1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_compat.h"
41
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 (low - 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 ostat oub;
1036 struct stat ub;
1037 int error;
1038
1039 error = kern_fstat(td, uap->fd, &ub);
1040 if (error == 0) {
1041 cvtstat(&ub, &oub);
1042 error = copyout(&oub, uap->sb, sizeof(oub));
1043 }
1044 return (error);
1045 }
1046 #endif /* COMPAT_43 */
1047
1048 /*
1049 * Return status information about a file descriptor.
1050 */
1051 #ifndef _SYS_SYSPROTO_H_
1052 struct fstat_args {
1053 int fd;
1054 struct stat *sb;
1055 };
1056 #endif
1057 /*
1058 * MPSAFE
1059 */
1060 /* ARGSUSED */
1061 int
1062 fstat(struct thread *td, struct fstat_args *uap)
1063 {
1064 struct stat ub;
1065 int error;
1066
1067 error = kern_fstat(td, uap->fd, &ub);
1068 if (error == 0)
1069 error = copyout(&ub, uap->sb, sizeof(ub));
1070 return (error);
1071 }
1072
1073 int
1074 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1075 {
1076 struct file *fp;
1077 int error;
1078
1079 if ((error = fget(td, fd, &fp)) != 0)
1080 return (error);
1081 error = fo_stat(fp, sbp, td->td_ucred, td);
1082 fdrop(fp, td);
1083 return (error);
1084 }
1085
1086 /*
1087 * Return status information about a file descriptor.
1088 */
1089 #ifndef _SYS_SYSPROTO_H_
1090 struct nfstat_args {
1091 int fd;
1092 struct nstat *sb;
1093 };
1094 #endif
1095 /*
1096 * MPSAFE
1097 */
1098 /* ARGSUSED */
1099 int
1100 nfstat(struct thread *td, struct nfstat_args *uap)
1101 {
1102 struct nstat nub;
1103 struct stat ub;
1104 int error;
1105
1106 error = kern_fstat(td, uap->fd, &ub);
1107 if (error == 0) {
1108 cvtnstat(&ub, &nub);
1109 error = copyout(&nub, uap->sb, sizeof(nub));
1110 }
1111 return (error);
1112 }
1113
1114 /*
1115 * Return pathconf information about a file descriptor.
1116 */
1117 #ifndef _SYS_SYSPROTO_H_
1118 struct fpathconf_args {
1119 int fd;
1120 int name;
1121 };
1122 #endif
1123 /*
1124 * MPSAFE
1125 */
1126 /* ARGSUSED */
1127 int
1128 fpathconf(struct thread *td, struct fpathconf_args *uap)
1129 {
1130 struct file *fp;
1131 struct vnode *vp;
1132 int error;
1133
1134 if ((error = fget(td, uap->fd, &fp)) != 0)
1135 return (error);
1136
1137 /* If asynchronous I/O is available, it works for all descriptors. */
1138 if (uap->name == _PC_ASYNC_IO) {
1139 td->td_retval[0] = async_io_version;
1140 goto out;
1141 }
1142 vp = fp->f_vnode;
1143 if (vp != NULL) {
1144 mtx_lock(&Giant);
1145 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1146 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1147 VOP_UNLOCK(vp, 0, td);
1148 mtx_unlock(&Giant);
1149 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1150 if (uap->name != _PC_PIPE_BUF) {
1151 error = EINVAL;
1152 } else {
1153 td->td_retval[0] = PIPE_BUF;
1154 error = 0;
1155 }
1156 } else {
1157 error = EOPNOTSUPP;
1158 }
1159 out:
1160 fdrop(fp, td);
1161 return (error);
1162 }
1163
1164 /*
1165 * Grow the file table to accomodate (at least) nfd descriptors. This may
1166 * block and drop the filedesc lock, but it will reacquire it before
1167 * returing.
1168 */
1169 static void
1170 fdgrowtable(struct filedesc *fdp, int nfd)
1171 {
1172 struct file **ntable;
1173 char *nfileflags;
1174 int nnfiles, onfiles;
1175 NDSLOTTYPE *nmap;
1176
1177 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1178
1179 KASSERT(fdp->fd_nfiles > 0,
1180 ("zero-length file table"));
1181
1182 /* compute the size of the new table */
1183 onfiles = fdp->fd_nfiles;
1184 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1185 if (nnfiles <= onfiles)
1186 /* the table is already large enough */
1187 return;
1188
1189 /* allocate a new table and (if required) new bitmaps */
1190 FILEDESC_UNLOCK(fdp);
1191 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1192 M_FILEDESC, M_ZERO | M_WAITOK);
1193 nfileflags = (char *)&ntable[nnfiles];
1194 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1195 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1196 M_FILEDESC, M_ZERO | M_WAITOK);
1197 else
1198 nmap = NULL;
1199 FILEDESC_LOCK(fdp);
1200
1201 /*
1202 * We now have new tables ready to go. Since we dropped the
1203 * filedesc lock to call malloc(), watch out for a race.
1204 */
1205 onfiles = fdp->fd_nfiles;
1206 if (onfiles >= nnfiles) {
1207 /* we lost the race, but that's OK */
1208 free(ntable, M_FILEDESC);
1209 if (nmap != NULL)
1210 free(nmap, M_FILEDESC);
1211 return;
1212 }
1213 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1214 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1215 if (onfiles > NDFILE)
1216 free(fdp->fd_ofiles, M_FILEDESC);
1217 fdp->fd_ofiles = ntable;
1218 fdp->fd_ofileflags = nfileflags;
1219 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1220 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1221 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1222 free(fdp->fd_map, M_FILEDESC);
1223 fdp->fd_map = nmap;
1224 }
1225 fdp->fd_nfiles = nnfiles;
1226 }
1227
1228 /*
1229 * Allocate a file descriptor for the process.
1230 */
1231 int
1232 fdalloc(struct thread *td, int minfd, int *result)
1233 {
1234 struct proc *p = td->td_proc;
1235 struct filedesc *fdp = p->p_fd;
1236 int fd = -1, maxfd;
1237
1238 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1239
1240 if (fdp->fd_freefile > minfd)
1241 minfd = fdp->fd_freefile;
1242
1243 PROC_LOCK(p);
1244 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1245 PROC_UNLOCK(p);
1246
1247 /*
1248 * Search the bitmap for a free descriptor. If none is found, try
1249 * to grow the file table. Keep at it until we either get a file
1250 * descriptor or run into process or system limits; fdgrowtable()
1251 * may drop the filedesc lock, so we're in a race.
1252 */
1253 for (;;) {
1254 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1255 if (fd >= maxfd)
1256 return (EMFILE);
1257 if (fd < fdp->fd_nfiles)
1258 break;
1259 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1260 }
1261
1262 /*
1263 * Perform some sanity checks, then mark the file descriptor as
1264 * used and return it to the caller.
1265 */
1266 KASSERT(!fdisused(fdp, fd),
1267 ("fd_first_free() returned non-free descriptor"));
1268 KASSERT(fdp->fd_ofiles[fd] == NULL,
1269 ("free descriptor isn't"));
1270 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1271 fdused(fdp, fd);
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 newfdp->fd_fd.fd_lastfile = -1;
1406 return (&newfdp->fd_fd);
1407 }
1408
1409 static struct filedesc *
1410 fdhold(struct proc *p)
1411 {
1412 struct filedesc *fdp;
1413
1414 mtx_lock(&fdesc_mtx);
1415 fdp = p->p_fd;
1416 if (fdp != NULL)
1417 fdp->fd_holdcnt++;
1418 mtx_unlock(&fdesc_mtx);
1419 return (fdp);
1420 }
1421
1422 static void
1423 fddrop(struct filedesc *fdp)
1424 {
1425 int i;
1426
1427 mtx_lock(&fdesc_mtx);
1428 i = --fdp->fd_holdcnt;
1429 mtx_unlock(&fdesc_mtx);
1430 if (i > 0)
1431 return;
1432
1433 mtx_destroy(&fdp->fd_mtx);
1434 FREE(fdp, M_FILEDESC);
1435 }
1436
1437 /*
1438 * Share a filedesc structure.
1439 */
1440 struct filedesc *
1441 fdshare(struct filedesc *fdp)
1442 {
1443 FILEDESC_LOCK_FAST(fdp);
1444 fdp->fd_refcnt++;
1445 FILEDESC_UNLOCK_FAST(fdp);
1446 return (fdp);
1447 }
1448
1449 /*
1450 * Unshare a filedesc structure, if necessary by making a copy
1451 */
1452 void
1453 fdunshare(struct proc *p, struct thread *td)
1454 {
1455
1456 FILEDESC_LOCK_FAST(p->p_fd);
1457 if (p->p_fd->fd_refcnt > 1) {
1458 struct filedesc *tmp;
1459
1460 FILEDESC_UNLOCK_FAST(p->p_fd);
1461 tmp = fdcopy(p->p_fd);
1462 fdfree(td);
1463 p->p_fd = tmp;
1464 } else
1465 FILEDESC_UNLOCK_FAST(p->p_fd);
1466 }
1467
1468 /*
1469 * Copy a filedesc structure.
1470 * A NULL pointer in returns a NULL reference, this is to ease callers,
1471 * not catch errors.
1472 */
1473 struct filedesc *
1474 fdcopy(struct filedesc *fdp)
1475 {
1476 struct filedesc *newfdp;
1477 int i;
1478
1479 /* Certain daemons might not have file descriptors. */
1480 if (fdp == NULL)
1481 return (NULL);
1482
1483 newfdp = fdinit(fdp);
1484 FILEDESC_LOCK_FAST(fdp);
1485 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1486 FILEDESC_UNLOCK_FAST(fdp);
1487 FILEDESC_LOCK(newfdp);
1488 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1489 FILEDESC_UNLOCK(newfdp);
1490 FILEDESC_LOCK_FAST(fdp);
1491 }
1492 /* copy everything except kqueue descriptors */
1493 newfdp->fd_freefile = -1;
1494 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1495 if (fdisused(fdp, i) &&
1496 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1497 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1498 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1499 fhold(newfdp->fd_ofiles[i]);
1500 newfdp->fd_lastfile = i;
1501 } else {
1502 if (newfdp->fd_freefile == -1)
1503 newfdp->fd_freefile = i;
1504 }
1505 }
1506 FILEDESC_UNLOCK_FAST(fdp);
1507 FILEDESC_LOCK(newfdp);
1508 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1509 if (newfdp->fd_ofiles[i] != NULL)
1510 fdused(newfdp, i);
1511 FILEDESC_UNLOCK(newfdp);
1512 FILEDESC_LOCK_FAST(fdp);
1513 if (newfdp->fd_freefile == -1)
1514 newfdp->fd_freefile = i;
1515 newfdp->fd_cmask = fdp->fd_cmask;
1516 FILEDESC_UNLOCK_FAST(fdp);
1517 return (newfdp);
1518 }
1519
1520 /*
1521 * Release a filedesc structure.
1522 */
1523 void
1524 fdfree(struct thread *td)
1525 {
1526 struct filedesc *fdp;
1527 struct file **fpp;
1528 int i;
1529 struct filedesc_to_leader *fdtol;
1530 struct file *fp;
1531 struct vnode *vp;
1532 struct flock lf;
1533
1534 GIANT_REQUIRED; /* VFS */
1535
1536 /* Certain daemons might not have file descriptors. */
1537 fdp = td->td_proc->p_fd;
1538 if (fdp == NULL)
1539 return;
1540
1541 /* Check for special need to clear POSIX style locks */
1542 fdtol = td->td_proc->p_fdtol;
1543 if (fdtol != NULL) {
1544 FILEDESC_LOCK(fdp);
1545 KASSERT(fdtol->fdl_refcount > 0,
1546 ("filedesc_to_refcount botch: fdl_refcount=%d",
1547 fdtol->fdl_refcount));
1548 if (fdtol->fdl_refcount == 1 &&
1549 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1550 i = 0;
1551 fpp = fdp->fd_ofiles;
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 (void) VOP_ADVLOCK(vp,
1567 (caddr_t)td->td_proc->
1568 p_leader,
1569 F_UNLCK,
1570 &lf,
1571 F_POSIX);
1572 FILEDESC_LOCK(fdp);
1573 fdrop(fp, td);
1574 fpp = fdp->fd_ofiles + i;
1575 }
1576 }
1577 retry:
1578 if (fdtol->fdl_refcount == 1) {
1579 if (fdp->fd_holdleaderscount > 0 &&
1580 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1581 /*
1582 * close() or do_dup() has cleared a reference
1583 * in a shared file descriptor table.
1584 */
1585 fdp->fd_holdleaderswakeup = 1;
1586 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1587 PLOCK, "fdlhold", 0);
1588 goto retry;
1589 }
1590 if (fdtol->fdl_holdcount > 0) {
1591 /*
1592 * Ensure that fdtol->fdl_leader
1593 * remains valid in closef().
1594 */
1595 fdtol->fdl_wakeup = 1;
1596 msleep(fdtol, &fdp->fd_mtx,
1597 PLOCK, "fdlhold", 0);
1598 goto retry;
1599 }
1600 }
1601 fdtol->fdl_refcount--;
1602 if (fdtol->fdl_refcount == 0 &&
1603 fdtol->fdl_holdcount == 0) {
1604 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1605 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1606 } else
1607 fdtol = NULL;
1608 td->td_proc->p_fdtol = NULL;
1609 FILEDESC_UNLOCK(fdp);
1610 if (fdtol != NULL)
1611 FREE(fdtol, M_FILEDESC_TO_LEADER);
1612 }
1613 FILEDESC_LOCK_FAST(fdp);
1614 i = --fdp->fd_refcnt;
1615 FILEDESC_UNLOCK_FAST(fdp);
1616 if (i > 0)
1617 return;
1618 /*
1619 * We are the last reference to the structure, so we can
1620 * safely assume it will not change out from under us.
1621 */
1622 fpp = fdp->fd_ofiles;
1623 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1624 if (*fpp)
1625 (void) closef(*fpp, td);
1626 }
1627 FILEDESC_LOCK(fdp);
1628
1629 /* XXX This should happen earlier. */
1630 mtx_lock(&fdesc_mtx);
1631 td->td_proc->p_fd = NULL;
1632 mtx_unlock(&fdesc_mtx);
1633
1634 if (fdp->fd_nfiles > NDFILE)
1635 FREE(fdp->fd_ofiles, M_FILEDESC);
1636 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1637 FREE(fdp->fd_map, M_FILEDESC);
1638
1639 fdp->fd_nfiles = 0;
1640
1641 if (fdp->fd_cdir)
1642 vrele(fdp->fd_cdir);
1643 fdp->fd_cdir = NULL;
1644 if (fdp->fd_rdir)
1645 vrele(fdp->fd_rdir);
1646 fdp->fd_rdir = NULL;
1647 if (fdp->fd_jdir)
1648 vrele(fdp->fd_jdir);
1649 fdp->fd_jdir = NULL;
1650
1651 FILEDESC_UNLOCK(fdp);
1652
1653 fddrop(fdp);
1654 }
1655
1656 /*
1657 * For setugid programs, we don't want to people to use that setugidness
1658 * to generate error messages which write to a file which otherwise would
1659 * otherwise be off-limits to the process. We check for filesystems where
1660 * the vnode can change out from under us after execve (like [lin]procfs).
1661 *
1662 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1663 * sufficient. We also don't for check setugidness since we know we are.
1664 */
1665 static int
1666 is_unsafe(struct file *fp)
1667 {
1668 if (fp->f_type == DTYPE_VNODE) {
1669 struct vnode *vp = fp->f_vnode;
1670
1671 if ((vp->v_vflag & VV_PROCDEP) != 0)
1672 return (1);
1673 }
1674 return (0);
1675 }
1676
1677 /*
1678 * Make this setguid thing safe, if at all possible.
1679 */
1680 void
1681 setugidsafety(struct thread *td)
1682 {
1683 struct filedesc *fdp;
1684 int i;
1685
1686 /* Certain daemons might not have file descriptors. */
1687 fdp = td->td_proc->p_fd;
1688 if (fdp == NULL)
1689 return;
1690
1691 /*
1692 * Note: fdp->fd_ofiles may be reallocated out from under us while
1693 * we are blocked in a close. Be careful!
1694 */
1695 FILEDESC_LOCK(fdp);
1696 for (i = 0; i <= fdp->fd_lastfile; i++) {
1697 if (i > 2)
1698 break;
1699 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1700 struct file *fp;
1701
1702 knote_fdclose(td, i);
1703 /*
1704 * NULL-out descriptor prior to close to avoid
1705 * a race while close blocks.
1706 */
1707 fp = fdp->fd_ofiles[i];
1708 fdp->fd_ofiles[i] = NULL;
1709 fdp->fd_ofileflags[i] = 0;
1710 fdunused(fdp, i);
1711 FILEDESC_UNLOCK(fdp);
1712 (void) closef(fp, td);
1713 FILEDESC_LOCK(fdp);
1714 }
1715 }
1716 FILEDESC_UNLOCK(fdp);
1717 }
1718
1719 void
1720 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1721 {
1722
1723 FILEDESC_LOCK(fdp);
1724 if (fdp->fd_ofiles[idx] == fp) {
1725 fdp->fd_ofiles[idx] = NULL;
1726 fdunused(fdp, idx);
1727 FILEDESC_UNLOCK(fdp);
1728 fdrop(fp, td);
1729 } else {
1730 FILEDESC_UNLOCK(fdp);
1731 }
1732 }
1733
1734 /*
1735 * Close any files on exec?
1736 */
1737 void
1738 fdcloseexec(struct thread *td)
1739 {
1740 struct filedesc *fdp;
1741 int i;
1742
1743 /* Certain daemons might not have file descriptors. */
1744 fdp = td->td_proc->p_fd;
1745 if (fdp == NULL)
1746 return;
1747
1748 FILEDESC_LOCK(fdp);
1749
1750 /*
1751 * We cannot cache fd_ofiles or fd_ofileflags since operations
1752 * may block and rip them out from under us.
1753 */
1754 for (i = 0; i <= fdp->fd_lastfile; i++) {
1755 if (fdp->fd_ofiles[i] != NULL &&
1756 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1757 struct file *fp;
1758
1759 knote_fdclose(td, i);
1760 /*
1761 * NULL-out descriptor prior to close to avoid
1762 * a race while close blocks.
1763 */
1764 fp = fdp->fd_ofiles[i];
1765 fdp->fd_ofiles[i] = NULL;
1766 fdp->fd_ofileflags[i] = 0;
1767 fdunused(fdp, i);
1768 FILEDESC_UNLOCK(fdp);
1769 (void) closef(fp, td);
1770 FILEDESC_LOCK(fdp);
1771 }
1772 }
1773 FILEDESC_UNLOCK(fdp);
1774 }
1775
1776 /*
1777 * It is unsafe for set[ug]id processes to be started with file
1778 * descriptors 0..2 closed, as these descriptors are given implicit
1779 * significance in the Standard C library. fdcheckstd() will create a
1780 * descriptor referencing /dev/null for each of stdin, stdout, and
1781 * stderr that is not already open.
1782 */
1783 int
1784 fdcheckstd(struct thread *td)
1785 {
1786 struct nameidata nd;
1787 struct filedesc *fdp;
1788 struct file *fp;
1789 register_t retval;
1790 int fd, i, error, flags, devnull;
1791
1792 GIANT_REQUIRED; /* VFS */
1793
1794 fdp = td->td_proc->p_fd;
1795 if (fdp == NULL)
1796 return (0);
1797 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1798 devnull = -1;
1799 error = 0;
1800 for (i = 0; i < 3; i++) {
1801 if (fdp->fd_ofiles[i] != NULL)
1802 continue;
1803 if (devnull < 0) {
1804 error = falloc(td, &fp, &fd);
1805 if (error != 0)
1806 break;
1807 /* Note extra ref on `fp' held for us by falloc(). */
1808 KASSERT(fd == i, ("oof, we didn't get our fd"));
1809 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null",
1810 td);
1811 flags = FREAD | FWRITE;
1812 error = vn_open(&nd, &flags, 0, -1);
1813 if (error != 0) {
1814 /*
1815 * Someone may have closed the entry in the
1816 * file descriptor table, so check it hasn't
1817 * changed before dropping the reference count.
1818 */
1819 FILEDESC_LOCK(fdp);
1820 KASSERT(fdp->fd_ofiles[fd] == fp,
1821 ("table not shared, how did it change?"));
1822 fdp->fd_ofiles[fd] = NULL;
1823 fdunused(fdp, fd);
1824 FILEDESC_UNLOCK(fdp);
1825 fdrop(fp, td);
1826 fdrop(fp, td);
1827 break;
1828 }
1829 NDFREE(&nd, NDF_ONLY_PNBUF);
1830 fp->f_flag = flags;
1831 fp->f_vnode = nd.ni_vp;
1832 if (fp->f_data == NULL)
1833 fp->f_data = nd.ni_vp;
1834 if (fp->f_ops == &badfileops)
1835 fp->f_ops = &vnops;
1836 fp->f_type = DTYPE_VNODE;
1837 VOP_UNLOCK(nd.ni_vp, 0, td);
1838 devnull = fd;
1839 fdrop(fp, td);
1840 } else {
1841 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1842 if (error != 0)
1843 break;
1844 }
1845 }
1846 return (error);
1847 }
1848
1849 /*
1850 * Internal form of close.
1851 * Decrement reference count on file structure.
1852 * Note: td may be NULL when closing a file that was being passed in a
1853 * message.
1854 *
1855 * XXXRW: Giant is not required for the caller, but often will be held; this
1856 * makes it moderately likely the Giant will be recursed in the VFS case.
1857 */
1858 int
1859 closef(struct file *fp, struct thread *td)
1860 {
1861 struct vnode *vp;
1862 struct flock lf;
1863 struct filedesc_to_leader *fdtol;
1864 struct filedesc *fdp;
1865
1866 /*
1867 * POSIX record locking dictates that any close releases ALL
1868 * locks owned by this process. This is handled by setting
1869 * a flag in the unlock to free ONLY locks obeying POSIX
1870 * semantics, and not to free BSD-style file locks.
1871 * If the descriptor was in a message, POSIX-style locks
1872 * aren't passed with the descriptor, and the thread pointer
1873 * will be NULL. Callers should be careful only to pass a
1874 * NULL thread pointer when there really is no owning
1875 * context that might have locks, or the locks will be
1876 * leaked.
1877 */
1878 if (fp->f_type == DTYPE_VNODE && td != NULL) {
1879 mtx_lock(&Giant);
1880 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1881 lf.l_whence = SEEK_SET;
1882 lf.l_start = 0;
1883 lf.l_len = 0;
1884 lf.l_type = F_UNLCK;
1885 vp = fp->f_vnode;
1886 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1887 F_UNLCK, &lf, F_POSIX);
1888 }
1889 fdtol = td->td_proc->p_fdtol;
1890 if (fdtol != NULL) {
1891 /*
1892 * Handle special case where file descriptor table
1893 * is shared between multiple process leaders.
1894 */
1895 fdp = td->td_proc->p_fd;
1896 FILEDESC_LOCK(fdp);
1897 for (fdtol = fdtol->fdl_next;
1898 fdtol != td->td_proc->p_fdtol;
1899 fdtol = fdtol->fdl_next) {
1900 if ((fdtol->fdl_leader->p_flag &
1901 P_ADVLOCK) == 0)
1902 continue;
1903 fdtol->fdl_holdcount++;
1904 FILEDESC_UNLOCK(fdp);
1905 lf.l_whence = SEEK_SET;
1906 lf.l_start = 0;
1907 lf.l_len = 0;
1908 lf.l_type = F_UNLCK;
1909 vp = fp->f_vnode;
1910 (void) VOP_ADVLOCK(vp,
1911 (caddr_t)fdtol->fdl_leader,
1912 F_UNLCK, &lf, F_POSIX);
1913 FILEDESC_LOCK(fdp);
1914 fdtol->fdl_holdcount--;
1915 if (fdtol->fdl_holdcount == 0 &&
1916 fdtol->fdl_wakeup != 0) {
1917 fdtol->fdl_wakeup = 0;
1918 wakeup(fdtol);
1919 }
1920 }
1921 FILEDESC_UNLOCK(fdp);
1922 }
1923 mtx_unlock(&Giant);
1924 }
1925 return (fdrop(fp, td));
1926 }
1927
1928 /*
1929 * Extract the file pointer associated with the specified descriptor for
1930 * the current user process.
1931 *
1932 * If the descriptor doesn't exist, EBADF is returned.
1933 *
1934 * If the descriptor exists but doesn't match 'flags' then
1935 * return EBADF for read attempts and EINVAL for write attempts.
1936 *
1937 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1938 * It should be droped with fdrop().
1939 * If it is not set, then the refcount will not be bumped however the
1940 * thread's filedesc struct will be returned locked (for fgetsock).
1941 *
1942 * If an error occured the non-zero error is returned and *fpp is set to NULL.
1943 * Otherwise *fpp is set and zero is returned.
1944 */
1945 static __inline int
1946 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1947 {
1948 struct filedesc *fdp;
1949 struct file *fp;
1950
1951 *fpp = NULL;
1952 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1953 return (EBADF);
1954 FILEDESC_LOCK(fdp);
1955 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
1956 FILEDESC_UNLOCK(fdp);
1957 return (EBADF);
1958 }
1959
1960 /*
1961 * Note: FREAD failures returns EBADF to maintain backwards
1962 * compatibility with what routines returned before.
1963 *
1964 * Only one flag, or 0, may be specified.
1965 */
1966 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
1967 FILEDESC_UNLOCK(fdp);
1968 return (EBADF);
1969 }
1970 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
1971 FILEDESC_UNLOCK(fdp);
1972 return (EINVAL);
1973 }
1974 if (hold) {
1975 fhold(fp);
1976 FILEDESC_UNLOCK(fdp);
1977 }
1978 *fpp = fp;
1979 return (0);
1980 }
1981
1982 int
1983 fget(struct thread *td, int fd, struct file **fpp)
1984 {
1985
1986 return(_fget(td, fd, fpp, 0, 1));
1987 }
1988
1989 int
1990 fget_read(struct thread *td, int fd, struct file **fpp)
1991 {
1992
1993 return(_fget(td, fd, fpp, FREAD, 1));
1994 }
1995
1996 int
1997 fget_write(struct thread *td, int fd, struct file **fpp)
1998 {
1999
2000 return(_fget(td, fd, fpp, FWRITE, 1));
2001 }
2002
2003 /*
2004 * Like fget() but loads the underlying vnode, or returns an error if
2005 * the descriptor does not represent a vnode. Note that pipes use vnodes
2006 * but never have VM objects (so VOP_GETVOBJECT() calls will return an
2007 * error). The returned vnode will be vref()d.
2008 *
2009 * XXX: what about the unused flags ?
2010 */
2011 static __inline int
2012 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2013 {
2014 struct file *fp;
2015 int error;
2016
2017 GIANT_REQUIRED; /* VFS */
2018
2019 *vpp = NULL;
2020 if ((error = _fget(td, fd, &fp, flags, 0)) != 0)
2021 return (error);
2022 if (fp->f_vnode == NULL) {
2023 error = EINVAL;
2024 } else {
2025 *vpp = fp->f_vnode;
2026 vref(*vpp);
2027 }
2028 FILEDESC_UNLOCK(td->td_proc->p_fd);
2029 return (error);
2030 }
2031
2032 int
2033 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2034 {
2035
2036 return (_fgetvp(td, fd, vpp, 0));
2037 }
2038
2039 int
2040 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2041 {
2042
2043 return (_fgetvp(td, fd, vpp, FREAD));
2044 }
2045
2046 #ifdef notyet
2047 int
2048 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2049 {
2050
2051 return (_fgetvp(td, fd, vpp, FWRITE));
2052 }
2053 #endif
2054
2055 /*
2056 * Like fget() but loads the underlying socket, or returns an error if
2057 * the descriptor does not represent a socket.
2058 *
2059 * We bump the ref count on the returned socket. XXX Also obtain the SX
2060 * lock in the future.
2061 */
2062 int
2063 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2064 {
2065 struct file *fp;
2066 int error;
2067
2068 NET_ASSERT_GIANT();
2069
2070 *spp = NULL;
2071 if (fflagp != NULL)
2072 *fflagp = 0;
2073 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2074 return (error);
2075 if (fp->f_type != DTYPE_SOCKET) {
2076 error = ENOTSOCK;
2077 } else {
2078 *spp = fp->f_data;
2079 if (fflagp)
2080 *fflagp = fp->f_flag;
2081 SOCK_LOCK(*spp);
2082 soref(*spp);
2083 SOCK_UNLOCK(*spp);
2084 }
2085 FILEDESC_UNLOCK(td->td_proc->p_fd);
2086 return (error);
2087 }
2088
2089 /*
2090 * Drop the reference count on the the socket and XXX release the SX lock in
2091 * the future. The last reference closes the socket.
2092 */
2093 void
2094 fputsock(struct socket *so)
2095 {
2096
2097 NET_ASSERT_GIANT();
2098 ACCEPT_LOCK();
2099 SOCK_LOCK(so);
2100 sorele(so);
2101 }
2102
2103 int
2104 fdrop(struct file *fp, struct thread *td)
2105 {
2106
2107 FILE_LOCK(fp);
2108 return (fdrop_locked(fp, td));
2109 }
2110
2111 /*
2112 * Drop reference on struct file passed in, may call closef if the
2113 * reference hits zero.
2114 * Expects struct file locked, and will unlock it.
2115 */
2116 static int
2117 fdrop_locked(struct file *fp, struct thread *td)
2118 {
2119 int error;
2120
2121 FILE_LOCK_ASSERT(fp, MA_OWNED);
2122
2123 if (--fp->f_count > 0) {
2124 FILE_UNLOCK(fp);
2125 return (0);
2126 }
2127 /* We have the last ref so we can proceed without the file lock. */
2128 FILE_UNLOCK(fp);
2129 if (fp->f_count < 0)
2130 panic("fdrop: count < 0");
2131 if (fp->f_ops != &badfileops)
2132 error = fo_close(fp, td);
2133 else
2134 error = 0;
2135
2136 sx_xlock(&filelist_lock);
2137 LIST_REMOVE(fp, f_list);
2138 nfiles--;
2139 sx_xunlock(&filelist_lock);
2140 crfree(fp->f_cred);
2141 uma_zfree(file_zone, fp);
2142
2143 return (error);
2144 }
2145
2146 /*
2147 * Apply an advisory lock on a file descriptor.
2148 *
2149 * Just attempt to get a record lock of the requested type on
2150 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2151 */
2152 #ifndef _SYS_SYSPROTO_H_
2153 struct flock_args {
2154 int fd;
2155 int how;
2156 };
2157 #endif
2158 /*
2159 * MPSAFE
2160 */
2161 /* ARGSUSED */
2162 int
2163 flock(struct thread *td, struct flock_args *uap)
2164 {
2165 struct file *fp;
2166 struct vnode *vp;
2167 struct flock lf;
2168 int error;
2169
2170 if ((error = fget(td, uap->fd, &fp)) != 0)
2171 return (error);
2172 if (fp->f_type != DTYPE_VNODE) {
2173 fdrop(fp, td);
2174 return (EOPNOTSUPP);
2175 }
2176
2177 mtx_lock(&Giant);
2178 vp = fp->f_vnode;
2179 lf.l_whence = SEEK_SET;
2180 lf.l_start = 0;
2181 lf.l_len = 0;
2182 if (uap->how & LOCK_UN) {
2183 lf.l_type = F_UNLCK;
2184 FILE_LOCK(fp);
2185 fp->f_flag &= ~FHASLOCK;
2186 FILE_UNLOCK(fp);
2187 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2188 goto done2;
2189 }
2190 if (uap->how & LOCK_EX)
2191 lf.l_type = F_WRLCK;
2192 else if (uap->how & LOCK_SH)
2193 lf.l_type = F_RDLCK;
2194 else {
2195 error = EBADF;
2196 goto done2;
2197 }
2198 FILE_LOCK(fp);
2199 fp->f_flag |= FHASLOCK;
2200 FILE_UNLOCK(fp);
2201 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2202 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2203 done2:
2204 fdrop(fp, td);
2205 mtx_unlock(&Giant);
2206 return (error);
2207 }
2208 /*
2209 * Duplicate the specified descriptor to a free descriptor.
2210 */
2211 int
2212 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2213 {
2214 struct file *wfp;
2215 struct file *fp;
2216
2217 /*
2218 * If the to-be-dup'd fd number is greater than the allowed number
2219 * of file descriptors, or the fd to be dup'd has already been
2220 * closed, then reject.
2221 */
2222 FILEDESC_LOCK(fdp);
2223 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2224 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2225 FILEDESC_UNLOCK(fdp);
2226 return (EBADF);
2227 }
2228
2229 /*
2230 * There are two cases of interest here.
2231 *
2232 * For ENODEV simply dup (dfd) to file descriptor
2233 * (indx) and return.
2234 *
2235 * For ENXIO steal away the file structure from (dfd) and
2236 * store it in (indx). (dfd) is effectively closed by
2237 * this operation.
2238 *
2239 * Any other error code is just returned.
2240 */
2241 switch (error) {
2242 case ENODEV:
2243 /*
2244 * Check that the mode the file is being opened for is a
2245 * subset of the mode of the existing descriptor.
2246 */
2247 FILE_LOCK(wfp);
2248 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2249 FILE_UNLOCK(wfp);
2250 FILEDESC_UNLOCK(fdp);
2251 return (EACCES);
2252 }
2253 fp = fdp->fd_ofiles[indx];
2254 fdp->fd_ofiles[indx] = wfp;
2255 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2256 if (fp == NULL)
2257 fdused(fdp, indx);
2258 fhold_locked(wfp);
2259 FILE_UNLOCK(wfp);
2260 FILEDESC_UNLOCK(fdp);
2261 if (fp != NULL) {
2262 /*
2263 * We now own the reference to fp that the ofiles[]
2264 * array used to own. Release it.
2265 */
2266 FILE_LOCK(fp);
2267 fdrop_locked(fp, td);
2268 }
2269 return (0);
2270
2271 case ENXIO:
2272 /*
2273 * Steal away the file pointer from dfd and stuff it into indx.
2274 */
2275 fp = fdp->fd_ofiles[indx];
2276 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2277 fdp->fd_ofiles[dfd] = NULL;
2278 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2279 fdp->fd_ofileflags[dfd] = 0;
2280 fdunused(fdp, dfd);
2281 if (fp == NULL)
2282 fdused(fdp, indx);
2283 if (fp != NULL)
2284 FILE_LOCK(fp);
2285
2286 /*
2287 * we now own the reference to fp that the ofiles[] array
2288 * used to own. Release it.
2289 */
2290 if (fp != NULL)
2291 fdrop_locked(fp, td);
2292
2293 FILEDESC_UNLOCK(fdp);
2294
2295 return (0);
2296
2297 default:
2298 FILEDESC_UNLOCK(fdp);
2299 return (error);
2300 }
2301 /* NOTREACHED */
2302 }
2303 /*
2304 * Scan all active processes to see if any of them have a current
2305 * or root directory of `olddp'. If so, replace them with the new
2306 * mount point.
2307 */
2308 void
2309 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2310 {
2311 struct filedesc *fdp;
2312 struct proc *p;
2313 int nrele;
2314
2315 if (vrefcnt(olddp) == 1)
2316 return;
2317 sx_slock(&allproc_lock);
2318 LIST_FOREACH(p, &allproc, p_list) {
2319 fdp = fdhold(p);
2320 if (fdp == NULL)
2321 continue;
2322 nrele = 0;
2323 FILEDESC_LOCK_FAST(fdp);
2324 if (fdp->fd_cdir == olddp) {
2325 vref(newdp);
2326 fdp->fd_cdir = newdp;
2327 nrele++;
2328 }
2329 if (fdp->fd_rdir == olddp) {
2330 vref(newdp);
2331 fdp->fd_rdir = newdp;
2332 nrele++;
2333 }
2334 FILEDESC_UNLOCK_FAST(fdp);
2335 fddrop(fdp);
2336 while (nrele--)
2337 vrele(olddp);
2338 }
2339 sx_sunlock(&allproc_lock);
2340 if (rootvnode == olddp) {
2341 vrele(rootvnode);
2342 vref(newdp);
2343 rootvnode = newdp;
2344 }
2345 }
2346
2347 struct filedesc_to_leader *
2348 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2349 {
2350 struct filedesc_to_leader *fdtol;
2351
2352 MALLOC(fdtol, struct filedesc_to_leader *,
2353 sizeof(struct filedesc_to_leader),
2354 M_FILEDESC_TO_LEADER,
2355 M_WAITOK);
2356 fdtol->fdl_refcount = 1;
2357 fdtol->fdl_holdcount = 0;
2358 fdtol->fdl_wakeup = 0;
2359 fdtol->fdl_leader = leader;
2360 if (old != NULL) {
2361 FILEDESC_LOCK(fdp);
2362 fdtol->fdl_next = old->fdl_next;
2363 fdtol->fdl_prev = old;
2364 old->fdl_next = fdtol;
2365 fdtol->fdl_next->fdl_prev = fdtol;
2366 FILEDESC_UNLOCK(fdp);
2367 } else {
2368 fdtol->fdl_next = fdtol;
2369 fdtol->fdl_prev = fdtol;
2370 }
2371 return (fdtol);
2372 }
2373
2374 /*
2375 * Get file structures.
2376 */
2377 static int
2378 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2379 {
2380 struct xfile xf;
2381 struct filedesc *fdp;
2382 struct file *fp;
2383 struct proc *p;
2384 int error, n;
2385
2386 /*
2387 * Note: because the number of file descriptors is calculated
2388 * in different ways for sizing vs returning the data,
2389 * there is information leakage from the first loop. However,
2390 * it is of a similar order of magnitude to the leakage from
2391 * global system statistics such as kern.openfiles.
2392 */
2393 error = sysctl_wire_old_buffer(req, 0);
2394 if (error != 0)
2395 return (error);
2396 if (req->oldptr == NULL) {
2397 n = 16; /* A slight overestimate. */
2398 sx_slock(&filelist_lock);
2399 LIST_FOREACH(fp, &filehead, f_list) {
2400 /*
2401 * We should grab the lock, but this is an
2402 * estimate, so does it really matter?
2403 */
2404 /* mtx_lock(fp->f_mtxp); */
2405 n += fp->f_count;
2406 /* mtx_unlock(f->f_mtxp); */
2407 }
2408 sx_sunlock(&filelist_lock);
2409 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2410 }
2411 error = 0;
2412 bzero(&xf, sizeof(xf));
2413 xf.xf_size = sizeof(xf);
2414 sx_slock(&allproc_lock);
2415 LIST_FOREACH(p, &allproc, p_list) {
2416 if (p->p_state == PRS_NEW)
2417 continue;
2418 PROC_LOCK(p);
2419 if (p_cansee(req->td, p) != 0) {
2420 PROC_UNLOCK(p);
2421 continue;
2422 }
2423 xf.xf_pid = p->p_pid;
2424 xf.xf_uid = p->p_ucred->cr_uid;
2425 PROC_UNLOCK(p);
2426 fdp = fdhold(p);
2427 if (fdp == NULL)
2428 continue;
2429 FILEDESC_LOCK_FAST(fdp);
2430 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2431 if ((fp = fdp->fd_ofiles[n]) == NULL)
2432 continue;
2433 xf.xf_fd = n;
2434 xf.xf_file = fp;
2435 xf.xf_data = fp->f_data;
2436 xf.xf_vnode = fp->f_vnode;
2437 xf.xf_type = fp->f_type;
2438 xf.xf_count = fp->f_count;
2439 xf.xf_msgcount = fp->f_msgcount;
2440 xf.xf_offset = fp->f_offset;
2441 xf.xf_flag = fp->f_flag;
2442 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2443 if (error)
2444 break;
2445 }
2446 FILEDESC_UNLOCK_FAST(fdp);
2447 fddrop(fdp);
2448 if (error)
2449 break;
2450 }
2451 sx_sunlock(&allproc_lock);
2452 return (error);
2453 }
2454
2455 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2456 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2457
2458 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2459 &maxfilesperproc, 0, "Maximum files allowed open per process");
2460
2461 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2462 &maxfiles, 0, "Maximum number of files");
2463
2464 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2465 &nfiles, 0, "System-wide number of open files");
2466
2467 /* ARGSUSED*/
2468 static void
2469 filelistinit(void *dummy)
2470 {
2471
2472 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2473 NULL, NULL, UMA_ALIGN_PTR, 0);
2474 sx_init(&filelist_lock, "filelist lock");
2475 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2476 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2477 }
2478 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2479
2480 /*-------------------------------------------------------------------*/
2481
2482 static int
2483 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2484 {
2485
2486 return (EBADF);
2487 }
2488
2489 static int
2490 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2491 {
2492
2493 return (EBADF);
2494 }
2495
2496 static int
2497 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2498 {
2499
2500 return (0);
2501 }
2502
2503 static int
2504 badfo_kqfilter(struct file *fp, struct knote *kn)
2505 {
2506
2507 return (0);
2508 }
2509
2510 static int
2511 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2512 {
2513
2514 return (EBADF);
2515 }
2516
2517 static int
2518 badfo_close(struct file *fp, struct thread *td)
2519 {
2520
2521 return (EBADF);
2522 }
2523
2524 struct fileops badfileops = {
2525 .fo_read = badfo_readwrite,
2526 .fo_write = badfo_readwrite,
2527 .fo_ioctl = badfo_ioctl,
2528 .fo_poll = badfo_poll,
2529 .fo_kqfilter = badfo_kqfilter,
2530 .fo_stat = badfo_stat,
2531 .fo_close = badfo_close,
2532 };
2533
2534
2535 /*-------------------------------------------------------------------*/
2536
2537 /*
2538 * File Descriptor pseudo-device driver (/dev/fd/).
2539 *
2540 * Opening minor device N dup()s the file (if any) connected to file
2541 * descriptor N belonging to the calling process. Note that this driver
2542 * consists of only the ``open()'' routine, because all subsequent
2543 * references to this file will be direct to the other driver.
2544 *
2545 * XXX: we could give this one a cloning event handler if necessary.
2546 */
2547
2548 /* ARGSUSED */
2549 static int
2550 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2551 {
2552
2553 /*
2554 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2555 * the file descriptor being sought for duplication. The error
2556 * return ensures that the vnode for this device will be released
2557 * by vn_open. Open will detect this special error and take the
2558 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2559 * will simply report the error.
2560 */
2561 td->td_dupfd = dev2unit(dev);
2562 return (ENODEV);
2563 }
2564
2565 static struct cdevsw fildesc_cdevsw = {
2566 .d_version = D_VERSION,
2567 .d_flags = D_NEEDGIANT,
2568 .d_open = fdopen,
2569 .d_name = "FD",
2570 };
2571
2572 static void
2573 fildesc_drvinit(void *unused)
2574 {
2575 struct cdev *dev;
2576
2577 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2578 make_dev_alias(dev, "stdin");
2579 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2580 make_dev_alias(dev, "stdout");
2581 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2582 make_dev_alias(dev, "stderr");
2583 }
2584
2585 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
Cache object: 242ca7a59d89b015b2ec4fbdc8000f7d
|