1 /*
2 * Copyright (c) 2005-2018 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Jeffrey Hsu and Matthew Dillon.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 *
35 * Copyright (c) 1982, 1986, 1989, 1991, 1993
36 * The Regents of the University of California. All rights reserved.
37 * (c) UNIX System Laboratories, Inc.
38 * All or some portions of this file are derived from material licensed
39 * to the University of California by American Telephone and Telegraph
40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41 * the permission of UNIX System Laboratories, Inc.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. Neither the name of the University nor the names of its contributors
52 * may be used to endorse or promote products derived from this software
53 * without specific prior written permission.
54 *
55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65 * SUCH DAMAGE.
66 *
67 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
68 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
69 */
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/malloc.h>
74 #include <sys/sysmsg.h>
75 #include <sys/conf.h>
76 #include <sys/device.h>
77 #include <sys/file.h>
78 #include <sys/filedesc.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81 #include <sys/vnode.h>
82 #include <sys/proc.h>
83 #include <sys/nlookup.h>
84 #include <sys/stat.h>
85 #include <sys/filio.h>
86 #include <sys/fcntl.h>
87 #include <sys/unistd.h>
88 #include <sys/resourcevar.h>
89 #include <sys/event.h>
90 #include <sys/kern_syscall.h>
91 #include <sys/kcore.h>
92 #include <sys/kinfo.h>
93 #include <sys/un.h>
94 #include <sys/objcache.h>
95
96 #include <vm/vm.h>
97 #include <vm/vm_extern.h>
98
99 #include <sys/file2.h>
100 #include <sys/spinlock2.h>
101
102 static int fdalloc_locked(struct proc *p, struct filedesc *fdp,
103 int want, int *result);
104 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
105 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
106 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
107 static void ffree(struct file *fp);
108
109 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
110 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
111 "file desc to leader structures");
112 static MALLOC_DEFINE_OBJ(M_FILE, sizeof(struct file),
113 "file", "Open file structure");
114 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
115
116 static struct krate krate_uidinfo = { .freq = 1 };
117
118 static d_open_t fdopen;
119 #define NUMFDESC 64
120
121 #define CDEV_MAJOR 22
122 static struct dev_ops fildesc_ops = {
123 { "FD", 0, 0 },
124 .d_open = fdopen,
125 };
126
127 /*
128 * Descriptor management.
129 */
130 #ifndef NFILELIST_HEADS
131 #define NFILELIST_HEADS 257 /* primary number */
132 #endif
133
134 struct filelist_head {
135 struct spinlock spin;
136 struct filelist list;
137 } __cachealign;
138
139 static struct filelist_head filelist_heads[NFILELIST_HEADS];
140
141 static int nfiles; /* actual number of open files */
142 extern int cmask;
143
144 struct lwkt_token revoke_token = LWKT_TOKEN_INITIALIZER(revoke_token);
145
146 /*
147 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
148 *
149 * must be called with fdp->fd_spin exclusively held
150 */
151 static __inline
152 void
fdfixup_locked(struct filedesc * fdp,int fd)153 fdfixup_locked(struct filedesc *fdp, int fd)
154 {
155 if (fd < fdp->fd_freefile) {
156 fdp->fd_freefile = fd;
157 }
158 while (fdp->fd_lastfile >= 0 &&
159 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
160 fdp->fd_files[fdp->fd_lastfile].reserved == 0
161 ) {
162 --fdp->fd_lastfile;
163 }
164 }
165
166 /*
167 * Clear the fd thread caches for this fdnode.
168 *
169 * If match_fdc is NULL, all thread caches of fdn will be cleared.
170 * The caller must hold fdp->fd_spin exclusively. The threads caching
171 * the descriptor do not have to be the current thread. The (status)
172 * argument is ignored.
173 *
174 * If match_fdc is not NULL, only the match_fdc's cache will be cleared.
175 * The caller must hold fdp->fd_spin shared and match_fdc must match a
176 * fdcache entry in curthread. match_fdc has been locked by the caller
177 * and had the specified (status).
178 *
179 * Since we are matching against a fp in the fdp (which must still be present
180 * at this time), fp will have at least two refs on any match and we can
181 * decrement the count trivially.
182 */
183 static
184 void
fclearcache(struct fdnode * fdn,struct fdcache * match_fdc,int status)185 fclearcache(struct fdnode *fdn, struct fdcache *match_fdc, int status)
186 {
187 struct fdcache *fdc;
188 struct file *fp;
189 int i;
190
191 /*
192 * match_fdc == NULL We are cleaning out all tdcache entries
193 * for the fdn and hold fdp->fd_spin exclusively.
194 * This can race against the target threads
195 * cleaning out specific entries.
196 *
197 * match_fdc != NULL We are cleaning out a specific tdcache
198 * entry on behalf of the owning thread
199 * and hold fdp->fd_spin shared. The thread
200 * has already locked the entry. This cannot
201 * race.
202 */
203 fp = fdn->fp;
204 for (i = 0; i < NTDCACHEFD; ++i) {
205 if ((fdc = fdn->tdcache[i]) == NULL)
206 continue;
207
208 /*
209 * If match_fdc is non-NULL we are being asked to
210 * clear a specific fdc owned by curthread. There must
211 * be exactly one match. The caller has already locked
212 * the cache entry and will dispose of the lock after
213 * we return.
214 *
215 * Since we also have a shared lock on fdp, we
216 * can do this without atomic ops.
217 */
218 if (match_fdc) {
219 if (fdc != match_fdc)
220 continue;
221 fdn->tdcache[i] = NULL;
222 KASSERT(fp == fdc->fp,
223 ("fclearcache(1): fp mismatch %p/%p\n",
224 fp, fdc->fp));
225 fdc->fp = NULL;
226 fdc->fd = -1;
227
228 /*
229 * status can be 0 or 2. If 2 the ref is borrowed,
230 * if 0 the ref is not borrowed and we have to drop
231 * it.
232 */
233 if (status == 0)
234 atomic_add_int(&fp->f_count, -1);
235 fdn->isfull = 0; /* heuristic */
236 return;
237 }
238
239 /*
240 * Otherwise we hold an exclusive spin-lock and can only
241 * race thread consumers borrowing cache entries.
242 *
243 * Acquire the lock and dispose of the entry. We have to
244 * spin until we get the lock.
245 */
246 for (;;) {
247 status = atomic_swap_int(&fdc->locked, 1);
248 if (status == 1) { /* foreign lock, retry */
249 cpu_pause();
250 continue;
251 }
252 fdn->tdcache[i] = NULL;
253 KASSERT(fp == fdc->fp,
254 ("fclearcache(2): fp mismatch %p/%p\n",
255 fp, fdc->fp));
256 fdc->fp = NULL;
257 fdc->fd = -1;
258 if (status == 0)
259 atomic_add_int(&fp->f_count, -1);
260 fdn->isfull = 0; /* heuristic */
261 atomic_swap_int(&fdc->locked, 0);
262 break;
263 }
264 }
265 KKASSERT(match_fdc == NULL);
266 }
267
268 /*
269 * Retrieve the fp for the specified fd given the specified file descriptor
270 * table. The fdp does not have to be owned by the current process.
271 * If flags != -1, fp->f_flag must contain at least one of the flags.
272 *
273 * This function is not able to cache the fp.
274 */
275 struct file *
holdfp_fdp(struct filedesc * fdp,int fd,int flag)276 holdfp_fdp(struct filedesc *fdp, int fd, int flag)
277 {
278 struct file *fp;
279
280 spin_lock_shared(&fdp->fd_spin);
281 if (((u_int)fd) < fdp->fd_nfiles) {
282 fp = fdp->fd_files[fd].fp; /* can be NULL */
283 if (fp) {
284 if ((fp->f_flag & flag) == 0 && flag != -1) {
285 fp = NULL;
286 } else {
287 fhold(fp);
288 }
289 }
290 } else {
291 fp = NULL;
292 }
293 spin_unlock_shared(&fdp->fd_spin);
294
295 return fp;
296 }
297
298 struct file *
holdfp_fdp_locked(struct filedesc * fdp,int fd,int flag)299 holdfp_fdp_locked(struct filedesc *fdp, int fd, int flag)
300 {
301 struct file *fp;
302
303 if (((u_int)fd) < fdp->fd_nfiles) {
304 fp = fdp->fd_files[fd].fp; /* can be NULL */
305 if (fp) {
306 if ((fp->f_flag & flag) == 0 && flag != -1) {
307 fp = NULL;
308 } else {
309 fhold(fp);
310 }
311 }
312 } else {
313 fp = NULL;
314 }
315 return fp;
316 }
317
318 /*
319 * Acquire the fp for the specified file descriptor, using the thread
320 * cache if possible and caching it if possible.
321 *
322 * td must be the curren thread.
323 */
324 static
325 struct file *
_holdfp_cache(thread_t td,int fd)326 _holdfp_cache(thread_t td, int fd)
327 {
328 struct filedesc *fdp;
329 struct fdcache *fdc;
330 struct fdcache *best;
331 struct fdnode *fdn;
332 struct file *fp;
333 int status;
334 int delta;
335 int i;
336
337 /*
338 * Fast
339 */
340 for (fdc = &td->td_fdcache[0]; fdc < &td->td_fdcache[NFDCACHE]; ++fdc) {
341 if (fdc->fd != fd || fdc->fp == NULL)
342 continue;
343 status = atomic_swap_int(&fdc->locked, 1);
344
345 /*
346 * If someone else has locked our cache entry they are in
347 * the middle of clearing it, skip the entry.
348 */
349 if (status == 1)
350 continue;
351
352 /*
353 * We have locked the entry, but if it no longer matches
354 * restore the previous state (0 or 2) and skip the entry.
355 */
356 if (fdc->fd != fd || fdc->fp == NULL) {
357 atomic_swap_int(&fdc->locked, status);
358 continue;
359 }
360
361 /*
362 * We have locked a valid entry. We can borrow the ref
363 * for a mode 0 entry. We can get a valid fp for a mode
364 * 2 entry but not borrow the ref.
365 */
366 if (status == 0) {
367 fp = fdc->fp;
368 fdc->lru = ++td->td_fdcache_lru;
369 atomic_swap_int(&fdc->locked, 2);
370
371 return fp;
372 }
373 if (status == 2) {
374 fp = fdc->fp;
375 fhold(fp);
376 fdc->lru = ++td->td_fdcache_lru;
377 atomic_swap_int(&fdc->locked, 2);
378
379 return fp;
380 }
381 KKASSERT(0);
382 }
383
384 /*
385 * Lookup the descriptor the slow way. This can contend against
386 * modifying operations in a multi-threaded environment and cause
387 * cache line ping ponging otherwise.
388 */
389 fdp = td->td_proc->p_fd;
390 spin_lock_shared(&fdp->fd_spin);
391
392 if (((u_int)fd) < fdp->fd_nfiles) {
393 fp = fdp->fd_files[fd].fp; /* can be NULL */
394 if (fp) {
395 fhold(fp);
396 if (fdp->fd_files[fd].isfull == 0)
397 goto enter;
398 }
399 } else {
400 fp = NULL;
401 }
402 spin_unlock_shared(&fdp->fd_spin);
403
404 return fp;
405
406 /*
407 * We found a valid fp and held it, fdp is still shared locked.
408 * Enter the fp into the per-thread cache. Find the oldest entry
409 * via lru, or an empty entry.
410 *
411 * Because fdp's spinlock is held (shared is fine), no other
412 * thread should be in the middle of clearing our selected entry.
413 */
414 enter:
415 best = &td->td_fdcache[0];
416 for (fdc = &td->td_fdcache[0]; fdc < &td->td_fdcache[NFDCACHE]; ++fdc) {
417 if (fdc->fp == NULL) {
418 best = fdc;
419 break;
420 }
421 delta = fdc->lru - best->lru;
422 if (delta < 0)
423 best = fdc;
424 }
425
426 /*
427 * Replace best
428 *
429 * Don't enter into the cache if we cannot get the lock.
430 */
431 status = atomic_swap_int(&best->locked, 1);
432 if (status == 1)
433 goto done;
434
435 /*
436 * Clear the previous cache entry if present
437 */
438 if (best->fp) {
439 KKASSERT(best->fd >= 0);
440 fclearcache(&fdp->fd_files[best->fd], best, status);
441 }
442
443 /*
444 * Create our new cache entry. This entry is 'safe' until we tie
445 * into the fdnode. If we cannot tie in, we will clear the entry.
446 */
447 best->fd = fd;
448 best->fp = fp;
449 best->lru = ++td->td_fdcache_lru;
450 best->locked = 2; /* borrowed ref */
451
452 fdn = &fdp->fd_files[fd];
453 for (i = 0; i < NTDCACHEFD; ++i) {
454 if (fdn->tdcache[i] == NULL &&
455 atomic_cmpset_ptr((void **)&fdn->tdcache[i], NULL, best)) {
456 goto done;
457 }
458 }
459 fdn->isfull = 1; /* no space */
460 best->fd = -1;
461 best->fp = NULL;
462 best->locked = 0;
463 done:
464 spin_unlock_shared(&fdp->fd_spin);
465
466 return fp;
467 }
468
469 /*
470 * holdfp(), bypassing the cache in order to also be able to return
471 * the descriptor flags. A bit of a hack.
472 */
473 static
474 struct file *
_holdfp2(thread_t td,int fd,char * fflagsp)475 _holdfp2(thread_t td, int fd, char *fflagsp)
476 {
477 struct filedesc *fdp;
478 struct file *fp;
479
480 /*
481 * Lookup the descriptor the slow way. This can contend against
482 * modifying operations in a multi-threaded environment and cause
483 * cache line ping ponging otherwise.
484 */
485 fdp = td->td_proc->p_fd;
486 spin_lock_shared(&fdp->fd_spin);
487
488 if (((u_int)fd) < fdp->fd_nfiles) {
489 fp = fdp->fd_files[fd].fp; /* can be NULL */
490 if (fp) {
491 *fflagsp = fdp->fd_files[fd].fileflags;
492 fhold(fp);
493 }
494 } else {
495 fp = NULL;
496 }
497 spin_unlock_shared(&fdp->fd_spin);
498
499 return fp;
500 }
501
502
503 /*
504 * Drop the file pointer and return to the thread cache if possible.
505 *
506 * Caller must not hold fdp's spin lock.
507 * td must be the current thread.
508 */
509 void
dropfp(thread_t td,int fd,struct file * fp)510 dropfp(thread_t td, int fd, struct file *fp)
511 {
512 struct filedesc *fdp;
513 struct fdcache *fdc;
514 int status;
515
516 fdp = td->td_proc->p_fd;
517
518 /*
519 * If our placeholder is still present we can re-cache the ref.
520 *
521 * Note that we can race an fclearcache().
522 */
523 for (fdc = &td->td_fdcache[0]; fdc < &td->td_fdcache[NFDCACHE]; ++fdc) {
524 if (fdc->fp != fp || fdc->fd != fd)
525 continue;
526 status = atomic_swap_int(&fdc->locked, 1);
527 switch(status) {
528 case 0:
529 /*
530 * Not in mode 2, fdrop fp without caching.
531 */
532 atomic_swap_int(&fdc->locked, 0);
533 break;
534 case 1:
535 /*
536 * Not in mode 2, locked by someone else.
537 * fdrop fp without caching.
538 */
539 break;
540 case 2:
541 /*
542 * Intact borrowed ref, return to mode 0
543 * indicating that we have returned the ref.
544 *
545 * Return the borrowed ref (2->1->0)
546 */
547 if (fdc->fp == fp && fdc->fd == fd) {
548 atomic_swap_int(&fdc->locked, 0);
549 return;
550 }
551 atomic_swap_int(&fdc->locked, 2);
552 break;
553 }
554 }
555
556 /*
557 * Failed to re-cache, drop the fp without caching.
558 */
559 fdrop(fp);
560 }
561
562 /*
563 * Clear all descriptors cached in the per-thread fd cache for
564 * the specified thread.
565 *
566 * Caller must not hold p_fd->spin. This function will temporarily
567 * obtain a shared spin lock.
568 */
569 void
fexitcache(thread_t td)570 fexitcache(thread_t td)
571 {
572 struct filedesc *fdp;
573 struct fdcache *fdc;
574 int status;
575 int i;
576
577 if (td->td_proc == NULL)
578 return;
579 fdp = td->td_proc->p_fd;
580 if (fdp == NULL)
581 return;
582
583 /*
584 * A shared lock is sufficient as the caller controls td and we
585 * are only clearing td's cache.
586 */
587 spin_lock_shared(&fdp->fd_spin);
588 for (i = 0; i < NFDCACHE; ++i) {
589 fdc = &td->td_fdcache[i];
590 if (fdc->fp) {
591 status = atomic_swap_int(&fdc->locked, 1);
592 if (status == 1) {
593 cpu_pause();
594 --i;
595 continue;
596 }
597 if (fdc->fp) {
598 KKASSERT(fdc->fd >= 0);
599 fclearcache(&fdp->fd_files[fdc->fd], fdc,
600 status);
601 }
602 atomic_swap_int(&fdc->locked, 0);
603 }
604 }
605 spin_unlock_shared(&fdp->fd_spin);
606 }
607
608 static __inline struct filelist_head *
fp2filelist(const struct file * fp)609 fp2filelist(const struct file *fp)
610 {
611 u_int i;
612
613 i = (u_int)(uintptr_t)fp % NFILELIST_HEADS;
614 return &filelist_heads[i];
615 }
616
617 static __inline
618 struct plimit *
readplimits(struct proc * p)619 readplimits(struct proc *p)
620 {
621 thread_t td = curthread;
622 struct plimit *limit;
623
624 limit = td->td_limit;
625 if (limit != p->p_limit) {
626 spin_lock_shared(&p->p_spin);
627 limit = p->p_limit;
628 atomic_add_int(&limit->p_refcnt, 1);
629 spin_unlock_shared(&p->p_spin);
630 if (td->td_limit)
631 plimit_free(td->td_limit);
632 td->td_limit = limit;
633 }
634 return limit;
635 }
636
637 /*
638 * System calls on descriptors.
639 */
640 int
sys_getdtablesize(struct sysmsg * sysmsg,const struct getdtablesize_args * uap)641 sys_getdtablesize(struct sysmsg *sysmsg, const struct getdtablesize_args *uap)
642 {
643 struct proc *p = curproc;
644 struct plimit *limit = readplimits(p);
645 int dtsize;
646
647 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
648 dtsize = INT_MAX;
649 else
650 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
651
652 if (dtsize > maxfilesperproc)
653 dtsize = maxfilesperproc;
654 if (dtsize < minfilesperproc)
655 dtsize = minfilesperproc;
656 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
657 dtsize = maxfilesperuser;
658 sysmsg->sysmsg_result = dtsize;
659 return (0);
660 }
661
662 /*
663 * Duplicate a file descriptor to a particular value.
664 *
665 * note: keep in mind that a potential race condition exists when closing
666 * descriptors from a shared descriptor table (via rfork).
667 */
668 int
sys_dup2(struct sysmsg * sysmsg,const struct dup2_args * uap)669 sys_dup2(struct sysmsg *sysmsg, const struct dup2_args *uap)
670 {
671 int error;
672 int fd = 0;
673
674 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
675 sysmsg->sysmsg_fds[0] = fd;
676
677 return (error);
678 }
679
680 /*
681 * Duplicate a file descriptor.
682 */
683 int
sys_dup(struct sysmsg * sysmsg,const struct dup_args * uap)684 sys_dup(struct sysmsg *sysmsg, const struct dup_args *uap)
685 {
686 int error;
687 int fd = 0;
688
689 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
690 sysmsg->sysmsg_fds[0] = fd;
691
692 return (error);
693 }
694
695 /*
696 * MPALMOSTSAFE - acquires mplock for fp operations
697 */
698 int
kern_fcntl(int fd,int cmd,union fcntl_dat * dat,struct ucred * cred)699 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
700 {
701 struct thread *td = curthread;
702 struct proc *p = td->td_proc;
703 struct file *fp;
704 struct vnode *vp;
705 u_int newmin;
706 u_int oflags;
707 u_int nflags;
708 int closedcounter;
709 int tmp, error, flg = F_POSIX;
710
711 KKASSERT(p);
712
713 /*
714 * Operations on file descriptors that do not require a file pointer.
715 */
716 switch (cmd) {
717 case F_GETFD:
718 error = fgetfdflags(p->p_fd, fd, &tmp);
719 if (error == 0)
720 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
721 return (error);
722
723 case F_SETFD:
724 if (dat->fc_cloexec & FD_CLOEXEC)
725 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
726 else
727 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
728 return (error);
729 case F_DUPFD:
730 newmin = dat->fc_fd;
731 error = kern_dup(DUP_VARIABLE | DUP_FCNTL, fd, newmin,
732 &dat->fc_fd);
733 return (error);
734 case F_DUPFD_CLOEXEC:
735 newmin = dat->fc_fd;
736 error = kern_dup(DUP_VARIABLE | DUP_CLOEXEC | DUP_FCNTL,
737 fd, newmin, &dat->fc_fd);
738 return (error);
739 case F_DUP2FD:
740 newmin = dat->fc_fd;
741 error = kern_dup(DUP_FIXED, fd, newmin, &dat->fc_fd);
742 return (error);
743 case F_DUP2FD_CLOEXEC:
744 newmin = dat->fc_fd;
745 error = kern_dup(DUP_FIXED | DUP_CLOEXEC, fd, newmin,
746 &dat->fc_fd);
747 return (error);
748 default:
749 break;
750 }
751
752 /*
753 * Operations on file pointers
754 */
755 closedcounter = p->p_fd->fd_closedcounter;
756 if ((fp = holdfp(td, fd, -1)) == NULL)
757 return (EBADF);
758
759 switch (cmd) {
760 case F_GETFL:
761 dat->fc_flags = OFLAGS(fp->f_flag);
762 error = 0;
763 break;
764
765 case F_SETFL:
766 oflags = fp->f_flag;
767 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
768 nflags |= oflags & ~FCNTLFLAGS;
769
770 error = 0;
771 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
772 error = EINVAL;
773 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
774 tmp = nflags & FASYNC;
775 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
776 cred, NULL);
777 }
778
779 /*
780 * If no error, must be atomically set.
781 */
782 while (error == 0) {
783 oflags = fp->f_flag;
784 cpu_ccfence();
785 nflags = (oflags & ~FCNTLFLAGS) | (nflags & FCNTLFLAGS);
786 if (atomic_cmpset_int(&fp->f_flag, oflags, nflags))
787 break;
788 cpu_pause();
789 }
790 break;
791
792 case F_GETOWN:
793 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
794 cred, NULL);
795 break;
796
797 case F_SETOWN:
798 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
799 cred, NULL);
800 break;
801
802 case F_SETLKW:
803 flg |= F_WAIT;
804 /* Fall into F_SETLK */
805
806 case F_SETLK:
807 if (fp->f_type != DTYPE_VNODE) {
808 error = EBADF;
809 break;
810 }
811 vp = (struct vnode *)fp->f_data;
812
813 /*
814 * copyin/lockop may block
815 */
816 if (dat->fc_flock.l_whence == SEEK_CUR)
817 dat->fc_flock.l_start += fp->f_offset;
818
819 switch (dat->fc_flock.l_type) {
820 case F_RDLCK:
821 if ((fp->f_flag & FREAD) == 0) {
822 error = EBADF;
823 break;
824 }
825 if (p->p_leader->p_advlock_flag == 0)
826 p->p_leader->p_advlock_flag = 1;
827 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
828 &dat->fc_flock, flg);
829 break;
830 case F_WRLCK:
831 if ((fp->f_flag & FWRITE) == 0) {
832 error = EBADF;
833 break;
834 }
835 if (p->p_leader->p_advlock_flag == 0)
836 p->p_leader->p_advlock_flag = 1;
837 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
838 &dat->fc_flock, flg);
839 break;
840 case F_UNLCK:
841 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
842 &dat->fc_flock, F_POSIX);
843 break;
844 default:
845 error = EINVAL;
846 break;
847 }
848
849 /*
850 * It is possible to race a close() on the descriptor while
851 * we were blocked getting the lock. If this occurs the
852 * close might not have caught the lock.
853 */
854 if (checkfdclosed(td, p->p_fd, fd, fp, closedcounter)) {
855 dat->fc_flock.l_whence = SEEK_SET;
856 dat->fc_flock.l_start = 0;
857 dat->fc_flock.l_len = 0;
858 dat->fc_flock.l_type = F_UNLCK;
859 VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
860 F_UNLCK, &dat->fc_flock, F_POSIX);
861 }
862 break;
863
864 case F_GETLK:
865 if (fp->f_type != DTYPE_VNODE) {
866 error = EBADF;
867 break;
868 }
869 vp = (struct vnode *)fp->f_data;
870 /*
871 * copyin/lockop may block
872 */
873 if (dat->fc_flock.l_type != F_RDLCK &&
874 dat->fc_flock.l_type != F_WRLCK &&
875 dat->fc_flock.l_type != F_UNLCK) {
876 error = EINVAL;
877 break;
878 }
879 if (dat->fc_flock.l_whence == SEEK_CUR)
880 dat->fc_flock.l_start += fp->f_offset;
881 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
882 &dat->fc_flock, F_POSIX);
883 break;
884
885 case F_GETPATH:
886 if (fp->f_type != DTYPE_VNODE) {
887 error = EBADF;
888 break;
889 }
890
891 /*
892 * cache_fullpath() itself is limited to MAXPATHLEN so we
893 * do not need an explicit length check, but we do have
894 * to munge the error to ERANGE as per fcntl.2
895 */
896 error = cache_fullpath(p, &fp->f_nchandle, NULL,
897 &dat->fc_path.ptr, &dat->fc_path.buf, 1);
898 if (error == ENOMEM)
899 error = ERANGE;
900 break;
901
902 default:
903 error = EINVAL;
904 break;
905 }
906
907 fdrop(fp);
908 return (error);
909 }
910
911 /*
912 * The file control system call.
913 */
914 int
sys_fcntl(struct sysmsg * sysmsg,const struct fcntl_args * uap)915 sys_fcntl(struct sysmsg *sysmsg, const struct fcntl_args *uap)
916 {
917 union fcntl_dat dat;
918 int error;
919
920 switch (uap->cmd) {
921 case F_DUPFD:
922 case F_DUP2FD:
923 case F_DUPFD_CLOEXEC:
924 case F_DUP2FD_CLOEXEC:
925 dat.fc_fd = uap->arg;
926 break;
927 case F_SETFD:
928 dat.fc_cloexec = uap->arg;
929 break;
930 case F_SETFL:
931 dat.fc_flags = uap->arg;
932 break;
933 case F_SETOWN:
934 dat.fc_owner = uap->arg;
935 break;
936 case F_SETLKW:
937 case F_SETLK:
938 case F_GETLK:
939 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
940 sizeof(struct flock));
941 if (error)
942 return (error);
943 break;
944 }
945
946 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
947
948 if (error == 0) {
949 switch (uap->cmd) {
950 case F_DUPFD:
951 case F_DUP2FD:
952 case F_DUPFD_CLOEXEC:
953 case F_DUP2FD_CLOEXEC:
954 sysmsg->sysmsg_result = dat.fc_fd;
955 break;
956 case F_GETFD:
957 sysmsg->sysmsg_result = dat.fc_cloexec;
958 break;
959 case F_GETFL:
960 sysmsg->sysmsg_result = dat.fc_flags;
961 break;
962 case F_GETOWN:
963 sysmsg->sysmsg_result = dat.fc_owner;
964 break;
965 case F_GETLK:
966 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
967 sizeof(struct flock));
968 break;
969 case F_GETPATH:
970 error = copyout(dat.fc_path.ptr, (caddr_t)uap->arg,
971 strlen(dat.fc_path.ptr) + 1);
972 kfree(dat.fc_path.buf, M_TEMP);
973 break;
974 }
975 }
976
977 return (error);
978 }
979
980 /*
981 * Common code for dup, dup2, and fcntl(F_DUPFD).
982 *
983 * There are four type flags: DUP_FCNTL, DUP_FIXED, DUP_VARIABLE, and
984 * DUP_CLOEXEC.
985 *
986 * DUP_FCNTL is for handling EINVAL vs. EBADF differences between
987 * fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC and dup2() (per POSIX).
988 * The next two flags are mutually exclusive, and the fourth is optional.
989 * DUP_FIXED tells kern_dup() to destructively dup over an existing file
990 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup()
991 * to find the lowest unused file descriptor that is greater than or
992 * equal to "new". DUP_CLOEXEC, which works with either of the first
993 * two flags, sets the close-on-exec flag on the "new" file descriptor.
994 */
995 int
kern_dup(int flags,int old,int new,int * res)996 kern_dup(int flags, int old, int new, int *res)
997 {
998 struct thread *td = curthread;
999 struct proc *p = td->td_proc;
1000 struct plimit *limit = readplimits(p);
1001 struct filedesc *fdp = p->p_fd;
1002 struct file *fp;
1003 struct file *delfp;
1004 int oldflags;
1005 int holdleaders;
1006 int dtsize;
1007 int error, newfd;
1008
1009 /*
1010 * Verify that we have a valid descriptor to dup from and
1011 * possibly to dup to. When the new descriptor is out of
1012 * bounds, fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC must
1013 * return EINVAL, while dup2() returns EBADF in
1014 * this case.
1015 *
1016 * NOTE: maxfilesperuser is not applicable to dup()
1017 */
1018 retry:
1019 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1020 dtsize = INT_MAX;
1021 else
1022 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
1023 if (dtsize > maxfilesperproc)
1024 dtsize = maxfilesperproc;
1025 if (dtsize < minfilesperproc)
1026 dtsize = minfilesperproc;
1027
1028 if (new < 0 || new >= dtsize)
1029 return (flags & DUP_FCNTL ? EINVAL : EBADF);
1030
1031 spin_lock(&fdp->fd_spin);
1032 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
1033 spin_unlock(&fdp->fd_spin);
1034 return (EBADF);
1035 }
1036 if ((flags & DUP_FIXED) && old == new) {
1037 *res = new;
1038 if (flags & DUP_CLOEXEC)
1039 fdp->fd_files[new].fileflags |= UF_EXCLOSE;
1040 spin_unlock(&fdp->fd_spin);
1041 return (0);
1042 }
1043 fp = fdp->fd_files[old].fp;
1044 oldflags = fdp->fd_files[old].fileflags;
1045 fhold(fp);
1046
1047 /*
1048 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
1049 * if the requested descriptor is beyond the current table size.
1050 *
1051 * This can block. Retry if the source descriptor no longer matches
1052 * or if our expectation in the expansion case races.
1053 *
1054 * If we are not expanding or allocating a new decriptor, then reset
1055 * the target descriptor to a reserved state so we have a uniform
1056 * setup for the next code block.
1057 */
1058 if ((flags & DUP_VARIABLE) || new >= fdp->fd_nfiles) {
1059 error = fdalloc_locked(p, fdp, new, &newfd);
1060 if (error) {
1061 spin_unlock(&fdp->fd_spin);
1062 fdrop(fp);
1063 return (error);
1064 }
1065 /*
1066 * Check for ripout
1067 */
1068 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
1069 fsetfd_locked(fdp, NULL, newfd);
1070 spin_unlock(&fdp->fd_spin);
1071 fdrop(fp);
1072 goto retry;
1073 }
1074 /*
1075 * Check for expansion race
1076 */
1077 if ((flags & DUP_VARIABLE) == 0 && new != newfd) {
1078 fsetfd_locked(fdp, NULL, newfd);
1079 spin_unlock(&fdp->fd_spin);
1080 fdrop(fp);
1081 goto retry;
1082 }
1083 /*
1084 * Check for ripout, newfd reused old (this case probably
1085 * can't occur).
1086 */
1087 if (old == newfd) {
1088 fsetfd_locked(fdp, NULL, newfd);
1089 spin_unlock(&fdp->fd_spin);
1090 fdrop(fp);
1091 goto retry;
1092 }
1093 new = newfd;
1094 delfp = NULL;
1095 } else {
1096 if (fdp->fd_files[new].reserved) {
1097 spin_unlock(&fdp->fd_spin);
1098 fdrop(fp);
1099 kprintf("Warning: dup(): target descriptor %d is "
1100 "reserved, waiting for it to be resolved\n",
1101 new);
1102 tsleep(fdp, 0, "fdres", hz);
1103 goto retry;
1104 }
1105
1106 /*
1107 * If the target descriptor was never allocated we have
1108 * to allocate it. If it was we have to clean out the
1109 * old descriptor. delfp inherits the ref from the
1110 * descriptor table.
1111 */
1112 ++fdp->fd_closedcounter;
1113 fclearcache(&fdp->fd_files[new], NULL, 0);
1114 ++fdp->fd_closedcounter;
1115 delfp = fdp->fd_files[new].fp;
1116 fdp->fd_files[new].fp = NULL;
1117 fdp->fd_files[new].reserved = 1;
1118 if (delfp == NULL) {
1119 fdreserve_locked(fdp, new, 1);
1120 if (new > fdp->fd_lastfile)
1121 fdp->fd_lastfile = new;
1122 }
1123
1124 }
1125
1126 /*
1127 * NOTE: still holding an exclusive spinlock
1128 */
1129
1130 /*
1131 * If a descriptor is being overwritten we may hve to tell
1132 * fdfree() to sleep to ensure that all relevant process
1133 * leaders can be traversed in closef().
1134 */
1135 if (delfp != NULL && p->p_fdtol != NULL) {
1136 fdp->fd_holdleaderscount++;
1137 holdleaders = 1;
1138 } else {
1139 holdleaders = 0;
1140 }
1141 KASSERT(delfp == NULL || (flags & DUP_FIXED),
1142 ("dup() picked an open file"));
1143
1144 /*
1145 * Duplicate the source descriptor, update lastfile. If the new
1146 * descriptor was not allocated and we aren't replacing an existing
1147 * descriptor we have to mark the descriptor as being in use.
1148 *
1149 * The fd_files[] array inherits fp's hold reference.
1150 */
1151 fsetfd_locked(fdp, fp, new);
1152 if ((flags & DUP_CLOEXEC) != 0)
1153 fdp->fd_files[new].fileflags = oldflags | UF_EXCLOSE;
1154 else
1155 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
1156 spin_unlock(&fdp->fd_spin);
1157 fdrop(fp);
1158 *res = new;
1159
1160 /*
1161 * If we dup'd over a valid file, we now own the reference to it
1162 * and must dispose of it using closef() semantics (as if a
1163 * close() were performed on it).
1164 */
1165 if (delfp) {
1166 if (SLIST_FIRST(&delfp->f_klist))
1167 knote_fdclose(delfp, fdp, new);
1168 closef(delfp, p);
1169 if (holdleaders) {
1170 spin_lock(&fdp->fd_spin);
1171 fdp->fd_holdleaderscount--;
1172 if (fdp->fd_holdleaderscount == 0 &&
1173 fdp->fd_holdleaderswakeup != 0) {
1174 fdp->fd_holdleaderswakeup = 0;
1175 spin_unlock(&fdp->fd_spin);
1176 wakeup(&fdp->fd_holdleaderscount);
1177 } else {
1178 spin_unlock(&fdp->fd_spin);
1179 }
1180 }
1181 }
1182 return (0);
1183 }
1184
1185 /*
1186 * If sigio is on the list associated with a process or process group,
1187 * disable signalling from the device, remove sigio from the list and
1188 * free sigio.
1189 */
1190 void
funsetown(struct sigio ** sigiop)1191 funsetown(struct sigio **sigiop)
1192 {
1193 struct pgrp *pgrp;
1194 struct proc *p;
1195 struct sigio *sigio;
1196
1197 if ((sigio = *sigiop) != NULL) {
1198 lwkt_gettoken(&sigio_token); /* protect sigio */
1199 KKASSERT(sigiop == sigio->sio_myref);
1200 sigio = *sigiop;
1201 *sigiop = NULL;
1202 lwkt_reltoken(&sigio_token);
1203 }
1204 if (sigio == NULL)
1205 return;
1206
1207 if (sigio->sio_pgid < 0) {
1208 pgrp = sigio->sio_pgrp;
1209 sigio->sio_pgrp = NULL;
1210 lwkt_gettoken(&pgrp->pg_token);
1211 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio);
1212 lwkt_reltoken(&pgrp->pg_token);
1213 pgrel(pgrp);
1214 } else /* if ((*sigiop)->sio_pgid > 0) */ {
1215 p = sigio->sio_proc;
1216 sigio->sio_proc = NULL;
1217 PHOLD(p);
1218 lwkt_gettoken(&p->p_token);
1219 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1220 lwkt_reltoken(&p->p_token);
1221 PRELE(p);
1222 }
1223 crfree(sigio->sio_ucred);
1224 sigio->sio_ucred = NULL;
1225 kfree(sigio, M_SIGIO);
1226 }
1227
1228 /*
1229 * Free a list of sigio structures. Caller is responsible for ensuring
1230 * that the list is MPSAFE.
1231 */
1232 void
funsetownlst(struct sigiolst * sigiolst)1233 funsetownlst(struct sigiolst *sigiolst)
1234 {
1235 struct sigio *sigio;
1236
1237 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
1238 funsetown(sigio->sio_myref);
1239 }
1240
1241 /*
1242 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1243 *
1244 * After permission checking, add a sigio structure to the sigio list for
1245 * the process or process group.
1246 */
1247 int
fsetown(pid_t pgid,struct sigio ** sigiop)1248 fsetown(pid_t pgid, struct sigio **sigiop)
1249 {
1250 struct proc *proc = NULL;
1251 struct pgrp *pgrp = NULL;
1252 struct sigio *sigio;
1253 int error;
1254
1255 if (pgid == 0) {
1256 funsetown(sigiop);
1257 return (0);
1258 }
1259
1260 if (pgid > 0) {
1261 proc = pfind(pgid);
1262 if (proc == NULL) {
1263 error = ESRCH;
1264 goto done;
1265 }
1266
1267 /*
1268 * Policy - Don't allow a process to FSETOWN a process
1269 * in another session.
1270 *
1271 * Remove this test to allow maximum flexibility or
1272 * restrict FSETOWN to the current process or process
1273 * group for maximum safety.
1274 */
1275 if (proc->p_session != curproc->p_session) {
1276 error = EPERM;
1277 goto done;
1278 }
1279 } else /* if (pgid < 0) */ {
1280 pgrp = pgfind(-pgid);
1281 if (pgrp == NULL) {
1282 error = ESRCH;
1283 goto done;
1284 }
1285
1286 /*
1287 * Policy - Don't allow a process to FSETOWN a process
1288 * in another session.
1289 *
1290 * Remove this test to allow maximum flexibility or
1291 * restrict FSETOWN to the current process or process
1292 * group for maximum safety.
1293 */
1294 if (pgrp->pg_session != curproc->p_session) {
1295 error = EPERM;
1296 goto done;
1297 }
1298 }
1299 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO);
1300 if (pgid > 0) {
1301 KKASSERT(pgrp == NULL);
1302 lwkt_gettoken(&proc->p_token);
1303 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1304 sigio->sio_proc = proc;
1305 lwkt_reltoken(&proc->p_token);
1306 } else {
1307 KKASSERT(proc == NULL);
1308 lwkt_gettoken(&pgrp->pg_token);
1309 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1310 sigio->sio_pgrp = pgrp;
1311 lwkt_reltoken(&pgrp->pg_token);
1312 pgrp = NULL;
1313 }
1314 sigio->sio_pgid = pgid;
1315 sigio->sio_ucred = crhold(curthread->td_ucred);
1316 /* It would be convenient if p_ruid was in ucred. */
1317 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
1318 sigio->sio_myref = sigiop;
1319
1320 lwkt_gettoken(&sigio_token);
1321 while (*sigiop)
1322 funsetown(sigiop);
1323 *sigiop = sigio;
1324 lwkt_reltoken(&sigio_token);
1325 error = 0;
1326 done:
1327 if (pgrp)
1328 pgrel(pgrp);
1329 if (proc)
1330 PRELE(proc);
1331 return (error);
1332 }
1333
1334 /*
1335 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1336 */
1337 pid_t
fgetown(struct sigio ** sigiop)1338 fgetown(struct sigio **sigiop)
1339 {
1340 struct sigio *sigio;
1341 pid_t own;
1342
1343 lwkt_gettoken_shared(&sigio_token);
1344 sigio = *sigiop;
1345 own = (sigio != NULL ? sigio->sio_pgid : 0);
1346 lwkt_reltoken(&sigio_token);
1347
1348 return (own);
1349 }
1350
1351 /*
1352 * Close many file descriptors.
1353 */
1354 int
sys_closefrom(struct sysmsg * sysmsg,const struct closefrom_args * uap)1355 sys_closefrom(struct sysmsg *sysmsg, const struct closefrom_args *uap)
1356 {
1357 return(kern_closefrom(uap->fd));
1358 }
1359
1360 /*
1361 * Close all file descriptors greater then or equal to fd
1362 */
1363 int
kern_closefrom(int fd)1364 kern_closefrom(int fd)
1365 {
1366 struct thread *td = curthread;
1367 struct proc *p = td->td_proc;
1368 struct filedesc *fdp;
1369 int error;
1370 int e2;
1371
1372 KKASSERT(p);
1373 fdp = p->p_fd;
1374
1375 if (fd < 0)
1376 return (EINVAL);
1377
1378 /*
1379 * NOTE: This function will skip unassociated descriptors and
1380 * reserved descriptors that have not yet been assigned.
1381 * fd_lastfile can change as a side effect of kern_close().
1382 *
1383 * NOTE: We accumulate EINTR errors and return EINTR if any
1384 * close() returned EINTR. However, the descriptor is
1385 * still closed and we do not break out of the loop.
1386 */
1387 error = 0;
1388 spin_lock(&fdp->fd_spin);
1389 while (fd <= fdp->fd_lastfile) {
1390 if (fdp->fd_files[fd].fp != NULL) {
1391 spin_unlock(&fdp->fd_spin);
1392 /* ok if this races another close */
1393 e2 = kern_close(fd);
1394 if (e2 == EINTR)
1395 error = EINTR;
1396 spin_lock(&fdp->fd_spin);
1397 }
1398 ++fd;
1399 }
1400 spin_unlock(&fdp->fd_spin);
1401
1402 return error;
1403 }
1404
1405 /*
1406 * Close a file descriptor.
1407 */
1408 int
sys_close(struct sysmsg * sysmsg,const struct close_args * uap)1409 sys_close(struct sysmsg *sysmsg, const struct close_args *uap)
1410 {
1411 return(kern_close(uap->fd));
1412 }
1413
1414 /*
1415 * close() helper
1416 */
1417 int
kern_close(int fd)1418 kern_close(int fd)
1419 {
1420 struct thread *td = curthread;
1421 struct proc *p = td->td_proc;
1422 struct filedesc *fdp;
1423 struct file *fp;
1424 int error;
1425 int holdleaders;
1426
1427 KKASSERT(p);
1428 fdp = p->p_fd;
1429
1430 /*
1431 * funsetfd*() also clears the fd cache
1432 */
1433 spin_lock(&fdp->fd_spin);
1434 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
1435 spin_unlock(&fdp->fd_spin);
1436 return (EBADF);
1437 }
1438 holdleaders = 0;
1439 if (p->p_fdtol != NULL) {
1440 /*
1441 * Ask fdfree() to sleep to ensure that all relevant
1442 * process leaders can be traversed in closef().
1443 */
1444 fdp->fd_holdleaderscount++;
1445 holdleaders = 1;
1446 }
1447
1448 /*
1449 * we now hold the fp reference that used to be owned by the descriptor
1450 * array.
1451 */
1452 spin_unlock(&fdp->fd_spin);
1453 if (SLIST_FIRST(&fp->f_klist))
1454 knote_fdclose(fp, fdp, fd);
1455 error = closef(fp, p);
1456 if (holdleaders) {
1457 spin_lock(&fdp->fd_spin);
1458 fdp->fd_holdleaderscount--;
1459 if (fdp->fd_holdleaderscount == 0 &&
1460 fdp->fd_holdleaderswakeup != 0) {
1461 fdp->fd_holdleaderswakeup = 0;
1462 spin_unlock(&fdp->fd_spin);
1463 wakeup(&fdp->fd_holdleaderscount);
1464 } else {
1465 spin_unlock(&fdp->fd_spin);
1466 }
1467 }
1468 return (error);
1469 }
1470
1471 /*
1472 * shutdown_args(int fd, int how)
1473 */
1474 int
kern_shutdown(int fd,int how)1475 kern_shutdown(int fd, int how)
1476 {
1477 struct thread *td = curthread;
1478 struct file *fp;
1479 int error;
1480
1481 if ((fp = holdfp(td, fd, -1)) == NULL)
1482 return (EBADF);
1483 error = fo_shutdown(fp, how);
1484 fdrop(fp);
1485
1486 return (error);
1487 }
1488
1489 /*
1490 * MPALMOSTSAFE
1491 */
1492 int
sys_shutdown(struct sysmsg * sysmsg,const struct shutdown_args * uap)1493 sys_shutdown(struct sysmsg *sysmsg, const struct shutdown_args *uap)
1494 {
1495 int error;
1496
1497 error = kern_shutdown(uap->s, uap->how);
1498
1499 return (error);
1500 }
1501
1502 /*
1503 * fstat() helper
1504 */
1505 int
kern_fstat(int fd,struct stat * ub)1506 kern_fstat(int fd, struct stat *ub)
1507 {
1508 struct thread *td = curthread;
1509 struct file *fp;
1510 int error;
1511
1512 if ((fp = holdfp(td, fd, -1)) == NULL)
1513 return (EBADF);
1514 error = fo_stat(fp, ub, td->td_ucred);
1515 fdrop(fp);
1516
1517 return (error);
1518 }
1519
1520 /*
1521 * Return status information about a file descriptor.
1522 */
1523 int
sys_fstat(struct sysmsg * sysmsg,const struct fstat_args * uap)1524 sys_fstat(struct sysmsg *sysmsg, const struct fstat_args *uap)
1525 {
1526 struct stat st;
1527 int error;
1528
1529 error = kern_fstat(uap->fd, &st);
1530
1531 if (error == 0)
1532 error = copyout(&st, uap->sb, sizeof(st));
1533 return (error);
1534 }
1535
1536 /*
1537 * Return pathconf information about a file descriptor.
1538 *
1539 * MPALMOSTSAFE
1540 */
1541 int
sys_fpathconf(struct sysmsg * sysmsg,const struct fpathconf_args * uap)1542 sys_fpathconf(struct sysmsg *sysmsg, const struct fpathconf_args *uap)
1543 {
1544 struct thread *td = curthread;
1545 struct file *fp;
1546 struct vnode *vp;
1547 int error = 0;
1548
1549 if ((fp = holdfp(td, uap->fd, -1)) == NULL)
1550 return (EBADF);
1551
1552 switch (fp->f_type) {
1553 case DTYPE_PIPE:
1554 case DTYPE_SOCKET:
1555 if (uap->name != _PC_PIPE_BUF) {
1556 error = EINVAL;
1557 } else {
1558 sysmsg->sysmsg_result = PIPE_BUF;
1559 error = 0;
1560 }
1561 break;
1562 case DTYPE_FIFO:
1563 case DTYPE_VNODE:
1564 vp = (struct vnode *)fp->f_data;
1565 error = VOP_PATHCONF(vp, uap->name, &sysmsg->sysmsg_reg);
1566 break;
1567 default:
1568 error = EOPNOTSUPP;
1569 break;
1570 }
1571 fdrop(fp);
1572 return(error);
1573 }
1574
1575 /*
1576 * Grow the file table so it can hold through descriptor (want).
1577 *
1578 * The fdp's spinlock must be held exclusively on entry and may be held
1579 * exclusively on return. The spinlock may be cycled by the routine.
1580 */
1581 static void
fdgrow_locked(struct filedesc * fdp,int want)1582 fdgrow_locked(struct filedesc *fdp, int want)
1583 {
1584 struct fdnode *newfiles;
1585 struct fdnode *oldfiles;
1586 int nf, extra;
1587
1588 nf = fdp->fd_nfiles;
1589 do {
1590 /* nf has to be of the form 2^n - 1 */
1591 nf = 2 * nf + 1;
1592 } while (nf <= want);
1593
1594 spin_unlock(&fdp->fd_spin);
1595 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
1596 spin_lock(&fdp->fd_spin);
1597
1598 /*
1599 * We could have raced another extend while we were not holding
1600 * the spinlock.
1601 */
1602 if (fdp->fd_nfiles >= nf) {
1603 spin_unlock(&fdp->fd_spin);
1604 kfree(newfiles, M_FILEDESC);
1605 spin_lock(&fdp->fd_spin);
1606 return;
1607 }
1608 /*
1609 * Copy the existing ofile and ofileflags arrays
1610 * and zero the new portion of each array.
1611 */
1612 extra = nf - fdp->fd_nfiles;
1613 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1614 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1615
1616 oldfiles = fdp->fd_files;
1617 fdp->fd_files = newfiles;
1618 fdp->fd_nfiles = nf;
1619
1620 if (oldfiles != fdp->fd_builtin_files) {
1621 spin_unlock(&fdp->fd_spin);
1622 kfree(oldfiles, M_FILEDESC);
1623 spin_lock(&fdp->fd_spin);
1624 }
1625 }
1626
1627 /*
1628 * Number of nodes in right subtree, including the root.
1629 */
1630 static __inline int
right_subtree_size(int n)1631 right_subtree_size(int n)
1632 {
1633 return (n ^ (n | (n + 1)));
1634 }
1635
1636 /*
1637 * Bigger ancestor.
1638 */
1639 static __inline int
right_ancestor(int n)1640 right_ancestor(int n)
1641 {
1642 return (n | (n + 1));
1643 }
1644
1645 /*
1646 * Smaller ancestor.
1647 */
1648 static __inline int
left_ancestor(int n)1649 left_ancestor(int n)
1650 {
1651 return ((n & (n + 1)) - 1);
1652 }
1653
1654 /*
1655 * Traverse the in-place binary tree buttom-up adjusting the allocation
1656 * count so scans can determine where free descriptors are located.
1657 *
1658 * caller must be holding an exclusive spinlock on fdp
1659 */
1660 static
1661 void
fdreserve_locked(struct filedesc * fdp,int fd,int incr)1662 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1663 {
1664 while (fd >= 0) {
1665 fdp->fd_files[fd].allocated += incr;
1666 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1667 fd = left_ancestor(fd);
1668 }
1669 }
1670
1671 /*
1672 * Reserve a file descriptor for the process. If no error occurs, the
1673 * caller MUST at some point call fsetfd() or assign a file pointer
1674 * or dispose of the reservation.
1675 */
1676 static
1677 int
fdalloc_locked(struct proc * p,struct filedesc * fdp,int want,int * result)1678 fdalloc_locked(struct proc *p, struct filedesc *fdp, int want, int *result)
1679 {
1680 struct plimit *limit = readplimits(p);
1681 struct uidinfo *uip;
1682 int fd, rsize, rsum, node, lim;
1683
1684 /*
1685 * Check dtable size limit
1686 */
1687 *result = -1; /* avoid gcc warnings */
1688 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1689 lim = INT_MAX;
1690 else
1691 lim = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
1692
1693 if (lim > maxfilesperproc)
1694 lim = maxfilesperproc;
1695 if (lim < minfilesperproc)
1696 lim = minfilesperproc;
1697 if (want >= lim)
1698 return (EINVAL);
1699
1700 /*
1701 * Check that the user has not run out of descriptors (non-root only).
1702 * As a safety measure the dtable is allowed to have at least
1703 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1704 *
1705 * This isn't as loose a spec as ui_posixlocks, so we use atomic
1706 * ops to force synchronize and recheck if we would otherwise
1707 * error.
1708 */
1709 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1710 uip = p->p_ucred->cr_uidinfo;
1711 if (uip->ui_openfiles > maxfilesperuser) {
1712 int n;
1713 int count;
1714
1715 count = 0;
1716 for (n = 0; n < ncpus; ++n) {
1717 count += atomic_swap_int(
1718 &uip->ui_pcpu[n].pu_openfiles, 0);
1719 }
1720 atomic_add_int(&uip->ui_openfiles, count);
1721 if (uip->ui_openfiles > maxfilesperuser) {
1722 krateprintf(&krate_uidinfo,
1723 "Warning: user %d pid %d (%s) "
1724 "ran out of file descriptors "
1725 "(%d/%d)\n",
1726 p->p_ucred->cr_uid, (int)p->p_pid,
1727 p->p_comm,
1728 uip->ui_openfiles, maxfilesperuser);
1729 return(ENFILE);
1730 }
1731 }
1732 }
1733
1734 /*
1735 * Grow the dtable if necessary
1736 */
1737 if (want >= fdp->fd_nfiles)
1738 fdgrow_locked(fdp, want);
1739
1740 /*
1741 * Search for a free descriptor starting at the higher
1742 * of want or fd_freefile. If that fails, consider
1743 * expanding the ofile array.
1744 *
1745 * NOTE! the 'allocated' field is a cumulative recursive allocation
1746 * count. If we happen to see a value of 0 then we can shortcut
1747 * our search. Otherwise we run through through the tree going
1748 * down branches we know have free descriptor(s) until we hit a
1749 * leaf node. The leaf node will be free but will not necessarily
1750 * have an allocated field of 0.
1751 */
1752 retry:
1753 /* move up the tree looking for a subtree with a free node */
1754 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1755 fd = right_ancestor(fd)) {
1756 if (fdp->fd_files[fd].allocated == 0)
1757 goto found;
1758
1759 rsize = right_subtree_size(fd);
1760 if (fdp->fd_files[fd].allocated == rsize)
1761 continue; /* right subtree full */
1762
1763 /*
1764 * Free fd is in the right subtree of the tree rooted at fd.
1765 * Call that subtree R. Look for the smallest (leftmost)
1766 * subtree of R with an unallocated fd: continue moving
1767 * down the left branch until encountering a full left
1768 * subtree, then move to the right.
1769 */
1770 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1771 node = fd + rsize;
1772 rsum += fdp->fd_files[node].allocated;
1773 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1774 fd = node; /* move to the right */
1775 if (fdp->fd_files[node].allocated == 0)
1776 goto found;
1777 rsum = 0;
1778 }
1779 }
1780 goto found;
1781 }
1782
1783 /*
1784 * No space in current array. Expand?
1785 */
1786 if (fdp->fd_nfiles >= lim) {
1787 return (EMFILE);
1788 }
1789 fdgrow_locked(fdp, want);
1790 goto retry;
1791
1792 found:
1793 KKASSERT(fd < fdp->fd_nfiles);
1794 if (fd > fdp->fd_lastfile)
1795 fdp->fd_lastfile = fd;
1796 if (want <= fdp->fd_freefile)
1797 fdp->fd_freefile = fd;
1798 *result = fd;
1799 KKASSERT(fdp->fd_files[fd].fp == NULL);
1800 KKASSERT(fdp->fd_files[fd].reserved == 0);
1801 fdp->fd_files[fd].fileflags = 0;
1802 fdp->fd_files[fd].reserved = 1;
1803 fdreserve_locked(fdp, fd, 1);
1804
1805 return (0);
1806 }
1807
1808 int
fdalloc(struct proc * p,int want,int * result)1809 fdalloc(struct proc *p, int want, int *result)
1810 {
1811 struct filedesc *fdp = p->p_fd;
1812 int error;
1813
1814 spin_lock(&fdp->fd_spin);
1815 error = fdalloc_locked(p, fdp, want, result);
1816 spin_unlock(&fdp->fd_spin);
1817
1818 return error;
1819 }
1820
1821 /*
1822 * Check to see whether n user file descriptors
1823 * are available to the process p.
1824 */
1825 int
fdavail(struct proc * p,int n)1826 fdavail(struct proc *p, int n)
1827 {
1828 struct plimit *limit = readplimits(p);
1829 struct filedesc *fdp = p->p_fd;
1830 struct fdnode *fdnode;
1831 int i, lim, last;
1832
1833 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1834 lim = INT_MAX;
1835 else
1836 lim = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
1837
1838 if (lim > maxfilesperproc)
1839 lim = maxfilesperproc;
1840 if (lim < minfilesperproc)
1841 lim = minfilesperproc;
1842
1843 spin_lock(&fdp->fd_spin);
1844 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1845 spin_unlock(&fdp->fd_spin);
1846 return (1);
1847 }
1848 last = min(fdp->fd_nfiles, lim);
1849 fdnode = &fdp->fd_files[fdp->fd_freefile];
1850 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1851 if (fdnode->fp == NULL && --n <= 0) {
1852 spin_unlock(&fdp->fd_spin);
1853 return (1);
1854 }
1855 }
1856 spin_unlock(&fdp->fd_spin);
1857 return (0);
1858 }
1859
1860 /*
1861 * Revoke open descriptors referencing (f_data, f_type)
1862 *
1863 * Any revoke executed within a prison is only able to
1864 * revoke descriptors for processes within that prison.
1865 *
1866 * Returns 0 on success or an error code.
1867 */
1868 struct fdrevoke_info {
1869 void *data;
1870 short type;
1871 short unused;
1872 int found;
1873 struct ucred *cred;
1874 struct file *nfp;
1875 };
1876
1877 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1878 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1879
1880 int
fdrevoke(void * f_data,short f_type,struct ucred * cred)1881 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1882 {
1883 struct fdrevoke_info info;
1884 int error;
1885
1886 bzero(&info, sizeof(info));
1887 info.data = f_data;
1888 info.type = f_type;
1889 info.cred = cred;
1890 error = falloc(NULL, &info.nfp, NULL);
1891 if (error)
1892 return (error);
1893
1894 /*
1895 * Scan the file pointer table once. dups do not dup file pointers,
1896 * only descriptors, so there is no leak. Set FREVOKED on the fps
1897 * being revoked.
1898 *
1899 * Any fps sent over unix-domain sockets will be revoked by the
1900 * socket code checking for FREVOKED when the fps are externialized.
1901 * revoke_token is used to make sure that fps marked FREVOKED and
1902 * externalized will be picked up by the following allproc_scan().
1903 */
1904 lwkt_gettoken(&revoke_token);
1905 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1906 lwkt_reltoken(&revoke_token);
1907
1908 /*
1909 * If any fps were marked track down the related descriptors
1910 * and close them. Any dup()s at this point will notice
1911 * the FREVOKED already set in the fp and do the right thing.
1912 */
1913 if (info.found)
1914 allproc_scan(fdrevoke_proc_callback, &info, 0);
1915 fdrop(info.nfp);
1916 return(0);
1917 }
1918
1919 /*
1920 * Locate matching file pointers directly.
1921 *
1922 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1923 */
1924 static int
fdrevoke_check_callback(struct file * fp,void * vinfo)1925 fdrevoke_check_callback(struct file *fp, void *vinfo)
1926 {
1927 struct fdrevoke_info *info = vinfo;
1928
1929 /*
1930 * File pointers already flagged for revokation are skipped.
1931 */
1932 if (fp->f_flag & FREVOKED)
1933 return(0);
1934
1935 /*
1936 * If revoking from a prison file pointers created outside of
1937 * that prison, or file pointers without creds, cannot be revoked.
1938 */
1939 if (info->cred->cr_prison &&
1940 (fp->f_cred == NULL ||
1941 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1942 return(0);
1943 }
1944
1945 /*
1946 * If the file pointer matches then mark it for revocation. The
1947 * flag is currently only used by unp_revoke_gc().
1948 *
1949 * info->found is a heuristic and can race in a SMP environment.
1950 */
1951 if (info->data == fp->f_data && info->type == fp->f_type) {
1952 atomic_set_int(&fp->f_flag, FREVOKED);
1953 info->found = 1;
1954 }
1955 return(0);
1956 }
1957
1958 /*
1959 * Locate matching file pointers via process descriptor tables.
1960 */
1961 static int
fdrevoke_proc_callback(struct proc * p,void * vinfo)1962 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1963 {
1964 struct fdrevoke_info *info = vinfo;
1965 struct filedesc *fdp;
1966 struct file *fp;
1967 int n;
1968
1969 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1970 return(0);
1971 if (info->cred->cr_prison &&
1972 info->cred->cr_prison != p->p_ucred->cr_prison) {
1973 return(0);
1974 }
1975
1976 /*
1977 * If the controlling terminal of the process matches the
1978 * vnode being revoked we clear the controlling terminal.
1979 *
1980 * The normal spec_close() may not catch this because it
1981 * uses curproc instead of p.
1982 */
1983 if (p->p_session && info->type == DTYPE_VNODE &&
1984 info->data == p->p_session->s_ttyvp) {
1985 p->p_session->s_ttyvp = NULL;
1986 vrele(info->data);
1987 }
1988
1989 /*
1990 * Softref the fdp to prevent it from being destroyed
1991 */
1992 spin_lock(&p->p_spin);
1993 if ((fdp = p->p_fd) == NULL) {
1994 spin_unlock(&p->p_spin);
1995 return(0);
1996 }
1997 atomic_add_int(&fdp->fd_softrefs, 1);
1998 spin_unlock(&p->p_spin);
1999
2000 /*
2001 * Locate and close any matching file descriptors, replacing
2002 * them with info->nfp.
2003 */
2004 spin_lock(&fdp->fd_spin);
2005 for (n = 0; n < fdp->fd_nfiles; ++n) {
2006 if ((fp = fdp->fd_files[n].fp) == NULL)
2007 continue;
2008 if (fp->f_flag & FREVOKED) {
2009 ++fdp->fd_closedcounter;
2010 fclearcache(&fdp->fd_files[n], NULL, 0);
2011 ++fdp->fd_closedcounter;
2012 fhold(info->nfp);
2013 fdp->fd_files[n].fp = info->nfp;
2014 spin_unlock(&fdp->fd_spin);
2015 knote_fdclose(fp, fdp, n); /* XXX */
2016 closef(fp, p);
2017 spin_lock(&fdp->fd_spin);
2018 }
2019 }
2020 spin_unlock(&fdp->fd_spin);
2021 atomic_subtract_int(&fdp->fd_softrefs, 1);
2022 return(0);
2023 }
2024
2025 /*
2026 * falloc:
2027 * Create a new open file structure and reserve a file decriptor
2028 * for the process that refers to it.
2029 *
2030 * Root creds are checked using lp, or assumed if lp is NULL. If
2031 * resultfd is non-NULL then lp must also be non-NULL. No file
2032 * descriptor is reserved (and no process context is needed) if
2033 * resultfd is NULL.
2034 *
2035 * A file pointer with a refcount of 1 is returned. Note that the
2036 * file pointer is NOT associated with the descriptor. If falloc
2037 * returns success, fsetfd() MUST be called to either associate the
2038 * file pointer or clear the reservation.
2039 */
2040 int
falloc(struct lwp * lp,struct file ** resultfp,int * resultfd)2041 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
2042 {
2043 static struct timeval lastfail;
2044 static int curfail;
2045 struct filelist_head *head;
2046 struct file *fp;
2047 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
2048 int error;
2049
2050 fp = NULL;
2051
2052 /*
2053 * Handle filetable full issues and root overfill.
2054 */
2055 if (nfiles >= maxfiles - maxfilesrootres &&
2056 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
2057 if (ppsratecheck(&lastfail, &curfail, 1)) {
2058 kprintf("kern.maxfiles limit exceeded by uid %d, "
2059 "please see tuning(7).\n",
2060 cred->cr_ruid);
2061 }
2062 error = ENFILE;
2063 goto done;
2064 }
2065
2066 /*
2067 * Allocate a new file descriptor.
2068 */
2069 fp = kmalloc_obj(sizeof(*fp), M_FILE, M_WAITOK|M_ZERO);
2070 spin_init(&fp->f_spin, "falloc");
2071 SLIST_INIT(&fp->f_klist);
2072 fp->f_count = 1;
2073 fp->f_ops = &badfileops;
2074 fp->f_seqcount = 1;
2075 fsetcred(fp, cred);
2076 atomic_add_int(&nfiles, 1);
2077
2078 head = fp2filelist(fp);
2079 spin_lock(&head->spin);
2080 LIST_INSERT_HEAD(&head->list, fp, f_list);
2081 spin_unlock(&head->spin);
2082
2083 if (resultfd) {
2084 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
2085 fdrop(fp);
2086 fp = NULL;
2087 }
2088 } else {
2089 error = 0;
2090 }
2091 done:
2092 *resultfp = fp;
2093 return (error);
2094 }
2095
2096 /*
2097 * Check for races against a file descriptor by determining that the
2098 * file pointer is still associated with the specified file descriptor,
2099 * and a close is not currently in progress.
2100 */
2101 int
checkfdclosed(thread_t td,struct filedesc * fdp,int fd,struct file * fp,int closedcounter)2102 checkfdclosed(thread_t td, struct filedesc *fdp, int fd, struct file *fp,
2103 int closedcounter)
2104 {
2105 struct fdcache *fdc;
2106 int error;
2107
2108 cpu_lfence();
2109 if (fdp->fd_closedcounter == closedcounter)
2110 return 0;
2111
2112 if (td->td_proc && td->td_proc->p_fd == fdp) {
2113 for (fdc = &td->td_fdcache[0];
2114 fdc < &td->td_fdcache[NFDCACHE]; ++fdc) {
2115 if (fdc->fd == fd && fdc->fp == fp)
2116 return 0;
2117 }
2118 }
2119
2120 spin_lock_shared(&fdp->fd_spin);
2121 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
2122 error = EBADF;
2123 else
2124 error = 0;
2125 spin_unlock_shared(&fdp->fd_spin);
2126 return (error);
2127 }
2128
2129 /*
2130 * Associate a file pointer with a previously reserved file descriptor.
2131 * This function always succeeds.
2132 *
2133 * If fp is NULL, the file descriptor is returned to the pool.
2134 *
2135 * Caller must hold an exclusive spinlock on fdp->fd_spin.
2136 */
2137 static void
fsetfd_locked(struct filedesc * fdp,struct file * fp,int fd)2138 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
2139 {
2140 KKASSERT((unsigned)fd < fdp->fd_nfiles);
2141 KKASSERT(fdp->fd_files[fd].reserved != 0);
2142 if (fp) {
2143 fhold(fp);
2144 /* fclearcache(&fdp->fd_files[fd], NULL, 0); */
2145 fdp->fd_files[fd].fp = fp;
2146 fdp->fd_files[fd].reserved = 0;
2147 } else {
2148 fdp->fd_files[fd].reserved = 0;
2149 fdreserve_locked(fdp, fd, -1);
2150 fdfixup_locked(fdp, fd);
2151 }
2152 }
2153
2154 /*
2155 * Caller must hold an exclusive spinlock on fdp->fd_spin.
2156 */
2157 void
fsetfd(struct filedesc * fdp,struct file * fp,int fd)2158 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
2159 {
2160 spin_lock(&fdp->fd_spin);
2161 fsetfd_locked(fdp, fp, fd);
2162 spin_unlock(&fdp->fd_spin);
2163 }
2164
2165 /*
2166 * Caller must hold an exclusive spinlock on fdp->fd_spin.
2167 */
2168 static
2169 struct file *
funsetfd_locked(struct filedesc * fdp,int fd)2170 funsetfd_locked(struct filedesc *fdp, int fd)
2171 {
2172 struct file *fp;
2173
2174 if ((unsigned)fd >= fdp->fd_nfiles)
2175 return (NULL);
2176 if ((fp = fdp->fd_files[fd].fp) == NULL)
2177 return (NULL);
2178 ++fdp->fd_closedcounter;
2179 fclearcache(&fdp->fd_files[fd], NULL, 0);
2180 fdp->fd_files[fd].fp = NULL;
2181 fdp->fd_files[fd].fileflags = 0;
2182 ++fdp->fd_closedcounter;
2183
2184 fdreserve_locked(fdp, fd, -1);
2185 fdfixup_locked(fdp, fd);
2186
2187 return(fp);
2188 }
2189
2190 /*
2191 * WARNING: May not be called before initial fsetfd().
2192 */
2193 int
fgetfdflags(struct filedesc * fdp,int fd,int * flagsp)2194 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
2195 {
2196 int error;
2197
2198 spin_lock_shared(&fdp->fd_spin);
2199 if (((u_int)fd) >= fdp->fd_nfiles) {
2200 error = EBADF;
2201 } else if (fdp->fd_files[fd].fp == NULL) {
2202 error = EBADF;
2203 } else {
2204 *flagsp = fdp->fd_files[fd].fileflags;
2205 error = 0;
2206 }
2207 spin_unlock_shared(&fdp->fd_spin);
2208
2209 return (error);
2210 }
2211
2212 /*
2213 * WARNING: May not be called before initial fsetfd().
2214 */
2215 int
fsetfdflags(struct filedesc * fdp,int fd,int add_flags)2216 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
2217 {
2218 int error;
2219
2220 spin_lock(&fdp->fd_spin);
2221 if (((u_int)fd) >= fdp->fd_nfiles) {
2222 error = EBADF;
2223 } else if (fdp->fd_files[fd].fp == NULL) {
2224 error = EBADF;
2225 } else {
2226 fdp->fd_files[fd].fileflags |= add_flags;
2227 error = 0;
2228 }
2229 spin_unlock(&fdp->fd_spin);
2230
2231 return (error);
2232 }
2233
2234 /*
2235 * WARNING: May not be called before initial fsetfd().
2236 */
2237 int
fclrfdflags(struct filedesc * fdp,int fd,int rem_flags)2238 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
2239 {
2240 int error;
2241
2242 spin_lock(&fdp->fd_spin);
2243 if (((u_int)fd) >= fdp->fd_nfiles) {
2244 error = EBADF;
2245 } else if (fdp->fd_files[fd].fp == NULL) {
2246 error = EBADF;
2247 } else {
2248 fdp->fd_files[fd].fileflags &= ~rem_flags;
2249 error = 0;
2250 }
2251 spin_unlock(&fdp->fd_spin);
2252
2253 return (error);
2254 }
2255
2256 /*
2257 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
2258 */
2259 void
fsetcred(struct file * fp,struct ucred * ncr)2260 fsetcred(struct file *fp, struct ucred *ncr)
2261 {
2262 struct ucred *ocr;
2263 struct uidinfo *uip;
2264 struct uidcount *pup;
2265 int cpu = mycpuid;
2266 int count;
2267
2268 ocr = fp->f_cred;
2269 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
2270 if (ocr) {
2271 uip = ocr->cr_uidinfo;
2272 pup = &uip->ui_pcpu[cpu];
2273 atomic_add_int(&pup->pu_openfiles, -1);
2274 if (pup->pu_openfiles < -PUP_LIMIT ||
2275 pup->pu_openfiles > PUP_LIMIT) {
2276 count = atomic_swap_int(&pup->pu_openfiles, 0);
2277 atomic_add_int(&uip->ui_openfiles, count);
2278 }
2279 }
2280 if (ncr) {
2281 uip = ncr->cr_uidinfo;
2282 pup = &uip->ui_pcpu[cpu];
2283 atomic_add_int(&pup->pu_openfiles, 1);
2284 if (pup->pu_openfiles < -PUP_LIMIT ||
2285 pup->pu_openfiles > PUP_LIMIT) {
2286 count = atomic_swap_int(&pup->pu_openfiles, 0);
2287 atomic_add_int(&uip->ui_openfiles, count);
2288 }
2289 }
2290 }
2291 if (ncr)
2292 crhold(ncr);
2293 fp->f_cred = ncr;
2294 if (ocr)
2295 crfree(ocr);
2296 }
2297
2298 /*
2299 * Free a file descriptor.
2300 */
2301 static
2302 void
ffree(struct file * fp)2303 ffree(struct file *fp)
2304 {
2305 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
2306 fsetcred(fp, NULL);
2307 if (fp->f_nchandle.ncp)
2308 cache_drop(&fp->f_nchandle);
2309 kfree_obj(fp, M_FILE);
2310 }
2311
2312 /*
2313 * called from init_main, initialize filedesc0 for proc0.
2314 */
2315 void
fdinit_bootstrap(struct proc * p0,struct filedesc * fdp0,int cmask)2316 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
2317 {
2318 p0->p_fd = fdp0;
2319 p0->p_fdtol = NULL;
2320 fdp0->fd_refcnt = 1;
2321 fdp0->fd_cmask = cmask;
2322 fdp0->fd_files = fdp0->fd_builtin_files;
2323 fdp0->fd_nfiles = NDFILE;
2324 fdp0->fd_lastfile = -1;
2325 spin_init(&fdp0->fd_spin, "fdinitbootstrap");
2326 }
2327
2328 /*
2329 * Build a new filedesc structure.
2330 */
2331 struct filedesc *
fdinit(struct proc * p)2332 fdinit(struct proc *p)
2333 {
2334 struct filedesc *newfdp;
2335 struct filedesc *fdp = p->p_fd;
2336
2337 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
2338 spin_lock(&fdp->fd_spin);
2339 if (fdp->fd_cdir) {
2340 newfdp->fd_cdir = fdp->fd_cdir;
2341 vref(newfdp->fd_cdir);
2342 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
2343 }
2344
2345 /*
2346 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
2347 * proc0, but should unconditionally exist in other processes.
2348 */
2349 if (fdp->fd_rdir) {
2350 newfdp->fd_rdir = fdp->fd_rdir;
2351 vref(newfdp->fd_rdir);
2352 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
2353 }
2354 if (fdp->fd_jdir) {
2355 newfdp->fd_jdir = fdp->fd_jdir;
2356 vref(newfdp->fd_jdir);
2357 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
2358 }
2359 spin_unlock(&fdp->fd_spin);
2360
2361 /* Create the file descriptor table. */
2362 newfdp->fd_refcnt = 1;
2363 newfdp->fd_cmask = cmask;
2364 newfdp->fd_files = newfdp->fd_builtin_files;
2365 newfdp->fd_nfiles = NDFILE;
2366 newfdp->fd_lastfile = -1;
2367 spin_init(&newfdp->fd_spin, "fdinit");
2368
2369 return (newfdp);
2370 }
2371
2372 /*
2373 * Share a filedesc structure.
2374 */
2375 struct filedesc *
fdshare(struct proc * p)2376 fdshare(struct proc *p)
2377 {
2378 struct filedesc *fdp;
2379
2380 fdp = p->p_fd;
2381 spin_lock(&fdp->fd_spin);
2382 fdp->fd_refcnt++;
2383 spin_unlock(&fdp->fd_spin);
2384 return (fdp);
2385 }
2386
2387 /*
2388 * Copy a filedesc structure.
2389 */
2390 int
fdcopy(struct proc * p,struct filedesc ** fpp)2391 fdcopy(struct proc *p, struct filedesc **fpp)
2392 {
2393 struct filedesc *fdp = p->p_fd;
2394 struct filedesc *newfdp;
2395 struct fdnode *fdnode;
2396 int i;
2397 int ni;
2398
2399 /*
2400 * Certain daemons might not have file descriptors.
2401 */
2402 if (fdp == NULL)
2403 return (0);
2404
2405 /*
2406 * Allocate the new filedesc and fd_files[] array. This can race
2407 * with operations by other threads on the fdp so we have to be
2408 * careful.
2409 */
2410 newfdp = kmalloc(sizeof(struct filedesc),
2411 M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK);
2412 if (newfdp == NULL) {
2413 *fpp = NULL;
2414 return (-1);
2415 }
2416 again:
2417 spin_lock(&fdp->fd_spin);
2418 if (fdp->fd_lastfile < NDFILE) {
2419 newfdp->fd_files = newfdp->fd_builtin_files;
2420 i = NDFILE;
2421 } else {
2422 /*
2423 * We have to allocate (N^2-1) entries for our in-place
2424 * binary tree. Allow the table to shrink.
2425 */
2426 i = fdp->fd_nfiles;
2427 ni = (i - 1) / 2;
2428 while (ni > fdp->fd_lastfile && ni > NDFILE) {
2429 i = ni;
2430 ni = (i - 1) / 2;
2431 }
2432 spin_unlock(&fdp->fd_spin);
2433 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
2434 M_FILEDESC, M_WAITOK | M_ZERO);
2435
2436 /*
2437 * Check for race, retry
2438 */
2439 spin_lock(&fdp->fd_spin);
2440 if (i <= fdp->fd_lastfile) {
2441 spin_unlock(&fdp->fd_spin);
2442 kfree(newfdp->fd_files, M_FILEDESC);
2443 goto again;
2444 }
2445 }
2446
2447 /*
2448 * Dup the remaining fields. vref() and cache_hold() can be
2449 * safely called while holding the read spinlock on fdp.
2450 *
2451 * The read spinlock on fdp is still being held.
2452 *
2453 * NOTE: vref and cache_hold calls for the case where the vnode
2454 * or cache entry already has at least one ref may be called
2455 * while holding spin locks.
2456 */
2457 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
2458 vref(newfdp->fd_cdir);
2459 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
2460 }
2461 /*
2462 * We must check for fd_rdir here, at least for now because
2463 * the init process is created before we have access to the
2464 * rootvode to take a reference to it.
2465 */
2466 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
2467 vref(newfdp->fd_rdir);
2468 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
2469 }
2470 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
2471 vref(newfdp->fd_jdir);
2472 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
2473 }
2474 newfdp->fd_refcnt = 1;
2475 newfdp->fd_nfiles = i;
2476 newfdp->fd_lastfile = fdp->fd_lastfile;
2477 newfdp->fd_freefile = fdp->fd_freefile;
2478 newfdp->fd_cmask = fdp->fd_cmask;
2479 spin_init(&newfdp->fd_spin, "fdcopy");
2480
2481 /*
2482 * Copy the descriptor table through (i). This also copies the
2483 * allocation state. Then go through and ref the file pointers
2484 * and clean up any KQ descriptors.
2485 *
2486 * kq descriptors cannot be copied. Since we haven't ref'd the
2487 * copied files yet we can ignore the return value from funsetfd().
2488 *
2489 * The read spinlock on fdp is still being held.
2490 *
2491 * Be sure to clean out fdnode->tdcache, otherwise bad things will
2492 * happen.
2493 */
2494 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
2495 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
2496 fdnode = &newfdp->fd_files[i];
2497 if (fdnode->reserved) {
2498 fdreserve_locked(newfdp, i, -1);
2499 fdnode->reserved = 0;
2500 fdfixup_locked(newfdp, i);
2501 } else if (fdnode->fp) {
2502 bzero(&fdnode->tdcache, sizeof(fdnode->tdcache));
2503 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
2504 (void)funsetfd_locked(newfdp, i);
2505 } else {
2506 fhold(fdnode->fp);
2507 }
2508 }
2509 }
2510 spin_unlock(&fdp->fd_spin);
2511 *fpp = newfdp;
2512 return (0);
2513 }
2514
2515 /*
2516 * Release a filedesc structure.
2517 *
2518 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
2519 */
2520 void
fdfree(struct proc * p,struct filedesc * repl)2521 fdfree(struct proc *p, struct filedesc *repl)
2522 {
2523 struct filedesc *fdp;
2524 struct fdnode *fdnode;
2525 int i;
2526 struct filedesc_to_leader *fdtol;
2527 struct file *fp;
2528 struct vnode *vp;
2529 struct flock lf;
2530
2531 /*
2532 * Before destroying or replacing p->p_fd we must be sure to
2533 * clean out the cache of the last thread, which should be
2534 * curthread.
2535 */
2536 fexitcache(curthread);
2537
2538 /*
2539 * Certain daemons might not have file descriptors.
2540 */
2541 fdp = p->p_fd;
2542 if (fdp == NULL) {
2543 p->p_fd = repl;
2544 return;
2545 }
2546
2547 /*
2548 * Severe messing around to follow.
2549 */
2550 spin_lock(&fdp->fd_spin);
2551
2552 /* Check for special need to clear POSIX style locks */
2553 fdtol = p->p_fdtol;
2554 if (fdtol != NULL) {
2555 KASSERT(fdtol->fdl_refcount > 0,
2556 ("filedesc_to_refcount botch: fdl_refcount=%d",
2557 fdtol->fdl_refcount));
2558 if (fdtol->fdl_refcount == 1 && p->p_leader->p_advlock_flag) {
2559 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2560 fdnode = &fdp->fd_files[i];
2561 if (fdnode->fp == NULL ||
2562 fdnode->fp->f_type != DTYPE_VNODE) {
2563 continue;
2564 }
2565 fp = fdnode->fp;
2566 fhold(fp);
2567 spin_unlock(&fdp->fd_spin);
2568
2569 lf.l_whence = SEEK_SET;
2570 lf.l_start = 0;
2571 lf.l_len = 0;
2572 lf.l_type = F_UNLCK;
2573 vp = (struct vnode *)fp->f_data;
2574 VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
2575 F_UNLCK, &lf, F_POSIX);
2576 fdrop(fp);
2577 spin_lock(&fdp->fd_spin);
2578 }
2579 }
2580 retry:
2581 if (fdtol->fdl_refcount == 1) {
2582 if (fdp->fd_holdleaderscount > 0 &&
2583 p->p_leader->p_advlock_flag) {
2584 /*
2585 * close() or do_dup() has cleared a reference
2586 * in a shared file descriptor table.
2587 */
2588 fdp->fd_holdleaderswakeup = 1;
2589 ssleep(&fdp->fd_holdleaderscount,
2590 &fdp->fd_spin, 0, "fdlhold", 0);
2591 goto retry;
2592 }
2593 if (fdtol->fdl_holdcount > 0) {
2594 /*
2595 * Ensure that fdtol->fdl_leader
2596 * remains valid in closef().
2597 */
2598 fdtol->fdl_wakeup = 1;
2599 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
2600 goto retry;
2601 }
2602 }
2603 fdtol->fdl_refcount--;
2604 if (fdtol->fdl_refcount == 0 &&
2605 fdtol->fdl_holdcount == 0) {
2606 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2607 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2608 } else {
2609 fdtol = NULL;
2610 }
2611 p->p_fdtol = NULL;
2612 if (fdtol != NULL) {
2613 spin_unlock(&fdp->fd_spin);
2614 kfree(fdtol, M_FILEDESC_TO_LEADER);
2615 spin_lock(&fdp->fd_spin);
2616 }
2617 }
2618 if (--fdp->fd_refcnt > 0) {
2619 spin_unlock(&fdp->fd_spin);
2620 spin_lock(&p->p_spin);
2621 p->p_fd = repl;
2622 spin_unlock(&p->p_spin);
2623 return;
2624 }
2625
2626 /*
2627 * Even though we are the last reference to the structure allproc
2628 * scans may still reference the structure. Maintain proper
2629 * locks until we can replace p->p_fd.
2630 *
2631 * Also note that kqueue's closef still needs to reference the
2632 * fdp via p->p_fd, so we have to close the descriptors before
2633 * we replace p->p_fd.
2634 */
2635 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2636 if (fdp->fd_files[i].fp) {
2637 fp = funsetfd_locked(fdp, i);
2638 if (fp) {
2639 spin_unlock(&fdp->fd_spin);
2640 if (SLIST_FIRST(&fp->f_klist))
2641 knote_fdclose(fp, fdp, i);
2642 closef(fp, p);
2643 spin_lock(&fdp->fd_spin);
2644 }
2645 }
2646 }
2647 spin_unlock(&fdp->fd_spin);
2648
2649 /*
2650 * Interlock against an allproc scan operations (typically frevoke).
2651 */
2652 spin_lock(&p->p_spin);
2653 p->p_fd = repl;
2654 spin_unlock(&p->p_spin);
2655
2656 /*
2657 * Wait for any softrefs to go away. This race rarely occurs so
2658 * we can use a non-critical-path style poll/sleep loop. The
2659 * race only occurs against allproc scans.
2660 *
2661 * No new softrefs can occur with the fdp disconnected from the
2662 * process.
2663 */
2664 if (fdp->fd_softrefs) {
2665 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2666 while (fdp->fd_softrefs)
2667 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2668 }
2669
2670 if (fdp->fd_files != fdp->fd_builtin_files)
2671 kfree(fdp->fd_files, M_FILEDESC);
2672 if (fdp->fd_cdir) {
2673 cache_drop(&fdp->fd_ncdir);
2674 vrele(fdp->fd_cdir);
2675 }
2676 if (fdp->fd_rdir) {
2677 cache_drop(&fdp->fd_nrdir);
2678 vrele(fdp->fd_rdir);
2679 }
2680 if (fdp->fd_jdir) {
2681 cache_drop(&fdp->fd_njdir);
2682 vrele(fdp->fd_jdir);
2683 }
2684 kfree(fdp, M_FILEDESC);
2685 }
2686
2687 /*
2688 * Retrieve and reference the file pointer associated with a descriptor.
2689 *
2690 * td must be the current thread.
2691 */
2692 struct file *
holdfp(thread_t td,int fd,int flag)2693 holdfp(thread_t td, int fd, int flag)
2694 {
2695 struct file *fp;
2696
2697 fp = _holdfp_cache(td, fd);
2698 if (fp) {
2699 if ((fp->f_flag & flag) == 0 && flag != -1) {
2700 fdrop(fp);
2701 fp = NULL;
2702 }
2703 }
2704 return fp;
2705 }
2706
2707 /*
2708 * holdsock() - load the struct file pointer associated
2709 * with a socket into *fpp. If an error occurs, non-zero
2710 * will be returned and *fpp will be set to NULL.
2711 *
2712 * td must be the current thread.
2713 */
2714 int
holdsock(thread_t td,int fd,struct file ** fpp)2715 holdsock(thread_t td, int fd, struct file **fpp)
2716 {
2717 struct file *fp;
2718 int error;
2719
2720 /*
2721 * Lockless shortcut
2722 */
2723 fp = _holdfp_cache(td, fd);
2724 if (fp) {
2725 if (fp->f_type != DTYPE_SOCKET) {
2726 fdrop(fp);
2727 fp = NULL;
2728 error = ENOTSOCK;
2729 } else {
2730 error = 0;
2731 }
2732 } else {
2733 error = EBADF;
2734 }
2735 *fpp = fp;
2736
2737 return (error);
2738 }
2739
2740 /*
2741 * Convert a user file descriptor to a held file pointer.
2742 *
2743 * td must be the current thread.
2744 */
2745 int
holdvnode(thread_t td,int fd,struct file ** fpp)2746 holdvnode(thread_t td, int fd, struct file **fpp)
2747 {
2748 struct file *fp;
2749 int error;
2750
2751 fp = _holdfp_cache(td, fd);
2752 if (fp) {
2753 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2754 fdrop(fp);
2755 fp = NULL;
2756 error = EINVAL;
2757 } else {
2758 error = 0;
2759 }
2760 } else {
2761 error = EBADF;
2762 }
2763 *fpp = fp;
2764
2765 return (error);
2766 }
2767
2768 /*
2769 * Convert a user file descriptor to a held file pointer.
2770 *
2771 * td must be the current thread.
2772 */
2773 int
holdvnode2(thread_t td,int fd,struct file ** fpp,char * fflagsp)2774 holdvnode2(thread_t td, int fd, struct file **fpp, char *fflagsp)
2775 {
2776 struct file *fp;
2777 int error;
2778
2779 fp = _holdfp2(td, fd, fflagsp);
2780 if (fp) {
2781 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2782 fdrop(fp);
2783 fp = NULL;
2784 error = EINVAL;
2785 } else {
2786 error = 0;
2787 }
2788 } else {
2789 error = EBADF;
2790 }
2791 *fpp = fp;
2792
2793 return (error);
2794 }
2795
2796 /*
2797 * For setugid programs, we don't want to people to use that setugidness
2798 * to generate error messages which write to a file which otherwise would
2799 * otherwise be off-limits to the process.
2800 *
2801 * This is a gross hack to plug the hole. A better solution would involve
2802 * a special vop or other form of generalized access control mechanism. We
2803 * go ahead and just reject all procfs file systems accesses as dangerous.
2804 *
2805 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2806 * sufficient. We also don't for check setugidness since we know we are.
2807 */
2808 static int
is_unsafe(struct file * fp)2809 is_unsafe(struct file *fp)
2810 {
2811 if (fp->f_type == DTYPE_VNODE &&
2812 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2813 return (1);
2814 return (0);
2815 }
2816
2817 /*
2818 * Make this setguid thing safe, if at all possible.
2819 *
2820 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2821 */
2822 void
setugidsafety(struct proc * p)2823 setugidsafety(struct proc *p)
2824 {
2825 struct filedesc *fdp = p->p_fd;
2826 int i;
2827
2828 /* Certain daemons might not have file descriptors. */
2829 if (fdp == NULL)
2830 return;
2831
2832 /*
2833 * note: fdp->fd_files may be reallocated out from under us while
2834 * we are blocked in a close. Be careful!
2835 */
2836 for (i = 0; i <= fdp->fd_lastfile; i++) {
2837 if (i > 2)
2838 break;
2839 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2840 struct file *fp;
2841
2842 /*
2843 * NULL-out descriptor prior to close to avoid
2844 * a race while close blocks.
2845 */
2846 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2847 knote_fdclose(fp, fdp, i);
2848 closef(fp, p);
2849 }
2850 }
2851 }
2852 }
2853
2854 /*
2855 * Close all CLOEXEC files on exec.
2856 *
2857 * Only a single thread remains for the current process.
2858 *
2859 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2860 */
2861 void
fdcloseexec(struct proc * p)2862 fdcloseexec(struct proc *p)
2863 {
2864 struct filedesc *fdp = p->p_fd;
2865 int i;
2866
2867 /* Certain daemons might not have file descriptors. */
2868 if (fdp == NULL)
2869 return;
2870
2871 /*
2872 * We cannot cache fd_files since operations may block and rip
2873 * them out from under us.
2874 */
2875 for (i = 0; i <= fdp->fd_lastfile; i++) {
2876 if (fdp->fd_files[i].fp != NULL &&
2877 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2878 struct file *fp;
2879
2880 /*
2881 * NULL-out descriptor prior to close to avoid
2882 * a race while close blocks.
2883 *
2884 * (funsetfd*() also clears the fd cache)
2885 */
2886 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2887 knote_fdclose(fp, fdp, i);
2888 closef(fp, p);
2889 }
2890 }
2891 }
2892 }
2893
2894 /*
2895 * It is unsafe for set[ug]id processes to be started with file
2896 * descriptors 0..2 closed, as these descriptors are given implicit
2897 * significance in the Standard C library. fdcheckstd() will create a
2898 * descriptor referencing /dev/null for each of stdin, stdout, and
2899 * stderr that is not already open.
2900 *
2901 * NOT MPSAFE - calls falloc, vn_open, etc
2902 */
2903 int
fdcheckstd(struct lwp * lp)2904 fdcheckstd(struct lwp *lp)
2905 {
2906 struct nlookupdata nd;
2907 struct filedesc *fdp;
2908 struct file *fp;
2909 int retval;
2910 int i, error, flags, devnull;
2911
2912 fdp = lp->lwp_proc->p_fd;
2913 if (fdp == NULL)
2914 return (0);
2915 devnull = -1;
2916 error = 0;
2917 for (i = 0; i < 3; i++) {
2918 if (fdp->fd_files[i].fp != NULL)
2919 continue;
2920 if (devnull < 0) {
2921 if ((error = falloc(lp, &fp, &devnull)) != 0)
2922 break;
2923
2924 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2925 NLC_FOLLOW|NLC_LOCKVP);
2926 flags = FREAD | FWRITE;
2927 if (error == 0)
2928 error = vn_open(&nd, &fp, flags, 0);
2929 if (error == 0)
2930 fsetfd(fdp, fp, devnull);
2931 else
2932 fsetfd(fdp, NULL, devnull);
2933 fdrop(fp);
2934 nlookup_done(&nd);
2935 if (error)
2936 break;
2937 KKASSERT(i == devnull);
2938 } else {
2939 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2940 if (error != 0)
2941 break;
2942 }
2943 }
2944 return (error);
2945 }
2946
2947 /*
2948 * Internal form of close.
2949 * Decrement reference count on file structure.
2950 * Note: td and/or p may be NULL when closing a file
2951 * that was being passed in a message.
2952 *
2953 * MPALMOSTSAFE - acquires mplock for VOP operations
2954 */
2955 int
closef(struct file * fp,struct proc * p)2956 closef(struct file *fp, struct proc *p)
2957 {
2958 struct vnode *vp;
2959 struct flock lf;
2960 struct filedesc_to_leader *fdtol;
2961
2962 if (fp == NULL)
2963 return (0);
2964
2965 /*
2966 * POSIX record locking dictates that any close releases ALL
2967 * locks owned by this process. This is handled by setting
2968 * a flag in the unlock to free ONLY locks obeying POSIX
2969 * semantics, and not to free BSD-style file locks.
2970 * If the descriptor was in a message, POSIX-style locks
2971 * aren't passed with the descriptor.
2972 */
2973 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2974 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2975 ) {
2976 if (p->p_leader->p_advlock_flag) {
2977 lf.l_whence = SEEK_SET;
2978 lf.l_start = 0;
2979 lf.l_len = 0;
2980 lf.l_type = F_UNLCK;
2981 vp = (struct vnode *)fp->f_data;
2982 VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2983 &lf, F_POSIX);
2984 }
2985 fdtol = p->p_fdtol;
2986 if (fdtol != NULL) {
2987 lwkt_gettoken(&p->p_token);
2988
2989 /*
2990 * Handle special case where file descriptor table
2991 * is shared between multiple process leaders.
2992 */
2993 for (fdtol = fdtol->fdl_next;
2994 fdtol != p->p_fdtol;
2995 fdtol = fdtol->fdl_next) {
2996 if (fdtol->fdl_leader->p_advlock_flag == 0)
2997 continue;
2998 fdtol->fdl_holdcount++;
2999 lf.l_whence = SEEK_SET;
3000 lf.l_start = 0;
3001 lf.l_len = 0;
3002 lf.l_type = F_UNLCK;
3003 vp = (struct vnode *)fp->f_data;
3004 VOP_ADVLOCK(vp, (caddr_t)fdtol->fdl_leader,
3005 F_UNLCK, &lf, F_POSIX);
3006 fdtol->fdl_holdcount--;
3007 if (fdtol->fdl_holdcount == 0 &&
3008 fdtol->fdl_wakeup != 0) {
3009 fdtol->fdl_wakeup = 0;
3010 wakeup(fdtol);
3011 }
3012 }
3013 lwkt_reltoken(&p->p_token);
3014 }
3015 }
3016 return (fdrop(fp));
3017 }
3018
3019 /*
3020 * fhold() can only be called if f_count is already at least 1 (i.e. the
3021 * caller of fhold() already has a reference to the file pointer in some
3022 * manner or other).
3023 *
3024 * Atomic ops are used for incrementing and decrementing f_count before
3025 * the 1->0 transition. f_count 1->0 transition is special, see the
3026 * comment in fdrop().
3027 */
3028 void
fhold(struct file * fp)3029 fhold(struct file *fp)
3030 {
3031 /* 0->1 transition will never work */
3032 KASSERT(fp->f_count > 0, ("fhold: invalid f_count %d", fp->f_count));
3033 atomic_add_int(&fp->f_count, 1);
3034 }
3035
3036 /*
3037 * fdrop() - drop a reference to a descriptor
3038 */
3039 int
fdrop(struct file * fp)3040 fdrop(struct file *fp)
3041 {
3042 struct flock lf;
3043 struct vnode *vp;
3044 int error, do_free = 0;
3045
3046 /*
3047 * NOTE:
3048 * Simple atomic_fetchadd_int(f_count, -1) here will cause use-
3049 * after-free or double free (due to f_count 0->1 transition), if
3050 * fhold() is called on the fps found through filehead iteration.
3051 */
3052 for (;;) {
3053 int count = fp->f_count;
3054
3055 cpu_ccfence();
3056 KASSERT(count > 0, ("fdrop: invalid f_count %d", count));
3057 if (count == 1) {
3058 struct filelist_head *head = fp2filelist(fp);
3059
3060 /*
3061 * About to drop the last reference, hold the
3062 * filehead spin lock and drop it, so that no
3063 * one could see this fp through filehead anymore,
3064 * let alone fhold() this fp.
3065 */
3066 spin_lock(&head->spin);
3067 if (atomic_cmpset_int(&fp->f_count, count, 0)) {
3068 LIST_REMOVE(fp, f_list);
3069 spin_unlock(&head->spin);
3070 atomic_subtract_int(&nfiles, 1);
3071 do_free = 1; /* free this fp */
3072 break;
3073 }
3074 spin_unlock(&head->spin);
3075 /* retry */
3076 } else if (atomic_cmpset_int(&fp->f_count, count, count - 1)) {
3077 break;
3078 }
3079 /* retry */
3080 }
3081 if (!do_free)
3082 return (0);
3083
3084 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
3085
3086 /*
3087 * The last reference has gone away, we own the fp structure free
3088 * and clear.
3089 */
3090 if (fp->f_count < 0)
3091 panic("fdrop: count < 0");
3092 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
3093 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
3094 ) {
3095 lf.l_whence = SEEK_SET;
3096 lf.l_start = 0;
3097 lf.l_len = 0;
3098 lf.l_type = F_UNLCK;
3099 vp = (struct vnode *)fp->f_data;
3100 VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
3101 }
3102 if (fp->f_ops != &badfileops)
3103 error = fo_close(fp);
3104 else
3105 error = 0;
3106 ffree(fp);
3107 return (error);
3108 }
3109
3110 /*
3111 * Apply an advisory lock on a file descriptor.
3112 *
3113 * Just attempt to get a record lock of the requested type on
3114 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3115 *
3116 * MPALMOSTSAFE
3117 */
3118 int
sys_flock(struct sysmsg * sysmsg,const struct flock_args * uap)3119 sys_flock(struct sysmsg *sysmsg, const struct flock_args *uap)
3120 {
3121 thread_t td = curthread;
3122 struct file *fp;
3123 struct vnode *vp;
3124 struct flock lf;
3125 int error;
3126
3127 if ((fp = holdfp(td, uap->fd, -1)) == NULL)
3128 return (EBADF);
3129 if (fp->f_type != DTYPE_VNODE) {
3130 error = EOPNOTSUPP;
3131 goto done;
3132 }
3133 vp = (struct vnode *)fp->f_data;
3134 lf.l_whence = SEEK_SET;
3135 lf.l_start = 0;
3136 lf.l_len = 0;
3137 if (uap->how & LOCK_UN) {
3138 lf.l_type = F_UNLCK;
3139 atomic_clear_int(&fp->f_flag, FHASLOCK); /* race ok */
3140 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
3141 goto done;
3142 }
3143 if (uap->how & LOCK_EX)
3144 lf.l_type = F_WRLCK;
3145 else if (uap->how & LOCK_SH)
3146 lf.l_type = F_RDLCK;
3147 else {
3148 error = EBADF;
3149 goto done;
3150 }
3151 if (uap->how & LOCK_NB)
3152 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
3153 else
3154 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
3155 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */
3156 done:
3157 fdrop(fp);
3158 return (error);
3159 }
3160
3161 /*
3162 * File Descriptor pseudo-device driver ( /dev/fd/N ).
3163 *
3164 * This interface is now a bit more linux-compatible and attempts to not
3165 * share seek positions by not sharing the fp of the descriptor when
3166 * possible.
3167 *
3168 * Probably a good idea anyhow, but now particularly important for
3169 * fexecve() which uses /dev/fd/N.
3170 *
3171 * The original interface effectively dup()d the descriptor.
3172 */
3173 static int
fdopen(struct dev_open_args * ap)3174 fdopen(struct dev_open_args *ap)
3175 {
3176 struct file *wfp;
3177 thread_t td;
3178 int error;
3179 int sfd;
3180
3181 td = curthread;
3182 KKASSERT(td->td_lwp != NULL);
3183
3184 /*
3185 * Get the fp for /dev/fd/N
3186 */
3187 sfd = minor(ap->a_head.a_dev);
3188 if ((wfp = holdfp(td, sfd, -1)) == NULL)
3189 return (EBADF);
3190
3191 /*
3192 * Close a revoke/dup race. Duping a descriptor marked as revoked
3193 * will dup a dummy descriptor instead of the real one.
3194 */
3195 if (wfp->f_flag & FREVOKED) {
3196 kprintf("Warning: attempt to dup() a revoked descriptor\n");
3197 fdrop(wfp);
3198 wfp = NULL;
3199 error = falloc(NULL, &wfp, NULL);
3200 if (error)
3201 return (error);
3202 }
3203
3204 /*
3205 * Check that the mode the file is being opened for is a
3206 * subset of the mode of the existing descriptor.
3207 */
3208 if (ap->a_fpp == NULL) {
3209 fdrop(wfp);
3210 return EINVAL;
3211 }
3212 if (((ap->a_oflags & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
3213 fdrop(wfp);
3214 return EACCES;
3215 }
3216 if (wfp->f_type == DTYPE_VNODE && wfp->f_data) {
3217 /*
3218 * If wfp is a vnode create a new fp so things like the
3219 * seek position (etc) are not shared with the original.
3220 *
3221 * Don't try to call VOP_OPEN(). Adjust the open-count
3222 * ourselves.
3223 */
3224 struct vnode *vp;
3225 struct file *fp;
3226
3227 vp = wfp->f_data;
3228 fp = *ap->a_fpp;
3229
3230 /*
3231 * Yah... this wouldn't be good.
3232 */
3233 if ((ap->a_oflags & (FWRITE|O_TRUNC)) && vp->v_type == VDIR) {
3234 fdrop(wfp);
3235 return EISDIR;
3236 }
3237
3238 /*
3239 * Setup the new fp and simulate an open(), but for now do
3240 * not actually call VOP_OPEN() though we probably could.
3241 */
3242 fp->f_type = DTYPE_VNODE;
3243 /* retain flags not to be copied */
3244 fp->f_flag = (fp->f_flag & ~FMASK) | (ap->a_oflags & FMASK);
3245 fp->f_ops = &vnode_fileops;
3246 fp->f_data = vp;
3247 vref(vp);
3248
3249 if (ap->a_oflags & FWRITE)
3250 atomic_add_int(&vp->v_writecount, 1);
3251 KKASSERT(vp->v_opencount >= 0 && vp->v_opencount != INT_MAX);
3252 atomic_add_int(&vp->v_opencount, 1);
3253 fdrop(wfp);
3254 } else {
3255 /*
3256 * If wfp is not a vnode we have to share it directly.
3257 */
3258 fdrop(*ap->a_fpp);
3259 *ap->a_fpp = wfp; /* transfer hold count */
3260 }
3261 return EALREADY;
3262 }
3263
3264 /*
3265 * NOT MPSAFE - I think these refer to a common file descriptor table
3266 * and we need to spinlock that to link fdtol in.
3267 */
3268 struct filedesc_to_leader *
filedesc_to_leader_alloc(struct filedesc_to_leader * old,struct proc * leader)3269 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
3270 struct proc *leader)
3271 {
3272 struct filedesc_to_leader *fdtol;
3273
3274 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
3275 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO);
3276 fdtol->fdl_refcount = 1;
3277 fdtol->fdl_holdcount = 0;
3278 fdtol->fdl_wakeup = 0;
3279 fdtol->fdl_leader = leader;
3280 if (old != NULL) {
3281 fdtol->fdl_next = old->fdl_next;
3282 fdtol->fdl_prev = old;
3283 old->fdl_next = fdtol;
3284 fdtol->fdl_next->fdl_prev = fdtol;
3285 } else {
3286 fdtol->fdl_next = fdtol;
3287 fdtol->fdl_prev = fdtol;
3288 }
3289 return fdtol;
3290 }
3291
3292 /*
3293 * Scan all file pointers in the system. The callback is made with
3294 * the master list spinlock held exclusively.
3295 */
3296 void
allfiles_scan_exclusive(int (* callback)(struct file *,void *),void * data)3297 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
3298 {
3299 int i;
3300
3301 for (i = 0; i < NFILELIST_HEADS; ++i) {
3302 struct filelist_head *head = &filelist_heads[i];
3303 struct file *fp;
3304
3305 spin_lock(&head->spin);
3306 LIST_FOREACH(fp, &head->list, f_list) {
3307 int res;
3308
3309 res = callback(fp, data);
3310 if (res < 0)
3311 break;
3312 }
3313 spin_unlock(&head->spin);
3314 }
3315 }
3316
3317 /*
3318 * Get file structures.
3319 *
3320 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
3321 */
3322
3323 struct sysctl_kern_file_info {
3324 int count;
3325 int error;
3326 struct sysctl_req *req;
3327 };
3328
3329 static int sysctl_kern_file_callback(struct proc *p, void *data);
3330
3331 static int
sysctl_kern_file(SYSCTL_HANDLER_ARGS)3332 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3333 {
3334 struct sysctl_kern_file_info info;
3335
3336 /*
3337 * Note: because the number of file descriptors is calculated
3338 * in different ways for sizing vs returning the data,
3339 * there is information leakage from the first loop. However,
3340 * it is of a similar order of magnitude to the leakage from
3341 * global system statistics such as kern.openfiles.
3342 *
3343 * When just doing a count, note that we cannot just count
3344 * the elements and add f_count via the filehead list because
3345 * threaded processes share their descriptor table and f_count might
3346 * still be '1' in that case.
3347 *
3348 * Since the SYSCTL op can block, we must hold the process to
3349 * prevent it being ripped out from under us either in the
3350 * file descriptor loop or in the greater LIST_FOREACH. The
3351 * process may be in varying states of disrepair. If the process
3352 * is in SZOMB we may have caught it just as it is being removed
3353 * from the allproc list, we must skip it in that case to maintain
3354 * an unbroken chain through the allproc list.
3355 */
3356 info.count = 0;
3357 info.error = 0;
3358 info.req = req;
3359 allproc_scan(sysctl_kern_file_callback, &info, 0);
3360
3361 /*
3362 * When just calculating the size, overestimate a bit to try to
3363 * prevent system activity from causing the buffer-fill call
3364 * to fail later on.
3365 */
3366 if (req->oldptr == NULL) {
3367 info.count = (info.count + 16) + (info.count / 10);
3368 info.error = SYSCTL_OUT(req, NULL,
3369 info.count * sizeof(struct kinfo_file));
3370 }
3371 return (info.error);
3372 }
3373
3374 static int
sysctl_kern_file_callback(struct proc * p,void * data)3375 sysctl_kern_file_callback(struct proc *p, void *data)
3376 {
3377 struct sysctl_kern_file_info *info = data;
3378 struct kinfo_file kf;
3379 struct filedesc *fdp;
3380 struct file *fp;
3381 uid_t uid;
3382 int n;
3383
3384 if (p->p_stat == SIDL || p->p_stat == SZOMB)
3385 return(0);
3386 if (!(PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0))
3387 return(0);
3388
3389 /*
3390 * Softref the fdp to prevent it from being destroyed
3391 */
3392 spin_lock(&p->p_spin);
3393 if ((fdp = p->p_fd) == NULL) {
3394 spin_unlock(&p->p_spin);
3395 return(0);
3396 }
3397 atomic_add_int(&fdp->fd_softrefs, 1);
3398 spin_unlock(&p->p_spin);
3399
3400 /*
3401 * The fdp's own spinlock prevents the contents from being
3402 * modified.
3403 */
3404 spin_lock_shared(&fdp->fd_spin);
3405 for (n = 0; n < fdp->fd_nfiles; ++n) {
3406 if ((fp = fdp->fd_files[n].fp) == NULL)
3407 continue;
3408 if (info->req->oldptr == NULL) {
3409 ++info->count;
3410 } else {
3411 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
3412 kcore_make_file(&kf, fp, p->p_pid, uid, n);
3413 spin_unlock_shared(&fdp->fd_spin);
3414 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
3415 spin_lock_shared(&fdp->fd_spin);
3416 if (info->error)
3417 break;
3418 }
3419 }
3420 spin_unlock_shared(&fdp->fd_spin);
3421 atomic_subtract_int(&fdp->fd_softrefs, 1);
3422 if (info->error)
3423 return(-1);
3424 return(0);
3425 }
3426
3427 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
3428 0, 0, sysctl_kern_file, "S,file", "Entire file table");
3429
3430 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
3431 &minfilesperproc, 0, "Minimum files allowed open per process");
3432 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3433 &maxfilesperproc, 0, "Maximum files allowed open per process");
3434 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
3435 &maxfilesperuser, 0, "Maximum files allowed open per user");
3436
3437 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3438 &maxfiles, 0, "Maximum number of files");
3439
3440 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
3441 &maxfilesrootres, 0, "Descriptors reserved for root use");
3442
3443 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3444 &nfiles, 0, "System-wide number of open files");
3445
3446 static void
fildesc_drvinit(void * unused)3447 fildesc_drvinit(void *unused)
3448 {
3449 int fd;
3450
3451 for (fd = 0; fd < NUMFDESC; fd++) {
3452 make_dev(&fildesc_ops, fd,
3453 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
3454 }
3455
3456 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
3457 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
3458 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
3459 }
3460
3461 struct fileops badfileops = {
3462 .fo_read = badfo_readwrite,
3463 .fo_write = badfo_readwrite,
3464 .fo_ioctl = badfo_ioctl,
3465 .fo_kqfilter = badfo_kqfilter,
3466 .fo_stat = badfo_stat,
3467 .fo_close = badfo_close,
3468 .fo_shutdown = badfo_shutdown,
3469 .fo_seek = badfo_seek
3470 };
3471
3472 int
badfo_readwrite(struct file * fp,struct uio * uio,struct ucred * cred,int flags)3473 badfo_readwrite(
3474 struct file *fp,
3475 struct uio *uio,
3476 struct ucred *cred,
3477 int flags
3478 ) {
3479 return (EBADF);
3480 }
3481
3482 int
badfo_ioctl(struct file * fp,u_long com,caddr_t data,struct ucred * cred,struct sysmsg * msgv)3483 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
3484 struct ucred *cred, struct sysmsg *msgv)
3485 {
3486 return (EBADF);
3487 }
3488
3489 /*
3490 * Must return an error to prevent registration, typically
3491 * due to a revoked descriptor (file_filtops assigned).
3492 */
3493 int
badfo_kqfilter(struct file * fp,struct knote * kn)3494 badfo_kqfilter(struct file *fp, struct knote *kn)
3495 {
3496 return (EOPNOTSUPP);
3497 }
3498
3499 int
badfo_stat(struct file * fp,struct stat * sb,struct ucred * cred)3500 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
3501 {
3502 return (EBADF);
3503 }
3504
3505 int
badfo_close(struct file * fp)3506 badfo_close(struct file *fp)
3507 {
3508 return (EBADF);
3509 }
3510
3511 int
badfo_shutdown(struct file * fp,int how)3512 badfo_shutdown(struct file *fp, int how)
3513 {
3514 return (EBADF);
3515 }
3516
3517 int
nofo_shutdown(struct file * fp,int how)3518 nofo_shutdown(struct file *fp, int how)
3519 {
3520 return (EOPNOTSUPP);
3521 }
3522
3523 int
badfo_seek(struct file * fp,off_t offset,int whence,off_t * res)3524 badfo_seek(struct file *fp, off_t offset, int whence, off_t *res)
3525 {
3526 return (ESPIPE);
3527 }
3528
3529 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE + CDEV_MAJOR,
3530 fildesc_drvinit,NULL);
3531
3532 static void
filelist_heads_init(void * arg __unused)3533 filelist_heads_init(void *arg __unused)
3534 {
3535 int i;
3536
3537 for (i = 0; i < NFILELIST_HEADS; ++i) {
3538 struct filelist_head *head = &filelist_heads[i];
3539
3540 spin_init(&head->spin, "filehead_spin");
3541 LIST_INIT(&head->list);
3542 }
3543 }
3544
3545 SYSINIT(filelistheads, SI_BOOT1_LOCK, SI_ORDER_ANY, filelist_heads_init, NULL);
3546