1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/6.0/sys/kern/kern_descrip.c 151168 2005-10-09 08:13:56Z rik $");
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 openfiles; /* 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 * inaccessible 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 int vfslocked;
1145 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1146 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1147 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1148 VOP_UNLOCK(vp, 0, td);
1149 VFS_UNLOCK_GIANT(vfslocked);
1150 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1151 if (uap->name != _PC_PIPE_BUF) {
1152 error = EINVAL;
1153 } else {
1154 td->td_retval[0] = PIPE_BUF;
1155 error = 0;
1156 }
1157 } else {
1158 error = EOPNOTSUPP;
1159 }
1160 out:
1161 fdrop(fp, td);
1162 return (error);
1163 }
1164
1165 /*
1166 * Grow the file table to accomodate (at least) nfd descriptors. This may
1167 * block and drop the filedesc lock, but it will reacquire it before
1168 * returning.
1169 */
1170 static void
1171 fdgrowtable(struct filedesc *fdp, int nfd)
1172 {
1173 struct file **ntable;
1174 char *nfileflags;
1175 int nnfiles, onfiles;
1176 NDSLOTTYPE *nmap;
1177
1178 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1179
1180 KASSERT(fdp->fd_nfiles > 0,
1181 ("zero-length file table"));
1182
1183 /* compute the size of the new table */
1184 onfiles = fdp->fd_nfiles;
1185 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1186 if (nnfiles <= onfiles)
1187 /* the table is already large enough */
1188 return;
1189
1190 /* allocate a new table and (if required) new bitmaps */
1191 FILEDESC_UNLOCK(fdp);
1192 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1193 M_FILEDESC, M_ZERO | M_WAITOK);
1194 nfileflags = (char *)&ntable[nnfiles];
1195 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1196 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1197 M_FILEDESC, M_ZERO | M_WAITOK);
1198 else
1199 nmap = NULL;
1200 FILEDESC_LOCK(fdp);
1201
1202 /*
1203 * We now have new tables ready to go. Since we dropped the
1204 * filedesc lock to call malloc(), watch out for a race.
1205 */
1206 onfiles = fdp->fd_nfiles;
1207 if (onfiles >= nnfiles) {
1208 /* we lost the race, but that's OK */
1209 free(ntable, M_FILEDESC);
1210 if (nmap != NULL)
1211 free(nmap, M_FILEDESC);
1212 return;
1213 }
1214 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1215 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1216 if (onfiles > NDFILE)
1217 free(fdp->fd_ofiles, M_FILEDESC);
1218 fdp->fd_ofiles = ntable;
1219 fdp->fd_ofileflags = nfileflags;
1220 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1221 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1222 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1223 free(fdp->fd_map, M_FILEDESC);
1224 fdp->fd_map = nmap;
1225 }
1226 fdp->fd_nfiles = nnfiles;
1227 }
1228
1229 /*
1230 * Allocate a file descriptor for the process.
1231 */
1232 int
1233 fdalloc(struct thread *td, int minfd, int *result)
1234 {
1235 struct proc *p = td->td_proc;
1236 struct filedesc *fdp = p->p_fd;
1237 int fd = -1, maxfd;
1238
1239 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1240
1241 if (fdp->fd_freefile > minfd)
1242 minfd = fdp->fd_freefile;
1243
1244 PROC_LOCK(p);
1245 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1246 PROC_UNLOCK(p);
1247
1248 /*
1249 * Search the bitmap for a free descriptor. If none is found, try
1250 * to grow the file table. Keep at it until we either get a file
1251 * descriptor or run into process or system limits; fdgrowtable()
1252 * may drop the filedesc lock, so we're in a race.
1253 */
1254 for (;;) {
1255 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1256 if (fd >= maxfd)
1257 return (EMFILE);
1258 if (fd < fdp->fd_nfiles)
1259 break;
1260 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1261 }
1262
1263 /*
1264 * Perform some sanity checks, then mark the file descriptor as
1265 * used and return it to the caller.
1266 */
1267 KASSERT(!fdisused(fdp, fd),
1268 ("fd_first_free() returned non-free descriptor"));
1269 KASSERT(fdp->fd_ofiles[fd] == NULL,
1270 ("free descriptor isn't"));
1271 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1272 fdused(fdp, fd);
1273 *result = fd;
1274 return (0);
1275 }
1276
1277 /*
1278 * Check to see whether n user file descriptors
1279 * are available to the process p.
1280 */
1281 int
1282 fdavail(struct thread *td, int n)
1283 {
1284 struct proc *p = td->td_proc;
1285 struct filedesc *fdp = td->td_proc->p_fd;
1286 struct file **fpp;
1287 int i, lim, last;
1288
1289 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1290
1291 PROC_LOCK(p);
1292 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1293 PROC_UNLOCK(p);
1294 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1295 return (1);
1296 last = min(fdp->fd_nfiles, lim);
1297 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1298 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1299 if (*fpp == NULL && --n <= 0)
1300 return (1);
1301 }
1302 return (0);
1303 }
1304
1305 /*
1306 * Create a new open file structure and allocate
1307 * a file decriptor for the process that refers to it.
1308 * We add one reference to the file for the descriptor table
1309 * and one reference for resultfp. This is to prevent us being
1310 * preempted and the entry in the descriptor table closed after
1311 * we release the FILEDESC lock.
1312 */
1313 int
1314 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1315 {
1316 struct proc *p = td->td_proc;
1317 struct file *fp, *fq;
1318 int error, i;
1319 int maxuserfiles = maxfiles - (maxfiles / 20);
1320 static struct timeval lastfail;
1321 static int curfail;
1322
1323 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1324 sx_xlock(&filelist_lock);
1325 if ((openfiles >= maxuserfiles && (td->td_ucred->cr_ruid != 0 ||
1326 jailed(td->td_ucred))) || openfiles >= maxfiles) {
1327 if (ppsratecheck(&lastfail, &curfail, 1)) {
1328 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1329 td->td_ucred->cr_ruid);
1330 }
1331 sx_xunlock(&filelist_lock);
1332 uma_zfree(file_zone, fp);
1333 return (ENFILE);
1334 }
1335 openfiles++;
1336
1337 /*
1338 * If the process has file descriptor zero open, add the new file
1339 * descriptor to the list of open files at that point, otherwise
1340 * put it at the front of the list of open files.
1341 */
1342 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1343 fp->f_count = 1;
1344 if (resultfp)
1345 fp->f_count++;
1346 fp->f_cred = crhold(td->td_ucred);
1347 fp->f_ops = &badfileops;
1348 fp->f_data = NULL;
1349 fp->f_vnode = NULL;
1350 FILEDESC_LOCK(p->p_fd);
1351 if ((fq = p->p_fd->fd_ofiles[0])) {
1352 LIST_INSERT_AFTER(fq, fp, f_list);
1353 } else {
1354 LIST_INSERT_HEAD(&filehead, fp, f_list);
1355 }
1356 sx_xunlock(&filelist_lock);
1357 if ((error = fdalloc(td, 0, &i))) {
1358 FILEDESC_UNLOCK(p->p_fd);
1359 fdrop(fp, td);
1360 if (resultfp)
1361 fdrop(fp, td);
1362 return (error);
1363 }
1364 p->p_fd->fd_ofiles[i] = fp;
1365 FILEDESC_UNLOCK(p->p_fd);
1366 if (resultfp)
1367 *resultfp = fp;
1368 if (resultfd)
1369 *resultfd = i;
1370 return (0);
1371 }
1372
1373 /*
1374 * Build a new filedesc structure from another.
1375 * Copy the current, root, and jail root vnode references.
1376 */
1377 struct filedesc *
1378 fdinit(struct filedesc *fdp)
1379 {
1380 struct filedesc0 *newfdp;
1381
1382 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1383 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1384 if (fdp != NULL) {
1385 FILEDESC_LOCK(fdp);
1386 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1387 if (newfdp->fd_fd.fd_cdir)
1388 VREF(newfdp->fd_fd.fd_cdir);
1389 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1390 if (newfdp->fd_fd.fd_rdir)
1391 VREF(newfdp->fd_fd.fd_rdir);
1392 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1393 if (newfdp->fd_fd.fd_jdir)
1394 VREF(newfdp->fd_fd.fd_jdir);
1395 FILEDESC_UNLOCK(fdp);
1396 }
1397
1398 /* Create the file descriptor table. */
1399 newfdp->fd_fd.fd_refcnt = 1;
1400 newfdp->fd_fd.fd_holdcnt = 1;
1401 newfdp->fd_fd.fd_cmask = CMASK;
1402 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1403 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1404 newfdp->fd_fd.fd_nfiles = NDFILE;
1405 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
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 /* Certain daemons might not have file descriptors. */
1535 fdp = td->td_proc->p_fd;
1536 if (fdp == NULL)
1537 return;
1538
1539 /* Check for special need to clear POSIX style locks */
1540 fdtol = td->td_proc->p_fdtol;
1541 if (fdtol != NULL) {
1542 FILEDESC_LOCK(fdp);
1543 KASSERT(fdtol->fdl_refcount > 0,
1544 ("filedesc_to_refcount botch: fdl_refcount=%d",
1545 fdtol->fdl_refcount));
1546 if (fdtol->fdl_refcount == 1 &&
1547 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1548 for (i = 0, fpp = fdp->fd_ofiles;
1549 i <= fdp->fd_lastfile;
1550 i++, fpp++) {
1551 if (*fpp == NULL ||
1552 (*fpp)->f_type != DTYPE_VNODE)
1553 continue;
1554 fp = *fpp;
1555 fhold(fp);
1556 FILEDESC_UNLOCK(fdp);
1557 lf.l_whence = SEEK_SET;
1558 lf.l_start = 0;
1559 lf.l_len = 0;
1560 lf.l_type = F_UNLCK;
1561 vp = fp->f_vnode;
1562 VFS_ASSERT_GIANT(vp->v_mount);
1563 (void) VOP_ADVLOCK(vp,
1564 (caddr_t)td->td_proc->
1565 p_leader,
1566 F_UNLCK,
1567 &lf,
1568 F_POSIX);
1569 FILEDESC_LOCK(fdp);
1570 fdrop(fp, td);
1571 fpp = fdp->fd_ofiles + i;
1572 }
1573 }
1574 retry:
1575 if (fdtol->fdl_refcount == 1) {
1576 if (fdp->fd_holdleaderscount > 0 &&
1577 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1578 /*
1579 * close() or do_dup() has cleared a reference
1580 * in a shared file descriptor table.
1581 */
1582 fdp->fd_holdleaderswakeup = 1;
1583 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1584 PLOCK, "fdlhold", 0);
1585 goto retry;
1586 }
1587 if (fdtol->fdl_holdcount > 0) {
1588 /*
1589 * Ensure that fdtol->fdl_leader
1590 * remains valid in closef().
1591 */
1592 fdtol->fdl_wakeup = 1;
1593 msleep(fdtol, &fdp->fd_mtx,
1594 PLOCK, "fdlhold", 0);
1595 goto retry;
1596 }
1597 }
1598 fdtol->fdl_refcount--;
1599 if (fdtol->fdl_refcount == 0 &&
1600 fdtol->fdl_holdcount == 0) {
1601 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1602 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1603 } else
1604 fdtol = NULL;
1605 td->td_proc->p_fdtol = NULL;
1606 FILEDESC_UNLOCK(fdp);
1607 if (fdtol != NULL)
1608 FREE(fdtol, M_FILEDESC_TO_LEADER);
1609 }
1610 FILEDESC_LOCK_FAST(fdp);
1611 i = --fdp->fd_refcnt;
1612 FILEDESC_UNLOCK_FAST(fdp);
1613 if (i > 0)
1614 return;
1615 /*
1616 * We are the last reference to the structure, so we can
1617 * safely assume it will not change out from under us.
1618 */
1619 fpp = fdp->fd_ofiles;
1620 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1621 if (*fpp)
1622 (void) closef(*fpp, td);
1623 }
1624 FILEDESC_LOCK(fdp);
1625
1626 /* XXX This should happen earlier. */
1627 mtx_lock(&fdesc_mtx);
1628 td->td_proc->p_fd = NULL;
1629 mtx_unlock(&fdesc_mtx);
1630
1631 if (fdp->fd_nfiles > NDFILE)
1632 FREE(fdp->fd_ofiles, M_FILEDESC);
1633 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1634 FREE(fdp->fd_map, M_FILEDESC);
1635
1636 fdp->fd_nfiles = 0;
1637
1638 if (fdp->fd_cdir)
1639 vrele(fdp->fd_cdir);
1640 fdp->fd_cdir = NULL;
1641 if (fdp->fd_rdir)
1642 vrele(fdp->fd_rdir);
1643 fdp->fd_rdir = NULL;
1644 if (fdp->fd_jdir)
1645 vrele(fdp->fd_jdir);
1646 fdp->fd_jdir = NULL;
1647
1648 FILEDESC_UNLOCK(fdp);
1649
1650 fddrop(fdp);
1651 }
1652
1653 /*
1654 * For setugid programs, we don't want to people to use that setugidness
1655 * to generate error messages which write to a file which otherwise would
1656 * otherwise be off-limits to the process. We check for filesystems where
1657 * the vnode can change out from under us after execve (like [lin]procfs).
1658 *
1659 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1660 * sufficient. We also don't check for setugidness since we know we are.
1661 */
1662 static int
1663 is_unsafe(struct file *fp)
1664 {
1665 if (fp->f_type == DTYPE_VNODE) {
1666 struct vnode *vp = fp->f_vnode;
1667
1668 if ((vp->v_vflag & VV_PROCDEP) != 0)
1669 return (1);
1670 }
1671 return (0);
1672 }
1673
1674 /*
1675 * Make this setguid thing safe, if at all possible.
1676 */
1677 void
1678 setugidsafety(struct thread *td)
1679 {
1680 struct filedesc *fdp;
1681 int i;
1682
1683 /* Certain daemons might not have file descriptors. */
1684 fdp = td->td_proc->p_fd;
1685 if (fdp == NULL)
1686 return;
1687
1688 /*
1689 * Note: fdp->fd_ofiles may be reallocated out from under us while
1690 * we are blocked in a close. Be careful!
1691 */
1692 FILEDESC_LOCK(fdp);
1693 for (i = 0; i <= fdp->fd_lastfile; i++) {
1694 if (i > 2)
1695 break;
1696 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1697 struct file *fp;
1698
1699 knote_fdclose(td, i);
1700 /*
1701 * NULL-out descriptor prior to close to avoid
1702 * a race while close blocks.
1703 */
1704 fp = fdp->fd_ofiles[i];
1705 fdp->fd_ofiles[i] = NULL;
1706 fdp->fd_ofileflags[i] = 0;
1707 fdunused(fdp, i);
1708 FILEDESC_UNLOCK(fdp);
1709 (void) closef(fp, td);
1710 FILEDESC_LOCK(fdp);
1711 }
1712 }
1713 FILEDESC_UNLOCK(fdp);
1714 }
1715
1716 void
1717 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1718 {
1719
1720 FILEDESC_LOCK(fdp);
1721 if (fdp->fd_ofiles[idx] == fp) {
1722 fdp->fd_ofiles[idx] = NULL;
1723 fdunused(fdp, idx);
1724 FILEDESC_UNLOCK(fdp);
1725 fdrop(fp, td);
1726 } else {
1727 FILEDESC_UNLOCK(fdp);
1728 }
1729 }
1730
1731 /*
1732 * Close any files on exec?
1733 */
1734 void
1735 fdcloseexec(struct thread *td)
1736 {
1737 struct filedesc *fdp;
1738 int i;
1739
1740 /* Certain daemons might not have file descriptors. */
1741 fdp = td->td_proc->p_fd;
1742 if (fdp == NULL)
1743 return;
1744
1745 FILEDESC_LOCK(fdp);
1746
1747 /*
1748 * We cannot cache fd_ofiles or fd_ofileflags since operations
1749 * may block and rip them out from under us.
1750 */
1751 for (i = 0; i <= fdp->fd_lastfile; i++) {
1752 if (fdp->fd_ofiles[i] != NULL &&
1753 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1754 struct file *fp;
1755
1756 knote_fdclose(td, i);
1757 /*
1758 * NULL-out descriptor prior to close to avoid
1759 * a race while close blocks.
1760 */
1761 fp = fdp->fd_ofiles[i];
1762 fdp->fd_ofiles[i] = NULL;
1763 fdp->fd_ofileflags[i] = 0;
1764 fdunused(fdp, i);
1765 FILEDESC_UNLOCK(fdp);
1766 (void) closef(fp, td);
1767 FILEDESC_LOCK(fdp);
1768 }
1769 }
1770 FILEDESC_UNLOCK(fdp);
1771 }
1772
1773 /*
1774 * It is unsafe for set[ug]id processes to be started with file
1775 * descriptors 0..2 closed, as these descriptors are given implicit
1776 * significance in the Standard C library. fdcheckstd() will create a
1777 * descriptor referencing /dev/null for each of stdin, stdout, and
1778 * stderr that is not already open.
1779 */
1780 int
1781 fdcheckstd(struct thread *td)
1782 {
1783 struct nameidata nd;
1784 struct filedesc *fdp;
1785 struct file *fp;
1786 register_t retval;
1787 int fd, i, error, flags, devnull;
1788
1789 fdp = td->td_proc->p_fd;
1790 if (fdp == NULL)
1791 return (0);
1792 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1793 devnull = -1;
1794 error = 0;
1795 for (i = 0; i < 3; i++) {
1796 if (fdp->fd_ofiles[i] != NULL)
1797 continue;
1798 if (devnull < 0) {
1799 int vfslocked;
1800 error = falloc(td, &fp, &fd);
1801 if (error != 0)
1802 break;
1803 /* Note extra ref on `fp' held for us by falloc(). */
1804 KASSERT(fd == i, ("oof, we didn't get our fd"));
1805 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE,
1806 "/dev/null", td);
1807 flags = FREAD | FWRITE;
1808 error = vn_open(&nd, &flags, 0, fd);
1809 if (error != 0) {
1810 /*
1811 * Someone may have closed the entry in the
1812 * file descriptor table, so check it hasn't
1813 * changed before dropping the reference count.
1814 */
1815 FILEDESC_LOCK(fdp);
1816 KASSERT(fdp->fd_ofiles[fd] == fp,
1817 ("table not shared, how did it change?"));
1818 fdp->fd_ofiles[fd] = NULL;
1819 fdunused(fdp, fd);
1820 FILEDESC_UNLOCK(fdp);
1821 fdrop(fp, td);
1822 fdrop(fp, td);
1823 break;
1824 }
1825 vfslocked = NDHASGIANT(&nd);
1826 NDFREE(&nd, NDF_ONLY_PNBUF);
1827 fp->f_flag = flags;
1828 fp->f_vnode = nd.ni_vp;
1829 if (fp->f_data == NULL)
1830 fp->f_data = nd.ni_vp;
1831 if (fp->f_ops == &badfileops)
1832 fp->f_ops = &vnops;
1833 fp->f_type = DTYPE_VNODE;
1834 VOP_UNLOCK(nd.ni_vp, 0, td);
1835 VFS_UNLOCK_GIANT(vfslocked);
1836 devnull = fd;
1837 fdrop(fp, td);
1838 } else {
1839 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1840 if (error != 0)
1841 break;
1842 }
1843 }
1844 return (error);
1845 }
1846
1847 /*
1848 * Internal form of close.
1849 * Decrement reference count on file structure.
1850 * Note: td may be NULL when closing a file that was being passed in a
1851 * message.
1852 *
1853 * XXXRW: Giant is not required for the caller, but often will be held; this
1854 * makes it moderately likely the Giant will be recursed in the VFS case.
1855 */
1856 int
1857 closef(struct file *fp, struct thread *td)
1858 {
1859 struct vnode *vp;
1860 struct flock lf;
1861 struct filedesc_to_leader *fdtol;
1862 struct filedesc *fdp;
1863
1864 /*
1865 * POSIX record locking dictates that any close releases ALL
1866 * locks owned by this process. This is handled by setting
1867 * a flag in the unlock to free ONLY locks obeying POSIX
1868 * semantics, and not to free BSD-style file locks.
1869 * If the descriptor was in a message, POSIX-style locks
1870 * aren't passed with the descriptor.
1871 */
1872 if (fp->f_type == DTYPE_VNODE) {
1873 int vfslocked;
1874
1875 vp = fp->f_vnode;
1876 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1877 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1878 lf.l_whence = SEEK_SET;
1879 lf.l_start = 0;
1880 lf.l_len = 0;
1881 lf.l_type = F_UNLCK;
1882 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1883 F_UNLCK, &lf, F_POSIX);
1884 }
1885 fdtol = td->td_proc->p_fdtol;
1886 if (fdtol != NULL) {
1887 /*
1888 * Handle special case where file descriptor table
1889 * is shared between multiple process leaders.
1890 */
1891 fdp = td->td_proc->p_fd;
1892 FILEDESC_LOCK(fdp);
1893 for (fdtol = fdtol->fdl_next;
1894 fdtol != td->td_proc->p_fdtol;
1895 fdtol = fdtol->fdl_next) {
1896 if ((fdtol->fdl_leader->p_flag &
1897 P_ADVLOCK) == 0)
1898 continue;
1899 fdtol->fdl_holdcount++;
1900 FILEDESC_UNLOCK(fdp);
1901 lf.l_whence = SEEK_SET;
1902 lf.l_start = 0;
1903 lf.l_len = 0;
1904 lf.l_type = F_UNLCK;
1905 vp = fp->f_vnode;
1906 (void) VOP_ADVLOCK(vp,
1907 (caddr_t)fdtol->fdl_leader,
1908 F_UNLCK, &lf, F_POSIX);
1909 FILEDESC_LOCK(fdp);
1910 fdtol->fdl_holdcount--;
1911 if (fdtol->fdl_holdcount == 0 &&
1912 fdtol->fdl_wakeup != 0) {
1913 fdtol->fdl_wakeup = 0;
1914 wakeup(fdtol);
1915 }
1916 }
1917 FILEDESC_UNLOCK(fdp);
1918 }
1919 VFS_UNLOCK_GIANT(vfslocked);
1920 }
1921 return (fdrop(fp, td));
1922 }
1923
1924 /*
1925 * Extract the file pointer associated with the specified descriptor for
1926 * the current user process.
1927 *
1928 * If the descriptor doesn't exist, EBADF is returned.
1929 *
1930 * If the descriptor exists but doesn't match 'flags' then
1931 * return EBADF for read attempts and EINVAL for write attempts.
1932 *
1933 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1934 * It should be dropped with fdrop().
1935 * If it is not set, then the refcount will not be bumped however the
1936 * thread's filedesc struct will be returned locked (for fgetsock).
1937 *
1938 * If an error occured the non-zero error is returned and *fpp is set to NULL.
1939 * Otherwise *fpp is set and zero is returned.
1940 */
1941 static __inline int
1942 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1943 {
1944 struct filedesc *fdp;
1945 struct file *fp;
1946
1947 *fpp = NULL;
1948 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1949 return (EBADF);
1950 FILEDESC_LOCK(fdp);
1951 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
1952 FILEDESC_UNLOCK(fdp);
1953 return (EBADF);
1954 }
1955
1956 /*
1957 * Note: FREAD failure returns EBADF to maintain backwards
1958 * compatibility with what routines returned before.
1959 *
1960 * Only one flag, or 0, may be specified.
1961 */
1962 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
1963 FILEDESC_UNLOCK(fdp);
1964 return (EBADF);
1965 }
1966 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
1967 FILEDESC_UNLOCK(fdp);
1968 return (EINVAL);
1969 }
1970 if (hold) {
1971 fhold(fp);
1972 FILEDESC_UNLOCK(fdp);
1973 }
1974 *fpp = fp;
1975 return (0);
1976 }
1977
1978 int
1979 fget(struct thread *td, int fd, struct file **fpp)
1980 {
1981
1982 return(_fget(td, fd, fpp, 0, 1));
1983 }
1984
1985 int
1986 fget_read(struct thread *td, int fd, struct file **fpp)
1987 {
1988
1989 return(_fget(td, fd, fpp, FREAD, 1));
1990 }
1991
1992 int
1993 fget_write(struct thread *td, int fd, struct file **fpp)
1994 {
1995
1996 return(_fget(td, fd, fpp, FWRITE, 1));
1997 }
1998
1999 /*
2000 * Like fget() but loads the underlying vnode, or returns an error if
2001 * the descriptor does not represent a vnode. Note that pipes use vnodes
2002 * but never have VM objects. 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 *vpp = NULL;
2013 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2014 return (error);
2015 if (fp->f_vnode == NULL) {
2016 error = EINVAL;
2017 } else {
2018 *vpp = fp->f_vnode;
2019 vref(*vpp);
2020 }
2021 FILEDESC_UNLOCK(td->td_proc->p_fd);
2022 return (error);
2023 }
2024
2025 int
2026 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2027 {
2028
2029 return (_fgetvp(td, fd, vpp, 0));
2030 }
2031
2032 int
2033 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2034 {
2035
2036 return (_fgetvp(td, fd, vpp, FREAD));
2037 }
2038
2039 #ifdef notyet
2040 int
2041 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2042 {
2043
2044 return (_fgetvp(td, fd, vpp, FWRITE));
2045 }
2046 #endif
2047
2048 /*
2049 * Like fget() but loads the underlying socket, or returns an error if
2050 * the descriptor does not represent a socket.
2051 *
2052 * We bump the ref count on the returned socket. XXX Also obtain the SX
2053 * lock in the future.
2054 */
2055 int
2056 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2057 {
2058 struct file *fp;
2059 int error;
2060
2061 NET_ASSERT_GIANT();
2062
2063 *spp = NULL;
2064 if (fflagp != NULL)
2065 *fflagp = 0;
2066 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2067 return (error);
2068 if (fp->f_type != DTYPE_SOCKET) {
2069 error = ENOTSOCK;
2070 } else {
2071 *spp = fp->f_data;
2072 if (fflagp)
2073 *fflagp = fp->f_flag;
2074 SOCK_LOCK(*spp);
2075 soref(*spp);
2076 SOCK_UNLOCK(*spp);
2077 }
2078 FILEDESC_UNLOCK(td->td_proc->p_fd);
2079 return (error);
2080 }
2081
2082 /*
2083 * Drop the reference count on the socket and XXX release the SX lock in
2084 * the future. The last reference closes the socket.
2085 */
2086 void
2087 fputsock(struct socket *so)
2088 {
2089
2090 NET_ASSERT_GIANT();
2091 ACCEPT_LOCK();
2092 SOCK_LOCK(so);
2093 sorele(so);
2094 }
2095
2096 int
2097 fdrop(struct file *fp, struct thread *td)
2098 {
2099
2100 FILE_LOCK(fp);
2101 return (fdrop_locked(fp, td));
2102 }
2103
2104 /*
2105 * Drop reference on struct file passed in, may call closef if the
2106 * reference hits zero.
2107 * Expects struct file locked, and will unlock it.
2108 */
2109 static int
2110 fdrop_locked(struct file *fp, struct thread *td)
2111 {
2112 int error;
2113
2114 FILE_LOCK_ASSERT(fp, MA_OWNED);
2115
2116 if (--fp->f_count > 0) {
2117 FILE_UNLOCK(fp);
2118 return (0);
2119 }
2120 /* We have the last ref so we can proceed without the file lock. */
2121 FILE_UNLOCK(fp);
2122 if (fp->f_count < 0)
2123 panic("fdrop: count < 0");
2124 if (fp->f_ops != &badfileops)
2125 error = fo_close(fp, td);
2126 else
2127 error = 0;
2128
2129 sx_xlock(&filelist_lock);
2130 LIST_REMOVE(fp, f_list);
2131 openfiles--;
2132 sx_xunlock(&filelist_lock);
2133 crfree(fp->f_cred);
2134 uma_zfree(file_zone, fp);
2135
2136 return (error);
2137 }
2138
2139 /*
2140 * Apply an advisory lock on a file descriptor.
2141 *
2142 * Just attempt to get a record lock of the requested type on
2143 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2144 */
2145 #ifndef _SYS_SYSPROTO_H_
2146 struct flock_args {
2147 int fd;
2148 int how;
2149 };
2150 #endif
2151 /*
2152 * MPSAFE
2153 */
2154 /* ARGSUSED */
2155 int
2156 flock(struct thread *td, struct flock_args *uap)
2157 {
2158 struct file *fp;
2159 struct vnode *vp;
2160 struct flock lf;
2161 int error;
2162
2163 if ((error = fget(td, uap->fd, &fp)) != 0)
2164 return (error);
2165 if (fp->f_type != DTYPE_VNODE) {
2166 fdrop(fp, td);
2167 return (EOPNOTSUPP);
2168 }
2169
2170 mtx_lock(&Giant);
2171 vp = fp->f_vnode;
2172 lf.l_whence = SEEK_SET;
2173 lf.l_start = 0;
2174 lf.l_len = 0;
2175 if (uap->how & LOCK_UN) {
2176 lf.l_type = F_UNLCK;
2177 FILE_LOCK(fp);
2178 fp->f_flag &= ~FHASLOCK;
2179 FILE_UNLOCK(fp);
2180 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2181 goto done2;
2182 }
2183 if (uap->how & LOCK_EX)
2184 lf.l_type = F_WRLCK;
2185 else if (uap->how & LOCK_SH)
2186 lf.l_type = F_RDLCK;
2187 else {
2188 error = EBADF;
2189 goto done2;
2190 }
2191 FILE_LOCK(fp);
2192 fp->f_flag |= FHASLOCK;
2193 FILE_UNLOCK(fp);
2194 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2195 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2196 done2:
2197 fdrop(fp, td);
2198 mtx_unlock(&Giant);
2199 return (error);
2200 }
2201 /*
2202 * Duplicate the specified descriptor to a free descriptor.
2203 */
2204 int
2205 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2206 {
2207 struct file *wfp;
2208 struct file *fp;
2209
2210 /*
2211 * If the to-be-dup'd fd number is greater than the allowed number
2212 * of file descriptors, or the fd to be dup'd has already been
2213 * closed, then reject.
2214 */
2215 FILEDESC_LOCK(fdp);
2216 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2217 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2218 FILEDESC_UNLOCK(fdp);
2219 return (EBADF);
2220 }
2221
2222 /*
2223 * There are two cases of interest here.
2224 *
2225 * For ENODEV simply dup (dfd) to file descriptor
2226 * (indx) and return.
2227 *
2228 * For ENXIO steal away the file structure from (dfd) and
2229 * store it in (indx). (dfd) is effectively closed by
2230 * this operation.
2231 *
2232 * Any other error code is just returned.
2233 */
2234 switch (error) {
2235 case ENODEV:
2236 /*
2237 * Check that the mode the file is being opened for is a
2238 * subset of the mode of the existing descriptor.
2239 */
2240 FILE_LOCK(wfp);
2241 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2242 FILE_UNLOCK(wfp);
2243 FILEDESC_UNLOCK(fdp);
2244 return (EACCES);
2245 }
2246 fp = fdp->fd_ofiles[indx];
2247 fdp->fd_ofiles[indx] = wfp;
2248 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2249 if (fp == NULL)
2250 fdused(fdp, indx);
2251 fhold_locked(wfp);
2252 FILE_UNLOCK(wfp);
2253 FILEDESC_UNLOCK(fdp);
2254 if (fp != NULL) {
2255 /*
2256 * We now own the reference to fp that the ofiles[]
2257 * array used to own. Release it.
2258 */
2259 FILE_LOCK(fp);
2260 fdrop_locked(fp, td);
2261 }
2262 return (0);
2263
2264 case ENXIO:
2265 /*
2266 * Steal away the file pointer from dfd and stuff it into indx.
2267 */
2268 fp = fdp->fd_ofiles[indx];
2269 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2270 fdp->fd_ofiles[dfd] = NULL;
2271 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2272 fdp->fd_ofileflags[dfd] = 0;
2273 fdunused(fdp, dfd);
2274 if (fp == NULL)
2275 fdused(fdp, indx);
2276 if (fp != NULL)
2277 FILE_LOCK(fp);
2278
2279 /*
2280 * We now own the reference to fp that the ofiles[] array
2281 * used to own. Release it.
2282 */
2283 if (fp != NULL)
2284 fdrop_locked(fp, td);
2285
2286 FILEDESC_UNLOCK(fdp);
2287
2288 return (0);
2289
2290 default:
2291 FILEDESC_UNLOCK(fdp);
2292 return (error);
2293 }
2294 /* NOTREACHED */
2295 }
2296
2297 /*
2298 * Scan all active processes to see if any of them have a current
2299 * or root directory of `olddp'. If so, replace them with the new
2300 * mount point.
2301 */
2302 void
2303 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2304 {
2305 struct filedesc *fdp;
2306 struct proc *p;
2307 int nrele;
2308
2309 if (vrefcnt(olddp) == 1)
2310 return;
2311 sx_slock(&allproc_lock);
2312 LIST_FOREACH(p, &allproc, p_list) {
2313 fdp = fdhold(p);
2314 if (fdp == NULL)
2315 continue;
2316 nrele = 0;
2317 FILEDESC_LOCK_FAST(fdp);
2318 if (fdp->fd_cdir == olddp) {
2319 vref(newdp);
2320 fdp->fd_cdir = newdp;
2321 nrele++;
2322 }
2323 if (fdp->fd_rdir == olddp) {
2324 vref(newdp);
2325 fdp->fd_rdir = newdp;
2326 nrele++;
2327 }
2328 FILEDESC_UNLOCK_FAST(fdp);
2329 fddrop(fdp);
2330 while (nrele--)
2331 vrele(olddp);
2332 }
2333 sx_sunlock(&allproc_lock);
2334 if (rootvnode == olddp) {
2335 vrele(rootvnode);
2336 vref(newdp);
2337 rootvnode = newdp;
2338 }
2339 }
2340
2341 struct filedesc_to_leader *
2342 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2343 {
2344 struct filedesc_to_leader *fdtol;
2345
2346 MALLOC(fdtol, struct filedesc_to_leader *,
2347 sizeof(struct filedesc_to_leader),
2348 M_FILEDESC_TO_LEADER,
2349 M_WAITOK);
2350 fdtol->fdl_refcount = 1;
2351 fdtol->fdl_holdcount = 0;
2352 fdtol->fdl_wakeup = 0;
2353 fdtol->fdl_leader = leader;
2354 if (old != NULL) {
2355 FILEDESC_LOCK(fdp);
2356 fdtol->fdl_next = old->fdl_next;
2357 fdtol->fdl_prev = old;
2358 old->fdl_next = fdtol;
2359 fdtol->fdl_next->fdl_prev = fdtol;
2360 FILEDESC_UNLOCK(fdp);
2361 } else {
2362 fdtol->fdl_next = fdtol;
2363 fdtol->fdl_prev = fdtol;
2364 }
2365 return (fdtol);
2366 }
2367
2368 /*
2369 * Get file structures.
2370 */
2371 static int
2372 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2373 {
2374 struct xfile xf;
2375 struct filedesc *fdp;
2376 struct file *fp;
2377 struct proc *p;
2378 int error, n;
2379
2380 /*
2381 * Note: because the number of file descriptors is calculated
2382 * in different ways for sizing vs returning the data,
2383 * there is information leakage from the first loop. However,
2384 * it is of a similar order of magnitude to the leakage from
2385 * global system statistics such as kern.openfiles.
2386 */
2387 error = sysctl_wire_old_buffer(req, 0);
2388 if (error != 0)
2389 return (error);
2390 if (req->oldptr == NULL) {
2391 n = 16; /* A slight overestimate. */
2392 sx_slock(&filelist_lock);
2393 LIST_FOREACH(fp, &filehead, f_list) {
2394 /*
2395 * We should grab the lock, but this is an
2396 * estimate, so does it really matter?
2397 */
2398 /* mtx_lock(fp->f_mtxp); */
2399 n += fp->f_count;
2400 /* mtx_unlock(f->f_mtxp); */
2401 }
2402 sx_sunlock(&filelist_lock);
2403 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2404 }
2405 error = 0;
2406 bzero(&xf, sizeof(xf));
2407 xf.xf_size = sizeof(xf);
2408 sx_slock(&allproc_lock);
2409 LIST_FOREACH(p, &allproc, p_list) {
2410 if (p->p_state == PRS_NEW)
2411 continue;
2412 PROC_LOCK(p);
2413 if (p_cansee(req->td, p) != 0) {
2414 PROC_UNLOCK(p);
2415 continue;
2416 }
2417 xf.xf_pid = p->p_pid;
2418 xf.xf_uid = p->p_ucred->cr_uid;
2419 PROC_UNLOCK(p);
2420 fdp = fdhold(p);
2421 if (fdp == NULL)
2422 continue;
2423 FILEDESC_LOCK_FAST(fdp);
2424 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2425 if ((fp = fdp->fd_ofiles[n]) == NULL)
2426 continue;
2427 xf.xf_fd = n;
2428 xf.xf_file = fp;
2429 xf.xf_data = fp->f_data;
2430 xf.xf_vnode = fp->f_vnode;
2431 xf.xf_type = fp->f_type;
2432 xf.xf_count = fp->f_count;
2433 xf.xf_msgcount = fp->f_msgcount;
2434 xf.xf_offset = fp->f_offset;
2435 xf.xf_flag = fp->f_flag;
2436 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2437 if (error)
2438 break;
2439 }
2440 FILEDESC_UNLOCK_FAST(fdp);
2441 fddrop(fdp);
2442 if (error)
2443 break;
2444 }
2445 sx_sunlock(&allproc_lock);
2446 return (error);
2447 }
2448
2449 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2450 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2451
2452 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2453 &maxfilesperproc, 0, "Maximum files allowed open per process");
2454
2455 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2456 &maxfiles, 0, "Maximum number of files");
2457
2458 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2459 &openfiles, 0, "System-wide number of open files");
2460
2461 /* ARGSUSED*/
2462 static void
2463 filelistinit(void *dummy)
2464 {
2465
2466 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2467 NULL, NULL, UMA_ALIGN_PTR, 0);
2468 sx_init(&filelist_lock, "filelist lock");
2469 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2470 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2471 }
2472 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2473
2474 /*-------------------------------------------------------------------*/
2475
2476 static int
2477 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2478 {
2479
2480 return (EBADF);
2481 }
2482
2483 static int
2484 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2485 {
2486
2487 return (EBADF);
2488 }
2489
2490 static int
2491 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2492 {
2493
2494 return (0);
2495 }
2496
2497 static int
2498 badfo_kqfilter(struct file *fp, struct knote *kn)
2499 {
2500
2501 return (0);
2502 }
2503
2504 static int
2505 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2506 {
2507
2508 return (EBADF);
2509 }
2510
2511 static int
2512 badfo_close(struct file *fp, struct thread *td)
2513 {
2514
2515 return (EBADF);
2516 }
2517
2518 struct fileops badfileops = {
2519 .fo_read = badfo_readwrite,
2520 .fo_write = badfo_readwrite,
2521 .fo_ioctl = badfo_ioctl,
2522 .fo_poll = badfo_poll,
2523 .fo_kqfilter = badfo_kqfilter,
2524 .fo_stat = badfo_stat,
2525 .fo_close = badfo_close,
2526 };
2527
2528
2529 /*-------------------------------------------------------------------*/
2530
2531 /*
2532 * File Descriptor pseudo-device driver (/dev/fd/).
2533 *
2534 * Opening minor device N dup()s the file (if any) connected to file
2535 * descriptor N belonging to the calling process. Note that this driver
2536 * consists of only the ``open()'' routine, because all subsequent
2537 * references to this file will be direct to the other driver.
2538 *
2539 * XXX: we could give this one a cloning event handler if necessary.
2540 */
2541
2542 /* ARGSUSED */
2543 static int
2544 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2545 {
2546
2547 /*
2548 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2549 * the file descriptor being sought for duplication. The error
2550 * return ensures that the vnode for this device will be released
2551 * by vn_open. Open will detect this special error and take the
2552 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2553 * will simply report the error.
2554 */
2555 td->td_dupfd = dev2unit(dev);
2556 return (ENODEV);
2557 }
2558
2559 static struct cdevsw fildesc_cdevsw = {
2560 .d_version = D_VERSION,
2561 .d_flags = D_NEEDGIANT,
2562 .d_open = fdopen,
2563 .d_name = "FD",
2564 };
2565
2566 static void
2567 fildesc_drvinit(void *unused)
2568 {
2569 struct cdev *dev;
2570
2571 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2572 make_dev_alias(dev, "stdin");
2573 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2574 make_dev_alias(dev, "stdout");
2575 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2576 make_dev_alias(dev, "stderr");
2577 }
2578
2579 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
Cache object: 49d3e58a6665f7a123ae0b083fbb7804
|