2 * Copyright (c) 1982, 1986, 1989, 1993, 1995
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 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
39 * $FreeBSD: src/sys/ufs/ufs/ufs_vnops.c,v 1.131.2.8 2003/01/02 17:26:19 bde Exp $
40 * $DragonFly: src/sys/vfs/ufs/ufs_vnops.c,v 1.67 2008/09/28 05:04:22 dillon Exp $
43 #include "opt_quota.h"
44 #include "opt_suiddir.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
55 #include <sys/namei.h>
56 #include <sys/mount.h>
57 #include <sys/unistd.h>
58 #include <sys/vnode.h>
59 #include <sys/malloc.h>
60 #include <sys/dirent.h>
61 #include <sys/lockf.h>
62 #include <sys/event.h>
65 #include <sys/file.h> /* XXX */
69 #include <vm/vm_extern.h>
71 #include <vfs/fifofs/fifo.h>
77 #include "ufs_extern.h"
78 #include "ffs_extern.h"
83 static int ufs_access (struct vop_access_args *);
84 static int ufs_advlock (struct vop_advlock_args *);
85 static int ufs_chmod (struct vnode *, int, struct ucred *);
86 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *);
87 static int ufs_close (struct vop_close_args *);
88 static int ufs_create (struct vop_old_create_args *);
89 static int ufs_getattr (struct vop_getattr_args *);
90 static int ufs_link (struct vop_old_link_args *);
91 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *);
92 static int ufs_markatime (struct vop_markatime_args *);
93 static int ufs_missingop (struct vop_generic_args *ap);
94 static int ufs_mkdir (struct vop_old_mkdir_args *);
95 static int ufs_mknod (struct vop_old_mknod_args *);
96 static int ufs_mmap (struct vop_mmap_args *);
97 static int ufs_print (struct vop_print_args *);
98 static int ufs_readdir (struct vop_readdir_args *);
99 static int ufs_readlink (struct vop_readlink_args *);
100 static int ufs_remove (struct vop_old_remove_args *);
101 static int ufs_rename (struct vop_old_rename_args *);
102 static int ufs_rmdir (struct vop_old_rmdir_args *);
103 static int ufs_setattr (struct vop_setattr_args *);
104 static int ufs_strategy (struct vop_strategy_args *);
105 static int ufs_symlink (struct vop_old_symlink_args *);
106 static int ufs_whiteout (struct vop_old_whiteout_args *);
107 static int ufsfifo_close (struct vop_close_args *);
108 static int ufsfifo_kqfilter (struct vop_kqfilter_args *);
109 static int ufsfifo_read (struct vop_read_args *);
110 static int ufsfifo_write (struct vop_write_args *);
111 static int filt_ufsread (struct knote *kn, long hint);
112 static int filt_ufswrite (struct knote *kn, long hint);
113 static int filt_ufsvnode (struct knote *kn, long hint);
114 static void filt_ufsdetach (struct knote *kn);
115 static int ufs_kqfilter (struct vop_kqfilter_args *ap);
121 #define SETHIGH(q, h) { \
124 tmp.val[_QUAD_HIGHWORD] = (h); \
127 #define SETLOW(q, l) { \
130 tmp.val[_QUAD_LOWWORD] = (l); \
133 #define VN_KNOTE(vp, b) \
134 KNOTE(&vp->v_pollinfo.vpi_selinfo.si_note, (b))
136 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
139 * A virgin directory (no blushing please).
141 static struct dirtemplate mastertemplate = {
142 0, 12, DT_DIR, 1, ".",
143 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
145 static struct odirtemplate omastertemplate = {
147 0, DIRBLKSIZ - 12, 2, ".."
151 ufs_itimes(struct vnode *vp)
157 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
159 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
160 ip->i_flag |= IN_LAZYMOD;
162 ip->i_flag |= IN_MODIFIED;
163 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
165 if (ip->i_flag & IN_ACCESS) {
166 ip->i_atime = ts.tv_sec;
167 ip->i_atimensec = ts.tv_nsec;
169 if (ip->i_flag & IN_UPDATE) {
170 ip->i_mtime = ts.tv_sec;
171 ip->i_mtimensec = ts.tv_nsec;
174 if (ip->i_flag & IN_CHANGE) {
175 ip->i_ctime = ts.tv_sec;
176 ip->i_ctimensec = ts.tv_nsec;
179 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
183 * Create a regular file
185 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp,
186 * struct componentname *a_cnp, struct vattr *a_vap)
190 ufs_create(struct vop_old_create_args *ap)
195 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
196 ap->a_dvp, ap->a_vpp, ap->a_cnp);
199 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
206 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
207 * struct componentname *a_cnp, struct vattr *a_vap)
212 ufs_mknod(struct vop_old_mknod_args *ap)
214 struct vattr *vap = ap->a_vap;
215 struct vnode **vpp = ap->a_vpp;
221 * UFS cannot represent the entire major/minor range supported by
224 if (vap->va_rmajor != VNOVAL &&
225 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) {
229 /* no special directory support */
230 if (vap->va_type == VDIR)
233 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
234 ap->a_dvp, vpp, ap->a_cnp);
237 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
239 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
240 if (vap->va_rmajor != VNOVAL) {
242 * Want to be able to use this to make badblock
243 * inodes, so don't truncate the dev number.
245 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
248 * Remove inode, then reload it through VFS_VGET so it is
249 * checked to see if it is an alias of an existing entry in
252 (*vpp)->v_type = VNON;
253 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
254 vgone_vxlocked(*vpp);
256 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp);
267 * Update the times on the inode.
269 * ufs_close(struct vnode *a_vp, int a_fflag)
274 ufs_close(struct vop_close_args *ap)
276 struct vnode *vp = ap->a_vp;
278 if (vp->v_sysref.refcnt > 1)
280 return (vop_stdclose(ap));
284 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
288 ufs_access(struct vop_access_args *ap)
290 struct vnode *vp = ap->a_vp;
291 struct inode *ip = VTOI(vp);
295 if (ap->a_mode & VWRITE) {
296 switch (vp->v_type) {
300 if ((error = ufs_getinoquota(ip)) != 0)
309 error = vop_helper_access(ap, ip->i_uid, ip->i_gid, ip->i_mode, 0);
314 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap)
319 ufs_getattr(struct vop_getattr_args *ap)
321 struct vnode *vp = ap->a_vp;
322 struct inode *ip = VTOI(vp);
323 struct vattr *vap = ap->a_vap;
326 * This may cause i_fsmid to be updated even if no change (0)
327 * is returned, but we should only write out the inode if non-zero
328 * is returned and if the mount is read-write.
330 if (cache_check_fsmid_vp(vp, &ip->i_fsmid) &&
331 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0
333 ip->i_flag |= IN_LAZYMOD;
338 * Copy from inode table
340 vap->va_fsid = dev2udev(ip->i_dev);
341 vap->va_fileid = ip->i_number;
342 vap->va_mode = ip->i_mode & ~IFMT;
343 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ?
344 ip->i_effnlink : ip->i_nlink;
345 vap->va_uid = ip->i_uid;
346 vap->va_gid = ip->i_gid;
347 vap->va_rmajor = umajor(ip->i_rdev);
348 vap->va_rminor = uminor(ip->i_rdev);
349 vap->va_size = ip->i_din.di_size;
350 vap->va_atime.tv_sec = ip->i_atime;
351 vap->va_atime.tv_nsec = ip->i_atimensec;
352 vap->va_mtime.tv_sec = ip->i_mtime;
353 vap->va_mtime.tv_nsec = ip->i_mtimensec;
354 vap->va_ctime.tv_sec = ip->i_ctime;
355 vap->va_ctime.tv_nsec = ip->i_ctimensec;
356 vap->va_flags = ip->i_flags;
357 vap->va_gen = ip->i_gen;
358 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
359 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
360 vap->va_type = IFTOVT(ip->i_mode);
361 vap->va_filerev = ip->i_modrev;
362 vap->va_fsmid = ip->i_fsmid;
368 ufs_markatime(struct vop_markatime_args *ap)
370 struct vnode *vp = ap->a_vp;
371 struct inode *ip = VTOI(vp);
373 if (vp->v_mount->mnt_flag & MNT_RDONLY)
375 if (vp->v_mount->mnt_flag & MNT_NOATIME)
377 ip->i_flag |= IN_ACCESS;
378 VN_KNOTE(vp, NOTE_ATTRIB);
383 * Set attribute vnode op. called from several syscalls
385 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap,
386 * struct ucred *a_cred)
390 ufs_setattr(struct vop_setattr_args *ap)
392 struct vattr *vap = ap->a_vap;
393 struct vnode *vp = ap->a_vp;
394 struct inode *ip = VTOI(vp);
395 struct ucred *cred = ap->a_cred;
399 * Check for unsettable attributes.
401 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
402 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
403 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
404 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
407 if (vap->va_flags != VNOVAL) {
408 if (vp->v_mount->mnt_flag & MNT_RDONLY)
410 if (cred->cr_uid != ip->i_uid &&
411 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)))
414 * Note that a root chflags becomes a user chflags when
415 * we are jailed, unless the jail.chflags_allowed sysctl
418 if (cred->cr_uid == 0 &&
419 (!jailed(cred) || jail_chflags_allowed)) {
421 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) &&
424 ip->i_flags = vap->va_flags;
427 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
428 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
430 ip->i_flags &= SF_SETTABLE;
431 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
433 ip->i_flag |= IN_CHANGE;
434 if (vap->va_flags & (IMMUTABLE | APPEND))
437 if (ip->i_flags & (IMMUTABLE | APPEND))
440 * Go through the fields and update iff not VNOVAL.
442 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
443 if (vp->v_mount->mnt_flag & MNT_RDONLY)
445 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0)
448 if (vap->va_size != VNOVAL) {
450 * Disallow write attempts on read-only filesystems;
451 * unless the file is a socket, fifo, or a block or
452 * character device resident on the filesystem.
454 switch (vp->v_type) {
459 if (vp->v_mount->mnt_flag & MNT_RDONLY)
465 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0)
469 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
470 if (vp->v_mount->mnt_flag & MNT_RDONLY)
472 if (cred->cr_uid != ip->i_uid &&
473 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) &&
474 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
475 (error = VOP_EACCESS(vp, VWRITE, cred))))
477 if (vap->va_atime.tv_sec != VNOVAL)
478 ip->i_flag |= IN_ACCESS;
479 if (vap->va_mtime.tv_sec != VNOVAL)
480 ip->i_flag |= IN_CHANGE | IN_UPDATE;
482 if (vap->va_atime.tv_sec != VNOVAL) {
483 ip->i_atime = vap->va_atime.tv_sec;
484 ip->i_atimensec = vap->va_atime.tv_nsec;
486 if (vap->va_mtime.tv_sec != VNOVAL) {
487 ip->i_mtime = vap->va_mtime.tv_sec;
488 ip->i_mtimensec = vap->va_mtime.tv_nsec;
490 error = ffs_update(vp, 0);
495 if (vap->va_mode != (mode_t)VNOVAL) {
496 if (vp->v_mount->mnt_flag & MNT_RDONLY)
498 error = ufs_chmod(vp, (int)vap->va_mode, cred);
500 VN_KNOTE(vp, NOTE_ATTRIB);
505 * Change the mode on a file.
506 * Inode must be locked before calling.
509 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred)
511 struct inode *ip = VTOI(vp);
513 mode_t cur_mode = ip->i_mode;
515 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid,
520 if (cred->cr_uid != ip->i_uid) {
521 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0);
526 if (vp->v_type != VDIR && (mode & S_ISTXT))
528 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
532 ip->i_mode = cur_mode;
533 ip->i_flag |= IN_CHANGE;
538 * Perform chown operation on inode ip;
539 * inode must be locked prior to call.
542 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
544 struct inode *ip = VTOI(vp);
553 if (uid == (uid_t)VNOVAL)
555 if (gid == (gid_t)VNOVAL)
558 * If we don't own the file, are trying to change the owner
559 * of the file, or are not a member of the target group,
560 * the caller must be superuser or the call fails.
562 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
563 (gid != ip->i_gid && !(cred->cr_gid == gid ||
564 groupmember((gid_t)gid, cred)))) &&
565 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
570 if ((error = ufs_getinoquota(ip)) != 0)
573 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
574 ip->i_dquot[USRQUOTA] = NODQUOT;
577 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
578 ip->i_dquot[GRPQUOTA] = NODQUOT;
580 change = ip->i_blocks;
581 (void) ufs_chkdq(ip, -change, cred, CHOWN);
582 (void) ufs_chkiq(ip, -1, cred, CHOWN);
583 for (i = 0; i < MAXQUOTAS; i++) {
584 ufs_dqrele(vp, ip->i_dquot[i]);
585 ip->i_dquot[i] = NODQUOT;
591 if ((error = ufs_getinoquota(ip)) == 0) {
593 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
594 ip->i_dquot[USRQUOTA] = NODQUOT;
597 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
598 ip->i_dquot[GRPQUOTA] = NODQUOT;
600 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) {
601 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0)
604 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE);
606 for (i = 0; i < MAXQUOTAS; i++) {
607 ufs_dqrele(vp, ip->i_dquot[i]);
608 ip->i_dquot[i] = NODQUOT;
613 if (ufs_getinoquota(ip) == 0) {
615 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
616 ip->i_dquot[USRQUOTA] = NODQUOT;
619 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
620 ip->i_dquot[GRPQUOTA] = NODQUOT;
622 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN);
623 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN);
624 (void) ufs_getinoquota(ip);
628 if (ufs_getinoquota(ip))
629 panic("ufs_chown: lost quota");
631 ip->i_flag |= IN_CHANGE;
632 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid))
633 ip->i_mode &= ~(ISUID | ISGID);
640 * NB Currently unsupported.
642 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
647 ufs_mmap(struct vop_mmap_args *ap)
653 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp,
654 * struct componentname *a_cnp)
658 ufs_remove(struct vop_old_remove_args *ap)
661 struct vnode *vp = ap->a_vp;
662 struct vnode *dvp = ap->a_dvp;
666 #if 0 /* handled by kernel now */
667 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
668 (VTOI(dvp)->i_flags & APPEND)) {
673 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
674 VN_KNOTE(vp, NOTE_DELETE);
675 VN_KNOTE(dvp, NOTE_WRITE);
685 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp,
686 * struct componentname *a_cnp)
690 ufs_link(struct vop_old_link_args *ap)
692 struct vnode *vp = ap->a_vp;
693 struct vnode *tdvp = ap->a_tdvp;
694 struct componentname *cnp = ap->a_cnp;
696 struct direct newdir;
699 if (tdvp->v_mount != vp->v_mount) {
703 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) {
707 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
711 #if 0 /* handled by kernel now, also DragonFly allows this */
712 if (ip->i_flags & (IMMUTABLE | APPEND)) {
719 ip->i_flag |= IN_CHANGE;
720 if (DOINGSOFTDEP(vp))
721 softdep_change_linkcnt(ip);
722 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
724 ufs_makedirentry(ip, cnp, &newdir);
725 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
731 ip->i_flag |= IN_CHANGE;
732 if (DOINGSOFTDEP(vp))
733 softdep_change_linkcnt(ip);
739 VN_KNOTE(vp, NOTE_LINK);
740 VN_KNOTE(tdvp, NOTE_WRITE);
745 * whiteout vnode call
747 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags)
751 ufs_whiteout(struct vop_old_whiteout_args *ap)
753 struct vnode *dvp = ap->a_dvp;
754 struct componentname *cnp = ap->a_cnp;
755 struct direct newdir;
758 switch (ap->a_flags) {
760 /* 4.4 format directories support whiteout operations */
761 if (dvp->v_mount->mnt_maxsymlinklen > 0)
766 /* create a new directory whiteout */
768 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
769 panic("ufs_whiteout: old format filesystem");
773 newdir.d_namlen = cnp->cn_namelen;
774 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
775 newdir.d_type = DT_WHT;
776 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
780 /* remove an existing directory whiteout */
782 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
783 panic("ufs_whiteout: old format filesystem");
786 cnp->cn_flags &= ~CNP_DOWHITEOUT;
787 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
790 panic("ufs_whiteout: unknown op");
796 * Rename system call.
797 * rename("foo", "bar");
800 * link("foo", "bar");
802 * but ``atomically''. Can't do full commit without saving state in the
803 * inode on disk which isn't feasible at this time. Best we can do is
804 * always guarantee the target exists.
806 * Basic algorithm is:
808 * 1) Bump link count on source while we're linking it to the
809 * target. This also ensure the inode won't be deleted out
810 * from underneath us while we work (it may be truncated by
811 * a concurrent `trunc' or `open' for creation).
812 * 2) Link source to destination. If destination already exists,
814 * 3) Unlink source reference to inode if still around. If a
815 * directory was moved and the parent of the destination
816 * is different from the source, patch the ".." entry in the
819 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
820 * struct componentname *a_fcnp, struct vnode *a_tdvp,
821 * struct vnode *a_tvp, struct componentname *a_tcnp)
825 ufs_rename(struct vop_old_rename_args *ap)
827 struct vnode *tvp = ap->a_tvp;
828 struct vnode *tdvp = ap->a_tdvp;
829 struct vnode *fvp = ap->a_fvp;
830 struct vnode *fdvp = ap->a_fdvp;
831 struct componentname *tcnp = ap->a_tcnp;
832 struct componentname *fcnp = ap->a_fcnp;
833 struct inode *ip, *xp, *dp;
834 struct direct newdir;
835 ino_t oldparent = 0, newparent = 0;
836 int doingdirectory = 0;
837 int error = 0, ioflag;
840 * Check for cross-device rename.
842 if ((fvp->v_mount != tdvp->v_mount) ||
843 (tvp && (fvp->v_mount != tvp->v_mount))) {
857 #if 0 /* handled by kernel now */
858 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
859 (VTOI(tdvp)->i_flags & APPEND))) {
866 * Renaming a file to itself has no effect. The upper layers should
867 * not call us in that case. Temporarily just warn if they do.
870 kprintf("ufs_rename: fvp == tvp (can't happen)\n");
875 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
879 * Note: now that fvp is locked we have to be sure to unlock it before
880 * using the 'abortit' target.
884 if (ip->i_nlink >= LINK_MAX) {
889 #if 0 /* handled by kernel now */
890 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
891 || (dp->i_flags & APPEND)) {
897 if ((ip->i_mode & IFMT) == IFDIR) {
899 * Avoid ".", "..", and aliases of "." for obvious reasons.
901 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
902 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT ||
903 (ip->i_flag & IN_RENAME)) {
908 ip->i_flag |= IN_RENAME;
909 oldparent = dp->i_number;
912 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
915 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory.
916 * Cleanup fvp requirements so we can unlock it.
918 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp
919 * is setup we can use the 'bad' target if we unlock fvp. We cannot
920 * use the abortit target anymore because of IN_RENAME.
929 * 1) Bump link count while we're moving stuff
930 * around. If we crash somewhere before
931 * completing our work, the link count
932 * may be wrong, but correctable.
936 ip->i_flag |= IN_CHANGE;
937 if (DOINGSOFTDEP(fvp))
938 softdep_change_linkcnt(ip);
939 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) |
940 DOINGASYNC(fvp)))) != 0) {
946 * If ".." must be changed (ie the directory gets a new
947 * parent) then the source directory must not be in the
948 * directory heirarchy above the target, as this would
949 * orphan everything below the source directory. Also
950 * the user must have write permission in the source so
951 * as to be able to change "..". We must repeat the call
952 * to namei, as the parent directory is unlocked by the
953 * call to checkpath().
955 error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred);
959 * We are now back to where we were in that fvp, fdvp are unlocked
960 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be
961 * set. Only the bad target or, if we clean up tvp and tdvp, the
962 * out target, may be used.
964 if (oldparent != dp->i_number)
965 newparent = dp->i_number;
966 if (doingdirectory && newparent) {
967 if (error) /* write access check above */
971 * Once we start messing with tvp and tdvp we cannot use the
972 * 'bad' target, only finish cleaning tdvp and tvp up and
973 * use the 'out' target.
975 * This cleans up tvp.
983 * This is a real mess. ufs_checkpath vput's the target
984 * directory so retain an extra ref and note that tdvp will
985 * lose its lock on return. This leaves us with one good
986 * ref after ufs_checkpath returns.
989 error = ufs_checkpath(ip, dp, tcnp->cn_cred);
990 tcnp->cn_flags |= CNP_PDIRUNLOCK;
997 * relookup no longer messes with tdvp's refs. tdvp must be
998 * unlocked on entry and will be locked on a successful
1001 error = relookup(tdvp, &tvp, tcnp);
1003 if (tcnp->cn_flags & CNP_PDIRUNLOCK)
1009 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1016 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may
1017 * be NULL (xp will also be NULL in that case), and IN_RENAME will
1018 * be set if doingdirectory. This means we can use the 'bad' target
1023 * 2) If target doesn't exist, link the target
1024 * to the source and unlink the source.
1025 * Otherwise, rewrite the target directory
1026 * entry to reference the source inode and
1027 * expunge the original entry's existence.
1030 if (dp->i_dev != ip->i_dev)
1031 panic("ufs_rename: EXDEV");
1033 * Account for ".." in new directory.
1034 * When source and destination have the same
1035 * parent we don't fool with the link count.
1037 if (doingdirectory && newparent) {
1038 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1044 dp->i_flag |= IN_CHANGE;
1045 if (DOINGSOFTDEP(tdvp))
1046 softdep_change_linkcnt(dp);
1047 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) |
1052 ufs_makedirentry(ip, tcnp, &newdir);
1053 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1055 if (doingdirectory && newparent) {
1058 dp->i_flag |= IN_CHANGE;
1059 if (DOINGSOFTDEP(tdvp))
1060 softdep_change_linkcnt(dp);
1061 (void)ffs_update(tdvp, 1);
1065 VN_KNOTE(tdvp, NOTE_WRITE);
1068 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1069 panic("ufs_rename: EXDEV");
1071 * Short circuit rename(foo, foo).
1073 if (xp->i_number == ip->i_number)
1074 panic("ufs_rename: same file");
1076 * If the parent directory is "sticky", then the user must
1077 * own the parent directory, or the destination of the rename,
1078 * otherwise the destination may not be changed (except by
1079 * root). This implements append-only directories.
1081 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1082 tcnp->cn_cred->cr_uid != dp->i_uid &&
1083 xp->i_uid != tcnp->cn_cred->cr_uid) {
1088 * Target must be empty if a directory and have no links
1089 * to it. Also, ensure source and target are compatible
1090 * (both directories, or both not directories).
1092 * Purge the file or directory being replaced from the
1095 if ((xp->i_mode&IFMT) == IFDIR) {
1096 if ((xp->i_effnlink > 2) ||
1097 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1101 if (!doingdirectory) {
1105 /* cache_purge removed - handled by VFS compat layer */
1106 } else if (doingdirectory == 0) {
1107 /* cache_purge removed - handled by VFS compat layer */
1113 * note: inode passed to ufs_dirrewrite() is 0 for a
1114 * non-directory file rename, 1 for a directory rename
1115 * in the same directory, and > 1 for an inode representing
1116 * the new directory.
1118 error = ufs_dirrewrite(dp, xp, ip->i_number,
1120 (doingdirectory && newparent) ?
1121 newparent : (ino_t)doingdirectory);
1124 if (doingdirectory) {
1127 if (DOINGSOFTDEP(tdvp))
1128 softdep_change_linkcnt(dp);
1131 if (DOINGSOFTDEP(tvp))
1132 softdep_change_linkcnt(xp);
1134 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1136 * Truncate inode. The only stuff left in the directory
1137 * is "." and "..". The "." reference is inconsequential
1138 * since we are quashing it. We have removed the "."
1139 * reference and the reference in the parent directory,
1140 * but there may be other hard links. The soft
1141 * dependency code will arrange to do these operations
1142 * after the parent directory entry has been deleted on
1143 * disk, so when running with that code we avoid doing
1148 dp->i_flag |= IN_CHANGE;
1151 xp->i_flag |= IN_CHANGE;
1152 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC;
1153 error = ffs_truncate(tvp, (off_t)0, ioflag,
1158 VN_KNOTE(tdvp, NOTE_WRITE);
1160 VN_KNOTE(tvp, NOTE_DELETE);
1166 * tvp and tdvp have been cleaned up. only fvp and fdvp (both
1167 * unlocked) remain. We are about to overwrite fvp but we have to
1168 * keep 'ip' intact so we cannot release the old fvp, which is still
1169 * refd and accessible via ap->a_fvp.
1171 * This means we cannot use either 'bad' or 'out' to cleanup any
1176 * 3) Unlink the source.
1178 fcnp->cn_flags &= ~CNP_MODMASK;
1179 fcnp->cn_flags |= CNP_LOCKPARENT;
1180 error = relookup(fdvp, &fvp, fcnp);
1181 if (error || fvp == NULL) {
1183 * From name has disappeared. IN_RENAME will not be set if
1184 * we get past the panic so we don't have to clean it up.
1187 panic("ufs_rename: lost dir entry");
1189 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1195 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1198 * fdvp and fvp are locked.
1204 * Ensure that the directory entry still exists and has not
1205 * changed while the new name has been entered. If the source is
1206 * a file then the entry may have been unlinked or renamed. In
1207 * either case there is no further work to be done. If the source
1208 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1209 * flag ensures that it cannot be moved by another rename or removed
1210 * by a rmdir. Cleanup IN_RENAME.
1214 panic("ufs_rename: lost dir entry");
1217 * If the source is a directory with a
1218 * new parent, the link count of the old
1219 * parent directory must be decremented
1220 * and ".." set to point to the new parent.
1222 if (doingdirectory && newparent) {
1223 xp->i_offset = mastertemplate.dot_reclen;
1224 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1225 /* cache_purge removed - handled by VFS compat layer */
1227 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1228 xp->i_flag &= ~IN_RENAME;
1231 VN_KNOTE(fvp, NOTE_RENAME);
1243 ip->i_flag &= ~IN_RENAME;
1244 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1247 ip->i_flag |= IN_CHANGE;
1248 ip->i_flag &= ~IN_RENAME;
1249 if (DOINGSOFTDEP(fvp))
1250 softdep_change_linkcnt(ip);
1261 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1262 * struct componentname *a_cnp, struct vattr *a_vap)
1266 ufs_mkdir(struct vop_old_mkdir_args *ap)
1268 struct vnode *dvp = ap->a_dvp;
1269 struct vattr *vap = ap->a_vap;
1270 struct componentname *cnp = ap->a_cnp;
1271 struct inode *ip, *dp;
1274 struct dirtemplate dirtemplate, *dtp;
1275 struct direct newdir;
1280 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1284 dmode = vap->va_mode & 0777;
1287 * Must simulate part of ufs_makeinode here to acquire the inode,
1288 * but not have it entered in the parent directory. The entry is
1289 * made later after writing "." and ".." entries.
1291 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1295 ip->i_gid = dp->i_gid;
1299 struct ucred ucred, *ucp;
1303 * If we are hacking owners here, (only do this where told to)
1304 * and we are not giving it TO root, (would subvert quotas)
1305 * then go ahead and give it to the other user.
1306 * The new directory also inherits the SUID bit.
1307 * If user's UID and dir UID are the same,
1308 * 'give it away' so that the SUID is still forced on.
1310 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1311 (dp->i_mode & ISUID) && dp->i_uid) {
1313 ip->i_uid = dp->i_uid;
1315 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1317 * Make sure the correct user gets charged
1319 * Make a dummy credential for the victim.
1320 * XXX This seems to never be accessed out of
1321 * our context so a stack variable is ok.
1324 ucred.cr_uid = ip->i_uid;
1325 ucred.cr_ngroups = 1;
1326 ucred.cr_groups[0] = dp->i_gid;
1331 ip->i_uid = cnp->cn_cred->cr_uid;
1333 if ((error = ufs_getinoquota(ip)) ||
1334 (error = ufs_chkiq(ip, 1, ucp, 0))) {
1335 ffs_vfree(tvp, ip->i_number, dmode);
1341 #else /* !SUIDDIR */
1342 ip->i_uid = cnp->cn_cred->cr_uid;
1344 if ((error = ufs_getinoquota(ip)) ||
1345 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
1346 ffs_vfree(tvp, ip->i_number, dmode);
1351 #endif /* !SUIDDIR */
1352 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1354 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1357 if (DOINGSOFTDEP(tvp))
1358 softdep_change_linkcnt(ip);
1359 if (cnp->cn_flags & CNP_ISWHITEOUT)
1360 ip->i_flags |= UF_OPAQUE;
1363 * Bump link count in parent directory to reflect work done below.
1364 * Should be done before reference is created so cleanup is
1365 * possible if we crash.
1369 dp->i_flag |= IN_CHANGE;
1370 if (DOINGSOFTDEP(dvp))
1371 softdep_change_linkcnt(dp);
1372 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1377 * The vnode must have a VM object in order to issue buffer cache
1380 vinitvmio(tvp, DIRBLKSIZ);
1383 * Initialize directory with "." and ".." from static template.
1385 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1386 dtp = &mastertemplate;
1388 dtp = (struct dirtemplate *)&omastertemplate;
1390 dirtemplate.dot_ino = ip->i_number;
1391 dirtemplate.dotdot_ino = dp->i_number;
1392 vnode_pager_setsize(tvp, DIRBLKSIZ);
1393 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp);
1396 ip->i_size = DIRBLKSIZ;
1397 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1398 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1399 if (DOINGSOFTDEP(tvp)) {
1401 * Ensure that the entire newly allocated block is a
1402 * valid directory so that future growth within the
1403 * block does not have to ensure that the block is
1404 * written before the inode.
1407 while (blkoff < bp->b_bcount) {
1409 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1410 blkoff += DIRBLKSIZ;
1413 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) |
1414 DOINGASYNC(tvp)))) != 0) {
1419 * Directory set up, now install its entry in the parent directory.
1421 * If we are not doing soft dependencies, then we must write out the
1422 * buffer containing the new directory body before entering the new
1423 * name in the parent. If we are doing soft dependencies, then the
1424 * buffer containing the new directory body will be passed to and
1425 * released in the soft dependency code after the code has attached
1426 * an appropriate ordering dependency to the buffer which ensures that
1427 * the buffer is written before the new name is written in the parent.
1429 if (DOINGASYNC(dvp))
1431 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0)
1433 ufs_makedirentry(ip, cnp, &newdir);
1434 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1438 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1443 dp->i_flag |= IN_CHANGE;
1444 if (DOINGSOFTDEP(dvp))
1445 softdep_change_linkcnt(dp);
1447 * No need to do an explicit VOP_TRUNCATE here, vrele will
1448 * do this for us because we set the link count to 0.
1452 ip->i_flag |= IN_CHANGE;
1453 if (DOINGSOFTDEP(tvp))
1454 softdep_change_linkcnt(ip);
1462 * Rmdir system call.
1464 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1465 * struct componentname *a_cnp)
1469 ufs_rmdir(struct vop_old_rmdir_args *ap)
1471 struct vnode *vp = ap->a_vp;
1472 struct vnode *dvp = ap->a_dvp;
1473 struct componentname *cnp = ap->a_cnp;
1474 struct inode *ip, *dp;
1481 * Do not remove a directory that is in the process of being renamed.
1482 * Verify the directory is empty (and valid). Rmdir ".." will not be
1483 * valid since ".." will contain a reference to the current directory
1484 * and thus be non-empty. Do not allow the removal of mounted on
1485 * directories (this can happen when an NFS exported filesystem
1486 * tries to remove a locally mounted on directory).
1489 if (ip->i_flag & IN_RENAME) {
1493 if (ip->i_effnlink != 2 ||
1494 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1498 #if 0 /* handled by kernel now */
1499 if ((dp->i_flags & APPEND)
1500 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1506 * Delete reference to directory before purging
1507 * inode. If we crash in between, the directory
1508 * will be reattached to lost+found,
1512 if (DOINGSOFTDEP(vp)) {
1513 softdep_change_linkcnt(dp);
1514 softdep_change_linkcnt(ip);
1516 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1520 if (DOINGSOFTDEP(vp)) {
1521 softdep_change_linkcnt(dp);
1522 softdep_change_linkcnt(ip);
1526 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1528 * Truncate inode. The only stuff left in the directory is "." and
1529 * "..". The "." reference is inconsequential since we are quashing
1530 * it. The soft dependency code will arrange to do these operations
1531 * after the parent directory entry has been deleted on disk, so
1532 * when running with that code we avoid doing them now.
1534 if (!DOINGSOFTDEP(vp)) {
1536 dp->i_flag |= IN_CHANGE;
1538 ip->i_flag |= IN_CHANGE;
1539 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
1540 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred);
1542 /* cache_purge removed - handled by VFS compat layer */
1544 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1545 if (ip->i_dirhash != NULL)
1546 ufsdirhash_free(ip);
1549 VN_KNOTE(vp, NOTE_DELETE);
1554 * symlink -- make a symbolic link
1556 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1557 * struct componentname *a_cnp, struct vattr *a_vap,
1562 ufs_symlink(struct vop_old_symlink_args *ap)
1564 struct vnode *vp, **vpp = ap->a_vpp;
1568 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1572 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1574 len = strlen(ap->a_target);
1575 if (len < vp->v_mount->mnt_maxsymlinklen) {
1577 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1579 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1582 * Make sure we have a VM object in order to use
1585 if (vp->v_object == NULL)
1587 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1588 UIO_SYSSPACE, IO_NODELOCKED,
1589 ap->a_cnp->cn_cred, NULL);
1597 * Vnode op for reading directories.
1599 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
1600 * int *a_eofflag, int *ncookies, off_t **a_cookies)
1604 ufs_readdir(struct vop_readdir_args *ap)
1606 struct uio *uio = ap->a_uio;
1607 struct vnode *vp = ap->a_vp;
1612 int offset; /* offset into buffer cache buffer */
1613 int eoffset; /* end of buffer clipped to file EOF */
1614 int pickup; /* pickup point */
1619 if (uio->uio_offset < 0)
1622 * Guess the number of cookies needed. Make sure we compute at
1623 * least 1, and no more then a reasonable limit.
1625 if (ap->a_ncookies) {
1626 ncookies = uio->uio_resid / 16 + 1;
1627 if (ncookies > 1024)
1629 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1631 ncookies = -1; /* force conditionals below */
1636 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
1642 if (uio->uio_offset >= VTOI(vp)->i_size) {
1645 if (ap->a_ncookies) {
1646 *ap->a_ncookies = cookie_index;
1647 *ap->a_cookies = cookies;
1653 * Loop until we run out of cookies, we run out of user buffer,
1654 * or we hit the directory EOF.
1656 * Always start scans at the beginning of the buffer, don't trust
1657 * the offset supplied by userland.
1659 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) {
1660 pickup = (int)(uio->uio_offset - bp->b_loffset);
1663 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size)
1664 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset);
1666 eoffset = bp->b_bcount;
1668 while (offset < eoffset) {
1669 dp = (struct direct *)(bp->b_data + offset);
1670 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) ||
1671 offset + dp->d_reclen > bp->b_bcount) {
1675 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) {
1679 if (offset < pickup) {
1680 offset += dp->d_reclen;
1683 #if BYTE_ORDER == LITTLE_ENDIAN
1685 retval = vop_write_dirent(&error, uio,
1686 dp->d_ino, dp->d_namlen, dp->d_type,
1691 retval = vop_write_dirent(&error, uio,
1692 dp->d_ino, dp->d_type, dp->d_namlen,
1698 cookies[cookie_index] = bp->b_loffset + offset;
1700 offset += dp->d_reclen;
1701 if (cookie_index == ncookies)
1706 * This will align the next loop to the beginning of the
1707 * next block, and pickup will calculate to 0.
1709 uio->uio_offset = bp->b_loffset + offset;
1712 if (retval || error || cookie_index == ncookies ||
1713 uio->uio_offset >= VTOI(vp)->i_size) {
1718 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset;
1721 * Report errors only if we didn't manage to read anything
1723 if (error && cookie_index == 0) {
1725 kfree(cookies, M_TEMP);
1726 *ap->a_ncookies = 0;
1727 *ap->a_cookies = NULL;
1732 *ap->a_ncookies = cookie_index;
1733 *ap->a_cookies = cookies;
1742 * Return target name of a symbolic link
1744 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1748 ufs_readlink(struct vop_readlink_args *ap)
1750 struct vnode *vp = ap->a_vp;
1751 struct inode *ip = VTOI(vp);
1755 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1756 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */
1757 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1762 * Perform the equivalent of an OPEN on vp so we can issue a
1765 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1769 * Calculate the logical to physical mapping if not done already,
1770 * then call the device strategy routine.
1772 * In order to be able to swap to a file, the VOP_BMAP operation may not
1773 * deadlock on memory. See ufs_bmap() for details.
1775 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio)
1779 ufs_strategy(struct vop_strategy_args *ap)
1781 struct bio *bio = ap->a_bio;
1783 struct buf *bp = bio->bio_buf;
1784 struct vnode *vp = ap->a_vp;
1789 if (vp->v_type == VBLK || vp->v_type == VCHR)
1790 panic("ufs_strategy: spec");
1791 nbio = push_bio(bio);
1792 if (nbio->bio_offset == NOOFFSET) {
1793 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset,
1794 NULL, NULL, bp->b_cmd);
1796 bp->b_error = error;
1797 bp->b_flags |= B_ERROR;
1798 /* I/O was never started on nbio, must biodone(bio) */
1802 if (nbio->bio_offset == NOOFFSET)
1805 if (nbio->bio_offset == NOOFFSET) {
1807 * We hit a hole in the file. The buffer has been zero-filled
1808 * so just biodone() it.
1812 vn_strategy(ip->i_devvp, nbio);
1818 * Print out the contents of an inode.
1820 * ufs_print(struct vnode *a_vp)
1824 ufs_print(struct vop_print_args *ap)
1826 struct vnode *vp = ap->a_vp;
1827 struct inode *ip = VTOI(vp);
1829 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)",
1830 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev),
1832 if (vp->v_type == VFIFO)
1834 lockmgr_printinfo(&vp->v_lock);
1840 * Read wrapper for fifos.
1842 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1843 * struct ucred *a_cred)
1847 ufsfifo_read(struct vop_read_args *ap)
1854 resid = uio->uio_resid;
1855 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1856 ip = VTOI(ap->a_vp);
1857 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL &&
1858 (uio->uio_resid != resid || (error == 0 && resid != 0)))
1859 VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1864 * Write wrapper for fifos.
1866 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1867 * struct ucred *a_cred)
1871 ufsfifo_write(struct vop_write_args *ap)
1878 resid = uio->uio_resid;
1879 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1880 ip = VTOI(ap->a_vp);
1881 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1882 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1887 * Close wrapper for fifos.
1889 * Update the times on the inode then do device close.
1891 * ufsfifo_close(struct vnode *a_vp, int a_fflag)
1895 ufsfifo_close(struct vop_close_args *ap)
1897 struct vnode *vp = ap->a_vp;
1899 if (vp->v_sysref.refcnt > 1)
1901 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1905 * Kqfilter wrapper for fifos.
1907 * Fall through to ufs kqfilter routines if needed
1911 ufsfifo_kqfilter(struct vop_kqfilter_args *ap)
1915 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1917 error = ufs_kqfilter(ap);
1922 * Advisory record locking support
1924 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
1929 ufs_advlock(struct vop_advlock_args *ap)
1931 struct inode *ip = VTOI(ap->a_vp);
1933 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size));
1937 * Initialize the vnode associated with a new inode, handle aliased
1940 * Make sure directories have their VM object now rather then later,
1941 * saving us from having to check on all the myrid directory VOPs
1942 * that might be executed without a VOP_OPEN being performed.
1945 ufs_vinit(struct mount *mntp, struct vnode **vpp)
1954 vp->v_type = IFTOVT(ip->i_mode);
1956 switch(vp->v_type) {
1959 vp->v_ops = &mntp->mnt_vn_spec_ops;
1960 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev));
1963 vp->v_ops = &mntp->mnt_vn_fifo_ops;
1967 vinitvmio(vp, ip->i_size);
1970 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen)
1971 vinitvmio(vp, ip->i_size);
1978 if (ip->i_number == ROOTINO)
1979 vp->v_flag |= VROOT;
1981 * Initialize modrev times
1983 getmicrouptime(&tv);
1984 SETHIGH(ip->i_modrev, tv.tv_sec);
1985 SETLOW(ip->i_modrev, tv.tv_usec * 4294);
1991 * Allocate a new inode.
1995 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1996 struct componentname *cnp)
1998 struct inode *ip, *pdir;
1999 struct direct newdir;
2005 if ((mode & IFMT) == 0)
2008 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp);
2012 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP);
2013 ip->i_gid = pdir->i_gid;
2017 struct ucred ucred, *ucp;
2021 * If we are not the owner of the directory,
2022 * and we are hacking owners here, (only do this where told to)
2023 * and we are not giving it TO root, (would subvert quotas)
2024 * then go ahead and give it to the other user.
2025 * Note that this drops off the execute bits for security.
2027 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2028 (pdir->i_mode & ISUID) &&
2029 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2030 ip->i_uid = pdir->i_uid;
2034 * Make sure the correct user gets charged
2036 * Quickly knock up a dummy credential for the victim.
2037 * XXX This seems to never be accessed out of our
2038 * context so a stack variable is ok.
2041 ucred.cr_uid = ip->i_uid;
2042 ucred.cr_ngroups = 1;
2043 ucred.cr_groups[0] = pdir->i_gid;
2047 ip->i_uid = cnp->cn_cred->cr_uid;
2050 if ((error = ufs_getinoquota(ip)) ||
2051 (error = ufs_chkiq(ip, 1, ucp, 0))) {
2052 ffs_vfree(tvp, ip->i_number, mode);
2058 #else /* !SUIDDIR */
2059 ip->i_uid = cnp->cn_cred->cr_uid;
2061 if ((error = ufs_getinoquota(ip)) ||
2062 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
2063 ffs_vfree(tvp, ip->i_number, mode);
2068 #endif /* !SUIDDIR */
2069 ip->i_fsmid = cache_getnewfsmid();
2070 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2072 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2075 if (DOINGSOFTDEP(tvp))
2076 softdep_change_linkcnt(ip);
2077 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2078 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) {
2079 ip->i_mode &= ~ISGID;
2082 if (cnp->cn_flags & CNP_ISWHITEOUT)
2083 ip->i_flags |= UF_OPAQUE;
2086 * Regular files and directories need VM objects. Softlinks do
2087 * not (not immediately anyway).
2089 if (tvp->v_type == VREG || tvp->v_type == VDIR)
2093 * Make sure inode goes to disk before directory entry.
2095 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2098 ufs_makedirentry(ip, cnp, &newdir);
2099 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2107 * Write error occurred trying to update the inode
2108 * or the directory so must deallocate the inode.
2112 ip->i_flag |= IN_CHANGE;
2113 if (DOINGSOFTDEP(tvp))
2114 softdep_change_linkcnt(ip);
2120 ufs_missingop(struct vop_generic_args *ap)
2122 panic("no vop function for %s in ufs child", ap->a_desc->sd_name);
2123 return (EOPNOTSUPP);
2126 static struct filterops ufsread_filtops =
2127 { 1, NULL, filt_ufsdetach, filt_ufsread };
2128 static struct filterops ufswrite_filtops =
2129 { 1, NULL, filt_ufsdetach, filt_ufswrite };
2130 static struct filterops ufsvnode_filtops =
2131 { 1, NULL, filt_ufsdetach, filt_ufsvnode };
2134 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn)
2137 ufs_kqfilter(struct vop_kqfilter_args *ap)
2139 struct vnode *vp = ap->a_vp;
2140 struct knote *kn = ap->a_kn;
2143 switch (kn->kn_filter) {
2145 kn->kn_fop = &ufsread_filtops;
2148 kn->kn_fop = &ufswrite_filtops;
2151 kn->kn_fop = &ufsvnode_filtops;
2157 kn->kn_hook = (caddr_t)vp;
2159 lwkt_gettoken(&vlock, &vp->v_token);
2160 SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext);
2161 lwkt_reltoken(&vlock);
2167 filt_ufsdetach(struct knote *kn)
2169 struct vnode *vp = (struct vnode *)kn->kn_hook;
2172 lwkt_gettoken(&vlock, &vp->v_token);
2173 SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note,
2174 kn, knote, kn_selnext);
2175 lwkt_reltoken(&vlock);
2180 filt_ufsread(struct knote *kn, long hint)
2182 struct vnode *vp = (struct vnode *)kn->kn_hook;
2183 struct inode *ip = VTOI(vp);
2186 * filesystem is gone, so set the EOF flag and schedule
2187 * the knote for deletion.
2189 if (hint == NOTE_REVOKE) {
2190 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2194 kn->kn_data = ip->i_size - kn->kn_fp->f_offset;
2195 return (kn->kn_data != 0);
2200 filt_ufswrite(struct knote *kn, long hint)
2203 * filesystem is gone, so set the EOF flag and schedule
2204 * the knote for deletion.
2206 if (hint == NOTE_REVOKE)
2207 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2214 filt_ufsvnode(struct knote *kn, long hint)
2216 if (kn->kn_sfflags & hint)
2217 kn->kn_fflags |= hint;
2218 if (hint == NOTE_REVOKE) {
2219 kn->kn_flags |= EV_EOF;
2222 return (kn->kn_fflags != 0);
2225 /* Global vfs data structures for ufs. */
2226 static struct vop_ops ufs_vnode_vops = {
2227 .vop_default = vop_defaultop,
2228 .vop_fsync = (void *)ufs_missingop,
2229 .vop_read = (void *)ufs_missingop,
2230 .vop_reallocblks = (void *)ufs_missingop,
2231 .vop_write = (void *)ufs_missingop,
2232 .vop_access = ufs_access,
2233 .vop_advlock = ufs_advlock,
2234 .vop_bmap = ufs_bmap,
2235 .vop_old_lookup = ufs_lookup,
2236 .vop_close = ufs_close,
2237 .vop_old_create = ufs_create,
2238 .vop_getattr = ufs_getattr,
2239 .vop_inactive = ufs_inactive,
2240 .vop_old_link = ufs_link,
2241 .vop_old_mkdir = ufs_mkdir,
2242 .vop_old_mknod = ufs_mknod,
2243 .vop_mmap = ufs_mmap,
2244 .vop_open = vop_stdopen,
2245 .vop_pathconf = vop_stdpathconf,
2246 .vop_poll = vop_stdpoll,
2247 .vop_kqfilter = ufs_kqfilter,
2248 .vop_print = ufs_print,
2249 .vop_readdir = ufs_readdir,
2250 .vop_readlink = ufs_readlink,
2251 .vop_reclaim = ufs_reclaim,
2252 .vop_old_remove = ufs_remove,
2253 .vop_old_rename = ufs_rename,
2254 .vop_old_rmdir = ufs_rmdir,
2255 .vop_setattr = ufs_setattr,
2256 .vop_markatime = ufs_markatime,
2257 .vop_strategy = ufs_strategy,
2258 .vop_old_symlink = ufs_symlink,
2259 .vop_old_whiteout = ufs_whiteout
2262 static struct vop_ops ufs_spec_vops = {
2263 .vop_default = vop_defaultop,
2264 .vop_fsync = (void *)ufs_missingop,
2265 .vop_access = ufs_access,
2266 .vop_close = ufs_close,
2267 .vop_getattr = ufs_getattr,
2268 .vop_inactive = ufs_inactive,
2269 .vop_print = ufs_print,
2270 .vop_read = vop_stdnoread,
2271 .vop_reclaim = ufs_reclaim,
2272 .vop_setattr = ufs_setattr,
2273 .vop_markatime = ufs_markatime,
2274 .vop_write = vop_stdnowrite
2277 static struct vop_ops ufs_fifo_vops = {
2278 .vop_default = fifo_vnoperate,
2279 .vop_fsync = (void *)ufs_missingop,
2280 .vop_access = ufs_access,
2281 .vop_close = ufsfifo_close,
2282 .vop_getattr = ufs_getattr,
2283 .vop_inactive = ufs_inactive,
2284 .vop_kqfilter = ufsfifo_kqfilter,
2285 .vop_print = ufs_print,
2286 .vop_read = ufsfifo_read,
2287 .vop_reclaim = ufs_reclaim,
2288 .vop_setattr = ufs_setattr,
2289 .vop_markatime = ufs_markatime,
2290 .vop_write = ufsfifo_write
2293 VNODEOP_SET(ufs_vnode_vops);
2294 VNODEOP_SET(ufs_spec_vops);
2295 VNODEOP_SET(ufs_fifo_vops);
2301 ufs_vnoperate(struct vop_generic_args *ap)
2303 return (VOCALL(&ufs_vnode_vops, ap));
2307 * ufs_vnoperatefifo()
2310 ufs_vnoperatefifo(struct vop_generic_args *ap)
2312 return (VOCALL(&ufs_fifo_vops, ap));
2316 * ufs_vnoperatespec()
2319 ufs_vnoperatespec(struct vop_generic_args *ap)
2321 return (VOCALL(&ufs_spec_vops, ap));