1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_compat.h"
41 #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 #if defined(COMPAT_43)
1140 /*
1141 * Return status information about a file descriptor.
1142 */
1143 #ifndef _SYS_SYSPROTO_H_
1144 struct ofstat_args {
1145 int fd;
1146 struct ostat *sb;
1147 };
1148 #endif
1149 /* ARGSUSED */
1150 int
1151 ofstat(struct thread *td, struct ofstat_args *uap)
1152 {
1153 struct ostat oub;
1154 struct stat ub;
1155 int error;
1156
1157 error = kern_fstat(td, uap->fd, &ub);
1158 if (error == 0) {
1159 cvtstat(&ub, &oub);
1160 error = copyout(&oub, uap->sb, sizeof(oub));
1161 }
1162 return (error);
1163 }
1164 #endif /* COMPAT_43 */
1165
1166 /*
1167 * Return status information about a file descriptor.
1168 */
1169 #ifndef _SYS_SYSPROTO_H_
1170 struct fstat_args {
1171 int fd;
1172 struct stat *sb;
1173 };
1174 #endif
1175 /* ARGSUSED */
1176 int
1177 fstat(struct thread *td, struct fstat_args *uap)
1178 {
1179 struct stat ub;
1180 int error;
1181
1182 error = kern_fstat(td, uap->fd, &ub);
1183 if (error == 0)
1184 error = copyout(&ub, uap->sb, sizeof(ub));
1185 return (error);
1186 }
1187
1188 int
1189 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1190 {
1191 struct file *fp;
1192 int error;
1193
1194 AUDIT_ARG(fd, fd);
1195
1196 if ((error = fget(td, fd, &fp)) != 0)
1197 return (error);
1198
1199 AUDIT_ARG(file, td->td_proc, fp);
1200
1201 error = fo_stat(fp, sbp, td->td_ucred, td);
1202 fdrop(fp, td);
1203 #ifdef KTRACE
1204 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1205 ktrstat(sbp);
1206 #endif
1207 return (error);
1208 }
1209
1210 /*
1211 * Return status information about a file descriptor.
1212 */
1213 #ifndef _SYS_SYSPROTO_H_
1214 struct nfstat_args {
1215 int fd;
1216 struct nstat *sb;
1217 };
1218 #endif
1219 /* ARGSUSED */
1220 int
1221 nfstat(struct thread *td, struct nfstat_args *uap)
1222 {
1223 struct nstat nub;
1224 struct stat ub;
1225 int error;
1226
1227 error = kern_fstat(td, uap->fd, &ub);
1228 if (error == 0) {
1229 cvtnstat(&ub, &nub);
1230 error = copyout(&nub, uap->sb, sizeof(nub));
1231 }
1232 return (error);
1233 }
1234
1235 /*
1236 * Return pathconf information about a file descriptor.
1237 */
1238 #ifndef _SYS_SYSPROTO_H_
1239 struct fpathconf_args {
1240 int fd;
1241 int name;
1242 };
1243 #endif
1244 /* ARGSUSED */
1245 int
1246 fpathconf(struct thread *td, struct fpathconf_args *uap)
1247 {
1248 struct file *fp;
1249 struct vnode *vp;
1250 int error;
1251
1252 if ((error = fget(td, uap->fd, &fp)) != 0)
1253 return (error);
1254
1255 /* If asynchronous I/O is available, it works for all descriptors. */
1256 if (uap->name == _PC_ASYNC_IO) {
1257 td->td_retval[0] = async_io_version;
1258 goto out;
1259 }
1260 vp = fp->f_vnode;
1261 if (vp != NULL) {
1262 int vfslocked;
1263 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1264 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1265 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1266 VOP_UNLOCK(vp, 0, td);
1267 VFS_UNLOCK_GIANT(vfslocked);
1268 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1269 if (uap->name != _PC_PIPE_BUF) {
1270 error = EINVAL;
1271 } else {
1272 td->td_retval[0] = PIPE_BUF;
1273 error = 0;
1274 }
1275 } else {
1276 error = EOPNOTSUPP;
1277 }
1278 out:
1279 fdrop(fp, td);
1280 return (error);
1281 }
1282
1283 /*
1284 * Grow the file table to accomodate (at least) nfd descriptors. This may
1285 * block and drop the filedesc lock, but it will reacquire it before
1286 * returning.
1287 */
1288 static void
1289 fdgrowtable(struct filedesc *fdp, int nfd)
1290 {
1291 struct file **ntable;
1292 char *nfileflags;
1293 int nnfiles, onfiles;
1294 NDSLOTTYPE *nmap;
1295
1296 FILEDESC_XLOCK_ASSERT(fdp);
1297
1298 KASSERT(fdp->fd_nfiles > 0,
1299 ("zero-length file table"));
1300
1301 /* compute the size of the new table */
1302 onfiles = fdp->fd_nfiles;
1303 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1304 if (nnfiles <= onfiles)
1305 /* the table is already large enough */
1306 return;
1307
1308 /* allocate a new table and (if required) new bitmaps */
1309 FILEDESC_XUNLOCK(fdp);
1310 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1311 M_FILEDESC, M_ZERO | M_WAITOK);
1312 nfileflags = (char *)&ntable[nnfiles];
1313 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1314 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1315 M_FILEDESC, M_ZERO | M_WAITOK);
1316 else
1317 nmap = NULL;
1318 FILEDESC_XLOCK(fdp);
1319
1320 /*
1321 * We now have new tables ready to go. Since we dropped the
1322 * filedesc lock to call malloc(), watch out for a race.
1323 */
1324 onfiles = fdp->fd_nfiles;
1325 if (onfiles >= nnfiles) {
1326 /* we lost the race, but that's OK */
1327 free(ntable, M_FILEDESC);
1328 if (nmap != NULL)
1329 free(nmap, M_FILEDESC);
1330 return;
1331 }
1332 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1333 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1334 if (onfiles > NDFILE)
1335 free(fdp->fd_ofiles, M_FILEDESC);
1336 fdp->fd_ofiles = ntable;
1337 fdp->fd_ofileflags = nfileflags;
1338 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1339 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1340 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1341 free(fdp->fd_map, M_FILEDESC);
1342 fdp->fd_map = nmap;
1343 }
1344 fdp->fd_nfiles = nnfiles;
1345 }
1346
1347 /*
1348 * Allocate a file descriptor for the process.
1349 */
1350 int
1351 fdalloc(struct thread *td, int minfd, int *result)
1352 {
1353 struct proc *p = td->td_proc;
1354 struct filedesc *fdp = p->p_fd;
1355 int fd = -1, maxfd;
1356
1357 FILEDESC_XLOCK_ASSERT(fdp);
1358
1359 if (fdp->fd_freefile > minfd)
1360 minfd = fdp->fd_freefile;
1361
1362 PROC_LOCK(p);
1363 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1364 PROC_UNLOCK(p);
1365
1366 /*
1367 * Search the bitmap for a free descriptor. If none is found, try
1368 * to grow the file table. Keep at it until we either get a file
1369 * descriptor or run into process or system limits; fdgrowtable()
1370 * may drop the filedesc lock, so we're in a race.
1371 */
1372 for (;;) {
1373 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1374 if (fd >= maxfd)
1375 return (EMFILE);
1376 if (fd < fdp->fd_nfiles)
1377 break;
1378 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1379 }
1380
1381 /*
1382 * Perform some sanity checks, then mark the file descriptor as
1383 * used and return it to the caller.
1384 */
1385 KASSERT(!fdisused(fdp, fd),
1386 ("fd_first_free() returned non-free descriptor"));
1387 KASSERT(fdp->fd_ofiles[fd] == NULL,
1388 ("free descriptor isn't"));
1389 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1390 fdused(fdp, fd);
1391 *result = fd;
1392 return (0);
1393 }
1394
1395 /*
1396 * Check to see whether n user file descriptors are available to the process
1397 * p.
1398 */
1399 int
1400 fdavail(struct thread *td, int n)
1401 {
1402 struct proc *p = td->td_proc;
1403 struct filedesc *fdp = td->td_proc->p_fd;
1404 struct file **fpp;
1405 int i, lim, last;
1406
1407 FILEDESC_LOCK_ASSERT(fdp);
1408
1409 PROC_LOCK(p);
1410 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1411 PROC_UNLOCK(p);
1412 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1413 return (1);
1414 last = min(fdp->fd_nfiles, lim);
1415 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1416 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1417 if (*fpp == NULL && --n <= 0)
1418 return (1);
1419 }
1420 return (0);
1421 }
1422
1423 /*
1424 * Create a new open file structure and allocate a file decriptor for the
1425 * process that refers to it. We add one reference to the file for the
1426 * descriptor table and one reference for resultfp. This is to prevent us
1427 * being preempted and the entry in the descriptor table closed after we
1428 * release the FILEDESC lock.
1429 */
1430 int
1431 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1432 {
1433 struct proc *p = td->td_proc;
1434 struct file *fp, *fq;
1435 int error, i;
1436 int maxuserfiles = maxfiles - (maxfiles / 20);
1437 static struct timeval lastfail;
1438 static int curfail;
1439
1440 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1441 sx_xlock(&filelist_lock);
1442
1443 if ((openfiles >= maxuserfiles &&
1444 priv_check(td, PRIV_MAXFILES) != 0) ||
1445 openfiles >= maxfiles) {
1446 if (ppsratecheck(&lastfail, &curfail, 1)) {
1447 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1448 td->td_ucred->cr_ruid);
1449 }
1450 sx_xunlock(&filelist_lock);
1451 uma_zfree(file_zone, fp);
1452 return (ENFILE);
1453 }
1454 openfiles++;
1455
1456 /*
1457 * If the process has file descriptor zero open, add the new file
1458 * descriptor to the list of open files at that point, otherwise
1459 * put it at the front of the list of open files.
1460 */
1461 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1462 fp->f_count = 1;
1463 if (resultfp)
1464 fp->f_count++;
1465 fp->f_cred = crhold(td->td_ucred);
1466 fp->f_ops = &badfileops;
1467 fp->f_data = NULL;
1468 fp->f_vnode = NULL;
1469 FILEDESC_XLOCK(p->p_fd);
1470 if ((fq = p->p_fd->fd_ofiles[0])) {
1471 LIST_INSERT_AFTER(fq, fp, f_list);
1472 } else {
1473 LIST_INSERT_HEAD(&filehead, fp, f_list);
1474 }
1475 sx_xunlock(&filelist_lock);
1476 if ((error = fdalloc(td, 0, &i))) {
1477 FILEDESC_XUNLOCK(p->p_fd);
1478 fdrop(fp, td);
1479 if (resultfp)
1480 fdrop(fp, td);
1481 return (error);
1482 }
1483 p->p_fd->fd_ofiles[i] = fp;
1484 FILEDESC_XUNLOCK(p->p_fd);
1485 if (resultfp)
1486 *resultfp = fp;
1487 if (resultfd)
1488 *resultfd = i;
1489 return (0);
1490 }
1491
1492 /*
1493 * Build a new filedesc structure from another.
1494 * Copy the current, root, and jail root vnode references.
1495 */
1496 struct filedesc *
1497 fdinit(struct filedesc *fdp)
1498 {
1499 struct filedesc0 *newfdp;
1500
1501 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1502 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1503 if (fdp != NULL) {
1504 FILEDESC_XLOCK(fdp);
1505 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1506 if (newfdp->fd_fd.fd_cdir)
1507 VREF(newfdp->fd_fd.fd_cdir);
1508 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1509 if (newfdp->fd_fd.fd_rdir)
1510 VREF(newfdp->fd_fd.fd_rdir);
1511 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1512 if (newfdp->fd_fd.fd_jdir)
1513 VREF(newfdp->fd_fd.fd_jdir);
1514 FILEDESC_XUNLOCK(fdp);
1515 }
1516
1517 /* Create the file descriptor table. */
1518 newfdp->fd_fd.fd_refcnt = 1;
1519 newfdp->fd_fd.fd_holdcnt = 1;
1520 newfdp->fd_fd.fd_cmask = CMASK;
1521 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1522 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1523 newfdp->fd_fd.fd_nfiles = NDFILE;
1524 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1525 newfdp->fd_fd.fd_lastfile = -1;
1526 return (&newfdp->fd_fd);
1527 }
1528
1529 static struct filedesc *
1530 fdhold(struct proc *p)
1531 {
1532 struct filedesc *fdp;
1533
1534 mtx_lock(&fdesc_mtx);
1535 fdp = p->p_fd;
1536 if (fdp != NULL)
1537 fdp->fd_holdcnt++;
1538 mtx_unlock(&fdesc_mtx);
1539 return (fdp);
1540 }
1541
1542 static void
1543 fddrop(struct filedesc *fdp)
1544 {
1545 int i;
1546
1547 mtx_lock(&fdesc_mtx);
1548 i = --fdp->fd_holdcnt;
1549 mtx_unlock(&fdesc_mtx);
1550 if (i > 0)
1551 return;
1552
1553 FILEDESC_LOCK_DESTROY(fdp);
1554 FREE(fdp, M_FILEDESC);
1555 }
1556
1557 /*
1558 * Share a filedesc structure.
1559 */
1560 struct filedesc *
1561 fdshare(struct filedesc *fdp)
1562 {
1563
1564 FILEDESC_XLOCK(fdp);
1565 fdp->fd_refcnt++;
1566 FILEDESC_XUNLOCK(fdp);
1567 return (fdp);
1568 }
1569
1570 /*
1571 * Unshare a filedesc structure, if necessary by making a copy
1572 */
1573 void
1574 fdunshare(struct proc *p, struct thread *td)
1575 {
1576
1577 FILEDESC_XLOCK(p->p_fd);
1578 if (p->p_fd->fd_refcnt > 1) {
1579 struct filedesc *tmp;
1580
1581 FILEDESC_XUNLOCK(p->p_fd);
1582 tmp = fdcopy(p->p_fd);
1583 fdfree(td);
1584 p->p_fd = tmp;
1585 } else
1586 FILEDESC_XUNLOCK(p->p_fd);
1587 }
1588
1589 /*
1590 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1591 * this is to ease callers, not catch errors.
1592 */
1593 struct filedesc *
1594 fdcopy(struct filedesc *fdp)
1595 {
1596 struct filedesc *newfdp;
1597 int i;
1598
1599 /* Certain daemons might not have file descriptors. */
1600 if (fdp == NULL)
1601 return (NULL);
1602
1603 newfdp = fdinit(fdp);
1604 FILEDESC_SLOCK(fdp);
1605 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1606 FILEDESC_SUNLOCK(fdp);
1607 FILEDESC_XLOCK(newfdp);
1608 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1609 FILEDESC_XUNLOCK(newfdp);
1610 FILEDESC_SLOCK(fdp);
1611 }
1612 /* copy everything except kqueue descriptors */
1613 newfdp->fd_freefile = -1;
1614 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1615 if (fdisused(fdp, i) &&
1616 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1617 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1618 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1619 fhold(newfdp->fd_ofiles[i]);
1620 newfdp->fd_lastfile = i;
1621 } else {
1622 if (newfdp->fd_freefile == -1)
1623 newfdp->fd_freefile = i;
1624 }
1625 }
1626 FILEDESC_SUNLOCK(fdp);
1627 FILEDESC_XLOCK(newfdp);
1628 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1629 if (newfdp->fd_ofiles[i] != NULL)
1630 fdused(newfdp, i);
1631 FILEDESC_XUNLOCK(newfdp);
1632 FILEDESC_SLOCK(fdp);
1633 if (newfdp->fd_freefile == -1)
1634 newfdp->fd_freefile = i;
1635 newfdp->fd_cmask = fdp->fd_cmask;
1636 FILEDESC_SUNLOCK(fdp);
1637 return (newfdp);
1638 }
1639
1640 /*
1641 * Release a filedesc structure.
1642 */
1643 void
1644 fdfree(struct thread *td)
1645 {
1646 struct filedesc *fdp;
1647 struct file **fpp;
1648 int i, locked;
1649 struct filedesc_to_leader *fdtol;
1650 struct file *fp;
1651 struct vnode *cdir, *jdir, *rdir, *vp;
1652 struct flock lf;
1653
1654 /* Certain daemons might not have file descriptors. */
1655 fdp = td->td_proc->p_fd;
1656 if (fdp == NULL)
1657 return;
1658
1659 /* Check for special need to clear POSIX style locks */
1660 fdtol = td->td_proc->p_fdtol;
1661 if (fdtol != NULL) {
1662 FILEDESC_XLOCK(fdp);
1663 KASSERT(fdtol->fdl_refcount > 0,
1664 ("filedesc_to_refcount botch: fdl_refcount=%d",
1665 fdtol->fdl_refcount));
1666 if (fdtol->fdl_refcount == 1 &&
1667 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1668 for (i = 0, fpp = fdp->fd_ofiles;
1669 i <= fdp->fd_lastfile;
1670 i++, fpp++) {
1671 if (*fpp == NULL ||
1672 (*fpp)->f_type != DTYPE_VNODE)
1673 continue;
1674 fp = *fpp;
1675 fhold(fp);
1676 FILEDESC_XUNLOCK(fdp);
1677 lf.l_whence = SEEK_SET;
1678 lf.l_start = 0;
1679 lf.l_len = 0;
1680 lf.l_type = F_UNLCK;
1681 vp = fp->f_vnode;
1682 locked = VFS_LOCK_GIANT(vp->v_mount);
1683 (void) VOP_ADVLOCK(vp,
1684 (caddr_t)td->td_proc->
1685 p_leader,
1686 F_UNLCK,
1687 &lf,
1688 F_POSIX);
1689 VFS_UNLOCK_GIANT(locked);
1690 FILEDESC_XLOCK(fdp);
1691 fdrop(fp, td);
1692 fpp = fdp->fd_ofiles + i;
1693 }
1694 }
1695 retry:
1696 if (fdtol->fdl_refcount == 1) {
1697 if (fdp->fd_holdleaderscount > 0 &&
1698 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1699 /*
1700 * close() or do_dup() has cleared a reference
1701 * in a shared file descriptor table.
1702 */
1703 fdp->fd_holdleaderswakeup = 1;
1704 sx_sleep(&fdp->fd_holdleaderscount,
1705 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1706 goto retry;
1707 }
1708 if (fdtol->fdl_holdcount > 0) {
1709 /*
1710 * Ensure that fdtol->fdl_leader remains
1711 * valid in closef().
1712 */
1713 fdtol->fdl_wakeup = 1;
1714 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1715 "fdlhold", 0);
1716 goto retry;
1717 }
1718 }
1719 fdtol->fdl_refcount--;
1720 if (fdtol->fdl_refcount == 0 &&
1721 fdtol->fdl_holdcount == 0) {
1722 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1723 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1724 } else
1725 fdtol = NULL;
1726 td->td_proc->p_fdtol = NULL;
1727 FILEDESC_XUNLOCK(fdp);
1728 if (fdtol != NULL)
1729 FREE(fdtol, M_FILEDESC_TO_LEADER);
1730 }
1731 FILEDESC_XLOCK(fdp);
1732 i = --fdp->fd_refcnt;
1733 FILEDESC_XUNLOCK(fdp);
1734 if (i > 0)
1735 return;
1736 /*
1737 * We are the last reference to the structure, so we can
1738 * safely assume it will not change out from under us.
1739 */
1740 fpp = fdp->fd_ofiles;
1741 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1742 if (*fpp)
1743 (void) closef(*fpp, td);
1744 }
1745 FILEDESC_XLOCK(fdp);
1746
1747 /* XXX This should happen earlier. */
1748 mtx_lock(&fdesc_mtx);
1749 td->td_proc->p_fd = NULL;
1750 mtx_unlock(&fdesc_mtx);
1751
1752 if (fdp->fd_nfiles > NDFILE)
1753 FREE(fdp->fd_ofiles, M_FILEDESC);
1754 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1755 FREE(fdp->fd_map, M_FILEDESC);
1756
1757 fdp->fd_nfiles = 0;
1758
1759 cdir = fdp->fd_cdir;
1760 fdp->fd_cdir = NULL;
1761 rdir = fdp->fd_rdir;
1762 fdp->fd_rdir = NULL;
1763 jdir = fdp->fd_jdir;
1764 fdp->fd_jdir = NULL;
1765 FILEDESC_XUNLOCK(fdp);
1766
1767 if (cdir) {
1768 locked = VFS_LOCK_GIANT(cdir->v_mount);
1769 vrele(cdir);
1770 VFS_UNLOCK_GIANT(locked);
1771 }
1772 if (rdir) {
1773 locked = VFS_LOCK_GIANT(rdir->v_mount);
1774 vrele(rdir);
1775 VFS_UNLOCK_GIANT(locked);
1776 }
1777 if (jdir) {
1778 locked = VFS_LOCK_GIANT(jdir->v_mount);
1779 vrele(jdir);
1780 VFS_UNLOCK_GIANT(locked);
1781 }
1782
1783 fddrop(fdp);
1784 }
1785
1786 /*
1787 * For setugid programs, we don't want to people to use that setugidness
1788 * to generate error messages which write to a file which otherwise would
1789 * otherwise be off-limits to the process. We check for filesystems where
1790 * the vnode can change out from under us after execve (like [lin]procfs).
1791 *
1792 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1793 * sufficient. We also don't check for setugidness since we know we are.
1794 */
1795 static int
1796 is_unsafe(struct file *fp)
1797 {
1798 if (fp->f_type == DTYPE_VNODE) {
1799 struct vnode *vp = fp->f_vnode;
1800
1801 if ((vp->v_vflag & VV_PROCDEP) != 0)
1802 return (1);
1803 }
1804 return (0);
1805 }
1806
1807 /*
1808 * Make this setguid thing safe, if at all possible.
1809 */
1810 void
1811 setugidsafety(struct thread *td)
1812 {
1813 struct filedesc *fdp;
1814 int i;
1815
1816 /* Certain daemons might not have file descriptors. */
1817 fdp = td->td_proc->p_fd;
1818 if (fdp == NULL)
1819 return;
1820
1821 /*
1822 * Note: fdp->fd_ofiles may be reallocated out from under us while
1823 * we are blocked in a close. Be careful!
1824 */
1825 FILEDESC_XLOCK(fdp);
1826 for (i = 0; i <= fdp->fd_lastfile; i++) {
1827 if (i > 2)
1828 break;
1829 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1830 struct file *fp;
1831
1832 knote_fdclose(td, i);
1833 /*
1834 * NULL-out descriptor prior to close to avoid
1835 * a race while close blocks.
1836 */
1837 fp = fdp->fd_ofiles[i];
1838 fdp->fd_ofiles[i] = NULL;
1839 fdp->fd_ofileflags[i] = 0;
1840 fdunused(fdp, i);
1841 FILEDESC_XUNLOCK(fdp);
1842 (void) closef(fp, td);
1843 FILEDESC_XLOCK(fdp);
1844 }
1845 }
1846 FILEDESC_XUNLOCK(fdp);
1847 }
1848
1849 /*
1850 * If a specific file object occupies a specific file descriptor, close the
1851 * file descriptor entry and drop a reference on the file object. This is a
1852 * convenience function to handle a subsequent error in a function that calls
1853 * falloc() that handles the race that another thread might have closed the
1854 * file descriptor out from under the thread creating the file object.
1855 */
1856 void
1857 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1858 {
1859
1860 FILEDESC_XLOCK(fdp);
1861 if (fdp->fd_ofiles[idx] == fp) {
1862 fdp->fd_ofiles[idx] = NULL;
1863 fdunused(fdp, idx);
1864 FILEDESC_XUNLOCK(fdp);
1865 fdrop(fp, td);
1866 } else
1867 FILEDESC_XUNLOCK(fdp);
1868 }
1869
1870 /*
1871 * Close any files on exec?
1872 */
1873 void
1874 fdcloseexec(struct thread *td)
1875 {
1876 struct filedesc *fdp;
1877 int i;
1878
1879 /* Certain daemons might not have file descriptors. */
1880 fdp = td->td_proc->p_fd;
1881 if (fdp == NULL)
1882 return;
1883
1884 FILEDESC_XLOCK(fdp);
1885
1886 /*
1887 * We cannot cache fd_ofiles or fd_ofileflags since operations
1888 * may block and rip them out from under us.
1889 */
1890 for (i = 0; i <= fdp->fd_lastfile; i++) {
1891 if (fdp->fd_ofiles[i] != NULL &&
1892 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1893 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1894 struct file *fp;
1895
1896 knote_fdclose(td, i);
1897 /*
1898 * NULL-out descriptor prior to close to avoid
1899 * a race while close blocks.
1900 */
1901 fp = fdp->fd_ofiles[i];
1902 fdp->fd_ofiles[i] = NULL;
1903 fdp->fd_ofileflags[i] = 0;
1904 fdunused(fdp, i);
1905 if (fp->f_type == DTYPE_MQUEUE)
1906 mq_fdclose(td, i, fp);
1907 FILEDESC_XUNLOCK(fdp);
1908 (void) closef(fp, td);
1909 FILEDESC_XLOCK(fdp);
1910 }
1911 }
1912 FILEDESC_XUNLOCK(fdp);
1913 }
1914
1915 /*
1916 * It is unsafe for set[ug]id processes to be started with file
1917 * descriptors 0..2 closed, as these descriptors are given implicit
1918 * significance in the Standard C library. fdcheckstd() will create a
1919 * descriptor referencing /dev/null for each of stdin, stdout, and
1920 * stderr that is not already open.
1921 */
1922 int
1923 fdcheckstd(struct thread *td)
1924 {
1925 struct filedesc *fdp;
1926 register_t retval, save;
1927 int i, error, devnull;
1928
1929 fdp = td->td_proc->p_fd;
1930 if (fdp == NULL)
1931 return (0);
1932 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1933 devnull = -1;
1934 error = 0;
1935 for (i = 0; i < 3; i++) {
1936 if (fdp->fd_ofiles[i] != NULL)
1937 continue;
1938 if (devnull < 0) {
1939 save = td->td_retval[0];
1940 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
1941 O_RDWR, 0);
1942 devnull = td->td_retval[0];
1943 KASSERT(devnull == i, ("oof, we didn't get our fd"));
1944 td->td_retval[0] = save;
1945 if (error)
1946 break;
1947 } else {
1948 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1949 if (error != 0)
1950 break;
1951 }
1952 }
1953 return (error);
1954 }
1955
1956 /*
1957 * Internal form of close. Decrement reference count on file structure.
1958 * Note: td may be NULL when closing a file that was being passed in a
1959 * message.
1960 *
1961 * XXXRW: Giant is not required for the caller, but often will be held; this
1962 * makes it moderately likely the Giant will be recursed in the VFS case.
1963 */
1964 int
1965 closef(struct file *fp, struct thread *td)
1966 {
1967 struct vnode *vp;
1968 struct flock lf;
1969 struct filedesc_to_leader *fdtol;
1970 struct filedesc *fdp;
1971
1972 /*
1973 * POSIX record locking dictates that any close releases ALL
1974 * locks owned by this process. This is handled by setting
1975 * a flag in the unlock to free ONLY locks obeying POSIX
1976 * semantics, and not to free BSD-style file locks.
1977 * If the descriptor was in a message, POSIX-style locks
1978 * aren't passed with the descriptor, and the thread pointer
1979 * will be NULL. Callers should be careful only to pass a
1980 * NULL thread pointer when there really is no owning
1981 * context that might have locks, or the locks will be
1982 * leaked.
1983 */
1984 if (fp->f_type == DTYPE_VNODE && td != NULL) {
1985 int vfslocked;
1986
1987 vp = fp->f_vnode;
1988 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1989 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1990 lf.l_whence = SEEK_SET;
1991 lf.l_start = 0;
1992 lf.l_len = 0;
1993 lf.l_type = F_UNLCK;
1994 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1995 F_UNLCK, &lf, F_POSIX);
1996 }
1997 fdtol = td->td_proc->p_fdtol;
1998 if (fdtol != NULL) {
1999 /*
2000 * Handle special case where file descriptor table is
2001 * shared between multiple process leaders.
2002 */
2003 fdp = td->td_proc->p_fd;
2004 FILEDESC_XLOCK(fdp);
2005 for (fdtol = fdtol->fdl_next;
2006 fdtol != td->td_proc->p_fdtol;
2007 fdtol = fdtol->fdl_next) {
2008 if ((fdtol->fdl_leader->p_flag &
2009 P_ADVLOCK) == 0)
2010 continue;
2011 fdtol->fdl_holdcount++;
2012 FILEDESC_XUNLOCK(fdp);
2013 lf.l_whence = SEEK_SET;
2014 lf.l_start = 0;
2015 lf.l_len = 0;
2016 lf.l_type = F_UNLCK;
2017 vp = fp->f_vnode;
2018 (void) VOP_ADVLOCK(vp,
2019 (caddr_t)fdtol->fdl_leader,
2020 F_UNLCK, &lf, F_POSIX);
2021 FILEDESC_XLOCK(fdp);
2022 fdtol->fdl_holdcount--;
2023 if (fdtol->fdl_holdcount == 0 &&
2024 fdtol->fdl_wakeup != 0) {
2025 fdtol->fdl_wakeup = 0;
2026 wakeup(fdtol);
2027 }
2028 }
2029 FILEDESC_XUNLOCK(fdp);
2030 }
2031 VFS_UNLOCK_GIANT(vfslocked);
2032 }
2033 return (fdrop(fp, td));
2034 }
2035
2036 /*
2037 * Extract the file pointer associated with the specified descriptor for the
2038 * current user process.
2039 *
2040 * If the descriptor doesn't exist, EBADF is returned.
2041 *
2042 * If the descriptor exists but doesn't match 'flags' then return EBADF for
2043 * read attempts and EINVAL for write attempts.
2044 *
2045 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
2046 * It should be dropped with fdrop(). If it is not set, then the refcount
2047 * will not be bumped however the thread's filedesc struct will be returned
2048 * locked (for fgetsock).
2049 *
2050 * If an error occured the non-zero error is returned and *fpp is set to
2051 * NULL. Otherwise *fpp is set and zero is returned.
2052 */
2053 static __inline int
2054 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
2055 {
2056 struct filedesc *fdp;
2057 struct file *fp;
2058
2059 *fpp = NULL;
2060 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2061 return (EBADF);
2062 FILEDESC_SLOCK(fdp);
2063 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
2064 FILEDESC_SUNLOCK(fdp);
2065 return (EBADF);
2066 }
2067
2068 /*
2069 * FREAD and FWRITE failure return EBADF as per POSIX.
2070 *
2071 * Only one flag, or 0, may be specified.
2072 */
2073 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2074 FILEDESC_SUNLOCK(fdp);
2075 return (EBADF);
2076 }
2077 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2078 FILEDESC_SUNLOCK(fdp);
2079 return (EBADF);
2080 }
2081 if (hold) {
2082 fhold(fp);
2083 FILEDESC_SUNLOCK(fdp);
2084 }
2085 *fpp = fp;
2086 return (0);
2087 }
2088
2089 int
2090 fget(struct thread *td, int fd, struct file **fpp)
2091 {
2092
2093 return(_fget(td, fd, fpp, 0, 1));
2094 }
2095
2096 int
2097 fget_read(struct thread *td, int fd, struct file **fpp)
2098 {
2099
2100 return(_fget(td, fd, fpp, FREAD, 1));
2101 }
2102
2103 int
2104 fget_write(struct thread *td, int fd, struct file **fpp)
2105 {
2106
2107 return(_fget(td, fd, fpp, FWRITE, 1));
2108 }
2109
2110 /*
2111 * Like fget() but loads the underlying vnode, or returns an error if the
2112 * descriptor does not represent a vnode. Note that pipes use vnodes but
2113 * never have VM objects. The returned vnode will be vref()'d.
2114 *
2115 * XXX: what about the unused flags ?
2116 */
2117 static __inline int
2118 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2119 {
2120 struct file *fp;
2121 int error;
2122
2123 *vpp = NULL;
2124 if ((error = _fget(td, fd, &fp, flags, 0)) != 0)
2125 return (error);
2126 if (fp->f_vnode == NULL) {
2127 error = EINVAL;
2128 } else {
2129 *vpp = fp->f_vnode;
2130 vref(*vpp);
2131 }
2132 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2133 return (error);
2134 }
2135
2136 int
2137 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2138 {
2139
2140 return (_fgetvp(td, fd, vpp, 0));
2141 }
2142
2143 int
2144 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2145 {
2146
2147 return (_fgetvp(td, fd, vpp, FREAD));
2148 }
2149
2150 #ifdef notyet
2151 int
2152 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2153 {
2154
2155 return (_fgetvp(td, fd, vpp, FWRITE));
2156 }
2157 #endif
2158
2159 /*
2160 * Like fget() but loads the underlying socket, or returns an error if the
2161 * descriptor does not represent a socket.
2162 *
2163 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2164 * in the future.
2165 *
2166 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2167 * on their file descriptor reference to prevent the socket from being free'd
2168 * during use.
2169 */
2170 int
2171 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2172 {
2173 struct file *fp;
2174 int error;
2175
2176 *spp = NULL;
2177 if (fflagp != NULL)
2178 *fflagp = 0;
2179 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2180 return (error);
2181 if (fp->f_type != DTYPE_SOCKET) {
2182 error = ENOTSOCK;
2183 } else {
2184 *spp = fp->f_data;
2185 if (fflagp)
2186 *fflagp = fp->f_flag;
2187 SOCK_LOCK(*spp);
2188 soref(*spp);
2189 SOCK_UNLOCK(*spp);
2190 }
2191 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2192 return (error);
2193 }
2194
2195 /*
2196 * Drop the reference count on the socket and XXX release the SX lock in the
2197 * future. The last reference closes the socket.
2198 *
2199 * Note: fputsock() is deprecated, see comment for fgetsock().
2200 */
2201 void
2202 fputsock(struct socket *so)
2203 {
2204
2205 ACCEPT_LOCK();
2206 SOCK_LOCK(so);
2207 sorele(so);
2208 }
2209
2210 int
2211 fdrop(struct file *fp, struct thread *td)
2212 {
2213
2214 FILE_LOCK(fp);
2215 return (fdrop_locked(fp, td));
2216 }
2217
2218 /*
2219 * Drop reference on struct file passed in, may call closef if the
2220 * reference hits zero.
2221 * Expects struct file locked, and will unlock it.
2222 */
2223 static int
2224 fdrop_locked(struct file *fp, struct thread *td)
2225 {
2226 int error;
2227
2228 FILE_LOCK_ASSERT(fp, MA_OWNED);
2229
2230 if (--fp->f_count > 0) {
2231 FILE_UNLOCK(fp);
2232 return (0);
2233 }
2234
2235 /*
2236 * We might have just dropped the last reference to a file
2237 * object that is for a UNIX domain socket whose message
2238 * buffers are being examined in unp_gc(). If that is the
2239 * case, FWAIT will be set in f_gcflag and we need to wait for
2240 * unp_gc() to finish its scan.
2241 */
2242 while (fp->f_gcflag & FWAIT)
2243 msleep(&fp->f_gcflag, fp->f_mtxp, 0, "fpdrop", 0);
2244
2245 /* We have the last ref so we can proceed without the file lock. */
2246 FILE_UNLOCK(fp);
2247 if (fp->f_count < 0)
2248 panic("fdrop: count < 0");
2249 if (fp->f_ops != &badfileops)
2250 error = fo_close(fp, td);
2251 else
2252 error = 0;
2253
2254 sx_xlock(&filelist_lock);
2255 LIST_REMOVE(fp, f_list);
2256 openfiles--;
2257 sx_xunlock(&filelist_lock);
2258
2259 /*
2260 * The f_cdevpriv cannot be assigned non-NULL value while we
2261 * are destroying the file.
2262 */
2263 if (fp->f_cdevpriv != NULL)
2264 devfs_fpdrop(fp);
2265 crfree(fp->f_cred);
2266 uma_zfree(file_zone, fp);
2267
2268 return (error);
2269 }
2270
2271 /*
2272 * Apply an advisory lock on a file descriptor.
2273 *
2274 * Just attempt to get a record lock of the requested type on the entire file
2275 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2276 */
2277 #ifndef _SYS_SYSPROTO_H_
2278 struct flock_args {
2279 int fd;
2280 int how;
2281 };
2282 #endif
2283 /* ARGSUSED */
2284 int
2285 flock(struct thread *td, struct flock_args *uap)
2286 {
2287 struct file *fp;
2288 struct vnode *vp;
2289 struct flock lf;
2290 int vfslocked;
2291 int error;
2292
2293 if ((error = fget(td, uap->fd, &fp)) != 0)
2294 return (error);
2295 if (fp->f_type != DTYPE_VNODE) {
2296 fdrop(fp, td);
2297 return (EOPNOTSUPP);
2298 }
2299
2300 vp = fp->f_vnode;
2301 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2302 lf.l_whence = SEEK_SET;
2303 lf.l_start = 0;
2304 lf.l_len = 0;
2305 if (uap->how & LOCK_UN) {
2306 lf.l_type = F_UNLCK;
2307 FILE_LOCK(fp);
2308 fp->f_flag &= ~FHASLOCK;
2309 FILE_UNLOCK(fp);
2310 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2311 goto done2;
2312 }
2313 if (uap->how & LOCK_EX)
2314 lf.l_type = F_WRLCK;
2315 else if (uap->how & LOCK_SH)
2316 lf.l_type = F_RDLCK;
2317 else {
2318 error = EBADF;
2319 goto done2;
2320 }
2321 FILE_LOCK(fp);
2322 fp->f_flag |= FHASLOCK;
2323 FILE_UNLOCK(fp);
2324 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2325 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2326 done2:
2327 fdrop(fp, td);
2328 VFS_UNLOCK_GIANT(vfslocked);
2329 return (error);
2330 }
2331 /*
2332 * Duplicate the specified descriptor to a free descriptor.
2333 */
2334 int
2335 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2336 {
2337 struct file *wfp;
2338 struct file *fp;
2339
2340 /*
2341 * If the to-be-dup'd fd number is greater than the allowed number
2342 * of file descriptors, or the fd to be dup'd has already been
2343 * closed, then reject.
2344 */
2345 FILEDESC_XLOCK(fdp);
2346 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2347 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2348 FILEDESC_XUNLOCK(fdp);
2349 return (EBADF);
2350 }
2351
2352 /*
2353 * There are two cases of interest here.
2354 *
2355 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2356 *
2357 * For ENXIO steal away the file structure from (dfd) and store it in
2358 * (indx). (dfd) is effectively closed by this operation.
2359 *
2360 * Any other error code is just returned.
2361 */
2362 switch (error) {
2363 case ENODEV:
2364 /*
2365 * Check that the mode the file is being opened for is a
2366 * subset of the mode of the existing descriptor.
2367 */
2368 FILE_LOCK(wfp);
2369 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2370 FILE_UNLOCK(wfp);
2371 FILEDESC_XUNLOCK(fdp);
2372 return (EACCES);
2373 }
2374 fp = fdp->fd_ofiles[indx];
2375 fdp->fd_ofiles[indx] = wfp;
2376 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2377 if (fp == NULL)
2378 fdused(fdp, indx);
2379 fhold_locked(wfp);
2380 FILE_UNLOCK(wfp);
2381 FILEDESC_XUNLOCK(fdp);
2382 if (fp != NULL)
2383 /*
2384 * We now own the reference to fp that the ofiles[]
2385 * array used to own. Release it.
2386 */
2387 fdrop(fp, td);
2388 return (0);
2389
2390 case ENXIO:
2391 /*
2392 * Steal away the file pointer from dfd and stuff it into indx.
2393 */
2394 fp = fdp->fd_ofiles[indx];
2395 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2396 fdp->fd_ofiles[dfd] = NULL;
2397 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2398 fdp->fd_ofileflags[dfd] = 0;
2399 fdunused(fdp, dfd);
2400 if (fp == NULL)
2401 fdused(fdp, indx);
2402 FILEDESC_XUNLOCK(fdp);
2403
2404 /*
2405 * We now own the reference to fp that the ofiles[] array
2406 * used to own. Release it.
2407 */
2408 if (fp != NULL)
2409 fdrop(fp, td);
2410 return (0);
2411
2412 default:
2413 FILEDESC_XUNLOCK(fdp);
2414 return (error);
2415 }
2416 /* NOTREACHED */
2417 }
2418
2419 /*
2420 * Scan all active processes to see if any of them have a current or root
2421 * directory of `olddp'. If so, replace them with the new mount point.
2422 */
2423 void
2424 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2425 {
2426 struct filedesc *fdp;
2427 struct proc *p;
2428 int nrele;
2429
2430 if (vrefcnt(olddp) == 1)
2431 return;
2432 sx_slock(&allproc_lock);
2433 FOREACH_PROC_IN_SYSTEM(p) {
2434 fdp = fdhold(p);
2435 if (fdp == NULL)
2436 continue;
2437 nrele = 0;
2438 FILEDESC_XLOCK(fdp);
2439 if (fdp->fd_cdir == olddp) {
2440 vref(newdp);
2441 fdp->fd_cdir = newdp;
2442 nrele++;
2443 }
2444 if (fdp->fd_rdir == olddp) {
2445 vref(newdp);
2446 fdp->fd_rdir = newdp;
2447 nrele++;
2448 }
2449 FILEDESC_XUNLOCK(fdp);
2450 fddrop(fdp);
2451 while (nrele--)
2452 vrele(olddp);
2453 }
2454 sx_sunlock(&allproc_lock);
2455 if (rootvnode == olddp) {
2456 vrele(rootvnode);
2457 vref(newdp);
2458 rootvnode = newdp;
2459 }
2460 }
2461
2462 struct filedesc_to_leader *
2463 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2464 {
2465 struct filedesc_to_leader *fdtol;
2466
2467 MALLOC(fdtol, struct filedesc_to_leader *,
2468 sizeof(struct filedesc_to_leader),
2469 M_FILEDESC_TO_LEADER,
2470 M_WAITOK);
2471 fdtol->fdl_refcount = 1;
2472 fdtol->fdl_holdcount = 0;
2473 fdtol->fdl_wakeup = 0;
2474 fdtol->fdl_leader = leader;
2475 if (old != NULL) {
2476 FILEDESC_XLOCK(fdp);
2477 fdtol->fdl_next = old->fdl_next;
2478 fdtol->fdl_prev = old;
2479 old->fdl_next = fdtol;
2480 fdtol->fdl_next->fdl_prev = fdtol;
2481 FILEDESC_XUNLOCK(fdp);
2482 } else {
2483 fdtol->fdl_next = fdtol;
2484 fdtol->fdl_prev = fdtol;
2485 }
2486 return (fdtol);
2487 }
2488
2489 /*
2490 * Get file structures globally.
2491 */
2492 static int
2493 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2494 {
2495 struct xfile xf;
2496 struct filedesc *fdp;
2497 struct file *fp;
2498 struct proc *p;
2499 int error, n;
2500
2501 /*
2502 * Note: because the number of file descriptors is calculated
2503 * in different ways for sizing vs returning the data,
2504 * there is information leakage from the first loop. However,
2505 * it is of a similar order of magnitude to the leakage from
2506 * global system statistics such as kern.openfiles.
2507 */
2508 error = sysctl_wire_old_buffer(req, 0);
2509 if (error != 0)
2510 return (error);
2511 if (req->oldptr == NULL) {
2512 n = 16; /* A slight overestimate. */
2513 sx_slock(&filelist_lock);
2514 LIST_FOREACH(fp, &filehead, f_list) {
2515 /*
2516 * We should grab the lock, but this is an
2517 * estimate, so does it really matter?
2518 */
2519 /* mtx_lock(fp->f_mtxp); */
2520 n += fp->f_count;
2521 /* mtx_unlock(f->f_mtxp); */
2522 }
2523 sx_sunlock(&filelist_lock);
2524 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2525 }
2526 error = 0;
2527 bzero(&xf, sizeof(xf));
2528 xf.xf_size = sizeof(xf);
2529 sx_slock(&allproc_lock);
2530 FOREACH_PROC_IN_SYSTEM(p) {
2531 if (p->p_state == PRS_NEW)
2532 continue;
2533 PROC_LOCK(p);
2534 if (p_cansee(req->td, p) != 0) {
2535 PROC_UNLOCK(p);
2536 continue;
2537 }
2538 xf.xf_pid = p->p_pid;
2539 xf.xf_uid = p->p_ucred->cr_uid;
2540 PROC_UNLOCK(p);
2541 fdp = fdhold(p);
2542 if (fdp == NULL)
2543 continue;
2544 FILEDESC_SLOCK(fdp);
2545 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2546 if ((fp = fdp->fd_ofiles[n]) == NULL)
2547 continue;
2548 xf.xf_fd = n;
2549 xf.xf_file = fp;
2550 xf.xf_data = fp->f_data;
2551 xf.xf_vnode = fp->f_vnode;
2552 xf.xf_type = fp->f_type;
2553 xf.xf_count = fp->f_count;
2554 xf.xf_msgcount = fp->f_msgcount;
2555 xf.xf_offset = fp->f_offset;
2556 xf.xf_flag = fp->f_flag;
2557 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2558 if (error)
2559 break;
2560 }
2561 FILEDESC_SUNLOCK(fdp);
2562 fddrop(fdp);
2563 if (error)
2564 break;
2565 }
2566 sx_sunlock(&allproc_lock);
2567 return (error);
2568 }
2569
2570 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2571 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2572
2573 #ifdef KINFO_OFILE_SIZE
2574 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2575 #endif
2576
2577 /* Compatability with early 7-stable */
2578 static int
2579 export_vnode_for_osysctl(struct vnode *vp, int type,
2580 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2581 {
2582 int error;
2583 char *fullpath, *freepath;
2584 int vfslocked;
2585
2586 bzero(kif, sizeof(*kif));
2587 kif->kf_structsize = sizeof(*kif);
2588
2589 vref(vp);
2590 kif->kf_fd = type;
2591 kif->kf_type = KF_TYPE_VNODE;
2592 /* This function only handles directories. */
2593 KASSERT(vp->v_type == VDIR, ("export_vnode_for_osysctl: vnode not directory"));
2594 kif->kf_vnode_type = KF_VTYPE_VDIR;
2595
2596 /*
2597 * This is not a true file descriptor, so we set a bogus refcount
2598 * and offset to indicate these fields should be ignored.
2599 */
2600 kif->kf_ref_count = -1;
2601 kif->kf_offset = -1;
2602
2603 freepath = NULL;
2604 fullpath = "-";
2605 FILEDESC_SUNLOCK(fdp);
2606 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2607 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
2608 vn_fullpath(curthread, vp, &fullpath, &freepath);
2609 vput(vp);
2610 VFS_UNLOCK_GIANT(vfslocked);
2611 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2612 if (freepath != NULL)
2613 free(freepath, M_TEMP);
2614 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2615 FILEDESC_SLOCK(fdp);
2616 return (error);
2617 }
2618
2619 /*
2620 * Get per-process file descriptors for use by procstat(1), et al.
2621 */
2622 static int
2623 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
2624 {
2625 char *fullpath, *freepath;
2626 struct kinfo_ofile *kif;
2627 struct filedesc *fdp;
2628 int error, i, *name;
2629 struct socket *so;
2630 struct vnode *vp;
2631 struct file *fp;
2632 struct proc *p;
2633 int vfslocked;
2634
2635 name = (int *)arg1;
2636 if ((p = pfind((pid_t)name[0])) == NULL)
2637 return (ESRCH);
2638 if ((error = p_candebug(curthread, p))) {
2639 PROC_UNLOCK(p);
2640 return (error);
2641 }
2642 fdp = fdhold(p);
2643 PROC_UNLOCK(p);
2644 if (fdp == NULL)
2645 return (ENOENT);
2646 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2647 FILEDESC_SLOCK(fdp);
2648 if (fdp->fd_cdir != NULL)
2649 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2650 fdp, req);
2651 if (fdp->fd_rdir != NULL)
2652 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2653 fdp, req);
2654 if (fdp->fd_jdir != NULL)
2655 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2656 fdp, req);
2657 for (i = 0; i < fdp->fd_nfiles; i++) {
2658 if ((fp = fdp->fd_ofiles[i]) == NULL)
2659 continue;
2660 bzero(kif, sizeof(*kif));
2661 kif->kf_structsize = sizeof(*kif);
2662 FILE_LOCK(fp);
2663 vp = NULL;
2664 so = NULL;
2665 kif->kf_fd = i;
2666 switch (fp->f_type) {
2667 case DTYPE_VNODE:
2668 kif->kf_type = KF_TYPE_VNODE;
2669 vp = fp->f_vnode;
2670 vref(vp);
2671 break;
2672
2673 case DTYPE_SOCKET:
2674 kif->kf_type = KF_TYPE_SOCKET;
2675 so = fp->f_data;
2676 break;
2677
2678 case DTYPE_PIPE:
2679 kif->kf_type = KF_TYPE_PIPE;
2680 break;
2681
2682 case DTYPE_FIFO:
2683 kif->kf_type = KF_TYPE_FIFO;
2684 vp = fp->f_vnode;
2685 vref(vp);
2686 break;
2687
2688 case DTYPE_KQUEUE:
2689 kif->kf_type = KF_TYPE_KQUEUE;
2690 break;
2691
2692 case DTYPE_CRYPTO:
2693 kif->kf_type = KF_TYPE_CRYPTO;
2694 break;
2695
2696 case DTYPE_MQUEUE:
2697 kif->kf_type = KF_TYPE_MQUEUE;
2698 break;
2699
2700 default:
2701 kif->kf_type = KF_TYPE_UNKNOWN;
2702 break;
2703 }
2704 kif->kf_ref_count = fp->f_count;
2705 if (fp->f_flag & FREAD)
2706 kif->kf_flags |= KF_FLAG_READ;
2707 if (fp->f_flag & FWRITE)
2708 kif->kf_flags |= KF_FLAG_WRITE;
2709 if (fp->f_flag & FAPPEND)
2710 kif->kf_flags |= KF_FLAG_APPEND;
2711 if (fp->f_flag & FASYNC)
2712 kif->kf_flags |= KF_FLAG_ASYNC;
2713 if (fp->f_flag & FFSYNC)
2714 kif->kf_flags |= KF_FLAG_FSYNC;
2715 if (fp->f_flag & FNONBLOCK)
2716 kif->kf_flags |= KF_FLAG_NONBLOCK;
2717 if (fp->f_flag & O_DIRECT)
2718 kif->kf_flags |= KF_FLAG_DIRECT;
2719 if (fp->f_flag & FHASLOCK)
2720 kif->kf_flags |= KF_FLAG_HASLOCK;
2721 kif->kf_offset = fp->f_offset;
2722 FILE_UNLOCK(fp);
2723 if (vp != NULL) {
2724 switch (vp->v_type) {
2725 case VNON:
2726 kif->kf_vnode_type = KF_VTYPE_VNON;
2727 break;
2728 case VREG:
2729 kif->kf_vnode_type = KF_VTYPE_VREG;
2730 break;
2731 case VDIR:
2732 kif->kf_vnode_type = KF_VTYPE_VDIR;
2733 break;
2734 case VBLK:
2735 kif->kf_vnode_type = KF_VTYPE_VBLK;
2736 break;
2737 case VCHR:
2738 kif->kf_vnode_type = KF_VTYPE_VCHR;
2739 break;
2740 case VLNK:
2741 kif->kf_vnode_type = KF_VTYPE_VLNK;
2742 break;
2743 case VSOCK:
2744 kif->kf_vnode_type = KF_VTYPE_VSOCK;
2745 break;
2746 case VFIFO:
2747 kif->kf_vnode_type = KF_VTYPE_VFIFO;
2748 break;
2749 case VBAD:
2750 kif->kf_vnode_type = KF_VTYPE_VBAD;
2751 break;
2752 default:
2753 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2754 break;
2755 }
2756 /*
2757 * It is OK to drop the filedesc lock here as we will
2758 * re-validate and re-evaluate its properties when
2759 * the loop continues.
2760 */
2761 freepath = NULL;
2762 fullpath = "-";
2763 FILEDESC_SUNLOCK(fdp);
2764 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2765 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
2766 vn_fullpath(curthread, vp, &fullpath, &freepath);
2767 vput(vp);
2768 VFS_UNLOCK_GIANT(vfslocked);
2769 strlcpy(kif->kf_path, fullpath,
2770 sizeof(kif->kf_path));
2771 if (freepath != NULL)
2772 free(freepath, M_TEMP);
2773 FILEDESC_SLOCK(fdp);
2774 }
2775 if (so != NULL) {
2776 struct sockaddr *sa;
2777
2778 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
2779 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
2780 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
2781 free(sa, M_SONAME);
2782 }
2783 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
2784 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
2785 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
2786 free(sa, M_SONAME);
2787 }
2788 kif->kf_sock_domain =
2789 so->so_proto->pr_domain->dom_family;
2790 kif->kf_sock_type = so->so_type;
2791 kif->kf_sock_protocol = so->so_proto->pr_protocol;
2792 }
2793 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2794 if (error)
2795 break;
2796 }
2797 FILEDESC_SUNLOCK(fdp);
2798 fddrop(fdp);
2799 free(kif, M_TEMP);
2800 return (0);
2801 }
2802
2803 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
2804 sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
2805
2806 #ifdef KINFO_FILE_SIZE
2807 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2808 #endif
2809
2810 static int
2811 export_vnode_for_sysctl(struct vnode *vp, int type,
2812 struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req)
2813 {
2814 int error;
2815 char *fullpath, *freepath;
2816 int vfslocked;
2817
2818 bzero(kif, sizeof(*kif));
2819
2820 vref(vp);
2821 kif->kf_fd = type;
2822 kif->kf_type = KF_TYPE_VNODE;
2823 /* This function only handles directories. */
2824 KASSERT(vp->v_type == VDIR, ("export_vnode_for_sysctl: vnode not directory"));
2825 kif->kf_vnode_type = KF_VTYPE_VDIR;
2826
2827 /*
2828 * This is not a true file descriptor, so we set a bogus refcount
2829 * and offset to indicate these fields should be ignored.
2830 */
2831 kif->kf_ref_count = -1;
2832 kif->kf_offset = -1;
2833
2834 freepath = NULL;
2835 fullpath = "-";
2836 FILEDESC_SUNLOCK(fdp);
2837 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2838 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
2839 vn_fullpath(curthread, vp, &fullpath, &freepath);
2840 vput(vp);
2841 VFS_UNLOCK_GIANT(vfslocked);
2842 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2843 if (freepath != NULL)
2844 free(freepath, M_TEMP);
2845 /* Pack record size down */
2846 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
2847 strlen(kif->kf_path) + 1;
2848 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
2849 error = SYSCTL_OUT(req, kif, kif->kf_structsize);
2850 FILEDESC_SLOCK(fdp);
2851 return (error);
2852 }
2853
2854 /*
2855 * Get per-process file descriptors for use by procstat(1), et al.
2856 */
2857 static int
2858 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
2859 {
2860 char *fullpath, *freepath;
2861 struct kinfo_file *kif;
2862 struct filedesc *fdp;
2863 int error, i, *name;
2864 struct socket *so;
2865 struct vnode *vp;
2866 struct file *fp;
2867 struct proc *p;
2868 int vfslocked;
2869
2870 name = (int *)arg1;
2871 if ((p = pfind((pid_t)name[0])) == NULL)
2872 return (ESRCH);
2873 if ((error = p_candebug(curthread, p))) {
2874 PROC_UNLOCK(p);
2875 return (error);
2876 }
2877 fdp = fdhold(p);
2878 PROC_UNLOCK(p);
2879 if (fdp == NULL)
2880 return (ENOENT);
2881 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2882 FILEDESC_SLOCK(fdp);
2883 if (fdp->fd_cdir != NULL)
2884 export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2885 fdp, req);
2886 if (fdp->fd_rdir != NULL)
2887 export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2888 fdp, req);
2889 if (fdp->fd_jdir != NULL)
2890 export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2891 fdp, req);
2892 for (i = 0; i < fdp->fd_nfiles; i++) {
2893 if ((fp = fdp->fd_ofiles[i]) == NULL)
2894 continue;
2895 bzero(kif, sizeof(*kif));
2896 FILE_LOCK(fp);
2897 vp = NULL;
2898 so = NULL;
2899 kif->kf_fd = i;
2900 switch (fp->f_type) {
2901 case DTYPE_VNODE:
2902 kif->kf_type = KF_TYPE_VNODE;
2903 vp = fp->f_vnode;
2904 vref(vp);
2905 break;
2906
2907 case DTYPE_SOCKET:
2908 kif->kf_type = KF_TYPE_SOCKET;
2909 so = fp->f_data;
2910 break;
2911
2912 case DTYPE_PIPE:
2913 kif->kf_type = KF_TYPE_PIPE;
2914 break;
2915
2916 case DTYPE_FIFO:
2917 kif->kf_type = KF_TYPE_FIFO;
2918 vp = fp->f_vnode;
2919 vref(vp);
2920 break;
2921
2922 case DTYPE_KQUEUE:
2923 kif->kf_type = KF_TYPE_KQUEUE;
2924 break;
2925
2926 case DTYPE_CRYPTO:
2927 kif->kf_type = KF_TYPE_CRYPTO;
2928 break;
2929
2930 case DTYPE_MQUEUE:
2931 kif->kf_type = KF_TYPE_MQUEUE;
2932 break;
2933
2934 default:
2935 kif->kf_type = KF_TYPE_UNKNOWN;
2936 break;
2937 }
2938 kif->kf_ref_count = fp->f_count;
2939 if (fp->f_flag & FREAD)
2940 kif->kf_flags |= KF_FLAG_READ;
2941 if (fp->f_flag & FWRITE)
2942 kif->kf_flags |= KF_FLAG_WRITE;
2943 if (fp->f_flag & FAPPEND)
2944 kif->kf_flags |= KF_FLAG_APPEND;
2945 if (fp->f_flag & FASYNC)
2946 kif->kf_flags |= KF_FLAG_ASYNC;
2947 if (fp->f_flag & FFSYNC)
2948 kif->kf_flags |= KF_FLAG_FSYNC;
2949 if (fp->f_flag & FNONBLOCK)
2950 kif->kf_flags |= KF_FLAG_NONBLOCK;
2951 if (fp->f_flag & O_DIRECT)
2952 kif->kf_flags |= KF_FLAG_DIRECT;
2953 if (fp->f_flag & FHASLOCK)
2954 kif->kf_flags |= KF_FLAG_HASLOCK;
2955 kif->kf_offset = fp->f_offset;
2956 FILE_UNLOCK(fp);
2957 if (vp != NULL) {
2958 switch (vp->v_type) {
2959 case VNON:
2960 kif->kf_vnode_type = KF_VTYPE_VNON;
2961 break;
2962 case VREG:
2963 kif->kf_vnode_type = KF_VTYPE_VREG;
2964 break;
2965 case VDIR:
2966 kif->kf_vnode_type = KF_VTYPE_VDIR;
2967 break;
2968 case VBLK:
2969 kif->kf_vnode_type = KF_VTYPE_VBLK;
2970 break;
2971 case VCHR:
2972 kif->kf_vnode_type = KF_VTYPE_VCHR;
2973 break;
2974 case VLNK:
2975 kif->kf_vnode_type = KF_VTYPE_VLNK;
2976 break;
2977 case VSOCK:
2978 kif->kf_vnode_type = KF_VTYPE_VSOCK;
2979 break;
2980 case VFIFO:
2981 kif->kf_vnode_type = KF_VTYPE_VFIFO;
2982 break;
2983 case VBAD:
2984 kif->kf_vnode_type = KF_VTYPE_VBAD;
2985 break;
2986 default:
2987 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2988 break;
2989 }
2990 /*
2991 * It is OK to drop the filedesc lock here as we will
2992 * re-validate and re-evaluate its properties when
2993 * the loop continues.
2994 */
2995 freepath = NULL;
2996 fullpath = "-";
2997 FILEDESC_SUNLOCK(fdp);
2998 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2999 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
3000 vn_fullpath(curthread, vp, &fullpath, &freepath);
3001 vput(vp);
3002 VFS_UNLOCK_GIANT(vfslocked);
3003 strlcpy(kif->kf_path, fullpath,
3004 sizeof(kif->kf_path));
3005 if (freepath != NULL)
3006 free(freepath, M_TEMP);
3007 FILEDESC_SLOCK(fdp);
3008 }
3009 if (so != NULL) {
3010 struct sockaddr *sa;
3011
3012 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3013 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3014 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3015 free(sa, M_SONAME);
3016 }
3017 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3018 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3019 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3020 free(sa, M_SONAME);
3021 }
3022 kif->kf_sock_domain =
3023 so->so_proto->pr_domain->dom_family;
3024 kif->kf_sock_type = so->so_type;
3025 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3026 }
3027 /* Pack record size down */
3028 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3029 strlen(kif->kf_path) + 1;
3030 kif->kf_structsize = roundup(kif->kf_structsize,
3031 sizeof(uint64_t));
3032 error = SYSCTL_OUT(req, kif, kif->kf_structsize);
3033 if (error)
3034 break;
3035 }
3036 FILEDESC_SUNLOCK(fdp);
3037 fddrop(fdp);
3038 free(kif, M_TEMP);
3039 return (0);
3040 }
3041
3042 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3043 sysctl_kern_proc_filedesc, "Process filedesc entries");
3044
3045 #ifdef DDB
3046 /*
3047 * For the purposes of debugging, generate a human-readable string for the
3048 * file type.
3049 */
3050 static const char *
3051 file_type_to_name(short type)
3052 {
3053
3054 switch (type) {
3055 case 0:
3056 return ("zero");
3057 case DTYPE_VNODE:
3058 return ("vnod");
3059 case DTYPE_SOCKET:
3060 return ("sock");
3061 case DTYPE_PIPE:
3062 return ("pipe");
3063 case DTYPE_FIFO:
3064 return ("fifo");
3065 case DTYPE_KQUEUE:
3066 return ("kque");
3067 case DTYPE_CRYPTO:
3068 return ("crpt");
3069 case DTYPE_MQUEUE:
3070 return ("mque");
3071 default:
3072 return ("unkn");
3073 }
3074 }
3075
3076 /*
3077 * For the purposes of debugging, identify a process (if any, perhaps one of
3078 * many) that references the passed file in its file descriptor array. Return
3079 * NULL if none.
3080 */
3081 static struct proc *
3082 file_to_first_proc(struct file *fp)
3083 {
3084 struct filedesc *fdp;
3085 struct proc *p;
3086 int n;
3087
3088 FOREACH_PROC_IN_SYSTEM(p) {
3089 if (p->p_state == PRS_NEW)
3090 continue;
3091 fdp = p->p_fd;
3092 if (fdp == NULL)
3093 continue;
3094 for (n = 0; n < fdp->fd_nfiles; n++) {
3095 if (fp == fdp->fd_ofiles[n])
3096 return (p);
3097 }
3098 }
3099 return (NULL);
3100 }
3101
3102 static void
3103 db_print_file(struct file *fp, int header)
3104 {
3105 struct proc *p;
3106
3107 if (header)
3108 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3109 "File", "Type", "Data", "Flag", "GCFl", "Count",
3110 "MCount", "Vnode", "FPID", "FCmd");
3111 p = file_to_first_proc(fp);
3112 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3113 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3114 fp->f_gcflag, fp->f_count, fp->f_msgcount, fp->f_vnode,
3115 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3116 }
3117
3118 DB_SHOW_COMMAND(file, db_show_file)
3119 {
3120 struct file *fp;
3121
3122 if (!have_addr) {
3123 db_printf("usage: show file <addr>\n");
3124 return;
3125 }
3126 fp = (struct file *)addr;
3127 db_print_file(fp, 1);
3128 }
3129
3130 DB_SHOW_COMMAND(files, db_show_files)
3131 {
3132 struct file *fp;
3133 int header;
3134
3135 header = 1;
3136 LIST_FOREACH(fp, &filehead, f_list) {
3137 db_print_file(fp, header);
3138 header = 0;
3139 }
3140 }
3141 #endif
3142
3143 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3144 &maxfilesperproc, 0, "Maximum files allowed open per process");
3145
3146 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3147 &maxfiles, 0, "Maximum number of files");
3148
3149 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3150 &openfiles, 0, "System-wide number of open files");
3151
3152 /* ARGSUSED*/
3153 static void
3154 filelistinit(void *dummy)
3155 {
3156
3157 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3158 NULL, NULL, UMA_ALIGN_PTR, 0);
3159 sx_init(&filelist_lock, "filelist lock");
3160 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3161 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3162 }
3163 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3164
3165 /*-------------------------------------------------------------------*/
3166
3167 static int
3168 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
3169 {
3170
3171 return (EBADF);
3172 }
3173
3174 static int
3175 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
3176 {
3177
3178 return (EBADF);
3179 }
3180
3181 static int
3182 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
3183 {
3184
3185 return (0);
3186 }
3187
3188 static int
3189 badfo_kqfilter(struct file *fp, struct knote *kn)
3190 {
3191
3192 return (EBADF);
3193 }
3194
3195 static int
3196 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
3197 {
3198
3199 return (EBADF);
3200 }
3201
3202 static int
3203 badfo_close(struct file *fp, struct thread *td)
3204 {
3205
3206 return (EBADF);
3207 }
3208
3209 struct fileops badfileops = {
3210 .fo_read = badfo_readwrite,
3211 .fo_write = badfo_readwrite,
3212 .fo_ioctl = badfo_ioctl,
3213 .fo_poll = badfo_poll,
3214 .fo_kqfilter = badfo_kqfilter,
3215 .fo_stat = badfo_stat,
3216 .fo_close = badfo_close,
3217 };
3218
3219
3220 /*-------------------------------------------------------------------*/
3221
3222 /*
3223 * File Descriptor pseudo-device driver (/dev/fd/).
3224 *
3225 * Opening minor device N dup()s the file (if any) connected to file
3226 * descriptor N belonging to the calling process. Note that this driver
3227 * consists of only the ``open()'' routine, because all subsequent
3228 * references to this file will be direct to the other driver.
3229 *
3230 * XXX: we could give this one a cloning event handler if necessary.
3231 */
3232
3233 /* ARGSUSED */
3234 static int
3235 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3236 {
3237
3238 /*
3239 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3240 * the file descriptor being sought for duplication. The error
3241 * return ensures that the vnode for this device will be released
3242 * by vn_open. Open will detect this special error and take the
3243 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3244 * will simply report the error.
3245 */
3246 td->td_dupfd = dev2unit(dev);
3247 return (ENODEV);
3248 }
3249
3250 static struct cdevsw fildesc_cdevsw = {
3251 .d_version = D_VERSION,
3252 .d_flags = D_NEEDGIANT,
3253 .d_open = fdopen,
3254 .d_name = "FD",
3255 };
3256
3257 static void
3258 fildesc_drvinit(void *unused)
3259 {
3260 struct cdev *dev;
3261
3262 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
3263 make_dev_alias(dev, "stdin");
3264 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
3265 make_dev_alias(dev, "stdout");
3266 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
3267 make_dev_alias(dev, "stderr");
3268 }
3269
3270 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
Cache object: 8cea87a698d7bac64bd3a863e20dc227
|