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/7.3/sys/kern/kern_descrip.c 195694 2009-07-14 19:48:31Z jhb $");
39
40 #include "opt_compat.h"
41 #include "opt_ddb.h"
42 #include "opt_ktrace.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46
47 #include <sys/conf.h>
48 #include <sys/domain.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/filedesc.h>
52 #include <sys/filio.h>
53 #include <sys/jail.h>
54 #include <sys/kernel.h>
55 #include <sys/limits.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/mqueue.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/priv.h>
63 #include <sys/proc.h>
64 #include <sys/protosw.h>
65 #include <sys/resourcevar.h>
66 #include <sys/signalvar.h>
67 #include <sys/socketvar.h>
68 #include <sys/stat.h>
69 #include <sys/sx.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/sysctl.h>
72 #include <sys/sysproto.h>
73 #include <sys/unistd.h>
74 #include <sys/user.h>
75 #include <sys/vnode.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79
80 #include <security/audit/audit.h>
81
82 #include <vm/uma.h>
83
84 #include <ddb/ddb.h>
85
86 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
87 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
88 "file desc to leader structures");
89 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
90
91 static uma_zone_t file_zone;
92
93
94 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
95 enum dup_type { DUP_VARIABLE, DUP_FIXED };
96
97 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
98 register_t *retval);
99 static int fd_first_free(struct filedesc *, int, int);
100 static int fd_last_used(struct filedesc *, int, int);
101 static void fdgrowtable(struct filedesc *, int);
102 static int fdrop_locked(struct file *fp, struct thread *td);
103 static void fdunused(struct filedesc *fdp, int fd);
104 static void fdused(struct filedesc *fdp, int fd);
105
106 /*
107 * A process is initially started out with NDFILE descriptors stored within
108 * this structure, selected to be enough for typical applications based on
109 * the historical limit of 20 open files (and the usage of descriptors by
110 * shells). If these descriptors are exhausted, a larger descriptor table
111 * may be allocated, up to a process' resource limit; the internal arrays
112 * are then unused.
113 */
114 #define NDFILE 20
115 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
116 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
117 #define NDSLOT(x) ((x) / NDENTRIES)
118 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
119 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
120
121 /*
122 * Storage required per open file descriptor.
123 */
124 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
125
126 /*
127 * Basic allocation of descriptors:
128 * one of the above, plus arrays for NDFILE descriptors.
129 */
130 struct filedesc0 {
131 struct filedesc fd_fd;
132 /*
133 * These arrays are used when the number of open files is
134 * <= NDFILE, and are then pointed to by the pointers above.
135 */
136 struct file *fd_dfiles[NDFILE];
137 char fd_dfileflags[NDFILE];
138 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
139 };
140
141 /*
142 * Descriptor management.
143 */
144 struct filelist filehead; /* head of list of open files */
145 int openfiles; /* actual number of open files */
146 struct sx filelist_lock; /* sx to protect filelist */
147 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
148 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
149
150 /* A mutex to protect the association between a proc and filedesc. */
151 static struct mtx fdesc_mtx;
152
153 /*
154 * Find the first zero bit in the given bitmap, starting at low and not
155 * exceeding size - 1.
156 */
157 static int
158 fd_first_free(struct filedesc *fdp, int low, int size)
159 {
160 NDSLOTTYPE *map = fdp->fd_map;
161 NDSLOTTYPE mask;
162 int off, maxoff;
163
164 if (low >= size)
165 return (low);
166
167 off = NDSLOT(low);
168 if (low % NDENTRIES) {
169 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
170 if ((mask &= ~map[off]) != 0UL)
171 return (off * NDENTRIES + ffsl(mask) - 1);
172 ++off;
173 }
174 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
175 if (map[off] != ~0UL)
176 return (off * NDENTRIES + ffsl(~map[off]) - 1);
177 return (size);
178 }
179
180 /*
181 * Find the highest non-zero bit in the given bitmap, starting at low and
182 * not exceeding size - 1.
183 */
184 static int
185 fd_last_used(struct filedesc *fdp, int low, int size)
186 {
187 NDSLOTTYPE *map = fdp->fd_map;
188 NDSLOTTYPE mask;
189 int off, minoff;
190
191 if (low >= size)
192 return (-1);
193
194 off = NDSLOT(size);
195 if (size % NDENTRIES) {
196 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
197 if ((mask &= map[off]) != 0)
198 return (off * NDENTRIES + flsl(mask) - 1);
199 --off;
200 }
201 for (minoff = NDSLOT(low); off >= minoff; --off)
202 if (map[off] != 0)
203 return (off * NDENTRIES + flsl(map[off]) - 1);
204 return (low - 1);
205 }
206
207 static int
208 fdisused(struct filedesc *fdp, int fd)
209 {
210 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
211 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
212 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
213 }
214
215 /*
216 * Mark a file descriptor as used.
217 */
218 static void
219 fdused(struct filedesc *fdp, int fd)
220 {
221
222 FILEDESC_XLOCK_ASSERT(fdp);
223 KASSERT(!fdisused(fdp, fd),
224 ("fd already used"));
225
226 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
227 if (fd > fdp->fd_lastfile)
228 fdp->fd_lastfile = fd;
229 if (fd == fdp->fd_freefile)
230 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
231 }
232
233 /*
234 * Mark a file descriptor as unused.
235 */
236 static void
237 fdunused(struct filedesc *fdp, int fd)
238 {
239
240 FILEDESC_XLOCK_ASSERT(fdp);
241 KASSERT(fdisused(fdp, fd),
242 ("fd is already unused"));
243 KASSERT(fdp->fd_ofiles[fd] == NULL,
244 ("fd is still in use"));
245
246 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
247 if (fd < fdp->fd_freefile)
248 fdp->fd_freefile = fd;
249 if (fd == fdp->fd_lastfile)
250 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
251 }
252
253 /*
254 * System calls on descriptors.
255 */
256 #ifndef _SYS_SYSPROTO_H_
257 struct getdtablesize_args {
258 int dummy;
259 };
260 #endif
261 /* ARGSUSED */
262 int
263 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
264 {
265 struct proc *p = td->td_proc;
266
267 PROC_LOCK(p);
268 td->td_retval[0] =
269 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
270 PROC_UNLOCK(p);
271 return (0);
272 }
273
274 /*
275 * Duplicate a file descriptor to a particular value.
276 *
277 * Note: keep in mind that a potential race condition exists when closing
278 * descriptors from a shared descriptor table (via rfork).
279 */
280 #ifndef _SYS_SYSPROTO_H_
281 struct dup2_args {
282 u_int from;
283 u_int to;
284 };
285 #endif
286 /* ARGSUSED */
287 int
288 dup2(struct thread *td, struct dup2_args *uap)
289 {
290
291 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
292 td->td_retval));
293 }
294
295 /*
296 * Duplicate a file descriptor.
297 */
298 #ifndef _SYS_SYSPROTO_H_
299 struct dup_args {
300 u_int fd;
301 };
302 #endif
303 /* ARGSUSED */
304 int
305 dup(struct thread *td, struct dup_args *uap)
306 {
307
308 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
309 }
310
311 /*
312 * The file control system call.
313 */
314 #ifndef _SYS_SYSPROTO_H_
315 struct fcntl_args {
316 int fd;
317 int cmd;
318 long arg;
319 };
320 #endif
321 /* ARGSUSED */
322 int
323 fcntl(struct thread *td, struct fcntl_args *uap)
324 {
325 struct flock fl;
326 struct oflock ofl;
327 intptr_t arg;
328 int error;
329 int cmd;
330
331 error = 0;
332 cmd = uap->cmd;
333 switch (uap->cmd) {
334 case F_OGETLK:
335 case F_OSETLK:
336 case F_OSETLKW:
337 /*
338 * Convert old flock structure to new.
339 */
340 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
341 fl.l_start = ofl.l_start;
342 fl.l_len = ofl.l_len;
343 fl.l_pid = ofl.l_pid;
344 fl.l_type = ofl.l_type;
345 fl.l_whence = ofl.l_whence;
346 fl.l_sysid = 0;
347
348 switch (uap->cmd) {
349 case F_OGETLK:
350 cmd = F_GETLK;
351 break;
352 case F_OSETLK:
353 cmd = F_SETLK;
354 break;
355 case F_OSETLKW:
356 cmd = F_SETLKW;
357 break;
358 }
359 arg = (intptr_t)&fl;
360 break;
361 case F_GETLK:
362 case F_SETLK:
363 case F_SETLKW:
364 case F_SETLK_REMOTE:
365 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
366 arg = (intptr_t)&fl;
367 break;
368 default:
369 arg = uap->arg;
370 break;
371 }
372 if (error)
373 return (error);
374 error = kern_fcntl(td, uap->fd, cmd, arg);
375 if (error)
376 return (error);
377 if (uap->cmd == F_OGETLK) {
378 ofl.l_start = fl.l_start;
379 ofl.l_len = fl.l_len;
380 ofl.l_pid = fl.l_pid;
381 ofl.l_type = fl.l_type;
382 ofl.l_whence = fl.l_whence;
383 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
384 } else if (uap->cmd == F_GETLK) {
385 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
386 }
387 return (error);
388 }
389
390 static inline struct file *
391 fdtofp(int fd, struct filedesc *fdp)
392 {
393 struct file *fp;
394
395 FILEDESC_LOCK_ASSERT(fdp);
396 if ((unsigned)fd >= fdp->fd_nfiles ||
397 (fp = fdp->fd_ofiles[fd]) == NULL)
398 return (NULL);
399 return (fp);
400 }
401
402 int
403 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
404 {
405 struct filedesc *fdp;
406 struct flock *flp;
407 struct file *fp;
408 struct proc *p;
409 char *pop;
410 struct vnode *vp;
411 u_int newmin;
412 int error, flg, tmp;
413 int vfslocked;
414
415 vfslocked = 0;
416 error = 0;
417 flg = F_POSIX;
418 p = td->td_proc;
419 fdp = p->p_fd;
420
421 switch (cmd) {
422 case F_DUPFD:
423 FILEDESC_SLOCK(fdp);
424 if ((fp = fdtofp(fd, fdp)) == NULL) {
425 FILEDESC_SUNLOCK(fdp);
426 error = EBADF;
427 break;
428 }
429 FILEDESC_SUNLOCK(fdp);
430 newmin = arg;
431 PROC_LOCK(p);
432 if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
433 newmin >= maxfilesperproc) {
434 PROC_UNLOCK(p);
435 error = EINVAL;
436 break;
437 }
438 PROC_UNLOCK(p);
439 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
440 break;
441
442 case F_DUP2FD:
443 tmp = arg;
444 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
445 break;
446
447 case F_GETFD:
448 FILEDESC_SLOCK(fdp);
449 if ((fp = fdtofp(fd, fdp)) == NULL) {
450 FILEDESC_SUNLOCK(fdp);
451 error = EBADF;
452 break;
453 }
454 pop = &fdp->fd_ofileflags[fd];
455 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
456 FILEDESC_SUNLOCK(fdp);
457 break;
458
459 case F_SETFD:
460 FILEDESC_XLOCK(fdp);
461 if ((fp = fdtofp(fd, fdp)) == NULL) {
462 FILEDESC_XUNLOCK(fdp);
463 error = EBADF;
464 break;
465 }
466 pop = &fdp->fd_ofileflags[fd];
467 *pop = (*pop &~ UF_EXCLOSE) |
468 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
469 FILEDESC_XUNLOCK(fdp);
470 break;
471
472 case F_GETFL:
473 FILEDESC_SLOCK(fdp);
474 if ((fp = fdtofp(fd, fdp)) == NULL) {
475 FILEDESC_SUNLOCK(fdp);
476 error = EBADF;
477 break;
478 }
479 FILE_LOCK(fp);
480 td->td_retval[0] = OFLAGS(fp->f_flag);
481 FILE_UNLOCK(fp);
482 FILEDESC_SUNLOCK(fdp);
483 break;
484
485 case F_SETFL:
486 FILEDESC_SLOCK(fdp);
487 if ((fp = fdtofp(fd, fdp)) == NULL) {
488 FILEDESC_SUNLOCK(fdp);
489 error = EBADF;
490 break;
491 }
492 FILE_LOCK(fp);
493 fhold_locked(fp);
494 fp->f_flag &= ~FCNTLFLAGS;
495 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
496 FILE_UNLOCK(fp);
497 FILEDESC_SUNLOCK(fdp);
498 tmp = fp->f_flag & FNONBLOCK;
499 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
500 if (error) {
501 fdrop(fp, td);
502 break;
503 }
504 tmp = fp->f_flag & FASYNC;
505 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
506 if (error == 0) {
507 fdrop(fp, td);
508 break;
509 }
510 FILE_LOCK(fp);
511 fp->f_flag &= ~FNONBLOCK;
512 FILE_UNLOCK(fp);
513 tmp = 0;
514 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
515 fdrop(fp, td);
516 break;
517
518 case F_GETOWN:
519 FILEDESC_SLOCK(fdp);
520 if ((fp = fdtofp(fd, fdp)) == NULL) {
521 FILEDESC_SUNLOCK(fdp);
522 error = EBADF;
523 break;
524 }
525 fhold(fp);
526 FILEDESC_SUNLOCK(fdp);
527 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
528 if (error == 0)
529 td->td_retval[0] = tmp;
530 fdrop(fp, td);
531 break;
532
533 case F_SETOWN:
534 FILEDESC_SLOCK(fdp);
535 if ((fp = fdtofp(fd, fdp)) == NULL) {
536 FILEDESC_SUNLOCK(fdp);
537 error = EBADF;
538 break;
539 }
540 fhold(fp);
541 FILEDESC_SUNLOCK(fdp);
542 tmp = arg;
543 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
544 fdrop(fp, td);
545 break;
546
547 case F_SETLK_REMOTE:
548 error = priv_check(td, PRIV_NFS_LOCKD);
549 if (error)
550 return (error);
551 flg = F_REMOTE;
552 goto do_setlk;
553
554 case F_SETLKW:
555 flg |= F_WAIT;
556 /* FALLTHROUGH F_SETLK */
557
558 case F_SETLK:
559 do_setlk:
560 FILEDESC_SLOCK(fdp);
561 if ((fp = fdtofp(fd, fdp)) == NULL) {
562 FILEDESC_SUNLOCK(fdp);
563 error = EBADF;
564 break;
565 }
566 if (fp->f_type != DTYPE_VNODE) {
567 FILEDESC_SUNLOCK(fdp);
568 error = EBADF;
569 break;
570 }
571 flp = (struct flock *)arg;
572 if (flp->l_whence == SEEK_CUR) {
573 if (fp->f_offset < 0 ||
574 (flp->l_start > 0 &&
575 fp->f_offset > OFF_MAX - flp->l_start)) {
576 FILEDESC_SUNLOCK(fdp);
577 error = EOVERFLOW;
578 break;
579 }
580 flp->l_start += fp->f_offset;
581 }
582
583 /*
584 * VOP_ADVLOCK() may block.
585 */
586 fhold(fp);
587 FILEDESC_SUNLOCK(fdp);
588 vp = fp->f_vnode;
589 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
590 switch (flp->l_type) {
591 case F_RDLCK:
592 if ((fp->f_flag & FREAD) == 0) {
593 error = EBADF;
594 break;
595 }
596 PROC_LOCK(p->p_leader);
597 p->p_leader->p_flag |= P_ADVLOCK;
598 PROC_UNLOCK(p->p_leader);
599 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
600 flp, flg);
601 break;
602 case F_WRLCK:
603 if ((fp->f_flag & FWRITE) == 0) {
604 error = EBADF;
605 break;
606 }
607 PROC_LOCK(p->p_leader);
608 p->p_leader->p_flag |= P_ADVLOCK;
609 PROC_UNLOCK(p->p_leader);
610 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
611 flp, flg);
612 break;
613 case F_UNLCK:
614 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
615 flp, flg);
616 break;
617 case F_UNLCKSYS:
618 /*
619 * Temporary api for testing remote lock
620 * infrastructure.
621 */
622 if (flg != F_REMOTE) {
623 error = EINVAL;
624 break;
625 }
626 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
627 F_UNLCKSYS, flp, flg);
628 break;
629 default:
630 error = EINVAL;
631 break;
632 }
633 VFS_UNLOCK_GIANT(vfslocked);
634 vfslocked = 0;
635 /* Check for race with close */
636 FILEDESC_SLOCK(fdp);
637 if ((unsigned) fd >= fdp->fd_nfiles ||
638 fp != fdp->fd_ofiles[fd]) {
639 FILEDESC_SUNLOCK(fdp);
640 flp->l_whence = SEEK_SET;
641 flp->l_start = 0;
642 flp->l_len = 0;
643 flp->l_type = F_UNLCK;
644 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
645 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
646 F_UNLCK, flp, F_POSIX);
647 VFS_UNLOCK_GIANT(vfslocked);
648 vfslocked = 0;
649 } else
650 FILEDESC_SUNLOCK(fdp);
651 fdrop(fp, td);
652 break;
653
654 case F_GETLK:
655 FILEDESC_SLOCK(fdp);
656 if ((fp = fdtofp(fd, fdp)) == NULL) {
657 FILEDESC_SUNLOCK(fdp);
658 error = EBADF;
659 break;
660 }
661 if (fp->f_type != DTYPE_VNODE) {
662 FILEDESC_SUNLOCK(fdp);
663 error = EBADF;
664 break;
665 }
666 flp = (struct flock *)arg;
667 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
668 flp->l_type != F_UNLCK) {
669 FILEDESC_SUNLOCK(fdp);
670 error = EINVAL;
671 break;
672 }
673 if (flp->l_whence == SEEK_CUR) {
674 if ((flp->l_start > 0 &&
675 fp->f_offset > OFF_MAX - flp->l_start) ||
676 (flp->l_start < 0 &&
677 fp->f_offset < OFF_MIN - flp->l_start)) {
678 FILEDESC_SUNLOCK(fdp);
679 error = EOVERFLOW;
680 break;
681 }
682 flp->l_start += fp->f_offset;
683 }
684 /*
685 * VOP_ADVLOCK() may block.
686 */
687 fhold(fp);
688 FILEDESC_SUNLOCK(fdp);
689 vp = fp->f_vnode;
690 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
691 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
692 F_POSIX);
693 VFS_UNLOCK_GIANT(vfslocked);
694 vfslocked = 0;
695 fdrop(fp, td);
696 break;
697 default:
698 error = EINVAL;
699 break;
700 }
701 VFS_UNLOCK_GIANT(vfslocked);
702 return (error);
703 }
704
705 /*
706 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
707 */
708 static int
709 do_dup(struct thread *td, enum dup_type type, int old, int new,
710 register_t *retval)
711 {
712 struct filedesc *fdp;
713 struct proc *p;
714 struct file *fp;
715 struct file *delfp;
716 int error, holdleaders, maxfd;
717
718 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
719 ("invalid dup type %d", type));
720
721 p = td->td_proc;
722 fdp = p->p_fd;
723
724 /*
725 * Verify we have a valid descriptor to dup from and possibly to
726 * dup to.
727 */
728 if (old < 0 || new < 0)
729 return (EBADF);
730 PROC_LOCK(p);
731 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
732 PROC_UNLOCK(p);
733 if (new >= maxfd)
734 return (EMFILE);
735
736 FILEDESC_XLOCK(fdp);
737 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
738 FILEDESC_XUNLOCK(fdp);
739 return (EBADF);
740 }
741 if (type == DUP_FIXED && old == new) {
742 *retval = new;
743 FILEDESC_XUNLOCK(fdp);
744 return (0);
745 }
746 fp = fdp->fd_ofiles[old];
747 fhold(fp);
748
749 /*
750 * If the caller specified a file descriptor, make sure the file
751 * table is large enough to hold it, and grab it. Otherwise, just
752 * allocate a new descriptor the usual way. Since the filedesc
753 * lock may be temporarily dropped in the process, we have to look
754 * out for a race.
755 */
756 if (type == DUP_FIXED) {
757 if (new >= fdp->fd_nfiles)
758 fdgrowtable(fdp, new + 1);
759 if (fdp->fd_ofiles[new] == NULL)
760 fdused(fdp, new);
761 } else {
762 if ((error = fdalloc(td, new, &new)) != 0) {
763 FILEDESC_XUNLOCK(fdp);
764 fdrop(fp, td);
765 return (error);
766 }
767 }
768
769 /*
770 * If the old file changed out from under us then treat it as a
771 * bad file descriptor. Userland should do its own locking to
772 * avoid this case.
773 */
774 if (fdp->fd_ofiles[old] != fp) {
775 /* we've allocated a descriptor which we won't use */
776 if (fdp->fd_ofiles[new] == NULL)
777 fdunused(fdp, new);
778 FILEDESC_XUNLOCK(fdp);
779 fdrop(fp, td);
780 return (EBADF);
781 }
782 KASSERT(old != new,
783 ("new fd is same as old"));
784
785 /*
786 * Save info on the descriptor being overwritten. We cannot close
787 * it without introducing an ownership race for the slot, since we
788 * need to drop the filedesc lock to call closef().
789 *
790 * XXX this duplicates parts of close().
791 */
792 delfp = fdp->fd_ofiles[new];
793 holdleaders = 0;
794 if (delfp != NULL) {
795 if (td->td_proc->p_fdtol != NULL) {
796 /*
797 * Ask fdfree() to sleep to ensure that all relevant
798 * process leaders can be traversed in closef().
799 */
800 fdp->fd_holdleaderscount++;
801 holdleaders = 1;
802 }
803 }
804
805 /*
806 * Duplicate the source descriptor
807 */
808 fdp->fd_ofiles[new] = fp;
809 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
810 if (new > fdp->fd_lastfile)
811 fdp->fd_lastfile = new;
812 *retval = new;
813
814 /*
815 * If we dup'd over a valid file, we now own the reference to it
816 * and must dispose of it using closef() semantics (as if a
817 * close() were performed on it).
818 *
819 * XXX this duplicates parts of close().
820 */
821 if (delfp != NULL) {
822 knote_fdclose(td, new);
823 if (delfp->f_type == DTYPE_MQUEUE)
824 mq_fdclose(td, new, delfp);
825 FILEDESC_XUNLOCK(fdp);
826 (void) closef(delfp, td);
827 if (holdleaders) {
828 FILEDESC_XLOCK(fdp);
829 fdp->fd_holdleaderscount--;
830 if (fdp->fd_holdleaderscount == 0 &&
831 fdp->fd_holdleaderswakeup != 0) {
832 fdp->fd_holdleaderswakeup = 0;
833 wakeup(&fdp->fd_holdleaderscount);
834 }
835 FILEDESC_XUNLOCK(fdp);
836 }
837 } else {
838 FILEDESC_XUNLOCK(fdp);
839 }
840 return (0);
841 }
842
843 /*
844 * If sigio is on the list associated with a process or process group,
845 * disable signalling from the device, remove sigio from the list and
846 * free sigio.
847 */
848 void
849 funsetown(struct sigio **sigiop)
850 {
851 struct sigio *sigio;
852
853 SIGIO_LOCK();
854 sigio = *sigiop;
855 if (sigio == NULL) {
856 SIGIO_UNLOCK();
857 return;
858 }
859 *(sigio->sio_myref) = NULL;
860 if ((sigio)->sio_pgid < 0) {
861 struct pgrp *pg = (sigio)->sio_pgrp;
862 PGRP_LOCK(pg);
863 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
864 sigio, sio_pgsigio);
865 PGRP_UNLOCK(pg);
866 } else {
867 struct proc *p = (sigio)->sio_proc;
868 PROC_LOCK(p);
869 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
870 sigio, sio_pgsigio);
871 PROC_UNLOCK(p);
872 }
873 SIGIO_UNLOCK();
874 crfree(sigio->sio_ucred);
875 FREE(sigio, M_SIGIO);
876 }
877
878 /*
879 * Free a list of sigio structures.
880 * We only need to lock the SIGIO_LOCK because we have made ourselves
881 * inaccessible to callers of fsetown and therefore do not need to lock
882 * the proc or pgrp struct for the list manipulation.
883 */
884 void
885 funsetownlst(struct sigiolst *sigiolst)
886 {
887 struct proc *p;
888 struct pgrp *pg;
889 struct sigio *sigio;
890
891 sigio = SLIST_FIRST(sigiolst);
892 if (sigio == NULL)
893 return;
894 p = NULL;
895 pg = NULL;
896
897 /*
898 * Every entry of the list should belong
899 * to a single proc or pgrp.
900 */
901 if (sigio->sio_pgid < 0) {
902 pg = sigio->sio_pgrp;
903 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
904 } else /* if (sigio->sio_pgid > 0) */ {
905 p = sigio->sio_proc;
906 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
907 }
908
909 SIGIO_LOCK();
910 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
911 *(sigio->sio_myref) = NULL;
912 if (pg != NULL) {
913 KASSERT(sigio->sio_pgid < 0,
914 ("Proc sigio in pgrp sigio list"));
915 KASSERT(sigio->sio_pgrp == pg,
916 ("Bogus pgrp in sigio list"));
917 PGRP_LOCK(pg);
918 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
919 sio_pgsigio);
920 PGRP_UNLOCK(pg);
921 } else /* if (p != NULL) */ {
922 KASSERT(sigio->sio_pgid > 0,
923 ("Pgrp sigio in proc sigio list"));
924 KASSERT(sigio->sio_proc == p,
925 ("Bogus proc in sigio list"));
926 PROC_LOCK(p);
927 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
928 sio_pgsigio);
929 PROC_UNLOCK(p);
930 }
931 SIGIO_UNLOCK();
932 crfree(sigio->sio_ucred);
933 FREE(sigio, M_SIGIO);
934 SIGIO_LOCK();
935 }
936 SIGIO_UNLOCK();
937 }
938
939 /*
940 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
941 *
942 * After permission checking, add a sigio structure to the sigio list for
943 * the process or process group.
944 */
945 int
946 fsetown(pid_t pgid, struct sigio **sigiop)
947 {
948 struct proc *proc;
949 struct pgrp *pgrp;
950 struct sigio *sigio;
951 int ret;
952
953 if (pgid == 0) {
954 funsetown(sigiop);
955 return (0);
956 }
957
958 ret = 0;
959
960 /* Allocate and fill in the new sigio out of locks. */
961 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
962 sigio->sio_pgid = pgid;
963 sigio->sio_ucred = crhold(curthread->td_ucred);
964 sigio->sio_myref = sigiop;
965
966 sx_slock(&proctree_lock);
967 if (pgid > 0) {
968 proc = pfind(pgid);
969 if (proc == NULL) {
970 ret = ESRCH;
971 goto fail;
972 }
973
974 /*
975 * Policy - Don't allow a process to FSETOWN a process
976 * in another session.
977 *
978 * Remove this test to allow maximum flexibility or
979 * restrict FSETOWN to the current process or process
980 * group for maximum safety.
981 */
982 PROC_UNLOCK(proc);
983 if (proc->p_session != curthread->td_proc->p_session) {
984 ret = EPERM;
985 goto fail;
986 }
987
988 pgrp = NULL;
989 } else /* if (pgid < 0) */ {
990 pgrp = pgfind(-pgid);
991 if (pgrp == NULL) {
992 ret = ESRCH;
993 goto fail;
994 }
995 PGRP_UNLOCK(pgrp);
996
997 /*
998 * Policy - Don't allow a process to FSETOWN a process
999 * in another session.
1000 *
1001 * Remove this test to allow maximum flexibility or
1002 * restrict FSETOWN to the current process or process
1003 * group for maximum safety.
1004 */
1005 if (pgrp->pg_session != curthread->td_proc->p_session) {
1006 ret = EPERM;
1007 goto fail;
1008 }
1009
1010 proc = NULL;
1011 }
1012 funsetown(sigiop);
1013 if (pgid > 0) {
1014 PROC_LOCK(proc);
1015 /*
1016 * Since funsetownlst() is called without the proctree
1017 * locked, we need to check for P_WEXIT.
1018 * XXX: is ESRCH correct?
1019 */
1020 if ((proc->p_flag & P_WEXIT) != 0) {
1021 PROC_UNLOCK(proc);
1022 ret = ESRCH;
1023 goto fail;
1024 }
1025 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1026 sigio->sio_proc = proc;
1027 PROC_UNLOCK(proc);
1028 } else {
1029 PGRP_LOCK(pgrp);
1030 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1031 sigio->sio_pgrp = pgrp;
1032 PGRP_UNLOCK(pgrp);
1033 }
1034 sx_sunlock(&proctree_lock);
1035 SIGIO_LOCK();
1036 *sigiop = sigio;
1037 SIGIO_UNLOCK();
1038 return (0);
1039
1040 fail:
1041 sx_sunlock(&proctree_lock);
1042 crfree(sigio->sio_ucred);
1043 FREE(sigio, M_SIGIO);
1044 return (ret);
1045 }
1046
1047 /*
1048 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1049 */
1050 pid_t
1051 fgetown(sigiop)
1052 struct sigio **sigiop;
1053 {
1054 pid_t pgid;
1055
1056 SIGIO_LOCK();
1057 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1058 SIGIO_UNLOCK();
1059 return (pgid);
1060 }
1061
1062 /*
1063 * Close a file descriptor.
1064 */
1065 #ifndef _SYS_SYSPROTO_H_
1066 struct close_args {
1067 int fd;
1068 };
1069 #endif
1070 /* ARGSUSED */
1071 int
1072 close(td, uap)
1073 struct thread *td;
1074 struct close_args *uap;
1075 {
1076
1077 return (kern_close(td, uap->fd));
1078 }
1079
1080 int
1081 kern_close(td, fd)
1082 struct thread *td;
1083 int fd;
1084 {
1085 struct filedesc *fdp;
1086 struct file *fp;
1087 int error;
1088 int holdleaders;
1089
1090 error = 0;
1091 holdleaders = 0;
1092 fdp = td->td_proc->p_fd;
1093
1094 AUDIT_SYSCLOSE(td, fd);
1095
1096 FILEDESC_XLOCK(fdp);
1097 if ((unsigned)fd >= fdp->fd_nfiles ||
1098 (fp = fdp->fd_ofiles[fd]) == NULL) {
1099 FILEDESC_XUNLOCK(fdp);
1100 return (EBADF);
1101 }
1102 fdp->fd_ofiles[fd] = NULL;
1103 fdp->fd_ofileflags[fd] = 0;
1104 fdunused(fdp, fd);
1105 if (td->td_proc->p_fdtol != NULL) {
1106 /*
1107 * Ask fdfree() to sleep to ensure that all relevant
1108 * process leaders can be traversed in closef().
1109 */
1110 fdp->fd_holdleaderscount++;
1111 holdleaders = 1;
1112 }
1113
1114 /*
1115 * We now hold the fp reference that used to be owned by the
1116 * descriptor array. We have to unlock the FILEDESC *AFTER*
1117 * knote_fdclose to prevent a race of the fd getting opened, a knote
1118 * added, and deleteing a knote for the new fd.
1119 */
1120 knote_fdclose(td, fd);
1121 if (fp->f_type == DTYPE_MQUEUE)
1122 mq_fdclose(td, fd, fp);
1123 FILEDESC_XUNLOCK(fdp);
1124
1125 error = closef(fp, td);
1126 if (holdleaders) {
1127 FILEDESC_XLOCK(fdp);
1128 fdp->fd_holdleaderscount--;
1129 if (fdp->fd_holdleaderscount == 0 &&
1130 fdp->fd_holdleaderswakeup != 0) {
1131 fdp->fd_holdleaderswakeup = 0;
1132 wakeup(&fdp->fd_holdleaderscount);
1133 }
1134 FILEDESC_XUNLOCK(fdp);
1135 }
1136 return (error);
1137 }
1138
1139 /*
1140 * Close open file descriptors.
1141 */
1142 #ifndef _SYS_SYSPROTO_H_
1143 struct closefrom_args {
1144 int lowfd;
1145 };
1146 #endif
1147 /* ARGSUSED */
1148 int
1149 closefrom(struct thread *td, struct closefrom_args *uap)
1150 {
1151 struct filedesc *fdp;
1152 int fd;
1153
1154 fdp = td->td_proc->p_fd;
1155 AUDIT_ARG(fd, uap->lowfd);
1156
1157 /*
1158 * Treat negative starting file descriptor values identical to
1159 * closefrom(0) which closes all files.
1160 */
1161 if (uap->lowfd < 0)
1162 uap->lowfd = 0;
1163 FILEDESC_SLOCK(fdp);
1164 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
1165 if (fdp->fd_ofiles[fd] != NULL) {
1166 FILEDESC_SUNLOCK(fdp);
1167 (void)kern_close(td, fd);
1168 FILEDESC_SLOCK(fdp);
1169 }
1170 }
1171 FILEDESC_SUNLOCK(fdp);
1172 return (0);
1173 }
1174
1175 #if defined(COMPAT_43)
1176 /*
1177 * Return status information about a file descriptor.
1178 */
1179 #ifndef _SYS_SYSPROTO_H_
1180 struct ofstat_args {
1181 int fd;
1182 struct ostat *sb;
1183 };
1184 #endif
1185 /* ARGSUSED */
1186 int
1187 ofstat(struct thread *td, struct ofstat_args *uap)
1188 {
1189 struct ostat oub;
1190 struct stat ub;
1191 int error;
1192
1193 error = kern_fstat(td, uap->fd, &ub);
1194 if (error == 0) {
1195 cvtstat(&ub, &oub);
1196 error = copyout(&oub, uap->sb, sizeof(oub));
1197 }
1198 return (error);
1199 }
1200 #endif /* COMPAT_43 */
1201
1202 /*
1203 * Return status information about a file descriptor.
1204 */
1205 #ifndef _SYS_SYSPROTO_H_
1206 struct fstat_args {
1207 int fd;
1208 struct stat *sb;
1209 };
1210 #endif
1211 /* ARGSUSED */
1212 int
1213 fstat(struct thread *td, struct fstat_args *uap)
1214 {
1215 struct stat ub;
1216 int error;
1217
1218 error = kern_fstat(td, uap->fd, &ub);
1219 if (error == 0)
1220 error = copyout(&ub, uap->sb, sizeof(ub));
1221 return (error);
1222 }
1223
1224 int
1225 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1226 {
1227 struct file *fp;
1228 int error;
1229
1230 AUDIT_ARG(fd, fd);
1231
1232 if ((error = fget(td, fd, &fp)) != 0)
1233 return (error);
1234
1235 AUDIT_ARG(file, td->td_proc, fp);
1236
1237 error = fo_stat(fp, sbp, td->td_ucred, td);
1238 fdrop(fp, td);
1239 #ifdef KTRACE
1240 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1241 ktrstat(sbp);
1242 #endif
1243 return (error);
1244 }
1245
1246 /*
1247 * Return status information about a file descriptor.
1248 */
1249 #ifndef _SYS_SYSPROTO_H_
1250 struct nfstat_args {
1251 int fd;
1252 struct nstat *sb;
1253 };
1254 #endif
1255 /* ARGSUSED */
1256 int
1257 nfstat(struct thread *td, struct nfstat_args *uap)
1258 {
1259 struct nstat nub;
1260 struct stat ub;
1261 int error;
1262
1263 error = kern_fstat(td, uap->fd, &ub);
1264 if (error == 0) {
1265 cvtnstat(&ub, &nub);
1266 error = copyout(&nub, uap->sb, sizeof(nub));
1267 }
1268 return (error);
1269 }
1270
1271 /*
1272 * Return pathconf information about a file descriptor.
1273 */
1274 #ifndef _SYS_SYSPROTO_H_
1275 struct fpathconf_args {
1276 int fd;
1277 int name;
1278 };
1279 #endif
1280 /* ARGSUSED */
1281 int
1282 fpathconf(struct thread *td, struct fpathconf_args *uap)
1283 {
1284 struct file *fp;
1285 struct vnode *vp;
1286 int error;
1287
1288 if ((error = fget(td, uap->fd, &fp)) != 0)
1289 return (error);
1290
1291 /* If asynchronous I/O is available, it works for all descriptors. */
1292 if (uap->name == _PC_ASYNC_IO) {
1293 td->td_retval[0] = async_io_version;
1294 goto out;
1295 }
1296 vp = fp->f_vnode;
1297 if (vp != NULL) {
1298 int vfslocked;
1299 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1300 vn_lock(vp, LK_SHARED | LK_RETRY, td);
1301 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1302 VOP_UNLOCK(vp, 0, td);
1303 VFS_UNLOCK_GIANT(vfslocked);
1304 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1305 if (uap->name != _PC_PIPE_BUF) {
1306 error = EINVAL;
1307 } else {
1308 td->td_retval[0] = PIPE_BUF;
1309 error = 0;
1310 }
1311 } else {
1312 error = EOPNOTSUPP;
1313 }
1314 out:
1315 fdrop(fp, td);
1316 return (error);
1317 }
1318
1319 /*
1320 * Grow the file table to accomodate (at least) nfd descriptors. This may
1321 * block and drop the filedesc lock, but it will reacquire it before
1322 * returning.
1323 */
1324 static void
1325 fdgrowtable(struct filedesc *fdp, int nfd)
1326 {
1327 struct file **ntable;
1328 char *nfileflags;
1329 int nnfiles, onfiles;
1330 NDSLOTTYPE *nmap;
1331
1332 FILEDESC_XLOCK_ASSERT(fdp);
1333
1334 KASSERT(fdp->fd_nfiles > 0,
1335 ("zero-length file table"));
1336
1337 /* compute the size of the new table */
1338 onfiles = fdp->fd_nfiles;
1339 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1340 if (nnfiles <= onfiles)
1341 /* the table is already large enough */
1342 return;
1343
1344 /* allocate a new table and (if required) new bitmaps */
1345 FILEDESC_XUNLOCK(fdp);
1346 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1347 M_FILEDESC, M_ZERO | M_WAITOK);
1348 nfileflags = (char *)&ntable[nnfiles];
1349 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1350 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1351 M_FILEDESC, M_ZERO | M_WAITOK);
1352 else
1353 nmap = NULL;
1354 FILEDESC_XLOCK(fdp);
1355
1356 /*
1357 * We now have new tables ready to go. Since we dropped the
1358 * filedesc lock to call malloc(), watch out for a race.
1359 */
1360 onfiles = fdp->fd_nfiles;
1361 if (onfiles >= nnfiles) {
1362 /* we lost the race, but that's OK */
1363 free(ntable, M_FILEDESC);
1364 if (nmap != NULL)
1365 free(nmap, M_FILEDESC);
1366 return;
1367 }
1368 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1369 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1370 if (onfiles > NDFILE)
1371 free(fdp->fd_ofiles, M_FILEDESC);
1372 fdp->fd_ofiles = ntable;
1373 fdp->fd_ofileflags = nfileflags;
1374 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1375 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1376 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1377 free(fdp->fd_map, M_FILEDESC);
1378 fdp->fd_map = nmap;
1379 }
1380 fdp->fd_nfiles = nnfiles;
1381 }
1382
1383 /*
1384 * Allocate a file descriptor for the process.
1385 */
1386 int
1387 fdalloc(struct thread *td, int minfd, int *result)
1388 {
1389 struct proc *p = td->td_proc;
1390 struct filedesc *fdp = p->p_fd;
1391 int fd = -1, maxfd;
1392
1393 FILEDESC_XLOCK_ASSERT(fdp);
1394
1395 if (fdp->fd_freefile > minfd)
1396 minfd = fdp->fd_freefile;
1397
1398 PROC_LOCK(p);
1399 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1400 PROC_UNLOCK(p);
1401
1402 /*
1403 * Search the bitmap for a free descriptor. If none is found, try
1404 * to grow the file table. Keep at it until we either get a file
1405 * descriptor or run into process or system limits; fdgrowtable()
1406 * may drop the filedesc lock, so we're in a race.
1407 */
1408 for (;;) {
1409 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1410 if (fd >= maxfd)
1411 return (EMFILE);
1412 if (fd < fdp->fd_nfiles)
1413 break;
1414 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1415 }
1416
1417 /*
1418 * Perform some sanity checks, then mark the file descriptor as
1419 * used and return it to the caller.
1420 */
1421 KASSERT(!fdisused(fdp, fd),
1422 ("fd_first_free() returned non-free descriptor"));
1423 KASSERT(fdp->fd_ofiles[fd] == NULL,
1424 ("free descriptor isn't"));
1425 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1426 fdused(fdp, fd);
1427 *result = fd;
1428 return (0);
1429 }
1430
1431 /*
1432 * Check to see whether n user file descriptors are available to the process
1433 * p.
1434 */
1435 int
1436 fdavail(struct thread *td, int n)
1437 {
1438 struct proc *p = td->td_proc;
1439 struct filedesc *fdp = td->td_proc->p_fd;
1440 struct file **fpp;
1441 int i, lim, last;
1442
1443 FILEDESC_LOCK_ASSERT(fdp);
1444
1445 PROC_LOCK(p);
1446 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1447 PROC_UNLOCK(p);
1448 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1449 return (1);
1450 last = min(fdp->fd_nfiles, lim);
1451 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1452 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1453 if (*fpp == NULL && --n <= 0)
1454 return (1);
1455 }
1456 return (0);
1457 }
1458
1459 /*
1460 * Create a new open file structure and allocate a file decriptor for the
1461 * process that refers to it. We add one reference to the file for the
1462 * descriptor table and one reference for resultfp. This is to prevent us
1463 * being preempted and the entry in the descriptor table closed after we
1464 * release the FILEDESC lock.
1465 */
1466 int
1467 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1468 {
1469 struct proc *p = td->td_proc;
1470 struct file *fp, *fq;
1471 int error, i;
1472 int maxuserfiles = maxfiles - (maxfiles / 20);
1473 static struct timeval lastfail;
1474 static int curfail;
1475
1476 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1477 sx_xlock(&filelist_lock);
1478
1479 if ((openfiles >= maxuserfiles &&
1480 priv_check(td, PRIV_MAXFILES) != 0) ||
1481 openfiles >= maxfiles) {
1482 if (ppsratecheck(&lastfail, &curfail, 1)) {
1483 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1484 td->td_ucred->cr_ruid);
1485 }
1486 sx_xunlock(&filelist_lock);
1487 uma_zfree(file_zone, fp);
1488 return (ENFILE);
1489 }
1490 openfiles++;
1491
1492 /*
1493 * If the process has file descriptor zero open, add the new file
1494 * descriptor to the list of open files at that point, otherwise
1495 * put it at the front of the list of open files.
1496 */
1497 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1498 fp->f_count = 1;
1499 if (resultfp)
1500 fp->f_count++;
1501 fp->f_cred = crhold(td->td_ucred);
1502 fp->f_ops = &badfileops;
1503 fp->f_data = NULL;
1504 fp->f_vnode = NULL;
1505 FILEDESC_XLOCK(p->p_fd);
1506 if ((fq = p->p_fd->fd_ofiles[0])) {
1507 LIST_INSERT_AFTER(fq, fp, f_list);
1508 } else {
1509 LIST_INSERT_HEAD(&filehead, fp, f_list);
1510 }
1511 sx_xunlock(&filelist_lock);
1512 if ((error = fdalloc(td, 0, &i))) {
1513 FILEDESC_XUNLOCK(p->p_fd);
1514 fdrop(fp, td);
1515 if (resultfp)
1516 fdrop(fp, td);
1517 return (error);
1518 }
1519 p->p_fd->fd_ofiles[i] = fp;
1520 FILEDESC_XUNLOCK(p->p_fd);
1521 if (resultfp)
1522 *resultfp = fp;
1523 if (resultfd)
1524 *resultfd = i;
1525 return (0);
1526 }
1527
1528 /*
1529 * Build a new filedesc structure from another.
1530 * Copy the current, root, and jail root vnode references.
1531 */
1532 struct filedesc *
1533 fdinit(struct filedesc *fdp)
1534 {
1535 struct filedesc0 *newfdp;
1536
1537 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1538 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1539 if (fdp != NULL) {
1540 FILEDESC_XLOCK(fdp);
1541 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1542 if (newfdp->fd_fd.fd_cdir)
1543 VREF(newfdp->fd_fd.fd_cdir);
1544 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1545 if (newfdp->fd_fd.fd_rdir)
1546 VREF(newfdp->fd_fd.fd_rdir);
1547 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1548 if (newfdp->fd_fd.fd_jdir)
1549 VREF(newfdp->fd_fd.fd_jdir);
1550 FILEDESC_XUNLOCK(fdp);
1551 }
1552
1553 /* Create the file descriptor table. */
1554 newfdp->fd_fd.fd_refcnt = 1;
1555 newfdp->fd_fd.fd_holdcnt = 1;
1556 newfdp->fd_fd.fd_cmask = CMASK;
1557 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1558 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1559 newfdp->fd_fd.fd_nfiles = NDFILE;
1560 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1561 newfdp->fd_fd.fd_lastfile = -1;
1562 return (&newfdp->fd_fd);
1563 }
1564
1565 static struct filedesc *
1566 fdhold(struct proc *p)
1567 {
1568 struct filedesc *fdp;
1569
1570 mtx_lock(&fdesc_mtx);
1571 fdp = p->p_fd;
1572 if (fdp != NULL)
1573 fdp->fd_holdcnt++;
1574 mtx_unlock(&fdesc_mtx);
1575 return (fdp);
1576 }
1577
1578 static void
1579 fddrop(struct filedesc *fdp)
1580 {
1581 int i;
1582
1583 mtx_lock(&fdesc_mtx);
1584 i = --fdp->fd_holdcnt;
1585 mtx_unlock(&fdesc_mtx);
1586 if (i > 0)
1587 return;
1588
1589 FILEDESC_LOCK_DESTROY(fdp);
1590 FREE(fdp, M_FILEDESC);
1591 }
1592
1593 /*
1594 * Share a filedesc structure.
1595 */
1596 struct filedesc *
1597 fdshare(struct filedesc *fdp)
1598 {
1599
1600 FILEDESC_XLOCK(fdp);
1601 fdp->fd_refcnt++;
1602 FILEDESC_XUNLOCK(fdp);
1603 return (fdp);
1604 }
1605
1606 /*
1607 * Unshare a filedesc structure, if necessary by making a copy
1608 */
1609 void
1610 fdunshare(struct proc *p, struct thread *td)
1611 {
1612
1613 FILEDESC_XLOCK(p->p_fd);
1614 if (p->p_fd->fd_refcnt > 1) {
1615 struct filedesc *tmp;
1616
1617 FILEDESC_XUNLOCK(p->p_fd);
1618 tmp = fdcopy(p->p_fd);
1619 fdfree(td);
1620 p->p_fd = tmp;
1621 } else
1622 FILEDESC_XUNLOCK(p->p_fd);
1623 }
1624
1625 /*
1626 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1627 * this is to ease callers, not catch errors.
1628 */
1629 struct filedesc *
1630 fdcopy(struct filedesc *fdp)
1631 {
1632 struct filedesc *newfdp;
1633 int i;
1634
1635 /* Certain daemons might not have file descriptors. */
1636 if (fdp == NULL)
1637 return (NULL);
1638
1639 newfdp = fdinit(fdp);
1640 FILEDESC_SLOCK(fdp);
1641 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1642 FILEDESC_SUNLOCK(fdp);
1643 FILEDESC_XLOCK(newfdp);
1644 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1645 FILEDESC_XUNLOCK(newfdp);
1646 FILEDESC_SLOCK(fdp);
1647 }
1648 /* copy everything except kqueue descriptors */
1649 newfdp->fd_freefile = -1;
1650 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1651 if (fdisused(fdp, i) &&
1652 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE &&
1653 fdp->fd_ofiles[i]->f_ops != &badfileops) {
1654 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1655 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1656 fhold(newfdp->fd_ofiles[i]);
1657 newfdp->fd_lastfile = i;
1658 } else {
1659 if (newfdp->fd_freefile == -1)
1660 newfdp->fd_freefile = i;
1661 }
1662 }
1663 FILEDESC_SUNLOCK(fdp);
1664 FILEDESC_XLOCK(newfdp);
1665 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1666 if (newfdp->fd_ofiles[i] != NULL)
1667 fdused(newfdp, i);
1668 FILEDESC_XUNLOCK(newfdp);
1669 FILEDESC_SLOCK(fdp);
1670 if (newfdp->fd_freefile == -1)
1671 newfdp->fd_freefile = i;
1672 newfdp->fd_cmask = fdp->fd_cmask;
1673 FILEDESC_SUNLOCK(fdp);
1674 return (newfdp);
1675 }
1676
1677 /*
1678 * Release a filedesc structure.
1679 */
1680 void
1681 fdfree(struct thread *td)
1682 {
1683 struct filedesc *fdp;
1684 struct file **fpp;
1685 int i, locked;
1686 struct filedesc_to_leader *fdtol;
1687 struct file *fp;
1688 struct vnode *cdir, *jdir, *rdir, *vp;
1689 struct flock lf;
1690
1691 /* Certain daemons might not have file descriptors. */
1692 fdp = td->td_proc->p_fd;
1693 if (fdp == NULL)
1694 return;
1695
1696 /* Check for special need to clear POSIX style locks */
1697 fdtol = td->td_proc->p_fdtol;
1698 if (fdtol != NULL) {
1699 FILEDESC_XLOCK(fdp);
1700 KASSERT(fdtol->fdl_refcount > 0,
1701 ("filedesc_to_refcount botch: fdl_refcount=%d",
1702 fdtol->fdl_refcount));
1703 if (fdtol->fdl_refcount == 1 &&
1704 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1705 for (i = 0, fpp = fdp->fd_ofiles;
1706 i <= fdp->fd_lastfile;
1707 i++, fpp++) {
1708 if (*fpp == NULL ||
1709 (*fpp)->f_type != DTYPE_VNODE)
1710 continue;
1711 fp = *fpp;
1712 fhold(fp);
1713 FILEDESC_XUNLOCK(fdp);
1714 lf.l_whence = SEEK_SET;
1715 lf.l_start = 0;
1716 lf.l_len = 0;
1717 lf.l_type = F_UNLCK;
1718 vp = fp->f_vnode;
1719 locked = VFS_LOCK_GIANT(vp->v_mount);
1720 (void) VOP_ADVLOCK(vp,
1721 (caddr_t)td->td_proc->
1722 p_leader,
1723 F_UNLCK,
1724 &lf,
1725 F_POSIX);
1726 VFS_UNLOCK_GIANT(locked);
1727 FILEDESC_XLOCK(fdp);
1728 fdrop(fp, td);
1729 fpp = fdp->fd_ofiles + i;
1730 }
1731 }
1732 retry:
1733 if (fdtol->fdl_refcount == 1) {
1734 if (fdp->fd_holdleaderscount > 0 &&
1735 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1736 /*
1737 * close() or do_dup() has cleared a reference
1738 * in a shared file descriptor table.
1739 */
1740 fdp->fd_holdleaderswakeup = 1;
1741 sx_sleep(&fdp->fd_holdleaderscount,
1742 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1743 goto retry;
1744 }
1745 if (fdtol->fdl_holdcount > 0) {
1746 /*
1747 * Ensure that fdtol->fdl_leader remains
1748 * valid in closef().
1749 */
1750 fdtol->fdl_wakeup = 1;
1751 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1752 "fdlhold", 0);
1753 goto retry;
1754 }
1755 }
1756 fdtol->fdl_refcount--;
1757 if (fdtol->fdl_refcount == 0 &&
1758 fdtol->fdl_holdcount == 0) {
1759 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1760 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1761 } else
1762 fdtol = NULL;
1763 td->td_proc->p_fdtol = NULL;
1764 FILEDESC_XUNLOCK(fdp);
1765 if (fdtol != NULL)
1766 FREE(fdtol, M_FILEDESC_TO_LEADER);
1767 }
1768 FILEDESC_XLOCK(fdp);
1769 i = --fdp->fd_refcnt;
1770 FILEDESC_XUNLOCK(fdp);
1771 if (i > 0)
1772 return;
1773
1774 fpp = fdp->fd_ofiles;
1775 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1776 if (*fpp) {
1777 FILEDESC_XLOCK(fdp);
1778 fp = *fpp;
1779 *fpp = NULL;
1780 FILEDESC_XUNLOCK(fdp);
1781 (void) closef(fp, td);
1782 }
1783 }
1784 FILEDESC_XLOCK(fdp);
1785
1786 /* XXX This should happen earlier. */
1787 mtx_lock(&fdesc_mtx);
1788 td->td_proc->p_fd = NULL;
1789 mtx_unlock(&fdesc_mtx);
1790
1791 if (fdp->fd_nfiles > NDFILE)
1792 FREE(fdp->fd_ofiles, M_FILEDESC);
1793 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1794 FREE(fdp->fd_map, M_FILEDESC);
1795
1796 fdp->fd_nfiles = 0;
1797
1798 cdir = fdp->fd_cdir;
1799 fdp->fd_cdir = NULL;
1800 rdir = fdp->fd_rdir;
1801 fdp->fd_rdir = NULL;
1802 jdir = fdp->fd_jdir;
1803 fdp->fd_jdir = NULL;
1804 FILEDESC_XUNLOCK(fdp);
1805
1806 if (cdir) {
1807 locked = VFS_LOCK_GIANT(cdir->v_mount);
1808 vrele(cdir);
1809 VFS_UNLOCK_GIANT(locked);
1810 }
1811 if (rdir) {
1812 locked = VFS_LOCK_GIANT(rdir->v_mount);
1813 vrele(rdir);
1814 VFS_UNLOCK_GIANT(locked);
1815 }
1816 if (jdir) {
1817 locked = VFS_LOCK_GIANT(jdir->v_mount);
1818 vrele(jdir);
1819 VFS_UNLOCK_GIANT(locked);
1820 }
1821
1822 fddrop(fdp);
1823 }
1824
1825 /*
1826 * For setugid programs, we don't want to people to use that setugidness
1827 * to generate error messages which write to a file which otherwise would
1828 * otherwise be off-limits to the process. We check for filesystems where
1829 * the vnode can change out from under us after execve (like [lin]procfs).
1830 *
1831 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1832 * sufficient. We also don't check for setugidness since we know we are.
1833 */
1834 static int
1835 is_unsafe(struct file *fp)
1836 {
1837 if (fp->f_type == DTYPE_VNODE) {
1838 struct vnode *vp = fp->f_vnode;
1839
1840 if ((vp->v_vflag & VV_PROCDEP) != 0)
1841 return (1);
1842 }
1843 return (0);
1844 }
1845
1846 /*
1847 * Make this setguid thing safe, if at all possible.
1848 */
1849 void
1850 setugidsafety(struct thread *td)
1851 {
1852 struct filedesc *fdp;
1853 int i;
1854
1855 /* Certain daemons might not have file descriptors. */
1856 fdp = td->td_proc->p_fd;
1857 if (fdp == NULL)
1858 return;
1859
1860 /*
1861 * Note: fdp->fd_ofiles may be reallocated out from under us while
1862 * we are blocked in a close. Be careful!
1863 */
1864 FILEDESC_XLOCK(fdp);
1865 for (i = 0; i <= fdp->fd_lastfile; i++) {
1866 if (i > 2)
1867 break;
1868 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1869 struct file *fp;
1870
1871 knote_fdclose(td, i);
1872 /*
1873 * NULL-out descriptor prior to close to avoid
1874 * a race while close blocks.
1875 */
1876 fp = fdp->fd_ofiles[i];
1877 fdp->fd_ofiles[i] = NULL;
1878 fdp->fd_ofileflags[i] = 0;
1879 fdunused(fdp, i);
1880 FILEDESC_XUNLOCK(fdp);
1881 (void) closef(fp, td);
1882 FILEDESC_XLOCK(fdp);
1883 }
1884 }
1885 FILEDESC_XUNLOCK(fdp);
1886 }
1887
1888 /*
1889 * If a specific file object occupies a specific file descriptor, close the
1890 * file descriptor entry and drop a reference on the file object. This is a
1891 * convenience function to handle a subsequent error in a function that calls
1892 * falloc() that handles the race that another thread might have closed the
1893 * file descriptor out from under the thread creating the file object.
1894 */
1895 void
1896 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1897 {
1898
1899 FILEDESC_XLOCK(fdp);
1900 if (fdp->fd_ofiles[idx] == fp) {
1901 fdp->fd_ofiles[idx] = NULL;
1902 fdunused(fdp, idx);
1903 FILEDESC_XUNLOCK(fdp);
1904 fdrop(fp, td);
1905 } else
1906 FILEDESC_XUNLOCK(fdp);
1907 }
1908
1909 /*
1910 * Close any files on exec?
1911 */
1912 void
1913 fdcloseexec(struct thread *td)
1914 {
1915 struct filedesc *fdp;
1916 int i;
1917
1918 /* Certain daemons might not have file descriptors. */
1919 fdp = td->td_proc->p_fd;
1920 if (fdp == NULL)
1921 return;
1922
1923 FILEDESC_XLOCK(fdp);
1924
1925 /*
1926 * We cannot cache fd_ofiles or fd_ofileflags since operations
1927 * may block and rip them out from under us.
1928 */
1929 for (i = 0; i <= fdp->fd_lastfile; i++) {
1930 if (fdp->fd_ofiles[i] != NULL &&
1931 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1932 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1933 struct file *fp;
1934
1935 knote_fdclose(td, i);
1936 /*
1937 * NULL-out descriptor prior to close to avoid
1938 * a race while close blocks.
1939 */
1940 fp = fdp->fd_ofiles[i];
1941 fdp->fd_ofiles[i] = NULL;
1942 fdp->fd_ofileflags[i] = 0;
1943 fdunused(fdp, i);
1944 if (fp->f_type == DTYPE_MQUEUE)
1945 mq_fdclose(td, i, fp);
1946 FILEDESC_XUNLOCK(fdp);
1947 (void) closef(fp, td);
1948 FILEDESC_XLOCK(fdp);
1949 }
1950 }
1951 FILEDESC_XUNLOCK(fdp);
1952 }
1953
1954 /*
1955 * It is unsafe for set[ug]id processes to be started with file
1956 * descriptors 0..2 closed, as these descriptors are given implicit
1957 * significance in the Standard C library. fdcheckstd() will create a
1958 * descriptor referencing /dev/null for each of stdin, stdout, and
1959 * stderr that is not already open.
1960 */
1961 int
1962 fdcheckstd(struct thread *td)
1963 {
1964 struct filedesc *fdp;
1965 register_t retval, save;
1966 int i, error, devnull;
1967
1968 fdp = td->td_proc->p_fd;
1969 if (fdp == NULL)
1970 return (0);
1971 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1972 devnull = -1;
1973 error = 0;
1974 for (i = 0; i < 3; i++) {
1975 if (fdp->fd_ofiles[i] != NULL)
1976 continue;
1977 if (devnull < 0) {
1978 save = td->td_retval[0];
1979 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
1980 O_RDWR, 0);
1981 devnull = td->td_retval[0];
1982 KASSERT(devnull == i, ("oof, we didn't get our fd"));
1983 td->td_retval[0] = save;
1984 if (error)
1985 break;
1986 } else {
1987 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1988 if (error != 0)
1989 break;
1990 }
1991 }
1992 return (error);
1993 }
1994
1995 /*
1996 * Internal form of close. Decrement reference count on file structure.
1997 * Note: td may be NULL when closing a file that was being passed in a
1998 * message.
1999 *
2000 * XXXRW: Giant is not required for the caller, but often will be held; this
2001 * makes it moderately likely the Giant will be recursed in the VFS case.
2002 */
2003 int
2004 closef(struct file *fp, struct thread *td)
2005 {
2006 struct vnode *vp;
2007 struct flock lf;
2008 struct filedesc_to_leader *fdtol;
2009 struct filedesc *fdp;
2010
2011 /*
2012 * POSIX record locking dictates that any close releases ALL
2013 * locks owned by this process. This is handled by setting
2014 * a flag in the unlock to free ONLY locks obeying POSIX
2015 * semantics, and not to free BSD-style file locks.
2016 * If the descriptor was in a message, POSIX-style locks
2017 * aren't passed with the descriptor, and the thread pointer
2018 * will be NULL. Callers should be careful only to pass a
2019 * NULL thread pointer when there really is no owning
2020 * context that might have locks, or the locks will be
2021 * leaked.
2022 */
2023 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2024 int vfslocked;
2025
2026 vp = fp->f_vnode;
2027 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2028 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2029 lf.l_whence = SEEK_SET;
2030 lf.l_start = 0;
2031 lf.l_len = 0;
2032 lf.l_type = F_UNLCK;
2033 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2034 F_UNLCK, &lf, F_POSIX);
2035 }
2036 fdtol = td->td_proc->p_fdtol;
2037 if (fdtol != NULL) {
2038 /*
2039 * Handle special case where file descriptor table is
2040 * shared between multiple process leaders.
2041 */
2042 fdp = td->td_proc->p_fd;
2043 FILEDESC_XLOCK(fdp);
2044 for (fdtol = fdtol->fdl_next;
2045 fdtol != td->td_proc->p_fdtol;
2046 fdtol = fdtol->fdl_next) {
2047 if ((fdtol->fdl_leader->p_flag &
2048 P_ADVLOCK) == 0)
2049 continue;
2050 fdtol->fdl_holdcount++;
2051 FILEDESC_XUNLOCK(fdp);
2052 lf.l_whence = SEEK_SET;
2053 lf.l_start = 0;
2054 lf.l_len = 0;
2055 lf.l_type = F_UNLCK;
2056 vp = fp->f_vnode;
2057 (void) VOP_ADVLOCK(vp,
2058 (caddr_t)fdtol->fdl_leader,
2059 F_UNLCK, &lf, F_POSIX);
2060 FILEDESC_XLOCK(fdp);
2061 fdtol->fdl_holdcount--;
2062 if (fdtol->fdl_holdcount == 0 &&
2063 fdtol->fdl_wakeup != 0) {
2064 fdtol->fdl_wakeup = 0;
2065 wakeup(fdtol);
2066 }
2067 }
2068 FILEDESC_XUNLOCK(fdp);
2069 }
2070 VFS_UNLOCK_GIANT(vfslocked);
2071 }
2072 return (fdrop(fp, td));
2073 }
2074
2075 /*
2076 * Extract the file pointer associated with the specified descriptor for the
2077 * current user process.
2078 *
2079 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2080 * returned.
2081 *
2082 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
2083 * It should be dropped with fdrop(). If it is not set, then the refcount
2084 * will not be bumped however the thread's filedesc struct will be returned
2085 * locked (for fgetsock).
2086 *
2087 * If an error occured the non-zero error is returned and *fpp is set to
2088 * NULL. Otherwise *fpp is set and zero is returned.
2089 */
2090 static __inline int
2091 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
2092 {
2093 struct filedesc *fdp;
2094 struct file *fp;
2095
2096 *fpp = NULL;
2097 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2098 return (EBADF);
2099 FILEDESC_SLOCK(fdp);
2100 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
2101 FILEDESC_SUNLOCK(fdp);
2102 return (EBADF);
2103 }
2104
2105 /*
2106 * FREAD and FWRITE failure return EBADF as per POSIX.
2107 *
2108 * Only one flag, or 0, may be specified.
2109 */
2110 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2111 FILEDESC_SUNLOCK(fdp);
2112 return (EBADF);
2113 }
2114 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2115 FILEDESC_SUNLOCK(fdp);
2116 return (EBADF);
2117 }
2118 if (hold) {
2119 fhold(fp);
2120 FILEDESC_SUNLOCK(fdp);
2121 }
2122 *fpp = fp;
2123 return (0);
2124 }
2125
2126 int
2127 fget(struct thread *td, int fd, struct file **fpp)
2128 {
2129
2130 return(_fget(td, fd, fpp, 0, 1));
2131 }
2132
2133 int
2134 fget_read(struct thread *td, int fd, struct file **fpp)
2135 {
2136
2137 return(_fget(td, fd, fpp, FREAD, 1));
2138 }
2139
2140 int
2141 fget_write(struct thread *td, int fd, struct file **fpp)
2142 {
2143
2144 return(_fget(td, fd, fpp, FWRITE, 1));
2145 }
2146
2147 /*
2148 * Like fget() but loads the underlying vnode, or returns an error if the
2149 * descriptor does not represent a vnode. Note that pipes use vnodes but
2150 * never have VM objects. The returned vnode will be vref()'d.
2151 *
2152 * XXX: what about the unused flags ?
2153 */
2154 static __inline int
2155 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2156 {
2157 struct file *fp;
2158 int error;
2159
2160 *vpp = NULL;
2161 if ((error = _fget(td, fd, &fp, flags, 0)) != 0)
2162 return (error);
2163 if (fp->f_vnode == NULL) {
2164 error = EINVAL;
2165 } else {
2166 *vpp = fp->f_vnode;
2167 vref(*vpp);
2168 }
2169 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2170 return (error);
2171 }
2172
2173 int
2174 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2175 {
2176
2177 return (_fgetvp(td, fd, vpp, 0));
2178 }
2179
2180 int
2181 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2182 {
2183
2184 return (_fgetvp(td, fd, vpp, FREAD));
2185 }
2186
2187 #ifdef notyet
2188 int
2189 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2190 {
2191
2192 return (_fgetvp(td, fd, vpp, FWRITE));
2193 }
2194 #endif
2195
2196 /*
2197 * Like fget() but loads the underlying socket, or returns an error if the
2198 * descriptor does not represent a socket.
2199 *
2200 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2201 * in the future.
2202 *
2203 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2204 * on their file descriptor reference to prevent the socket from being free'd
2205 * during use.
2206 */
2207 int
2208 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2209 {
2210 struct file *fp;
2211 int error;
2212
2213 *spp = NULL;
2214 if (fflagp != NULL)
2215 *fflagp = 0;
2216 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2217 return (error);
2218 if (fp->f_type != DTYPE_SOCKET) {
2219 error = ENOTSOCK;
2220 } else {
2221 *spp = fp->f_data;
2222 if (fflagp)
2223 *fflagp = fp->f_flag;
2224 SOCK_LOCK(*spp);
2225 soref(*spp);
2226 SOCK_UNLOCK(*spp);
2227 }
2228 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2229 return (error);
2230 }
2231
2232 /*
2233 * Drop the reference count on the socket and XXX release the SX lock in the
2234 * future. The last reference closes the socket.
2235 *
2236 * Note: fputsock() is deprecated, see comment for fgetsock().
2237 */
2238 void
2239 fputsock(struct socket *so)
2240 {
2241
2242 ACCEPT_LOCK();
2243 SOCK_LOCK(so);
2244 sorele(so);
2245 }
2246
2247 int
2248 fdrop(struct file *fp, struct thread *td)
2249 {
2250
2251 FILE_LOCK(fp);
2252 return (fdrop_locked(fp, td));
2253 }
2254
2255 /*
2256 * Drop reference on struct file passed in, may call closef if the
2257 * reference hits zero.
2258 * Expects struct file locked, and will unlock it.
2259 */
2260 static int
2261 fdrop_locked(struct file *fp, struct thread *td)
2262 {
2263 int error;
2264
2265 FILE_LOCK_ASSERT(fp, MA_OWNED);
2266
2267 if (--fp->f_count > 0) {
2268 FILE_UNLOCK(fp);
2269 return (0);
2270 }
2271
2272 /*
2273 * We might have just dropped the last reference to a file
2274 * object that is for a UNIX domain socket whose message
2275 * buffers are being examined in unp_gc(). If that is the
2276 * case, FWAIT will be set in f_gcflag and we need to wait for
2277 * unp_gc() to finish its scan.
2278 */
2279 while (fp->f_gcflag & FWAIT)
2280 msleep(&fp->f_gcflag, fp->f_mtxp, 0, "fpdrop", 0);
2281
2282 /* We have the last ref so we can proceed without the file lock. */
2283 FILE_UNLOCK(fp);
2284 if (fp->f_count < 0)
2285 panic("fdrop: count < 0");
2286 if (fp->f_ops != &badfileops)
2287 error = fo_close(fp, td);
2288 else
2289 error = 0;
2290
2291 sx_xlock(&filelist_lock);
2292 LIST_REMOVE(fp, f_list);
2293 openfiles--;
2294 sx_xunlock(&filelist_lock);
2295
2296 /*
2297 * The f_cdevpriv cannot be assigned non-NULL value while we
2298 * are destroying the file.
2299 */
2300 if (fp->f_cdevpriv != NULL)
2301 devfs_fpdrop(fp);
2302 crfree(fp->f_cred);
2303 uma_zfree(file_zone, fp);
2304
2305 return (error);
2306 }
2307
2308 /*
2309 * Apply an advisory lock on a file descriptor.
2310 *
2311 * Just attempt to get a record lock of the requested type on the entire file
2312 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2313 */
2314 #ifndef _SYS_SYSPROTO_H_
2315 struct flock_args {
2316 int fd;
2317 int how;
2318 };
2319 #endif
2320 /* ARGSUSED */
2321 int
2322 flock(struct thread *td, struct flock_args *uap)
2323 {
2324 struct file *fp;
2325 struct vnode *vp;
2326 struct flock lf;
2327 int vfslocked;
2328 int error;
2329
2330 if ((error = fget(td, uap->fd, &fp)) != 0)
2331 return (error);
2332 if (fp->f_type != DTYPE_VNODE) {
2333 fdrop(fp, td);
2334 return (EOPNOTSUPP);
2335 }
2336
2337 vp = fp->f_vnode;
2338 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2339 lf.l_whence = SEEK_SET;
2340 lf.l_start = 0;
2341 lf.l_len = 0;
2342 if (uap->how & LOCK_UN) {
2343 lf.l_type = F_UNLCK;
2344 FILE_LOCK(fp);
2345 fp->f_flag &= ~FHASLOCK;
2346 FILE_UNLOCK(fp);
2347 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2348 goto done2;
2349 }
2350 if (uap->how & LOCK_EX)
2351 lf.l_type = F_WRLCK;
2352 else if (uap->how & LOCK_SH)
2353 lf.l_type = F_RDLCK;
2354 else {
2355 error = EBADF;
2356 goto done2;
2357 }
2358 FILE_LOCK(fp);
2359 fp->f_flag |= FHASLOCK;
2360 FILE_UNLOCK(fp);
2361 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2362 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2363 done2:
2364 fdrop(fp, td);
2365 VFS_UNLOCK_GIANT(vfslocked);
2366 return (error);
2367 }
2368 /*
2369 * Duplicate the specified descriptor to a free descriptor.
2370 */
2371 int
2372 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2373 {
2374 struct file *wfp;
2375 struct file *fp;
2376
2377 /*
2378 * If the to-be-dup'd fd number is greater than the allowed number
2379 * of file descriptors, or the fd to be dup'd has already been
2380 * closed, then reject.
2381 */
2382 FILEDESC_XLOCK(fdp);
2383 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2384 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2385 FILEDESC_XUNLOCK(fdp);
2386 return (EBADF);
2387 }
2388
2389 /*
2390 * There are two cases of interest here.
2391 *
2392 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2393 *
2394 * For ENXIO steal away the file structure from (dfd) and store it in
2395 * (indx). (dfd) is effectively closed by this operation.
2396 *
2397 * Any other error code is just returned.
2398 */
2399 switch (error) {
2400 case ENODEV:
2401 /*
2402 * Check that the mode the file is being opened for is a
2403 * subset of the mode of the existing descriptor.
2404 */
2405 FILE_LOCK(wfp);
2406 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2407 FILE_UNLOCK(wfp);
2408 FILEDESC_XUNLOCK(fdp);
2409 return (EACCES);
2410 }
2411 fp = fdp->fd_ofiles[indx];
2412 fdp->fd_ofiles[indx] = wfp;
2413 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2414 if (fp == NULL)
2415 fdused(fdp, indx);
2416 fhold_locked(wfp);
2417 FILE_UNLOCK(wfp);
2418 FILEDESC_XUNLOCK(fdp);
2419 if (fp != NULL)
2420 /*
2421 * We now own the reference to fp that the ofiles[]
2422 * array used to own. Release it.
2423 */
2424 fdrop(fp, td);
2425 return (0);
2426
2427 case ENXIO:
2428 /*
2429 * Steal away the file pointer from dfd and stuff it into indx.
2430 */
2431 fp = fdp->fd_ofiles[indx];
2432 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2433 fdp->fd_ofiles[dfd] = NULL;
2434 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2435 fdp->fd_ofileflags[dfd] = 0;
2436 fdunused(fdp, dfd);
2437 if (fp == NULL)
2438 fdused(fdp, indx);
2439 FILEDESC_XUNLOCK(fdp);
2440
2441 /*
2442 * We now own the reference to fp that the ofiles[] array
2443 * used to own. Release it.
2444 */
2445 if (fp != NULL)
2446 fdrop(fp, td);
2447 return (0);
2448
2449 default:
2450 FILEDESC_XUNLOCK(fdp);
2451 return (error);
2452 }
2453 /* NOTREACHED */
2454 }
2455
2456 /*
2457 * Scan all active processes to see if any of them have a current or root
2458 * directory of `olddp'. If so, replace them with the new mount point.
2459 */
2460 void
2461 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2462 {
2463 struct filedesc *fdp;
2464 struct proc *p;
2465 int nrele;
2466
2467 if (vrefcnt(olddp) == 1)
2468 return;
2469 sx_slock(&allproc_lock);
2470 FOREACH_PROC_IN_SYSTEM(p) {
2471 fdp = fdhold(p);
2472 if (fdp == NULL)
2473 continue;
2474 nrele = 0;
2475 FILEDESC_XLOCK(fdp);
2476 if (fdp->fd_cdir == olddp) {
2477 vref(newdp);
2478 fdp->fd_cdir = newdp;
2479 nrele++;
2480 }
2481 if (fdp->fd_rdir == olddp) {
2482 vref(newdp);
2483 fdp->fd_rdir = newdp;
2484 nrele++;
2485 }
2486 FILEDESC_XUNLOCK(fdp);
2487 fddrop(fdp);
2488 while (nrele--)
2489 vrele(olddp);
2490 }
2491 sx_sunlock(&allproc_lock);
2492 if (rootvnode == olddp) {
2493 vrele(rootvnode);
2494 vref(newdp);
2495 rootvnode = newdp;
2496 }
2497 }
2498
2499 struct filedesc_to_leader *
2500 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2501 {
2502 struct filedesc_to_leader *fdtol;
2503
2504 MALLOC(fdtol, struct filedesc_to_leader *,
2505 sizeof(struct filedesc_to_leader),
2506 M_FILEDESC_TO_LEADER,
2507 M_WAITOK);
2508 fdtol->fdl_refcount = 1;
2509 fdtol->fdl_holdcount = 0;
2510 fdtol->fdl_wakeup = 0;
2511 fdtol->fdl_leader = leader;
2512 if (old != NULL) {
2513 FILEDESC_XLOCK(fdp);
2514 fdtol->fdl_next = old->fdl_next;
2515 fdtol->fdl_prev = old;
2516 old->fdl_next = fdtol;
2517 fdtol->fdl_next->fdl_prev = fdtol;
2518 FILEDESC_XUNLOCK(fdp);
2519 } else {
2520 fdtol->fdl_next = fdtol;
2521 fdtol->fdl_prev = fdtol;
2522 }
2523 return (fdtol);
2524 }
2525
2526 /*
2527 * Get file structures globally.
2528 */
2529 static int
2530 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2531 {
2532 struct xfile xf;
2533 struct filedesc *fdp;
2534 struct file *fp;
2535 struct proc *p;
2536 int error, n;
2537
2538 /*
2539 * Note: because the number of file descriptors is calculated
2540 * in different ways for sizing vs returning the data,
2541 * there is information leakage from the first loop. However,
2542 * it is of a similar order of magnitude to the leakage from
2543 * global system statistics such as kern.openfiles.
2544 */
2545 error = sysctl_wire_old_buffer(req, 0);
2546 if (error != 0)
2547 return (error);
2548 if (req->oldptr == NULL) {
2549 n = 16; /* A slight overestimate. */
2550 sx_slock(&filelist_lock);
2551 LIST_FOREACH(fp, &filehead, f_list) {
2552 /*
2553 * We should grab the lock, but this is an
2554 * estimate, so does it really matter?
2555 */
2556 /* mtx_lock(fp->f_mtxp); */
2557 n += fp->f_count;
2558 /* mtx_unlock(f->f_mtxp); */
2559 }
2560 sx_sunlock(&filelist_lock);
2561 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2562 }
2563 error = 0;
2564 bzero(&xf, sizeof(xf));
2565 xf.xf_size = sizeof(xf);
2566 sx_slock(&allproc_lock);
2567 FOREACH_PROC_IN_SYSTEM(p) {
2568 if (p->p_state == PRS_NEW)
2569 continue;
2570 PROC_LOCK(p);
2571 if (p_cansee(req->td, p) != 0) {
2572 PROC_UNLOCK(p);
2573 continue;
2574 }
2575 xf.xf_pid = p->p_pid;
2576 xf.xf_uid = p->p_ucred->cr_uid;
2577 PROC_UNLOCK(p);
2578 fdp = fdhold(p);
2579 if (fdp == NULL)
2580 continue;
2581 FILEDESC_SLOCK(fdp);
2582 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2583 if ((fp = fdp->fd_ofiles[n]) == NULL)
2584 continue;
2585 xf.xf_fd = n;
2586 xf.xf_file = fp;
2587 xf.xf_data = fp->f_data;
2588 xf.xf_vnode = fp->f_vnode;
2589 xf.xf_type = fp->f_type;
2590 xf.xf_count = fp->f_count;
2591 xf.xf_msgcount = fp->f_msgcount;
2592 xf.xf_offset = fp->f_offset;
2593 xf.xf_flag = fp->f_flag;
2594 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2595 if (error)
2596 break;
2597 }
2598 FILEDESC_SUNLOCK(fdp);
2599 fddrop(fdp);
2600 if (error)
2601 break;
2602 }
2603 sx_sunlock(&allproc_lock);
2604 return (error);
2605 }
2606
2607 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2608 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2609
2610 #ifdef KINFO_OFILE_SIZE
2611 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2612 #endif
2613
2614 /* Compatability with early 7-stable */
2615 static int
2616 export_vnode_for_osysctl(struct vnode *vp, int type,
2617 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2618 {
2619 int error;
2620 char *fullpath, *freepath;
2621 int vfslocked;
2622
2623 bzero(kif, sizeof(*kif));
2624 kif->kf_structsize = sizeof(*kif);
2625
2626 vref(vp);
2627 kif->kf_fd = type;
2628 kif->kf_type = KF_TYPE_VNODE;
2629 /* This function only handles directories. */
2630 KASSERT(vp->v_type == VDIR, ("export_vnode_for_osysctl: vnode not directory"));
2631 kif->kf_vnode_type = KF_VTYPE_VDIR;
2632
2633 /*
2634 * This is not a true file descriptor, so we set a bogus refcount
2635 * and offset to indicate these fields should be ignored.
2636 */
2637 kif->kf_ref_count = -1;
2638 kif->kf_offset = -1;
2639
2640 freepath = NULL;
2641 fullpath = "-";
2642 FILEDESC_SUNLOCK(fdp);
2643 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2644 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
2645 vn_fullpath(curthread, vp, &fullpath, &freepath);
2646 vput(vp);
2647 VFS_UNLOCK_GIANT(vfslocked);
2648 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2649 if (freepath != NULL)
2650 free(freepath, M_TEMP);
2651 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2652 FILEDESC_SLOCK(fdp);
2653 return (error);
2654 }
2655
2656 /*
2657 * Get per-process file descriptors for use by procstat(1), et al.
2658 */
2659 static int
2660 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
2661 {
2662 char *fullpath, *freepath;
2663 struct kinfo_ofile *kif;
2664 struct filedesc *fdp;
2665 int error, i, *name;
2666 struct socket *so;
2667 struct vnode *vp;
2668 struct file *fp;
2669 struct proc *p;
2670 int vfslocked;
2671
2672 name = (int *)arg1;
2673 if ((p = pfind((pid_t)name[0])) == NULL)
2674 return (ESRCH);
2675 if ((error = p_candebug(curthread, p))) {
2676 PROC_UNLOCK(p);
2677 return (error);
2678 }
2679 fdp = fdhold(p);
2680 PROC_UNLOCK(p);
2681 if (fdp == NULL)
2682 return (ENOENT);
2683 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2684 FILEDESC_SLOCK(fdp);
2685 if (fdp->fd_cdir != NULL)
2686 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2687 fdp, req);
2688 if (fdp->fd_rdir != NULL)
2689 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2690 fdp, req);
2691 if (fdp->fd_jdir != NULL)
2692 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2693 fdp, req);
2694 for (i = 0; i < fdp->fd_nfiles; i++) {
2695 if ((fp = fdp->fd_ofiles[i]) == NULL)
2696 continue;
2697 bzero(kif, sizeof(*kif));
2698 kif->kf_structsize = sizeof(*kif);
2699 FILE_LOCK(fp);
2700 vp = NULL;
2701 so = NULL;
2702 kif->kf_fd = i;
2703 switch (fp->f_type) {
2704 case DTYPE_VNODE:
2705 kif->kf_type = KF_TYPE_VNODE;
2706 vp = fp->f_vnode;
2707 vref(vp);
2708 break;
2709
2710 case DTYPE_SOCKET:
2711 kif->kf_type = KF_TYPE_SOCKET;
2712 so = fp->f_data;
2713 break;
2714
2715 case DTYPE_PIPE:
2716 kif->kf_type = KF_TYPE_PIPE;
2717 break;
2718
2719 case DTYPE_FIFO:
2720 kif->kf_type = KF_TYPE_FIFO;
2721 vp = fp->f_vnode;
2722 vref(vp);
2723 break;
2724
2725 case DTYPE_KQUEUE:
2726 kif->kf_type = KF_TYPE_KQUEUE;
2727 break;
2728
2729 case DTYPE_CRYPTO:
2730 kif->kf_type = KF_TYPE_CRYPTO;
2731 break;
2732
2733 case DTYPE_MQUEUE:
2734 kif->kf_type = KF_TYPE_MQUEUE;
2735 break;
2736
2737 default:
2738 kif->kf_type = KF_TYPE_UNKNOWN;
2739 break;
2740 }
2741 kif->kf_ref_count = fp->f_count;
2742 if (fp->f_flag & FREAD)
2743 kif->kf_flags |= KF_FLAG_READ;
2744 if (fp->f_flag & FWRITE)
2745 kif->kf_flags |= KF_FLAG_WRITE;
2746 if (fp->f_flag & FAPPEND)
2747 kif->kf_flags |= KF_FLAG_APPEND;
2748 if (fp->f_flag & FASYNC)
2749 kif->kf_flags |= KF_FLAG_ASYNC;
2750 if (fp->f_flag & FFSYNC)
2751 kif->kf_flags |= KF_FLAG_FSYNC;
2752 if (fp->f_flag & FNONBLOCK)
2753 kif->kf_flags |= KF_FLAG_NONBLOCK;
2754 if (fp->f_flag & O_DIRECT)
2755 kif->kf_flags |= KF_FLAG_DIRECT;
2756 if (fp->f_flag & FHASLOCK)
2757 kif->kf_flags |= KF_FLAG_HASLOCK;
2758 kif->kf_offset = fp->f_offset;
2759 FILE_UNLOCK(fp);
2760 if (vp != NULL) {
2761 switch (vp->v_type) {
2762 case VNON:
2763 kif->kf_vnode_type = KF_VTYPE_VNON;
2764 break;
2765 case VREG:
2766 kif->kf_vnode_type = KF_VTYPE_VREG;
2767 break;
2768 case VDIR:
2769 kif->kf_vnode_type = KF_VTYPE_VDIR;
2770 break;
2771 case VBLK:
2772 kif->kf_vnode_type = KF_VTYPE_VBLK;
2773 break;
2774 case VCHR:
2775 kif->kf_vnode_type = KF_VTYPE_VCHR;
2776 break;
2777 case VLNK:
2778 kif->kf_vnode_type = KF_VTYPE_VLNK;
2779 break;
2780 case VSOCK:
2781 kif->kf_vnode_type = KF_VTYPE_VSOCK;
2782 break;
2783 case VFIFO:
2784 kif->kf_vnode_type = KF_VTYPE_VFIFO;
2785 break;
2786 case VBAD:
2787 kif->kf_vnode_type = KF_VTYPE_VBAD;
2788 break;
2789 default:
2790 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2791 break;
2792 }
2793 /*
2794 * It is OK to drop the filedesc lock here as we will
2795 * re-validate and re-evaluate its properties when
2796 * the loop continues.
2797 */
2798 freepath = NULL;
2799 fullpath = "-";
2800 FILEDESC_SUNLOCK(fdp);
2801 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2802 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
2803 vn_fullpath(curthread, vp, &fullpath, &freepath);
2804 vput(vp);
2805 VFS_UNLOCK_GIANT(vfslocked);
2806 strlcpy(kif->kf_path, fullpath,
2807 sizeof(kif->kf_path));
2808 if (freepath != NULL)
2809 free(freepath, M_TEMP);
2810 FILEDESC_SLOCK(fdp);
2811 }
2812 if (so != NULL) {
2813 struct sockaddr *sa;
2814
2815 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
2816 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
2817 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
2818 free(sa, M_SONAME);
2819 }
2820 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
2821 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
2822 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
2823 free(sa, M_SONAME);
2824 }
2825 kif->kf_sock_domain =
2826 so->so_proto->pr_domain->dom_family;
2827 kif->kf_sock_type = so->so_type;
2828 kif->kf_sock_protocol = so->so_proto->pr_protocol;
2829 }
2830 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2831 if (error)
2832 break;
2833 }
2834 FILEDESC_SUNLOCK(fdp);
2835 fddrop(fdp);
2836 free(kif, M_TEMP);
2837 return (0);
2838 }
2839
2840 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
2841 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
2842
2843 #ifdef KINFO_FILE_SIZE
2844 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2845 #endif
2846
2847 static int
2848 export_vnode_for_sysctl(struct vnode *vp, int type,
2849 struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req)
2850 {
2851 int error;
2852 char *fullpath, *freepath;
2853 int vfslocked;
2854
2855 bzero(kif, sizeof(*kif));
2856
2857 vref(vp);
2858 kif->kf_fd = type;
2859 kif->kf_type = KF_TYPE_VNODE;
2860 /* This function only handles directories. */
2861 KASSERT(vp->v_type == VDIR, ("export_vnode_for_sysctl: vnode not directory"));
2862 kif->kf_vnode_type = KF_VTYPE_VDIR;
2863
2864 /*
2865 * This is not a true file descriptor, so we set a bogus refcount
2866 * and offset to indicate these fields should be ignored.
2867 */
2868 kif->kf_ref_count = -1;
2869 kif->kf_offset = -1;
2870
2871 freepath = NULL;
2872 fullpath = "-";
2873 FILEDESC_SUNLOCK(fdp);
2874 vn_fullpath(curthread, vp, &fullpath, &freepath);
2875 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2876 vrele(vp);
2877 VFS_UNLOCK_GIANT(vfslocked);
2878 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2879 if (freepath != NULL)
2880 free(freepath, M_TEMP);
2881 /* Pack record size down */
2882 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
2883 strlen(kif->kf_path) + 1;
2884 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
2885 error = SYSCTL_OUT(req, kif, kif->kf_structsize);
2886 FILEDESC_SLOCK(fdp);
2887 return (error);
2888 }
2889
2890 /*
2891 * Get per-process file descriptors for use by procstat(1), et al.
2892 */
2893 static int
2894 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
2895 {
2896 char *fullpath, *freepath;
2897 struct kinfo_file *kif;
2898 struct filedesc *fdp;
2899 int error, i, *name;
2900 struct socket *so;
2901 struct vnode *vp;
2902 struct file *fp;
2903 struct proc *p;
2904 int vfslocked;
2905 size_t oldidx;
2906
2907 name = (int *)arg1;
2908 if ((p = pfind((pid_t)name[0])) == NULL)
2909 return (ESRCH);
2910 if ((error = p_candebug(curthread, p))) {
2911 PROC_UNLOCK(p);
2912 return (error);
2913 }
2914 fdp = fdhold(p);
2915 PROC_UNLOCK(p);
2916 if (fdp == NULL)
2917 return (ENOENT);
2918 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2919 FILEDESC_SLOCK(fdp);
2920 if (fdp->fd_cdir != NULL)
2921 export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2922 fdp, req);
2923 if (fdp->fd_rdir != NULL)
2924 export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2925 fdp, req);
2926 if (fdp->fd_jdir != NULL)
2927 export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2928 fdp, req);
2929 for (i = 0; i < fdp->fd_nfiles; i++) {
2930 if ((fp = fdp->fd_ofiles[i]) == NULL)
2931 continue;
2932 bzero(kif, sizeof(*kif));
2933 FILE_LOCK(fp);
2934 vp = NULL;
2935 so = NULL;
2936 kif->kf_fd = i;
2937 switch (fp->f_type) {
2938 case DTYPE_VNODE:
2939 kif->kf_type = KF_TYPE_VNODE;
2940 vp = fp->f_vnode;
2941 vref(vp);
2942 break;
2943
2944 case DTYPE_SOCKET:
2945 kif->kf_type = KF_TYPE_SOCKET;
2946 so = fp->f_data;
2947 break;
2948
2949 case DTYPE_PIPE:
2950 kif->kf_type = KF_TYPE_PIPE;
2951 break;
2952
2953 case DTYPE_FIFO:
2954 kif->kf_type = KF_TYPE_FIFO;
2955 vp = fp->f_vnode;
2956 vref(vp);
2957 break;
2958
2959 case DTYPE_KQUEUE:
2960 kif->kf_type = KF_TYPE_KQUEUE;
2961 break;
2962
2963 case DTYPE_CRYPTO:
2964 kif->kf_type = KF_TYPE_CRYPTO;
2965 break;
2966
2967 case DTYPE_MQUEUE:
2968 kif->kf_type = KF_TYPE_MQUEUE;
2969 break;
2970
2971 case DTYPE_SEM:
2972 kif->kf_type = KF_TYPE_SEM;
2973 break;
2974
2975 default:
2976 kif->kf_type = KF_TYPE_UNKNOWN;
2977 break;
2978 }
2979 kif->kf_ref_count = fp->f_count;
2980 if (fp->f_flag & FREAD)
2981 kif->kf_flags |= KF_FLAG_READ;
2982 if (fp->f_flag & FWRITE)
2983 kif->kf_flags |= KF_FLAG_WRITE;
2984 if (fp->f_flag & FAPPEND)
2985 kif->kf_flags |= KF_FLAG_APPEND;
2986 if (fp->f_flag & FASYNC)
2987 kif->kf_flags |= KF_FLAG_ASYNC;
2988 if (fp->f_flag & FFSYNC)
2989 kif->kf_flags |= KF_FLAG_FSYNC;
2990 if (fp->f_flag & FNONBLOCK)
2991 kif->kf_flags |= KF_FLAG_NONBLOCK;
2992 if (fp->f_flag & O_DIRECT)
2993 kif->kf_flags |= KF_FLAG_DIRECT;
2994 if (fp->f_flag & FHASLOCK)
2995 kif->kf_flags |= KF_FLAG_HASLOCK;
2996 kif->kf_offset = fp->f_offset;
2997 FILE_UNLOCK(fp);
2998 if (vp != NULL) {
2999 switch (vp->v_type) {
3000 case VNON:
3001 kif->kf_vnode_type = KF_VTYPE_VNON;
3002 break;
3003 case VREG:
3004 kif->kf_vnode_type = KF_VTYPE_VREG;
3005 break;
3006 case VDIR:
3007 kif->kf_vnode_type = KF_VTYPE_VDIR;
3008 break;
3009 case VBLK:
3010 kif->kf_vnode_type = KF_VTYPE_VBLK;
3011 break;
3012 case VCHR:
3013 kif->kf_vnode_type = KF_VTYPE_VCHR;
3014 break;
3015 case VLNK:
3016 kif->kf_vnode_type = KF_VTYPE_VLNK;
3017 break;
3018 case VSOCK:
3019 kif->kf_vnode_type = KF_VTYPE_VSOCK;
3020 break;
3021 case VFIFO:
3022 kif->kf_vnode_type = KF_VTYPE_VFIFO;
3023 break;
3024 case VBAD:
3025 kif->kf_vnode_type = KF_VTYPE_VBAD;
3026 break;
3027 default:
3028 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3029 break;
3030 }
3031 /*
3032 * It is OK to drop the filedesc lock here as we will
3033 * re-validate and re-evaluate its properties when
3034 * the loop continues.
3035 */
3036 freepath = NULL;
3037 fullpath = "-";
3038 FILEDESC_SUNLOCK(fdp);
3039 vn_fullpath(curthread, vp, &fullpath, &freepath);
3040 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
3041 vrele(vp);
3042 VFS_UNLOCK_GIANT(vfslocked);
3043 strlcpy(kif->kf_path, fullpath,
3044 sizeof(kif->kf_path));
3045 if (freepath != NULL)
3046 free(freepath, M_TEMP);
3047 FILEDESC_SLOCK(fdp);
3048 }
3049 if (so != NULL) {
3050 struct sockaddr *sa;
3051
3052 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3053 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3054 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3055 free(sa, M_SONAME);
3056 }
3057 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3058 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3059 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3060 free(sa, M_SONAME);
3061 }
3062 kif->kf_sock_domain =
3063 so->so_proto->pr_domain->dom_family;
3064 kif->kf_sock_type = so->so_type;
3065 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3066 }
3067 /* Pack record size down */
3068 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3069 strlen(kif->kf_path) + 1;
3070 kif->kf_structsize = roundup(kif->kf_structsize,
3071 sizeof(uint64_t));
3072 oldidx = req->oldidx;
3073 error = SYSCTL_OUT(req, kif, kif->kf_structsize);
3074 if (error) {
3075 if (error == ENOMEM) {
3076 /*
3077 * The hack to keep the ABI of sysctl
3078 * kern.proc.filedesc intact, but not
3079 * to account a partially copied
3080 * kinfo_file into the oldidx.
3081 */
3082 req->oldidx = oldidx;
3083 error = 0;
3084 }
3085 break;
3086 }
3087 }
3088 FILEDESC_SUNLOCK(fdp);
3089 fddrop(fdp);
3090 free(kif, M_TEMP);
3091 return (error);
3092 }
3093
3094 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3095 sysctl_kern_proc_filedesc, "Process filedesc entries");
3096
3097 #ifdef DDB
3098 /*
3099 * For the purposes of debugging, generate a human-readable string for the
3100 * file type.
3101 */
3102 static const char *
3103 file_type_to_name(short type)
3104 {
3105
3106 switch (type) {
3107 case 0:
3108 return ("zero");
3109 case DTYPE_VNODE:
3110 return ("vnod");
3111 case DTYPE_SOCKET:
3112 return ("sock");
3113 case DTYPE_PIPE:
3114 return ("pipe");
3115 case DTYPE_FIFO:
3116 return ("fifo");
3117 case DTYPE_KQUEUE:
3118 return ("kque");
3119 case DTYPE_CRYPTO:
3120 return ("crpt");
3121 case DTYPE_MQUEUE:
3122 return ("mque");
3123 case DTYPE_SEM:
3124 return ("ksem");
3125 default:
3126 return ("unkn");
3127 }
3128 }
3129
3130 /*
3131 * For the purposes of debugging, identify a process (if any, perhaps one of
3132 * many) that references the passed file in its file descriptor array. Return
3133 * NULL if none.
3134 */
3135 static struct proc *
3136 file_to_first_proc(struct file *fp)
3137 {
3138 struct filedesc *fdp;
3139 struct proc *p;
3140 int n;
3141
3142 FOREACH_PROC_IN_SYSTEM(p) {
3143 if (p->p_state == PRS_NEW)
3144 continue;
3145 fdp = p->p_fd;
3146 if (fdp == NULL)
3147 continue;
3148 for (n = 0; n < fdp->fd_nfiles; n++) {
3149 if (fp == fdp->fd_ofiles[n])
3150 return (p);
3151 }
3152 }
3153 return (NULL);
3154 }
3155
3156 static void
3157 db_print_file(struct file *fp, int header)
3158 {
3159 struct proc *p;
3160
3161 if (header)
3162 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3163 "File", "Type", "Data", "Flag", "GCFl", "Count",
3164 "MCount", "Vnode", "FPID", "FCmd");
3165 p = file_to_first_proc(fp);
3166 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3167 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3168 fp->f_gcflag, fp->f_count, fp->f_msgcount, fp->f_vnode,
3169 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3170 }
3171
3172 DB_SHOW_COMMAND(file, db_show_file)
3173 {
3174 struct file *fp;
3175
3176 if (!have_addr) {
3177 db_printf("usage: show file <addr>\n");
3178 return;
3179 }
3180 fp = (struct file *)addr;
3181 db_print_file(fp, 1);
3182 }
3183
3184 DB_SHOW_COMMAND(files, db_show_files)
3185 {
3186 struct file *fp;
3187 int header;
3188
3189 header = 1;
3190 LIST_FOREACH(fp, &filehead, f_list) {
3191 db_print_file(fp, header);
3192 header = 0;
3193 }
3194 }
3195 #endif
3196
3197 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3198 &maxfilesperproc, 0, "Maximum files allowed open per process");
3199
3200 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3201 &maxfiles, 0, "Maximum number of files");
3202
3203 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3204 &openfiles, 0, "System-wide number of open files");
3205
3206 /* ARGSUSED*/
3207 static void
3208 filelistinit(void *dummy)
3209 {
3210
3211 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3212 NULL, NULL, UMA_ALIGN_PTR, 0);
3213 sx_init(&filelist_lock, "filelist lock");
3214 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3215 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3216 }
3217 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3218
3219 /*-------------------------------------------------------------------*/
3220
3221 static int
3222 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
3223 {
3224
3225 return (EBADF);
3226 }
3227
3228 static int
3229 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
3230 {
3231
3232 return (EBADF);
3233 }
3234
3235 static int
3236 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
3237 {
3238
3239 return (0);
3240 }
3241
3242 static int
3243 badfo_kqfilter(struct file *fp, struct knote *kn)
3244 {
3245
3246 return (EBADF);
3247 }
3248
3249 static int
3250 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
3251 {
3252
3253 return (EBADF);
3254 }
3255
3256 static int
3257 badfo_close(struct file *fp, struct thread *td)
3258 {
3259
3260 return (EBADF);
3261 }
3262
3263 struct fileops badfileops = {
3264 .fo_read = badfo_readwrite,
3265 .fo_write = badfo_readwrite,
3266 .fo_ioctl = badfo_ioctl,
3267 .fo_poll = badfo_poll,
3268 .fo_kqfilter = badfo_kqfilter,
3269 .fo_stat = badfo_stat,
3270 .fo_close = badfo_close,
3271 };
3272
3273
3274 /*-------------------------------------------------------------------*/
3275
3276 /*
3277 * File Descriptor pseudo-device driver (/dev/fd/).
3278 *
3279 * Opening minor device N dup()s the file (if any) connected to file
3280 * descriptor N belonging to the calling process. Note that this driver
3281 * consists of only the ``open()'' routine, because all subsequent
3282 * references to this file will be direct to the other driver.
3283 *
3284 * XXX: we could give this one a cloning event handler if necessary.
3285 */
3286
3287 /* ARGSUSED */
3288 static int
3289 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3290 {
3291
3292 /*
3293 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3294 * the file descriptor being sought for duplication. The error
3295 * return ensures that the vnode for this device will be released
3296 * by vn_open. Open will detect this special error and take the
3297 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3298 * will simply report the error.
3299 */
3300 td->td_dupfd = dev2unit(dev);
3301 return (ENODEV);
3302 }
3303
3304 static struct cdevsw fildesc_cdevsw = {
3305 .d_version = D_VERSION,
3306 .d_flags = D_NEEDGIANT,
3307 .d_open = fdopen,
3308 .d_name = "FD",
3309 };
3310
3311 static void
3312 fildesc_drvinit(void *unused)
3313 {
3314 struct cdev *dev;
3315
3316 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
3317 make_dev_alias(dev, "stdin");
3318 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
3319 make_dev_alias(dev, "stdout");
3320 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
3321 make_dev_alias(dev, "stderr");
3322 }
3323
3324 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
Cache object: a89c77e22c31b7b17181acaa2819fdd0
|