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/11.1/sys/kern/kern_descrip.c 315396 2017-03-16 08:37:05Z mjg $");
39
40 #include "opt_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ddb.h"
43 #include "opt_ktrace.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47
48 #include <sys/capsicum.h>
49 #include <sys/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/file.h>
52 #include <sys/filedesc.h>
53 #include <sys/filio.h>
54 #include <sys/jail.h>
55 #include <sys/kernel.h>
56 #include <sys/limits.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/selinfo.h>
63 #include <sys/priv.h>
64 #include <sys/proc.h>
65 #include <sys/protosw.h>
66 #include <sys/racct.h>
67 #include <sys/resourcevar.h>
68 #include <sys/sbuf.h>
69 #include <sys/signalvar.h>
70 #include <sys/kdb.h>
71 #include <sys/stat.h>
72 #include <sys/sx.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysctl.h>
75 #include <sys/sysproto.h>
76 #include <sys/unistd.h>
77 #include <sys/user.h>
78 #include <sys/vnode.h>
79 #ifdef KTRACE
80 #include <sys/ktrace.h>
81 #endif
82
83 #include <net/vnet.h>
84
85 #include <security/audit/audit.h>
86
87 #include <vm/uma.h>
88 #include <vm/vm.h>
89
90 #include <ddb/ddb.h>
91
92 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
93 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
94 "file desc to leader structures");
95 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
96 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
97
98 MALLOC_DECLARE(M_FADVISE);
99
100 static __read_mostly uma_zone_t file_zone;
101 static __read_mostly uma_zone_t filedesc0_zone;
102
103 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
104 struct thread *td, int holdleaders);
105 static int fd_first_free(struct filedesc *fdp, int low, int size);
106 static int fd_last_used(struct filedesc *fdp, int size);
107 static void fdgrowtable(struct filedesc *fdp, int nfd);
108 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
109 static void fdunused(struct filedesc *fdp, int fd);
110 static void fdused(struct filedesc *fdp, int fd);
111 static int getmaxfd(struct thread *td);
112
113 /*
114 * Each process has:
115 *
116 * - An array of open file descriptors (fd_ofiles)
117 * - An array of file flags (fd_ofileflags)
118 * - A bitmap recording which descriptors are in use (fd_map)
119 *
120 * A process starts out with NDFILE descriptors. The value of NDFILE has
121 * been selected based the historical limit of 20 open files, and an
122 * assumption that the majority of processes, especially short-lived
123 * processes like shells, will never need more.
124 *
125 * If this initial allocation is exhausted, a larger descriptor table and
126 * map are allocated dynamically, and the pointers in the process's struct
127 * filedesc are updated to point to those. This is repeated every time
128 * the process runs out of file descriptors (provided it hasn't hit its
129 * resource limit).
130 *
131 * Since threads may hold references to individual descriptor table
132 * entries, the tables are never freed. Instead, they are placed on a
133 * linked list and freed only when the struct filedesc is released.
134 */
135 #define NDFILE 20
136 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
137 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
138 #define NDSLOT(x) ((x) / NDENTRIES)
139 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
140 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
141
142 /*
143 * SLIST entry used to keep track of ofiles which must be reclaimed when
144 * the process exits.
145 */
146 struct freetable {
147 struct fdescenttbl *ft_table;
148 SLIST_ENTRY(freetable) ft_next;
149 };
150
151 /*
152 * Initial allocation: a filedesc structure + the head of SLIST used to
153 * keep track of old ofiles + enough space for NDFILE descriptors.
154 */
155
156 struct fdescenttbl0 {
157 int fdt_nfiles;
158 struct filedescent fdt_ofiles[NDFILE];
159 };
160
161 struct filedesc0 {
162 struct filedesc fd_fd;
163 SLIST_HEAD(, freetable) fd_free;
164 struct fdescenttbl0 fd_dfiles;
165 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
166 };
167
168 /*
169 * Descriptor management.
170 */
171 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
172 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
173 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
174
175 /*
176 * If low >= size, just return low. Otherwise find the first zero bit in the
177 * given bitmap, starting at low and not exceeding size - 1. Return size if
178 * not found.
179 */
180 static int
181 fd_first_free(struct filedesc *fdp, int low, int size)
182 {
183 NDSLOTTYPE *map = fdp->fd_map;
184 NDSLOTTYPE mask;
185 int off, maxoff;
186
187 if (low >= size)
188 return (low);
189
190 off = NDSLOT(low);
191 if (low % NDENTRIES) {
192 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
193 if ((mask &= ~map[off]) != 0UL)
194 return (off * NDENTRIES + ffsl(mask) - 1);
195 ++off;
196 }
197 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
198 if (map[off] != ~0UL)
199 return (off * NDENTRIES + ffsl(~map[off]) - 1);
200 return (size);
201 }
202
203 /*
204 * Find the highest non-zero bit in the given bitmap, starting at 0 and
205 * not exceeding size - 1. Return -1 if not found.
206 */
207 static int
208 fd_last_used(struct filedesc *fdp, int size)
209 {
210 NDSLOTTYPE *map = fdp->fd_map;
211 NDSLOTTYPE mask;
212 int off, minoff;
213
214 off = NDSLOT(size);
215 if (size % NDENTRIES) {
216 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
217 if ((mask &= map[off]) != 0)
218 return (off * NDENTRIES + flsl(mask) - 1);
219 --off;
220 }
221 for (minoff = NDSLOT(0); off >= minoff; --off)
222 if (map[off] != 0)
223 return (off * NDENTRIES + flsl(map[off]) - 1);
224 return (-1);
225 }
226
227 static int
228 fdisused(struct filedesc *fdp, int fd)
229 {
230
231 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
232 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
233
234 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
235 }
236
237 /*
238 * Mark a file descriptor as used.
239 */
240 static void
241 fdused_init(struct filedesc *fdp, int fd)
242 {
243
244 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
245
246 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
247 }
248
249 static void
250 fdused(struct filedesc *fdp, int fd)
251 {
252
253 FILEDESC_XLOCK_ASSERT(fdp);
254
255 fdused_init(fdp, fd);
256 if (fd > fdp->fd_lastfile)
257 fdp->fd_lastfile = fd;
258 if (fd == fdp->fd_freefile)
259 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
260 }
261
262 /*
263 * Mark a file descriptor as unused.
264 */
265 static void
266 fdunused(struct filedesc *fdp, int fd)
267 {
268
269 FILEDESC_XLOCK_ASSERT(fdp);
270
271 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
272 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
273 ("fd=%d is still in use", fd));
274
275 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
276 if (fd < fdp->fd_freefile)
277 fdp->fd_freefile = fd;
278 if (fd == fdp->fd_lastfile)
279 fdp->fd_lastfile = fd_last_used(fdp, fd);
280 }
281
282 /*
283 * Free a file descriptor.
284 *
285 * Avoid some work if fdp is about to be destroyed.
286 */
287 static inline void
288 fdefree_last(struct filedescent *fde)
289 {
290
291 filecaps_free(&fde->fde_caps);
292 }
293
294 static inline void
295 fdfree(struct filedesc *fdp, int fd)
296 {
297 struct filedescent *fde;
298
299 fde = &fdp->fd_ofiles[fd];
300 #ifdef CAPABILITIES
301 seq_write_begin(&fde->fde_seq);
302 #endif
303 fdefree_last(fde);
304 fde->fde_file = NULL;
305 fdunused(fdp, fd);
306 #ifdef CAPABILITIES
307 seq_write_end(&fde->fde_seq);
308 #endif
309 }
310
311 void
312 pwd_ensure_dirs(void)
313 {
314 struct filedesc *fdp;
315
316 fdp = curproc->p_fd;
317 FILEDESC_XLOCK(fdp);
318 if (fdp->fd_cdir == NULL) {
319 fdp->fd_cdir = rootvnode;
320 vrefact(rootvnode);
321 }
322 if (fdp->fd_rdir == NULL) {
323 fdp->fd_rdir = rootvnode;
324 vrefact(rootvnode);
325 }
326 FILEDESC_XUNLOCK(fdp);
327 }
328
329 /*
330 * System calls on descriptors.
331 */
332 #ifndef _SYS_SYSPROTO_H_
333 struct getdtablesize_args {
334 int dummy;
335 };
336 #endif
337 /* ARGSUSED */
338 int
339 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
340 {
341 #ifdef RACCT
342 uint64_t lim;
343 #endif
344
345 td->td_retval[0] =
346 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
347 #ifdef RACCT
348 PROC_LOCK(td->td_proc);
349 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
350 PROC_UNLOCK(td->td_proc);
351 if (lim < td->td_retval[0])
352 td->td_retval[0] = lim;
353 #endif
354 return (0);
355 }
356
357 /*
358 * Duplicate a file descriptor to a particular value.
359 *
360 * Note: keep in mind that a potential race condition exists when closing
361 * descriptors from a shared descriptor table (via rfork).
362 */
363 #ifndef _SYS_SYSPROTO_H_
364 struct dup2_args {
365 u_int from;
366 u_int to;
367 };
368 #endif
369 /* ARGSUSED */
370 int
371 sys_dup2(struct thread *td, struct dup2_args *uap)
372 {
373
374 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
375 }
376
377 /*
378 * Duplicate a file descriptor.
379 */
380 #ifndef _SYS_SYSPROTO_H_
381 struct dup_args {
382 u_int fd;
383 };
384 #endif
385 /* ARGSUSED */
386 int
387 sys_dup(struct thread *td, struct dup_args *uap)
388 {
389
390 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
391 }
392
393 /*
394 * The file control system call.
395 */
396 #ifndef _SYS_SYSPROTO_H_
397 struct fcntl_args {
398 int fd;
399 int cmd;
400 long arg;
401 };
402 #endif
403 /* ARGSUSED */
404 int
405 sys_fcntl(struct thread *td, struct fcntl_args *uap)
406 {
407
408 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
409 }
410
411 int
412 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
413 {
414 struct flock fl;
415 struct __oflock ofl;
416 intptr_t arg1;
417 int error, newcmd;
418
419 error = 0;
420 newcmd = cmd;
421 switch (cmd) {
422 case F_OGETLK:
423 case F_OSETLK:
424 case F_OSETLKW:
425 /*
426 * Convert old flock structure to new.
427 */
428 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
429 fl.l_start = ofl.l_start;
430 fl.l_len = ofl.l_len;
431 fl.l_pid = ofl.l_pid;
432 fl.l_type = ofl.l_type;
433 fl.l_whence = ofl.l_whence;
434 fl.l_sysid = 0;
435
436 switch (cmd) {
437 case F_OGETLK:
438 newcmd = F_GETLK;
439 break;
440 case F_OSETLK:
441 newcmd = F_SETLK;
442 break;
443 case F_OSETLKW:
444 newcmd = F_SETLKW;
445 break;
446 }
447 arg1 = (intptr_t)&fl;
448 break;
449 case F_GETLK:
450 case F_SETLK:
451 case F_SETLKW:
452 case F_SETLK_REMOTE:
453 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
454 arg1 = (intptr_t)&fl;
455 break;
456 default:
457 arg1 = arg;
458 break;
459 }
460 if (error)
461 return (error);
462 error = kern_fcntl(td, fd, newcmd, arg1);
463 if (error)
464 return (error);
465 if (cmd == F_OGETLK) {
466 ofl.l_start = fl.l_start;
467 ofl.l_len = fl.l_len;
468 ofl.l_pid = fl.l_pid;
469 ofl.l_type = fl.l_type;
470 ofl.l_whence = fl.l_whence;
471 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
472 } else if (cmd == F_GETLK) {
473 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
474 }
475 return (error);
476 }
477
478 int
479 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
480 {
481 struct filedesc *fdp;
482 struct flock *flp;
483 struct file *fp, *fp2;
484 struct filedescent *fde;
485 struct proc *p;
486 struct vnode *vp;
487 cap_rights_t rights;
488 int error, flg, tmp;
489 uint64_t bsize;
490 off_t foffset;
491
492 error = 0;
493 flg = F_POSIX;
494 p = td->td_proc;
495 fdp = p->p_fd;
496
497 switch (cmd) {
498 case F_DUPFD:
499 tmp = arg;
500 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
501 break;
502
503 case F_DUPFD_CLOEXEC:
504 tmp = arg;
505 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
506 break;
507
508 case F_DUP2FD:
509 tmp = arg;
510 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
511 break;
512
513 case F_DUP2FD_CLOEXEC:
514 tmp = arg;
515 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
516 break;
517
518 case F_GETFD:
519 error = EBADF;
520 FILEDESC_SLOCK(fdp);
521 fde = fdeget_locked(fdp, fd);
522 if (fde != NULL) {
523 td->td_retval[0] =
524 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
525 error = 0;
526 }
527 FILEDESC_SUNLOCK(fdp);
528 break;
529
530 case F_SETFD:
531 error = EBADF;
532 FILEDESC_XLOCK(fdp);
533 fde = fdeget_locked(fdp, fd);
534 if (fde != NULL) {
535 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
536 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
537 error = 0;
538 }
539 FILEDESC_XUNLOCK(fdp);
540 break;
541
542 case F_GETFL:
543 error = fget_fcntl(td, fd,
544 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp);
545 if (error != 0)
546 break;
547 td->td_retval[0] = OFLAGS(fp->f_flag);
548 fdrop(fp, td);
549 break;
550
551 case F_SETFL:
552 error = fget_fcntl(td, fd,
553 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp);
554 if (error != 0)
555 break;
556 do {
557 tmp = flg = fp->f_flag;
558 tmp &= ~FCNTLFLAGS;
559 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
560 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
561 tmp = fp->f_flag & FNONBLOCK;
562 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
563 if (error != 0) {
564 fdrop(fp, td);
565 break;
566 }
567 tmp = fp->f_flag & FASYNC;
568 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
569 if (error == 0) {
570 fdrop(fp, td);
571 break;
572 }
573 atomic_clear_int(&fp->f_flag, FNONBLOCK);
574 tmp = 0;
575 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
576 fdrop(fp, td);
577 break;
578
579 case F_GETOWN:
580 error = fget_fcntl(td, fd,
581 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp);
582 if (error != 0)
583 break;
584 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
585 if (error == 0)
586 td->td_retval[0] = tmp;
587 fdrop(fp, td);
588 break;
589
590 case F_SETOWN:
591 error = fget_fcntl(td, fd,
592 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp);
593 if (error != 0)
594 break;
595 tmp = arg;
596 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
597 fdrop(fp, td);
598 break;
599
600 case F_SETLK_REMOTE:
601 error = priv_check(td, PRIV_NFS_LOCKD);
602 if (error)
603 return (error);
604 flg = F_REMOTE;
605 goto do_setlk;
606
607 case F_SETLKW:
608 flg |= F_WAIT;
609 /* FALLTHROUGH F_SETLK */
610
611 case F_SETLK:
612 do_setlk:
613 cap_rights_init(&rights, CAP_FLOCK);
614 error = fget_unlocked(fdp, fd, &rights, &fp, NULL);
615 if (error != 0)
616 break;
617 if (fp->f_type != DTYPE_VNODE) {
618 error = EBADF;
619 fdrop(fp, td);
620 break;
621 }
622
623 flp = (struct flock *)arg;
624 if (flp->l_whence == SEEK_CUR) {
625 foffset = foffset_get(fp);
626 if (foffset < 0 ||
627 (flp->l_start > 0 &&
628 foffset > OFF_MAX - flp->l_start)) {
629 error = EOVERFLOW;
630 fdrop(fp, td);
631 break;
632 }
633 flp->l_start += foffset;
634 }
635
636 vp = fp->f_vnode;
637 switch (flp->l_type) {
638 case F_RDLCK:
639 if ((fp->f_flag & FREAD) == 0) {
640 error = EBADF;
641 break;
642 }
643 PROC_LOCK(p->p_leader);
644 p->p_leader->p_flag |= P_ADVLOCK;
645 PROC_UNLOCK(p->p_leader);
646 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
647 flp, flg);
648 break;
649 case F_WRLCK:
650 if ((fp->f_flag & FWRITE) == 0) {
651 error = EBADF;
652 break;
653 }
654 PROC_LOCK(p->p_leader);
655 p->p_leader->p_flag |= P_ADVLOCK;
656 PROC_UNLOCK(p->p_leader);
657 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
658 flp, flg);
659 break;
660 case F_UNLCK:
661 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
662 flp, flg);
663 break;
664 case F_UNLCKSYS:
665 /*
666 * Temporary api for testing remote lock
667 * infrastructure.
668 */
669 if (flg != F_REMOTE) {
670 error = EINVAL;
671 break;
672 }
673 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
674 F_UNLCKSYS, flp, flg);
675 break;
676 default:
677 error = EINVAL;
678 break;
679 }
680 if (error != 0 || flp->l_type == F_UNLCK ||
681 flp->l_type == F_UNLCKSYS) {
682 fdrop(fp, td);
683 break;
684 }
685
686 /*
687 * Check for a race with close.
688 *
689 * The vnode is now advisory locked (or unlocked, but this case
690 * is not really important) as the caller requested.
691 * We had to drop the filedesc lock, so we need to recheck if
692 * the descriptor is still valid, because if it was closed
693 * in the meantime we need to remove advisory lock from the
694 * vnode - close on any descriptor leading to an advisory
695 * locked vnode, removes that lock.
696 * We will return 0 on purpose in that case, as the result of
697 * successful advisory lock might have been externally visible
698 * already. This is fine - effectively we pretend to the caller
699 * that the closing thread was a bit slower and that the
700 * advisory lock succeeded before the close.
701 */
702 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL);
703 if (error != 0) {
704 fdrop(fp, td);
705 break;
706 }
707 if (fp != fp2) {
708 flp->l_whence = SEEK_SET;
709 flp->l_start = 0;
710 flp->l_len = 0;
711 flp->l_type = F_UNLCK;
712 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
713 F_UNLCK, flp, F_POSIX);
714 }
715 fdrop(fp, td);
716 fdrop(fp2, td);
717 break;
718
719 case F_GETLK:
720 error = fget_unlocked(fdp, fd,
721 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL);
722 if (error != 0)
723 break;
724 if (fp->f_type != DTYPE_VNODE) {
725 error = EBADF;
726 fdrop(fp, td);
727 break;
728 }
729 flp = (struct flock *)arg;
730 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
731 flp->l_type != F_UNLCK) {
732 error = EINVAL;
733 fdrop(fp, td);
734 break;
735 }
736 if (flp->l_whence == SEEK_CUR) {
737 foffset = foffset_get(fp);
738 if ((flp->l_start > 0 &&
739 foffset > OFF_MAX - flp->l_start) ||
740 (flp->l_start < 0 &&
741 foffset < OFF_MIN - flp->l_start)) {
742 error = EOVERFLOW;
743 fdrop(fp, td);
744 break;
745 }
746 flp->l_start += foffset;
747 }
748 vp = fp->f_vnode;
749 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
750 F_POSIX);
751 fdrop(fp, td);
752 break;
753
754 case F_RDAHEAD:
755 arg = arg ? 128 * 1024: 0;
756 /* FALLTHROUGH */
757 case F_READAHEAD:
758 error = fget_unlocked(fdp, fd,
759 cap_rights_init(&rights), &fp, NULL);
760 if (error != 0)
761 break;
762 if (fp->f_type != DTYPE_VNODE) {
763 fdrop(fp, td);
764 error = EBADF;
765 break;
766 }
767 vp = fp->f_vnode;
768 /*
769 * Exclusive lock synchronizes against f_seqcount reads and
770 * writes in sequential_heuristic().
771 */
772 error = vn_lock(vp, LK_EXCLUSIVE);
773 if (error != 0) {
774 fdrop(fp, td);
775 break;
776 }
777 if (arg >= 0) {
778 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
779 fp->f_seqcount = (arg + bsize - 1) / bsize;
780 atomic_set_int(&fp->f_flag, FRDAHEAD);
781 } else {
782 atomic_clear_int(&fp->f_flag, FRDAHEAD);
783 }
784 VOP_UNLOCK(vp, 0);
785 fdrop(fp, td);
786 break;
787
788 default:
789 error = EINVAL;
790 break;
791 }
792 return (error);
793 }
794
795 static int
796 getmaxfd(struct thread *td)
797 {
798
799 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
800 }
801
802 /*
803 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
804 */
805 int
806 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
807 {
808 struct filedesc *fdp;
809 struct filedescent *oldfde, *newfde;
810 struct proc *p;
811 struct file *delfp;
812 int error, maxfd;
813
814 p = td->td_proc;
815 fdp = p->p_fd;
816
817 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
818 MPASS(mode < FDDUP_LASTMODE);
819
820 AUDIT_ARG_FD(old);
821 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
822
823 /*
824 * Verify we have a valid descriptor to dup from and possibly to
825 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
826 * return EINVAL when the new descriptor is out of bounds.
827 */
828 if (old < 0)
829 return (EBADF);
830 if (new < 0)
831 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
832 maxfd = getmaxfd(td);
833 if (new >= maxfd)
834 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
835
836 error = EBADF;
837 FILEDESC_XLOCK(fdp);
838 if (fget_locked(fdp, old) == NULL)
839 goto unlock;
840 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
841 td->td_retval[0] = new;
842 if (flags & FDDUP_FLAG_CLOEXEC)
843 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
844 error = 0;
845 goto unlock;
846 }
847
848 /*
849 * If the caller specified a file descriptor, make sure the file
850 * table is large enough to hold it, and grab it. Otherwise, just
851 * allocate a new descriptor the usual way.
852 */
853 switch (mode) {
854 case FDDUP_NORMAL:
855 case FDDUP_FCNTL:
856 if ((error = fdalloc(td, new, &new)) != 0)
857 goto unlock;
858 break;
859 case FDDUP_MUSTREPLACE:
860 /* Target file descriptor must exist. */
861 if (fget_locked(fdp, new) == NULL)
862 goto unlock;
863 break;
864 case FDDUP_FIXED:
865 if (new >= fdp->fd_nfiles) {
866 /*
867 * The resource limits are here instead of e.g.
868 * fdalloc(), because the file descriptor table may be
869 * shared between processes, so we can't really use
870 * racct_add()/racct_sub(). Instead of counting the
871 * number of actually allocated descriptors, just put
872 * the limit on the size of the file descriptor table.
873 */
874 #ifdef RACCT
875 if (racct_enable) {
876 PROC_LOCK(p);
877 error = racct_set(p, RACCT_NOFILE, new + 1);
878 PROC_UNLOCK(p);
879 if (error != 0) {
880 error = EMFILE;
881 goto unlock;
882 }
883 }
884 #endif
885 fdgrowtable_exp(fdp, new + 1);
886 }
887 if (!fdisused(fdp, new))
888 fdused(fdp, new);
889 break;
890 default:
891 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
892 }
893
894 KASSERT(old != new, ("new fd is same as old"));
895
896 oldfde = &fdp->fd_ofiles[old];
897 fhold(oldfde->fde_file);
898 newfde = &fdp->fd_ofiles[new];
899 delfp = newfde->fde_file;
900
901 /*
902 * Duplicate the source descriptor.
903 */
904 #ifdef CAPABILITIES
905 seq_write_begin(&newfde->fde_seq);
906 #endif
907 filecaps_free(&newfde->fde_caps);
908 memcpy(newfde, oldfde, fde_change_size);
909 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
910 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
911 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
912 else
913 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
914 #ifdef CAPABILITIES
915 seq_write_end(&newfde->fde_seq);
916 #endif
917 td->td_retval[0] = new;
918
919 error = 0;
920
921 if (delfp != NULL) {
922 (void) closefp(fdp, new, delfp, td, 1);
923 FILEDESC_UNLOCK_ASSERT(fdp);
924 } else {
925 unlock:
926 FILEDESC_XUNLOCK(fdp);
927 }
928
929 return (error);
930 }
931
932 /*
933 * If sigio is on the list associated with a process or process group,
934 * disable signalling from the device, remove sigio from the list and
935 * free sigio.
936 */
937 void
938 funsetown(struct sigio **sigiop)
939 {
940 struct sigio *sigio;
941
942 if (*sigiop == NULL)
943 return;
944 SIGIO_LOCK();
945 sigio = *sigiop;
946 if (sigio == NULL) {
947 SIGIO_UNLOCK();
948 return;
949 }
950 *(sigio->sio_myref) = NULL;
951 if ((sigio)->sio_pgid < 0) {
952 struct pgrp *pg = (sigio)->sio_pgrp;
953 PGRP_LOCK(pg);
954 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
955 sigio, sio_pgsigio);
956 PGRP_UNLOCK(pg);
957 } else {
958 struct proc *p = (sigio)->sio_proc;
959 PROC_LOCK(p);
960 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
961 sigio, sio_pgsigio);
962 PROC_UNLOCK(p);
963 }
964 SIGIO_UNLOCK();
965 crfree(sigio->sio_ucred);
966 free(sigio, M_SIGIO);
967 }
968
969 /*
970 * Free a list of sigio structures.
971 * We only need to lock the SIGIO_LOCK because we have made ourselves
972 * inaccessible to callers of fsetown and therefore do not need to lock
973 * the proc or pgrp struct for the list manipulation.
974 */
975 void
976 funsetownlst(struct sigiolst *sigiolst)
977 {
978 struct proc *p;
979 struct pgrp *pg;
980 struct sigio *sigio;
981
982 sigio = SLIST_FIRST(sigiolst);
983 if (sigio == NULL)
984 return;
985 p = NULL;
986 pg = NULL;
987
988 /*
989 * Every entry of the list should belong
990 * to a single proc or pgrp.
991 */
992 if (sigio->sio_pgid < 0) {
993 pg = sigio->sio_pgrp;
994 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
995 } else /* if (sigio->sio_pgid > 0) */ {
996 p = sigio->sio_proc;
997 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
998 }
999
1000 SIGIO_LOCK();
1001 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1002 *(sigio->sio_myref) = NULL;
1003 if (pg != NULL) {
1004 KASSERT(sigio->sio_pgid < 0,
1005 ("Proc sigio in pgrp sigio list"));
1006 KASSERT(sigio->sio_pgrp == pg,
1007 ("Bogus pgrp in sigio list"));
1008 PGRP_LOCK(pg);
1009 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1010 sio_pgsigio);
1011 PGRP_UNLOCK(pg);
1012 } else /* if (p != NULL) */ {
1013 KASSERT(sigio->sio_pgid > 0,
1014 ("Pgrp sigio in proc sigio list"));
1015 KASSERT(sigio->sio_proc == p,
1016 ("Bogus proc in sigio list"));
1017 PROC_LOCK(p);
1018 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1019 sio_pgsigio);
1020 PROC_UNLOCK(p);
1021 }
1022 SIGIO_UNLOCK();
1023 crfree(sigio->sio_ucred);
1024 free(sigio, M_SIGIO);
1025 SIGIO_LOCK();
1026 }
1027 SIGIO_UNLOCK();
1028 }
1029
1030 /*
1031 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1032 *
1033 * After permission checking, add a sigio structure to the sigio list for
1034 * the process or process group.
1035 */
1036 int
1037 fsetown(pid_t pgid, struct sigio **sigiop)
1038 {
1039 struct proc *proc;
1040 struct pgrp *pgrp;
1041 struct sigio *sigio;
1042 int ret;
1043
1044 if (pgid == 0) {
1045 funsetown(sigiop);
1046 return (0);
1047 }
1048
1049 ret = 0;
1050
1051 /* Allocate and fill in the new sigio out of locks. */
1052 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1053 sigio->sio_pgid = pgid;
1054 sigio->sio_ucred = crhold(curthread->td_ucred);
1055 sigio->sio_myref = sigiop;
1056
1057 sx_slock(&proctree_lock);
1058 if (pgid > 0) {
1059 proc = pfind(pgid);
1060 if (proc == NULL) {
1061 ret = ESRCH;
1062 goto fail;
1063 }
1064
1065 /*
1066 * Policy - Don't allow a process to FSETOWN a process
1067 * in another session.
1068 *
1069 * Remove this test to allow maximum flexibility or
1070 * restrict FSETOWN to the current process or process
1071 * group for maximum safety.
1072 */
1073 PROC_UNLOCK(proc);
1074 if (proc->p_session != curthread->td_proc->p_session) {
1075 ret = EPERM;
1076 goto fail;
1077 }
1078
1079 pgrp = NULL;
1080 } else /* if (pgid < 0) */ {
1081 pgrp = pgfind(-pgid);
1082 if (pgrp == NULL) {
1083 ret = ESRCH;
1084 goto fail;
1085 }
1086 PGRP_UNLOCK(pgrp);
1087
1088 /*
1089 * Policy - Don't allow a process to FSETOWN a process
1090 * in another session.
1091 *
1092 * Remove this test to allow maximum flexibility or
1093 * restrict FSETOWN to the current process or process
1094 * group for maximum safety.
1095 */
1096 if (pgrp->pg_session != curthread->td_proc->p_session) {
1097 ret = EPERM;
1098 goto fail;
1099 }
1100
1101 proc = NULL;
1102 }
1103 funsetown(sigiop);
1104 if (pgid > 0) {
1105 PROC_LOCK(proc);
1106 /*
1107 * Since funsetownlst() is called without the proctree
1108 * locked, we need to check for P_WEXIT.
1109 * XXX: is ESRCH correct?
1110 */
1111 if ((proc->p_flag & P_WEXIT) != 0) {
1112 PROC_UNLOCK(proc);
1113 ret = ESRCH;
1114 goto fail;
1115 }
1116 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1117 sigio->sio_proc = proc;
1118 PROC_UNLOCK(proc);
1119 } else {
1120 PGRP_LOCK(pgrp);
1121 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1122 sigio->sio_pgrp = pgrp;
1123 PGRP_UNLOCK(pgrp);
1124 }
1125 sx_sunlock(&proctree_lock);
1126 SIGIO_LOCK();
1127 *sigiop = sigio;
1128 SIGIO_UNLOCK();
1129 return (0);
1130
1131 fail:
1132 sx_sunlock(&proctree_lock);
1133 crfree(sigio->sio_ucred);
1134 free(sigio, M_SIGIO);
1135 return (ret);
1136 }
1137
1138 /*
1139 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1140 */
1141 pid_t
1142 fgetown(sigiop)
1143 struct sigio **sigiop;
1144 {
1145 pid_t pgid;
1146
1147 SIGIO_LOCK();
1148 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1149 SIGIO_UNLOCK();
1150 return (pgid);
1151 }
1152
1153 /*
1154 * Function drops the filedesc lock on return.
1155 */
1156 static int
1157 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1158 int holdleaders)
1159 {
1160 int error;
1161
1162 FILEDESC_XLOCK_ASSERT(fdp);
1163
1164 if (holdleaders) {
1165 if (td->td_proc->p_fdtol != NULL) {
1166 /*
1167 * Ask fdfree() to sleep to ensure that all relevant
1168 * process leaders can be traversed in closef().
1169 */
1170 fdp->fd_holdleaderscount++;
1171 } else {
1172 holdleaders = 0;
1173 }
1174 }
1175
1176 /*
1177 * We now hold the fp reference that used to be owned by the
1178 * descriptor array. We have to unlock the FILEDESC *AFTER*
1179 * knote_fdclose to prevent a race of the fd getting opened, a knote
1180 * added, and deleteing a knote for the new fd.
1181 */
1182 knote_fdclose(td, fd);
1183
1184 /*
1185 * We need to notify mqueue if the object is of type mqueue.
1186 */
1187 if (fp->f_type == DTYPE_MQUEUE)
1188 mq_fdclose(td, fd, fp);
1189 FILEDESC_XUNLOCK(fdp);
1190
1191 error = closef(fp, td);
1192 if (holdleaders) {
1193 FILEDESC_XLOCK(fdp);
1194 fdp->fd_holdleaderscount--;
1195 if (fdp->fd_holdleaderscount == 0 &&
1196 fdp->fd_holdleaderswakeup != 0) {
1197 fdp->fd_holdleaderswakeup = 0;
1198 wakeup(&fdp->fd_holdleaderscount);
1199 }
1200 FILEDESC_XUNLOCK(fdp);
1201 }
1202 return (error);
1203 }
1204
1205 /*
1206 * Close a file descriptor.
1207 */
1208 #ifndef _SYS_SYSPROTO_H_
1209 struct close_args {
1210 int fd;
1211 };
1212 #endif
1213 /* ARGSUSED */
1214 int
1215 sys_close(struct thread *td, struct close_args *uap)
1216 {
1217
1218 return (kern_close(td, uap->fd));
1219 }
1220
1221 int
1222 kern_close(struct thread *td, int fd)
1223 {
1224 struct filedesc *fdp;
1225 struct file *fp;
1226
1227 fdp = td->td_proc->p_fd;
1228
1229 AUDIT_SYSCLOSE(td, fd);
1230
1231 FILEDESC_XLOCK(fdp);
1232 if ((fp = fget_locked(fdp, fd)) == NULL) {
1233 FILEDESC_XUNLOCK(fdp);
1234 return (EBADF);
1235 }
1236 fdfree(fdp, fd);
1237
1238 /* closefp() drops the FILEDESC lock for us. */
1239 return (closefp(fdp, fd, fp, td, 1));
1240 }
1241
1242 /*
1243 * Close open file descriptors.
1244 */
1245 #ifndef _SYS_SYSPROTO_H_
1246 struct closefrom_args {
1247 int lowfd;
1248 };
1249 #endif
1250 /* ARGSUSED */
1251 int
1252 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1253 {
1254 struct filedesc *fdp;
1255 int fd;
1256
1257 fdp = td->td_proc->p_fd;
1258 AUDIT_ARG_FD(uap->lowfd);
1259
1260 /*
1261 * Treat negative starting file descriptor values identical to
1262 * closefrom(0) which closes all files.
1263 */
1264 if (uap->lowfd < 0)
1265 uap->lowfd = 0;
1266 FILEDESC_SLOCK(fdp);
1267 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1268 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1269 FILEDESC_SUNLOCK(fdp);
1270 (void)kern_close(td, fd);
1271 FILEDESC_SLOCK(fdp);
1272 }
1273 }
1274 FILEDESC_SUNLOCK(fdp);
1275 return (0);
1276 }
1277
1278 #if defined(COMPAT_43)
1279 /*
1280 * Return status information about a file descriptor.
1281 */
1282 #ifndef _SYS_SYSPROTO_H_
1283 struct ofstat_args {
1284 int fd;
1285 struct ostat *sb;
1286 };
1287 #endif
1288 /* ARGSUSED */
1289 int
1290 ofstat(struct thread *td, struct ofstat_args *uap)
1291 {
1292 struct ostat oub;
1293 struct stat ub;
1294 int error;
1295
1296 error = kern_fstat(td, uap->fd, &ub);
1297 if (error == 0) {
1298 cvtstat(&ub, &oub);
1299 error = copyout(&oub, uap->sb, sizeof(oub));
1300 }
1301 return (error);
1302 }
1303 #endif /* COMPAT_43 */
1304
1305 /*
1306 * Return status information about a file descriptor.
1307 */
1308 #ifndef _SYS_SYSPROTO_H_
1309 struct fstat_args {
1310 int fd;
1311 struct stat *sb;
1312 };
1313 #endif
1314 /* ARGSUSED */
1315 int
1316 sys_fstat(struct thread *td, struct fstat_args *uap)
1317 {
1318 struct stat ub;
1319 int error;
1320
1321 error = kern_fstat(td, uap->fd, &ub);
1322 if (error == 0)
1323 error = copyout(&ub, uap->sb, sizeof(ub));
1324 return (error);
1325 }
1326
1327 int
1328 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1329 {
1330 struct file *fp;
1331 cap_rights_t rights;
1332 int error;
1333
1334 AUDIT_ARG_FD(fd);
1335
1336 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1337 if (error != 0)
1338 return (error);
1339
1340 AUDIT_ARG_FILE(td->td_proc, fp);
1341
1342 error = fo_stat(fp, sbp, td->td_ucred, td);
1343 fdrop(fp, td);
1344 #ifdef KTRACE
1345 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1346 ktrstat(sbp);
1347 #endif
1348 return (error);
1349 }
1350
1351 /*
1352 * Return status information about a file descriptor.
1353 */
1354 #ifndef _SYS_SYSPROTO_H_
1355 struct nfstat_args {
1356 int fd;
1357 struct nstat *sb;
1358 };
1359 #endif
1360 /* ARGSUSED */
1361 int
1362 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1363 {
1364 struct nstat nub;
1365 struct stat ub;
1366 int error;
1367
1368 error = kern_fstat(td, uap->fd, &ub);
1369 if (error == 0) {
1370 cvtnstat(&ub, &nub);
1371 error = copyout(&nub, uap->sb, sizeof(nub));
1372 }
1373 return (error);
1374 }
1375
1376 /*
1377 * Return pathconf information about a file descriptor.
1378 */
1379 #ifndef _SYS_SYSPROTO_H_
1380 struct fpathconf_args {
1381 int fd;
1382 int name;
1383 };
1384 #endif
1385 /* ARGSUSED */
1386 int
1387 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1388 {
1389 struct file *fp;
1390 struct vnode *vp;
1391 cap_rights_t rights;
1392 int error;
1393
1394 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1395 if (error != 0)
1396 return (error);
1397
1398 if (uap->name == _PC_ASYNC_IO) {
1399 td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1400 goto out;
1401 }
1402 vp = fp->f_vnode;
1403 if (vp != NULL) {
1404 vn_lock(vp, LK_SHARED | LK_RETRY);
1405 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1406 VOP_UNLOCK(vp, 0);
1407 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1408 if (uap->name != _PC_PIPE_BUF) {
1409 error = EINVAL;
1410 } else {
1411 td->td_retval[0] = PIPE_BUF;
1412 error = 0;
1413 }
1414 } else {
1415 error = EOPNOTSUPP;
1416 }
1417 out:
1418 fdrop(fp, td);
1419 return (error);
1420 }
1421
1422 /*
1423 * Initialize filecaps structure.
1424 */
1425 void
1426 filecaps_init(struct filecaps *fcaps)
1427 {
1428
1429 bzero(fcaps, sizeof(*fcaps));
1430 fcaps->fc_nioctls = -1;
1431 }
1432
1433 /*
1434 * Copy filecaps structure allocating memory for ioctls array if needed.
1435 *
1436 * The last parameter indicates whether the fdtable is locked. If it is not and
1437 * ioctls are encountered, copying fails and the caller must lock the table.
1438 *
1439 * Note that if the table was not locked, the caller has to check the relevant
1440 * sequence counter to determine whether the operation was successful.
1441 */
1442 int
1443 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1444 {
1445 size_t size;
1446
1447 *dst = *src;
1448 if (src->fc_ioctls == NULL)
1449 return (0);
1450 if (!locked)
1451 return (1);
1452
1453 KASSERT(src->fc_nioctls > 0,
1454 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1455
1456 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1457 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1458 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1459 return (0);
1460 }
1461
1462 /*
1463 * Move filecaps structure to the new place and clear the old place.
1464 */
1465 void
1466 filecaps_move(struct filecaps *src, struct filecaps *dst)
1467 {
1468
1469 *dst = *src;
1470 bzero(src, sizeof(*src));
1471 }
1472
1473 /*
1474 * Fill the given filecaps structure with full rights.
1475 */
1476 static void
1477 filecaps_fill(struct filecaps *fcaps)
1478 {
1479
1480 CAP_ALL(&fcaps->fc_rights);
1481 fcaps->fc_ioctls = NULL;
1482 fcaps->fc_nioctls = -1;
1483 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1484 }
1485
1486 /*
1487 * Free memory allocated within filecaps structure.
1488 */
1489 void
1490 filecaps_free(struct filecaps *fcaps)
1491 {
1492
1493 free(fcaps->fc_ioctls, M_FILECAPS);
1494 bzero(fcaps, sizeof(*fcaps));
1495 }
1496
1497 /*
1498 * Validate the given filecaps structure.
1499 */
1500 static void
1501 filecaps_validate(const struct filecaps *fcaps, const char *func)
1502 {
1503
1504 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1505 ("%s: invalid rights", func));
1506 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1507 ("%s: invalid fcntls", func));
1508 KASSERT(fcaps->fc_fcntls == 0 ||
1509 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1510 ("%s: fcntls without CAP_FCNTL", func));
1511 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1512 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1513 ("%s: invalid ioctls", func));
1514 KASSERT(fcaps->fc_nioctls == 0 ||
1515 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1516 ("%s: ioctls without CAP_IOCTL", func));
1517 }
1518
1519 static void
1520 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1521 {
1522 int nfd1;
1523
1524 FILEDESC_XLOCK_ASSERT(fdp);
1525
1526 nfd1 = fdp->fd_nfiles * 2;
1527 if (nfd1 < nfd)
1528 nfd1 = nfd;
1529 fdgrowtable(fdp, nfd1);
1530 }
1531
1532 /*
1533 * Grow the file table to accommodate (at least) nfd descriptors.
1534 */
1535 static void
1536 fdgrowtable(struct filedesc *fdp, int nfd)
1537 {
1538 struct filedesc0 *fdp0;
1539 struct freetable *ft;
1540 struct fdescenttbl *ntable;
1541 struct fdescenttbl *otable;
1542 int nnfiles, onfiles;
1543 NDSLOTTYPE *nmap, *omap;
1544
1545 /*
1546 * If lastfile is -1 this struct filedesc was just allocated and we are
1547 * growing it to accommodate for the one we are going to copy from. There
1548 * is no need to have a lock on this one as it's not visible to anyone.
1549 */
1550 if (fdp->fd_lastfile != -1)
1551 FILEDESC_XLOCK_ASSERT(fdp);
1552
1553 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1554
1555 /* save old values */
1556 onfiles = fdp->fd_nfiles;
1557 otable = fdp->fd_files;
1558 omap = fdp->fd_map;
1559
1560 /* compute the size of the new table */
1561 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1562 if (nnfiles <= onfiles)
1563 /* the table is already large enough */
1564 return;
1565
1566 /*
1567 * Allocate a new table. We need enough space for the number of
1568 * entries, file entries themselves and the struct freetable we will use
1569 * when we decommission the table and place it on the freelist.
1570 * We place the struct freetable in the middle so we don't have
1571 * to worry about padding.
1572 */
1573 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1574 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1575 sizeof(struct freetable),
1576 M_FILEDESC, M_ZERO | M_WAITOK);
1577 /* copy the old data */
1578 ntable->fdt_nfiles = nnfiles;
1579 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1580 onfiles * sizeof(ntable->fdt_ofiles[0]));
1581
1582 /*
1583 * Allocate a new map only if the old is not large enough. It will
1584 * grow at a slower rate than the table as it can map more
1585 * entries than the table can hold.
1586 */
1587 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1588 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1589 M_ZERO | M_WAITOK);
1590 /* copy over the old data and update the pointer */
1591 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1592 fdp->fd_map = nmap;
1593 }
1594
1595 /*
1596 * Make sure that ntable is correctly initialized before we replace
1597 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1598 * data.
1599 */
1600 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1601
1602 /*
1603 * Do not free the old file table, as some threads may still
1604 * reference entries within it. Instead, place it on a freelist
1605 * which will be processed when the struct filedesc is released.
1606 *
1607 * Note that if onfiles == NDFILE, we're dealing with the original
1608 * static allocation contained within (struct filedesc0 *)fdp,
1609 * which must not be freed.
1610 */
1611 if (onfiles > NDFILE) {
1612 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1613 fdp0 = (struct filedesc0 *)fdp;
1614 ft->ft_table = otable;
1615 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1616 }
1617 /*
1618 * The map does not have the same possibility of threads still
1619 * holding references to it. So always free it as long as it
1620 * does not reference the original static allocation.
1621 */
1622 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1623 free(omap, M_FILEDESC);
1624 }
1625
1626 /*
1627 * Allocate a file descriptor for the process.
1628 */
1629 int
1630 fdalloc(struct thread *td, int minfd, int *result)
1631 {
1632 struct proc *p = td->td_proc;
1633 struct filedesc *fdp = p->p_fd;
1634 int fd, maxfd, allocfd;
1635 #ifdef RACCT
1636 int error;
1637 #endif
1638
1639 FILEDESC_XLOCK_ASSERT(fdp);
1640
1641 if (fdp->fd_freefile > minfd)
1642 minfd = fdp->fd_freefile;
1643
1644 maxfd = getmaxfd(td);
1645
1646 /*
1647 * Search the bitmap for a free descriptor starting at minfd.
1648 * If none is found, grow the file table.
1649 */
1650 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1651 if (fd >= maxfd)
1652 return (EMFILE);
1653 if (fd >= fdp->fd_nfiles) {
1654 allocfd = min(fd * 2, maxfd);
1655 #ifdef RACCT
1656 if (racct_enable) {
1657 PROC_LOCK(p);
1658 error = racct_set(p, RACCT_NOFILE, allocfd);
1659 PROC_UNLOCK(p);
1660 if (error != 0)
1661 return (EMFILE);
1662 }
1663 #endif
1664 /*
1665 * fd is already equal to first free descriptor >= minfd, so
1666 * we only need to grow the table and we are done.
1667 */
1668 fdgrowtable_exp(fdp, allocfd);
1669 }
1670
1671 /*
1672 * Perform some sanity checks, then mark the file descriptor as
1673 * used and return it to the caller.
1674 */
1675 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1676 ("invalid descriptor %d", fd));
1677 KASSERT(!fdisused(fdp, fd),
1678 ("fd_first_free() returned non-free descriptor"));
1679 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1680 ("file descriptor isn't free"));
1681 fdused(fdp, fd);
1682 *result = fd;
1683 return (0);
1684 }
1685
1686 /*
1687 * Allocate n file descriptors for the process.
1688 */
1689 int
1690 fdallocn(struct thread *td, int minfd, int *fds, int n)
1691 {
1692 struct proc *p = td->td_proc;
1693 struct filedesc *fdp = p->p_fd;
1694 int i;
1695
1696 FILEDESC_XLOCK_ASSERT(fdp);
1697
1698 for (i = 0; i < n; i++)
1699 if (fdalloc(td, 0, &fds[i]) != 0)
1700 break;
1701
1702 if (i < n) {
1703 for (i--; i >= 0; i--)
1704 fdunused(fdp, fds[i]);
1705 return (EMFILE);
1706 }
1707
1708 return (0);
1709 }
1710
1711 /*
1712 * Create a new open file structure and allocate a file descriptor for the
1713 * process that refers to it. We add one reference to the file for the
1714 * descriptor table and one reference for resultfp. This is to prevent us
1715 * being preempted and the entry in the descriptor table closed after we
1716 * release the FILEDESC lock.
1717 */
1718 int
1719 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1720 struct filecaps *fcaps)
1721 {
1722 struct file *fp;
1723 int error, fd;
1724
1725 error = falloc_noinstall(td, &fp);
1726 if (error)
1727 return (error); /* no reference held on error */
1728
1729 error = finstall(td, fp, &fd, flags, fcaps);
1730 if (error) {
1731 fdrop(fp, td); /* one reference (fp only) */
1732 return (error);
1733 }
1734
1735 if (resultfp != NULL)
1736 *resultfp = fp; /* copy out result */
1737 else
1738 fdrop(fp, td); /* release local reference */
1739
1740 if (resultfd != NULL)
1741 *resultfd = fd;
1742
1743 return (0);
1744 }
1745
1746 /*
1747 * Create a new open file structure without allocating a file descriptor.
1748 */
1749 int
1750 falloc_noinstall(struct thread *td, struct file **resultfp)
1751 {
1752 struct file *fp;
1753 int maxuserfiles = maxfiles - (maxfiles / 20);
1754 int openfiles_new;
1755 static struct timeval lastfail;
1756 static int curfail;
1757
1758 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1759
1760 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1761 if ((openfiles_new >= maxuserfiles &&
1762 priv_check(td, PRIV_MAXFILES) != 0) ||
1763 openfiles_new >= maxfiles) {
1764 atomic_subtract_int(&openfiles, 1);
1765 if (ppsratecheck(&lastfail, &curfail, 1)) {
1766 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1767 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1768 }
1769 return (ENFILE);
1770 }
1771 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1772 refcount_init(&fp->f_count, 1);
1773 fp->f_cred = crhold(td->td_ucred);
1774 fp->f_ops = &badfileops;
1775 *resultfp = fp;
1776 return (0);
1777 }
1778
1779 /*
1780 * Install a file in a file descriptor table.
1781 */
1782 void
1783 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1784 struct filecaps *fcaps)
1785 {
1786 struct filedescent *fde;
1787
1788 MPASS(fp != NULL);
1789 if (fcaps != NULL)
1790 filecaps_validate(fcaps, __func__);
1791 FILEDESC_XLOCK_ASSERT(fdp);
1792
1793 fde = &fdp->fd_ofiles[fd];
1794 #ifdef CAPABILITIES
1795 seq_write_begin(&fde->fde_seq);
1796 #endif
1797 fde->fde_file = fp;
1798 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1799 if (fcaps != NULL)
1800 filecaps_move(fcaps, &fde->fde_caps);
1801 else
1802 filecaps_fill(&fde->fde_caps);
1803 #ifdef CAPABILITIES
1804 seq_write_end(&fde->fde_seq);
1805 #endif
1806 }
1807
1808 int
1809 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1810 struct filecaps *fcaps)
1811 {
1812 struct filedesc *fdp = td->td_proc->p_fd;
1813 int error;
1814
1815 MPASS(fd != NULL);
1816
1817 FILEDESC_XLOCK(fdp);
1818 if ((error = fdalloc(td, 0, fd))) {
1819 FILEDESC_XUNLOCK(fdp);
1820 return (error);
1821 }
1822 fhold(fp);
1823 _finstall(fdp, fp, *fd, flags, fcaps);
1824 FILEDESC_XUNLOCK(fdp);
1825 return (0);
1826 }
1827
1828 /*
1829 * Build a new filedesc structure from another.
1830 * Copy the current, root, and jail root vnode references.
1831 *
1832 * If fdp is not NULL, return with it shared locked.
1833 */
1834 struct filedesc *
1835 fdinit(struct filedesc *fdp, bool prepfiles)
1836 {
1837 struct filedesc0 *newfdp0;
1838 struct filedesc *newfdp;
1839
1840 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1841 newfdp = &newfdp0->fd_fd;
1842
1843 /* Create the file descriptor table. */
1844 FILEDESC_LOCK_INIT(newfdp);
1845 refcount_init(&newfdp->fd_refcnt, 1);
1846 refcount_init(&newfdp->fd_holdcnt, 1);
1847 newfdp->fd_cmask = CMASK;
1848 newfdp->fd_map = newfdp0->fd_dmap;
1849 newfdp->fd_lastfile = -1;
1850 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1851 newfdp->fd_files->fdt_nfiles = NDFILE;
1852
1853 if (fdp == NULL)
1854 return (newfdp);
1855
1856 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1857 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1858
1859 FILEDESC_SLOCK(fdp);
1860 newfdp->fd_cdir = fdp->fd_cdir;
1861 if (newfdp->fd_cdir)
1862 vrefact(newfdp->fd_cdir);
1863 newfdp->fd_rdir = fdp->fd_rdir;
1864 if (newfdp->fd_rdir)
1865 vrefact(newfdp->fd_rdir);
1866 newfdp->fd_jdir = fdp->fd_jdir;
1867 if (newfdp->fd_jdir)
1868 vrefact(newfdp->fd_jdir);
1869
1870 if (!prepfiles) {
1871 FILEDESC_SUNLOCK(fdp);
1872 } else {
1873 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1874 FILEDESC_SUNLOCK(fdp);
1875 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1876 FILEDESC_SLOCK(fdp);
1877 }
1878 }
1879
1880 return (newfdp);
1881 }
1882
1883 static struct filedesc *
1884 fdhold(struct proc *p)
1885 {
1886 struct filedesc *fdp;
1887
1888 PROC_LOCK_ASSERT(p, MA_OWNED);
1889 fdp = p->p_fd;
1890 if (fdp != NULL)
1891 refcount_acquire(&fdp->fd_holdcnt);
1892 return (fdp);
1893 }
1894
1895 static void
1896 fddrop(struct filedesc *fdp)
1897 {
1898
1899 if (fdp->fd_holdcnt > 1) {
1900 if (refcount_release(&fdp->fd_holdcnt) == 0)
1901 return;
1902 }
1903
1904 FILEDESC_LOCK_DESTROY(fdp);
1905 uma_zfree(filedesc0_zone, fdp);
1906 }
1907
1908 /*
1909 * Share a filedesc structure.
1910 */
1911 struct filedesc *
1912 fdshare(struct filedesc *fdp)
1913 {
1914
1915 refcount_acquire(&fdp->fd_refcnt);
1916 return (fdp);
1917 }
1918
1919 /*
1920 * Unshare a filedesc structure, if necessary by making a copy
1921 */
1922 void
1923 fdunshare(struct thread *td)
1924 {
1925 struct filedesc *tmp;
1926 struct proc *p = td->td_proc;
1927
1928 if (p->p_fd->fd_refcnt == 1)
1929 return;
1930
1931 tmp = fdcopy(p->p_fd);
1932 fdescfree(td);
1933 p->p_fd = tmp;
1934 }
1935
1936 void
1937 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1938 {
1939
1940 fdescfree(td);
1941 td->td_proc->p_fd = fdp;
1942 }
1943
1944 /*
1945 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1946 * this is to ease callers, not catch errors.
1947 */
1948 struct filedesc *
1949 fdcopy(struct filedesc *fdp)
1950 {
1951 struct filedesc *newfdp;
1952 struct filedescent *nfde, *ofde;
1953 int i;
1954
1955 MPASS(fdp != NULL);
1956
1957 newfdp = fdinit(fdp, true);
1958 /* copy all passable descriptors (i.e. not kqueue) */
1959 newfdp->fd_freefile = -1;
1960 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1961 ofde = &fdp->fd_ofiles[i];
1962 if (ofde->fde_file == NULL ||
1963 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1964 if (newfdp->fd_freefile == -1)
1965 newfdp->fd_freefile = i;
1966 continue;
1967 }
1968 nfde = &newfdp->fd_ofiles[i];
1969 *nfde = *ofde;
1970 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
1971 fhold(nfde->fde_file);
1972 fdused_init(newfdp, i);
1973 newfdp->fd_lastfile = i;
1974 }
1975 if (newfdp->fd_freefile == -1)
1976 newfdp->fd_freefile = i;
1977 newfdp->fd_cmask = fdp->fd_cmask;
1978 FILEDESC_SUNLOCK(fdp);
1979 return (newfdp);
1980 }
1981
1982 /*
1983 * Copies a filedesc structure, while remapping all file descriptors
1984 * stored inside using a translation table.
1985 *
1986 * File descriptors are copied over to the new file descriptor table,
1987 * regardless of whether the close-on-exec flag is set.
1988 */
1989 int
1990 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
1991 struct filedesc **ret)
1992 {
1993 struct filedesc *newfdp;
1994 struct filedescent *nfde, *ofde;
1995 int error, i;
1996
1997 MPASS(fdp != NULL);
1998
1999 newfdp = fdinit(fdp, true);
2000 if (nfds > fdp->fd_lastfile + 1) {
2001 /* New table cannot be larger than the old one. */
2002 error = E2BIG;
2003 goto bad;
2004 }
2005 /* Copy all passable descriptors (i.e. not kqueue). */
2006 newfdp->fd_freefile = nfds;
2007 for (i = 0; i < nfds; ++i) {
2008 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2009 /* File descriptor out of bounds. */
2010 error = EBADF;
2011 goto bad;
2012 }
2013 ofde = &fdp->fd_ofiles[fds[i]];
2014 if (ofde->fde_file == NULL) {
2015 /* Unused file descriptor. */
2016 error = EBADF;
2017 goto bad;
2018 }
2019 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2020 /* File descriptor cannot be passed. */
2021 error = EINVAL;
2022 goto bad;
2023 }
2024 nfde = &newfdp->fd_ofiles[i];
2025 *nfde = *ofde;
2026 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2027 fhold(nfde->fde_file);
2028 fdused_init(newfdp, i);
2029 newfdp->fd_lastfile = i;
2030 }
2031 newfdp->fd_cmask = fdp->fd_cmask;
2032 FILEDESC_SUNLOCK(fdp);
2033 *ret = newfdp;
2034 return (0);
2035 bad:
2036 FILEDESC_SUNLOCK(fdp);
2037 fdescfree_remapped(newfdp);
2038 return (error);
2039 }
2040
2041 /*
2042 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2043 * one of processes using it exits) and the table used to be shared.
2044 */
2045 static void
2046 fdclearlocks(struct thread *td)
2047 {
2048 struct filedesc *fdp;
2049 struct filedesc_to_leader *fdtol;
2050 struct flock lf;
2051 struct file *fp;
2052 struct proc *p;
2053 struct vnode *vp;
2054 int i;
2055
2056 p = td->td_proc;
2057 fdp = p->p_fd;
2058 fdtol = p->p_fdtol;
2059 MPASS(fdtol != NULL);
2060
2061 FILEDESC_XLOCK(fdp);
2062 KASSERT(fdtol->fdl_refcount > 0,
2063 ("filedesc_to_refcount botch: fdl_refcount=%d",
2064 fdtol->fdl_refcount));
2065 if (fdtol->fdl_refcount == 1 &&
2066 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2067 for (i = 0; i <= fdp->fd_lastfile; i++) {
2068 fp = fdp->fd_ofiles[i].fde_file;
2069 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2070 continue;
2071 fhold(fp);
2072 FILEDESC_XUNLOCK(fdp);
2073 lf.l_whence = SEEK_SET;
2074 lf.l_start = 0;
2075 lf.l_len = 0;
2076 lf.l_type = F_UNLCK;
2077 vp = fp->f_vnode;
2078 (void) VOP_ADVLOCK(vp,
2079 (caddr_t)p->p_leader, F_UNLCK,
2080 &lf, F_POSIX);
2081 FILEDESC_XLOCK(fdp);
2082 fdrop(fp, td);
2083 }
2084 }
2085 retry:
2086 if (fdtol->fdl_refcount == 1) {
2087 if (fdp->fd_holdleaderscount > 0 &&
2088 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2089 /*
2090 * close() or kern_dup() has cleared a reference
2091 * in a shared file descriptor table.
2092 */
2093 fdp->fd_holdleaderswakeup = 1;
2094 sx_sleep(&fdp->fd_holdleaderscount,
2095 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2096 goto retry;
2097 }
2098 if (fdtol->fdl_holdcount > 0) {
2099 /*
2100 * Ensure that fdtol->fdl_leader remains
2101 * valid in closef().
2102 */
2103 fdtol->fdl_wakeup = 1;
2104 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2105 "fdlhold", 0);
2106 goto retry;
2107 }
2108 }
2109 fdtol->fdl_refcount--;
2110 if (fdtol->fdl_refcount == 0 &&
2111 fdtol->fdl_holdcount == 0) {
2112 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2113 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2114 } else
2115 fdtol = NULL;
2116 p->p_fdtol = NULL;
2117 FILEDESC_XUNLOCK(fdp);
2118 if (fdtol != NULL)
2119 free(fdtol, M_FILEDESC_TO_LEADER);
2120 }
2121
2122 /*
2123 * Release a filedesc structure.
2124 */
2125 static void
2126 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2127 {
2128 struct filedesc0 *fdp0;
2129 struct freetable *ft, *tft;
2130 struct filedescent *fde;
2131 struct file *fp;
2132 int i;
2133
2134 for (i = 0; i <= fdp->fd_lastfile; i++) {
2135 fde = &fdp->fd_ofiles[i];
2136 fp = fde->fde_file;
2137 if (fp != NULL) {
2138 fdefree_last(fde);
2139 if (needclose)
2140 (void) closef(fp, td);
2141 else
2142 fdrop(fp, td);
2143 }
2144 }
2145
2146 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2147 free(fdp->fd_map, M_FILEDESC);
2148 if (fdp->fd_nfiles > NDFILE)
2149 free(fdp->fd_files, M_FILEDESC);
2150
2151 fdp0 = (struct filedesc0 *)fdp;
2152 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2153 free(ft->ft_table, M_FILEDESC);
2154
2155 fddrop(fdp);
2156 }
2157
2158 void
2159 fdescfree(struct thread *td)
2160 {
2161 struct proc *p;
2162 struct filedesc *fdp;
2163 struct vnode *cdir, *jdir, *rdir;
2164
2165 p = td->td_proc;
2166 fdp = p->p_fd;
2167 MPASS(fdp != NULL);
2168
2169 #ifdef RACCT
2170 if (racct_enable) {
2171 PROC_LOCK(p);
2172 racct_set(p, RACCT_NOFILE, 0);
2173 PROC_UNLOCK(p);
2174 }
2175 #endif
2176
2177 if (p->p_fdtol != NULL)
2178 fdclearlocks(td);
2179
2180 PROC_LOCK(p);
2181 p->p_fd = NULL;
2182 PROC_UNLOCK(p);
2183
2184 if (refcount_release(&fdp->fd_refcnt) == 0)
2185 return;
2186
2187 FILEDESC_XLOCK(fdp);
2188 cdir = fdp->fd_cdir;
2189 fdp->fd_cdir = NULL;
2190 rdir = fdp->fd_rdir;
2191 fdp->fd_rdir = NULL;
2192 jdir = fdp->fd_jdir;
2193 fdp->fd_jdir = NULL;
2194 FILEDESC_XUNLOCK(fdp);
2195
2196 if (cdir != NULL)
2197 vrele(cdir);
2198 if (rdir != NULL)
2199 vrele(rdir);
2200 if (jdir != NULL)
2201 vrele(jdir);
2202
2203 fdescfree_fds(td, fdp, 1);
2204 }
2205
2206 void
2207 fdescfree_remapped(struct filedesc *fdp)
2208 {
2209
2210 if (fdp->fd_cdir != NULL)
2211 vrele(fdp->fd_cdir);
2212 if (fdp->fd_rdir != NULL)
2213 vrele(fdp->fd_rdir);
2214 if (fdp->fd_jdir != NULL)
2215 vrele(fdp->fd_jdir);
2216
2217 fdescfree_fds(curthread, fdp, 0);
2218 }
2219
2220 /*
2221 * For setugid programs, we don't want to people to use that setugidness
2222 * to generate error messages which write to a file which otherwise would
2223 * otherwise be off-limits to the process. We check for filesystems where
2224 * the vnode can change out from under us after execve (like [lin]procfs).
2225 *
2226 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2227 * sufficient. We also don't check for setugidness since we know we are.
2228 */
2229 static bool
2230 is_unsafe(struct file *fp)
2231 {
2232 struct vnode *vp;
2233
2234 if (fp->f_type != DTYPE_VNODE)
2235 return (false);
2236
2237 vp = fp->f_vnode;
2238 return ((vp->v_vflag & VV_PROCDEP) != 0);
2239 }
2240
2241 /*
2242 * Make this setguid thing safe, if at all possible.
2243 */
2244 void
2245 fdsetugidsafety(struct thread *td)
2246 {
2247 struct filedesc *fdp;
2248 struct file *fp;
2249 int i;
2250
2251 fdp = td->td_proc->p_fd;
2252 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2253 MPASS(fdp->fd_nfiles >= 3);
2254 for (i = 0; i <= 2; i++) {
2255 fp = fdp->fd_ofiles[i].fde_file;
2256 if (fp != NULL && is_unsafe(fp)) {
2257 FILEDESC_XLOCK(fdp);
2258 knote_fdclose(td, i);
2259 /*
2260 * NULL-out descriptor prior to close to avoid
2261 * a race while close blocks.
2262 */
2263 fdfree(fdp, i);
2264 FILEDESC_XUNLOCK(fdp);
2265 (void) closef(fp, td);
2266 }
2267 }
2268 }
2269
2270 /*
2271 * If a specific file object occupies a specific file descriptor, close the
2272 * file descriptor entry and drop a reference on the file object. This is a
2273 * convenience function to handle a subsequent error in a function that calls
2274 * falloc() that handles the race that another thread might have closed the
2275 * file descriptor out from under the thread creating the file object.
2276 */
2277 void
2278 fdclose(struct thread *td, struct file *fp, int idx)
2279 {
2280 struct filedesc *fdp = td->td_proc->p_fd;
2281
2282 FILEDESC_XLOCK(fdp);
2283 if (fdp->fd_ofiles[idx].fde_file == fp) {
2284 fdfree(fdp, idx);
2285 FILEDESC_XUNLOCK(fdp);
2286 fdrop(fp, td);
2287 } else
2288 FILEDESC_XUNLOCK(fdp);
2289 }
2290
2291 /*
2292 * Close any files on exec?
2293 */
2294 void
2295 fdcloseexec(struct thread *td)
2296 {
2297 struct filedesc *fdp;
2298 struct filedescent *fde;
2299 struct file *fp;
2300 int i;
2301
2302 fdp = td->td_proc->p_fd;
2303 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2304 for (i = 0; i <= fdp->fd_lastfile; i++) {
2305 fde = &fdp->fd_ofiles[i];
2306 fp = fde->fde_file;
2307 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2308 (fde->fde_flags & UF_EXCLOSE))) {
2309 FILEDESC_XLOCK(fdp);
2310 fdfree(fdp, i);
2311 (void) closefp(fdp, i, fp, td, 0);
2312 FILEDESC_UNLOCK_ASSERT(fdp);
2313 }
2314 }
2315 }
2316
2317 /*
2318 * It is unsafe for set[ug]id processes to be started with file
2319 * descriptors 0..2 closed, as these descriptors are given implicit
2320 * significance in the Standard C library. fdcheckstd() will create a
2321 * descriptor referencing /dev/null for each of stdin, stdout, and
2322 * stderr that is not already open.
2323 */
2324 int
2325 fdcheckstd(struct thread *td)
2326 {
2327 struct filedesc *fdp;
2328 register_t save;
2329 int i, error, devnull;
2330
2331 fdp = td->td_proc->p_fd;
2332 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2333 MPASS(fdp->fd_nfiles >= 3);
2334 devnull = -1;
2335 for (i = 0; i <= 2; i++) {
2336 if (fdp->fd_ofiles[i].fde_file != NULL)
2337 continue;
2338
2339 save = td->td_retval[0];
2340 if (devnull != -1) {
2341 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2342 } else {
2343 error = kern_openat(td, AT_FDCWD, "/dev/null",
2344 UIO_SYSSPACE, O_RDWR, 0);
2345 if (error == 0) {
2346 devnull = td->td_retval[0];
2347 KASSERT(devnull == i, ("we didn't get our fd"));
2348 }
2349 }
2350 td->td_retval[0] = save;
2351 if (error != 0)
2352 return (error);
2353 }
2354 return (0);
2355 }
2356
2357 /*
2358 * Internal form of close. Decrement reference count on file structure.
2359 * Note: td may be NULL when closing a file that was being passed in a
2360 * message.
2361 *
2362 * XXXRW: Giant is not required for the caller, but often will be held; this
2363 * makes it moderately likely the Giant will be recursed in the VFS case.
2364 */
2365 int
2366 closef(struct file *fp, struct thread *td)
2367 {
2368 struct vnode *vp;
2369 struct flock lf;
2370 struct filedesc_to_leader *fdtol;
2371 struct filedesc *fdp;
2372
2373 /*
2374 * POSIX record locking dictates that any close releases ALL
2375 * locks owned by this process. This is handled by setting
2376 * a flag in the unlock to free ONLY locks obeying POSIX
2377 * semantics, and not to free BSD-style file locks.
2378 * If the descriptor was in a message, POSIX-style locks
2379 * aren't passed with the descriptor, and the thread pointer
2380 * will be NULL. Callers should be careful only to pass a
2381 * NULL thread pointer when there really is no owning
2382 * context that might have locks, or the locks will be
2383 * leaked.
2384 */
2385 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2386 vp = fp->f_vnode;
2387 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2388 lf.l_whence = SEEK_SET;
2389 lf.l_start = 0;
2390 lf.l_len = 0;
2391 lf.l_type = F_UNLCK;
2392 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2393 F_UNLCK, &lf, F_POSIX);
2394 }
2395 fdtol = td->td_proc->p_fdtol;
2396 if (fdtol != NULL) {
2397 /*
2398 * Handle special case where file descriptor table is
2399 * shared between multiple process leaders.
2400 */
2401 fdp = td->td_proc->p_fd;
2402 FILEDESC_XLOCK(fdp);
2403 for (fdtol = fdtol->fdl_next;
2404 fdtol != td->td_proc->p_fdtol;
2405 fdtol = fdtol->fdl_next) {
2406 if ((fdtol->fdl_leader->p_flag &
2407 P_ADVLOCK) == 0)
2408 continue;
2409 fdtol->fdl_holdcount++;
2410 FILEDESC_XUNLOCK(fdp);
2411 lf.l_whence = SEEK_SET;
2412 lf.l_start = 0;
2413 lf.l_len = 0;
2414 lf.l_type = F_UNLCK;
2415 vp = fp->f_vnode;
2416 (void) VOP_ADVLOCK(vp,
2417 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2418 F_POSIX);
2419 FILEDESC_XLOCK(fdp);
2420 fdtol->fdl_holdcount--;
2421 if (fdtol->fdl_holdcount == 0 &&
2422 fdtol->fdl_wakeup != 0) {
2423 fdtol->fdl_wakeup = 0;
2424 wakeup(fdtol);
2425 }
2426 }
2427 FILEDESC_XUNLOCK(fdp);
2428 }
2429 }
2430 return (fdrop(fp, td));
2431 }
2432
2433 /*
2434 * Initialize the file pointer with the specified properties.
2435 *
2436 * The ops are set with release semantics to be certain that the flags, type,
2437 * and data are visible when ops is. This is to prevent ops methods from being
2438 * called with bad data.
2439 */
2440 void
2441 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2442 {
2443 fp->f_data = data;
2444 fp->f_flag = flag;
2445 fp->f_type = type;
2446 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2447 }
2448
2449 int
2450 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2451 struct file **fpp, struct filecaps *havecapsp)
2452 {
2453 struct filedescent *fde;
2454 int error;
2455
2456 FILEDESC_LOCK_ASSERT(fdp);
2457
2458 fde = fdeget_locked(fdp, fd);
2459 if (fde == NULL) {
2460 error = EBADF;
2461 goto out;
2462 }
2463
2464 #ifdef CAPABILITIES
2465 error = cap_check(cap_rights_fde(fde), needrightsp);
2466 if (error != 0)
2467 goto out;
2468 #endif
2469
2470 if (havecapsp != NULL)
2471 filecaps_copy(&fde->fde_caps, havecapsp, true);
2472
2473 *fpp = fde->fde_file;
2474
2475 error = 0;
2476 out:
2477 return (error);
2478 }
2479
2480 int
2481 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2482 struct file **fpp, struct filecaps *havecapsp)
2483 {
2484 struct filedesc *fdp = td->td_proc->p_fd;
2485 int error;
2486 #ifndef CAPABILITIES
2487 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2488 if (error == 0 && havecapsp != NULL)
2489 filecaps_fill(havecapsp);
2490 #else
2491 struct file *fp;
2492 seq_t seq;
2493
2494 for (;;) {
2495 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2496 if (error != 0)
2497 return (error);
2498
2499 if (havecapsp != NULL) {
2500 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2501 havecapsp, false)) {
2502 fdrop(fp, td);
2503 goto get_locked;
2504 }
2505 }
2506
2507 if (!fd_modified(fdp, fd, seq))
2508 break;
2509 fdrop(fp, td);
2510 }
2511
2512 *fpp = fp;
2513 return (0);
2514
2515 get_locked:
2516 FILEDESC_SLOCK(fdp);
2517 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2518 if (error == 0)
2519 fhold(*fpp);
2520 FILEDESC_SUNLOCK(fdp);
2521 #endif
2522 return (error);
2523 }
2524
2525 int
2526 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2527 struct file **fpp, seq_t *seqp)
2528 {
2529 #ifdef CAPABILITIES
2530 struct filedescent *fde;
2531 #endif
2532 struct fdescenttbl *fdt;
2533 struct file *fp;
2534 u_int count;
2535 #ifdef CAPABILITIES
2536 seq_t seq;
2537 cap_rights_t haverights;
2538 int error;
2539 #endif
2540
2541 fdt = fdp->fd_files;
2542 if ((u_int)fd >= fdt->fdt_nfiles)
2543 return (EBADF);
2544 /*
2545 * Fetch the descriptor locklessly. We avoid fdrop() races by
2546 * never raising a refcount above 0. To accomplish this we have
2547 * to use a cmpset loop rather than an atomic_add. The descriptor
2548 * must be re-verified once we acquire a reference to be certain
2549 * that the identity is still correct and we did not lose a race
2550 * due to preemption.
2551 */
2552 for (;;) {
2553 #ifdef CAPABILITIES
2554 seq = seq_read(fd_seq(fdt, fd));
2555 fde = &fdt->fdt_ofiles[fd];
2556 haverights = *cap_rights_fde(fde);
2557 fp = fde->fde_file;
2558 if (!seq_consistent(fd_seq(fdt, fd), seq))
2559 continue;
2560 #else
2561 fp = fdt->fdt_ofiles[fd].fde_file;
2562 #endif
2563 if (fp == NULL)
2564 return (EBADF);
2565 #ifdef CAPABILITIES
2566 error = cap_check(&haverights, needrightsp);
2567 if (error != 0)
2568 return (error);
2569 #endif
2570 count = fp->f_count;
2571 retry:
2572 if (count == 0) {
2573 /*
2574 * Force a reload. Other thread could reallocate the
2575 * table before this fd was closed, so it possible that
2576 * there is a stale fp pointer in cached version.
2577 */
2578 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2579 continue;
2580 }
2581 /*
2582 * Use an acquire barrier to force re-reading of fdt so it is
2583 * refreshed for verification.
2584 */
2585 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2586 goto retry;
2587 fdt = fdp->fd_files;
2588 #ifdef CAPABILITIES
2589 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2590 #else
2591 if (fp == fdt->fdt_ofiles[fd].fde_file)
2592 #endif
2593 break;
2594 fdrop(fp, curthread);
2595 }
2596 *fpp = fp;
2597 if (seqp != NULL) {
2598 #ifdef CAPABILITIES
2599 *seqp = seq;
2600 #endif
2601 }
2602 return (0);
2603 }
2604
2605 /*
2606 * Extract the file pointer associated with the specified descriptor for the
2607 * current user process.
2608 *
2609 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2610 * returned.
2611 *
2612 * File's rights will be checked against the capability rights mask.
2613 *
2614 * If an error occurred the non-zero error is returned and *fpp is set to
2615 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2616 * responsible for fdrop().
2617 */
2618 static __inline int
2619 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2620 cap_rights_t *needrightsp, seq_t *seqp)
2621 {
2622 struct filedesc *fdp;
2623 struct file *fp;
2624 int error;
2625
2626 *fpp = NULL;
2627 fdp = td->td_proc->p_fd;
2628 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2629 if (error != 0)
2630 return (error);
2631 if (fp->f_ops == &badfileops) {
2632 fdrop(fp, td);
2633 return (EBADF);
2634 }
2635
2636 /*
2637 * FREAD and FWRITE failure return EBADF as per POSIX.
2638 */
2639 error = 0;
2640 switch (flags) {
2641 case FREAD:
2642 case FWRITE:
2643 if ((fp->f_flag & flags) == 0)
2644 error = EBADF;
2645 break;
2646 case FEXEC:
2647 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2648 ((fp->f_flag & FWRITE) != 0))
2649 error = EBADF;
2650 break;
2651 case 0:
2652 break;
2653 default:
2654 KASSERT(0, ("wrong flags"));
2655 }
2656
2657 if (error != 0) {
2658 fdrop(fp, td);
2659 return (error);
2660 }
2661
2662 *fpp = fp;
2663 return (0);
2664 }
2665
2666 int
2667 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2668 {
2669
2670 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2671 }
2672
2673 int
2674 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2675 struct file **fpp)
2676 {
2677 int error;
2678 #ifndef CAPABILITIES
2679 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2680 if (maxprotp != NULL)
2681 *maxprotp = VM_PROT_ALL;
2682 #else
2683 struct filedesc *fdp = td->td_proc->p_fd;
2684 seq_t seq;
2685
2686 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2687 for (;;) {
2688 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2689 if (error != 0)
2690 return (error);
2691 /*
2692 * If requested, convert capability rights to access flags.
2693 */
2694 if (maxprotp != NULL)
2695 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2696 if (!fd_modified(fdp, fd, seq))
2697 break;
2698 fdrop(*fpp, td);
2699 }
2700 #endif
2701 return (error);
2702 }
2703
2704 int
2705 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2706 {
2707
2708 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2709 }
2710
2711 int
2712 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2713 {
2714
2715 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2716 }
2717
2718 int
2719 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2720 struct file **fpp)
2721 {
2722 struct filedesc *fdp = td->td_proc->p_fd;
2723 #ifndef CAPABILITIES
2724 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2725 #else
2726 int error;
2727 seq_t seq;
2728
2729 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2730 for (;;) {
2731 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2732 if (error != 0)
2733 return (error);
2734 error = cap_fcntl_check(fdp, fd, needfcntl);
2735 if (!fd_modified(fdp, fd, seq))
2736 break;
2737 fdrop(*fpp, td);
2738 }
2739 if (error != 0) {
2740 fdrop(*fpp, td);
2741 *fpp = NULL;
2742 }
2743 return (error);
2744 #endif
2745 }
2746
2747 /*
2748 * Like fget() but loads the underlying vnode, or returns an error if the
2749 * descriptor does not represent a vnode. Note that pipes use vnodes but
2750 * never have VM objects. The returned vnode will be vref()'d.
2751 *
2752 * XXX: what about the unused flags ?
2753 */
2754 static __inline int
2755 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2756 struct vnode **vpp)
2757 {
2758 struct file *fp;
2759 int error;
2760
2761 *vpp = NULL;
2762 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2763 if (error != 0)
2764 return (error);
2765 if (fp->f_vnode == NULL) {
2766 error = EINVAL;
2767 } else {
2768 *vpp = fp->f_vnode;
2769 vrefact(*vpp);
2770 }
2771 fdrop(fp, td);
2772
2773 return (error);
2774 }
2775
2776 int
2777 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2778 {
2779
2780 return (_fgetvp(td, fd, 0, rightsp, vpp));
2781 }
2782
2783 int
2784 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2785 struct filecaps *havecaps, struct vnode **vpp)
2786 {
2787 struct filedesc *fdp;
2788 struct filecaps caps;
2789 struct file *fp;
2790 int error;
2791
2792 fdp = td->td_proc->p_fd;
2793 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2794 if (error != 0)
2795 return (error);
2796 if (fp->f_ops == &badfileops) {
2797 error = EBADF;
2798 goto out;
2799 }
2800 if (fp->f_vnode == NULL) {
2801 error = EINVAL;
2802 goto out;
2803 }
2804
2805 *havecaps = caps;
2806 *vpp = fp->f_vnode;
2807 vrefact(*vpp);
2808
2809 return (0);
2810 out:
2811 filecaps_free(&caps);
2812 return (error);
2813 }
2814
2815 int
2816 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2817 {
2818
2819 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2820 }
2821
2822 int
2823 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2824 {
2825
2826 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2827 }
2828
2829 #ifdef notyet
2830 int
2831 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2832 struct vnode **vpp)
2833 {
2834
2835 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2836 }
2837 #endif
2838
2839 /*
2840 * Handle the last reference to a file being closed.
2841 */
2842 int
2843 _fdrop(struct file *fp, struct thread *td)
2844 {
2845 int error;
2846
2847 if (fp->f_count != 0)
2848 panic("fdrop: count %d", fp->f_count);
2849 error = fo_close(fp, td);
2850 atomic_subtract_int(&openfiles, 1);
2851 crfree(fp->f_cred);
2852 free(fp->f_advice, M_FADVISE);
2853 uma_zfree(file_zone, fp);
2854
2855 return (error);
2856 }
2857
2858 /*
2859 * Apply an advisory lock on a file descriptor.
2860 *
2861 * Just attempt to get a record lock of the requested type on the entire file
2862 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2863 */
2864 #ifndef _SYS_SYSPROTO_H_
2865 struct flock_args {
2866 int fd;
2867 int how;
2868 };
2869 #endif
2870 /* ARGSUSED */
2871 int
2872 sys_flock(struct thread *td, struct flock_args *uap)
2873 {
2874 struct file *fp;
2875 struct vnode *vp;
2876 struct flock lf;
2877 cap_rights_t rights;
2878 int error;
2879
2880 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2881 if (error != 0)
2882 return (error);
2883 if (fp->f_type != DTYPE_VNODE) {
2884 fdrop(fp, td);
2885 return (EOPNOTSUPP);
2886 }
2887
2888 vp = fp->f_vnode;
2889 lf.l_whence = SEEK_SET;
2890 lf.l_start = 0;
2891 lf.l_len = 0;
2892 if (uap->how & LOCK_UN) {
2893 lf.l_type = F_UNLCK;
2894 atomic_clear_int(&fp->f_flag, FHASLOCK);
2895 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2896 goto done2;
2897 }
2898 if (uap->how & LOCK_EX)
2899 lf.l_type = F_WRLCK;
2900 else if (uap->how & LOCK_SH)
2901 lf.l_type = F_RDLCK;
2902 else {
2903 error = EBADF;
2904 goto done2;
2905 }
2906 atomic_set_int(&fp->f_flag, FHASLOCK);
2907 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2908 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2909 done2:
2910 fdrop(fp, td);
2911 return (error);
2912 }
2913 /*
2914 * Duplicate the specified descriptor to a free descriptor.
2915 */
2916 int
2917 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2918 int openerror, int *indxp)
2919 {
2920 struct filedescent *newfde, *oldfde;
2921 struct file *fp;
2922 int error, indx;
2923
2924 KASSERT(openerror == ENODEV || openerror == ENXIO,
2925 ("unexpected error %d in %s", openerror, __func__));
2926
2927 /*
2928 * If the to-be-dup'd fd number is greater than the allowed number
2929 * of file descriptors, or the fd to be dup'd has already been
2930 * closed, then reject.
2931 */
2932 FILEDESC_XLOCK(fdp);
2933 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2934 FILEDESC_XUNLOCK(fdp);
2935 return (EBADF);
2936 }
2937
2938 error = fdalloc(td, 0, &indx);
2939 if (error != 0) {
2940 FILEDESC_XUNLOCK(fdp);
2941 return (error);
2942 }
2943
2944 /*
2945 * There are two cases of interest here.
2946 *
2947 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2948 *
2949 * For ENXIO steal away the file structure from (dfd) and store it in
2950 * (indx). (dfd) is effectively closed by this operation.
2951 */
2952 switch (openerror) {
2953 case ENODEV:
2954 /*
2955 * Check that the mode the file is being opened for is a
2956 * subset of the mode of the existing descriptor.
2957 */
2958 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2959 fdunused(fdp, indx);
2960 FILEDESC_XUNLOCK(fdp);
2961 return (EACCES);
2962 }
2963 fhold(fp);
2964 newfde = &fdp->fd_ofiles[indx];
2965 oldfde = &fdp->fd_ofiles[dfd];
2966 #ifdef CAPABILITIES
2967 seq_write_begin(&newfde->fde_seq);
2968 #endif
2969 memcpy(newfde, oldfde, fde_change_size);
2970 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
2971 #ifdef CAPABILITIES
2972 seq_write_end(&newfde->fde_seq);
2973 #endif
2974 break;
2975 case ENXIO:
2976 /*
2977 * Steal away the file pointer from dfd and stuff it into indx.
2978 */
2979 newfde = &fdp->fd_ofiles[indx];
2980 oldfde = &fdp->fd_ofiles[dfd];
2981 #ifdef CAPABILITIES
2982 seq_write_begin(&newfde->fde_seq);
2983 #endif
2984 memcpy(newfde, oldfde, fde_change_size);
2985 oldfde->fde_file = NULL;
2986 fdunused(fdp, dfd);
2987 #ifdef CAPABILITIES
2988 seq_write_end(&newfde->fde_seq);
2989 #endif
2990 break;
2991 }
2992 FILEDESC_XUNLOCK(fdp);
2993 *indxp = indx;
2994 return (0);
2995 }
2996
2997 /*
2998 * This sysctl determines if we will allow a process to chroot(2) if it
2999 * has a directory open:
3000 * 0: disallowed for all processes.
3001 * 1: allowed for processes that were not already chroot(2)'ed.
3002 * 2: allowed for all processes.
3003 */
3004
3005 static int chroot_allow_open_directories = 1;
3006
3007 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3008 &chroot_allow_open_directories, 0,
3009 "Allow a process to chroot(2) if it has a directory open");
3010
3011 /*
3012 * Helper function for raised chroot(2) security function: Refuse if
3013 * any filedescriptors are open directories.
3014 */
3015 static int
3016 chroot_refuse_vdir_fds(struct filedesc *fdp)
3017 {
3018 struct vnode *vp;
3019 struct file *fp;
3020 int fd;
3021
3022 FILEDESC_LOCK_ASSERT(fdp);
3023
3024 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3025 fp = fget_locked(fdp, fd);
3026 if (fp == NULL)
3027 continue;
3028 if (fp->f_type == DTYPE_VNODE) {
3029 vp = fp->f_vnode;
3030 if (vp->v_type == VDIR)
3031 return (EPERM);
3032 }
3033 }
3034 return (0);
3035 }
3036
3037 /*
3038 * Common routine for kern_chroot() and jail_attach(). The caller is
3039 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3040 * authorize this operation.
3041 */
3042 int
3043 pwd_chroot(struct thread *td, struct vnode *vp)
3044 {
3045 struct filedesc *fdp;
3046 struct vnode *oldvp;
3047 int error;
3048
3049 fdp = td->td_proc->p_fd;
3050 FILEDESC_XLOCK(fdp);
3051 if (chroot_allow_open_directories == 0 ||
3052 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3053 error = chroot_refuse_vdir_fds(fdp);
3054 if (error != 0) {
3055 FILEDESC_XUNLOCK(fdp);
3056 return (error);
3057 }
3058 }
3059 oldvp = fdp->fd_rdir;
3060 vrefact(vp);
3061 fdp->fd_rdir = vp;
3062 if (fdp->fd_jdir == NULL) {
3063 vrefact(vp);
3064 fdp->fd_jdir = vp;
3065 }
3066 FILEDESC_XUNLOCK(fdp);
3067 vrele(oldvp);
3068 return (0);
3069 }
3070
3071 void
3072 pwd_chdir(struct thread *td, struct vnode *vp)
3073 {
3074 struct filedesc *fdp;
3075 struct vnode *oldvp;
3076
3077 fdp = td->td_proc->p_fd;
3078 FILEDESC_XLOCK(fdp);
3079 VNASSERT(vp->v_usecount > 0, vp,
3080 ("chdir to a vnode with zero usecount"));
3081 oldvp = fdp->fd_cdir;
3082 fdp->fd_cdir = vp;
3083 FILEDESC_XUNLOCK(fdp);
3084 vrele(oldvp);
3085 }
3086
3087 /*
3088 * Scan all active processes and prisons to see if any of them have a current
3089 * or root directory of `olddp'. If so, replace them with the new mount point.
3090 */
3091 void
3092 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3093 {
3094 struct filedesc *fdp;
3095 struct prison *pr;
3096 struct proc *p;
3097 int nrele;
3098
3099 if (vrefcnt(olddp) == 1)
3100 return;
3101 nrele = 0;
3102 sx_slock(&allproc_lock);
3103 FOREACH_PROC_IN_SYSTEM(p) {
3104 PROC_LOCK(p);
3105 fdp = fdhold(p);
3106 PROC_UNLOCK(p);
3107 if (fdp == NULL)
3108 continue;
3109 FILEDESC_XLOCK(fdp);
3110 if (fdp->fd_cdir == olddp) {
3111 vrefact(newdp);
3112 fdp->fd_cdir = newdp;
3113 nrele++;
3114 }
3115 if (fdp->fd_rdir == olddp) {
3116 vrefact(newdp);
3117 fdp->fd_rdir = newdp;
3118 nrele++;
3119 }
3120 if (fdp->fd_jdir == olddp) {
3121 vrefact(newdp);
3122 fdp->fd_jdir = newdp;
3123 nrele++;
3124 }
3125 FILEDESC_XUNLOCK(fdp);
3126 fddrop(fdp);
3127 }
3128 sx_sunlock(&allproc_lock);
3129 if (rootvnode == olddp) {
3130 vrefact(newdp);
3131 rootvnode = newdp;
3132 nrele++;
3133 }
3134 mtx_lock(&prison0.pr_mtx);
3135 if (prison0.pr_root == olddp) {
3136 vrefact(newdp);
3137 prison0.pr_root = newdp;
3138 nrele++;
3139 }
3140 mtx_unlock(&prison0.pr_mtx);
3141 sx_slock(&allprison_lock);
3142 TAILQ_FOREACH(pr, &allprison, pr_list) {
3143 mtx_lock(&pr->pr_mtx);
3144 if (pr->pr_root == olddp) {
3145 vrefact(newdp);
3146 pr->pr_root = newdp;
3147 nrele++;
3148 }
3149 mtx_unlock(&pr->pr_mtx);
3150 }
3151 sx_sunlock(&allprison_lock);
3152 while (nrele--)
3153 vrele(olddp);
3154 }
3155
3156 struct filedesc_to_leader *
3157 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3158 {
3159 struct filedesc_to_leader *fdtol;
3160
3161 fdtol = malloc(sizeof(struct filedesc_to_leader),
3162 M_FILEDESC_TO_LEADER, M_WAITOK);
3163 fdtol->fdl_refcount = 1;
3164 fdtol->fdl_holdcount = 0;
3165 fdtol->fdl_wakeup = 0;
3166 fdtol->fdl_leader = leader;
3167 if (old != NULL) {
3168 FILEDESC_XLOCK(fdp);
3169 fdtol->fdl_next = old->fdl_next;
3170 fdtol->fdl_prev = old;
3171 old->fdl_next = fdtol;
3172 fdtol->fdl_next->fdl_prev = fdtol;
3173 FILEDESC_XUNLOCK(fdp);
3174 } else {
3175 fdtol->fdl_next = fdtol;
3176 fdtol->fdl_prev = fdtol;
3177 }
3178 return (fdtol);
3179 }
3180
3181 static int
3182 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3183 {
3184 struct filedesc *fdp;
3185 int i, count, slots;
3186
3187 if (*(int *)arg1 != 0)
3188 return (EINVAL);
3189
3190 fdp = curproc->p_fd;
3191 count = 0;
3192 FILEDESC_SLOCK(fdp);
3193 slots = NDSLOTS(fdp->fd_lastfile + 1);
3194 for (i = 0; i < slots; i++)
3195 count += bitcountl(fdp->fd_map[i]);
3196 FILEDESC_SUNLOCK(fdp);
3197
3198 return (SYSCTL_OUT(req, &count, sizeof(count)));
3199 }
3200
3201 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3202 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3203 "Number of open file descriptors");
3204
3205 /*
3206 * Get file structures globally.
3207 */
3208 static int
3209 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3210 {
3211 struct xfile xf;
3212 struct filedesc *fdp;
3213 struct file *fp;
3214 struct proc *p;
3215 int error, n;
3216
3217 error = sysctl_wire_old_buffer(req, 0);
3218 if (error != 0)
3219 return (error);
3220 if (req->oldptr == NULL) {
3221 n = 0;
3222 sx_slock(&allproc_lock);
3223 FOREACH_PROC_IN_SYSTEM(p) {
3224 PROC_LOCK(p);
3225 if (p->p_state == PRS_NEW) {
3226 PROC_UNLOCK(p);
3227 continue;
3228 }
3229 fdp = fdhold(p);
3230 PROC_UNLOCK(p);
3231 if (fdp == NULL)
3232 continue;
3233 /* overestimates sparse tables. */
3234 if (fdp->fd_lastfile > 0)
3235 n += fdp->fd_lastfile;
3236 fddrop(fdp);
3237 }
3238 sx_sunlock(&allproc_lock);
3239 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3240 }
3241 error = 0;
3242 bzero(&xf, sizeof(xf));
3243 xf.xf_size = sizeof(xf);
3244 sx_slock(&allproc_lock);
3245 FOREACH_PROC_IN_SYSTEM(p) {
3246 PROC_LOCK(p);
3247 if (p->p_state == PRS_NEW) {
3248 PROC_UNLOCK(p);
3249 continue;
3250 }
3251 if (p_cansee(req->td, p) != 0) {
3252 PROC_UNLOCK(p);
3253 continue;
3254 }
3255 xf.xf_pid = p->p_pid;
3256 xf.xf_uid = p->p_ucred->cr_uid;
3257 fdp = fdhold(p);
3258 PROC_UNLOCK(p);
3259 if (fdp == NULL)
3260 continue;
3261 FILEDESC_SLOCK(fdp);
3262 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3263 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3264 continue;
3265 xf.xf_fd = n;
3266 xf.xf_file = fp;
3267 xf.xf_data = fp->f_data;
3268 xf.xf_vnode = fp->f_vnode;
3269 xf.xf_type = fp->f_type;
3270 xf.xf_count = fp->f_count;
3271 xf.xf_msgcount = 0;
3272 xf.xf_offset = foffset_get(fp);
3273 xf.xf_flag = fp->f_flag;
3274 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3275 if (error)
3276 break;
3277 }
3278 FILEDESC_SUNLOCK(fdp);
3279 fddrop(fdp);
3280 if (error)
3281 break;
3282 }
3283 sx_sunlock(&allproc_lock);
3284 return (error);
3285 }
3286
3287 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3288 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3289
3290 #ifdef KINFO_FILE_SIZE
3291 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3292 #endif
3293
3294 static int
3295 xlate_fflags(int fflags)
3296 {
3297 static const struct {
3298 int fflag;
3299 int kf_fflag;
3300 } fflags_table[] = {
3301 { FAPPEND, KF_FLAG_APPEND },
3302 { FASYNC, KF_FLAG_ASYNC },
3303 { FFSYNC, KF_FLAG_FSYNC },
3304 { FHASLOCK, KF_FLAG_HASLOCK },
3305 { FNONBLOCK, KF_FLAG_NONBLOCK },
3306 { FREAD, KF_FLAG_READ },
3307 { FWRITE, KF_FLAG_WRITE },
3308 { O_CREAT, KF_FLAG_CREAT },
3309 { O_DIRECT, KF_FLAG_DIRECT },
3310 { O_EXCL, KF_FLAG_EXCL },
3311 { O_EXEC, KF_FLAG_EXEC },
3312 { O_EXLOCK, KF_FLAG_EXLOCK },
3313 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3314 { O_SHLOCK, KF_FLAG_SHLOCK },
3315 { O_TRUNC, KF_FLAG_TRUNC }
3316 };
3317 unsigned int i;
3318 int kflags;
3319
3320 kflags = 0;
3321 for (i = 0; i < nitems(fflags_table); i++)
3322 if (fflags & fflags_table[i].fflag)
3323 kflags |= fflags_table[i].kf_fflag;
3324 return (kflags);
3325 }
3326
3327 /* Trim unused data from kf_path by truncating the structure size. */
3328 static void
3329 pack_kinfo(struct kinfo_file *kif)
3330 {
3331
3332 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3333 strlen(kif->kf_path) + 1;
3334 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3335 }
3336
3337 static void
3338 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3339 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3340 {
3341 int error;
3342
3343 bzero(kif, sizeof(*kif));
3344
3345 /* Set a default type to allow for empty fill_kinfo() methods. */
3346 kif->kf_type = KF_TYPE_UNKNOWN;
3347 kif->kf_flags = xlate_fflags(fp->f_flag);
3348 if (rightsp != NULL)
3349 kif->kf_cap_rights = *rightsp;
3350 else
3351 cap_rights_init(&kif->kf_cap_rights);
3352 kif->kf_fd = fd;
3353 kif->kf_ref_count = fp->f_count;
3354 kif->kf_offset = foffset_get(fp);
3355
3356 /*
3357 * This may drop the filedesc lock, so the 'fp' cannot be
3358 * accessed after this call.
3359 */
3360 error = fo_fill_kinfo(fp, kif, fdp);
3361 if (error == 0)
3362 kif->kf_status |= KF_ATTR_VALID;
3363 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3364 pack_kinfo(kif);
3365 else
3366 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3367 }
3368
3369 static void
3370 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3371 struct kinfo_file *kif, int flags)
3372 {
3373 int error;
3374
3375 bzero(kif, sizeof(*kif));
3376
3377 kif->kf_type = KF_TYPE_VNODE;
3378 error = vn_fill_kinfo_vnode(vp, kif);
3379 if (error == 0)
3380 kif->kf_status |= KF_ATTR_VALID;
3381 kif->kf_flags = xlate_fflags(fflags);
3382 cap_rights_init(&kif->kf_cap_rights);
3383 kif->kf_fd = fd;
3384 kif->kf_ref_count = -1;
3385 kif->kf_offset = -1;
3386 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3387 pack_kinfo(kif);
3388 else
3389 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3390 vrele(vp);
3391 }
3392
3393 struct export_fd_buf {
3394 struct filedesc *fdp;
3395 struct sbuf *sb;
3396 ssize_t remainder;
3397 struct kinfo_file kif;
3398 int flags;
3399 };
3400
3401 static int
3402 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3403 {
3404 struct kinfo_file *kif;
3405
3406 kif = &efbuf->kif;
3407 if (efbuf->remainder != -1) {
3408 if (efbuf->remainder < kif->kf_structsize) {
3409 /* Terminate export. */
3410 efbuf->remainder = 0;
3411 return (0);
3412 }
3413 efbuf->remainder -= kif->kf_structsize;
3414 }
3415 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3416 }
3417
3418 static int
3419 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3420 struct export_fd_buf *efbuf)
3421 {
3422 int error;
3423
3424 if (efbuf->remainder == 0)
3425 return (0);
3426 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3427 efbuf->flags);
3428 FILEDESC_SUNLOCK(efbuf->fdp);
3429 error = export_kinfo_to_sb(efbuf);
3430 FILEDESC_SLOCK(efbuf->fdp);
3431 return (error);
3432 }
3433
3434 static int
3435 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3436 struct export_fd_buf *efbuf)
3437 {
3438 int error;
3439
3440 if (efbuf->remainder == 0)
3441 return (0);
3442 if (efbuf->fdp != NULL)
3443 FILEDESC_SUNLOCK(efbuf->fdp);
3444 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3445 error = export_kinfo_to_sb(efbuf);
3446 if (efbuf->fdp != NULL)
3447 FILEDESC_SLOCK(efbuf->fdp);
3448 return (error);
3449 }
3450
3451 /*
3452 * Store a process file descriptor information to sbuf.
3453 *
3454 * Takes a locked proc as argument, and returns with the proc unlocked.
3455 */
3456 int
3457 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3458 int flags)
3459 {
3460 struct file *fp;
3461 struct filedesc *fdp;
3462 struct export_fd_buf *efbuf;
3463 struct vnode *cttyvp, *textvp, *tracevp;
3464 int error, i;
3465 cap_rights_t rights;
3466
3467 PROC_LOCK_ASSERT(p, MA_OWNED);
3468
3469 /* ktrace vnode */
3470 tracevp = p->p_tracevp;
3471 if (tracevp != NULL)
3472 vrefact(tracevp);
3473 /* text vnode */
3474 textvp = p->p_textvp;
3475 if (textvp != NULL)
3476 vrefact(textvp);
3477 /* Controlling tty. */
3478 cttyvp = NULL;
3479 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3480 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3481 if (cttyvp != NULL)
3482 vrefact(cttyvp);
3483 }
3484 fdp = fdhold(p);
3485 PROC_UNLOCK(p);
3486 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3487 efbuf->fdp = NULL;
3488 efbuf->sb = sb;
3489 efbuf->remainder = maxlen;
3490 efbuf->flags = flags;
3491 if (tracevp != NULL)
3492 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3493 efbuf);
3494 if (textvp != NULL)
3495 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3496 if (cttyvp != NULL)
3497 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3498 efbuf);
3499 error = 0;
3500 if (fdp == NULL)
3501 goto fail;
3502 efbuf->fdp = fdp;
3503 FILEDESC_SLOCK(fdp);
3504 /* working directory */
3505 if (fdp->fd_cdir != NULL) {
3506 vrefact(fdp->fd_cdir);
3507 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3508 }
3509 /* root directory */
3510 if (fdp->fd_rdir != NULL) {
3511 vrefact(fdp->fd_rdir);
3512 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3513 }
3514 /* jail directory */
3515 if (fdp->fd_jdir != NULL) {
3516 vrefact(fdp->fd_jdir);
3517 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3518 }
3519 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3520 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3521 continue;
3522 #ifdef CAPABILITIES
3523 rights = *cap_rights(fdp, i);
3524 #else /* !CAPABILITIES */
3525 cap_rights_init(&rights);
3526 #endif
3527 /*
3528 * Create sysctl entry. It is OK to drop the filedesc
3529 * lock inside of export_file_to_sb() as we will
3530 * re-validate and re-evaluate its properties when the
3531 * loop continues.
3532 */
3533 error = export_file_to_sb(fp, i, &rights, efbuf);
3534 if (error != 0 || efbuf->remainder == 0)
3535 break;
3536 }
3537 FILEDESC_SUNLOCK(fdp);
3538 fddrop(fdp);
3539 fail:
3540 free(efbuf, M_TEMP);
3541 return (error);
3542 }
3543
3544 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3545
3546 /*
3547 * Get per-process file descriptors for use by procstat(1), et al.
3548 */
3549 static int
3550 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3551 {
3552 struct sbuf sb;
3553 struct proc *p;
3554 ssize_t maxlen;
3555 int error, error2, *name;
3556
3557 name = (int *)arg1;
3558
3559 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3560 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3561 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3562 if (error != 0) {
3563 sbuf_delete(&sb);
3564 return (error);
3565 }
3566 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3567 error = kern_proc_filedesc_out(p, &sb, maxlen,
3568 KERN_FILEDESC_PACK_KINFO);
3569 error2 = sbuf_finish(&sb);
3570 sbuf_delete(&sb);
3571 return (error != 0 ? error : error2);
3572 }
3573
3574 #ifdef KINFO_OFILE_SIZE
3575 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3576 #endif
3577
3578 #ifdef COMPAT_FREEBSD7
3579 static void
3580 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3581 {
3582
3583 okif->kf_structsize = sizeof(*okif);
3584 okif->kf_type = kif->kf_type;
3585 okif->kf_fd = kif->kf_fd;
3586 okif->kf_ref_count = kif->kf_ref_count;
3587 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3588 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3589 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3590 okif->kf_offset = kif->kf_offset;
3591 okif->kf_vnode_type = kif->kf_vnode_type;
3592 okif->kf_sock_domain = kif->kf_sock_domain;
3593 okif->kf_sock_type = kif->kf_sock_type;
3594 okif->kf_sock_protocol = kif->kf_sock_protocol;
3595 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3596 okif->kf_sa_local = kif->kf_sa_local;
3597 okif->kf_sa_peer = kif->kf_sa_peer;
3598 }
3599
3600 static int
3601 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3602 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3603 {
3604 int error;
3605
3606 vrefact(vp);
3607 FILEDESC_SUNLOCK(fdp);
3608 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3609 kinfo_to_okinfo(kif, okif);
3610 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3611 FILEDESC_SLOCK(fdp);
3612 return (error);
3613 }
3614
3615 /*
3616 * Get per-process file descriptors for use by procstat(1), et al.
3617 */
3618 static int
3619 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3620 {
3621 struct kinfo_ofile *okif;
3622 struct kinfo_file *kif;
3623 struct filedesc *fdp;
3624 int error, i, *name;
3625 struct file *fp;
3626 struct proc *p;
3627
3628 name = (int *)arg1;
3629 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3630 if (error != 0)
3631 return (error);
3632 fdp = fdhold(p);
3633 PROC_UNLOCK(p);
3634 if (fdp == NULL)
3635 return (ENOENT);
3636 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3637 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3638 FILEDESC_SLOCK(fdp);
3639 if (fdp->fd_cdir != NULL)
3640 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3641 okif, fdp, req);
3642 if (fdp->fd_rdir != NULL)
3643 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3644 okif, fdp, req);
3645 if (fdp->fd_jdir != NULL)
3646 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3647 okif, fdp, req);
3648 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3649 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3650 continue;
3651 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3652 KERN_FILEDESC_PACK_KINFO);
3653 FILEDESC_SUNLOCK(fdp);
3654 kinfo_to_okinfo(kif, okif);
3655 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3656 FILEDESC_SLOCK(fdp);
3657 if (error)
3658 break;
3659 }
3660 FILEDESC_SUNLOCK(fdp);
3661 fddrop(fdp);
3662 free(kif, M_TEMP);
3663 free(okif, M_TEMP);
3664 return (0);
3665 }
3666
3667 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3668 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3669 "Process ofiledesc entries");
3670 #endif /* COMPAT_FREEBSD7 */
3671
3672 int
3673 vntype_to_kinfo(int vtype)
3674 {
3675 struct {
3676 int vtype;
3677 int kf_vtype;
3678 } vtypes_table[] = {
3679 { VBAD, KF_VTYPE_VBAD },
3680 { VBLK, KF_VTYPE_VBLK },
3681 { VCHR, KF_VTYPE_VCHR },
3682 { VDIR, KF_VTYPE_VDIR },
3683 { VFIFO, KF_VTYPE_VFIFO },
3684 { VLNK, KF_VTYPE_VLNK },
3685 { VNON, KF_VTYPE_VNON },
3686 { VREG, KF_VTYPE_VREG },
3687 { VSOCK, KF_VTYPE_VSOCK }
3688 };
3689 unsigned int i;
3690
3691 /*
3692 * Perform vtype translation.
3693 */
3694 for (i = 0; i < nitems(vtypes_table); i++)
3695 if (vtypes_table[i].vtype == vtype)
3696 return (vtypes_table[i].kf_vtype);
3697
3698 return (KF_VTYPE_UNKNOWN);
3699 }
3700
3701 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3702 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3703 "Process filedesc entries");
3704
3705 /*
3706 * Store a process current working directory information to sbuf.
3707 *
3708 * Takes a locked proc as argument, and returns with the proc unlocked.
3709 */
3710 int
3711 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3712 {
3713 struct filedesc *fdp;
3714 struct export_fd_buf *efbuf;
3715 int error;
3716
3717 PROC_LOCK_ASSERT(p, MA_OWNED);
3718
3719 fdp = fdhold(p);
3720 PROC_UNLOCK(p);
3721 if (fdp == NULL)
3722 return (EINVAL);
3723
3724 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3725 efbuf->fdp = fdp;
3726 efbuf->sb = sb;
3727 efbuf->remainder = maxlen;
3728
3729 FILEDESC_SLOCK(fdp);
3730 if (fdp->fd_cdir == NULL)
3731 error = EINVAL;
3732 else {
3733 vrefact(fdp->fd_cdir);
3734 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3735 FREAD, efbuf);
3736 }
3737 FILEDESC_SUNLOCK(fdp);
3738 fddrop(fdp);
3739 free(efbuf, M_TEMP);
3740 return (error);
3741 }
3742
3743 /*
3744 * Get per-process current working directory.
3745 */
3746 static int
3747 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3748 {
3749 struct sbuf sb;
3750 struct proc *p;
3751 ssize_t maxlen;
3752 int error, error2, *name;
3753
3754 name = (int *)arg1;
3755
3756 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3757 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3758 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3759 if (error != 0) {
3760 sbuf_delete(&sb);
3761 return (error);
3762 }
3763 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3764 error = kern_proc_cwd_out(p, &sb, maxlen);
3765 error2 = sbuf_finish(&sb);
3766 sbuf_delete(&sb);
3767 return (error != 0 ? error : error2);
3768 }
3769
3770 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3771 sysctl_kern_proc_cwd, "Process current working directory");
3772
3773 #ifdef DDB
3774 /*
3775 * For the purposes of debugging, generate a human-readable string for the
3776 * file type.
3777 */
3778 static const char *
3779 file_type_to_name(short type)
3780 {
3781
3782 switch (type) {
3783 case 0:
3784 return ("zero");
3785 case DTYPE_VNODE:
3786 return ("vnod");
3787 case DTYPE_SOCKET:
3788 return ("sock");
3789 case DTYPE_PIPE:
3790 return ("pipe");
3791 case DTYPE_FIFO:
3792 return ("fifo");
3793 case DTYPE_KQUEUE:
3794 return ("kque");
3795 case DTYPE_CRYPTO:
3796 return ("crpt");
3797 case DTYPE_MQUEUE:
3798 return ("mque");
3799 case DTYPE_SHM:
3800 return ("shm");
3801 case DTYPE_SEM:
3802 return ("ksem");
3803 default:
3804 return ("unkn");
3805 }
3806 }
3807
3808 /*
3809 * For the purposes of debugging, identify a process (if any, perhaps one of
3810 * many) that references the passed file in its file descriptor array. Return
3811 * NULL if none.
3812 */
3813 static struct proc *
3814 file_to_first_proc(struct file *fp)
3815 {
3816 struct filedesc *fdp;
3817 struct proc *p;
3818 int n;
3819
3820 FOREACH_PROC_IN_SYSTEM(p) {
3821 if (p->p_state == PRS_NEW)
3822 continue;
3823 fdp = p->p_fd;
3824 if (fdp == NULL)
3825 continue;
3826 for (n = 0; n <= fdp->fd_lastfile; n++) {
3827 if (fp == fdp->fd_ofiles[n].fde_file)
3828 return (p);
3829 }
3830 }
3831 return (NULL);
3832 }
3833
3834 static void
3835 db_print_file(struct file *fp, int header)
3836 {
3837 struct proc *p;
3838
3839 if (header)
3840 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3841 "File", "Type", "Data", "Flag", "GCFl", "Count",
3842 "MCount", "Vnode", "FPID", "FCmd");
3843 p = file_to_first_proc(fp);
3844 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3845 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3846 0, fp->f_count, 0, fp->f_vnode,
3847 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3848 }
3849
3850 DB_SHOW_COMMAND(file, db_show_file)
3851 {
3852 struct file *fp;
3853
3854 if (!have_addr) {
3855 db_printf("usage: show file <addr>\n");
3856 return;
3857 }
3858 fp = (struct file *)addr;
3859 db_print_file(fp, 1);
3860 }
3861
3862 DB_SHOW_COMMAND(files, db_show_files)
3863 {
3864 struct filedesc *fdp;
3865 struct file *fp;
3866 struct proc *p;
3867 int header;
3868 int n;
3869
3870 header = 1;
3871 FOREACH_PROC_IN_SYSTEM(p) {
3872 if (p->p_state == PRS_NEW)
3873 continue;
3874 if ((fdp = p->p_fd) == NULL)
3875 continue;
3876 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3877 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3878 continue;
3879 db_print_file(fp, header);
3880 header = 0;
3881 }
3882 }
3883 }
3884 #endif
3885
3886 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3887 &maxfilesperproc, 0, "Maximum files allowed open per process");
3888
3889 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3890 &maxfiles, 0, "Maximum number of files");
3891
3892 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3893 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3894
3895 /* ARGSUSED*/
3896 static void
3897 filelistinit(void *dummy)
3898 {
3899
3900 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3901 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3902 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3903 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3904 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3905 }
3906 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3907
3908 /*-------------------------------------------------------------------*/
3909
3910 static int
3911 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3912 int flags, struct thread *td)
3913 {
3914
3915 return (EBADF);
3916 }
3917
3918 static int
3919 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3920 struct thread *td)
3921 {
3922
3923 return (EINVAL);
3924 }
3925
3926 static int
3927 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3928 struct thread *td)
3929 {
3930
3931 return (EBADF);
3932 }
3933
3934 static int
3935 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3936 struct thread *td)
3937 {
3938
3939 return (0);
3940 }
3941
3942 static int
3943 badfo_kqfilter(struct file *fp, struct knote *kn)
3944 {
3945
3946 return (EBADF);
3947 }
3948
3949 static int
3950 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3951 struct thread *td)
3952 {
3953
3954 return (EBADF);
3955 }
3956
3957 static int
3958 badfo_close(struct file *fp, struct thread *td)
3959 {
3960
3961 return (0);
3962 }
3963
3964 static int
3965 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3966 struct thread *td)
3967 {
3968
3969 return (EBADF);
3970 }
3971
3972 static int
3973 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3974 struct thread *td)
3975 {
3976
3977 return (EBADF);
3978 }
3979
3980 static int
3981 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3982 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3983 struct thread *td)
3984 {
3985
3986 return (EBADF);
3987 }
3988
3989 static int
3990 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
3991 {
3992
3993 return (0);
3994 }
3995
3996 struct fileops badfileops = {
3997 .fo_read = badfo_readwrite,
3998 .fo_write = badfo_readwrite,
3999 .fo_truncate = badfo_truncate,
4000 .fo_ioctl = badfo_ioctl,
4001 .fo_poll = badfo_poll,
4002 .fo_kqfilter = badfo_kqfilter,
4003 .fo_stat = badfo_stat,
4004 .fo_close = badfo_close,
4005 .fo_chmod = badfo_chmod,
4006 .fo_chown = badfo_chown,
4007 .fo_sendfile = badfo_sendfile,
4008 .fo_fill_kinfo = badfo_fill_kinfo,
4009 };
4010
4011 int
4012 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4013 int flags, struct thread *td)
4014 {
4015
4016 return (EOPNOTSUPP);
4017 }
4018
4019 int
4020 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4021 struct thread *td)
4022 {
4023
4024 return (EINVAL);
4025 }
4026
4027 int
4028 invfo_ioctl(struct file *fp, u_long com, void *data,
4029 struct ucred *active_cred, struct thread *td)
4030 {
4031
4032 return (ENOTTY);
4033 }
4034
4035 int
4036 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4037 struct thread *td)
4038 {
4039
4040 return (poll_no_poll(events));
4041 }
4042
4043 int
4044 invfo_kqfilter(struct file *fp, struct knote *kn)
4045 {
4046
4047 return (EINVAL);
4048 }
4049
4050 int
4051 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4052 struct thread *td)
4053 {
4054
4055 return (EINVAL);
4056 }
4057
4058 int
4059 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4060 struct thread *td)
4061 {
4062
4063 return (EINVAL);
4064 }
4065
4066 int
4067 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4068 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4069 struct thread *td)
4070 {
4071
4072 return (EINVAL);
4073 }
4074
4075 /*-------------------------------------------------------------------*/
4076
4077 /*
4078 * File Descriptor pseudo-device driver (/dev/fd/).
4079 *
4080 * Opening minor device N dup()s the file (if any) connected to file
4081 * descriptor N belonging to the calling process. Note that this driver
4082 * consists of only the ``open()'' routine, because all subsequent
4083 * references to this file will be direct to the other driver.
4084 *
4085 * XXX: we could give this one a cloning event handler if necessary.
4086 */
4087
4088 /* ARGSUSED */
4089 static int
4090 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4091 {
4092
4093 /*
4094 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4095 * the file descriptor being sought for duplication. The error
4096 * return ensures that the vnode for this device will be released
4097 * by vn_open. Open will detect this special error and take the
4098 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4099 * will simply report the error.
4100 */
4101 td->td_dupfd = dev2unit(dev);
4102 return (ENODEV);
4103 }
4104
4105 static struct cdevsw fildesc_cdevsw = {
4106 .d_version = D_VERSION,
4107 .d_open = fdopen,
4108 .d_name = "FD",
4109 };
4110
4111 static void
4112 fildesc_drvinit(void *unused)
4113 {
4114 struct cdev *dev;
4115
4116 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4117 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4118 make_dev_alias(dev, "stdin");
4119 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4120 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4121 make_dev_alias(dev, "stdout");
4122 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4123 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4124 make_dev_alias(dev, "stderr");
4125 }
4126
4127 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
Cache object: 1839009dadaf88e94eccb2c4af38b28c
|