdl_wakeup != 0) {
2064 fdtol->fdl_wakeup = 0;
2065 wakeup(fdtol);
2066 }
2067 }
2068 FILEDESC_XUNLOCK(fdp);
2069 }
2070 VFS_UNLOCK_GIANT(vfslocked);
2071 }
2072 return (fdrop(fp, td));
2073 }
2074
2075 /*
2076 * Initialize the file pointer with the specified properties.
2077 *
2078 * The ops are set with release semantics to be certain that the flags, type,
2079 * and data are visible when ops is. This is to prevent ops methods from being
2080 * called with bad data.
2081 */
2082 void
2083 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2084 {
2085 fp->f_data = data;
2086 fp->f_flag = flag;
2087 fp->f_type = type;
2088 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2089 }
2090
2091 struct file *
2092 fget_unlocked(struct filedesc *fdp, int fd)
2093 {
2094 struct file *fp;
2095 u_int count;
2096
2097 if (fd < 0 || fd >= fdp->fd_nfiles)
2098 return (NULL);
2099 /*
2100 * Fetch the descriptor locklessly. We avoid fdrop() races by
2101 * never raising a refcount above 0. To accomplish this we have
2102 * to use a cmpset loop rather than an atomic_add. The descriptor
2103 * must be re-verified once we acquire a reference to be certain
2104 * that the identity is still correct and we did not lose a race
2105 * due to preemption.
2106 */
2107 for (;;) {
2108 fp = fdp->fd_ofiles[fd];
2109 if (fp == NULL)
2110 break;
2111 count = fp->f_count;
2112 if (count == 0)
2113 continue;
2114 /*
2115 * Use an acquire barrier to prevent caching of fd_ofiles
2116 * so it is refreshed for verification.
2117 */
2118 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2119 continue;
2120 if (fp == fdp->fd_ofiles[fd])
2121 break;
2122 fdrop(fp, curthread);
2123 }
2124
2125 return (fp);
2126 }
2127
2128 /*
2129 * Extract the file pointer associated with the specified descriptor for the
2130 * current user process.
2131 *
2132 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2133 * returned.
2134 *
2135 * If an error occured the non-zero error is returned and *fpp is set to
2136 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2137 * responsible for fdrop().
2138 */
2139 static __inline int
2140 _fget(struct thread *td, int fd, struct file **fpp, int flags)
2141 {
2142 struct filedesc *fdp;
2143 struct file *fp;
2144
2145 *fpp = NULL;
2146 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2147 return (EBADF);
2148 if ((fp = fget_unlocked(fdp, fd)) == NULL)
2149 return (EBADF);
2150 if (fp->f_ops == &badfileops) {
2151 fdrop(fp, td);
2152 return (EBADF);
2153 }
2154 /*
2155 * FREAD and FWRITE failure return EBADF as per POSIX.
2156 *
2157 * Only one flag, or 0, may be specified.
2158 */
2159 if ((flags == FREAD && (fp->f_flag & FREAD) == 0) ||
2160 (flags == FWRITE && (fp->f_flag & FWRITE) == 0)) {
2161 fdrop(fp, td);
2162 return (EBADF);
2163 }
2164 *fpp = fp;
2165 return (0);
2166 }
2167
2168 int
2169 fget(struct thread *td, int fd, struct file **fpp)
2170 {
2171
2172 return(_fget(td, fd, fpp, 0));
2173 }
2174
2175 int
2176 fget_read(struct thread *td, int fd, struct file **fpp)
2177 {
2178
2179 return(_fget(td, fd, fpp, FREAD));
2180 }
2181
2182 int
2183 fget_write(struct thread *td, int fd, struct file **fpp)
2184 {
2185
2186 return(_fget(td, fd, fpp, FWRITE));
2187 }
2188
2189 /*
2190 * Like fget() but loads the underlying vnode, or returns an error if the
2191 * descriptor does not represent a vnode. Note that pipes use vnodes but
2192 * never have VM objects. The returned vnode will be vref()'d.
2193 *
2194 * XXX: what about the unused flags ?
2195 */
2196 static __inline int
2197 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2198 {
2199 struct file *fp;
2200 int error;
2201
2202 *vpp = NULL;
2203 if ((error = _fget(td, fd, &fp, flags)) != 0)
2204 return (error);
2205 if (fp->f_vnode == NULL) {
2206 error = EINVAL;
2207 } else {
2208 *vpp = fp->f_vnode;
2209 vref(*vpp);
2210 }
2211 fdrop(fp, td);
2212
2213 return (error);
2214 }
2215
2216 int
2217 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2218 {
2219
2220 return (_fgetvp(td, fd, vpp, 0));
2221 }
2222
2223 int
2224 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2225 {
2226
2227 return (_fgetvp(td, fd, vpp, FREAD));
2228 }
2229
2230 #ifdef notyet
2231 int
2232 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2233 {
2234
2235 return (_fgetvp(td, fd, vpp, FWRITE));
2236 }
2237 #endif
2238
2239 /*
2240 * Like fget() but loads the underlying socket, or returns an error if the
2241 * descriptor does not represent a socket.
2242 *
2243 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2244 * in the future.
2245 *
2246 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2247 * on their file descriptor reference to prevent the socket from being free'd
2248 * during use.
2249 */
2250 int
2251 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2252 {
2253 struct file *fp;
2254 int error;
2255
2256 *spp = NULL;
2257 if (fflagp != NULL)
2258 *fflagp = 0;
2259 if ((error = _fget(td, fd, &fp, 0)) != 0)
2260 return (error);
2261 if (fp->f_type != DTYPE_SOCKET) {
2262 error = ENOTSOCK;
2263 } else {
2264 *spp = fp->f_data;
2265 if (fflagp)
2266 *fflagp = fp->f_flag;
2267 SOCK_LOCK(*spp);
2268 soref(*spp);
2269 SOCK_UNLOCK(*spp);
2270 }
2271 fdrop(fp, td);
2272
2273 return (error);
2274 }
2275
2276 /*
2277 * Drop the reference count on the socket and XXX release the SX lock in the
2278 * future. The last reference closes the socket.
2279 *
2280 * Note: fputsock() is deprecated, see comment for fgetsock().
2281 */
2282 void
2283 fputsock(struct socket *so)
2284 {
2285
2286 ACCEPT_LOCK();
2287 SOCK_LOCK(so);
2288 sorele(so);
2289 }
2290
2291 /*
2292 * Handle the last reference to a file being closed.
2293 */
2294 int
2295 _fdrop(struct file *fp, struct thread *td)
2296 {
2297 int error;
2298
2299 error = 0;
2300 if (fp->f_count != 0)
2301 panic("fdrop: count %d", fp->f_count);
2302 if (fp->f_ops != &badfileops)
2303 error = fo_close(fp, td);
2304 /*
2305 * The f_cdevpriv cannot be assigned non-NULL value while we
2306 * are destroying the file.
2307 */
2308 if (fp->f_cdevpriv != NULL)
2309 devfs_fpdrop(fp);
2310 atomic_subtract_int(&openfiles, 1);
2311 crfree(fp->f_cred);
2312 uma_zfree(file_zone, fp);
2313
2314 return (error);
2315 }
2316
2317 /*
2318 * Apply an advisory lock on a file descriptor.
2319 *
2320 * Just attempt to get a record lock of the requested type on the entire file
2321 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2322 */
2323 #ifndef _SYS_SYSPROTO_H_
2324 struct flock_args {
2325 int fd;
2326 int how;
2327 };
2328 #endif
2329 /* ARGSUSED */
2330 int
2331 flock(struct thread *td, struct flock_args *uap)
2332 {
2333 struct file *fp;
2334 struct vnode *vp;
2335 struct flock lf;
2336 int vfslocked;
2337 int error;
2338
2339 if ((error = fget(td, uap->fd, &fp)) != 0)
2340 return (error);
2341 if (fp->f_type != DTYPE_VNODE) {
2342 fdrop(fp, td);
2343 return (EOPNOTSUPP);
2344 }
2345
2346 vp = fp->f_vnode;
2347 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2348 lf.l_whence = SEEK_SET;
2349 lf.l_start = 0;
2350 lf.l_len = 0;
2351 if (uap->how & LOCK_UN) {
2352 lf.l_type = F_UNLCK;
2353 atomic_clear_int(&fp->f_flag, FHASLOCK);
2354 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2355 goto done2;
2356 }
2357 if (uap->how & LOCK_EX)
2358 lf.l_type = F_WRLCK;
2359 else if (uap->how & LOCK_SH)
2360 lf.l_type = F_RDLCK;
2361 else {
2362 error = EBADF;
2363 goto done2;
2364 }
2365 atomic_set_int(&fp->f_flag, FHASLOCK);
2366 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2367 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2368 done2:
2369 fdrop(fp, td);
2370 VFS_UNLOCK_GIANT(vfslocked);
2371 return (error);
2372 }
2373 /*
2374 * Duplicate the specified descriptor to a free descriptor.
2375 */
2376 int
2377 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2378 {
2379 struct file *wfp;
2380 struct file *fp;
2381
2382 /*
2383 * If the to-be-dup'd fd number is greater than the allowed number
2384 * of file descriptors, or the fd to be dup'd has already been
2385 * closed, then reject.
2386 */
2387 FILEDESC_XLOCK(fdp);
2388 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2389 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2390 FILEDESC_XUNLOCK(fdp);
2391 return (EBADF);
2392 }
2393
2394 /*
2395 * There are two cases of interest here.
2396 *
2397 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2398 *
2399 * For ENXIO steal away the file structure from (dfd) and store it in
2400 * (indx). (dfd) is effectively closed by this operation.
2401 *
2402 * Any other error code is just returned.
2403 */
2404 switch (error) {
2405 case ENODEV:
2406 /*
2407 * Check that the mode the file is being opened for is a
2408 * subset of the mode of the existing descriptor.
2409 */
2410 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2411 FILEDESC_XUNLOCK(fdp);
2412 return (EACCES);
2413 }
2414 fp = fdp->fd_ofiles[indx];
2415 fdp->fd_ofiles[indx] = wfp;
2416 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2417 if (fp == NULL)
2418 fdused(fdp, indx);
2419 fhold(wfp);
2420 FILEDESC_XUNLOCK(fdp);
2421 if (fp != NULL)
2422 /*
2423 * We now own the reference to fp that the ofiles[]
2424 * array used to own. Release it.
2425 */
2426 fdrop(fp, td);
2427 return (0);
2428
2429 case ENXIO:
2430 /*
2431 * Steal away the file pointer from dfd and stuff it into indx.
2432 */
2433 fp = fdp->fd_ofiles[indx];
2434 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2435 fdp->fd_ofiles[dfd] = NULL;
2436 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2437 fdp->fd_ofileflags[dfd] = 0;
2438 fdunused(fdp, dfd);
2439 if (fp == NULL)
2440 fdused(fdp, indx);
2441 FILEDESC_XUNLOCK(fdp);
2442
2443 /*
2444 * We now own the reference to fp that the ofiles[] array
2445 * used to own. Release it.
2446 */
2447 if (fp != NULL)
2448 fdrop(fp, td);
2449 return (0);
2450
2451 default:
2452 FILEDESC_XUNLOCK(fdp);
2453 return (error);
2454 }
2455 /* NOTREACHED */
2456 }
2457
2458 /*
2459 * Scan all active processes and prisons to see if any of them have a current
2460 * or root directory of `olddp'. If so, replace them with the new mount point.
2461 */
2462 void
2463 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2464 {
2465 struct filedesc *fdp;
2466 struct prison *pr;
2467 struct proc *p;
2468 int nrele;
2469
2470 if (vrefcnt(olddp) == 1)
2471 return;
2472 nrele = 0;
2473 sx_slock(&allproc_lock);
2474 FOREACH_PROC_IN_SYSTEM(p) {
2475 fdp = fdhold(p);
2476 if (fdp == NULL)
2477 continue;
2478 FILEDESC_XLOCK(fdp);
2479 if (fdp->fd_cdir == olddp) {
2480 vref(newdp);
2481 fdp->fd_cdir = newdp;
2482 nrele++;
2483 }
2484 if (fdp->fd_rdir == olddp) {
2485 vref(newdp);
2486 fdp->fd_rdir = newdp;
2487 nrele++;
2488 }
2489 if (fdp->fd_jdir == olddp) {
2490 vref(newdp);
2491 fdp->fd_jdir = newdp;
2492 nrele++;
2493 }
2494 FILEDESC_XUNLOCK(fdp);
2495 fddrop(fdp);
2496 }
2497 sx_sunlock(&allproc_lock);
2498 if (rootvnode == olddp) {
2499 vref(newdp);
2500 rootvnode = newdp;
2501 nrele++;
2502 }
2503 mtx_lock(&prison0.pr_mtx);
2504 if (prison0.pr_root == olddp) {
2505 vref(newdp);
2506 prison0.pr_root = newdp;
2507 nrele++;
2508 }
2509 mtx_unlock(&prison0.pr_mtx);
2510 sx_slock(&allprison_lock);
2511 TAILQ_FOREACH(pr, &allprison, pr_list) {
2512 mtx_lock(&pr->pr_mtx);
2513 if (pr->pr_root == olddp) {
2514 vref(newdp);
2515 pr->pr_root = newdp;
2516 nrele++;
2517 }
2518 mtx_unlock(&pr->pr_mtx);
2519 }
2520 sx_sunlock(&allprison_lock);
2521 while (nrele--)
2522 vrele(olddp);
2523 }
2524
2525 struct filedesc_to_leader *
2526 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2527 {
2528 struct filedesc_to_leader *fdtol;
2529
2530 fdtol = malloc(sizeof(struct filedesc_to_leader),
2531 M_FILEDESC_TO_LEADER,
2532 M_WAITOK);
2533 fdtol->fdl_refcount = 1;
2534 fdtol->fdl_holdcount = 0;
2535 fdtol->fdl_wakeup = 0;
2536 fdtol->fdl_leader = leader;
2537 if (old != NULL) {
2538 FILEDESC_XLOCK(fdp);
2539 fdtol->fdl_next = old->fdl_next;
2540 fdtol->fdl_prev = old;
2541 old->fdl_next = fdtol;
2542 fdtol->fdl_next->fdl_prev = fdtol;
2543 FILEDESC_XUNLOCK(fdp);
2544 } else {
2545 fdtol->fdl_next = fdtol;
2546 fdtol->fdl_prev = fdtol;
2547 }
2548 return (fdtol);
2549 }
2550
2551 /*
2552 * Get file structures globally.
2553 */
2554 static int
2555 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2556 {
2557 struct xfile xf;
2558 struct filedesc *fdp;
2559 struct file *fp;
2560 struct proc *p;
2561 int error, n;
2562
2563 error = sysctl_wire_old_buffer(req, 0);
2564 if (error != 0)
2565 return (error);
2566 if (req->oldptr == NULL) {
2567 n = 0;
2568 sx_slock(&allproc_lock);
2569 FOREACH_PROC_IN_SYSTEM(p) {
2570 if (p->p_state == PRS_NEW)
2571 continue;
2572 fdp = fdhold(p);
2573 if (fdp == NULL)
2574 continue;
2575 /* overestimates sparse tables. */
2576 if (fdp->fd_lastfile > 0)
2577 n += fdp->fd_lastfile;
2578 fddrop(fdp);
2579 }
2580 sx_sunlock(&allproc_lock);
2581 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2582 }
2583 error = 0;
2584 bzero(&xf, sizeof(xf));
2585 xf.xf_size = sizeof(xf);
2586 sx_slock(&allproc_lock);
2587 FOREACH_PROC_IN_SYSTEM(p) {
2588 if (p->p_state == PRS_NEW)
2589 continue;
2590 PROC_LOCK(p);
2591 if (p_cansee(req->td, p) != 0) {
2592 PROC_UNLOCK(p);
2593 continue;
2594 }
2595 xf.xf_pid = p->p_pid;
2596 xf.xf_uid = p->p_ucred->cr_uid;
2597 PROC_UNLOCK(p);
2598 fdp = fdhold(p);
2599 if (fdp == NULL)
2600 continue;
2601 FILEDESC_SLOCK(fdp);
2602 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2603 if ((fp = fdp->fd_ofiles[n]) == NULL)
2604 continue;
2605 xf.xf_fd = n;
2606 xf.xf_file = fp;
2607 xf.xf_data = fp->f_data;
2608 xf.xf_vnode = fp->f_vnode;
2609 xf.xf_type = fp->f_type;
2610 xf.xf_count = fp->f_count;
2611 xf.xf_msgcount = 0;
2612 xf.xf_offset = fp->f_offset;
2613 xf.xf_flag = fp->f_flag;
2614 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2615 if (error)
2616 break;
2617 }
2618 FILEDESC_SUNLOCK(fdp);
2619 fddrop(fdp);
2620 if (error)
2621 break;
2622 }
2623 sx_sunlock(&allproc_lock);
2624 return (error);
2625 }
2626
2627 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2628 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2629
2630 #ifdef KINFO_OFILE_SIZE
2631 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2632 #endif
2633
2634 #ifdef COMPAT_FREEBSD7
2635 static int
2636 export_vnode_for_osysctl(struct vnode *vp, int type,
2637 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2638 {
2639 int error;
2640 char *fullpath, *freepath;
2641 int vfslocked;
2642
2643 bzero(kif, sizeof(*kif));
2644 kif->kf_structsize = sizeof(*kif);
2645
2646 vref(vp);
2647 kif->kf_fd = type;
2648 kif->kf_type = KF_TYPE_VNODE;
2649 /* This function only handles directories. */
2650 if (vp->v_type != VDIR) {
2651 vrele(vp);
2652 return (ENOTDIR);
2653 }
2654 kif->kf_vnode_type = KF_VTYPE_VDIR;
2655
2656 /*
2657 * This is not a true file descriptor, so we set a bogus refcount
2658 * and offset to indicate these fields should be ignored.
2659 */
2660 kif->kf_ref_count = -1;
2661 kif->kf_offset = -1;
2662
2663 freepath = NULL;
2664 fullpath = "-";
2665 FILEDESC_SUNLOCK(fdp);
2666 vn_fullpath(curthread, vp, &fullpath, &freepath);
2667 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2668 vrele(vp);
2669 VFS_UNLOCK_GIANT(vfslocked);
2670 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2671 if (freepath != NULL)
2672 free(freepath, M_TEMP);
2673 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2674 FILEDESC_SLOCK(fdp);
2675 return (error);
2676 }
2677
2678 /*
2679 * Get per-process file descriptors for use by procstat(1), et al.
2680 */
2681 static int
2682 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
2683 {
2684 char *fullpath, *freepath;
2685 struct kinfo_ofile *kif;
2686 struct filedesc *fdp;
2687 int error, i, *name;
2688 struct socket *so;
2689 struct vnode *vp;
2690 struct file *fp;
2691 struct proc *p;
2692 struct tty *tp;
2693 int vfslocked;
2694
2695 name = (int *)arg1;
2696 if ((p = pfind((pid_t)name[0])) == NULL)
2697 return (ESRCH);
2698 if ((error = p_candebug(curthread, p))) {
2699 PROC_UNLOCK(p);
2700 return (error);
2701 }
2702 fdp = fdhold(p);
2703 PROC_UNLOCK(p);
2704 if (fdp == NULL)
2705 return (ENOENT);
2706 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2707 FILEDESC_SLOCK(fdp);
2708 if (fdp->fd_cdir != NULL)
2709 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2710 fdp, req);
2711 if (fdp->fd_rdir != NULL)
2712 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2713 fdp, req);
2714 if (fdp->fd_jdir != NULL)
2715 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2716 fdp, req);
2717 for (i = 0; i < fdp->fd_nfiles; i++) {
2718 if ((fp = fdp->fd_ofiles[i]) == NULL)
2719 continue;
2720 bzero(kif, sizeof(*kif));
2721 kif->kf_structsize = sizeof(*kif);
2722 vp = NULL;
2723 so = NULL;
2724 tp = NULL;
2725 kif->kf_fd = i;
2726 switch (fp->f_type) {
2727 case DTYPE_VNODE:
2728 kif->kf_type = KF_TYPE_VNODE;
2729 vp = fp->f_vnode;
2730 break;
2731
2732 case DTYPE_SOCKET:
2733 kif->kf_type = KF_TYPE_SOCKET;
2734 so = fp->f_data;
2735 break;
2736
2737 case DTYPE_PIPE:
2738 kif->kf_type = KF_TYPE_PIPE;
2739 break;
2740
2741 case DTYPE_FIFO:
2742 kif->kf_type = KF_TYPE_FIFO;
2743 vp = fp->f_vnode;
2744 break;
2745
2746 case DTYPE_KQUEUE:
2747 kif->kf_type = KF_TYPE_KQUEUE;
2748 break;
2749
2750 case DTYPE_CRYPTO:
2751 kif->kf_type = KF_TYPE_CRYPTO;
2752 break;
2753
2754 case DTYPE_MQUEUE:
2755 kif->kf_type = KF_TYPE_MQUEUE;
2756 break;
2757
2758 case DTYPE_SHM:
2759 kif->kf_type = KF_TYPE_SHM;
2760 break;
2761
2762 case DTYPE_SEM:
2763 kif->kf_type = KF_TYPE_SEM;
2764 break;
2765
2766 case DTYPE_PTS:
2767 kif->kf_type = KF_TYPE_PTS;
2768 tp = fp->f_data;
2769 break;
2770
2771 default:
2772 kif->kf_type = KF_TYPE_UNKNOWN;
2773 break;
2774 }
2775 kif->kf_ref_count = fp->f_count;
2776 if (fp->f_flag & FREAD)
2777 kif->kf_flags |= KF_FLAG_READ;
2778 if (fp->f_flag & FWRITE)
2779 kif->kf_flags |= KF_FLAG_WRITE;
2780 if (fp->f_flag & FAPPEND)
2781 kif->kf_flags |= KF_FLAG_APPEND;
2782 if (fp->f_flag & FASYNC)
2783 kif->kf_flags |= KF_FLAG_ASYNC;
2784 if (fp->f_flag & FFSYNC)
2785 kif->kf_flags |= KF_FLAG_FSYNC;
2786 if (fp->f_flag & FNONBLOCK)
2787 kif->kf_flags |= KF_FLAG_NONBLOCK;
2788 if (fp->f_flag & O_DIRECT)
2789 kif->kf_flags |= KF_FLAG_DIRECT;
2790 if (fp->f_flag & FHASLOCK)
2791 kif->kf_flags |= KF_FLAG_HASLOCK;
2792 kif->kf_offset = fp->f_offset;
2793 if (vp != NULL) {
2794 vref(vp);
2795 switch (vp->v_type) {
2796 case VNON:
2797 kif->kf_vnode_type = KF_VTYPE_VNON;
2798 break;
2799 case VREG:
2800 kif->kf_vnode_type = KF_VTYPE_VREG;
2801 break;
2802 case VDIR:
2803 kif->kf_vnode_type = KF_VTYPE_VDIR;
2804 break;
2805 case VBLK:
2806 kif->kf_vnode_type = KF_VTYPE_VBLK;
2807 break;
2808 case VCHR:
2809 kif->kf_vnode_type = KF_VTYPE_VCHR;
2810 break;
2811 case VLNK:
2812 kif->kf_vnode_type = KF_VTYPE_VLNK;
2813 break;
2814 case VSOCK:
2815 kif->kf_vnode_type = KF_VTYPE_VSOCK;
2816 break;
2817 case VFIFO:
2818 kif->kf_vnode_type = KF_VTYPE_VFIFO;
2819 break;
2820 case VBAD:
2821 kif->kf_vnode_type = KF_VTYPE_VBAD;
2822 break;
2823 default:
2824 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2825 break;
2826 }
2827 /*
2828 * It is OK to drop the filedesc lock here as we will
2829 * re-validate and re-evaluate its properties when
2830 * the loop continues.
2831 */
2832 freepath = NULL;
2833 fullpath = "-";
2834 FILEDESC_SUNLOCK(fdp);
2835 vn_fullpath(curthread, vp, &fullpath, &freepath);
2836 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2837 vrele(vp);
2838 VFS_UNLOCK_GIANT(vfslocked);
2839 strlcpy(kif->kf_path, fullpath,
2840 sizeof(kif->kf_path));
2841 if (freepath != NULL)
2842 free(freepath, M_TEMP);
2843 FILEDESC_SLOCK(fdp);
2844 }
2845 if (so != NULL) {
2846 struct sockaddr *sa;
2847
2848 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
2849 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
2850 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
2851 free(sa, M_SONAME);
2852 }
2853 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
2854 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
2855 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
2856 free(sa, M_SONAME);
2857 }
2858 kif->kf_sock_domain =
2859 so->so_proto->pr_domain->dom_family;
2860 kif->kf_sock_type = so->so_type;
2861 kif->kf_sock_protocol = so->so_proto->pr_protocol;
2862 }
2863 if (tp != NULL) {
2864 strlcpy(kif->kf_path, tty_devname(tp),
2865 sizeof(kif->kf_path));
2866 }
2867 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2868 if (error)
2869 break;
2870 }
2871 FILEDESC_SUNLOCK(fdp);
2872 fddrop(fdp);
2873 free(kif, M_TEMP);
2874 return (0);
2875 }
2876
2877 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
2878 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
2879 #endif /* COMPAT_FREEBSD7 */
2880
2881 #ifdef KINFO_FILE_SIZE
2882 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2883 #endif
2884
2885 static int
2886 export_vnode_for_sysctl(struct vnode *vp, int type,
2887 struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req)
2888 {
2889 int error;
2890 char *fullpath, *freepath;
2891 int vfslocked;
2892
2893 bzero(kif, sizeof(*kif));
2894
2895 vref(vp);
2896 kif->kf_fd = type;
2897 kif->kf_type = KF_TYPE_VNODE;
2898 /* This function only handles directories. */
2899 if (vp->v_type != VDIR) {
2900 vrele(vp);
2901 return (ENOTDIR);
2902 }
2903 kif->kf_vnode_type = KF_VTYPE_VDIR;
2904
2905 /*
2906 * This is not a true file descriptor, so we set a bogus refcount
2907 * and offset to indicate these fields should be ignored.
2908 */
2909 kif->kf_ref_count = -1;
2910 kif->kf_offset = -1;
2911
2912 freepath = NULL;
2913 fullpath = "-";
2914 FILEDESC_SUNLOCK(fdp);
2915 vn_fullpath(curthread, vp, &fullpath, &freepath);
2916 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2917 vrele(vp);
2918 VFS_UNLOCK_GIANT(vfslocked);
2919 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2920 if (freepath != NULL)
2921 free(freepath, M_TEMP);
2922 /* Pack record size down */
2923 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
2924 strlen(kif->kf_path) + 1;
2925 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
2926 error = SYSCTL_OUT(req, kif, kif->kf_structsize);
2927 FILEDESC_SLOCK(fdp);
2928 return (error);
2929 }
2930
2931 /*
2932 * Get per-process file descriptors for use by procstat(1), et al.
2933 */
2934 static int
2935 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
2936 {
2937 char *fullpath, *freepath;
2938 struct kinfo_file *kif;
2939 struct filedesc *fdp;
2940 int error, i, *name;
2941 struct socket *so;
2942 struct vnode *vp;
2943 struct file *fp;
2944 struct proc *p;
2945 struct tty *tp;
2946 int vfslocked;
2947 size_t oldidx;
2948
2949 name = (int *)arg1;
2950 if ((p = pfind((pid_t)name[0])) == NULL)
2951 return (ESRCH);
2952 if ((error = p_candebug(curthread, p))) {
2953 PROC_UNLOCK(p);
2954 return (error);
2955 }
2956 fdp = fdhold(p);
2957 PROC_UNLOCK(p);
2958 if (fdp == NULL)
2959 return (ENOENT);
2960 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2961 FILEDESC_SLOCK(fdp);
2962 if (fdp->fd_cdir != NULL)
2963 export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2964 fdp, req);
2965 if (fdp->fd_rdir != NULL)
2966 export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2967 fdp, req);
2968 if (fdp->fd_jdir != NULL)
2969 export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2970 fdp, req);
2971 for (i = 0; i < fdp->fd_nfiles; i++) {
2972 if ((fp = fdp->fd_ofiles[i]) == NULL)
2973 continue;
2974 bzero(kif, sizeof(*kif));
2975 vp = NULL;
2976 so = NULL;
2977 tp = NULL;
2978 kif->kf_fd = i;
2979 switch (fp->f_type) {
2980 case DTYPE_VNODE:
2981 kif->kf_type = KF_TYPE_VNODE;
2982 vp = fp->f_vnode;
2983 break;
2984
2985 case DTYPE_SOCKET:
2986 kif->kf_type = KF_TYPE_SOCKET;
2987 so = fp->f_data;
2988 break;
2989
2990 case DTYPE_PIPE:
2991 kif->kf_type = KF_TYPE_PIPE;
2992 break;
2993
2994 case DTYPE_FIFO:
2995 kif->kf_type = KF_TYPE_FIFO;
2996 vp = fp->f_vnode;
2997 break;
2998
2999 case DTYPE_KQUEUE:
3000 kif->kf_type = KF_TYPE_KQUEUE;
3001 break;
3002
3003 case DTYPE_CRYPTO:
3004 kif->kf_type = KF_TYPE_CRYPTO;
3005 break;
3006
3007 case DTYPE_MQUEUE:
3008 kif->kf_type = KF_TYPE_MQUEUE;
3009 break;
3010
3011 case DTYPE_SHM:
3012 kif->kf_type = KF_TYPE_SHM;
3013 break;
3014
3015 case DTYPE_SEM:
3016 kif->kf_type = KF_TYPE_SEM;
3017 break;
3018
3019 case DTYPE_PTS:
3020 kif->kf_type = KF_TYPE_PTS;
3021 tp = fp->f_data;
3022 break;
3023
3024 default:
3025 kif->kf_type = KF_TYPE_UNKNOWN;
3026 break;
3027 }
3028 kif->kf_ref_count = fp->f_count;
3029 if (fp->f_flag & FREAD)
3030 kif->kf_flags |= KF_FLAG_READ;
3031 if (fp->f_flag & FWRITE)
3032 kif->kf_flags |= KF_FLAG_WRITE;
3033 if (fp->f_flag & FAPPEND)
3034 kif->kf_flags |= KF_FLAG_APPEND;
3035 if (fp->f_flag & FASYNC)
3036 kif->kf_flags |= KF_FLAG_ASYNC;
3037 if (fp->f_flag & FFSYNC)
3038 kif->kf_flags |= KF_FLAG_FSYNC;
3039 if (fp->f_flag & FNONBLOCK)
3040 kif->kf_flags |= KF_FLAG_NONBLOCK;
3041 if (fp->f_flag & O_DIRECT)
3042 kif->kf_flags |= KF_FLAG_DIRECT;
3043 if (fp->f_flag & FHASLOCK)
3044 kif->kf_flags |= KF_FLAG_HASLOCK;
3045 kif->kf_offset = fp->f_offset;
3046 if (vp != NULL) {
3047 vref(vp);
3048 switch (vp->v_type) {
3049 case VNON:
3050 kif->kf_vnode_type = KF_VTYPE_VNON;
3051 break;
3052 case VREG:
3053 kif->kf_vnode_type = KF_VTYPE_VREG;
3054 break;
3055 case VDIR:
3056 kif->kf_vnode_type = KF_VTYPE_VDIR;
3057 break;
3058 case VBLK:
3059 kif->kf_vnode_type = KF_VTYPE_VBLK;
3060 break;
3061 case VCHR:
3062 kif->kf_vnode_type = KF_VTYPE_VCHR;
3063 break;
3064 case VLNK:
3065 kif->kf_vnode_type = KF_VTYPE_VLNK;
3066 break;
3067 case VSOCK:
3068 kif->kf_vnode_type = KF_VTYPE_VSOCK;
3069 break;
3070 case VFIFO:
3071 kif->kf_vnode_type = KF_VTYPE_VFIFO;
3072 break;
3073 case VBAD:
3074 kif->kf_vnode_type = KF_VTYPE_VBAD;
3075 break;
3076 default:
3077 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3078 break;
3079 }
3080 /*
3081 * It is OK to drop the filedesc lock here as we will
3082 * re-validate and re-evaluate its properties when
3083 * the loop continues.
3084 */
3085 freepath = NULL;
3086 fullpath = "-";
3087 FILEDESC_SUNLOCK(fdp);
3088 vn_fullpath(curthread, vp, &fullpath, &freepath);
3089 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
3090 vrele(vp);
3091 VFS_UNLOCK_GIANT(vfslocked);
3092 strlcpy(kif->kf_path, fullpath,
3093 sizeof(kif->kf_path));
3094 if (freepath != NULL)
3095 free(freepath, M_TEMP);
3096 FILEDESC_SLOCK(fdp);
3097 }
3098 if (so != NULL) {
3099 struct sockaddr *sa;
3100
3101 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3102 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3103 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3104 free(sa, M_SONAME);
3105 }
3106 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3107 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3108 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3109 free(sa, M_SONAME);
3110 }
3111 kif->kf_sock_domain =
3112 so->so_proto->pr_domain->dom_family;
3113 kif->kf_sock_type = so->so_type;
3114 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3115 }
3116 if (tp != NULL) {
3117 strlcpy(kif->kf_path, tty_devname(tp),
3118 sizeof(kif->kf_path));
3119 }
3120 /* Pack record size down */
3121 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3122 strlen(kif->kf_path) + 1;
3123 kif->kf_structsize = roundup(kif->kf_structsize,
3124 sizeof(uint64_t));
3125 oldidx = req->oldidx;
3126 error = SYSCTL_OUT(req, kif, kif->kf_structsize);
3127 if (error) {
3128 if (error == ENOMEM) {
3129 /*
3130 * The hack to keep the ABI of sysctl
3131 * kern.proc.filedesc intact, but not
3132 * to account a partially copied
3133 * kinfo_file into the oldidx.
3134 */
3135 req->oldidx = oldidx;
3136 error = 0;
3137 }
3138 break;
3139 }
3140 }
3141 FILEDESC_SUNLOCK(fdp);
3142 fddrop(fdp);
3143 free(kif, M_TEMP);
3144 return (error);
3145 }
3146
3147 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3148 sysctl_kern_proc_filedesc, "Process filedesc entries");
3149
3150 #ifdef DDB
3151 /*
3152 * For the purposes of debugging, generate a human-readable string for the
3153 * file type.
3154 */
3155 static const char *
3156 file_type_to_name(short type)
3157 {
3158
3159 switch (type) {
3160 case 0:
3161 return ("zero");
3162 case DTYPE_VNODE:
3163 return ("vnod");
3164 case DTYPE_SOCKET:
3165 return ("sock");
3166 case DTYPE_PIPE:
3167 return ("pipe");
3168 case DTYPE_FIFO:
3169 return ("fifo");
3170 case DTYPE_KQUEUE:
3171 return ("kque");
3172 case DTYPE_CRYPTO:
3173 return ("crpt");
3174 case DTYPE_MQUEUE:
3175 return ("mque");
3176 case DTYPE_SHM:
3177 return ("shm");
3178 case DTYPE_SEM:
3179 return ("ksem");
3180 default:
3181 return ("unkn");
3182 }
3183 }
3184
3185 /*
3186 * For the purposes of debugging, identify a process (if any, perhaps one of
3187 * many) that references the passed file in its file descriptor array. Return
3188 * NULL if none.
3189 */
3190 static struct proc *
3191 file_to_first_proc(struct file *fp)
3192 {
3193 struct filedesc *fdp;
3194 struct proc *p;
3195 int n;
3196
3197 FOREACH_PROC_IN_SYSTEM(p) {
3198 if (p->p_state == PRS_NEW)
3199 continue;
3200 fdp = p->p_fd;
3201 if (fdp == NULL)
3202 continue;
3203 for (n = 0; n < fdp->fd_nfiles; n++) {
3204 if (fp == fdp->fd_ofiles[n])
3205 return (p);
3206 }
3207 }
3208 return (NULL);
3209 }
3210
3211 static void
3212 db_print_file(struct file *fp, int header)
3213 {
3214 struct proc *p;
3215
3216 if (header)
3217 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3218 "File", "Type", "Data", "Flag", "GCFl", "Count",
3219 "MCount", "Vnode", "FPID", "FCmd");
3220 p = file_to_first_proc(fp);
3221 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3222 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3223 0, fp->f_count, 0, fp->f_vnode,
3224 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3225 }
3226
3227 DB_SHOW_COMMAND(file, db_show_file)
3228 {
3229 struct file *fp;
3230
3231 if (!have_addr) {
3232 db_printf("usage: show file <addr>\n");
3233 return;
3234 }
3235 fp = (struct file *)addr;
3236 db_print_file(fp, 1);
3237 }
3238
3239 DB_SHOW_COMMAND(files, db_show_files)
3240 {
3241 struct filedesc *fdp;
3242 struct file *fp;
3243 struct proc *p;
3244 int header;
3245 int n;
3246
3247 header = 1;
3248 FOREACH_PROC_IN_SYSTEM(p) {
3249 if (p->p_state == PRS_NEW)
3250 continue;
3251 if ((fdp = p->p_fd) == NULL)
3252 continue;
3253 for (n = 0; n < fdp->fd_nfiles; ++n) {
3254 if ((fp = fdp->fd_ofiles[n]) == NULL)
3255 continue;
3256 db_print_file(fp, header);
3257 header = 0;
3258 }
3259 }
3260 }
3261 #endif
3262
3263 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3264 &maxfilesperproc, 0, "Maximum files allowed open per process");
3265
3266 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3267 &maxfiles, 0, "Maximum number of files");
3268
3269 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3270 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3271
3272 /* ARGSUSED*/
3273 static void
3274 filelistinit(void *dummy)
3275 {
3276
3277 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3278 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3279 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3280 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3281 }
3282 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3283
3284 /*-------------------------------------------------------------------*/
3285
3286 static int
3287 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
3288 {
3289
3290 return (EBADF);
3291 }
3292
3293 static int
3294 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td)
3295 {
3296
3297 return (EINVAL);
3298 }
3299
3300 static int
3301 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
3302 {
3303
3304 return (EBADF);
3305 }
3306
3307 static int
3308 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
3309 {
3310
3311 return (0);
3312 }
3313
3314 static int
3315 badfo_kqfilter(struct file *fp, struct knote *kn)
3316 {
3317
3318 return (EBADF);
3319 }
3320
3321 static int
3322 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
3323 {
3324
3325 return (EBADF);
3326 }
3327
3328 static int
3329 badfo_close(struct file *fp, struct thread *td)
3330 {
3331
3332 return (EBADF);
3333 }
3334
3335 struct fileops badfileops = {
3336 .fo_read = badfo_readwrite,
3337 .fo_write = badfo_readwrite,
3338 .fo_truncate = badfo_truncate,
3339 .fo_ioctl = badfo_ioctl,
3340 .fo_poll = badfo_poll,
3341 .fo_kqfilter = badfo_kqfilter,
3342 .fo_stat = badfo_stat,
3343 .fo_close = badfo_close,
3344 };
3345
3346
3347 /*-------------------------------------------------------------------*/
3348
3349 /*
3350 * File Descriptor pseudo-device driver (/dev/fd/).
3351 *
3352 * Opening minor device N dup()s the file (if any) connected to file
3353 * descriptor N belonging to the calling process. Note that this driver
3354 * consists of only the ``open()'' routine, because all subsequent
3355 * references to this file will be direct to the other driver.
3356 *
3357 * XXX: we could give this one a cloning event handler if necessary.
3358 */
3359
3360 /* ARGSUSED */
3361 static int
3362 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3363 {
3364
3365 /*
3366 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3367 * the file descriptor being sought for duplication. The error
3368 * return ensures that the vnode for this device will be released
3369 * by vn_open. Open will detect this special error and take the
3370 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3371 * will simply report the error.
3372 */
3373 td->td_dupfd = dev2unit(dev);
3374 return (ENODEV);
3375 }
3376
3377 static struct cdevsw fildesc_cdevsw = {
3378 .d_version = D_VERSION,
3379 .d_open = fdopen,
3380 .d_name = "FD",
3381 };
3382
3383 static void
3384 fildesc_drvinit(void *unused)
3385 {
3386 struct cdev *dev;
3387
3388 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
3389 make_dev_alias(dev, "stdin");
3390 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
3391 make_dev_alias(dev, "stdout");
3392 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
3393 make_dev_alias(dev, "stderr");
3394 }
3395
3396 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
Cache object: 469e5fc3ee1490962dd4b0862cdca3eb
|