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 "opt_capsicum.h"
38 #include "opt_ddb.h"
39 #include "opt_ktrace.h"
40
41 #define EXTERR_CATEGORY EXTERR_CAT_FILEDESC
42 #include <sys/systm.h>
43 #include <sys/capsicum.h>
44 #include <sys/conf.h>
45 #include <sys/exterrvar.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/limits.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mount.h>
56 #include <sys/mutex.h>
57 #include <sys/namei.h>
58 #include <sys/selinfo.h>
59 #include <sys/poll.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/protosw.h>
63 #include <sys/racct.h>
64 #include <sys/resourcevar.h>
65 #include <sys/sbuf.h>
66 #include <sys/signalvar.h>
67 #include <sys/kdb.h>
68 #include <sys/smr.h>
69 #include <sys/stat.h>
70 #include <sys/sx.h>
71 #include <sys/syscallsubr.h>
72 #include <sys/sysctl.h>
73 #include <sys/sysproto.h>
74 #include <sys/unistd.h>
75 #include <sys/user.h>
76 #include <sys/vnode.h>
77 #include <sys/ktrace.h>
78
79 #include <net/vnet.h>
80
81 #include <security/audit/audit.h>
82
83 #include <vm/uma.h>
84 #include <vm/vm.h>
85
86 #include <ddb/ddb.h>
87
88 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
89 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
90 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
91 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
92 "file desc to leader structures");
93 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
94 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
95
96 MALLOC_DECLARE(M_FADVISE);
97
98 static __read_mostly uma_zone_t file_zone;
99 static __read_mostly uma_zone_t filedesc0_zone;
100 __read_mostly uma_zone_t pwd_zone;
101 VFS_SMR_DECLARE;
102
103 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
104 struct thread *td, bool holdleaders, bool audit);
105 static void export_file_to_kinfo(struct file *fp, int fd,
106 cap_rights_t *rightsp, struct kinfo_file *kif,
107 struct filedesc *fdp, int flags);
108 static int fd_first_free(struct filedesc *fdp, int low, int size);
109 static void fdgrowtable(struct filedesc *fdp, int nfd);
110 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
111 static void fdunused(struct filedesc *fdp, int fd);
112 static void fdused(struct filedesc *fdp, int fd);
113 static int fget_unlocked_seq(struct thread *td, int fd,
114 const cap_rights_t *needrightsp, uint8_t *flagsp,
115 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 struct flags_trans_elem {
484 u_int f;
485 u_int t;
486 };
487
488 static u_int
flags_trans(const struct flags_trans_elem * ftes,int nitems,u_int from_flags)489 flags_trans(const struct flags_trans_elem *ftes, int nitems, u_int from_flags)
490 {
491 u_int res;
492 int i;
493
494 res = 0;
495 for (i = 0; i < nitems; i++) {
496 if ((from_flags & ftes[i].f) != 0)
497 res |= ftes[i].t;
498 }
499 return (res);
500 }
501
502 static uint8_t
fd_to_fde_flags(int fd_flags)503 fd_to_fde_flags(int fd_flags)
504 {
505 static const struct flags_trans_elem fd_to_fde_flags_s[] = {
506 { .f = FD_CLOEXEC, .t = UF_EXCLOSE },
507 { .f = FD_CLOFORK, .t = UF_FOCLOSE },
508 { .f = FD_RESOLVE_BENEATH, .t = UF_RESOLVE_BENEATH },
509 };
510
511 return (flags_trans(fd_to_fde_flags_s, nitems(fd_to_fde_flags_s),
512 fd_flags));
513 }
514
515 static int
fde_to_fd_flags(uint8_t fde_flags)516 fde_to_fd_flags(uint8_t fde_flags)
517 {
518 static const struct flags_trans_elem fde_to_fd_flags_s[] = {
519 { .f = UF_EXCLOSE, .t = FD_CLOEXEC },
520 { .f = UF_FOCLOSE, .t = FD_CLOFORK },
521 { .f = UF_RESOLVE_BENEATH, .t = FD_RESOLVE_BENEATH },
522 };
523
524 return (flags_trans(fde_to_fd_flags_s, nitems(fde_to_fd_flags_s),
525 fde_flags));
526 }
527
528 static uint8_t
fddup_to_fde_flags(int fddup_flags)529 fddup_to_fde_flags(int fddup_flags)
530 {
531 static const struct flags_trans_elem fddup_to_fde_flags_s[] = {
532 { .f = FDDUP_FLAG_CLOEXEC, .t = UF_EXCLOSE },
533 { .f = FDDUP_FLAG_CLOFORK, .t = UF_FOCLOSE },
534 };
535
536 return (flags_trans(fddup_to_fde_flags_s, nitems(fddup_to_fde_flags_s),
537 fddup_flags));
538 }
539
540 static uint8_t
close_range_to_fde_flags(int close_range_flags)541 close_range_to_fde_flags(int close_range_flags)
542 {
543 static const struct flags_trans_elem close_range_to_fde_flags_s[] = {
544 { .f = CLOSE_RANGE_CLOEXEC, .t = UF_EXCLOSE },
545 { .f = CLOSE_RANGE_CLOFORK, .t = UF_FOCLOSE },
546 };
547
548 return (flags_trans(close_range_to_fde_flags_s,
549 nitems(close_range_to_fde_flags_s), close_range_flags));
550 }
551
552 static uint8_t
open_to_fde_flags(int open_flags,bool sticky_orb)553 open_to_fde_flags(int open_flags, bool sticky_orb)
554 {
555 static const struct flags_trans_elem open_to_fde_flags_s[] = {
556 { .f = O_CLOEXEC, .t = UF_EXCLOSE },
557 { .f = O_CLOFORK, .t = UF_FOCLOSE },
558 { .f = O_RESOLVE_BENEATH, .t = UF_RESOLVE_BENEATH },
559 };
560 #if defined(__clang__) && __clang_major__ >= 19
561 _Static_assert(open_to_fde_flags_s[nitems(open_to_fde_flags_s) - 1].f ==
562 O_RESOLVE_BENEATH, "O_RESOLVE_BENEATH must be last, for sticky_orb");
563 #endif
564
565 return (flags_trans(open_to_fde_flags_s, nitems(open_to_fde_flags_s) -
566 (sticky_orb ? 0 : 1), open_flags));
567 }
568
569 int
kern_fcntl(struct thread * td,int fd,int cmd,intptr_t arg)570 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
571 {
572 struct filedesc *fdp;
573 struct flock *flp;
574 struct file *fp, *fp2;
575 struct filedescent *fde;
576 struct proc *p;
577 struct vnode *vp;
578 struct mount *mp;
579 struct kinfo_file *kif;
580 int error, flg, kif_sz, seals, tmp, got_set, got_cleared;
581 uint64_t bsize;
582 off_t foffset;
583 int flags;
584
585 error = 0;
586 flg = F_POSIX;
587 p = td->td_proc;
588 fdp = p->p_fd;
589
590 AUDIT_ARG_FD(cmd);
591 AUDIT_ARG_CMD(cmd);
592 switch (cmd) {
593 case F_DUPFD:
594 tmp = arg;
595 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
596 break;
597
598 case F_DUPFD_CLOEXEC:
599 tmp = arg;
600 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
601 break;
602
603 case F_DUPFD_CLOFORK:
604 tmp = arg;
605 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOFORK, fd, tmp);
606 break;
607
608 case F_DUP2FD:
609 tmp = arg;
610 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
611 break;
612
613 case F_DUP2FD_CLOEXEC:
614 tmp = arg;
615 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
616 break;
617
618 case F_GETFD:
619 error = EBADF;
620 FILEDESC_SLOCK(fdp);
621 fde = fdeget_noref(fdp, fd);
622 if (fde != NULL) {
623 td->td_retval[0] = fde_to_fd_flags(fde->fde_flags);
624 error = 0;
625 }
626 FILEDESC_SUNLOCK(fdp);
627 break;
628
629 case F_SETFD:
630 error = EBADF;
631 FILEDESC_XLOCK(fdp);
632 fde = fdeget_noref(fdp, fd);
633 if (fde != NULL) {
634 /*
635 * UF_RESOLVE_BENEATH is sticky and cannot be cleared.
636 */
637 fde->fde_flags = (fde->fde_flags &
638 ~(UF_EXCLOSE | UF_FOCLOSE)) | fd_to_fde_flags(arg);
639 error = 0;
640 }
641 FILEDESC_XUNLOCK(fdp);
642 break;
643
644 case F_GETFL:
645 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
646 if (error != 0)
647 break;
648 td->td_retval[0] = OFLAGS(fp->f_flag);
649 fdrop(fp, td);
650 break;
651
652 case F_SETFL:
653 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
654 if (error != 0)
655 break;
656 if (fp->f_ops == &path_fileops) {
657 fdrop(fp, td);
658 error = EBADF;
659 break;
660 }
661 fsetfl_lock(fp);
662 do {
663 tmp = flg = fp->f_flag;
664 tmp &= ~FCNTLFLAGS;
665 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
666 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
667 got_set = tmp & ~flg;
668 got_cleared = flg & ~tmp;
669 if (((got_set | got_cleared) & FNONBLOCK) != 0) {
670 tmp = fp->f_flag & FNONBLOCK;
671 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
672 if (error != 0)
673 goto revert_flags;
674 }
675 if (((got_set | got_cleared) & FASYNC) != 0) {
676 tmp = fp->f_flag & FASYNC;
677 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
678 if (error != 0)
679 goto revert_nonblock;
680 }
681 fsetfl_unlock(fp);
682 fdrop(fp, td);
683 break;
684 revert_nonblock:
685 if (((got_set | got_cleared) & FNONBLOCK) != 0) {
686 tmp = ~fp->f_flag & FNONBLOCK;
687 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
688 }
689 revert_flags:
690 do {
691 tmp = flg = fp->f_flag;
692 tmp &= ~FCNTLFLAGS;
693 tmp |= got_cleared;
694 tmp &= ~got_set;
695 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
696 fsetfl_unlock(fp);
697 fdrop(fp, td);
698 break;
699
700 case F_GETOWN:
701 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
702 if (error != 0)
703 break;
704 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
705 if (error == 0)
706 td->td_retval[0] = tmp;
707 fdrop(fp, td);
708 break;
709
710 case F_SETOWN:
711 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
712 if (error != 0)
713 break;
714 tmp = arg;
715 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
716 fdrop(fp, td);
717 break;
718
719 case F_SETLK_REMOTE:
720 error = priv_check(td, PRIV_NFS_LOCKD);
721 if (error != 0)
722 return (error);
723 flg = F_REMOTE;
724 goto do_setlk;
725
726 case F_SETLKW:
727 flg |= F_WAIT;
728 /* FALLTHROUGH F_SETLK */
729
730 case F_SETLK:
731 do_setlk:
732 flp = (struct flock *)arg;
733 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
734 error = EINVAL;
735 break;
736 }
737
738 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
739 if (error != 0)
740 break;
741 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
742 error = EBADF;
743 fdrop(fp, td);
744 break;
745 }
746
747 if (flp->l_whence == SEEK_CUR) {
748 foffset = foffset_get(fp);
749 if (foffset < 0 ||
750 (flp->l_start > 0 &&
751 foffset > OFF_MAX - flp->l_start)) {
752 error = EOVERFLOW;
753 fdrop(fp, td);
754 break;
755 }
756 flp->l_start += foffset;
757 }
758
759 vp = fp->f_vnode;
760 switch (flp->l_type) {
761 case F_RDLCK:
762 if ((fp->f_flag & FREAD) == 0) {
763 error = EBADF;
764 break;
765 }
766 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
767 PROC_LOCK(p->p_leader);
768 p->p_leader->p_flag |= P_ADVLOCK;
769 PROC_UNLOCK(p->p_leader);
770 }
771 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
772 flp, flg);
773 break;
774 case F_WRLCK:
775 if ((fp->f_flag & FWRITE) == 0) {
776 error = EBADF;
777 break;
778 }
779 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
780 PROC_LOCK(p->p_leader);
781 p->p_leader->p_flag |= P_ADVLOCK;
782 PROC_UNLOCK(p->p_leader);
783 }
784 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
785 flp, flg);
786 break;
787 case F_UNLCK:
788 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
789 flp, flg);
790 break;
791 case F_UNLCKSYS:
792 if (flg != F_REMOTE) {
793 error = EINVAL;
794 break;
795 }
796 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
797 F_UNLCKSYS, flp, flg);
798 break;
799 default:
800 error = EINVAL;
801 break;
802 }
803 if (error != 0 || flp->l_type == F_UNLCK ||
804 flp->l_type == F_UNLCKSYS) {
805 fdrop(fp, td);
806 break;
807 }
808
809 /*
810 * Check for a race with close.
811 *
812 * The vnode is now advisory locked (or unlocked, but this case
813 * is not really important) as the caller requested.
814 * We had to drop the filedesc lock, so we need to recheck if
815 * the descriptor is still valid, because if it was closed
816 * in the meantime we need to remove advisory lock from the
817 * vnode - close on any descriptor leading to an advisory
818 * locked vnode, removes that lock.
819 * We will return 0 on purpose in that case, as the result of
820 * successful advisory lock might have been externally visible
821 * already. This is fine - effectively we pretend to the caller
822 * that the closing thread was a bit slower and that the
823 * advisory lock succeeded before the close.
824 */
825 error = fget_unlocked(td, fd, &cap_no_rights, &fp2);
826 if (error != 0) {
827 fdrop(fp, td);
828 break;
829 }
830 if (fp != fp2) {
831 flp->l_whence = SEEK_SET;
832 flp->l_start = 0;
833 flp->l_len = 0;
834 flp->l_type = F_UNLCK;
835 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
836 F_UNLCK, flp, F_POSIX);
837 }
838 fdrop(fp, td);
839 fdrop(fp2, td);
840 break;
841
842 case F_GETLK:
843 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
844 if (error != 0)
845 break;
846 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
847 error = EBADF;
848 fdrop(fp, td);
849 break;
850 }
851 flp = (struct flock *)arg;
852 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
853 flp->l_type != F_UNLCK) {
854 error = EINVAL;
855 fdrop(fp, td);
856 break;
857 }
858 if (flp->l_whence == SEEK_CUR) {
859 foffset = foffset_get(fp);
860 if ((flp->l_start > 0 &&
861 foffset > OFF_MAX - flp->l_start) ||
862 (flp->l_start < 0 &&
863 foffset < OFF_MIN - flp->l_start)) {
864 error = EOVERFLOW;
865 fdrop(fp, td);
866 break;
867 }
868 flp->l_start += foffset;
869 }
870 vp = fp->f_vnode;
871 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
872 F_POSIX);
873 fdrop(fp, td);
874 break;
875
876 case F_ADD_SEALS:
877 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
878 if (error != 0)
879 break;
880 error = fo_add_seals(fp, arg);
881 fdrop(fp, td);
882 break;
883
884 case F_GET_SEALS:
885 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
886 if (error != 0)
887 break;
888 if (fo_get_seals(fp, &seals) == 0)
889 td->td_retval[0] = seals;
890 else
891 error = EINVAL;
892 fdrop(fp, td);
893 break;
894
895 case F_RDAHEAD:
896 arg = arg ? 128 * 1024: 0;
897 /* FALLTHROUGH */
898 case F_READAHEAD:
899 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
900 if (error != 0)
901 break;
902 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
903 fdrop(fp, td);
904 error = EBADF;
905 break;
906 }
907 vp = fp->f_vnode;
908 if (vp->v_type != VREG) {
909 fdrop(fp, td);
910 error = ENOTTY;
911 break;
912 }
913
914 /*
915 * Exclusive lock synchronizes against f_seqcount reads and
916 * writes in sequential_heuristic().
917 */
918 error = vn_lock(vp, LK_EXCLUSIVE);
919 if (error != 0) {
920 fdrop(fp, td);
921 break;
922 }
923 if (arg >= 0) {
924 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
925 arg = MIN(arg, INT_MAX - bsize + 1);
926 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
927 (arg + bsize - 1) / bsize);
928 atomic_set_int(&fp->f_flag, FRDAHEAD);
929 } else {
930 atomic_clear_int(&fp->f_flag, FRDAHEAD);
931 }
932 VOP_UNLOCK(vp);
933 fdrop(fp, td);
934 break;
935
936 case F_ISUNIONSTACK:
937 /*
938 * Check if the vnode is part of a union stack (either the
939 * "union" flag from mount(2) or unionfs).
940 *
941 * Prior to introduction of this op libc's readdir would call
942 * fstatfs(2), in effect unnecessarily copying kilobytes of
943 * data just to check fs name and a mount flag.
944 *
945 * Fixing the code to handle everything in the kernel instead
946 * is a non-trivial endeavor and has low priority, thus this
947 * horrible kludge facilitates the current behavior in a much
948 * cheaper manner until someone(tm) sorts this out.
949 */
950 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
951 if (error != 0)
952 break;
953 if (fp->f_type != DTYPE_VNODE) {
954 fdrop(fp, td);
955 error = EBADF;
956 break;
957 }
958 vp = fp->f_vnode;
959 /*
960 * Since we don't prevent dooming the vnode even non-null mp
961 * found can become immediately stale. This is tolerable since
962 * mount points are type-stable (providing safe memory access)
963 * and any vfs op on this vnode going forward will return an
964 * error (meaning return value in this case is meaningless).
965 */
966 mp = atomic_load_ptr(&vp->v_mount);
967 if (__predict_false(mp == NULL)) {
968 fdrop(fp, td);
969 error = EBADF;
970 break;
971 }
972 td->td_retval[0] = 0;
973 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
974 mp->mnt_flag & MNT_UNION)
975 td->td_retval[0] = 1;
976 fdrop(fp, td);
977 break;
978
979 case F_KINFO:
980 #ifdef CAPABILITY_MODE
981 if (CAP_TRACING(td))
982 ktrcapfail(CAPFAIL_SYSCALL, &cmd);
983 if (IN_CAPABILITY_MODE(td)) {
984 error = ECAPMODE;
985 break;
986 }
987 #endif
988 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
989 if (error != 0)
990 break;
991 if (kif_sz != sizeof(*kif)) {
992 error = EINVAL;
993 break;
994 }
995 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
996 FILEDESC_SLOCK(fdp);
997 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL);
998 if (error == 0 && fhold(fp)) {
999 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
1000 FILEDESC_SUNLOCK(fdp);
1001 fdrop(fp, td);
1002 if ((kif->kf_status & KF_ATTR_VALID) != 0) {
1003 kif->kf_structsize = sizeof(*kif);
1004 error = copyout(kif, (void *)arg, sizeof(*kif));
1005 } else {
1006 error = EBADF;
1007 }
1008 } else {
1009 FILEDESC_SUNLOCK(fdp);
1010 if (error == 0)
1011 error = EBADF;
1012 }
1013 free(kif, M_TEMP);
1014 break;
1015
1016 default:
1017 if ((cmd & ((1u << F_DUP3FD_SHIFT) - 1)) != F_DUP3FD)
1018 return (EXTERROR(EINVAL, "invalid fcntl cmd"));
1019 /* Handle F_DUP3FD */
1020 flags = (cmd >> F_DUP3FD_SHIFT);
1021 if ((flags & ~(FD_CLOEXEC | FD_CLOFORK)) != 0)
1022 return (EXTERROR(EINVAL, "invalid flags for F_DUP3FD"));
1023 tmp = arg;
1024 error = kern_dup(td, FDDUP_FIXED,
1025 ((flags & FD_CLOEXEC) != 0 ? FDDUP_FLAG_CLOEXEC : 0) |
1026 ((flags & FD_CLOFORK) != 0 ? FDDUP_FLAG_CLOFORK : 0),
1027 fd, tmp);
1028 break;
1029 }
1030 return (error);
1031 }
1032
1033 static int
getmaxfd(struct thread * td)1034 getmaxfd(struct thread *td)
1035 {
1036
1037 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
1038 }
1039
1040 /*
1041 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
1042 */
1043 int
kern_dup(struct thread * td,u_int mode,int flags,int old,int new)1044 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
1045 {
1046 struct filedesc *fdp;
1047 struct filedescent *oldfde, *newfde;
1048 struct proc *p;
1049 struct file *delfp, *oldfp;
1050 u_long *oioctls, *nioctls;
1051 int error, maxfd;
1052
1053 p = td->td_proc;
1054 fdp = p->p_fd;
1055 oioctls = NULL;
1056
1057 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC | FDDUP_FLAG_CLOFORK)) == 0);
1058 MPASS(mode < FDDUP_LASTMODE);
1059
1060 AUDIT_ARG_FD(old);
1061 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
1062
1063 /*
1064 * Verify we have a valid descriptor to dup from and possibly to
1065 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
1066 * return EINVAL when the new descriptor is out of bounds.
1067 */
1068 if (old < 0)
1069 return (EBADF);
1070 if (new < 0)
1071 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
1072 maxfd = getmaxfd(td);
1073 if (new >= maxfd)
1074 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
1075
1076 error = EBADF;
1077 FILEDESC_XLOCK(fdp);
1078 if (fget_noref(fdp, old) == NULL)
1079 goto unlock;
1080 if (mode == FDDUP_FIXED && old == new) {
1081 td->td_retval[0] = new;
1082 fdp->fd_ofiles[new].fde_flags |= fddup_to_fde_flags(flags);
1083 error = 0;
1084 goto unlock;
1085 }
1086
1087 oldfde = &fdp->fd_ofiles[old];
1088 oldfp = oldfde->fde_file;
1089 if (!fhold(oldfp))
1090 goto unlock;
1091
1092 /*
1093 * If the caller specified a file descriptor, make sure the file
1094 * table is large enough to hold it, and grab it. Otherwise, just
1095 * allocate a new descriptor the usual way.
1096 */
1097 switch (mode) {
1098 case FDDUP_NORMAL:
1099 case FDDUP_FCNTL:
1100 if ((error = fdalloc(td, new, &new)) != 0) {
1101 fdrop(oldfp, td);
1102 goto unlock;
1103 }
1104 break;
1105 case FDDUP_FIXED:
1106 if (new >= fdp->fd_nfiles) {
1107 /*
1108 * The resource limits are here instead of e.g.
1109 * fdalloc(), because the file descriptor table may be
1110 * shared between processes, so we can't really use
1111 * racct_add()/racct_sub(). Instead of counting the
1112 * number of actually allocated descriptors, just put
1113 * the limit on the size of the file descriptor table.
1114 */
1115 #ifdef RACCT
1116 if (RACCT_ENABLED()) {
1117 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
1118 if (error != 0) {
1119 error = EMFILE;
1120 fdrop(oldfp, td);
1121 goto unlock;
1122 }
1123 }
1124 #endif
1125 fdgrowtable_exp(fdp, new + 1);
1126 }
1127 if (!fdisused(fdp, new))
1128 fdused(fdp, new);
1129 break;
1130 default:
1131 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1132 }
1133
1134 KASSERT(old != new, ("new fd is same as old"));
1135
1136 /* Refetch oldfde because the table may have grown and old one freed. */
1137 oldfde = &fdp->fd_ofiles[old];
1138 KASSERT(oldfp == oldfde->fde_file,
1139 ("fdt_ofiles shift from growth observed at fd %d",
1140 old));
1141
1142 newfde = &fdp->fd_ofiles[new];
1143 delfp = newfde->fde_file;
1144
1145 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1146
1147 /*
1148 * Duplicate the source descriptor.
1149 */
1150 #ifdef CAPABILITIES
1151 seqc_write_begin(&newfde->fde_seqc);
1152 #endif
1153 oioctls = filecaps_free_prep(&newfde->fde_caps);
1154 fde_copy(oldfde, newfde);
1155 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1156 nioctls);
1157 newfde->fde_flags = (oldfde->fde_flags & ~(UF_EXCLOSE | UF_FOCLOSE)) |
1158 fddup_to_fde_flags(flags);
1159 #ifdef CAPABILITIES
1160 seqc_write_end(&newfde->fde_seqc);
1161 #endif
1162 td->td_retval[0] = new;
1163
1164 error = 0;
1165
1166 if (delfp != NULL) {
1167 (void) closefp(fdp, new, delfp, td, true, false);
1168 FILEDESC_UNLOCK_ASSERT(fdp);
1169 } else {
1170 unlock:
1171 FILEDESC_XUNLOCK(fdp);
1172 }
1173
1174 filecaps_free_finish(oioctls);
1175 return (error);
1176 }
1177
1178 static void
sigiofree(struct sigio * sigio)1179 sigiofree(struct sigio *sigio)
1180 {
1181 crfree(sigio->sio_ucred);
1182 free(sigio, M_SIGIO);
1183 }
1184
1185 static struct sigio *
funsetown_locked(struct sigio * sigio)1186 funsetown_locked(struct sigio *sigio)
1187 {
1188 struct proc *p;
1189 struct pgrp *pg;
1190
1191 SIGIO_ASSERT_LOCKED();
1192
1193 if (sigio == NULL)
1194 return (NULL);
1195 *sigio->sio_myref = NULL;
1196 if (sigio->sio_pgid < 0) {
1197 pg = sigio->sio_pgrp;
1198 PGRP_LOCK(pg);
1199 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1200 PGRP_UNLOCK(pg);
1201 } else {
1202 p = sigio->sio_proc;
1203 PROC_LOCK(p);
1204 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1205 PROC_UNLOCK(p);
1206 }
1207 return (sigio);
1208 }
1209
1210 /*
1211 * If sigio is on the list associated with a process or process group,
1212 * disable signalling from the device, remove sigio from the list and
1213 * free sigio.
1214 */
1215 void
funsetown(struct sigio ** sigiop)1216 funsetown(struct sigio **sigiop)
1217 {
1218 struct sigio *sigio;
1219
1220 /* Racy check, consumers must provide synchronization. */
1221 if (*sigiop == NULL)
1222 return;
1223
1224 SIGIO_LOCK();
1225 sigio = funsetown_locked(*sigiop);
1226 SIGIO_UNLOCK();
1227 if (sigio != NULL)
1228 sigiofree(sigio);
1229 }
1230
1231 /*
1232 * Free a list of sigio structures. The caller must ensure that new sigio
1233 * structures cannot be added after this point. For process groups this is
1234 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1235 * as an interlock.
1236 */
1237 void
funsetownlst(struct sigiolst * sigiolst)1238 funsetownlst(struct sigiolst *sigiolst)
1239 {
1240 struct proc *p;
1241 struct pgrp *pg;
1242 struct sigio *sigio, *tmp;
1243
1244 /* Racy check. */
1245 sigio = SLIST_FIRST(sigiolst);
1246 if (sigio == NULL)
1247 return;
1248
1249 p = NULL;
1250 pg = NULL;
1251
1252 SIGIO_LOCK();
1253 sigio = SLIST_FIRST(sigiolst);
1254 if (sigio == NULL) {
1255 SIGIO_UNLOCK();
1256 return;
1257 }
1258
1259 /*
1260 * Every entry of the list should belong to a single proc or pgrp.
1261 */
1262 if (sigio->sio_pgid < 0) {
1263 pg = sigio->sio_pgrp;
1264 sx_assert(&proctree_lock, SX_XLOCKED);
1265 PGRP_LOCK(pg);
1266 } else /* if (sigio->sio_pgid > 0) */ {
1267 p = sigio->sio_proc;
1268 PROC_LOCK(p);
1269 KASSERT((p->p_flag & P_WEXIT) != 0,
1270 ("%s: process %p is not exiting", __func__, p));
1271 }
1272
1273 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1274 *sigio->sio_myref = NULL;
1275 if (pg != NULL) {
1276 KASSERT(sigio->sio_pgid < 0,
1277 ("Proc sigio in pgrp sigio list"));
1278 KASSERT(sigio->sio_pgrp == pg,
1279 ("Bogus pgrp in sigio list"));
1280 } else /* if (p != NULL) */ {
1281 KASSERT(sigio->sio_pgid > 0,
1282 ("Pgrp sigio in proc sigio list"));
1283 KASSERT(sigio->sio_proc == p,
1284 ("Bogus proc in sigio list"));
1285 }
1286 }
1287
1288 if (pg != NULL)
1289 PGRP_UNLOCK(pg);
1290 else
1291 PROC_UNLOCK(p);
1292 SIGIO_UNLOCK();
1293
1294 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1295 sigiofree(sigio);
1296 }
1297
1298 /*
1299 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1300 *
1301 * After permission checking, add a sigio structure to the sigio list for
1302 * the process or process group.
1303 */
1304 int
fsetown(pid_t pgid,struct sigio ** sigiop)1305 fsetown(pid_t pgid, struct sigio **sigiop)
1306 {
1307 struct proc *proc;
1308 struct pgrp *pgrp;
1309 struct sigio *osigio, *sigio;
1310 int ret;
1311
1312 if (pgid == 0) {
1313 funsetown(sigiop);
1314 return (0);
1315 }
1316
1317 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1318 sigio->sio_pgid = pgid;
1319 sigio->sio_ucred = crhold(curthread->td_ucred);
1320 sigio->sio_myref = sigiop;
1321
1322 ret = 0;
1323 if (pgid > 0) {
1324 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1325 SIGIO_LOCK();
1326 osigio = funsetown_locked(*sigiop);
1327 if (ret == 0) {
1328 PROC_LOCK(proc);
1329 _PRELE(proc);
1330 if ((proc->p_flag & P_WEXIT) != 0) {
1331 ret = ESRCH;
1332 } else if (proc->p_session !=
1333 curthread->td_proc->p_session) {
1334 /*
1335 * Policy - Don't allow a process to FSETOWN a
1336 * process in another session.
1337 *
1338 * Remove this test to allow maximum flexibility
1339 * or restrict FSETOWN to the current process or
1340 * process group for maximum safety.
1341 */
1342 ret = EPERM;
1343 } else {
1344 sigio->sio_proc = proc;
1345 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1346 sio_pgsigio);
1347 }
1348 PROC_UNLOCK(proc);
1349 }
1350 } else /* if (pgid < 0) */ {
1351 sx_slock(&proctree_lock);
1352 SIGIO_LOCK();
1353 osigio = funsetown_locked(*sigiop);
1354 pgrp = pgfind(-pgid);
1355 if (pgrp == NULL) {
1356 ret = ESRCH;
1357 } else {
1358 if (pgrp->pg_session != curthread->td_proc->p_session) {
1359 /*
1360 * Policy - Don't allow a process to FSETOWN a
1361 * process in another session.
1362 *
1363 * Remove this test to allow maximum flexibility
1364 * or restrict FSETOWN to the current process or
1365 * process group for maximum safety.
1366 */
1367 ret = EPERM;
1368 } else {
1369 sigio->sio_pgrp = pgrp;
1370 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1371 sio_pgsigio);
1372 }
1373 PGRP_UNLOCK(pgrp);
1374 }
1375 sx_sunlock(&proctree_lock);
1376 }
1377 if (ret == 0)
1378 *sigiop = sigio;
1379 SIGIO_UNLOCK();
1380 if (osigio != NULL)
1381 sigiofree(osigio);
1382 return (ret);
1383 }
1384
1385 /*
1386 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1387 */
1388 pid_t
fgetown(struct sigio ** sigiop)1389 fgetown(struct sigio **sigiop)
1390 {
1391 pid_t pgid;
1392
1393 SIGIO_LOCK();
1394 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1395 SIGIO_UNLOCK();
1396 return (pgid);
1397 }
1398
1399 static int
closefp_impl(struct filedesc * fdp,int fd,struct file * fp,struct thread * td,bool audit)1400 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1401 bool audit)
1402 {
1403 int error;
1404
1405 FILEDESC_XLOCK_ASSERT(fdp);
1406
1407 /*
1408 * We now hold the fp reference that used to be owned by the
1409 * descriptor array. We have to unlock the FILEDESC *AFTER*
1410 * knote_fdclose to prevent a race of the fd getting opened, a knote
1411 * added, and deleteing a knote for the new fd.
1412 */
1413 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1414 knote_fdclose(td, fd);
1415
1416 /*
1417 * We need to notify mqueue if the object is of type mqueue.
1418 */
1419 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1420 mq_fdclose(td, fd, fp);
1421 FILEDESC_XUNLOCK(fdp);
1422
1423 #ifdef AUDIT
1424 if (AUDITING_TD(td) && audit)
1425 audit_sysclose(td, fd, fp);
1426 #endif
1427 error = closef(fp, td);
1428
1429 /*
1430 * All paths leading up to closefp() will have already removed or
1431 * replaced the fd in the filedesc table, so a restart would not
1432 * operate on the same file.
1433 */
1434 if (error == ERESTART)
1435 error = EINTR;
1436
1437 return (error);
1438 }
1439
1440 static int
closefp_hl(struct filedesc * fdp,int fd,struct file * fp,struct thread * td,bool holdleaders,bool audit)1441 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1442 bool holdleaders, bool audit)
1443 {
1444 int error;
1445
1446 FILEDESC_XLOCK_ASSERT(fdp);
1447
1448 if (holdleaders) {
1449 if (td->td_proc->p_fdtol != NULL) {
1450 /*
1451 * Ask fdfree() to sleep to ensure that all relevant
1452 * process leaders can be traversed in closef().
1453 */
1454 fdp->fd_holdleaderscount++;
1455 } else {
1456 holdleaders = false;
1457 }
1458 }
1459
1460 error = closefp_impl(fdp, fd, fp, td, audit);
1461 if (holdleaders) {
1462 FILEDESC_XLOCK(fdp);
1463 fdp->fd_holdleaderscount--;
1464 if (fdp->fd_holdleaderscount == 0 &&
1465 fdp->fd_holdleaderswakeup != 0) {
1466 fdp->fd_holdleaderswakeup = 0;
1467 wakeup(&fdp->fd_holdleaderscount);
1468 }
1469 FILEDESC_XUNLOCK(fdp);
1470 }
1471 return (error);
1472 }
1473
1474 static int
closefp(struct filedesc * fdp,int fd,struct file * fp,struct thread * td,bool holdleaders,bool audit)1475 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1476 bool holdleaders, bool audit)
1477 {
1478
1479 FILEDESC_XLOCK_ASSERT(fdp);
1480
1481 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1482 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1483 } else {
1484 return (closefp_impl(fdp, fd, fp, td, audit));
1485 }
1486 }
1487
1488 /*
1489 * Close a file descriptor.
1490 */
1491 #ifndef _SYS_SYSPROTO_H_
1492 struct close_args {
1493 int fd;
1494 };
1495 #endif
1496 /* ARGSUSED */
1497 int
sys_close(struct thread * td,struct close_args * uap)1498 sys_close(struct thread *td, struct close_args *uap)
1499 {
1500
1501 return (kern_close(td, uap->fd));
1502 }
1503
1504 int
kern_close(struct thread * td,int fd)1505 kern_close(struct thread *td, int fd)
1506 {
1507 struct filedesc *fdp;
1508 struct file *fp;
1509
1510 fdp = td->td_proc->p_fd;
1511
1512 FILEDESC_XLOCK(fdp);
1513 if ((fp = fget_noref(fdp, fd)) == NULL) {
1514 FILEDESC_XUNLOCK(fdp);
1515 return (EBADF);
1516 }
1517 fdfree(fdp, fd);
1518
1519 /* closefp() drops the FILEDESC lock for us. */
1520 return (closefp(fdp, fd, fp, td, true, true));
1521 }
1522
1523 static int
close_range_flags(struct thread * td,u_int lowfd,u_int highfd,int flags)1524 close_range_flags(struct thread *td, u_int lowfd, u_int highfd, int flags)
1525 {
1526 struct filedesc *fdp;
1527 struct fdescenttbl *fdt;
1528 struct filedescent *fde;
1529 int fd, fde_flags;
1530
1531 fde_flags = close_range_to_fde_flags(flags);
1532 fdp = td->td_proc->p_fd;
1533 FILEDESC_XLOCK(fdp);
1534 fdt = atomic_load_ptr(&fdp->fd_files);
1535 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1536 fd = lowfd;
1537 if (__predict_false(fd > highfd)) {
1538 goto out_locked;
1539 }
1540 for (; fd <= highfd; fd++) {
1541 fde = &fdt->fdt_ofiles[fd];
1542 if (fde->fde_file != NULL)
1543 fde->fde_flags |= fde_flags;
1544 }
1545 out_locked:
1546 FILEDESC_XUNLOCK(fdp);
1547 return (0);
1548 }
1549
1550 static int
close_range_impl(struct thread * td,u_int lowfd,u_int highfd)1551 close_range_impl(struct thread *td, u_int lowfd, u_int highfd)
1552 {
1553 struct filedesc *fdp;
1554 const struct fdescenttbl *fdt;
1555 struct file *fp;
1556 int fd;
1557
1558 fdp = td->td_proc->p_fd;
1559 FILEDESC_XLOCK(fdp);
1560 fdt = atomic_load_ptr(&fdp->fd_files);
1561 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1562 fd = lowfd;
1563 if (__predict_false(fd > highfd)) {
1564 goto out_locked;
1565 }
1566 for (;;) {
1567 fp = fdt->fdt_ofiles[fd].fde_file;
1568 if (fp == NULL) {
1569 if (fd == highfd)
1570 goto out_locked;
1571 } else {
1572 fdfree(fdp, fd);
1573 (void) closefp(fdp, fd, fp, td, true, true);
1574 if (fd == highfd)
1575 goto out_unlocked;
1576 FILEDESC_XLOCK(fdp);
1577 fdt = atomic_load_ptr(&fdp->fd_files);
1578 }
1579 fd++;
1580 }
1581 out_locked:
1582 FILEDESC_XUNLOCK(fdp);
1583 out_unlocked:
1584 return (0);
1585 }
1586
1587 int
kern_close_range(struct thread * td,int flags,u_int lowfd,u_int highfd)1588 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd)
1589 {
1590
1591 /*
1592 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1593 * open should not be a usage error. From a close_range() perspective,
1594 * close_range(3, ~0U, 0) in the same scenario should also likely not
1595 * be a usage error as all fd above 3 are in-fact already closed.
1596 */
1597 if (highfd < lowfd) {
1598 return (EINVAL);
1599 }
1600
1601 if ((flags & (CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_CLOFORK)) != 0)
1602 return (close_range_flags(td, lowfd, highfd, flags));
1603
1604 return (close_range_impl(td, lowfd, highfd));
1605 }
1606
1607 #ifndef _SYS_SYSPROTO_H_
1608 struct close_range_args {
1609 u_int lowfd;
1610 u_int highfd;
1611 int flags;
1612 };
1613 #endif
1614 int
sys_close_range(struct thread * td,struct close_range_args * uap)1615 sys_close_range(struct thread *td, struct close_range_args *uap)
1616 {
1617
1618 AUDIT_ARG_FD(uap->lowfd);
1619 AUDIT_ARG_CMD(uap->highfd);
1620 AUDIT_ARG_FFLAGS(uap->flags);
1621
1622 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_CLOFORK)) != 0)
1623 return (EINVAL);
1624 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd));
1625 }
1626
1627 #ifdef COMPAT_FREEBSD12
1628 /*
1629 * Close open file descriptors.
1630 */
1631 #ifndef _SYS_SYSPROTO_H_
1632 struct freebsd12_closefrom_args {
1633 int lowfd;
1634 };
1635 #endif
1636 /* ARGSUSED */
1637 int
freebsd12_closefrom(struct thread * td,struct freebsd12_closefrom_args * uap)1638 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1639 {
1640 u_int lowfd;
1641
1642 AUDIT_ARG_FD(uap->lowfd);
1643
1644 /*
1645 * Treat negative starting file descriptor values identical to
1646 * closefrom(0) which closes all files.
1647 */
1648 lowfd = MAX(0, uap->lowfd);
1649 return (kern_close_range(td, 0, lowfd, ~0U));
1650 }
1651 #endif /* COMPAT_FREEBSD12 */
1652
1653 #if defined(COMPAT_43)
1654 /*
1655 * Return status information about a file descriptor.
1656 */
1657 #ifndef _SYS_SYSPROTO_H_
1658 struct ofstat_args {
1659 int fd;
1660 struct ostat *sb;
1661 };
1662 #endif
1663 /* ARGSUSED */
1664 int
ofstat(struct thread * td,struct ofstat_args * uap)1665 ofstat(struct thread *td, struct ofstat_args *uap)
1666 {
1667 struct ostat oub;
1668 struct stat ub;
1669 int error;
1670
1671 error = kern_fstat(td, uap->fd, &ub);
1672 if (error == 0) {
1673 cvtstat(&ub, &oub);
1674 error = copyout(&oub, uap->sb, sizeof(oub));
1675 }
1676 return (error);
1677 }
1678 #endif /* COMPAT_43 */
1679
1680 #if defined(COMPAT_FREEBSD11)
1681 int
freebsd11_fstat(struct thread * td,struct freebsd11_fstat_args * uap)1682 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1683 {
1684 struct stat sb;
1685 struct freebsd11_stat osb;
1686 int error;
1687
1688 error = kern_fstat(td, uap->fd, &sb);
1689 if (error != 0)
1690 return (error);
1691 error = freebsd11_cvtstat(&sb, &osb);
1692 if (error == 0)
1693 error = copyout(&osb, uap->sb, sizeof(osb));
1694 return (error);
1695 }
1696 #endif /* COMPAT_FREEBSD11 */
1697
1698 /*
1699 * Return status information about a file descriptor.
1700 */
1701 #ifndef _SYS_SYSPROTO_H_
1702 struct fstat_args {
1703 int fd;
1704 struct stat *sb;
1705 };
1706 #endif
1707 /* ARGSUSED */
1708 int
sys_fstat(struct thread * td,struct fstat_args * uap)1709 sys_fstat(struct thread *td, struct fstat_args *uap)
1710 {
1711 struct stat ub;
1712 int error;
1713
1714 error = kern_fstat(td, uap->fd, &ub);
1715 if (error == 0)
1716 error = copyout(&ub, uap->sb, sizeof(ub));
1717 return (error);
1718 }
1719
1720 int
kern_fstat(struct thread * td,int fd,struct stat * sbp)1721 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1722 {
1723 struct file *fp;
1724 int error;
1725
1726 AUDIT_ARG_FD(fd);
1727
1728 error = fget(td, fd, &cap_fstat_rights, &fp);
1729 if (__predict_false(error != 0))
1730 return (error);
1731
1732 AUDIT_ARG_FILE(td->td_proc, fp);
1733
1734 sbp->st_filerev = 0;
1735 sbp->st_bsdflags = 0;
1736 error = fo_stat(fp, sbp, td->td_ucred);
1737 fdrop(fp, td);
1738 #ifdef __STAT_TIME_T_EXT
1739 sbp->st_atim_ext = 0;
1740 sbp->st_mtim_ext = 0;
1741 sbp->st_ctim_ext = 0;
1742 sbp->st_btim_ext = 0;
1743 #endif
1744 #ifdef KTRACE
1745 if (KTRPOINT(td, KTR_STRUCT))
1746 ktrstat_error(sbp, error);
1747 #endif
1748 return (error);
1749 }
1750
1751 #if defined(COMPAT_FREEBSD11)
1752 /*
1753 * Return status information about a file descriptor.
1754 */
1755 #ifndef _SYS_SYSPROTO_H_
1756 struct freebsd11_nfstat_args {
1757 int fd;
1758 struct nstat *sb;
1759 };
1760 #endif
1761 /* ARGSUSED */
1762 int
freebsd11_nfstat(struct thread * td,struct freebsd11_nfstat_args * uap)1763 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1764 {
1765 struct nstat nub;
1766 struct stat ub;
1767 int error;
1768
1769 error = kern_fstat(td, uap->fd, &ub);
1770 if (error != 0)
1771 return (error);
1772 error = freebsd11_cvtnstat(&ub, &nub);
1773 if (error != 0)
1774 error = copyout(&nub, uap->sb, sizeof(nub));
1775 return (error);
1776 }
1777 #endif /* COMPAT_FREEBSD11 */
1778
1779 /*
1780 * Return pathconf information about a file descriptor.
1781 */
1782 #ifndef _SYS_SYSPROTO_H_
1783 struct fpathconf_args {
1784 int fd;
1785 int name;
1786 };
1787 #endif
1788 /* ARGSUSED */
1789 int
sys_fpathconf(struct thread * td,struct fpathconf_args * uap)1790 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1791 {
1792 long value;
1793 int error;
1794
1795 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1796 if (error == 0)
1797 td->td_retval[0] = value;
1798 return (error);
1799 }
1800
1801 int
kern_fpathconf(struct thread * td,int fd,int name,long * valuep)1802 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1803 {
1804 struct file *fp;
1805 struct vnode *vp;
1806 int error;
1807
1808 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1809 if (error != 0)
1810 return (error);
1811
1812 if (name == _PC_ASYNC_IO) {
1813 *valuep = _POSIX_ASYNCHRONOUS_IO;
1814 goto out;
1815 }
1816 vp = fp->f_vnode;
1817 if (vp != NULL) {
1818 vn_lock(vp, LK_SHARED | LK_RETRY);
1819 error = VOP_PATHCONF(vp, name, valuep);
1820 VOP_UNLOCK(vp);
1821 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1822 if (name != _PC_PIPE_BUF) {
1823 error = EINVAL;
1824 } else {
1825 *valuep = PIPE_BUF;
1826 error = 0;
1827 }
1828 } else {
1829 error = EOPNOTSUPP;
1830 }
1831 out:
1832 fdrop(fp, td);
1833 return (error);
1834 }
1835
1836 /*
1837 * Copy filecaps structure allocating memory for ioctls array if needed.
1838 *
1839 * The last parameter indicates whether the fdtable is locked. If it is not and
1840 * ioctls are encountered, copying fails and the caller must lock the table.
1841 *
1842 * Note that if the table was not locked, the caller has to check the relevant
1843 * sequence counter to determine whether the operation was successful.
1844 */
1845 bool
filecaps_copy(const struct filecaps * src,struct filecaps * dst,bool locked)1846 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1847 {
1848 size_t size;
1849
1850 if (src->fc_ioctls != NULL && !locked)
1851 return (false);
1852 memcpy(dst, src, sizeof(*src));
1853 if (src->fc_ioctls == NULL)
1854 return (true);
1855
1856 KASSERT(src->fc_nioctls > 0,
1857 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1858
1859 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1860 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1861 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1862 return (true);
1863 }
1864
1865 static u_long *
filecaps_copy_prep(const struct filecaps * src)1866 filecaps_copy_prep(const struct filecaps *src)
1867 {
1868 u_long *ioctls;
1869 size_t size;
1870
1871 if (__predict_true(src->fc_ioctls == NULL))
1872 return (NULL);
1873
1874 KASSERT(src->fc_nioctls > 0,
1875 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1876
1877 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1878 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1879 return (ioctls);
1880 }
1881
1882 static void
filecaps_copy_finish(const struct filecaps * src,struct filecaps * dst,u_long * ioctls)1883 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1884 u_long *ioctls)
1885 {
1886 size_t size;
1887
1888 *dst = *src;
1889 if (__predict_true(src->fc_ioctls == NULL)) {
1890 MPASS(ioctls == NULL);
1891 return;
1892 }
1893
1894 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1895 dst->fc_ioctls = ioctls;
1896 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1897 }
1898
1899 /*
1900 * Move filecaps structure to the new place and clear the old place.
1901 */
1902 void
filecaps_move(struct filecaps * src,struct filecaps * dst)1903 filecaps_move(struct filecaps *src, struct filecaps *dst)
1904 {
1905
1906 *dst = *src;
1907 bzero(src, sizeof(*src));
1908 }
1909
1910 /*
1911 * Fill the given filecaps structure with full rights.
1912 */
1913 static void
filecaps_fill(struct filecaps * fcaps)1914 filecaps_fill(struct filecaps *fcaps)
1915 {
1916
1917 CAP_ALL(&fcaps->fc_rights);
1918 fcaps->fc_ioctls = NULL;
1919 fcaps->fc_nioctls = -1;
1920 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1921 }
1922
1923 /*
1924 * Free memory allocated within filecaps structure.
1925 */
1926 static void
filecaps_free_ioctl(struct filecaps * fcaps)1927 filecaps_free_ioctl(struct filecaps *fcaps)
1928 {
1929
1930 free(fcaps->fc_ioctls, M_FILECAPS);
1931 fcaps->fc_ioctls = NULL;
1932 }
1933
1934 void
filecaps_free(struct filecaps * fcaps)1935 filecaps_free(struct filecaps *fcaps)
1936 {
1937
1938 filecaps_free_ioctl(fcaps);
1939 bzero(fcaps, sizeof(*fcaps));
1940 }
1941
1942 static u_long *
filecaps_free_prep(struct filecaps * fcaps)1943 filecaps_free_prep(struct filecaps *fcaps)
1944 {
1945 u_long *ioctls;
1946
1947 ioctls = fcaps->fc_ioctls;
1948 bzero(fcaps, sizeof(*fcaps));
1949 return (ioctls);
1950 }
1951
1952 static void
filecaps_free_finish(u_long * ioctls)1953 filecaps_free_finish(u_long *ioctls)
1954 {
1955
1956 free(ioctls, M_FILECAPS);
1957 }
1958
1959 /*
1960 * Validate the given filecaps structure.
1961 */
1962 static void
filecaps_validate(const struct filecaps * fcaps,const char * func)1963 filecaps_validate(const struct filecaps *fcaps, const char *func)
1964 {
1965
1966 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1967 ("%s: invalid rights", func));
1968 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1969 ("%s: invalid fcntls", func));
1970 KASSERT(fcaps->fc_fcntls == 0 ||
1971 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1972 ("%s: fcntls without CAP_FCNTL", func));
1973 /*
1974 * open calls without WANTIOCTLCAPS free caps but leave the counter
1975 */
1976 #if 0
1977 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1978 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1979 ("%s: invalid ioctls", func));
1980 #endif
1981 KASSERT(fcaps->fc_nioctls == 0 ||
1982 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1983 ("%s: ioctls without CAP_IOCTL", func));
1984 }
1985
1986 static void
fdgrowtable_exp(struct filedesc * fdp,int nfd)1987 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1988 {
1989 int nfd1;
1990
1991 FILEDESC_XLOCK_ASSERT(fdp);
1992
1993 nfd1 = fdp->fd_nfiles * 2;
1994 if (nfd1 < nfd)
1995 nfd1 = nfd;
1996 fdgrowtable(fdp, nfd1);
1997 }
1998
1999 /*
2000 * Grow the file table to accommodate (at least) nfd descriptors.
2001 */
2002 static void
fdgrowtable(struct filedesc * fdp,int nfd)2003 fdgrowtable(struct filedesc *fdp, int nfd)
2004 {
2005 struct filedesc0 *fdp0;
2006 struct freetable *ft;
2007 struct fdescenttbl *ntable;
2008 struct fdescenttbl *otable;
2009 int nnfiles, onfiles;
2010 NDSLOTTYPE *nmap, *omap;
2011
2012 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
2013
2014 /* save old values */
2015 onfiles = fdp->fd_nfiles;
2016 otable = fdp->fd_files;
2017 omap = fdp->fd_map;
2018
2019 /* compute the size of the new table */
2020 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
2021 if (nnfiles <= onfiles)
2022 /* the table is already large enough */
2023 return;
2024
2025 /*
2026 * Allocate a new table. We need enough space for the number of
2027 * entries, file entries themselves and the struct freetable we will use
2028 * when we decommission the table and place it on the freelist.
2029 * We place the struct freetable in the middle so we don't have
2030 * to worry about padding.
2031 */
2032 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
2033 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
2034 sizeof(struct freetable),
2035 M_FILEDESC, M_ZERO | M_WAITOK);
2036 /* copy the old data */
2037 ntable->fdt_nfiles = nnfiles;
2038 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
2039 onfiles * sizeof(ntable->fdt_ofiles[0]));
2040
2041 /*
2042 * Allocate a new map only if the old is not large enough. It will
2043 * grow at a slower rate than the table as it can map more
2044 * entries than the table can hold.
2045 */
2046 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
2047 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
2048 M_ZERO | M_WAITOK);
2049 /* copy over the old data and update the pointer */
2050 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
2051 fdp->fd_map = nmap;
2052 }
2053
2054 /*
2055 * Make sure that ntable is correctly initialized before we replace
2056 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
2057 * data.
2058 */
2059 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
2060
2061 /*
2062 * Free the old file table when not shared by other threads or processes.
2063 * The old file table is considered to be shared when either are true:
2064 * - The process has more than one thread.
2065 * - The file descriptor table has been shared via fdshare().
2066 *
2067 * When shared, the old file table will be placed on a freelist
2068 * which will be processed when the struct filedesc is released.
2069 *
2070 * Note that if onfiles == NDFILE, we're dealing with the original
2071 * static allocation contained within (struct filedesc0 *)fdp,
2072 * which must not be freed.
2073 */
2074 if (onfiles > NDFILE) {
2075 /*
2076 * Note we may be called here from fdinit while allocating a
2077 * table for a new process in which case ->p_fd points
2078 * elsewhere.
2079 */
2080 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
2081 free(otable, M_FILEDESC);
2082 } else {
2083 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
2084 fdp0 = (struct filedesc0 *)fdp;
2085 ft->ft_table = otable;
2086 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
2087 }
2088 }
2089 /*
2090 * The map does not have the same possibility of threads still
2091 * holding references to it. So always free it as long as it
2092 * does not reference the original static allocation.
2093 */
2094 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
2095 free(omap, M_FILEDESC);
2096 }
2097
2098 /*
2099 * Allocate a file descriptor for the process.
2100 */
2101 int
fdalloc(struct thread * td,int minfd,int * result)2102 fdalloc(struct thread *td, int minfd, int *result)
2103 {
2104 struct proc *p = td->td_proc;
2105 struct filedesc *fdp = p->p_fd;
2106 int fd, maxfd, allocfd;
2107 #ifdef RACCT
2108 int error;
2109 #endif
2110
2111 FILEDESC_XLOCK_ASSERT(fdp);
2112
2113 if (fdp->fd_freefile > minfd)
2114 minfd = fdp->fd_freefile;
2115
2116 maxfd = getmaxfd(td);
2117
2118 /*
2119 * Search the bitmap for a free descriptor starting at minfd.
2120 * If none is found, grow the file table.
2121 */
2122 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
2123 if (__predict_false(fd >= maxfd))
2124 return (EMFILE);
2125 if (__predict_false(fd >= fdp->fd_nfiles)) {
2126 allocfd = min(fd * 2, maxfd);
2127 #ifdef RACCT
2128 if (RACCT_ENABLED()) {
2129 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
2130 if (error != 0)
2131 return (EMFILE);
2132 }
2133 #endif
2134 /*
2135 * fd is already equal to first free descriptor >= minfd, so
2136 * we only need to grow the table and we are done.
2137 */
2138 fdgrowtable_exp(fdp, allocfd);
2139 }
2140
2141 /*
2142 * Perform some sanity checks, then mark the file descriptor as
2143 * used and return it to the caller.
2144 */
2145 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
2146 ("invalid descriptor %d", fd));
2147 KASSERT(!fdisused(fdp, fd),
2148 ("fd_first_free() returned non-free descriptor"));
2149 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
2150 ("file descriptor isn't free"));
2151 fdused(fdp, fd);
2152 *result = fd;
2153 return (0);
2154 }
2155
2156 /*
2157 * Allocate n file descriptors for the process.
2158 */
2159 int
fdallocn(struct thread * td,int minfd,int * fds,int n)2160 fdallocn(struct thread *td, int minfd, int *fds, int n)
2161 {
2162 struct proc *p = td->td_proc;
2163 struct filedesc *fdp = p->p_fd;
2164 int i;
2165
2166 FILEDESC_XLOCK_ASSERT(fdp);
2167
2168 for (i = 0; i < n; i++)
2169 if (fdalloc(td, 0, &fds[i]) != 0)
2170 break;
2171
2172 if (i < n) {
2173 for (i--; i >= 0; i--)
2174 fdunused(fdp, fds[i]);
2175 return (EMFILE);
2176 }
2177
2178 return (0);
2179 }
2180
2181 /*
2182 * Create a new open file structure and allocate a file descriptor for the
2183 * process that refers to it. We add one reference to the file for the
2184 * descriptor table and one reference for resultfp. This is to prevent us
2185 * being preempted and the entry in the descriptor table closed after we
2186 * release the FILEDESC lock.
2187 */
2188 int
falloc_caps(struct thread * td,struct file ** resultfp,int * resultfd,int flags,struct filecaps * fcaps)2189 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2190 struct filecaps *fcaps)
2191 {
2192 struct file *fp;
2193 int error, fd;
2194
2195 MPASS(resultfp != NULL);
2196 MPASS(resultfd != NULL);
2197
2198 error = _falloc_noinstall(td, &fp, 2);
2199 if (__predict_false(error != 0)) {
2200 return (error);
2201 }
2202
2203 error = finstall_refed(td, fp, &fd, flags, fcaps);
2204 if (__predict_false(error != 0)) {
2205 falloc_abort(td, fp);
2206 return (error);
2207 }
2208
2209 *resultfp = fp;
2210 *resultfd = fd;
2211
2212 return (0);
2213 }
2214
2215 /*
2216 * Create a new open file structure without allocating a file descriptor.
2217 */
2218 int
_falloc_noinstall(struct thread * td,struct file ** resultfp,u_int n)2219 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2220 {
2221 struct file *fp;
2222 int maxuserfiles = maxfiles - (maxfiles / 20);
2223 int openfiles_new;
2224 static struct timeval lastfail;
2225 static int curfail;
2226
2227 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2228 MPASS(n > 0);
2229
2230 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2231 if ((openfiles_new >= maxuserfiles &&
2232 priv_check(td, PRIV_MAXFILES) != 0) ||
2233 openfiles_new >= maxfiles) {
2234 atomic_subtract_int(&openfiles, 1);
2235 if (ppsratecheck(&lastfail, &curfail, 1)) {
2236 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2237 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2238 }
2239 return (ENFILE);
2240 }
2241 fp = uma_zalloc(file_zone, M_WAITOK);
2242 bzero(fp, sizeof(*fp));
2243 refcount_init(&fp->f_count, n);
2244 fp->f_cred = crhold(td->td_ucred);
2245 fp->f_ops = &badfileops;
2246 *resultfp = fp;
2247 return (0);
2248 }
2249
2250 void
falloc_abort(struct thread * td,struct file * fp)2251 falloc_abort(struct thread *td, struct file *fp)
2252 {
2253
2254 /*
2255 * For assertion purposes.
2256 */
2257 refcount_init(&fp->f_count, 0);
2258 _fdrop(fp, td);
2259 }
2260
2261 /*
2262 * Install a file in a file descriptor table.
2263 */
2264 void
_finstall(struct filedesc * fdp,struct file * fp,int fd,int flags,struct filecaps * fcaps)2265 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2266 struct filecaps *fcaps)
2267 {
2268 struct filedescent *fde;
2269
2270 MPASS(fp != NULL);
2271 if (fcaps != NULL)
2272 filecaps_validate(fcaps, __func__);
2273 FILEDESC_XLOCK_ASSERT(fdp);
2274
2275 fde = &fdp->fd_ofiles[fd];
2276 #ifdef CAPABILITIES
2277 seqc_write_begin(&fde->fde_seqc);
2278 #endif
2279 fde->fde_file = fp;
2280 fde->fde_flags = open_to_fde_flags(flags, true);
2281 if (fcaps != NULL)
2282 filecaps_move(fcaps, &fde->fde_caps);
2283 else
2284 filecaps_fill(&fde->fde_caps);
2285 #ifdef CAPABILITIES
2286 seqc_write_end(&fde->fde_seqc);
2287 #endif
2288 }
2289
2290 int
finstall_refed(struct thread * td,struct file * fp,int * fd,int flags,struct filecaps * fcaps)2291 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2292 struct filecaps *fcaps)
2293 {
2294 struct filedesc *fdp = td->td_proc->p_fd;
2295 int error;
2296
2297 MPASS(fd != NULL);
2298
2299 FILEDESC_XLOCK(fdp);
2300 error = fdalloc(td, 0, fd);
2301 if (__predict_true(error == 0)) {
2302 _finstall(fdp, fp, *fd, flags, fcaps);
2303 }
2304 FILEDESC_XUNLOCK(fdp);
2305 return (error);
2306 }
2307
2308 int
finstall(struct thread * td,struct file * fp,int * fd,int flags,struct filecaps * fcaps)2309 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2310 struct filecaps *fcaps)
2311 {
2312 int error;
2313
2314 MPASS(fd != NULL);
2315
2316 if (!fhold(fp))
2317 return (EBADF);
2318 error = finstall_refed(td, fp, fd, flags, fcaps);
2319 if (__predict_false(error != 0)) {
2320 fdrop(fp, td);
2321 }
2322 return (error);
2323 }
2324
2325 /*
2326 * Build a new filedesc structure from another.
2327 *
2328 * If fdp is not NULL, return with it shared locked.
2329 */
2330 struct filedesc *
fdinit(void)2331 fdinit(void)
2332 {
2333 struct filedesc0 *newfdp0;
2334 struct filedesc *newfdp;
2335
2336 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2337 newfdp = &newfdp0->fd_fd;
2338
2339 /* Create the file descriptor table. */
2340 FILEDESC_LOCK_INIT(newfdp);
2341 refcount_init(&newfdp->fd_refcnt, 1);
2342 refcount_init(&newfdp->fd_holdcnt, 1);
2343 newfdp->fd_map = newfdp0->fd_dmap;
2344 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2345 newfdp->fd_files->fdt_nfiles = NDFILE;
2346
2347 return (newfdp);
2348 }
2349
2350 /*
2351 * Build a pwddesc structure from another.
2352 * Copy the current, root, and jail root vnode references.
2353 *
2354 * If pdp is not NULL and keeplock is true, return with it (exclusively) locked.
2355 */
2356 struct pwddesc *
pdinit(struct pwddesc * pdp,bool keeplock)2357 pdinit(struct pwddesc *pdp, bool keeplock)
2358 {
2359 struct pwddesc *newpdp;
2360 struct pwd *newpwd;
2361
2362 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2363
2364 PWDDESC_LOCK_INIT(newpdp);
2365 refcount_init(&newpdp->pd_refcount, 1);
2366 newpdp->pd_cmask = CMASK;
2367
2368 if (pdp == NULL) {
2369 newpwd = pwd_alloc();
2370 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2371 return (newpdp);
2372 }
2373
2374 PWDDESC_XLOCK(pdp);
2375 newpwd = pwd_hold_pwddesc(pdp);
2376 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2377 if (!keeplock)
2378 PWDDESC_XUNLOCK(pdp);
2379 return (newpdp);
2380 }
2381
2382 /*
2383 * Hold either filedesc or pwddesc of the passed process.
2384 *
2385 * The process lock is used to synchronize against the target exiting and
2386 * freeing the data.
2387 *
2388 * Clearing can be ilustrated in 3 steps:
2389 * 1. set the pointer to NULL. Either routine can race against it, hence
2390 * atomic_load_ptr.
2391 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2392 * race to either still see the pointer or find NULL. It is still safe to
2393 * grab a reference as clearing is stalled.
2394 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2395 * guaranteed to see NULL, making it safe to finish clearing
2396 */
2397 static struct filedesc *
fdhold(struct proc * p)2398 fdhold(struct proc *p)
2399 {
2400 struct filedesc *fdp;
2401
2402 PROC_LOCK_ASSERT(p, MA_OWNED);
2403 fdp = atomic_load_ptr(&p->p_fd);
2404 if (fdp != NULL)
2405 refcount_acquire(&fdp->fd_holdcnt);
2406 return (fdp);
2407 }
2408
2409 static struct pwddesc *
pdhold(struct proc * p)2410 pdhold(struct proc *p)
2411 {
2412 struct pwddesc *pdp;
2413
2414 PROC_LOCK_ASSERT(p, MA_OWNED);
2415 pdp = atomic_load_ptr(&p->p_pd);
2416 if (pdp != NULL)
2417 refcount_acquire(&pdp->pd_refcount);
2418 return (pdp);
2419 }
2420
2421 static void
fddrop(struct filedesc * fdp)2422 fddrop(struct filedesc *fdp)
2423 {
2424
2425 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2426 if (refcount_release(&fdp->fd_holdcnt) == 0)
2427 return;
2428 }
2429
2430 FILEDESC_LOCK_DESTROY(fdp);
2431 uma_zfree(filedesc0_zone, fdp);
2432 }
2433
2434 static void
pddrop(struct pwddesc * pdp)2435 pddrop(struct pwddesc *pdp)
2436 {
2437 struct pwd *pwd;
2438
2439 if (refcount_release_if_not_last(&pdp->pd_refcount))
2440 return;
2441
2442 PWDDESC_XLOCK(pdp);
2443 if (refcount_release(&pdp->pd_refcount) == 0) {
2444 PWDDESC_XUNLOCK(pdp);
2445 return;
2446 }
2447 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2448 pwd_set(pdp, NULL);
2449 PWDDESC_XUNLOCK(pdp);
2450 pwd_drop(pwd);
2451
2452 PWDDESC_LOCK_DESTROY(pdp);
2453 free(pdp, M_PWDDESC);
2454 }
2455
2456 /*
2457 * Share a filedesc structure.
2458 */
2459 struct filedesc *
fdshare(struct filedesc * fdp)2460 fdshare(struct filedesc *fdp)
2461 {
2462
2463 refcount_acquire(&fdp->fd_refcnt);
2464 return (fdp);
2465 }
2466
2467 /*
2468 * Share a pwddesc structure.
2469 */
2470 struct pwddesc *
pdshare(struct pwddesc * pdp)2471 pdshare(struct pwddesc *pdp)
2472 {
2473 refcount_acquire(&pdp->pd_refcount);
2474 return (pdp);
2475 }
2476
2477 /*
2478 * Unshare a filedesc structure, if necessary by making a copy
2479 */
2480 void
fdunshare(struct thread * td)2481 fdunshare(struct thread *td)
2482 {
2483 struct filedesc *tmp;
2484 struct proc *p = td->td_proc;
2485
2486 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2487 return;
2488
2489 tmp = fdcopy(p->p_fd, p);
2490 fdescfree(td);
2491 p->p_fd = tmp;
2492 }
2493
2494 /*
2495 * Unshare a pwddesc structure.
2496 */
2497 void
pdunshare(struct thread * td)2498 pdunshare(struct thread *td)
2499 {
2500 struct pwddesc *pdp;
2501 struct proc *p;
2502
2503 p = td->td_proc;
2504 /* Not shared. */
2505 if (refcount_load(&p->p_pd->pd_refcount) == 1)
2506 return;
2507
2508 pdp = pdcopy(p->p_pd);
2509 pdescfree(td);
2510 p->p_pd = pdp;
2511 }
2512
2513 /*
2514 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2515 * this is to ease callers, not catch errors.
2516 */
2517 struct filedesc *
fdcopy(struct filedesc * fdp,struct proc * p1)2518 fdcopy(struct filedesc *fdp, struct proc *p1)
2519 {
2520 struct filedesc *newfdp;
2521 struct filedescent *nfde, *ofde;
2522 struct file *fp;
2523 int i, lastfile;
2524 bool fork_pass;
2525
2526 MPASS(fdp != NULL);
2527
2528 fork_pass = false;
2529 newfdp = fdinit();
2530 FILEDESC_SLOCK(fdp);
2531 for (;;) {
2532 lastfile = fdlastfile(fdp);
2533 if (lastfile < newfdp->fd_nfiles)
2534 break;
2535 FILEDESC_SUNLOCK(fdp);
2536 fdgrowtable(newfdp, lastfile + 1);
2537 FILEDESC_SLOCK(fdp);
2538 }
2539
2540 /*
2541 * Copy all passable descriptors (i.e. not kqueue), and
2542 * prepare to handle copyable but not passable descriptors
2543 * (kqueues).
2544 *
2545 * The pass to handle copying is performed after all passable
2546 * files are installed into the new file descriptor's table,
2547 * since kqueues need all referenced file descriptors already
2548 * valid, including other kqueues. For the same reason the
2549 * copying is done in two passes by itself, first installing
2550 * not fully initialized ('empty') copyable files into the new
2551 * fd table, and then giving the subsystems a second chance to
2552 * really fill the copied file backing structure with the
2553 * content.
2554 */
2555 newfdp->fd_freefile = fdp->fd_freefile;
2556 FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2557 const struct fileops *ops;
2558
2559 ops = ofde->fde_file->f_ops;
2560 fp = NULL;
2561 if ((ops->fo_flags & DFLAG_FORK) != 0 &&
2562 (ofde->fde_flags & UF_FOCLOSE) == 0) {
2563 if (ops->fo_fork(newfdp, ofde->fde_file, &fp, p1,
2564 curthread) != 0)
2565 continue;
2566 fork_pass = true;
2567 } else if ((ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2568 (ofde->fde_flags & UF_FOCLOSE) != 0 ||
2569 !fhold(ofde->fde_file)) {
2570 if (newfdp->fd_freefile == fdp->fd_freefile)
2571 newfdp->fd_freefile = i;
2572 continue;
2573 }
2574 nfde = &newfdp->fd_ofiles[i];
2575 *nfde = *ofde;
2576 if (fp != NULL)
2577 nfde->fde_file = fp;
2578 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2579 fdused_init(newfdp, i);
2580 }
2581 MPASS(newfdp->fd_freefile != -1);
2582 FILEDESC_SUNLOCK(fdp);
2583
2584 /*
2585 * Now handle copying kqueues, since all fds, including
2586 * kqueues, are in place.
2587 */
2588 if (__predict_false(fork_pass)) {
2589 FILEDESC_FOREACH_FDE(newfdp, i, nfde) {
2590 const struct fileops *ops;
2591
2592 ops = nfde->fde_file->f_ops;
2593 if ((ops->fo_flags & DFLAG_FORK) == 0 ||
2594 nfde->fde_file == NULL)
2595 continue;
2596 ops->fo_fork(newfdp, NULL, &nfde->fde_file, p1,
2597 curthread);
2598 }
2599 }
2600 return (newfdp);
2601 }
2602
2603 /*
2604 * Copy a pwddesc structure.
2605 */
2606 struct pwddesc *
pdcopy(struct pwddesc * pdp)2607 pdcopy(struct pwddesc *pdp)
2608 {
2609 struct pwddesc *newpdp;
2610
2611 MPASS(pdp != NULL);
2612
2613 newpdp = pdinit(pdp, true);
2614 newpdp->pd_cmask = pdp->pd_cmask;
2615 PWDDESC_XUNLOCK(pdp);
2616 return (newpdp);
2617 }
2618
2619 /*
2620 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2621 * one of processes using it exits) and the table used to be shared.
2622 */
2623 static void
fdclearlocks(struct thread * td)2624 fdclearlocks(struct thread *td)
2625 {
2626 struct filedesc *fdp;
2627 struct filedesc_to_leader *fdtol;
2628 struct flock lf;
2629 struct file *fp;
2630 struct proc *p;
2631 struct vnode *vp;
2632 int i;
2633
2634 p = td->td_proc;
2635 fdp = p->p_fd;
2636 fdtol = p->p_fdtol;
2637 MPASS(fdtol != NULL);
2638
2639 FILEDESC_XLOCK(fdp);
2640 KASSERT(fdtol->fdl_refcount > 0,
2641 ("filedesc_to_refcount botch: fdl_refcount=%d",
2642 fdtol->fdl_refcount));
2643 if (fdtol->fdl_refcount == 1 &&
2644 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2645 FILEDESC_FOREACH_FP(fdp, i, fp) {
2646 if (fp->f_type != DTYPE_VNODE ||
2647 !fhold(fp))
2648 continue;
2649 FILEDESC_XUNLOCK(fdp);
2650 lf.l_whence = SEEK_SET;
2651 lf.l_start = 0;
2652 lf.l_len = 0;
2653 lf.l_type = F_UNLCK;
2654 vp = fp->f_vnode;
2655 (void) VOP_ADVLOCK(vp,
2656 (caddr_t)p->p_leader, F_UNLCK,
2657 &lf, F_POSIX);
2658 FILEDESC_XLOCK(fdp);
2659 fdrop(fp, td);
2660 }
2661 }
2662 retry:
2663 if (fdtol->fdl_refcount == 1) {
2664 if (fdp->fd_holdleaderscount > 0 &&
2665 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2666 /*
2667 * close() or kern_dup() has cleared a reference
2668 * in a shared file descriptor table.
2669 */
2670 fdp->fd_holdleaderswakeup = 1;
2671 sx_sleep(&fdp->fd_holdleaderscount,
2672 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2673 goto retry;
2674 }
2675 if (fdtol->fdl_holdcount > 0) {
2676 /*
2677 * Ensure that fdtol->fdl_leader remains
2678 * valid in closef().
2679 */
2680 fdtol->fdl_wakeup = 1;
2681 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2682 "fdlhold", 0);
2683 goto retry;
2684 }
2685 }
2686 fdtol->fdl_refcount--;
2687 if (fdtol->fdl_refcount == 0 &&
2688 fdtol->fdl_holdcount == 0) {
2689 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2690 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2691 } else
2692 fdtol = NULL;
2693 p->p_fdtol = NULL;
2694 FILEDESC_XUNLOCK(fdp);
2695 if (fdtol != NULL)
2696 free(fdtol, M_FILEDESC_TO_LEADER);
2697 }
2698
2699 /*
2700 * Release a filedesc structure.
2701 */
2702 static void
fdescfree_fds(struct thread * td,struct filedesc * fdp)2703 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2704 {
2705 struct filedesc0 *fdp0;
2706 struct freetable *ft, *tft;
2707 struct filedescent *fde;
2708 struct file *fp;
2709 int i;
2710
2711 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2712 ("%s: fd table %p carries references", __func__, fdp));
2713
2714 /*
2715 * Serialize with threads iterating over the table, if any.
2716 */
2717 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2718 FILEDESC_XLOCK(fdp);
2719 FILEDESC_XUNLOCK(fdp);
2720 }
2721
2722 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2723 fp = fde->fde_file;
2724 fdefree_last(fde);
2725 (void) closef(fp, td);
2726 }
2727
2728 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2729 free(fdp->fd_map, M_FILEDESC);
2730 if (fdp->fd_nfiles > NDFILE)
2731 free(fdp->fd_files, M_FILEDESC);
2732
2733 fdp0 = (struct filedesc0 *)fdp;
2734 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2735 free(ft->ft_table, M_FILEDESC);
2736
2737 fddrop(fdp);
2738 }
2739
2740 void
fdescfree(struct thread * td)2741 fdescfree(struct thread *td)
2742 {
2743 struct proc *p;
2744 struct filedesc *fdp;
2745
2746 p = td->td_proc;
2747 fdp = p->p_fd;
2748 MPASS(fdp != NULL);
2749
2750 #ifdef RACCT
2751 if (RACCT_ENABLED())
2752 racct_set_unlocked(p, RACCT_NOFILE, 0);
2753 #endif
2754
2755 if (p->p_fdtol != NULL)
2756 fdclearlocks(td);
2757
2758 /*
2759 * Check fdhold for an explanation.
2760 */
2761 atomic_store_ptr(&p->p_fd, NULL);
2762 atomic_thread_fence_seq_cst();
2763 PROC_WAIT_UNLOCKED(p);
2764
2765 if (refcount_release(&fdp->fd_refcnt) == 0)
2766 return;
2767
2768 fdescfree_fds(td, fdp);
2769 }
2770
2771 void
pdescfree(struct thread * td)2772 pdescfree(struct thread *td)
2773 {
2774 struct proc *p;
2775 struct pwddesc *pdp;
2776
2777 p = td->td_proc;
2778 pdp = p->p_pd;
2779 MPASS(pdp != NULL);
2780
2781 /*
2782 * Check pdhold for an explanation.
2783 */
2784 atomic_store_ptr(&p->p_pd, NULL);
2785 atomic_thread_fence_seq_cst();
2786 PROC_WAIT_UNLOCKED(p);
2787
2788 pddrop(pdp);
2789 }
2790
2791 /*
2792 * For setugid programs, we don't want to people to use that setugidness
2793 * to generate error messages which write to a file which otherwise would
2794 * otherwise be off-limits to the process. We check for filesystems where
2795 * the vnode can change out from under us after execve (like [lin]procfs).
2796 *
2797 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2798 * sufficient. We also don't check for setugidness since we know we are.
2799 */
2800 static bool
is_unsafe(struct file * fp)2801 is_unsafe(struct file *fp)
2802 {
2803 struct vnode *vp;
2804
2805 if (fp->f_type != DTYPE_VNODE)
2806 return (false);
2807
2808 vp = fp->f_vnode;
2809 return ((vp->v_vflag & VV_PROCDEP) != 0);
2810 }
2811
2812 /*
2813 * Make this setguid thing safe, if at all possible.
2814 */
2815 void
fdsetugidsafety(struct thread * td)2816 fdsetugidsafety(struct thread *td)
2817 {
2818 struct filedesc *fdp;
2819 struct file *fp;
2820 int i;
2821
2822 fdp = td->td_proc->p_fd;
2823 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2824 ("the fdtable should not be shared"));
2825 MPASS(fdp->fd_nfiles >= 3);
2826 for (i = 0; i <= 2; i++) {
2827 fp = fdp->fd_ofiles[i].fde_file;
2828 if (fp != NULL && is_unsafe(fp)) {
2829 FILEDESC_XLOCK(fdp);
2830 knote_fdclose(td, i);
2831 /*
2832 * NULL-out descriptor prior to close to avoid
2833 * a race while close blocks.
2834 */
2835 fdfree(fdp, i);
2836 FILEDESC_XUNLOCK(fdp);
2837 (void) closef(fp, td);
2838 }
2839 }
2840 }
2841
2842 /*
2843 * If a specific file object occupies a specific file descriptor, close the
2844 * file descriptor entry and drop a reference on the file object. This is a
2845 * convenience function to handle a subsequent error in a function that calls
2846 * falloc() that handles the race that another thread might have closed the
2847 * file descriptor out from under the thread creating the file object.
2848 */
2849 void
fdclose(struct thread * td,struct file * fp,int idx)2850 fdclose(struct thread *td, struct file *fp, int idx)
2851 {
2852 struct filedesc *fdp = td->td_proc->p_fd;
2853
2854 FILEDESC_XLOCK(fdp);
2855 if (fdp->fd_ofiles[idx].fde_file == fp) {
2856 fdfree(fdp, idx);
2857 FILEDESC_XUNLOCK(fdp);
2858 fdrop(fp, td);
2859 } else
2860 FILEDESC_XUNLOCK(fdp);
2861 }
2862
2863 /*
2864 * Close any files on exec?
2865 */
2866 void
fdcloseexec(struct thread * td)2867 fdcloseexec(struct thread *td)
2868 {
2869 struct filedesc *fdp;
2870 struct filedescent *fde;
2871 struct file *fp;
2872 int i;
2873
2874 fdp = td->td_proc->p_fd;
2875 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2876 ("the fdtable should not be shared"));
2877 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2878 fp = fde->fde_file;
2879 if (fp->f_type == DTYPE_MQUEUE ||
2880 (fde->fde_flags & UF_EXCLOSE)) {
2881 FILEDESC_XLOCK(fdp);
2882 fdfree(fdp, i);
2883 (void) closefp(fdp, i, fp, td, false, false);
2884 FILEDESC_UNLOCK_ASSERT(fdp);
2885 } else if (fde->fde_flags & UF_FOCLOSE) {
2886 /*
2887 * https://austingroupbugs.net/view.php?id=1851
2888 * FD_CLOFORK should not be preserved across exec
2889 */
2890 fde->fde_flags &= ~UF_FOCLOSE;
2891 }
2892 }
2893 }
2894
2895 /*
2896 * It is unsafe for set[ug]id processes to be started with file
2897 * descriptors 0..2 closed, as these descriptors are given implicit
2898 * significance in the Standard C library. fdcheckstd() will create a
2899 * descriptor referencing /dev/null for each of stdin, stdout, and
2900 * stderr that is not already open.
2901 */
2902 int
fdcheckstd(struct thread * td)2903 fdcheckstd(struct thread *td)
2904 {
2905 struct filedesc *fdp;
2906 register_t save;
2907 int i, error, devnull;
2908
2909 fdp = td->td_proc->p_fd;
2910 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2911 ("the fdtable should not be shared"));
2912 MPASS(fdp->fd_nfiles >= 3);
2913 devnull = -1;
2914 for (i = 0; i <= 2; i++) {
2915 if (fdp->fd_ofiles[i].fde_file != NULL)
2916 continue;
2917
2918 save = td->td_retval[0];
2919 if (devnull != -1) {
2920 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2921 } else {
2922 error = kern_openat(td, AT_FDCWD, "/dev/null",
2923 UIO_SYSSPACE, O_RDWR, 0);
2924 if (error == 0) {
2925 devnull = td->td_retval[0];
2926 KASSERT(devnull == i, ("we didn't get our fd"));
2927 }
2928 }
2929 td->td_retval[0] = save;
2930 if (error != 0)
2931 return (error);
2932 }
2933 return (0);
2934 }
2935
2936 /*
2937 * Internal form of close. Decrement reference count on file structure.
2938 * Note: td may be NULL when closing a file that was being passed in a
2939 * message.
2940 */
2941 int
closef(struct file * fp,struct thread * td)2942 closef(struct file *fp, struct thread *td)
2943 {
2944 struct vnode *vp;
2945 struct flock lf;
2946 struct filedesc_to_leader *fdtol;
2947 struct filedesc *fdp;
2948
2949 MPASS(td != NULL);
2950
2951 /*
2952 * POSIX record locking dictates that any close releases ALL
2953 * locks owned by this process. This is handled by setting
2954 * a flag in the unlock to free ONLY locks obeying POSIX
2955 * semantics, and not to free BSD-style file locks.
2956 * If the descriptor was in a message, POSIX-style locks
2957 * aren't passed with the descriptor, and the thread pointer
2958 * will be NULL. Callers should be careful only to pass a
2959 * NULL thread pointer when there really is no owning
2960 * context that might have locks, or the locks will be
2961 * leaked.
2962 */
2963 if (fp->f_type == DTYPE_VNODE) {
2964 vp = fp->f_vnode;
2965 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2966 lf.l_whence = SEEK_SET;
2967 lf.l_start = 0;
2968 lf.l_len = 0;
2969 lf.l_type = F_UNLCK;
2970 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2971 F_UNLCK, &lf, F_POSIX);
2972 }
2973 fdtol = td->td_proc->p_fdtol;
2974 if (fdtol != NULL) {
2975 /*
2976 * Handle special case where file descriptor table is
2977 * shared between multiple process leaders.
2978 */
2979 fdp = td->td_proc->p_fd;
2980 FILEDESC_XLOCK(fdp);
2981 for (fdtol = fdtol->fdl_next;
2982 fdtol != td->td_proc->p_fdtol;
2983 fdtol = fdtol->fdl_next) {
2984 if ((fdtol->fdl_leader->p_flag &
2985 P_ADVLOCK) == 0)
2986 continue;
2987 fdtol->fdl_holdcount++;
2988 FILEDESC_XUNLOCK(fdp);
2989 lf.l_whence = SEEK_SET;
2990 lf.l_start = 0;
2991 lf.l_len = 0;
2992 lf.l_type = F_UNLCK;
2993 vp = fp->f_vnode;
2994 (void) VOP_ADVLOCK(vp,
2995 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2996 F_POSIX);
2997 FILEDESC_XLOCK(fdp);
2998 fdtol->fdl_holdcount--;
2999 if (fdtol->fdl_holdcount == 0 &&
3000 fdtol->fdl_wakeup != 0) {
3001 fdtol->fdl_wakeup = 0;
3002 wakeup(fdtol);
3003 }
3004 }
3005 FILEDESC_XUNLOCK(fdp);
3006 }
3007 }
3008 return (fdrop_close(fp, td));
3009 }
3010
3011 /*
3012 * Hack for file descriptor passing code.
3013 */
3014 void
closef_nothread(struct file * fp)3015 closef_nothread(struct file *fp)
3016 {
3017
3018 fdrop(fp, NULL);
3019 }
3020
3021 /*
3022 * Initialize the file pointer with the specified properties.
3023 *
3024 * The ops are set with release semantics to be certain that the flags, type,
3025 * and data are visible when ops is. This is to prevent ops methods from being
3026 * called with bad data.
3027 */
3028 void
finit(struct file * fp,u_int flag,short type,void * data,const struct fileops * ops)3029 finit(struct file *fp, u_int flag, short type, void *data,
3030 const struct fileops *ops)
3031 {
3032 fp->f_data = data;
3033 fp->f_flag = flag;
3034 fp->f_type = type;
3035 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
3036 }
3037
3038 void
finit_vnode(struct file * fp,u_int flag,void * data,const struct fileops * ops)3039 finit_vnode(struct file *fp, u_int flag, void *data, const struct fileops *ops)
3040 {
3041 fp->f_seqcount[UIO_READ] = 1;
3042 fp->f_seqcount[UIO_WRITE] = 1;
3043 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
3044 data, ops);
3045 }
3046
3047 int
fget_cap_noref(struct filedesc * fdp,int fd,const cap_rights_t * needrightsp,struct file ** fpp,struct filecaps * havecapsp)3048 fget_cap_noref(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp,
3049 struct file **fpp, struct filecaps *havecapsp)
3050 {
3051 struct filedescent *fde;
3052 int error;
3053
3054 FILEDESC_LOCK_ASSERT(fdp);
3055
3056 *fpp = NULL;
3057 fde = fdeget_noref(fdp, fd);
3058 if (fde == NULL) {
3059 error = EBADF;
3060 goto out;
3061 }
3062
3063 #ifdef CAPABILITIES
3064 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
3065 if (error != 0)
3066 goto out;
3067 #endif
3068
3069 if (havecapsp != NULL)
3070 filecaps_copy(&fde->fde_caps, havecapsp, true);
3071
3072 *fpp = fde->fde_file;
3073
3074 error = 0;
3075 out:
3076 return (error);
3077 }
3078
3079 #ifdef CAPABILITIES
3080 int
fget_cap(struct thread * td,int fd,const cap_rights_t * needrightsp,uint8_t * flagsp,struct file ** fpp,struct filecaps * havecapsp)3081 fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp,
3082 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp)
3083 {
3084 struct filedesc *fdp = td->td_proc->p_fd;
3085 int error;
3086 struct file *fp;
3087 seqc_t seq;
3088
3089 *fpp = NULL;
3090 for (;;) {
3091 error = fget_unlocked_seq(td, fd, needrightsp, flagsp, &fp,
3092 &seq);
3093 if (error != 0)
3094 return (error);
3095
3096 if (havecapsp != NULL) {
3097 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
3098 havecapsp, false)) {
3099 fdrop(fp, td);
3100 goto get_locked;
3101 }
3102 }
3103
3104 if (!fd_modified(fdp, fd, seq))
3105 break;
3106 fdrop(fp, td);
3107 }
3108
3109 *fpp = fp;
3110 return (0);
3111
3112 get_locked:
3113 FILEDESC_SLOCK(fdp);
3114 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp);
3115 if (error == 0 && !fhold(*fpp))
3116 error = EBADF;
3117 FILEDESC_SUNLOCK(fdp);
3118 return (error);
3119 }
3120 #else
3121 int
fget_cap(struct thread * td,int fd,const cap_rights_t * needrightsp,uint8_t * flagsp,struct file ** fpp,struct filecaps * havecapsp)3122 fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp,
3123 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp)
3124 {
3125 int error;
3126 error = fget_unlocked_flags(td, fd, needrightsp, flagsp, fpp);
3127 if (havecapsp != NULL && error == 0)
3128 filecaps_fill(havecapsp);
3129
3130 return (error);
3131 }
3132 #endif
3133
3134 int
fget_remote(struct thread * td,struct proc * p,int fd,struct file ** fpp)3135 fget_remote(struct thread *td, struct proc *p, int fd, struct file **fpp)
3136 {
3137 struct filedesc *fdp;
3138 struct file *fp;
3139 int error;
3140
3141 if (p == td->td_proc) /* curproc */
3142 return (fget_unlocked(td, fd, &cap_no_rights, fpp));
3143
3144 PROC_LOCK(p);
3145 fdp = fdhold(p);
3146 PROC_UNLOCK(p);
3147 if (fdp == NULL)
3148 return (ENOENT);
3149 FILEDESC_SLOCK(fdp);
3150 if (refcount_load(&fdp->fd_refcnt) != 0) {
3151 fp = fget_noref(fdp, fd);
3152 if (fp != NULL && fhold(fp)) {
3153 *fpp = fp;
3154 error = 0;
3155 } else {
3156 error = EBADF;
3157 }
3158 } else {
3159 error = ENOENT;
3160 }
3161 FILEDESC_SUNLOCK(fdp);
3162 fddrop(fdp);
3163 return (error);
3164 }
3165
3166 int
fget_remote_foreach(struct thread * td,struct proc * p,int (* fn)(struct proc *,int,struct file *,void *),void * arg)3167 fget_remote_foreach(struct thread *td, struct proc *p,
3168 int (*fn)(struct proc *, int, struct file *, void *), void *arg)
3169 {
3170 struct filedesc *fdp;
3171 struct fdescenttbl *fdt;
3172 struct file *fp;
3173 int error, error1, fd, highfd;
3174
3175 error = 0;
3176 PROC_LOCK(p);
3177 fdp = fdhold(p);
3178 PROC_UNLOCK(p);
3179 if (fdp == NULL)
3180 return (ENOENT);
3181
3182 FILEDESC_SLOCK(fdp);
3183 if (refcount_load(&fdp->fd_refcnt) != 0) {
3184 fdt = atomic_load_ptr(&fdp->fd_files);
3185 highfd = fdt->fdt_nfiles - 1;
3186 FILEDESC_SUNLOCK(fdp);
3187 } else {
3188 error = ENOENT;
3189 FILEDESC_SUNLOCK(fdp);
3190 goto out;
3191 }
3192
3193 for (fd = 0; fd <= highfd; fd++) {
3194 error1 = fget_remote(td, p, fd, &fp);
3195 if (error1 != 0)
3196 continue;
3197 error = fn(p, fd, fp, arg);
3198 fdrop(fp, td);
3199 if (error != 0)
3200 break;
3201 }
3202 out:
3203 fddrop(fdp);
3204 return (error);
3205 }
3206
3207 #ifdef CAPABILITIES
3208 int
fgetvp_lookup_smr(struct nameidata * ndp,struct vnode ** vpp,int * flagsp)3209 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, int *flagsp)
3210 {
3211 const struct filedescent *fde;
3212 const struct fdescenttbl *fdt;
3213 struct filedesc *fdp;
3214 struct file *fp;
3215 struct vnode *vp;
3216 const cap_rights_t *haverights;
3217 cap_rights_t rights;
3218 seqc_t seq;
3219 int fd, flags;
3220
3221 VFS_SMR_ASSERT_ENTERED();
3222
3223 fd = ndp->ni_dirfd;
3224 rights = *ndp->ni_rightsneeded;
3225 cap_rights_set_one(&rights, CAP_LOOKUP);
3226
3227 fdp = curproc->p_fd;
3228 fdt = fdp->fd_files;
3229 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3230 return (EBADF);
3231 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3232 fde = &fdt->fdt_ofiles[fd];
3233 haverights = cap_rights_fde_inline(fde);
3234 fp = fde->fde_file;
3235 if (__predict_false(fp == NULL))
3236 return (EAGAIN);
3237 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
3238 return (EAGAIN);
3239 flags = fp->f_flag & FSEARCH;
3240 flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ?
3241 O_RESOLVE_BENEATH : 0;
3242 vp = fp->f_vnode;
3243 if (__predict_false(vp == NULL)) {
3244 return (EAGAIN);
3245 }
3246 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
3247 return (EAGAIN);
3248 }
3249 /*
3250 * Use an acquire barrier to force re-reading of fdt so it is
3251 * refreshed for verification.
3252 */
3253 atomic_thread_fence_acq();
3254 fdt = fdp->fd_files;
3255 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3256 return (EAGAIN);
3257 /*
3258 * If file descriptor doesn't have all rights,
3259 * all lookups relative to it must also be
3260 * strictly relative.
3261 *
3262 * Not yet supported by fast path.
3263 */
3264 CAP_ALL(&rights);
3265 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3266 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3267 ndp->ni_filecaps.fc_nioctls != -1) {
3268 #ifdef notyet
3269 ndp->ni_lcf |= NI_LCF_STRICTREL;
3270 #else
3271 return (EAGAIN);
3272 #endif
3273 }
3274 *vpp = vp;
3275 *flagsp = flags;
3276 return (0);
3277 }
3278 #else
3279 int
fgetvp_lookup_smr(struct nameidata * ndp,struct vnode ** vpp,int * flagsp)3280 fgetvp_lookup_smr(struct nameidata *ndp, struct vnode **vpp, int *flagsp)
3281 {
3282 const struct filedescent *fde;
3283 const struct fdescenttbl *fdt;
3284 struct filedesc *fdp;
3285 struct file *fp;
3286 struct vnode *vp;
3287 int fd, flags;
3288
3289 VFS_SMR_ASSERT_ENTERED();
3290
3291 fd = ndp->ni_dirfd;
3292 fdp = curproc->p_fd;
3293 fdt = fdp->fd_files;
3294 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3295 return (EBADF);
3296 fde = &fdt->fdt_ofiles[fd];
3297 fp = fde->fde_file;
3298 if (__predict_false(fp == NULL))
3299 return (EAGAIN);
3300 flags = fp->f_flag & FSEARCH;
3301 flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ?
3302 O_RESOLVE_BENEATH : 0;
3303 vp = fp->f_vnode;
3304 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3305 return (EAGAIN);
3306 }
3307 /*
3308 * Use an acquire barrier to force re-reading of fdt so it is
3309 * refreshed for verification.
3310 */
3311 atomic_thread_fence_acq();
3312 fdt = fdp->fd_files;
3313 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3314 return (EAGAIN);
3315 filecaps_fill(&ndp->ni_filecaps);
3316 *vpp = vp;
3317 *flagsp = flags;
3318 return (0);
3319 }
3320 #endif
3321
3322 int
fgetvp_lookup(struct nameidata * ndp,struct vnode ** vpp)3323 fgetvp_lookup(struct nameidata *ndp, struct vnode **vpp)
3324 {
3325 struct thread *td;
3326 struct file *fp;
3327 struct vnode *vp;
3328 struct componentname *cnp;
3329 cap_rights_t rights;
3330 int error;
3331 uint8_t flags;
3332
3333 td = curthread;
3334 rights = *ndp->ni_rightsneeded;
3335 cap_rights_set_one(&rights, CAP_LOOKUP);
3336 cnp = &ndp->ni_cnd;
3337
3338 error = fget_cap(td, ndp->ni_dirfd, &rights, &flags, &fp,
3339 &ndp->ni_filecaps);
3340 if (__predict_false(error != 0))
3341 return (error);
3342 if (__predict_false(fp->f_ops == &badfileops)) {
3343 error = EBADF;
3344 goto out_free;
3345 }
3346 vp = fp->f_vnode;
3347 if (__predict_false(vp == NULL)) {
3348 error = ENOTDIR;
3349 goto out_free;
3350 }
3351 vrefact(vp);
3352 /*
3353 * XXX does not check for VDIR, handled by namei_setup
3354 */
3355 if ((fp->f_flag & FSEARCH) != 0)
3356 cnp->cn_flags |= NOEXECCHECK;
3357 if ((flags & UF_RESOLVE_BENEATH) != 0) {
3358 cnp->cn_flags |= RBENEATH;
3359 ndp->ni_resflags |= NIRES_BENEATH;
3360 }
3361 fdrop(fp, td);
3362
3363 #ifdef CAPABILITIES
3364 /*
3365 * If file descriptor doesn't have all rights,
3366 * all lookups relative to it must also be
3367 * strictly relative.
3368 */
3369 CAP_ALL(&rights);
3370 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3371 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3372 ndp->ni_filecaps.fc_nioctls != -1) {
3373 ndp->ni_lcf |= NI_LCF_STRICTREL;
3374 ndp->ni_resflags |= NIRES_STRICTREL;
3375 }
3376 #endif
3377
3378 /*
3379 * TODO: avoid copying ioctl caps if it can be helped to begin with
3380 */
3381 if ((cnp->cn_flags & WANTIOCTLCAPS) == 0)
3382 filecaps_free_ioctl(&ndp->ni_filecaps);
3383
3384 *vpp = vp;
3385 return (0);
3386
3387 out_free:
3388 filecaps_free(&ndp->ni_filecaps);
3389 fdrop(fp, td);
3390 return (error);
3391 }
3392
3393 /*
3394 * Fetch the descriptor locklessly.
3395 *
3396 * We avoid fdrop() races by never raising a refcount above 0. To accomplish
3397 * this we have to use a cmpset loop rather than an atomic_add. The descriptor
3398 * must be re-verified once we acquire a reference to be certain that the
3399 * identity is still correct and we did not lose a race due to preemption.
3400 *
3401 * Force a reload of fdt when looping. Another thread could reallocate
3402 * the table before this fd was closed, so it is possible that there is
3403 * a stale fp pointer in cached version.
3404 */
3405 #ifdef CAPABILITIES
3406 static int
fget_unlocked_seq(struct thread * td,int fd,const cap_rights_t * needrightsp,uint8_t * flagsp,struct file ** fpp,seqc_t * seqp)3407 fget_unlocked_seq(struct thread *td, int fd, const cap_rights_t *needrightsp,
3408 uint8_t *flagsp, struct file **fpp, seqc_t *seqp)
3409 {
3410 struct filedesc *fdp;
3411 const struct filedescent *fde;
3412 const struct fdescenttbl *fdt;
3413 struct file *fp;
3414 seqc_t seq;
3415 cap_rights_t haverights;
3416 int error;
3417 uint8_t flags;
3418
3419 fdp = td->td_proc->p_fd;
3420 fdt = fdp->fd_files;
3421 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3422 return (EBADF);
3423
3424 for (;;) {
3425 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3426 fde = &fdt->fdt_ofiles[fd];
3427 haverights = *cap_rights_fde_inline(fde);
3428 fp = fde->fde_file;
3429 flags = fde->fde_flags;
3430 if (__predict_false(fp == NULL)) {
3431 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3432 return (EBADF);
3433 fdt = atomic_load_ptr(&fdp->fd_files);
3434 continue;
3435 }
3436 error = cap_check_inline(&haverights, needrightsp);
3437 if (__predict_false(error != 0)) {
3438 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3439 return (error);
3440 fdt = atomic_load_ptr(&fdp->fd_files);
3441 continue;
3442 }
3443 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3444 fdt = atomic_load_ptr(&fdp->fd_files);
3445 continue;
3446 }
3447 /*
3448 * Use an acquire barrier to force re-reading of fdt so it is
3449 * refreshed for verification.
3450 */
3451 atomic_thread_fence_acq();
3452 fdt = fdp->fd_files;
3453 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3454 break;
3455 fdrop(fp, td);
3456 }
3457 *fpp = fp;
3458 if (flagsp != NULL)
3459 *flagsp = flags;
3460 if (seqp != NULL)
3461 *seqp = seq;
3462 return (0);
3463 }
3464 #else
3465 static int
fget_unlocked_seq(struct thread * td,int fd,const cap_rights_t * needrightsp,uint8_t * flagsp,struct file ** fpp,seqc_t * seqp __unused)3466 fget_unlocked_seq(struct thread *td, int fd, const cap_rights_t *needrightsp,
3467 uint8_t *flagsp, struct file **fpp, seqc_t *seqp __unused)
3468 {
3469 struct filedesc *fdp;
3470 const struct fdescenttbl *fdt;
3471 struct file *fp;
3472 uint8_t flags;
3473
3474 fdp = td->td_proc->p_fd;
3475 fdt = fdp->fd_files;
3476 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3477 return (EBADF);
3478
3479 for (;;) {
3480 fp = fdt->fdt_ofiles[fd].fde_file;
3481 flags = fdt->fdt_ofiles[fd].fde_flags;
3482 if (__predict_false(fp == NULL))
3483 return (EBADF);
3484 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3485 fdt = atomic_load_ptr(&fdp->fd_files);
3486 continue;
3487 }
3488 /*
3489 * Use an acquire barrier to force re-reading of fdt so it is
3490 * refreshed for verification.
3491 */
3492 atomic_thread_fence_acq();
3493 fdt = fdp->fd_files;
3494 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3495 break;
3496 fdrop(fp, td);
3497 }
3498 if (flagsp != NULL)
3499 *flagsp = flags;
3500 *fpp = fp;
3501 return (0);
3502 }
3503 #endif
3504
3505 /*
3506 * See the comments in fget_unlocked_seq for an explanation of how this works.
3507 *
3508 * This is a simplified variant which bails out to the aforementioned routine
3509 * if anything goes wrong. In practice this only happens when userspace is
3510 * racing with itself.
3511 */
3512 int
fget_unlocked_flags(struct thread * td,int fd,const cap_rights_t * needrightsp,uint8_t * flagsp,struct file ** fpp)3513 fget_unlocked_flags(struct thread *td, int fd, const cap_rights_t *needrightsp,
3514 uint8_t *flagsp, struct file **fpp)
3515 {
3516 struct filedesc *fdp;
3517 #ifdef CAPABILITIES
3518 const struct filedescent *fde;
3519 #endif
3520 const struct fdescenttbl *fdt;
3521 struct file *fp;
3522 #ifdef CAPABILITIES
3523 seqc_t seq;
3524 const cap_rights_t *haverights;
3525 #endif
3526 uint8_t flags;
3527
3528 fdp = td->td_proc->p_fd;
3529 fdt = fdp->fd_files;
3530 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3531 *fpp = NULL;
3532 return (EBADF);
3533 }
3534 #ifdef CAPABILITIES
3535 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3536 fde = &fdt->fdt_ofiles[fd];
3537 haverights = cap_rights_fde_inline(fde);
3538 fp = fde->fde_file;
3539 flags = fde->fde_flags;
3540 #else
3541 fp = fdt->fdt_ofiles[fd].fde_file;
3542 flags = fdt->fdt_ofiles[fd].fde_flags;
3543 #endif
3544 if (__predict_false(fp == NULL))
3545 goto out_fallback;
3546 #ifdef CAPABILITIES
3547 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3548 goto out_fallback;
3549 #endif
3550 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3551 goto out_fallback;
3552
3553 /*
3554 * Use an acquire barrier to force re-reading of fdt so it is
3555 * refreshed for verification.
3556 */
3557 atomic_thread_fence_acq();
3558 fdt = fdp->fd_files;
3559 #ifdef CAPABILITIES
3560 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3561 #else
3562 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3563 #endif
3564 goto out_fdrop;
3565 *fpp = fp;
3566 if (flagsp != NULL)
3567 *flagsp = flags;
3568 return (0);
3569 out_fdrop:
3570 fdrop(fp, td);
3571 out_fallback:
3572 *fpp = NULL;
3573 return (fget_unlocked_seq(td, fd, needrightsp, flagsp, fpp, NULL));
3574 }
3575
3576 int
fget_unlocked(struct thread * td,int fd,const cap_rights_t * needrightsp,struct file ** fpp)3577 fget_unlocked(struct thread *td, int fd, const cap_rights_t *needrightsp,
3578 struct file **fpp)
3579 {
3580 return (fget_unlocked_flags(td, fd, needrightsp, NULL, fpp));
3581 }
3582
3583 /*
3584 * Translate fd -> file when the caller guarantees the file descriptor table
3585 * can't be changed by others.
3586 *
3587 * Note this does not mean the file object itself is only visible to the caller,
3588 * merely that it wont disappear without having to be referenced.
3589 *
3590 * Must be paired with fput_only_user.
3591 */
3592 #ifdef CAPABILITIES
3593 int
fget_only_user(struct filedesc * fdp,int fd,const cap_rights_t * needrightsp,struct file ** fpp)3594 fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp,
3595 struct file **fpp)
3596 {
3597 const struct filedescent *fde;
3598 const struct fdescenttbl *fdt;
3599 const cap_rights_t *haverights;
3600 struct file *fp;
3601 int error;
3602
3603 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3604
3605 *fpp = NULL;
3606 if (__predict_false(fd >= fdp->fd_nfiles))
3607 return (EBADF);
3608
3609 fdt = fdp->fd_files;
3610 fde = &fdt->fdt_ofiles[fd];
3611 fp = fde->fde_file;
3612 if (__predict_false(fp == NULL))
3613 return (EBADF);
3614 MPASS(refcount_load(&fp->f_count) > 0);
3615 haverights = cap_rights_fde_inline(fde);
3616 error = cap_check_inline(haverights, needrightsp);
3617 if (__predict_false(error != 0))
3618 return (error);
3619 *fpp = fp;
3620 return (0);
3621 }
3622 #else
3623 int
fget_only_user(struct filedesc * fdp,int fd,const cap_rights_t * needrightsp,struct file ** fpp)3624 fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp,
3625 struct file **fpp)
3626 {
3627 struct file *fp;
3628
3629 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3630
3631 *fpp = NULL;
3632 if (__predict_false(fd >= fdp->fd_nfiles))
3633 return (EBADF);
3634
3635 fp = fdp->fd_ofiles[fd].fde_file;
3636 if (__predict_false(fp == NULL))
3637 return (EBADF);
3638
3639 MPASS(refcount_load(&fp->f_count) > 0);
3640 *fpp = fp;
3641 return (0);
3642 }
3643 #endif
3644
3645 /*
3646 * Extract the file pointer associated with the specified descriptor for the
3647 * current user process.
3648 *
3649 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3650 * returned.
3651 *
3652 * File's rights will be checked against the capability rights mask.
3653 *
3654 * If an error occurred the non-zero error is returned and *fpp is set to
3655 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3656 * responsible for fdrop().
3657 */
3658 static __inline int
_fget(struct thread * td,int fd,struct file ** fpp,int flags,const cap_rights_t * needrightsp)3659 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3660 const cap_rights_t *needrightsp)
3661 {
3662 struct file *fp;
3663 int error;
3664
3665 *fpp = NULL;
3666 error = fget_unlocked(td, fd, needrightsp, &fp);
3667 if (__predict_false(error != 0))
3668 return (error);
3669 if (__predict_false(fp->f_ops == &badfileops)) {
3670 fdrop(fp, td);
3671 return (EBADF);
3672 }
3673
3674 /*
3675 * FREAD and FWRITE failure return EBADF as per POSIX.
3676 */
3677 error = 0;
3678 switch (flags) {
3679 case FREAD:
3680 case FWRITE:
3681 if ((fp->f_flag & flags) == 0)
3682 error = EBADF;
3683 break;
3684 case FEXEC:
3685 if (fp->f_ops != &path_fileops &&
3686 ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3687 (fp->f_flag & FWRITE) != 0))
3688 error = EBADF;
3689 break;
3690 case 0:
3691 break;
3692 default:
3693 KASSERT(0, ("wrong flags"));
3694 }
3695
3696 if (error != 0) {
3697 fdrop(fp, td);
3698 return (error);
3699 }
3700
3701 *fpp = fp;
3702 return (0);
3703 }
3704
3705 int
fget(struct thread * td,int fd,const cap_rights_t * rightsp,struct file ** fpp)3706 fget(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp)
3707 {
3708
3709 return (_fget(td, fd, fpp, 0, rightsp));
3710 }
3711
3712 int
fget_mmap(struct thread * td,int fd,const cap_rights_t * rightsp,vm_prot_t * maxprotp,struct file ** fpp)3713 fget_mmap(struct thread *td, int fd, const cap_rights_t *rightsp,
3714 vm_prot_t *maxprotp, struct file **fpp)
3715 {
3716 int error;
3717 #ifndef CAPABILITIES
3718 error = _fget(td, fd, fpp, 0, rightsp);
3719 if (maxprotp != NULL)
3720 *maxprotp = VM_PROT_ALL;
3721 return (error);
3722 #else
3723 cap_rights_t fdrights;
3724 struct filedesc *fdp;
3725 struct file *fp;
3726 seqc_t seq;
3727
3728 *fpp = NULL;
3729 fdp = td->td_proc->p_fd;
3730 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3731 for (;;) {
3732 error = fget_unlocked_seq(td, fd, rightsp, NULL, &fp, &seq);
3733 if (__predict_false(error != 0))
3734 return (error);
3735 if (__predict_false(fp->f_ops == &badfileops)) {
3736 fdrop(fp, td);
3737 return (EBADF);
3738 }
3739 if (maxprotp != NULL)
3740 fdrights = *cap_rights(fdp, fd);
3741 if (!fd_modified(fdp, fd, seq))
3742 break;
3743 fdrop(fp, td);
3744 }
3745
3746 /*
3747 * If requested, convert capability rights to access flags.
3748 */
3749 if (maxprotp != NULL)
3750 *maxprotp = cap_rights_to_vmprot(&fdrights);
3751 *fpp = fp;
3752 return (0);
3753 #endif
3754 }
3755
3756 int
fget_read(struct thread * td,int fd,const cap_rights_t * rightsp,struct file ** fpp)3757 fget_read(struct thread *td, int fd, const cap_rights_t *rightsp,
3758 struct file **fpp)
3759 {
3760
3761 return (_fget(td, fd, fpp, FREAD, rightsp));
3762 }
3763
3764 int
fget_write(struct thread * td,int fd,const cap_rights_t * rightsp,struct file ** fpp)3765 fget_write(struct thread *td, int fd, const cap_rights_t *rightsp,
3766 struct file **fpp)
3767 {
3768
3769 return (_fget(td, fd, fpp, FWRITE, rightsp));
3770 }
3771
3772 int
fget_fcntl(struct thread * td,int fd,const cap_rights_t * rightsp,int needfcntl,struct file ** fpp)3773 fget_fcntl(struct thread *td, int fd, const cap_rights_t *rightsp,
3774 int needfcntl, struct file **fpp)
3775 {
3776 #ifndef CAPABILITIES
3777 return (fget_unlocked(td, fd, rightsp, fpp));
3778 #else
3779 struct filedesc *fdp = td->td_proc->p_fd;
3780 struct file *fp;
3781 int error;
3782 seqc_t seq;
3783
3784 *fpp = NULL;
3785 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3786 for (;;) {
3787 error = fget_unlocked_seq(td, fd, rightsp, NULL, &fp, &seq);
3788 if (error != 0)
3789 return (error);
3790 error = cap_fcntl_check(fdp, fd, needfcntl);
3791 if (!fd_modified(fdp, fd, seq))
3792 break;
3793 fdrop(fp, td);
3794 }
3795 if (error != 0) {
3796 fdrop(fp, td);
3797 return (error);
3798 }
3799 *fpp = fp;
3800 return (0);
3801 #endif
3802 }
3803
3804 /*
3805 * Like fget() but loads the underlying vnode, or returns an error if the
3806 * descriptor does not represent a vnode. Note that pipes use vnodes but
3807 * never have VM objects. The returned vnode will be vref()'d.
3808 *
3809 * XXX: what about the unused flags ?
3810 */
3811 static __inline int
_fgetvp(struct thread * td,int fd,int flags,const cap_rights_t * needrightsp,struct vnode ** vpp)3812 _fgetvp(struct thread *td, int fd, int flags, const cap_rights_t *needrightsp,
3813 struct vnode **vpp)
3814 {
3815 struct file *fp;
3816 int error;
3817
3818 *vpp = NULL;
3819 error = _fget(td, fd, &fp, flags, needrightsp);
3820 if (error != 0)
3821 return (error);
3822 if (fp->f_vnode == NULL) {
3823 error = EINVAL;
3824 } else {
3825 *vpp = fp->f_vnode;
3826 vrefact(*vpp);
3827 }
3828 fdrop(fp, td);
3829
3830 return (error);
3831 }
3832
3833 int
fgetvp(struct thread * td,int fd,const cap_rights_t * rightsp,struct vnode ** vpp)3834 fgetvp(struct thread *td, int fd, const cap_rights_t *rightsp,
3835 struct vnode **vpp)
3836 {
3837
3838 return (_fgetvp(td, fd, 0, rightsp, vpp));
3839 }
3840
3841 int
fgetvp_rights(struct thread * td,int fd,const cap_rights_t * needrightsp,struct filecaps * havecaps,struct vnode ** vpp)3842 fgetvp_rights(struct thread *td, int fd, const cap_rights_t *needrightsp,
3843 struct filecaps *havecaps, struct vnode **vpp)
3844 {
3845 struct filecaps caps;
3846 struct file *fp;
3847 int error;
3848
3849 error = fget_cap(td, fd, needrightsp, NULL, &fp, &caps);
3850 if (error != 0)
3851 return (error);
3852 if (fp->f_ops == &badfileops) {
3853 error = EBADF;
3854 goto out;
3855 }
3856 if (fp->f_vnode == NULL) {
3857 error = EINVAL;
3858 goto out;
3859 }
3860
3861 *havecaps = caps;
3862 *vpp = fp->f_vnode;
3863 vrefact(*vpp);
3864 fdrop(fp, td);
3865
3866 return (0);
3867 out:
3868 filecaps_free(&caps);
3869 fdrop(fp, td);
3870 return (error);
3871 }
3872
3873 int
fgetvp_read(struct thread * td,int fd,const cap_rights_t * rightsp,struct vnode ** vpp)3874 fgetvp_read(struct thread *td, int fd, const cap_rights_t *rightsp,
3875 struct vnode **vpp)
3876 {
3877
3878 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3879 }
3880
3881 int
fgetvp_exec(struct thread * td,int fd,const cap_rights_t * rightsp,struct vnode ** vpp)3882 fgetvp_exec(struct thread *td, int fd, const cap_rights_t *rightsp,
3883 struct vnode **vpp)
3884 {
3885
3886 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3887 }
3888
3889 #ifdef notyet
3890 int
fgetvp_write(struct thread * td,int fd,const cap_rights_t * rightsp,struct vnode ** vpp)3891 fgetvp_write(struct thread *td, int fd, const cap_rights_t *rightsp,
3892 struct vnode **vpp)
3893 {
3894
3895 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3896 }
3897 #endif
3898
3899 /*
3900 * Handle the last reference to a file being closed.
3901 *
3902 * Without the noinline attribute clang keeps inlining the func thorough this
3903 * file when fdrop is used.
3904 */
3905 int __noinline
_fdrop(struct file * fp,struct thread * td)3906 _fdrop(struct file *fp, struct thread *td)
3907 {
3908 int error;
3909
3910 KASSERT(refcount_load(&fp->f_count) == 0,
3911 ("fdrop: fp %p count %d", fp, refcount_load(&fp->f_count)));
3912
3913 error = fo_close(fp, td);
3914 atomic_subtract_int(&openfiles, 1);
3915 crfree(fp->f_cred);
3916 free(fp->f_advice, M_FADVISE);
3917 uma_zfree(file_zone, fp);
3918
3919 return (error);
3920 }
3921
3922 /*
3923 * Apply an advisory lock on a file descriptor.
3924 *
3925 * Just attempt to get a record lock of the requested type on the entire file
3926 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3927 */
3928 #ifndef _SYS_SYSPROTO_H_
3929 struct flock_args {
3930 int fd;
3931 int how;
3932 };
3933 #endif
3934 /* ARGSUSED */
3935 int
sys_flock(struct thread * td,struct flock_args * uap)3936 sys_flock(struct thread *td, struct flock_args *uap)
3937 {
3938 struct file *fp;
3939 struct vnode *vp;
3940 struct flock lf;
3941 int error;
3942
3943 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3944 if (error != 0)
3945 return (error);
3946 error = EOPNOTSUPP;
3947 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3948 goto done;
3949 }
3950 if (fp->f_ops == &path_fileops) {
3951 goto done;
3952 }
3953
3954 error = 0;
3955 vp = fp->f_vnode;
3956 lf.l_whence = SEEK_SET;
3957 lf.l_start = 0;
3958 lf.l_len = 0;
3959 if (uap->how & LOCK_UN) {
3960 lf.l_type = F_UNLCK;
3961 atomic_clear_int(&fp->f_flag, FHASLOCK);
3962 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3963 goto done;
3964 }
3965 if (uap->how & LOCK_EX)
3966 lf.l_type = F_WRLCK;
3967 else if (uap->how & LOCK_SH)
3968 lf.l_type = F_RDLCK;
3969 else {
3970 error = EBADF;
3971 goto done;
3972 }
3973 atomic_set_int(&fp->f_flag, FHASLOCK);
3974 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3975 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3976 done:
3977 fdrop(fp, td);
3978 return (error);
3979 }
3980 /*
3981 * Duplicate the specified descriptor to a free descriptor.
3982 */
3983 int
dupfdopen(struct thread * td,struct filedesc * fdp,int dfd,int mode,int openerror,int * indxp)3984 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3985 int openerror, int *indxp)
3986 {
3987 struct filedescent *newfde, *oldfde;
3988 struct file *fp;
3989 u_long *ioctls;
3990 int error, indx;
3991
3992 KASSERT(openerror == ENODEV || openerror == ENXIO,
3993 ("unexpected error %d in %s", openerror, __func__));
3994
3995 /*
3996 * If the to-be-dup'd fd number is greater than the allowed number
3997 * of file descriptors, or the fd to be dup'd has already been
3998 * closed, then reject.
3999 */
4000 FILEDESC_XLOCK(fdp);
4001 if ((fp = fget_noref(fdp, dfd)) == NULL) {
4002 FILEDESC_XUNLOCK(fdp);
4003 return (EBADF);
4004 }
4005
4006 error = fdalloc(td, 0, &indx);
4007 if (error != 0) {
4008 FILEDESC_XUNLOCK(fdp);
4009 return (error);
4010 }
4011
4012 /*
4013 * There are two cases of interest here.
4014 *
4015 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
4016 *
4017 * For ENXIO steal away the file structure from (dfd) and store it in
4018 * (indx). (dfd) is effectively closed by this operation.
4019 */
4020 switch (openerror) {
4021 case ENODEV:
4022 /*
4023 * Check that the mode the file is being opened for is a
4024 * subset of the mode of the existing descriptor.
4025 */
4026 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
4027 fdunused(fdp, indx);
4028 FILEDESC_XUNLOCK(fdp);
4029 return (EACCES);
4030 }
4031 if (!fhold(fp)) {
4032 fdunused(fdp, indx);
4033 FILEDESC_XUNLOCK(fdp);
4034 return (EBADF);
4035 }
4036 newfde = &fdp->fd_ofiles[indx];
4037 oldfde = &fdp->fd_ofiles[dfd];
4038 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
4039 #ifdef CAPABILITIES
4040 seqc_write_begin(&newfde->fde_seqc);
4041 #endif
4042 fde_copy(oldfde, newfde);
4043 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
4044 ioctls);
4045 #ifdef CAPABILITIES
4046 seqc_write_end(&newfde->fde_seqc);
4047 #endif
4048 break;
4049 case ENXIO:
4050 /*
4051 * Steal away the file pointer from dfd and stuff it into indx.
4052 */
4053 newfde = &fdp->fd_ofiles[indx];
4054 oldfde = &fdp->fd_ofiles[dfd];
4055 #ifdef CAPABILITIES
4056 seqc_write_begin(&oldfde->fde_seqc);
4057 seqc_write_begin(&newfde->fde_seqc);
4058 #endif
4059 fde_copy(oldfde, newfde);
4060 oldfde->fde_file = NULL;
4061 fdunused(fdp, dfd);
4062 #ifdef CAPABILITIES
4063 seqc_write_end(&newfde->fde_seqc);
4064 seqc_write_end(&oldfde->fde_seqc);
4065 #endif
4066 break;
4067 }
4068 FILEDESC_XUNLOCK(fdp);
4069 *indxp = indx;
4070 return (0);
4071 }
4072
4073 /*
4074 * This sysctl determines if we will allow a process to chroot(2) if it
4075 * has a directory open:
4076 * 0: disallowed for all processes.
4077 * 1: allowed for processes that were not already chroot(2)'ed.
4078 * 2: allowed for all processes.
4079 */
4080
4081 static int chroot_allow_open_directories = 1;
4082
4083 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
4084 &chroot_allow_open_directories, 0,
4085 "Allow a process to chroot(2) if it has a directory open");
4086
4087 /*
4088 * Helper function for raised chroot(2) security function: Refuse if
4089 * any filedescriptors are open directories.
4090 */
4091 static int
chroot_refuse_vdir_fds(struct filedesc * fdp)4092 chroot_refuse_vdir_fds(struct filedesc *fdp)
4093 {
4094 struct vnode *vp;
4095 struct file *fp;
4096 int i;
4097
4098 FILEDESC_LOCK_ASSERT(fdp);
4099
4100 FILEDESC_FOREACH_FP(fdp, i, fp) {
4101 if (fp->f_type == DTYPE_VNODE) {
4102 vp = fp->f_vnode;
4103 if (vp->v_type == VDIR)
4104 return (EPERM);
4105 }
4106 }
4107 return (0);
4108 }
4109
4110 static void
pwd_fill(struct pwd * oldpwd,struct pwd * newpwd)4111 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
4112 {
4113
4114 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
4115 vrefact(oldpwd->pwd_cdir);
4116 newpwd->pwd_cdir = oldpwd->pwd_cdir;
4117 }
4118
4119 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
4120 vrefact(oldpwd->pwd_rdir);
4121 newpwd->pwd_rdir = oldpwd->pwd_rdir;
4122 }
4123
4124 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
4125 vrefact(oldpwd->pwd_jdir);
4126 newpwd->pwd_jdir = oldpwd->pwd_jdir;
4127 }
4128
4129 if (newpwd->pwd_adir == NULL && oldpwd->pwd_adir != NULL) {
4130 vrefact(oldpwd->pwd_adir);
4131 newpwd->pwd_adir = oldpwd->pwd_adir;
4132 }
4133 }
4134
4135 struct pwd *
pwd_hold_pwddesc(struct pwddesc * pdp)4136 pwd_hold_pwddesc(struct pwddesc *pdp)
4137 {
4138 struct pwd *pwd;
4139
4140 PWDDESC_ASSERT_XLOCKED(pdp);
4141 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4142 if (pwd != NULL)
4143 refcount_acquire(&pwd->pwd_refcount);
4144 return (pwd);
4145 }
4146
4147 bool
pwd_hold_smr(struct pwd * pwd)4148 pwd_hold_smr(struct pwd *pwd)
4149 {
4150
4151 MPASS(pwd != NULL);
4152 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
4153 return (true);
4154 }
4155 return (false);
4156 }
4157
4158 struct pwd *
pwd_hold(struct thread * td)4159 pwd_hold(struct thread *td)
4160 {
4161 struct pwddesc *pdp;
4162 struct pwd *pwd;
4163
4164 pdp = td->td_proc->p_pd;
4165
4166 vfs_smr_enter();
4167 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
4168 if (pwd_hold_smr(pwd)) {
4169 vfs_smr_exit();
4170 return (pwd);
4171 }
4172 vfs_smr_exit();
4173 PWDDESC_XLOCK(pdp);
4174 pwd = pwd_hold_pwddesc(pdp);
4175 MPASS(pwd != NULL);
4176 PWDDESC_XUNLOCK(pdp);
4177 return (pwd);
4178 }
4179
4180 struct pwd *
pwd_hold_proc(struct proc * p)4181 pwd_hold_proc(struct proc *p)
4182 {
4183 struct pwddesc *pdp;
4184 struct pwd *pwd;
4185
4186 PROC_ASSERT_HELD(p);
4187 PROC_LOCK(p);
4188 pdp = pdhold(p);
4189 MPASS(pdp != NULL);
4190 PROC_UNLOCK(p);
4191
4192 PWDDESC_XLOCK(pdp);
4193 pwd = pwd_hold_pwddesc(pdp);
4194 MPASS(pwd != NULL);
4195 PWDDESC_XUNLOCK(pdp);
4196 pddrop(pdp);
4197 return (pwd);
4198 }
4199
4200 static struct pwd *
pwd_alloc(void)4201 pwd_alloc(void)
4202 {
4203 struct pwd *pwd;
4204
4205 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
4206 bzero(pwd, sizeof(*pwd));
4207 refcount_init(&pwd->pwd_refcount, 1);
4208 return (pwd);
4209 }
4210
4211 void
pwd_drop(struct pwd * pwd)4212 pwd_drop(struct pwd *pwd)
4213 {
4214
4215 if (!refcount_release(&pwd->pwd_refcount))
4216 return;
4217
4218 if (pwd->pwd_cdir != NULL)
4219 vrele(pwd->pwd_cdir);
4220 if (pwd->pwd_rdir != NULL)
4221 vrele(pwd->pwd_rdir);
4222 if (pwd->pwd_jdir != NULL)
4223 vrele(pwd->pwd_jdir);
4224 if (pwd->pwd_adir != NULL)
4225 vrele(pwd->pwd_adir);
4226 uma_zfree_smr(pwd_zone, pwd);
4227 }
4228
4229 /*
4230 * The caller is responsible for invoking priv_check() and
4231 * mac_vnode_check_chroot() to authorize this operation.
4232 */
4233 int
pwd_chroot(struct thread * td,struct vnode * vp)4234 pwd_chroot(struct thread *td, struct vnode *vp)
4235 {
4236 struct pwddesc *pdp;
4237 struct filedesc *fdp;
4238 struct pwd *newpwd, *oldpwd;
4239 int error;
4240
4241 fdp = td->td_proc->p_fd;
4242 pdp = td->td_proc->p_pd;
4243 newpwd = pwd_alloc();
4244 FILEDESC_SLOCK(fdp);
4245 PWDDESC_XLOCK(pdp);
4246 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4247 if (chroot_allow_open_directories == 0 ||
4248 (chroot_allow_open_directories == 1 &&
4249 oldpwd->pwd_rdir != rootvnode)) {
4250 error = chroot_refuse_vdir_fds(fdp);
4251 FILEDESC_SUNLOCK(fdp);
4252 if (error != 0) {
4253 PWDDESC_XUNLOCK(pdp);
4254 pwd_drop(newpwd);
4255 return (error);
4256 }
4257 } else {
4258 FILEDESC_SUNLOCK(fdp);
4259 }
4260
4261 vrefact(vp);
4262 newpwd->pwd_rdir = vp;
4263 vrefact(vp);
4264 newpwd->pwd_adir = vp;
4265 if (oldpwd->pwd_jdir == NULL) {
4266 vrefact(vp);
4267 newpwd->pwd_jdir = vp;
4268 }
4269 pwd_fill(oldpwd, newpwd);
4270 pwd_set(pdp, newpwd);
4271 PWDDESC_XUNLOCK(pdp);
4272 pwd_drop(oldpwd);
4273 return (0);
4274 }
4275
4276 void
pwd_chdir(struct thread * td,struct vnode * vp)4277 pwd_chdir(struct thread *td, struct vnode *vp)
4278 {
4279 struct pwddesc *pdp;
4280 struct pwd *newpwd, *oldpwd;
4281
4282 VNPASS(vp->v_usecount > 0, vp);
4283
4284 newpwd = pwd_alloc();
4285 pdp = td->td_proc->p_pd;
4286 PWDDESC_XLOCK(pdp);
4287 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4288 newpwd->pwd_cdir = vp;
4289 pwd_fill(oldpwd, newpwd);
4290 pwd_set(pdp, newpwd);
4291 PWDDESC_XUNLOCK(pdp);
4292 pwd_drop(oldpwd);
4293 }
4294
4295 /*
4296 * Process is transitioning to/from a non-native ABI.
4297 */
4298 void
pwd_altroot(struct thread * td,struct vnode * altroot_vp)4299 pwd_altroot(struct thread *td, struct vnode *altroot_vp)
4300 {
4301 struct pwddesc *pdp;
4302 struct pwd *newpwd, *oldpwd;
4303
4304 newpwd = pwd_alloc();
4305 pdp = td->td_proc->p_pd;
4306 PWDDESC_XLOCK(pdp);
4307 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4308 if (altroot_vp != NULL) {
4309 /*
4310 * Native process to a non-native ABI.
4311 */
4312
4313 vrefact(altroot_vp);
4314 newpwd->pwd_adir = altroot_vp;
4315 } else {
4316 /*
4317 * Non-native process to the native ABI.
4318 */
4319
4320 vrefact(oldpwd->pwd_rdir);
4321 newpwd->pwd_adir = oldpwd->pwd_rdir;
4322 }
4323 pwd_fill(oldpwd, newpwd);
4324 pwd_set(pdp, newpwd);
4325 PWDDESC_XUNLOCK(pdp);
4326 pwd_drop(oldpwd);
4327 }
4328
4329 /*
4330 * jail_attach(2) changes both root and working directories.
4331 */
4332 int
pwd_chroot_chdir(struct thread * td,struct vnode * vp)4333 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
4334 {
4335 struct pwddesc *pdp;
4336 struct filedesc *fdp;
4337 struct pwd *newpwd, *oldpwd;
4338 int error;
4339
4340 fdp = td->td_proc->p_fd;
4341 pdp = td->td_proc->p_pd;
4342 newpwd = pwd_alloc();
4343 FILEDESC_SLOCK(fdp);
4344 PWDDESC_XLOCK(pdp);
4345 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4346 error = chroot_refuse_vdir_fds(fdp);
4347 FILEDESC_SUNLOCK(fdp);
4348 if (error != 0) {
4349 PWDDESC_XUNLOCK(pdp);
4350 pwd_drop(newpwd);
4351 return (error);
4352 }
4353
4354 vrefact(vp);
4355 newpwd->pwd_rdir = vp;
4356 vrefact(vp);
4357 newpwd->pwd_cdir = vp;
4358 if (oldpwd->pwd_jdir == NULL) {
4359 vrefact(vp);
4360 newpwd->pwd_jdir = vp;
4361 }
4362 vrefact(vp);
4363 newpwd->pwd_adir = vp;
4364 pwd_fill(oldpwd, newpwd);
4365 pwd_set(pdp, newpwd);
4366 PWDDESC_XUNLOCK(pdp);
4367 pwd_drop(oldpwd);
4368 return (0);
4369 }
4370
4371 void
pwd_ensure_dirs(void)4372 pwd_ensure_dirs(void)
4373 {
4374 struct pwddesc *pdp;
4375 struct pwd *oldpwd, *newpwd;
4376
4377 pdp = curproc->p_pd;
4378 PWDDESC_XLOCK(pdp);
4379 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4380 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL &&
4381 oldpwd->pwd_adir != NULL) {
4382 PWDDESC_XUNLOCK(pdp);
4383 return;
4384 }
4385 PWDDESC_XUNLOCK(pdp);
4386
4387 newpwd = pwd_alloc();
4388 PWDDESC_XLOCK(pdp);
4389 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4390 pwd_fill(oldpwd, newpwd);
4391 if (newpwd->pwd_cdir == NULL) {
4392 vrefact(rootvnode);
4393 newpwd->pwd_cdir = rootvnode;
4394 }
4395 if (newpwd->pwd_rdir == NULL) {
4396 vrefact(rootvnode);
4397 newpwd->pwd_rdir = rootvnode;
4398 }
4399 if (newpwd->pwd_adir == NULL) {
4400 vrefact(rootvnode);
4401 newpwd->pwd_adir = rootvnode;
4402 }
4403 pwd_set(pdp, newpwd);
4404 PWDDESC_XUNLOCK(pdp);
4405 pwd_drop(oldpwd);
4406 }
4407
4408 void
pwd_set_rootvnode(void)4409 pwd_set_rootvnode(void)
4410 {
4411 struct pwddesc *pdp;
4412 struct pwd *oldpwd, *newpwd;
4413
4414 pdp = curproc->p_pd;
4415
4416 newpwd = pwd_alloc();
4417 PWDDESC_XLOCK(pdp);
4418 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4419 vrefact(rootvnode);
4420 newpwd->pwd_cdir = rootvnode;
4421 vrefact(rootvnode);
4422 newpwd->pwd_rdir = rootvnode;
4423 vrefact(rootvnode);
4424 newpwd->pwd_adir = rootvnode;
4425 pwd_fill(oldpwd, newpwd);
4426 pwd_set(pdp, newpwd);
4427 PWDDESC_XUNLOCK(pdp);
4428 pwd_drop(oldpwd);
4429 }
4430
4431 /*
4432 * Scan all active processes and prisons to see if any of them have a current
4433 * or root directory of `olddp'. If so, replace them with the new mount point.
4434 */
4435 void
mountcheckdirs(struct vnode * olddp,struct vnode * newdp)4436 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
4437 {
4438 struct pwddesc *pdp;
4439 struct pwd *newpwd, *oldpwd;
4440 struct prison *pr;
4441 struct proc *p;
4442 int nrele;
4443
4444 if (vrefcnt(olddp) == 1)
4445 return;
4446 nrele = 0;
4447 newpwd = pwd_alloc();
4448 sx_slock(&allproc_lock);
4449 FOREACH_PROC_IN_SYSTEM(p) {
4450 PROC_LOCK(p);
4451 pdp = pdhold(p);
4452 PROC_UNLOCK(p);
4453 if (pdp == NULL)
4454 continue;
4455 PWDDESC_XLOCK(pdp);
4456 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4457 if (oldpwd == NULL ||
4458 (oldpwd->pwd_cdir != olddp &&
4459 oldpwd->pwd_rdir != olddp &&
4460 oldpwd->pwd_jdir != olddp &&
4461 oldpwd->pwd_adir != olddp)) {
4462 PWDDESC_XUNLOCK(pdp);
4463 pddrop(pdp);
4464 continue;
4465 }
4466 if (oldpwd->pwd_cdir == olddp) {
4467 vrefact(newdp);
4468 newpwd->pwd_cdir = newdp;
4469 }
4470 if (oldpwd->pwd_rdir == olddp) {
4471 vrefact(newdp);
4472 newpwd->pwd_rdir = newdp;
4473 }
4474 if (oldpwd->pwd_jdir == olddp) {
4475 vrefact(newdp);
4476 newpwd->pwd_jdir = newdp;
4477 }
4478 if (oldpwd->pwd_adir == olddp) {
4479 vrefact(newdp);
4480 newpwd->pwd_adir = newdp;
4481 }
4482 pwd_fill(oldpwd, newpwd);
4483 pwd_set(pdp, newpwd);
4484 PWDDESC_XUNLOCK(pdp);
4485 pwd_drop(oldpwd);
4486 pddrop(pdp);
4487 newpwd = pwd_alloc();
4488 }
4489 sx_sunlock(&allproc_lock);
4490 pwd_drop(newpwd);
4491 if (rootvnode == olddp) {
4492 vrefact(newdp);
4493 rootvnode = newdp;
4494 nrele++;
4495 }
4496 mtx_lock(&prison0.pr_mtx);
4497 if (prison0.pr_root == olddp) {
4498 vrefact(newdp);
4499 prison0.pr_root = newdp;
4500 nrele++;
4501 }
4502 mtx_unlock(&prison0.pr_mtx);
4503 sx_slock(&allprison_lock);
4504 TAILQ_FOREACH(pr, &allprison, pr_list) {
4505 mtx_lock(&pr->pr_mtx);
4506 if (pr->pr_root == olddp) {
4507 vrefact(newdp);
4508 pr->pr_root = newdp;
4509 nrele++;
4510 }
4511 mtx_unlock(&pr->pr_mtx);
4512 }
4513 sx_sunlock(&allprison_lock);
4514 while (nrele--)
4515 vrele(olddp);
4516 }
4517
4518 int
descrip_check_write_mp(struct filedesc * fdp,struct mount * mp)4519 descrip_check_write_mp(struct filedesc *fdp, struct mount *mp)
4520 {
4521 struct file *fp;
4522 struct vnode *vp;
4523 int error, i;
4524
4525 error = 0;
4526 FILEDESC_SLOCK(fdp);
4527 FILEDESC_FOREACH_FP(fdp, i, fp) {
4528 if (fp->f_type != DTYPE_VNODE ||
4529 (atomic_load_int(&fp->f_flag) & FWRITE) == 0)
4530 continue;
4531 vp = fp->f_vnode;
4532 if (vp->v_mount == mp) {
4533 error = EDEADLK;
4534 break;
4535 }
4536 }
4537 FILEDESC_SUNLOCK(fdp);
4538 return (error);
4539 }
4540
4541 struct filedesc_to_leader *
filedesc_to_leader_alloc(struct filedesc_to_leader * old,struct filedesc * fdp,struct proc * leader)4542 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp,
4543 struct proc *leader)
4544 {
4545 struct filedesc_to_leader *fdtol;
4546
4547 fdtol = malloc(sizeof(struct filedesc_to_leader),
4548 M_FILEDESC_TO_LEADER, M_WAITOK);
4549 fdtol->fdl_refcount = 1;
4550 fdtol->fdl_holdcount = 0;
4551 fdtol->fdl_wakeup = 0;
4552 fdtol->fdl_leader = leader;
4553 if (old != NULL) {
4554 FILEDESC_XLOCK(fdp);
4555 fdtol->fdl_next = old->fdl_next;
4556 fdtol->fdl_prev = old;
4557 old->fdl_next = fdtol;
4558 fdtol->fdl_next->fdl_prev = fdtol;
4559 FILEDESC_XUNLOCK(fdp);
4560 } else {
4561 fdtol->fdl_next = fdtol;
4562 fdtol->fdl_prev = fdtol;
4563 }
4564 return (fdtol);
4565 }
4566
4567 struct filedesc_to_leader *
filedesc_to_leader_share(struct filedesc_to_leader * fdtol,struct filedesc * fdp)4568 filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp)
4569 {
4570 FILEDESC_XLOCK(fdp);
4571 fdtol->fdl_refcount++;
4572 FILEDESC_XUNLOCK(fdp);
4573 return (fdtol);
4574 }
4575
4576 static int
filedesc_nfiles(struct filedesc * fdp)4577 filedesc_nfiles(struct filedesc *fdp)
4578 {
4579 NDSLOTTYPE *map;
4580 int count, off, minoff;
4581
4582 if (fdp == NULL)
4583 return (0);
4584 count = 0;
4585 FILEDESC_SLOCK(fdp);
4586 map = fdp->fd_map;
4587 off = NDSLOT(fdp->fd_nfiles - 1);
4588 for (minoff = NDSLOT(0); off >= minoff; --off)
4589 count += bitcountl(map[off]);
4590 FILEDESC_SUNLOCK(fdp);
4591 return (count);
4592 }
4593
4594 int
proc_nfiles(struct proc * p)4595 proc_nfiles(struct proc *p)
4596 {
4597 struct filedesc *fdp;
4598 int res;
4599
4600 PROC_LOCK(p);
4601 fdp = fdhold(p);
4602 PROC_UNLOCK(p);
4603 res = filedesc_nfiles(fdp);
4604 fddrop(fdp);
4605 return (res);
4606 }
4607
4608 static int
sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)4609 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4610 {
4611 u_int namelen;
4612 int count;
4613
4614 namelen = arg2;
4615 if (namelen != 1)
4616 return (EINVAL);
4617
4618 if (*(int *)arg1 != 0)
4619 return (EINVAL);
4620
4621 count = filedesc_nfiles(curproc->p_fd);
4622 return (SYSCTL_OUT(req, &count, sizeof(count)));
4623 }
4624
4625 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4626 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4627 "Number of open file descriptors");
4628
4629 /*
4630 * Get file structures globally.
4631 */
4632 static int
sysctl_kern_file(SYSCTL_HANDLER_ARGS)4633 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4634 {
4635 struct xfile xf;
4636 struct filedesc *fdp;
4637 struct file *fp;
4638 struct proc *p;
4639 int error, n;
4640
4641 error = sysctl_wire_old_buffer(req, 0);
4642 if (error != 0)
4643 return (error);
4644 if (req->oldptr == NULL) {
4645 n = 0;
4646 sx_slock(&allproc_lock);
4647 FOREACH_PROC_IN_SYSTEM(p) {
4648 PROC_LOCK(p);
4649 if (p->p_state == PRS_NEW) {
4650 PROC_UNLOCK(p);
4651 continue;
4652 }
4653 fdp = fdhold(p);
4654 PROC_UNLOCK(p);
4655 if (fdp == NULL)
4656 continue;
4657 /* overestimates sparse tables. */
4658 n += fdp->fd_nfiles;
4659 fddrop(fdp);
4660 }
4661 sx_sunlock(&allproc_lock);
4662 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4663 }
4664 error = 0;
4665 bzero(&xf, sizeof(xf));
4666 xf.xf_size = sizeof(xf);
4667 sx_slock(&allproc_lock);
4668 FOREACH_PROC_IN_SYSTEM(p) {
4669 PROC_LOCK(p);
4670 if (p->p_state == PRS_NEW) {
4671 PROC_UNLOCK(p);
4672 continue;
4673 }
4674 if (p_cansee(req->td, p) != 0) {
4675 PROC_UNLOCK(p);
4676 continue;
4677 }
4678 xf.xf_pid = p->p_pid;
4679 xf.xf_uid = p->p_ucred->cr_uid;
4680 fdp = fdhold(p);
4681 PROC_UNLOCK(p);
4682 if (fdp == NULL)
4683 continue;
4684 FILEDESC_SLOCK(fdp);
4685 if (refcount_load(&fdp->fd_refcnt) == 0)
4686 goto nextproc;
4687 FILEDESC_FOREACH_FP(fdp, n, fp) {
4688 xf.xf_fd = n;
4689 xf.xf_file = (uintptr_t)fp;
4690 xf.xf_data = (uintptr_t)fp->f_data;
4691 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4692 xf.xf_type = (uintptr_t)fp->f_type;
4693 xf.xf_count = refcount_load(&fp->f_count);
4694 xf.xf_msgcount = 0;
4695 xf.xf_offset = foffset_get(fp);
4696 xf.xf_flag = fp->f_flag;
4697 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4698
4699 /*
4700 * There is no need to re-check the fdtable refcount
4701 * here since the filedesc lock is not dropped in the
4702 * loop body.
4703 */
4704 if (error != 0)
4705 break;
4706 }
4707 nextproc:
4708 FILEDESC_SUNLOCK(fdp);
4709 fddrop(fdp);
4710 if (error)
4711 break;
4712 }
4713 sx_sunlock(&allproc_lock);
4714 return (error);
4715 }
4716
4717 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4718 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4719
4720 #ifdef KINFO_FILE_SIZE
4721 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4722 #endif
4723
4724 static int
xlate_fflags(int fflags)4725 xlate_fflags(int fflags)
4726 {
4727 static const struct {
4728 int fflag;
4729 int kf_fflag;
4730 } fflags_table[] = {
4731 { FAPPEND, KF_FLAG_APPEND },
4732 { FASYNC, KF_FLAG_ASYNC },
4733 { FFSYNC, KF_FLAG_FSYNC },
4734 { FHASLOCK, KF_FLAG_HASLOCK },
4735 { FNONBLOCK, KF_FLAG_NONBLOCK },
4736 { FREAD, KF_FLAG_READ },
4737 { FWRITE, KF_FLAG_WRITE },
4738 { O_CREAT, KF_FLAG_CREAT },
4739 { O_DIRECT, KF_FLAG_DIRECT },
4740 { O_EXCL, KF_FLAG_EXCL },
4741 { O_EXEC, KF_FLAG_EXEC },
4742 { O_EXLOCK, KF_FLAG_EXLOCK },
4743 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4744 { O_SHLOCK, KF_FLAG_SHLOCK },
4745 { O_TRUNC, KF_FLAG_TRUNC }
4746 };
4747 unsigned int i;
4748 int kflags;
4749
4750 kflags = 0;
4751 for (i = 0; i < nitems(fflags_table); i++)
4752 if (fflags & fflags_table[i].fflag)
4753 kflags |= fflags_table[i].kf_fflag;
4754 return (kflags);
4755 }
4756
4757 /* Trim unused data from kf_path by truncating the structure size. */
4758 void
pack_kinfo(struct kinfo_file * kif)4759 pack_kinfo(struct kinfo_file *kif)
4760 {
4761
4762 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4763 strlen(kif->kf_path) + 1;
4764 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4765 }
4766
4767 static void
export_file_to_kinfo(struct file * fp,int fd,cap_rights_t * rightsp,struct kinfo_file * kif,struct filedesc * fdp,int flags)4768 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4769 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4770 {
4771 int error;
4772
4773 bzero(kif, sizeof(*kif));
4774
4775 /* Set a default type to allow for empty fill_kinfo() methods. */
4776 kif->kf_type = KF_TYPE_UNKNOWN;
4777 kif->kf_flags = xlate_fflags(fp->f_flag);
4778 if (rightsp != NULL)
4779 kif->kf_cap_rights = *rightsp;
4780 else
4781 cap_rights_init_zero(&kif->kf_cap_rights);
4782 kif->kf_fd = fd;
4783 kif->kf_ref_count = refcount_load(&fp->f_count);
4784 kif->kf_offset = foffset_get(fp);
4785
4786 /*
4787 * This may drop the filedesc lock, so the 'fp' cannot be
4788 * accessed after this call.
4789 */
4790 error = fo_fill_kinfo(fp, kif, fdp);
4791 if (error == 0)
4792 kif->kf_status |= KF_ATTR_VALID;
4793 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4794 pack_kinfo(kif);
4795 else
4796 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4797 }
4798
4799 static void
export_vnode_to_kinfo(struct vnode * vp,int fd,int fflags,struct kinfo_file * kif,int flags)4800 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4801 struct kinfo_file *kif, int flags)
4802 {
4803 int error;
4804
4805 bzero(kif, sizeof(*kif));
4806
4807 kif->kf_type = KF_TYPE_VNODE;
4808 error = vn_fill_kinfo_vnode(vp, kif);
4809 if (error == 0)
4810 kif->kf_status |= KF_ATTR_VALID;
4811 kif->kf_flags = xlate_fflags(fflags);
4812 cap_rights_init_zero(&kif->kf_cap_rights);
4813 kif->kf_fd = fd;
4814 kif->kf_ref_count = -1;
4815 kif->kf_offset = -1;
4816 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4817 pack_kinfo(kif);
4818 else
4819 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4820 vrele(vp);
4821 }
4822
4823 struct export_fd_buf {
4824 struct filedesc *fdp;
4825 struct pwddesc *pdp;
4826 struct sbuf *sb;
4827 ssize_t remainder;
4828 struct kinfo_file kif;
4829 int flags;
4830 };
4831
4832 static int
export_kinfo_to_sb(struct export_fd_buf * efbuf)4833 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4834 {
4835 struct kinfo_file *kif;
4836
4837 kif = &efbuf->kif;
4838 if (efbuf->remainder != -1) {
4839 if (efbuf->remainder < kif->kf_structsize)
4840 return (ENOMEM);
4841 efbuf->remainder -= kif->kf_structsize;
4842 }
4843 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4844 return (sbuf_error(efbuf->sb));
4845 return (0);
4846 }
4847
4848 static int
export_file_to_sb(struct file * fp,int fd,cap_rights_t * rightsp,struct export_fd_buf * efbuf)4849 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4850 struct export_fd_buf *efbuf)
4851 {
4852 int error;
4853
4854 if (efbuf->remainder == 0)
4855 return (ENOMEM);
4856 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4857 efbuf->flags);
4858 FILEDESC_SUNLOCK(efbuf->fdp);
4859 error = export_kinfo_to_sb(efbuf);
4860 FILEDESC_SLOCK(efbuf->fdp);
4861 return (error);
4862 }
4863
4864 static int
export_vnode_to_sb(struct vnode * vp,int fd,int fflags,struct export_fd_buf * efbuf)4865 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4866 struct export_fd_buf *efbuf)
4867 {
4868 int error;
4869
4870 if (efbuf->remainder == 0)
4871 return (ENOMEM);
4872 if (efbuf->pdp != NULL)
4873 PWDDESC_XUNLOCK(efbuf->pdp);
4874 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4875 error = export_kinfo_to_sb(efbuf);
4876 if (efbuf->pdp != NULL)
4877 PWDDESC_XLOCK(efbuf->pdp);
4878 return (error);
4879 }
4880
4881 /*
4882 * Store a process file descriptor information to sbuf.
4883 *
4884 * Takes a locked proc as argument, and returns with the proc unlocked.
4885 */
4886 int
kern_proc_filedesc_out(struct proc * p,struct sbuf * sb,ssize_t maxlen,int flags)4887 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4888 int flags)
4889 {
4890 struct file *fp;
4891 struct filedesc *fdp;
4892 struct pwddesc *pdp;
4893 struct export_fd_buf *efbuf;
4894 struct vnode *cttyvp, *textvp, *tracevp;
4895 struct pwd *pwd;
4896 int error, i;
4897 cap_rights_t rights;
4898
4899 PROC_LOCK_ASSERT(p, MA_OWNED);
4900
4901 /* ktrace vnode */
4902 tracevp = ktr_get_tracevp(p, true);
4903 /* text vnode */
4904 textvp = p->p_textvp;
4905 if (textvp != NULL)
4906 vrefact(textvp);
4907 /* Controlling tty. */
4908 cttyvp = NULL;
4909 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4910 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4911 if (cttyvp != NULL)
4912 vrefact(cttyvp);
4913 }
4914 fdp = fdhold(p);
4915 pdp = pdhold(p);
4916 PROC_UNLOCK(p);
4917
4918 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4919 efbuf->fdp = NULL;
4920 efbuf->pdp = NULL;
4921 efbuf->sb = sb;
4922 efbuf->remainder = maxlen;
4923 efbuf->flags = flags;
4924
4925 error = 0;
4926 if (tracevp != NULL)
4927 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4928 FREAD | FWRITE, efbuf);
4929 if (error == 0 && textvp != NULL)
4930 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4931 efbuf);
4932 if (error == 0 && cttyvp != NULL)
4933 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4934 FREAD | FWRITE, efbuf);
4935 if (error != 0 || pdp == NULL || fdp == NULL)
4936 goto fail;
4937 efbuf->fdp = fdp;
4938 efbuf->pdp = pdp;
4939 PWDDESC_XLOCK(pdp);
4940 pwd = pwd_hold_pwddesc(pdp);
4941 if (pwd != NULL) {
4942 /* working directory */
4943 if (pwd->pwd_cdir != NULL) {
4944 vrefact(pwd->pwd_cdir);
4945 error = export_vnode_to_sb(pwd->pwd_cdir,
4946 KF_FD_TYPE_CWD, FREAD, efbuf);
4947 }
4948 /* root directory */
4949 if (error == 0 && pwd->pwd_rdir != NULL) {
4950 vrefact(pwd->pwd_rdir);
4951 error = export_vnode_to_sb(pwd->pwd_rdir,
4952 KF_FD_TYPE_ROOT, FREAD, efbuf);
4953 }
4954 /* jail directory */
4955 if (error == 0 && pwd->pwd_jdir != NULL) {
4956 vrefact(pwd->pwd_jdir);
4957 error = export_vnode_to_sb(pwd->pwd_jdir,
4958 KF_FD_TYPE_JAIL, FREAD, efbuf);
4959 }
4960 }
4961 PWDDESC_XUNLOCK(pdp);
4962 if (error != 0)
4963 goto fail;
4964 if (pwd != NULL)
4965 pwd_drop(pwd);
4966 FILEDESC_SLOCK(fdp);
4967 if (refcount_load(&fdp->fd_refcnt) == 0)
4968 goto skip;
4969 FILEDESC_FOREACH_FP(fdp, i, fp) {
4970 #ifdef CAPABILITIES
4971 rights = *cap_rights(fdp, i);
4972 #else /* !CAPABILITIES */
4973 rights = cap_no_rights;
4974 #endif
4975 /*
4976 * Create sysctl entry. It is OK to drop the filedesc
4977 * lock inside of export_file_to_sb() as we will
4978 * re-validate and re-evaluate its properties when the
4979 * loop continues.
4980 */
4981 error = export_file_to_sb(fp, i, &rights, efbuf);
4982 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4983 break;
4984 }
4985 skip:
4986 FILEDESC_SUNLOCK(fdp);
4987 fail:
4988 if (fdp != NULL)
4989 fddrop(fdp);
4990 if (pdp != NULL)
4991 pddrop(pdp);
4992 free(efbuf, M_TEMP);
4993 return (error);
4994 }
4995
4996 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4997
4998 /*
4999 * Get per-process file descriptors for use by procstat(1), et al.
5000 */
5001 static int
sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)5002 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
5003 {
5004 struct sbuf sb;
5005 struct proc *p;
5006 ssize_t maxlen;
5007 u_int namelen;
5008 int error, error2, *name;
5009
5010 namelen = arg2;
5011 if (namelen != 1)
5012 return (EINVAL);
5013
5014 name = (int *)arg1;
5015
5016 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
5017 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
5018 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
5019 if (error != 0) {
5020 sbuf_delete(&sb);
5021 return (error);
5022 }
5023 maxlen = req->oldptr != NULL ? req->oldlen : -1;
5024 error = kern_proc_filedesc_out(p, &sb, maxlen,
5025 KERN_FILEDESC_PACK_KINFO);
5026 error2 = sbuf_finish(&sb);
5027 sbuf_delete(&sb);
5028 return (error != 0 ? error : error2);
5029 }
5030
5031 #ifdef COMPAT_FREEBSD7
5032 #ifdef KINFO_OFILE_SIZE
5033 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
5034 #endif
5035
5036 static void
kinfo_to_okinfo(struct kinfo_file * kif,struct kinfo_ofile * okif)5037 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
5038 {
5039
5040 okif->kf_structsize = sizeof(*okif);
5041 okif->kf_type = kif->kf_type;
5042 okif->kf_fd = kif->kf_fd;
5043 okif->kf_ref_count = kif->kf_ref_count;
5044 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
5045 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
5046 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
5047 okif->kf_offset = kif->kf_offset;
5048 if (kif->kf_type == KF_TYPE_VNODE)
5049 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
5050 else
5051 okif->kf_vnode_type = KF_VTYPE_VNON;
5052 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
5053 if (kif->kf_type == KF_TYPE_SOCKET) {
5054 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
5055 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
5056 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
5057 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
5058 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
5059 } else {
5060 okif->kf_sa_local.ss_family = AF_UNSPEC;
5061 okif->kf_sa_peer.ss_family = AF_UNSPEC;
5062 }
5063 }
5064
5065 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)5066 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
5067 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
5068 {
5069 int error;
5070
5071 vrefact(vp);
5072 PWDDESC_XUNLOCK(pdp);
5073 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
5074 kinfo_to_okinfo(kif, okif);
5075 error = SYSCTL_OUT(req, okif, sizeof(*okif));
5076 PWDDESC_XLOCK(pdp);
5077 return (error);
5078 }
5079
5080 /*
5081 * Get per-process file descriptors for use by procstat(1), et al.
5082 */
5083 static int
sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)5084 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
5085 {
5086 struct kinfo_ofile *okif;
5087 struct kinfo_file *kif;
5088 struct filedesc *fdp;
5089 struct pwddesc *pdp;
5090 struct pwd *pwd;
5091 u_int namelen;
5092 int error, i, *name;
5093 struct file *fp;
5094 struct proc *p;
5095
5096 namelen = arg2;
5097 if (namelen != 1)
5098 return (EINVAL);
5099
5100 name = (int *)arg1;
5101 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
5102 if (error != 0)
5103 return (error);
5104 fdp = fdhold(p);
5105 if (fdp != NULL)
5106 pdp = pdhold(p);
5107 PROC_UNLOCK(p);
5108 if (fdp == NULL || pdp == NULL) {
5109 if (fdp != NULL)
5110 fddrop(fdp);
5111 return (ENOENT);
5112 }
5113 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
5114 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
5115 PWDDESC_XLOCK(pdp);
5116 pwd = pwd_hold_pwddesc(pdp);
5117 if (pwd != NULL) {
5118 if (pwd->pwd_cdir != NULL)
5119 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
5120 okif, pdp, req);
5121 if (pwd->pwd_rdir != NULL)
5122 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
5123 okif, pdp, req);
5124 if (pwd->pwd_jdir != NULL)
5125 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
5126 okif, pdp, req);
5127 }
5128 PWDDESC_XUNLOCK(pdp);
5129 if (pwd != NULL)
5130 pwd_drop(pwd);
5131 FILEDESC_SLOCK(fdp);
5132 if (refcount_load(&fdp->fd_refcnt) == 0)
5133 goto skip;
5134 FILEDESC_FOREACH_FP(fdp, i, fp) {
5135 export_file_to_kinfo(fp, i, NULL, kif, fdp,
5136 KERN_FILEDESC_PACK_KINFO);
5137 FILEDESC_SUNLOCK(fdp);
5138 kinfo_to_okinfo(kif, okif);
5139 error = SYSCTL_OUT(req, okif, sizeof(*okif));
5140 FILEDESC_SLOCK(fdp);
5141 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
5142 break;
5143 }
5144 skip:
5145 FILEDESC_SUNLOCK(fdp);
5146 fddrop(fdp);
5147 pddrop(pdp);
5148 free(kif, M_TEMP);
5149 free(okif, M_TEMP);
5150 return (0);
5151 }
5152
5153 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
5154 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
5155 "Process ofiledesc entries");
5156 #endif /* COMPAT_FREEBSD7 */
5157
5158 int
vntype_to_kinfo(int vtype)5159 vntype_to_kinfo(int vtype)
5160 {
5161 struct {
5162 int vtype;
5163 int kf_vtype;
5164 } vtypes_table[] = {
5165 { VBAD, KF_VTYPE_VBAD },
5166 { VBLK, KF_VTYPE_VBLK },
5167 { VCHR, KF_VTYPE_VCHR },
5168 { VDIR, KF_VTYPE_VDIR },
5169 { VFIFO, KF_VTYPE_VFIFO },
5170 { VLNK, KF_VTYPE_VLNK },
5171 { VNON, KF_VTYPE_VNON },
5172 { VREG, KF_VTYPE_VREG },
5173 { VSOCK, KF_VTYPE_VSOCK }
5174 };
5175 unsigned int i;
5176
5177 /*
5178 * Perform vtype translation.
5179 */
5180 for (i = 0; i < nitems(vtypes_table); i++)
5181 if (vtypes_table[i].vtype == vtype)
5182 return (vtypes_table[i].kf_vtype);
5183
5184 return (KF_VTYPE_UNKNOWN);
5185 }
5186
5187 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
5188 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
5189 "Process filedesc entries");
5190
5191 /*
5192 * Store a process current working directory information to sbuf.
5193 *
5194 * Takes a locked proc as argument, and returns with the proc unlocked.
5195 */
5196 int
kern_proc_cwd_out(struct proc * p,struct sbuf * sb,ssize_t maxlen)5197 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
5198 {
5199 struct pwddesc *pdp;
5200 struct pwd *pwd;
5201 struct export_fd_buf *efbuf;
5202 struct vnode *cdir;
5203 int error;
5204
5205 PROC_LOCK_ASSERT(p, MA_OWNED);
5206
5207 pdp = pdhold(p);
5208 PROC_UNLOCK(p);
5209 if (pdp == NULL)
5210 return (EINVAL);
5211
5212 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
5213 efbuf->fdp = NULL;
5214 efbuf->pdp = pdp;
5215 efbuf->sb = sb;
5216 efbuf->remainder = maxlen;
5217 efbuf->flags = 0;
5218
5219 PWDDESC_XLOCK(pdp);
5220 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
5221 cdir = pwd->pwd_cdir;
5222 if (cdir == NULL) {
5223 error = EINVAL;
5224 } else {
5225 vrefact(cdir);
5226 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
5227 }
5228 PWDDESC_XUNLOCK(pdp);
5229 pddrop(pdp);
5230 free(efbuf, M_TEMP);
5231 return (error);
5232 }
5233
5234 /*
5235 * Get per-process current working directory.
5236 */
5237 static int
sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)5238 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
5239 {
5240 struct sbuf sb;
5241 struct proc *p;
5242 ssize_t maxlen;
5243 u_int namelen;
5244 int error, error2, *name;
5245
5246 namelen = arg2;
5247 if (namelen != 1)
5248 return (EINVAL);
5249
5250 name = (int *)arg1;
5251
5252 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
5253 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
5254 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
5255 if (error != 0) {
5256 sbuf_delete(&sb);
5257 return (error);
5258 }
5259 maxlen = req->oldptr != NULL ? req->oldlen : -1;
5260 error = kern_proc_cwd_out(p, &sb, maxlen);
5261 error2 = sbuf_finish(&sb);
5262 sbuf_delete(&sb);
5263 return (error != 0 ? error : error2);
5264 }
5265
5266 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
5267 sysctl_kern_proc_cwd, "Process current working directory");
5268
5269 #ifdef DDB
5270 /*
5271 * For the purposes of debugging, generate a human-readable string for the
5272 * file type.
5273 */
5274 static const char *
file_type_to_name(short type)5275 file_type_to_name(short type)
5276 {
5277
5278 switch (type) {
5279 case 0:
5280 return ("zero");
5281 case DTYPE_VNODE:
5282 return ("vnode");
5283 case DTYPE_SOCKET:
5284 return ("socket");
5285 case DTYPE_PIPE:
5286 return ("pipe");
5287 case DTYPE_FIFO:
5288 return ("fifo");
5289 case DTYPE_KQUEUE:
5290 return ("kqueue");
5291 case DTYPE_CRYPTO:
5292 return ("crypto");
5293 case DTYPE_MQUEUE:
5294 return ("mqueue");
5295 case DTYPE_SHM:
5296 return ("shm");
5297 case DTYPE_SEM:
5298 return ("ksem");
5299 case DTYPE_PTS:
5300 return ("pts");
5301 case DTYPE_DEV:
5302 return ("dev");
5303 case DTYPE_PROCDESC:
5304 return ("proc");
5305 case DTYPE_EVENTFD:
5306 return ("eventfd");
5307 case DTYPE_TIMERFD:
5308 return ("timerfd");
5309 case DTYPE_JAILDESC:
5310 return ("jail");
5311 default:
5312 return ("unkn");
5313 }
5314 }
5315
5316 /*
5317 * For the purposes of debugging, identify a process (if any, perhaps one of
5318 * many) that references the passed file in its file descriptor array. Return
5319 * NULL if none.
5320 */
5321 static struct proc *
file_to_first_proc(struct file * fp)5322 file_to_first_proc(struct file *fp)
5323 {
5324 struct filedesc *fdp;
5325 struct proc *p;
5326 int n;
5327
5328 FOREACH_PROC_IN_SYSTEM(p) {
5329 if (p->p_state == PRS_NEW)
5330 continue;
5331 fdp = p->p_fd;
5332 if (fdp == NULL)
5333 continue;
5334 for (n = 0; n < fdp->fd_nfiles; n++) {
5335 if (fp == fdp->fd_ofiles[n].fde_file)
5336 return (p);
5337 }
5338 }
5339 return (NULL);
5340 }
5341
5342 static void
db_print_file(struct file * fp,int header)5343 db_print_file(struct file *fp, int header)
5344 {
5345 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
5346 struct proc *p;
5347
5348 if (header)
5349 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
5350 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
5351 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
5352 "FCmd");
5353 p = file_to_first_proc(fp);
5354 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
5355 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
5356 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
5357 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
5358
5359 #undef XPTRWIDTH
5360 }
5361
DB_SHOW_COMMAND(file,db_show_file)5362 DB_SHOW_COMMAND(file, db_show_file)
5363 {
5364 struct file *fp;
5365
5366 if (!have_addr) {
5367 db_printf("usage: show file <addr>\n");
5368 return;
5369 }
5370 fp = (struct file *)addr;
5371 db_print_file(fp, 1);
5372 }
5373
DB_SHOW_COMMAND_FLAGS(files,db_show_files,DB_CMD_MEMSAFE)5374 DB_SHOW_COMMAND_FLAGS(files, db_show_files, DB_CMD_MEMSAFE)
5375 {
5376 struct filedesc *fdp;
5377 struct file *fp;
5378 struct proc *p;
5379 int header;
5380 int n;
5381
5382 header = 1;
5383 FOREACH_PROC_IN_SYSTEM(p) {
5384 if (p->p_state == PRS_NEW)
5385 continue;
5386 if ((fdp = p->p_fd) == NULL)
5387 continue;
5388 for (n = 0; n < fdp->fd_nfiles; ++n) {
5389 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
5390 continue;
5391 db_print_file(fp, header);
5392 header = 0;
5393 }
5394 }
5395 }
5396 #endif
5397
5398 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc,
5399 CTLFLAG_RWTUN | CTLFLAG_NOFETCH,
5400 &maxfilesperproc, 0, "Maximum files allowed open per process");
5401
5402 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RWTUN | CTLFLAG_NOFETCH,
5403 &maxfiles, 0, "Maximum number of files");
5404
5405 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
5406 &openfiles, 0, "System-wide number of open files");
5407
5408 /* ARGSUSED*/
5409 static void
filelistinit(void * dummy)5410 filelistinit(void *dummy)
5411 {
5412
5413 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
5414 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
5415 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
5416 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
5417 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
5418 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
5419 /*
5420 * XXXMJG this is a temporary hack due to boot ordering issues against
5421 * the vnode zone.
5422 */
5423 vfs_smr = uma_zone_get_smr(pwd_zone);
5424 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
5425 }
5426 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
5427
5428 /*-------------------------------------------------------------------*/
5429
5430 static int
badfo_readwrite(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)5431 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
5432 int flags, struct thread *td)
5433 {
5434
5435 return (EBADF);
5436 }
5437
5438 static int
badfo_truncate(struct file * fp,off_t length,struct ucred * active_cred,struct thread * td)5439 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5440 struct thread *td)
5441 {
5442
5443 return (EINVAL);
5444 }
5445
5446 static int
badfo_ioctl(struct file * fp,u_long com,void * data,struct ucred * active_cred,struct thread * td)5447 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
5448 struct thread *td)
5449 {
5450
5451 return (EBADF);
5452 }
5453
5454 static int
badfo_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)5455 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
5456 struct thread *td)
5457 {
5458
5459 return (0);
5460 }
5461
5462 static int
badfo_kqfilter(struct file * fp,struct knote * kn)5463 badfo_kqfilter(struct file *fp, struct knote *kn)
5464 {
5465
5466 return (EBADF);
5467 }
5468
5469 static int
badfo_stat(struct file * fp,struct stat * sb,struct ucred * active_cred)5470 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
5471 {
5472
5473 return (EBADF);
5474 }
5475
5476 static int
badfo_close(struct file * fp,struct thread * td)5477 badfo_close(struct file *fp, struct thread *td)
5478 {
5479
5480 return (0);
5481 }
5482
5483 static int
badfo_chmod(struct file * fp,mode_t mode,struct ucred * active_cred,struct thread * td)5484 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5485 struct thread *td)
5486 {
5487
5488 return (EBADF);
5489 }
5490
5491 static int
badfo_chown(struct file * fp,uid_t uid,gid_t gid,struct ucred * active_cred,struct thread * td)5492 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5493 struct thread *td)
5494 {
5495
5496 return (EBADF);
5497 }
5498
5499 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)5500 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5501 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5502 struct thread *td)
5503 {
5504
5505 return (EBADF);
5506 }
5507
5508 static int
badfo_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)5509 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
5510 {
5511
5512 return (0);
5513 }
5514
5515 const struct fileops badfileops = {
5516 .fo_read = badfo_readwrite,
5517 .fo_write = badfo_readwrite,
5518 .fo_truncate = badfo_truncate,
5519 .fo_ioctl = badfo_ioctl,
5520 .fo_poll = badfo_poll,
5521 .fo_kqfilter = badfo_kqfilter,
5522 .fo_stat = badfo_stat,
5523 .fo_close = badfo_close,
5524 .fo_chmod = badfo_chmod,
5525 .fo_chown = badfo_chown,
5526 .fo_sendfile = badfo_sendfile,
5527 .fo_fill_kinfo = badfo_fill_kinfo,
5528 };
5529
5530 static int
path_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)5531 path_poll(struct file *fp, int events, struct ucred *active_cred,
5532 struct thread *td)
5533 {
5534 return (POLLNVAL);
5535 }
5536
5537 static int
path_close(struct file * fp,struct thread * td)5538 path_close(struct file *fp, struct thread *td)
5539 {
5540 MPASS(fp->f_type == DTYPE_VNODE);
5541 fp->f_ops = &badfileops;
5542 vrele(fp->f_vnode);
5543 return (0);
5544 }
5545
5546 const struct fileops path_fileops = {
5547 .fo_read = badfo_readwrite,
5548 .fo_write = badfo_readwrite,
5549 .fo_truncate = badfo_truncate,
5550 .fo_ioctl = badfo_ioctl,
5551 .fo_poll = path_poll,
5552 .fo_kqfilter = vn_kqfilter_opath,
5553 .fo_stat = vn_statfile,
5554 .fo_close = path_close,
5555 .fo_chmod = badfo_chmod,
5556 .fo_chown = badfo_chown,
5557 .fo_sendfile = badfo_sendfile,
5558 .fo_fill_kinfo = vn_fill_kinfo,
5559 .fo_cmp = vn_cmp,
5560 .fo_flags = DFLAG_PASSABLE,
5561 };
5562
5563 int
invfo_rdwr(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)5564 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5565 int flags, struct thread *td)
5566 {
5567
5568 return (EOPNOTSUPP);
5569 }
5570
5571 int
invfo_truncate(struct file * fp,off_t length,struct ucred * active_cred,struct thread * td)5572 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5573 struct thread *td)
5574 {
5575
5576 return (EINVAL);
5577 }
5578
5579 int
invfo_ioctl(struct file * fp,u_long com,void * data,struct ucred * active_cred,struct thread * td)5580 invfo_ioctl(struct file *fp, u_long com, void *data,
5581 struct ucred *active_cred, struct thread *td)
5582 {
5583
5584 return (ENOTTY);
5585 }
5586
5587 int
invfo_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)5588 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5589 struct thread *td)
5590 {
5591
5592 return (poll_no_poll(events));
5593 }
5594
5595 int
invfo_kqfilter(struct file * fp,struct knote * kn)5596 invfo_kqfilter(struct file *fp, struct knote *kn)
5597 {
5598
5599 return (EINVAL);
5600 }
5601
5602 int
invfo_chmod(struct file * fp,mode_t mode,struct ucred * active_cred,struct thread * td)5603 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5604 struct thread *td)
5605 {
5606
5607 return (EINVAL);
5608 }
5609
5610 int
invfo_chown(struct file * fp,uid_t uid,gid_t gid,struct ucred * active_cred,struct thread * td)5611 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5612 struct thread *td)
5613 {
5614
5615 return (EINVAL);
5616 }
5617
5618 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)5619 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5620 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5621 struct thread *td)
5622 {
5623
5624 return (EINVAL);
5625 }
5626
5627 /*-------------------------------------------------------------------*/
5628
5629 /*
5630 * File Descriptor pseudo-device driver (/dev/fd/).
5631 *
5632 * Opening minor device N dup()s the file (if any) connected to file
5633 * descriptor N belonging to the calling process. Note that this driver
5634 * consists of only the ``open()'' routine, because all subsequent
5635 * references to this file will be direct to the other driver.
5636 *
5637 * XXX: we could give this one a cloning event handler if necessary.
5638 */
5639
5640 /* ARGSUSED */
5641 static int
fdopen(struct cdev * dev,int mode,int type,struct thread * td)5642 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5643 {
5644
5645 /*
5646 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5647 * the file descriptor being sought for duplication. The error
5648 * return ensures that the vnode for this device will be released
5649 * by vn_open. Open will detect this special error and take the
5650 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5651 * will simply report the error.
5652 */
5653 td->td_dupfd = dev2unit(dev);
5654 return (ENODEV);
5655 }
5656
5657 static struct cdevsw fildesc_cdevsw = {
5658 .d_version = D_VERSION,
5659 .d_open = fdopen,
5660 .d_name = "FD",
5661 };
5662
5663 static void
fildesc_drvinit(void * unused)5664 fildesc_drvinit(void *unused)
5665 {
5666 struct cdev *dev;
5667
5668 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5669 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5670 make_dev_alias(dev, "stdin");
5671 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5672 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5673 make_dev_alias(dev, "stdout");
5674 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5675 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5676 make_dev_alias(dev, "stderr");
5677 }
5678
5679 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5680