1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36
37 #include <sys/cdefs.h>
38 #include "opt_capsicum.h"
39 #include "opt_ddb.h"
40 #include "opt_ktrace.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44
45 #include <sys/capsicum.h>
46 #include <sys/conf.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
58 #include <sys/namei.h>
59 #include <sys/selinfo.h>
60 #include <sys/poll.h>
61 #include <sys/priv.h>
62 #include <sys/proc.h>
63 #include <sys/protosw.h>
64 #include <sys/racct.h>
65 #include <sys/resourcevar.h>
66 #include <sys/sbuf.h>
67 #include <sys/signalvar.h>
68 #include <sys/kdb.h>
69 #include <sys/smr.h>
70 #include <sys/stat.h>
71 #include <sys/sx.h>
72 #include <sys/syscallsubr.h>
73 #include <sys/sysctl.h>
74 #include <sys/sysproto.h>
75 #include <sys/unistd.h>
76 #include <sys/user.h>
77 #include <sys/vnode.h>
78 #include <sys/ktrace.h>
79
80 #include <net/vnet.h>
81
82 #include <security/audit/audit.h>
83
84 #include <vm/uma.h>
85 #include <vm/vm.h>
86
87 #include <ddb/ddb.h>
88
89 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
90 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
91 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
92 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
93 "file desc to leader structures");
94 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
95 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
96
97 MALLOC_DECLARE(M_FADVISE);
98
99 static __read_mostly uma_zone_t file_zone;
100 static __read_mostly uma_zone_t filedesc0_zone;
101 __read_mostly uma_zone_t pwd_zone;
102 VFS_SMR_DECLARE;
103
104 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
105 struct thread *td, bool holdleaders, bool audit);
106 static void export_file_to_kinfo(struct file *fp, int fd,
107 cap_rights_t *rightsp, struct kinfo_file *kif,
108 struct filedesc *fdp, int flags);
109 static int fd_first_free(struct filedesc *fdp, int low, int size);
110 static void fdgrowtable(struct filedesc *fdp, int nfd);
111 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
112 static void fdunused(struct filedesc *fdp, int fd);
113 static void fdused(struct filedesc *fdp, int fd);
114 static int fget_unlocked_seq(struct thread *td, int fd,
115 cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp);
116 static int getmaxfd(struct thread *td);
117 static u_long *filecaps_copy_prep(const struct filecaps *src);
118 static void filecaps_copy_finish(const struct filecaps *src,
119 struct filecaps *dst, u_long *ioctls);
120 static u_long *filecaps_free_prep(struct filecaps *fcaps);
121 static void filecaps_free_finish(u_long *ioctls);
122
123 static struct pwd *pwd_alloc(void);
124
125 /*
126 * Each process has:
127 *
128 * - An array of open file descriptors (fd_ofiles)
129 * - An array of file flags (fd_ofileflags)
130 * - A bitmap recording which descriptors are in use (fd_map)
131 *
132 * A process starts out with NDFILE descriptors. The value of NDFILE has
133 * been selected based the historical limit of 20 open files, and an
134 * assumption that the majority of processes, especially short-lived
135 * processes like shells, will never need more.
136 *
137 * If this initial allocation is exhausted, a larger descriptor table and
138 * map are allocated dynamically, and the pointers in the process's struct
139 * filedesc are updated to point to those. This is repeated every time
140 * the process runs out of file descriptors (provided it hasn't hit its
141 * resource limit).
142 *
143 * Since threads may hold references to individual descriptor table
144 * entries, the tables are never freed. Instead, they are placed on a
145 * linked list and freed only when the struct filedesc is released.
146 */
147 #define NDFILE 20
148 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
149 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x) ((x) / NDENTRIES)
151 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
153
154 #define FILEDESC_FOREACH_FDE(fdp, _iterator, _fde) \
155 struct filedesc *_fdp = (fdp); \
156 int _lastfile = fdlastfile_single(_fdp); \
157 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \
158 if ((_fde = &_fdp->fd_ofiles[_iterator])->fde_file != NULL)
159
160 #define FILEDESC_FOREACH_FP(fdp, _iterator, _fp) \
161 struct filedesc *_fdp = (fdp); \
162 int _lastfile = fdlastfile_single(_fdp); \
163 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \
164 if ((_fp = _fdp->fd_ofiles[_iterator].fde_file) != NULL)
165
166 /*
167 * SLIST entry used to keep track of ofiles which must be reclaimed when
168 * the process exits.
169 */
170 struct freetable {
171 struct fdescenttbl *ft_table;
172 SLIST_ENTRY(freetable) ft_next;
173 };
174
175 /*
176 * Initial allocation: a filedesc structure + the head of SLIST used to
177 * keep track of old ofiles + enough space for NDFILE descriptors.
178 */
179
180 struct fdescenttbl0 {
181 int fdt_nfiles;
182 struct filedescent fdt_ofiles[NDFILE];
183 };
184
185 struct filedesc0 {
186 struct filedesc fd_fd;
187 SLIST_HEAD(, freetable) fd_free;
188 struct fdescenttbl0 fd_dfiles;
189 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
190 };
191
192 /*
193 * Descriptor management.
194 */
195 static int __exclusive_cache_line openfiles; /* actual number of open files */
196 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
197 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
198
199 /*
200 * If low >= size, just return low. Otherwise find the first zero bit in the
201 * given bitmap, starting at low and not exceeding size - 1. Return size if
202 * not found.
203 */
204 static int
fd_first_free(struct filedesc * fdp,int low,int size)205 fd_first_free(struct filedesc *fdp, int low, int size)
206 {
207 NDSLOTTYPE *map = fdp->fd_map;
208 NDSLOTTYPE mask;
209 int off, maxoff;
210
211 if (low >= size)
212 return (low);
213
214 off = NDSLOT(low);
215 if (low % NDENTRIES) {
216 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
217 if ((mask &= ~map[off]) != 0UL)
218 return (off * NDENTRIES + ffsl(mask) - 1);
219 ++off;
220 }
221 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
222 if (map[off] != ~0UL)
223 return (off * NDENTRIES + ffsl(~map[off]) - 1);
224 return (size);
225 }
226
227 /*
228 * Find the last used fd.
229 *
230 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
231 * Otherwise use fdlastfile.
232 */
233 int
fdlastfile_single(struct filedesc * fdp)234 fdlastfile_single(struct filedesc *fdp)
235 {
236 NDSLOTTYPE *map = fdp->fd_map;
237 int off, minoff;
238
239 off = NDSLOT(fdp->fd_nfiles - 1);
240 for (minoff = NDSLOT(0); off >= minoff; --off)
241 if (map[off] != 0)
242 return (off * NDENTRIES + flsl(map[off]) - 1);
243 return (-1);
244 }
245
246 int
fdlastfile(struct filedesc * fdp)247 fdlastfile(struct filedesc *fdp)
248 {
249
250 FILEDESC_LOCK_ASSERT(fdp);
251 return (fdlastfile_single(fdp));
252 }
253
254 static int
fdisused(struct filedesc * fdp,int fd)255 fdisused(struct filedesc *fdp, int fd)
256 {
257
258 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
259 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
260
261 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
262 }
263
264 /*
265 * Mark a file descriptor as used.
266 */
267 static void
fdused_init(struct filedesc * fdp,int fd)268 fdused_init(struct filedesc *fdp, int fd)
269 {
270
271 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
272
273 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
274 }
275
276 static void
fdused(struct filedesc * fdp,int fd)277 fdused(struct filedesc *fdp, int fd)
278 {
279
280 FILEDESC_XLOCK_ASSERT(fdp);
281
282 fdused_init(fdp, fd);
283 if (fd == fdp->fd_freefile)
284 fdp->fd_freefile++;
285 }
286
287 /*
288 * Mark a file descriptor as unused.
289 */
290 static void
fdunused(struct filedesc * fdp,int fd)291 fdunused(struct filedesc *fdp, int fd)
292 {
293
294 FILEDESC_XLOCK_ASSERT(fdp);
295
296 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
297 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
298 ("fd=%d is still in use", fd));
299
300 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
301 if (fd < fdp->fd_freefile)
302 fdp->fd_freefile = fd;
303 }
304
305 /*
306 * Free a file descriptor.
307 *
308 * Avoid some work if fdp is about to be destroyed.
309 */
310 static inline void
fdefree_last(struct filedescent * fde)311 fdefree_last(struct filedescent *fde)
312 {
313
314 filecaps_free(&fde->fde_caps);
315 }
316
317 static inline void
fdfree(struct filedesc * fdp,int fd)318 fdfree(struct filedesc *fdp, int fd)
319 {
320 struct filedescent *fde;
321
322 FILEDESC_XLOCK_ASSERT(fdp);
323 fde = &fdp->fd_ofiles[fd];
324 #ifdef CAPABILITIES
325 seqc_write_begin(&fde->fde_seqc);
326 #endif
327 fde->fde_file = NULL;
328 #ifdef CAPABILITIES
329 seqc_write_end(&fde->fde_seqc);
330 #endif
331 fdefree_last(fde);
332 fdunused(fdp, fd);
333 }
334
335 /*
336 * System calls on descriptors.
337 */
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
340 int dummy;
341 };
342 #endif
343 /* ARGSUSED */
344 int
sys_getdtablesize(struct thread * td,struct getdtablesize_args * uap)345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
346 {
347 #ifdef RACCT
348 uint64_t lim;
349 #endif
350
351 td->td_retval[0] = getmaxfd(td);
352 #ifdef RACCT
353 PROC_LOCK(td->td_proc);
354 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
355 PROC_UNLOCK(td->td_proc);
356 if (lim < td->td_retval[0])
357 td->td_retval[0] = lim;
358 #endif
359 return (0);
360 }
361
362 /*
363 * Duplicate a file descriptor to a particular value.
364 *
365 * Note: keep in mind that a potential race condition exists when closing
366 * descriptors from a shared descriptor table (via rfork).
367 */
368 #ifndef _SYS_SYSPROTO_H_
369 struct dup2_args {
370 u_int from;
371 u_int to;
372 };
373 #endif
374 /* ARGSUSED */
375 int
sys_dup2(struct thread * td,struct dup2_args * uap)376 sys_dup2(struct thread *td, struct dup2_args *uap)
377 {
378
379 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
380 }
381
382 /*
383 * Duplicate a file descriptor.
384 */
385 #ifndef _SYS_SYSPROTO_H_
386 struct dup_args {
387 u_int fd;
388 };
389 #endif
390 /* ARGSUSED */
391 int
sys_dup(struct thread * td,struct dup_args * uap)392 sys_dup(struct thread *td, struct dup_args *uap)
393 {
394
395 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
396 }
397
398 /*
399 * The file control system call.
400 */
401 #ifndef _SYS_SYSPROTO_H_
402 struct fcntl_args {
403 int fd;
404 int cmd;
405 long arg;
406 };
407 #endif
408 /* ARGSUSED */
409 int
sys_fcntl(struct thread * td,struct fcntl_args * uap)410 sys_fcntl(struct thread *td, struct fcntl_args *uap)
411 {
412
413 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
414 }
415
416 int
kern_fcntl_freebsd(struct thread * td,int fd,int cmd,intptr_t arg)417 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, intptr_t arg)
418 {
419 struct flock fl;
420 struct __oflock ofl;
421 intptr_t arg1;
422 int error, newcmd;
423
424 error = 0;
425 newcmd = cmd;
426 switch (cmd) {
427 case F_OGETLK:
428 case F_OSETLK:
429 case F_OSETLKW:
430 /*
431 * Convert old flock structure to new.
432 */
433 error = copyin((void *)arg, &ofl, sizeof(ofl));
434 fl.l_start = ofl.l_start;
435 fl.l_len = ofl.l_len;
436 fl.l_pid = ofl.l_pid;
437 fl.l_type = ofl.l_type;
438 fl.l_whence = ofl.l_whence;
439 fl.l_sysid = 0;
440
441 switch (cmd) {
442 case F_OGETLK:
443 newcmd = F_GETLK;
444 break;
445 case F_OSETLK:
446 newcmd = F_SETLK;
447 break;
448 case F_OSETLKW:
449 newcmd = F_SETLKW;
450 break;
451 }
452 arg1 = (intptr_t)&fl;
453 break;
454 case F_GETLK:
455 case F_SETLK:
456 case F_SETLKW:
457 case F_SETLK_REMOTE:
458 error = copyin((void *)arg, &fl, sizeof(fl));
459 arg1 = (intptr_t)&fl;
460 break;
461 default:
462 arg1 = arg;
463 break;
464 }
465 if (error)
466 return (error);
467 error = kern_fcntl(td, fd, newcmd, arg1);
468 if (error)
469 return (error);
470 if (cmd == F_OGETLK) {
471 ofl.l_start = fl.l_start;
472 ofl.l_len = fl.l_len;
473 ofl.l_pid = fl.l_pid;
474 ofl.l_type = fl.l_type;
475 ofl.l_whence = fl.l_whence;
476 error = copyout(&ofl, (void *)arg, sizeof(ofl));
477 } else if (cmd == F_GETLK) {
478 error = copyout(&fl, (void *)arg, sizeof(fl));
479 }
480 return (error);
481 }
482
483 int
kern_fcntl(struct thread * td,int fd,int cmd,intptr_t arg)484 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
485 {
486 struct filedesc *fdp;
487 struct flock *flp;
488 struct file *fp, *fp2;
489 struct filedescent *fde;
490 struct proc *p;
491 struct vnode *vp;
492 struct mount *mp;
493 struct kinfo_file *kif;
494 int error, flg, kif_sz, seals, tmp, got_set, got_cleared;
495 uint64_t bsize;
496 off_t foffset;
497
498 error = 0;
499 flg = F_POSIX;
500 p = td->td_proc;
501 fdp = p->p_fd;
502
503 AUDIT_ARG_FD(cmd);
504 AUDIT_ARG_CMD(cmd);
505 switch (cmd) {
506 case F_DUPFD:
507 tmp = arg;
508 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
509 break;
510
511 case F_DUPFD_CLOEXEC:
512 tmp = arg;
513 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
514 break;
515
516 case F_DUP2FD:
517 tmp = arg;
518 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
519 break;
520
521 case F_DUP2FD_CLOEXEC:
522 tmp = arg;
523 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
524 break;
525
526 case F_GETFD:
527 error = EBADF;
528 FILEDESC_SLOCK(fdp);
529 fde = fdeget_noref(fdp, fd);
530 if (fde != NULL) {
531 td->td_retval[0] =
532 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
533 error = 0;
534 }
535 FILEDESC_SUNLOCK(fdp);
536 break;
537
538 case F_SETFD:
539 error = EBADF;
540 FILEDESC_XLOCK(fdp);
541 fde = fdeget_noref(fdp, fd);
542 if (fde != NULL) {
543 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
544 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
545 error = 0;
546 }
547 FILEDESC_XUNLOCK(fdp);
548 break;
549
550 case F_GETFL:
551 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
552 if (error != 0)
553 break;
554 td->td_retval[0] = OFLAGS(fp->f_flag);
555 fdrop(fp, td);
556 break;
557
558 case F_SETFL:
559 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
560 if (error != 0)
561 break;
562 if (fp->f_ops == &path_fileops) {
563 fdrop(fp, td);
564 error = EBADF;
565 break;
566 }
567 do {
568 tmp = flg = fp->f_flag;
569 tmp &= ~FCNTLFLAGS;
570 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
571 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
572 got_set = tmp & ~flg;
573 got_cleared = flg & ~tmp;
574 tmp = fp->f_flag & FNONBLOCK;
575 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
576 if (error != 0)
577 goto revert_f_setfl;
578 tmp = fp->f_flag & FASYNC;
579 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
580 if (error == 0) {
581 fdrop(fp, td);
582 break;
583 }
584 atomic_clear_int(&fp->f_flag, FNONBLOCK);
585 tmp = 0;
586 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
587 revert_f_setfl:
588 do {
589 tmp = flg = fp->f_flag;
590 tmp &= ~FCNTLFLAGS;
591 tmp |= got_cleared;
592 tmp &= ~got_set;
593 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
594 fdrop(fp, td);
595 break;
596
597 case F_GETOWN:
598 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
599 if (error != 0)
600 break;
601 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
602 if (error == 0)
603 td->td_retval[0] = tmp;
604 fdrop(fp, td);
605 break;
606
607 case F_SETOWN:
608 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
609 if (error != 0)
610 break;
611 tmp = arg;
612 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
613 fdrop(fp, td);
614 break;
615
616 case F_SETLK_REMOTE:
617 error = priv_check(td, PRIV_NFS_LOCKD);
618 if (error != 0)
619 return (error);
620 flg = F_REMOTE;
621 goto do_setlk;
622
623 case F_SETLKW:
624 flg |= F_WAIT;
625 /* FALLTHROUGH F_SETLK */
626
627 case F_SETLK:
628 do_setlk:
629 flp = (struct flock *)arg;
630 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
631 error = EINVAL;
632 break;
633 }
634
635 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
636 if (error != 0)
637 break;
638 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
639 error = EBADF;
640 fdrop(fp, td);
641 break;
642 }
643
644 if (flp->l_whence == SEEK_CUR) {
645 foffset = foffset_get(fp);
646 if (foffset < 0 ||
647 (flp->l_start > 0 &&
648 foffset > OFF_MAX - flp->l_start)) {
649 error = EOVERFLOW;
650 fdrop(fp, td);
651 break;
652 }
653 flp->l_start += foffset;
654 }
655
656 vp = fp->f_vnode;
657 switch (flp->l_type) {
658 case F_RDLCK:
659 if ((fp->f_flag & FREAD) == 0) {
660 error = EBADF;
661 break;
662 }
663 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
664 PROC_LOCK(p->p_leader);
665 p->p_leader->p_flag |= P_ADVLOCK;
666 PROC_UNLOCK(p->p_leader);
667 }
668 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
669 flp, flg);
670 break;
671 case F_WRLCK:
672 if ((fp->f_flag & FWRITE) == 0) {
673 error = EBADF;
674 break;
675 }
676 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
677 PROC_LOCK(p->p_leader);
678 p->p_leader->p_flag |= P_ADVLOCK;
679 PROC_UNLOCK(p->p_leader);
680 }
681 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
682 flp, flg);
683 break;
684 case F_UNLCK:
685 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
686 flp, flg);
687 break;
688 case F_UNLCKSYS:
689 if (flg != F_REMOTE) {
690 error = EINVAL;
691 break;
692 }
693 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
694 F_UNLCKSYS, flp, flg);
695 break;
696 default:
697 error = EINVAL;
698 break;
699 }
700 if (error != 0 || flp->l_type == F_UNLCK ||
701 flp->l_type == F_UNLCKSYS) {
702 fdrop(fp, td);
703 break;
704 }
705
706 /*
707 * Check for a race with close.
708 *
709 * The vnode is now advisory locked (or unlocked, but this case
710 * is not really important) as the caller requested.
711 * We had to drop the filedesc lock, so we need to recheck if
712 * the descriptor is still valid, because if it was closed
713 * in the meantime we need to remove advisory lock from the
714 * vnode - close on any descriptor leading to an advisory
715 * locked vnode, removes that lock.
716 * We will return 0 on purpose in that case, as the result of
717 * successful advisory lock might have been externally visible
718 * already. This is fine - effectively we pretend to the caller
719 * that the closing thread was a bit slower and that the
720 * advisory lock succeeded before the close.
721 */
722 error = fget_unlocked(td, fd, &cap_no_rights, &fp2);
723 if (error != 0) {
724 fdrop(fp, td);
725 break;
726 }
727 if (fp != fp2) {
728 flp->l_whence = SEEK_SET;
729 flp->l_start = 0;
730 flp->l_len = 0;
731 flp->l_type = F_UNLCK;
732 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
733 F_UNLCK, flp, F_POSIX);
734 }
735 fdrop(fp, td);
736 fdrop(fp2, td);
737 break;
738
739 case F_GETLK:
740 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
741 if (error != 0)
742 break;
743 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
744 error = EBADF;
745 fdrop(fp, td);
746 break;
747 }
748 flp = (struct flock *)arg;
749 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
750 flp->l_type != F_UNLCK) {
751 error = EINVAL;
752 fdrop(fp, td);
753 break;
754 }
755 if (flp->l_whence == SEEK_CUR) {
756 foffset = foffset_get(fp);
757 if ((flp->l_start > 0 &&
758 foffset > OFF_MAX - flp->l_start) ||
759 (flp->l_start < 0 &&
760 foffset < OFF_MIN - flp->l_start)) {
761 error = EOVERFLOW;
762 fdrop(fp, td);
763 break;
764 }
765 flp->l_start += foffset;
766 }
767 vp = fp->f_vnode;
768 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
769 F_POSIX);
770 fdrop(fp, td);
771 break;
772
773 case F_ADD_SEALS:
774 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
775 if (error != 0)
776 break;
777 error = fo_add_seals(fp, arg);
778 fdrop(fp, td);
779 break;
780
781 case F_GET_SEALS:
782 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
783 if (error != 0)
784 break;
785 if (fo_get_seals(fp, &seals) == 0)
786 td->td_retval[0] = seals;
787 else
788 error = EINVAL;
789 fdrop(fp, td);
790 break;
791
792 case F_RDAHEAD:
793 arg = arg ? 128 * 1024: 0;
794 /* FALLTHROUGH */
795 case F_READAHEAD:
796 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
797 if (error != 0)
798 break;
799 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
800 fdrop(fp, td);
801 error = EBADF;
802 break;
803 }
804 vp = fp->f_vnode;
805 if (vp->v_type != VREG) {
806 fdrop(fp, td);
807 error = ENOTTY;
808 break;
809 }
810
811 /*
812 * Exclusive lock synchronizes against f_seqcount reads and
813 * writes in sequential_heuristic().
814 */
815 error = vn_lock(vp, LK_EXCLUSIVE);
816 if (error != 0) {
817 fdrop(fp, td);
818 break;
819 }
820 if (arg >= 0) {
821 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
822 arg = MIN(arg, INT_MAX - bsize + 1);
823 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
824 (arg + bsize - 1) / bsize);
825 atomic_set_int(&fp->f_flag, FRDAHEAD);
826 } else {
827 atomic_clear_int(&fp->f_flag, FRDAHEAD);
828 }
829 VOP_UNLOCK(vp);
830 fdrop(fp, td);
831 break;
832
833 case F_ISUNIONSTACK:
834 /*
835 * Check if the vnode is part of a union stack (either the
836 * "union" flag from mount(2) or unionfs).
837 *
838 * Prior to introduction of this op libc's readdir would call
839 * fstatfs(2), in effect unnecessarily copying kilobytes of
840 * data just to check fs name and a mount flag.
841 *
842 * Fixing the code to handle everything in the kernel instead
843 * is a non-trivial endeavor and has low priority, thus this
844 * horrible kludge facilitates the current behavior in a much
845 * cheaper manner until someone(tm) sorts this out.
846 */
847 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
848 if (error != 0)
849 break;
850 if (fp->f_type != DTYPE_VNODE) {
851 fdrop(fp, td);
852 error = EBADF;
853 break;
854 }
855 vp = fp->f_vnode;
856 /*
857 * Since we don't prevent dooming the vnode even non-null mp
858 * found can become immediately stale. This is tolerable since
859 * mount points are type-stable (providing safe memory access)
860 * and any vfs op on this vnode going forward will return an
861 * error (meaning return value in this case is meaningless).
862 */
863 mp = atomic_load_ptr(&vp->v_mount);
864 if (__predict_false(mp == NULL)) {
865 fdrop(fp, td);
866 error = EBADF;
867 break;
868 }
869 td->td_retval[0] = 0;
870 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
871 mp->mnt_flag & MNT_UNION)
872 td->td_retval[0] = 1;
873 fdrop(fp, td);
874 break;
875
876 case F_KINFO:
877 #ifdef CAPABILITY_MODE
878 if (CAP_TRACING(td))
879 ktrcapfail(CAPFAIL_SYSCALL, &cmd);
880 if (IN_CAPABILITY_MODE(td)) {
881 error = ECAPMODE;
882 break;
883 }
884 #endif
885 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
886 if (error != 0)
887 break;
888 if (kif_sz != sizeof(*kif)) {
889 error = EINVAL;
890 break;
891 }
892 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
893 FILEDESC_SLOCK(fdp);
894 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL);
895 if (error == 0 && fhold(fp)) {
896 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
897 FILEDESC_SUNLOCK(fdp);
898 fdrop(fp, td);
899 if ((kif->kf_status & KF_ATTR_VALID) != 0) {
900 kif->kf_structsize = sizeof(*kif);
901 error = copyout(kif, (void *)arg, sizeof(*kif));
902 } else {
903 error = EBADF;
904 }
905 } else {
906 FILEDESC_SUNLOCK(fdp);
907 if (error == 0)
908 error = EBADF;
909 }
910 free(kif, M_TEMP);
911 break;
912
913 default:
914 error = EINVAL;
915 break;
916 }
917 return (error);
918 }
919
920 static int
getmaxfd(struct thread * td)921 getmaxfd(struct thread *td)
922 {
923
924 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
925 }
926
927 /*
928 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
929 */
930 int
kern_dup(struct thread * td,u_int mode,int flags,int old,int new)931 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
932 {
933 struct filedesc *fdp;
934 struct filedescent *oldfde, *newfde;
935 struct proc *p;
936 struct file *delfp, *oldfp;
937 u_long *oioctls, *nioctls;
938 int error, maxfd;
939
940 p = td->td_proc;
941 fdp = p->p_fd;
942 oioctls = NULL;
943
944 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
945 MPASS(mode < FDDUP_LASTMODE);
946
947 AUDIT_ARG_FD(old);
948 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
949
950 /*
951 * Verify we have a valid descriptor to dup from and possibly to
952 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
953 * return EINVAL when the new descriptor is out of bounds.
954 */
955 if (old < 0)
956 return (EBADF);
957 if (new < 0)
958 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
959 maxfd = getmaxfd(td);
960 if (new >= maxfd)
961 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
962
963 error = EBADF;
964 FILEDESC_XLOCK(fdp);
965 if (fget_noref(fdp, old) == NULL)
966 goto unlock;
967 if (mode == FDDUP_FIXED && old == new) {
968 td->td_retval[0] = new;
969 if (flags & FDDUP_FLAG_CLOEXEC)
970 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
971 error = 0;
972 goto unlock;
973 }
974
975 oldfde = &fdp->fd_ofiles[old];
976 oldfp = oldfde->fde_file;
977 if (!fhold(oldfp))
978 goto unlock;
979
980 /*
981 * If the caller specified a file descriptor, make sure the file
982 * table is large enough to hold it, and grab it. Otherwise, just
983 * allocate a new descriptor the usual way.
984 */
985 switch (mode) {
986 case FDDUP_NORMAL:
987 case FDDUP_FCNTL:
988 if ((error = fdalloc(td, new, &new)) != 0) {
989 fdrop(oldfp, td);
990 goto unlock;
991 }
992 break;
993 case FDDUP_FIXED:
994 if (new >= fdp->fd_nfiles) {
995 /*
996 * The resource limits are here instead of e.g.
997 * fdalloc(), because the file descriptor table may be
998 * shared between processes, so we can't really use
999 * racct_add()/racct_sub(). Instead of counting the
1000 * number of actually allocated descriptors, just put
1001 * the limit on the size of the file descriptor table.
1002 */
1003 #ifdef RACCT
1004 if (RACCT_ENABLED()) {
1005 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
1006 if (error != 0) {
1007 error = EMFILE;
1008 fdrop(oldfp, td);
1009 goto unlock;
1010 }
1011 }
1012 #endif
1013 fdgrowtable_exp(fdp, new + 1);
1014 }
1015 if (!fdisused(fdp, new))
1016 fdused(fdp, new);
1017 break;
1018 default:
1019 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1020 }
1021
1022 KASSERT(old != new, ("new fd is same as old"));
1023
1024 /* Refetch oldfde because the table may have grown and old one freed. */
1025 oldfde = &fdp->fd_ofiles[old];
1026 KASSERT(oldfp == oldfde->fde_file,
1027 ("fdt_ofiles shift from growth observed at fd %d",
1028 old));
1029
1030 newfde = &fdp->fd_ofiles[new];
1031 delfp = newfde->fde_file;
1032
1033 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1034
1035 /*
1036 * Duplicate the source descriptor.
1037 */
1038 #ifdef CAPABILITIES
1039 seqc_write_begin(&newfde->fde_seqc);
1040 #endif
1041 oioctls = filecaps_free_prep(&newfde->fde_caps);
1042 fde_copy(oldfde, newfde);
1043 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1044 nioctls);
1045 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1046 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1047 else
1048 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1049 #ifdef CAPABILITIES
1050 seqc_write_end(&newfde->fde_seqc);
1051 #endif
1052 td->td_retval[0] = new;
1053
1054 error = 0;
1055
1056 if (delfp != NULL) {
1057 (void) closefp(fdp, new, delfp, td, true, false);
1058 FILEDESC_UNLOCK_ASSERT(fdp);
1059 } else {
1060 unlock:
1061 FILEDESC_XUNLOCK(fdp);
1062 }
1063
1064 filecaps_free_finish(oioctls);
1065 return (error);
1066 }
1067
1068 static void
sigiofree(struct sigio * sigio)1069 sigiofree(struct sigio *sigio)
1070 {
1071 crfree(sigio->sio_ucred);
1072 free(sigio, M_SIGIO);
1073 }
1074
1075 static struct sigio *
funsetown_locked(struct sigio * sigio)1076 funsetown_locked(struct sigio *sigio)
1077 {
1078 struct proc *p;
1079 struct pgrp *pg;
1080
1081 SIGIO_ASSERT_LOCKED();
1082
1083 if (sigio == NULL)
1084 return (NULL);
1085 *sigio->sio_myref = NULL;
1086 if (sigio->sio_pgid < 0) {
1087 pg = sigio->sio_pgrp;
1088 PGRP_LOCK(pg);
1089 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1090 PGRP_UNLOCK(pg);
1091 } else {
1092 p = sigio->sio_proc;
1093 PROC_LOCK(p);
1094 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1095 PROC_UNLOCK(p);
1096 }
1097 return (sigio);
1098 }
1099
1100 /*
1101 * If sigio is on the list associated with a process or process group,
1102 * disable signalling from the device, remove sigio from the list and
1103 * free sigio.
1104 */
1105 void
funsetown(struct sigio ** sigiop)1106 funsetown(struct sigio **sigiop)
1107 {
1108 struct sigio *sigio;
1109
1110 /* Racy check, consumers must provide synchronization. */
1111 if (*sigiop == NULL)
1112 return;
1113
1114 SIGIO_LOCK();
1115 sigio = funsetown_locked(*sigiop);
1116 SIGIO_UNLOCK();
1117 if (sigio != NULL)
1118 sigiofree(sigio);
1119 }
1120
1121 /*
1122 * Free a list of sigio structures. The caller must ensure that new sigio
1123 * structures cannot be added after this point. For process groups this is
1124 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1125 * as an interlock.
1126 */
1127 void
funsetownlst(struct sigiolst * sigiolst)1128 funsetownlst(struct sigiolst *sigiolst)
1129 {
1130 struct proc *p;
1131 struct pgrp *pg;
1132 struct sigio *sigio, *tmp;
1133
1134 /* Racy check. */
1135 sigio = SLIST_FIRST(sigiolst);
1136 if (sigio == NULL)
1137 return;
1138
1139 p = NULL;
1140 pg = NULL;
1141
1142 SIGIO_LOCK();
1143 sigio = SLIST_FIRST(sigiolst);
1144 if (sigio == NULL) {
1145 SIGIO_UNLOCK();
1146 return;
1147 }
1148
1149 /*
1150 * Every entry of the list should belong to a single proc or pgrp.
1151 */
1152 if (sigio->sio_pgid < 0) {
1153 pg = sigio->sio_pgrp;
1154 sx_assert(&proctree_lock, SX_XLOCKED);
1155 PGRP_LOCK(pg);
1156 } else /* if (sigio->sio_pgid > 0) */ {
1157 p = sigio->sio_proc;
1158 PROC_LOCK(p);
1159 KASSERT((p->p_flag & P_WEXIT) != 0,
1160 ("%s: process %p is not exiting", __func__, p));
1161 }
1162
1163 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1164 *sigio->sio_myref = NULL;
1165 if (pg != NULL) {
1166 KASSERT(sigio->sio_pgid < 0,
1167 ("Proc sigio in pgrp sigio list"));
1168 KASSERT(sigio->sio_pgrp == pg,
1169 ("Bogus pgrp in sigio list"));
1170 } else /* if (p != NULL) */ {
1171 KASSERT(sigio->sio_pgid > 0,
1172 ("Pgrp sigio in proc sigio list"));
1173 KASSERT(sigio->sio_proc == p,
1174 ("Bogus proc in sigio list"));
1175 }
1176 }
1177
1178 if (pg != NULL)
1179 PGRP_UNLOCK(pg);
1180 else
1181 PROC_UNLOCK(p);
1182 SIGIO_UNLOCK();
1183
1184 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1185 sigiofree(sigio);
1186 }
1187
1188 /*
1189 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1190 *
1191 * After permission checking, add a sigio structure to the sigio list for
1192 * the process or process group.
1193 */
1194 int
fsetown(pid_t pgid,struct sigio ** sigiop)1195 fsetown(pid_t pgid, struct sigio **sigiop)
1196 {
1197 struct proc *proc;
1198 struct pgrp *pgrp;
1199 struct sigio *osigio, *sigio;
1200 int ret;
1201
1202 if (pgid == 0) {
1203 funsetown(sigiop);
1204 return (0);
1205 }
1206
1207 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1208 sigio->sio_pgid = pgid;
1209 sigio->sio_ucred = crhold(curthread->td_ucred);
1210 sigio->sio_myref = sigiop;
1211
1212 ret = 0;
1213 if (pgid > 0) {
1214 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1215 SIGIO_LOCK();
1216 osigio = funsetown_locked(*sigiop);
1217 if (ret == 0) {
1218 PROC_LOCK(proc);
1219 _PRELE(proc);
1220 if ((proc->p_flag & P_WEXIT) != 0) {
1221 ret = ESRCH;
1222 } else if (proc->p_session !=
1223 curthread->td_proc->p_session) {
1224 /*
1225 * Policy - Don't allow a process to FSETOWN a
1226 * process in another session.
1227 *
1228 * Remove this test to allow maximum flexibility
1229 * or restrict FSETOWN to the current process or
1230 * process group for maximum safety.
1231 */
1232 ret = EPERM;
1233 } else {
1234 sigio->sio_proc = proc;
1235 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1236 sio_pgsigio);
1237 }
1238 PROC_UNLOCK(proc);
1239 }
1240 } else /* if (pgid < 0) */ {
1241 sx_slock(&proctree_lock);
1242 SIGIO_LOCK();
1243 osigio = funsetown_locked(*sigiop);
1244 pgrp = pgfind(-pgid);
1245 if (pgrp == NULL) {
1246 ret = ESRCH;
1247 } else {
1248 if (pgrp->pg_session != curthread->td_proc->p_session) {
1249 /*
1250 * Policy - Don't allow a process to FSETOWN a
1251 * process in another session.
1252 *
1253 * Remove this test to allow maximum flexibility
1254 * or restrict FSETOWN to the current process or
1255 * process group for maximum safety.
1256 */
1257 ret = EPERM;
1258 } else {
1259 sigio->sio_pgrp = pgrp;
1260 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1261 sio_pgsigio);
1262 }
1263 PGRP_UNLOCK(pgrp);
1264 }
1265 sx_sunlock(&proctree_lock);
1266 }
1267 if (ret == 0)
1268 *sigiop = sigio;
1269 SIGIO_UNLOCK();
1270 if (osigio != NULL)
1271 sigiofree(osigio);
1272 return (ret);
1273 }
1274
1275 /*
1276 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1277 */
1278 pid_t
fgetown(struct sigio ** sigiop)1279 fgetown(struct sigio **sigiop)
1280 {
1281 pid_t pgid;
1282
1283 SIGIO_LOCK();
1284 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1285 SIGIO_UNLOCK();
1286 return (pgid);
1287 }
1288
1289 static int
closefp_impl(struct filedesc * fdp,int fd,struct file * fp,struct thread * td,bool audit)1290 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1291 bool audit)
1292 {
1293 int error;
1294
1295 FILEDESC_XLOCK_ASSERT(fdp);
1296
1297 /*
1298 * We now hold the fp reference that used to be owned by the
1299 * descriptor array. We have to unlock the FILEDESC *AFTER*
1300 * knote_fdclose to prevent a race of the fd getting opened, a knote
1301 * added, and deleteing a knote for the new fd.
1302 */
1303 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1304 knote_fdclose(td, fd);
1305
1306 /*
1307 * We need to notify mqueue if the object is of type mqueue.
1308 */
1309 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1310 mq_fdclose(td, fd, fp);
1311 FILEDESC_XUNLOCK(fdp);
1312
1313 #ifdef AUDIT
1314 if (AUDITING_TD(td) && audit)
1315 audit_sysclose(td, fd, fp);
1316 #endif
1317 error = closef(fp, td);
1318
1319 /*
1320 * All paths leading up to closefp() will have already removed or
1321 * replaced the fd in the filedesc table, so a restart would not
1322 * operate on the same file.
1323 */
1324 if (error == ERESTART)
1325 error = EINTR;
1326
1327 return (error);
1328 }
1329
1330 static int
closefp_hl(struct filedesc * fdp,int fd,struct file * fp,struct thread * td,bool holdleaders,bool audit)1331 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1332 bool holdleaders, bool audit)
1333 {
1334 int error;
1335
1336 FILEDESC_XLOCK_ASSERT(fdp);
1337
1338 if (holdleaders) {
1339 if (td->td_proc->p_fdtol != NULL) {
1340 /*
1341 * Ask fdfree() to sleep to ensure that all relevant
1342 * process leaders can be traversed in closef().
1343 */
1344 fdp->fd_holdleaderscount++;
1345 } else {
1346 holdleaders = false;
1347 }
1348 }
1349
1350 error = closefp_impl(fdp, fd, fp, td, audit);
1351 if (holdleaders) {
1352 FILEDESC_XLOCK(fdp);
1353 fdp->fd_holdleaderscount--;
1354 if (fdp->fd_holdleaderscount == 0 &&
1355 fdp->fd_holdleaderswakeup != 0) {
1356 fdp->fd_holdleaderswakeup = 0;
1357 wakeup(&fdp->fd_holdleaderscount);
1358 }
1359 FILEDESC_XUNLOCK(fdp);
1360 }
1361 return (error);
1362 }
1363
1364 static int
closefp(struct filedesc * fdp,int fd,struct file * fp,struct thread * td,bool holdleaders,bool audit)1365 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1366 bool holdleaders, bool audit)
1367 {
1368
1369 FILEDESC_XLOCK_ASSERT(fdp);
1370
1371 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1372 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1373 } else {
1374 return (closefp_impl(fdp, fd, fp, td, audit));
1375 }
1376 }
1377
1378 /*
1379 * Close a file descriptor.
1380 */
1381 #ifndef _SYS_SYSPROTO_H_
1382 struct close_args {
1383 int fd;
1384 };
1385 #endif
1386 /* ARGSUSED */
1387 int
sys_close(struct thread * td,struct close_args * uap)1388 sys_close(struct thread *td, struct close_args *uap)
1389 {
1390
1391 return (kern_close(td, uap->fd));
1392 }
1393
1394 int
kern_close(struct thread * td,int fd)1395 kern_close(struct thread *td, int fd)
1396 {
1397 struct filedesc *fdp;
1398 struct file *fp;
1399
1400 fdp = td->td_proc->p_fd;
1401
1402 FILEDESC_XLOCK(fdp);
1403 if ((fp = fget_noref(fdp, fd)) == NULL) {
1404 FILEDESC_XUNLOCK(fdp);
1405 return (EBADF);
1406 }
1407 fdfree(fdp, fd);
1408
1409 /* closefp() drops the FILEDESC lock for us. */
1410 return (closefp(fdp, fd, fp, td, true, true));
1411 }
1412
1413 static int
close_range_cloexec(struct thread * td,u_int lowfd,u_int highfd)1414 close_range_cloexec(struct thread *td, u_int lowfd, u_int highfd)
1415 {
1416 struct filedesc *fdp;
1417 struct fdescenttbl *fdt;
1418 struct filedescent *fde;
1419 int fd;
1420
1421 fdp = td->td_proc->p_fd;
1422 FILEDESC_XLOCK(fdp);
1423 fdt = atomic_load_ptr(&fdp->fd_files);
1424 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1425 fd = lowfd;
1426 if (__predict_false(fd > highfd)) {
1427 goto out_locked;
1428 }
1429 for (; fd <= highfd; fd++) {
1430 fde = &fdt->fdt_ofiles[fd];
1431 if (fde->fde_file != NULL)
1432 fde->fde_flags |= UF_EXCLOSE;
1433 }
1434 out_locked:
1435 FILEDESC_XUNLOCK(fdp);
1436 return (0);
1437 }
1438
1439 static int
close_range_impl(struct thread * td,u_int lowfd,u_int highfd)1440 close_range_impl(struct thread *td, u_int lowfd, u_int highfd)
1441 {
1442 struct filedesc *fdp;
1443 const struct fdescenttbl *fdt;
1444 struct file *fp;
1445 int fd;
1446
1447 fdp = td->td_proc->p_fd;
1448 FILEDESC_XLOCK(fdp);
1449 fdt = atomic_load_ptr(&fdp->fd_files);
1450 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1451 fd = lowfd;
1452 if (__predict_false(fd > highfd)) {
1453 goto out_locked;
1454 }
1455 for (;;) {
1456 fp = fdt->fdt_ofiles[fd].fde_file;
1457 if (fp == NULL) {
1458 if (fd == highfd)
1459 goto out_locked;
1460 } else {
1461 fdfree(fdp, fd);
1462 (void) closefp(fdp, fd, fp, td, true, true);
1463 if (fd == highfd)
1464 goto out_unlocked;
1465 FILEDESC_XLOCK(fdp);
1466 fdt = atomic_load_ptr(&fdp->fd_files);
1467 }
1468 fd++;
1469 }
1470 out_locked:
1471 FILEDESC_XUNLOCK(fdp);
1472 out_unlocked:
1473 return (0);
1474 }
1475
1476 int
kern_close_range(struct thread * td,int flags,u_int lowfd,u_int highfd)1477 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd)
1478 {
1479
1480 /*
1481 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1482 * open should not be a usage error. From a close_range() perspective,
1483 * close_range(3, ~0U, 0) in the same scenario should also likely not
1484 * be a usage error as all fd above 3 are in-fact already closed.
1485 */
1486 if (highfd < lowfd) {
1487 return (EINVAL);
1488 }
1489
1490 if ((flags & CLOSE_RANGE_CLOEXEC) != 0)
1491 return (close_range_cloexec(td, lowfd, highfd));
1492
1493 return (close_range_impl(td, lowfd, highfd));
1494 }
1495
1496 #ifndef _SYS_SYSPROTO_H_
1497 struct close_range_args {
1498 u_int lowfd;
1499 u_int highfd;
1500 int flags;
1501 };
1502 #endif
1503 int
sys_close_range(struct thread * td,struct close_range_args * uap)1504 sys_close_range(struct thread *td, struct close_range_args *uap)
1505 {
1506
1507 AUDIT_ARG_FD(uap->lowfd);
1508 AUDIT_ARG_CMD(uap->highfd);
1509 AUDIT_ARG_FFLAGS(uap->flags);
1510
1511 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC)) != 0)
1512 return (EINVAL);
1513 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd));
1514 }
1515
1516 #ifdef COMPAT_FREEBSD12
1517 /*
1518 * Close open file descriptors.
1519 */
1520 #ifndef _SYS_SYSPROTO_H_
1521 struct freebsd12_closefrom_args {
1522 int lowfd;
1523 };
1524 #endif
1525 /* ARGSUSED */
1526 int
freebsd12_closefrom(struct thread * td,struct freebsd12_closefrom_args * uap)1527 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1528 {
1529 u_int lowfd;
1530
1531 AUDIT_ARG_FD(uap->lowfd);
1532
1533 /*
1534 * Treat negative starting file descriptor values identical to
1535 * closefrom(0) which closes all files.
1536 */
1537 lowfd = MAX(0, uap->lowfd);
1538 return (kern_close_range(td, 0, lowfd, ~0U));
1539 }
1540 #endif /* COMPAT_FREEBSD12 */
1541
1542 #if defined(COMPAT_43)
1543 /*
1544 * Return status information about a file descriptor.
1545 */
1546 #ifndef _SYS_SYSPROTO_H_
1547 struct ofstat_args {
1548 int fd;
1549 struct ostat *sb;
1550 };
1551 #endif
1552 /* ARGSUSED */
1553 int
ofstat(struct thread * td,struct ofstat_args * uap)1554 ofstat(struct thread *td, struct ofstat_args *uap)
1555 {
1556 struct ostat oub;
1557 struct stat ub;
1558 int error;
1559
1560 error = kern_fstat(td, uap->fd, &ub);
1561 if (error == 0) {
1562 cvtstat(&ub, &oub);
1563 error = copyout(&oub, uap->sb, sizeof(oub));
1564 }
1565 return (error);
1566 }
1567 #endif /* COMPAT_43 */
1568
1569 #if defined(COMPAT_FREEBSD11)
1570 int
freebsd11_fstat(struct thread * td,struct freebsd11_fstat_args * uap)1571 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1572 {
1573 struct stat sb;
1574 struct freebsd11_stat osb;
1575 int error;
1576
1577 error = kern_fstat(td, uap->fd, &sb);
1578 if (error != 0)
1579 return (error);
1580 error = freebsd11_cvtstat(&sb, &osb);
1581 if (error == 0)
1582 error = copyout(&osb, uap->sb, sizeof(osb));
1583 return (error);
1584 }
1585 #endif /* COMPAT_FREEBSD11 */
1586
1587 /*
1588 * Return status information about a file descriptor.
1589 */
1590 #ifndef _SYS_SYSPROTO_H_
1591 struct fstat_args {
1592 int fd;
1593 struct stat *sb;
1594 };
1595 #endif
1596 /* ARGSUSED */
1597 int
sys_fstat(struct thread * td,struct fstat_args * uap)1598 sys_fstat(struct thread *td, struct fstat_args *uap)
1599 {
1600 struct stat ub;
1601 int error;
1602
1603 error = kern_fstat(td, uap->fd, &ub);
1604 if (error == 0)
1605 error = copyout(&ub, uap->sb, sizeof(ub));
1606 return (error);
1607 }
1608
1609 int
kern_fstat(struct thread * td,int fd,struct stat * sbp)1610 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1611 {
1612 struct file *fp;
1613 int error;
1614
1615 AUDIT_ARG_FD(fd);
1616
1617 error = fget(td, fd, &cap_fstat_rights, &fp);
1618 if (__predict_false(error != 0))
1619 return (error);
1620
1621 AUDIT_ARG_FILE(td->td_proc, fp);
1622
1623 error = fo_stat(fp, sbp, td->td_ucred);
1624 fdrop(fp, td);
1625 #ifdef __STAT_TIME_T_EXT
1626 sbp->st_atim_ext = 0;
1627 sbp->st_mtim_ext = 0;
1628 sbp->st_ctim_ext = 0;
1629 sbp->st_btim_ext = 0;
1630 #endif
1631 #ifdef KTRACE
1632 if (KTRPOINT(td, KTR_STRUCT))
1633 ktrstat_error(sbp, error);
1634 #endif
1635 return (error);
1636 }
1637
1638 #if defined(COMPAT_FREEBSD11)
1639 /*
1640 * Return status information about a file descriptor.
1641 */
1642 #ifndef _SYS_SYSPROTO_H_
1643 struct freebsd11_nfstat_args {
1644 int fd;
1645 struct nstat *sb;
1646 };
1647 #endif
1648 /* ARGSUSED */
1649 int
freebsd11_nfstat(struct thread * td,struct freebsd11_nfstat_args * uap)1650 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1651 {
1652 struct nstat nub;
1653 struct stat ub;
1654 int error;
1655
1656 error = kern_fstat(td, uap->fd, &ub);
1657 if (error != 0)
1658 return (error);
1659 error = freebsd11_cvtnstat(&ub, &nub);
1660 if (error != 0)
1661 error = copyout(&nub, uap->sb, sizeof(nub));
1662 return (error);
1663 }
1664 #endif /* COMPAT_FREEBSD11 */
1665
1666 /*
1667 * Return pathconf information about a file descriptor.
1668 */
1669 #ifndef _SYS_SYSPROTO_H_
1670 struct fpathconf_args {
1671 int fd;
1672 int name;
1673 };
1674 #endif
1675 /* ARGSUSED */
1676 int
sys_fpathconf(struct thread * td,struct fpathconf_args * uap)1677 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1678 {
1679 long value;
1680 int error;
1681
1682 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1683 if (error == 0)
1684 td->td_retval[0] = value;
1685 return (error);
1686 }
1687
1688 int
kern_fpathconf(struct thread * td,int fd,int name,long * valuep)1689 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1690 {
1691 struct file *fp;
1692 struct vnode *vp;
1693 int error;
1694
1695 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1696 if (error != 0)
1697 return (error);
1698
1699 if (name == _PC_ASYNC_IO) {
1700 *valuep = _POSIX_ASYNCHRONOUS_IO;
1701 goto out;
1702 }
1703 vp = fp->f_vnode;
1704 if (vp != NULL) {
1705 vn_lock(vp, LK_SHARED | LK_RETRY);
1706 error = VOP_PATHCONF(vp, name, valuep);
1707 VOP_UNLOCK(vp);
1708 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1709 if (name != _PC_PIPE_BUF) {
1710 error = EINVAL;
1711 } else {
1712 *valuep = PIPE_BUF;
1713 error = 0;
1714 }
1715 } else {
1716 error = EOPNOTSUPP;
1717 }
1718 out:
1719 fdrop(fp, td);
1720 return (error);
1721 }
1722
1723 /*
1724 * Copy filecaps structure allocating memory for ioctls array if needed.
1725 *
1726 * The last parameter indicates whether the fdtable is locked. If it is not and
1727 * ioctls are encountered, copying fails and the caller must lock the table.
1728 *
1729 * Note that if the table was not locked, the caller has to check the relevant
1730 * sequence counter to determine whether the operation was successful.
1731 */
1732 bool
filecaps_copy(const struct filecaps * src,struct filecaps * dst,bool locked)1733 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1734 {
1735 size_t size;
1736
1737 if (src->fc_ioctls != NULL && !locked)
1738 return (false);
1739 memcpy(dst, src, sizeof(*src));
1740 if (src->fc_ioctls == NULL)
1741 return (true);
1742
1743 KASSERT(src->fc_nioctls > 0,
1744 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1745
1746 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1747 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1748 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1749 return (true);
1750 }
1751
1752 static u_long *
filecaps_copy_prep(const struct filecaps * src)1753 filecaps_copy_prep(const struct filecaps *src)
1754 {
1755 u_long *ioctls;
1756 size_t size;
1757
1758 if (__predict_true(src->fc_ioctls == NULL))
1759 return (NULL);
1760
1761 KASSERT(src->fc_nioctls > 0,
1762 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1763
1764 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1765 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1766 return (ioctls);
1767 }
1768
1769 static void
filecaps_copy_finish(const struct filecaps * src,struct filecaps * dst,u_long * ioctls)1770 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1771 u_long *ioctls)
1772 {
1773 size_t size;
1774
1775 *dst = *src;
1776 if (__predict_true(src->fc_ioctls == NULL)) {
1777 MPASS(ioctls == NULL);
1778 return;
1779 }
1780
1781 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1782 dst->fc_ioctls = ioctls;
1783 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1784 }
1785
1786 /*
1787 * Move filecaps structure to the new place and clear the old place.
1788 */
1789 void
filecaps_move(struct filecaps * src,struct filecaps * dst)1790 filecaps_move(struct filecaps *src, struct filecaps *dst)
1791 {
1792
1793 *dst = *src;
1794 bzero(src, sizeof(*src));
1795 }
1796
1797 /*
1798 * Fill the given filecaps structure with full rights.
1799 */
1800 static void
filecaps_fill(struct filecaps * fcaps)1801 filecaps_fill(struct filecaps *fcaps)
1802 {
1803
1804 CAP_ALL(&fcaps->fc_rights);
1805 fcaps->fc_ioctls = NULL;
1806 fcaps->fc_nioctls = -1;
1807 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1808 }
1809
1810 /*
1811 * Free memory allocated within filecaps structure.
1812 */
1813 static void
filecaps_free_ioctl(struct filecaps * fcaps)1814 filecaps_free_ioctl(struct filecaps *fcaps)
1815 {
1816
1817 free(fcaps->fc_ioctls, M_FILECAPS);
1818 fcaps->fc_ioctls = NULL;
1819 }
1820
1821 void
filecaps_free(struct filecaps * fcaps)1822 filecaps_free(struct filecaps *fcaps)
1823 {
1824
1825 filecaps_free_ioctl(fcaps);
1826 bzero(fcaps, sizeof(*fcaps));
1827 }
1828
1829 static u_long *
filecaps_free_prep(struct filecaps * fcaps)1830 filecaps_free_prep(struct filecaps *fcaps)
1831 {
1832 u_long *ioctls;
1833
1834 ioctls = fcaps->fc_ioctls;
1835 bzero(fcaps, sizeof(*fcaps));
1836 return (ioctls);
1837 }
1838
1839 static void
filecaps_free_finish(u_long * ioctls)1840 filecaps_free_finish(u_long *ioctls)
1841 {
1842
1843 free(ioctls, M_FILECAPS);
1844 }
1845
1846 /*
1847 * Validate the given filecaps structure.
1848 */
1849 static void
filecaps_validate(const struct filecaps * fcaps,const char * func)1850 filecaps_validate(const struct filecaps *fcaps, const char *func)
1851 {
1852
1853 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1854 ("%s: invalid rights", func));
1855 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1856 ("%s: invalid fcntls", func));
1857 KASSERT(fcaps->fc_fcntls == 0 ||
1858 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1859 ("%s: fcntls without CAP_FCNTL", func));
1860 /*
1861 * open calls without WANTIOCTLCAPS free caps but leave the counter
1862 */
1863 #if 0
1864 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1865 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1866 ("%s: invalid ioctls", func));
1867 #endif
1868 KASSERT(fcaps->fc_nioctls == 0 ||
1869 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1870 ("%s: ioctls without CAP_IOCTL", func));
1871 }
1872
1873 static void
fdgrowtable_exp(struct filedesc * fdp,int nfd)1874 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1875 {
1876 int nfd1;
1877
1878 FILEDESC_XLOCK_ASSERT(fdp);
1879
1880 nfd1 = fdp->fd_nfiles * 2;
1881 if (nfd1 < nfd)
1882 nfd1 = nfd;
1883 fdgrowtable(fdp, nfd1);
1884 }
1885
1886 /*
1887 * Grow the file table to accommodate (at least) nfd descriptors.
1888 */
1889 static void
fdgrowtable(struct filedesc * fdp,int nfd)1890 fdgrowtable(struct filedesc *fdp, int nfd)
1891 {
1892 struct filedesc0 *fdp0;
1893 struct freetable *ft;
1894 struct fdescenttbl *ntable;
1895 struct fdescenttbl *otable;
1896 int nnfiles, onfiles;
1897 NDSLOTTYPE *nmap, *omap;
1898
1899 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1900
1901 /* save old values */
1902 onfiles = fdp->fd_nfiles;
1903 otable = fdp->fd_files;
1904 omap = fdp->fd_map;
1905
1906 /* compute the size of the new table */
1907 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1908 if (nnfiles <= onfiles)
1909 /* the table is already large enough */
1910 return;
1911
1912 /*
1913 * Allocate a new table. We need enough space for the number of
1914 * entries, file entries themselves and the struct freetable we will use
1915 * when we decommission the table and place it on the freelist.
1916 * We place the struct freetable in the middle so we don't have
1917 * to worry about padding.
1918 */
1919 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1920 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1921 sizeof(struct freetable),
1922 M_FILEDESC, M_ZERO | M_WAITOK);
1923 /* copy the old data */
1924 ntable->fdt_nfiles = nnfiles;
1925 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1926 onfiles * sizeof(ntable->fdt_ofiles[0]));
1927
1928 /*
1929 * Allocate a new map only if the old is not large enough. It will
1930 * grow at a slower rate than the table as it can map more
1931 * entries than the table can hold.
1932 */
1933 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1934 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1935 M_ZERO | M_WAITOK);
1936 /* copy over the old data and update the pointer */
1937 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1938 fdp->fd_map = nmap;
1939 }
1940
1941 /*
1942 * Make sure that ntable is correctly initialized before we replace
1943 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1944 * data.
1945 */
1946 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1947
1948 /*
1949 * Free the old file table when not shared by other threads or processes.
1950 * The old file table is considered to be shared when either are true:
1951 * - The process has more than one thread.
1952 * - The file descriptor table has been shared via fdshare().
1953 *
1954 * When shared, the old file table will be placed on a freelist
1955 * which will be processed when the struct filedesc is released.
1956 *
1957 * Note that if onfiles == NDFILE, we're dealing with the original
1958 * static allocation contained within (struct filedesc0 *)fdp,
1959 * which must not be freed.
1960 */
1961 if (onfiles > NDFILE) {
1962 /*
1963 * Note we may be called here from fdinit while allocating a
1964 * table for a new process in which case ->p_fd points
1965 * elsewhere.
1966 */
1967 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1968 free(otable, M_FILEDESC);
1969 } else {
1970 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1971 fdp0 = (struct filedesc0 *)fdp;
1972 ft->ft_table = otable;
1973 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1974 }
1975 }
1976 /*
1977 * The map does not have the same possibility of threads still
1978 * holding references to it. So always free it as long as it
1979 * does not reference the original static allocation.
1980 */
1981 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1982 free(omap, M_FILEDESC);
1983 }
1984
1985 /*
1986 * Allocate a file descriptor for the process.
1987 */
1988 int
fdalloc(struct thread * td,int minfd,int * result)1989 fdalloc(struct thread *td, int minfd, int *result)
1990 {
1991 struct proc *p = td->td_proc;
1992 struct filedesc *fdp = p->p_fd;
1993 int fd, maxfd, allocfd;
1994 #ifdef RACCT
1995 int error;
1996 #endif
1997
1998 FILEDESC_XLOCK_ASSERT(fdp);
1999
2000 if (fdp->fd_freefile > minfd)
2001 minfd = fdp->fd_freefile;
2002
2003 maxfd = getmaxfd(td);
2004
2005 /*
2006 * Search the bitmap for a free descriptor starting at minfd.
2007 * If none is found, grow the file table.
2008 */
2009 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
2010 if (__predict_false(fd >= maxfd))
2011 return (EMFILE);
2012 if (__predict_false(fd >= fdp->fd_nfiles)) {
2013 allocfd = min(fd * 2, maxfd);
2014 #ifdef RACCT
2015 if (RACCT_ENABLED()) {
2016 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
2017 if (error != 0)
2018 return (EMFILE);
2019 }
2020 #endif
2021 /*
2022 * fd is already equal to first free descriptor >= minfd, so
2023 * we only need to grow the table and we are done.
2024 */
2025 fdgrowtable_exp(fdp, allocfd);
2026 }
2027
2028 /*
2029 * Perform some sanity checks, then mark the file descriptor as
2030 * used and return it to the caller.
2031 */
2032 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
2033 ("invalid descriptor %d", fd));
2034 KASSERT(!fdisused(fdp, fd),
2035 ("fd_first_free() returned non-free descriptor"));
2036 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
2037 ("file descriptor isn't free"));
2038 fdused(fdp, fd);
2039 *result = fd;
2040 return (0);
2041 }
2042
2043 /*
2044 * Allocate n file descriptors for the process.
2045 */
2046 int
fdallocn(struct thread * td,int minfd,int * fds,int n)2047 fdallocn(struct thread *td, int minfd, int *fds, int n)
2048 {
2049 struct proc *p = td->td_proc;
2050 struct filedesc *fdp = p->p_fd;
2051 int i;
2052
2053 FILEDESC_XLOCK_ASSERT(fdp);
2054
2055 for (i = 0; i < n; i++)
2056 if (fdalloc(td, 0, &fds[i]) != 0)
2057 break;
2058
2059 if (i < n) {
2060 for (i--; i >= 0; i--)
2061 fdunused(fdp, fds[i]);
2062 return (EMFILE);
2063 }
2064
2065 return (0);
2066 }
2067
2068 /*
2069 * Create a new open file structure and allocate a file descriptor for the
2070 * process that refers to it. We add one reference to the file for the
2071 * descriptor table and one reference for resultfp. This is to prevent us
2072 * being preempted and the entry in the descriptor table closed after we
2073 * release the FILEDESC lock.
2074 */
2075 int
falloc_caps(struct thread * td,struct file ** resultfp,int * resultfd,int flags,struct filecaps * fcaps)2076 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2077 struct filecaps *fcaps)
2078 {
2079 struct file *fp;
2080 int error, fd;
2081
2082 MPASS(resultfp != NULL);
2083 MPASS(resultfd != NULL);
2084
2085 error = _falloc_noinstall(td, &fp, 2);
2086 if (__predict_false(error != 0)) {
2087 return (error);
2088 }
2089
2090 error = finstall_refed(td, fp, &fd, flags, fcaps);
2091 if (__predict_false(error != 0)) {
2092 falloc_abort(td, fp);
2093 return (error);
2094 }
2095
2096 *resultfp = fp;
2097 *resultfd = fd;
2098
2099 return (0);
2100 }
2101
2102 /*
2103 * Create a new open file structure without allocating a file descriptor.
2104 */
2105 int
_falloc_noinstall(struct thread * td,struct file ** resultfp,u_int n)2106 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2107 {
2108 struct file *fp;
2109 int maxuserfiles = maxfiles - (maxfiles / 20);
2110 int openfiles_new;
2111 static struct timeval lastfail;
2112 static int curfail;
2113
2114 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2115 MPASS(n > 0);
2116
2117 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2118 if ((openfiles_new >= maxuserfiles &&
2119 priv_check(td, PRIV_MAXFILES) != 0) ||
2120 openfiles_new >= maxfiles) {
2121 atomic_subtract_int(&openfiles, 1);
2122 if (ppsratecheck(&lastfail, &curfail, 1)) {
2123 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2124 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2125 }
2126 return (ENFILE);
2127 }
2128 fp = uma_zalloc(file_zone, M_WAITOK);
2129 bzero(fp, sizeof(*fp));
2130 refcount_init(&fp->f_count, n);
2131 fp->f_cred = crhold(td->td_ucred);
2132 fp->f_ops = &badfileops;
2133 *resultfp = fp;
2134 return (0);
2135 }
2136
2137 void
falloc_abort(struct thread * td,struct file * fp)2138 falloc_abort(struct thread *td, struct file *fp)
2139 {
2140
2141 /*
2142 * For assertion purposes.
2143 */
2144 refcount_init(&fp->f_count, 0);
2145 _fdrop(fp, td);
2146 }
2147
2148 /*
2149 * Install a file in a file descriptor table.
2150 */
2151 void
_finstall(struct filedesc * fdp,struct file * fp,int fd,int flags,struct filecaps * fcaps)2152 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2153 struct filecaps *fcaps)
2154 {
2155 struct filedescent *fde;
2156
2157 MPASS(fp != NULL);
2158 if (fcaps != NULL)
2159 filecaps_validate(fcaps, __func__);
2160 FILEDESC_XLOCK_ASSERT(fdp);
2161
2162 fde = &fdp->fd_ofiles[fd];
2163 #ifdef CAPABILITIES
2164 seqc_write_begin(&fde->fde_seqc);
2165 #endif
2166 fde->fde_file = fp;
2167 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2168 if (fcaps != NULL)
2169 filecaps_move(fcaps, &fde->fde_caps);
2170 else
2171 filecaps_fill(&fde->fde_caps);
2172 #ifdef CAPABILITIES
2173 seqc_write_end(&fde->fde_seqc);
2174 #endif
2175 }
2176
2177 int
finstall_refed(struct thread * td,struct file * fp,int * fd,int flags,struct filecaps * fcaps)2178 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2179 struct filecaps *fcaps)
2180 {
2181 struct filedesc *fdp = td->td_proc->p_fd;
2182 int error;
2183
2184 MPASS(fd != NULL);
2185
2186 FILEDESC_XLOCK(fdp);
2187 error = fdalloc(td, 0, fd);
2188 if (__predict_true(error == 0)) {
2189 _finstall(fdp, fp, *fd, flags, fcaps);
2190 }
2191 FILEDESC_XUNLOCK(fdp);
2192 return (error);
2193 }
2194
2195 int
finstall(struct thread * td,struct file * fp,int * fd,int flags,struct filecaps * fcaps)2196 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2197 struct filecaps *fcaps)
2198 {
2199 int error;
2200
2201 MPASS(fd != NULL);
2202
2203 if (!fhold(fp))
2204 return (EBADF);
2205 error = finstall_refed(td, fp, fd, flags, fcaps);
2206 if (__predict_false(error != 0)) {
2207 fdrop(fp, td);
2208 }
2209 return (error);
2210 }
2211
2212 /*
2213 * Build a new filedesc structure from another.
2214 *
2215 * If fdp is not NULL, return with it shared locked.
2216 */
2217 struct filedesc *
fdinit(void)2218 fdinit(void)
2219 {
2220 struct filedesc0 *newfdp0;
2221 struct filedesc *newfdp;
2222
2223 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2224 newfdp = &newfdp0->fd_fd;
2225
2226 /* Create the file descriptor table. */
2227 FILEDESC_LOCK_INIT(newfdp);
2228 refcount_init(&newfdp->fd_refcnt, 1);
2229 refcount_init(&newfdp->fd_holdcnt, 1);
2230 newfdp->fd_map = newfdp0->fd_dmap;
2231 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2232 newfdp->fd_files->fdt_nfiles = NDFILE;
2233
2234 return (newfdp);
2235 }
2236
2237 /*
2238 * Build a pwddesc structure from another.
2239 * Copy the current, root, and jail root vnode references.
2240 *
2241 * If pdp is not NULL and keeplock is true, return with it (exclusively) locked.
2242 */
2243 struct pwddesc *
pdinit(struct pwddesc * pdp,bool keeplock)2244 pdinit(struct pwddesc *pdp, bool keeplock)
2245 {
2246 struct pwddesc *newpdp;
2247 struct pwd *newpwd;
2248
2249 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2250
2251 PWDDESC_LOCK_INIT(newpdp);
2252 refcount_init(&newpdp->pd_refcount, 1);
2253 newpdp->pd_cmask = CMASK;
2254
2255 if (pdp == NULL) {
2256 newpwd = pwd_alloc();
2257 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2258 return (newpdp);
2259 }
2260
2261 PWDDESC_XLOCK(pdp);
2262 newpwd = pwd_hold_pwddesc(pdp);
2263 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2264 if (!keeplock)
2265 PWDDESC_XUNLOCK(pdp);
2266 return (newpdp);
2267 }
2268
2269 /*
2270 * Hold either filedesc or pwddesc of the passed process.
2271 *
2272 * The process lock is used to synchronize against the target exiting and
2273 * freeing the data.
2274 *
2275 * Clearing can be ilustrated in 3 steps:
2276 * 1. set the pointer to NULL. Either routine can race against it, hence
2277 * atomic_load_ptr.
2278 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2279 * race to either still see the pointer or find NULL. It is still safe to
2280 * grab a reference as clearing is stalled.
2281 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2282 * guaranteed to see NULL, making it safe to finish clearing
2283 */
2284 static struct filedesc *
fdhold(struct proc * p)2285 fdhold(struct proc *p)
2286 {
2287 struct filedesc *fdp;
2288
2289 PROC_LOCK_ASSERT(p, MA_OWNED);
2290 fdp = atomic_load_ptr(&p->p_fd);
2291 if (fdp != NULL)
2292 refcount_acquire(&fdp->fd_holdcnt);
2293 return (fdp);
2294 }
2295
2296 static struct pwddesc *
pdhold(struct proc * p)2297 pdhold(struct proc *p)
2298 {
2299 struct pwddesc *pdp;
2300
2301 PROC_LOCK_ASSERT(p, MA_OWNED);
2302 pdp = atomic_load_ptr(&p->p_pd);
2303 if (pdp != NULL)
2304 refcount_acquire(&pdp->pd_refcount);
2305 return (pdp);
2306 }
2307
2308 static void
fddrop(struct filedesc * fdp)2309 fddrop(struct filedesc *fdp)
2310 {
2311
2312 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2313 if (refcount_release(&fdp->fd_holdcnt) == 0)
2314 return;
2315 }
2316
2317 FILEDESC_LOCK_DESTROY(fdp);
2318 uma_zfree(filedesc0_zone, fdp);
2319 }
2320
2321 static void
pddrop(struct pwddesc * pdp)2322 pddrop(struct pwddesc *pdp)
2323 {
2324 struct pwd *pwd;
2325
2326 if (refcount_release_if_not_last(&pdp->pd_refcount))
2327 return;
2328
2329 PWDDESC_XLOCK(pdp);
2330 if (refcount_release(&pdp->pd_refcount) == 0) {
2331 PWDDESC_XUNLOCK(pdp);
2332 return;
2333 }
2334 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2335 pwd_set(pdp, NULL);
2336 PWDDESC_XUNLOCK(pdp);
2337 pwd_drop(pwd);
2338
2339 PWDDESC_LOCK_DESTROY(pdp);
2340 free(pdp, M_PWDDESC);
2341 }
2342
2343 /*
2344 * Share a filedesc structure.
2345 */
2346 struct filedesc *
fdshare(struct filedesc * fdp)2347 fdshare(struct filedesc *fdp)
2348 {
2349
2350 refcount_acquire(&fdp->fd_refcnt);
2351 return (fdp);
2352 }
2353
2354 /*
2355 * Share a pwddesc structure.
2356 */
2357 struct pwddesc *
pdshare(struct pwddesc * pdp)2358 pdshare(struct pwddesc *pdp)
2359 {
2360 refcount_acquire(&pdp->pd_refcount);
2361 return (pdp);
2362 }
2363
2364 /*
2365 * Unshare a filedesc structure, if necessary by making a copy
2366 */
2367 void
fdunshare(struct thread * td)2368 fdunshare(struct thread *td)
2369 {
2370 struct filedesc *tmp;
2371 struct proc *p = td->td_proc;
2372
2373 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2374 return;
2375
2376 tmp = fdcopy(p->p_fd);
2377 fdescfree(td);
2378 p->p_fd = tmp;
2379 }
2380
2381 /*
2382 * Unshare a pwddesc structure.
2383 */
2384 void
pdunshare(struct thread * td)2385 pdunshare(struct thread *td)
2386 {
2387 struct pwddesc *pdp;
2388 struct proc *p;
2389
2390 p = td->td_proc;
2391 /* Not shared. */
2392 if (refcount_load(&p->p_pd->pd_refcount) == 1)
2393 return;
2394
2395 pdp = pdcopy(p->p_pd);
2396 pdescfree(td);
2397 p->p_pd = pdp;
2398 }
2399
2400 /*
2401 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2402 * this is to ease callers, not catch errors.
2403 */
2404 struct filedesc *
fdcopy(struct filedesc * fdp)2405 fdcopy(struct filedesc *fdp)
2406 {
2407 struct filedesc *newfdp;
2408 struct filedescent *nfde, *ofde;
2409 int i, lastfile;
2410
2411 MPASS(fdp != NULL);
2412
2413 newfdp = fdinit();
2414 FILEDESC_SLOCK(fdp);
2415 for (;;) {
2416 lastfile = fdlastfile(fdp);
2417 if (lastfile < newfdp->fd_nfiles)
2418 break;
2419 FILEDESC_SUNLOCK(fdp);
2420 fdgrowtable(newfdp, lastfile + 1);
2421 FILEDESC_SLOCK(fdp);
2422 }
2423 /* copy all passable descriptors (i.e. not kqueue) */
2424 newfdp->fd_freefile = fdp->fd_freefile;
2425 FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2426 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2427 !fhold(ofde->fde_file)) {
2428 if (newfdp->fd_freefile == fdp->fd_freefile)
2429 newfdp->fd_freefile = i;
2430 continue;
2431 }
2432 nfde = &newfdp->fd_ofiles[i];
2433 *nfde = *ofde;
2434 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2435 fdused_init(newfdp, i);
2436 }
2437 MPASS(newfdp->fd_freefile != -1);
2438 FILEDESC_SUNLOCK(fdp);
2439 return (newfdp);
2440 }
2441
2442 /*
2443 * Copy a pwddesc structure.
2444 */
2445 struct pwddesc *
pdcopy(struct pwddesc * pdp)2446 pdcopy(struct pwddesc *pdp)
2447 {
2448 struct pwddesc *newpdp;
2449
2450 MPASS(pdp != NULL);
2451
2452 newpdp = pdinit(pdp, true);
2453 newpdp->pd_cmask = pdp->pd_cmask;
2454 PWDDESC_XUNLOCK(pdp);
2455 return (newpdp);
2456 }
2457
2458 /*
2459 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2460 * one of processes using it exits) and the table used to be shared.
2461 */
2462 static void
fdclearlocks(struct thread * td)2463 fdclearlocks(struct thread *td)
2464 {
2465 struct filedesc *fdp;
2466 struct filedesc_to_leader *fdtol;
2467 struct flock lf;
2468 struct file *fp;
2469 struct proc *p;
2470 struct vnode *vp;
2471 int i;
2472
2473 p = td->td_proc;
2474 fdp = p->p_fd;
2475 fdtol = p->p_fdtol;
2476 MPASS(fdtol != NULL);
2477
2478 FILEDESC_XLOCK(fdp);
2479 KASSERT(fdtol->fdl_refcount > 0,
2480 ("filedesc_to_refcount botch: fdl_refcount=%d",
2481 fdtol->fdl_refcount));
2482 if (fdtol->fdl_refcount == 1 &&
2483 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2484 FILEDESC_FOREACH_FP(fdp, i, fp) {
2485 if (fp->f_type != DTYPE_VNODE ||
2486 !fhold(fp))
2487 continue;
2488 FILEDESC_XUNLOCK(fdp);
2489 lf.l_whence = SEEK_SET;
2490 lf.l_start = 0;
2491 lf.l_len = 0;
2492 lf.l_type = F_UNLCK;
2493 vp = fp->f_vnode;
2494 (void) VOP_ADVLOCK(vp,
2495 (caddr_t)p->p_leader, F_UNLCK,
2496 &lf, F_POSIX);
2497 FILEDESC_XLOCK(fdp);
2498 fdrop(fp, td);
2499 }
2500 }
2501 retry:
2502 if (fdtol->fdl_refcount == 1) {
2503 if (fdp->fd_holdleaderscount > 0 &&
2504 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2505 /*
2506 * close() or kern_dup() has cleared a reference
2507 * in a shared file descriptor table.
2508 */
2509 fdp->fd_holdleaderswakeup = 1;
2510 sx_sleep(&fdp->fd_holdleaderscount,
2511 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2512 goto retry;
2513 }
2514 if (fdtol->fdl_holdcount > 0) {
2515 /*
2516 * Ensure that fdtol->fdl_leader remains
2517 * valid in closef().
2518 */
2519 fdtol->fdl_wakeup = 1;
2520 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2521 "fdlhold", 0);
2522 goto retry;
2523 }
2524 }
2525 fdtol->fdl_refcount--;
2526 if (fdtol->fdl_refcount == 0 &&
2527 fdtol->fdl_holdcount == 0) {
2528 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2529 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2530 } else
2531 fdtol = NULL;
2532 p->p_fdtol = NULL;
2533 FILEDESC_XUNLOCK(fdp);
2534 if (fdtol != NULL)
2535 free(fdtol, M_FILEDESC_TO_LEADER);
2536 }
2537
2538 /*
2539 * Release a filedesc structure.
2540 */
2541 static void
fdescfree_fds(struct thread * td,struct filedesc * fdp)2542 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2543 {
2544 struct filedesc0 *fdp0;
2545 struct freetable *ft, *tft;
2546 struct filedescent *fde;
2547 struct file *fp;
2548 int i;
2549
2550 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2551 ("%s: fd table %p carries references", __func__, fdp));
2552
2553 /*
2554 * Serialize with threads iterating over the table, if any.
2555 */
2556 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2557 FILEDESC_XLOCK(fdp);
2558 FILEDESC_XUNLOCK(fdp);
2559 }
2560
2561 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2562 fp = fde->fde_file;
2563 fdefree_last(fde);
2564 (void) closef(fp, td);
2565 }
2566
2567 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2568 free(fdp->fd_map, M_FILEDESC);
2569 if (fdp->fd_nfiles > NDFILE)
2570 free(fdp->fd_files, M_FILEDESC);
2571
2572 fdp0 = (struct filedesc0 *)fdp;
2573 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2574 free(ft->ft_table, M_FILEDESC);
2575
2576 fddrop(fdp);
2577 }
2578
2579 void
fdescfree(struct thread * td)2580 fdescfree(struct thread *td)
2581 {
2582 struct proc *p;
2583 struct filedesc *fdp;
2584
2585 p = td->td_proc;
2586 fdp = p->p_fd;
2587 MPASS(fdp != NULL);
2588
2589 #ifdef RACCT
2590 if (RACCT_ENABLED())
2591 racct_set_unlocked(p, RACCT_NOFILE, 0);
2592 #endif
2593
2594 if (p->p_fdtol != NULL)
2595 fdclearlocks(td);
2596
2597 /*
2598 * Check fdhold for an explanation.
2599 */
2600 atomic_store_ptr(&p->p_fd, NULL);
2601 atomic_thread_fence_seq_cst();
2602 PROC_WAIT_UNLOCKED(p);
2603
2604 if (refcount_release(&fdp->fd_refcnt) == 0)
2605 return;
2606
2607 fdescfree_fds(td, fdp);
2608 }
2609
2610 void
pdescfree(struct thread * td)2611 pdescfree(struct thread *td)
2612 {
2613 struct proc *p;
2614 struct pwddesc *pdp;
2615
2616 p = td->td_proc;
2617 pdp = p->p_pd;
2618 MPASS(pdp != NULL);
2619
2620 /*
2621 * Check pdhold for an explanation.
2622 */
2623 atomic_store_ptr(&p->p_pd, NULL);
2624 atomic_thread_fence_seq_cst();
2625 PROC_WAIT_UNLOCKED(p);
2626
2627 pddrop(pdp);
2628 }
2629
2630 /*
2631 * For setugid programs, we don't want to people to use that setugidness
2632 * to generate error messages which write to a file which otherwise would
2633 * otherwise be off-limits to the process. We check for filesystems where
2634 * the vnode can change out from under us after execve (like [lin]procfs).
2635 *
2636 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2637 * sufficient. We also don't check for setugidness since we know we are.
2638 */
2639 static bool
is_unsafe(struct file * fp)2640 is_unsafe(struct file *fp)
2641 {
2642 struct vnode *vp;
2643
2644 if (fp->f_type != DTYPE_VNODE)
2645 return (false);
2646
2647 vp = fp->f_vnode;
2648 return ((vp->v_vflag & VV_PROCDEP) != 0);
2649 }
2650
2651 /*
2652 * Make this setguid thing safe, if at all possible.
2653 */
2654 void
fdsetugidsafety(struct thread * td)2655 fdsetugidsafety(struct thread *td)
2656 {
2657 struct filedesc *fdp;
2658 struct file *fp;
2659 int i;
2660
2661 fdp = td->td_proc->p_fd;
2662 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2663 ("the fdtable should not be shared"));
2664 MPASS(fdp->fd_nfiles >= 3);
2665 for (i = 0; i <= 2; i++) {
2666 fp = fdp->fd_ofiles[i].fde_file;
2667 if (fp != NULL && is_unsafe(fp)) {
2668 FILEDESC_XLOCK(fdp);
2669 knote_fdclose(td, i);
2670 /*
2671 * NULL-out descriptor prior to close to avoid
2672 * a race while close blocks.
2673 */
2674 fdfree(fdp, i);
2675 FILEDESC_XUNLOCK(fdp);
2676 (void) closef(fp, td);
2677 }
2678 }
2679 }
2680
2681 /*
2682 * If a specific file object occupies a specific file descriptor, close the
2683 * file descriptor entry and drop a reference on the file object. This is a
2684 * convenience function to handle a subsequent error in a function that calls
2685 * falloc() that handles the race that another thread might have closed the
2686 * file descriptor out from under the thread creating the file object.
2687 */
2688 void
fdclose(struct thread * td,struct file * fp,int idx)2689 fdclose(struct thread *td, struct file *fp, int idx)
2690 {
2691 struct filedesc *fdp = td->td_proc->p_fd;
2692
2693 FILEDESC_XLOCK(fdp);
2694 if (fdp->fd_ofiles[idx].fde_file == fp) {
2695 fdfree(fdp, idx);
2696 FILEDESC_XUNLOCK(fdp);
2697 fdrop(fp, td);
2698 } else
2699 FILEDESC_XUNLOCK(fdp);
2700 }
2701
2702 /*
2703 * Close any files on exec?
2704 */
2705 void
fdcloseexec(struct thread * td)2706 fdcloseexec(struct thread *td)
2707 {
2708 struct filedesc *fdp;
2709 struct filedescent *fde;
2710 struct file *fp;
2711 int i;
2712
2713 fdp = td->td_proc->p_fd;
2714 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2715 ("the fdtable should not be shared"));
2716 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2717 fp = fde->fde_file;
2718 if (fp->f_type == DTYPE_MQUEUE ||
2719 (fde->fde_flags & UF_EXCLOSE)) {
2720 FILEDESC_XLOCK(fdp);
2721 fdfree(fdp, i);
2722 (void) closefp(fdp, i, fp, td, false, false);
2723 FILEDESC_UNLOCK_ASSERT(fdp);
2724 }
2725 }
2726 }
2727
2728 /*
2729 * It is unsafe for set[ug]id processes to be started with file
2730 * descriptors 0..2 closed, as these descriptors are given implicit
2731 * significance in the Standard C library. fdcheckstd() will create a
2732 * descriptor referencing /dev/null for each of stdin, stdout, and
2733 * stderr that is not already open.
2734 */
2735 int
fdcheckstd(struct thread * td)2736 fdcheckstd(struct thread *td)
2737 {
2738 struct filedesc *fdp;
2739 register_t save;
2740 int i, error, devnull;
2741
2742 fdp = td->td_proc->p_fd;
2743 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2744 ("the fdtable should not be shared"));
2745 MPASS(fdp->fd_nfiles >= 3);
2746 devnull = -1;
2747 for (i = 0; i <= 2; i++) {
2748 if (fdp->fd_ofiles[i].fde_file != NULL)
2749 continue;
2750
2751 save = td->td_retval[0];
2752 if (devnull != -1) {
2753 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2754 } else {
2755 error = kern_openat(td, AT_FDCWD, "/dev/null",
2756 UIO_SYSSPACE, O_RDWR, 0);
2757 if (error == 0) {
2758 devnull = td->td_retval[0];
2759 KASSERT(devnull == i, ("we didn't get our fd"));
2760 }
2761 }
2762 td->td_retval[0] = save;
2763 if (error != 0)
2764 return (error);
2765 }
2766 return (0);
2767 }
2768
2769 /*
2770 * Internal form of close. Decrement reference count on file structure.
2771 * Note: td may be NULL when closing a file that was being passed in a
2772 * message.
2773 */
2774 int
closef(struct file * fp,struct thread * td)2775 closef(struct file *fp, struct thread *td)
2776 {
2777 struct vnode *vp;
2778 struct flock lf;
2779 struct filedesc_to_leader *fdtol;
2780 struct filedesc *fdp;
2781
2782 MPASS(td != NULL);
2783
2784 /*
2785 * POSIX record locking dictates that any close releases ALL
2786 * locks owned by this process. This is handled by setting
2787 * a flag in the unlock to free ONLY locks obeying POSIX
2788 * semantics, and not to free BSD-style file locks.
2789 * If the descriptor was in a message, POSIX-style locks
2790 * aren't passed with the descriptor, and the thread pointer
2791 * will be NULL. Callers should be careful only to pass a
2792 * NULL thread pointer when there really is no owning
2793 * context that might have locks, or the locks will be
2794 * leaked.
2795 */
2796 if (fp->f_type == DTYPE_VNODE) {
2797 vp = fp->f_vnode;
2798 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2799 lf.l_whence = SEEK_SET;
2800 lf.l_start = 0;
2801 lf.l_len = 0;
2802 lf.l_type = F_UNLCK;
2803 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2804 F_UNLCK, &lf, F_POSIX);
2805 }
2806 fdtol = td->td_proc->p_fdtol;
2807 if (fdtol != NULL) {
2808 /*
2809 * Handle special case where file descriptor table is
2810 * shared between multiple process leaders.
2811 */
2812 fdp = td->td_proc->p_fd;
2813 FILEDESC_XLOCK(fdp);
2814 for (fdtol = fdtol->fdl_next;
2815 fdtol != td->td_proc->p_fdtol;
2816 fdtol = fdtol->fdl_next) {
2817 if ((fdtol->fdl_leader->p_flag &
2818 P_ADVLOCK) == 0)
2819 continue;
2820 fdtol->fdl_holdcount++;
2821 FILEDESC_XUNLOCK(fdp);
2822 lf.l_whence = SEEK_SET;
2823 lf.l_start = 0;
2824 lf.l_len = 0;
2825 lf.l_type = F_UNLCK;
2826 vp = fp->f_vnode;
2827 (void) VOP_ADVLOCK(vp,
2828 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2829 F_POSIX);
2830 FILEDESC_XLOCK(fdp);
2831 fdtol->fdl_holdcount--;
2832 if (fdtol->fdl_holdcount == 0 &&
2833 fdtol->fdl_wakeup != 0) {
2834 fdtol->fdl_wakeup = 0;
2835 wakeup(fdtol);
2836 }
2837 }
2838 FILEDESC_XUNLOCK(fdp);
2839 }
2840 }
2841 return (fdrop_close(fp, td));
2842 }
2843
2844 /*
2845 * Hack for file descriptor passing code.
2846 */
2847 void
closef_nothread(struct file * fp)2848 closef_nothread(struct file *fp)
2849 {
2850
2851 fdrop(fp, NULL);
2852 }
2853
2854 /*
2855 * Initialize the file pointer with the specified properties.
2856 *
2857 * The ops are set with release semantics to be certain that the flags, type,
2858 * and data are visible when ops is. This is to prevent ops methods from being
2859 * called with bad data.
2860 */
2861 void
finit(struct file * fp,u_int flag,short type,void * data,const struct fileops * ops)2862 finit(struct file *fp, u_int flag, short type, void *data,
2863 const struct fileops *ops)
2864 {
2865 fp->f_data = data;
2866 fp->f_flag = flag;
2867 fp->f_type = type;
2868 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2869 }
2870
2871 void
finit_vnode(struct file * fp,u_int flag,void * data,const struct fileops * ops)2872 finit_vnode(struct file *fp, u_int flag, void *data, const struct fileops *ops)
2873 {
2874 fp->f_seqcount[UIO_READ] = 1;
2875 fp->f_seqcount[UIO_WRITE] = 1;
2876 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2877 data, ops);
2878 }
2879
2880 int
fget_cap_noref(struct filedesc * fdp,int fd,cap_rights_t * needrightsp,struct file ** fpp,struct filecaps * havecapsp)2881 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2882 struct file **fpp, struct filecaps *havecapsp)
2883 {
2884 struct filedescent *fde;
2885 int error;
2886
2887 FILEDESC_LOCK_ASSERT(fdp);
2888
2889 *fpp = NULL;
2890 fde = fdeget_noref(fdp, fd);
2891 if (fde == NULL) {
2892 error = EBADF;
2893 goto out;
2894 }
2895
2896 #ifdef CAPABILITIES
2897 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2898 if (error != 0)
2899 goto out;
2900 #endif
2901
2902 if (havecapsp != NULL)
2903 filecaps_copy(&fde->fde_caps, havecapsp, true);
2904
2905 *fpp = fde->fde_file;
2906
2907 error = 0;
2908 out:
2909 return (error);
2910 }
2911
2912 #ifdef CAPABILITIES
2913 int
fget_cap(struct thread * td,int fd,cap_rights_t * needrightsp,struct file ** fpp,struct filecaps * havecapsp)2914 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2915 struct file **fpp, struct filecaps *havecapsp)
2916 {
2917 struct filedesc *fdp = td->td_proc->p_fd;
2918 int error;
2919 struct file *fp;
2920 seqc_t seq;
2921
2922 *fpp = NULL;
2923 for (;;) {
2924 error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq);
2925 if (error != 0)
2926 return (error);
2927
2928 if (havecapsp != NULL) {
2929 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2930 havecapsp, false)) {
2931 fdrop(fp, td);
2932 goto get_locked;
2933 }
2934 }
2935
2936 if (!fd_modified(fdp, fd, seq))
2937 break;
2938 fdrop(fp, td);
2939 }
2940
2941 *fpp = fp;
2942 return (0);
2943
2944 get_locked:
2945 FILEDESC_SLOCK(fdp);
2946 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp);
2947 if (error == 0 && !fhold(*fpp))
2948 error = EBADF;
2949 FILEDESC_SUNLOCK(fdp);
2950 return (error);
2951 }
2952 #else
2953 int
fget_cap(struct thread * td,int fd,cap_rights_t * needrightsp,struct file ** fpp,struct filecaps * havecapsp)2954 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2955 struct file **fpp, struct filecaps *havecapsp)
2956 {
2957 int error;
2958 error = fget_unlocked(td, fd, needrightsp, fpp);
2959 if (havecapsp != NULL && error == 0)
2960 filecaps_fill(havecapsp);
2961
2962 return (error);
2963 }
2964 #endif
2965
2966 int
fget_remote(struct thread * td,struct proc * p,int fd,struct file ** fpp)2967 fget_remote(struct thread *td, struct proc *p, int fd, struct file **fpp)
2968 {
2969 struct filedesc *fdp;
2970 struct file *fp;
2971 int error;
2972
2973 if (p == td->td_proc) /* curproc */
2974 return (fget_unlocked(td, fd, &cap_no_rights, fpp));
2975
2976 PROC_LOCK(p);
2977 fdp = fdhold(p);
2978 PROC_UNLOCK(p);
2979 if (fdp == NULL)
2980 return (ENOENT);
2981 FILEDESC_SLOCK(fdp);
2982 if (refcount_load(&fdp->fd_refcnt) != 0) {
2983 fp = fget_noref(fdp, fd);
2984 if (fp != NULL && fhold(fp)) {
2985 *fpp = fp;
2986 error = 0;
2987 } else {
2988 error = EBADF;
2989 }
2990 } else {
2991 error = ENOENT;
2992 }
2993 FILEDESC_SUNLOCK(fdp);
2994 fddrop(fdp);
2995 return (error);
2996 }
2997
2998 #ifdef CAPABILITIES
2999 int
fgetvp_lookup_smr(struct nameidata * ndp,struct vnode ** vpp,bool * fsearch)3000 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3001 {
3002 const struct filedescent *fde;
3003 const struct fdescenttbl *fdt;
3004 struct filedesc *fdp;
3005 struct file *fp;
3006 struct vnode *vp;
3007 const cap_rights_t *haverights;
3008 cap_rights_t rights;
3009 seqc_t seq;
3010 int fd;
3011
3012 VFS_SMR_ASSERT_ENTERED();
3013
3014 fd = ndp->ni_dirfd;
3015 rights = *ndp->ni_rightsneeded;
3016 cap_rights_set_one(&rights, CAP_LOOKUP);
3017
3018 fdp = curproc->p_fd;
3019 fdt = fdp->fd_files;
3020 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3021 return (EBADF);
3022 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3023 fde = &fdt->fdt_ofiles[fd];
3024 haverights = cap_rights_fde_inline(fde);
3025 fp = fde->fde_file;
3026 if (__predict_false(fp == NULL))
3027 return (EAGAIN);
3028 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
3029 return (EAGAIN);
3030 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3031 vp = fp->f_vnode;
3032 if (__predict_false(vp == NULL)) {
3033 return (EAGAIN);
3034 }
3035 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
3036 return (EAGAIN);
3037 }
3038 /*
3039 * Use an acquire barrier to force re-reading of fdt so it is
3040 * refreshed for verification.
3041 */
3042 atomic_thread_fence_acq();
3043 fdt = fdp->fd_files;
3044 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3045 return (EAGAIN);
3046 /*
3047 * If file descriptor doesn't have all rights,
3048 * all lookups relative to it must also be
3049 * strictly relative.
3050 *
3051 * Not yet supported by fast path.
3052 */
3053 CAP_ALL(&rights);
3054 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3055 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3056 ndp->ni_filecaps.fc_nioctls != -1) {
3057 #ifdef notyet
3058 ndp->ni_lcf |= NI_LCF_STRICTREL;
3059 #else
3060 return (EAGAIN);
3061 #endif
3062 }
3063 *vpp = vp;
3064 return (0);
3065 }
3066 #else
3067 int
fgetvp_lookup_smr(struct nameidata * ndp,struct vnode ** vpp,bool * fsearch)3068 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3069 {
3070 const struct fdescenttbl *fdt;
3071 struct filedesc *fdp;
3072 struct file *fp;
3073 struct vnode *vp;
3074 int fd;
3075
3076 VFS_SMR_ASSERT_ENTERED();
3077
3078 fd = ndp->ni_dirfd;
3079 fdp = curproc->p_fd;
3080 fdt = fdp->fd_files;
3081 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3082 return (EBADF);
3083 fp = fdt->fdt_ofiles[fd].fde_file;
3084 if (__predict_false(fp == NULL))
3085 return (EAGAIN);
3086 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3087 vp = fp->f_vnode;
3088 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3089 return (EAGAIN);
3090 }
3091 /*
3092 * Use an acquire barrier to force re-reading of fdt so it is
3093 * refreshed for verification.
3094 */
3095 atomic_thread_fence_acq();
3096 fdt = fdp->fd_files;
3097 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3098 return (EAGAIN);
3099 filecaps_fill(&ndp->ni_filecaps);
3100 *vpp = vp;
3101 return (0);
3102 }
3103 #endif
3104
3105 int
fgetvp_lookup(struct nameidata * ndp,struct vnode ** vpp)3106 fgetvp_lookup(struct nameidata *ndp, struct vnode **vpp)
3107 {
3108 struct thread *td;
3109 struct file *fp;
3110 struct vnode *vp;
3111 struct componentname *cnp;
3112 cap_rights_t rights;
3113 int error;
3114
3115 td = curthread;
3116 rights = *ndp->ni_rightsneeded;
3117 cap_rights_set_one(&rights, CAP_LOOKUP);
3118 cnp = &ndp->ni_cnd;
3119
3120 error = fget_cap(td, ndp->ni_dirfd, &rights, &fp, &ndp->ni_filecaps);
3121 if (__predict_false(error != 0))
3122 return (error);
3123 if (__predict_false(fp->f_ops == &badfileops)) {
3124 error = EBADF;
3125 goto out_free;
3126 }
3127 vp = fp->f_vnode;
3128 if (__predict_false(vp == NULL)) {
3129 error = ENOTDIR;
3130 goto out_free;
3131 }
3132 vrefact(vp);
3133 /*
3134 * XXX does not check for VDIR, handled by namei_setup
3135 */
3136 if ((fp->f_flag & FSEARCH) != 0)
3137 cnp->cn_flags |= NOEXECCHECK;
3138 fdrop(fp, td);
3139
3140 #ifdef CAPABILITIES
3141 /*
3142 * If file descriptor doesn't have all rights,
3143 * all lookups relative to it must also be
3144 * strictly relative.
3145 */
3146 CAP_ALL(&rights);
3147 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3148 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3149 ndp->ni_filecaps.fc_nioctls != -1) {
3150 ndp->ni_lcf |= NI_LCF_STRICTREL;
3151 ndp->ni_resflags |= NIRES_STRICTREL;
3152 }
3153 #endif
3154
3155 /*
3156 * TODO: avoid copying ioctl caps if it can be helped to begin with
3157 */
3158 if ((cnp->cn_flags & WANTIOCTLCAPS) == 0)
3159 filecaps_free_ioctl(&ndp->ni_filecaps);
3160
3161 *vpp = vp;
3162 return (0);
3163
3164 out_free:
3165 filecaps_free(&ndp->ni_filecaps);
3166 fdrop(fp, td);
3167 return (error);
3168 }
3169
3170 /*
3171 * Fetch the descriptor locklessly.
3172 *
3173 * We avoid fdrop() races by never raising a refcount above 0. To accomplish
3174 * this we have to use a cmpset loop rather than an atomic_add. The descriptor
3175 * must be re-verified once we acquire a reference to be certain that the
3176 * identity is still correct and we did not lose a race due to preemption.
3177 *
3178 * Force a reload of fdt when looping. Another thread could reallocate
3179 * the table before this fd was closed, so it is possible that there is
3180 * a stale fp pointer in cached version.
3181 */
3182 #ifdef CAPABILITIES
3183 static int
fget_unlocked_seq(struct thread * td,int fd,cap_rights_t * needrightsp,struct file ** fpp,seqc_t * seqp)3184 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3185 struct file **fpp, seqc_t *seqp)
3186 {
3187 struct filedesc *fdp;
3188 const struct filedescent *fde;
3189 const struct fdescenttbl *fdt;
3190 struct file *fp;
3191 seqc_t seq;
3192 cap_rights_t haverights;
3193 int error;
3194
3195 fdp = td->td_proc->p_fd;
3196 fdt = fdp->fd_files;
3197 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3198 return (EBADF);
3199
3200 for (;;) {
3201 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3202 fde = &fdt->fdt_ofiles[fd];
3203 haverights = *cap_rights_fde_inline(fde);
3204 fp = fde->fde_file;
3205 if (__predict_false(fp == NULL)) {
3206 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3207 return (EBADF);
3208 fdt = atomic_load_ptr(&fdp->fd_files);
3209 continue;
3210 }
3211 error = cap_check_inline(&haverights, needrightsp);
3212 if (__predict_false(error != 0)) {
3213 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3214 return (error);
3215 fdt = atomic_load_ptr(&fdp->fd_files);
3216 continue;
3217 }
3218 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3219 fdt = atomic_load_ptr(&fdp->fd_files);
3220 continue;
3221 }
3222 /*
3223 * Use an acquire barrier to force re-reading of fdt so it is
3224 * refreshed for verification.
3225 */
3226 atomic_thread_fence_acq();
3227 fdt = fdp->fd_files;
3228 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3229 break;
3230 fdrop(fp, td);
3231 }
3232 *fpp = fp;
3233 if (seqp != NULL) {
3234 *seqp = seq;
3235 }
3236 return (0);
3237 }
3238 #else
3239 static int
fget_unlocked_seq(struct thread * td,int fd,cap_rights_t * needrightsp,struct file ** fpp,seqc_t * seqp __unused)3240 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3241 struct file **fpp, seqc_t *seqp __unused)
3242 {
3243 struct filedesc *fdp;
3244 const struct fdescenttbl *fdt;
3245 struct file *fp;
3246
3247 fdp = td->td_proc->p_fd;
3248 fdt = fdp->fd_files;
3249 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3250 return (EBADF);
3251
3252 for (;;) {
3253 fp = fdt->fdt_ofiles[fd].fde_file;
3254 if (__predict_false(fp == NULL))
3255 return (EBADF);
3256 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3257 fdt = atomic_load_ptr(&fdp->fd_files);
3258 continue;
3259 }
3260 /*
3261 * Use an acquire barrier to force re-reading of fdt so it is
3262 * refreshed for verification.
3263 */
3264 atomic_thread_fence_acq();
3265 fdt = fdp->fd_files;
3266 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3267 break;
3268 fdrop(fp, td);
3269 }
3270 *fpp = fp;
3271 return (0);
3272 }
3273 #endif
3274
3275 /*
3276 * See the comments in fget_unlocked_seq for an explanation of how this works.
3277 *
3278 * This is a simplified variant which bails out to the aforementioned routine
3279 * if anything goes wrong. In practice this only happens when userspace is
3280 * racing with itself.
3281 */
3282 int
fget_unlocked(struct thread * td,int fd,cap_rights_t * needrightsp,struct file ** fpp)3283 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp,
3284 struct file **fpp)
3285 {
3286 struct filedesc *fdp;
3287 #ifdef CAPABILITIES
3288 const struct filedescent *fde;
3289 #endif
3290 const struct fdescenttbl *fdt;
3291 struct file *fp;
3292 #ifdef CAPABILITIES
3293 seqc_t seq;
3294 const cap_rights_t *haverights;
3295 #endif
3296
3297 fdp = td->td_proc->p_fd;
3298 fdt = fdp->fd_files;
3299 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3300 *fpp = NULL;
3301 return (EBADF);
3302 }
3303 #ifdef CAPABILITIES
3304 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3305 fde = &fdt->fdt_ofiles[fd];
3306 haverights = cap_rights_fde_inline(fde);
3307 fp = fde->fde_file;
3308 #else
3309 fp = fdt->fdt_ofiles[fd].fde_file;
3310 #endif
3311 if (__predict_false(fp == NULL))
3312 goto out_fallback;
3313 #ifdef CAPABILITIES
3314 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3315 goto out_fallback;
3316 #endif
3317 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3318 goto out_fallback;
3319
3320 /*
3321 * Use an acquire barrier to force re-reading of fdt so it is
3322 * refreshed for verification.
3323 */
3324 atomic_thread_fence_acq();
3325 fdt = fdp->fd_files;
3326 #ifdef CAPABILITIES
3327 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3328 #else
3329 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3330 #endif
3331 goto out_fdrop;
3332 *fpp = fp;
3333 return (0);
3334 out_fdrop:
3335 fdrop(fp, td);
3336 out_fallback:
3337 *fpp = NULL;
3338 return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL));
3339 }
3340
3341 /*
3342 * Translate fd -> file when the caller guarantees the file descriptor table
3343 * can't be changed by others.
3344 *
3345 * Note this does not mean the file object itself is only visible to the caller,
3346 * merely that it wont disappear without having to be referenced.
3347 *
3348 * Must be paired with fput_only_user.
3349 */
3350 #ifdef CAPABILITIES
3351 int
fget_only_user(struct filedesc * fdp,int fd,cap_rights_t * needrightsp,struct file ** fpp)3352 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3353 struct file **fpp)
3354 {
3355 const struct filedescent *fde;
3356 const struct fdescenttbl *fdt;
3357 const cap_rights_t *haverights;
3358 struct file *fp;
3359 int error;
3360
3361 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3362
3363 *fpp = NULL;
3364 if (__predict_false(fd >= fdp->fd_nfiles))
3365 return (EBADF);
3366
3367 fdt = fdp->fd_files;
3368 fde = &fdt->fdt_ofiles[fd];
3369 fp = fde->fde_file;
3370 if (__predict_false(fp == NULL))
3371 return (EBADF);
3372 MPASS(refcount_load(&fp->f_count) > 0);
3373 haverights = cap_rights_fde_inline(fde);
3374 error = cap_check_inline(haverights, needrightsp);
3375 if (__predict_false(error != 0))
3376 return (error);
3377 *fpp = fp;
3378 return (0);
3379 }
3380 #else
3381 int
fget_only_user(struct filedesc * fdp,int fd,cap_rights_t * needrightsp,struct file ** fpp)3382 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3383 struct file **fpp)
3384 {
3385 struct file *fp;
3386
3387 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3388
3389 *fpp = NULL;
3390 if (__predict_false(fd >= fdp->fd_nfiles))
3391 return (EBADF);
3392
3393 fp = fdp->fd_ofiles[fd].fde_file;
3394 if (__predict_false(fp == NULL))
3395 return (EBADF);
3396
3397 MPASS(refcount_load(&fp->f_count) > 0);
3398 *fpp = fp;
3399 return (0);
3400 }
3401 #endif
3402
3403 /*
3404 * Extract the file pointer associated with the specified descriptor for the
3405 * current user process.
3406 *
3407 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3408 * returned.
3409 *
3410 * File's rights will be checked against the capability rights mask.
3411 *
3412 * If an error occurred the non-zero error is returned and *fpp is set to
3413 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3414 * responsible for fdrop().
3415 */
3416 static __inline int
_fget(struct thread * td,int fd,struct file ** fpp,int flags,cap_rights_t * needrightsp)3417 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3418 cap_rights_t *needrightsp)
3419 {
3420 struct file *fp;
3421 int error;
3422
3423 *fpp = NULL;
3424 error = fget_unlocked(td, fd, needrightsp, &fp);
3425 if (__predict_false(error != 0))
3426 return (error);
3427 if (__predict_false(fp->f_ops == &badfileops)) {
3428 fdrop(fp, td);
3429 return (EBADF);
3430 }
3431
3432 /*
3433 * FREAD and FWRITE failure return EBADF as per POSIX.
3434 */
3435 error = 0;
3436 switch (flags) {
3437 case FREAD:
3438 case FWRITE:
3439 if ((fp->f_flag & flags) == 0)
3440 error = EBADF;
3441 break;
3442 case FEXEC:
3443 if (fp->f_ops != &path_fileops &&
3444 ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3445 (fp->f_flag & FWRITE) != 0))
3446 error = EBADF;
3447 break;
3448 case 0:
3449 break;
3450 default:
3451 KASSERT(0, ("wrong flags"));
3452 }
3453
3454 if (error != 0) {
3455 fdrop(fp, td);
3456 return (error);
3457 }
3458
3459 *fpp = fp;
3460 return (0);
3461 }
3462
3463 int
fget(struct thread * td,int fd,cap_rights_t * rightsp,struct file ** fpp)3464 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3465 {
3466
3467 return (_fget(td, fd, fpp, 0, rightsp));
3468 }
3469
3470 int
fget_mmap(struct thread * td,int fd,cap_rights_t * rightsp,vm_prot_t * maxprotp,struct file ** fpp)3471 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3472 struct file **fpp)
3473 {
3474 int error;
3475 #ifndef CAPABILITIES
3476 error = _fget(td, fd, fpp, 0, rightsp);
3477 if (maxprotp != NULL)
3478 *maxprotp = VM_PROT_ALL;
3479 return (error);
3480 #else
3481 cap_rights_t fdrights;
3482 struct filedesc *fdp;
3483 struct file *fp;
3484 seqc_t seq;
3485
3486 *fpp = NULL;
3487 fdp = td->td_proc->p_fd;
3488 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3489 for (;;) {
3490 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3491 if (__predict_false(error != 0))
3492 return (error);
3493 if (__predict_false(fp->f_ops == &badfileops)) {
3494 fdrop(fp, td);
3495 return (EBADF);
3496 }
3497 if (maxprotp != NULL)
3498 fdrights = *cap_rights(fdp, fd);
3499 if (!fd_modified(fdp, fd, seq))
3500 break;
3501 fdrop(fp, td);
3502 }
3503
3504 /*
3505 * If requested, convert capability rights to access flags.
3506 */
3507 if (maxprotp != NULL)
3508 *maxprotp = cap_rights_to_vmprot(&fdrights);
3509 *fpp = fp;
3510 return (0);
3511 #endif
3512 }
3513
3514 int
fget_read(struct thread * td,int fd,cap_rights_t * rightsp,struct file ** fpp)3515 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3516 {
3517
3518 return (_fget(td, fd, fpp, FREAD, rightsp));
3519 }
3520
3521 int
fget_write(struct thread * td,int fd,cap_rights_t * rightsp,struct file ** fpp)3522 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3523 {
3524
3525 return (_fget(td, fd, fpp, FWRITE, rightsp));
3526 }
3527
3528 int
fget_fcntl(struct thread * td,int fd,cap_rights_t * rightsp,int needfcntl,struct file ** fpp)3529 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3530 struct file **fpp)
3531 {
3532 #ifndef CAPABILITIES
3533 return (fget_unlocked(td, fd, rightsp, fpp));
3534 #else
3535 struct filedesc *fdp = td->td_proc->p_fd;
3536 struct file *fp;
3537 int error;
3538 seqc_t seq;
3539
3540 *fpp = NULL;
3541 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3542 for (;;) {
3543 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3544 if (error != 0)
3545 return (error);
3546 error = cap_fcntl_check(fdp, fd, needfcntl);
3547 if (!fd_modified(fdp, fd, seq))
3548 break;
3549 fdrop(fp, td);
3550 }
3551 if (error != 0) {
3552 fdrop(fp, td);
3553 return (error);
3554 }
3555 *fpp = fp;
3556 return (0);
3557 #endif
3558 }
3559
3560 /*
3561 * Like fget() but loads the underlying vnode, or returns an error if the
3562 * descriptor does not represent a vnode. Note that pipes use vnodes but
3563 * never have VM objects. The returned vnode will be vref()'d.
3564 *
3565 * XXX: what about the unused flags ?
3566 */
3567 static __inline int
_fgetvp(struct thread * td,int fd,int flags,cap_rights_t * needrightsp,struct vnode ** vpp)3568 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3569 struct vnode **vpp)
3570 {
3571 struct file *fp;
3572 int error;
3573
3574 *vpp = NULL;
3575 error = _fget(td, fd, &fp, flags, needrightsp);
3576 if (error != 0)
3577 return (error);
3578 if (fp->f_vnode == NULL) {
3579 error = EINVAL;
3580 } else {
3581 *vpp = fp->f_vnode;
3582 vrefact(*vpp);
3583 }
3584 fdrop(fp, td);
3585
3586 return (error);
3587 }
3588
3589 int
fgetvp(struct thread * td,int fd,cap_rights_t * rightsp,struct vnode ** vpp)3590 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3591 {
3592
3593 return (_fgetvp(td, fd, 0, rightsp, vpp));
3594 }
3595
3596 int
fgetvp_rights(struct thread * td,int fd,cap_rights_t * needrightsp,struct filecaps * havecaps,struct vnode ** vpp)3597 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3598 struct filecaps *havecaps, struct vnode **vpp)
3599 {
3600 struct filecaps caps;
3601 struct file *fp;
3602 int error;
3603
3604 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3605 if (error != 0)
3606 return (error);
3607 if (fp->f_ops == &badfileops) {
3608 error = EBADF;
3609 goto out;
3610 }
3611 if (fp->f_vnode == NULL) {
3612 error = EINVAL;
3613 goto out;
3614 }
3615
3616 *havecaps = caps;
3617 *vpp = fp->f_vnode;
3618 vrefact(*vpp);
3619 fdrop(fp, td);
3620
3621 return (0);
3622 out:
3623 filecaps_free(&caps);
3624 fdrop(fp, td);
3625 return (error);
3626 }
3627
3628 int
fgetvp_read(struct thread * td,int fd,cap_rights_t * rightsp,struct vnode ** vpp)3629 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3630 {
3631
3632 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3633 }
3634
3635 int
fgetvp_exec(struct thread * td,int fd,cap_rights_t * rightsp,struct vnode ** vpp)3636 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3637 {
3638
3639 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3640 }
3641
3642 #ifdef notyet
3643 int
fgetvp_write(struct thread * td,int fd,cap_rights_t * rightsp,struct vnode ** vpp)3644 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3645 struct vnode **vpp)
3646 {
3647
3648 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3649 }
3650 #endif
3651
3652 /*
3653 * Handle the last reference to a file being closed.
3654 *
3655 * Without the noinline attribute clang keeps inlining the func thorough this
3656 * file when fdrop is used.
3657 */
3658 int __noinline
_fdrop(struct file * fp,struct thread * td)3659 _fdrop(struct file *fp, struct thread *td)
3660 {
3661 int error;
3662 #ifdef INVARIANTS
3663 int count;
3664
3665 count = refcount_load(&fp->f_count);
3666 if (count != 0)
3667 panic("fdrop: fp %p count %d", fp, count);
3668 #endif
3669 error = fo_close(fp, td);
3670 atomic_subtract_int(&openfiles, 1);
3671 crfree(fp->f_cred);
3672 free(fp->f_advice, M_FADVISE);
3673 uma_zfree(file_zone, fp);
3674
3675 return (error);
3676 }
3677
3678 /*
3679 * Apply an advisory lock on a file descriptor.
3680 *
3681 * Just attempt to get a record lock of the requested type on the entire file
3682 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3683 */
3684 #ifndef _SYS_SYSPROTO_H_
3685 struct flock_args {
3686 int fd;
3687 int how;
3688 };
3689 #endif
3690 /* ARGSUSED */
3691 int
sys_flock(struct thread * td,struct flock_args * uap)3692 sys_flock(struct thread *td, struct flock_args *uap)
3693 {
3694 struct file *fp;
3695 struct vnode *vp;
3696 struct flock lf;
3697 int error;
3698
3699 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3700 if (error != 0)
3701 return (error);
3702 error = EOPNOTSUPP;
3703 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3704 goto done;
3705 }
3706 if (fp->f_ops == &path_fileops) {
3707 goto done;
3708 }
3709
3710 error = 0;
3711 vp = fp->f_vnode;
3712 lf.l_whence = SEEK_SET;
3713 lf.l_start = 0;
3714 lf.l_len = 0;
3715 if (uap->how & LOCK_UN) {
3716 lf.l_type = F_UNLCK;
3717 atomic_clear_int(&fp->f_flag, FHASLOCK);
3718 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3719 goto done;
3720 }
3721 if (uap->how & LOCK_EX)
3722 lf.l_type = F_WRLCK;
3723 else if (uap->how & LOCK_SH)
3724 lf.l_type = F_RDLCK;
3725 else {
3726 error = EBADF;
3727 goto done;
3728 }
3729 atomic_set_int(&fp->f_flag, FHASLOCK);
3730 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3731 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3732 done:
3733 fdrop(fp, td);
3734 return (error);
3735 }
3736 /*
3737 * Duplicate the specified descriptor to a free descriptor.
3738 */
3739 int
dupfdopen(struct thread * td,struct filedesc * fdp,int dfd,int mode,int openerror,int * indxp)3740 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3741 int openerror, int *indxp)
3742 {
3743 struct filedescent *newfde, *oldfde;
3744 struct file *fp;
3745 u_long *ioctls;
3746 int error, indx;
3747
3748 KASSERT(openerror == ENODEV || openerror == ENXIO,
3749 ("unexpected error %d in %s", openerror, __func__));
3750
3751 /*
3752 * If the to-be-dup'd fd number is greater than the allowed number
3753 * of file descriptors, or the fd to be dup'd has already been
3754 * closed, then reject.
3755 */
3756 FILEDESC_XLOCK(fdp);
3757 if ((fp = fget_noref(fdp, dfd)) == NULL) {
3758 FILEDESC_XUNLOCK(fdp);
3759 return (EBADF);
3760 }
3761
3762 error = fdalloc(td, 0, &indx);
3763 if (error != 0) {
3764 FILEDESC_XUNLOCK(fdp);
3765 return (error);
3766 }
3767
3768 /*
3769 * There are two cases of interest here.
3770 *
3771 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3772 *
3773 * For ENXIO steal away the file structure from (dfd) and store it in
3774 * (indx). (dfd) is effectively closed by this operation.
3775 */
3776 switch (openerror) {
3777 case ENODEV:
3778 /*
3779 * Check that the mode the file is being opened for is a
3780 * subset of the mode of the existing descriptor.
3781 */
3782 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3783 fdunused(fdp, indx);
3784 FILEDESC_XUNLOCK(fdp);
3785 return (EACCES);
3786 }
3787 if (!fhold(fp)) {
3788 fdunused(fdp, indx);
3789 FILEDESC_XUNLOCK(fdp);
3790 return (EBADF);
3791 }
3792 newfde = &fdp->fd_ofiles[indx];
3793 oldfde = &fdp->fd_ofiles[dfd];
3794 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3795 #ifdef CAPABILITIES
3796 seqc_write_begin(&newfde->fde_seqc);
3797 #endif
3798 fde_copy(oldfde, newfde);
3799 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3800 ioctls);
3801 #ifdef CAPABILITIES
3802 seqc_write_end(&newfde->fde_seqc);
3803 #endif
3804 break;
3805 case ENXIO:
3806 /*
3807 * Steal away the file pointer from dfd and stuff it into indx.
3808 */
3809 newfde = &fdp->fd_ofiles[indx];
3810 oldfde = &fdp->fd_ofiles[dfd];
3811 #ifdef CAPABILITIES
3812 seqc_write_begin(&oldfde->fde_seqc);
3813 seqc_write_begin(&newfde->fde_seqc);
3814 #endif
3815 fde_copy(oldfde, newfde);
3816 oldfde->fde_file = NULL;
3817 fdunused(fdp, dfd);
3818 #ifdef CAPABILITIES
3819 seqc_write_end(&newfde->fde_seqc);
3820 seqc_write_end(&oldfde->fde_seqc);
3821 #endif
3822 break;
3823 }
3824 FILEDESC_XUNLOCK(fdp);
3825 *indxp = indx;
3826 return (0);
3827 }
3828
3829 /*
3830 * This sysctl determines if we will allow a process to chroot(2) if it
3831 * has a directory open:
3832 * 0: disallowed for all processes.
3833 * 1: allowed for processes that were not already chroot(2)'ed.
3834 * 2: allowed for all processes.
3835 */
3836
3837 static int chroot_allow_open_directories = 1;
3838
3839 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3840 &chroot_allow_open_directories, 0,
3841 "Allow a process to chroot(2) if it has a directory open");
3842
3843 /*
3844 * Helper function for raised chroot(2) security function: Refuse if
3845 * any filedescriptors are open directories.
3846 */
3847 static int
chroot_refuse_vdir_fds(struct filedesc * fdp)3848 chroot_refuse_vdir_fds(struct filedesc *fdp)
3849 {
3850 struct vnode *vp;
3851 struct file *fp;
3852 int i;
3853
3854 FILEDESC_LOCK_ASSERT(fdp);
3855
3856 FILEDESC_FOREACH_FP(fdp, i, fp) {
3857 if (fp->f_type == DTYPE_VNODE) {
3858 vp = fp->f_vnode;
3859 if (vp->v_type == VDIR)
3860 return (EPERM);
3861 }
3862 }
3863 return (0);
3864 }
3865
3866 static void
pwd_fill(struct pwd * oldpwd,struct pwd * newpwd)3867 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3868 {
3869
3870 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3871 vrefact(oldpwd->pwd_cdir);
3872 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3873 }
3874
3875 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3876 vrefact(oldpwd->pwd_rdir);
3877 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3878 }
3879
3880 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3881 vrefact(oldpwd->pwd_jdir);
3882 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3883 }
3884
3885 if (newpwd->pwd_adir == NULL && oldpwd->pwd_adir != NULL) {
3886 vrefact(oldpwd->pwd_adir);
3887 newpwd->pwd_adir = oldpwd->pwd_adir;
3888 }
3889 }
3890
3891 struct pwd *
pwd_hold_pwddesc(struct pwddesc * pdp)3892 pwd_hold_pwddesc(struct pwddesc *pdp)
3893 {
3894 struct pwd *pwd;
3895
3896 PWDDESC_ASSERT_XLOCKED(pdp);
3897 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3898 if (pwd != NULL)
3899 refcount_acquire(&pwd->pwd_refcount);
3900 return (pwd);
3901 }
3902
3903 bool
pwd_hold_smr(struct pwd * pwd)3904 pwd_hold_smr(struct pwd *pwd)
3905 {
3906
3907 MPASS(pwd != NULL);
3908 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3909 return (true);
3910 }
3911 return (false);
3912 }
3913
3914 struct pwd *
pwd_hold(struct thread * td)3915 pwd_hold(struct thread *td)
3916 {
3917 struct pwddesc *pdp;
3918 struct pwd *pwd;
3919
3920 pdp = td->td_proc->p_pd;
3921
3922 vfs_smr_enter();
3923 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3924 if (pwd_hold_smr(pwd)) {
3925 vfs_smr_exit();
3926 return (pwd);
3927 }
3928 vfs_smr_exit();
3929 PWDDESC_XLOCK(pdp);
3930 pwd = pwd_hold_pwddesc(pdp);
3931 MPASS(pwd != NULL);
3932 PWDDESC_XUNLOCK(pdp);
3933 return (pwd);
3934 }
3935
3936 struct pwd *
pwd_hold_proc(struct proc * p)3937 pwd_hold_proc(struct proc *p)
3938 {
3939 struct pwddesc *pdp;
3940 struct pwd *pwd;
3941
3942 PROC_ASSERT_HELD(p);
3943 PROC_LOCK(p);
3944 pdp = pdhold(p);
3945 MPASS(pdp != NULL);
3946 PROC_UNLOCK(p);
3947
3948 PWDDESC_XLOCK(pdp);
3949 pwd = pwd_hold_pwddesc(pdp);
3950 MPASS(pwd != NULL);
3951 PWDDESC_XUNLOCK(pdp);
3952 pddrop(pdp);
3953 return (pwd);
3954 }
3955
3956 static struct pwd *
pwd_alloc(void)3957 pwd_alloc(void)
3958 {
3959 struct pwd *pwd;
3960
3961 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3962 bzero(pwd, sizeof(*pwd));
3963 refcount_init(&pwd->pwd_refcount, 1);
3964 return (pwd);
3965 }
3966
3967 void
pwd_drop(struct pwd * pwd)3968 pwd_drop(struct pwd *pwd)
3969 {
3970
3971 if (!refcount_release(&pwd->pwd_refcount))
3972 return;
3973
3974 if (pwd->pwd_cdir != NULL)
3975 vrele(pwd->pwd_cdir);
3976 if (pwd->pwd_rdir != NULL)
3977 vrele(pwd->pwd_rdir);
3978 if (pwd->pwd_jdir != NULL)
3979 vrele(pwd->pwd_jdir);
3980 if (pwd->pwd_adir != NULL)
3981 vrele(pwd->pwd_adir);
3982 uma_zfree_smr(pwd_zone, pwd);
3983 }
3984
3985 /*
3986 * The caller is responsible for invoking priv_check() and
3987 * mac_vnode_check_chroot() to authorize this operation.
3988 */
3989 int
pwd_chroot(struct thread * td,struct vnode * vp)3990 pwd_chroot(struct thread *td, struct vnode *vp)
3991 {
3992 struct pwddesc *pdp;
3993 struct filedesc *fdp;
3994 struct pwd *newpwd, *oldpwd;
3995 int error;
3996
3997 fdp = td->td_proc->p_fd;
3998 pdp = td->td_proc->p_pd;
3999 newpwd = pwd_alloc();
4000 FILEDESC_SLOCK(fdp);
4001 PWDDESC_XLOCK(pdp);
4002 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4003 if (chroot_allow_open_directories == 0 ||
4004 (chroot_allow_open_directories == 1 &&
4005 oldpwd->pwd_rdir != rootvnode)) {
4006 error = chroot_refuse_vdir_fds(fdp);
4007 FILEDESC_SUNLOCK(fdp);
4008 if (error != 0) {
4009 PWDDESC_XUNLOCK(pdp);
4010 pwd_drop(newpwd);
4011 return (error);
4012 }
4013 } else {
4014 FILEDESC_SUNLOCK(fdp);
4015 }
4016
4017 vrefact(vp);
4018 newpwd->pwd_rdir = vp;
4019 vrefact(vp);
4020 newpwd->pwd_adir = vp;
4021 if (oldpwd->pwd_jdir == NULL) {
4022 vrefact(vp);
4023 newpwd->pwd_jdir = vp;
4024 }
4025 pwd_fill(oldpwd, newpwd);
4026 pwd_set(pdp, newpwd);
4027 PWDDESC_XUNLOCK(pdp);
4028 pwd_drop(oldpwd);
4029 return (0);
4030 }
4031
4032 void
pwd_chdir(struct thread * td,struct vnode * vp)4033 pwd_chdir(struct thread *td, struct vnode *vp)
4034 {
4035 struct pwddesc *pdp;
4036 struct pwd *newpwd, *oldpwd;
4037
4038 VNPASS(vp->v_usecount > 0, vp);
4039
4040 newpwd = pwd_alloc();
4041 pdp = td->td_proc->p_pd;
4042 PWDDESC_XLOCK(pdp);
4043 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4044 newpwd->pwd_cdir = vp;
4045 pwd_fill(oldpwd, newpwd);
4046 pwd_set(pdp, newpwd);
4047 PWDDESC_XUNLOCK(pdp);
4048 pwd_drop(oldpwd);
4049 }
4050
4051 /*
4052 * Process is transitioning to/from a non-native ABI.
4053 */
4054 void
pwd_altroot(struct thread * td,struct vnode * altroot_vp)4055 pwd_altroot(struct thread *td, struct vnode *altroot_vp)
4056 {
4057 struct pwddesc *pdp;
4058 struct pwd *newpwd, *oldpwd;
4059
4060 newpwd = pwd_alloc();
4061 pdp = td->td_proc->p_pd;
4062 PWDDESC_XLOCK(pdp);
4063 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4064 if (altroot_vp != NULL) {
4065 /*
4066 * Native process to a non-native ABI.
4067 */
4068
4069 vrefact(altroot_vp);
4070 newpwd->pwd_adir = altroot_vp;
4071 } else {
4072 /*
4073 * Non-native process to the native ABI.
4074 */
4075
4076 vrefact(oldpwd->pwd_rdir);
4077 newpwd->pwd_adir = oldpwd->pwd_rdir;
4078 }
4079 pwd_fill(oldpwd, newpwd);
4080 pwd_set(pdp, newpwd);
4081 PWDDESC_XUNLOCK(pdp);
4082 pwd_drop(oldpwd);
4083 }
4084
4085 /*
4086 * jail_attach(2) changes both root and working directories.
4087 */
4088 int
pwd_chroot_chdir(struct thread * td,struct vnode * vp)4089 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
4090 {
4091 struct pwddesc *pdp;
4092 struct filedesc *fdp;
4093 struct pwd *newpwd, *oldpwd;
4094 int error;
4095
4096 fdp = td->td_proc->p_fd;
4097 pdp = td->td_proc->p_pd;
4098 newpwd = pwd_alloc();
4099 FILEDESC_SLOCK(fdp);
4100 PWDDESC_XLOCK(pdp);
4101 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4102 error = chroot_refuse_vdir_fds(fdp);
4103 FILEDESC_SUNLOCK(fdp);
4104 if (error != 0) {
4105 PWDDESC_XUNLOCK(pdp);
4106 pwd_drop(newpwd);
4107 return (error);
4108 }
4109
4110 vrefact(vp);
4111 newpwd->pwd_rdir = vp;
4112 vrefact(vp);
4113 newpwd->pwd_cdir = vp;
4114 if (oldpwd->pwd_jdir == NULL) {
4115 vrefact(vp);
4116 newpwd->pwd_jdir = vp;
4117 }
4118 vrefact(vp);
4119 newpwd->pwd_adir = vp;
4120 pwd_fill(oldpwd, newpwd);
4121 pwd_set(pdp, newpwd);
4122 PWDDESC_XUNLOCK(pdp);
4123 pwd_drop(oldpwd);
4124 return (0);
4125 }
4126
4127 void
pwd_ensure_dirs(void)4128 pwd_ensure_dirs(void)
4129 {
4130 struct pwddesc *pdp;
4131 struct pwd *oldpwd, *newpwd;
4132
4133 pdp = curproc->p_pd;
4134 PWDDESC_XLOCK(pdp);
4135 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4136 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL &&
4137 oldpwd->pwd_adir != NULL) {
4138 PWDDESC_XUNLOCK(pdp);
4139 return;
4140 }
4141 PWDDESC_XUNLOCK(pdp);
4142
4143 newpwd = pwd_alloc();
4144 PWDDESC_XLOCK(pdp);
4145 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4146 pwd_fill(oldpwd, newpwd);
4147 if (newpwd->pwd_cdir == NULL) {
4148 vrefact(rootvnode);
4149 newpwd->pwd_cdir = rootvnode;
4150 }
4151 if (newpwd->pwd_rdir == NULL) {
4152 vrefact(rootvnode);
4153 newpwd->pwd_rdir = rootvnode;
4154 }
4155 if (newpwd->pwd_adir == NULL) {
4156 vrefact(rootvnode);
4157 newpwd->pwd_adir = rootvnode;
4158 }
4159 pwd_set(pdp, newpwd);
4160 PWDDESC_XUNLOCK(pdp);
4161 pwd_drop(oldpwd);
4162 }
4163
4164 void
pwd_set_rootvnode(void)4165 pwd_set_rootvnode(void)
4166 {
4167 struct pwddesc *pdp;
4168 struct pwd *oldpwd, *newpwd;
4169
4170 pdp = curproc->p_pd;
4171
4172 newpwd = pwd_alloc();
4173 PWDDESC_XLOCK(pdp);
4174 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4175 vrefact(rootvnode);
4176 newpwd->pwd_cdir = rootvnode;
4177 vrefact(rootvnode);
4178 newpwd->pwd_rdir = rootvnode;
4179 vrefact(rootvnode);
4180 newpwd->pwd_adir = rootvnode;
4181 pwd_fill(oldpwd, newpwd);
4182 pwd_set(pdp, newpwd);
4183 PWDDESC_XUNLOCK(pdp);
4184 pwd_drop(oldpwd);
4185 }
4186
4187 /*
4188 * Scan all active processes and prisons to see if any of them have a current
4189 * or root directory of `olddp'. If so, replace them with the new mount point.
4190 */
4191 void
mountcheckdirs(struct vnode * olddp,struct vnode * newdp)4192 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
4193 {
4194 struct pwddesc *pdp;
4195 struct pwd *newpwd, *oldpwd;
4196 struct prison *pr;
4197 struct proc *p;
4198 int nrele;
4199
4200 if (vrefcnt(olddp) == 1)
4201 return;
4202 nrele = 0;
4203 newpwd = pwd_alloc();
4204 sx_slock(&allproc_lock);
4205 FOREACH_PROC_IN_SYSTEM(p) {
4206 PROC_LOCK(p);
4207 pdp = pdhold(p);
4208 PROC_UNLOCK(p);
4209 if (pdp == NULL)
4210 continue;
4211 PWDDESC_XLOCK(pdp);
4212 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4213 if (oldpwd == NULL ||
4214 (oldpwd->pwd_cdir != olddp &&
4215 oldpwd->pwd_rdir != olddp &&
4216 oldpwd->pwd_jdir != olddp &&
4217 oldpwd->pwd_adir != olddp)) {
4218 PWDDESC_XUNLOCK(pdp);
4219 pddrop(pdp);
4220 continue;
4221 }
4222 if (oldpwd->pwd_cdir == olddp) {
4223 vrefact(newdp);
4224 newpwd->pwd_cdir = newdp;
4225 }
4226 if (oldpwd->pwd_rdir == olddp) {
4227 vrefact(newdp);
4228 newpwd->pwd_rdir = newdp;
4229 }
4230 if (oldpwd->pwd_jdir == olddp) {
4231 vrefact(newdp);
4232 newpwd->pwd_jdir = newdp;
4233 }
4234 if (oldpwd->pwd_adir == olddp) {
4235 vrefact(newdp);
4236 newpwd->pwd_adir = newdp;
4237 }
4238 pwd_fill(oldpwd, newpwd);
4239 pwd_set(pdp, newpwd);
4240 PWDDESC_XUNLOCK(pdp);
4241 pwd_drop(oldpwd);
4242 pddrop(pdp);
4243 newpwd = pwd_alloc();
4244 }
4245 sx_sunlock(&allproc_lock);
4246 pwd_drop(newpwd);
4247 if (rootvnode == olddp) {
4248 vrefact(newdp);
4249 rootvnode = newdp;
4250 nrele++;
4251 }
4252 mtx_lock(&prison0.pr_mtx);
4253 if (prison0.pr_root == olddp) {
4254 vrefact(newdp);
4255 prison0.pr_root = newdp;
4256 nrele++;
4257 }
4258 mtx_unlock(&prison0.pr_mtx);
4259 sx_slock(&allprison_lock);
4260 TAILQ_FOREACH(pr, &allprison, pr_list) {
4261 mtx_lock(&pr->pr_mtx);
4262 if (pr->pr_root == olddp) {
4263 vrefact(newdp);
4264 pr->pr_root = newdp;
4265 nrele++;
4266 }
4267 mtx_unlock(&pr->pr_mtx);
4268 }
4269 sx_sunlock(&allprison_lock);
4270 while (nrele--)
4271 vrele(olddp);
4272 }
4273
4274 int
descrip_check_write_mp(struct filedesc * fdp,struct mount * mp)4275 descrip_check_write_mp(struct filedesc *fdp, struct mount *mp)
4276 {
4277 struct file *fp;
4278 struct vnode *vp;
4279 int error, i;
4280
4281 error = 0;
4282 FILEDESC_SLOCK(fdp);
4283 FILEDESC_FOREACH_FP(fdp, i, fp) {
4284 if (fp->f_type != DTYPE_VNODE ||
4285 (atomic_load_int(&fp->f_flag) & FWRITE) == 0)
4286 continue;
4287 vp = fp->f_vnode;
4288 if (vp->v_mount == mp) {
4289 error = EDEADLK;
4290 break;
4291 }
4292 }
4293 FILEDESC_SUNLOCK(fdp);
4294 return (error);
4295 }
4296
4297 struct filedesc_to_leader *
filedesc_to_leader_alloc(struct filedesc_to_leader * old,struct filedesc * fdp,struct proc * leader)4298 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp,
4299 struct proc *leader)
4300 {
4301 struct filedesc_to_leader *fdtol;
4302
4303 fdtol = malloc(sizeof(struct filedesc_to_leader),
4304 M_FILEDESC_TO_LEADER, M_WAITOK);
4305 fdtol->fdl_refcount = 1;
4306 fdtol->fdl_holdcount = 0;
4307 fdtol->fdl_wakeup = 0;
4308 fdtol->fdl_leader = leader;
4309 if (old != NULL) {
4310 FILEDESC_XLOCK(fdp);
4311 fdtol->fdl_next = old->fdl_next;
4312 fdtol->fdl_prev = old;
4313 old->fdl_next = fdtol;
4314 fdtol->fdl_next->fdl_prev = fdtol;
4315 FILEDESC_XUNLOCK(fdp);
4316 } else {
4317 fdtol->fdl_next = fdtol;
4318 fdtol->fdl_prev = fdtol;
4319 }
4320 return (fdtol);
4321 }
4322
4323 struct filedesc_to_leader *
filedesc_to_leader_share(struct filedesc_to_leader * fdtol,struct filedesc * fdp)4324 filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp)
4325 {
4326 FILEDESC_XLOCK(fdp);
4327 fdtol->fdl_refcount++;
4328 FILEDESC_XUNLOCK(fdp);
4329 return (fdtol);
4330 }
4331
4332 static int
filedesc_nfiles(struct filedesc * fdp)4333 filedesc_nfiles(struct filedesc *fdp)
4334 {
4335 NDSLOTTYPE *map;
4336 int count, off, minoff;
4337
4338 if (fdp == NULL)
4339 return (0);
4340 count = 0;
4341 FILEDESC_SLOCK(fdp);
4342 map = fdp->fd_map;
4343 off = NDSLOT(fdp->fd_nfiles - 1);
4344 for (minoff = NDSLOT(0); off >= minoff; --off)
4345 count += bitcountl(map[off]);
4346 FILEDESC_SUNLOCK(fdp);
4347 return (count);
4348 }
4349
4350 int
proc_nfiles(struct proc * p)4351 proc_nfiles(struct proc *p)
4352 {
4353 struct filedesc *fdp;
4354 int res;
4355
4356 PROC_LOCK(p);
4357 fdp = fdhold(p);
4358 PROC_UNLOCK(p);
4359 res = filedesc_nfiles(fdp);
4360 fddrop(fdp);
4361 return (res);
4362 }
4363
4364 static int
sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)4365 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4366 {
4367 u_int namelen;
4368 int count;
4369
4370 namelen = arg2;
4371 if (namelen != 1)
4372 return (EINVAL);
4373
4374 if (*(int *)arg1 != 0)
4375 return (EINVAL);
4376
4377 count = filedesc_nfiles(curproc->p_fd);
4378 return (SYSCTL_OUT(req, &count, sizeof(count)));
4379 }
4380
4381 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4382 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4383 "Number of open file descriptors");
4384
4385 /*
4386 * Get file structures globally.
4387 */
4388 static int
sysctl_kern_file(SYSCTL_HANDLER_ARGS)4389 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4390 {
4391 struct xfile xf;
4392 struct filedesc *fdp;
4393 struct file *fp;
4394 struct proc *p;
4395 int error, n;
4396
4397 error = sysctl_wire_old_buffer(req, 0);
4398 if (error != 0)
4399 return (error);
4400 if (req->oldptr == NULL) {
4401 n = 0;
4402 sx_slock(&allproc_lock);
4403 FOREACH_PROC_IN_SYSTEM(p) {
4404 PROC_LOCK(p);
4405 if (p->p_state == PRS_NEW) {
4406 PROC_UNLOCK(p);
4407 continue;
4408 }
4409 fdp = fdhold(p);
4410 PROC_UNLOCK(p);
4411 if (fdp == NULL)
4412 continue;
4413 /* overestimates sparse tables. */
4414 n += fdp->fd_nfiles;
4415 fddrop(fdp);
4416 }
4417 sx_sunlock(&allproc_lock);
4418 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4419 }
4420 error = 0;
4421 bzero(&xf, sizeof(xf));
4422 xf.xf_size = sizeof(xf);
4423 sx_slock(&allproc_lock);
4424 FOREACH_PROC_IN_SYSTEM(p) {
4425 PROC_LOCK(p);
4426 if (p->p_state == PRS_NEW) {
4427 PROC_UNLOCK(p);
4428 continue;
4429 }
4430 if (p_cansee(req->td, p) != 0) {
4431 PROC_UNLOCK(p);
4432 continue;
4433 }
4434 xf.xf_pid = p->p_pid;
4435 xf.xf_uid = p->p_ucred->cr_uid;
4436 fdp = fdhold(p);
4437 PROC_UNLOCK(p);
4438 if (fdp == NULL)
4439 continue;
4440 FILEDESC_SLOCK(fdp);
4441 if (refcount_load(&fdp->fd_refcnt) == 0)
4442 goto nextproc;
4443 FILEDESC_FOREACH_FP(fdp, n, fp) {
4444 xf.xf_fd = n;
4445 xf.xf_file = (uintptr_t)fp;
4446 xf.xf_data = (uintptr_t)fp->f_data;
4447 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4448 xf.xf_type = (uintptr_t)fp->f_type;
4449 xf.xf_count = refcount_load(&fp->f_count);
4450 xf.xf_msgcount = 0;
4451 xf.xf_offset = foffset_get(fp);
4452 xf.xf_flag = fp->f_flag;
4453 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4454
4455 /*
4456 * There is no need to re-check the fdtable refcount
4457 * here since the filedesc lock is not dropped in the
4458 * loop body.
4459 */
4460 if (error != 0)
4461 break;
4462 }
4463 nextproc:
4464 FILEDESC_SUNLOCK(fdp);
4465 fddrop(fdp);
4466 if (error)
4467 break;
4468 }
4469 sx_sunlock(&allproc_lock);
4470 return (error);
4471 }
4472
4473 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4474 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4475
4476 #ifdef KINFO_FILE_SIZE
4477 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4478 #endif
4479
4480 static int
xlate_fflags(int fflags)4481 xlate_fflags(int fflags)
4482 {
4483 static const struct {
4484 int fflag;
4485 int kf_fflag;
4486 } fflags_table[] = {
4487 { FAPPEND, KF_FLAG_APPEND },
4488 { FASYNC, KF_FLAG_ASYNC },
4489 { FFSYNC, KF_FLAG_FSYNC },
4490 { FHASLOCK, KF_FLAG_HASLOCK },
4491 { FNONBLOCK, KF_FLAG_NONBLOCK },
4492 { FREAD, KF_FLAG_READ },
4493 { FWRITE, KF_FLAG_WRITE },
4494 { O_CREAT, KF_FLAG_CREAT },
4495 { O_DIRECT, KF_FLAG_DIRECT },
4496 { O_EXCL, KF_FLAG_EXCL },
4497 { O_EXEC, KF_FLAG_EXEC },
4498 { O_EXLOCK, KF_FLAG_EXLOCK },
4499 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4500 { O_SHLOCK, KF_FLAG_SHLOCK },
4501 { O_TRUNC, KF_FLAG_TRUNC }
4502 };
4503 unsigned int i;
4504 int kflags;
4505
4506 kflags = 0;
4507 for (i = 0; i < nitems(fflags_table); i++)
4508 if (fflags & fflags_table[i].fflag)
4509 kflags |= fflags_table[i].kf_fflag;
4510 return (kflags);
4511 }
4512
4513 /* Trim unused data from kf_path by truncating the structure size. */
4514 void
pack_kinfo(struct kinfo_file * kif)4515 pack_kinfo(struct kinfo_file *kif)
4516 {
4517
4518 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4519 strlen(kif->kf_path) + 1;
4520 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4521 }
4522
4523 static void
export_file_to_kinfo(struct file * fp,int fd,cap_rights_t * rightsp,struct kinfo_file * kif,struct filedesc * fdp,int flags)4524 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4525 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4526 {
4527 int error;
4528
4529 bzero(kif, sizeof(*kif));
4530
4531 /* Set a default type to allow for empty fill_kinfo() methods. */
4532 kif->kf_type = KF_TYPE_UNKNOWN;
4533 kif->kf_flags = xlate_fflags(fp->f_flag);
4534 if (rightsp != NULL)
4535 kif->kf_cap_rights = *rightsp;
4536 else
4537 cap_rights_init_zero(&kif->kf_cap_rights);
4538 kif->kf_fd = fd;
4539 kif->kf_ref_count = refcount_load(&fp->f_count);
4540 kif->kf_offset = foffset_get(fp);
4541
4542 /*
4543 * This may drop the filedesc lock, so the 'fp' cannot be
4544 * accessed after this call.
4545 */
4546 error = fo_fill_kinfo(fp, kif, fdp);
4547 if (error == 0)
4548 kif->kf_status |= KF_ATTR_VALID;
4549 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4550 pack_kinfo(kif);
4551 else
4552 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4553 }
4554
4555 static void
export_vnode_to_kinfo(struct vnode * vp,int fd,int fflags,struct kinfo_file * kif,int flags)4556 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4557 struct kinfo_file *kif, int flags)
4558 {
4559 int error;
4560
4561 bzero(kif, sizeof(*kif));
4562
4563 kif->kf_type = KF_TYPE_VNODE;
4564 error = vn_fill_kinfo_vnode(vp, kif);
4565 if (error == 0)
4566 kif->kf_status |= KF_ATTR_VALID;
4567 kif->kf_flags = xlate_fflags(fflags);
4568 cap_rights_init_zero(&kif->kf_cap_rights);
4569 kif->kf_fd = fd;
4570 kif->kf_ref_count = -1;
4571 kif->kf_offset = -1;
4572 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4573 pack_kinfo(kif);
4574 else
4575 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4576 vrele(vp);
4577 }
4578
4579 struct export_fd_buf {
4580 struct filedesc *fdp;
4581 struct pwddesc *pdp;
4582 struct sbuf *sb;
4583 ssize_t remainder;
4584 struct kinfo_file kif;
4585 int flags;
4586 };
4587
4588 static int
export_kinfo_to_sb(struct export_fd_buf * efbuf)4589 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4590 {
4591 struct kinfo_file *kif;
4592
4593 kif = &efbuf->kif;
4594 if (efbuf->remainder != -1) {
4595 if (efbuf->remainder < kif->kf_structsize)
4596 return (ENOMEM);
4597 efbuf->remainder -= kif->kf_structsize;
4598 }
4599 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4600 return (sbuf_error(efbuf->sb));
4601 return (0);
4602 }
4603
4604 static int
export_file_to_sb(struct file * fp,int fd,cap_rights_t * rightsp,struct export_fd_buf * efbuf)4605 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4606 struct export_fd_buf *efbuf)
4607 {
4608 int error;
4609
4610 if (efbuf->remainder == 0)
4611 return (ENOMEM);
4612 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4613 efbuf->flags);
4614 FILEDESC_SUNLOCK(efbuf->fdp);
4615 error = export_kinfo_to_sb(efbuf);
4616 FILEDESC_SLOCK(efbuf->fdp);
4617 return (error);
4618 }
4619
4620 static int
export_vnode_to_sb(struct vnode * vp,int fd,int fflags,struct export_fd_buf * efbuf)4621 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4622 struct export_fd_buf *efbuf)
4623 {
4624 int error;
4625
4626 if (efbuf->remainder == 0)
4627 return (ENOMEM);
4628 if (efbuf->pdp != NULL)
4629 PWDDESC_XUNLOCK(efbuf->pdp);
4630 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4631 error = export_kinfo_to_sb(efbuf);
4632 if (efbuf->pdp != NULL)
4633 PWDDESC_XLOCK(efbuf->pdp);
4634 return (error);
4635 }
4636
4637 /*
4638 * Store a process file descriptor information to sbuf.
4639 *
4640 * Takes a locked proc as argument, and returns with the proc unlocked.
4641 */
4642 int
kern_proc_filedesc_out(struct proc * p,struct sbuf * sb,ssize_t maxlen,int flags)4643 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4644 int flags)
4645 {
4646 struct file *fp;
4647 struct filedesc *fdp;
4648 struct pwddesc *pdp;
4649 struct export_fd_buf *efbuf;
4650 struct vnode *cttyvp, *textvp, *tracevp;
4651 struct pwd *pwd;
4652 int error, i;
4653 cap_rights_t rights;
4654
4655 PROC_LOCK_ASSERT(p, MA_OWNED);
4656
4657 /* ktrace vnode */
4658 tracevp = ktr_get_tracevp(p, true);
4659 /* text vnode */
4660 textvp = p->p_textvp;
4661 if (textvp != NULL)
4662 vrefact(textvp);
4663 /* Controlling tty. */
4664 cttyvp = NULL;
4665 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4666 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4667 if (cttyvp != NULL)
4668 vrefact(cttyvp);
4669 }
4670 fdp = fdhold(p);
4671 pdp = pdhold(p);
4672 PROC_UNLOCK(p);
4673
4674 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4675 efbuf->fdp = NULL;
4676 efbuf->pdp = NULL;
4677 efbuf->sb = sb;
4678 efbuf->remainder = maxlen;
4679 efbuf->flags = flags;
4680
4681 error = 0;
4682 if (tracevp != NULL)
4683 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4684 FREAD | FWRITE, efbuf);
4685 if (error == 0 && textvp != NULL)
4686 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4687 efbuf);
4688 if (error == 0 && cttyvp != NULL)
4689 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4690 FREAD | FWRITE, efbuf);
4691 if (error != 0 || pdp == NULL || fdp == NULL)
4692 goto fail;
4693 efbuf->fdp = fdp;
4694 efbuf->pdp = pdp;
4695 PWDDESC_XLOCK(pdp);
4696 pwd = pwd_hold_pwddesc(pdp);
4697 if (pwd != NULL) {
4698 /* working directory */
4699 if (pwd->pwd_cdir != NULL) {
4700 vrefact(pwd->pwd_cdir);
4701 error = export_vnode_to_sb(pwd->pwd_cdir,
4702 KF_FD_TYPE_CWD, FREAD, efbuf);
4703 }
4704 /* root directory */
4705 if (error == 0 && pwd->pwd_rdir != NULL) {
4706 vrefact(pwd->pwd_rdir);
4707 error = export_vnode_to_sb(pwd->pwd_rdir,
4708 KF_FD_TYPE_ROOT, FREAD, efbuf);
4709 }
4710 /* jail directory */
4711 if (error == 0 && pwd->pwd_jdir != NULL) {
4712 vrefact(pwd->pwd_jdir);
4713 error = export_vnode_to_sb(pwd->pwd_jdir,
4714 KF_FD_TYPE_JAIL, FREAD, efbuf);
4715 }
4716 }
4717 PWDDESC_XUNLOCK(pdp);
4718 if (error != 0)
4719 goto fail;
4720 if (pwd != NULL)
4721 pwd_drop(pwd);
4722 FILEDESC_SLOCK(fdp);
4723 if (refcount_load(&fdp->fd_refcnt) == 0)
4724 goto skip;
4725 FILEDESC_FOREACH_FP(fdp, i, fp) {
4726 #ifdef CAPABILITIES
4727 rights = *cap_rights(fdp, i);
4728 #else /* !CAPABILITIES */
4729 rights = cap_no_rights;
4730 #endif
4731 /*
4732 * Create sysctl entry. It is OK to drop the filedesc
4733 * lock inside of export_file_to_sb() as we will
4734 * re-validate and re-evaluate its properties when the
4735 * loop continues.
4736 */
4737 error = export_file_to_sb(fp, i, &rights, efbuf);
4738 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4739 break;
4740 }
4741 skip:
4742 FILEDESC_SUNLOCK(fdp);
4743 fail:
4744 if (fdp != NULL)
4745 fddrop(fdp);
4746 if (pdp != NULL)
4747 pddrop(pdp);
4748 free(efbuf, M_TEMP);
4749 return (error);
4750 }
4751
4752 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4753
4754 /*
4755 * Get per-process file descriptors for use by procstat(1), et al.
4756 */
4757 static int
sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)4758 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4759 {
4760 struct sbuf sb;
4761 struct proc *p;
4762 ssize_t maxlen;
4763 u_int namelen;
4764 int error, error2, *name;
4765
4766 namelen = arg2;
4767 if (namelen != 1)
4768 return (EINVAL);
4769
4770 name = (int *)arg1;
4771
4772 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4773 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4774 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4775 if (error != 0) {
4776 sbuf_delete(&sb);
4777 return (error);
4778 }
4779 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4780 error = kern_proc_filedesc_out(p, &sb, maxlen,
4781 KERN_FILEDESC_PACK_KINFO);
4782 error2 = sbuf_finish(&sb);
4783 sbuf_delete(&sb);
4784 return (error != 0 ? error : error2);
4785 }
4786
4787 #ifdef COMPAT_FREEBSD7
4788 #ifdef KINFO_OFILE_SIZE
4789 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4790 #endif
4791
4792 static void
kinfo_to_okinfo(struct kinfo_file * kif,struct kinfo_ofile * okif)4793 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4794 {
4795
4796 okif->kf_structsize = sizeof(*okif);
4797 okif->kf_type = kif->kf_type;
4798 okif->kf_fd = kif->kf_fd;
4799 okif->kf_ref_count = kif->kf_ref_count;
4800 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4801 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4802 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4803 okif->kf_offset = kif->kf_offset;
4804 if (kif->kf_type == KF_TYPE_VNODE)
4805 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4806 else
4807 okif->kf_vnode_type = KF_VTYPE_VNON;
4808 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4809 if (kif->kf_type == KF_TYPE_SOCKET) {
4810 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4811 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4812 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4813 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4814 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4815 } else {
4816 okif->kf_sa_local.ss_family = AF_UNSPEC;
4817 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4818 }
4819 }
4820
4821 static int
export_vnode_for_osysctl(struct vnode * vp,int type,struct kinfo_file * kif,struct kinfo_ofile * okif,struct pwddesc * pdp,struct sysctl_req * req)4822 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4823 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4824 {
4825 int error;
4826
4827 vrefact(vp);
4828 PWDDESC_XUNLOCK(pdp);
4829 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4830 kinfo_to_okinfo(kif, okif);
4831 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4832 PWDDESC_XLOCK(pdp);
4833 return (error);
4834 }
4835
4836 /*
4837 * Get per-process file descriptors for use by procstat(1), et al.
4838 */
4839 static int
sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)4840 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4841 {
4842 struct kinfo_ofile *okif;
4843 struct kinfo_file *kif;
4844 struct filedesc *fdp;
4845 struct pwddesc *pdp;
4846 struct pwd *pwd;
4847 u_int namelen;
4848 int error, i, *name;
4849 struct file *fp;
4850 struct proc *p;
4851
4852 namelen = arg2;
4853 if (namelen != 1)
4854 return (EINVAL);
4855
4856 name = (int *)arg1;
4857 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4858 if (error != 0)
4859 return (error);
4860 fdp = fdhold(p);
4861 if (fdp != NULL)
4862 pdp = pdhold(p);
4863 PROC_UNLOCK(p);
4864 if (fdp == NULL || pdp == NULL) {
4865 if (fdp != NULL)
4866 fddrop(fdp);
4867 return (ENOENT);
4868 }
4869 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4870 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4871 PWDDESC_XLOCK(pdp);
4872 pwd = pwd_hold_pwddesc(pdp);
4873 if (pwd != NULL) {
4874 if (pwd->pwd_cdir != NULL)
4875 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4876 okif, pdp, req);
4877 if (pwd->pwd_rdir != NULL)
4878 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4879 okif, pdp, req);
4880 if (pwd->pwd_jdir != NULL)
4881 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4882 okif, pdp, req);
4883 }
4884 PWDDESC_XUNLOCK(pdp);
4885 if (pwd != NULL)
4886 pwd_drop(pwd);
4887 FILEDESC_SLOCK(fdp);
4888 if (refcount_load(&fdp->fd_refcnt) == 0)
4889 goto skip;
4890 FILEDESC_FOREACH_FP(fdp, i, fp) {
4891 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4892 KERN_FILEDESC_PACK_KINFO);
4893 FILEDESC_SUNLOCK(fdp);
4894 kinfo_to_okinfo(kif, okif);
4895 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4896 FILEDESC_SLOCK(fdp);
4897 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4898 break;
4899 }
4900 skip:
4901 FILEDESC_SUNLOCK(fdp);
4902 fddrop(fdp);
4903 pddrop(pdp);
4904 free(kif, M_TEMP);
4905 free(okif, M_TEMP);
4906 return (0);
4907 }
4908
4909 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4910 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4911 "Process ofiledesc entries");
4912 #endif /* COMPAT_FREEBSD7 */
4913
4914 int
vntype_to_kinfo(int vtype)4915 vntype_to_kinfo(int vtype)
4916 {
4917 struct {
4918 int vtype;
4919 int kf_vtype;
4920 } vtypes_table[] = {
4921 { VBAD, KF_VTYPE_VBAD },
4922 { VBLK, KF_VTYPE_VBLK },
4923 { VCHR, KF_VTYPE_VCHR },
4924 { VDIR, KF_VTYPE_VDIR },
4925 { VFIFO, KF_VTYPE_VFIFO },
4926 { VLNK, KF_VTYPE_VLNK },
4927 { VNON, KF_VTYPE_VNON },
4928 { VREG, KF_VTYPE_VREG },
4929 { VSOCK, KF_VTYPE_VSOCK }
4930 };
4931 unsigned int i;
4932
4933 /*
4934 * Perform vtype translation.
4935 */
4936 for (i = 0; i < nitems(vtypes_table); i++)
4937 if (vtypes_table[i].vtype == vtype)
4938 return (vtypes_table[i].kf_vtype);
4939
4940 return (KF_VTYPE_UNKNOWN);
4941 }
4942
4943 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4944 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4945 "Process filedesc entries");
4946
4947 /*
4948 * Store a process current working directory information to sbuf.
4949 *
4950 * Takes a locked proc as argument, and returns with the proc unlocked.
4951 */
4952 int
kern_proc_cwd_out(struct proc * p,struct sbuf * sb,ssize_t maxlen)4953 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4954 {
4955 struct pwddesc *pdp;
4956 struct pwd *pwd;
4957 struct export_fd_buf *efbuf;
4958 struct vnode *cdir;
4959 int error;
4960
4961 PROC_LOCK_ASSERT(p, MA_OWNED);
4962
4963 pdp = pdhold(p);
4964 PROC_UNLOCK(p);
4965 if (pdp == NULL)
4966 return (EINVAL);
4967
4968 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4969 efbuf->fdp = NULL;
4970 efbuf->pdp = pdp;
4971 efbuf->sb = sb;
4972 efbuf->remainder = maxlen;
4973 efbuf->flags = 0;
4974
4975 PWDDESC_XLOCK(pdp);
4976 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4977 cdir = pwd->pwd_cdir;
4978 if (cdir == NULL) {
4979 error = EINVAL;
4980 } else {
4981 vrefact(cdir);
4982 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4983 }
4984 PWDDESC_XUNLOCK(pdp);
4985 pddrop(pdp);
4986 free(efbuf, M_TEMP);
4987 return (error);
4988 }
4989
4990 /*
4991 * Get per-process current working directory.
4992 */
4993 static int
sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)4994 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4995 {
4996 struct sbuf sb;
4997 struct proc *p;
4998 ssize_t maxlen;
4999 u_int namelen;
5000 int error, error2, *name;
5001
5002 namelen = arg2;
5003 if (namelen != 1)
5004 return (EINVAL);
5005
5006 name = (int *)arg1;
5007
5008 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
5009 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
5010 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
5011 if (error != 0) {
5012 sbuf_delete(&sb);
5013 return (error);
5014 }
5015 maxlen = req->oldptr != NULL ? req->oldlen : -1;
5016 error = kern_proc_cwd_out(p, &sb, maxlen);
5017 error2 = sbuf_finish(&sb);
5018 sbuf_delete(&sb);
5019 return (error != 0 ? error : error2);
5020 }
5021
5022 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
5023 sysctl_kern_proc_cwd, "Process current working directory");
5024
5025 #ifdef DDB
5026 /*
5027 * For the purposes of debugging, generate a human-readable string for the
5028 * file type.
5029 */
5030 static const char *
file_type_to_name(short type)5031 file_type_to_name(short type)
5032 {
5033
5034 switch (type) {
5035 case 0:
5036 return ("zero");
5037 case DTYPE_VNODE:
5038 return ("vnode");
5039 case DTYPE_SOCKET:
5040 return ("socket");
5041 case DTYPE_PIPE:
5042 return ("pipe");
5043 case DTYPE_FIFO:
5044 return ("fifo");
5045 case DTYPE_KQUEUE:
5046 return ("kqueue");
5047 case DTYPE_CRYPTO:
5048 return ("crypto");
5049 case DTYPE_MQUEUE:
5050 return ("mqueue");
5051 case DTYPE_SHM:
5052 return ("shm");
5053 case DTYPE_SEM:
5054 return ("ksem");
5055 case DTYPE_PTS:
5056 return ("pts");
5057 case DTYPE_DEV:
5058 return ("dev");
5059 case DTYPE_PROCDESC:
5060 return ("proc");
5061 case DTYPE_EVENTFD:
5062 return ("eventfd");
5063 case DTYPE_TIMERFD:
5064 return ("timerfd");
5065 default:
5066 return ("unkn");
5067 }
5068 }
5069
5070 /*
5071 * For the purposes of debugging, identify a process (if any, perhaps one of
5072 * many) that references the passed file in its file descriptor array. Return
5073 * NULL if none.
5074 */
5075 static struct proc *
file_to_first_proc(struct file * fp)5076 file_to_first_proc(struct file *fp)
5077 {
5078 struct filedesc *fdp;
5079 struct proc *p;
5080 int n;
5081
5082 FOREACH_PROC_IN_SYSTEM(p) {
5083 if (p->p_state == PRS_NEW)
5084 continue;
5085 fdp = p->p_fd;
5086 if (fdp == NULL)
5087 continue;
5088 for (n = 0; n < fdp->fd_nfiles; n++) {
5089 if (fp == fdp->fd_ofiles[n].fde_file)
5090 return (p);
5091 }
5092 }
5093 return (NULL);
5094 }
5095
5096 static void
db_print_file(struct file * fp,int header)5097 db_print_file(struct file *fp, int header)
5098 {
5099 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
5100 struct proc *p;
5101
5102 if (header)
5103 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
5104 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
5105 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
5106 "FCmd");
5107 p = file_to_first_proc(fp);
5108 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
5109 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
5110 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
5111 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
5112
5113 #undef XPTRWIDTH
5114 }
5115
DB_SHOW_COMMAND(file,db_show_file)5116 DB_SHOW_COMMAND(file, db_show_file)
5117 {
5118 struct file *fp;
5119
5120 if (!have_addr) {
5121 db_printf("usage: show file <addr>\n");
5122 return;
5123 }
5124 fp = (struct file *)addr;
5125 db_print_file(fp, 1);
5126 }
5127
DB_SHOW_COMMAND_FLAGS(files,db_show_files,DB_CMD_MEMSAFE)5128 DB_SHOW_COMMAND_FLAGS(files, db_show_files, DB_CMD_MEMSAFE)
5129 {
5130 struct filedesc *fdp;
5131 struct file *fp;
5132 struct proc *p;
5133 int header;
5134 int n;
5135
5136 header = 1;
5137 FOREACH_PROC_IN_SYSTEM(p) {
5138 if (p->p_state == PRS_NEW)
5139 continue;
5140 if ((fdp = p->p_fd) == NULL)
5141 continue;
5142 for (n = 0; n < fdp->fd_nfiles; ++n) {
5143 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
5144 continue;
5145 db_print_file(fp, header);
5146 header = 0;
5147 }
5148 }
5149 }
5150 #endif
5151
5152 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc,
5153 CTLFLAG_RWTUN | CTLFLAG_NOFETCH,
5154 &maxfilesperproc, 0, "Maximum files allowed open per process");
5155
5156 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RWTUN | CTLFLAG_NOFETCH,
5157 &maxfiles, 0, "Maximum number of files");
5158
5159 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
5160 &openfiles, 0, "System-wide number of open files");
5161
5162 /* ARGSUSED*/
5163 static void
filelistinit(void * dummy)5164 filelistinit(void *dummy)
5165 {
5166
5167 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
5168 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
5169 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
5170 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
5171 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
5172 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
5173 /*
5174 * XXXMJG this is a temporary hack due to boot ordering issues against
5175 * the vnode zone.
5176 */
5177 vfs_smr = uma_zone_get_smr(pwd_zone);
5178 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
5179 }
5180 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
5181
5182 /*-------------------------------------------------------------------*/
5183
5184 static int
badfo_readwrite(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)5185 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
5186 int flags, struct thread *td)
5187 {
5188
5189 return (EBADF);
5190 }
5191
5192 static int
badfo_truncate(struct file * fp,off_t length,struct ucred * active_cred,struct thread * td)5193 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5194 struct thread *td)
5195 {
5196
5197 return (EINVAL);
5198 }
5199
5200 static int
badfo_ioctl(struct file * fp,u_long com,void * data,struct ucred * active_cred,struct thread * td)5201 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
5202 struct thread *td)
5203 {
5204
5205 return (EBADF);
5206 }
5207
5208 static int
badfo_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)5209 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
5210 struct thread *td)
5211 {
5212
5213 return (0);
5214 }
5215
5216 static int
badfo_kqfilter(struct file * fp,struct knote * kn)5217 badfo_kqfilter(struct file *fp, struct knote *kn)
5218 {
5219
5220 return (EBADF);
5221 }
5222
5223 static int
badfo_stat(struct file * fp,struct stat * sb,struct ucred * active_cred)5224 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
5225 {
5226
5227 return (EBADF);
5228 }
5229
5230 static int
badfo_close(struct file * fp,struct thread * td)5231 badfo_close(struct file *fp, struct thread *td)
5232 {
5233
5234 return (0);
5235 }
5236
5237 static int
badfo_chmod(struct file * fp,mode_t mode,struct ucred * active_cred,struct thread * td)5238 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5239 struct thread *td)
5240 {
5241
5242 return (EBADF);
5243 }
5244
5245 static int
badfo_chown(struct file * fp,uid_t uid,gid_t gid,struct ucred * active_cred,struct thread * td)5246 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5247 struct thread *td)
5248 {
5249
5250 return (EBADF);
5251 }
5252
5253 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)5254 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5255 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5256 struct thread *td)
5257 {
5258
5259 return (EBADF);
5260 }
5261
5262 static int
badfo_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)5263 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
5264 {
5265
5266 return (0);
5267 }
5268
5269 const struct fileops badfileops = {
5270 .fo_read = badfo_readwrite,
5271 .fo_write = badfo_readwrite,
5272 .fo_truncate = badfo_truncate,
5273 .fo_ioctl = badfo_ioctl,
5274 .fo_poll = badfo_poll,
5275 .fo_kqfilter = badfo_kqfilter,
5276 .fo_stat = badfo_stat,
5277 .fo_close = badfo_close,
5278 .fo_chmod = badfo_chmod,
5279 .fo_chown = badfo_chown,
5280 .fo_sendfile = badfo_sendfile,
5281 .fo_fill_kinfo = badfo_fill_kinfo,
5282 };
5283
5284 static int
path_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)5285 path_poll(struct file *fp, int events, struct ucred *active_cred,
5286 struct thread *td)
5287 {
5288 return (POLLNVAL);
5289 }
5290
5291 static int
path_close(struct file * fp,struct thread * td)5292 path_close(struct file *fp, struct thread *td)
5293 {
5294 MPASS(fp->f_type == DTYPE_VNODE);
5295 fp->f_ops = &badfileops;
5296 vrele(fp->f_vnode);
5297 return (0);
5298 }
5299
5300 const struct fileops path_fileops = {
5301 .fo_read = badfo_readwrite,
5302 .fo_write = badfo_readwrite,
5303 .fo_truncate = badfo_truncate,
5304 .fo_ioctl = badfo_ioctl,
5305 .fo_poll = path_poll,
5306 .fo_kqfilter = vn_kqfilter_opath,
5307 .fo_stat = vn_statfile,
5308 .fo_close = path_close,
5309 .fo_chmod = badfo_chmod,
5310 .fo_chown = badfo_chown,
5311 .fo_sendfile = badfo_sendfile,
5312 .fo_fill_kinfo = vn_fill_kinfo,
5313 .fo_cmp = vn_cmp,
5314 .fo_flags = DFLAG_PASSABLE,
5315 };
5316
5317 int
invfo_rdwr(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)5318 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5319 int flags, struct thread *td)
5320 {
5321
5322 return (EOPNOTSUPP);
5323 }
5324
5325 int
invfo_truncate(struct file * fp,off_t length,struct ucred * active_cred,struct thread * td)5326 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5327 struct thread *td)
5328 {
5329
5330 return (EINVAL);
5331 }
5332
5333 int
invfo_ioctl(struct file * fp,u_long com,void * data,struct ucred * active_cred,struct thread * td)5334 invfo_ioctl(struct file *fp, u_long com, void *data,
5335 struct ucred *active_cred, struct thread *td)
5336 {
5337
5338 return (ENOTTY);
5339 }
5340
5341 int
invfo_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)5342 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5343 struct thread *td)
5344 {
5345
5346 return (poll_no_poll(events));
5347 }
5348
5349 int
invfo_kqfilter(struct file * fp,struct knote * kn)5350 invfo_kqfilter(struct file *fp, struct knote *kn)
5351 {
5352
5353 return (EINVAL);
5354 }
5355
5356 int
invfo_chmod(struct file * fp,mode_t mode,struct ucred * active_cred,struct thread * td)5357 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5358 struct thread *td)
5359 {
5360
5361 return (EINVAL);
5362 }
5363
5364 int
invfo_chown(struct file * fp,uid_t uid,gid_t gid,struct ucred * active_cred,struct thread * td)5365 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5366 struct thread *td)
5367 {
5368
5369 return (EINVAL);
5370 }
5371
5372 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)5373 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5374 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5375 struct thread *td)
5376 {
5377
5378 return (EINVAL);
5379 }
5380
5381 /*-------------------------------------------------------------------*/
5382
5383 /*
5384 * File Descriptor pseudo-device driver (/dev/fd/).
5385 *
5386 * Opening minor device N dup()s the file (if any) connected to file
5387 * descriptor N belonging to the calling process. Note that this driver
5388 * consists of only the ``open()'' routine, because all subsequent
5389 * references to this file will be direct to the other driver.
5390 *
5391 * XXX: we could give this one a cloning event handler if necessary.
5392 */
5393
5394 /* ARGSUSED */
5395 static int
fdopen(struct cdev * dev,int mode,int type,struct thread * td)5396 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5397 {
5398
5399 /*
5400 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5401 * the file descriptor being sought for duplication. The error
5402 * return ensures that the vnode for this device will be released
5403 * by vn_open. Open will detect this special error and take the
5404 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5405 * will simply report the error.
5406 */
5407 td->td_dupfd = dev2unit(dev);
5408 return (ENODEV);
5409 }
5410
5411 static struct cdevsw fildesc_cdevsw = {
5412 .d_version = D_VERSION,
5413 .d_open = fdopen,
5414 .d_name = "FD",
5415 };
5416
5417 static void
fildesc_drvinit(void * unused)5418 fildesc_drvinit(void *unused)
5419 {
5420 struct cdev *dev;
5421
5422 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5423 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5424 make_dev_alias(dev, "stdin");
5425 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5426 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5427 make_dev_alias(dev, "stdout");
5428 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5429 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5430 make_dev_alias(dev, "stderr");
5431 }
5432
5433 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5434