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("$FreeBSD$");
39
40 #include "opt_compat.h"
41 #include "opt_ddb.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45
46 #include <sys/conf.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
58 #include <sys/namei.h>
59 #include <sys/proc.h>
60 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/socketvar.h>
63 #include <sys/stat.h>
64 #include <sys/sx.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysproto.h>
68 #include <sys/unistd.h>
69 #include <sys/vnode.h>
70
71 #include <security/audit/audit.h>
72
73 #include <vm/uma.h>
74
75 #include <ddb/ddb.h>
76
77 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
78 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
79 "file desc to leader structures");
80 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
81
82 static uma_zone_t file_zone;
83
84
85 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
86 enum dup_type { DUP_VARIABLE, DUP_FIXED };
87
88 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
89 register_t *retval);
90 static int fd_first_free(struct filedesc *, int, int);
91 static int fd_last_used(struct filedesc *, int, int);
92 static void fdgrowtable(struct filedesc *, int);
93 static int fdrop_locked(struct file *fp, struct thread *td);
94 static void fdunused(struct filedesc *fdp, int fd);
95 static void fdused(struct filedesc *fdp, int fd);
96
97 /*
98 * A process is initially started out with NDFILE descriptors stored within
99 * this structure, selected to be enough for typical applications based on
100 * the historical limit of 20 open files (and the usage of descriptors by
101 * shells). If these descriptors are exhausted, a larger descriptor table
102 * may be allocated, up to a process' resource limit; the internal arrays
103 * are then unused.
104 */
105 #define NDFILE 20
106 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
107 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
108 #define NDSLOT(x) ((x) / NDENTRIES)
109 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
110 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
111
112 /*
113 * Storage required per open file descriptor.
114 */
115 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
116
117 /*
118 * Basic allocation of descriptors:
119 * one of the above, plus arrays for NDFILE descriptors.
120 */
121 struct filedesc0 {
122 struct filedesc fd_fd;
123 /*
124 * These arrays are used when the number of open files is
125 * <= NDFILE, and are then pointed to by the pointers above.
126 */
127 struct file *fd_dfiles[NDFILE];
128 char fd_dfileflags[NDFILE];
129 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
130 };
131
132 /*
133 * Descriptor management.
134 */
135 struct filelist filehead; /* head of list of open files */
136 int openfiles; /* actual number of open files */
137 struct sx filelist_lock; /* sx to protect filelist */
138 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
139
140 /* A mutex to protect the association between a proc and filedesc. */
141 static struct mtx fdesc_mtx;
142
143 /*
144 * Find the first zero bit in the given bitmap, starting at low and not
145 * exceeding size - 1.
146 */
147 static int
148 fd_first_free(struct filedesc *fdp, int low, int size)
149 {
150 NDSLOTTYPE *map = fdp->fd_map;
151 NDSLOTTYPE mask;
152 int off, maxoff;
153
154 if (low >= size)
155 return (low);
156
157 off = NDSLOT(low);
158 if (low % NDENTRIES) {
159 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
160 if ((mask &= ~map[off]) != 0UL)
161 return (off * NDENTRIES + ffsl(mask) - 1);
162 ++off;
163 }
164 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
165 if (map[off] != ~0UL)
166 return (off * NDENTRIES + ffsl(~map[off]) - 1);
167 return (size);
168 }
169
170 /*
171 * Find the highest non-zero bit in the given bitmap, starting at low and
172 * not exceeding size - 1.
173 */
174 static int
175 fd_last_used(struct filedesc *fdp, int low, int size)
176 {
177 NDSLOTTYPE *map = fdp->fd_map;
178 NDSLOTTYPE mask;
179 int off, minoff;
180
181 if (low >= size)
182 return (-1);
183
184 off = NDSLOT(size);
185 if (size % NDENTRIES) {
186 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
187 if ((mask &= map[off]) != 0)
188 return (off * NDENTRIES + flsl(mask) - 1);
189 --off;
190 }
191 for (minoff = NDSLOT(low); off >= minoff; --off)
192 if (map[off] != 0)
193 return (off * NDENTRIES + flsl(map[off]) - 1);
194 return (low - 1);
195 }
196
197 static int
198 fdisused(struct filedesc *fdp, int fd)
199 {
200 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
201 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
202 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
203 }
204
205 /*
206 * Mark a file descriptor as used.
207 */
208 static void
209 fdused(struct filedesc *fdp, int fd)
210 {
211 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
212 KASSERT(!fdisused(fdp, fd),
213 ("fd already used"));
214 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
215 if (fd > fdp->fd_lastfile)
216 fdp->fd_lastfile = fd;
217 if (fd == fdp->fd_freefile)
218 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
219 }
220
221 /*
222 * Mark a file descriptor as unused.
223 */
224 static void
225 fdunused(struct filedesc *fdp, int fd)
226 {
227 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
228 KASSERT(fdisused(fdp, fd),
229 ("fd is already unused"));
230 KASSERT(fdp->fd_ofiles[fd] == NULL,
231 ("fd is still in use"));
232 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
233 if (fd < fdp->fd_freefile)
234 fdp->fd_freefile = fd;
235 if (fd == fdp->fd_lastfile)
236 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
237 }
238
239 /*
240 * System calls on descriptors.
241 */
242 #ifndef _SYS_SYSPROTO_H_
243 struct getdtablesize_args {
244 int dummy;
245 };
246 #endif
247 /*
248 * MPSAFE
249 */
250 /* ARGSUSED */
251 int
252 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
253 {
254 struct proc *p = td->td_proc;
255
256 PROC_LOCK(p);
257 td->td_retval[0] =
258 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
259 PROC_UNLOCK(p);
260 return (0);
261 }
262
263 /*
264 * Duplicate a file descriptor to a particular value.
265 *
266 * note: keep in mind that a potential race condition exists when closing
267 * descriptors from a shared descriptor table (via rfork).
268 */
269 #ifndef _SYS_SYSPROTO_H_
270 struct dup2_args {
271 u_int from;
272 u_int to;
273 };
274 #endif
275 /*
276 * MPSAFE
277 */
278 /* ARGSUSED */
279 int
280 dup2(struct thread *td, struct dup2_args *uap)
281 {
282
283 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
284 td->td_retval));
285 }
286
287 /*
288 * Duplicate a file descriptor.
289 */
290 #ifndef _SYS_SYSPROTO_H_
291 struct dup_args {
292 u_int fd;
293 };
294 #endif
295 /*
296 * MPSAFE
297 */
298 /* ARGSUSED */
299 int
300 dup(struct thread *td, struct dup_args *uap)
301 {
302
303 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
304 }
305
306 /*
307 * The file control system call.
308 */
309 #ifndef _SYS_SYSPROTO_H_
310 struct fcntl_args {
311 int fd;
312 int cmd;
313 long arg;
314 };
315 #endif
316 /*
317 * MPSAFE
318 */
319 /* ARGSUSED */
320 int
321 fcntl(struct thread *td, struct fcntl_args *uap)
322 {
323 struct flock fl;
324 struct oflock ofl;
325 intptr_t arg;
326 int error;
327 int cmd;
328
329 error = 0;
330 cmd = uap->cmd;
331 switch (uap->cmd) {
332 case F_OGETLK:
333 case F_OSETLK:
334 case F_OSETLKW:
335 /*
336 * Convert old flock structure to new.
337 */
338 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
339 fl.l_start = ofl.l_start;
340 fl.l_len = ofl.l_len;
341 fl.l_pid = ofl.l_pid;
342 fl.l_type = ofl.l_type;
343 fl.l_whence = ofl.l_whence;
344 fl.l_sysid = 0;
345
346 switch (uap->cmd) {
347 case F_OGETLK:
348 cmd = F_GETLK;
349 break;
350 case F_OSETLK:
351 cmd = F_SETLK;
352 break;
353 case F_OSETLKW:
354 cmd = F_SETLKW;
355 break;
356 }
357 arg = (intptr_t)&fl;
358 break;
359 case F_GETLK:
360 case F_SETLK:
361 case F_SETLKW:
362 case F_SETLK_REMOTE:
363 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
364 arg = (intptr_t)&fl;
365 break;
366 default:
367 arg = uap->arg;
368 break;
369 }
370 if (error)
371 return (error);
372 error = kern_fcntl(td, uap->fd, cmd, arg);
373 if (error)
374 return (error);
375 if (uap->cmd == F_OGETLK) {
376 ofl.l_start = fl.l_start;
377 ofl.l_len = fl.l_len;
378 ofl.l_pid = fl.l_pid;
379 ofl.l_type = fl.l_type;
380 ofl.l_whence = fl.l_whence;
381 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
382 } else if (uap->cmd == F_GETLK) {
383 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
384 }
385 return (error);
386 }
387
388 int
389 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
390 {
391 struct filedesc *fdp;
392 struct flock *flp;
393 struct file *fp;
394 struct proc *p;
395 char *pop;
396 struct vnode *vp;
397 u_int newmin;
398 int error, flg, tmp;
399 int giant_locked;
400
401 /*
402 * XXXRW: Some fcntl() calls require Giant -- others don't. Try to
403 * avoid grabbing Giant for calls we know don't need it.
404 */
405 switch (cmd) {
406 case F_DUPFD:
407 case F_GETFD:
408 case F_SETFD:
409 case F_GETFL:
410 giant_locked = 0;
411 break;
412
413 default:
414 giant_locked = 1;
415 mtx_lock(&Giant);
416 }
417
418 error = 0;
419 flg = F_POSIX;
420 p = td->td_proc;
421 fdp = p->p_fd;
422 FILEDESC_LOCK(fdp);
423 if ((unsigned)fd >= fdp->fd_nfiles ||
424 (fp = fdp->fd_ofiles[fd]) == NULL) {
425 FILEDESC_UNLOCK(fdp);
426 error = EBADF;
427 goto done2;
428 }
429 pop = &fdp->fd_ofileflags[fd];
430
431 switch (cmd) {
432 case F_DUPFD:
433 /* mtx_assert(&Giant, MA_NOTOWNED); */
434 FILEDESC_UNLOCK(fdp);
435 newmin = arg;
436 PROC_LOCK(p);
437 if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
438 newmin >= maxfilesperproc) {
439 PROC_UNLOCK(p);
440 error = EINVAL;
441 break;
442 }
443 PROC_UNLOCK(p);
444 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
445 break;
446
447 case F_GETFD:
448 /* mtx_assert(&Giant, MA_NOTOWNED); */
449 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
450 FILEDESC_UNLOCK(fdp);
451 break;
452
453 case F_SETFD:
454 /* mtx_assert(&Giant, MA_NOTOWNED); */
455 *pop = (*pop &~ UF_EXCLOSE) |
456 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
457 FILEDESC_UNLOCK(fdp);
458 break;
459
460 case F_GETFL:
461 /* mtx_assert(&Giant, MA_NOTOWNED); */
462 FILE_LOCK(fp);
463 td->td_retval[0] = OFLAGS(fp->f_flag);
464 FILE_UNLOCK(fp);
465 FILEDESC_UNLOCK(fdp);
466 break;
467
468 case F_SETFL:
469 mtx_assert(&Giant, MA_OWNED);
470 FILE_LOCK(fp);
471 fhold_locked(fp);
472 fp->f_flag &= ~FCNTLFLAGS;
473 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
474 FILE_UNLOCK(fp);
475 FILEDESC_UNLOCK(fdp);
476 tmp = fp->f_flag & FNONBLOCK;
477 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
478 if (error) {
479 fdrop(fp, td);
480 break;
481 }
482 tmp = fp->f_flag & FASYNC;
483 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
484 if (error == 0) {
485 fdrop(fp, td);
486 break;
487 }
488 FILE_LOCK(fp);
489 fp->f_flag &= ~FNONBLOCK;
490 FILE_UNLOCK(fp);
491 tmp = 0;
492 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
493 fdrop(fp, td);
494 break;
495
496 case F_GETOWN:
497 mtx_assert(&Giant, MA_OWNED);
498 fhold(fp);
499 FILEDESC_UNLOCK(fdp);
500 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
501 if (error == 0)
502 td->td_retval[0] = tmp;
503 fdrop(fp, td);
504 break;
505
506 case F_SETOWN:
507 mtx_assert(&Giant, MA_OWNED);
508 fhold(fp);
509 FILEDESC_UNLOCK(fdp);
510 tmp = arg;
511 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
512 fdrop(fp, td);
513 break;
514
515 case F_SETLK_REMOTE:
516 mtx_assert(&Giant, MA_OWNED);
517 error = suser(td);
518 if (error)
519 return (error);
520 flg = F_REMOTE;
521 goto do_setlk;
522
523 case F_SETLKW:
524 mtx_assert(&Giant, MA_OWNED);
525 flg |= F_WAIT;
526 /* FALLTHROUGH F_SETLK */
527
528 case F_SETLK:
529 do_setlk:
530 mtx_assert(&Giant, MA_OWNED);
531 if (fp->f_type != DTYPE_VNODE) {
532 FILEDESC_UNLOCK(fdp);
533 error = EBADF;
534 break;
535 }
536
537 flp = (struct flock *)arg;
538 if (flp->l_whence == SEEK_CUR) {
539 if (fp->f_offset < 0 ||
540 (flp->l_start > 0 &&
541 fp->f_offset > OFF_MAX - flp->l_start)) {
542 FILEDESC_UNLOCK(fdp);
543 error = EOVERFLOW;
544 break;
545 }
546 flp->l_start += fp->f_offset;
547 }
548
549 /*
550 * VOP_ADVLOCK() may block.
551 */
552 fhold(fp);
553 FILEDESC_UNLOCK(fdp);
554 vp = fp->f_vnode;
555
556 switch (flp->l_type) {
557 case F_RDLCK:
558 if ((fp->f_flag & FREAD) == 0) {
559 error = EBADF;
560 break;
561 }
562 PROC_LOCK(p->p_leader);
563 p->p_leader->p_flag |= P_ADVLOCK;
564 PROC_UNLOCK(p->p_leader);
565 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
566 flp, flg);
567 break;
568 case F_WRLCK:
569 if ((fp->f_flag & FWRITE) == 0) {
570 error = EBADF;
571 break;
572 }
573 PROC_LOCK(p->p_leader);
574 p->p_leader->p_flag |= P_ADVLOCK;
575 PROC_UNLOCK(p->p_leader);
576 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
577 flp, flg);
578 break;
579 case F_UNLCK:
580 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
581 flp, flg);
582 break;
583 case F_UNLCKSYS:
584 /*
585 * Temporary api for testing remote lock
586 * infrastructure.
587 */
588 if (flg != F_REMOTE) {
589 error = EINVAL;
590 break;
591 }
592 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
593 F_UNLCKSYS, flp, flg);
594 break;
595 default:
596 error = EINVAL;
597 break;
598 }
599 /* Check for race with close */
600 FILEDESC_LOCK_FAST(fdp);
601 if ((unsigned) fd >= fdp->fd_nfiles ||
602 fp != fdp->fd_ofiles[fd]) {
603 FILEDESC_UNLOCK_FAST(fdp);
604 flp->l_whence = SEEK_SET;
605 flp->l_start = 0;
606 flp->l_len = 0;
607 flp->l_type = F_UNLCK;
608 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
609 F_UNLCK, flp, F_POSIX);
610 } else
611 FILEDESC_UNLOCK_FAST(fdp);
612 fdrop(fp, td);
613 break;
614
615 case F_GETLK:
616 mtx_assert(&Giant, MA_OWNED);
617 if (fp->f_type != DTYPE_VNODE) {
618 FILEDESC_UNLOCK(fdp);
619 error = EBADF;
620 break;
621 }
622 flp = (struct flock *)arg;
623 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
624 flp->l_type != F_UNLCK) {
625 FILEDESC_UNLOCK(fdp);
626 error = EINVAL;
627 break;
628 }
629 if (flp->l_whence == SEEK_CUR) {
630 if ((flp->l_start > 0 &&
631 fp->f_offset > OFF_MAX - flp->l_start) ||
632 (flp->l_start < 0 &&
633 fp->f_offset < OFF_MIN - flp->l_start)) {
634 FILEDESC_UNLOCK(fdp);
635 error = EOVERFLOW;
636 break;
637 }
638 flp->l_start += fp->f_offset;
639 }
640 /*
641 * VOP_ADVLOCK() may block.
642 */
643 fhold(fp);
644 FILEDESC_UNLOCK(fdp);
645 vp = fp->f_vnode;
646 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
647 F_POSIX);
648 fdrop(fp, td);
649 break;
650 default:
651 FILEDESC_UNLOCK(fdp);
652 error = EINVAL;
653 break;
654 }
655 done2:
656 if (giant_locked)
657 mtx_unlock(&Giant);
658 return (error);
659 }
660
661 /*
662 * Common code for dup, dup2, and fcntl(F_DUPFD).
663 */
664 static int
665 do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval)
666 {
667 struct filedesc *fdp;
668 struct proc *p;
669 struct file *fp;
670 struct file *delfp;
671 int error, holdleaders, maxfd;
672
673 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
674 ("invalid dup type %d", type));
675
676 p = td->td_proc;
677 fdp = p->p_fd;
678
679 /*
680 * Verify we have a valid descriptor to dup from and possibly to
681 * dup to.
682 */
683 if (old < 0 || new < 0)
684 return (EBADF);
685 PROC_LOCK(p);
686 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
687 PROC_UNLOCK(p);
688 if (new >= maxfd)
689 return (EMFILE);
690
691 FILEDESC_LOCK(fdp);
692 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
693 FILEDESC_UNLOCK(fdp);
694 return (EBADF);
695 }
696 if (type == DUP_FIXED && old == new) {
697 *retval = new;
698 FILEDESC_UNLOCK(fdp);
699 return (0);
700 }
701 fp = fdp->fd_ofiles[old];
702 fhold(fp);
703
704 /*
705 * If the caller specified a file descriptor, make sure the file
706 * table is large enough to hold it, and grab it. Otherwise, just
707 * allocate a new descriptor the usual way. Since the filedesc
708 * lock may be temporarily dropped in the process, we have to look
709 * out for a race.
710 */
711 if (type == DUP_FIXED) {
712 if (new >= fdp->fd_nfiles)
713 fdgrowtable(fdp, new + 1);
714 if (fdp->fd_ofiles[new] == NULL)
715 fdused(fdp, new);
716 } else {
717 if ((error = fdalloc(td, new, &new)) != 0) {
718 FILEDESC_UNLOCK(fdp);
719 fdrop(fp, td);
720 return (error);
721 }
722 }
723
724 /*
725 * If the old file changed out from under us then treat it as a
726 * bad file descriptor. Userland should do its own locking to
727 * avoid this case.
728 */
729 if (fdp->fd_ofiles[old] != fp ||
730 (fdp->fd_ofileflags[old] & UF_OPENING) != 0 ||
731 (fdp->fd_ofileflags[new] & UF_OPENING) != 0) {
732 /* we've allocated a descriptor which we won't use */
733 if (fdp->fd_ofiles[new] == NULL)
734 fdunused(fdp, new);
735 FILEDESC_UNLOCK(fdp);
736 fdrop(fp, td);
737 return (EBADF);
738 }
739 KASSERT(old != new,
740 ("new fd is same as old"));
741
742 /*
743 * Save info on the descriptor being overwritten. We cannot close
744 * it without introducing an ownership race for the slot, since we
745 * need to drop the filedesc lock to call closef().
746 *
747 * XXX this duplicates parts of close().
748 */
749 delfp = fdp->fd_ofiles[new];
750 holdleaders = 0;
751 if (delfp != NULL) {
752 if (td->td_proc->p_fdtol != NULL) {
753 /*
754 * Ask fdfree() to sleep to ensure that all relevant
755 * process leaders can be traversed in closef().
756 */
757 fdp->fd_holdleaderscount++;
758 holdleaders = 1;
759 }
760 }
761
762 /*
763 * Duplicate the source descriptor
764 */
765 fdp->fd_ofiles[new] = fp;
766 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
767 if (new > fdp->fd_lastfile)
768 fdp->fd_lastfile = new;
769 *retval = new;
770
771 /*
772 * If we dup'd over a valid file, we now own the reference to it
773 * and must dispose of it using closef() semantics (as if a
774 * close() were performed on it).
775 *
776 * XXX this duplicates parts of close().
777 */
778 if (delfp != NULL) {
779 knote_fdclose(td, new);
780 FILEDESC_UNLOCK(fdp);
781 (void) closef(delfp, td);
782 if (holdleaders) {
783 FILEDESC_LOCK_FAST(fdp);
784 fdp->fd_holdleaderscount--;
785 if (fdp->fd_holdleaderscount == 0 &&
786 fdp->fd_holdleaderswakeup != 0) {
787 fdp->fd_holdleaderswakeup = 0;
788 wakeup(&fdp->fd_holdleaderscount);
789 }
790 FILEDESC_UNLOCK_FAST(fdp);
791 }
792 } else {
793 FILEDESC_UNLOCK(fdp);
794 }
795 return (0);
796 }
797
798 /*
799 * If sigio is on the list associated with a process or process group,
800 * disable signalling from the device, remove sigio from the list and
801 * free sigio.
802 */
803 void
804 funsetown(struct sigio **sigiop)
805 {
806 struct sigio *sigio;
807
808 SIGIO_LOCK();
809 sigio = *sigiop;
810 if (sigio == NULL) {
811 SIGIO_UNLOCK();
812 return;
813 }
814 *(sigio->sio_myref) = NULL;
815 if ((sigio)->sio_pgid < 0) {
816 struct pgrp *pg = (sigio)->sio_pgrp;
817 PGRP_LOCK(pg);
818 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
819 sigio, sio_pgsigio);
820 PGRP_UNLOCK(pg);
821 } else {
822 struct proc *p = (sigio)->sio_proc;
823 PROC_LOCK(p);
824 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
825 sigio, sio_pgsigio);
826 PROC_UNLOCK(p);
827 }
828 SIGIO_UNLOCK();
829 crfree(sigio->sio_ucred);
830 FREE(sigio, M_SIGIO);
831 }
832
833 /*
834 * Free a list of sigio structures.
835 * We only need to lock the SIGIO_LOCK because we have made ourselves
836 * inaccessible to callers of fsetown and therefore do not need to lock
837 * the proc or pgrp struct for the list manipulation.
838 */
839 void
840 funsetownlst(struct sigiolst *sigiolst)
841 {
842 struct proc *p;
843 struct pgrp *pg;
844 struct sigio *sigio;
845
846 sigio = SLIST_FIRST(sigiolst);
847 if (sigio == NULL)
848 return;
849 p = NULL;
850 pg = NULL;
851
852 /*
853 * Every entry of the list should belong
854 * to a single proc or pgrp.
855 */
856 if (sigio->sio_pgid < 0) {
857 pg = sigio->sio_pgrp;
858 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
859 } else /* if (sigio->sio_pgid > 0) */ {
860 p = sigio->sio_proc;
861 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
862 }
863
864 SIGIO_LOCK();
865 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
866 *(sigio->sio_myref) = NULL;
867 if (pg != NULL) {
868 KASSERT(sigio->sio_pgid < 0,
869 ("Proc sigio in pgrp sigio list"));
870 KASSERT(sigio->sio_pgrp == pg,
871 ("Bogus pgrp in sigio list"));
872 PGRP_LOCK(pg);
873 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
874 sio_pgsigio);
875 PGRP_UNLOCK(pg);
876 } else /* if (p != NULL) */ {
877 KASSERT(sigio->sio_pgid > 0,
878 ("Pgrp sigio in proc sigio list"));
879 KASSERT(sigio->sio_proc == p,
880 ("Bogus proc in sigio list"));
881 PROC_LOCK(p);
882 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
883 sio_pgsigio);
884 PROC_UNLOCK(p);
885 }
886 SIGIO_UNLOCK();
887 crfree(sigio->sio_ucred);
888 FREE(sigio, M_SIGIO);
889 SIGIO_LOCK();
890 }
891 SIGIO_UNLOCK();
892 }
893
894 /*
895 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
896 *
897 * After permission checking, add a sigio structure to the sigio list for
898 * the process or process group.
899 */
900 int
901 fsetown(pid_t pgid, struct sigio **sigiop)
902 {
903 struct proc *proc;
904 struct pgrp *pgrp;
905 struct sigio *sigio;
906 int ret;
907
908 if (pgid == 0) {
909 funsetown(sigiop);
910 return (0);
911 }
912
913 ret = 0;
914
915 /* Allocate and fill in the new sigio out of locks. */
916 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
917 sigio->sio_pgid = pgid;
918 sigio->sio_ucred = crhold(curthread->td_ucred);
919 sigio->sio_myref = sigiop;
920
921 sx_slock(&proctree_lock);
922 if (pgid > 0) {
923 proc = pfind(pgid);
924 if (proc == NULL) {
925 ret = ESRCH;
926 goto fail;
927 }
928
929 /*
930 * Policy - Don't allow a process to FSETOWN a process
931 * in another session.
932 *
933 * Remove this test to allow maximum flexibility or
934 * restrict FSETOWN to the current process or process
935 * group for maximum safety.
936 */
937 PROC_UNLOCK(proc);
938 if (proc->p_session != curthread->td_proc->p_session) {
939 ret = EPERM;
940 goto fail;
941 }
942
943 pgrp = NULL;
944 } else /* if (pgid < 0) */ {
945 pgrp = pgfind(-pgid);
946 if (pgrp == NULL) {
947 ret = ESRCH;
948 goto fail;
949 }
950 PGRP_UNLOCK(pgrp);
951
952 /*
953 * Policy - Don't allow a process to FSETOWN a process
954 * in another session.
955 *
956 * Remove this test to allow maximum flexibility or
957 * restrict FSETOWN to the current process or process
958 * group for maximum safety.
959 */
960 if (pgrp->pg_session != curthread->td_proc->p_session) {
961 ret = EPERM;
962 goto fail;
963 }
964
965 proc = NULL;
966 }
967 funsetown(sigiop);
968 if (pgid > 0) {
969 PROC_LOCK(proc);
970 /*
971 * Since funsetownlst() is called without the proctree
972 * locked, we need to check for P_WEXIT.
973 * XXX: is ESRCH correct?
974 */
975 if ((proc->p_flag & P_WEXIT) != 0) {
976 PROC_UNLOCK(proc);
977 ret = ESRCH;
978 goto fail;
979 }
980 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
981 sigio->sio_proc = proc;
982 PROC_UNLOCK(proc);
983 } else {
984 PGRP_LOCK(pgrp);
985 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
986 sigio->sio_pgrp = pgrp;
987 PGRP_UNLOCK(pgrp);
988 }
989 sx_sunlock(&proctree_lock);
990 SIGIO_LOCK();
991 *sigiop = sigio;
992 SIGIO_UNLOCK();
993 return (0);
994
995 fail:
996 sx_sunlock(&proctree_lock);
997 crfree(sigio->sio_ucred);
998 FREE(sigio, M_SIGIO);
999 return (ret);
1000 }
1001
1002 /*
1003 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1004 */
1005 pid_t
1006 fgetown(sigiop)
1007 struct sigio **sigiop;
1008 {
1009 pid_t pgid;
1010
1011 SIGIO_LOCK();
1012 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1013 SIGIO_UNLOCK();
1014 return (pgid);
1015 }
1016
1017 /*
1018 * Close a file descriptor.
1019 */
1020 #ifndef _SYS_SYSPROTO_H_
1021 struct close_args {
1022 int fd;
1023 };
1024 #endif
1025 /*
1026 * MPSAFE
1027 */
1028 /* ARGSUSED */
1029 int
1030 close(td, uap)
1031 struct thread *td;
1032 struct close_args *uap;
1033 {
1034
1035 return (kern_close(td, uap->fd));
1036 }
1037
1038 int
1039 kern_close(td, fd)
1040 struct thread *td;
1041 int fd;
1042 {
1043 struct filedesc *fdp;
1044 struct file *fp;
1045 int error;
1046 int holdleaders;
1047
1048 error = 0;
1049 holdleaders = 0;
1050 fdp = td->td_proc->p_fd;
1051
1052 AUDIT_SYSCLOSE(td, fd);
1053
1054 FILEDESC_LOCK(fdp);
1055 if ((unsigned)fd >= fdp->fd_nfiles ||
1056 (fp = fdp->fd_ofiles[fd]) == NULL ||
1057 (fdp->fd_ofileflags[fd] & UF_OPENING) != 0) {
1058 FILEDESC_UNLOCK(fdp);
1059 return (EBADF);
1060 }
1061 fdp->fd_ofiles[fd] = NULL;
1062 fdp->fd_ofileflags[fd] = 0;
1063 fdunused(fdp, fd);
1064 if (td->td_proc->p_fdtol != NULL) {
1065 /*
1066 * Ask fdfree() to sleep to ensure that all relevant
1067 * process leaders can be traversed in closef().
1068 */
1069 fdp->fd_holdleaderscount++;
1070 holdleaders = 1;
1071 }
1072
1073 /*
1074 * We now hold the fp reference that used to be owned by the descriptor
1075 * array.
1076 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a
1077 * race of the fd getting opened, a knote added, and deleteing a knote
1078 * for the new fd.
1079 */
1080 knote_fdclose(td, fd);
1081 FILEDESC_UNLOCK(fdp);
1082
1083 error = closef(fp, td);
1084 if (holdleaders) {
1085 FILEDESC_LOCK_FAST(fdp);
1086 fdp->fd_holdleaderscount--;
1087 if (fdp->fd_holdleaderscount == 0 &&
1088 fdp->fd_holdleaderswakeup != 0) {
1089 fdp->fd_holdleaderswakeup = 0;
1090 wakeup(&fdp->fd_holdleaderscount);
1091 }
1092 FILEDESC_UNLOCK_FAST(fdp);
1093 }
1094 return (error);
1095 }
1096
1097 #if defined(COMPAT_43)
1098 /*
1099 * Return status information about a file descriptor.
1100 */
1101 #ifndef _SYS_SYSPROTO_H_
1102 struct ofstat_args {
1103 int fd;
1104 struct ostat *sb;
1105 };
1106 #endif
1107 /*
1108 * MPSAFE
1109 */
1110 /* ARGSUSED */
1111 int
1112 ofstat(struct thread *td, struct ofstat_args *uap)
1113 {
1114 struct ostat oub;
1115 struct stat ub;
1116 int error;
1117
1118 error = kern_fstat(td, uap->fd, &ub);
1119 if (error == 0) {
1120 cvtstat(&ub, &oub);
1121 error = copyout(&oub, uap->sb, sizeof(oub));
1122 }
1123 return (error);
1124 }
1125 #endif /* COMPAT_43 */
1126
1127 /*
1128 * Return status information about a file descriptor.
1129 */
1130 #ifndef _SYS_SYSPROTO_H_
1131 struct fstat_args {
1132 int fd;
1133 struct stat *sb;
1134 };
1135 #endif
1136 /*
1137 * MPSAFE
1138 */
1139 /* ARGSUSED */
1140 int
1141 fstat(struct thread *td, struct fstat_args *uap)
1142 {
1143 struct stat ub;
1144 int error;
1145
1146 error = kern_fstat(td, uap->fd, &ub);
1147 if (error == 0)
1148 error = copyout(&ub, uap->sb, sizeof(ub));
1149 return (error);
1150 }
1151
1152 int
1153 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1154 {
1155 struct file *fp;
1156 int error;
1157
1158 AUDIT_ARG(fd, fd);
1159
1160 if ((error = fget(td, fd, &fp)) != 0)
1161 return (error);
1162
1163 AUDIT_ARG(file, td->td_proc, fp);
1164
1165 error = fo_stat(fp, sbp, td->td_ucred, td);
1166 fdrop(fp, td);
1167 return (error);
1168 }
1169
1170 /*
1171 * Return status information about a file descriptor.
1172 */
1173 #ifndef _SYS_SYSPROTO_H_
1174 struct nfstat_args {
1175 int fd;
1176 struct nstat *sb;
1177 };
1178 #endif
1179 /*
1180 * MPSAFE
1181 */
1182 /* ARGSUSED */
1183 int
1184 nfstat(struct thread *td, struct nfstat_args *uap)
1185 {
1186 struct nstat nub;
1187 struct stat ub;
1188 int error;
1189
1190 error = kern_fstat(td, uap->fd, &ub);
1191 if (error == 0) {
1192 cvtnstat(&ub, &nub);
1193 error = copyout(&nub, uap->sb, sizeof(nub));
1194 }
1195 return (error);
1196 }
1197
1198 /*
1199 * Return pathconf information about a file descriptor.
1200 */
1201 #ifndef _SYS_SYSPROTO_H_
1202 struct fpathconf_args {
1203 int fd;
1204 int name;
1205 };
1206 #endif
1207 /*
1208 * MPSAFE
1209 */
1210 /* ARGSUSED */
1211 int
1212 fpathconf(struct thread *td, struct fpathconf_args *uap)
1213 {
1214 struct file *fp;
1215 struct vnode *vp;
1216 int error;
1217
1218 if ((error = fget(td, uap->fd, &fp)) != 0)
1219 return (error);
1220
1221 /* If asynchronous I/O is available, it works for all descriptors. */
1222 if (uap->name == _PC_ASYNC_IO) {
1223 td->td_retval[0] = async_io_version;
1224 goto out;
1225 }
1226 vp = fp->f_vnode;
1227 if (vp != NULL) {
1228 int vfslocked;
1229 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1230 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1231 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1232 VOP_UNLOCK(vp, 0, td);
1233 VFS_UNLOCK_GIANT(vfslocked);
1234 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1235 if (uap->name != _PC_PIPE_BUF) {
1236 error = EINVAL;
1237 } else {
1238 td->td_retval[0] = PIPE_BUF;
1239 error = 0;
1240 }
1241 } else {
1242 error = EOPNOTSUPP;
1243 }
1244 out:
1245 fdrop(fp, td);
1246 return (error);
1247 }
1248
1249 /*
1250 * Grow the file table to accomodate (at least) nfd descriptors. This may
1251 * block and drop the filedesc lock, but it will reacquire it before
1252 * returning.
1253 */
1254 static void
1255 fdgrowtable(struct filedesc *fdp, int nfd)
1256 {
1257 struct file **ntable;
1258 char *nfileflags;
1259 int nnfiles, onfiles;
1260 NDSLOTTYPE *nmap;
1261
1262 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1263
1264 KASSERT(fdp->fd_nfiles > 0,
1265 ("zero-length file table"));
1266
1267 /* compute the size of the new table */
1268 onfiles = fdp->fd_nfiles;
1269 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1270 if (nnfiles <= onfiles)
1271 /* the table is already large enough */
1272 return;
1273
1274 /* allocate a new table and (if required) new bitmaps */
1275 FILEDESC_UNLOCK(fdp);
1276 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1277 M_FILEDESC, M_ZERO | M_WAITOK);
1278 nfileflags = (char *)&ntable[nnfiles];
1279 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1280 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1281 M_FILEDESC, M_ZERO | M_WAITOK);
1282 else
1283 nmap = NULL;
1284 FILEDESC_LOCK(fdp);
1285
1286 /*
1287 * We now have new tables ready to go. Since we dropped the
1288 * filedesc lock to call malloc(), watch out for a race.
1289 */
1290 onfiles = fdp->fd_nfiles;
1291 if (onfiles >= nnfiles) {
1292 /* we lost the race, but that's OK */
1293 free(ntable, M_FILEDESC);
1294 if (nmap != NULL)
1295 free(nmap, M_FILEDESC);
1296 return;
1297 }
1298 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1299 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1300 if (onfiles > NDFILE)
1301 free(fdp->fd_ofiles, M_FILEDESC);
1302 fdp->fd_ofiles = ntable;
1303 fdp->fd_ofileflags = nfileflags;
1304 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1305 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1306 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1307 free(fdp->fd_map, M_FILEDESC);
1308 fdp->fd_map = nmap;
1309 }
1310 fdp->fd_nfiles = nnfiles;
1311 }
1312
1313 /*
1314 * Allocate a file descriptor for the process.
1315 */
1316 int
1317 fdalloc(struct thread *td, int minfd, int *result)
1318 {
1319 struct proc *p = td->td_proc;
1320 struct filedesc *fdp = p->p_fd;
1321 int fd = -1, maxfd;
1322
1323 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1324
1325 if (fdp->fd_freefile > minfd)
1326 minfd = fdp->fd_freefile;
1327
1328 PROC_LOCK(p);
1329 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1330 PROC_UNLOCK(p);
1331
1332 /*
1333 * Search the bitmap for a free descriptor. If none is found, try
1334 * to grow the file table. Keep at it until we either get a file
1335 * descriptor or run into process or system limits; fdgrowtable()
1336 * may drop the filedesc lock, so we're in a race.
1337 */
1338 for (;;) {
1339 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1340 if (fd >= maxfd)
1341 return (EMFILE);
1342 if (fd < fdp->fd_nfiles)
1343 break;
1344 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1345 }
1346
1347 /*
1348 * Perform some sanity checks, then mark the file descriptor as
1349 * used and return it to the caller.
1350 */
1351 KASSERT(!fdisused(fdp, fd),
1352 ("fd_first_free() returned non-free descriptor"));
1353 KASSERT(fdp->fd_ofiles[fd] == NULL,
1354 ("free descriptor isn't"));
1355 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1356 fdused(fdp, fd);
1357 *result = fd;
1358 return (0);
1359 }
1360
1361 /*
1362 * Check to see whether n user file descriptors
1363 * are available to the process p.
1364 */
1365 int
1366 fdavail(struct thread *td, int n)
1367 {
1368 struct proc *p = td->td_proc;
1369 struct filedesc *fdp = td->td_proc->p_fd;
1370 struct file **fpp;
1371 int i, lim, last;
1372
1373 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1374
1375 PROC_LOCK(p);
1376 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1377 PROC_UNLOCK(p);
1378 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1379 return (1);
1380 last = min(fdp->fd_nfiles, lim);
1381 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1382 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1383 if (*fpp == NULL && --n <= 0)
1384 return (1);
1385 }
1386 return (0);
1387 }
1388
1389 /*
1390 * Create a new open file structure and allocate
1391 * a file decriptor for the process that refers to it.
1392 * We add one reference to the file for the descriptor table
1393 * and one reference for resultfp. This is to prevent us being
1394 * preempted and the entry in the descriptor table closed after
1395 * we release the FILEDESC lock.
1396 */
1397 int
1398 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1399 {
1400 struct proc *p = td->td_proc;
1401 struct file *fp, *fq;
1402 int error, i;
1403 int maxuserfiles = maxfiles - (maxfiles / 20);
1404 static struct timeval lastfail;
1405 static int curfail;
1406
1407 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1408 sx_xlock(&filelist_lock);
1409
1410 if ((openfiles >= maxuserfiles &&
1411 suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
1412 openfiles >= maxfiles) {
1413 if (ppsratecheck(&lastfail, &curfail, 1)) {
1414 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1415 td->td_ucred->cr_ruid);
1416 }
1417 sx_xunlock(&filelist_lock);
1418 uma_zfree(file_zone, fp);
1419 return (ENFILE);
1420 }
1421 openfiles++;
1422
1423 /*
1424 * If the process has file descriptor zero open, add the new file
1425 * descriptor to the list of open files at that point, otherwise
1426 * put it at the front of the list of open files.
1427 */
1428 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1429 fp->f_count = 1;
1430 if (resultfp)
1431 fp->f_count++;
1432 fp->f_cred = crhold(td->td_ucred);
1433 fp->f_ops = &badfileops;
1434 fp->f_data = NULL;
1435 fp->f_vnode = NULL;
1436 FILEDESC_LOCK(p->p_fd);
1437 if ((fq = p->p_fd->fd_ofiles[0])) {
1438 LIST_INSERT_AFTER(fq, fp, f_list);
1439 } else {
1440 LIST_INSERT_HEAD(&filehead, fp, f_list);
1441 }
1442 sx_xunlock(&filelist_lock);
1443 if ((error = fdalloc(td, 0, &i))) {
1444 FILEDESC_UNLOCK(p->p_fd);
1445 fdrop(fp, td);
1446 if (resultfp)
1447 fdrop(fp, td);
1448 return (error);
1449 }
1450 p->p_fd->fd_ofiles[i] = fp;
1451 FILEDESC_UNLOCK(p->p_fd);
1452 if (resultfp)
1453 *resultfp = fp;
1454 if (resultfd)
1455 *resultfd = i;
1456 return (0);
1457 }
1458
1459 /*
1460 * Build a new filedesc structure from another.
1461 * Copy the current, root, and jail root vnode references.
1462 */
1463 struct filedesc *
1464 fdinit(struct filedesc *fdp)
1465 {
1466 struct filedesc0 *newfdp;
1467
1468 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1469 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1470 if (fdp != NULL) {
1471 FILEDESC_LOCK(fdp);
1472 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1473 if (newfdp->fd_fd.fd_cdir)
1474 VREF(newfdp->fd_fd.fd_cdir);
1475 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1476 if (newfdp->fd_fd.fd_rdir)
1477 VREF(newfdp->fd_fd.fd_rdir);
1478 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1479 if (newfdp->fd_fd.fd_jdir)
1480 VREF(newfdp->fd_fd.fd_jdir);
1481 FILEDESC_UNLOCK(fdp);
1482 }
1483
1484 /* Create the file descriptor table. */
1485 newfdp->fd_fd.fd_refcnt = 1;
1486 newfdp->fd_fd.fd_holdcnt = 1;
1487 newfdp->fd_fd.fd_cmask = CMASK;
1488 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1489 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1490 newfdp->fd_fd.fd_nfiles = NDFILE;
1491 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1492 newfdp->fd_fd.fd_lastfile = -1;
1493 return (&newfdp->fd_fd);
1494 }
1495
1496 static struct filedesc *
1497 fdhold(struct proc *p)
1498 {
1499 struct filedesc *fdp;
1500
1501 mtx_lock(&fdesc_mtx);
1502 fdp = p->p_fd;
1503 if (fdp != NULL)
1504 fdp->fd_holdcnt++;
1505 mtx_unlock(&fdesc_mtx);
1506 return (fdp);
1507 }
1508
1509 static void
1510 fddrop(struct filedesc *fdp)
1511 {
1512 int i;
1513
1514 mtx_lock(&fdesc_mtx);
1515 i = --fdp->fd_holdcnt;
1516 mtx_unlock(&fdesc_mtx);
1517 if (i > 0)
1518 return;
1519
1520 mtx_destroy(&fdp->fd_mtx);
1521 FREE(fdp, M_FILEDESC);
1522 }
1523
1524 /*
1525 * Share a filedesc structure.
1526 */
1527 struct filedesc *
1528 fdshare(struct filedesc *fdp)
1529 {
1530 FILEDESC_LOCK_FAST(fdp);
1531 fdp->fd_refcnt++;
1532 FILEDESC_UNLOCK_FAST(fdp);
1533 return (fdp);
1534 }
1535
1536 /*
1537 * Unshare a filedesc structure, if necessary by making a copy
1538 */
1539 void
1540 fdunshare(struct proc *p, struct thread *td)
1541 {
1542
1543 FILEDESC_LOCK_FAST(p->p_fd);
1544 if (p->p_fd->fd_refcnt > 1) {
1545 struct filedesc *tmp;
1546
1547 FILEDESC_UNLOCK_FAST(p->p_fd);
1548 tmp = fdcopy(p->p_fd);
1549 fdfree(td);
1550 p->p_fd = tmp;
1551 } else
1552 FILEDESC_UNLOCK_FAST(p->p_fd);
1553 }
1554
1555 /*
1556 * Copy a filedesc structure.
1557 * A NULL pointer in returns a NULL reference, this is to ease callers,
1558 * not catch errors.
1559 */
1560 struct filedesc *
1561 fdcopy(struct filedesc *fdp)
1562 {
1563 struct filedesc *newfdp;
1564 int i;
1565
1566 /* Certain daemons might not have file descriptors. */
1567 if (fdp == NULL)
1568 return (NULL);
1569
1570 newfdp = fdinit(fdp);
1571 FILEDESC_LOCK_FAST(fdp);
1572 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1573 FILEDESC_UNLOCK_FAST(fdp);
1574 FILEDESC_LOCK(newfdp);
1575 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1576 FILEDESC_UNLOCK(newfdp);
1577 FILEDESC_LOCK_FAST(fdp);
1578 }
1579 /* copy everything except kqueue descriptors */
1580 newfdp->fd_freefile = -1;
1581 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1582 if (fdisused(fdp, i) &&
1583 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE &&
1584 (fdp->fd_ofileflags[i] & UF_OPENING) == 0) {
1585 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1586 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1587 fhold(newfdp->fd_ofiles[i]);
1588 newfdp->fd_lastfile = i;
1589 } else {
1590 if (newfdp->fd_freefile == -1)
1591 newfdp->fd_freefile = i;
1592 }
1593 }
1594 FILEDESC_UNLOCK_FAST(fdp);
1595 FILEDESC_LOCK(newfdp);
1596 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1597 if (newfdp->fd_ofiles[i] != NULL)
1598 fdused(newfdp, i);
1599 FILEDESC_UNLOCK(newfdp);
1600 FILEDESC_LOCK_FAST(fdp);
1601 if (newfdp->fd_freefile == -1)
1602 newfdp->fd_freefile = i;
1603 newfdp->fd_cmask = fdp->fd_cmask;
1604 FILEDESC_UNLOCK_FAST(fdp);
1605 return (newfdp);
1606 }
1607
1608 /*
1609 * Release a filedesc structure.
1610 */
1611 void
1612 fdfree(struct thread *td)
1613 {
1614 struct filedesc *fdp;
1615 struct file **fpp;
1616 int i, locked;
1617 struct filedesc_to_leader *fdtol;
1618 struct file *fp;
1619 struct vnode *cdir, *jdir, *rdir, *vp;
1620 struct flock lf;
1621
1622 /* Certain daemons might not have file descriptors. */
1623 fdp = td->td_proc->p_fd;
1624 if (fdp == NULL)
1625 return;
1626
1627 /* Check for special need to clear POSIX style locks */
1628 fdtol = td->td_proc->p_fdtol;
1629 if (fdtol != NULL) {
1630 FILEDESC_LOCK(fdp);
1631 KASSERT(fdtol->fdl_refcount > 0,
1632 ("filedesc_to_refcount botch: fdl_refcount=%d",
1633 fdtol->fdl_refcount));
1634 if (fdtol->fdl_refcount == 1 &&
1635 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1636 for (i = 0, fpp = fdp->fd_ofiles;
1637 i <= fdp->fd_lastfile;
1638 i++, fpp++) {
1639 if (*fpp == NULL ||
1640 (*fpp)->f_type != DTYPE_VNODE)
1641 continue;
1642 fp = *fpp;
1643 fhold(fp);
1644 FILEDESC_UNLOCK(fdp);
1645 lf.l_whence = SEEK_SET;
1646 lf.l_start = 0;
1647 lf.l_len = 0;
1648 lf.l_type = F_UNLCK;
1649 vp = fp->f_vnode;
1650 locked = VFS_LOCK_GIANT(vp->v_mount);
1651 (void) VOP_ADVLOCK(vp,
1652 (caddr_t)td->td_proc->
1653 p_leader,
1654 F_UNLCK,
1655 &lf,
1656 F_POSIX);
1657 VFS_UNLOCK_GIANT(locked);
1658 FILEDESC_LOCK(fdp);
1659 fdrop(fp, td);
1660 fpp = fdp->fd_ofiles + i;
1661 }
1662 }
1663 retry:
1664 if (fdtol->fdl_refcount == 1) {
1665 if (fdp->fd_holdleaderscount > 0 &&
1666 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1667 /*
1668 * close() or do_dup() has cleared a reference
1669 * in a shared file descriptor table.
1670 */
1671 fdp->fd_holdleaderswakeup = 1;
1672 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1673 PLOCK, "fdlhold", 0);
1674 goto retry;
1675 }
1676 if (fdtol->fdl_holdcount > 0) {
1677 /*
1678 * Ensure that fdtol->fdl_leader
1679 * remains valid in closef().
1680 */
1681 fdtol->fdl_wakeup = 1;
1682 msleep(fdtol, &fdp->fd_mtx,
1683 PLOCK, "fdlhold", 0);
1684 goto retry;
1685 }
1686 }
1687 fdtol->fdl_refcount--;
1688 if (fdtol->fdl_refcount == 0 &&
1689 fdtol->fdl_holdcount == 0) {
1690 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1691 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1692 } else
1693 fdtol = NULL;
1694 td->td_proc->p_fdtol = NULL;
1695 FILEDESC_UNLOCK(fdp);
1696 if (fdtol != NULL)
1697 FREE(fdtol, M_FILEDESC_TO_LEADER);
1698 }
1699 FILEDESC_LOCK_FAST(fdp);
1700 i = --fdp->fd_refcnt;
1701 FILEDESC_UNLOCK_FAST(fdp);
1702 if (i > 0)
1703 return;
1704 /*
1705 * We are the last reference to the structure, so we can
1706 * safely assume it will not change out from under us.
1707 */
1708 fpp = fdp->fd_ofiles;
1709 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1710 if (*fpp)
1711 (void) closef(*fpp, td);
1712 }
1713 FILEDESC_LOCK(fdp);
1714
1715 /* XXX This should happen earlier. */
1716 mtx_lock(&fdesc_mtx);
1717 td->td_proc->p_fd = NULL;
1718 mtx_unlock(&fdesc_mtx);
1719
1720 if (fdp->fd_nfiles > NDFILE)
1721 FREE(fdp->fd_ofiles, M_FILEDESC);
1722 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1723 FREE(fdp->fd_map, M_FILEDESC);
1724
1725 fdp->fd_nfiles = 0;
1726
1727 cdir = fdp->fd_cdir;
1728 fdp->fd_cdir = NULL;
1729 rdir = fdp->fd_rdir;
1730 fdp->fd_rdir = NULL;
1731 jdir = fdp->fd_jdir;
1732 fdp->fd_jdir = NULL;
1733 FILEDESC_UNLOCK(fdp);
1734
1735 if (cdir) {
1736 locked = VFS_LOCK_GIANT(cdir->v_mount);
1737 vrele(cdir);
1738 VFS_UNLOCK_GIANT(locked);
1739 }
1740 if (rdir) {
1741 locked = VFS_LOCK_GIANT(rdir->v_mount);
1742 vrele(rdir);
1743 VFS_UNLOCK_GIANT(locked);
1744 }
1745 if (jdir) {
1746 locked = VFS_LOCK_GIANT(jdir->v_mount);
1747 vrele(jdir);
1748 VFS_UNLOCK_GIANT(locked);
1749 }
1750
1751 fddrop(fdp);
1752 }
1753
1754 /*
1755 * For setugid programs, we don't want to people to use that setugidness
1756 * to generate error messages which write to a file which otherwise would
1757 * otherwise be off-limits to the process. We check for filesystems where
1758 * the vnode can change out from under us after execve (like [lin]procfs).
1759 *
1760 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1761 * sufficient. We also don't check for setugidness since we know we are.
1762 */
1763 static int
1764 is_unsafe(struct file *fp)
1765 {
1766 if (fp->f_type == DTYPE_VNODE) {
1767 struct vnode *vp = fp->f_vnode;
1768
1769 if ((vp->v_vflag & VV_PROCDEP) != 0)
1770 return (1);
1771 }
1772 return (0);
1773 }
1774
1775 /*
1776 * Make this setguid thing safe, if at all possible.
1777 */
1778 void
1779 setugidsafety(struct thread *td)
1780 {
1781 struct filedesc *fdp;
1782 int i;
1783
1784 /* Certain daemons might not have file descriptors. */
1785 fdp = td->td_proc->p_fd;
1786 if (fdp == NULL)
1787 return;
1788
1789 /*
1790 * Note: fdp->fd_ofiles may be reallocated out from under us while
1791 * we are blocked in a close. Be careful!
1792 */
1793 FILEDESC_LOCK(fdp);
1794 for (i = 0; i <= fdp->fd_lastfile; i++) {
1795 if (i > 2)
1796 break;
1797 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1798 struct file *fp;
1799
1800 knote_fdclose(td, i);
1801 /*
1802 * NULL-out descriptor prior to close to avoid
1803 * a race while close blocks.
1804 */
1805 fp = fdp->fd_ofiles[i];
1806 fdp->fd_ofiles[i] = NULL;
1807 fdp->fd_ofileflags[i] = 0;
1808 fdunused(fdp, i);
1809 FILEDESC_UNLOCK(fdp);
1810 (void) closef(fp, td);
1811 FILEDESC_LOCK(fdp);
1812 }
1813 }
1814 FILEDESC_UNLOCK(fdp);
1815 }
1816
1817 void
1818 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1819 {
1820
1821 FILEDESC_LOCK(fdp);
1822 if (fdp->fd_ofiles[idx] == fp) {
1823 fdp->fd_ofiles[idx] = NULL;
1824 fdunused(fdp, idx);
1825 FILEDESC_UNLOCK(fdp);
1826 fdrop(fp, td);
1827 } else {
1828 FILEDESC_UNLOCK(fdp);
1829 }
1830 }
1831
1832 /*
1833 * Close any files on exec?
1834 */
1835 void
1836 fdcloseexec(struct thread *td)
1837 {
1838 struct filedesc *fdp;
1839 int i;
1840
1841 /* Certain daemons might not have file descriptors. */
1842 fdp = td->td_proc->p_fd;
1843 if (fdp == NULL)
1844 return;
1845
1846 FILEDESC_LOCK(fdp);
1847
1848 /*
1849 * We cannot cache fd_ofiles or fd_ofileflags since operations
1850 * may block and rip them out from under us.
1851 */
1852 for (i = 0; i <= fdp->fd_lastfile; i++) {
1853 if (fdp->fd_ofiles[i] != NULL &&
1854 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1855 struct file *fp;
1856
1857 knote_fdclose(td, i);
1858 /*
1859 * NULL-out descriptor prior to close to avoid
1860 * a race while close blocks.
1861 */
1862 fp = fdp->fd_ofiles[i];
1863 fdp->fd_ofiles[i] = NULL;
1864 fdp->fd_ofileflags[i] = 0;
1865 fdunused(fdp, i);
1866 FILEDESC_UNLOCK(fdp);
1867 (void) closef(fp, td);
1868 FILEDESC_LOCK(fdp);
1869 }
1870 }
1871 FILEDESC_UNLOCK(fdp);
1872 }
1873
1874 /*
1875 * It is unsafe for set[ug]id processes to be started with file
1876 * descriptors 0..2 closed, as these descriptors are given implicit
1877 * significance in the Standard C library. fdcheckstd() will create a
1878 * descriptor referencing /dev/null for each of stdin, stdout, and
1879 * stderr that is not already open.
1880 */
1881 int
1882 fdcheckstd(struct thread *td)
1883 {
1884 struct filedesc *fdp;
1885 register_t retval, save;
1886 int i, error, devnull;
1887
1888 fdp = td->td_proc->p_fd;
1889 if (fdp == NULL)
1890 return (0);
1891 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1892 devnull = -1;
1893 error = 0;
1894 for (i = 0; i < 3; i++) {
1895 if (fdp->fd_ofiles[i] != NULL)
1896 continue;
1897 if (devnull < 0) {
1898 save = td->td_retval[0];
1899 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
1900 O_RDWR, 0);
1901 devnull = td->td_retval[0];
1902 KASSERT(devnull == i, ("oof, we didn't get our fd"));
1903 td->td_retval[0] = save;
1904 if (error)
1905 break;
1906 } else {
1907 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1908 if (error != 0)
1909 break;
1910 }
1911 }
1912 return (error);
1913 }
1914
1915 /*
1916 * Internal form of close.
1917 * Decrement reference count on file structure.
1918 * Note: td may be NULL when closing a file that was being passed in a
1919 * message.
1920 *
1921 * XXXRW: Giant is not required for the caller, but often will be held; this
1922 * makes it moderately likely the Giant will be recursed in the VFS case.
1923 */
1924 int
1925 closef(struct file *fp, struct thread *td)
1926 {
1927 struct vnode *vp;
1928 struct flock lf;
1929 struct filedesc_to_leader *fdtol;
1930 struct filedesc *fdp;
1931
1932 /*
1933 * POSIX record locking dictates that any close releases ALL
1934 * locks owned by this process. This is handled by setting
1935 * a flag in the unlock to free ONLY locks obeying POSIX
1936 * semantics, and not to free BSD-style file locks.
1937 * If the descriptor was in a message, POSIX-style locks
1938 * aren't passed with the descriptor, and the thread pointer
1939 * will be NULL. Callers should be careful only to pass a
1940 * NULL thread pointer when there really is no owning
1941 * context that might have locks, or the locks will be
1942 * leaked.
1943 */
1944 if (fp->f_type == DTYPE_VNODE && td != NULL) {
1945 int vfslocked;
1946
1947 vp = fp->f_vnode;
1948 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1949 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1950 lf.l_whence = SEEK_SET;
1951 lf.l_start = 0;
1952 lf.l_len = 0;
1953 lf.l_type = F_UNLCK;
1954 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1955 F_UNLCK, &lf, F_POSIX);
1956 }
1957 fdtol = td->td_proc->p_fdtol;
1958 if (fdtol != NULL) {
1959 /*
1960 * Handle special case where file descriptor table
1961 * is shared between multiple process leaders.
1962 */
1963 fdp = td->td_proc->p_fd;
1964 FILEDESC_LOCK(fdp);
1965 for (fdtol = fdtol->fdl_next;
1966 fdtol != td->td_proc->p_fdtol;
1967 fdtol = fdtol->fdl_next) {
1968 if ((fdtol->fdl_leader->p_flag &
1969 P_ADVLOCK) == 0)
1970 continue;
1971 fdtol->fdl_holdcount++;
1972 FILEDESC_UNLOCK(fdp);
1973 lf.l_whence = SEEK_SET;
1974 lf.l_start = 0;
1975 lf.l_len = 0;
1976 lf.l_type = F_UNLCK;
1977 vp = fp->f_vnode;
1978 (void) VOP_ADVLOCK(vp,
1979 (caddr_t)fdtol->fdl_leader,
1980 F_UNLCK, &lf, F_POSIX);
1981 FILEDESC_LOCK(fdp);
1982 fdtol->fdl_holdcount--;
1983 if (fdtol->fdl_holdcount == 0 &&
1984 fdtol->fdl_wakeup != 0) {
1985 fdtol->fdl_wakeup = 0;
1986 wakeup(fdtol);
1987 }
1988 }
1989 FILEDESC_UNLOCK(fdp);
1990 }
1991 VFS_UNLOCK_GIANT(vfslocked);
1992 }
1993 return (fdrop(fp, td));
1994 }
1995
1996 /*
1997 * Extract the file pointer associated with the specified descriptor for
1998 * the current user process.
1999 *
2000 * If the descriptor doesn't exist, EBADF is returned.
2001 *
2002 * If the descriptor exists but doesn't match 'flags' then
2003 * return EBADF for read attempts and EINVAL for write attempts.
2004 *
2005 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
2006 * It should be dropped with fdrop().
2007 * If it is not set, then the refcount will not be bumped however the
2008 * thread's filedesc struct will be returned locked (for fgetsock).
2009 *
2010 * If an error occured the non-zero error is returned and *fpp is set to NULL.
2011 * Otherwise *fpp is set and zero is returned.
2012 */
2013 static __inline int
2014 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
2015 {
2016 struct filedesc *fdp;
2017 struct file *fp;
2018
2019 *fpp = NULL;
2020 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2021 return (EBADF);
2022 FILEDESC_LOCK(fdp);
2023 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
2024 FILEDESC_UNLOCK(fdp);
2025 return (EBADF);
2026 }
2027
2028 /*
2029 * Note: FREAD failure returns EBADF to maintain backwards
2030 * compatibility with what routines returned before.
2031 *
2032 * Only one flag, or 0, may be specified.
2033 */
2034 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2035 FILEDESC_UNLOCK(fdp);
2036 return (EBADF);
2037 }
2038 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2039 FILEDESC_UNLOCK(fdp);
2040 return (EINVAL);
2041 }
2042 if (hold) {
2043 fhold(fp);
2044 FILEDESC_UNLOCK(fdp);
2045 }
2046 *fpp = fp;
2047 return (0);
2048 }
2049
2050 int
2051 fget(struct thread *td, int fd, struct file **fpp)
2052 {
2053
2054 return(_fget(td, fd, fpp, 0, 1));
2055 }
2056
2057 int
2058 fget_read(struct thread *td, int fd, struct file **fpp)
2059 {
2060
2061 return(_fget(td, fd, fpp, FREAD, 1));
2062 }
2063
2064 int
2065 fget_write(struct thread *td, int fd, struct file **fpp)
2066 {
2067
2068 return(_fget(td, fd, fpp, FWRITE, 1));
2069 }
2070
2071 /*
2072 * Like fget() but loads the underlying vnode, or returns an error if
2073 * the descriptor does not represent a vnode. Note that pipes use vnodes
2074 * but never have VM objects. The returned vnode will be vref()d.
2075 *
2076 * XXX: what about the unused flags ?
2077 */
2078 static __inline int
2079 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2080 {
2081 struct file *fp;
2082 int error;
2083
2084 *vpp = NULL;
2085 if ((error = _fget(td, fd, &fp, flags, 0)) != 0)
2086 return (error);
2087 if (fp->f_vnode == NULL) {
2088 error = EINVAL;
2089 } else {
2090 *vpp = fp->f_vnode;
2091 vref(*vpp);
2092 }
2093 FILEDESC_UNLOCK(td->td_proc->p_fd);
2094 return (error);
2095 }
2096
2097 int
2098 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2099 {
2100
2101 return (_fgetvp(td, fd, vpp, 0));
2102 }
2103
2104 int
2105 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2106 {
2107
2108 return (_fgetvp(td, fd, vpp, FREAD));
2109 }
2110
2111 #ifdef notyet
2112 int
2113 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2114 {
2115
2116 return (_fgetvp(td, fd, vpp, FWRITE));
2117 }
2118 #endif
2119
2120 /*
2121 * Like fget() but loads the underlying socket, or returns an error if
2122 * the descriptor does not represent a socket.
2123 *
2124 * We bump the ref count on the returned socket. XXX Also obtain the SX
2125 * lock in the future.
2126 */
2127 int
2128 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2129 {
2130 struct file *fp;
2131 int error;
2132
2133 NET_ASSERT_GIANT();
2134
2135 *spp = NULL;
2136 if (fflagp != NULL)
2137 *fflagp = 0;
2138 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2139 return (error);
2140 if (fp->f_type != DTYPE_SOCKET) {
2141 error = ENOTSOCK;
2142 } else {
2143 *spp = fp->f_data;
2144 if (fflagp)
2145 *fflagp = fp->f_flag;
2146 SOCK_LOCK(*spp);
2147 soref(*spp);
2148 SOCK_UNLOCK(*spp);
2149 }
2150 FILEDESC_UNLOCK(td->td_proc->p_fd);
2151 return (error);
2152 }
2153
2154 /*
2155 * Drop the reference count on the socket and XXX release the SX lock in
2156 * the future. The last reference closes the socket.
2157 */
2158 void
2159 fputsock(struct socket *so)
2160 {
2161
2162 NET_ASSERT_GIANT();
2163 ACCEPT_LOCK();
2164 SOCK_LOCK(so);
2165 sorele(so);
2166 }
2167
2168 int
2169 fdrop(struct file *fp, struct thread *td)
2170 {
2171
2172 FILE_LOCK(fp);
2173 return (fdrop_locked(fp, td));
2174 }
2175
2176 /*
2177 * Drop reference on struct file passed in, may call closef if the
2178 * reference hits zero.
2179 * Expects struct file locked, and will unlock it.
2180 */
2181 static int
2182 fdrop_locked(struct file *fp, struct thread *td)
2183 {
2184 int error;
2185
2186 FILE_LOCK_ASSERT(fp, MA_OWNED);
2187
2188 if (--fp->f_count > 0) {
2189 FILE_UNLOCK(fp);
2190 return (0);
2191 }
2192
2193 /*
2194 * We might have just dropped the last reference to a file
2195 * object that is for a UNIX domain socket whose message
2196 * buffers are being examined in unp_gc(). If that is the
2197 * case, FWAIT will be set in f_gcflag and we need to wait for
2198 * unp_gc() to finish its scan.
2199 */
2200 while (fp->f_gcflag & FWAIT)
2201 msleep(&fp->f_gcflag, fp->f_mtxp, 0, "fpdrop", 0);
2202
2203 /* We have the last ref so we can proceed without the file lock. */
2204 FILE_UNLOCK(fp);
2205 if (fp->f_count < 0)
2206 panic("fdrop: count < 0");
2207 if (fp->f_ops != &badfileops)
2208 error = fo_close(fp, td);
2209 else
2210 error = 0;
2211
2212 sx_xlock(&filelist_lock);
2213 LIST_REMOVE(fp, f_list);
2214 openfiles--;
2215 sx_xunlock(&filelist_lock);
2216 crfree(fp->f_cred);
2217 uma_zfree(file_zone, fp);
2218
2219 return (error);
2220 }
2221
2222 /*
2223 * Apply an advisory lock on a file descriptor.
2224 *
2225 * Just attempt to get a record lock of the requested type on
2226 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2227 */
2228 #ifndef _SYS_SYSPROTO_H_
2229 struct flock_args {
2230 int fd;
2231 int how;
2232 };
2233 #endif
2234 /*
2235 * MPSAFE
2236 */
2237 /* ARGSUSED */
2238 int
2239 flock(struct thread *td, struct flock_args *uap)
2240 {
2241 struct file *fp;
2242 struct vnode *vp;
2243 struct flock lf;
2244 int error;
2245
2246 if ((error = fget(td, uap->fd, &fp)) != 0)
2247 return (error);
2248 if (fp->f_type != DTYPE_VNODE) {
2249 fdrop(fp, td);
2250 return (EOPNOTSUPP);
2251 }
2252
2253 mtx_lock(&Giant);
2254 vp = fp->f_vnode;
2255 lf.l_whence = SEEK_SET;
2256 lf.l_start = 0;
2257 lf.l_len = 0;
2258 if (uap->how & LOCK_UN) {
2259 lf.l_type = F_UNLCK;
2260 FILE_LOCK(fp);
2261 fp->f_flag &= ~FHASLOCK;
2262 FILE_UNLOCK(fp);
2263 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2264 goto done2;
2265 }
2266 if (uap->how & LOCK_EX)
2267 lf.l_type = F_WRLCK;
2268 else if (uap->how & LOCK_SH)
2269 lf.l_type = F_RDLCK;
2270 else {
2271 error = EBADF;
2272 goto done2;
2273 }
2274 FILE_LOCK(fp);
2275 fp->f_flag |= FHASLOCK;
2276 FILE_UNLOCK(fp);
2277 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2278 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2279 done2:
2280 fdrop(fp, td);
2281 mtx_unlock(&Giant);
2282 return (error);
2283 }
2284 /*
2285 * Duplicate the specified descriptor to a free descriptor.
2286 */
2287 int
2288 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2289 {
2290 struct file *wfp;
2291 struct file *fp;
2292
2293 /*
2294 * If the to-be-dup'd fd number is greater than the allowed number
2295 * of file descriptors, or the fd to be dup'd has already been
2296 * closed, then reject.
2297 */
2298 FILEDESC_LOCK(fdp);
2299 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2300 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2301 FILEDESC_UNLOCK(fdp);
2302 return (EBADF);
2303 }
2304
2305 /*
2306 * There are two cases of interest here.
2307 *
2308 * For ENODEV simply dup (dfd) to file descriptor
2309 * (indx) and return.
2310 *
2311 * For ENXIO steal away the file structure from (dfd) and
2312 * store it in (indx). (dfd) is effectively closed by
2313 * this operation.
2314 *
2315 * Any other error code is just returned.
2316 */
2317 switch (error) {
2318 case ENODEV:
2319 /*
2320 * Check that the mode the file is being opened for is a
2321 * subset of the mode of the existing descriptor.
2322 */
2323 FILE_LOCK(wfp);
2324 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2325 FILE_UNLOCK(wfp);
2326 FILEDESC_UNLOCK(fdp);
2327 return (EACCES);
2328 }
2329 fp = fdp->fd_ofiles[indx];
2330 fdp->fd_ofiles[indx] = wfp;
2331 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2332 if (fp == NULL)
2333 fdused(fdp, indx);
2334 fhold_locked(wfp);
2335 FILE_UNLOCK(wfp);
2336 FILEDESC_UNLOCK(fdp);
2337 if (fp != NULL) {
2338 /*
2339 * We now own the reference to fp that the ofiles[]
2340 * array used to own. Release it.
2341 */
2342 FILE_LOCK(fp);
2343 fdrop_locked(fp, td);
2344 }
2345 return (0);
2346
2347 case ENXIO:
2348 /*
2349 * Steal away the file pointer from dfd and stuff it into indx.
2350 */
2351 fp = fdp->fd_ofiles[indx];
2352 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2353 fdp->fd_ofiles[dfd] = NULL;
2354 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2355 fdp->fd_ofileflags[dfd] = 0;
2356 fdunused(fdp, dfd);
2357 if (fp == NULL)
2358 fdused(fdp, indx);
2359 if (fp != NULL)
2360 FILE_LOCK(fp);
2361
2362 /*
2363 * We now own the reference to fp that the ofiles[] array
2364 * used to own. Release it.
2365 */
2366 if (fp != NULL)
2367 fdrop_locked(fp, td);
2368
2369 FILEDESC_UNLOCK(fdp);
2370
2371 return (0);
2372
2373 default:
2374 FILEDESC_UNLOCK(fdp);
2375 return (error);
2376 }
2377 /* NOTREACHED */
2378 }
2379
2380 /*
2381 * Scan all active processes to see if any of them have a current
2382 * or root directory of `olddp'. If so, replace them with the new
2383 * mount point.
2384 */
2385 void
2386 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2387 {
2388 struct filedesc *fdp;
2389 struct proc *p;
2390 int nrele;
2391
2392 if (vrefcnt(olddp) == 1)
2393 return;
2394 sx_slock(&allproc_lock);
2395 LIST_FOREACH(p, &allproc, p_list) {
2396 fdp = fdhold(p);
2397 if (fdp == NULL)
2398 continue;
2399 nrele = 0;
2400 FILEDESC_LOCK_FAST(fdp);
2401 if (fdp->fd_cdir == olddp) {
2402 vref(newdp);
2403 fdp->fd_cdir = newdp;
2404 nrele++;
2405 }
2406 if (fdp->fd_rdir == olddp) {
2407 vref(newdp);
2408 fdp->fd_rdir = newdp;
2409 nrele++;
2410 }
2411 FILEDESC_UNLOCK_FAST(fdp);
2412 fddrop(fdp);
2413 while (nrele--)
2414 vrele(olddp);
2415 }
2416 sx_sunlock(&allproc_lock);
2417 if (rootvnode == olddp) {
2418 vrele(rootvnode);
2419 vref(newdp);
2420 rootvnode = newdp;
2421 }
2422 }
2423
2424 struct filedesc_to_leader *
2425 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2426 {
2427 struct filedesc_to_leader *fdtol;
2428
2429 MALLOC(fdtol, struct filedesc_to_leader *,
2430 sizeof(struct filedesc_to_leader),
2431 M_FILEDESC_TO_LEADER,
2432 M_WAITOK);
2433 fdtol->fdl_refcount = 1;
2434 fdtol->fdl_holdcount = 0;
2435 fdtol->fdl_wakeup = 0;
2436 fdtol->fdl_leader = leader;
2437 if (old != NULL) {
2438 FILEDESC_LOCK(fdp);
2439 fdtol->fdl_next = old->fdl_next;
2440 fdtol->fdl_prev = old;
2441 old->fdl_next = fdtol;
2442 fdtol->fdl_next->fdl_prev = fdtol;
2443 FILEDESC_UNLOCK(fdp);
2444 } else {
2445 fdtol->fdl_next = fdtol;
2446 fdtol->fdl_prev = fdtol;
2447 }
2448 return (fdtol);
2449 }
2450
2451 /*
2452 * Get file structures.
2453 */
2454 static int
2455 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2456 {
2457 struct xfile xf;
2458 struct filedesc *fdp;
2459 struct file *fp;
2460 struct proc *p;
2461 int error, n;
2462
2463 /*
2464 * Note: because the number of file descriptors is calculated
2465 * in different ways for sizing vs returning the data,
2466 * there is information leakage from the first loop. However,
2467 * it is of a similar order of magnitude to the leakage from
2468 * global system statistics such as kern.openfiles.
2469 */
2470 error = sysctl_wire_old_buffer(req, 0);
2471 if (error != 0)
2472 return (error);
2473 if (req->oldptr == NULL) {
2474 n = 16; /* A slight overestimate. */
2475 sx_slock(&filelist_lock);
2476 LIST_FOREACH(fp, &filehead, f_list) {
2477 /*
2478 * We should grab the lock, but this is an
2479 * estimate, so does it really matter?
2480 */
2481 /* mtx_lock(fp->f_mtxp); */
2482 n += fp->f_count;
2483 /* mtx_unlock(f->f_mtxp); */
2484 }
2485 sx_sunlock(&filelist_lock);
2486 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2487 }
2488 error = 0;
2489 bzero(&xf, sizeof(xf));
2490 xf.xf_size = sizeof(xf);
2491 sx_slock(&allproc_lock);
2492 LIST_FOREACH(p, &allproc, p_list) {
2493 if (p->p_state == PRS_NEW)
2494 continue;
2495 PROC_LOCK(p);
2496 if (p_cansee(req->td, p) != 0) {
2497 PROC_UNLOCK(p);
2498 continue;
2499 }
2500 xf.xf_pid = p->p_pid;
2501 xf.xf_uid = p->p_ucred->cr_uid;
2502 PROC_UNLOCK(p);
2503 fdp = fdhold(p);
2504 if (fdp == NULL)
2505 continue;
2506 FILEDESC_LOCK_FAST(fdp);
2507 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2508 if ((fp = fdp->fd_ofiles[n]) == NULL)
2509 continue;
2510 xf.xf_fd = n;
2511 xf.xf_file = fp;
2512 xf.xf_data = fp->f_data;
2513 xf.xf_vnode = fp->f_vnode;
2514 xf.xf_type = fp->f_type;
2515 xf.xf_count = fp->f_count;
2516 xf.xf_msgcount = fp->f_msgcount;
2517 xf.xf_offset = fp->f_offset;
2518 xf.xf_flag = fp->f_flag;
2519 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2520 if (error)
2521 break;
2522 }
2523 FILEDESC_UNLOCK_FAST(fdp);
2524 fddrop(fdp);
2525 if (error)
2526 break;
2527 }
2528 sx_sunlock(&allproc_lock);
2529 return (error);
2530 }
2531
2532 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2533 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2534
2535 #ifdef DDB
2536 /*
2537 * For the purposes of debugging, generate a human-readable string for the
2538 * file type.
2539 */
2540 static const char *
2541 file_type_to_name(short type)
2542 {
2543
2544 switch (type) {
2545 case 0:
2546 return ("zero");
2547 case DTYPE_VNODE:
2548 return ("vnod");
2549 case DTYPE_SOCKET:
2550 return ("sock");
2551 case DTYPE_PIPE:
2552 return ("pipe");
2553 case DTYPE_FIFO:
2554 return ("fifo");
2555 case DTYPE_CRYPTO:
2556 return ("crpt");
2557 default:
2558 return ("unkn");
2559 }
2560 }
2561
2562 /*
2563 * For the purposes of debugging, identify a process (if any, perhaps one of
2564 * many) that references the passed file in its file descriptor array. Return
2565 * NULL if none.
2566 */
2567 static struct proc *
2568 file_to_first_proc(struct file *fp)
2569 {
2570 struct filedesc *fdp;
2571 struct proc *p;
2572 int n;
2573
2574 LIST_FOREACH(p, &allproc, p_list) {
2575 if (p->p_state == PRS_NEW)
2576 continue;
2577 fdp = p->p_fd;
2578 if (fdp == NULL)
2579 continue;
2580 for (n = 0; n < fdp->fd_nfiles; n++) {
2581 if (fp == fdp->fd_ofiles[n])
2582 return (p);
2583 }
2584 }
2585 return (NULL);
2586 }
2587
2588 DB_SHOW_COMMAND(files, db_show_files)
2589 {
2590 struct file *fp;
2591 struct proc *p;
2592
2593 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", "File",
2594 "Type", "Data", "Flag", "GCFl", "Count", "MCount", "Vnode",
2595 "FPID", "FCmd");
2596 LIST_FOREACH(fp, &filehead, f_list) {
2597 p = file_to_first_proc(fp);
2598 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2599 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2600 fp->f_gcflag, fp->f_count, fp->f_msgcount, fp->f_vnode,
2601 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2602 }
2603 }
2604 #endif
2605
2606 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2607 &maxfilesperproc, 0, "Maximum files allowed open per process");
2608
2609 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2610 &maxfiles, 0, "Maximum number of files");
2611
2612 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2613 &openfiles, 0, "System-wide number of open files");
2614
2615 /* ARGSUSED*/
2616 static void
2617 filelistinit(void *dummy)
2618 {
2619
2620 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2621 NULL, NULL, UMA_ALIGN_PTR, 0);
2622 sx_init(&filelist_lock, "filelist lock");
2623 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2624 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2625 }
2626 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2627
2628 /*-------------------------------------------------------------------*/
2629
2630 static int
2631 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2632 {
2633
2634 return (EBADF);
2635 }
2636
2637 static int
2638 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2639 {
2640
2641 return (EBADF);
2642 }
2643
2644 static int
2645 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2646 {
2647
2648 return (0);
2649 }
2650
2651 static int
2652 badfo_kqfilter(struct file *fp, struct knote *kn)
2653 {
2654
2655 return (EBADF);
2656 }
2657
2658 static int
2659 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2660 {
2661
2662 return (EBADF);
2663 }
2664
2665 static int
2666 badfo_close(struct file *fp, struct thread *td)
2667 {
2668
2669 return (EBADF);
2670 }
2671
2672 struct fileops badfileops = {
2673 .fo_read = badfo_readwrite,
2674 .fo_write = badfo_readwrite,
2675 .fo_ioctl = badfo_ioctl,
2676 .fo_poll = badfo_poll,
2677 .fo_kqfilter = badfo_kqfilter,
2678 .fo_stat = badfo_stat,
2679 .fo_close = badfo_close,
2680 };
2681
2682
2683 /*-------------------------------------------------------------------*/
2684
2685 /*
2686 * File Descriptor pseudo-device driver (/dev/fd/).
2687 *
2688 * Opening minor device N dup()s the file (if any) connected to file
2689 * descriptor N belonging to the calling process. Note that this driver
2690 * consists of only the ``open()'' routine, because all subsequent
2691 * references to this file will be direct to the other driver.
2692 *
2693 * XXX: we could give this one a cloning event handler if necessary.
2694 */
2695
2696 /* ARGSUSED */
2697 static int
2698 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2699 {
2700
2701 /*
2702 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2703 * the file descriptor being sought for duplication. The error
2704 * return ensures that the vnode for this device will be released
2705 * by vn_open. Open will detect this special error and take the
2706 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2707 * will simply report the error.
2708 */
2709 td->td_dupfd = dev2unit(dev);
2710 return (ENODEV);
2711 }
2712
2713 static struct cdevsw fildesc_cdevsw = {
2714 .d_version = D_VERSION,
2715 .d_flags = D_NEEDGIANT,
2716 .d_open = fdopen,
2717 .d_name = "FD",
2718 };
2719
2720 static void
2721 fildesc_drvinit(void *unused)
2722 {
2723 struct cdev *dev;
2724
2725 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2726 make_dev_alias(dev, "stdin");
2727 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2728 make_dev_alias(dev, "stdout");
2729 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2730 make_dev_alias(dev, "stderr");
2731 }
2732
2733 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
Cache object: d2d2e6bc53b3c6e6451284b0423c4dd1
|