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/9.1/sys/kern/kern_descrip.c 238560 2012-07-18 04:52:37Z mjg $");
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/capability.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/limits.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mman.h>
62 #include <sys/mount.h>
63 #include <sys/mqueue.h>
64 #include <sys/mutex.h>
65 #include <sys/namei.h>
66 #include <sys/selinfo.h>
67 #include <sys/pipe.h>
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/procdesc.h>
71 #include <sys/protosw.h>
72 #include <sys/racct.h>
73 #include <sys/resourcevar.h>
74 #include <sys/signalvar.h>
75 #include <sys/socketvar.h>
76 #include <sys/stat.h>
77 #include <sys/sx.h>
78 #include <sys/syscallsubr.h>
79 #include <sys/sysctl.h>
80 #include <sys/sysproto.h>
81 #include <sys/tty.h>
82 #include <sys/unistd.h>
83 #include <sys/un.h>
84 #include <sys/unpcb.h>
85 #include <sys/user.h>
86 #include <sys/vnode.h>
87 #ifdef KTRACE
88 #include <sys/ktrace.h>
89 #endif
90
91 #include <net/vnet.h>
92
93 #include <netinet/in.h>
94 #include <netinet/in_pcb.h>
95
96 #include <security/audit/audit.h>
97
98 #include <vm/uma.h>
99 #include <vm/vm.h>
100
101 #include <ddb/ddb.h>
102
103 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
104 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
105 "file desc to leader structures");
106 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
107
108 MALLOC_DECLARE(M_FADVISE);
109
110 static uma_zone_t file_zone;
111
112
113 /* Flags for do_dup() */
114 #define DUP_FIXED 0x1 /* Force fixed allocation */
115 #define DUP_FCNTL 0x2 /* fcntl()-style errors */
116
117 static int do_dup(struct thread *td, int flags, int old, int new,
118 register_t *retval);
119 static int fd_first_free(struct filedesc *, int, int);
120 static int fd_last_used(struct filedesc *, int, int);
121 static void fdgrowtable(struct filedesc *, int);
122 static void fdunused(struct filedesc *fdp, int fd);
123 static void fdused(struct filedesc *fdp, int fd);
124 static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
125 static int fill_socket_info(struct socket *so, struct kinfo_file *kif);
126 static int fill_pts_info(struct tty *tp, struct kinfo_file *kif);
127 static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
128 static int fill_procdesc_info(struct procdesc *pdp,
129 struct kinfo_file *kif);
130 static int fill_shm_info(struct file *fp, struct kinfo_file *kif);
131
132 /*
133 * A process is initially started out with NDFILE descriptors stored within
134 * this structure, selected to be enough for typical applications based on
135 * the historical limit of 20 open files (and the usage of descriptors by
136 * shells). If these descriptors are exhausted, a larger descriptor table
137 * may be allocated, up to a process' resource limit; the internal arrays
138 * are then unused.
139 */
140 #define NDFILE 20
141 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
142 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
143 #define NDSLOT(x) ((x) / NDENTRIES)
144 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
145 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
146
147 /*
148 * Storage required per open file descriptor.
149 */
150 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
151
152 /*
153 * Storage to hold unused ofiles that need to be reclaimed.
154 */
155 struct freetable {
156 struct file **ft_table;
157 SLIST_ENTRY(freetable) ft_next;
158 };
159
160 /*
161 * Basic allocation of descriptors:
162 * one of the above, plus arrays for NDFILE descriptors.
163 */
164 struct filedesc0 {
165 struct filedesc fd_fd;
166 /*
167 * ofiles which need to be reclaimed on free.
168 */
169 SLIST_HEAD(,freetable) fd_free;
170 /*
171 * These arrays are used when the number of open files is
172 * <= NDFILE, and are then pointed to by the pointers above.
173 */
174 struct file *fd_dfiles[NDFILE];
175 char fd_dfileflags[NDFILE];
176 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
177 };
178
179 /*
180 * Descriptor management.
181 */
182 volatile int openfiles; /* actual number of open files */
183 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
184 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
185
186 /* A mutex to protect the association between a proc and filedesc. */
187 static struct mtx fdesc_mtx;
188
189 /*
190 * Find the first zero bit in the given bitmap, starting at low and not
191 * exceeding size - 1.
192 */
193 static int
194 fd_first_free(struct filedesc *fdp, int low, int size)
195 {
196 NDSLOTTYPE *map = fdp->fd_map;
197 NDSLOTTYPE mask;
198 int off, maxoff;
199
200 if (low >= size)
201 return (low);
202
203 off = NDSLOT(low);
204 if (low % NDENTRIES) {
205 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
206 if ((mask &= ~map[off]) != 0UL)
207 return (off * NDENTRIES + ffsl(mask) - 1);
208 ++off;
209 }
210 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
211 if (map[off] != ~0UL)
212 return (off * NDENTRIES + ffsl(~map[off]) - 1);
213 return (size);
214 }
215
216 /*
217 * Find the highest non-zero bit in the given bitmap, starting at low and
218 * not exceeding size - 1.
219 */
220 static int
221 fd_last_used(struct filedesc *fdp, int low, int size)
222 {
223 NDSLOTTYPE *map = fdp->fd_map;
224 NDSLOTTYPE mask;
225 int off, minoff;
226
227 if (low >= size)
228 return (-1);
229
230 off = NDSLOT(size);
231 if (size % NDENTRIES) {
232 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
233 if ((mask &= map[off]) != 0)
234 return (off * NDENTRIES + flsl(mask) - 1);
235 --off;
236 }
237 for (minoff = NDSLOT(low); off >= minoff; --off)
238 if (map[off] != 0)
239 return (off * NDENTRIES + flsl(map[off]) - 1);
240 return (low - 1);
241 }
242
243 static int
244 fdisused(struct filedesc *fdp, int fd)
245 {
246 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
249 }
250
251 /*
252 * Mark a file descriptor as used.
253 */
254 static void
255 fdused(struct filedesc *fdp, int fd)
256 {
257
258 FILEDESC_XLOCK_ASSERT(fdp);
259 KASSERT(!fdisused(fdp, fd),
260 ("fd already used"));
261
262 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
263 if (fd > fdp->fd_lastfile)
264 fdp->fd_lastfile = fd;
265 if (fd == fdp->fd_freefile)
266 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
267 }
268
269 /*
270 * Mark a file descriptor as unused.
271 */
272 static void
273 fdunused(struct filedesc *fdp, int fd)
274 {
275
276 FILEDESC_XLOCK_ASSERT(fdp);
277 KASSERT(fdisused(fdp, fd),
278 ("fd is already unused"));
279 KASSERT(fdp->fd_ofiles[fd] == NULL,
280 ("fd is still in use"));
281
282 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
283 if (fd < fdp->fd_freefile)
284 fdp->fd_freefile = fd;
285 if (fd == fdp->fd_lastfile)
286 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
287 }
288
289 /*
290 * System calls on descriptors.
291 */
292 #ifndef _SYS_SYSPROTO_H_
293 struct getdtablesize_args {
294 int dummy;
295 };
296 #endif
297 /* ARGSUSED */
298 int
299 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
300 {
301 struct proc *p = td->td_proc;
302 uint64_t lim;
303
304 PROC_LOCK(p);
305 td->td_retval[0] =
306 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
307 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
308 PROC_UNLOCK(p);
309 if (lim < td->td_retval[0])
310 td->td_retval[0] = lim;
311 return (0);
312 }
313
314 /*
315 * Duplicate a file descriptor to a particular value.
316 *
317 * Note: keep in mind that a potential race condition exists when closing
318 * descriptors from a shared descriptor table (via rfork).
319 */
320 #ifndef _SYS_SYSPROTO_H_
321 struct dup2_args {
322 u_int from;
323 u_int to;
324 };
325 #endif
326 /* ARGSUSED */
327 int
328 sys_dup2(struct thread *td, struct dup2_args *uap)
329 {
330
331 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
332 td->td_retval));
333 }
334
335 /*
336 * Duplicate a file descriptor.
337 */
338 #ifndef _SYS_SYSPROTO_H_
339 struct dup_args {
340 u_int fd;
341 };
342 #endif
343 /* ARGSUSED */
344 int
345 sys_dup(struct thread *td, struct dup_args *uap)
346 {
347
348 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
349 }
350
351 /*
352 * The file control system call.
353 */
354 #ifndef _SYS_SYSPROTO_H_
355 struct fcntl_args {
356 int fd;
357 int cmd;
358 long arg;
359 };
360 #endif
361 /* ARGSUSED */
362 int
363 sys_fcntl(struct thread *td, struct fcntl_args *uap)
364 {
365 struct flock fl;
366 struct oflock ofl;
367 intptr_t arg;
368 int error;
369 int cmd;
370
371 error = 0;
372 cmd = uap->cmd;
373 switch (uap->cmd) {
374 case F_OGETLK:
375 case F_OSETLK:
376 case F_OSETLKW:
377 /*
378 * Convert old flock structure to new.
379 */
380 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
381 fl.l_start = ofl.l_start;
382 fl.l_len = ofl.l_len;
383 fl.l_pid = ofl.l_pid;
384 fl.l_type = ofl.l_type;
385 fl.l_whence = ofl.l_whence;
386 fl.l_sysid = 0;
387
388 switch (uap->cmd) {
389 case F_OGETLK:
390 cmd = F_GETLK;
391 break;
392 case F_OSETLK:
393 cmd = F_SETLK;
394 break;
395 case F_OSETLKW:
396 cmd = F_SETLKW;
397 break;
398 }
399 arg = (intptr_t)&fl;
400 break;
401 case F_GETLK:
402 case F_SETLK:
403 case F_SETLKW:
404 case F_SETLK_REMOTE:
405 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
406 arg = (intptr_t)&fl;
407 break;
408 default:
409 arg = uap->arg;
410 break;
411 }
412 if (error)
413 return (error);
414 error = kern_fcntl(td, uap->fd, cmd, arg);
415 if (error)
416 return (error);
417 if (uap->cmd == F_OGETLK) {
418 ofl.l_start = fl.l_start;
419 ofl.l_len = fl.l_len;
420 ofl.l_pid = fl.l_pid;
421 ofl.l_type = fl.l_type;
422 ofl.l_whence = fl.l_whence;
423 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
424 } else if (uap->cmd == F_GETLK) {
425 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
426 }
427 return (error);
428 }
429
430 static inline struct file *
431 fdtofp(int fd, struct filedesc *fdp)
432 {
433 struct file *fp;
434
435 FILEDESC_LOCK_ASSERT(fdp);
436 if ((unsigned)fd >= fdp->fd_nfiles ||
437 (fp = fdp->fd_ofiles[fd]) == NULL)
438 return (NULL);
439 return (fp);
440 }
441
442 static inline int
443 fdunwrap(int fd, cap_rights_t rights, struct filedesc *fdp, struct file **fpp)
444 {
445
446 *fpp = fdtofp(fd, fdp);
447 if (*fpp == NULL)
448 return (EBADF);
449
450 #ifdef CAPABILITIES
451 if ((*fpp)->f_type == DTYPE_CAPABILITY) {
452 int err = cap_funwrap(*fpp, rights, fpp);
453 if (err != 0) {
454 *fpp = NULL;
455 return (err);
456 }
457 }
458 #endif /* CAPABILITIES */
459 return (0);
460 }
461
462 int
463 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
464 {
465 struct filedesc *fdp;
466 struct flock *flp;
467 struct file *fp;
468 struct proc *p;
469 char *pop;
470 struct vnode *vp;
471 int error, flg, tmp;
472 int vfslocked;
473 u_int old, new;
474 uint64_t bsize;
475
476 vfslocked = 0;
477 error = 0;
478 flg = F_POSIX;
479 p = td->td_proc;
480 fdp = p->p_fd;
481
482 switch (cmd) {
483 case F_DUPFD:
484 tmp = arg;
485 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
486 break;
487
488 case F_DUP2FD:
489 tmp = arg;
490 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
491 break;
492
493 case F_GETFD:
494 FILEDESC_SLOCK(fdp);
495 if ((fp = fdtofp(fd, fdp)) == NULL) {
496 FILEDESC_SUNLOCK(fdp);
497 error = EBADF;
498 break;
499 }
500 pop = &fdp->fd_ofileflags[fd];
501 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
502 FILEDESC_SUNLOCK(fdp);
503 break;
504
505 case F_SETFD:
506 FILEDESC_XLOCK(fdp);
507 if ((fp = fdtofp(fd, fdp)) == NULL) {
508 FILEDESC_XUNLOCK(fdp);
509 error = EBADF;
510 break;
511 }
512 pop = &fdp->fd_ofileflags[fd];
513 *pop = (*pop &~ UF_EXCLOSE) |
514 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
515 FILEDESC_XUNLOCK(fdp);
516 break;
517
518 case F_GETFL:
519 FILEDESC_SLOCK(fdp);
520 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
521 if (error != 0) {
522 FILEDESC_SUNLOCK(fdp);
523 break;
524 }
525 td->td_retval[0] = OFLAGS(fp->f_flag);
526 FILEDESC_SUNLOCK(fdp);
527 break;
528
529 case F_SETFL:
530 FILEDESC_SLOCK(fdp);
531 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
532 if (error != 0) {
533 FILEDESC_SUNLOCK(fdp);
534 break;
535 }
536 fhold(fp);
537 FILEDESC_SUNLOCK(fdp);
538 do {
539 tmp = flg = fp->f_flag;
540 tmp &= ~FCNTLFLAGS;
541 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
542 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
543 tmp = fp->f_flag & FNONBLOCK;
544 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
545 if (error) {
546 fdrop(fp, td);
547 break;
548 }
549 tmp = fp->f_flag & FASYNC;
550 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
551 if (error == 0) {
552 fdrop(fp, td);
553 break;
554 }
555 atomic_clear_int(&fp->f_flag, FNONBLOCK);
556 tmp = 0;
557 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
558 fdrop(fp, td);
559 break;
560
561 case F_GETOWN:
562 FILEDESC_SLOCK(fdp);
563 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
564 if (error != 0) {
565 FILEDESC_SUNLOCK(fdp);
566 break;
567 }
568 fhold(fp);
569 FILEDESC_SUNLOCK(fdp);
570 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
571 if (error == 0)
572 td->td_retval[0] = tmp;
573 fdrop(fp, td);
574 break;
575
576 case F_SETOWN:
577 FILEDESC_SLOCK(fdp);
578 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
579 if (error != 0) {
580 FILEDESC_SUNLOCK(fdp);
581 break;
582 }
583 fhold(fp);
584 FILEDESC_SUNLOCK(fdp);
585 tmp = arg;
586 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
587 fdrop(fp, td);
588 break;
589
590 case F_SETLK_REMOTE:
591 error = priv_check(td, PRIV_NFS_LOCKD);
592 if (error)
593 return (error);
594 flg = F_REMOTE;
595 goto do_setlk;
596
597 case F_SETLKW:
598 flg |= F_WAIT;
599 /* FALLTHROUGH F_SETLK */
600
601 case F_SETLK:
602 do_setlk:
603 FILEDESC_SLOCK(fdp);
604 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp);
605 if (error != 0) {
606 FILEDESC_SUNLOCK(fdp);
607 break;
608 }
609 if (fp->f_type != DTYPE_VNODE) {
610 FILEDESC_SUNLOCK(fdp);
611 error = EBADF;
612 break;
613 }
614 flp = (struct flock *)arg;
615 if (flp->l_whence == SEEK_CUR) {
616 if (fp->f_offset < 0 ||
617 (flp->l_start > 0 &&
618 fp->f_offset > OFF_MAX - flp->l_start)) {
619 FILEDESC_SUNLOCK(fdp);
620 error = EOVERFLOW;
621 break;
622 }
623 flp->l_start += fp->f_offset;
624 }
625
626 /*
627 * VOP_ADVLOCK() may block.
628 */
629 fhold(fp);
630 FILEDESC_SUNLOCK(fdp);
631 vp = fp->f_vnode;
632 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
633 switch (flp->l_type) {
634 case F_RDLCK:
635 if ((fp->f_flag & FREAD) == 0) {
636 error = EBADF;
637 break;
638 }
639 PROC_LOCK(p->p_leader);
640 p->p_leader->p_flag |= P_ADVLOCK;
641 PROC_UNLOCK(p->p_leader);
642 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
643 flp, flg);
644 break;
645 case F_WRLCK:
646 if ((fp->f_flag & FWRITE) == 0) {
647 error = EBADF;
648 break;
649 }
650 PROC_LOCK(p->p_leader);
651 p->p_leader->p_flag |= P_ADVLOCK;
652 PROC_UNLOCK(p->p_leader);
653 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
654 flp, flg);
655 break;
656 case F_UNLCK:
657 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
658 flp, flg);
659 break;
660 case F_UNLCKSYS:
661 /*
662 * Temporary api for testing remote lock
663 * infrastructure.
664 */
665 if (flg != F_REMOTE) {
666 error = EINVAL;
667 break;
668 }
669 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
670 F_UNLCKSYS, flp, flg);
671 break;
672 default:
673 error = EINVAL;
674 break;
675 }
676 VFS_UNLOCK_GIANT(vfslocked);
677 vfslocked = 0;
678 /* Check for race with close */
679 FILEDESC_SLOCK(fdp);
680 if ((unsigned) fd >= fdp->fd_nfiles ||
681 fp != fdp->fd_ofiles[fd]) {
682 FILEDESC_SUNLOCK(fdp);
683 flp->l_whence = SEEK_SET;
684 flp->l_start = 0;
685 flp->l_len = 0;
686 flp->l_type = F_UNLCK;
687 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
688 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
689 F_UNLCK, flp, F_POSIX);
690 VFS_UNLOCK_GIANT(vfslocked);
691 vfslocked = 0;
692 } else
693 FILEDESC_SUNLOCK(fdp);
694 fdrop(fp, td);
695 break;
696
697 case F_GETLK:
698 FILEDESC_SLOCK(fdp);
699 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp);
700 if (error != 0) {
701 FILEDESC_SUNLOCK(fdp);
702 break;
703 }
704 if (fp->f_type != DTYPE_VNODE) {
705 FILEDESC_SUNLOCK(fdp);
706 error = EBADF;
707 break;
708 }
709 flp = (struct flock *)arg;
710 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
711 flp->l_type != F_UNLCK) {
712 FILEDESC_SUNLOCK(fdp);
713 error = EINVAL;
714 break;
715 }
716 if (flp->l_whence == SEEK_CUR) {
717 if ((flp->l_start > 0 &&
718 fp->f_offset > OFF_MAX - flp->l_start) ||
719 (flp->l_start < 0 &&
720 fp->f_offset < OFF_MIN - flp->l_start)) {
721 FILEDESC_SUNLOCK(fdp);
722 error = EOVERFLOW;
723 break;
724 }
725 flp->l_start += fp->f_offset;
726 }
727 /*
728 * VOP_ADVLOCK() may block.
729 */
730 fhold(fp);
731 FILEDESC_SUNLOCK(fdp);
732 vp = fp->f_vnode;
733 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
734 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
735 F_POSIX);
736 VFS_UNLOCK_GIANT(vfslocked);
737 vfslocked = 0;
738 fdrop(fp, td);
739 break;
740
741 case F_RDAHEAD:
742 arg = arg ? 128 * 1024: 0;
743 /* FALLTHROUGH */
744 case F_READAHEAD:
745 FILEDESC_SLOCK(fdp);
746 if ((fp = fdtofp(fd, fdp)) == NULL) {
747 FILEDESC_SUNLOCK(fdp);
748 error = EBADF;
749 break;
750 }
751 if (fp->f_type != DTYPE_VNODE) {
752 FILEDESC_SUNLOCK(fdp);
753 error = EBADF;
754 break;
755 }
756 fhold(fp);
757 FILEDESC_SUNLOCK(fdp);
758 if (arg != 0) {
759 vp = fp->f_vnode;
760 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
761 error = vn_lock(vp, LK_SHARED);
762 if (error != 0)
763 goto readahead_vnlock_fail;
764 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
765 VOP_UNLOCK(vp, 0);
766 fp->f_seqcount = (arg + bsize - 1) / bsize;
767 do {
768 new = old = fp->f_flag;
769 new |= FRDAHEAD;
770 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
771 readahead_vnlock_fail:
772 VFS_UNLOCK_GIANT(vfslocked);
773 vfslocked = 0;
774 } else {
775 do {
776 new = old = fp->f_flag;
777 new &= ~FRDAHEAD;
778 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
779 }
780 fdrop(fp, td);
781 break;
782
783 default:
784 error = EINVAL;
785 break;
786 }
787 VFS_UNLOCK_GIANT(vfslocked);
788 return (error);
789 }
790
791 /*
792 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
793 */
794 static int
795 do_dup(struct thread *td, int flags, int old, int new,
796 register_t *retval)
797 {
798 struct filedesc *fdp;
799 struct proc *p;
800 struct file *fp;
801 struct file *delfp;
802 int error, holdleaders, maxfd;
803
804 p = td->td_proc;
805 fdp = p->p_fd;
806
807 /*
808 * Verify we have a valid descriptor to dup from and possibly to
809 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
810 * return EINVAL when the new descriptor is out of bounds.
811 */
812 if (old < 0)
813 return (EBADF);
814 if (new < 0)
815 return (flags & DUP_FCNTL ? EINVAL : EBADF);
816 PROC_LOCK(p);
817 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
818 PROC_UNLOCK(p);
819 if (new >= maxfd)
820 return (flags & DUP_FCNTL ? EINVAL : EBADF);
821
822 FILEDESC_XLOCK(fdp);
823 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
824 FILEDESC_XUNLOCK(fdp);
825 return (EBADF);
826 }
827 if (flags & DUP_FIXED && old == new) {
828 *retval = new;
829 FILEDESC_XUNLOCK(fdp);
830 return (0);
831 }
832 fp = fdp->fd_ofiles[old];
833 fhold(fp);
834
835 /*
836 * If the caller specified a file descriptor, make sure the file
837 * table is large enough to hold it, and grab it. Otherwise, just
838 * allocate a new descriptor the usual way. Since the filedesc
839 * lock may be temporarily dropped in the process, we have to look
840 * out for a race.
841 */
842 if (flags & DUP_FIXED) {
843 if (new >= fdp->fd_nfiles) {
844 /*
845 * The resource limits are here instead of e.g.
846 * fdalloc(), because the file descriptor table may be
847 * shared between processes, so we can't really use
848 * racct_add()/racct_sub(). Instead of counting the
849 * number of actually allocated descriptors, just put
850 * the limit on the size of the file descriptor table.
851 */
852 #ifdef RACCT
853 PROC_LOCK(p);
854 error = racct_set(p, RACCT_NOFILE, new + 1);
855 PROC_UNLOCK(p);
856 if (error != 0) {
857 FILEDESC_XUNLOCK(fdp);
858 fdrop(fp, td);
859 return (EMFILE);
860 }
861 #endif
862 fdgrowtable(fdp, new + 1);
863 }
864 if (fdp->fd_ofiles[new] == NULL)
865 fdused(fdp, new);
866 } else {
867 if ((error = fdalloc(td, new, &new)) != 0) {
868 FILEDESC_XUNLOCK(fdp);
869 fdrop(fp, td);
870 return (error);
871 }
872 }
873
874 /*
875 * If the old file changed out from under us then treat it as a
876 * bad file descriptor. Userland should do its own locking to
877 * avoid this case.
878 */
879 if (fdp->fd_ofiles[old] != fp) {
880 /* we've allocated a descriptor which we won't use */
881 if (fdp->fd_ofiles[new] == NULL)
882 fdunused(fdp, new);
883 FILEDESC_XUNLOCK(fdp);
884 fdrop(fp, td);
885 return (EBADF);
886 }
887 KASSERT(old != new,
888 ("new fd is same as old"));
889
890 /*
891 * Save info on the descriptor being overwritten. We cannot close
892 * it without introducing an ownership race for the slot, since we
893 * need to drop the filedesc lock to call closef().
894 *
895 * XXX this duplicates parts of close().
896 */
897 delfp = fdp->fd_ofiles[new];
898 holdleaders = 0;
899 if (delfp != NULL) {
900 if (td->td_proc->p_fdtol != NULL) {
901 /*
902 * Ask fdfree() to sleep to ensure that all relevant
903 * process leaders can be traversed in closef().
904 */
905 fdp->fd_holdleaderscount++;
906 holdleaders = 1;
907 }
908 }
909
910 /*
911 * Duplicate the source descriptor
912 */
913 fdp->fd_ofiles[new] = fp;
914 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
915 if (new > fdp->fd_lastfile)
916 fdp->fd_lastfile = new;
917 *retval = new;
918
919 /*
920 * If we dup'd over a valid file, we now own the reference to it
921 * and must dispose of it using closef() semantics (as if a
922 * close() were performed on it).
923 *
924 * XXX this duplicates parts of close().
925 */
926 if (delfp != NULL) {
927 knote_fdclose(td, new);
928 if (delfp->f_type == DTYPE_MQUEUE)
929 mq_fdclose(td, new, delfp);
930 FILEDESC_XUNLOCK(fdp);
931 (void) closef(delfp, td);
932 if (holdleaders) {
933 FILEDESC_XLOCK(fdp);
934 fdp->fd_holdleaderscount--;
935 if (fdp->fd_holdleaderscount == 0 &&
936 fdp->fd_holdleaderswakeup != 0) {
937 fdp->fd_holdleaderswakeup = 0;
938 wakeup(&fdp->fd_holdleaderscount);
939 }
940 FILEDESC_XUNLOCK(fdp);
941 }
942 } else {
943 FILEDESC_XUNLOCK(fdp);
944 }
945 return (0);
946 }
947
948 /*
949 * If sigio is on the list associated with a process or process group,
950 * disable signalling from the device, remove sigio from the list and
951 * free sigio.
952 */
953 void
954 funsetown(struct sigio **sigiop)
955 {
956 struct sigio *sigio;
957
958 SIGIO_LOCK();
959 sigio = *sigiop;
960 if (sigio == NULL) {
961 SIGIO_UNLOCK();
962 return;
963 }
964 *(sigio->sio_myref) = NULL;
965 if ((sigio)->sio_pgid < 0) {
966 struct pgrp *pg = (sigio)->sio_pgrp;
967 PGRP_LOCK(pg);
968 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
969 sigio, sio_pgsigio);
970 PGRP_UNLOCK(pg);
971 } else {
972 struct proc *p = (sigio)->sio_proc;
973 PROC_LOCK(p);
974 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
975 sigio, sio_pgsigio);
976 PROC_UNLOCK(p);
977 }
978 SIGIO_UNLOCK();
979 crfree(sigio->sio_ucred);
980 free(sigio, M_SIGIO);
981 }
982
983 /*
984 * Free a list of sigio structures.
985 * We only need to lock the SIGIO_LOCK because we have made ourselves
986 * inaccessible to callers of fsetown and therefore do not need to lock
987 * the proc or pgrp struct for the list manipulation.
988 */
989 void
990 funsetownlst(struct sigiolst *sigiolst)
991 {
992 struct proc *p;
993 struct pgrp *pg;
994 struct sigio *sigio;
995
996 sigio = SLIST_FIRST(sigiolst);
997 if (sigio == NULL)
998 return;
999 p = NULL;
1000 pg = NULL;
1001
1002 /*
1003 * Every entry of the list should belong
1004 * to a single proc or pgrp.
1005 */
1006 if (sigio->sio_pgid < 0) {
1007 pg = sigio->sio_pgrp;
1008 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1009 } else /* if (sigio->sio_pgid > 0) */ {
1010 p = sigio->sio_proc;
1011 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1012 }
1013
1014 SIGIO_LOCK();
1015 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1016 *(sigio->sio_myref) = NULL;
1017 if (pg != NULL) {
1018 KASSERT(sigio->sio_pgid < 0,
1019 ("Proc sigio in pgrp sigio list"));
1020 KASSERT(sigio->sio_pgrp == pg,
1021 ("Bogus pgrp in sigio list"));
1022 PGRP_LOCK(pg);
1023 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1024 sio_pgsigio);
1025 PGRP_UNLOCK(pg);
1026 } else /* if (p != NULL) */ {
1027 KASSERT(sigio->sio_pgid > 0,
1028 ("Pgrp sigio in proc sigio list"));
1029 KASSERT(sigio->sio_proc == p,
1030 ("Bogus proc in sigio list"));
1031 PROC_LOCK(p);
1032 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1033 sio_pgsigio);
1034 PROC_UNLOCK(p);
1035 }
1036 SIGIO_UNLOCK();
1037 crfree(sigio->sio_ucred);
1038 free(sigio, M_SIGIO);
1039 SIGIO_LOCK();
1040 }
1041 SIGIO_UNLOCK();
1042 }
1043
1044 /*
1045 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1046 *
1047 * After permission checking, add a sigio structure to the sigio list for
1048 * the process or process group.
1049 */
1050 int
1051 fsetown(pid_t pgid, struct sigio **sigiop)
1052 {
1053 struct proc *proc;
1054 struct pgrp *pgrp;
1055 struct sigio *sigio;
1056 int ret;
1057
1058 if (pgid == 0) {
1059 funsetown(sigiop);
1060 return (0);
1061 }
1062
1063 ret = 0;
1064
1065 /* Allocate and fill in the new sigio out of locks. */
1066 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1067 sigio->sio_pgid = pgid;
1068 sigio->sio_ucred = crhold(curthread->td_ucred);
1069 sigio->sio_myref = sigiop;
1070
1071 sx_slock(&proctree_lock);
1072 if (pgid > 0) {
1073 proc = pfind(pgid);
1074 if (proc == NULL) {
1075 ret = ESRCH;
1076 goto fail;
1077 }
1078
1079 /*
1080 * Policy - Don't allow a process to FSETOWN a process
1081 * in another session.
1082 *
1083 * Remove this test to allow maximum flexibility or
1084 * restrict FSETOWN to the current process or process
1085 * group for maximum safety.
1086 */
1087 PROC_UNLOCK(proc);
1088 if (proc->p_session != curthread->td_proc->p_session) {
1089 ret = EPERM;
1090 goto fail;
1091 }
1092
1093 pgrp = NULL;
1094 } else /* if (pgid < 0) */ {
1095 pgrp = pgfind(-pgid);
1096 if (pgrp == NULL) {
1097 ret = ESRCH;
1098 goto fail;
1099 }
1100 PGRP_UNLOCK(pgrp);
1101
1102 /*
1103 * Policy - Don't allow a process to FSETOWN a process
1104 * in another session.
1105 *
1106 * Remove this test to allow maximum flexibility or
1107 * restrict FSETOWN to the current process or process
1108 * group for maximum safety.
1109 */
1110 if (pgrp->pg_session != curthread->td_proc->p_session) {
1111 ret = EPERM;
1112 goto fail;
1113 }
1114
1115 proc = NULL;
1116 }
1117 funsetown(sigiop);
1118 if (pgid > 0) {
1119 PROC_LOCK(proc);
1120 /*
1121 * Since funsetownlst() is called without the proctree
1122 * locked, we need to check for P_WEXIT.
1123 * XXX: is ESRCH correct?
1124 */
1125 if ((proc->p_flag & P_WEXIT) != 0) {
1126 PROC_UNLOCK(proc);
1127 ret = ESRCH;
1128 goto fail;
1129 }
1130 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1131 sigio->sio_proc = proc;
1132 PROC_UNLOCK(proc);
1133 } else {
1134 PGRP_LOCK(pgrp);
1135 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1136 sigio->sio_pgrp = pgrp;
1137 PGRP_UNLOCK(pgrp);
1138 }
1139 sx_sunlock(&proctree_lock);
1140 SIGIO_LOCK();
1141 *sigiop = sigio;
1142 SIGIO_UNLOCK();
1143 return (0);
1144
1145 fail:
1146 sx_sunlock(&proctree_lock);
1147 crfree(sigio->sio_ucred);
1148 free(sigio, M_SIGIO);
1149 return (ret);
1150 }
1151
1152 /*
1153 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1154 */
1155 pid_t
1156 fgetown(sigiop)
1157 struct sigio **sigiop;
1158 {
1159 pid_t pgid;
1160
1161 SIGIO_LOCK();
1162 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1163 SIGIO_UNLOCK();
1164 return (pgid);
1165 }
1166
1167 /*
1168 * Close a file descriptor.
1169 */
1170 #ifndef _SYS_SYSPROTO_H_
1171 struct close_args {
1172 int fd;
1173 };
1174 #endif
1175 /* ARGSUSED */
1176 int
1177 sys_close(td, uap)
1178 struct thread *td;
1179 struct close_args *uap;
1180 {
1181
1182 return (kern_close(td, uap->fd));
1183 }
1184
1185 int
1186 kern_close(td, fd)
1187 struct thread *td;
1188 int fd;
1189 {
1190 struct filedesc *fdp;
1191 struct file *fp, *fp_object;
1192 int error;
1193 int holdleaders;
1194
1195 error = 0;
1196 holdleaders = 0;
1197 fdp = td->td_proc->p_fd;
1198
1199 AUDIT_SYSCLOSE(td, fd);
1200
1201 FILEDESC_XLOCK(fdp);
1202 if ((unsigned)fd >= fdp->fd_nfiles ||
1203 (fp = fdp->fd_ofiles[fd]) == NULL) {
1204 FILEDESC_XUNLOCK(fdp);
1205 return (EBADF);
1206 }
1207 fdp->fd_ofiles[fd] = NULL;
1208 fdp->fd_ofileflags[fd] = 0;
1209 fdunused(fdp, fd);
1210 if (td->td_proc->p_fdtol != NULL) {
1211 /*
1212 * Ask fdfree() to sleep to ensure that all relevant
1213 * process leaders can be traversed in closef().
1214 */
1215 fdp->fd_holdleaderscount++;
1216 holdleaders = 1;
1217 }
1218
1219 /*
1220 * We now hold the fp reference that used to be owned by the
1221 * descriptor array. We have to unlock the FILEDESC *AFTER*
1222 * knote_fdclose to prevent a race of the fd getting opened, a knote
1223 * added, and deleteing a knote for the new fd.
1224 */
1225 knote_fdclose(td, fd);
1226
1227 /*
1228 * When we're closing an fd with a capability, we need to notify
1229 * mqueue if the underlying object is of type mqueue.
1230 */
1231 (void)cap_funwrap(fp, 0, &fp_object);
1232 if (fp_object->f_type == DTYPE_MQUEUE)
1233 mq_fdclose(td, fd, fp_object);
1234 FILEDESC_XUNLOCK(fdp);
1235
1236 error = closef(fp, td);
1237 if (holdleaders) {
1238 FILEDESC_XLOCK(fdp);
1239 fdp->fd_holdleaderscount--;
1240 if (fdp->fd_holdleaderscount == 0 &&
1241 fdp->fd_holdleaderswakeup != 0) {
1242 fdp->fd_holdleaderswakeup = 0;
1243 wakeup(&fdp->fd_holdleaderscount);
1244 }
1245 FILEDESC_XUNLOCK(fdp);
1246 }
1247 return (error);
1248 }
1249
1250 /*
1251 * Close open file descriptors.
1252 */
1253 #ifndef _SYS_SYSPROTO_H_
1254 struct closefrom_args {
1255 int lowfd;
1256 };
1257 #endif
1258 /* ARGSUSED */
1259 int
1260 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1261 {
1262 struct filedesc *fdp;
1263 int fd;
1264
1265 fdp = td->td_proc->p_fd;
1266 AUDIT_ARG_FD(uap->lowfd);
1267
1268 /*
1269 * Treat negative starting file descriptor values identical to
1270 * closefrom(0) which closes all files.
1271 */
1272 if (uap->lowfd < 0)
1273 uap->lowfd = 0;
1274 FILEDESC_SLOCK(fdp);
1275 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
1276 if (fdp->fd_ofiles[fd] != NULL) {
1277 FILEDESC_SUNLOCK(fdp);
1278 (void)kern_close(td, fd);
1279 FILEDESC_SLOCK(fdp);
1280 }
1281 }
1282 FILEDESC_SUNLOCK(fdp);
1283 return (0);
1284 }
1285
1286 #if defined(COMPAT_43)
1287 /*
1288 * Return status information about a file descriptor.
1289 */
1290 #ifndef _SYS_SYSPROTO_H_
1291 struct ofstat_args {
1292 int fd;
1293 struct ostat *sb;
1294 };
1295 #endif
1296 /* ARGSUSED */
1297 int
1298 ofstat(struct thread *td, struct ofstat_args *uap)
1299 {
1300 struct ostat oub;
1301 struct stat ub;
1302 int error;
1303
1304 error = kern_fstat(td, uap->fd, &ub);
1305 if (error == 0) {
1306 cvtstat(&ub, &oub);
1307 error = copyout(&oub, uap->sb, sizeof(oub));
1308 }
1309 return (error);
1310 }
1311 #endif /* COMPAT_43 */
1312
1313 /*
1314 * Return status information about a file descriptor.
1315 */
1316 #ifndef _SYS_SYSPROTO_H_
1317 struct fstat_args {
1318 int fd;
1319 struct stat *sb;
1320 };
1321 #endif
1322 /* ARGSUSED */
1323 int
1324 sys_fstat(struct thread *td, struct fstat_args *uap)
1325 {
1326 struct stat ub;
1327 int error;
1328
1329 error = kern_fstat(td, uap->fd, &ub);
1330 if (error == 0)
1331 error = copyout(&ub, uap->sb, sizeof(ub));
1332 return (error);
1333 }
1334
1335 int
1336 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1337 {
1338 struct file *fp;
1339 int error;
1340
1341 AUDIT_ARG_FD(fd);
1342
1343 if ((error = fget(td, fd, CAP_FSTAT, &fp)) != 0)
1344 return (error);
1345
1346 AUDIT_ARG_FILE(td->td_proc, fp);
1347
1348 error = fo_stat(fp, sbp, td->td_ucred, td);
1349 fdrop(fp, td);
1350 #ifdef KTRACE
1351 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1352 ktrstat(sbp);
1353 #endif
1354 return (error);
1355 }
1356
1357 /*
1358 * Return status information about a file descriptor.
1359 */
1360 #ifndef _SYS_SYSPROTO_H_
1361 struct nfstat_args {
1362 int fd;
1363 struct nstat *sb;
1364 };
1365 #endif
1366 /* ARGSUSED */
1367 int
1368 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1369 {
1370 struct nstat nub;
1371 struct stat ub;
1372 int error;
1373
1374 error = kern_fstat(td, uap->fd, &ub);
1375 if (error == 0) {
1376 cvtnstat(&ub, &nub);
1377 error = copyout(&nub, uap->sb, sizeof(nub));
1378 }
1379 return (error);
1380 }
1381
1382 /*
1383 * Return pathconf information about a file descriptor.
1384 */
1385 #ifndef _SYS_SYSPROTO_H_
1386 struct fpathconf_args {
1387 int fd;
1388 int name;
1389 };
1390 #endif
1391 /* ARGSUSED */
1392 int
1393 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1394 {
1395 struct file *fp;
1396 struct vnode *vp;
1397 int error;
1398
1399 if ((error = fget(td, uap->fd, CAP_FPATHCONF, &fp)) != 0)
1400 return (error);
1401
1402 /* If asynchronous I/O is available, it works for all descriptors. */
1403 if (uap->name == _PC_ASYNC_IO) {
1404 td->td_retval[0] = async_io_version;
1405 goto out;
1406 }
1407 vp = fp->f_vnode;
1408 if (vp != NULL) {
1409 int vfslocked;
1410 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1411 vn_lock(vp, LK_SHARED | LK_RETRY);
1412 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1413 VOP_UNLOCK(vp, 0);
1414 VFS_UNLOCK_GIANT(vfslocked);
1415 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1416 if (uap->name != _PC_PIPE_BUF) {
1417 error = EINVAL;
1418 } else {
1419 td->td_retval[0] = PIPE_BUF;
1420 error = 0;
1421 }
1422 } else {
1423 error = EOPNOTSUPP;
1424 }
1425 out:
1426 fdrop(fp, td);
1427 return (error);
1428 }
1429
1430 /*
1431 * Grow the file table to accomodate (at least) nfd descriptors. This may
1432 * block and drop the filedesc lock, but it will reacquire it before
1433 * returning.
1434 */
1435 static void
1436 fdgrowtable(struct filedesc *fdp, int nfd)
1437 {
1438 struct filedesc0 *fdp0;
1439 struct freetable *fo;
1440 struct file **ntable;
1441 struct file **otable;
1442 char *nfileflags;
1443 int nnfiles, onfiles;
1444 NDSLOTTYPE *nmap;
1445
1446 FILEDESC_XLOCK_ASSERT(fdp);
1447
1448 KASSERT(fdp->fd_nfiles > 0,
1449 ("zero-length file table"));
1450
1451 /* compute the size of the new table */
1452 onfiles = fdp->fd_nfiles;
1453 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1454 if (nnfiles <= onfiles)
1455 /* the table is already large enough */
1456 return;
1457
1458 /* allocate a new table and (if required) new bitmaps */
1459 FILEDESC_XUNLOCK(fdp);
1460 ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable),
1461 M_FILEDESC, M_ZERO | M_WAITOK);
1462 nfileflags = (char *)&ntable[nnfiles];
1463 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1464 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE,
1465 M_FILEDESC, M_ZERO | M_WAITOK);
1466 else
1467 nmap = NULL;
1468 FILEDESC_XLOCK(fdp);
1469
1470 /*
1471 * We now have new tables ready to go. Since we dropped the
1472 * filedesc lock to call malloc(), watch out for a race.
1473 */
1474 onfiles = fdp->fd_nfiles;
1475 if (onfiles >= nnfiles) {
1476 /* we lost the race, but that's OK */
1477 free(ntable, M_FILEDESC);
1478 if (nmap != NULL)
1479 free(nmap, M_FILEDESC);
1480 return;
1481 }
1482 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1483 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1484 otable = fdp->fd_ofiles;
1485 fdp->fd_ofileflags = nfileflags;
1486 fdp->fd_ofiles = ntable;
1487 /*
1488 * We must preserve ofiles until the process exits because we can't
1489 * be certain that no threads have references to the old table via
1490 * _fget().
1491 */
1492 if (onfiles > NDFILE) {
1493 fo = (struct freetable *)&otable[onfiles];
1494 fdp0 = (struct filedesc0 *)fdp;
1495 fo->ft_table = otable;
1496 SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next);
1497 }
1498 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1499 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1500 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1501 free(fdp->fd_map, M_FILEDESC);
1502 fdp->fd_map = nmap;
1503 }
1504 fdp->fd_nfiles = nnfiles;
1505 }
1506
1507 /*
1508 * Allocate a file descriptor for the process.
1509 */
1510 int
1511 fdalloc(struct thread *td, int minfd, int *result)
1512 {
1513 struct proc *p = td->td_proc;
1514 struct filedesc *fdp = p->p_fd;
1515 int fd = -1, maxfd;
1516 #ifdef RACCT
1517 int error;
1518 #endif
1519
1520 FILEDESC_XLOCK_ASSERT(fdp);
1521
1522 if (fdp->fd_freefile > minfd)
1523 minfd = fdp->fd_freefile;
1524
1525 PROC_LOCK(p);
1526 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1527 PROC_UNLOCK(p);
1528
1529 /*
1530 * Search the bitmap for a free descriptor. If none is found, try
1531 * to grow the file table. Keep at it until we either get a file
1532 * descriptor or run into process or system limits; fdgrowtable()
1533 * may drop the filedesc lock, so we're in a race.
1534 */
1535 for (;;) {
1536 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1537 if (fd >= maxfd)
1538 return (EMFILE);
1539 if (fd < fdp->fd_nfiles)
1540 break;
1541 #ifdef RACCT
1542 PROC_LOCK(p);
1543 error = racct_set(p, RACCT_NOFILE, min(fdp->fd_nfiles * 2, maxfd));
1544 PROC_UNLOCK(p);
1545 if (error != 0)
1546 return (EMFILE);
1547 #endif
1548 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1549 }
1550
1551 /*
1552 * Perform some sanity checks, then mark the file descriptor as
1553 * used and return it to the caller.
1554 */
1555 KASSERT(!fdisused(fdp, fd),
1556 ("fd_first_free() returned non-free descriptor"));
1557 KASSERT(fdp->fd_ofiles[fd] == NULL,
1558 ("free descriptor isn't"));
1559 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1560 fdused(fdp, fd);
1561 *result = fd;
1562 return (0);
1563 }
1564
1565 /*
1566 * Check to see whether n user file descriptors are available to the process
1567 * p.
1568 */
1569 int
1570 fdavail(struct thread *td, int n)
1571 {
1572 struct proc *p = td->td_proc;
1573 struct filedesc *fdp = td->td_proc->p_fd;
1574 struct file **fpp;
1575 int i, lim, last;
1576
1577 FILEDESC_LOCK_ASSERT(fdp);
1578
1579 /*
1580 * XXX: This is only called from uipc_usrreq.c:unp_externalize();
1581 * call racct_add() from there instead of dealing with containers
1582 * here.
1583 */
1584 PROC_LOCK(p);
1585 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1586 PROC_UNLOCK(p);
1587 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1588 return (1);
1589 last = min(fdp->fd_nfiles, lim);
1590 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1591 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1592 if (*fpp == NULL && --n <= 0)
1593 return (1);
1594 }
1595 return (0);
1596 }
1597
1598 /*
1599 * Create a new open file structure and allocate a file decriptor for the
1600 * process that refers to it. We add one reference to the file for the
1601 * descriptor table and one reference for resultfp. This is to prevent us
1602 * being preempted and the entry in the descriptor table closed after we
1603 * release the FILEDESC lock.
1604 */
1605 int
1606 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1607 {
1608 struct file *fp;
1609 int error, fd;
1610
1611 error = falloc_noinstall(td, &fp);
1612 if (error)
1613 return (error); /* no reference held on error */
1614
1615 error = finstall(td, fp, &fd, flags);
1616 if (error) {
1617 fdrop(fp, td); /* one reference (fp only) */
1618 return (error);
1619 }
1620
1621 if (resultfp != NULL)
1622 *resultfp = fp; /* copy out result */
1623 else
1624 fdrop(fp, td); /* release local reference */
1625
1626 if (resultfd != NULL)
1627 *resultfd = fd;
1628
1629 return (0);
1630 }
1631
1632 /*
1633 * Create a new open file structure without allocating a file descriptor.
1634 */
1635 int
1636 falloc_noinstall(struct thread *td, struct file **resultfp)
1637 {
1638 struct file *fp;
1639 int maxuserfiles = maxfiles - (maxfiles / 20);
1640 static struct timeval lastfail;
1641 static int curfail;
1642
1643 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1644
1645 if ((openfiles >= maxuserfiles &&
1646 priv_check(td, PRIV_MAXFILES) != 0) ||
1647 openfiles >= maxfiles) {
1648 if (ppsratecheck(&lastfail, &curfail, 1)) {
1649 printf("kern.maxfiles limit exceeded by uid %i, "
1650 "please see tuning(7).\n", td->td_ucred->cr_ruid);
1651 }
1652 return (ENFILE);
1653 }
1654 atomic_add_int(&openfiles, 1);
1655 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1656 refcount_init(&fp->f_count, 1);
1657 fp->f_cred = crhold(td->td_ucred);
1658 fp->f_ops = &badfileops;
1659 fp->f_data = NULL;
1660 fp->f_vnode = NULL;
1661 *resultfp = fp;
1662 return (0);
1663 }
1664
1665 /*
1666 * Install a file in a file descriptor table.
1667 */
1668 int
1669 finstall(struct thread *td, struct file *fp, int *fd, int flags)
1670 {
1671 struct filedesc *fdp = td->td_proc->p_fd;
1672 int error;
1673
1674 KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1675 KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1676
1677 FILEDESC_XLOCK(fdp);
1678 if ((error = fdalloc(td, 0, fd))) {
1679 FILEDESC_XUNLOCK(fdp);
1680 return (error);
1681 }
1682 fhold(fp);
1683 fdp->fd_ofiles[*fd] = fp;
1684 if ((flags & O_CLOEXEC) != 0)
1685 fdp->fd_ofileflags[*fd] |= UF_EXCLOSE;
1686 FILEDESC_XUNLOCK(fdp);
1687 return (0);
1688 }
1689
1690 /*
1691 * Build a new filedesc structure from another.
1692 * Copy the current, root, and jail root vnode references.
1693 */
1694 struct filedesc *
1695 fdinit(struct filedesc *fdp)
1696 {
1697 struct filedesc0 *newfdp;
1698
1699 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1700 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1701 if (fdp != NULL) {
1702 FILEDESC_XLOCK(fdp);
1703 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1704 if (newfdp->fd_fd.fd_cdir)
1705 VREF(newfdp->fd_fd.fd_cdir);
1706 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1707 if (newfdp->fd_fd.fd_rdir)
1708 VREF(newfdp->fd_fd.fd_rdir);
1709 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1710 if (newfdp->fd_fd.fd_jdir)
1711 VREF(newfdp->fd_fd.fd_jdir);
1712 FILEDESC_XUNLOCK(fdp);
1713 }
1714
1715 /* Create the file descriptor table. */
1716 newfdp->fd_fd.fd_refcnt = 1;
1717 newfdp->fd_fd.fd_holdcnt = 1;
1718 newfdp->fd_fd.fd_cmask = CMASK;
1719 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1720 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1721 newfdp->fd_fd.fd_nfiles = NDFILE;
1722 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1723 newfdp->fd_fd.fd_lastfile = -1;
1724 return (&newfdp->fd_fd);
1725 }
1726
1727 static struct filedesc *
1728 fdhold(struct proc *p)
1729 {
1730 struct filedesc *fdp;
1731
1732 mtx_lock(&fdesc_mtx);
1733 fdp = p->p_fd;
1734 if (fdp != NULL)
1735 fdp->fd_holdcnt++;
1736 mtx_unlock(&fdesc_mtx);
1737 return (fdp);
1738 }
1739
1740 static void
1741 fddrop(struct filedesc *fdp)
1742 {
1743 struct filedesc0 *fdp0;
1744 struct freetable *ft;
1745 int i;
1746
1747 mtx_lock(&fdesc_mtx);
1748 i = --fdp->fd_holdcnt;
1749 mtx_unlock(&fdesc_mtx);
1750 if (i > 0)
1751 return;
1752
1753 FILEDESC_LOCK_DESTROY(fdp);
1754 fdp0 = (struct filedesc0 *)fdp;
1755 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1756 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1757 free(ft->ft_table, M_FILEDESC);
1758 }
1759 free(fdp, M_FILEDESC);
1760 }
1761
1762 /*
1763 * Share a filedesc structure.
1764 */
1765 struct filedesc *
1766 fdshare(struct filedesc *fdp)
1767 {
1768
1769 FILEDESC_XLOCK(fdp);
1770 fdp->fd_refcnt++;
1771 FILEDESC_XUNLOCK(fdp);
1772 return (fdp);
1773 }
1774
1775 /*
1776 * Unshare a filedesc structure, if necessary by making a copy
1777 */
1778 void
1779 fdunshare(struct proc *p, struct thread *td)
1780 {
1781
1782 FILEDESC_XLOCK(p->p_fd);
1783 if (p->p_fd->fd_refcnt > 1) {
1784 struct filedesc *tmp;
1785
1786 FILEDESC_XUNLOCK(p->p_fd);
1787 tmp = fdcopy(p->p_fd);
1788 fdfree(td);
1789 p->p_fd = tmp;
1790 } else
1791 FILEDESC_XUNLOCK(p->p_fd);
1792 }
1793
1794 /*
1795 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1796 * this is to ease callers, not catch errors.
1797 */
1798 struct filedesc *
1799 fdcopy(struct filedesc *fdp)
1800 {
1801 struct filedesc *newfdp;
1802 int i;
1803
1804 /* Certain daemons might not have file descriptors. */
1805 if (fdp == NULL)
1806 return (NULL);
1807
1808 newfdp = fdinit(fdp);
1809 FILEDESC_SLOCK(fdp);
1810 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1811 FILEDESC_SUNLOCK(fdp);
1812 FILEDESC_XLOCK(newfdp);
1813 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1814 FILEDESC_XUNLOCK(newfdp);
1815 FILEDESC_SLOCK(fdp);
1816 }
1817 /* copy all passable descriptors (i.e. not kqueue) */
1818 newfdp->fd_freefile = -1;
1819 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1820 if (fdisused(fdp, i) &&
1821 (fdp->fd_ofiles[i]->f_ops->fo_flags & DFLAG_PASSABLE) &&
1822 fdp->fd_ofiles[i]->f_ops != &badfileops) {
1823 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1824 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1825 fhold(newfdp->fd_ofiles[i]);
1826 newfdp->fd_lastfile = i;
1827 } else {
1828 if (newfdp->fd_freefile == -1)
1829 newfdp->fd_freefile = i;
1830 }
1831 }
1832 newfdp->fd_cmask = fdp->fd_cmask;
1833 FILEDESC_SUNLOCK(fdp);
1834 FILEDESC_XLOCK(newfdp);
1835 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1836 if (newfdp->fd_ofiles[i] != NULL)
1837 fdused(newfdp, i);
1838 if (newfdp->fd_freefile == -1)
1839 newfdp->fd_freefile = i;
1840 FILEDESC_XUNLOCK(newfdp);
1841 return (newfdp);
1842 }
1843
1844 /*
1845 * Release a filedesc structure.
1846 */
1847 void
1848 fdfree(struct thread *td)
1849 {
1850 struct filedesc *fdp;
1851 struct file **fpp;
1852 int i, locked;
1853 struct filedesc_to_leader *fdtol;
1854 struct file *fp;
1855 struct vnode *cdir, *jdir, *rdir, *vp;
1856 struct flock lf;
1857
1858 /* Certain daemons might not have file descriptors. */
1859 fdp = td->td_proc->p_fd;
1860 if (fdp == NULL)
1861 return;
1862
1863 #ifdef RACCT
1864 PROC_LOCK(td->td_proc);
1865 racct_set(td->td_proc, RACCT_NOFILE, 0);
1866 PROC_UNLOCK(td->td_proc);
1867 #endif
1868
1869 /* Check for special need to clear POSIX style locks */
1870 fdtol = td->td_proc->p_fdtol;
1871 if (fdtol != NULL) {
1872 FILEDESC_XLOCK(fdp);
1873 KASSERT(fdtol->fdl_refcount > 0,
1874 ("filedesc_to_refcount botch: fdl_refcount=%d",
1875 fdtol->fdl_refcount));
1876 if (fdtol->fdl_refcount == 1 &&
1877 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1878 for (i = 0, fpp = fdp->fd_ofiles;
1879 i <= fdp->fd_lastfile;
1880 i++, fpp++) {
1881 if (*fpp == NULL ||
1882 (*fpp)->f_type != DTYPE_VNODE)
1883 continue;
1884 fp = *fpp;
1885 fhold(fp);
1886 FILEDESC_XUNLOCK(fdp);
1887 lf.l_whence = SEEK_SET;
1888 lf.l_start = 0;
1889 lf.l_len = 0;
1890 lf.l_type = F_UNLCK;
1891 vp = fp->f_vnode;
1892 locked = VFS_LOCK_GIANT(vp->v_mount);
1893 (void) VOP_ADVLOCK(vp,
1894 (caddr_t)td->td_proc->
1895 p_leader,
1896 F_UNLCK,
1897 &lf,
1898 F_POSIX);
1899 VFS_UNLOCK_GIANT(locked);
1900 FILEDESC_XLOCK(fdp);
1901 fdrop(fp, td);
1902 fpp = fdp->fd_ofiles + i;
1903 }
1904 }
1905 retry:
1906 if (fdtol->fdl_refcount == 1) {
1907 if (fdp->fd_holdleaderscount > 0 &&
1908 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1909 /*
1910 * close() or do_dup() has cleared a reference
1911 * in a shared file descriptor table.
1912 */
1913 fdp->fd_holdleaderswakeup = 1;
1914 sx_sleep(&fdp->fd_holdleaderscount,
1915 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1916 goto retry;
1917 }
1918 if (fdtol->fdl_holdcount > 0) {
1919 /*
1920 * Ensure that fdtol->fdl_leader remains
1921 * valid in closef().
1922 */
1923 fdtol->fdl_wakeup = 1;
1924 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1925 "fdlhold", 0);
1926 goto retry;
1927 }
1928 }
1929 fdtol->fdl_refcount--;
1930 if (fdtol->fdl_refcount == 0 &&
1931 fdtol->fdl_holdcount == 0) {
1932 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1933 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1934 } else
1935 fdtol = NULL;
1936 td->td_proc->p_fdtol = NULL;
1937 FILEDESC_XUNLOCK(fdp);
1938 if (fdtol != NULL)
1939 free(fdtol, M_FILEDESC_TO_LEADER);
1940 }
1941 FILEDESC_XLOCK(fdp);
1942 i = --fdp->fd_refcnt;
1943 FILEDESC_XUNLOCK(fdp);
1944 if (i > 0)
1945 return;
1946
1947 fpp = fdp->fd_ofiles;
1948 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1949 if (*fpp) {
1950 FILEDESC_XLOCK(fdp);
1951 fp = *fpp;
1952 *fpp = NULL;
1953 FILEDESC_XUNLOCK(fdp);
1954 (void) closef(fp, td);
1955 }
1956 }
1957 FILEDESC_XLOCK(fdp);
1958
1959 /* XXX This should happen earlier. */
1960 mtx_lock(&fdesc_mtx);
1961 td->td_proc->p_fd = NULL;
1962 mtx_unlock(&fdesc_mtx);
1963
1964 if (fdp->fd_nfiles > NDFILE)
1965 free(fdp->fd_ofiles, M_FILEDESC);
1966 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1967 free(fdp->fd_map, M_FILEDESC);
1968
1969 fdp->fd_nfiles = 0;
1970
1971 cdir = fdp->fd_cdir;
1972 fdp->fd_cdir = NULL;
1973 rdir = fdp->fd_rdir;
1974 fdp->fd_rdir = NULL;
1975 jdir = fdp->fd_jdir;
1976 fdp->fd_jdir = NULL;
1977 FILEDESC_XUNLOCK(fdp);
1978
1979 if (cdir) {
1980 locked = VFS_LOCK_GIANT(cdir->v_mount);
1981 vrele(cdir);
1982 VFS_UNLOCK_GIANT(locked);
1983 }
1984 if (rdir) {
1985 locked = VFS_LOCK_GIANT(rdir->v_mount);
1986 vrele(rdir);
1987 VFS_UNLOCK_GIANT(locked);
1988 }
1989 if (jdir) {
1990 locked = VFS_LOCK_GIANT(jdir->v_mount);
1991 vrele(jdir);
1992 VFS_UNLOCK_GIANT(locked);
1993 }
1994
1995 fddrop(fdp);
1996 }
1997
1998 /*
1999 * For setugid programs, we don't want to people to use that setugidness
2000 * to generate error messages which write to a file which otherwise would
2001 * otherwise be off-limits to the process. We check for filesystems where
2002 * the vnode can change out from under us after execve (like [lin]procfs).
2003 *
2004 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2005 * sufficient. We also don't check for setugidness since we know we are.
2006 */
2007 static int
2008 is_unsafe(struct file *fp)
2009 {
2010 if (fp->f_type == DTYPE_VNODE) {
2011 struct vnode *vp = fp->f_vnode;
2012
2013 if ((vp->v_vflag & VV_PROCDEP) != 0)
2014 return (1);
2015 }
2016 return (0);
2017 }
2018
2019 /*
2020 * Make this setguid thing safe, if at all possible.
2021 */
2022 void
2023 setugidsafety(struct thread *td)
2024 {
2025 struct filedesc *fdp;
2026 int i;
2027
2028 /* Certain daemons might not have file descriptors. */
2029 fdp = td->td_proc->p_fd;
2030 if (fdp == NULL)
2031 return;
2032
2033 /*
2034 * Note: fdp->fd_ofiles may be reallocated out from under us while
2035 * we are blocked in a close. Be careful!
2036 */
2037 FILEDESC_XLOCK(fdp);
2038 for (i = 0; i <= fdp->fd_lastfile; i++) {
2039 if (i > 2)
2040 break;
2041 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
2042 struct file *fp;
2043
2044 knote_fdclose(td, i);
2045 /*
2046 * NULL-out descriptor prior to close to avoid
2047 * a race while close blocks.
2048 */
2049 fp = fdp->fd_ofiles[i];
2050 fdp->fd_ofiles[i] = NULL;
2051 fdp->fd_ofileflags[i] = 0;
2052 fdunused(fdp, i);
2053 FILEDESC_XUNLOCK(fdp);
2054 (void) closef(fp, td);
2055 FILEDESC_XLOCK(fdp);
2056 }
2057 }
2058 FILEDESC_XUNLOCK(fdp);
2059 }
2060
2061 /*
2062 * If a specific file object occupies a specific file descriptor, close the
2063 * file descriptor entry and drop a reference on the file object. This is a
2064 * convenience function to handle a subsequent error in a function that calls
2065 * falloc() that handles the race that another thread might have closed the
2066 * file descriptor out from under the thread creating the file object.
2067 */
2068 void
2069 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2070 {
2071
2072 FILEDESC_XLOCK(fdp);
2073 if (fdp->fd_ofiles[idx] == fp) {
2074 fdp->fd_ofiles[idx] = NULL;
2075 fdunused(fdp, idx);
2076 FILEDESC_XUNLOCK(fdp);
2077 fdrop(fp, td);
2078 } else
2079 FILEDESC_XUNLOCK(fdp);
2080 }
2081
2082 /*
2083 * Close any files on exec?
2084 */
2085 void
2086 fdcloseexec(struct thread *td)
2087 {
2088 struct filedesc *fdp;
2089 int i;
2090
2091 /* Certain daemons might not have file descriptors. */
2092 fdp = td->td_proc->p_fd;
2093 if (fdp == NULL)
2094 return;
2095
2096 FILEDESC_XLOCK(fdp);
2097
2098 /*
2099 * We cannot cache fd_ofiles or fd_ofileflags since operations
2100 * may block and rip them out from under us.
2101 */
2102 for (i = 0; i <= fdp->fd_lastfile; i++) {
2103 if (fdp->fd_ofiles[i] != NULL &&
2104 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
2105 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
2106 struct file *fp;
2107
2108 knote_fdclose(td, i);
2109 /*
2110 * NULL-out descriptor prior to close to avoid
2111 * a race while close blocks.
2112 */
2113 fp = fdp->fd_ofiles[i];
2114 fdp->fd_ofiles[i] = NULL;
2115 fdp->fd_ofileflags[i] = 0;
2116 fdunused(fdp, i);
2117 if (fp->f_type == DTYPE_MQUEUE)
2118 mq_fdclose(td, i, fp);
2119 FILEDESC_XUNLOCK(fdp);
2120 (void) closef(fp, td);
2121 FILEDESC_XLOCK(fdp);
2122 }
2123 }
2124 FILEDESC_XUNLOCK(fdp);
2125 }
2126
2127 /*
2128 * It is unsafe for set[ug]id processes to be started with file
2129 * descriptors 0..2 closed, as these descriptors are given implicit
2130 * significance in the Standard C library. fdcheckstd() will create a
2131 * descriptor referencing /dev/null for each of stdin, stdout, and
2132 * stderr that is not already open.
2133 */
2134 int
2135 fdcheckstd(struct thread *td)
2136 {
2137 struct filedesc *fdp;
2138 register_t retval, save;
2139 int i, error, devnull;
2140
2141 fdp = td->td_proc->p_fd;
2142 if (fdp == NULL)
2143 return (0);
2144 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2145 devnull = -1;
2146 error = 0;
2147 for (i = 0; i < 3; i++) {
2148 if (fdp->fd_ofiles[i] != NULL)
2149 continue;
2150 if (devnull < 0) {
2151 save = td->td_retval[0];
2152 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2153 O_RDWR, 0);
2154 devnull = td->td_retval[0];
2155 td->td_retval[0] = save;
2156 if (error)
2157 break;
2158 KASSERT(devnull == i, ("oof, we didn't get our fd"));
2159 } else {
2160 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2161 if (error != 0)
2162 break;
2163 }
2164 }
2165 return (error);
2166 }
2167
2168 /*
2169 * Internal form of close. Decrement reference count on file structure.
2170 * Note: td may be NULL when closing a file that was being passed in a
2171 * message.
2172 *
2173 * XXXRW: Giant is not required for the caller, but often will be held; this
2174 * makes it moderately likely the Giant will be recursed in the VFS case.
2175 */
2176 int
2177 closef(struct file *fp, struct thread *td)
2178 {
2179 struct vnode *vp;
2180 struct flock lf;
2181 struct filedesc_to_leader *fdtol;
2182 struct filedesc *fdp;
2183 struct file *fp_object;
2184
2185 /*
2186 * POSIX record locking dictates that any close releases ALL
2187 * locks owned by this process. This is handled by setting
2188 * a flag in the unlock to free ONLY locks obeying POSIX
2189 * semantics, and not to free BSD-style file locks.
2190 * If the descriptor was in a message, POSIX-style locks
2191 * aren't passed with the descriptor, and the thread pointer
2192 * will be NULL. Callers should be careful only to pass a
2193 * NULL thread pointer when there really is no owning
2194 * context that might have locks, or the locks will be
2195 * leaked.
2196 *
2197 * If this is a capability, we do lock processing under the underlying
2198 * node, not the capability itself.
2199 */
2200 (void)cap_funwrap(fp, 0, &fp_object);
2201 if ((fp_object->f_type == DTYPE_VNODE) && (td != NULL)) {
2202 int vfslocked;
2203
2204 vp = fp_object->f_vnode;
2205 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2206 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2207 lf.l_whence = SEEK_SET;
2208 lf.l_start = 0;
2209 lf.l_len = 0;
2210 lf.l_type = F_UNLCK;
2211 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2212 F_UNLCK, &lf, F_POSIX);
2213 }
2214 fdtol = td->td_proc->p_fdtol;
2215 if (fdtol != NULL) {
2216 /*
2217 * Handle special case where file descriptor table is
2218 * shared between multiple process leaders.
2219 */
2220 fdp = td->td_proc->p_fd;
2221 FILEDESC_XLOCK(fdp);
2222 for (fdtol = fdtol->fdl_next;
2223 fdtol != td->td_proc->p_fdtol;
2224 fdtol = fdtol->fdl_next) {
2225 if ((fdtol->fdl_leader->p_flag &
2226 P_ADVLOCK) == 0)
2227 continue;
2228 fdtol->fdl_holdcount++;
2229 FILEDESC_XUNLOCK(fdp);
2230 lf.l_whence = SEEK_SET;
2231 lf.l_start = 0;
2232 lf.l_len = 0;
2233 lf.l_type = F_UNLCK;
2234 vp = fp_object->f_vnode;
2235 (void) VOP_ADVLOCK(vp,
2236 (caddr_t)fdtol->fdl_leader,
2237 F_UNLCK, &lf, F_POSIX);
2238 FILEDESC_XLOCK(fdp);
2239 fdtol->fdl_holdcount--;
2240 if (fdtol->fdl_holdcount == 0 &&
2241 fdtol->fdl_wakeup != 0) {
2242 fdtol->fdl_wakeup = 0;
2243 wakeup(fdtol);
2244 }
2245 }
2246 FILEDESC_XUNLOCK(fdp);
2247 }
2248 VFS_UNLOCK_GIANT(vfslocked);
2249 }
2250 return (fdrop(fp, td));
2251 }
2252
2253 /*
2254 * Initialize the file pointer with the specified properties.
2255 *
2256 * The ops are set with release semantics to be certain that the flags, type,
2257 * and data are visible when ops is. This is to prevent ops methods from being
2258 * called with bad data.
2259 */
2260 void
2261 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2262 {
2263 fp->f_data = data;
2264 fp->f_flag = flag;
2265 fp->f_type = type;
2266 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2267 }
2268
2269 struct file *
2270 fget_unlocked(struct filedesc *fdp, int fd)
2271 {
2272 struct file *fp;
2273 u_int count;
2274
2275 if (fd < 0 || fd >= fdp->fd_nfiles)
2276 return (NULL);
2277 /*
2278 * Fetch the descriptor locklessly. We avoid fdrop() races by
2279 * never raising a refcount above 0. To accomplish this we have
2280 * to use a cmpset loop rather than an atomic_add. The descriptor
2281 * must be re-verified once we acquire a reference to be certain
2282 * that the identity is still correct and we did not lose a race
2283 * due to preemption.
2284 */
2285 for (;;) {
2286 fp = fdp->fd_ofiles[fd];
2287 if (fp == NULL)
2288 break;
2289 count = fp->f_count;
2290 if (count == 0)
2291 continue;
2292 /*
2293 * Use an acquire barrier to prevent caching of fd_ofiles
2294 * so it is refreshed for verification.
2295 */
2296 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2297 continue;
2298 if (fp == fdp->fd_ofiles[fd])
2299 break;
2300 fdrop(fp, curthread);
2301 }
2302
2303 return (fp);
2304 }
2305
2306 /*
2307 * Extract the file pointer associated with the specified descriptor for the
2308 * current user process.
2309 *
2310 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2311 * returned.
2312 *
2313 * If the FGET_GETCAP flag is set, the capability itself will be returned.
2314 * Calling _fget() with FGET_GETCAP on a non-capability will return EINVAL.
2315 * Otherwise, if the file is a capability, its rights will be checked against
2316 * the capability rights mask, and if successful, the object will be unwrapped.
2317 *
2318 * If an error occured the non-zero error is returned and *fpp is set to
2319 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2320 * responsible for fdrop().
2321 */
2322 #define FGET_GETCAP 0x00000001
2323 static __inline int
2324 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2325 cap_rights_t needrights, cap_rights_t *haverightsp, u_char *maxprotp,
2326 int fget_flags)
2327 {
2328 struct filedesc *fdp;
2329 struct file *fp;
2330 #ifdef CAPABILITIES
2331 struct file *fp_fromcap;
2332 #endif
2333 int error;
2334
2335 *fpp = NULL;
2336 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2337 return (EBADF);
2338 if ((fp = fget_unlocked(fdp, fd)) == NULL)
2339 return (EBADF);
2340 if (fp->f_ops == &badfileops) {
2341 fdrop(fp, td);
2342 return (EBADF);
2343 }
2344
2345 #ifdef CAPABILITIES
2346 /*
2347 * If this is a capability, what rights does it have?
2348 */
2349 if (haverightsp != NULL) {
2350 if (fp->f_type == DTYPE_CAPABILITY)
2351 *haverightsp = cap_rights(fp);
2352 else
2353 *haverightsp = CAP_MASK_VALID;
2354 }
2355
2356 /*
2357 * If a capability has been requested, return the capability directly.
2358 * Otherwise, check capability rights, extract the underlying object,
2359 * and check its access flags.
2360 */
2361 if (fget_flags & FGET_GETCAP) {
2362 if (fp->f_type != DTYPE_CAPABILITY) {
2363 fdrop(fp, td);
2364 return (EINVAL);
2365 }
2366 } else {
2367 if (maxprotp == NULL)
2368 error = cap_funwrap(fp, needrights, &fp_fromcap);
2369 else
2370 error = cap_funwrap_mmap(fp, needrights, maxprotp,
2371 &fp_fromcap);
2372 if (error != 0) {
2373 fdrop(fp, td);
2374 return (error);
2375 }
2376
2377 /*
2378 * If we've unwrapped a file, drop the original capability
2379 * and hold the new descriptor. fp after this point refers to
2380 * the actual (unwrapped) object, not the capability.
2381 */
2382 if (fp != fp_fromcap) {
2383 fhold(fp_fromcap);
2384 fdrop(fp, td);
2385 fp = fp_fromcap;
2386 }
2387 }
2388 #else /* !CAPABILITIES */
2389 KASSERT(fp->f_type != DTYPE_CAPABILITY,
2390 ("%s: saw capability", __func__));
2391 if (maxprotp != NULL)
2392 *maxprotp = VM_PROT_ALL;
2393 #endif /* CAPABILITIES */
2394
2395 /*
2396 * FREAD and FWRITE failure return EBADF as per POSIX.
2397 */
2398 error = 0;
2399 switch (flags) {
2400 case FREAD:
2401 case FWRITE:
2402 if ((fp->f_flag & flags) == 0)
2403 error = EBADF;
2404 break;
2405 case FEXEC:
2406 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2407 ((fp->f_flag & FWRITE) != 0))
2408 error = EBADF;
2409 break;
2410 case 0:
2411 break;
2412 default:
2413 KASSERT(0, ("wrong flags"));
2414 }
2415
2416 if (error != 0) {
2417 fdrop(fp, td);
2418 return (error);
2419 }
2420
2421 *fpp = fp;
2422 return (0);
2423 }
2424
2425 int
2426 fget(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2427 {
2428
2429 return(_fget(td, fd, fpp, 0, rights, NULL, NULL, 0));
2430 }
2431
2432 int
2433 fget_mmap(struct thread *td, int fd, cap_rights_t rights, u_char *maxprotp,
2434 struct file **fpp)
2435 {
2436
2437 return (_fget(td, fd, fpp, 0, rights, NULL, maxprotp, 0));
2438 }
2439
2440 int
2441 fget_read(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2442 {
2443
2444 return(_fget(td, fd, fpp, FREAD, rights, NULL, NULL, 0));
2445 }
2446
2447 int
2448 fget_write(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2449 {
2450
2451 return (_fget(td, fd, fpp, FWRITE, rights, NULL, NULL, 0));
2452 }
2453
2454 /*
2455 * Unlike the other fget() calls, which accept and check capability rights
2456 * but never return capabilities, fgetcap() returns the capability but doesn't
2457 * check capability rights.
2458 */
2459 int
2460 fgetcap(struct thread *td, int fd, struct file **fpp)
2461 {
2462
2463 return (_fget(td, fd, fpp, 0, 0, NULL, NULL, FGET_GETCAP));
2464 }
2465
2466
2467 /*
2468 * Like fget() but loads the underlying vnode, or returns an error if the
2469 * descriptor does not represent a vnode. Note that pipes use vnodes but
2470 * never have VM objects. The returned vnode will be vref()'d.
2471 *
2472 * XXX: what about the unused flags ?
2473 */
2474 static __inline int
2475 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t needrights,
2476 cap_rights_t *haverightsp, struct vnode **vpp)
2477 {
2478 struct file *fp;
2479 int error;
2480
2481 *vpp = NULL;
2482 if ((error = _fget(td, fd, &fp, flags, needrights, haverightsp,
2483 NULL, 0)) != 0)
2484 return (error);
2485 if (fp->f_vnode == NULL) {
2486 error = EINVAL;
2487 } else {
2488 *vpp = fp->f_vnode;
2489 vref(*vpp);
2490 }
2491 fdrop(fp, td);
2492
2493 return (error);
2494 }
2495
2496 int
2497 fgetvp(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2498 {
2499
2500 return (_fgetvp(td, fd, 0, rights, NULL, vpp));
2501 }
2502
2503 int
2504 fgetvp_rights(struct thread *td, int fd, cap_rights_t need, cap_rights_t *have,
2505 struct vnode **vpp)
2506 {
2507 return (_fgetvp(td, fd, 0, need, have, vpp));
2508 }
2509
2510 int
2511 fgetvp_read(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2512 {
2513
2514 return (_fgetvp(td, fd, FREAD, rights, NULL, vpp));
2515 }
2516
2517 int
2518 fgetvp_exec(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2519 {
2520
2521 return (_fgetvp(td, fd, FEXEC, rights, NULL, vpp));
2522 }
2523
2524 #ifdef notyet
2525 int
2526 fgetvp_write(struct thread *td, int fd, cap_rights_t rights,
2527 struct vnode **vpp)
2528 {
2529
2530 return (_fgetvp(td, fd, FWRITE, rights, NULL, vpp));
2531 }
2532 #endif
2533
2534 /*
2535 * Like fget() but loads the underlying socket, or returns an error if the
2536 * descriptor does not represent a socket.
2537 *
2538 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2539 * in the future.
2540 *
2541 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2542 * on their file descriptor reference to prevent the socket from being free'd
2543 * during use.
2544 */
2545 int
2546 fgetsock(struct thread *td, int fd, cap_rights_t rights, struct socket **spp,
2547 u_int *fflagp)
2548 {
2549 struct file *fp;
2550 int error;
2551
2552 *spp = NULL;
2553 if (fflagp != NULL)
2554 *fflagp = 0;
2555 if ((error = _fget(td, fd, &fp, 0, rights, NULL, NULL, 0)) != 0)
2556 return (error);
2557 if (fp->f_type != DTYPE_SOCKET) {
2558 error = ENOTSOCK;
2559 } else {
2560 *spp = fp->f_data;
2561 if (fflagp)
2562 *fflagp = fp->f_flag;
2563 SOCK_LOCK(*spp);
2564 soref(*spp);
2565 SOCK_UNLOCK(*spp);
2566 }
2567 fdrop(fp, td);
2568
2569 return (error);
2570 }
2571
2572 /*
2573 * Drop the reference count on the socket and XXX release the SX lock in the
2574 * future. The last reference closes the socket.
2575 *
2576 * Note: fputsock() is deprecated, see comment for fgetsock().
2577 */
2578 void
2579 fputsock(struct socket *so)
2580 {
2581
2582 ACCEPT_LOCK();
2583 SOCK_LOCK(so);
2584 CURVNET_SET(so->so_vnet);
2585 sorele(so);
2586 CURVNET_RESTORE();
2587 }
2588
2589 /*
2590 * Handle the last reference to a file being closed.
2591 *
2592 * No special capability handling here, as the capability's fo_close will run
2593 * instead of the object here, and perform any necessary drop on the object.
2594 */
2595 int
2596 _fdrop(struct file *fp, struct thread *td)
2597 {
2598 int error;
2599
2600 error = 0;
2601 if (fp->f_count != 0)
2602 panic("fdrop: count %d", fp->f_count);
2603 if (fp->f_ops != &badfileops)
2604 error = fo_close(fp, td);
2605 atomic_subtract_int(&openfiles, 1);
2606 crfree(fp->f_cred);
2607 free(fp->f_advice, M_FADVISE);
2608 uma_zfree(file_zone, fp);
2609
2610 return (error);
2611 }
2612
2613 /*
2614 * Apply an advisory lock on a file descriptor.
2615 *
2616 * Just attempt to get a record lock of the requested type on the entire file
2617 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2618 */
2619 #ifndef _SYS_SYSPROTO_H_
2620 struct flock_args {
2621 int fd;
2622 int how;
2623 };
2624 #endif
2625 /* ARGSUSED */
2626 int
2627 sys_flock(struct thread *td, struct flock_args *uap)
2628 {
2629 struct file *fp;
2630 struct vnode *vp;
2631 struct flock lf;
2632 int vfslocked;
2633 int error;
2634
2635 if ((error = fget(td, uap->fd, CAP_FLOCK, &fp)) != 0)
2636 return (error);
2637 if (fp->f_type != DTYPE_VNODE) {
2638 fdrop(fp, td);
2639 return (EOPNOTSUPP);
2640 }
2641
2642 vp = fp->f_vnode;
2643 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2644 lf.l_whence = SEEK_SET;
2645 lf.l_start = 0;
2646 lf.l_len = 0;
2647 if (uap->how & LOCK_UN) {
2648 lf.l_type = F_UNLCK;
2649 atomic_clear_int(&fp->f_flag, FHASLOCK);
2650 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2651 goto done2;
2652 }
2653 if (uap->how & LOCK_EX)
2654 lf.l_type = F_WRLCK;
2655 else if (uap->how & LOCK_SH)
2656 lf.l_type = F_RDLCK;
2657 else {
2658 error = EBADF;
2659 goto done2;
2660 }
2661 atomic_set_int(&fp->f_flag, FHASLOCK);
2662 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2663 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2664 done2:
2665 fdrop(fp, td);
2666 VFS_UNLOCK_GIANT(vfslocked);
2667 return (error);
2668 }
2669 /*
2670 * Duplicate the specified descriptor to a free descriptor.
2671 */
2672 int
2673 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2674 {
2675 struct file *wfp;
2676 struct file *fp;
2677
2678 /*
2679 * If the to-be-dup'd fd number is greater than the allowed number
2680 * of file descriptors, or the fd to be dup'd has already been
2681 * closed, then reject.
2682 */
2683 FILEDESC_XLOCK(fdp);
2684 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2685 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2686 FILEDESC_XUNLOCK(fdp);
2687 return (EBADF);
2688 }
2689
2690 /*
2691 * There are two cases of interest here.
2692 *
2693 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2694 *
2695 * For ENXIO steal away the file structure from (dfd) and store it in
2696 * (indx). (dfd) is effectively closed by this operation.
2697 *
2698 * Any other error code is just returned.
2699 */
2700 switch (error) {
2701 case ENODEV:
2702 /*
2703 * Check that the mode the file is being opened for is a
2704 * subset of the mode of the existing descriptor.
2705 */
2706 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2707 FILEDESC_XUNLOCK(fdp);
2708 return (EACCES);
2709 }
2710 fp = fdp->fd_ofiles[indx];
2711 fdp->fd_ofiles[indx] = wfp;
2712 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2713 if (fp == NULL)
2714 fdused(fdp, indx);
2715 fhold(wfp);
2716 FILEDESC_XUNLOCK(fdp);
2717 if (fp != NULL)
2718 /*
2719 * We now own the reference to fp that the ofiles[]
2720 * array used to own. Release it.
2721 */
2722 fdrop(fp, td);
2723 return (0);
2724
2725 case ENXIO:
2726 /*
2727 * Steal away the file pointer from dfd and stuff it into indx.
2728 */
2729 fp = fdp->fd_ofiles[indx];
2730 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2731 fdp->fd_ofiles[dfd] = NULL;
2732 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2733 fdp->fd_ofileflags[dfd] = 0;
2734 fdunused(fdp, dfd);
2735 if (fp == NULL)
2736 fdused(fdp, indx);
2737 FILEDESC_XUNLOCK(fdp);
2738
2739 /*
2740 * We now own the reference to fp that the ofiles[] array
2741 * used to own. Release it.
2742 */
2743 if (fp != NULL)
2744 fdrop(fp, td);
2745 return (0);
2746
2747 default:
2748 FILEDESC_XUNLOCK(fdp);
2749 return (error);
2750 }
2751 /* NOTREACHED */
2752 }
2753
2754 /*
2755 * Scan all active processes and prisons to see if any of them have a current
2756 * or root directory of `olddp'. If so, replace them with the new mount point.
2757 */
2758 void
2759 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2760 {
2761 struct filedesc *fdp;
2762 struct prison *pr;
2763 struct proc *p;
2764 int nrele;
2765
2766 if (vrefcnt(olddp) == 1)
2767 return;
2768 nrele = 0;
2769 sx_slock(&allproc_lock);
2770 FOREACH_PROC_IN_SYSTEM(p) {
2771 fdp = fdhold(p);
2772 if (fdp == NULL)
2773 continue;
2774 FILEDESC_XLOCK(fdp);
2775 if (fdp->fd_cdir == olddp) {
2776 vref(newdp);
2777 fdp->fd_cdir = newdp;
2778 nrele++;
2779 }
2780 if (fdp->fd_rdir == olddp) {
2781 vref(newdp);
2782 fdp->fd_rdir = newdp;
2783 nrele++;
2784 }
2785 if (fdp->fd_jdir == olddp) {
2786 vref(newdp);
2787 fdp->fd_jdir = newdp;
2788 nrele++;
2789 }
2790 FILEDESC_XUNLOCK(fdp);
2791 fddrop(fdp);
2792 }
2793 sx_sunlock(&allproc_lock);
2794 if (rootvnode == olddp) {
2795 vref(newdp);
2796 rootvnode = newdp;
2797 nrele++;
2798 }
2799 mtx_lock(&prison0.pr_mtx);
2800 if (prison0.pr_root == olddp) {
2801 vref(newdp);
2802 prison0.pr_root = newdp;
2803 nrele++;
2804 }
2805 mtx_unlock(&prison0.pr_mtx);
2806 sx_slock(&allprison_lock);
2807 TAILQ_FOREACH(pr, &allprison, pr_list) {
2808 mtx_lock(&pr->pr_mtx);
2809 if (pr->pr_root == olddp) {
2810 vref(newdp);
2811 pr->pr_root = newdp;
2812 nrele++;
2813 }
2814 mtx_unlock(&pr->pr_mtx);
2815 }
2816 sx_sunlock(&allprison_lock);
2817 while (nrele--)
2818 vrele(olddp);
2819 }
2820
2821 struct filedesc_to_leader *
2822 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2823 {
2824 struct filedesc_to_leader *fdtol;
2825
2826 fdtol = malloc(sizeof(struct filedesc_to_leader),
2827 M_FILEDESC_TO_LEADER,
2828 M_WAITOK);
2829 fdtol->fdl_refcount = 1;
2830 fdtol->fdl_holdcount = 0;
2831 fdtol->fdl_wakeup = 0;
2832 fdtol->fdl_leader = leader;
2833 if (old != NULL) {
2834 FILEDESC_XLOCK(fdp);
2835 fdtol->fdl_next = old->fdl_next;
2836 fdtol->fdl_prev = old;
2837 old->fdl_next = fdtol;
2838 fdtol->fdl_next->fdl_prev = fdtol;
2839 FILEDESC_XUNLOCK(fdp);
2840 } else {
2841 fdtol->fdl_next = fdtol;
2842 fdtol->fdl_prev = fdtol;
2843 }
2844 return (fdtol);
2845 }
2846
2847 /*
2848 * Get file structures globally.
2849 */
2850 static int
2851 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2852 {
2853 struct xfile xf;
2854 struct filedesc *fdp;
2855 struct file *fp;
2856 struct proc *p;
2857 int error, n;
2858
2859 error = sysctl_wire_old_buffer(req, 0);
2860 if (error != 0)
2861 return (error);
2862 if (req->oldptr == NULL) {
2863 n = 0;
2864 sx_slock(&allproc_lock);
2865 FOREACH_PROC_IN_SYSTEM(p) {
2866 if (p->p_state == PRS_NEW)
2867 continue;
2868 fdp = fdhold(p);
2869 if (fdp == NULL)
2870 continue;
2871 /* overestimates sparse tables. */
2872 if (fdp->fd_lastfile > 0)
2873 n += fdp->fd_lastfile;
2874 fddrop(fdp);
2875 }
2876 sx_sunlock(&allproc_lock);
2877 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2878 }
2879 error = 0;
2880 bzero(&xf, sizeof(xf));
2881 xf.xf_size = sizeof(xf);
2882 sx_slock(&allproc_lock);
2883 FOREACH_PROC_IN_SYSTEM(p) {
2884 PROC_LOCK(p);
2885 if (p->p_state == PRS_NEW) {
2886 PROC_UNLOCK(p);
2887 continue;
2888 }
2889 if (p_cansee(req->td, p) != 0) {
2890 PROC_UNLOCK(p);
2891 continue;
2892 }
2893 xf.xf_pid = p->p_pid;
2894 xf.xf_uid = p->p_ucred->cr_uid;
2895 PROC_UNLOCK(p);
2896 fdp = fdhold(p);
2897 if (fdp == NULL)
2898 continue;
2899 FILEDESC_SLOCK(fdp);
2900 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2901 if ((fp = fdp->fd_ofiles[n]) == NULL)
2902 continue;
2903 xf.xf_fd = n;
2904 xf.xf_file = fp;
2905 xf.xf_data = fp->f_data;
2906 xf.xf_vnode = fp->f_vnode;
2907 xf.xf_type = fp->f_type;
2908 xf.xf_count = fp->f_count;
2909 xf.xf_msgcount = 0;
2910 xf.xf_offset = fp->f_offset;
2911 xf.xf_flag = fp->f_flag;
2912 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2913 if (error)
2914 break;
2915 }
2916 FILEDESC_SUNLOCK(fdp);
2917 fddrop(fdp);
2918 if (error)
2919 break;
2920 }
2921 sx_sunlock(&allproc_lock);
2922 return (error);
2923 }
2924
2925 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2926 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2927
2928 #ifdef KINFO_OFILE_SIZE
2929 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2930 #endif
2931
2932 #ifdef COMPAT_FREEBSD7
2933 static int
2934 export_vnode_for_osysctl(struct vnode *vp, int type,
2935 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2936 {
2937 int error;
2938 char *fullpath, *freepath;
2939 int vfslocked;
2940
2941 bzero(kif, sizeof(*kif));
2942 kif->kf_structsize = sizeof(*kif);
2943
2944 vref(vp);
2945 kif->kf_fd = type;
2946 kif->kf_type = KF_TYPE_VNODE;
2947 /* This function only handles directories. */
2948 if (vp->v_type != VDIR) {
2949 vrele(vp);
2950 return (ENOTDIR);
2951 }
2952 kif->kf_vnode_type = KF_VTYPE_VDIR;
2953
2954 /*
2955 * This is not a true file descriptor, so we set a bogus refcount
2956 * and offset to indicate these fields should be ignored.
2957 */
2958 kif->kf_ref_count = -1;
2959 kif->kf_offset = -1;
2960
2961 freepath = NULL;
2962 fullpath = "-";
2963 FILEDESC_SUNLOCK(fdp);
2964 vn_fullpath(curthread, vp, &fullpath, &freepath);
2965 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2966 vrele(vp);
2967 VFS_UNLOCK_GIANT(vfslocked);
2968 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2969 if (freepath != NULL)
2970 free(freepath, M_TEMP);
2971 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2972 FILEDESC_SLOCK(fdp);
2973 return (error);
2974 }
2975
2976 /*
2977 * Get per-process file descriptors for use by procstat(1), et al.
2978 */
2979 static int
2980 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
2981 {
2982 char *fullpath, *freepath;
2983 struct kinfo_ofile *kif;
2984 struct filedesc *fdp;
2985 int error, i, *name;
2986 struct shmfd *shmfd;
2987 struct socket *so;
2988 struct vnode *vp;
2989 struct file *fp;
2990 struct proc *p;
2991 struct tty *tp;
2992 int vfslocked;
2993
2994 name = (int *)arg1;
2995 if ((p = pfind((pid_t)name[0])) == NULL)
2996 return (ESRCH);
2997 if ((error = p_candebug(curthread, p))) {
2998 PROC_UNLOCK(p);
2999 return (error);
3000 }
3001 fdp = fdhold(p);
3002 PROC_UNLOCK(p);
3003 if (fdp == NULL)
3004 return (ENOENT);
3005 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3006 FILEDESC_SLOCK(fdp);
3007 if (fdp->fd_cdir != NULL)
3008 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3009 fdp, req);
3010 if (fdp->fd_rdir != NULL)
3011 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3012 fdp, req);
3013 if (fdp->fd_jdir != NULL)
3014 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3015 fdp, req);
3016 for (i = 0; i < fdp->fd_nfiles; i++) {
3017 if ((fp = fdp->fd_ofiles[i]) == NULL)
3018 continue;
3019 bzero(kif, sizeof(*kif));
3020 kif->kf_structsize = sizeof(*kif);
3021 vp = NULL;
3022 so = NULL;
3023 tp = NULL;
3024 shmfd = NULL;
3025 kif->kf_fd = i;
3026
3027 #ifdef CAPABILITIES
3028 /*
3029 * When reporting a capability, most fields will be from the
3030 * underlying object, but do mark as a capability. With
3031 * ofiledesc, we don't have a field to export the cap_rights_t,
3032 * but we do with the new filedesc.
3033 */
3034 if (fp->f_type == DTYPE_CAPABILITY) {
3035 kif->kf_flags |= KF_FLAG_CAPABILITY;
3036 (void)cap_funwrap(fp, 0, &fp);
3037 }
3038 #else
3039 KASSERT(fp->f_type != DTYPE_CAPABILITY,
3040 ("sysctl_kern_proc_ofiledesc: saw capability"));
3041 #endif
3042 switch (fp->f_type) {
3043 case DTYPE_VNODE:
3044 kif->kf_type = KF_TYPE_VNODE;
3045 vp = fp->f_vnode;
3046 break;
3047
3048 case DTYPE_SOCKET:
3049 kif->kf_type = KF_TYPE_SOCKET;
3050 so = fp->f_data;
3051 break;
3052
3053 case DTYPE_PIPE:
3054 kif->kf_type = KF_TYPE_PIPE;
3055 break;
3056
3057 case DTYPE_FIFO:
3058 kif->kf_type = KF_TYPE_FIFO;
3059 vp = fp->f_vnode;
3060 break;
3061
3062 case DTYPE_KQUEUE:
3063 kif->kf_type = KF_TYPE_KQUEUE;
3064 break;
3065
3066 case DTYPE_CRYPTO:
3067 kif->kf_type = KF_TYPE_CRYPTO;
3068 break;
3069
3070 case DTYPE_MQUEUE:
3071 kif->kf_type = KF_TYPE_MQUEUE;
3072 break;
3073
3074 case DTYPE_SHM:
3075 kif->kf_type = KF_TYPE_SHM;
3076 shmfd = fp->f_data;
3077 break;
3078
3079 case DTYPE_SEM:
3080 kif->kf_type = KF_TYPE_SEM;
3081 break;
3082
3083 case DTYPE_PTS:
3084 kif->kf_type = KF_TYPE_PTS;
3085 tp = fp->f_data;
3086 break;
3087
3088 #ifdef PROCDESC
3089 case DTYPE_PROCDESC:
3090 kif->kf_type = KF_TYPE_PROCDESC;
3091 break;
3092 #endif
3093
3094 default:
3095 kif->kf_type = KF_TYPE_UNKNOWN;
3096 break;
3097 }
3098 kif->kf_ref_count = fp->f_count;
3099 if (fp->f_flag & FREAD)
3100 kif->kf_flags |= KF_FLAG_READ;
3101 if (fp->f_flag & FWRITE)
3102 kif->kf_flags |= KF_FLAG_WRITE;
3103 if (fp->f_flag & FAPPEND)
3104 kif->kf_flags |= KF_FLAG_APPEND;
3105 if (fp->f_flag & FASYNC)
3106 kif->kf_flags |= KF_FLAG_ASYNC;
3107 if (fp->f_flag & FFSYNC)
3108 kif->kf_flags |= KF_FLAG_FSYNC;
3109 if (fp->f_flag & FNONBLOCK)
3110 kif->kf_flags |= KF_FLAG_NONBLOCK;
3111 if (fp->f_flag & O_DIRECT)
3112 kif->kf_flags |= KF_FLAG_DIRECT;
3113 if (fp->f_flag & FHASLOCK)
3114 kif->kf_flags |= KF_FLAG_HASLOCK;
3115 kif->kf_offset = fp->f_offset;
3116 if (vp != NULL) {
3117 vref(vp);
3118 switch (vp->v_type) {
3119 case VNON:
3120 kif->kf_vnode_type = KF_VTYPE_VNON;
3121 break;
3122 case VREG:
3123 kif->kf_vnode_type = KF_VTYPE_VREG;
3124 break;
3125 case VDIR:
3126 kif->kf_vnode_type = KF_VTYPE_VDIR;
3127 break;
3128 case VBLK:
3129 kif->kf_vnode_type = KF_VTYPE_VBLK;
3130 break;
3131 case VCHR:
3132 kif->kf_vnode_type = KF_VTYPE_VCHR;
3133 break;
3134 case VLNK:
3135 kif->kf_vnode_type = KF_VTYPE_VLNK;
3136 break;
3137 case VSOCK:
3138 kif->kf_vnode_type = KF_VTYPE_VSOCK;
3139 break;
3140 case VFIFO:
3141 kif->kf_vnode_type = KF_VTYPE_VFIFO;
3142 break;
3143 case VBAD:
3144 kif->kf_vnode_type = KF_VTYPE_VBAD;
3145 break;
3146 default:
3147 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3148 break;
3149 }
3150 /*
3151 * It is OK to drop the filedesc lock here as we will
3152 * re-validate and re-evaluate its properties when
3153 * the loop continues.
3154 */
3155 freepath = NULL;
3156 fullpath = "-";
3157 FILEDESC_SUNLOCK(fdp);
3158 vn_fullpath(curthread, vp, &fullpath, &freepath);
3159 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
3160 vrele(vp);
3161 VFS_UNLOCK_GIANT(vfslocked);
3162 strlcpy(kif->kf_path, fullpath,
3163 sizeof(kif->kf_path));
3164 if (freepath != NULL)
3165 free(freepath, M_TEMP);
3166 FILEDESC_SLOCK(fdp);
3167 }
3168 if (so != NULL) {
3169 struct sockaddr *sa;
3170
3171 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3172 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3173 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3174 free(sa, M_SONAME);
3175 }
3176 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3177 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3178 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3179 free(sa, M_SONAME);
3180 }
3181 kif->kf_sock_domain =
3182 so->so_proto->pr_domain->dom_family;
3183 kif->kf_sock_type = so->so_type;
3184 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3185 }
3186 if (tp != NULL) {
3187 strlcpy(kif->kf_path, tty_devname(tp),
3188 sizeof(kif->kf_path));
3189 }
3190 if (shmfd != NULL)
3191 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
3192 error = SYSCTL_OUT(req, kif, sizeof(*kif));
3193 if (error)
3194 break;
3195 }
3196 FILEDESC_SUNLOCK(fdp);
3197 fddrop(fdp);
3198 free(kif, M_TEMP);
3199 return (0);
3200 }
3201
3202 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
3203 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
3204 #endif /* COMPAT_FREEBSD7 */
3205
3206 #ifdef KINFO_FILE_SIZE
3207 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3208 #endif
3209
3210 static int
3211 export_fd_for_sysctl(void *data, int type, int fd, int fflags, int refcnt,
3212 int64_t offset, int fd_is_cap, cap_rights_t fd_cap_rights,
3213 struct kinfo_file *kif, struct sysctl_req *req)
3214 {
3215 struct {
3216 int fflag;
3217 int kf_fflag;
3218 } fflags_table[] = {
3219 { FAPPEND, KF_FLAG_APPEND },
3220 { FASYNC, KF_FLAG_ASYNC },
3221 { FFSYNC, KF_FLAG_FSYNC },
3222 { FHASLOCK, KF_FLAG_HASLOCK },
3223 { FNONBLOCK, KF_FLAG_NONBLOCK },
3224 { FREAD, KF_FLAG_READ },
3225 { FWRITE, KF_FLAG_WRITE },
3226 { O_CREAT, KF_FLAG_CREAT },
3227 { O_DIRECT, KF_FLAG_DIRECT },
3228 { O_EXCL, KF_FLAG_EXCL },
3229 { O_EXEC, KF_FLAG_EXEC },
3230 { O_EXLOCK, KF_FLAG_EXLOCK },
3231 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3232 { O_SHLOCK, KF_FLAG_SHLOCK },
3233 { O_TRUNC, KF_FLAG_TRUNC }
3234 };
3235 #define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table))
3236 struct vnode *vp;
3237 int error, vfslocked;
3238 unsigned int i;
3239
3240 bzero(kif, sizeof(*kif));
3241 switch (type) {
3242 case KF_TYPE_FIFO:
3243 case KF_TYPE_VNODE:
3244 vp = (struct vnode *)data;
3245 error = fill_vnode_info(vp, kif);
3246 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
3247 vrele(vp);
3248 VFS_UNLOCK_GIANT(vfslocked);
3249 break;
3250 case KF_TYPE_SOCKET:
3251 error = fill_socket_info((struct socket *)data, kif);
3252 break;
3253 case KF_TYPE_PIPE:
3254 error = fill_pipe_info((struct pipe *)data, kif);
3255 break;
3256 case KF_TYPE_PTS:
3257 error = fill_pts_info((struct tty *)data, kif);
3258 break;
3259 case KF_TYPE_PROCDESC:
3260 error = fill_procdesc_info((struct procdesc *)data, kif);
3261 break;
3262 case KF_TYPE_SHM:
3263 error = fill_shm_info((struct file *)data, kif);
3264 break;
3265 default:
3266 error = 0;
3267 }
3268 if (error == 0)
3269 kif->kf_status |= KF_ATTR_VALID;
3270
3271 /*
3272 * Translate file access flags.
3273 */
3274 for (i = 0; i < NFFLAGS; i++)
3275 if (fflags & fflags_table[i].fflag)
3276 kif->kf_flags |= fflags_table[i].kf_fflag;
3277 if (fd_is_cap)
3278 kif->kf_flags |= KF_FLAG_CAPABILITY;
3279 if (fd_is_cap)
3280 kif->kf_cap_rights = fd_cap_rights;
3281 kif->kf_fd = fd;
3282 kif->kf_type = type;
3283 kif->kf_ref_count = refcnt;
3284 kif->kf_offset = offset;
3285 /* Pack record size down */
3286 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3287 strlen(kif->kf_path) + 1;
3288 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3289 error = SYSCTL_OUT(req, kif, kif->kf_structsize);
3290 return (error);
3291 }
3292
3293 /*
3294 * Get per-process file descriptors for use by procstat(1), et al.
3295 */
3296 static int
3297 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3298 {
3299 struct file *fp;
3300 struct filedesc *fdp;
3301 struct kinfo_file *kif;
3302 struct proc *p;
3303 struct vnode *cttyvp, *textvp, *tracevp;
3304 size_t oldidx;
3305 int64_t offset;
3306 void *data;
3307 int error, i, *name;
3308 int fd_is_cap, type, refcnt, fflags;
3309 cap_rights_t fd_cap_rights;
3310
3311 name = (int *)arg1;
3312 if ((p = pfind((pid_t)name[0])) == NULL)
3313 return (ESRCH);
3314 if ((error = p_candebug(curthread, p))) {
3315 PROC_UNLOCK(p);
3316 return (error);
3317 }
3318 /* ktrace vnode */
3319 tracevp = p->p_tracevp;
3320 if (tracevp != NULL)
3321 vref(tracevp);
3322 /* text vnode */
3323 textvp = p->p_textvp;
3324 if (textvp != NULL)
3325 vref(textvp);
3326 /* Controlling tty. */
3327 cttyvp = NULL;
3328 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3329 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3330 if (cttyvp != NULL)
3331 vref(cttyvp);
3332 }
3333 fdp = fdhold(p);
3334 PROC_UNLOCK(p);
3335 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3336 if (tracevp != NULL)
3337 export_fd_for_sysctl(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
3338 FREAD | FWRITE, -1, -1, 0, 0, kif, req);
3339 if (textvp != NULL)
3340 export_fd_for_sysctl(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
3341 FREAD, -1, -1, 0, 0, kif, req);
3342 if (cttyvp != NULL)
3343 export_fd_for_sysctl(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
3344 FREAD | FWRITE, -1, -1, 0, 0, kif, req);
3345 if (fdp == NULL)
3346 goto fail;
3347 FILEDESC_SLOCK(fdp);
3348 /* working directory */
3349 if (fdp->fd_cdir != NULL) {
3350 vref(fdp->fd_cdir);
3351 data = fdp->fd_cdir;
3352 FILEDESC_SUNLOCK(fdp);
3353 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
3354 FREAD, -1, -1, 0, 0, kif, req);
3355 FILEDESC_SLOCK(fdp);
3356 }
3357 /* root directory */
3358 if (fdp->fd_rdir != NULL) {
3359 vref(fdp->fd_rdir);
3360 data = fdp->fd_rdir;
3361 FILEDESC_SUNLOCK(fdp);
3362 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
3363 FREAD, -1, -1, 0, 0, kif, req);
3364 FILEDESC_SLOCK(fdp);
3365 }
3366 /* jail directory */
3367 if (fdp->fd_jdir != NULL) {
3368 vref(fdp->fd_jdir);
3369 data = fdp->fd_jdir;
3370 FILEDESC_SUNLOCK(fdp);
3371 export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
3372 FREAD, -1, -1, 0, 0, kif, req);
3373 FILEDESC_SLOCK(fdp);
3374 }
3375 for (i = 0; i < fdp->fd_nfiles; i++) {
3376 if ((fp = fdp->fd_ofiles[i]) == NULL)
3377 continue;
3378 data = NULL;
3379 fd_is_cap = 0;
3380 fd_cap_rights = 0;
3381
3382 #ifdef CAPABILITIES
3383 /*
3384 * When reporting a capability, most fields will be from the
3385 * underlying object, but do mark as a capability and export
3386 * the capability rights mask.
3387 */
3388 if (fp->f_type == DTYPE_CAPABILITY) {
3389 fd_is_cap = 1;
3390 fd_cap_rights = cap_rights(fp);
3391 (void)cap_funwrap(fp, 0, &fp);
3392 }
3393 #else /* !CAPABILITIES */
3394 KASSERT(fp->f_type != DTYPE_CAPABILITY,
3395 ("sysctl_kern_proc_filedesc: saw capability"));
3396 #endif
3397 switch (fp->f_type) {
3398 case DTYPE_VNODE:
3399 type = KF_TYPE_VNODE;
3400 vref(fp->f_vnode);
3401 data = fp->f_vnode;
3402 break;
3403
3404 case DTYPE_SOCKET:
3405 type = KF_TYPE_SOCKET;
3406 data = fp->f_data;
3407 break;
3408
3409 case DTYPE_PIPE:
3410 type = KF_TYPE_PIPE;
3411 data = fp->f_data;
3412 break;
3413
3414 case DTYPE_FIFO:
3415 type = KF_TYPE_FIFO;
3416 vref(fp->f_vnode);
3417 data = fp->f_vnode;
3418 break;
3419
3420 case DTYPE_KQUEUE:
3421 type = KF_TYPE_KQUEUE;
3422 break;
3423
3424 case DTYPE_CRYPTO:
3425 type = KF_TYPE_CRYPTO;
3426 break;
3427
3428 case DTYPE_MQUEUE:
3429 type = KF_TYPE_MQUEUE;
3430 break;
3431
3432 case DTYPE_SHM:
3433 type = KF_TYPE_SHM;
3434 data = fp;
3435 break;
3436
3437 case DTYPE_SEM:
3438 type = KF_TYPE_SEM;
3439 break;
3440
3441 case DTYPE_PTS:
3442 type = KF_TYPE_PTS;
3443 data = fp->f_data;
3444 break;
3445
3446 #ifdef PROCDESC
3447 case DTYPE_PROCDESC:
3448 type = KF_TYPE_PROCDESC;
3449 data = fp->f_data;
3450 break;
3451 #endif
3452
3453 default:
3454 type = KF_TYPE_UNKNOWN;
3455 break;
3456 }
3457 refcnt = fp->f_count;
3458 fflags = fp->f_flag;
3459 offset = fp->f_offset;
3460
3461 /*
3462 * Create sysctl entry.
3463 * It is OK to drop the filedesc lock here as we will
3464 * re-validate and re-evaluate its properties when
3465 * the loop continues.
3466 */
3467 oldidx = req->oldidx;
3468 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO)
3469 FILEDESC_SUNLOCK(fdp);
3470 error = export_fd_for_sysctl(data, type, i, fflags, refcnt,
3471 offset, fd_is_cap, fd_cap_rights, kif, req);
3472 if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO)
3473 FILEDESC_SLOCK(fdp);
3474 if (error) {
3475 if (error == ENOMEM) {
3476 /*
3477 * The hack to keep the ABI of sysctl
3478 * kern.proc.filedesc intact, but not
3479 * to account a partially copied
3480 * kinfo_file into the oldidx.
3481 */
3482 req->oldidx = oldidx;
3483 error = 0;
3484 }
3485 break;
3486 }
3487 }
3488 FILEDESC_SUNLOCK(fdp);
3489 fail:
3490 if (fdp != NULL)
3491 fddrop(fdp);
3492 free(kif, M_TEMP);
3493 return (error);
3494 }
3495
3496 int
3497 vntype_to_kinfo(int vtype)
3498 {
3499 struct {
3500 int vtype;
3501 int kf_vtype;
3502 } vtypes_table[] = {
3503 { VBAD, KF_VTYPE_VBAD },
3504 { VBLK, KF_VTYPE_VBLK },
3505 { VCHR, KF_VTYPE_VCHR },
3506 { VDIR, KF_VTYPE_VDIR },
3507 { VFIFO, KF_VTYPE_VFIFO },
3508 { VLNK, KF_VTYPE_VLNK },
3509 { VNON, KF_VTYPE_VNON },
3510 { VREG, KF_VTYPE_VREG },
3511 { VSOCK, KF_VTYPE_VSOCK }
3512 };
3513 #define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table))
3514 unsigned int i;
3515
3516 /*
3517 * Perform vtype translation.
3518 */
3519 for (i = 0; i < NVTYPES; i++)
3520 if (vtypes_table[i].vtype == vtype)
3521 break;
3522 if (i < NVTYPES)
3523 return (vtypes_table[i].kf_vtype);
3524
3525 return (KF_VTYPE_UNKNOWN);
3526 }
3527
3528 static int
3529 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
3530 {
3531 struct vattr va;
3532 char *fullpath, *freepath;
3533 int error, vfslocked;
3534
3535 if (vp == NULL)
3536 return (1);
3537 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
3538 freepath = NULL;
3539 fullpath = "-";
3540 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
3541 if (error == 0) {
3542 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3543 }
3544 if (freepath != NULL)
3545 free(freepath, M_TEMP);
3546
3547 /*
3548 * Retrieve vnode attributes.
3549 */
3550 va.va_fsid = VNOVAL;
3551 va.va_rdev = NODEV;
3552 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
3553 vn_lock(vp, LK_SHARED | LK_RETRY);
3554 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
3555 VOP_UNLOCK(vp, 0);
3556 VFS_UNLOCK_GIANT(vfslocked);
3557 if (error != 0)
3558 return (error);
3559 if (va.va_fsid != VNOVAL)
3560 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
3561 else
3562 kif->kf_un.kf_file.kf_file_fsid =
3563 vp->v_mount->mnt_stat.f_fsid.val[0];
3564 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
3565 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
3566 kif->kf_un.kf_file.kf_file_size = va.va_size;
3567 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
3568 return (0);
3569 }
3570
3571 static int
3572 fill_socket_info(struct socket *so, struct kinfo_file *kif)
3573 {
3574 struct sockaddr *sa;
3575 struct inpcb *inpcb;
3576 struct unpcb *unpcb;
3577 int error;
3578
3579 if (so == NULL)
3580 return (1);
3581 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
3582 kif->kf_sock_type = so->so_type;
3583 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3584 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
3585 switch(kif->kf_sock_domain) {
3586 case AF_INET:
3587 case AF_INET6:
3588 if (kif->kf_sock_protocol == IPPROTO_TCP) {
3589 if (so->so_pcb != NULL) {
3590 inpcb = (struct inpcb *)(so->so_pcb);
3591 kif->kf_un.kf_sock.kf_sock_inpcb =
3592 (uintptr_t)inpcb->inp_ppcb;
3593 }
3594 }
3595 break;
3596 case AF_UNIX:
3597 if (so->so_pcb != NULL) {
3598 unpcb = (struct unpcb *)(so->so_pcb);
3599 if (unpcb->unp_conn) {
3600 kif->kf_un.kf_sock.kf_sock_unpconn =
3601 (uintptr_t)unpcb->unp_conn;
3602 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
3603 so->so_rcv.sb_state;
3604 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
3605 so->so_snd.sb_state;
3606 }
3607 }
3608 break;
3609 }
3610 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
3611 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3612 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3613 free(sa, M_SONAME);
3614 }
3615 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
3616 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3617 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3618 free(sa, M_SONAME);
3619 }
3620 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
3621 sizeof(kif->kf_path));
3622 return (0);
3623 }
3624
3625 static int
3626 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
3627 {
3628
3629 if (tp == NULL)
3630 return (1);
3631 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
3632 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
3633 return (0);
3634 }
3635
3636 static int
3637 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
3638 {
3639
3640 if (pi == NULL)
3641 return (1);
3642 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
3643 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
3644 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
3645 return (0);
3646 }
3647
3648 static int
3649 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
3650 {
3651
3652 if (pdp == NULL)
3653 return (1);
3654 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
3655 return (0);
3656 }
3657
3658 static int
3659 fill_shm_info(struct file *fp, struct kinfo_file *kif)
3660 {
3661 struct thread *td;
3662 struct stat sb;
3663
3664 td = curthread;
3665 if (fp->f_data == NULL)
3666 return (1);
3667 if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3668 return (1);
3669 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
3670 kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
3671 kif->kf_un.kf_file.kf_file_size = sb.st_size;
3672 return (0);
3673 }
3674
3675 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3676 sysctl_kern_proc_filedesc, "Process filedesc entries");
3677
3678 #ifdef DDB
3679 /*
3680 * For the purposes of debugging, generate a human-readable string for the
3681 * file type.
3682 */
3683 static const char *
3684 file_type_to_name(short type)
3685 {
3686
3687 switch (type) {
3688 case 0:
3689 return ("zero");
3690 case DTYPE_VNODE:
3691 return ("vnod");
3692 case DTYPE_SOCKET:
3693 return ("sock");
3694 case DTYPE_PIPE:
3695 return ("pipe");
3696 case DTYPE_FIFO:
3697 return ("fifo");
3698 case DTYPE_KQUEUE:
3699 return ("kque");
3700 case DTYPE_CRYPTO:
3701 return ("crpt");
3702 case DTYPE_MQUEUE:
3703 return ("mque");
3704 case DTYPE_SHM:
3705 return ("shm");
3706 case DTYPE_SEM:
3707 return ("ksem");
3708 default:
3709 return ("unkn");
3710 }
3711 }
3712
3713 /*
3714 * For the purposes of debugging, identify a process (if any, perhaps one of
3715 * many) that references the passed file in its file descriptor array. Return
3716 * NULL if none.
3717 */
3718 static struct proc *
3719 file_to_first_proc(struct file *fp)
3720 {
3721 struct filedesc *fdp;
3722 struct proc *p;
3723 int n;
3724
3725 FOREACH_PROC_IN_SYSTEM(p) {
3726 if (p->p_state == PRS_NEW)
3727 continue;
3728 fdp = p->p_fd;
3729 if (fdp == NULL)
3730 continue;
3731 for (n = 0; n < fdp->fd_nfiles; n++) {
3732 if (fp == fdp->fd_ofiles[n])
3733 return (p);
3734 }
3735 }
3736 return (NULL);
3737 }
3738
3739 static void
3740 db_print_file(struct file *fp, int header)
3741 {
3742 struct proc *p;
3743
3744 if (header)
3745 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3746 "File", "Type", "Data", "Flag", "GCFl", "Count",
3747 "MCount", "Vnode", "FPID", "FCmd");
3748 p = file_to_first_proc(fp);
3749 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3750 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3751 0, fp->f_count, 0, fp->f_vnode,
3752 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3753 }
3754
3755 DB_SHOW_COMMAND(file, db_show_file)
3756 {
3757 struct file *fp;
3758
3759 if (!have_addr) {
3760 db_printf("usage: show file <addr>\n");
3761 return;
3762 }
3763 fp = (struct file *)addr;
3764 db_print_file(fp, 1);
3765 }
3766
3767 DB_SHOW_COMMAND(files, db_show_files)
3768 {
3769 struct filedesc *fdp;
3770 struct file *fp;
3771 struct proc *p;
3772 int header;
3773 int n;
3774
3775 header = 1;
3776 FOREACH_PROC_IN_SYSTEM(p) {
3777 if (p->p_state == PRS_NEW)
3778 continue;
3779 if ((fdp = p->p_fd) == NULL)
3780 continue;
3781 for (n = 0; n < fdp->fd_nfiles; ++n) {
3782 if ((fp = fdp->fd_ofiles[n]) == NULL)
3783 continue;
3784 db_print_file(fp, header);
3785 header = 0;
3786 }
3787 }
3788 }
3789 #endif
3790
3791 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3792 &maxfilesperproc, 0, "Maximum files allowed open per process");
3793
3794 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3795 &maxfiles, 0, "Maximum number of files");
3796
3797 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3798 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3799
3800 /* ARGSUSED*/
3801 static void
3802 filelistinit(void *dummy)
3803 {
3804
3805 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3806 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3807 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3808 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3809 }
3810 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3811
3812 /*-------------------------------------------------------------------*/
3813
3814 static int
3815 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3816 int flags, struct thread *td)
3817 {
3818
3819 return (EBADF);
3820 }
3821
3822 static int
3823 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3824 struct thread *td)
3825 {
3826
3827 return (EINVAL);
3828 }
3829
3830 static int
3831 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3832 struct thread *td)
3833 {
3834
3835 return (EBADF);
3836 }
3837
3838 static int
3839 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3840 struct thread *td)
3841 {
3842
3843 return (0);
3844 }
3845
3846 static int
3847 badfo_kqfilter(struct file *fp, struct knote *kn)
3848 {
3849
3850 return (EBADF);
3851 }
3852
3853 static int
3854 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3855 struct thread *td)
3856 {
3857
3858 return (EBADF);
3859 }
3860
3861 static int
3862 badfo_close(struct file *fp, struct thread *td)
3863 {
3864
3865 return (EBADF);
3866 }
3867
3868 static int
3869 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3870 struct thread *td)
3871 {
3872
3873 return (EBADF);
3874 }
3875
3876 static int
3877 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3878 struct thread *td)
3879 {
3880
3881 return (EBADF);
3882 }
3883
3884 struct fileops badfileops = {
3885 .fo_read = badfo_readwrite,
3886 .fo_write = badfo_readwrite,
3887 .fo_truncate = badfo_truncate,
3888 .fo_ioctl = badfo_ioctl,
3889 .fo_poll = badfo_poll,
3890 .fo_kqfilter = badfo_kqfilter,
3891 .fo_stat = badfo_stat,
3892 .fo_close = badfo_close,
3893 .fo_chmod = badfo_chmod,
3894 .fo_chown = badfo_chown,
3895 };
3896
3897 int
3898 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3899 struct thread *td)
3900 {
3901
3902 return (EINVAL);
3903 }
3904
3905 int
3906 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3907 struct thread *td)
3908 {
3909
3910 return (EINVAL);
3911 }
3912
3913 /*-------------------------------------------------------------------*/
3914
3915 /*
3916 * File Descriptor pseudo-device driver (/dev/fd/).
3917 *
3918 * Opening minor device N dup()s the file (if any) connected to file
3919 * descriptor N belonging to the calling process. Note that this driver
3920 * consists of only the ``open()'' routine, because all subsequent
3921 * references to this file will be direct to the other driver.
3922 *
3923 * XXX: we could give this one a cloning event handler if necessary.
3924 */
3925
3926 /* ARGSUSED */
3927 static int
3928 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3929 {
3930
3931 /*
3932 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3933 * the file descriptor being sought for duplication. The error
3934 * return ensures that the vnode for this device will be released
3935 * by vn_open. Open will detect this special error and take the
3936 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3937 * will simply report the error.
3938 */
3939 td->td_dupfd = dev2unit(dev);
3940 return (ENODEV);
3941 }
3942
3943 static struct cdevsw fildesc_cdevsw = {
3944 .d_version = D_VERSION,
3945 .d_open = fdopen,
3946 .d_name = "FD",
3947 };
3948
3949 static void
3950 fildesc_drvinit(void *unused)
3951 {
3952 struct cdev *dev;
3953
3954 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
3955 UID_ROOT, GID_WHEEL, 0666, "fd/0");
3956 make_dev_alias(dev, "stdin");
3957 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
3958 UID_ROOT, GID_WHEEL, 0666, "fd/1");
3959 make_dev_alias(dev, "stdout");
3960 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
3961 UID_ROOT, GID_WHEEL, 0666, "fd/2");
3962 make_dev_alias(dev, "stderr");
3963 }
3964
3965 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
Cache object: bc47a59ef599e1b9f1e1b9fd6904950b
|