1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_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
fd_first_free(struct filedesc * fdp,int low,int size)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
fd_last_used(struct filedesc * fdp,int size)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
fdisused(struct filedesc * fdp,int fd)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
fdused_init(struct filedesc * fdp,int fd)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
fdused(struct filedesc * fdp,int fd)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
fdunused(struct filedesc * fdp,int fd)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
fdefree_last(struct filedescent * fde)288 fdefree_last(struct filedescent *fde)
289 {
290
291 filecaps_free(&fde->fde_caps);
292 }
293
294 static inline void
fdfree(struct filedesc * fdp,int fd)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
pwd_ensure_dirs(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
sys_getdtablesize(struct thread * td,struct getdtablesize_args * uap)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
sys_dup2(struct thread * td,struct dup2_args * uap)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
sys_dup(struct thread * td,struct dup_args * uap)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
sys_fcntl(struct thread * td,struct fcntl_args * uap)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
kern_fcntl_freebsd(struct thread * td,int fd,int cmd,long arg)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
kern_fcntl(struct thread * td,int fd,int cmd,intptr_t arg)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 if (vp->v_type != VREG) {
769 fdrop(fp, td);
770 error = ENOTTY;
771 break;
772 }
773
774 /*
775 * Exclusive lock synchronizes against f_seqcount reads and
776 * writes in sequential_heuristic().
777 */
778 error = vn_lock(vp, LK_EXCLUSIVE);
779 if (error != 0) {
780 fdrop(fp, td);
781 break;
782 }
783 if (arg >= 0) {
784 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
785 fp->f_seqcount = (arg + bsize - 1) / bsize;
786 atomic_set_int(&fp->f_flag, FRDAHEAD);
787 } else {
788 atomic_clear_int(&fp->f_flag, FRDAHEAD);
789 }
790 VOP_UNLOCK(vp, 0);
791 fdrop(fp, td);
792 break;
793
794 default:
795 error = EINVAL;
796 break;
797 }
798 return (error);
799 }
800
801 static int
getmaxfd(struct thread * td)802 getmaxfd(struct thread *td)
803 {
804
805 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
806 }
807
808 /*
809 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
810 */
811 int
kern_dup(struct thread * td,u_int mode,int flags,int old,int new)812 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
813 {
814 struct filedesc *fdp;
815 struct filedescent *oldfde, *newfde;
816 struct proc *p;
817 struct file *delfp;
818 int error, maxfd;
819
820 p = td->td_proc;
821 fdp = p->p_fd;
822
823 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
824 MPASS(mode < FDDUP_LASTMODE);
825
826 AUDIT_ARG_FD(old);
827 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
828
829 /*
830 * Verify we have a valid descriptor to dup from and possibly to
831 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
832 * return EINVAL when the new descriptor is out of bounds.
833 */
834 if (old < 0)
835 return (EBADF);
836 if (new < 0)
837 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
838 maxfd = getmaxfd(td);
839 if (new >= maxfd)
840 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
841
842 error = EBADF;
843 FILEDESC_XLOCK(fdp);
844 if (fget_locked(fdp, old) == NULL)
845 goto unlock;
846 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
847 td->td_retval[0] = new;
848 if (flags & FDDUP_FLAG_CLOEXEC)
849 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
850 error = 0;
851 goto unlock;
852 }
853
854 /*
855 * If the caller specified a file descriptor, make sure the file
856 * table is large enough to hold it, and grab it. Otherwise, just
857 * allocate a new descriptor the usual way.
858 */
859 switch (mode) {
860 case FDDUP_NORMAL:
861 case FDDUP_FCNTL:
862 if ((error = fdalloc(td, new, &new)) != 0)
863 goto unlock;
864 break;
865 case FDDUP_MUSTREPLACE:
866 /* Target file descriptor must exist. */
867 if (fget_locked(fdp, new) == NULL)
868 goto unlock;
869 break;
870 case FDDUP_FIXED:
871 if (new >= fdp->fd_nfiles) {
872 /*
873 * The resource limits are here instead of e.g.
874 * fdalloc(), because the file descriptor table may be
875 * shared between processes, so we can't really use
876 * racct_add()/racct_sub(). Instead of counting the
877 * number of actually allocated descriptors, just put
878 * the limit on the size of the file descriptor table.
879 */
880 #ifdef RACCT
881 if (racct_enable) {
882 PROC_LOCK(p);
883 error = racct_set(p, RACCT_NOFILE, new + 1);
884 PROC_UNLOCK(p);
885 if (error != 0) {
886 error = EMFILE;
887 goto unlock;
888 }
889 }
890 #endif
891 fdgrowtable_exp(fdp, new + 1);
892 }
893 if (!fdisused(fdp, new))
894 fdused(fdp, new);
895 break;
896 default:
897 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
898 }
899
900 KASSERT(old != new, ("new fd is same as old"));
901
902 oldfde = &fdp->fd_ofiles[old];
903 fhold(oldfde->fde_file);
904 newfde = &fdp->fd_ofiles[new];
905 delfp = newfde->fde_file;
906
907 /*
908 * Duplicate the source descriptor.
909 */
910 #ifdef CAPABILITIES
911 seq_write_begin(&newfde->fde_seq);
912 #endif
913 filecaps_free(&newfde->fde_caps);
914 memcpy(newfde, oldfde, fde_change_size);
915 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
916 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
917 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
918 else
919 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
920 #ifdef CAPABILITIES
921 seq_write_end(&newfde->fde_seq);
922 #endif
923 td->td_retval[0] = new;
924
925 error = 0;
926
927 if (delfp != NULL) {
928 (void) closefp(fdp, new, delfp, td, 1);
929 FILEDESC_UNLOCK_ASSERT(fdp);
930 } else {
931 unlock:
932 FILEDESC_XUNLOCK(fdp);
933 }
934
935 return (error);
936 }
937
938 /*
939 * If sigio is on the list associated with a process or process group,
940 * disable signalling from the device, remove sigio from the list and
941 * free sigio.
942 */
943 void
funsetown(struct sigio ** sigiop)944 funsetown(struct sigio **sigiop)
945 {
946 struct sigio *sigio;
947
948 if (*sigiop == NULL)
949 return;
950 SIGIO_LOCK();
951 sigio = *sigiop;
952 if (sigio == NULL) {
953 SIGIO_UNLOCK();
954 return;
955 }
956 *(sigio->sio_myref) = NULL;
957 if ((sigio)->sio_pgid < 0) {
958 struct pgrp *pg = (sigio)->sio_pgrp;
959 PGRP_LOCK(pg);
960 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
961 sigio, sio_pgsigio);
962 PGRP_UNLOCK(pg);
963 } else {
964 struct proc *p = (sigio)->sio_proc;
965 PROC_LOCK(p);
966 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
967 sigio, sio_pgsigio);
968 PROC_UNLOCK(p);
969 }
970 SIGIO_UNLOCK();
971 crfree(sigio->sio_ucred);
972 free(sigio, M_SIGIO);
973 }
974
975 /*
976 * Free a list of sigio structures.
977 * We only need to lock the SIGIO_LOCK because we have made ourselves
978 * inaccessible to callers of fsetown and therefore do not need to lock
979 * the proc or pgrp struct for the list manipulation.
980 */
981 void
funsetownlst(struct sigiolst * sigiolst)982 funsetownlst(struct sigiolst *sigiolst)
983 {
984 struct proc *p;
985 struct pgrp *pg;
986 struct sigio *sigio;
987
988 sigio = SLIST_FIRST(sigiolst);
989 if (sigio == NULL)
990 return;
991 p = NULL;
992 pg = NULL;
993
994 /*
995 * Every entry of the list should belong
996 * to a single proc or pgrp.
997 */
998 if (sigio->sio_pgid < 0) {
999 pg = sigio->sio_pgrp;
1000 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1001 } else /* if (sigio->sio_pgid > 0) */ {
1002 p = sigio->sio_proc;
1003 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1004 }
1005
1006 SIGIO_LOCK();
1007 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1008 *(sigio->sio_myref) = NULL;
1009 if (pg != NULL) {
1010 KASSERT(sigio->sio_pgid < 0,
1011 ("Proc sigio in pgrp sigio list"));
1012 KASSERT(sigio->sio_pgrp == pg,
1013 ("Bogus pgrp in sigio list"));
1014 PGRP_LOCK(pg);
1015 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1016 sio_pgsigio);
1017 PGRP_UNLOCK(pg);
1018 } else /* if (p != NULL) */ {
1019 KASSERT(sigio->sio_pgid > 0,
1020 ("Pgrp sigio in proc sigio list"));
1021 KASSERT(sigio->sio_proc == p,
1022 ("Bogus proc in sigio list"));
1023 PROC_LOCK(p);
1024 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1025 sio_pgsigio);
1026 PROC_UNLOCK(p);
1027 }
1028 SIGIO_UNLOCK();
1029 crfree(sigio->sio_ucred);
1030 free(sigio, M_SIGIO);
1031 SIGIO_LOCK();
1032 }
1033 SIGIO_UNLOCK();
1034 }
1035
1036 /*
1037 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1038 *
1039 * After permission checking, add a sigio structure to the sigio list for
1040 * the process or process group.
1041 */
1042 int
fsetown(pid_t pgid,struct sigio ** sigiop)1043 fsetown(pid_t pgid, struct sigio **sigiop)
1044 {
1045 struct proc *proc;
1046 struct pgrp *pgrp;
1047 struct sigio *sigio;
1048 int ret;
1049
1050 if (pgid == 0) {
1051 funsetown(sigiop);
1052 return (0);
1053 }
1054
1055 ret = 0;
1056
1057 /* Allocate and fill in the new sigio out of locks. */
1058 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1059 sigio->sio_pgid = pgid;
1060 sigio->sio_ucred = crhold(curthread->td_ucred);
1061 sigio->sio_myref = sigiop;
1062
1063 sx_slock(&proctree_lock);
1064 if (pgid > 0) {
1065 proc = pfind(pgid);
1066 if (proc == NULL) {
1067 ret = ESRCH;
1068 goto fail;
1069 }
1070
1071 /*
1072 * Policy - Don't allow a process to FSETOWN a process
1073 * in another session.
1074 *
1075 * Remove this test to allow maximum flexibility or
1076 * restrict FSETOWN to the current process or process
1077 * group for maximum safety.
1078 */
1079 PROC_UNLOCK(proc);
1080 if (proc->p_session != curthread->td_proc->p_session) {
1081 ret = EPERM;
1082 goto fail;
1083 }
1084
1085 pgrp = NULL;
1086 } else /* if (pgid < 0) */ {
1087 pgrp = pgfind(-pgid);
1088 if (pgrp == NULL) {
1089 ret = ESRCH;
1090 goto fail;
1091 }
1092 PGRP_UNLOCK(pgrp);
1093
1094 /*
1095 * Policy - Don't allow a process to FSETOWN a process
1096 * in another session.
1097 *
1098 * Remove this test to allow maximum flexibility or
1099 * restrict FSETOWN to the current process or process
1100 * group for maximum safety.
1101 */
1102 if (pgrp->pg_session != curthread->td_proc->p_session) {
1103 ret = EPERM;
1104 goto fail;
1105 }
1106
1107 proc = NULL;
1108 }
1109 funsetown(sigiop);
1110 if (pgid > 0) {
1111 PROC_LOCK(proc);
1112 /*
1113 * Since funsetownlst() is called without the proctree
1114 * locked, we need to check for P_WEXIT.
1115 * XXX: is ESRCH correct?
1116 */
1117 if ((proc->p_flag & P_WEXIT) != 0) {
1118 PROC_UNLOCK(proc);
1119 ret = ESRCH;
1120 goto fail;
1121 }
1122 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1123 sigio->sio_proc = proc;
1124 PROC_UNLOCK(proc);
1125 } else {
1126 PGRP_LOCK(pgrp);
1127 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1128 sigio->sio_pgrp = pgrp;
1129 PGRP_UNLOCK(pgrp);
1130 }
1131 sx_sunlock(&proctree_lock);
1132 SIGIO_LOCK();
1133 *sigiop = sigio;
1134 SIGIO_UNLOCK();
1135 return (0);
1136
1137 fail:
1138 sx_sunlock(&proctree_lock);
1139 crfree(sigio->sio_ucred);
1140 free(sigio, M_SIGIO);
1141 return (ret);
1142 }
1143
1144 /*
1145 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1146 */
1147 pid_t
fgetown(sigiop)1148 fgetown(sigiop)
1149 struct sigio **sigiop;
1150 {
1151 pid_t pgid;
1152
1153 SIGIO_LOCK();
1154 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1155 SIGIO_UNLOCK();
1156 return (pgid);
1157 }
1158
1159 /*
1160 * Function drops the filedesc lock on return.
1161 */
1162 static int
closefp(struct filedesc * fdp,int fd,struct file * fp,struct thread * td,int holdleaders)1163 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1164 int holdleaders)
1165 {
1166 int error;
1167
1168 FILEDESC_XLOCK_ASSERT(fdp);
1169
1170 if (holdleaders) {
1171 if (td->td_proc->p_fdtol != NULL) {
1172 /*
1173 * Ask fdfree() to sleep to ensure that all relevant
1174 * process leaders can be traversed in closef().
1175 */
1176 fdp->fd_holdleaderscount++;
1177 } else {
1178 holdleaders = 0;
1179 }
1180 }
1181
1182 /*
1183 * We now hold the fp reference that used to be owned by the
1184 * descriptor array. We have to unlock the FILEDESC *AFTER*
1185 * knote_fdclose to prevent a race of the fd getting opened, a knote
1186 * added, and deleteing a knote for the new fd.
1187 */
1188 knote_fdclose(td, fd);
1189
1190 /*
1191 * We need to notify mqueue if the object is of type mqueue.
1192 */
1193 if (fp->f_type == DTYPE_MQUEUE)
1194 mq_fdclose(td, fd, fp);
1195 FILEDESC_XUNLOCK(fdp);
1196
1197 error = closef(fp, td);
1198 if (holdleaders) {
1199 FILEDESC_XLOCK(fdp);
1200 fdp->fd_holdleaderscount--;
1201 if (fdp->fd_holdleaderscount == 0 &&
1202 fdp->fd_holdleaderswakeup != 0) {
1203 fdp->fd_holdleaderswakeup = 0;
1204 wakeup(&fdp->fd_holdleaderscount);
1205 }
1206 FILEDESC_XUNLOCK(fdp);
1207 }
1208 return (error);
1209 }
1210
1211 /*
1212 * Close a file descriptor.
1213 */
1214 #ifndef _SYS_SYSPROTO_H_
1215 struct close_args {
1216 int fd;
1217 };
1218 #endif
1219 /* ARGSUSED */
1220 int
sys_close(struct thread * td,struct close_args * uap)1221 sys_close(struct thread *td, struct close_args *uap)
1222 {
1223
1224 return (kern_close(td, uap->fd));
1225 }
1226
1227 int
kern_close(struct thread * td,int fd)1228 kern_close(struct thread *td, int fd)
1229 {
1230 struct filedesc *fdp;
1231 struct file *fp;
1232
1233 fdp = td->td_proc->p_fd;
1234
1235 AUDIT_SYSCLOSE(td, fd);
1236
1237 FILEDESC_XLOCK(fdp);
1238 if ((fp = fget_locked(fdp, fd)) == NULL) {
1239 FILEDESC_XUNLOCK(fdp);
1240 return (EBADF);
1241 }
1242 fdfree(fdp, fd);
1243
1244 /* closefp() drops the FILEDESC lock for us. */
1245 return (closefp(fdp, fd, fp, td, 1));
1246 }
1247
1248 /*
1249 * Close open file descriptors.
1250 */
1251 #ifndef _SYS_SYSPROTO_H_
1252 struct closefrom_args {
1253 int lowfd;
1254 };
1255 #endif
1256 /* ARGSUSED */
1257 int
sys_closefrom(struct thread * td,struct closefrom_args * uap)1258 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1259 {
1260 struct filedesc *fdp;
1261 int fd;
1262
1263 fdp = td->td_proc->p_fd;
1264 AUDIT_ARG_FD(uap->lowfd);
1265
1266 /*
1267 * Treat negative starting file descriptor values identical to
1268 * closefrom(0) which closes all files.
1269 */
1270 if (uap->lowfd < 0)
1271 uap->lowfd = 0;
1272 FILEDESC_SLOCK(fdp);
1273 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1274 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1275 FILEDESC_SUNLOCK(fdp);
1276 (void)kern_close(td, fd);
1277 FILEDESC_SLOCK(fdp);
1278 }
1279 }
1280 FILEDESC_SUNLOCK(fdp);
1281 return (0);
1282 }
1283
1284 #if defined(COMPAT_43)
1285 /*
1286 * Return status information about a file descriptor.
1287 */
1288 #ifndef _SYS_SYSPROTO_H_
1289 struct ofstat_args {
1290 int fd;
1291 struct ostat *sb;
1292 };
1293 #endif
1294 /* ARGSUSED */
1295 int
ofstat(struct thread * td,struct ofstat_args * uap)1296 ofstat(struct thread *td, struct ofstat_args *uap)
1297 {
1298 struct ostat oub;
1299 struct stat ub;
1300 int error;
1301
1302 error = kern_fstat(td, uap->fd, &ub);
1303 if (error == 0) {
1304 cvtstat(&ub, &oub);
1305 error = copyout(&oub, uap->sb, sizeof(oub));
1306 }
1307 return (error);
1308 }
1309 #endif /* COMPAT_43 */
1310
1311 /*
1312 * Return status information about a file descriptor.
1313 */
1314 #ifndef _SYS_SYSPROTO_H_
1315 struct fstat_args {
1316 int fd;
1317 struct stat *sb;
1318 };
1319 #endif
1320 /* ARGSUSED */
1321 int
sys_fstat(struct thread * td,struct fstat_args * uap)1322 sys_fstat(struct thread *td, struct fstat_args *uap)
1323 {
1324 struct stat ub;
1325 int error;
1326
1327 error = kern_fstat(td, uap->fd, &ub);
1328 if (error == 0)
1329 error = copyout(&ub, uap->sb, sizeof(ub));
1330 return (error);
1331 }
1332
1333 int
kern_fstat(struct thread * td,int fd,struct stat * sbp)1334 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1335 {
1336 struct file *fp;
1337 cap_rights_t rights;
1338 int error;
1339
1340 AUDIT_ARG_FD(fd);
1341
1342 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1343 if (error != 0)
1344 return (error);
1345
1346 AUDIT_ARG_FILE(td->td_proc, fp);
1347
1348 error = fo_stat(fp, sbp, td->td_ucred, td);
1349 fdrop(fp, td);
1350 #ifdef KTRACE
1351 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1352 ktrstat(sbp);
1353 #endif
1354 return (error);
1355 }
1356
1357 /*
1358 * Return status information about a file descriptor.
1359 */
1360 #ifndef _SYS_SYSPROTO_H_
1361 struct nfstat_args {
1362 int fd;
1363 struct nstat *sb;
1364 };
1365 #endif
1366 /* ARGSUSED */
1367 int
sys_nfstat(struct thread * td,struct nfstat_args * uap)1368 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1369 {
1370 struct nstat nub;
1371 struct stat ub;
1372 int error;
1373
1374 error = kern_fstat(td, uap->fd, &ub);
1375 if (error == 0) {
1376 cvtnstat(&ub, &nub);
1377 error = copyout(&nub, uap->sb, sizeof(nub));
1378 }
1379 return (error);
1380 }
1381
1382 /*
1383 * Return pathconf information about a file descriptor.
1384 */
1385 #ifndef _SYS_SYSPROTO_H_
1386 struct fpathconf_args {
1387 int fd;
1388 int name;
1389 };
1390 #endif
1391 /* ARGSUSED */
1392 int
sys_fpathconf(struct thread * td,struct fpathconf_args * uap)1393 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1394 {
1395
1396 return (kern_fpathconf(td, uap->fd, uap->name));
1397 }
1398
1399 int
kern_fpathconf(struct thread * td,int fd,int name)1400 kern_fpathconf(struct thread *td, int fd, int name)
1401 {
1402 struct file *fp;
1403 struct vnode *vp;
1404 cap_rights_t rights;
1405 int error;
1406
1407 error = fget(td, fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1408 if (error != 0)
1409 return (error);
1410
1411 if (name == _PC_ASYNC_IO) {
1412 td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1413 goto out;
1414 }
1415 vp = fp->f_vnode;
1416 if (vp != NULL) {
1417 vn_lock(vp, LK_SHARED | LK_RETRY);
1418 error = VOP_PATHCONF(vp, name, td->td_retval);
1419 VOP_UNLOCK(vp, 0);
1420 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1421 if (name != _PC_PIPE_BUF) {
1422 error = EINVAL;
1423 } else {
1424 td->td_retval[0] = PIPE_BUF;
1425 error = 0;
1426 }
1427 } else {
1428 error = EOPNOTSUPP;
1429 }
1430 out:
1431 fdrop(fp, td);
1432 return (error);
1433 }
1434
1435 /*
1436 * Initialize filecaps structure.
1437 */
1438 void
filecaps_init(struct filecaps * fcaps)1439 filecaps_init(struct filecaps *fcaps)
1440 {
1441
1442 bzero(fcaps, sizeof(*fcaps));
1443 fcaps->fc_nioctls = -1;
1444 }
1445
1446 /*
1447 * Copy filecaps structure allocating memory for ioctls array if needed.
1448 *
1449 * The last parameter indicates whether the fdtable is locked. If it is not and
1450 * ioctls are encountered, copying fails and the caller must lock the table.
1451 *
1452 * Note that if the table was not locked, the caller has to check the relevant
1453 * sequence counter to determine whether the operation was successful.
1454 */
1455 bool
filecaps_copy(const struct filecaps * src,struct filecaps * dst,bool locked)1456 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1457 {
1458 size_t size;
1459
1460 if (src->fc_ioctls != NULL && !locked)
1461 return (false);
1462 *dst = *src;
1463 if (src->fc_ioctls == NULL)
1464 return (true);
1465
1466 KASSERT(src->fc_nioctls > 0,
1467 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1468
1469 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1470 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1471 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1472 return (true);
1473 }
1474
1475 /*
1476 * Move filecaps structure to the new place and clear the old place.
1477 */
1478 void
filecaps_move(struct filecaps * src,struct filecaps * dst)1479 filecaps_move(struct filecaps *src, struct filecaps *dst)
1480 {
1481
1482 *dst = *src;
1483 bzero(src, sizeof(*src));
1484 }
1485
1486 /*
1487 * Fill the given filecaps structure with full rights.
1488 */
1489 static void
filecaps_fill(struct filecaps * fcaps)1490 filecaps_fill(struct filecaps *fcaps)
1491 {
1492
1493 CAP_ALL(&fcaps->fc_rights);
1494 fcaps->fc_ioctls = NULL;
1495 fcaps->fc_nioctls = -1;
1496 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1497 }
1498
1499 /*
1500 * Free memory allocated within filecaps structure.
1501 */
1502 void
filecaps_free(struct filecaps * fcaps)1503 filecaps_free(struct filecaps *fcaps)
1504 {
1505
1506 free(fcaps->fc_ioctls, M_FILECAPS);
1507 bzero(fcaps, sizeof(*fcaps));
1508 }
1509
1510 /*
1511 * Validate the given filecaps structure.
1512 */
1513 static void
filecaps_validate(const struct filecaps * fcaps,const char * func)1514 filecaps_validate(const struct filecaps *fcaps, const char *func)
1515 {
1516
1517 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1518 ("%s: invalid rights", func));
1519 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1520 ("%s: invalid fcntls", func));
1521 KASSERT(fcaps->fc_fcntls == 0 ||
1522 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1523 ("%s: fcntls without CAP_FCNTL", func));
1524 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1525 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1526 ("%s: invalid ioctls", func));
1527 KASSERT(fcaps->fc_nioctls == 0 ||
1528 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1529 ("%s: ioctls without CAP_IOCTL", func));
1530 }
1531
1532 static void
fdgrowtable_exp(struct filedesc * fdp,int nfd)1533 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1534 {
1535 int nfd1;
1536
1537 FILEDESC_XLOCK_ASSERT(fdp);
1538
1539 nfd1 = fdp->fd_nfiles * 2;
1540 if (nfd1 < nfd)
1541 nfd1 = nfd;
1542 fdgrowtable(fdp, nfd1);
1543 }
1544
1545 /*
1546 * Grow the file table to accommodate (at least) nfd descriptors.
1547 */
1548 static void
fdgrowtable(struct filedesc * fdp,int nfd)1549 fdgrowtable(struct filedesc *fdp, int nfd)
1550 {
1551 struct filedesc0 *fdp0;
1552 struct freetable *ft;
1553 struct fdescenttbl *ntable;
1554 struct fdescenttbl *otable;
1555 int nnfiles, onfiles;
1556 NDSLOTTYPE *nmap, *omap;
1557
1558 /*
1559 * If lastfile is -1 this struct filedesc was just allocated and we are
1560 * growing it to accommodate for the one we are going to copy from. There
1561 * is no need to have a lock on this one as it's not visible to anyone.
1562 */
1563 if (fdp->fd_lastfile != -1)
1564 FILEDESC_XLOCK_ASSERT(fdp);
1565
1566 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1567
1568 /* save old values */
1569 onfiles = fdp->fd_nfiles;
1570 otable = fdp->fd_files;
1571 omap = fdp->fd_map;
1572
1573 /* compute the size of the new table */
1574 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1575 if (nnfiles <= onfiles)
1576 /* the table is already large enough */
1577 return;
1578
1579 /*
1580 * Allocate a new table. We need enough space for the number of
1581 * entries, file entries themselves and the struct freetable we will use
1582 * when we decommission the table and place it on the freelist.
1583 * We place the struct freetable in the middle so we don't have
1584 * to worry about padding.
1585 */
1586 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1587 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1588 sizeof(struct freetable),
1589 M_FILEDESC, M_ZERO | M_WAITOK);
1590 /* copy the old data */
1591 ntable->fdt_nfiles = nnfiles;
1592 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1593 onfiles * sizeof(ntable->fdt_ofiles[0]));
1594
1595 /*
1596 * Allocate a new map only if the old is not large enough. It will
1597 * grow at a slower rate than the table as it can map more
1598 * entries than the table can hold.
1599 */
1600 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1601 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1602 M_ZERO | M_WAITOK);
1603 /* copy over the old data and update the pointer */
1604 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1605 fdp->fd_map = nmap;
1606 }
1607
1608 /*
1609 * Make sure that ntable is correctly initialized before we replace
1610 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1611 * data.
1612 */
1613 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1614
1615 /*
1616 * Do not free the old file table, as some threads may still
1617 * reference entries within it. Instead, place it on a freelist
1618 * which will be processed when the struct filedesc is released.
1619 *
1620 * Note that if onfiles == NDFILE, we're dealing with the original
1621 * static allocation contained within (struct filedesc0 *)fdp,
1622 * which must not be freed.
1623 */
1624 if (onfiles > NDFILE) {
1625 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1626 fdp0 = (struct filedesc0 *)fdp;
1627 ft->ft_table = otable;
1628 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1629 }
1630 /*
1631 * The map does not have the same possibility of threads still
1632 * holding references to it. So always free it as long as it
1633 * does not reference the original static allocation.
1634 */
1635 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1636 free(omap, M_FILEDESC);
1637 }
1638
1639 /*
1640 * Allocate a file descriptor for the process.
1641 */
1642 int
fdalloc(struct thread * td,int minfd,int * result)1643 fdalloc(struct thread *td, int minfd, int *result)
1644 {
1645 struct proc *p = td->td_proc;
1646 struct filedesc *fdp = p->p_fd;
1647 int fd, maxfd, allocfd;
1648 #ifdef RACCT
1649 int error;
1650 #endif
1651
1652 FILEDESC_XLOCK_ASSERT(fdp);
1653
1654 if (fdp->fd_freefile > minfd)
1655 minfd = fdp->fd_freefile;
1656
1657 maxfd = getmaxfd(td);
1658
1659 /*
1660 * Search the bitmap for a free descriptor starting at minfd.
1661 * If none is found, grow the file table.
1662 */
1663 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1664 if (fd >= maxfd)
1665 return (EMFILE);
1666 if (fd >= fdp->fd_nfiles) {
1667 allocfd = min(fd * 2, maxfd);
1668 #ifdef RACCT
1669 if (racct_enable) {
1670 PROC_LOCK(p);
1671 error = racct_set(p, RACCT_NOFILE, allocfd);
1672 PROC_UNLOCK(p);
1673 if (error != 0)
1674 return (EMFILE);
1675 }
1676 #endif
1677 /*
1678 * fd is already equal to first free descriptor >= minfd, so
1679 * we only need to grow the table and we are done.
1680 */
1681 fdgrowtable_exp(fdp, allocfd);
1682 }
1683
1684 /*
1685 * Perform some sanity checks, then mark the file descriptor as
1686 * used and return it to the caller.
1687 */
1688 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1689 ("invalid descriptor %d", fd));
1690 KASSERT(!fdisused(fdp, fd),
1691 ("fd_first_free() returned non-free descriptor"));
1692 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1693 ("file descriptor isn't free"));
1694 fdused(fdp, fd);
1695 *result = fd;
1696 return (0);
1697 }
1698
1699 /*
1700 * Allocate n file descriptors for the process.
1701 */
1702 int
fdallocn(struct thread * td,int minfd,int * fds,int n)1703 fdallocn(struct thread *td, int minfd, int *fds, int n)
1704 {
1705 struct proc *p = td->td_proc;
1706 struct filedesc *fdp = p->p_fd;
1707 int i;
1708
1709 FILEDESC_XLOCK_ASSERT(fdp);
1710
1711 for (i = 0; i < n; i++)
1712 if (fdalloc(td, 0, &fds[i]) != 0)
1713 break;
1714
1715 if (i < n) {
1716 for (i--; i >= 0; i--)
1717 fdunused(fdp, fds[i]);
1718 return (EMFILE);
1719 }
1720
1721 return (0);
1722 }
1723
1724 /*
1725 * Create a new open file structure and allocate a file descriptor for the
1726 * process that refers to it. We add one reference to the file for the
1727 * descriptor table and one reference for resultfp. This is to prevent us
1728 * being preempted and the entry in the descriptor table closed after we
1729 * release the FILEDESC lock.
1730 */
1731 int
falloc_caps(struct thread * td,struct file ** resultfp,int * resultfd,int flags,struct filecaps * fcaps)1732 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1733 struct filecaps *fcaps)
1734 {
1735 struct file *fp;
1736 int error, fd;
1737
1738 error = falloc_noinstall(td, &fp);
1739 if (error)
1740 return (error); /* no reference held on error */
1741
1742 error = finstall(td, fp, &fd, flags, fcaps);
1743 if (error) {
1744 fdrop(fp, td); /* one reference (fp only) */
1745 return (error);
1746 }
1747
1748 if (resultfp != NULL)
1749 *resultfp = fp; /* copy out result */
1750 else
1751 fdrop(fp, td); /* release local reference */
1752
1753 if (resultfd != NULL)
1754 *resultfd = fd;
1755
1756 return (0);
1757 }
1758
1759 /*
1760 * Create a new open file structure without allocating a file descriptor.
1761 */
1762 int
falloc_noinstall(struct thread * td,struct file ** resultfp)1763 falloc_noinstall(struct thread *td, struct file **resultfp)
1764 {
1765 struct file *fp;
1766 int maxuserfiles = maxfiles - (maxfiles / 20);
1767 int openfiles_new;
1768 static struct timeval lastfail;
1769 static int curfail;
1770
1771 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1772
1773 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1774 if ((openfiles_new >= maxuserfiles &&
1775 priv_check(td, PRIV_MAXFILES) != 0) ||
1776 openfiles_new >= maxfiles) {
1777 atomic_subtract_int(&openfiles, 1);
1778 if (ppsratecheck(&lastfail, &curfail, 1)) {
1779 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1780 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1781 }
1782 return (ENFILE);
1783 }
1784 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1785 refcount_init(&fp->f_count, 1);
1786 fp->f_cred = crhold(td->td_ucred);
1787 fp->f_ops = &badfileops;
1788 *resultfp = fp;
1789 return (0);
1790 }
1791
1792 /*
1793 * Install a file in a file descriptor table.
1794 */
1795 void
_finstall(struct filedesc * fdp,struct file * fp,int fd,int flags,struct filecaps * fcaps)1796 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1797 struct filecaps *fcaps)
1798 {
1799 struct filedescent *fde;
1800
1801 MPASS(fp != NULL);
1802 if (fcaps != NULL)
1803 filecaps_validate(fcaps, __func__);
1804 FILEDESC_XLOCK_ASSERT(fdp);
1805
1806 fde = &fdp->fd_ofiles[fd];
1807 #ifdef CAPABILITIES
1808 seq_write_begin(&fde->fde_seq);
1809 #endif
1810 fde->fde_file = fp;
1811 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1812 if (fcaps != NULL)
1813 filecaps_move(fcaps, &fde->fde_caps);
1814 else
1815 filecaps_fill(&fde->fde_caps);
1816 #ifdef CAPABILITIES
1817 seq_write_end(&fde->fde_seq);
1818 #endif
1819 }
1820
1821 int
finstall(struct thread * td,struct file * fp,int * fd,int flags,struct filecaps * fcaps)1822 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1823 struct filecaps *fcaps)
1824 {
1825 struct filedesc *fdp = td->td_proc->p_fd;
1826 int error;
1827
1828 MPASS(fd != NULL);
1829
1830 FILEDESC_XLOCK(fdp);
1831 if ((error = fdalloc(td, 0, fd))) {
1832 FILEDESC_XUNLOCK(fdp);
1833 return (error);
1834 }
1835 fhold(fp);
1836 _finstall(fdp, fp, *fd, flags, fcaps);
1837 FILEDESC_XUNLOCK(fdp);
1838 return (0);
1839 }
1840
1841 /*
1842 * Build a new filedesc structure from another.
1843 * Copy the current, root, and jail root vnode references.
1844 *
1845 * If fdp is not NULL, return with it shared locked.
1846 */
1847 struct filedesc *
fdinit(struct filedesc * fdp,bool prepfiles)1848 fdinit(struct filedesc *fdp, bool prepfiles)
1849 {
1850 struct filedesc0 *newfdp0;
1851 struct filedesc *newfdp;
1852
1853 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1854 newfdp = &newfdp0->fd_fd;
1855
1856 /* Create the file descriptor table. */
1857 FILEDESC_LOCK_INIT(newfdp);
1858 refcount_init(&newfdp->fd_refcnt, 1);
1859 refcount_init(&newfdp->fd_holdcnt, 1);
1860 newfdp->fd_cmask = CMASK;
1861 newfdp->fd_map = newfdp0->fd_dmap;
1862 newfdp->fd_lastfile = -1;
1863 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1864 newfdp->fd_files->fdt_nfiles = NDFILE;
1865
1866 if (fdp == NULL)
1867 return (newfdp);
1868
1869 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1870 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1871
1872 FILEDESC_SLOCK(fdp);
1873 newfdp->fd_cdir = fdp->fd_cdir;
1874 if (newfdp->fd_cdir)
1875 vrefact(newfdp->fd_cdir);
1876 newfdp->fd_rdir = fdp->fd_rdir;
1877 if (newfdp->fd_rdir)
1878 vrefact(newfdp->fd_rdir);
1879 newfdp->fd_jdir = fdp->fd_jdir;
1880 if (newfdp->fd_jdir)
1881 vrefact(newfdp->fd_jdir);
1882
1883 if (!prepfiles) {
1884 FILEDESC_SUNLOCK(fdp);
1885 } else {
1886 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1887 FILEDESC_SUNLOCK(fdp);
1888 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1889 FILEDESC_SLOCK(fdp);
1890 }
1891 }
1892
1893 return (newfdp);
1894 }
1895
1896 static struct filedesc *
fdhold(struct proc * p)1897 fdhold(struct proc *p)
1898 {
1899 struct filedesc *fdp;
1900
1901 PROC_LOCK_ASSERT(p, MA_OWNED);
1902 fdp = p->p_fd;
1903 if (fdp != NULL)
1904 refcount_acquire(&fdp->fd_holdcnt);
1905 return (fdp);
1906 }
1907
1908 static void
fddrop(struct filedesc * fdp)1909 fddrop(struct filedesc *fdp)
1910 {
1911
1912 if (fdp->fd_holdcnt > 1) {
1913 if (refcount_release(&fdp->fd_holdcnt) == 0)
1914 return;
1915 }
1916
1917 FILEDESC_LOCK_DESTROY(fdp);
1918 uma_zfree(filedesc0_zone, fdp);
1919 }
1920
1921 /*
1922 * Share a filedesc structure.
1923 */
1924 struct filedesc *
fdshare(struct filedesc * fdp)1925 fdshare(struct filedesc *fdp)
1926 {
1927
1928 refcount_acquire(&fdp->fd_refcnt);
1929 return (fdp);
1930 }
1931
1932 /*
1933 * Unshare a filedesc structure, if necessary by making a copy
1934 */
1935 void
fdunshare(struct thread * td)1936 fdunshare(struct thread *td)
1937 {
1938 struct filedesc *tmp;
1939 struct proc *p = td->td_proc;
1940
1941 if (p->p_fd->fd_refcnt == 1)
1942 return;
1943
1944 tmp = fdcopy(p->p_fd);
1945 fdescfree(td);
1946 p->p_fd = tmp;
1947 }
1948
1949 void
fdinstall_remapped(struct thread * td,struct filedesc * fdp)1950 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1951 {
1952
1953 fdescfree(td);
1954 td->td_proc->p_fd = fdp;
1955 }
1956
1957 /*
1958 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1959 * this is to ease callers, not catch errors.
1960 */
1961 struct filedesc *
fdcopy(struct filedesc * fdp)1962 fdcopy(struct filedesc *fdp)
1963 {
1964 struct filedesc *newfdp;
1965 struct filedescent *nfde, *ofde;
1966 int i;
1967
1968 MPASS(fdp != NULL);
1969
1970 newfdp = fdinit(fdp, true);
1971 /* copy all passable descriptors (i.e. not kqueue) */
1972 newfdp->fd_freefile = -1;
1973 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1974 ofde = &fdp->fd_ofiles[i];
1975 if (ofde->fde_file == NULL ||
1976 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1977 if (newfdp->fd_freefile == -1)
1978 newfdp->fd_freefile = i;
1979 continue;
1980 }
1981 nfde = &newfdp->fd_ofiles[i];
1982 *nfde = *ofde;
1983 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
1984 fhold(nfde->fde_file);
1985 fdused_init(newfdp, i);
1986 newfdp->fd_lastfile = i;
1987 }
1988 if (newfdp->fd_freefile == -1)
1989 newfdp->fd_freefile = i;
1990 newfdp->fd_cmask = fdp->fd_cmask;
1991 FILEDESC_SUNLOCK(fdp);
1992 return (newfdp);
1993 }
1994
1995 /*
1996 * Copies a filedesc structure, while remapping all file descriptors
1997 * stored inside using a translation table.
1998 *
1999 * File descriptors are copied over to the new file descriptor table,
2000 * regardless of whether the close-on-exec flag is set.
2001 */
2002 int
fdcopy_remapped(struct filedesc * fdp,const int * fds,size_t nfds,struct filedesc ** ret)2003 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2004 struct filedesc **ret)
2005 {
2006 struct filedesc *newfdp;
2007 struct filedescent *nfde, *ofde;
2008 int error, i;
2009
2010 MPASS(fdp != NULL);
2011
2012 newfdp = fdinit(fdp, true);
2013 if (nfds > fdp->fd_lastfile + 1) {
2014 /* New table cannot be larger than the old one. */
2015 error = E2BIG;
2016 goto bad;
2017 }
2018 /* Copy all passable descriptors (i.e. not kqueue). */
2019 newfdp->fd_freefile = nfds;
2020 for (i = 0; i < nfds; ++i) {
2021 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2022 /* File descriptor out of bounds. */
2023 error = EBADF;
2024 goto bad;
2025 }
2026 ofde = &fdp->fd_ofiles[fds[i]];
2027 if (ofde->fde_file == NULL) {
2028 /* Unused file descriptor. */
2029 error = EBADF;
2030 goto bad;
2031 }
2032 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2033 /* File descriptor cannot be passed. */
2034 error = EINVAL;
2035 goto bad;
2036 }
2037 nfde = &newfdp->fd_ofiles[i];
2038 *nfde = *ofde;
2039 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2040 fhold(nfde->fde_file);
2041 fdused_init(newfdp, i);
2042 newfdp->fd_lastfile = i;
2043 }
2044 newfdp->fd_cmask = fdp->fd_cmask;
2045 FILEDESC_SUNLOCK(fdp);
2046 *ret = newfdp;
2047 return (0);
2048 bad:
2049 FILEDESC_SUNLOCK(fdp);
2050 fdescfree_remapped(newfdp);
2051 return (error);
2052 }
2053
2054 /*
2055 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2056 * one of processes using it exits) and the table used to be shared.
2057 */
2058 static void
fdclearlocks(struct thread * td)2059 fdclearlocks(struct thread *td)
2060 {
2061 struct filedesc *fdp;
2062 struct filedesc_to_leader *fdtol;
2063 struct flock lf;
2064 struct file *fp;
2065 struct proc *p;
2066 struct vnode *vp;
2067 int i;
2068
2069 p = td->td_proc;
2070 fdp = p->p_fd;
2071 fdtol = p->p_fdtol;
2072 MPASS(fdtol != NULL);
2073
2074 FILEDESC_XLOCK(fdp);
2075 KASSERT(fdtol->fdl_refcount > 0,
2076 ("filedesc_to_refcount botch: fdl_refcount=%d",
2077 fdtol->fdl_refcount));
2078 if (fdtol->fdl_refcount == 1 &&
2079 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2080 for (i = 0; i <= fdp->fd_lastfile; i++) {
2081 fp = fdp->fd_ofiles[i].fde_file;
2082 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2083 continue;
2084 fhold(fp);
2085 FILEDESC_XUNLOCK(fdp);
2086 lf.l_whence = SEEK_SET;
2087 lf.l_start = 0;
2088 lf.l_len = 0;
2089 lf.l_type = F_UNLCK;
2090 vp = fp->f_vnode;
2091 (void) VOP_ADVLOCK(vp,
2092 (caddr_t)p->p_leader, F_UNLCK,
2093 &lf, F_POSIX);
2094 FILEDESC_XLOCK(fdp);
2095 fdrop(fp, td);
2096 }
2097 }
2098 retry:
2099 if (fdtol->fdl_refcount == 1) {
2100 if (fdp->fd_holdleaderscount > 0 &&
2101 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2102 /*
2103 * close() or kern_dup() has cleared a reference
2104 * in a shared file descriptor table.
2105 */
2106 fdp->fd_holdleaderswakeup = 1;
2107 sx_sleep(&fdp->fd_holdleaderscount,
2108 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2109 goto retry;
2110 }
2111 if (fdtol->fdl_holdcount > 0) {
2112 /*
2113 * Ensure that fdtol->fdl_leader remains
2114 * valid in closef().
2115 */
2116 fdtol->fdl_wakeup = 1;
2117 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2118 "fdlhold", 0);
2119 goto retry;
2120 }
2121 }
2122 fdtol->fdl_refcount--;
2123 if (fdtol->fdl_refcount == 0 &&
2124 fdtol->fdl_holdcount == 0) {
2125 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2126 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2127 } else
2128 fdtol = NULL;
2129 p->p_fdtol = NULL;
2130 FILEDESC_XUNLOCK(fdp);
2131 if (fdtol != NULL)
2132 free(fdtol, M_FILEDESC_TO_LEADER);
2133 }
2134
2135 /*
2136 * Release a filedesc structure.
2137 */
2138 static void
fdescfree_fds(struct thread * td,struct filedesc * fdp,bool needclose)2139 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2140 {
2141 struct filedesc0 *fdp0;
2142 struct freetable *ft, *tft;
2143 struct filedescent *fde;
2144 struct file *fp;
2145 int i;
2146
2147 for (i = 0; i <= fdp->fd_lastfile; i++) {
2148 fde = &fdp->fd_ofiles[i];
2149 fp = fde->fde_file;
2150 if (fp != NULL) {
2151 fdefree_last(fde);
2152 if (needclose)
2153 (void) closef(fp, td);
2154 else
2155 fdrop(fp, td);
2156 }
2157 }
2158
2159 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2160 free(fdp->fd_map, M_FILEDESC);
2161 if (fdp->fd_nfiles > NDFILE)
2162 free(fdp->fd_files, M_FILEDESC);
2163
2164 fdp0 = (struct filedesc0 *)fdp;
2165 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2166 free(ft->ft_table, M_FILEDESC);
2167
2168 fddrop(fdp);
2169 }
2170
2171 void
fdescfree(struct thread * td)2172 fdescfree(struct thread *td)
2173 {
2174 struct proc *p;
2175 struct filedesc *fdp;
2176 struct vnode *cdir, *jdir, *rdir;
2177
2178 p = td->td_proc;
2179 fdp = p->p_fd;
2180 MPASS(fdp != NULL);
2181
2182 #ifdef RACCT
2183 if (racct_enable) {
2184 PROC_LOCK(p);
2185 racct_set(p, RACCT_NOFILE, 0);
2186 PROC_UNLOCK(p);
2187 }
2188 #endif
2189
2190 if (p->p_fdtol != NULL)
2191 fdclearlocks(td);
2192
2193 PROC_LOCK(p);
2194 p->p_fd = NULL;
2195 PROC_UNLOCK(p);
2196
2197 if (refcount_release(&fdp->fd_refcnt) == 0)
2198 return;
2199
2200 FILEDESC_XLOCK(fdp);
2201 cdir = fdp->fd_cdir;
2202 fdp->fd_cdir = NULL;
2203 rdir = fdp->fd_rdir;
2204 fdp->fd_rdir = NULL;
2205 jdir = fdp->fd_jdir;
2206 fdp->fd_jdir = NULL;
2207 FILEDESC_XUNLOCK(fdp);
2208
2209 if (cdir != NULL)
2210 vrele(cdir);
2211 if (rdir != NULL)
2212 vrele(rdir);
2213 if (jdir != NULL)
2214 vrele(jdir);
2215
2216 fdescfree_fds(td, fdp, 1);
2217 }
2218
2219 void
fdescfree_remapped(struct filedesc * fdp)2220 fdescfree_remapped(struct filedesc *fdp)
2221 {
2222
2223 if (fdp->fd_cdir != NULL)
2224 vrele(fdp->fd_cdir);
2225 if (fdp->fd_rdir != NULL)
2226 vrele(fdp->fd_rdir);
2227 if (fdp->fd_jdir != NULL)
2228 vrele(fdp->fd_jdir);
2229
2230 fdescfree_fds(curthread, fdp, 0);
2231 }
2232
2233 /*
2234 * For setugid programs, we don't want to people to use that setugidness
2235 * to generate error messages which write to a file which otherwise would
2236 * otherwise be off-limits to the process. We check for filesystems where
2237 * the vnode can change out from under us after execve (like [lin]procfs).
2238 *
2239 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2240 * sufficient. We also don't check for setugidness since we know we are.
2241 */
2242 static bool
is_unsafe(struct file * fp)2243 is_unsafe(struct file *fp)
2244 {
2245 struct vnode *vp;
2246
2247 if (fp->f_type != DTYPE_VNODE)
2248 return (false);
2249
2250 vp = fp->f_vnode;
2251 return ((vp->v_vflag & VV_PROCDEP) != 0);
2252 }
2253
2254 /*
2255 * Make this setguid thing safe, if at all possible.
2256 */
2257 void
fdsetugidsafety(struct thread * td)2258 fdsetugidsafety(struct thread *td)
2259 {
2260 struct filedesc *fdp;
2261 struct file *fp;
2262 int i;
2263
2264 fdp = td->td_proc->p_fd;
2265 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2266 MPASS(fdp->fd_nfiles >= 3);
2267 for (i = 0; i <= 2; i++) {
2268 fp = fdp->fd_ofiles[i].fde_file;
2269 if (fp != NULL && is_unsafe(fp)) {
2270 FILEDESC_XLOCK(fdp);
2271 knote_fdclose(td, i);
2272 /*
2273 * NULL-out descriptor prior to close to avoid
2274 * a race while close blocks.
2275 */
2276 fdfree(fdp, i);
2277 FILEDESC_XUNLOCK(fdp);
2278 (void) closef(fp, td);
2279 }
2280 }
2281 }
2282
2283 /*
2284 * If a specific file object occupies a specific file descriptor, close the
2285 * file descriptor entry and drop a reference on the file object. This is a
2286 * convenience function to handle a subsequent error in a function that calls
2287 * falloc() that handles the race that another thread might have closed the
2288 * file descriptor out from under the thread creating the file object.
2289 */
2290 void
fdclose(struct thread * td,struct file * fp,int idx)2291 fdclose(struct thread *td, struct file *fp, int idx)
2292 {
2293 struct filedesc *fdp = td->td_proc->p_fd;
2294
2295 FILEDESC_XLOCK(fdp);
2296 if (fdp->fd_ofiles[idx].fde_file == fp) {
2297 fdfree(fdp, idx);
2298 FILEDESC_XUNLOCK(fdp);
2299 fdrop(fp, td);
2300 } else
2301 FILEDESC_XUNLOCK(fdp);
2302 }
2303
2304 /*
2305 * Close any files on exec?
2306 */
2307 void
fdcloseexec(struct thread * td)2308 fdcloseexec(struct thread *td)
2309 {
2310 struct filedesc *fdp;
2311 struct filedescent *fde;
2312 struct file *fp;
2313 int i;
2314
2315 fdp = td->td_proc->p_fd;
2316 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2317 for (i = 0; i <= fdp->fd_lastfile; i++) {
2318 fde = &fdp->fd_ofiles[i];
2319 fp = fde->fde_file;
2320 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2321 (fde->fde_flags & UF_EXCLOSE))) {
2322 FILEDESC_XLOCK(fdp);
2323 fdfree(fdp, i);
2324 (void) closefp(fdp, i, fp, td, 0);
2325 FILEDESC_UNLOCK_ASSERT(fdp);
2326 }
2327 }
2328 }
2329
2330 /*
2331 * It is unsafe for set[ug]id processes to be started with file
2332 * descriptors 0..2 closed, as these descriptors are given implicit
2333 * significance in the Standard C library. fdcheckstd() will create a
2334 * descriptor referencing /dev/null for each of stdin, stdout, and
2335 * stderr that is not already open.
2336 */
2337 int
fdcheckstd(struct thread * td)2338 fdcheckstd(struct thread *td)
2339 {
2340 struct filedesc *fdp;
2341 register_t save;
2342 int i, error, devnull;
2343
2344 fdp = td->td_proc->p_fd;
2345 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2346 MPASS(fdp->fd_nfiles >= 3);
2347 devnull = -1;
2348 for (i = 0; i <= 2; i++) {
2349 if (fdp->fd_ofiles[i].fde_file != NULL)
2350 continue;
2351
2352 save = td->td_retval[0];
2353 if (devnull != -1) {
2354 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2355 } else {
2356 error = kern_openat(td, AT_FDCWD, "/dev/null",
2357 UIO_SYSSPACE, O_RDWR, 0);
2358 if (error == 0) {
2359 devnull = td->td_retval[0];
2360 KASSERT(devnull == i, ("we didn't get our fd"));
2361 }
2362 }
2363 td->td_retval[0] = save;
2364 if (error != 0)
2365 return (error);
2366 }
2367 return (0);
2368 }
2369
2370 /*
2371 * Internal form of close. Decrement reference count on file structure.
2372 * Note: td may be NULL when closing a file that was being passed in a
2373 * message.
2374 *
2375 * XXXRW: Giant is not required for the caller, but often will be held; this
2376 * makes it moderately likely the Giant will be recursed in the VFS case.
2377 */
2378 int
closef(struct file * fp,struct thread * td)2379 closef(struct file *fp, struct thread *td)
2380 {
2381 struct vnode *vp;
2382 struct flock lf;
2383 struct filedesc_to_leader *fdtol;
2384 struct filedesc *fdp;
2385
2386 /*
2387 * POSIX record locking dictates that any close releases ALL
2388 * locks owned by this process. This is handled by setting
2389 * a flag in the unlock to free ONLY locks obeying POSIX
2390 * semantics, and not to free BSD-style file locks.
2391 * If the descriptor was in a message, POSIX-style locks
2392 * aren't passed with the descriptor, and the thread pointer
2393 * will be NULL. Callers should be careful only to pass a
2394 * NULL thread pointer when there really is no owning
2395 * context that might have locks, or the locks will be
2396 * leaked.
2397 */
2398 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2399 vp = fp->f_vnode;
2400 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2401 lf.l_whence = SEEK_SET;
2402 lf.l_start = 0;
2403 lf.l_len = 0;
2404 lf.l_type = F_UNLCK;
2405 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2406 F_UNLCK, &lf, F_POSIX);
2407 }
2408 fdtol = td->td_proc->p_fdtol;
2409 if (fdtol != NULL) {
2410 /*
2411 * Handle special case where file descriptor table is
2412 * shared between multiple process leaders.
2413 */
2414 fdp = td->td_proc->p_fd;
2415 FILEDESC_XLOCK(fdp);
2416 for (fdtol = fdtol->fdl_next;
2417 fdtol != td->td_proc->p_fdtol;
2418 fdtol = fdtol->fdl_next) {
2419 if ((fdtol->fdl_leader->p_flag &
2420 P_ADVLOCK) == 0)
2421 continue;
2422 fdtol->fdl_holdcount++;
2423 FILEDESC_XUNLOCK(fdp);
2424 lf.l_whence = SEEK_SET;
2425 lf.l_start = 0;
2426 lf.l_len = 0;
2427 lf.l_type = F_UNLCK;
2428 vp = fp->f_vnode;
2429 (void) VOP_ADVLOCK(vp,
2430 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2431 F_POSIX);
2432 FILEDESC_XLOCK(fdp);
2433 fdtol->fdl_holdcount--;
2434 if (fdtol->fdl_holdcount == 0 &&
2435 fdtol->fdl_wakeup != 0) {
2436 fdtol->fdl_wakeup = 0;
2437 wakeup(fdtol);
2438 }
2439 }
2440 FILEDESC_XUNLOCK(fdp);
2441 }
2442 }
2443 return (fdrop(fp, td));
2444 }
2445
2446 /*
2447 * Initialize the file pointer with the specified properties.
2448 *
2449 * The ops are set with release semantics to be certain that the flags, type,
2450 * and data are visible when ops is. This is to prevent ops methods from being
2451 * called with bad data.
2452 */
2453 void
finit(struct file * fp,u_int flag,short type,void * data,struct fileops * ops)2454 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2455 {
2456 fp->f_data = data;
2457 fp->f_flag = flag;
2458 fp->f_type = type;
2459 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2460 }
2461
2462 int
fget_cap_locked(struct filedesc * fdp,int fd,cap_rights_t * needrightsp,struct file ** fpp,struct filecaps * havecapsp)2463 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2464 struct file **fpp, struct filecaps *havecapsp)
2465 {
2466 struct filedescent *fde;
2467 int error;
2468
2469 FILEDESC_LOCK_ASSERT(fdp);
2470
2471 fde = fdeget_locked(fdp, fd);
2472 if (fde == NULL) {
2473 error = EBADF;
2474 goto out;
2475 }
2476
2477 #ifdef CAPABILITIES
2478 error = cap_check(cap_rights_fde(fde), needrightsp);
2479 if (error != 0)
2480 goto out;
2481 #endif
2482
2483 if (havecapsp != NULL)
2484 filecaps_copy(&fde->fde_caps, havecapsp, true);
2485
2486 *fpp = fde->fde_file;
2487
2488 error = 0;
2489 out:
2490 return (error);
2491 }
2492
2493 int
fget_cap(struct thread * td,int fd,cap_rights_t * needrightsp,struct file ** fpp,struct filecaps * havecapsp)2494 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2495 struct file **fpp, struct filecaps *havecapsp)
2496 {
2497 struct filedesc *fdp = td->td_proc->p_fd;
2498 int error;
2499 #ifndef CAPABILITIES
2500 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2501 if (error == 0 && havecapsp != NULL)
2502 filecaps_fill(havecapsp);
2503 #else
2504 struct file *fp;
2505 seq_t seq;
2506
2507 for (;;) {
2508 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2509 if (error != 0)
2510 return (error);
2511
2512 if (havecapsp != NULL) {
2513 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2514 havecapsp, false)) {
2515 fdrop(fp, td);
2516 goto get_locked;
2517 }
2518 }
2519
2520 if (!fd_modified(fdp, fd, seq))
2521 break;
2522 fdrop(fp, td);
2523 }
2524
2525 *fpp = fp;
2526 return (0);
2527
2528 get_locked:
2529 FILEDESC_SLOCK(fdp);
2530 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2531 if (error == 0)
2532 fhold(*fpp);
2533 FILEDESC_SUNLOCK(fdp);
2534 #endif
2535 return (error);
2536 }
2537
2538 int
fget_unlocked(struct filedesc * fdp,int fd,cap_rights_t * needrightsp,struct file ** fpp,seq_t * seqp)2539 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2540 struct file **fpp, seq_t *seqp)
2541 {
2542 #ifdef CAPABILITIES
2543 struct filedescent *fde;
2544 #endif
2545 struct fdescenttbl *fdt;
2546 struct file *fp;
2547 u_int count;
2548 #ifdef CAPABILITIES
2549 seq_t seq;
2550 cap_rights_t haverights;
2551 int error;
2552 #endif
2553
2554 fdt = fdp->fd_files;
2555 if ((u_int)fd >= fdt->fdt_nfiles)
2556 return (EBADF);
2557 /*
2558 * Fetch the descriptor locklessly. We avoid fdrop() races by
2559 * never raising a refcount above 0. To accomplish this we have
2560 * to use a cmpset loop rather than an atomic_add. The descriptor
2561 * must be re-verified once we acquire a reference to be certain
2562 * that the identity is still correct and we did not lose a race
2563 * due to preemption.
2564 */
2565 for (;;) {
2566 #ifdef CAPABILITIES
2567 seq = seq_read(fd_seq(fdt, fd));
2568 fde = &fdt->fdt_ofiles[fd];
2569 haverights = *cap_rights_fde(fde);
2570 fp = fde->fde_file;
2571 if (!seq_consistent(fd_seq(fdt, fd), seq))
2572 continue;
2573 #else
2574 fp = fdt->fdt_ofiles[fd].fde_file;
2575 #endif
2576 if (fp == NULL)
2577 return (EBADF);
2578 #ifdef CAPABILITIES
2579 error = cap_check(&haverights, needrightsp);
2580 if (error != 0)
2581 return (error);
2582 #endif
2583 count = fp->f_count;
2584 retry:
2585 if (count == 0) {
2586 /*
2587 * Force a reload. Other thread could reallocate the
2588 * table before this fd was closed, so it possible that
2589 * there is a stale fp pointer in cached version.
2590 */
2591 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2592 continue;
2593 }
2594 /*
2595 * Use an acquire barrier to force re-reading of fdt so it is
2596 * refreshed for verification.
2597 */
2598 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2599 goto retry;
2600 fdt = fdp->fd_files;
2601 #ifdef CAPABILITIES
2602 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2603 #else
2604 if (fp == fdt->fdt_ofiles[fd].fde_file)
2605 #endif
2606 break;
2607 fdrop(fp, curthread);
2608 }
2609 *fpp = fp;
2610 if (seqp != NULL) {
2611 #ifdef CAPABILITIES
2612 *seqp = seq;
2613 #endif
2614 }
2615 return (0);
2616 }
2617
2618 /*
2619 * Extract the file pointer associated with the specified descriptor for the
2620 * current user process.
2621 *
2622 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2623 * returned.
2624 *
2625 * File's rights will be checked against the capability rights mask.
2626 *
2627 * If an error occurred the non-zero error is returned and *fpp is set to
2628 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2629 * responsible for fdrop().
2630 */
2631 static __inline int
_fget(struct thread * td,int fd,struct file ** fpp,int flags,cap_rights_t * needrightsp,seq_t * seqp)2632 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2633 cap_rights_t *needrightsp, seq_t *seqp)
2634 {
2635 struct filedesc *fdp;
2636 struct file *fp;
2637 int error;
2638
2639 *fpp = NULL;
2640 fdp = td->td_proc->p_fd;
2641 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2642 if (error != 0)
2643 return (error);
2644 if (fp->f_ops == &badfileops) {
2645 fdrop(fp, td);
2646 return (EBADF);
2647 }
2648
2649 /*
2650 * FREAD and FWRITE failure return EBADF as per POSIX.
2651 */
2652 error = 0;
2653 switch (flags) {
2654 case FREAD:
2655 case FWRITE:
2656 if ((fp->f_flag & flags) == 0)
2657 error = EBADF;
2658 break;
2659 case FEXEC:
2660 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2661 ((fp->f_flag & FWRITE) != 0))
2662 error = EBADF;
2663 break;
2664 case 0:
2665 break;
2666 default:
2667 KASSERT(0, ("wrong flags"));
2668 }
2669
2670 if (error != 0) {
2671 fdrop(fp, td);
2672 return (error);
2673 }
2674
2675 *fpp = fp;
2676 return (0);
2677 }
2678
2679 int
fget(struct thread * td,int fd,cap_rights_t * rightsp,struct file ** fpp)2680 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2681 {
2682
2683 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2684 }
2685
2686 int
fget_mmap(struct thread * td,int fd,cap_rights_t * rightsp,vm_prot_t * maxprotp,struct file ** fpp)2687 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
2688 struct file **fpp)
2689 {
2690 int error;
2691 #ifndef CAPABILITIES
2692 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2693 if (maxprotp != NULL)
2694 *maxprotp = VM_PROT_ALL;
2695 #else
2696 cap_rights_t fdrights;
2697 struct filedesc *fdp = td->td_proc->p_fd;
2698 seq_t seq;
2699
2700 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2701 for (;;) {
2702 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2703 if (error != 0)
2704 return (error);
2705 if (maxprotp != NULL)
2706 fdrights = *cap_rights(fdp, fd);
2707 if (!fd_modified(fdp, fd, seq))
2708 break;
2709 fdrop(*fpp, td);
2710 }
2711
2712 /*
2713 * If requested, convert capability rights to access flags.
2714 */
2715 if (maxprotp != NULL)
2716 *maxprotp = cap_rights_to_vmprot(&fdrights);
2717 #endif
2718 return (error);
2719 }
2720
2721 int
fget_read(struct thread * td,int fd,cap_rights_t * rightsp,struct file ** fpp)2722 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2723 {
2724
2725 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2726 }
2727
2728 int
fget_write(struct thread * td,int fd,cap_rights_t * rightsp,struct file ** fpp)2729 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2730 {
2731
2732 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2733 }
2734
2735 int
fget_fcntl(struct thread * td,int fd,cap_rights_t * rightsp,int needfcntl,struct file ** fpp)2736 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2737 struct file **fpp)
2738 {
2739 struct filedesc *fdp = td->td_proc->p_fd;
2740 #ifndef CAPABILITIES
2741 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2742 #else
2743 int error;
2744 seq_t seq;
2745
2746 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2747 for (;;) {
2748 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2749 if (error != 0)
2750 return (error);
2751 error = cap_fcntl_check(fdp, fd, needfcntl);
2752 if (!fd_modified(fdp, fd, seq))
2753 break;
2754 fdrop(*fpp, td);
2755 }
2756 if (error != 0) {
2757 fdrop(*fpp, td);
2758 *fpp = NULL;
2759 }
2760 return (error);
2761 #endif
2762 }
2763
2764 /*
2765 * Like fget() but loads the underlying vnode, or returns an error if the
2766 * descriptor does not represent a vnode. Note that pipes use vnodes but
2767 * never have VM objects. The returned vnode will be vref()'d.
2768 *
2769 * XXX: what about the unused flags ?
2770 */
2771 static __inline int
_fgetvp(struct thread * td,int fd,int flags,cap_rights_t * needrightsp,struct vnode ** vpp)2772 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2773 struct vnode **vpp)
2774 {
2775 struct file *fp;
2776 int error;
2777
2778 *vpp = NULL;
2779 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2780 if (error != 0)
2781 return (error);
2782 if (fp->f_vnode == NULL) {
2783 error = EINVAL;
2784 } else {
2785 *vpp = fp->f_vnode;
2786 vrefact(*vpp);
2787 }
2788 fdrop(fp, td);
2789
2790 return (error);
2791 }
2792
2793 int
fgetvp(struct thread * td,int fd,cap_rights_t * rightsp,struct vnode ** vpp)2794 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2795 {
2796
2797 return (_fgetvp(td, fd, 0, rightsp, vpp));
2798 }
2799
2800 int
fgetvp_rights(struct thread * td,int fd,cap_rights_t * needrightsp,struct filecaps * havecaps,struct vnode ** vpp)2801 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2802 struct filecaps *havecaps, struct vnode **vpp)
2803 {
2804 struct filedesc *fdp;
2805 struct filecaps caps;
2806 struct file *fp;
2807 int error;
2808
2809 fdp = td->td_proc->p_fd;
2810 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2811 if (error != 0)
2812 return (error);
2813 if (fp->f_ops == &badfileops) {
2814 error = EBADF;
2815 goto out;
2816 }
2817 if (fp->f_vnode == NULL) {
2818 error = EINVAL;
2819 goto out;
2820 }
2821
2822 *havecaps = caps;
2823 *vpp = fp->f_vnode;
2824 vrefact(*vpp);
2825
2826 return (0);
2827 out:
2828 filecaps_free(&caps);
2829 return (error);
2830 }
2831
2832 int
fgetvp_read(struct thread * td,int fd,cap_rights_t * rightsp,struct vnode ** vpp)2833 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2834 {
2835
2836 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2837 }
2838
2839 int
fgetvp_exec(struct thread * td,int fd,cap_rights_t * rightsp,struct vnode ** vpp)2840 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2841 {
2842
2843 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2844 }
2845
2846 #ifdef notyet
2847 int
fgetvp_write(struct thread * td,int fd,cap_rights_t * rightsp,struct vnode ** vpp)2848 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2849 struct vnode **vpp)
2850 {
2851
2852 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2853 }
2854 #endif
2855
2856 /*
2857 * Handle the last reference to a file being closed.
2858 */
2859 int
_fdrop(struct file * fp,struct thread * td)2860 _fdrop(struct file *fp, struct thread *td)
2861 {
2862 int error;
2863
2864 if (fp->f_count != 0)
2865 panic("fdrop: count %d", fp->f_count);
2866 error = fo_close(fp, td);
2867 atomic_subtract_int(&openfiles, 1);
2868 crfree(fp->f_cred);
2869 free(fp->f_advice, M_FADVISE);
2870 uma_zfree(file_zone, fp);
2871
2872 return (error);
2873 }
2874
2875 /*
2876 * Apply an advisory lock on a file descriptor.
2877 *
2878 * Just attempt to get a record lock of the requested type on the entire file
2879 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2880 */
2881 #ifndef _SYS_SYSPROTO_H_
2882 struct flock_args {
2883 int fd;
2884 int how;
2885 };
2886 #endif
2887 /* ARGSUSED */
2888 int
sys_flock(struct thread * td,struct flock_args * uap)2889 sys_flock(struct thread *td, struct flock_args *uap)
2890 {
2891 struct file *fp;
2892 struct vnode *vp;
2893 struct flock lf;
2894 cap_rights_t rights;
2895 int error;
2896
2897 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2898 if (error != 0)
2899 return (error);
2900 if (fp->f_type != DTYPE_VNODE) {
2901 fdrop(fp, td);
2902 return (EOPNOTSUPP);
2903 }
2904
2905 vp = fp->f_vnode;
2906 lf.l_whence = SEEK_SET;
2907 lf.l_start = 0;
2908 lf.l_len = 0;
2909 if (uap->how & LOCK_UN) {
2910 lf.l_type = F_UNLCK;
2911 atomic_clear_int(&fp->f_flag, FHASLOCK);
2912 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2913 goto done2;
2914 }
2915 if (uap->how & LOCK_EX)
2916 lf.l_type = F_WRLCK;
2917 else if (uap->how & LOCK_SH)
2918 lf.l_type = F_RDLCK;
2919 else {
2920 error = EBADF;
2921 goto done2;
2922 }
2923 atomic_set_int(&fp->f_flag, FHASLOCK);
2924 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2925 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2926 done2:
2927 fdrop(fp, td);
2928 return (error);
2929 }
2930 /*
2931 * Duplicate the specified descriptor to a free descriptor.
2932 */
2933 int
dupfdopen(struct thread * td,struct filedesc * fdp,int dfd,int mode,int openerror,int * indxp)2934 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2935 int openerror, int *indxp)
2936 {
2937 struct filedescent *newfde, *oldfde;
2938 struct file *fp;
2939 int error, indx;
2940
2941 KASSERT(openerror == ENODEV || openerror == ENXIO,
2942 ("unexpected error %d in %s", openerror, __func__));
2943
2944 /*
2945 * If the to-be-dup'd fd number is greater than the allowed number
2946 * of file descriptors, or the fd to be dup'd has already been
2947 * closed, then reject.
2948 */
2949 FILEDESC_XLOCK(fdp);
2950 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2951 FILEDESC_XUNLOCK(fdp);
2952 return (EBADF);
2953 }
2954
2955 error = fdalloc(td, 0, &indx);
2956 if (error != 0) {
2957 FILEDESC_XUNLOCK(fdp);
2958 return (error);
2959 }
2960
2961 /*
2962 * There are two cases of interest here.
2963 *
2964 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2965 *
2966 * For ENXIO steal away the file structure from (dfd) and store it in
2967 * (indx). (dfd) is effectively closed by this operation.
2968 */
2969 switch (openerror) {
2970 case ENODEV:
2971 /*
2972 * Check that the mode the file is being opened for is a
2973 * subset of the mode of the existing descriptor.
2974 */
2975 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2976 fdunused(fdp, indx);
2977 FILEDESC_XUNLOCK(fdp);
2978 return (EACCES);
2979 }
2980 fhold(fp);
2981 newfde = &fdp->fd_ofiles[indx];
2982 oldfde = &fdp->fd_ofiles[dfd];
2983 #ifdef CAPABILITIES
2984 seq_write_begin(&newfde->fde_seq);
2985 #endif
2986 memcpy(newfde, oldfde, fde_change_size);
2987 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
2988 #ifdef CAPABILITIES
2989 seq_write_end(&newfde->fde_seq);
2990 #endif
2991 break;
2992 case ENXIO:
2993 /*
2994 * Steal away the file pointer from dfd and stuff it into indx.
2995 */
2996 newfde = &fdp->fd_ofiles[indx];
2997 oldfde = &fdp->fd_ofiles[dfd];
2998 #ifdef CAPABILITIES
2999 seq_write_begin(&newfde->fde_seq);
3000 #endif
3001 memcpy(newfde, oldfde, fde_change_size);
3002 oldfde->fde_file = NULL;
3003 fdunused(fdp, dfd);
3004 #ifdef CAPABILITIES
3005 seq_write_end(&newfde->fde_seq);
3006 #endif
3007 break;
3008 }
3009 FILEDESC_XUNLOCK(fdp);
3010 *indxp = indx;
3011 return (0);
3012 }
3013
3014 /*
3015 * This sysctl determines if we will allow a process to chroot(2) if it
3016 * has a directory open:
3017 * 0: disallowed for all processes.
3018 * 1: allowed for processes that were not already chroot(2)'ed.
3019 * 2: allowed for all processes.
3020 */
3021
3022 static int chroot_allow_open_directories = 1;
3023
3024 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3025 &chroot_allow_open_directories, 0,
3026 "Allow a process to chroot(2) if it has a directory open");
3027
3028 /*
3029 * Helper function for raised chroot(2) security function: Refuse if
3030 * any filedescriptors are open directories.
3031 */
3032 static int
chroot_refuse_vdir_fds(struct filedesc * fdp)3033 chroot_refuse_vdir_fds(struct filedesc *fdp)
3034 {
3035 struct vnode *vp;
3036 struct file *fp;
3037 int fd;
3038
3039 FILEDESC_LOCK_ASSERT(fdp);
3040
3041 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3042 fp = fget_locked(fdp, fd);
3043 if (fp == NULL)
3044 continue;
3045 if (fp->f_type == DTYPE_VNODE) {
3046 vp = fp->f_vnode;
3047 if (vp->v_type == VDIR)
3048 return (EPERM);
3049 }
3050 }
3051 return (0);
3052 }
3053
3054 /*
3055 * The caller is responsible for invoking priv_check() and
3056 * mac_vnode_check_chroot() to authorize this operation.
3057 */
3058 int
pwd_chroot(struct thread * td,struct vnode * vp)3059 pwd_chroot(struct thread *td, struct vnode *vp)
3060 {
3061 struct filedesc *fdp;
3062 struct vnode *oldvp;
3063 int error;
3064
3065 fdp = td->td_proc->p_fd;
3066 FILEDESC_XLOCK(fdp);
3067 if (chroot_allow_open_directories == 0 ||
3068 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3069 error = chroot_refuse_vdir_fds(fdp);
3070 if (error != 0) {
3071 FILEDESC_XUNLOCK(fdp);
3072 return (error);
3073 }
3074 }
3075 oldvp = fdp->fd_rdir;
3076 vrefact(vp);
3077 fdp->fd_rdir = vp;
3078 if (fdp->fd_jdir == NULL) {
3079 vrefact(vp);
3080 fdp->fd_jdir = vp;
3081 }
3082 FILEDESC_XUNLOCK(fdp);
3083 vrele(oldvp);
3084 return (0);
3085 }
3086
3087 void
pwd_chdir(struct thread * td,struct vnode * vp)3088 pwd_chdir(struct thread *td, struct vnode *vp)
3089 {
3090 struct filedesc *fdp;
3091 struct vnode *oldvp;
3092
3093 fdp = td->td_proc->p_fd;
3094 FILEDESC_XLOCK(fdp);
3095 VNASSERT(vp->v_usecount > 0, vp,
3096 ("chdir to a vnode with zero usecount"));
3097 oldvp = fdp->fd_cdir;
3098 fdp->fd_cdir = vp;
3099 FILEDESC_XUNLOCK(fdp);
3100 vrele(oldvp);
3101 }
3102
3103 /*
3104 * jail_attach(2) changes both root and working directories.
3105 */
3106 int
pwd_chroot_chdir(struct thread * td,struct vnode * vp)3107 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3108 {
3109 struct filedesc *fdp;
3110 struct vnode *oldvrp, *oldvcp;
3111 int error;
3112
3113 fdp = td->td_proc->p_fd;
3114 FILEDESC_XLOCK(fdp);
3115 error = chroot_refuse_vdir_fds(fdp);
3116 if (error != 0) {
3117 FILEDESC_XUNLOCK(fdp);
3118 return (error);
3119 }
3120 oldvrp = fdp->fd_rdir;
3121 vrefact(vp);
3122 fdp->fd_rdir = vp;
3123 oldvcp = fdp->fd_cdir;
3124 vrefact(vp);
3125 fdp->fd_cdir = vp;
3126 if (fdp->fd_jdir == NULL) {
3127 vrefact(vp);
3128 fdp->fd_jdir = vp;
3129 }
3130 FILEDESC_XUNLOCK(fdp);
3131 vrele(oldvrp);
3132 vrele(oldvcp);
3133 return (0);
3134 }
3135
3136 /*
3137 * Scan all active processes and prisons to see if any of them have a current
3138 * or root directory of `olddp'. If so, replace them with the new mount point.
3139 */
3140 void
mountcheckdirs(struct vnode * olddp,struct vnode * newdp)3141 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3142 {
3143 struct filedesc *fdp;
3144 struct prison *pr;
3145 struct proc *p;
3146 int nrele;
3147
3148 if (vrefcnt(olddp) == 1)
3149 return;
3150 nrele = 0;
3151 sx_slock(&allproc_lock);
3152 FOREACH_PROC_IN_SYSTEM(p) {
3153 PROC_LOCK(p);
3154 fdp = fdhold(p);
3155 PROC_UNLOCK(p);
3156 if (fdp == NULL)
3157 continue;
3158 FILEDESC_XLOCK(fdp);
3159 if (fdp->fd_cdir == olddp) {
3160 vrefact(newdp);
3161 fdp->fd_cdir = newdp;
3162 nrele++;
3163 }
3164 if (fdp->fd_rdir == olddp) {
3165 vrefact(newdp);
3166 fdp->fd_rdir = newdp;
3167 nrele++;
3168 }
3169 if (fdp->fd_jdir == olddp) {
3170 vrefact(newdp);
3171 fdp->fd_jdir = newdp;
3172 nrele++;
3173 }
3174 FILEDESC_XUNLOCK(fdp);
3175 fddrop(fdp);
3176 }
3177 sx_sunlock(&allproc_lock);
3178 if (rootvnode == olddp) {
3179 vrefact(newdp);
3180 rootvnode = newdp;
3181 nrele++;
3182 }
3183 mtx_lock(&prison0.pr_mtx);
3184 if (prison0.pr_root == olddp) {
3185 vrefact(newdp);
3186 prison0.pr_root = newdp;
3187 nrele++;
3188 }
3189 mtx_unlock(&prison0.pr_mtx);
3190 sx_slock(&allprison_lock);
3191 TAILQ_FOREACH(pr, &allprison, pr_list) {
3192 mtx_lock(&pr->pr_mtx);
3193 if (pr->pr_root == olddp) {
3194 vrefact(newdp);
3195 pr->pr_root = newdp;
3196 nrele++;
3197 }
3198 mtx_unlock(&pr->pr_mtx);
3199 }
3200 sx_sunlock(&allprison_lock);
3201 while (nrele--)
3202 vrele(olddp);
3203 }
3204
3205 struct filedesc_to_leader *
filedesc_to_leader_alloc(struct filedesc_to_leader * old,struct filedesc * fdp,struct proc * leader)3206 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3207 {
3208 struct filedesc_to_leader *fdtol;
3209
3210 fdtol = malloc(sizeof(struct filedesc_to_leader),
3211 M_FILEDESC_TO_LEADER, M_WAITOK);
3212 fdtol->fdl_refcount = 1;
3213 fdtol->fdl_holdcount = 0;
3214 fdtol->fdl_wakeup = 0;
3215 fdtol->fdl_leader = leader;
3216 if (old != NULL) {
3217 FILEDESC_XLOCK(fdp);
3218 fdtol->fdl_next = old->fdl_next;
3219 fdtol->fdl_prev = old;
3220 old->fdl_next = fdtol;
3221 fdtol->fdl_next->fdl_prev = fdtol;
3222 FILEDESC_XUNLOCK(fdp);
3223 } else {
3224 fdtol->fdl_next = fdtol;
3225 fdtol->fdl_prev = fdtol;
3226 }
3227 return (fdtol);
3228 }
3229
3230 static int
sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)3231 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3232 {
3233 struct filedesc *fdp;
3234 int i, count, slots;
3235
3236 if (*(int *)arg1 != 0)
3237 return (EINVAL);
3238
3239 fdp = curproc->p_fd;
3240 count = 0;
3241 FILEDESC_SLOCK(fdp);
3242 slots = NDSLOTS(fdp->fd_lastfile + 1);
3243 for (i = 0; i < slots; i++)
3244 count += bitcountl(fdp->fd_map[i]);
3245 FILEDESC_SUNLOCK(fdp);
3246
3247 return (SYSCTL_OUT(req, &count, sizeof(count)));
3248 }
3249
3250 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3251 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3252 "Number of open file descriptors");
3253
3254 /*
3255 * Get file structures globally.
3256 */
3257 static int
sysctl_kern_file(SYSCTL_HANDLER_ARGS)3258 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3259 {
3260 struct xfile xf;
3261 struct filedesc *fdp;
3262 struct file *fp;
3263 struct proc *p;
3264 int error, n;
3265
3266 error = sysctl_wire_old_buffer(req, 0);
3267 if (error != 0)
3268 return (error);
3269 if (req->oldptr == NULL) {
3270 n = 0;
3271 sx_slock(&allproc_lock);
3272 FOREACH_PROC_IN_SYSTEM(p) {
3273 PROC_LOCK(p);
3274 if (p->p_state == PRS_NEW) {
3275 PROC_UNLOCK(p);
3276 continue;
3277 }
3278 fdp = fdhold(p);
3279 PROC_UNLOCK(p);
3280 if (fdp == NULL)
3281 continue;
3282 /* overestimates sparse tables. */
3283 if (fdp->fd_lastfile > 0)
3284 n += fdp->fd_lastfile;
3285 fddrop(fdp);
3286 }
3287 sx_sunlock(&allproc_lock);
3288 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3289 }
3290 error = 0;
3291 bzero(&xf, sizeof(xf));
3292 xf.xf_size = sizeof(xf);
3293 sx_slock(&allproc_lock);
3294 FOREACH_PROC_IN_SYSTEM(p) {
3295 PROC_LOCK(p);
3296 if (p->p_state == PRS_NEW) {
3297 PROC_UNLOCK(p);
3298 continue;
3299 }
3300 if (p_cansee(req->td, p) != 0) {
3301 PROC_UNLOCK(p);
3302 continue;
3303 }
3304 xf.xf_pid = p->p_pid;
3305 xf.xf_uid = p->p_ucred->cr_uid;
3306 fdp = fdhold(p);
3307 PROC_UNLOCK(p);
3308 if (fdp == NULL)
3309 continue;
3310 FILEDESC_SLOCK(fdp);
3311 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3312 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3313 continue;
3314 xf.xf_fd = n;
3315 xf.xf_file = fp;
3316 xf.xf_data = fp->f_data;
3317 xf.xf_vnode = fp->f_vnode;
3318 xf.xf_type = fp->f_type;
3319 xf.xf_count = fp->f_count;
3320 xf.xf_msgcount = 0;
3321 xf.xf_offset = foffset_get(fp);
3322 xf.xf_flag = fp->f_flag;
3323 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3324 if (error)
3325 break;
3326 }
3327 FILEDESC_SUNLOCK(fdp);
3328 fddrop(fdp);
3329 if (error)
3330 break;
3331 }
3332 sx_sunlock(&allproc_lock);
3333 return (error);
3334 }
3335
3336 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3337 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3338
3339 #ifdef KINFO_FILE_SIZE
3340 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3341 #endif
3342
3343 static int
xlate_fflags(int fflags)3344 xlate_fflags(int fflags)
3345 {
3346 static const struct {
3347 int fflag;
3348 int kf_fflag;
3349 } fflags_table[] = {
3350 { FAPPEND, KF_FLAG_APPEND },
3351 { FASYNC, KF_FLAG_ASYNC },
3352 { FFSYNC, KF_FLAG_FSYNC },
3353 { FHASLOCK, KF_FLAG_HASLOCK },
3354 { FNONBLOCK, KF_FLAG_NONBLOCK },
3355 { FREAD, KF_FLAG_READ },
3356 { FWRITE, KF_FLAG_WRITE },
3357 { O_CREAT, KF_FLAG_CREAT },
3358 { O_DIRECT, KF_FLAG_DIRECT },
3359 { O_EXCL, KF_FLAG_EXCL },
3360 { O_EXEC, KF_FLAG_EXEC },
3361 { O_EXLOCK, KF_FLAG_EXLOCK },
3362 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3363 { O_SHLOCK, KF_FLAG_SHLOCK },
3364 { O_TRUNC, KF_FLAG_TRUNC }
3365 };
3366 unsigned int i;
3367 int kflags;
3368
3369 kflags = 0;
3370 for (i = 0; i < nitems(fflags_table); i++)
3371 if (fflags & fflags_table[i].fflag)
3372 kflags |= fflags_table[i].kf_fflag;
3373 return (kflags);
3374 }
3375
3376 /* Trim unused data from kf_path by truncating the structure size. */
3377 static void
pack_kinfo(struct kinfo_file * kif)3378 pack_kinfo(struct kinfo_file *kif)
3379 {
3380
3381 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3382 strlen(kif->kf_path) + 1;
3383 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3384 }
3385
3386 static void
export_file_to_kinfo(struct file * fp,int fd,cap_rights_t * rightsp,struct kinfo_file * kif,struct filedesc * fdp,int flags)3387 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3388 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3389 {
3390 int error;
3391
3392 bzero(kif, sizeof(*kif));
3393
3394 /* Set a default type to allow for empty fill_kinfo() methods. */
3395 kif->kf_type = KF_TYPE_UNKNOWN;
3396 kif->kf_flags = xlate_fflags(fp->f_flag);
3397 if (rightsp != NULL)
3398 kif->kf_cap_rights = *rightsp;
3399 else
3400 cap_rights_init(&kif->kf_cap_rights);
3401 kif->kf_fd = fd;
3402 kif->kf_ref_count = fp->f_count;
3403 kif->kf_offset = foffset_get(fp);
3404
3405 /*
3406 * This may drop the filedesc lock, so the 'fp' cannot be
3407 * accessed after this call.
3408 */
3409 error = fo_fill_kinfo(fp, kif, fdp);
3410 if (error == 0)
3411 kif->kf_status |= KF_ATTR_VALID;
3412 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3413 pack_kinfo(kif);
3414 else
3415 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3416 }
3417
3418 static void
export_vnode_to_kinfo(struct vnode * vp,int fd,int fflags,struct kinfo_file * kif,int flags)3419 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3420 struct kinfo_file *kif, int flags)
3421 {
3422 int error;
3423
3424 bzero(kif, sizeof(*kif));
3425
3426 kif->kf_type = KF_TYPE_VNODE;
3427 error = vn_fill_kinfo_vnode(vp, kif);
3428 if (error == 0)
3429 kif->kf_status |= KF_ATTR_VALID;
3430 kif->kf_flags = xlate_fflags(fflags);
3431 cap_rights_init(&kif->kf_cap_rights);
3432 kif->kf_fd = fd;
3433 kif->kf_ref_count = -1;
3434 kif->kf_offset = -1;
3435 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3436 pack_kinfo(kif);
3437 else
3438 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3439 vrele(vp);
3440 }
3441
3442 struct export_fd_buf {
3443 struct filedesc *fdp;
3444 struct sbuf *sb;
3445 ssize_t remainder;
3446 struct kinfo_file kif;
3447 int flags;
3448 };
3449
3450 static int
export_kinfo_to_sb(struct export_fd_buf * efbuf)3451 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3452 {
3453 struct kinfo_file *kif;
3454
3455 kif = &efbuf->kif;
3456 if (efbuf->remainder != -1) {
3457 if (efbuf->remainder < kif->kf_structsize) {
3458 /* Terminate export. */
3459 efbuf->remainder = 0;
3460 return (0);
3461 }
3462 efbuf->remainder -= kif->kf_structsize;
3463 }
3464 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3465 }
3466
3467 static int
export_file_to_sb(struct file * fp,int fd,cap_rights_t * rightsp,struct export_fd_buf * efbuf)3468 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3469 struct export_fd_buf *efbuf)
3470 {
3471 int error;
3472
3473 if (efbuf->remainder == 0)
3474 return (0);
3475 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3476 efbuf->flags);
3477 FILEDESC_SUNLOCK(efbuf->fdp);
3478 error = export_kinfo_to_sb(efbuf);
3479 FILEDESC_SLOCK(efbuf->fdp);
3480 return (error);
3481 }
3482
3483 static int
export_vnode_to_sb(struct vnode * vp,int fd,int fflags,struct export_fd_buf * efbuf)3484 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3485 struct export_fd_buf *efbuf)
3486 {
3487 int error;
3488
3489 if (efbuf->remainder == 0)
3490 return (0);
3491 if (efbuf->fdp != NULL)
3492 FILEDESC_SUNLOCK(efbuf->fdp);
3493 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3494 error = export_kinfo_to_sb(efbuf);
3495 if (efbuf->fdp != NULL)
3496 FILEDESC_SLOCK(efbuf->fdp);
3497 return (error);
3498 }
3499
3500 /*
3501 * Store a process file descriptor information to sbuf.
3502 *
3503 * Takes a locked proc as argument, and returns with the proc unlocked.
3504 */
3505 int
kern_proc_filedesc_out(struct proc * p,struct sbuf * sb,ssize_t maxlen,int flags)3506 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3507 int flags)
3508 {
3509 struct file *fp;
3510 struct filedesc *fdp;
3511 struct export_fd_buf *efbuf;
3512 struct vnode *cttyvp, *textvp, *tracevp;
3513 int error, i;
3514 cap_rights_t rights;
3515
3516 PROC_LOCK_ASSERT(p, MA_OWNED);
3517
3518 /* ktrace vnode */
3519 tracevp = p->p_tracevp;
3520 if (tracevp != NULL)
3521 vrefact(tracevp);
3522 /* text vnode */
3523 textvp = p->p_textvp;
3524 if (textvp != NULL)
3525 vrefact(textvp);
3526 /* Controlling tty. */
3527 cttyvp = NULL;
3528 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3529 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3530 if (cttyvp != NULL)
3531 vrefact(cttyvp);
3532 }
3533 fdp = fdhold(p);
3534 PROC_UNLOCK(p);
3535 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3536 efbuf->fdp = NULL;
3537 efbuf->sb = sb;
3538 efbuf->remainder = maxlen;
3539 efbuf->flags = flags;
3540 if (tracevp != NULL)
3541 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3542 efbuf);
3543 if (textvp != NULL)
3544 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3545 if (cttyvp != NULL)
3546 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3547 efbuf);
3548 error = 0;
3549 if (fdp == NULL)
3550 goto fail;
3551 efbuf->fdp = fdp;
3552 FILEDESC_SLOCK(fdp);
3553 /* working directory */
3554 if (fdp->fd_cdir != NULL) {
3555 vrefact(fdp->fd_cdir);
3556 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3557 }
3558 /* root directory */
3559 if (fdp->fd_rdir != NULL) {
3560 vrefact(fdp->fd_rdir);
3561 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3562 }
3563 /* jail directory */
3564 if (fdp->fd_jdir != NULL) {
3565 vrefact(fdp->fd_jdir);
3566 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3567 }
3568 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3569 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3570 continue;
3571 #ifdef CAPABILITIES
3572 rights = *cap_rights(fdp, i);
3573 #else /* !CAPABILITIES */
3574 cap_rights_init(&rights);
3575 #endif
3576 /*
3577 * Create sysctl entry. It is OK to drop the filedesc
3578 * lock inside of export_file_to_sb() as we will
3579 * re-validate and re-evaluate its properties when the
3580 * loop continues.
3581 */
3582 error = export_file_to_sb(fp, i, &rights, efbuf);
3583 if (error != 0 || efbuf->remainder == 0)
3584 break;
3585 }
3586 FILEDESC_SUNLOCK(fdp);
3587 fddrop(fdp);
3588 fail:
3589 free(efbuf, M_TEMP);
3590 return (error);
3591 }
3592
3593 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3594
3595 /*
3596 * Get per-process file descriptors for use by procstat(1), et al.
3597 */
3598 static int
sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)3599 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3600 {
3601 struct sbuf sb;
3602 struct proc *p;
3603 ssize_t maxlen;
3604 int error, error2, *name;
3605
3606 name = (int *)arg1;
3607
3608 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3609 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3610 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3611 if (error != 0) {
3612 sbuf_delete(&sb);
3613 return (error);
3614 }
3615 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3616 error = kern_proc_filedesc_out(p, &sb, maxlen,
3617 KERN_FILEDESC_PACK_KINFO);
3618 error2 = sbuf_finish(&sb);
3619 sbuf_delete(&sb);
3620 return (error != 0 ? error : error2);
3621 }
3622
3623 #ifdef KINFO_OFILE_SIZE
3624 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3625 #endif
3626
3627 #ifdef COMPAT_FREEBSD7
3628 static void
kinfo_to_okinfo(struct kinfo_file * kif,struct kinfo_ofile * okif)3629 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3630 {
3631
3632 okif->kf_structsize = sizeof(*okif);
3633 okif->kf_type = kif->kf_type;
3634 okif->kf_fd = kif->kf_fd;
3635 okif->kf_ref_count = kif->kf_ref_count;
3636 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3637 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3638 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3639 okif->kf_offset = kif->kf_offset;
3640 okif->kf_vnode_type = kif->kf_vnode_type;
3641 okif->kf_sock_domain = kif->kf_sock_domain;
3642 okif->kf_sock_type = kif->kf_sock_type;
3643 okif->kf_sock_protocol = kif->kf_sock_protocol;
3644 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3645 okif->kf_sa_local = kif->kf_sa_local;
3646 okif->kf_sa_peer = kif->kf_sa_peer;
3647 }
3648
3649 static int
export_vnode_for_osysctl(struct vnode * vp,int type,struct kinfo_file * kif,struct kinfo_ofile * okif,struct filedesc * fdp,struct sysctl_req * req)3650 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3651 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3652 {
3653 int error;
3654
3655 vrefact(vp);
3656 FILEDESC_SUNLOCK(fdp);
3657 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3658 kinfo_to_okinfo(kif, okif);
3659 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3660 FILEDESC_SLOCK(fdp);
3661 return (error);
3662 }
3663
3664 /*
3665 * Get per-process file descriptors for use by procstat(1), et al.
3666 */
3667 static int
sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)3668 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3669 {
3670 struct kinfo_ofile *okif;
3671 struct kinfo_file *kif;
3672 struct filedesc *fdp;
3673 int error, i, *name;
3674 struct file *fp;
3675 struct proc *p;
3676
3677 name = (int *)arg1;
3678 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3679 if (error != 0)
3680 return (error);
3681 fdp = fdhold(p);
3682 PROC_UNLOCK(p);
3683 if (fdp == NULL)
3684 return (ENOENT);
3685 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3686 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3687 FILEDESC_SLOCK(fdp);
3688 if (fdp->fd_cdir != NULL)
3689 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3690 okif, fdp, req);
3691 if (fdp->fd_rdir != NULL)
3692 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3693 okif, fdp, req);
3694 if (fdp->fd_jdir != NULL)
3695 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3696 okif, fdp, req);
3697 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3698 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3699 continue;
3700 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3701 KERN_FILEDESC_PACK_KINFO);
3702 FILEDESC_SUNLOCK(fdp);
3703 kinfo_to_okinfo(kif, okif);
3704 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3705 FILEDESC_SLOCK(fdp);
3706 if (error)
3707 break;
3708 }
3709 FILEDESC_SUNLOCK(fdp);
3710 fddrop(fdp);
3711 free(kif, M_TEMP);
3712 free(okif, M_TEMP);
3713 return (0);
3714 }
3715
3716 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3717 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3718 "Process ofiledesc entries");
3719 #endif /* COMPAT_FREEBSD7 */
3720
3721 int
vntype_to_kinfo(int vtype)3722 vntype_to_kinfo(int vtype)
3723 {
3724 struct {
3725 int vtype;
3726 int kf_vtype;
3727 } vtypes_table[] = {
3728 { VBAD, KF_VTYPE_VBAD },
3729 { VBLK, KF_VTYPE_VBLK },
3730 { VCHR, KF_VTYPE_VCHR },
3731 { VDIR, KF_VTYPE_VDIR },
3732 { VFIFO, KF_VTYPE_VFIFO },
3733 { VLNK, KF_VTYPE_VLNK },
3734 { VNON, KF_VTYPE_VNON },
3735 { VREG, KF_VTYPE_VREG },
3736 { VSOCK, KF_VTYPE_VSOCK }
3737 };
3738 unsigned int i;
3739
3740 /*
3741 * Perform vtype translation.
3742 */
3743 for (i = 0; i < nitems(vtypes_table); i++)
3744 if (vtypes_table[i].vtype == vtype)
3745 return (vtypes_table[i].kf_vtype);
3746
3747 return (KF_VTYPE_UNKNOWN);
3748 }
3749
3750 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3751 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3752 "Process filedesc entries");
3753
3754 /*
3755 * Store a process current working directory information to sbuf.
3756 *
3757 * Takes a locked proc as argument, and returns with the proc unlocked.
3758 */
3759 int
kern_proc_cwd_out(struct proc * p,struct sbuf * sb,ssize_t maxlen)3760 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3761 {
3762 struct filedesc *fdp;
3763 struct export_fd_buf *efbuf;
3764 int error;
3765
3766 PROC_LOCK_ASSERT(p, MA_OWNED);
3767
3768 fdp = fdhold(p);
3769 PROC_UNLOCK(p);
3770 if (fdp == NULL)
3771 return (EINVAL);
3772
3773 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3774 efbuf->fdp = fdp;
3775 efbuf->sb = sb;
3776 efbuf->remainder = maxlen;
3777
3778 FILEDESC_SLOCK(fdp);
3779 if (fdp->fd_cdir == NULL)
3780 error = EINVAL;
3781 else {
3782 vrefact(fdp->fd_cdir);
3783 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3784 FREAD, efbuf);
3785 }
3786 FILEDESC_SUNLOCK(fdp);
3787 fddrop(fdp);
3788 free(efbuf, M_TEMP);
3789 return (error);
3790 }
3791
3792 /*
3793 * Get per-process current working directory.
3794 */
3795 static int
sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)3796 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3797 {
3798 struct sbuf sb;
3799 struct proc *p;
3800 ssize_t maxlen;
3801 int error, error2, *name;
3802
3803 name = (int *)arg1;
3804
3805 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3806 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3807 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3808 if (error != 0) {
3809 sbuf_delete(&sb);
3810 return (error);
3811 }
3812 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3813 error = kern_proc_cwd_out(p, &sb, maxlen);
3814 error2 = sbuf_finish(&sb);
3815 sbuf_delete(&sb);
3816 return (error != 0 ? error : error2);
3817 }
3818
3819 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3820 sysctl_kern_proc_cwd, "Process current working directory");
3821
3822 #ifdef DDB
3823 /*
3824 * For the purposes of debugging, generate a human-readable string for the
3825 * file type.
3826 */
3827 static const char *
file_type_to_name(short type)3828 file_type_to_name(short type)
3829 {
3830
3831 switch (type) {
3832 case 0:
3833 return ("zero");
3834 case DTYPE_VNODE:
3835 return ("vnod");
3836 case DTYPE_SOCKET:
3837 return ("sock");
3838 case DTYPE_PIPE:
3839 return ("pipe");
3840 case DTYPE_FIFO:
3841 return ("fifo");
3842 case DTYPE_KQUEUE:
3843 return ("kque");
3844 case DTYPE_CRYPTO:
3845 return ("crpt");
3846 case DTYPE_MQUEUE:
3847 return ("mque");
3848 case DTYPE_SHM:
3849 return ("shm");
3850 case DTYPE_SEM:
3851 return ("ksem");
3852 default:
3853 return ("unkn");
3854 }
3855 }
3856
3857 /*
3858 * For the purposes of debugging, identify a process (if any, perhaps one of
3859 * many) that references the passed file in its file descriptor array. Return
3860 * NULL if none.
3861 */
3862 static struct proc *
file_to_first_proc(struct file * fp)3863 file_to_first_proc(struct file *fp)
3864 {
3865 struct filedesc *fdp;
3866 struct proc *p;
3867 int n;
3868
3869 FOREACH_PROC_IN_SYSTEM(p) {
3870 if (p->p_state == PRS_NEW)
3871 continue;
3872 fdp = p->p_fd;
3873 if (fdp == NULL)
3874 continue;
3875 for (n = 0; n <= fdp->fd_lastfile; n++) {
3876 if (fp == fdp->fd_ofiles[n].fde_file)
3877 return (p);
3878 }
3879 }
3880 return (NULL);
3881 }
3882
3883 static void
db_print_file(struct file * fp,int header)3884 db_print_file(struct file *fp, int header)
3885 {
3886 struct proc *p;
3887
3888 if (header)
3889 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3890 "File", "Type", "Data", "Flag", "GCFl", "Count",
3891 "MCount", "Vnode", "FPID", "FCmd");
3892 p = file_to_first_proc(fp);
3893 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3894 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3895 0, fp->f_count, 0, fp->f_vnode,
3896 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3897 }
3898
DB_SHOW_COMMAND(file,db_show_file)3899 DB_SHOW_COMMAND(file, db_show_file)
3900 {
3901 struct file *fp;
3902
3903 if (!have_addr) {
3904 db_printf("usage: show file <addr>\n");
3905 return;
3906 }
3907 fp = (struct file *)addr;
3908 db_print_file(fp, 1);
3909 }
3910
DB_SHOW_COMMAND(files,db_show_files)3911 DB_SHOW_COMMAND(files, db_show_files)
3912 {
3913 struct filedesc *fdp;
3914 struct file *fp;
3915 struct proc *p;
3916 int header;
3917 int n;
3918
3919 header = 1;
3920 FOREACH_PROC_IN_SYSTEM(p) {
3921 if (p->p_state == PRS_NEW)
3922 continue;
3923 if ((fdp = p->p_fd) == NULL)
3924 continue;
3925 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3926 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3927 continue;
3928 db_print_file(fp, header);
3929 header = 0;
3930 }
3931 }
3932 }
3933 #endif
3934
3935 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3936 &maxfilesperproc, 0, "Maximum files allowed open per process");
3937
3938 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3939 &maxfiles, 0, "Maximum number of files");
3940
3941 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3942 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3943
3944 /* ARGSUSED*/
3945 static void
filelistinit(void * dummy)3946 filelistinit(void *dummy)
3947 {
3948
3949 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3950 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3951 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3952 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3953 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3954 }
3955 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3956
3957 /*-------------------------------------------------------------------*/
3958
3959 static int
badfo_readwrite(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)3960 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3961 int flags, struct thread *td)
3962 {
3963
3964 return (EBADF);
3965 }
3966
3967 static int
badfo_truncate(struct file * fp,off_t length,struct ucred * active_cred,struct thread * td)3968 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3969 struct thread *td)
3970 {
3971
3972 return (EINVAL);
3973 }
3974
3975 static int
badfo_ioctl(struct file * fp,u_long com,void * data,struct ucred * active_cred,struct thread * td)3976 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3977 struct thread *td)
3978 {
3979
3980 return (EBADF);
3981 }
3982
3983 static int
badfo_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)3984 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3985 struct thread *td)
3986 {
3987
3988 return (0);
3989 }
3990
3991 static int
badfo_kqfilter(struct file * fp,struct knote * kn)3992 badfo_kqfilter(struct file *fp, struct knote *kn)
3993 {
3994
3995 return (EBADF);
3996 }
3997
3998 static int
badfo_stat(struct file * fp,struct stat * sb,struct ucred * active_cred,struct thread * td)3999 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4000 struct thread *td)
4001 {
4002
4003 return (EBADF);
4004 }
4005
4006 static int
badfo_close(struct file * fp,struct thread * td)4007 badfo_close(struct file *fp, struct thread *td)
4008 {
4009
4010 return (0);
4011 }
4012
4013 static int
badfo_chmod(struct file * fp,mode_t mode,struct ucred * active_cred,struct thread * td)4014 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4015 struct thread *td)
4016 {
4017
4018 return (EBADF);
4019 }
4020
4021 static int
badfo_chown(struct file * fp,uid_t uid,gid_t gid,struct ucred * active_cred,struct thread * td)4022 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4023 struct thread *td)
4024 {
4025
4026 return (EBADF);
4027 }
4028
4029 static int
badfo_sendfile(struct file * fp,int sockfd,struct uio * hdr_uio,struct uio * trl_uio,off_t offset,size_t nbytes,off_t * sent,int flags,struct thread * td)4030 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4031 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4032 struct thread *td)
4033 {
4034
4035 return (EBADF);
4036 }
4037
4038 static int
badfo_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)4039 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4040 {
4041
4042 return (0);
4043 }
4044
4045 struct fileops badfileops = {
4046 .fo_read = badfo_readwrite,
4047 .fo_write = badfo_readwrite,
4048 .fo_truncate = badfo_truncate,
4049 .fo_ioctl = badfo_ioctl,
4050 .fo_poll = badfo_poll,
4051 .fo_kqfilter = badfo_kqfilter,
4052 .fo_stat = badfo_stat,
4053 .fo_close = badfo_close,
4054 .fo_chmod = badfo_chmod,
4055 .fo_chown = badfo_chown,
4056 .fo_sendfile = badfo_sendfile,
4057 .fo_fill_kinfo = badfo_fill_kinfo,
4058 };
4059
4060 int
invfo_rdwr(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)4061 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4062 int flags, struct thread *td)
4063 {
4064
4065 return (EOPNOTSUPP);
4066 }
4067
4068 int
invfo_truncate(struct file * fp,off_t length,struct ucred * active_cred,struct thread * td)4069 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4070 struct thread *td)
4071 {
4072
4073 return (EINVAL);
4074 }
4075
4076 int
invfo_ioctl(struct file * fp,u_long com,void * data,struct ucred * active_cred,struct thread * td)4077 invfo_ioctl(struct file *fp, u_long com, void *data,
4078 struct ucred *active_cred, struct thread *td)
4079 {
4080
4081 return (ENOTTY);
4082 }
4083
4084 int
invfo_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)4085 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4086 struct thread *td)
4087 {
4088
4089 return (poll_no_poll(events));
4090 }
4091
4092 int
invfo_kqfilter(struct file * fp,struct knote * kn)4093 invfo_kqfilter(struct file *fp, struct knote *kn)
4094 {
4095
4096 return (EINVAL);
4097 }
4098
4099 int
invfo_chmod(struct file * fp,mode_t mode,struct ucred * active_cred,struct thread * td)4100 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4101 struct thread *td)
4102 {
4103
4104 return (EINVAL);
4105 }
4106
4107 int
invfo_chown(struct file * fp,uid_t uid,gid_t gid,struct ucred * active_cred,struct thread * td)4108 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4109 struct thread *td)
4110 {
4111
4112 return (EINVAL);
4113 }
4114
4115 int
invfo_sendfile(struct file * fp,int sockfd,struct uio * hdr_uio,struct uio * trl_uio,off_t offset,size_t nbytes,off_t * sent,int flags,struct thread * td)4116 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4117 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4118 struct thread *td)
4119 {
4120
4121 return (EINVAL);
4122 }
4123
4124 /*-------------------------------------------------------------------*/
4125
4126 /*
4127 * File Descriptor pseudo-device driver (/dev/fd/).
4128 *
4129 * Opening minor device N dup()s the file (if any) connected to file
4130 * descriptor N belonging to the calling process. Note that this driver
4131 * consists of only the ``open()'' routine, because all subsequent
4132 * references to this file will be direct to the other driver.
4133 *
4134 * XXX: we could give this one a cloning event handler if necessary.
4135 */
4136
4137 /* ARGSUSED */
4138 static int
fdopen(struct cdev * dev,int mode,int type,struct thread * td)4139 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4140 {
4141
4142 /*
4143 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4144 * the file descriptor being sought for duplication. The error
4145 * return ensures that the vnode for this device will be released
4146 * by vn_open. Open will detect this special error and take the
4147 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4148 * will simply report the error.
4149 */
4150 td->td_dupfd = dev2unit(dev);
4151 return (ENODEV);
4152 }
4153
4154 static struct cdevsw fildesc_cdevsw = {
4155 .d_version = D_VERSION,
4156 .d_open = fdopen,
4157 .d_name = "FD",
4158 };
4159
4160 static void
fildesc_drvinit(void * unused)4161 fildesc_drvinit(void *unused)
4162 {
4163 struct cdev *dev;
4164
4165 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4166 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4167 make_dev_alias(dev, "stdin");
4168 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4169 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4170 make_dev_alias(dev, "stdout");
4171 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4172 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4173 make_dev_alias(dev, "stderr");
4174 }
4175
4176 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4177