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/6.1/sys/kern/kern_descrip.c 176272 2008-02-14 11:47:39Z simon $");
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 <vm/uma.h>
72
73 #include <ddb/ddb.h>
74
75 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
76 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
77 "file desc to leader structures");
78 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
79
80 static uma_zone_t file_zone;
81
82
83 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
84 enum dup_type { DUP_VARIABLE, DUP_FIXED };
85
86 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
87 register_t *retval);
88 static int fd_first_free(struct filedesc *, int, int);
89 static int fd_last_used(struct filedesc *, int, int);
90 static void fdgrowtable(struct filedesc *, int);
91 static int fdrop_locked(struct file *fp, struct thread *td);
92 static void fdunused(struct filedesc *fdp, int fd);
93 static void fdused(struct filedesc *fdp, int fd);
94
95 /*
96 * A process is initially started out with NDFILE descriptors stored within
97 * this structure, selected to be enough for typical applications based on
98 * the historical limit of 20 open files (and the usage of descriptors by
99 * shells). If these descriptors are exhausted, a larger descriptor table
100 * may be allocated, up to a process' resource limit; the internal arrays
101 * are then unused.
102 */
103 #define NDFILE 20
104 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
105 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
106 #define NDSLOT(x) ((x) / NDENTRIES)
107 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
108 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
109
110 /*
111 * Storage required per open file descriptor.
112 */
113 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
114
115 /*
116 * Basic allocation of descriptors:
117 * one of the above, plus arrays for NDFILE descriptors.
118 */
119 struct filedesc0 {
120 struct filedesc fd_fd;
121 /*
122 * These arrays are used when the number of open files is
123 * <= NDFILE, and are then pointed to by the pointers above.
124 */
125 struct file *fd_dfiles[NDFILE];
126 char fd_dfileflags[NDFILE];
127 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
128 };
129
130 /*
131 * Descriptor management.
132 */
133 struct filelist filehead; /* head of list of open files */
134 int openfiles; /* actual number of open files */
135 struct sx filelist_lock; /* sx to protect filelist */
136 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
137
138 /* A mutex to protect the association between a proc and filedesc. */
139 static struct mtx fdesc_mtx;
140
141 /*
142 * Find the first zero bit in the given bitmap, starting at low and not
143 * exceeding size - 1.
144 */
145 static int
146 fd_first_free(struct filedesc *fdp, int low, int size)
147 {
148 NDSLOTTYPE *map = fdp->fd_map;
149 NDSLOTTYPE mask;
150 int off, maxoff;
151
152 if (low >= size)
153 return (low);
154
155 off = NDSLOT(low);
156 if (low % NDENTRIES) {
157 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
158 if ((mask &= ~map[off]) != 0UL)
159 return (off * NDENTRIES + ffsl(mask) - 1);
160 ++off;
161 }
162 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
163 if (map[off] != ~0UL)
164 return (off * NDENTRIES + ffsl(~map[off]) - 1);
165 return (size);
166 }
167
168 /*
169 * Find the highest non-zero bit in the given bitmap, starting at low and
170 * not exceeding size - 1.
171 */
172 static int
173 fd_last_used(struct filedesc *fdp, int low, int size)
174 {
175 NDSLOTTYPE *map = fdp->fd_map;
176 NDSLOTTYPE mask;
177 int off, minoff;
178
179 if (low >= size)
180 return (-1);
181
182 off = NDSLOT(size);
183 if (size % NDENTRIES) {
184 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
185 if ((mask &= map[off]) != 0)
186 return (off * NDENTRIES + flsl(mask) - 1);
187 --off;
188 }
189 for (minoff = NDSLOT(low); off >= minoff; --off)
190 if (map[off] != 0)
191 return (off * NDENTRIES + flsl(map[off]) - 1);
192 return (low - 1);
193 }
194
195 static int
196 fdisused(struct filedesc *fdp, int fd)
197 {
198 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
199 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
200 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
201 }
202
203 /*
204 * Mark a file descriptor as used.
205 */
206 static void
207 fdused(struct filedesc *fdp, int fd)
208 {
209 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
210 KASSERT(!fdisused(fdp, fd),
211 ("fd already used"));
212 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
213 if (fd > fdp->fd_lastfile)
214 fdp->fd_lastfile = fd;
215 if (fd == fdp->fd_freefile)
216 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
217 }
218
219 /*
220 * Mark a file descriptor as unused.
221 */
222 static void
223 fdunused(struct filedesc *fdp, int fd)
224 {
225 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
226 KASSERT(fdisused(fdp, fd),
227 ("fd is already unused"));
228 KASSERT(fdp->fd_ofiles[fd] == NULL,
229 ("fd is still in use"));
230 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
231 if (fd < fdp->fd_freefile)
232 fdp->fd_freefile = fd;
233 if (fd == fdp->fd_lastfile)
234 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
235 }
236
237 /*
238 * System calls on descriptors.
239 */
240 #ifndef _SYS_SYSPROTO_H_
241 struct getdtablesize_args {
242 int dummy;
243 };
244 #endif
245 /*
246 * MPSAFE
247 */
248 /* ARGSUSED */
249 int
250 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
251 {
252 struct proc *p = td->td_proc;
253
254 PROC_LOCK(p);
255 td->td_retval[0] =
256 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
257 PROC_UNLOCK(p);
258 return (0);
259 }
260
261 /*
262 * Duplicate a file descriptor to a particular value.
263 *
264 * note: keep in mind that a potential race condition exists when closing
265 * descriptors from a shared descriptor table (via rfork).
266 */
267 #ifndef _SYS_SYSPROTO_H_
268 struct dup2_args {
269 u_int from;
270 u_int to;
271 };
272 #endif
273 /*
274 * MPSAFE
275 */
276 /* ARGSUSED */
277 int
278 dup2(struct thread *td, struct dup2_args *uap)
279 {
280
281 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
282 td->td_retval));
283 }
284
285 /*
286 * Duplicate a file descriptor.
287 */
288 #ifndef _SYS_SYSPROTO_H_
289 struct dup_args {
290 u_int fd;
291 };
292 #endif
293 /*
294 * MPSAFE
295 */
296 /* ARGSUSED */
297 int
298 dup(struct thread *td, struct dup_args *uap)
299 {
300
301 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
302 }
303
304 /*
305 * The file control system call.
306 */
307 #ifndef _SYS_SYSPROTO_H_
308 struct fcntl_args {
309 int fd;
310 int cmd;
311 long arg;
312 };
313 #endif
314 /*
315 * MPSAFE
316 */
317 /* ARGSUSED */
318 int
319 fcntl(struct thread *td, struct fcntl_args *uap)
320 {
321 struct flock fl;
322 intptr_t arg;
323 int error;
324
325 error = 0;
326 switch (uap->cmd) {
327 case F_GETLK:
328 case F_SETLK:
329 case F_SETLKW:
330 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
331 arg = (intptr_t)&fl;
332 break;
333 default:
334 arg = uap->arg;
335 break;
336 }
337 if (error)
338 return (error);
339 error = kern_fcntl(td, uap->fd, uap->cmd, arg);
340 if (error)
341 return (error);
342 if (uap->cmd == F_GETLK)
343 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
344 return (error);
345 }
346
347 int
348 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
349 {
350 struct filedesc *fdp;
351 struct flock *flp;
352 struct file *fp;
353 struct proc *p;
354 char *pop;
355 struct vnode *vp;
356 u_int newmin;
357 int error, flg, tmp;
358 int giant_locked;
359
360 /*
361 * XXXRW: Some fcntl() calls require Giant -- others don't. Try to
362 * avoid grabbing Giant for calls we know don't need it.
363 */
364 switch (cmd) {
365 case F_DUPFD:
366 case F_GETFD:
367 case F_SETFD:
368 case F_GETFL:
369 giant_locked = 0;
370 break;
371
372 default:
373 giant_locked = 1;
374 mtx_lock(&Giant);
375 }
376
377 error = 0;
378 flg = F_POSIX;
379 p = td->td_proc;
380 fdp = p->p_fd;
381 FILEDESC_LOCK(fdp);
382 if ((unsigned)fd >= fdp->fd_nfiles ||
383 (fp = fdp->fd_ofiles[fd]) == NULL) {
384 FILEDESC_UNLOCK(fdp);
385 error = EBADF;
386 goto done2;
387 }
388 pop = &fdp->fd_ofileflags[fd];
389
390 switch (cmd) {
391 case F_DUPFD:
392 /* mtx_assert(&Giant, MA_NOTOWNED); */
393 FILEDESC_UNLOCK(fdp);
394 newmin = arg;
395 PROC_LOCK(p);
396 if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
397 newmin >= maxfilesperproc) {
398 PROC_UNLOCK(p);
399 error = EINVAL;
400 break;
401 }
402 PROC_UNLOCK(p);
403 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
404 break;
405
406 case F_GETFD:
407 /* mtx_assert(&Giant, MA_NOTOWNED); */
408 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
409 FILEDESC_UNLOCK(fdp);
410 break;
411
412 case F_SETFD:
413 /* mtx_assert(&Giant, MA_NOTOWNED); */
414 *pop = (*pop &~ UF_EXCLOSE) |
415 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
416 FILEDESC_UNLOCK(fdp);
417 break;
418
419 case F_GETFL:
420 /* mtx_assert(&Giant, MA_NOTOWNED); */
421 FILE_LOCK(fp);
422 td->td_retval[0] = OFLAGS(fp->f_flag);
423 FILE_UNLOCK(fp);
424 FILEDESC_UNLOCK(fdp);
425 break;
426
427 case F_SETFL:
428 mtx_assert(&Giant, MA_OWNED);
429 FILE_LOCK(fp);
430 fhold_locked(fp);
431 fp->f_flag &= ~FCNTLFLAGS;
432 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
433 FILE_UNLOCK(fp);
434 FILEDESC_UNLOCK(fdp);
435 tmp = fp->f_flag & FNONBLOCK;
436 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
437 if (error) {
438 fdrop(fp, td);
439 break;
440 }
441 tmp = fp->f_flag & FASYNC;
442 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
443 if (error == 0) {
444 fdrop(fp, td);
445 break;
446 }
447 FILE_LOCK(fp);
448 fp->f_flag &= ~FNONBLOCK;
449 FILE_UNLOCK(fp);
450 tmp = 0;
451 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
452 fdrop(fp, td);
453 break;
454
455 case F_GETOWN:
456 mtx_assert(&Giant, MA_OWNED);
457 fhold(fp);
458 FILEDESC_UNLOCK(fdp);
459 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
460 if (error == 0)
461 td->td_retval[0] = tmp;
462 fdrop(fp, td);
463 break;
464
465 case F_SETOWN:
466 mtx_assert(&Giant, MA_OWNED);
467 fhold(fp);
468 FILEDESC_UNLOCK(fdp);
469 tmp = arg;
470 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
471 fdrop(fp, td);
472 break;
473
474 case F_SETLKW:
475 mtx_assert(&Giant, MA_OWNED);
476 flg |= F_WAIT;
477 /* FALLTHROUGH F_SETLK */
478
479 case F_SETLK:
480 mtx_assert(&Giant, MA_OWNED);
481 if (fp->f_type != DTYPE_VNODE) {
482 FILEDESC_UNLOCK(fdp);
483 error = EBADF;
484 break;
485 }
486
487 flp = (struct flock *)arg;
488 if (flp->l_whence == SEEK_CUR) {
489 if (fp->f_offset < 0 ||
490 (flp->l_start > 0 &&
491 fp->f_offset > OFF_MAX - flp->l_start)) {
492 FILEDESC_UNLOCK(fdp);
493 error = EOVERFLOW;
494 break;
495 }
496 flp->l_start += fp->f_offset;
497 }
498
499 /*
500 * VOP_ADVLOCK() may block.
501 */
502 fhold(fp);
503 FILEDESC_UNLOCK(fdp);
504 vp = fp->f_vnode;
505
506 switch (flp->l_type) {
507 case F_RDLCK:
508 if ((fp->f_flag & FREAD) == 0) {
509 error = EBADF;
510 break;
511 }
512 PROC_LOCK(p->p_leader);
513 p->p_leader->p_flag |= P_ADVLOCK;
514 PROC_UNLOCK(p->p_leader);
515 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
516 flp, flg);
517 break;
518 case F_WRLCK:
519 if ((fp->f_flag & FWRITE) == 0) {
520 error = EBADF;
521 break;
522 }
523 PROC_LOCK(p->p_leader);
524 p->p_leader->p_flag |= P_ADVLOCK;
525 PROC_UNLOCK(p->p_leader);
526 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
527 flp, flg);
528 break;
529 case F_UNLCK:
530 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
531 flp, F_POSIX);
532 break;
533 default:
534 error = EINVAL;
535 break;
536 }
537 /* Check for race with close */
538 FILEDESC_LOCK_FAST(fdp);
539 if ((unsigned) fd >= fdp->fd_nfiles ||
540 fp != fdp->fd_ofiles[fd]) {
541 FILEDESC_UNLOCK_FAST(fdp);
542 flp->l_whence = SEEK_SET;
543 flp->l_start = 0;
544 flp->l_len = 0;
545 flp->l_type = F_UNLCK;
546 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
547 F_UNLCK, flp, F_POSIX);
548 } else
549 FILEDESC_UNLOCK_FAST(fdp);
550 fdrop(fp, td);
551 break;
552
553 case F_GETLK:
554 mtx_assert(&Giant, MA_OWNED);
555 if (fp->f_type != DTYPE_VNODE) {
556 FILEDESC_UNLOCK(fdp);
557 error = EBADF;
558 break;
559 }
560 flp = (struct flock *)arg;
561 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
562 flp->l_type != F_UNLCK) {
563 FILEDESC_UNLOCK(fdp);
564 error = EINVAL;
565 break;
566 }
567 if (flp->l_whence == SEEK_CUR) {
568 if ((flp->l_start > 0 &&
569 fp->f_offset > OFF_MAX - flp->l_start) ||
570 (flp->l_start < 0 &&
571 fp->f_offset < OFF_MIN - flp->l_start)) {
572 FILEDESC_UNLOCK(fdp);
573 error = EOVERFLOW;
574 break;
575 }
576 flp->l_start += fp->f_offset;
577 }
578 /*
579 * VOP_ADVLOCK() may block.
580 */
581 fhold(fp);
582 FILEDESC_UNLOCK(fdp);
583 vp = fp->f_vnode;
584 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
585 F_POSIX);
586 fdrop(fp, td);
587 break;
588 default:
589 FILEDESC_UNLOCK(fdp);
590 error = EINVAL;
591 break;
592 }
593 done2:
594 if (giant_locked)
595 mtx_unlock(&Giant);
596 return (error);
597 }
598
599 /*
600 * Common code for dup, dup2, and fcntl(F_DUPFD).
601 */
602 static int
603 do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval)
604 {
605 struct filedesc *fdp;
606 struct proc *p;
607 struct file *fp;
608 struct file *delfp;
609 int error, holdleaders, maxfd;
610
611 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
612 ("invalid dup type %d", type));
613
614 p = td->td_proc;
615 fdp = p->p_fd;
616
617 /*
618 * Verify we have a valid descriptor to dup from and possibly to
619 * dup to.
620 */
621 if (old < 0 || new < 0)
622 return (EBADF);
623 PROC_LOCK(p);
624 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
625 PROC_UNLOCK(p);
626 if (new >= maxfd)
627 return (EMFILE);
628
629 FILEDESC_LOCK(fdp);
630 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
631 FILEDESC_UNLOCK(fdp);
632 return (EBADF);
633 }
634 if (type == DUP_FIXED && old == new) {
635 *retval = new;
636 FILEDESC_UNLOCK(fdp);
637 return (0);
638 }
639 fp = fdp->fd_ofiles[old];
640 fhold(fp);
641
642 /*
643 * If the caller specified a file descriptor, make sure the file
644 * table is large enough to hold it, and grab it. Otherwise, just
645 * allocate a new descriptor the usual way. Since the filedesc
646 * lock may be temporarily dropped in the process, we have to look
647 * out for a race.
648 */
649 if (type == DUP_FIXED) {
650 if (new >= fdp->fd_nfiles)
651 fdgrowtable(fdp, new + 1);
652 if (fdp->fd_ofiles[new] == NULL)
653 fdused(fdp, new);
654 } else {
655 if ((error = fdalloc(td, new, &new)) != 0) {
656 FILEDESC_UNLOCK(fdp);
657 fdrop(fp, td);
658 return (error);
659 }
660 }
661
662 /*
663 * If the old file changed out from under us then treat it as a
664 * bad file descriptor. Userland should do its own locking to
665 * avoid this case.
666 */
667 if (fdp->fd_ofiles[old] != fp) {
668 /* we've allocated a descriptor which we won't use */
669 if (fdp->fd_ofiles[new] == NULL)
670 fdunused(fdp, new);
671 FILEDESC_UNLOCK(fdp);
672 fdrop(fp, td);
673 return (EBADF);
674 }
675 KASSERT(old != new,
676 ("new fd is same as old"));
677
678 /*
679 * Save info on the descriptor being overwritten. We cannot close
680 * it without introducing an ownership race for the slot, since we
681 * need to drop the filedesc lock to call closef().
682 *
683 * XXX this duplicates parts of close().
684 */
685 delfp = fdp->fd_ofiles[new];
686 holdleaders = 0;
687 if (delfp != NULL) {
688 if (td->td_proc->p_fdtol != NULL) {
689 /*
690 * Ask fdfree() to sleep to ensure that all relevant
691 * process leaders can be traversed in closef().
692 */
693 fdp->fd_holdleaderscount++;
694 holdleaders = 1;
695 }
696 }
697
698 /*
699 * Duplicate the source descriptor
700 */
701 fdp->fd_ofiles[new] = fp;
702 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
703 if (new > fdp->fd_lastfile)
704 fdp->fd_lastfile = new;
705 *retval = new;
706
707 /*
708 * If we dup'd over a valid file, we now own the reference to it
709 * and must dispose of it using closef() semantics (as if a
710 * close() were performed on it).
711 *
712 * XXX this duplicates parts of close().
713 */
714 if (delfp != NULL) {
715 knote_fdclose(td, new);
716 FILEDESC_UNLOCK(fdp);
717 (void) closef(delfp, td);
718 if (holdleaders) {
719 FILEDESC_LOCK_FAST(fdp);
720 fdp->fd_holdleaderscount--;
721 if (fdp->fd_holdleaderscount == 0 &&
722 fdp->fd_holdleaderswakeup != 0) {
723 fdp->fd_holdleaderswakeup = 0;
724 wakeup(&fdp->fd_holdleaderscount);
725 }
726 FILEDESC_UNLOCK_FAST(fdp);
727 }
728 } else {
729 FILEDESC_UNLOCK(fdp);
730 }
731 return (0);
732 }
733
734 /*
735 * If sigio is on the list associated with a process or process group,
736 * disable signalling from the device, remove sigio from the list and
737 * free sigio.
738 */
739 void
740 funsetown(struct sigio **sigiop)
741 {
742 struct sigio *sigio;
743
744 SIGIO_LOCK();
745 sigio = *sigiop;
746 if (sigio == NULL) {
747 SIGIO_UNLOCK();
748 return;
749 }
750 *(sigio->sio_myref) = NULL;
751 if ((sigio)->sio_pgid < 0) {
752 struct pgrp *pg = (sigio)->sio_pgrp;
753 PGRP_LOCK(pg);
754 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
755 sigio, sio_pgsigio);
756 PGRP_UNLOCK(pg);
757 } else {
758 struct proc *p = (sigio)->sio_proc;
759 PROC_LOCK(p);
760 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
761 sigio, sio_pgsigio);
762 PROC_UNLOCK(p);
763 }
764 SIGIO_UNLOCK();
765 crfree(sigio->sio_ucred);
766 FREE(sigio, M_SIGIO);
767 }
768
769 /*
770 * Free a list of sigio structures.
771 * We only need to lock the SIGIO_LOCK because we have made ourselves
772 * inaccessible to callers of fsetown and therefore do not need to lock
773 * the proc or pgrp struct for the list manipulation.
774 */
775 void
776 funsetownlst(struct sigiolst *sigiolst)
777 {
778 struct proc *p;
779 struct pgrp *pg;
780 struct sigio *sigio;
781
782 sigio = SLIST_FIRST(sigiolst);
783 if (sigio == NULL)
784 return;
785 p = NULL;
786 pg = NULL;
787
788 /*
789 * Every entry of the list should belong
790 * to a single proc or pgrp.
791 */
792 if (sigio->sio_pgid < 0) {
793 pg = sigio->sio_pgrp;
794 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
795 } else /* if (sigio->sio_pgid > 0) */ {
796 p = sigio->sio_proc;
797 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
798 }
799
800 SIGIO_LOCK();
801 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
802 *(sigio->sio_myref) = NULL;
803 if (pg != NULL) {
804 KASSERT(sigio->sio_pgid < 0,
805 ("Proc sigio in pgrp sigio list"));
806 KASSERT(sigio->sio_pgrp == pg,
807 ("Bogus pgrp in sigio list"));
808 PGRP_LOCK(pg);
809 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
810 sio_pgsigio);
811 PGRP_UNLOCK(pg);
812 } else /* if (p != NULL) */ {
813 KASSERT(sigio->sio_pgid > 0,
814 ("Pgrp sigio in proc sigio list"));
815 KASSERT(sigio->sio_proc == p,
816 ("Bogus proc in sigio list"));
817 PROC_LOCK(p);
818 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
819 sio_pgsigio);
820 PROC_UNLOCK(p);
821 }
822 SIGIO_UNLOCK();
823 crfree(sigio->sio_ucred);
824 FREE(sigio, M_SIGIO);
825 SIGIO_LOCK();
826 }
827 SIGIO_UNLOCK();
828 }
829
830 /*
831 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
832 *
833 * After permission checking, add a sigio structure to the sigio list for
834 * the process or process group.
835 */
836 int
837 fsetown(pid_t pgid, struct sigio **sigiop)
838 {
839 struct proc *proc;
840 struct pgrp *pgrp;
841 struct sigio *sigio;
842 int ret;
843
844 if (pgid == 0) {
845 funsetown(sigiop);
846 return (0);
847 }
848
849 ret = 0;
850
851 /* Allocate and fill in the new sigio out of locks. */
852 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
853 sigio->sio_pgid = pgid;
854 sigio->sio_ucred = crhold(curthread->td_ucred);
855 sigio->sio_myref = sigiop;
856
857 sx_slock(&proctree_lock);
858 if (pgid > 0) {
859 proc = pfind(pgid);
860 if (proc == NULL) {
861 ret = ESRCH;
862 goto fail;
863 }
864
865 /*
866 * Policy - Don't allow a process to FSETOWN a process
867 * in another session.
868 *
869 * Remove this test to allow maximum flexibility or
870 * restrict FSETOWN to the current process or process
871 * group for maximum safety.
872 */
873 PROC_UNLOCK(proc);
874 if (proc->p_session != curthread->td_proc->p_session) {
875 ret = EPERM;
876 goto fail;
877 }
878
879 pgrp = NULL;
880 } else /* if (pgid < 0) */ {
881 pgrp = pgfind(-pgid);
882 if (pgrp == NULL) {
883 ret = ESRCH;
884 goto fail;
885 }
886 PGRP_UNLOCK(pgrp);
887
888 /*
889 * Policy - Don't allow a process to FSETOWN a process
890 * in another session.
891 *
892 * Remove this test to allow maximum flexibility or
893 * restrict FSETOWN to the current process or process
894 * group for maximum safety.
895 */
896 if (pgrp->pg_session != curthread->td_proc->p_session) {
897 ret = EPERM;
898 goto fail;
899 }
900
901 proc = NULL;
902 }
903 funsetown(sigiop);
904 if (pgid > 0) {
905 PROC_LOCK(proc);
906 /*
907 * Since funsetownlst() is called without the proctree
908 * locked, we need to check for P_WEXIT.
909 * XXX: is ESRCH correct?
910 */
911 if ((proc->p_flag & P_WEXIT) != 0) {
912 PROC_UNLOCK(proc);
913 ret = ESRCH;
914 goto fail;
915 }
916 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
917 sigio->sio_proc = proc;
918 PROC_UNLOCK(proc);
919 } else {
920 PGRP_LOCK(pgrp);
921 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
922 sigio->sio_pgrp = pgrp;
923 PGRP_UNLOCK(pgrp);
924 }
925 sx_sunlock(&proctree_lock);
926 SIGIO_LOCK();
927 *sigiop = sigio;
928 SIGIO_UNLOCK();
929 return (0);
930
931 fail:
932 sx_sunlock(&proctree_lock);
933 crfree(sigio->sio_ucred);
934 FREE(sigio, M_SIGIO);
935 return (ret);
936 }
937
938 /*
939 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
940 */
941 pid_t
942 fgetown(sigiop)
943 struct sigio **sigiop;
944 {
945 pid_t pgid;
946
947 SIGIO_LOCK();
948 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
949 SIGIO_UNLOCK();
950 return (pgid);
951 }
952
953 /*
954 * Close a file descriptor.
955 */
956 #ifndef _SYS_SYSPROTO_H_
957 struct close_args {
958 int fd;
959 };
960 #endif
961 /*
962 * MPSAFE
963 */
964 /* ARGSUSED */
965 int
966 close(td, uap)
967 struct thread *td;
968 struct close_args *uap;
969 {
970 struct filedesc *fdp;
971 struct file *fp;
972 int fd, error;
973 int holdleaders;
974
975 fd = uap->fd;
976 error = 0;
977 holdleaders = 0;
978 fdp = td->td_proc->p_fd;
979 FILEDESC_LOCK(fdp);
980 if ((unsigned)fd >= fdp->fd_nfiles ||
981 (fp = fdp->fd_ofiles[fd]) == NULL) {
982 FILEDESC_UNLOCK(fdp);
983 return (EBADF);
984 }
985 fdp->fd_ofiles[fd] = NULL;
986 fdp->fd_ofileflags[fd] = 0;
987 fdunused(fdp, fd);
988 if (td->td_proc->p_fdtol != NULL) {
989 /*
990 * Ask fdfree() to sleep to ensure that all relevant
991 * process leaders can be traversed in closef().
992 */
993 fdp->fd_holdleaderscount++;
994 holdleaders = 1;
995 }
996
997 /*
998 * We now hold the fp reference that used to be owned by the descriptor
999 * array.
1000 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a
1001 * race of the fd getting opened, a knote added, and deleteing a knote
1002 * for the new fd.
1003 */
1004 knote_fdclose(td, fd);
1005 FILEDESC_UNLOCK(fdp);
1006
1007 error = closef(fp, td);
1008 if (holdleaders) {
1009 FILEDESC_LOCK_FAST(fdp);
1010 fdp->fd_holdleaderscount--;
1011 if (fdp->fd_holdleaderscount == 0 &&
1012 fdp->fd_holdleaderswakeup != 0) {
1013 fdp->fd_holdleaderswakeup = 0;
1014 wakeup(&fdp->fd_holdleaderscount);
1015 }
1016 FILEDESC_UNLOCK_FAST(fdp);
1017 }
1018 return (error);
1019 }
1020
1021 #if defined(COMPAT_43)
1022 /*
1023 * Return status information about a file descriptor.
1024 */
1025 #ifndef _SYS_SYSPROTO_H_
1026 struct ofstat_args {
1027 int fd;
1028 struct ostat *sb;
1029 };
1030 #endif
1031 /*
1032 * MPSAFE
1033 */
1034 /* ARGSUSED */
1035 int
1036 ofstat(struct thread *td, struct ofstat_args *uap)
1037 {
1038 struct ostat oub;
1039 struct stat ub;
1040 int error;
1041
1042 error = kern_fstat(td, uap->fd, &ub);
1043 if (error == 0) {
1044 cvtstat(&ub, &oub);
1045 error = copyout(&oub, uap->sb, sizeof(oub));
1046 }
1047 return (error);
1048 }
1049 #endif /* COMPAT_43 */
1050
1051 /*
1052 * Return status information about a file descriptor.
1053 */
1054 #ifndef _SYS_SYSPROTO_H_
1055 struct fstat_args {
1056 int fd;
1057 struct stat *sb;
1058 };
1059 #endif
1060 /*
1061 * MPSAFE
1062 */
1063 /* ARGSUSED */
1064 int
1065 fstat(struct thread *td, struct fstat_args *uap)
1066 {
1067 struct stat ub;
1068 int error;
1069
1070 error = kern_fstat(td, uap->fd, &ub);
1071 if (error == 0)
1072 error = copyout(&ub, uap->sb, sizeof(ub));
1073 return (error);
1074 }
1075
1076 int
1077 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1078 {
1079 struct file *fp;
1080 int error;
1081
1082 if ((error = fget(td, fd, &fp)) != 0)
1083 return (error);
1084 error = fo_stat(fp, sbp, td->td_ucred, td);
1085 fdrop(fp, td);
1086 return (error);
1087 }
1088
1089 /*
1090 * Return status information about a file descriptor.
1091 */
1092 #ifndef _SYS_SYSPROTO_H_
1093 struct nfstat_args {
1094 int fd;
1095 struct nstat *sb;
1096 };
1097 #endif
1098 /*
1099 * MPSAFE
1100 */
1101 /* ARGSUSED */
1102 int
1103 nfstat(struct thread *td, struct nfstat_args *uap)
1104 {
1105 struct nstat nub;
1106 struct stat ub;
1107 int error;
1108
1109 error = kern_fstat(td, uap->fd, &ub);
1110 if (error == 0) {
1111 cvtnstat(&ub, &nub);
1112 error = copyout(&nub, uap->sb, sizeof(nub));
1113 }
1114 return (error);
1115 }
1116
1117 /*
1118 * Return pathconf information about a file descriptor.
1119 */
1120 #ifndef _SYS_SYSPROTO_H_
1121 struct fpathconf_args {
1122 int fd;
1123 int name;
1124 };
1125 #endif
1126 /*
1127 * MPSAFE
1128 */
1129 /* ARGSUSED */
1130 int
1131 fpathconf(struct thread *td, struct fpathconf_args *uap)
1132 {
1133 struct file *fp;
1134 struct vnode *vp;
1135 int error;
1136
1137 if ((error = fget(td, uap->fd, &fp)) != 0)
1138 return (error);
1139
1140 /* If asynchronous I/O is available, it works for all descriptors. */
1141 if (uap->name == _PC_ASYNC_IO) {
1142 td->td_retval[0] = async_io_version;
1143 goto out;
1144 }
1145 vp = fp->f_vnode;
1146 if (vp != NULL) {
1147 int vfslocked;
1148 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1149 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1150 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1151 VOP_UNLOCK(vp, 0, td);
1152 VFS_UNLOCK_GIANT(vfslocked);
1153 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1154 if (uap->name != _PC_PIPE_BUF) {
1155 error = EINVAL;
1156 } else {
1157 td->td_retval[0] = PIPE_BUF;
1158 error = 0;
1159 }
1160 } else {
1161 error = EOPNOTSUPP;
1162 }
1163 out:
1164 fdrop(fp, td);
1165 return (error);
1166 }
1167
1168 /*
1169 * Grow the file table to accomodate (at least) nfd descriptors. This may
1170 * block and drop the filedesc lock, but it will reacquire it before
1171 * returning.
1172 */
1173 static void
1174 fdgrowtable(struct filedesc *fdp, int nfd)
1175 {
1176 struct file **ntable;
1177 char *nfileflags;
1178 int nnfiles, onfiles;
1179 NDSLOTTYPE *nmap;
1180
1181 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1182
1183 KASSERT(fdp->fd_nfiles > 0,
1184 ("zero-length file table"));
1185
1186 /* compute the size of the new table */
1187 onfiles = fdp->fd_nfiles;
1188 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1189 if (nnfiles <= onfiles)
1190 /* the table is already large enough */
1191 return;
1192
1193 /* allocate a new table and (if required) new bitmaps */
1194 FILEDESC_UNLOCK(fdp);
1195 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1196 M_FILEDESC, M_ZERO | M_WAITOK);
1197 nfileflags = (char *)&ntable[nnfiles];
1198 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1199 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1200 M_FILEDESC, M_ZERO | M_WAITOK);
1201 else
1202 nmap = NULL;
1203 FILEDESC_LOCK(fdp);
1204
1205 /*
1206 * We now have new tables ready to go. Since we dropped the
1207 * filedesc lock to call malloc(), watch out for a race.
1208 */
1209 onfiles = fdp->fd_nfiles;
1210 if (onfiles >= nnfiles) {
1211 /* we lost the race, but that's OK */
1212 free(ntable, M_FILEDESC);
1213 if (nmap != NULL)
1214 free(nmap, M_FILEDESC);
1215 return;
1216 }
1217 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1218 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1219 if (onfiles > NDFILE)
1220 free(fdp->fd_ofiles, M_FILEDESC);
1221 fdp->fd_ofiles = ntable;
1222 fdp->fd_ofileflags = nfileflags;
1223 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1224 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1225 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1226 free(fdp->fd_map, M_FILEDESC);
1227 fdp->fd_map = nmap;
1228 }
1229 fdp->fd_nfiles = nnfiles;
1230 }
1231
1232 /*
1233 * Allocate a file descriptor for the process.
1234 */
1235 int
1236 fdalloc(struct thread *td, int minfd, int *result)
1237 {
1238 struct proc *p = td->td_proc;
1239 struct filedesc *fdp = p->p_fd;
1240 int fd = -1, maxfd;
1241
1242 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1243
1244 if (fdp->fd_freefile > minfd)
1245 minfd = fdp->fd_freefile;
1246
1247 PROC_LOCK(p);
1248 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1249 PROC_UNLOCK(p);
1250
1251 /*
1252 * Search the bitmap for a free descriptor. If none is found, try
1253 * to grow the file table. Keep at it until we either get a file
1254 * descriptor or run into process or system limits; fdgrowtable()
1255 * may drop the filedesc lock, so we're in a race.
1256 */
1257 for (;;) {
1258 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1259 if (fd >= maxfd)
1260 return (EMFILE);
1261 if (fd < fdp->fd_nfiles)
1262 break;
1263 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1264 }
1265
1266 /*
1267 * Perform some sanity checks, then mark the file descriptor as
1268 * used and return it to the caller.
1269 */
1270 KASSERT(!fdisused(fdp, fd),
1271 ("fd_first_free() returned non-free descriptor"));
1272 KASSERT(fdp->fd_ofiles[fd] == NULL,
1273 ("free descriptor isn't"));
1274 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1275 fdused(fdp, fd);
1276 *result = fd;
1277 return (0);
1278 }
1279
1280 /*
1281 * Check to see whether n user file descriptors
1282 * are available to the process p.
1283 */
1284 int
1285 fdavail(struct thread *td, int n)
1286 {
1287 struct proc *p = td->td_proc;
1288 struct filedesc *fdp = td->td_proc->p_fd;
1289 struct file **fpp;
1290 int i, lim, last;
1291
1292 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1293
1294 PROC_LOCK(p);
1295 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1296 PROC_UNLOCK(p);
1297 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1298 return (1);
1299 last = min(fdp->fd_nfiles, lim);
1300 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1301 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1302 if (*fpp == NULL && --n <= 0)
1303 return (1);
1304 }
1305 return (0);
1306 }
1307
1308 /*
1309 * Create a new open file structure and allocate
1310 * a file decriptor for the process that refers to it.
1311 * We add one reference to the file for the descriptor table
1312 * and one reference for resultfp. This is to prevent us being
1313 * preempted and the entry in the descriptor table closed after
1314 * we release the FILEDESC lock.
1315 */
1316 int
1317 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1318 {
1319 struct proc *p = td->td_proc;
1320 struct file *fp, *fq;
1321 int error, i;
1322 int maxuserfiles = maxfiles - (maxfiles / 20);
1323 static struct timeval lastfail;
1324 static int curfail;
1325
1326 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1327 sx_xlock(&filelist_lock);
1328 if ((openfiles >= maxuserfiles && (td->td_ucred->cr_ruid != 0 ||
1329 jailed(td->td_ucred))) || openfiles >= maxfiles) {
1330 if (ppsratecheck(&lastfail, &curfail, 1)) {
1331 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1332 td->td_ucred->cr_ruid);
1333 }
1334 sx_xunlock(&filelist_lock);
1335 uma_zfree(file_zone, fp);
1336 return (ENFILE);
1337 }
1338 openfiles++;
1339
1340 /*
1341 * If the process has file descriptor zero open, add the new file
1342 * descriptor to the list of open files at that point, otherwise
1343 * put it at the front of the list of open files.
1344 */
1345 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1346 fp->f_count = 1;
1347 if (resultfp)
1348 fp->f_count++;
1349 fp->f_cred = crhold(td->td_ucred);
1350 fp->f_ops = &badfileops;
1351 fp->f_data = NULL;
1352 fp->f_vnode = NULL;
1353 FILEDESC_LOCK(p->p_fd);
1354 if ((fq = p->p_fd->fd_ofiles[0])) {
1355 LIST_INSERT_AFTER(fq, fp, f_list);
1356 } else {
1357 LIST_INSERT_HEAD(&filehead, fp, f_list);
1358 }
1359 sx_xunlock(&filelist_lock);
1360 if ((error = fdalloc(td, 0, &i))) {
1361 FILEDESC_UNLOCK(p->p_fd);
1362 fdrop(fp, td);
1363 if (resultfp)
1364 fdrop(fp, td);
1365 return (error);
1366 }
1367 p->p_fd->fd_ofiles[i] = fp;
1368 FILEDESC_UNLOCK(p->p_fd);
1369 if (resultfp)
1370 *resultfp = fp;
1371 if (resultfd)
1372 *resultfd = i;
1373 return (0);
1374 }
1375
1376 /*
1377 * Build a new filedesc structure from another.
1378 * Copy the current, root, and jail root vnode references.
1379 */
1380 struct filedesc *
1381 fdinit(struct filedesc *fdp)
1382 {
1383 struct filedesc0 *newfdp;
1384
1385 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1386 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1387 if (fdp != NULL) {
1388 FILEDESC_LOCK(fdp);
1389 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1390 if (newfdp->fd_fd.fd_cdir)
1391 VREF(newfdp->fd_fd.fd_cdir);
1392 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1393 if (newfdp->fd_fd.fd_rdir)
1394 VREF(newfdp->fd_fd.fd_rdir);
1395 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1396 if (newfdp->fd_fd.fd_jdir)
1397 VREF(newfdp->fd_fd.fd_jdir);
1398 FILEDESC_UNLOCK(fdp);
1399 }
1400
1401 /* Create the file descriptor table. */
1402 newfdp->fd_fd.fd_refcnt = 1;
1403 newfdp->fd_fd.fd_holdcnt = 1;
1404 newfdp->fd_fd.fd_cmask = CMASK;
1405 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1406 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1407 newfdp->fd_fd.fd_nfiles = NDFILE;
1408 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1409 newfdp->fd_fd.fd_lastfile = -1;
1410 return (&newfdp->fd_fd);
1411 }
1412
1413 static struct filedesc *
1414 fdhold(struct proc *p)
1415 {
1416 struct filedesc *fdp;
1417
1418 mtx_lock(&fdesc_mtx);
1419 fdp = p->p_fd;
1420 if (fdp != NULL)
1421 fdp->fd_holdcnt++;
1422 mtx_unlock(&fdesc_mtx);
1423 return (fdp);
1424 }
1425
1426 static void
1427 fddrop(struct filedesc *fdp)
1428 {
1429 int i;
1430
1431 mtx_lock(&fdesc_mtx);
1432 i = --fdp->fd_holdcnt;
1433 mtx_unlock(&fdesc_mtx);
1434 if (i > 0)
1435 return;
1436
1437 mtx_destroy(&fdp->fd_mtx);
1438 FREE(fdp, M_FILEDESC);
1439 }
1440
1441 /*
1442 * Share a filedesc structure.
1443 */
1444 struct filedesc *
1445 fdshare(struct filedesc *fdp)
1446 {
1447 FILEDESC_LOCK_FAST(fdp);
1448 fdp->fd_refcnt++;
1449 FILEDESC_UNLOCK_FAST(fdp);
1450 return (fdp);
1451 }
1452
1453 /*
1454 * Unshare a filedesc structure, if necessary by making a copy
1455 */
1456 void
1457 fdunshare(struct proc *p, struct thread *td)
1458 {
1459
1460 FILEDESC_LOCK_FAST(p->p_fd);
1461 if (p->p_fd->fd_refcnt > 1) {
1462 struct filedesc *tmp;
1463
1464 FILEDESC_UNLOCK_FAST(p->p_fd);
1465 tmp = fdcopy(p->p_fd);
1466 fdfree(td);
1467 p->p_fd = tmp;
1468 } else
1469 FILEDESC_UNLOCK_FAST(p->p_fd);
1470 }
1471
1472 /*
1473 * Copy a filedesc structure.
1474 * A NULL pointer in returns a NULL reference, this is to ease callers,
1475 * not catch errors.
1476 */
1477 struct filedesc *
1478 fdcopy(struct filedesc *fdp)
1479 {
1480 struct filedesc *newfdp;
1481 int i;
1482
1483 /* Certain daemons might not have file descriptors. */
1484 if (fdp == NULL)
1485 return (NULL);
1486
1487 newfdp = fdinit(fdp);
1488 FILEDESC_LOCK_FAST(fdp);
1489 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1490 FILEDESC_UNLOCK_FAST(fdp);
1491 FILEDESC_LOCK(newfdp);
1492 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1493 FILEDESC_UNLOCK(newfdp);
1494 FILEDESC_LOCK_FAST(fdp);
1495 }
1496 /* copy everything except kqueue descriptors */
1497 newfdp->fd_freefile = -1;
1498 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1499 if (fdisused(fdp, i) &&
1500 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1501 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1502 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1503 fhold(newfdp->fd_ofiles[i]);
1504 newfdp->fd_lastfile = i;
1505 } else {
1506 if (newfdp->fd_freefile == -1)
1507 newfdp->fd_freefile = i;
1508 }
1509 }
1510 FILEDESC_UNLOCK_FAST(fdp);
1511 FILEDESC_LOCK(newfdp);
1512 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1513 if (newfdp->fd_ofiles[i] != NULL)
1514 fdused(newfdp, i);
1515 FILEDESC_UNLOCK(newfdp);
1516 FILEDESC_LOCK_FAST(fdp);
1517 if (newfdp->fd_freefile == -1)
1518 newfdp->fd_freefile = i;
1519 newfdp->fd_cmask = fdp->fd_cmask;
1520 FILEDESC_UNLOCK_FAST(fdp);
1521 return (newfdp);
1522 }
1523
1524 /*
1525 * Release a filedesc structure.
1526 */
1527 void
1528 fdfree(struct thread *td)
1529 {
1530 struct filedesc *fdp;
1531 struct file **fpp;
1532 int i, locked;
1533 struct filedesc_to_leader *fdtol;
1534 struct file *fp;
1535 struct vnode *cdir, *jdir, *rdir, *vp;
1536 struct flock lf;
1537
1538 /* Certain daemons might not have file descriptors. */
1539 fdp = td->td_proc->p_fd;
1540 if (fdp == NULL)
1541 return;
1542
1543 /* Check for special need to clear POSIX style locks */
1544 fdtol = td->td_proc->p_fdtol;
1545 if (fdtol != NULL) {
1546 FILEDESC_LOCK(fdp);
1547 KASSERT(fdtol->fdl_refcount > 0,
1548 ("filedesc_to_refcount botch: fdl_refcount=%d",
1549 fdtol->fdl_refcount));
1550 if (fdtol->fdl_refcount == 1 &&
1551 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1552 for (i = 0, fpp = fdp->fd_ofiles;
1553 i <= fdp->fd_lastfile;
1554 i++, fpp++) {
1555 if (*fpp == NULL ||
1556 (*fpp)->f_type != DTYPE_VNODE)
1557 continue;
1558 fp = *fpp;
1559 fhold(fp);
1560 FILEDESC_UNLOCK(fdp);
1561 lf.l_whence = SEEK_SET;
1562 lf.l_start = 0;
1563 lf.l_len = 0;
1564 lf.l_type = F_UNLCK;
1565 vp = fp->f_vnode;
1566 locked = VFS_LOCK_GIANT(vp->v_mount);
1567 (void) VOP_ADVLOCK(vp,
1568 (caddr_t)td->td_proc->
1569 p_leader,
1570 F_UNLCK,
1571 &lf,
1572 F_POSIX);
1573 VFS_UNLOCK_GIANT(locked);
1574 FILEDESC_LOCK(fdp);
1575 fdrop(fp, td);
1576 fpp = fdp->fd_ofiles + i;
1577 }
1578 }
1579 retry:
1580 if (fdtol->fdl_refcount == 1) {
1581 if (fdp->fd_holdleaderscount > 0 &&
1582 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1583 /*
1584 * close() or do_dup() has cleared a reference
1585 * in a shared file descriptor table.
1586 */
1587 fdp->fd_holdleaderswakeup = 1;
1588 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1589 PLOCK, "fdlhold", 0);
1590 goto retry;
1591 }
1592 if (fdtol->fdl_holdcount > 0) {
1593 /*
1594 * Ensure that fdtol->fdl_leader
1595 * remains valid in closef().
1596 */
1597 fdtol->fdl_wakeup = 1;
1598 msleep(fdtol, &fdp->fd_mtx,
1599 PLOCK, "fdlhold", 0);
1600 goto retry;
1601 }
1602 }
1603 fdtol->fdl_refcount--;
1604 if (fdtol->fdl_refcount == 0 &&
1605 fdtol->fdl_holdcount == 0) {
1606 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1607 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1608 } else
1609 fdtol = NULL;
1610 td->td_proc->p_fdtol = NULL;
1611 FILEDESC_UNLOCK(fdp);
1612 if (fdtol != NULL)
1613 FREE(fdtol, M_FILEDESC_TO_LEADER);
1614 }
1615 FILEDESC_LOCK_FAST(fdp);
1616 i = --fdp->fd_refcnt;
1617 FILEDESC_UNLOCK_FAST(fdp);
1618 if (i > 0)
1619 return;
1620 /*
1621 * We are the last reference to the structure, so we can
1622 * safely assume it will not change out from under us.
1623 */
1624 fpp = fdp->fd_ofiles;
1625 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1626 if (*fpp)
1627 (void) closef(*fpp, td);
1628 }
1629 FILEDESC_LOCK(fdp);
1630
1631 /* XXX This should happen earlier. */
1632 mtx_lock(&fdesc_mtx);
1633 td->td_proc->p_fd = NULL;
1634 mtx_unlock(&fdesc_mtx);
1635
1636 if (fdp->fd_nfiles > NDFILE)
1637 FREE(fdp->fd_ofiles, M_FILEDESC);
1638 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1639 FREE(fdp->fd_map, M_FILEDESC);
1640
1641 fdp->fd_nfiles = 0;
1642
1643 cdir = fdp->fd_cdir;
1644 fdp->fd_cdir = NULL;
1645 rdir = fdp->fd_rdir;
1646 fdp->fd_rdir = NULL;
1647 jdir = fdp->fd_jdir;
1648 fdp->fd_jdir = NULL;
1649 FILEDESC_UNLOCK(fdp);
1650
1651 if (cdir) {
1652 locked = VFS_LOCK_GIANT(cdir->v_mount);
1653 vrele(cdir);
1654 VFS_UNLOCK_GIANT(locked);
1655 }
1656 if (rdir) {
1657 locked = VFS_LOCK_GIANT(rdir->v_mount);
1658 vrele(rdir);
1659 VFS_UNLOCK_GIANT(locked);
1660 }
1661 if (jdir) {
1662 locked = VFS_LOCK_GIANT(jdir->v_mount);
1663 vrele(jdir);
1664 VFS_UNLOCK_GIANT(locked);
1665 }
1666
1667 fddrop(fdp);
1668 }
1669
1670 /*
1671 * For setugid programs, we don't want to people to use that setugidness
1672 * to generate error messages which write to a file which otherwise would
1673 * otherwise be off-limits to the process. We check for filesystems where
1674 * the vnode can change out from under us after execve (like [lin]procfs).
1675 *
1676 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1677 * sufficient. We also don't check for setugidness since we know we are.
1678 */
1679 static int
1680 is_unsafe(struct file *fp)
1681 {
1682 if (fp->f_type == DTYPE_VNODE) {
1683 struct vnode *vp = fp->f_vnode;
1684
1685 if ((vp->v_vflag & VV_PROCDEP) != 0)
1686 return (1);
1687 }
1688 return (0);
1689 }
1690
1691 /*
1692 * Make this setguid thing safe, if at all possible.
1693 */
1694 void
1695 setugidsafety(struct thread *td)
1696 {
1697 struct filedesc *fdp;
1698 int i;
1699
1700 /* Certain daemons might not have file descriptors. */
1701 fdp = td->td_proc->p_fd;
1702 if (fdp == NULL)
1703 return;
1704
1705 /*
1706 * Note: fdp->fd_ofiles may be reallocated out from under us while
1707 * we are blocked in a close. Be careful!
1708 */
1709 FILEDESC_LOCK(fdp);
1710 for (i = 0; i <= fdp->fd_lastfile; i++) {
1711 if (i > 2)
1712 break;
1713 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1714 struct file *fp;
1715
1716 knote_fdclose(td, i);
1717 /*
1718 * NULL-out descriptor prior to close to avoid
1719 * a race while close blocks.
1720 */
1721 fp = fdp->fd_ofiles[i];
1722 fdp->fd_ofiles[i] = NULL;
1723 fdp->fd_ofileflags[i] = 0;
1724 fdunused(fdp, i);
1725 FILEDESC_UNLOCK(fdp);
1726 (void) closef(fp, td);
1727 FILEDESC_LOCK(fdp);
1728 }
1729 }
1730 FILEDESC_UNLOCK(fdp);
1731 }
1732
1733 void
1734 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1735 {
1736
1737 FILEDESC_LOCK(fdp);
1738 if (fdp->fd_ofiles[idx] == fp) {
1739 fdp->fd_ofiles[idx] = NULL;
1740 fdunused(fdp, idx);
1741 FILEDESC_UNLOCK(fdp);
1742 fdrop(fp, td);
1743 } else {
1744 FILEDESC_UNLOCK(fdp);
1745 }
1746 }
1747
1748 /*
1749 * Close any files on exec?
1750 */
1751 void
1752 fdcloseexec(struct thread *td)
1753 {
1754 struct filedesc *fdp;
1755 int i;
1756
1757 /* Certain daemons might not have file descriptors. */
1758 fdp = td->td_proc->p_fd;
1759 if (fdp == NULL)
1760 return;
1761
1762 FILEDESC_LOCK(fdp);
1763
1764 /*
1765 * We cannot cache fd_ofiles or fd_ofileflags since operations
1766 * may block and rip them out from under us.
1767 */
1768 for (i = 0; i <= fdp->fd_lastfile; i++) {
1769 if (fdp->fd_ofiles[i] != NULL &&
1770 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1771 struct file *fp;
1772
1773 knote_fdclose(td, i);
1774 /*
1775 * NULL-out descriptor prior to close to avoid
1776 * a race while close blocks.
1777 */
1778 fp = fdp->fd_ofiles[i];
1779 fdp->fd_ofiles[i] = NULL;
1780 fdp->fd_ofileflags[i] = 0;
1781 fdunused(fdp, i);
1782 FILEDESC_UNLOCK(fdp);
1783 (void) closef(fp, td);
1784 FILEDESC_LOCK(fdp);
1785 }
1786 }
1787 FILEDESC_UNLOCK(fdp);
1788 }
1789
1790 /*
1791 * It is unsafe for set[ug]id processes to be started with file
1792 * descriptors 0..2 closed, as these descriptors are given implicit
1793 * significance in the Standard C library. fdcheckstd() will create a
1794 * descriptor referencing /dev/null for each of stdin, stdout, and
1795 * stderr that is not already open.
1796 */
1797 int
1798 fdcheckstd(struct thread *td)
1799 {
1800 struct nameidata nd;
1801 struct filedesc *fdp;
1802 struct file *fp;
1803 register_t retval;
1804 int fd, i, error, flags, devnull;
1805
1806 fdp = td->td_proc->p_fd;
1807 if (fdp == NULL)
1808 return (0);
1809 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1810 devnull = -1;
1811 error = 0;
1812 for (i = 0; i < 3; i++) {
1813 if (fdp->fd_ofiles[i] != NULL)
1814 continue;
1815 if (devnull < 0) {
1816 int vfslocked;
1817 error = falloc(td, &fp, &fd);
1818 if (error != 0)
1819 break;
1820 /* Note extra ref on `fp' held for us by falloc(). */
1821 KASSERT(fd == i, ("oof, we didn't get our fd"));
1822 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE,
1823 "/dev/null", td);
1824 flags = FREAD | FWRITE;
1825 error = vn_open(&nd, &flags, 0, fd);
1826 if (error != 0) {
1827 /*
1828 * Someone may have closed the entry in the
1829 * file descriptor table, so check it hasn't
1830 * changed before dropping the reference count.
1831 */
1832 FILEDESC_LOCK(fdp);
1833 KASSERT(fdp->fd_ofiles[fd] == fp,
1834 ("table not shared, how did it change?"));
1835 fdp->fd_ofiles[fd] = NULL;
1836 fdunused(fdp, fd);
1837 FILEDESC_UNLOCK(fdp);
1838 fdrop(fp, td);
1839 fdrop(fp, td);
1840 break;
1841 }
1842 vfslocked = NDHASGIANT(&nd);
1843 NDFREE(&nd, NDF_ONLY_PNBUF);
1844 fp->f_flag = flags;
1845 fp->f_vnode = nd.ni_vp;
1846 if (fp->f_data == NULL)
1847 fp->f_data = nd.ni_vp;
1848 if (fp->f_ops == &badfileops)
1849 fp->f_ops = &vnops;
1850 fp->f_type = DTYPE_VNODE;
1851 VOP_UNLOCK(nd.ni_vp, 0, td);
1852 VFS_UNLOCK_GIANT(vfslocked);
1853 devnull = fd;
1854 fdrop(fp, td);
1855 } else {
1856 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1857 if (error != 0)
1858 break;
1859 }
1860 }
1861 return (error);
1862 }
1863
1864 /*
1865 * Internal form of close.
1866 * Decrement reference count on file structure.
1867 * Note: td may be NULL when closing a file that was being passed in a
1868 * message.
1869 *
1870 * XXXRW: Giant is not required for the caller, but often will be held; this
1871 * makes it moderately likely the Giant will be recursed in the VFS case.
1872 */
1873 int
1874 closef(struct file *fp, struct thread *td)
1875 {
1876 struct vnode *vp;
1877 struct flock lf;
1878 struct filedesc_to_leader *fdtol;
1879 struct filedesc *fdp;
1880
1881 /*
1882 * POSIX record locking dictates that any close releases ALL
1883 * locks owned by this process. This is handled by setting
1884 * a flag in the unlock to free ONLY locks obeying POSIX
1885 * semantics, and not to free BSD-style file locks.
1886 * If the descriptor was in a message, POSIX-style locks
1887 * aren't passed with the descriptor, and the thread pointer
1888 * will be NULL. Callers should be careful only to pass a
1889 * NULL thread pointer when there really is no owning
1890 * context that might have locks, or the locks will be
1891 * leaked.
1892 */
1893 if (fp->f_type == DTYPE_VNODE && td != NULL) {
1894 int vfslocked;
1895
1896 vp = fp->f_vnode;
1897 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1898 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1899 lf.l_whence = SEEK_SET;
1900 lf.l_start = 0;
1901 lf.l_len = 0;
1902 lf.l_type = F_UNLCK;
1903 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1904 F_UNLCK, &lf, F_POSIX);
1905 }
1906 fdtol = td->td_proc->p_fdtol;
1907 if (fdtol != NULL) {
1908 /*
1909 * Handle special case where file descriptor table
1910 * is shared between multiple process leaders.
1911 */
1912 fdp = td->td_proc->p_fd;
1913 FILEDESC_LOCK(fdp);
1914 for (fdtol = fdtol->fdl_next;
1915 fdtol != td->td_proc->p_fdtol;
1916 fdtol = fdtol->fdl_next) {
1917 if ((fdtol->fdl_leader->p_flag &
1918 P_ADVLOCK) == 0)
1919 continue;
1920 fdtol->fdl_holdcount++;
1921 FILEDESC_UNLOCK(fdp);
1922 lf.l_whence = SEEK_SET;
1923 lf.l_start = 0;
1924 lf.l_len = 0;
1925 lf.l_type = F_UNLCK;
1926 vp = fp->f_vnode;
1927 (void) VOP_ADVLOCK(vp,
1928 (caddr_t)fdtol->fdl_leader,
1929 F_UNLCK, &lf, F_POSIX);
1930 FILEDESC_LOCK(fdp);
1931 fdtol->fdl_holdcount--;
1932 if (fdtol->fdl_holdcount == 0 &&
1933 fdtol->fdl_wakeup != 0) {
1934 fdtol->fdl_wakeup = 0;
1935 wakeup(fdtol);
1936 }
1937 }
1938 FILEDESC_UNLOCK(fdp);
1939 }
1940 VFS_UNLOCK_GIANT(vfslocked);
1941 }
1942 return (fdrop(fp, td));
1943 }
1944
1945 /*
1946 * Extract the file pointer associated with the specified descriptor for
1947 * the current user process.
1948 *
1949 * If the descriptor doesn't exist, EBADF is returned.
1950 *
1951 * If the descriptor exists but doesn't match 'flags' then
1952 * return EBADF for read attempts and EINVAL for write attempts.
1953 *
1954 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1955 * It should be dropped with fdrop().
1956 * If it is not set, then the refcount will not be bumped however the
1957 * thread's filedesc struct will be returned locked (for fgetsock).
1958 *
1959 * If an error occured the non-zero error is returned and *fpp is set to NULL.
1960 * Otherwise *fpp is set and zero is returned.
1961 */
1962 static __inline int
1963 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1964 {
1965 struct filedesc *fdp;
1966 struct file *fp;
1967
1968 *fpp = NULL;
1969 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1970 return (EBADF);
1971 FILEDESC_LOCK(fdp);
1972 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
1973 FILEDESC_UNLOCK(fdp);
1974 return (EBADF);
1975 }
1976
1977 /*
1978 * Note: FREAD failure returns EBADF to maintain backwards
1979 * compatibility with what routines returned before.
1980 *
1981 * Only one flag, or 0, may be specified.
1982 */
1983 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
1984 FILEDESC_UNLOCK(fdp);
1985 return (EBADF);
1986 }
1987 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
1988 FILEDESC_UNLOCK(fdp);
1989 return (EINVAL);
1990 }
1991 if (hold) {
1992 fhold(fp);
1993 FILEDESC_UNLOCK(fdp);
1994 }
1995 *fpp = fp;
1996 return (0);
1997 }
1998
1999 int
2000 fget(struct thread *td, int fd, struct file **fpp)
2001 {
2002
2003 return(_fget(td, fd, fpp, 0, 1));
2004 }
2005
2006 int
2007 fget_read(struct thread *td, int fd, struct file **fpp)
2008 {
2009
2010 return(_fget(td, fd, fpp, FREAD, 1));
2011 }
2012
2013 int
2014 fget_write(struct thread *td, int fd, struct file **fpp)
2015 {
2016
2017 return(_fget(td, fd, fpp, FWRITE, 1));
2018 }
2019
2020 /*
2021 * Like fget() but loads the underlying vnode, or returns an error if
2022 * the descriptor does not represent a vnode. Note that pipes use vnodes
2023 * but never have VM objects. The returned vnode will be vref()d.
2024 *
2025 * XXX: what about the unused flags ?
2026 */
2027 static __inline int
2028 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2029 {
2030 struct file *fp;
2031 int error;
2032
2033 *vpp = NULL;
2034 if ((error = _fget(td, fd, &fp, flags, 0)) != 0)
2035 return (error);
2036 if (fp->f_vnode == NULL) {
2037 error = EINVAL;
2038 } else {
2039 *vpp = fp->f_vnode;
2040 vref(*vpp);
2041 }
2042 FILEDESC_UNLOCK(td->td_proc->p_fd);
2043 return (error);
2044 }
2045
2046 int
2047 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2048 {
2049
2050 return (_fgetvp(td, fd, vpp, 0));
2051 }
2052
2053 int
2054 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2055 {
2056
2057 return (_fgetvp(td, fd, vpp, FREAD));
2058 }
2059
2060 #ifdef notyet
2061 int
2062 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2063 {
2064
2065 return (_fgetvp(td, fd, vpp, FWRITE));
2066 }
2067 #endif
2068
2069 /*
2070 * Like fget() but loads the underlying socket, or returns an error if
2071 * the descriptor does not represent a socket.
2072 *
2073 * We bump the ref count on the returned socket. XXX Also obtain the SX
2074 * lock in the future.
2075 */
2076 int
2077 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2078 {
2079 struct file *fp;
2080 int error;
2081
2082 NET_ASSERT_GIANT();
2083
2084 *spp = NULL;
2085 if (fflagp != NULL)
2086 *fflagp = 0;
2087 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2088 return (error);
2089 if (fp->f_type != DTYPE_SOCKET) {
2090 error = ENOTSOCK;
2091 } else {
2092 *spp = fp->f_data;
2093 if (fflagp)
2094 *fflagp = fp->f_flag;
2095 SOCK_LOCK(*spp);
2096 soref(*spp);
2097 SOCK_UNLOCK(*spp);
2098 }
2099 FILEDESC_UNLOCK(td->td_proc->p_fd);
2100 return (error);
2101 }
2102
2103 /*
2104 * Drop the reference count on the socket and XXX release the SX lock in
2105 * the future. The last reference closes the socket.
2106 */
2107 void
2108 fputsock(struct socket *so)
2109 {
2110
2111 NET_ASSERT_GIANT();
2112 ACCEPT_LOCK();
2113 SOCK_LOCK(so);
2114 sorele(so);
2115 }
2116
2117 int
2118 fdrop(struct file *fp, struct thread *td)
2119 {
2120
2121 FILE_LOCK(fp);
2122 return (fdrop_locked(fp, td));
2123 }
2124
2125 /*
2126 * Drop reference on struct file passed in, may call closef if the
2127 * reference hits zero.
2128 * Expects struct file locked, and will unlock it.
2129 */
2130 static int
2131 fdrop_locked(struct file *fp, struct thread *td)
2132 {
2133 int error;
2134
2135 FILE_LOCK_ASSERT(fp, MA_OWNED);
2136
2137 if (--fp->f_count > 0) {
2138 FILE_UNLOCK(fp);
2139 return (0);
2140 }
2141 /* We have the last ref so we can proceed without the file lock. */
2142 FILE_UNLOCK(fp);
2143 if (fp->f_count < 0)
2144 panic("fdrop: count < 0");
2145 if (fp->f_ops != &badfileops)
2146 error = fo_close(fp, td);
2147 else
2148 error = 0;
2149
2150 sx_xlock(&filelist_lock);
2151 LIST_REMOVE(fp, f_list);
2152 openfiles--;
2153 sx_xunlock(&filelist_lock);
2154 crfree(fp->f_cred);
2155 uma_zfree(file_zone, fp);
2156
2157 return (error);
2158 }
2159
2160 /*
2161 * Apply an advisory lock on a file descriptor.
2162 *
2163 * Just attempt to get a record lock of the requested type on
2164 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2165 */
2166 #ifndef _SYS_SYSPROTO_H_
2167 struct flock_args {
2168 int fd;
2169 int how;
2170 };
2171 #endif
2172 /*
2173 * MPSAFE
2174 */
2175 /* ARGSUSED */
2176 int
2177 flock(struct thread *td, struct flock_args *uap)
2178 {
2179 struct file *fp;
2180 struct vnode *vp;
2181 struct flock lf;
2182 int error;
2183
2184 if ((error = fget(td, uap->fd, &fp)) != 0)
2185 return (error);
2186 if (fp->f_type != DTYPE_VNODE) {
2187 fdrop(fp, td);
2188 return (EOPNOTSUPP);
2189 }
2190
2191 mtx_lock(&Giant);
2192 vp = fp->f_vnode;
2193 lf.l_whence = SEEK_SET;
2194 lf.l_start = 0;
2195 lf.l_len = 0;
2196 if (uap->how & LOCK_UN) {
2197 lf.l_type = F_UNLCK;
2198 FILE_LOCK(fp);
2199 fp->f_flag &= ~FHASLOCK;
2200 FILE_UNLOCK(fp);
2201 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2202 goto done2;
2203 }
2204 if (uap->how & LOCK_EX)
2205 lf.l_type = F_WRLCK;
2206 else if (uap->how & LOCK_SH)
2207 lf.l_type = F_RDLCK;
2208 else {
2209 error = EBADF;
2210 goto done2;
2211 }
2212 FILE_LOCK(fp);
2213 fp->f_flag |= FHASLOCK;
2214 FILE_UNLOCK(fp);
2215 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2216 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2217 done2:
2218 fdrop(fp, td);
2219 mtx_unlock(&Giant);
2220 return (error);
2221 }
2222 /*
2223 * Duplicate the specified descriptor to a free descriptor.
2224 */
2225 int
2226 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2227 {
2228 struct file *wfp;
2229 struct file *fp;
2230
2231 /*
2232 * If the to-be-dup'd fd number is greater than the allowed number
2233 * of file descriptors, or the fd to be dup'd has already been
2234 * closed, then reject.
2235 */
2236 FILEDESC_LOCK(fdp);
2237 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2238 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2239 FILEDESC_UNLOCK(fdp);
2240 return (EBADF);
2241 }
2242
2243 /*
2244 * There are two cases of interest here.
2245 *
2246 * For ENODEV simply dup (dfd) to file descriptor
2247 * (indx) and return.
2248 *
2249 * For ENXIO steal away the file structure from (dfd) and
2250 * store it in (indx). (dfd) is effectively closed by
2251 * this operation.
2252 *
2253 * Any other error code is just returned.
2254 */
2255 switch (error) {
2256 case ENODEV:
2257 /*
2258 * Check that the mode the file is being opened for is a
2259 * subset of the mode of the existing descriptor.
2260 */
2261 FILE_LOCK(wfp);
2262 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2263 FILE_UNLOCK(wfp);
2264 FILEDESC_UNLOCK(fdp);
2265 return (EACCES);
2266 }
2267 fp = fdp->fd_ofiles[indx];
2268 fdp->fd_ofiles[indx] = wfp;
2269 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2270 if (fp == NULL)
2271 fdused(fdp, indx);
2272 fhold_locked(wfp);
2273 FILE_UNLOCK(wfp);
2274 FILEDESC_UNLOCK(fdp);
2275 if (fp != NULL) {
2276 /*
2277 * We now own the reference to fp that the ofiles[]
2278 * array used to own. Release it.
2279 */
2280 FILE_LOCK(fp);
2281 fdrop_locked(fp, td);
2282 }
2283 return (0);
2284
2285 case ENXIO:
2286 /*
2287 * Steal away the file pointer from dfd and stuff it into indx.
2288 */
2289 fp = fdp->fd_ofiles[indx];
2290 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2291 fdp->fd_ofiles[dfd] = NULL;
2292 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2293 fdp->fd_ofileflags[dfd] = 0;
2294 fdunused(fdp, dfd);
2295 if (fp == NULL)
2296 fdused(fdp, indx);
2297 if (fp != NULL)
2298 FILE_LOCK(fp);
2299
2300 /*
2301 * We now own the reference to fp that the ofiles[] array
2302 * used to own. Release it.
2303 */
2304 if (fp != NULL)
2305 fdrop_locked(fp, td);
2306
2307 FILEDESC_UNLOCK(fdp);
2308
2309 return (0);
2310
2311 default:
2312 FILEDESC_UNLOCK(fdp);
2313 return (error);
2314 }
2315 /* NOTREACHED */
2316 }
2317
2318 /*
2319 * Scan all active processes to see if any of them have a current
2320 * or root directory of `olddp'. If so, replace them with the new
2321 * mount point.
2322 */
2323 void
2324 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2325 {
2326 struct filedesc *fdp;
2327 struct proc *p;
2328 int nrele;
2329
2330 if (vrefcnt(olddp) == 1)
2331 return;
2332 sx_slock(&allproc_lock);
2333 LIST_FOREACH(p, &allproc, p_list) {
2334 fdp = fdhold(p);
2335 if (fdp == NULL)
2336 continue;
2337 nrele = 0;
2338 FILEDESC_LOCK_FAST(fdp);
2339 if (fdp->fd_cdir == olddp) {
2340 vref(newdp);
2341 fdp->fd_cdir = newdp;
2342 nrele++;
2343 }
2344 if (fdp->fd_rdir == olddp) {
2345 vref(newdp);
2346 fdp->fd_rdir = newdp;
2347 nrele++;
2348 }
2349 FILEDESC_UNLOCK_FAST(fdp);
2350 fddrop(fdp);
2351 while (nrele--)
2352 vrele(olddp);
2353 }
2354 sx_sunlock(&allproc_lock);
2355 if (rootvnode == olddp) {
2356 vrele(rootvnode);
2357 vref(newdp);
2358 rootvnode = newdp;
2359 }
2360 }
2361
2362 struct filedesc_to_leader *
2363 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2364 {
2365 struct filedesc_to_leader *fdtol;
2366
2367 MALLOC(fdtol, struct filedesc_to_leader *,
2368 sizeof(struct filedesc_to_leader),
2369 M_FILEDESC_TO_LEADER,
2370 M_WAITOK);
2371 fdtol->fdl_refcount = 1;
2372 fdtol->fdl_holdcount = 0;
2373 fdtol->fdl_wakeup = 0;
2374 fdtol->fdl_leader = leader;
2375 if (old != NULL) {
2376 FILEDESC_LOCK(fdp);
2377 fdtol->fdl_next = old->fdl_next;
2378 fdtol->fdl_prev = old;
2379 old->fdl_next = fdtol;
2380 fdtol->fdl_next->fdl_prev = fdtol;
2381 FILEDESC_UNLOCK(fdp);
2382 } else {
2383 fdtol->fdl_next = fdtol;
2384 fdtol->fdl_prev = fdtol;
2385 }
2386 return (fdtol);
2387 }
2388
2389 /*
2390 * Get file structures.
2391 */
2392 static int
2393 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2394 {
2395 struct xfile xf;
2396 struct filedesc *fdp;
2397 struct file *fp;
2398 struct proc *p;
2399 int error, n;
2400
2401 /*
2402 * Note: because the number of file descriptors is calculated
2403 * in different ways for sizing vs returning the data,
2404 * there is information leakage from the first loop. However,
2405 * it is of a similar order of magnitude to the leakage from
2406 * global system statistics such as kern.openfiles.
2407 */
2408 error = sysctl_wire_old_buffer(req, 0);
2409 if (error != 0)
2410 return (error);
2411 if (req->oldptr == NULL) {
2412 n = 16; /* A slight overestimate. */
2413 sx_slock(&filelist_lock);
2414 LIST_FOREACH(fp, &filehead, f_list) {
2415 /*
2416 * We should grab the lock, but this is an
2417 * estimate, so does it really matter?
2418 */
2419 /* mtx_lock(fp->f_mtxp); */
2420 n += fp->f_count;
2421 /* mtx_unlock(f->f_mtxp); */
2422 }
2423 sx_sunlock(&filelist_lock);
2424 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2425 }
2426 error = 0;
2427 bzero(&xf, sizeof(xf));
2428 xf.xf_size = sizeof(xf);
2429 sx_slock(&allproc_lock);
2430 LIST_FOREACH(p, &allproc, p_list) {
2431 if (p->p_state == PRS_NEW)
2432 continue;
2433 PROC_LOCK(p);
2434 if (p_cansee(req->td, p) != 0) {
2435 PROC_UNLOCK(p);
2436 continue;
2437 }
2438 xf.xf_pid = p->p_pid;
2439 xf.xf_uid = p->p_ucred->cr_uid;
2440 PROC_UNLOCK(p);
2441 fdp = fdhold(p);
2442 if (fdp == NULL)
2443 continue;
2444 FILEDESC_LOCK_FAST(fdp);
2445 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2446 if ((fp = fdp->fd_ofiles[n]) == NULL)
2447 continue;
2448 xf.xf_fd = n;
2449 xf.xf_file = fp;
2450 xf.xf_data = fp->f_data;
2451 xf.xf_vnode = fp->f_vnode;
2452 xf.xf_type = fp->f_type;
2453 xf.xf_count = fp->f_count;
2454 xf.xf_msgcount = fp->f_msgcount;
2455 xf.xf_offset = fp->f_offset;
2456 xf.xf_flag = fp->f_flag;
2457 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2458 if (error)
2459 break;
2460 }
2461 FILEDESC_UNLOCK_FAST(fdp);
2462 fddrop(fdp);
2463 if (error)
2464 break;
2465 }
2466 sx_sunlock(&allproc_lock);
2467 return (error);
2468 }
2469
2470 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2471 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2472
2473 #ifdef DDB
2474 /*
2475 * For the purposes of debugging, generate a human-readable string for the
2476 * file type.
2477 */
2478 static const char *
2479 file_type_to_name(short type)
2480 {
2481
2482 switch (type) {
2483 case 0:
2484 return ("zero");
2485 case DTYPE_VNODE:
2486 return ("vnod");
2487 case DTYPE_SOCKET:
2488 return ("sock");
2489 case DTYPE_PIPE:
2490 return ("pipe");
2491 case DTYPE_FIFO:
2492 return ("fifo");
2493 case DTYPE_CRYPTO:
2494 return ("crpt");
2495 default:
2496 return ("unkn");
2497 }
2498 }
2499
2500 /*
2501 * For the purposes of debugging, identify a process (if any, perhaps one of
2502 * many) that references the passed file in its file descriptor array. Return
2503 * NULL if none.
2504 */
2505 static struct proc *
2506 file_to_first_proc(struct file *fp)
2507 {
2508 struct filedesc *fdp;
2509 struct proc *p;
2510 int n;
2511
2512 LIST_FOREACH(p, &allproc, p_list) {
2513 if (p->p_state == PRS_NEW)
2514 continue;
2515 fdp = p->p_fd;
2516 if (fdp == NULL)
2517 continue;
2518 for (n = 0; n < fdp->fd_nfiles; n++) {
2519 if (fp == fdp->fd_ofiles[n])
2520 return (p);
2521 }
2522 }
2523 return (NULL);
2524 }
2525
2526 DB_SHOW_COMMAND(files, db_show_files)
2527 {
2528 struct file *fp;
2529 struct proc *p;
2530
2531 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", "File",
2532 "Type", "Data", "Flag", "GCFl", "Count", "MCount", "Vnode",
2533 "FPID", "FCmd");
2534 LIST_FOREACH(fp, &filehead, f_list) {
2535 p = file_to_first_proc(fp);
2536 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2537 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2538 fp->f_gcflag, fp->f_count, fp->f_msgcount, fp->f_vnode,
2539 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2540 }
2541 }
2542 #endif
2543
2544 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2545 &maxfilesperproc, 0, "Maximum files allowed open per process");
2546
2547 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2548 &maxfiles, 0, "Maximum number of files");
2549
2550 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2551 &openfiles, 0, "System-wide number of open files");
2552
2553 /* ARGSUSED*/
2554 static void
2555 filelistinit(void *dummy)
2556 {
2557
2558 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2559 NULL, NULL, UMA_ALIGN_PTR, 0);
2560 sx_init(&filelist_lock, "filelist lock");
2561 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2562 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2563 }
2564 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2565
2566 /*-------------------------------------------------------------------*/
2567
2568 static int
2569 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2570 {
2571
2572 return (EBADF);
2573 }
2574
2575 static int
2576 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2577 {
2578
2579 return (EBADF);
2580 }
2581
2582 static int
2583 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2584 {
2585
2586 return (0);
2587 }
2588
2589 static int
2590 badfo_kqfilter(struct file *fp, struct knote *kn)
2591 {
2592
2593 return (0);
2594 }
2595
2596 static int
2597 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2598 {
2599
2600 return (EBADF);
2601 }
2602
2603 static int
2604 badfo_close(struct file *fp, struct thread *td)
2605 {
2606
2607 return (EBADF);
2608 }
2609
2610 struct fileops badfileops = {
2611 .fo_read = badfo_readwrite,
2612 .fo_write = badfo_readwrite,
2613 .fo_ioctl = badfo_ioctl,
2614 .fo_poll = badfo_poll,
2615 .fo_kqfilter = badfo_kqfilter,
2616 .fo_stat = badfo_stat,
2617 .fo_close = badfo_close,
2618 };
2619
2620
2621 /*-------------------------------------------------------------------*/
2622
2623 /*
2624 * File Descriptor pseudo-device driver (/dev/fd/).
2625 *
2626 * Opening minor device N dup()s the file (if any) connected to file
2627 * descriptor N belonging to the calling process. Note that this driver
2628 * consists of only the ``open()'' routine, because all subsequent
2629 * references to this file will be direct to the other driver.
2630 *
2631 * XXX: we could give this one a cloning event handler if necessary.
2632 */
2633
2634 /* ARGSUSED */
2635 static int
2636 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2637 {
2638
2639 /*
2640 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2641 * the file descriptor being sought for duplication. The error
2642 * return ensures that the vnode for this device will be released
2643 * by vn_open. Open will detect this special error and take the
2644 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2645 * will simply report the error.
2646 */
2647 td->td_dupfd = dev2unit(dev);
2648 return (ENODEV);
2649 }
2650
2651 static struct cdevsw fildesc_cdevsw = {
2652 .d_version = D_VERSION,
2653 .d_flags = D_NEEDGIANT,
2654 .d_open = fdopen,
2655 .d_name = "FD",
2656 };
2657
2658 static void
2659 fildesc_drvinit(void *unused)
2660 {
2661 struct cdev *dev;
2662
2663 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2664 make_dev_alias(dev, "stdin");
2665 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2666 make_dev_alias(dev, "stdout");
2667 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2668 make_dev_alias(dev, "stderr");
2669 }
2670
2671 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
Cache object: 4ce9d8f514d05d99ce5d013f30cfe203
|