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/5.4/sys/kern/kern_descrip.c 145335 2005-04-20 19:11:07Z cvs2svn $");
39
40 #include "opt_compat.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44
45 #include <sys/conf.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/limits.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mount.h>
56 #include <sys/mutex.h>
57 #include <sys/namei.h>
58 #include <sys/proc.h>
59 #include <sys/resourcevar.h>
60 #include <sys/signalvar.h>
61 #include <sys/socketvar.h>
62 #include <sys/stat.h>
63 #include <sys/sx.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysproto.h>
67 #include <sys/unistd.h>
68 #include <sys/vnode.h>
69
70 #include <vm/uma.h>
71
72 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
73 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
74 "file desc to leader structures");
75 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
76
77 static uma_zone_t file_zone;
78
79
80 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
81 enum dup_type { DUP_VARIABLE, DUP_FIXED };
82
83 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
84 register_t *retval);
85 static int fd_first_free(struct filedesc *, int, int);
86 static int fd_last_used(struct filedesc *, int, int);
87 static void fdgrowtable(struct filedesc *, int);
88 static int fdrop_locked(struct file *fp, struct thread *td);
89 static void fdunused(struct filedesc *fdp, int fd);
90 static void fdused(struct filedesc *fdp, int fd);
91
92 /*
93 * A process is initially started out with NDFILE descriptors stored within
94 * this structure, selected to be enough for typical applications based on
95 * the historical limit of 20 open files (and the usage of descriptors by
96 * shells). If these descriptors are exhausted, a larger descriptor table
97 * may be allocated, up to a process' resource limit; the internal arrays
98 * are then unused.
99 */
100 #define NDFILE 20
101 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
102 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
103 #define NDSLOT(x) ((x) / NDENTRIES)
104 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
105 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
106
107 /*
108 * Storage required per open file descriptor.
109 */
110 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
111
112 /*
113 * Basic allocation of descriptors:
114 * one of the above, plus arrays for NDFILE descriptors.
115 */
116 struct filedesc0 {
117 struct filedesc fd_fd;
118 /*
119 * These arrays are used when the number of open files is
120 * <= NDFILE, and are then pointed to by the pointers above.
121 */
122 struct file *fd_dfiles[NDFILE];
123 char fd_dfileflags[NDFILE];
124 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
125 };
126
127 /*
128 * Descriptor management.
129 */
130 struct filelist filehead; /* head of list of open files */
131 int nfiles; /* actual number of open files */
132 struct sx filelist_lock; /* sx to protect filelist */
133 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
134
135 /* A mutex to protect the association between a proc and filedesc. */
136 static struct mtx fdesc_mtx;
137
138 /*
139 * Find the first zero bit in the given bitmap, starting at low and not
140 * exceeding size - 1.
141 */
142 static int
143 fd_first_free(struct filedesc *fdp, int low, int size)
144 {
145 NDSLOTTYPE *map = fdp->fd_map;
146 NDSLOTTYPE mask;
147 int off, maxoff;
148
149 if (low >= size)
150 return (low);
151
152 off = NDSLOT(low);
153 if (low % NDENTRIES) {
154 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
155 if ((mask &= ~map[off]) != 0UL)
156 return (off * NDENTRIES + ffsl(mask) - 1);
157 ++off;
158 }
159 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
160 if (map[off] != ~0UL)
161 return (off * NDENTRIES + ffsl(~map[off]) - 1);
162 return (size);
163 }
164
165 /*
166 * Find the highest non-zero bit in the given bitmap, starting at low and
167 * not exceeding size - 1.
168 */
169 static int
170 fd_last_used(struct filedesc *fdp, int low, int size)
171 {
172 NDSLOTTYPE *map = fdp->fd_map;
173 NDSLOTTYPE mask;
174 int off, minoff;
175
176 if (low >= size)
177 return (-1);
178
179 off = NDSLOT(size);
180 if (size % NDENTRIES) {
181 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
182 if ((mask &= map[off]) != 0)
183 return (off * NDENTRIES + flsl(mask) - 1);
184 --off;
185 }
186 for (minoff = NDSLOT(low); off >= minoff; --off)
187 if (map[off] != 0)
188 return (off * NDENTRIES + flsl(map[off]) - 1);
189 return (size - 1);
190 }
191
192 static int
193 fdisused(struct filedesc *fdp, int fd)
194 {
195 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
196 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
197 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
198 }
199
200 /*
201 * Mark a file descriptor as used.
202 */
203 static void
204 fdused(struct filedesc *fdp, int fd)
205 {
206 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
207 KASSERT(!fdisused(fdp, fd),
208 ("fd already used"));
209 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
210 if (fd > fdp->fd_lastfile)
211 fdp->fd_lastfile = fd;
212 if (fd == fdp->fd_freefile)
213 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
214 }
215
216 /*
217 * Mark a file descriptor as unused.
218 */
219 static void
220 fdunused(struct filedesc *fdp, int fd)
221 {
222 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
223 KASSERT(fdisused(fdp, fd),
224 ("fd is already unused"));
225 KASSERT(fdp->fd_ofiles[fd] == NULL,
226 ("fd is still in use"));
227 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
228 if (fd < fdp->fd_freefile)
229 fdp->fd_freefile = fd;
230 if (fd == fdp->fd_lastfile)
231 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
232 }
233
234 /*
235 * System calls on descriptors.
236 */
237 #ifndef _SYS_SYSPROTO_H_
238 struct getdtablesize_args {
239 int dummy;
240 };
241 #endif
242 /*
243 * MPSAFE
244 */
245 /* ARGSUSED */
246 int
247 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
248 {
249 struct proc *p = td->td_proc;
250
251 PROC_LOCK(p);
252 td->td_retval[0] =
253 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
254 PROC_UNLOCK(p);
255 return (0);
256 }
257
258 /*
259 * Duplicate a file descriptor to a particular value.
260 *
261 * note: keep in mind that a potential race condition exists when closing
262 * descriptors from a shared descriptor table (via rfork).
263 */
264 #ifndef _SYS_SYSPROTO_H_
265 struct dup2_args {
266 u_int from;
267 u_int to;
268 };
269 #endif
270 /*
271 * MPSAFE
272 */
273 /* ARGSUSED */
274 int
275 dup2(struct thread *td, struct dup2_args *uap)
276 {
277
278 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
279 td->td_retval));
280 }
281
282 /*
283 * Duplicate a file descriptor.
284 */
285 #ifndef _SYS_SYSPROTO_H_
286 struct dup_args {
287 u_int fd;
288 };
289 #endif
290 /*
291 * MPSAFE
292 */
293 /* ARGSUSED */
294 int
295 dup(struct thread *td, struct dup_args *uap)
296 {
297
298 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
299 }
300
301 /*
302 * The file control system call.
303 */
304 #ifndef _SYS_SYSPROTO_H_
305 struct fcntl_args {
306 int fd;
307 int cmd;
308 long arg;
309 };
310 #endif
311 /*
312 * MPSAFE
313 */
314 /* ARGSUSED */
315 int
316 fcntl(struct thread *td, struct fcntl_args *uap)
317 {
318 struct flock fl;
319 intptr_t arg;
320 int error;
321
322 error = 0;
323 switch (uap->cmd) {
324 case F_GETLK:
325 case F_SETLK:
326 case F_SETLKW:
327 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
328 arg = (intptr_t)&fl;
329 break;
330 default:
331 arg = uap->arg;
332 break;
333 }
334 if (error)
335 return (error);
336 error = kern_fcntl(td, uap->fd, uap->cmd, arg);
337 if (error)
338 return (error);
339 if (uap->cmd == F_GETLK)
340 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
341 return (error);
342 }
343
344 int
345 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
346 {
347 struct filedesc *fdp;
348 struct flock *flp;
349 struct file *fp;
350 struct proc *p;
351 char *pop;
352 struct vnode *vp;
353 u_int newmin;
354 int error, flg, tmp;
355 int giant_locked;
356
357 /*
358 * XXXRW: Some fcntl() calls require Giant -- others don't. Try to
359 * avoid grabbing Giant for calls we know don't need it.
360 */
361 switch (cmd) {
362 case F_DUPFD:
363 case F_GETFD:
364 case F_SETFD:
365 case F_GETFL:
366 giant_locked = 0;
367 break;
368
369 default:
370 giant_locked = 1;
371 mtx_lock(&Giant);
372 }
373
374 error = 0;
375 flg = F_POSIX;
376 p = td->td_proc;
377 fdp = p->p_fd;
378 FILEDESC_LOCK(fdp);
379 if ((unsigned)fd >= fdp->fd_nfiles ||
380 (fp = fdp->fd_ofiles[fd]) == NULL) {
381 FILEDESC_UNLOCK(fdp);
382 error = EBADF;
383 goto done2;
384 }
385 pop = &fdp->fd_ofileflags[fd];
386
387 switch (cmd) {
388 case F_DUPFD:
389 /* mtx_assert(&Giant, MA_NOTOWNED); */
390 FILEDESC_UNLOCK(fdp);
391 newmin = arg;
392 PROC_LOCK(p);
393 if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
394 newmin >= maxfilesperproc) {
395 PROC_UNLOCK(p);
396 error = EINVAL;
397 break;
398 }
399 PROC_UNLOCK(p);
400 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
401 break;
402
403 case F_GETFD:
404 /* mtx_assert(&Giant, MA_NOTOWNED); */
405 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
406 FILEDESC_UNLOCK(fdp);
407 break;
408
409 case F_SETFD:
410 /* mtx_assert(&Giant, MA_NOTOWNED); */
411 *pop = (*pop &~ UF_EXCLOSE) |
412 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
413 FILEDESC_UNLOCK(fdp);
414 break;
415
416 case F_GETFL:
417 /* mtx_assert(&Giant, MA_NOTOWNED); */
418 FILE_LOCK(fp);
419 td->td_retval[0] = OFLAGS(fp->f_flag);
420 FILE_UNLOCK(fp);
421 FILEDESC_UNLOCK(fdp);
422 break;
423
424 case F_SETFL:
425 mtx_assert(&Giant, MA_OWNED);
426 FILE_LOCK(fp);
427 fhold_locked(fp);
428 fp->f_flag &= ~FCNTLFLAGS;
429 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
430 FILE_UNLOCK(fp);
431 FILEDESC_UNLOCK(fdp);
432 tmp = fp->f_flag & FNONBLOCK;
433 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
434 if (error) {
435 fdrop(fp, td);
436 break;
437 }
438 tmp = fp->f_flag & FASYNC;
439 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
440 if (error == 0) {
441 fdrop(fp, td);
442 break;
443 }
444 FILE_LOCK(fp);
445 fp->f_flag &= ~FNONBLOCK;
446 FILE_UNLOCK(fp);
447 tmp = 0;
448 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
449 fdrop(fp, td);
450 break;
451
452 case F_GETOWN:
453 mtx_assert(&Giant, MA_OWNED);
454 fhold(fp);
455 FILEDESC_UNLOCK(fdp);
456 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
457 if (error == 0)
458 td->td_retval[0] = tmp;
459 fdrop(fp, td);
460 break;
461
462 case F_SETOWN:
463 mtx_assert(&Giant, MA_OWNED);
464 fhold(fp);
465 FILEDESC_UNLOCK(fdp);
466 tmp = arg;
467 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
468 fdrop(fp, td);
469 break;
470
471 case F_SETLKW:
472 mtx_assert(&Giant, MA_OWNED);
473 flg |= F_WAIT;
474 /* FALLTHROUGH F_SETLK */
475
476 case F_SETLK:
477 mtx_assert(&Giant, MA_OWNED);
478 if (fp->f_type != DTYPE_VNODE) {
479 FILEDESC_UNLOCK(fdp);
480 error = EBADF;
481 break;
482 }
483
484 flp = (struct flock *)arg;
485 if (flp->l_whence == SEEK_CUR) {
486 if (fp->f_offset < 0 ||
487 (flp->l_start > 0 &&
488 fp->f_offset > OFF_MAX - flp->l_start)) {
489 FILEDESC_UNLOCK(fdp);
490 error = EOVERFLOW;
491 break;
492 }
493 flp->l_start += fp->f_offset;
494 }
495
496 /*
497 * VOP_ADVLOCK() may block.
498 */
499 fhold(fp);
500 FILEDESC_UNLOCK(fdp);
501 vp = fp->f_vnode;
502
503 switch (flp->l_type) {
504 case F_RDLCK:
505 if ((fp->f_flag & FREAD) == 0) {
506 error = EBADF;
507 break;
508 }
509 PROC_LOCK(p->p_leader);
510 p->p_leader->p_flag |= P_ADVLOCK;
511 PROC_UNLOCK(p->p_leader);
512 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
513 flp, flg);
514 break;
515 case F_WRLCK:
516 if ((fp->f_flag & FWRITE) == 0) {
517 error = EBADF;
518 break;
519 }
520 PROC_LOCK(p->p_leader);
521 p->p_leader->p_flag |= P_ADVLOCK;
522 PROC_UNLOCK(p->p_leader);
523 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
524 flp, flg);
525 break;
526 case F_UNLCK:
527 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
528 flp, F_POSIX);
529 break;
530 default:
531 error = EINVAL;
532 break;
533 }
534 /* Check for race with close */
535 FILEDESC_LOCK_FAST(fdp);
536 if ((unsigned) fd >= fdp->fd_nfiles ||
537 fp != fdp->fd_ofiles[fd]) {
538 FILEDESC_UNLOCK_FAST(fdp);
539 flp->l_whence = SEEK_SET;
540 flp->l_start = 0;
541 flp->l_len = 0;
542 flp->l_type = F_UNLCK;
543 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
544 F_UNLCK, flp, F_POSIX);
545 } else
546 FILEDESC_UNLOCK_FAST(fdp);
547 fdrop(fp, td);
548 break;
549
550 case F_GETLK:
551 mtx_assert(&Giant, MA_OWNED);
552 if (fp->f_type != DTYPE_VNODE) {
553 FILEDESC_UNLOCK(fdp);
554 error = EBADF;
555 break;
556 }
557 flp = (struct flock *)arg;
558 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
559 flp->l_type != F_UNLCK) {
560 FILEDESC_UNLOCK(fdp);
561 error = EINVAL;
562 break;
563 }
564 if (flp->l_whence == SEEK_CUR) {
565 if ((flp->l_start > 0 &&
566 fp->f_offset > OFF_MAX - flp->l_start) ||
567 (flp->l_start < 0 &&
568 fp->f_offset < OFF_MIN - flp->l_start)) {
569 FILEDESC_UNLOCK(fdp);
570 error = EOVERFLOW;
571 break;
572 }
573 flp->l_start += fp->f_offset;
574 }
575 /*
576 * VOP_ADVLOCK() may block.
577 */
578 fhold(fp);
579 FILEDESC_UNLOCK(fdp);
580 vp = fp->f_vnode;
581 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
582 F_POSIX);
583 fdrop(fp, td);
584 break;
585 default:
586 FILEDESC_UNLOCK(fdp);
587 error = EINVAL;
588 break;
589 }
590 done2:
591 if (giant_locked)
592 mtx_unlock(&Giant);
593 return (error);
594 }
595
596 /*
597 * Common code for dup, dup2, and fcntl(F_DUPFD).
598 */
599 static int
600 do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval)
601 {
602 struct filedesc *fdp;
603 struct proc *p;
604 struct file *fp;
605 struct file *delfp;
606 int error, holdleaders, maxfd;
607
608 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
609 ("invalid dup type %d", type));
610
611 p = td->td_proc;
612 fdp = p->p_fd;
613
614 /*
615 * Verify we have a valid descriptor to dup from and possibly to
616 * dup to.
617 */
618 if (old < 0 || new < 0)
619 return (EBADF);
620 PROC_LOCK(p);
621 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
622 PROC_UNLOCK(p);
623 if (new >= maxfd)
624 return (EMFILE);
625
626 FILEDESC_LOCK(fdp);
627 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
628 FILEDESC_UNLOCK(fdp);
629 return (EBADF);
630 }
631 if (type == DUP_FIXED && old == new) {
632 *retval = new;
633 FILEDESC_UNLOCK(fdp);
634 return (0);
635 }
636 fp = fdp->fd_ofiles[old];
637 fhold(fp);
638
639 /*
640 * If the caller specified a file descriptor, make sure the file
641 * table is large enough to hold it, and grab it. Otherwise, just
642 * allocate a new descriptor the usual way. Since the filedesc
643 * lock may be temporarily dropped in the process, we have to look
644 * out for a race.
645 */
646 if (type == DUP_FIXED) {
647 if (new >= fdp->fd_nfiles)
648 fdgrowtable(fdp, new + 1);
649 if (fdp->fd_ofiles[new] == NULL)
650 fdused(fdp, new);
651 } else {
652 if ((error = fdalloc(td, new, &new)) != 0) {
653 FILEDESC_UNLOCK(fdp);
654 fdrop(fp, td);
655 return (error);
656 }
657 }
658
659 /*
660 * If the old file changed out from under us then treat it as a
661 * bad file descriptor. Userland should do its own locking to
662 * avoid this case.
663 */
664 if (fdp->fd_ofiles[old] != fp) {
665 /* we've allocated a descriptor which we won't use */
666 if (fdp->fd_ofiles[new] == NULL)
667 fdunused(fdp, new);
668 FILEDESC_UNLOCK(fdp);
669 fdrop(fp, td);
670 return (EBADF);
671 }
672 KASSERT(old != new,
673 ("new fd is same as old"));
674
675 /*
676 * Save info on the descriptor being overwritten. We cannot close
677 * it without introducing an ownership race for the slot, since we
678 * need to drop the filedesc lock to call closef().
679 *
680 * XXX this duplicates parts of close().
681 */
682 delfp = fdp->fd_ofiles[new];
683 holdleaders = 0;
684 if (delfp != NULL) {
685 if (td->td_proc->p_fdtol != NULL) {
686 /*
687 * Ask fdfree() to sleep to ensure that all relevant
688 * process leaders can be traversed in closef().
689 */
690 fdp->fd_holdleaderscount++;
691 holdleaders = 1;
692 }
693 }
694
695 /*
696 * Duplicate the source descriptor
697 */
698 fdp->fd_ofiles[new] = fp;
699 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
700 if (new > fdp->fd_lastfile)
701 fdp->fd_lastfile = new;
702 *retval = new;
703
704 /*
705 * If we dup'd over a valid file, we now own the reference to it
706 * and must dispose of it using closef() semantics (as if a
707 * close() were performed on it).
708 *
709 * XXX this duplicates parts of close().
710 */
711 if (delfp != NULL) {
712 knote_fdclose(td, new);
713 FILEDESC_UNLOCK(fdp);
714 (void) closef(delfp, td);
715 if (holdleaders) {
716 FILEDESC_LOCK_FAST(fdp);
717 fdp->fd_holdleaderscount--;
718 if (fdp->fd_holdleaderscount == 0 &&
719 fdp->fd_holdleaderswakeup != 0) {
720 fdp->fd_holdleaderswakeup = 0;
721 wakeup(&fdp->fd_holdleaderscount);
722 }
723 FILEDESC_UNLOCK_FAST(fdp);
724 }
725 } else {
726 FILEDESC_UNLOCK(fdp);
727 }
728 return (0);
729 }
730
731 /*
732 * If sigio is on the list associated with a process or process group,
733 * disable signalling from the device, remove sigio from the list and
734 * free sigio.
735 */
736 void
737 funsetown(struct sigio **sigiop)
738 {
739 struct sigio *sigio;
740
741 SIGIO_LOCK();
742 sigio = *sigiop;
743 if (sigio == NULL) {
744 SIGIO_UNLOCK();
745 return;
746 }
747 *(sigio->sio_myref) = NULL;
748 if ((sigio)->sio_pgid < 0) {
749 struct pgrp *pg = (sigio)->sio_pgrp;
750 PGRP_LOCK(pg);
751 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
752 sigio, sio_pgsigio);
753 PGRP_UNLOCK(pg);
754 } else {
755 struct proc *p = (sigio)->sio_proc;
756 PROC_LOCK(p);
757 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
758 sigio, sio_pgsigio);
759 PROC_UNLOCK(p);
760 }
761 SIGIO_UNLOCK();
762 crfree(sigio->sio_ucred);
763 FREE(sigio, M_SIGIO);
764 }
765
766 /*
767 * Free a list of sigio structures.
768 * We only need to lock the SIGIO_LOCK because we have made ourselves
769 * inaccessable to callers of fsetown and therefore do not need to lock
770 * the proc or pgrp struct for the list manipulation.
771 */
772 void
773 funsetownlst(struct sigiolst *sigiolst)
774 {
775 struct proc *p;
776 struct pgrp *pg;
777 struct sigio *sigio;
778
779 sigio = SLIST_FIRST(sigiolst);
780 if (sigio == NULL)
781 return;
782 p = NULL;
783 pg = NULL;
784
785 /*
786 * Every entry of the list should belong
787 * to a single proc or pgrp.
788 */
789 if (sigio->sio_pgid < 0) {
790 pg = sigio->sio_pgrp;
791 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
792 } else /* if (sigio->sio_pgid > 0) */ {
793 p = sigio->sio_proc;
794 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
795 }
796
797 SIGIO_LOCK();
798 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
799 *(sigio->sio_myref) = NULL;
800 if (pg != NULL) {
801 KASSERT(sigio->sio_pgid < 0,
802 ("Proc sigio in pgrp sigio list"));
803 KASSERT(sigio->sio_pgrp == pg,
804 ("Bogus pgrp in sigio list"));
805 PGRP_LOCK(pg);
806 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
807 sio_pgsigio);
808 PGRP_UNLOCK(pg);
809 } else /* if (p != NULL) */ {
810 KASSERT(sigio->sio_pgid > 0,
811 ("Pgrp sigio in proc sigio list"));
812 KASSERT(sigio->sio_proc == p,
813 ("Bogus proc in sigio list"));
814 PROC_LOCK(p);
815 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
816 sio_pgsigio);
817 PROC_UNLOCK(p);
818 }
819 SIGIO_UNLOCK();
820 crfree(sigio->sio_ucred);
821 FREE(sigio, M_SIGIO);
822 SIGIO_LOCK();
823 }
824 SIGIO_UNLOCK();
825 }
826
827 /*
828 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
829 *
830 * After permission checking, add a sigio structure to the sigio list for
831 * the process or process group.
832 */
833 int
834 fsetown(pid_t pgid, struct sigio **sigiop)
835 {
836 struct proc *proc;
837 struct pgrp *pgrp;
838 struct sigio *sigio;
839 int ret;
840
841 if (pgid == 0) {
842 funsetown(sigiop);
843 return (0);
844 }
845
846 ret = 0;
847
848 /* Allocate and fill in the new sigio out of locks. */
849 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
850 sigio->sio_pgid = pgid;
851 sigio->sio_ucred = crhold(curthread->td_ucred);
852 sigio->sio_myref = sigiop;
853
854 sx_slock(&proctree_lock);
855 if (pgid > 0) {
856 proc = pfind(pgid);
857 if (proc == NULL) {
858 ret = ESRCH;
859 goto fail;
860 }
861
862 /*
863 * Policy - Don't allow a process to FSETOWN a process
864 * in another session.
865 *
866 * Remove this test to allow maximum flexibility or
867 * restrict FSETOWN to the current process or process
868 * group for maximum safety.
869 */
870 PROC_UNLOCK(proc);
871 if (proc->p_session != curthread->td_proc->p_session) {
872 ret = EPERM;
873 goto fail;
874 }
875
876 pgrp = NULL;
877 } else /* if (pgid < 0) */ {
878 pgrp = pgfind(-pgid);
879 if (pgrp == NULL) {
880 ret = ESRCH;
881 goto fail;
882 }
883 PGRP_UNLOCK(pgrp);
884
885 /*
886 * Policy - Don't allow a process to FSETOWN a process
887 * in another session.
888 *
889 * Remove this test to allow maximum flexibility or
890 * restrict FSETOWN to the current process or process
891 * group for maximum safety.
892 */
893 if (pgrp->pg_session != curthread->td_proc->p_session) {
894 ret = EPERM;
895 goto fail;
896 }
897
898 proc = NULL;
899 }
900 funsetown(sigiop);
901 if (pgid > 0) {
902 PROC_LOCK(proc);
903 /*
904 * Since funsetownlst() is called without the proctree
905 * locked, we need to check for P_WEXIT.
906 * XXX: is ESRCH correct?
907 */
908 if ((proc->p_flag & P_WEXIT) != 0) {
909 PROC_UNLOCK(proc);
910 ret = ESRCH;
911 goto fail;
912 }
913 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
914 sigio->sio_proc = proc;
915 PROC_UNLOCK(proc);
916 } else {
917 PGRP_LOCK(pgrp);
918 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
919 sigio->sio_pgrp = pgrp;
920 PGRP_UNLOCK(pgrp);
921 }
922 sx_sunlock(&proctree_lock);
923 SIGIO_LOCK();
924 *sigiop = sigio;
925 SIGIO_UNLOCK();
926 return (0);
927
928 fail:
929 sx_sunlock(&proctree_lock);
930 crfree(sigio->sio_ucred);
931 FREE(sigio, M_SIGIO);
932 return (ret);
933 }
934
935 /*
936 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
937 */
938 pid_t
939 fgetown(sigiop)
940 struct sigio **sigiop;
941 {
942 pid_t pgid;
943
944 SIGIO_LOCK();
945 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
946 SIGIO_UNLOCK();
947 return (pgid);
948 }
949
950 /*
951 * Close a file descriptor.
952 */
953 #ifndef _SYS_SYSPROTO_H_
954 struct close_args {
955 int fd;
956 };
957 #endif
958 /*
959 * MPSAFE
960 */
961 /* ARGSUSED */
962 int
963 close(td, uap)
964 struct thread *td;
965 struct close_args *uap;
966 {
967 struct filedesc *fdp;
968 struct file *fp;
969 int fd, error;
970 int holdleaders;
971
972 fd = uap->fd;
973 error = 0;
974 holdleaders = 0;
975 fdp = td->td_proc->p_fd;
976 FILEDESC_LOCK(fdp);
977 if ((unsigned)fd >= fdp->fd_nfiles ||
978 (fp = fdp->fd_ofiles[fd]) == NULL) {
979 FILEDESC_UNLOCK(fdp);
980 return (EBADF);
981 }
982 fdp->fd_ofiles[fd] = NULL;
983 fdp->fd_ofileflags[fd] = 0;
984 fdunused(fdp, fd);
985 if (td->td_proc->p_fdtol != NULL) {
986 /*
987 * Ask fdfree() to sleep to ensure that all relevant
988 * process leaders can be traversed in closef().
989 */
990 fdp->fd_holdleaderscount++;
991 holdleaders = 1;
992 }
993
994 /*
995 * we now hold the fp reference that used to be owned by the descriptor
996 * array.
997 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a
998 * race of the fd getting opened, a knote added, and deleteing a knote
999 * for the new fd.
1000 */
1001 knote_fdclose(td, fd);
1002 FILEDESC_UNLOCK(fdp);
1003
1004 error = closef(fp, td);
1005 if (holdleaders) {
1006 FILEDESC_LOCK_FAST(fdp);
1007 fdp->fd_holdleaderscount--;
1008 if (fdp->fd_holdleaderscount == 0 &&
1009 fdp->fd_holdleaderswakeup != 0) {
1010 fdp->fd_holdleaderswakeup = 0;
1011 wakeup(&fdp->fd_holdleaderscount);
1012 }
1013 FILEDESC_UNLOCK_FAST(fdp);
1014 }
1015 return (error);
1016 }
1017
1018 #if defined(COMPAT_43)
1019 /*
1020 * Return status information about a file descriptor.
1021 */
1022 #ifndef _SYS_SYSPROTO_H_
1023 struct ofstat_args {
1024 int fd;
1025 struct ostat *sb;
1026 };
1027 #endif
1028 /*
1029 * MPSAFE
1030 */
1031 /* ARGSUSED */
1032 int
1033 ofstat(struct thread *td, struct ofstat_args *uap)
1034 {
1035 struct file *fp;
1036 struct stat ub;
1037 struct ostat oub;
1038 int error;
1039
1040 if ((error = fget(td, uap->fd, &fp)) != 0)
1041 goto done2;
1042 error = fo_stat(fp, &ub, td->td_ucred, td);
1043 if (error == 0) {
1044 cvtstat(&ub, &oub);
1045 error = copyout(&oub, uap->sb, sizeof(oub));
1046 }
1047 fdrop(fp, td);
1048 done2:
1049 return (error);
1050 }
1051 #endif /* COMPAT_43 */
1052
1053 /*
1054 * Return status information about a file descriptor.
1055 */
1056 #ifndef _SYS_SYSPROTO_H_
1057 struct fstat_args {
1058 int fd;
1059 struct stat *sb;
1060 };
1061 #endif
1062 /*
1063 * MPSAFE
1064 */
1065 /* ARGSUSED */
1066 int
1067 fstat(struct thread *td, struct fstat_args *uap)
1068 {
1069 struct file *fp;
1070 struct stat ub;
1071 int error;
1072
1073 if ((error = fget(td, uap->fd, &fp)) != 0)
1074 goto done2;
1075 error = fo_stat(fp, &ub, td->td_ucred, td);
1076 if (error == 0)
1077 error = copyout(&ub, uap->sb, sizeof(ub));
1078 fdrop(fp, td);
1079 done2:
1080 return (error);
1081 }
1082
1083 /*
1084 * Return status information about a file descriptor.
1085 */
1086 #ifndef _SYS_SYSPROTO_H_
1087 struct nfstat_args {
1088 int fd;
1089 struct nstat *sb;
1090 };
1091 #endif
1092 /*
1093 * MPSAFE
1094 */
1095 /* ARGSUSED */
1096 int
1097 nfstat(struct thread *td, struct nfstat_args *uap)
1098 {
1099 struct file *fp;
1100 struct stat ub;
1101 struct nstat nub;
1102 int error;
1103
1104 if ((error = fget(td, uap->fd, &fp)) != 0)
1105 goto done2;
1106 error = fo_stat(fp, &ub, td->td_ucred, td);
1107 if (error == 0) {
1108 cvtnstat(&ub, &nub);
1109 error = copyout(&nub, uap->sb, sizeof(nub));
1110 }
1111 fdrop(fp, td);
1112 done2:
1113 return (error);
1114 }
1115
1116 /*
1117 * Return pathconf information about a file descriptor.
1118 */
1119 #ifndef _SYS_SYSPROTO_H_
1120 struct fpathconf_args {
1121 int fd;
1122 int name;
1123 };
1124 #endif
1125 /*
1126 * MPSAFE
1127 */
1128 /* ARGSUSED */
1129 int
1130 fpathconf(struct thread *td, struct fpathconf_args *uap)
1131 {
1132 struct file *fp;
1133 struct vnode *vp;
1134 int error;
1135
1136 if ((error = fget(td, uap->fd, &fp)) != 0)
1137 return (error);
1138
1139 /* If asynchronous I/O is available, it works for all descriptors. */
1140 if (uap->name == _PC_ASYNC_IO) {
1141 td->td_retval[0] = async_io_version;
1142 goto out;
1143 }
1144 vp = fp->f_vnode;
1145 if (vp != NULL) {
1146 mtx_lock(&Giant);
1147 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1148 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1149 VOP_UNLOCK(vp, 0, td);
1150 mtx_unlock(&Giant);
1151 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1152 if (uap->name != _PC_PIPE_BUF) {
1153 error = EINVAL;
1154 } else {
1155 td->td_retval[0] = PIPE_BUF;
1156 error = 0;
1157 }
1158 } else {
1159 error = EOPNOTSUPP;
1160 }
1161 out:
1162 fdrop(fp, td);
1163 return (error);
1164 }
1165
1166 /*
1167 * Grow the file table to accomodate (at least) nfd descriptors. This may
1168 * block and drop the filedesc lock, but it will reacquire it before
1169 * returing.
1170 */
1171 static void
1172 fdgrowtable(struct filedesc *fdp, int nfd)
1173 {
1174 struct file **ntable;
1175 char *nfileflags;
1176 int nnfiles, onfiles;
1177 NDSLOTTYPE *nmap;
1178
1179 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1180
1181 KASSERT(fdp->fd_nfiles > 0,
1182 ("zero-length file table"));
1183
1184 /* compute the size of the new table */
1185 onfiles = fdp->fd_nfiles;
1186 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1187 if (nnfiles <= onfiles)
1188 /* the table is already large enough */
1189 return;
1190
1191 /* allocate a new table and (if required) new bitmaps */
1192 FILEDESC_UNLOCK(fdp);
1193 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1194 M_FILEDESC, M_ZERO | M_WAITOK);
1195 nfileflags = (char *)&ntable[nnfiles];
1196 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1197 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1198 M_FILEDESC, M_ZERO | M_WAITOK);
1199 else
1200 nmap = NULL;
1201 FILEDESC_LOCK(fdp);
1202
1203 /*
1204 * We now have new tables ready to go. Since we dropped the
1205 * filedesc lock to call malloc(), watch out for a race.
1206 */
1207 onfiles = fdp->fd_nfiles;
1208 if (onfiles >= nnfiles) {
1209 /* we lost the race, but that's OK */
1210 free(ntable, M_FILEDESC);
1211 if (nmap != NULL)
1212 free(nmap, M_FILEDESC);
1213 return;
1214 }
1215 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1216 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1217 if (onfiles > NDFILE)
1218 free(fdp->fd_ofiles, M_FILEDESC);
1219 fdp->fd_ofiles = ntable;
1220 fdp->fd_ofileflags = nfileflags;
1221 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1222 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1223 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1224 free(fdp->fd_map, M_FILEDESC);
1225 fdp->fd_map = nmap;
1226 }
1227 fdp->fd_nfiles = nnfiles;
1228 }
1229
1230 /*
1231 * Allocate a file descriptor for the process.
1232 */
1233 int
1234 fdalloc(struct thread *td, int minfd, int *result)
1235 {
1236 struct proc *p = td->td_proc;
1237 struct filedesc *fdp = p->p_fd;
1238 int fd = -1, maxfd;
1239
1240 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1241
1242 PROC_LOCK(p);
1243 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1244 PROC_UNLOCK(p);
1245
1246 /*
1247 * Search the bitmap for a free descriptor. If none is found, try
1248 * to grow the file table. Keep at it until we either get a file
1249 * descriptor or run into process or system limits; fdgrowtable()
1250 * may drop the filedesc lock, so we're in a race.
1251 */
1252 for (;;) {
1253 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1254 if (fd >= maxfd)
1255 return (EMFILE);
1256 if (fd < fdp->fd_nfiles)
1257 break;
1258 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1259 }
1260
1261 /*
1262 * Perform some sanity checks, then mark the file descriptor as
1263 * used and return it to the caller.
1264 */
1265 KASSERT(!fdisused(fdp, fd),
1266 ("fd_first_free() returned non-free descriptor"));
1267 KASSERT(fdp->fd_ofiles[fd] == NULL,
1268 ("free descriptor isn't"));
1269 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1270 fdused(fdp, fd);
1271 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
1272 *result = fd;
1273 return (0);
1274 }
1275
1276 /*
1277 * Check to see whether n user file descriptors
1278 * are available to the process p.
1279 */
1280 int
1281 fdavail(struct thread *td, int n)
1282 {
1283 struct proc *p = td->td_proc;
1284 struct filedesc *fdp = td->td_proc->p_fd;
1285 struct file **fpp;
1286 int i, lim, last;
1287
1288 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1289
1290 PROC_LOCK(p);
1291 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1292 PROC_UNLOCK(p);
1293 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1294 return (1);
1295 last = min(fdp->fd_nfiles, lim);
1296 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1297 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1298 if (*fpp == NULL && --n <= 0)
1299 return (1);
1300 }
1301 return (0);
1302 }
1303
1304 /*
1305 * Create a new open file structure and allocate
1306 * a file decriptor for the process that refers to it.
1307 * We add one reference to the file for the descriptor table
1308 * and one reference for resultfp. This is to prevent us being
1309 * prempted and the entry in the descriptor table closed after
1310 * we release the FILEDESC lock.
1311 */
1312 int
1313 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1314 {
1315 struct proc *p = td->td_proc;
1316 struct file *fp, *fq;
1317 int error, i;
1318 int maxuserfiles = maxfiles - (maxfiles / 20);
1319 static struct timeval lastfail;
1320 static int curfail;
1321
1322 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1323 sx_xlock(&filelist_lock);
1324 if ((nfiles >= maxuserfiles && (td->td_ucred->cr_ruid != 0 ||
1325 jailed(td->td_ucred))) || nfiles >= maxfiles) {
1326 if (ppsratecheck(&lastfail, &curfail, 1)) {
1327 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1328 td->td_ucred->cr_ruid);
1329 }
1330 sx_xunlock(&filelist_lock);
1331 uma_zfree(file_zone, fp);
1332 return (ENFILE);
1333 }
1334 nfiles++;
1335
1336 /*
1337 * If the process has file descriptor zero open, add the new file
1338 * descriptor to the list of open files at that point, otherwise
1339 * put it at the front of the list of open files.
1340 */
1341 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1342 fp->f_count = 1;
1343 if (resultfp)
1344 fp->f_count++;
1345 fp->f_cred = crhold(td->td_ucred);
1346 fp->f_ops = &badfileops;
1347 fp->f_data = NULL;
1348 fp->f_vnode = NULL;
1349 FILEDESC_LOCK(p->p_fd);
1350 if ((fq = p->p_fd->fd_ofiles[0])) {
1351 LIST_INSERT_AFTER(fq, fp, f_list);
1352 } else {
1353 LIST_INSERT_HEAD(&filehead, fp, f_list);
1354 }
1355 sx_xunlock(&filelist_lock);
1356 if ((error = fdalloc(td, 0, &i))) {
1357 FILEDESC_UNLOCK(p->p_fd);
1358 fdrop(fp, td);
1359 if (resultfp)
1360 fdrop(fp, td);
1361 return (error);
1362 }
1363 p->p_fd->fd_ofiles[i] = fp;
1364 FILEDESC_UNLOCK(p->p_fd);
1365 if (resultfp)
1366 *resultfp = fp;
1367 if (resultfd)
1368 *resultfd = i;
1369 return (0);
1370 }
1371
1372 /*
1373 * Build a new filedesc structure from another.
1374 * Copy the current, root, and jail root vnode references.
1375 */
1376 struct filedesc *
1377 fdinit(struct filedesc *fdp)
1378 {
1379 struct filedesc0 *newfdp;
1380
1381 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1382 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1383 if (fdp != NULL) {
1384 FILEDESC_LOCK(fdp);
1385 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1386 if (newfdp->fd_fd.fd_cdir)
1387 VREF(newfdp->fd_fd.fd_cdir);
1388 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1389 if (newfdp->fd_fd.fd_rdir)
1390 VREF(newfdp->fd_fd.fd_rdir);
1391 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1392 if (newfdp->fd_fd.fd_jdir)
1393 VREF(newfdp->fd_fd.fd_jdir);
1394 FILEDESC_UNLOCK(fdp);
1395 }
1396
1397 /* Create the file descriptor table. */
1398 newfdp->fd_fd.fd_refcnt = 1;
1399 newfdp->fd_fd.fd_holdcnt = 1;
1400 newfdp->fd_fd.fd_cmask = CMASK;
1401 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1402 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1403 newfdp->fd_fd.fd_nfiles = NDFILE;
1404 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1405 return (&newfdp->fd_fd);
1406 }
1407
1408 static struct filedesc *
1409 fdhold(struct proc *p)
1410 {
1411 struct filedesc *fdp;
1412
1413 mtx_lock(&fdesc_mtx);
1414 fdp = p->p_fd;
1415 if (fdp != NULL)
1416 fdp->fd_holdcnt++;
1417 mtx_unlock(&fdesc_mtx);
1418 return (fdp);
1419 }
1420
1421 static void
1422 fddrop(struct filedesc *fdp)
1423 {
1424 int i;
1425
1426 mtx_lock(&fdesc_mtx);
1427 i = --fdp->fd_holdcnt;
1428 mtx_unlock(&fdesc_mtx);
1429 if (i > 0)
1430 return;
1431
1432 mtx_destroy(&fdp->fd_mtx);
1433 FREE(fdp, M_FILEDESC);
1434 }
1435
1436 /*
1437 * Share a filedesc structure.
1438 */
1439 struct filedesc *
1440 fdshare(struct filedesc *fdp)
1441 {
1442 FILEDESC_LOCK_FAST(fdp);
1443 fdp->fd_refcnt++;
1444 FILEDESC_UNLOCK_FAST(fdp);
1445 return (fdp);
1446 }
1447
1448 /*
1449 * Unshare a filedesc structure, if necessary by making a copy
1450 */
1451 void
1452 fdunshare(struct proc *p, struct thread *td)
1453 {
1454
1455 FILEDESC_LOCK_FAST(p->p_fd);
1456 if (p->p_fd->fd_refcnt > 1) {
1457 struct filedesc *tmp;
1458
1459 FILEDESC_UNLOCK_FAST(p->p_fd);
1460 tmp = fdcopy(p->p_fd);
1461 fdfree(td);
1462 p->p_fd = tmp;
1463 } else
1464 FILEDESC_UNLOCK_FAST(p->p_fd);
1465 }
1466
1467 /*
1468 * Copy a filedesc structure.
1469 * A NULL pointer in returns a NULL reference, this is to ease callers,
1470 * not catch errors.
1471 */
1472 struct filedesc *
1473 fdcopy(struct filedesc *fdp)
1474 {
1475 struct filedesc *newfdp;
1476 int i;
1477
1478 /* Certain daemons might not have file descriptors. */
1479 if (fdp == NULL)
1480 return (NULL);
1481
1482 newfdp = fdinit(fdp);
1483 FILEDESC_LOCK_FAST(fdp);
1484 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1485 FILEDESC_UNLOCK_FAST(fdp);
1486 FILEDESC_LOCK(newfdp);
1487 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1488 FILEDESC_UNLOCK(newfdp);
1489 FILEDESC_LOCK_FAST(fdp);
1490 }
1491 /* copy everything except kqueue descriptors */
1492 newfdp->fd_freefile = -1;
1493 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1494 if (fdisused(fdp, i) &&
1495 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1496 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1497 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1498 fhold(newfdp->fd_ofiles[i]);
1499 newfdp->fd_lastfile = i;
1500 } else {
1501 if (newfdp->fd_freefile == -1)
1502 newfdp->fd_freefile = i;
1503 }
1504 }
1505 FILEDESC_UNLOCK_FAST(fdp);
1506 FILEDESC_LOCK(newfdp);
1507 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1508 if (newfdp->fd_ofiles[i] != NULL)
1509 fdused(newfdp, i);
1510 FILEDESC_UNLOCK(newfdp);
1511 FILEDESC_LOCK_FAST(fdp);
1512 if (newfdp->fd_freefile == -1)
1513 newfdp->fd_freefile = i;
1514 newfdp->fd_cmask = fdp->fd_cmask;
1515 FILEDESC_UNLOCK_FAST(fdp);
1516 return (newfdp);
1517 }
1518
1519 /*
1520 * Release a filedesc structure.
1521 */
1522 void
1523 fdfree(struct thread *td)
1524 {
1525 struct filedesc *fdp;
1526 struct file **fpp;
1527 int i;
1528 struct filedesc_to_leader *fdtol;
1529 struct file *fp;
1530 struct vnode *vp;
1531 struct flock lf;
1532
1533 GIANT_REQUIRED; /* VFS */
1534
1535 /* Certain daemons might not have file descriptors. */
1536 fdp = td->td_proc->p_fd;
1537 if (fdp == NULL)
1538 return;
1539
1540 /* Check for special need to clear POSIX style locks */
1541 fdtol = td->td_proc->p_fdtol;
1542 if (fdtol != NULL) {
1543 FILEDESC_LOCK(fdp);
1544 KASSERT(fdtol->fdl_refcount > 0,
1545 ("filedesc_to_refcount botch: fdl_refcount=%d",
1546 fdtol->fdl_refcount));
1547 if (fdtol->fdl_refcount == 1 &&
1548 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1549 i = 0;
1550 fpp = fdp->fd_ofiles;
1551 for (i = 0, fpp = fdp->fd_ofiles;
1552 i <= fdp->fd_lastfile;
1553 i++, fpp++) {
1554 if (*fpp == NULL ||
1555 (*fpp)->f_type != DTYPE_VNODE)
1556 continue;
1557 fp = *fpp;
1558 fhold(fp);
1559 FILEDESC_UNLOCK(fdp);
1560 lf.l_whence = SEEK_SET;
1561 lf.l_start = 0;
1562 lf.l_len = 0;
1563 lf.l_type = F_UNLCK;
1564 vp = fp->f_vnode;
1565 (void) VOP_ADVLOCK(vp,
1566 (caddr_t)td->td_proc->
1567 p_leader,
1568 F_UNLCK,
1569 &lf,
1570 F_POSIX);
1571 FILEDESC_LOCK(fdp);
1572 fdrop(fp, td);
1573 fpp = fdp->fd_ofiles + i;
1574 }
1575 }
1576 retry:
1577 if (fdtol->fdl_refcount == 1) {
1578 if (fdp->fd_holdleaderscount > 0 &&
1579 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1580 /*
1581 * close() or do_dup() has cleared a reference
1582 * in a shared file descriptor table.
1583 */
1584 fdp->fd_holdleaderswakeup = 1;
1585 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1586 PLOCK, "fdlhold", 0);
1587 goto retry;
1588 }
1589 if (fdtol->fdl_holdcount > 0) {
1590 /*
1591 * Ensure that fdtol->fdl_leader
1592 * remains valid in closef().
1593 */
1594 fdtol->fdl_wakeup = 1;
1595 msleep(fdtol, &fdp->fd_mtx,
1596 PLOCK, "fdlhold", 0);
1597 goto retry;
1598 }
1599 }
1600 fdtol->fdl_refcount--;
1601 if (fdtol->fdl_refcount == 0 &&
1602 fdtol->fdl_holdcount == 0) {
1603 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1604 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1605 } else
1606 fdtol = NULL;
1607 td->td_proc->p_fdtol = NULL;
1608 FILEDESC_UNLOCK(fdp);
1609 if (fdtol != NULL)
1610 FREE(fdtol, M_FILEDESC_TO_LEADER);
1611 }
1612 FILEDESC_LOCK_FAST(fdp);
1613 i = --fdp->fd_refcnt;
1614 FILEDESC_UNLOCK_FAST(fdp);
1615 if (i > 0)
1616 return;
1617 /*
1618 * We are the last reference to the structure, so we can
1619 * safely assume it will not change out from under us.
1620 */
1621 fpp = fdp->fd_ofiles;
1622 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1623 if (*fpp)
1624 (void) closef(*fpp, td);
1625 }
1626 FILEDESC_LOCK(fdp);
1627
1628 /* XXX This should happen earlier. */
1629 mtx_lock(&fdesc_mtx);
1630 td->td_proc->p_fd = NULL;
1631 mtx_unlock(&fdesc_mtx);
1632
1633 if (fdp->fd_nfiles > NDFILE)
1634 FREE(fdp->fd_ofiles, M_FILEDESC);
1635 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1636 FREE(fdp->fd_map, M_FILEDESC);
1637
1638 fdp->fd_nfiles = 0;
1639
1640 if (fdp->fd_cdir)
1641 vrele(fdp->fd_cdir);
1642 fdp->fd_cdir = NULL;
1643 if (fdp->fd_rdir)
1644 vrele(fdp->fd_rdir);
1645 fdp->fd_rdir = NULL;
1646 if (fdp->fd_jdir)
1647 vrele(fdp->fd_jdir);
1648 fdp->fd_jdir = NULL;
1649
1650 FILEDESC_UNLOCK(fdp);
1651
1652 fddrop(fdp);
1653 }
1654
1655 /*
1656 * For setugid programs, we don't want to people to use that setugidness
1657 * to generate error messages which write to a file which otherwise would
1658 * otherwise be off-limits to the process. We check for filesystems where
1659 * the vnode can change out from under us after execve (like [lin]procfs).
1660 *
1661 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1662 * sufficient. We also don't for check setugidness since we know we are.
1663 */
1664 static int
1665 is_unsafe(struct file *fp)
1666 {
1667 if (fp->f_type == DTYPE_VNODE) {
1668 struct vnode *vp = fp->f_vnode;
1669
1670 if ((vp->v_vflag & VV_PROCDEP) != 0)
1671 return (1);
1672 }
1673 return (0);
1674 }
1675
1676 /*
1677 * Make this setguid thing safe, if at all possible.
1678 */
1679 void
1680 setugidsafety(struct thread *td)
1681 {
1682 struct filedesc *fdp;
1683 int i;
1684
1685 /* Certain daemons might not have file descriptors. */
1686 fdp = td->td_proc->p_fd;
1687 if (fdp == NULL)
1688 return;
1689
1690 /*
1691 * Note: fdp->fd_ofiles may be reallocated out from under us while
1692 * we are blocked in a close. Be careful!
1693 */
1694 FILEDESC_LOCK(fdp);
1695 for (i = 0; i <= fdp->fd_lastfile; i++) {
1696 if (i > 2)
1697 break;
1698 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1699 struct file *fp;
1700
1701 knote_fdclose(td, i);
1702 /*
1703 * NULL-out descriptor prior to close to avoid
1704 * a race while close blocks.
1705 */
1706 fp = fdp->fd_ofiles[i];
1707 fdp->fd_ofiles[i] = NULL;
1708 fdp->fd_ofileflags[i] = 0;
1709 fdunused(fdp, i);
1710 FILEDESC_UNLOCK(fdp);
1711 (void) closef(fp, td);
1712 FILEDESC_LOCK(fdp);
1713 }
1714 }
1715 FILEDESC_UNLOCK(fdp);
1716 }
1717
1718 void
1719 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1720 {
1721
1722 FILEDESC_LOCK(fdp);
1723 if (fdp->fd_ofiles[idx] == fp) {
1724 fdp->fd_ofiles[idx] = NULL;
1725 fdunused(fdp, idx);
1726 FILEDESC_UNLOCK(fdp);
1727 fdrop(fp, td);
1728 } else {
1729 FILEDESC_UNLOCK(fdp);
1730 }
1731 }
1732
1733 /*
1734 * Close any files on exec?
1735 */
1736 void
1737 fdcloseexec(struct thread *td)
1738 {
1739 struct filedesc *fdp;
1740 int i;
1741
1742 /* Certain daemons might not have file descriptors. */
1743 fdp = td->td_proc->p_fd;
1744 if (fdp == NULL)
1745 return;
1746
1747 FILEDESC_LOCK(fdp);
1748
1749 /*
1750 * We cannot cache fd_ofiles or fd_ofileflags since operations
1751 * may block and rip them out from under us.
1752 */
1753 for (i = 0; i <= fdp->fd_lastfile; i++) {
1754 if (fdp->fd_ofiles[i] != NULL &&
1755 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1756 struct file *fp;
1757
1758 knote_fdclose(td, i);
1759 /*
1760 * NULL-out descriptor prior to close to avoid
1761 * a race while close blocks.
1762 */
1763 fp = fdp->fd_ofiles[i];
1764 fdp->fd_ofiles[i] = NULL;
1765 fdp->fd_ofileflags[i] = 0;
1766 fdunused(fdp, i);
1767 FILEDESC_UNLOCK(fdp);
1768 (void) closef(fp, td);
1769 FILEDESC_LOCK(fdp);
1770 }
1771 }
1772 FILEDESC_UNLOCK(fdp);
1773 }
1774
1775 /*
1776 * It is unsafe for set[ug]id processes to be started with file
1777 * descriptors 0..2 closed, as these descriptors are given implicit
1778 * significance in the Standard C library. fdcheckstd() will create a
1779 * descriptor referencing /dev/null for each of stdin, stdout, and
1780 * stderr that is not already open.
1781 */
1782 int
1783 fdcheckstd(struct thread *td)
1784 {
1785 struct nameidata nd;
1786 struct filedesc *fdp;
1787 struct file *fp;
1788 register_t retval;
1789 int fd, i, error, flags, devnull;
1790
1791 GIANT_REQUIRED; /* VFS */
1792
1793 fdp = td->td_proc->p_fd;
1794 if (fdp == NULL)
1795 return (0);
1796 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1797 devnull = -1;
1798 error = 0;
1799 for (i = 0; i < 3; i++) {
1800 if (fdp->fd_ofiles[i] != NULL)
1801 continue;
1802 if (devnull < 0) {
1803 error = falloc(td, &fp, &fd);
1804 if (error != 0)
1805 break;
1806 /* Note extra ref on `fp' held for us by falloc(). */
1807 KASSERT(fd == i, ("oof, we didn't get our fd"));
1808 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null",
1809 td);
1810 flags = FREAD | FWRITE;
1811 error = vn_open(&nd, &flags, 0, -1);
1812 if (error != 0) {
1813 /*
1814 * Someone may have closed the entry in the
1815 * file descriptor table, so check it hasn't
1816 * changed before dropping the reference count.
1817 */
1818 FILEDESC_LOCK(fdp);
1819 KASSERT(fdp->fd_ofiles[fd] == fp,
1820 ("table not shared, how did it change?"));
1821 fdp->fd_ofiles[fd] = NULL;
1822 fdunused(fdp, fd);
1823 FILEDESC_UNLOCK(fdp);
1824 fdrop(fp, td);
1825 fdrop(fp, td);
1826 break;
1827 }
1828 NDFREE(&nd, NDF_ONLY_PNBUF);
1829 fp->f_flag = flags;
1830 fp->f_vnode = nd.ni_vp;
1831 if (fp->f_data == NULL)
1832 fp->f_data = nd.ni_vp;
1833 if (fp->f_ops == &badfileops)
1834 fp->f_ops = &vnops;
1835 fp->f_type = DTYPE_VNODE;
1836 VOP_UNLOCK(nd.ni_vp, 0, td);
1837 devnull = fd;
1838 fdrop(fp, td);
1839 } else {
1840 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1841 if (error != 0)
1842 break;
1843 }
1844 }
1845 return (error);
1846 }
1847
1848 /*
1849 * Internal form of close.
1850 * Decrement reference count on file structure.
1851 * Note: td may be NULL when closing a file that was being passed in a
1852 * message.
1853 *
1854 * XXXRW: Giant is not required for the caller, but often will be held; this
1855 * makes it moderately likely the Giant will be recursed in the VFS case.
1856 */
1857 int
1858 closef(struct file *fp, struct thread *td)
1859 {
1860 struct vnode *vp;
1861 struct flock lf;
1862 struct filedesc_to_leader *fdtol;
1863 struct filedesc *fdp;
1864
1865 /*
1866 * POSIX record locking dictates that any close releases ALL
1867 * locks owned by this process. This is handled by setting
1868 * a flag in the unlock to free ONLY locks obeying POSIX
1869 * semantics, and not to free BSD-style file locks.
1870 * If the descriptor was in a message, POSIX-style locks
1871 * aren't passed with the descriptor.
1872 */
1873 if (fp->f_type == DTYPE_VNODE) {
1874 mtx_lock(&Giant);
1875 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1876 lf.l_whence = SEEK_SET;
1877 lf.l_start = 0;
1878 lf.l_len = 0;
1879 lf.l_type = F_UNLCK;
1880 vp = fp->f_vnode;
1881 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1882 F_UNLCK, &lf, F_POSIX);
1883 }
1884 fdtol = td->td_proc->p_fdtol;
1885 if (fdtol != NULL) {
1886 /*
1887 * Handle special case where file descriptor table
1888 * is shared between multiple process leaders.
1889 */
1890 fdp = td->td_proc->p_fd;
1891 FILEDESC_LOCK(fdp);
1892 for (fdtol = fdtol->fdl_next;
1893 fdtol != td->td_proc->p_fdtol;
1894 fdtol = fdtol->fdl_next) {
1895 if ((fdtol->fdl_leader->p_flag &
1896 P_ADVLOCK) == 0)
1897 continue;
1898 fdtol->fdl_holdcount++;
1899 FILEDESC_UNLOCK(fdp);
1900 lf.l_whence = SEEK_SET;
1901 lf.l_start = 0;
1902 lf.l_len = 0;
1903 lf.l_type = F_UNLCK;
1904 vp = fp->f_vnode;
1905 (void) VOP_ADVLOCK(vp,
1906 (caddr_t)fdtol->fdl_leader,
1907 F_UNLCK, &lf, F_POSIX);
1908 FILEDESC_LOCK(fdp);
1909 fdtol->fdl_holdcount--;
1910 if (fdtol->fdl_holdcount == 0 &&
1911 fdtol->fdl_wakeup != 0) {
1912 fdtol->fdl_wakeup = 0;
1913 wakeup(fdtol);
1914 }
1915 }
1916 FILEDESC_UNLOCK(fdp);
1917 }
1918 mtx_unlock(&Giant);
1919 }
1920 return (fdrop(fp, td));
1921 }
1922
1923 /*
1924 * Extract the file pointer associated with the specified descriptor for
1925 * the current user process.
1926 *
1927 * If the descriptor doesn't exist, EBADF is returned.
1928 *
1929 * If the descriptor exists but doesn't match 'flags' then
1930 * return EBADF for read attempts and EINVAL for write attempts.
1931 *
1932 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1933 * It should be droped with fdrop().
1934 * If it is not set, then the refcount will not be bumped however the
1935 * thread's filedesc struct will be returned locked (for fgetsock).
1936 *
1937 * If an error occured the non-zero error is returned and *fpp is set to NULL.
1938 * Otherwise *fpp is set and zero is returned.
1939 */
1940 static __inline int
1941 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1942 {
1943 struct filedesc *fdp;
1944 struct file *fp;
1945
1946 *fpp = NULL;
1947 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1948 return (EBADF);
1949 FILEDESC_LOCK(fdp);
1950 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
1951 FILEDESC_UNLOCK(fdp);
1952 return (EBADF);
1953 }
1954
1955 /*
1956 * Note: FREAD failures returns EBADF to maintain backwards
1957 * compatibility with what routines returned before.
1958 *
1959 * Only one flag, or 0, may be specified.
1960 */
1961 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
1962 FILEDESC_UNLOCK(fdp);
1963 return (EBADF);
1964 }
1965 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
1966 FILEDESC_UNLOCK(fdp);
1967 return (EINVAL);
1968 }
1969 if (hold) {
1970 fhold(fp);
1971 FILEDESC_UNLOCK(fdp);
1972 }
1973 *fpp = fp;
1974 return (0);
1975 }
1976
1977 int
1978 fget(struct thread *td, int fd, struct file **fpp)
1979 {
1980
1981 return(_fget(td, fd, fpp, 0, 1));
1982 }
1983
1984 int
1985 fget_read(struct thread *td, int fd, struct file **fpp)
1986 {
1987
1988 return(_fget(td, fd, fpp, FREAD, 1));
1989 }
1990
1991 int
1992 fget_write(struct thread *td, int fd, struct file **fpp)
1993 {
1994
1995 return(_fget(td, fd, fpp, FWRITE, 1));
1996 }
1997
1998 /*
1999 * Like fget() but loads the underlying vnode, or returns an error if
2000 * the descriptor does not represent a vnode. Note that pipes use vnodes
2001 * but never have VM objects (so VOP_GETVOBJECT() calls will return an
2002 * error). The returned vnode will be vref()d.
2003 *
2004 * XXX: what about the unused flags ?
2005 */
2006 static __inline int
2007 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2008 {
2009 struct file *fp;
2010 int error;
2011
2012 GIANT_REQUIRED; /* VFS */
2013
2014 *vpp = NULL;
2015 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2016 return (error);
2017 if (fp->f_vnode == NULL) {
2018 error = EINVAL;
2019 } else {
2020 *vpp = fp->f_vnode;
2021 vref(*vpp);
2022 }
2023 FILEDESC_UNLOCK(td->td_proc->p_fd);
2024 return (error);
2025 }
2026
2027 int
2028 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2029 {
2030
2031 return (_fgetvp(td, fd, vpp, 0));
2032 }
2033
2034 int
2035 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2036 {
2037
2038 return (_fgetvp(td, fd, vpp, FREAD));
2039 }
2040
2041 #ifdef notyet
2042 int
2043 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2044 {
2045
2046 return (_fgetvp(td, fd, vpp, FWRITE));
2047 }
2048 #endif
2049
2050 /*
2051 * Like fget() but loads the underlying socket, or returns an error if
2052 * the descriptor does not represent a socket.
2053 *
2054 * We bump the ref count on the returned socket. XXX Also obtain the SX
2055 * lock in the future.
2056 */
2057 int
2058 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2059 {
2060 struct file *fp;
2061 int error;
2062
2063 NET_ASSERT_GIANT();
2064
2065 *spp = NULL;
2066 if (fflagp != NULL)
2067 *fflagp = 0;
2068 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2069 return (error);
2070 if (fp->f_type != DTYPE_SOCKET) {
2071 error = ENOTSOCK;
2072 } else {
2073 *spp = fp->f_data;
2074 if (fflagp)
2075 *fflagp = fp->f_flag;
2076 SOCK_LOCK(*spp);
2077 soref(*spp);
2078 SOCK_UNLOCK(*spp);
2079 }
2080 FILEDESC_UNLOCK(td->td_proc->p_fd);
2081 return (error);
2082 }
2083
2084 /*
2085 * Drop the reference count on the the socket and XXX release the SX lock in
2086 * the future. The last reference closes the socket.
2087 */
2088 void
2089 fputsock(struct socket *so)
2090 {
2091
2092 NET_ASSERT_GIANT();
2093 ACCEPT_LOCK();
2094 SOCK_LOCK(so);
2095 sorele(so);
2096 }
2097
2098 int
2099 fdrop(struct file *fp, struct thread *td)
2100 {
2101
2102 FILE_LOCK(fp);
2103 return (fdrop_locked(fp, td));
2104 }
2105
2106 /*
2107 * Drop reference on struct file passed in, may call closef if the
2108 * reference hits zero.
2109 * Expects struct file locked, and will unlock it.
2110 */
2111 static int
2112 fdrop_locked(struct file *fp, struct thread *td)
2113 {
2114 int error;
2115
2116 FILE_LOCK_ASSERT(fp, MA_OWNED);
2117
2118 if (--fp->f_count > 0) {
2119 FILE_UNLOCK(fp);
2120 return (0);
2121 }
2122 /* We have the last ref so we can proceed without the file lock. */
2123 FILE_UNLOCK(fp);
2124 if (fp->f_count < 0)
2125 panic("fdrop: count < 0");
2126 if (fp->f_ops != &badfileops)
2127 error = fo_close(fp, td);
2128 else
2129 error = 0;
2130
2131 sx_xlock(&filelist_lock);
2132 LIST_REMOVE(fp, f_list);
2133 nfiles--;
2134 sx_xunlock(&filelist_lock);
2135 crfree(fp->f_cred);
2136 uma_zfree(file_zone, fp);
2137
2138 return (error);
2139 }
2140
2141 /*
2142 * Apply an advisory lock on a file descriptor.
2143 *
2144 * Just attempt to get a record lock of the requested type on
2145 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2146 */
2147 #ifndef _SYS_SYSPROTO_H_
2148 struct flock_args {
2149 int fd;
2150 int how;
2151 };
2152 #endif
2153 /*
2154 * MPSAFE
2155 */
2156 /* ARGSUSED */
2157 int
2158 flock(struct thread *td, struct flock_args *uap)
2159 {
2160 struct file *fp;
2161 struct vnode *vp;
2162 struct flock lf;
2163 int error;
2164
2165 if ((error = fget(td, uap->fd, &fp)) != 0)
2166 return (error);
2167 if (fp->f_type != DTYPE_VNODE) {
2168 fdrop(fp, td);
2169 return (EOPNOTSUPP);
2170 }
2171
2172 mtx_lock(&Giant);
2173 vp = fp->f_vnode;
2174 lf.l_whence = SEEK_SET;
2175 lf.l_start = 0;
2176 lf.l_len = 0;
2177 if (uap->how & LOCK_UN) {
2178 lf.l_type = F_UNLCK;
2179 FILE_LOCK(fp);
2180 fp->f_flag &= ~FHASLOCK;
2181 FILE_UNLOCK(fp);
2182 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2183 goto done2;
2184 }
2185 if (uap->how & LOCK_EX)
2186 lf.l_type = F_WRLCK;
2187 else if (uap->how & LOCK_SH)
2188 lf.l_type = F_RDLCK;
2189 else {
2190 error = EBADF;
2191 goto done2;
2192 }
2193 FILE_LOCK(fp);
2194 fp->f_flag |= FHASLOCK;
2195 FILE_UNLOCK(fp);
2196 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2197 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2198 done2:
2199 fdrop(fp, td);
2200 mtx_unlock(&Giant);
2201 return (error);
2202 }
2203 /*
2204 * Duplicate the specified descriptor to a free descriptor.
2205 */
2206 int
2207 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2208 {
2209 struct file *wfp;
2210 struct file *fp;
2211
2212 /*
2213 * If the to-be-dup'd fd number is greater than the allowed number
2214 * of file descriptors, or the fd to be dup'd has already been
2215 * closed, then reject.
2216 */
2217 FILEDESC_LOCK(fdp);
2218 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2219 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2220 FILEDESC_UNLOCK(fdp);
2221 return (EBADF);
2222 }
2223
2224 /*
2225 * There are two cases of interest here.
2226 *
2227 * For ENODEV simply dup (dfd) to file descriptor
2228 * (indx) and return.
2229 *
2230 * For ENXIO steal away the file structure from (dfd) and
2231 * store it in (indx). (dfd) is effectively closed by
2232 * this operation.
2233 *
2234 * Any other error code is just returned.
2235 */
2236 switch (error) {
2237 case ENODEV:
2238 /*
2239 * Check that the mode the file is being opened for is a
2240 * subset of the mode of the existing descriptor.
2241 */
2242 FILE_LOCK(wfp);
2243 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2244 FILE_UNLOCK(wfp);
2245 FILEDESC_UNLOCK(fdp);
2246 return (EACCES);
2247 }
2248 fp = fdp->fd_ofiles[indx];
2249 fdp->fd_ofiles[indx] = wfp;
2250 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2251 if (fp == NULL)
2252 fdused(fdp, indx);
2253 fhold_locked(wfp);
2254 FILE_UNLOCK(wfp);
2255 FILEDESC_UNLOCK(fdp);
2256 if (fp != NULL) {
2257 /*
2258 * We now own the reference to fp that the ofiles[]
2259 * array used to own. Release it.
2260 */
2261 FILE_LOCK(fp);
2262 fdrop_locked(fp, td);
2263 }
2264 return (0);
2265
2266 case ENXIO:
2267 /*
2268 * Steal away the file pointer from dfd and stuff it into indx.
2269 */
2270 fp = fdp->fd_ofiles[indx];
2271 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2272 fdp->fd_ofiles[dfd] = NULL;
2273 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2274 fdp->fd_ofileflags[dfd] = 0;
2275 fdunused(fdp, dfd);
2276 if (fp == NULL)
2277 fdused(fdp, indx);
2278 if (fp != NULL)
2279 FILE_LOCK(fp);
2280 FILEDESC_UNLOCK(fdp);
2281
2282 /*
2283 * we now own the reference to fp that the ofiles[] array
2284 * used to own. Release it.
2285 */
2286 if (fp != NULL)
2287 fdrop_locked(fp, td);
2288 return (0);
2289
2290 default:
2291 FILEDESC_UNLOCK(fdp);
2292 return (error);
2293 }
2294 /* NOTREACHED */
2295 }
2296 /*
2297 * Scan all active processes to see if any of them have a current
2298 * or root directory of `olddp'. If so, replace them with the new
2299 * mount point.
2300 */
2301 void
2302 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2303 {
2304 struct filedesc *fdp;
2305 struct proc *p;
2306 int nrele;
2307
2308 if (vrefcnt(olddp) == 1)
2309 return;
2310 sx_slock(&allproc_lock);
2311 LIST_FOREACH(p, &allproc, p_list) {
2312 fdp = fdhold(p);
2313 if (fdp == NULL)
2314 continue;
2315 nrele = 0;
2316 FILEDESC_LOCK_FAST(fdp);
2317 if (fdp->fd_cdir == olddp) {
2318 vref(newdp);
2319 fdp->fd_cdir = newdp;
2320 nrele++;
2321 }
2322 if (fdp->fd_rdir == olddp) {
2323 vref(newdp);
2324 fdp->fd_rdir = newdp;
2325 nrele++;
2326 }
2327 FILEDESC_UNLOCK_FAST(fdp);
2328 fddrop(fdp);
2329 while (nrele--)
2330 vrele(olddp);
2331 }
2332 sx_sunlock(&allproc_lock);
2333 if (rootvnode == olddp) {
2334 vrele(rootvnode);
2335 vref(newdp);
2336 rootvnode = newdp;
2337 }
2338 }
2339
2340 struct filedesc_to_leader *
2341 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2342 {
2343 struct filedesc_to_leader *fdtol;
2344
2345 MALLOC(fdtol, struct filedesc_to_leader *,
2346 sizeof(struct filedesc_to_leader),
2347 M_FILEDESC_TO_LEADER,
2348 M_WAITOK);
2349 fdtol->fdl_refcount = 1;
2350 fdtol->fdl_holdcount = 0;
2351 fdtol->fdl_wakeup = 0;
2352 fdtol->fdl_leader = leader;
2353 if (old != NULL) {
2354 FILEDESC_LOCK(fdp);
2355 fdtol->fdl_next = old->fdl_next;
2356 fdtol->fdl_prev = old;
2357 old->fdl_next = fdtol;
2358 fdtol->fdl_next->fdl_prev = fdtol;
2359 FILEDESC_UNLOCK(fdp);
2360 } else {
2361 fdtol->fdl_next = fdtol;
2362 fdtol->fdl_prev = fdtol;
2363 }
2364 return (fdtol);
2365 }
2366
2367 /*
2368 * Get file structures.
2369 */
2370 static int
2371 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2372 {
2373 struct xfile xf;
2374 struct filedesc *fdp;
2375 struct file *fp;
2376 struct proc *p;
2377 int error, n;
2378
2379 /*
2380 * Note: because the number of file descriptors is calculated
2381 * in different ways for sizing vs returning the data,
2382 * there is information leakage from the first loop. However,
2383 * it is of a similar order of magnitude to the leakage from
2384 * global system statistics such as kern.openfiles.
2385 */
2386 error = sysctl_wire_old_buffer(req, 0);
2387 if (error != 0)
2388 return (error);
2389 if (req->oldptr == NULL) {
2390 n = 16; /* A slight overestimate. */
2391 sx_slock(&filelist_lock);
2392 LIST_FOREACH(fp, &filehead, f_list) {
2393 /*
2394 * We should grab the lock, but this is an
2395 * estimate, so does it really matter?
2396 */
2397 /* mtx_lock(fp->f_mtxp); */
2398 n += fp->f_count;
2399 /* mtx_unlock(f->f_mtxp); */
2400 }
2401 sx_sunlock(&filelist_lock);
2402 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2403 }
2404 error = 0;
2405 bzero(&xf, sizeof(xf));
2406 xf.xf_size = sizeof(xf);
2407 sx_slock(&allproc_lock);
2408 LIST_FOREACH(p, &allproc, p_list) {
2409 if (p->p_state == PRS_NEW)
2410 continue;
2411 PROC_LOCK(p);
2412 if (p_cansee(req->td, p) != 0) {
2413 PROC_UNLOCK(p);
2414 continue;
2415 }
2416 xf.xf_pid = p->p_pid;
2417 xf.xf_uid = p->p_ucred->cr_uid;
2418 PROC_UNLOCK(p);
2419 fdp = fdhold(p);
2420 if (fdp == NULL)
2421 continue;
2422 FILEDESC_LOCK_FAST(fdp);
2423 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2424 if ((fp = fdp->fd_ofiles[n]) == NULL)
2425 continue;
2426 xf.xf_fd = n;
2427 xf.xf_file = fp;
2428 xf.xf_data = fp->f_data;
2429 xf.xf_vnode = fp->f_vnode;
2430 xf.xf_type = fp->f_type;
2431 xf.xf_count = fp->f_count;
2432 xf.xf_msgcount = fp->f_msgcount;
2433 xf.xf_offset = fp->f_offset;
2434 xf.xf_flag = fp->f_flag;
2435 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2436 if (error)
2437 break;
2438 }
2439 FILEDESC_UNLOCK_FAST(fdp);
2440 fddrop(fdp);
2441 if (error)
2442 break;
2443 }
2444 sx_sunlock(&allproc_lock);
2445 return (error);
2446 }
2447
2448 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2449 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2450
2451 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2452 &maxfilesperproc, 0, "Maximum files allowed open per process");
2453
2454 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2455 &maxfiles, 0, "Maximum number of files");
2456
2457 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2458 &nfiles, 0, "System-wide number of open files");
2459
2460 /* ARGSUSED*/
2461 static void
2462 filelistinit(void *dummy)
2463 {
2464
2465 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2466 NULL, NULL, UMA_ALIGN_PTR, 0);
2467 sx_init(&filelist_lock, "filelist lock");
2468 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2469 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2470 }
2471 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2472
2473 /*-------------------------------------------------------------------*/
2474
2475 static int
2476 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2477 {
2478
2479 return (EBADF);
2480 }
2481
2482 static int
2483 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2484 {
2485
2486 return (EBADF);
2487 }
2488
2489 static int
2490 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2491 {
2492
2493 return (0);
2494 }
2495
2496 static int
2497 badfo_kqfilter(struct file *fp, struct knote *kn)
2498 {
2499
2500 return (0);
2501 }
2502
2503 static int
2504 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2505 {
2506
2507 return (EBADF);
2508 }
2509
2510 static int
2511 badfo_close(struct file *fp, struct thread *td)
2512 {
2513
2514 return (EBADF);
2515 }
2516
2517 struct fileops badfileops = {
2518 .fo_read = badfo_readwrite,
2519 .fo_write = badfo_readwrite,
2520 .fo_ioctl = badfo_ioctl,
2521 .fo_poll = badfo_poll,
2522 .fo_kqfilter = badfo_kqfilter,
2523 .fo_stat = badfo_stat,
2524 .fo_close = badfo_close,
2525 };
2526
2527
2528 /*-------------------------------------------------------------------*/
2529
2530 /*
2531 * File Descriptor pseudo-device driver (/dev/fd/).
2532 *
2533 * Opening minor device N dup()s the file (if any) connected to file
2534 * descriptor N belonging to the calling process. Note that this driver
2535 * consists of only the ``open()'' routine, because all subsequent
2536 * references to this file will be direct to the other driver.
2537 *
2538 * XXX: we could give this one a cloning event handler if necessary.
2539 */
2540
2541 /* ARGSUSED */
2542 static int
2543 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2544 {
2545
2546 /*
2547 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2548 * the file descriptor being sought for duplication. The error
2549 * return ensures that the vnode for this device will be released
2550 * by vn_open. Open will detect this special error and take the
2551 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2552 * will simply report the error.
2553 */
2554 td->td_dupfd = dev2unit(dev);
2555 return (ENODEV);
2556 }
2557
2558 static struct cdevsw fildesc_cdevsw = {
2559 .d_version = D_VERSION,
2560 .d_flags = D_NEEDGIANT,
2561 .d_open = fdopen,
2562 .d_name = "FD",
2563 };
2564
2565 static void
2566 fildesc_drvinit(void *unused)
2567 {
2568 struct cdev *dev;
2569
2570 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2571 make_dev_alias(dev, "stdin");
2572 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2573 make_dev_alias(dev, "stdout");
2574 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2575 make_dev_alias(dev, "stderr");
2576 }
2577
2578 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
Cache object: 10d4578afea74e29b51732b1856c32b2
|