2 * modified for EXT2FS support in Lites 1.1
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
8 * Copyright (c) 1982, 1986, 1989, 1993
9 * The Regents of the University of California. All rights reserved.
10 * (c) UNIX System Laboratories, Inc.
11 * All or some portions of this file are derived from material licensed
12 * to the University of California by American Telephone and Telegraph
13 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14 * the permission of UNIX System Laboratories, Inc.
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * @(#)ufs_vnops.c 8.7 (Berkeley) 2/3/94
41 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
45 #include "opt_suiddir.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
51 #include <sys/filio.h>
55 #include <sys/endian.h>
57 #include <sys/mount.h>
58 #include <sys/unistd.h>
60 #include <sys/vnode.h>
61 #include <sys/namei.h>
62 #include <sys/lockf.h>
63 #include <sys/event.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vnode_pager.h>
73 #include "opt_directio.h"
75 #include <fs/fifofs/fifo.h>
77 #include <ufs/ufs/dir.h>
79 #include <fs/ext2fs/fs.h>
80 #include <fs/ext2fs/inode.h>
81 #include <fs/ext2fs/ext2_extern.h>
82 #include <fs/ext2fs/ext2fs.h>
83 #include <fs/ext2fs/ext2_dinode.h>
84 #include <fs/ext2fs/ext2_dir.h>
85 #include <fs/ext2fs/ext2_mount.h>
87 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
88 static void ext2_itimes_locked(struct vnode *);
90 static vop_access_t ext2_access;
91 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
92 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
94 static vop_close_t ext2_close;
95 static vop_create_t ext2_create;
96 static vop_fsync_t ext2_fsync;
97 static vop_getattr_t ext2_getattr;
98 static vop_ioctl_t ext2_ioctl;
99 static vop_link_t ext2_link;
100 static vop_mkdir_t ext2_mkdir;
101 static vop_mknod_t ext2_mknod;
102 static vop_open_t ext2_open;
103 static vop_pathconf_t ext2_pathconf;
104 static vop_print_t ext2_print;
105 static vop_read_t ext2_read;
106 static vop_readlink_t ext2_readlink;
107 static vop_remove_t ext2_remove;
108 static vop_rename_t ext2_rename;
109 static vop_rmdir_t ext2_rmdir;
110 static vop_setattr_t ext2_setattr;
111 static vop_strategy_t ext2_strategy;
112 static vop_symlink_t ext2_symlink;
113 static vop_write_t ext2_write;
114 static vop_vptofh_t ext2_vptofh;
115 static vop_close_t ext2fifo_close;
116 static vop_kqfilter_t ext2fifo_kqfilter;
118 /* Global vfs data structures for ext2. */
119 struct vop_vector ext2_vnodeops = {
120 .vop_default = &default_vnodeops,
121 .vop_access = ext2_access,
122 .vop_bmap = ext2_bmap,
123 .vop_cachedlookup = ext2_lookup,
124 .vop_close = ext2_close,
125 .vop_create = ext2_create,
126 .vop_fsync = ext2_fsync,
127 .vop_getattr = ext2_getattr,
128 .vop_inactive = ext2_inactive,
129 .vop_ioctl = ext2_ioctl,
130 .vop_link = ext2_link,
131 .vop_lookup = vfs_cache_lookup,
132 .vop_mkdir = ext2_mkdir,
133 .vop_mknod = ext2_mknod,
134 .vop_open = ext2_open,
135 .vop_pathconf = ext2_pathconf,
136 .vop_poll = vop_stdpoll,
137 .vop_print = ext2_print,
138 .vop_read = ext2_read,
139 .vop_readdir = ext2_readdir,
140 .vop_readlink = ext2_readlink,
141 .vop_reallocblks = ext2_reallocblks,
142 .vop_reclaim = ext2_reclaim,
143 .vop_remove = ext2_remove,
144 .vop_rename = ext2_rename,
145 .vop_rmdir = ext2_rmdir,
146 .vop_setattr = ext2_setattr,
147 .vop_strategy = ext2_strategy,
148 .vop_symlink = ext2_symlink,
149 .vop_write = ext2_write,
150 .vop_vptofh = ext2_vptofh,
153 struct vop_vector ext2_fifoops = {
154 .vop_default = &fifo_specops,
155 .vop_access = ext2_access,
156 .vop_close = ext2fifo_close,
157 .vop_fsync = ext2_fsync,
158 .vop_getattr = ext2_getattr,
159 .vop_inactive = ext2_inactive,
160 .vop_kqfilter = ext2fifo_kqfilter,
161 .vop_print = ext2_print,
162 .vop_read = VOP_PANIC,
163 .vop_reclaim = ext2_reclaim,
164 .vop_setattr = ext2_setattr,
165 .vop_write = VOP_PANIC,
166 .vop_vptofh = ext2_vptofh,
170 * A virgin directory (no blushing please).
171 * Note that the type and namlen fields are reversed relative to ext2.
172 * Also, we don't use `struct odirtemplate', since it would just cause
173 * endianness problems.
175 static struct dirtemplate mastertemplate = {
176 0, 12, 1, EXT2_FT_DIR, ".",
177 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
179 static struct dirtemplate omastertemplate = {
180 0, 12, 1, EXT2_FT_UNKNOWN, ".",
181 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
185 ext2_itimes_locked(struct vnode *vp)
190 ASSERT_VI_LOCKED(vp, __func__);
193 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
195 if ((vp->v_type == VBLK || vp->v_type == VCHR))
196 ip->i_flag |= IN_LAZYMOD;
198 ip->i_flag |= IN_MODIFIED;
199 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
201 if (ip->i_flag & IN_ACCESS) {
202 ip->i_atime = ts.tv_sec;
203 ip->i_atimensec = ts.tv_nsec;
205 if (ip->i_flag & IN_UPDATE) {
206 ip->i_mtime = ts.tv_sec;
207 ip->i_mtimensec = ts.tv_nsec;
210 if (ip->i_flag & IN_CHANGE) {
211 ip->i_ctime = ts.tv_sec;
212 ip->i_ctimensec = ts.tv_nsec;
215 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
219 ext2_itimes(struct vnode *vp)
223 ext2_itimes_locked(vp);
228 * Create a regular file
231 ext2_create(struct vop_create_args *ap)
236 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
237 ap->a_dvp, ap->a_vpp, ap->a_cnp);
244 ext2_open(struct vop_open_args *ap)
247 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
251 * Files marked append-only must be opened for appending.
253 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
254 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
257 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
265 * Update the times on the inode.
268 ext2_close(struct vop_close_args *ap)
270 struct vnode *vp = ap->a_vp;
273 if (vp->v_usecount > 1)
274 ext2_itimes_locked(vp);
280 ext2_access(struct vop_access_args *ap)
282 struct vnode *vp = ap->a_vp;
283 struct inode *ip = VTOI(vp);
284 accmode_t accmode = ap->a_accmode;
287 if (vp->v_type == VBLK || vp->v_type == VCHR)
291 * Disallow write attempts on read-only file systems;
292 * unless the file is a socket, fifo, or a block or
293 * character device resident on the file system.
295 if (accmode & VWRITE) {
296 switch (vp->v_type) {
300 if (vp->v_mount->mnt_flag & MNT_RDONLY)
308 /* If immutable bit set, nobody gets to write it. */
309 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
312 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
313 ap->a_accmode, ap->a_cred, NULL);
318 ext2_getattr(struct vop_getattr_args *ap)
320 struct vnode *vp = ap->a_vp;
321 struct inode *ip = VTOI(vp);
322 struct vattr *vap = ap->a_vap;
326 * Copy from inode table
328 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
329 vap->va_fileid = ip->i_number;
330 vap->va_mode = ip->i_mode & ~IFMT;
331 vap->va_nlink = ip->i_nlink;
332 vap->va_uid = ip->i_uid;
333 vap->va_gid = ip->i_gid;
334 vap->va_rdev = ip->i_rdev;
335 vap->va_size = ip->i_size;
336 vap->va_atime.tv_sec = ip->i_atime;
337 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
338 vap->va_mtime.tv_sec = ip->i_mtime;
339 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
340 vap->va_ctime.tv_sec = ip->i_ctime;
341 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
342 if E2DI_HAS_XTIME(ip) {
343 vap->va_birthtime.tv_sec = ip->i_birthtime;
344 vap->va_birthtime.tv_nsec = ip->i_birthnsec;
346 vap->va_flags = ip->i_flags;
347 vap->va_gen = ip->i_gen;
348 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
349 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
350 vap->va_type = IFTOVT(ip->i_mode);
351 vap->va_filerev = ip->i_modrev;
356 * Set attribute vnode op. called from several syscalls
359 ext2_setattr(struct vop_setattr_args *ap)
361 struct vattr *vap = ap->a_vap;
362 struct vnode *vp = ap->a_vp;
363 struct inode *ip = VTOI(vp);
364 struct ucred *cred = ap->a_cred;
365 struct thread *td = curthread;
369 * Check for unsettable attributes.
371 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
372 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
373 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
374 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
377 if (vap->va_flags != VNOVAL) {
378 /* Disallow flags not supported by ext2fs. */
379 if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
382 if (vp->v_mount->mnt_flag & MNT_RDONLY)
385 * Callers may only modify the file flags on objects they
386 * have VADMIN rights for.
388 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
391 * Unprivileged processes and privileged processes in
392 * jail() are not permitted to unset system flags, or
393 * modify flags if any system flags are set.
394 * Privileged non-jail processes may not modify system flags
395 * if securelevel > 0 and any existing system flags are set.
397 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
399 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
400 error = securelevel_gt(cred, 0);
404 ip->i_flags = vap->va_flags;
407 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
408 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
410 ip->i_flags &= SF_SETTABLE;
411 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
413 ip->i_flag |= IN_CHANGE;
414 if (ip->i_flags & (IMMUTABLE | APPEND))
417 if (ip->i_flags & (IMMUTABLE | APPEND))
420 * Go through the fields and update iff not VNOVAL.
422 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
423 if (vp->v_mount->mnt_flag & MNT_RDONLY)
425 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
429 if (vap->va_size != VNOVAL) {
431 * Disallow write attempts on read-only file systems;
432 * unless the file is a socket, fifo, or a block or
433 * character device resident on the file system.
435 switch (vp->v_type) {
440 if (vp->v_mount->mnt_flag & MNT_RDONLY)
446 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
449 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
450 if (vp->v_mount->mnt_flag & MNT_RDONLY)
454 * If times is NULL, ... The caller must be the owner of
455 * the file, have permission to write the file, or be the
457 * If times is non-NULL, ... The caller must be the owner of
458 * the file or be the super-user.
460 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
461 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
462 (error = VOP_ACCESS(vp, VWRITE, cred, td))))
464 if (vap->va_atime.tv_sec != VNOVAL)
465 ip->i_flag |= IN_ACCESS;
466 if (vap->va_mtime.tv_sec != VNOVAL)
467 ip->i_flag |= IN_CHANGE | IN_UPDATE;
469 if (vap->va_atime.tv_sec != VNOVAL) {
470 ip->i_atime = vap->va_atime.tv_sec;
471 ip->i_atimensec = vap->va_atime.tv_nsec;
473 if (vap->va_mtime.tv_sec != VNOVAL) {
474 ip->i_mtime = vap->va_mtime.tv_sec;
475 ip->i_mtimensec = vap->va_mtime.tv_nsec;
477 ip->i_birthtime = vap->va_birthtime.tv_sec;
478 ip->i_birthnsec = vap->va_birthtime.tv_nsec;
479 error = ext2_update(vp, 0);
484 if (vap->va_mode != (mode_t)VNOVAL) {
485 if (vp->v_mount->mnt_flag & MNT_RDONLY)
487 error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
493 * Change the mode on a file.
494 * Inode must be locked before calling.
497 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
499 struct inode *ip = VTOI(vp);
503 * To modify the permissions on a file, must possess VADMIN
506 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
509 * Privileged processes may set the sticky bit on non-directories,
510 * as well as set the setgid bit on a file with a group that the
511 * process is not a member of.
513 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
514 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
518 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
519 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
523 ip->i_mode &= ~ALLPERMS;
524 ip->i_mode |= (mode & ALLPERMS);
525 ip->i_flag |= IN_CHANGE;
530 * Perform chown operation on inode ip;
531 * inode must be locked prior to call.
534 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
537 struct inode *ip = VTOI(vp);
542 if (uid == (uid_t)VNOVAL)
544 if (gid == (gid_t)VNOVAL)
547 * To modify the ownership of a file, must possess VADMIN
550 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
553 * To change the owner of a file, or change the group of a file
554 * to a group of which we are not a member, the caller must
557 if (uid != ip->i_uid || (gid != ip->i_gid &&
558 !groupmember(gid, cred))) {
559 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
567 ip->i_flag |= IN_CHANGE;
568 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
569 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
570 ip->i_mode &= ~(ISUID | ISGID);
576 * Synch an open file.
580 ext2_fsync(struct vop_fsync_args *ap)
583 * Flush all dirty buffers associated with a vnode.
588 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
596 ext2_mknod(struct vop_mknod_args *ap)
598 struct vattr *vap = ap->a_vap;
599 struct vnode **vpp = ap->a_vpp;
604 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
605 ap->a_dvp, vpp, ap->a_cnp);
609 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
610 if (vap->va_rdev != VNOVAL) {
612 * Want to be able to use this to make badblock
613 * inodes, so don't truncate the dev number.
615 ip->i_rdev = vap->va_rdev;
618 * Remove inode, then reload it through VFS_VGET so it is
619 * checked to see if it is an alias of an existing entry in
620 * the inode cache. XXX I don't believe this is necessary now.
622 (*vpp)->v_type = VNON;
623 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
626 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
635 ext2_remove(struct vop_remove_args *ap)
638 struct vnode *vp = ap->a_vp;
639 struct vnode *dvp = ap->a_dvp;
643 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
644 (VTOI(dvp)->i_flags & APPEND)) {
648 error = ext2_dirremove(dvp, ap->a_cnp);
651 ip->i_flag |= IN_CHANGE;
661 ext2_link(struct vop_link_args *ap)
663 struct vnode *vp = ap->a_vp;
664 struct vnode *tdvp = ap->a_tdvp;
665 struct componentname *cnp = ap->a_cnp;
670 if ((cnp->cn_flags & HASBUF) == 0)
671 panic("ext2_link: no name");
673 if (tdvp->v_mount != vp->v_mount) {
678 if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
682 if (ip->i_flags & (IMMUTABLE | APPEND)) {
687 ip->i_flag |= IN_CHANGE;
688 error = ext2_update(vp, !DOINGASYNC(vp));
690 error = ext2_direnter(ip, tdvp, cnp);
693 ip->i_flag |= IN_CHANGE;
700 * Rename system call.
701 * rename("foo", "bar");
704 * link("foo", "bar");
706 * but ``atomically''. Can't do full commit without saving state in the
707 * inode on disk which isn't feasible at this time. Best we can do is
708 * always guarantee the target exists.
710 * Basic algorithm is:
712 * 1) Bump link count on source while we're linking it to the
713 * target. This also ensure the inode won't be deleted out
714 * from underneath us while we work (it may be truncated by
715 * a concurrent `trunc' or `open' for creation).
716 * 2) Link source to destination. If destination already exists,
718 * 3) Unlink source reference to inode if still around. If a
719 * directory was moved and the parent of the destination
720 * is different from the source, patch the ".." entry in the
724 ext2_rename(struct vop_rename_args *ap)
726 struct vnode *tvp = ap->a_tvp;
727 struct vnode *tdvp = ap->a_tdvp;
728 struct vnode *fvp = ap->a_fvp;
729 struct vnode *fdvp = ap->a_fdvp;
730 struct componentname *tcnp = ap->a_tcnp;
731 struct componentname *fcnp = ap->a_fcnp;
732 struct inode *ip, *xp, *dp;
733 struct dirtemplate dirbuf;
734 int doingdirectory = 0, oldparent = 0, newparent = 0;
739 if ((tcnp->cn_flags & HASBUF) == 0 ||
740 (fcnp->cn_flags & HASBUF) == 0)
741 panic("ext2_rename: no name");
744 * Check for cross-device rename.
746 if ((fvp->v_mount != tdvp->v_mount) ||
747 (tvp && (fvp->v_mount != tvp->v_mount))) {
761 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
762 (VTOI(tdvp)->i_flags & APPEND))) {
768 * Renaming a file to itself has no effect. The upper layers should
769 * not call us in that case. Temporarily just warn if they do.
772 printf("ext2_rename: fvp == tvp (can't happen)\n");
777 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
781 if (ip->i_nlink >= EXT2_LINK_MAX) {
786 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
787 || (dp->i_flags & APPEND)) {
792 if ((ip->i_mode & IFMT) == IFDIR) {
794 * Avoid ".", "..", and aliases of "." for obvious reasons.
796 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
797 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
798 (ip->i_flag & IN_RENAME)) {
803 ip->i_flag |= IN_RENAME;
804 oldparent = dp->i_number;
810 * When the target exists, both the directory
811 * and target vnodes are returned locked.
819 * 1) Bump link count while we're moving stuff
820 * around. If we crash somewhere before
821 * completing our work, the link count
822 * may be wrong, but correctable.
825 ip->i_flag |= IN_CHANGE;
826 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
832 * If ".." must be changed (ie the directory gets a new
833 * parent) then the source directory must not be in the
834 * directory hierarchy above the target, as this would
835 * orphan everything below the source directory. Also
836 * the user must have write permission in the source so
837 * as to be able to change "..". We must repeat the call
838 * to namei, as the parent directory is unlocked by the
839 * call to checkpath().
841 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
843 if (oldparent != dp->i_number)
844 newparent = dp->i_number;
845 if (doingdirectory && newparent) {
846 if (error) /* write access check above */
850 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
854 error = relookup(tdvp, &tvp, tcnp);
864 * 2) If target doesn't exist, link the target
865 * to the source and unlink the source.
866 * Otherwise, rewrite the target directory
867 * entry to reference the source inode and
868 * expunge the original entry's existence.
871 if (dp->i_devvp != ip->i_devvp)
872 panic("ext2_rename: EXDEV");
874 * Account for ".." in new directory.
875 * When source and destination have the same
876 * parent we don't fool with the link count.
878 if (doingdirectory && newparent) {
879 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
884 dp->i_flag |= IN_CHANGE;
885 error = ext2_update(tdvp, !DOINGASYNC(tdvp));
889 error = ext2_direnter(ip, tdvp, tcnp);
891 if (doingdirectory && newparent) {
893 dp->i_flag |= IN_CHANGE;
894 (void)ext2_update(tdvp, 1);
900 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
901 panic("ext2_rename: EXDEV");
903 * Short circuit rename(foo, foo).
905 if (xp->i_number == ip->i_number)
906 panic("ext2_rename: same file");
908 * If the parent directory is "sticky", then the user must
909 * own the parent directory, or the destination of the rename,
910 * otherwise the destination may not be changed (except by
911 * root). This implements append-only directories.
913 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
914 tcnp->cn_cred->cr_uid != dp->i_uid &&
915 xp->i_uid != tcnp->cn_cred->cr_uid) {
920 * Target must be empty if a directory and have no links
921 * to it. Also, ensure source and target are compatible
922 * (both directories, or both not directories).
924 if ((xp->i_mode&IFMT) == IFDIR) {
925 if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
930 if (!doingdirectory) {
935 } else if (doingdirectory) {
939 error = ext2_dirrewrite(dp, ip, tcnp);
943 * If the target directory is in the same
944 * directory as the source directory,
945 * decrement the link count on the parent
946 * of the target directory.
948 if (doingdirectory && !newparent) {
950 dp->i_flag |= IN_CHANGE;
954 * Adjust the link count of the target to
955 * reflect the dirrewrite above. If this is
956 * a directory it is empty and there are
957 * no links to it, so we can squash the inode and
958 * any space associated with it. We disallowed
959 * renaming over top of a directory with links to
960 * it above, as the remaining link would point to
961 * a directory without "." or ".." entries.
964 if (doingdirectory) {
965 if (--xp->i_nlink != 0)
966 panic("ext2_rename: linked directory");
967 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
968 tcnp->cn_cred, tcnp->cn_thread);
970 xp->i_flag |= IN_CHANGE;
976 * 3) Unlink the source.
978 fcnp->cn_flags &= ~MODMASK;
979 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
981 error = relookup(fdvp, &fvp, fcnp);
989 * From name has disappeared.
992 panic("ext2_rename: lost dir entry");
997 * Ensure that the directory entry still exists and has not
998 * changed while the new name has been entered. If the source is
999 * a file then the entry may have been unlinked or renamed. In
1000 * either case there is no further work to be done. If the source
1001 * is a directory then it cannot have been rmdir'ed; its link
1002 * count of three would cause a rmdir to fail with ENOTEMPTY.
1003 * The IN_RENAME flag ensures that it cannot be moved by another
1008 panic("ext2_rename: lost dir entry");
1011 * If the source is a directory with a
1012 * new parent, the link count of the old
1013 * parent directory must be decremented
1014 * and ".." set to point to the new parent.
1016 if (doingdirectory && newparent) {
1018 dp->i_flag |= IN_CHANGE;
1019 error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1020 sizeof(struct dirtemplate), (off_t)0,
1021 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1022 tcnp->cn_cred, NOCRED, NULL, NULL);
1024 /* Like ufs little-endian: */
1025 namlen = dirbuf.dotdot_type;
1027 dirbuf.dotdot_name[0] != '.' ||
1028 dirbuf.dotdot_name[1] != '.') {
1029 ext2_dirbad(xp, (doff_t)12,
1030 "rename: mangled dir");
1032 dirbuf.dotdot_ino = newparent;
1033 (void) vn_rdwr(UIO_WRITE, fvp,
1035 sizeof(struct dirtemplate),
1036 (off_t)0, UIO_SYSSPACE,
1037 IO_NODELOCKED | IO_SYNC |
1038 IO_NOMACCHECK, tcnp->cn_cred,
1039 NOCRED, NULL, NULL);
1044 error = ext2_dirremove(fdvp, fcnp);
1047 xp->i_flag |= IN_CHANGE;
1049 xp->i_flag &= ~IN_RENAME;
1064 ip->i_flag &= ~IN_RENAME;
1065 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1067 ip->i_flag |= IN_CHANGE;
1068 ip->i_flag &= ~IN_RENAME;
1079 ext2_mkdir(struct vop_mkdir_args *ap)
1081 struct vnode *dvp = ap->a_dvp;
1082 struct vattr *vap = ap->a_vap;
1083 struct componentname *cnp = ap->a_cnp;
1084 struct inode *ip, *dp;
1086 struct dirtemplate dirtemplate, *dtp;
1090 if ((cnp->cn_flags & HASBUF) == 0)
1091 panic("ext2_mkdir: no name");
1094 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1098 dmode = vap->va_mode & 0777;
1101 * Must simulate part of ext2_makeinode here to acquire the inode,
1102 * but not have it entered in the parent directory. The entry is
1103 * made later after writing "." and ".." entries.
1105 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1109 ip->i_gid = dp->i_gid;
1113 * if we are hacking owners here, (only do this where told to)
1114 * and we are not giving it TOO root, (would subvert quotas)
1115 * then go ahead and give it to the other user.
1116 * The new directory also inherits the SUID bit.
1117 * If user's UID and dir UID are the same,
1118 * 'give it away' so that the SUID is still forced on.
1120 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1121 (dp->i_mode & ISUID) && dp->i_uid) {
1123 ip->i_uid = dp->i_uid;
1125 ip->i_uid = cnp->cn_cred->cr_uid;
1129 ip->i_uid = cnp->cn_cred->cr_uid;
1131 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1133 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1135 if (cnp->cn_flags & ISWHITEOUT)
1136 ip->i_flags |= UF_OPAQUE;
1137 error = ext2_update(tvp, 1);
1140 * Bump link count in parent directory
1141 * to reflect work done below. Should
1142 * be done before reference is created
1143 * so reparation is possible if we crash.
1146 dp->i_flag |= IN_CHANGE;
1147 error = ext2_update(dvp, !DOINGASYNC(dvp));
1151 /* Initialize directory with "." and ".." from static template. */
1152 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1153 EXT2F_INCOMPAT_FTYPE))
1154 dtp = &mastertemplate;
1156 dtp = &omastertemplate;
1158 dirtemplate.dot_ino = ip->i_number;
1159 dirtemplate.dotdot_ino = dp->i_number;
1160 /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1161 * so let's just redefine it - for this function only
1164 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize
1165 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1166 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1167 sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1168 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1172 dp->i_flag |= IN_CHANGE;
1175 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1176 /* XXX should grow with balloc() */
1177 panic("ext2_mkdir: blksize");
1179 ip->i_size = DIRBLKSIZ;
1180 ip->i_flag |= IN_CHANGE;
1183 /* Directory set up, now install its entry in the parent directory. */
1184 error = ext2_direnter(ip, dvp, cnp);
1187 dp->i_flag |= IN_CHANGE;
1191 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1192 * for us because we set the link count to 0.
1196 ip->i_flag |= IN_CHANGE;
1203 #define DIRBLKSIZ DEV_BSIZE
1207 * Rmdir system call.
1210 ext2_rmdir(struct vop_rmdir_args *ap)
1212 struct vnode *vp = ap->a_vp;
1213 struct vnode *dvp = ap->a_dvp;
1214 struct componentname *cnp = ap->a_cnp;
1215 struct inode *ip, *dp;
1222 * Verify the directory is empty (and valid).
1223 * (Rmdir ".." won't be valid since
1224 * ".." will contain a reference to
1225 * the current directory and thus be
1229 if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1233 if ((dp->i_flags & APPEND)
1234 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1239 * Delete reference to directory before purging
1240 * inode. If we crash in between, the directory
1241 * will be reattached to lost+found,
1243 error = ext2_dirremove(dvp, cnp);
1247 dp->i_flag |= IN_CHANGE;
1251 * Truncate inode. The only stuff left
1252 * in the directory is "." and "..". The
1253 * "." reference is inconsequential since
1254 * we're quashing it. The ".." reference
1255 * has already been adjusted above. We've
1256 * removed the "." reference and the reference
1257 * in the parent directory, but there may be
1258 * other hard links so decrement by 2 and
1259 * worry about them later.
1262 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1264 cache_purge(ITOV(ip));
1265 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1267 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1268 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1275 * symlink -- make a symbolic link
1278 ext2_symlink(struct vop_symlink_args *ap)
1280 struct vnode *vp, **vpp = ap->a_vpp;
1284 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1289 len = strlen(ap->a_target);
1290 if (len < vp->v_mount->mnt_maxsymlinklen) {
1292 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1294 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1296 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1297 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1298 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1305 * Return target name of a symbolic link
1308 ext2_readlink(struct vop_readlink_args *ap)
1310 struct vnode *vp = ap->a_vp;
1311 struct inode *ip = VTOI(vp);
1315 if (isize < vp->v_mount->mnt_maxsymlinklen) {
1316 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1319 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1323 * Calculate the logical to physical mapping if not done already,
1324 * then call the device strategy routine.
1326 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1327 * deadlock on memory. See ext2_bmap() for details.
1330 ext2_strategy(struct vop_strategy_args *ap)
1332 struct buf *bp = ap->a_bp;
1333 struct vnode *vp = ap->a_vp;
1340 if (vp->v_type == VBLK || vp->v_type == VCHR)
1341 panic("ext2_strategy: spec");
1342 if (bp->b_blkno == bp->b_lblkno) {
1343 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1344 bp->b_blkno = blkno;
1346 bp->b_error = error;
1347 bp->b_ioflags |= BIO_ERROR;
1351 if ((long)bp->b_blkno == -1)
1354 if ((long)bp->b_blkno == -1) {
1358 bp->b_iooffset = dbtob(bp->b_blkno);
1359 bo = VFSTOEXT2(vp->v_mount)->um_bo;
1360 BO_STRATEGY(bo, bp);
1365 * Print out the contents of an inode.
1368 ext2_print(struct vop_print_args *ap)
1370 struct vnode *vp = ap->a_vp;
1371 struct inode *ip = VTOI(vp);
1373 vn_printf(ip->i_devvp, "\tino %lu", (u_long)ip->i_number);
1374 if (vp->v_type == VFIFO)
1381 * Close wrapper for fifos.
1383 * Update the times on the inode then do device close.
1386 ext2fifo_close(struct vop_close_args *ap)
1388 struct vnode *vp = ap->a_vp;
1391 if (vp->v_usecount > 1)
1392 ext2_itimes_locked(vp);
1394 return (fifo_specops.vop_close(ap));
1398 * Kqfilter wrapper for fifos.
1400 * Fall through to ext2 kqfilter routines if needed
1403 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1407 error = fifo_specops.vop_kqfilter(ap);
1409 error = vfs_kqfilter(ap);
1414 * Return POSIX pathconf information applicable to ext2 filesystems.
1417 ext2_pathconf(struct vop_pathconf_args *ap)
1420 switch (ap->a_name) {
1422 *ap->a_retval = EXT2_LINK_MAX;
1425 *ap->a_retval = NAME_MAX;
1428 *ap->a_retval = PATH_MAX;
1431 *ap->a_retval = PIPE_BUF;
1433 case _PC_CHOWN_RESTRICTED:
1439 case _PC_MIN_HOLE_SIZE:
1440 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1449 * Vnode pointer to File handle
1453 ext2_vptofh(struct vop_vptofh_args *ap)
1458 ip = VTOI(ap->a_vp);
1459 ufhp = (struct ufid *)ap->a_fhp;
1460 ufhp->ufid_len = sizeof(struct ufid);
1461 ufhp->ufid_ino = ip->i_number;
1462 ufhp->ufid_gen = ip->i_gen;
1467 * Initialize the vnode associated with a new inode, handle aliased
1471 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1478 vp->v_type = IFTOVT(ip->i_mode);
1479 if (vp->v_type == VFIFO)
1482 if (ip->i_number == EXT2_ROOTINO)
1483 vp->v_vflag |= VV_ROOT;
1484 ip->i_modrev = init_va_filerev();
1490 * Allocate a new inode.
1493 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1494 struct componentname *cnp)
1496 struct inode *ip, *pdir;
1502 if ((cnp->cn_flags & HASBUF) == 0)
1503 panic("ext2_makeinode: no name");
1506 if ((mode & IFMT) == 0)
1509 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1514 ip->i_gid = pdir->i_gid;
1519 * not the owner of the directory,
1520 * and we are hacking owners here, (only do this where told to)
1521 * and we are not giving it TOO root, (would subvert quotas)
1522 * then go ahead and give it to the other user.
1523 * Note that this drops off the execute bits for security.
1525 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1526 (pdir->i_mode & ISUID) &&
1527 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1528 ip->i_uid = pdir->i_uid;
1531 ip->i_uid = cnp->cn_cred->cr_uid;
1535 ip->i_uid = cnp->cn_cred->cr_uid;
1537 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1539 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
1541 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1542 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1543 ip->i_mode &= ~ISGID;
1546 if (cnp->cn_flags & ISWHITEOUT)
1547 ip->i_flags |= UF_OPAQUE;
1550 * Make sure inode goes to disk before directory entry.
1552 error = ext2_update(tvp, !DOINGASYNC(tvp));
1555 error = ext2_direnter(ip, dvp, cnp);
1564 * Write error occurred trying to update the inode
1565 * or the directory so must deallocate the inode.
1568 ip->i_flag |= IN_CHANGE;
1574 * Vnode op for reading.
1577 ext2_read(struct vop_read_args *ap)
1582 struct m_ext2fs *fs;
1584 daddr_t lbn, nextlbn;
1586 long size, xfersize, blkoffset;
1587 int error, orig_resid, seqcount;
1592 ioflag = ap->a_ioflag;
1594 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1598 if (uio->uio_rw != UIO_READ)
1599 panic("%s: mode", "ext2_read");
1601 if (vp->v_type == VLNK) {
1602 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1603 panic("%s: short symlink", "ext2_read");
1604 } else if (vp->v_type != VREG && vp->v_type != VDIR)
1605 panic("%s: type %d", "ext2_read", vp->v_type);
1607 orig_resid = uio->uio_resid;
1608 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1609 if (orig_resid == 0)
1611 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1613 if (uio->uio_offset < ip->i_size &&
1614 uio->uio_offset >= fs->e2fs_maxfilesize)
1617 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1618 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1620 lbn = lblkno(fs, uio->uio_offset);
1622 size = blksize(fs, ip, lbn);
1623 blkoffset = blkoff(fs, uio->uio_offset);
1625 xfersize = fs->e2fs_fsize - blkoffset;
1626 if (uio->uio_resid < xfersize)
1627 xfersize = uio->uio_resid;
1628 if (bytesinfile < xfersize)
1629 xfersize = bytesinfile;
1631 if (lblktosize(fs, nextlbn) >= ip->i_size)
1632 error = bread(vp, lbn, size, NOCRED, &bp);
1633 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0)
1634 error = cluster_read(vp, ip->i_size, lbn, size,
1635 NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
1636 else if (seqcount > 1) {
1637 int nextsize = blksize(fs, ip, nextlbn);
1638 error = breadn(vp, lbn,
1639 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1641 error = bread(vp, lbn, size, NOCRED, &bp);
1649 * If IO_DIRECT then set B_DIRECT for the buffer. This
1650 * will cause us to attempt to release the buffer later on
1651 * and will cause the buffer cache to attempt to free the
1654 if (ioflag & IO_DIRECT)
1655 bp->b_flags |= B_DIRECT;
1658 * We should only get non-zero b_resid when an I/O error
1659 * has occurred, which should cause us to break above.
1660 * However, if the short read did not cause an error,
1661 * then we want to ensure that we do not uiomove bad
1662 * or uninitialized data.
1664 size -= bp->b_resid;
1665 if (size < xfersize) {
1670 error = uiomove((char *)bp->b_data + blkoffset,
1671 (int)xfersize, uio);
1675 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1677 * If it's VMIO or direct I/O, then we don't
1678 * need the buf, mark it available for
1679 * freeing. If it's non-direct VMIO, the VM has
1682 bp->b_flags |= B_RELBUF;
1686 * Otherwise let whoever
1687 * made the request take care of
1688 * freeing it. We just queue
1689 * it onto another list.
1696 * This can only happen in the case of an error
1697 * because the loop above resets bp to NULL on each iteration
1698 * and on normal completion has not set a new value into it.
1699 * so it must have come from a 'break' statement
1702 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1703 bp->b_flags |= B_RELBUF;
1710 if ((error == 0 || uio->uio_resid != orig_resid) &&
1711 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1712 ip->i_flag |= IN_ACCESS;
1717 ext2_ioctl(struct vop_ioctl_args *ap)
1720 switch (ap->a_command) {
1723 return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1724 (off_t *)ap->a_data, ap->a_cred));
1731 * Vnode op for writing.
1734 ext2_write(struct vop_write_args *ap)
1739 struct m_ext2fs *fs;
1743 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1745 ioflag = ap->a_ioflag;
1749 seqcount = ioflag >> IO_SEQSHIFT;
1753 if (uio->uio_rw != UIO_WRITE)
1754 panic("%s: mode", "ext2_write");
1757 switch (vp->v_type) {
1759 if (ioflag & IO_APPEND)
1760 uio->uio_offset = ip->i_size;
1761 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1767 /* XXX differs from ffs -- this is called from ext2_mkdir(). */
1768 if ((ioflag & IO_SYNC) == 0)
1769 panic("ext2_write: nonsync dir write");
1772 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
1773 vp->v_type, (intmax_t)uio->uio_offset,
1774 (intmax_t)uio->uio_resid);
1777 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
1778 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
1780 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
1783 * Maybe this should be above the vnode op call, but so long as
1784 * file servers have no limits, I don't think it matters.
1786 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
1789 resid = uio->uio_resid;
1791 if (seqcount > BA_SEQMAX)
1792 flags = BA_SEQMAX << BA_SEQSHIFT;
1794 flags = seqcount << BA_SEQSHIFT;
1795 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1798 for (error = 0; uio->uio_resid > 0;) {
1799 lbn = lblkno(fs, uio->uio_offset);
1800 blkoffset = blkoff(fs, uio->uio_offset);
1801 xfersize = fs->e2fs_fsize - blkoffset;
1802 if (uio->uio_resid < xfersize)
1803 xfersize = uio->uio_resid;
1804 if (uio->uio_offset + xfersize > ip->i_size)
1805 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
1808 * We must perform a read-before-write if the transfer size
1809 * does not cover the entire buffer.
1811 if (fs->e2fs_bsize > xfersize)
1814 flags &= ~BA_CLRBUF;
1815 error = ext2_balloc(ip, lbn, blkoffset + xfersize,
1816 ap->a_cred, &bp, flags);
1820 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
1821 bp->b_flags |= B_NOCACHE;
1822 if (uio->uio_offset + xfersize > ip->i_size)
1823 ip->i_size = uio->uio_offset + xfersize;
1824 size = blksize(fs, ip, lbn) - bp->b_resid;
1825 if (size < xfersize)
1829 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1831 * If the buffer is not already filled and we encounter an
1832 * error while trying to fill it, we have to clear out any
1833 * garbage data from the pages instantiated for the buffer.
1834 * If we do not, a failed uiomove() during a write can leave
1835 * the prior contents of the pages exposed to a userland mmap.
1837 * Note that we need only clear buffers with a transfer size
1838 * equal to the block size because buffers with a shorter
1839 * transfer size were cleared above by the call to ext2_balloc()
1840 * with the BA_CLRBUF flag set.
1842 * If the source region for uiomove identically mmaps the
1843 * buffer, uiomove() performed the NOP copy, and the buffer
1844 * content remains valid because the page fault handler
1845 * validated the pages.
1847 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
1848 fs->e2fs_bsize == xfersize)
1850 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1851 bp->b_flags |= B_RELBUF;
1855 * If IO_SYNC each buffer is written synchronously. Otherwise
1856 * if we have a severe page deficiency write the buffer
1857 * asynchronously. Otherwise try to cluster, and if that
1858 * doesn't do it then either do an async write (if O_DIRECT),
1859 * or a delayed write (if not).
1861 if (ioflag & IO_SYNC) {
1863 } else if (vm_page_count_severe() ||
1864 buf_dirty_count_severe() ||
1865 (ioflag & IO_ASYNC)) {
1866 bp->b_flags |= B_CLUSTEROK;
1868 } else if (xfersize + blkoffset == fs->e2fs_fsize) {
1869 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
1870 bp->b_flags |= B_CLUSTEROK;
1871 cluster_write(vp, bp, ip->i_size, seqcount);
1875 } else if (ioflag & IO_DIRECT) {
1876 bp->b_flags |= B_CLUSTEROK;
1879 bp->b_flags |= B_CLUSTEROK;
1882 if (error || xfersize == 0)
1886 * If we successfully wrote any data, and we are not the superuser
1887 * we clear the setuid and setgid bits as a precaution against
1890 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
1892 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
1893 ip->i_mode &= ~(ISUID | ISGID);
1896 if (ioflag & IO_UNIT) {
1897 (void)ext2_truncate(vp, osize,
1898 ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
1899 uio->uio_offset -= resid - uio->uio_resid;
1900 uio->uio_resid = resid;
1903 if (uio->uio_resid != resid) {
1904 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1905 if (ioflag & IO_SYNC)
1906 error = ext2_update(vp, 1);