2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
39 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
40 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.27 2004/07/29 20:32:24 dillon Exp $
43 #include "opt_compat.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/malloc.h>
47 #include <sys/sysproto.h>
49 #include <sys/filedesc.h>
50 #include <sys/kernel.h>
51 #include <sys/sysctl.h>
52 #include <sys/vnode.h>
54 #include <sys/namei.h>
57 #include <sys/filio.h>
58 #include <sys/fcntl.h>
59 #include <sys/unistd.h>
60 #include <sys/resourcevar.h>
61 #include <sys/event.h>
62 #include <sys/kern_syscall.h>
65 #include <vm/vm_extern.h>
67 #include <sys/file2.h>
69 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
70 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
71 "file desc to leader structures");
72 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
73 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
75 static d_open_t fdopen;
79 static struct cdevsw fildesc_cdevsw = {
93 /* strategy */ nostrategy,
98 static int badfo_readwrite (struct file *fp, struct uio *uio,
99 struct ucred *cred, int flags, struct thread *td);
100 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data,
102 static int badfo_poll (struct file *fp, int events,
103 struct ucred *cred, struct thread *td);
104 static int badfo_kqfilter (struct file *fp, struct knote *kn);
105 static int badfo_stat (struct file *fp, struct stat *sb, struct thread *td);
106 static int badfo_close (struct file *fp, struct thread *td);
109 * Descriptor management.
111 struct filelist filehead; /* head of list of open files */
112 int nfiles; /* actual number of open files */
116 * System calls on descriptors.
120 getdtablesize(struct getdtablesize_args *uap)
122 struct proc *p = curproc;
125 min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
130 * Duplicate a file descriptor to a particular value.
132 * note: keep in mind that a potential race condition exists when closing
133 * descriptors from a shared descriptor table (via rfork).
137 dup2(struct dup2_args *uap)
141 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds);
147 * Duplicate a file descriptor.
151 dup(struct dup_args *uap)
155 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds);
161 kern_fcntl(int fd, int cmd, union fcntl_dat *dat)
163 struct thread *td = curthread;
164 struct proc *p = td->td_proc;
165 struct filedesc *fdp = p->p_fd;
170 int tmp, error, flg = F_POSIX;
174 if ((unsigned)fd >= fdp->fd_nfiles ||
175 (fp = fdp->fd_ofiles[fd]) == NULL)
177 pop = &fdp->fd_ofileflags[fd];
182 if (newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
183 newmin > maxfilesperproc)
185 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
189 dat->fc_cloexec = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
193 *pop = (*pop &~ UF_EXCLOSE) |
194 (dat->fc_cloexec & FD_CLOEXEC ? UF_EXCLOSE : 0);
198 dat->fc_flags = OFLAGS(fp->f_flag);
203 fp->f_flag &= ~FCNTLFLAGS;
204 fp->f_flag |= FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
205 tmp = fp->f_flag & FNONBLOCK;
206 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td);
211 tmp = fp->f_flag & FASYNC;
212 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, td);
217 fp->f_flag &= ~FNONBLOCK;
219 fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td);
225 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, td);
231 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, td);
237 /* Fall into F_SETLK */
240 if (fp->f_type != DTYPE_VNODE)
242 vp = (struct vnode *)fp->f_data;
245 * copyin/lockop may block
248 if (dat->fc_flock.l_whence == SEEK_CUR)
249 dat->fc_flock.l_start += fp->f_offset;
251 switch (dat->fc_flock.l_type) {
253 if ((fp->f_flag & FREAD) == 0) {
257 p->p_leader->p_flag |= P_ADVLOCK;
258 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
259 &dat->fc_flock, flg);
262 if ((fp->f_flag & FWRITE) == 0) {
266 p->p_leader->p_flag |= P_ADVLOCK;
267 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
268 &dat->fc_flock, flg);
271 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
272 &dat->fc_flock, F_POSIX);
278 /* Check for race with close */
279 if ((unsigned) fd >= fdp->fd_nfiles ||
280 fp != fdp->fd_ofiles[fd]) {
281 dat->fc_flock.l_whence = SEEK_SET;
282 dat->fc_flock.l_start = 0;
283 dat->fc_flock.l_len = 0;
284 dat->fc_flock.l_type = F_UNLCK;
285 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
286 F_UNLCK, &dat->fc_flock, F_POSIX);
292 if (fp->f_type != DTYPE_VNODE)
294 vp = (struct vnode *)fp->f_data;
296 * copyin/lockop may block
299 if (dat->fc_flock.l_type != F_RDLCK &&
300 dat->fc_flock.l_type != F_WRLCK &&
301 dat->fc_flock.l_type != F_UNLCK) {
305 if (dat->fc_flock.l_whence == SEEK_CUR)
306 dat->fc_flock.l_start += fp->f_offset;
307 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
308 &dat->fc_flock, F_POSIX);
318 * The file control system call.
321 fcntl(struct fcntl_args *uap)
328 dat.fc_fd = uap->arg;
331 dat.fc_cloexec = uap->arg;
334 dat.fc_flags = uap->arg;
337 dat.fc_owner = uap->arg;
342 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
343 sizeof(struct flock));
349 error = kern_fcntl(uap->fd, uap->cmd, &dat);
354 uap->sysmsg_result = dat.fc_fd;
357 uap->sysmsg_result = dat.fc_cloexec;
360 uap->sysmsg_result = dat.fc_flags;
363 uap->sysmsg_result = dat.fc_owner;
365 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
366 sizeof(struct flock));
375 * Common code for dup, dup2, and fcntl(F_DUPFD).
377 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
378 * kern_dup() to destructively dup over an existing file descriptor if new
379 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
380 * unused file descriptor that is greater than or equal to new.
383 kern_dup(enum dup_type type, int old, int new, int *res)
385 struct thread *td = curthread;
386 struct proc *p = td->td_proc;
387 struct filedesc *fdp = p->p_fd;
394 * Verify that we have a valid descriptor to dup from and
395 * possibly to dup to.
397 if (old < 0 || new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
398 new >= maxfilesperproc)
400 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL)
402 if (type == DUP_FIXED && old == new) {
406 fp = fdp->fd_ofiles[old];
410 * Expand the table for the new descriptor if needed. This may
411 * block and drop and reacquire the fidedesc lock.
413 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
414 error = fdalloc(p, new, &newfd);
420 if (type == DUP_VARIABLE)
424 * If the old file changed out from under us then treat it as a
425 * bad file descriptor. Userland should do its own locking to
428 if (fdp->fd_ofiles[old] != fp) {
429 if (fdp->fd_ofiles[new] == NULL) {
430 if (new < fdp->fd_freefile)
431 fdp->fd_freefile = new;
432 while (fdp->fd_lastfile > 0 &&
433 fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
439 KASSERT(old != new, ("new fd is same as old"));
442 * Save info on the descriptor being overwritten. We have
443 * to do the unmap now, but we cannot close it without
444 * introducing an ownership race for the slot.
446 delfp = fdp->fd_ofiles[new];
447 if (delfp != NULL && p->p_fdtol != NULL) {
449 * Ask fdfree() to sleep to ensure that all relevant
450 * process leaders can be traversed in closef().
452 fdp->fd_holdleaderscount++;
456 KASSERT(delfp == NULL || type == DUP_FIXED,
457 ("dup() picked an open file"));
459 if (delfp && (fdp->fd_ofileflags[new] & UF_MAPPED))
460 (void) munmapfd(p, new);
464 * Duplicate the source descriptor, update lastfile
466 fdp->fd_ofiles[new] = fp;
467 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
468 if (new > fdp->fd_lastfile)
469 fdp->fd_lastfile = new;
473 * If we dup'd over a valid file, we now own the reference to it
474 * and must dispose of it using closef() semantics (as if a
475 * close() were performed on it).
478 (void) closef(delfp, td);
480 fdp->fd_holdleaderscount--;
481 if (fdp->fd_holdleaderscount == 0 &&
482 fdp->fd_holdleaderswakeup != 0) {
483 fdp->fd_holdleaderswakeup = 0;
484 wakeup(&fdp->fd_holdleaderscount);
492 * If sigio is on the list associated with a process or process group,
493 * disable signalling from the device, remove sigio from the list and
497 funsetown(struct sigio *sigio)
504 *(sigio->sio_myref) = NULL;
506 if (sigio->sio_pgid < 0) {
507 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
509 } else /* if ((*sigiop)->sio_pgid > 0) */ {
510 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
513 crfree(sigio->sio_ucred);
514 free(sigio, M_SIGIO);
517 /* Free a list of sigio structures. */
519 funsetownlst(struct sigiolst *sigiolst)
523 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
528 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
530 * After permission checking, add a sigio structure to the sigio list for
531 * the process or process group.
534 fsetown(pid_t pgid, struct sigio **sigiop)
551 * Policy - Don't allow a process to FSETOWN a process
552 * in another session.
554 * Remove this test to allow maximum flexibility or
555 * restrict FSETOWN to the current process or process
556 * group for maximum safety.
558 if (proc->p_session != curproc->p_session)
562 } else /* if (pgid < 0) */ {
563 pgrp = pgfind(-pgid);
568 * Policy - Don't allow a process to FSETOWN a process
569 * in another session.
571 * Remove this test to allow maximum flexibility or
572 * restrict FSETOWN to the current process or process
573 * group for maximum safety.
575 if (pgrp->pg_session != curproc->p_session)
581 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
583 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
584 sigio->sio_proc = proc;
586 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
587 sigio->sio_pgrp = pgrp;
589 sigio->sio_pgid = pgid;
590 sigio->sio_ucred = crhold(curproc->p_ucred);
591 /* It would be convenient if p_ruid was in ucred. */
592 sigio->sio_ruid = curproc->p_ucred->cr_ruid;
593 sigio->sio_myref = sigiop;
601 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
604 fgetown(struct sigio *sigio)
606 return (sigio != NULL ? sigio->sio_pgid : 0);
610 * Close a file descriptor.
614 close(struct close_args *uap)
616 struct thread *td = curthread;
617 struct proc *p = td->td_proc;
618 struct filedesc *fdp;
627 if ((unsigned)fd >= fdp->fd_nfiles ||
628 (fp = fdp->fd_ofiles[fd]) == NULL)
631 if (fdp->fd_ofileflags[fd] & UF_MAPPED)
632 (void) munmapfd(p, fd);
634 fdp->fd_ofiles[fd] = NULL;
635 fdp->fd_ofileflags[fd] = 0;
637 if (p->p_fdtol != NULL) {
639 * Ask fdfree() to sleep to ensure that all relevant
640 * process leaders can be traversed in closef().
642 fdp->fd_holdleaderscount++;
647 * we now hold the fp reference that used to be owned by the descriptor
650 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
652 if (fd < fdp->fd_freefile)
653 fdp->fd_freefile = fd;
654 if (fd < fdp->fd_knlistsize)
655 knote_fdclose(p, fd);
656 error = closef(fp, td);
658 fdp->fd_holdleaderscount--;
659 if (fdp->fd_holdleaderscount == 0 &&
660 fdp->fd_holdleaderswakeup != 0) {
661 fdp->fd_holdleaderswakeup = 0;
662 wakeup(&fdp->fd_holdleaderscount);
669 kern_fstat(int fd, struct stat *ub)
671 struct thread *td = curthread;
672 struct proc *p = td->td_proc;
673 struct filedesc *fdp;
680 if ((unsigned)fd >= fdp->fd_nfiles ||
681 (fp = fdp->fd_ofiles[fd]) == NULL)
684 error = fo_stat(fp, ub, td);
691 * Return status information about a file descriptor.
694 fstat(struct fstat_args *uap)
699 error = kern_fstat(uap->fd, &st);
702 error = copyout(&st, uap->sb, sizeof(st));
707 * XXX: This is for source compatibility with NetBSD. Probably doesn't
711 nfstat(struct nfstat_args *uap)
717 error = kern_fstat(uap->fd, &st);
721 error = copyout(&nst, uap->sb, sizeof (nst));
727 * Return pathconf information about a file descriptor.
731 fpathconf(struct fpathconf_args *uap)
733 struct thread *td = curthread;
734 struct proc *p = td->td_proc;
735 struct filedesc *fdp;
742 if ((unsigned)uap->fd >= fdp->fd_nfiles ||
743 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
748 switch (fp->f_type) {
751 if (uap->name != _PC_PIPE_BUF) {
754 uap->sysmsg_result = PIPE_BUF;
760 vp = (struct vnode *)fp->f_data;
761 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
772 * Allocate a file descriptor for the process.
775 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
778 fdalloc(struct proc *p, int want, int *result)
780 struct filedesc *fdp = p->p_fd;
782 int lim, last, nfiles;
783 struct file **newofile;
787 * Search for a free descriptor starting at the higher
788 * of want or fd_freefile. If that fails, consider
789 * expanding the ofile array.
791 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
793 last = min(fdp->fd_nfiles, lim);
794 if ((i = want) < fdp->fd_freefile)
795 i = fdp->fd_freefile;
796 for (; i < last; i++) {
797 if (fdp->fd_ofiles[i] == NULL) {
798 fdp->fd_ofileflags[i] = 0;
799 if (i > fdp->fd_lastfile)
800 fdp->fd_lastfile = i;
801 if (want <= fdp->fd_freefile)
802 fdp->fd_freefile = i;
809 * No space in current array. Expand?
811 if (fdp->fd_nfiles >= lim)
813 if (fdp->fd_nfiles < NDEXTENT)
816 nfiles = 2 * fdp->fd_nfiles;
817 newofile = malloc(nfiles * OFILESIZE, M_FILEDESC, M_WAITOK);
820 * deal with file-table extend race that might have occured
821 * when malloc was blocked.
823 if (fdp->fd_nfiles >= nfiles) {
824 free(newofile, M_FILEDESC);
827 newofileflags = (char *) &newofile[nfiles];
829 * Copy the existing ofile and ofileflags arrays
830 * and zero the new portion of each array.
832 bcopy(fdp->fd_ofiles, newofile,
833 (i = sizeof(struct file *) * fdp->fd_nfiles));
834 bzero((char *)newofile + i, nfiles * sizeof(struct file *) - i);
835 bcopy(fdp->fd_ofileflags, newofileflags,
836 (i = sizeof(char) * fdp->fd_nfiles));
837 bzero(newofileflags + i, nfiles * sizeof(char) - i);
838 if (fdp->fd_nfiles > NDFILE)
839 free(fdp->fd_ofiles, M_FILEDESC);
840 fdp->fd_ofiles = newofile;
841 fdp->fd_ofileflags = newofileflags;
842 fdp->fd_nfiles = nfiles;
849 * Check to see whether n user file descriptors
850 * are available to the process p.
853 fdavail(struct proc *p, int n)
855 struct filedesc *fdp = p->p_fd;
859 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
860 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
863 last = min(fdp->fd_nfiles, lim);
864 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
865 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
866 if (*fpp == NULL && --n <= 0)
874 * Create a new open file structure and allocate a file decriptor
875 * for the process that refers to it. If p is NULL, no descriptor
876 * is allocated and the file pointer is returned unassociated with
877 * any process. resultfd is only used if p is not NULL and may
878 * separately be NULL indicating that you don't need the returned fd.
881 falloc(struct proc *p, struct file **resultfp, int *resultfd)
883 struct file *fp, *fq;
885 static struct timeval lastfail;
888 if (nfiles >= maxfiles - maxfilesrootres &&
889 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) {
890 if (ppsratecheck(&lastfail, &curfail, 1)) {
891 printf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n",
892 (p ? p->p_ucred->cr_ruid : -1));
897 * Allocate a new file descriptor.
898 * If the process has file descriptor zero open, add to the list
899 * of open files at that point, otherwise put it at the front of
900 * the list of open files.
903 fp = malloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
906 * wait until after malloc (which may have blocked) returns before
907 * allocating the slot, else a race might have shrunk it if we had
908 * allocated it before the malloc.
911 if (p && (error = fdalloc(p, 0, &i))) {
917 fp->f_ops = &badfileops;
920 fp->f_cred = crhold(p->p_ucred);
921 if ((fq = p->p_fd->fd_ofiles[0]) != NULL) {
922 LIST_INSERT_AFTER(fq, fp, f_list);
924 LIST_INSERT_HEAD(&filehead, fp, f_list);
926 p->p_fd->fd_ofiles[i] = fp;
928 fp->f_cred = crhold(proc0.p_ucred);
929 LIST_INSERT_HEAD(&filehead, fp, f_list);
939 fsetcred(struct file *fp, struct ucred *cr)
947 * Free a file descriptor.
950 ffree(struct file *fp)
952 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
953 LIST_REMOVE(fp, f_list);
960 * Build a new filedesc structure.
963 fdinit(struct proc *p)
965 struct filedesc0 *newfdp;
966 struct filedesc *fdp = p->p_fd;
968 newfdp = malloc(sizeof(struct filedesc0), M_FILEDESC, M_WAITOK|M_ZERO);
969 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
970 if (newfdp->fd_fd.fd_cdir)
971 vref(newfdp->fd_fd.fd_cdir);
972 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
973 vref(newfdp->fd_fd.fd_rdir);
974 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
975 if (newfdp->fd_fd.fd_jdir)
976 vref(newfdp->fd_fd.fd_jdir);
978 /* Create the file descriptor table. */
979 newfdp->fd_fd.fd_refcnt = 1;
980 newfdp->fd_fd.fd_cmask = cmask;
981 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
982 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
983 newfdp->fd_fd.fd_nfiles = NDFILE;
984 newfdp->fd_fd.fd_knlistsize = -1;
986 return (&newfdp->fd_fd);
990 * Share a filedesc structure.
993 fdshare(struct proc *p)
995 p->p_fd->fd_refcnt++;
1000 * Copy a filedesc structure.
1003 fdcopy(struct proc *p)
1005 struct filedesc *newfdp, *fdp = p->p_fd;
1009 /* Certain daemons might not have file descriptors. */
1013 newfdp = malloc(sizeof(struct filedesc0), M_FILEDESC, M_WAITOK);
1014 bcopy(fdp, newfdp, sizeof(struct filedesc));
1015 if (newfdp->fd_cdir)
1016 vref(newfdp->fd_cdir);
1018 * We must check for fd_rdir here, at least for now because
1019 * the init process is created before we have access to the
1020 * rootvode to take a reference to it.
1022 if (newfdp->fd_rdir)
1023 vref(newfdp->fd_rdir);
1024 if (newfdp->fd_jdir)
1025 vref(newfdp->fd_jdir);
1026 newfdp->fd_refcnt = 1;
1029 * If the number of open files fits in the internal arrays
1030 * of the open file structure, use them, otherwise allocate
1031 * additional memory for the number of descriptors currently
1034 if (newfdp->fd_lastfile < NDFILE) {
1035 newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
1036 newfdp->fd_ofileflags =
1037 ((struct filedesc0 *) newfdp)->fd_dfileflags;
1041 * Compute the smallest multiple of NDEXTENT needed
1042 * for the file descriptors currently in use,
1043 * allowing the table to shrink.
1045 i = newfdp->fd_nfiles;
1046 while (i > 2 * NDEXTENT && i > newfdp->fd_lastfile * 2)
1048 newfdp->fd_ofiles = malloc(i * OFILESIZE, M_FILEDESC, M_WAITOK);
1049 newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
1051 newfdp->fd_nfiles = i;
1052 bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **));
1053 bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char));
1056 * kq descriptors cannot be copied.
1058 if (newfdp->fd_knlistsize != -1) {
1059 fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile];
1060 for (i = newfdp->fd_lastfile; i >= 0; i--, fpp--) {
1061 if (*fpp != NULL && (*fpp)->f_type == DTYPE_KQUEUE) {
1063 if (i < newfdp->fd_freefile)
1064 newfdp->fd_freefile = i;
1066 if (*fpp == NULL && i == newfdp->fd_lastfile && i > 0)
1067 newfdp->fd_lastfile--;
1069 newfdp->fd_knlist = NULL;
1070 newfdp->fd_knlistsize = -1;
1071 newfdp->fd_knhash = NULL;
1072 newfdp->fd_knhashmask = 0;
1075 fpp = newfdp->fd_ofiles;
1076 for (i = newfdp->fd_lastfile; i-- >= 0; fpp++) {
1084 * Release a filedesc structure.
1087 fdfree(struct proc *p)
1089 struct thread *td = p->p_thread;
1090 struct filedesc *fdp = p->p_fd;
1093 struct filedesc_to_leader *fdtol;
1098 /* Certain daemons might not have file descriptors. */
1102 /* Check for special need to clear POSIX style locks */
1104 if (fdtol != NULL) {
1105 KASSERT(fdtol->fdl_refcount > 0,
1106 ("filedesc_to_refcount botch: fdl_refcount=%d",
1107 fdtol->fdl_refcount));
1108 if (fdtol->fdl_refcount == 1 &&
1109 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1111 fpp = fdp->fd_ofiles;
1112 for (i = 0, fpp = fdp->fd_ofiles;
1113 i <= fdp->fd_lastfile;
1116 (*fpp)->f_type != DTYPE_VNODE)
1120 lf.l_whence = SEEK_SET;
1123 lf.l_type = F_UNLCK;
1124 vp = (struct vnode *)fp->f_data;
1125 (void) VOP_ADVLOCK(vp,
1126 (caddr_t)p->p_leader,
1130 fdrop(fp, p->p_thread);
1131 fpp = fdp->fd_ofiles + i;
1135 if (fdtol->fdl_refcount == 1) {
1136 if (fdp->fd_holdleaderscount > 0 &&
1137 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1139 * close() or do_dup() has cleared a reference
1140 * in a shared file descriptor table.
1142 fdp->fd_holdleaderswakeup = 1;
1143 tsleep(&fdp->fd_holdleaderscount,
1147 if (fdtol->fdl_holdcount > 0) {
1149 * Ensure that fdtol->fdl_leader
1150 * remains valid in closef().
1152 fdtol->fdl_wakeup = 1;
1153 tsleep(fdtol, 0, "fdlhold", 0);
1157 fdtol->fdl_refcount--;
1158 if (fdtol->fdl_refcount == 0 &&
1159 fdtol->fdl_holdcount == 0) {
1160 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1161 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1166 free(fdtol, M_FILEDESC_TO_LEADER);
1168 if (--fdp->fd_refcnt > 0)
1171 * we are the last reference to the structure, we can
1172 * safely assume it will not change out from under us.
1174 fpp = fdp->fd_ofiles;
1175 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1177 (void) closef(*fpp, td);
1179 if (fdp->fd_nfiles > NDFILE)
1180 free(fdp->fd_ofiles, M_FILEDESC);
1182 vrele(fdp->fd_cdir);
1183 vrele(fdp->fd_rdir);
1185 vrele(fdp->fd_jdir);
1187 free(fdp->fd_knlist, M_KQUEUE);
1189 free(fdp->fd_knhash, M_KQUEUE);
1190 free(fdp, M_FILEDESC);
1194 * For setugid programs, we don't want to people to use that setugidness
1195 * to generate error messages which write to a file which otherwise would
1196 * otherwise be off-limits to the process.
1198 * This is a gross hack to plug the hole. A better solution would involve
1199 * a special vop or other form of generalized access control mechanism. We
1200 * go ahead and just reject all procfs file systems accesses as dangerous.
1202 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1203 * sufficient. We also don't for check setugidness since we know we are.
1206 is_unsafe(struct file *fp)
1208 if (fp->f_type == DTYPE_VNODE &&
1209 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
1215 * Make this setguid thing safe, if at all possible.
1218 setugidsafety(struct proc *p)
1220 struct thread *td = p->p_thread;
1221 struct filedesc *fdp = p->p_fd;
1224 /* Certain daemons might not have file descriptors. */
1229 * note: fdp->fd_ofiles may be reallocated out from under us while
1230 * we are blocked in a close. Be careful!
1232 for (i = 0; i <= fdp->fd_lastfile; i++) {
1235 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1239 if ((fdp->fd_ofileflags[i] & UF_MAPPED) != 0)
1240 (void) munmapfd(p, i);
1242 if (i < fdp->fd_knlistsize)
1243 knote_fdclose(p, i);
1245 * NULL-out descriptor prior to close to avoid
1246 * a race while close blocks.
1248 fp = fdp->fd_ofiles[i];
1249 fdp->fd_ofiles[i] = NULL;
1250 fdp->fd_ofileflags[i] = 0;
1251 if (i < fdp->fd_freefile)
1252 fdp->fd_freefile = i;
1253 (void) closef(fp, td);
1256 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
1261 * Close any files on exec?
1264 fdcloseexec(struct proc *p)
1266 struct thread *td = p->p_thread;
1267 struct filedesc *fdp = p->p_fd;
1270 /* Certain daemons might not have file descriptors. */
1275 * We cannot cache fd_ofiles or fd_ofileflags since operations
1276 * may block and rip them out from under us.
1278 for (i = 0; i <= fdp->fd_lastfile; i++) {
1279 if (fdp->fd_ofiles[i] != NULL &&
1280 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1284 if (fdp->fd_ofileflags[i] & UF_MAPPED)
1285 (void) munmapfd(p, i);
1287 if (i < fdp->fd_knlistsize)
1288 knote_fdclose(p, i);
1290 * NULL-out descriptor prior to close to avoid
1291 * a race while close blocks.
1293 fp = fdp->fd_ofiles[i];
1294 fdp->fd_ofiles[i] = NULL;
1295 fdp->fd_ofileflags[i] = 0;
1296 if (i < fdp->fd_freefile)
1297 fdp->fd_freefile = i;
1298 (void) closef(fp, td);
1301 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
1306 * It is unsafe for set[ug]id processes to be started with file
1307 * descriptors 0..2 closed, as these descriptors are given implicit
1308 * significance in the Standard C library. fdcheckstd() will create a
1309 * descriptor referencing /dev/null for each of stdin, stdout, and
1310 * stderr that is not already open.
1313 fdcheckstd(struct proc *p)
1315 struct thread *td = p->p_thread;
1316 struct nameidata nd;
1317 struct filedesc *fdp;
1320 int fd, i, error, flags, devnull;
1327 for (i = 0; i < 3; i++) {
1328 if (fdp->fd_ofiles[i] != NULL)
1331 error = falloc(p, &fp, &fd);
1334 NDINIT(&nd, NAMEI_LOOKUP, CNP_FOLLOW, UIO_SYSSPACE,
1336 flags = FREAD | FWRITE;
1337 error = vn_open(&nd, flags, 0);
1339 fdp->fd_ofiles[i] = NULL;
1343 NDFREE(&nd, NDF_ONLY_PNBUF);
1344 fp->f_data = (caddr_t)nd.ni_vp;
1347 fp->f_type = DTYPE_VNODE;
1348 VOP_UNLOCK(nd.ni_vp, NULL, 0, td);
1351 error = kern_dup(DUP_FIXED, devnull, i, &retval);
1360 * Internal form of close.
1361 * Decrement reference count on file structure.
1362 * Note: td and/or p may be NULL when closing a file
1363 * that was being passed in a message.
1366 closef(struct file *fp, struct thread *td)
1370 struct filedesc_to_leader *fdtol;
1377 p = NULL; /* allow no proc association */
1379 p = td->td_proc; /* can also be NULL */
1382 * POSIX record locking dictates that any close releases ALL
1383 * locks owned by this process. This is handled by setting
1384 * a flag in the unlock to free ONLY locks obeying POSIX
1385 * semantics, and not to free BSD-style file locks.
1386 * If the descriptor was in a message, POSIX-style locks
1387 * aren't passed with the descriptor.
1390 fp->f_type == DTYPE_VNODE) {
1391 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
1392 lf.l_whence = SEEK_SET;
1395 lf.l_type = F_UNLCK;
1396 vp = (struct vnode *)fp->f_data;
1397 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
1401 if (fdtol != NULL) {
1403 * Handle special case where file descriptor table
1404 * is shared between multiple process leaders.
1406 for (fdtol = fdtol->fdl_next;
1407 fdtol != p->p_fdtol;
1408 fdtol = fdtol->fdl_next) {
1409 if ((fdtol->fdl_leader->p_flag &
1412 fdtol->fdl_holdcount++;
1413 lf.l_whence = SEEK_SET;
1416 lf.l_type = F_UNLCK;
1417 vp = (struct vnode *)fp->f_data;
1418 (void) VOP_ADVLOCK(vp,
1419 (caddr_t)p->p_leader,
1420 F_UNLCK, &lf, F_POSIX);
1421 fdtol->fdl_holdcount--;
1422 if (fdtol->fdl_holdcount == 0 &&
1423 fdtol->fdl_wakeup != 0) {
1424 fdtol->fdl_wakeup = 0;
1430 return (fdrop(fp, td));
1434 fdrop(struct file *fp, struct thread *td)
1440 if (--fp->f_count > 0)
1442 if (fp->f_count < 0)
1443 panic("fdrop: count < 0");
1444 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
1445 lf.l_whence = SEEK_SET;
1448 lf.l_type = F_UNLCK;
1449 vp = (struct vnode *)fp->f_data;
1450 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
1452 if (fp->f_ops != &badfileops)
1453 error = fo_close(fp, td);
1461 * Apply an advisory lock on a file descriptor.
1463 * Just attempt to get a record lock of the requested type on
1464 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
1468 flock(struct flock_args *uap)
1470 struct proc *p = curproc;
1471 struct filedesc *fdp = p->p_fd;
1476 if ((unsigned)uap->fd >= fdp->fd_nfiles ||
1477 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
1479 if (fp->f_type != DTYPE_VNODE)
1480 return (EOPNOTSUPP);
1481 vp = (struct vnode *)fp->f_data;
1482 lf.l_whence = SEEK_SET;
1485 if (uap->how & LOCK_UN) {
1486 lf.l_type = F_UNLCK;
1487 fp->f_flag &= ~FHASLOCK;
1488 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK));
1490 if (uap->how & LOCK_EX)
1491 lf.l_type = F_WRLCK;
1492 else if (uap->how & LOCK_SH)
1493 lf.l_type = F_RDLCK;
1496 fp->f_flag |= FHASLOCK;
1497 if (uap->how & LOCK_NB)
1498 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK));
1499 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT));
1503 * File Descriptor pseudo-device driver (/dev/fd/).
1505 * Opening minor device N dup()s the file (if any) connected to file
1506 * descriptor N belonging to the calling process. Note that this driver
1507 * consists of only the ``open()'' routine, because all subsequent
1508 * references to this file will be direct to the other driver.
1512 fdopen(dev_t dev, int mode, int type, struct thread *td)
1514 KKASSERT(td->td_proc != NULL);
1517 * XXX Kludge: set curproc->p_dupfd to contain the value of the
1518 * the file descriptor being sought for duplication. The error
1519 * return ensures that the vnode for this device will be released
1520 * by vn_open. Open will detect this special error and take the
1521 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
1522 * will simply report the error.
1524 td->td_proc->p_dupfd = minor(dev);
1529 * Duplicate the specified descriptor to a free descriptor.
1532 dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error)
1538 * If the to-be-dup'd fd number is greater than the allowed number
1539 * of file descriptors, or the fd to be dup'd has already been
1540 * closed, then reject.
1542 if ((u_int)dfd >= fdp->fd_nfiles ||
1543 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
1548 * There are two cases of interest here.
1550 * For ENODEV simply dup (dfd) to file descriptor
1551 * (indx) and return.
1553 * For ENXIO steal away the file structure from (dfd) and
1554 * store it in (indx). (dfd) is effectively closed by
1557 * Any other error code is just returned.
1562 * Check that the mode the file is being opened for is a
1563 * subset of the mode of the existing descriptor.
1565 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag)
1567 fp = fdp->fd_ofiles[indx];
1569 if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED)
1570 (void) munmapfd(p, indx);
1572 fdp->fd_ofiles[indx] = wfp;
1573 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1575 if (indx > fdp->fd_lastfile)
1576 fdp->fd_lastfile = indx;
1578 * we now own the reference to fp that the ofiles[] array
1579 * used to own. Release it.
1582 fdrop(fp, curthread);
1587 * Steal away the file pointer from dfd, and stuff it into indx.
1589 fp = fdp->fd_ofiles[indx];
1591 if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED)
1592 (void) munmapfd(p, indx);
1594 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
1595 fdp->fd_ofiles[dfd] = NULL;
1596 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1597 fdp->fd_ofileflags[dfd] = 0;
1600 * we now own the reference to fp that the ofiles[] array
1601 * used to own. Release it.
1604 fdrop(fp, curthread);
1606 * Complete the clean up of the filedesc structure by
1607 * recomputing the various hints.
1609 if (indx > fdp->fd_lastfile) {
1610 fdp->fd_lastfile = indx;
1612 while (fdp->fd_lastfile > 0 &&
1613 fdp->fd_ofiles[fdp->fd_lastfile] == NULL) {
1616 if (dfd < fdp->fd_freefile)
1617 fdp->fd_freefile = dfd;
1628 struct filedesc_to_leader *
1629 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
1630 struct proc *leader)
1632 struct filedesc_to_leader *fdtol;
1634 fdtol = malloc(sizeof(struct filedesc_to_leader),
1635 M_FILEDESC_TO_LEADER, M_WAITOK);
1636 fdtol->fdl_refcount = 1;
1637 fdtol->fdl_holdcount = 0;
1638 fdtol->fdl_wakeup = 0;
1639 fdtol->fdl_leader = leader;
1641 fdtol->fdl_next = old->fdl_next;
1642 fdtol->fdl_prev = old;
1643 old->fdl_next = fdtol;
1644 fdtol->fdl_next->fdl_prev = fdtol;
1646 fdtol->fdl_next = fdtol;
1647 fdtol->fdl_prev = fdtol;
1653 * Get file structures.
1656 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
1663 * overestimate by 10 files
1665 return (SYSCTL_OUT(req, 0, sizeof(filehead) +
1666 (nfiles + 10) * sizeof(struct file)));
1669 error = SYSCTL_OUT(req, (caddr_t)&filehead, sizeof(filehead));
1674 * followed by an array of file structures
1676 LIST_FOREACH(fp, &filehead, f_list) {
1677 error = SYSCTL_OUT(req, (caddr_t)fp, sizeof (struct file));
1684 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
1685 0, 0, sysctl_kern_file, "S,file", "Entire file table");
1687 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
1688 &maxfilesperproc, 0, "Maximum files allowed open per process");
1690 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
1691 &maxfiles, 0, "Maximum number of files");
1693 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
1694 &maxfilesrootres, 0, "Descriptors reserved for root use");
1696 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
1697 &nfiles, 0, "System-wide number of open files");
1700 fildesc_drvinit(void *unused)
1704 cdevsw_add(&fildesc_cdevsw, 0, 0);
1705 for (fd = 0; fd < NUMFDESC; fd++) {
1706 make_dev(&fildesc_cdevsw, fd,
1707 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
1709 make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
1710 make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
1711 make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
1714 struct fileops badfileops = {
1738 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td)
1744 badfo_poll(struct file *fp, int events, struct ucred *cred, struct thread *td)
1750 badfo_kqfilter(struct file *fp, struct knote *kn)
1756 badfo_stat(struct file *fp, struct stat *sb, struct thread *td)
1762 badfo_close(struct file *fp, struct thread *td)
1767 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
1768 fildesc_drvinit,NULL)