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>
54 #include <sys/endian.h>
56 #include <sys/mount.h>
57 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/namei.h>
61 #include <sys/lockf.h>
62 #include <sys/event.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vnode_pager.h>
72 #include "opt_directio.h"
74 #include <fs/fifofs/fifo.h>
76 #include <ufs/ufs/dir.h>
78 #include <fs/ext2fs/fs.h>
79 #include <fs/ext2fs/inode.h>
80 #include <fs/ext2fs/ext2_extern.h>
81 #include <fs/ext2fs/ext2fs.h>
82 #include <fs/ext2fs/ext2_dinode.h>
83 #include <fs/ext2fs/ext2_dir.h>
84 #include <fs/ext2fs/ext2_mount.h>
86 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
87 static void ext2_itimes_locked(struct vnode *);
89 static vop_access_t ext2_access;
90 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
91 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
93 static vop_close_t ext2_close;
94 static vop_create_t ext2_create;
95 static vop_fsync_t ext2_fsync;
96 static vop_getattr_t ext2_getattr;
97 static vop_link_t ext2_link;
98 static vop_mkdir_t ext2_mkdir;
99 static vop_mknod_t ext2_mknod;
100 static vop_open_t ext2_open;
101 static vop_pathconf_t ext2_pathconf;
102 static vop_print_t ext2_print;
103 static vop_read_t ext2_read;
104 static vop_readlink_t ext2_readlink;
105 static vop_remove_t ext2_remove;
106 static vop_rename_t ext2_rename;
107 static vop_rmdir_t ext2_rmdir;
108 static vop_setattr_t ext2_setattr;
109 static vop_strategy_t ext2_strategy;
110 static vop_symlink_t ext2_symlink;
111 static vop_write_t ext2_write;
112 static vop_vptofh_t ext2_vptofh;
113 static vop_close_t ext2fifo_close;
114 static vop_kqfilter_t ext2fifo_kqfilter;
116 /* Global vfs data structures for ext2. */
117 struct vop_vector ext2_vnodeops = {
118 .vop_default = &default_vnodeops,
119 .vop_access = ext2_access,
120 .vop_bmap = ext2_bmap,
121 .vop_cachedlookup = ext2_lookup,
122 .vop_close = ext2_close,
123 .vop_create = ext2_create,
124 .vop_fsync = ext2_fsync,
125 .vop_getattr = ext2_getattr,
126 .vop_inactive = ext2_inactive,
127 .vop_link = ext2_link,
128 .vop_lookup = vfs_cache_lookup,
129 .vop_mkdir = ext2_mkdir,
130 .vop_mknod = ext2_mknod,
131 .vop_open = ext2_open,
132 .vop_pathconf = ext2_pathconf,
133 .vop_poll = vop_stdpoll,
134 .vop_print = ext2_print,
135 .vop_read = ext2_read,
136 .vop_readdir = ext2_readdir,
137 .vop_readlink = ext2_readlink,
138 .vop_reallocblks = ext2_reallocblks,
139 .vop_reclaim = ext2_reclaim,
140 .vop_remove = ext2_remove,
141 .vop_rename = ext2_rename,
142 .vop_rmdir = ext2_rmdir,
143 .vop_setattr = ext2_setattr,
144 .vop_strategy = ext2_strategy,
145 .vop_symlink = ext2_symlink,
146 .vop_write = ext2_write,
147 .vop_vptofh = ext2_vptofh,
150 struct vop_vector ext2_fifoops = {
151 .vop_default = &fifo_specops,
152 .vop_access = ext2_access,
153 .vop_close = ext2fifo_close,
154 .vop_fsync = ext2_fsync,
155 .vop_getattr = ext2_getattr,
156 .vop_inactive = ext2_inactive,
157 .vop_kqfilter = ext2fifo_kqfilter,
158 .vop_print = ext2_print,
159 .vop_read = VOP_PANIC,
160 .vop_reclaim = ext2_reclaim,
161 .vop_setattr = ext2_setattr,
162 .vop_write = VOP_PANIC,
163 .vop_vptofh = ext2_vptofh,
167 * A virgin directory (no blushing please).
168 * Note that the type and namlen fields are reversed relative to ext2.
169 * Also, we don't use `struct odirtemplate', since it would just cause
170 * endianness problems.
172 static struct dirtemplate mastertemplate = {
173 0, 12, 1, EXT2_FT_DIR, ".",
174 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
176 static struct dirtemplate omastertemplate = {
177 0, 12, 1, EXT2_FT_UNKNOWN, ".",
178 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
182 ext2_itimes_locked(struct vnode *vp)
187 ASSERT_VI_LOCKED(vp, __func__);
190 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
192 if ((vp->v_type == VBLK || vp->v_type == VCHR))
193 ip->i_flag |= IN_LAZYMOD;
195 ip->i_flag |= IN_MODIFIED;
196 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
198 if (ip->i_flag & IN_ACCESS) {
199 ip->i_atime = ts.tv_sec;
200 ip->i_atimensec = ts.tv_nsec;
202 if (ip->i_flag & IN_UPDATE) {
203 ip->i_mtime = ts.tv_sec;
204 ip->i_mtimensec = ts.tv_nsec;
207 if (ip->i_flag & IN_CHANGE) {
208 ip->i_ctime = ts.tv_sec;
209 ip->i_ctimensec = ts.tv_nsec;
212 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
216 ext2_itimes(struct vnode *vp)
220 ext2_itimes_locked(vp);
225 * Create a regular file
229 struct vop_create_args /* {
231 struct vnode **a_vpp;
232 struct componentname *a_cnp;
239 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
240 ap->a_dvp, ap->a_vpp, ap->a_cnp);
248 struct vop_open_args /* {
251 struct ucred *a_cred;
256 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
260 * Files marked append-only must be opened for appending.
262 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
263 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
266 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
274 * Update the times on the inode.
278 struct vop_close_args /* {
281 struct ucred *a_cred;
285 struct vnode *vp = ap->a_vp;
288 if (vp->v_usecount > 1)
289 ext2_itimes_locked(vp);
296 struct vop_access_args /* {
299 struct ucred *a_cred;
303 struct vnode *vp = ap->a_vp;
304 struct inode *ip = VTOI(vp);
305 accmode_t accmode = ap->a_accmode;
308 if (vp->v_type == VBLK || vp->v_type == VCHR)
312 * Disallow write attempts on read-only file systems;
313 * unless the file is a socket, fifo, or a block or
314 * character device resident on the file system.
316 if (accmode & VWRITE) {
317 switch (vp->v_type) {
321 if (vp->v_mount->mnt_flag & MNT_RDONLY)
329 /* If immutable bit set, nobody gets to write it. */
330 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
333 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
334 ap->a_accmode, ap->a_cred, NULL);
340 struct vop_getattr_args /* {
343 struct ucred *a_cred;
346 struct vnode *vp = ap->a_vp;
347 struct inode *ip = VTOI(vp);
348 struct vattr *vap = ap->a_vap;
352 * Copy from inode table
354 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
355 vap->va_fileid = ip->i_number;
356 vap->va_mode = ip->i_mode & ~IFMT;
357 vap->va_nlink = ip->i_nlink;
358 vap->va_uid = ip->i_uid;
359 vap->va_gid = ip->i_gid;
360 vap->va_rdev = ip->i_rdev;
361 vap->va_size = ip->i_size;
362 vap->va_atime.tv_sec = ip->i_atime;
363 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
364 vap->va_mtime.tv_sec = ip->i_mtime;
365 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
366 vap->va_ctime.tv_sec = ip->i_ctime;
367 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
368 if E2DI_HAS_XTIME(ip) {
369 vap->va_birthtime.tv_sec = ip->i_birthtime;
370 vap->va_birthtime.tv_nsec = ip->i_birthnsec;
372 vap->va_flags = ip->i_flags;
373 vap->va_gen = ip->i_gen;
374 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
375 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
376 vap->va_type = IFTOVT(ip->i_mode);
377 vap->va_filerev = ip->i_modrev;
382 * Set attribute vnode op. called from several syscalls
386 struct vop_setattr_args /* {
389 struct ucred *a_cred;
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;
396 struct thread *td = curthread;
400 * Check for unsettable attributes.
402 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
403 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
404 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
405 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
408 if (vap->va_flags != VNOVAL) {
409 /* Disallow flags not supported by ext2fs. */
410 if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
413 if (vp->v_mount->mnt_flag & MNT_RDONLY)
416 * Callers may only modify the file flags on objects they
417 * have VADMIN rights for.
419 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
422 * Unprivileged processes and privileged processes in
423 * jail() are not permitted to unset system flags, or
424 * modify flags if any system flags are set.
425 * Privileged non-jail processes may not modify system flags
426 * if securelevel > 0 and any existing system flags are set.
428 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
430 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
431 error = securelevel_gt(cred, 0);
435 ip->i_flags = vap->va_flags;
438 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
439 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
441 ip->i_flags &= SF_SETTABLE;
442 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
444 ip->i_flag |= IN_CHANGE;
445 if (ip->i_flags & (IMMUTABLE | APPEND))
448 if (ip->i_flags & (IMMUTABLE | APPEND))
451 * Go through the fields and update iff not VNOVAL.
453 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
454 if (vp->v_mount->mnt_flag & MNT_RDONLY)
456 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
460 if (vap->va_size != VNOVAL) {
462 * Disallow write attempts on read-only file systems;
463 * unless the file is a socket, fifo, or a block or
464 * character device resident on the file system.
466 switch (vp->v_type) {
471 if (vp->v_mount->mnt_flag & MNT_RDONLY)
477 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
480 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
481 if (vp->v_mount->mnt_flag & MNT_RDONLY)
485 * If times is NULL, ... The caller must be the owner of
486 * the file, have permission to write the file, or be the
488 * If times is non-NULL, ... The caller must be the owner of
489 * the file or be the super-user.
491 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
492 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
493 (error = VOP_ACCESS(vp, VWRITE, cred, td))))
495 if (vap->va_atime.tv_sec != VNOVAL)
496 ip->i_flag |= IN_ACCESS;
497 if (vap->va_mtime.tv_sec != VNOVAL)
498 ip->i_flag |= IN_CHANGE | IN_UPDATE;
500 if (vap->va_atime.tv_sec != VNOVAL) {
501 ip->i_atime = vap->va_atime.tv_sec;
502 ip->i_atimensec = vap->va_atime.tv_nsec;
504 if (vap->va_mtime.tv_sec != VNOVAL) {
505 ip->i_mtime = vap->va_mtime.tv_sec;
506 ip->i_mtimensec = vap->va_mtime.tv_nsec;
508 ip->i_birthtime = vap->va_birthtime.tv_sec;
509 ip->i_birthnsec = vap->va_birthtime.tv_nsec;
510 error = ext2_update(vp, 0);
515 if (vap->va_mode != (mode_t)VNOVAL) {
516 if (vp->v_mount->mnt_flag & MNT_RDONLY)
518 error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
524 * Change the mode on a file.
525 * Inode must be locked before calling.
528 ext2_chmod(vp, mode, cred, td)
534 struct inode *ip = VTOI(vp);
538 * To modify the permissions on a file, must possess VADMIN
541 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
544 * Privileged processes may set the sticky bit on non-directories,
545 * as well as set the setgid bit on a file with a group that the
546 * process is not a member of.
548 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
549 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
553 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
554 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
558 ip->i_mode &= ~ALLPERMS;
559 ip->i_mode |= (mode & ALLPERMS);
560 ip->i_flag |= IN_CHANGE;
565 * Perform chown operation on inode ip;
566 * inode must be locked prior to call.
569 ext2_chown(vp, uid, gid, cred, td)
576 struct inode *ip = VTOI(vp);
581 if (uid == (uid_t)VNOVAL)
583 if (gid == (gid_t)VNOVAL)
586 * To modify the ownership of a file, must possess VADMIN
589 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
592 * To change the owner of a file, or change the group of a file
593 * to a group of which we are not a member, the caller must
596 if (uid != ip->i_uid || (gid != ip->i_gid &&
597 !groupmember(gid, cred))) {
598 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
606 ip->i_flag |= IN_CHANGE;
607 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
608 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
609 ip->i_mode &= ~(ISUID | ISGID);
615 * Synch an open file.
620 struct vop_fsync_args /* {
622 struct ucred *a_cred;
628 * Flush all dirty buffers associated with a vnode.
633 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
642 struct vop_mknod_args /* {
644 struct vnode **a_vpp;
645 struct componentname *a_cnp;
649 struct vattr *vap = ap->a_vap;
650 struct vnode **vpp = ap->a_vpp;
655 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
656 ap->a_dvp, vpp, ap->a_cnp);
660 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
661 if (vap->va_rdev != VNOVAL) {
663 * Want to be able to use this to make badblock
664 * inodes, so don't truncate the dev number.
666 ip->i_rdev = vap->va_rdev;
669 * Remove inode, then reload it through VFS_VGET so it is
670 * checked to see if it is an alias of an existing entry in
671 * the inode cache. XXX I don't believe this is necessary now.
673 (*vpp)->v_type = VNON;
674 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
677 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
687 struct vop_remove_args /* {
690 struct componentname *a_cnp;
694 struct vnode *vp = ap->a_vp;
695 struct vnode *dvp = ap->a_dvp;
699 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
700 (VTOI(dvp)->i_flags & APPEND)) {
704 error = ext2_dirremove(dvp, ap->a_cnp);
707 ip->i_flag |= IN_CHANGE;
718 struct vop_link_args /* {
719 struct vnode *a_tdvp;
721 struct componentname *a_cnp;
724 struct vnode *vp = ap->a_vp;
725 struct vnode *tdvp = ap->a_tdvp;
726 struct componentname *cnp = ap->a_cnp;
731 if ((cnp->cn_flags & HASBUF) == 0)
732 panic("ext2_link: no name");
734 if (tdvp->v_mount != vp->v_mount) {
739 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
743 if (ip->i_flags & (IMMUTABLE | APPEND)) {
748 ip->i_flag |= IN_CHANGE;
749 error = ext2_update(vp, !DOINGASYNC(vp));
751 error = ext2_direnter(ip, tdvp, cnp);
754 ip->i_flag |= IN_CHANGE;
761 * Rename system call.
762 * rename("foo", "bar");
765 * link("foo", "bar");
767 * but ``atomically''. Can't do full commit without saving state in the
768 * inode on disk which isn't feasible at this time. Best we can do is
769 * always guarantee the target exists.
771 * Basic algorithm is:
773 * 1) Bump link count on source while we're linking it to the
774 * target. This also ensure the inode won't be deleted out
775 * from underneath us while we work (it may be truncated by
776 * a concurrent `trunc' or `open' for creation).
777 * 2) Link source to destination. If destination already exists,
779 * 3) Unlink source reference to inode if still around. If a
780 * directory was moved and the parent of the destination
781 * is different from the source, patch the ".." entry in the
786 struct vop_rename_args /* {
787 struct vnode *a_fdvp;
789 struct componentname *a_fcnp;
790 struct vnode *a_tdvp;
792 struct componentname *a_tcnp;
795 struct vnode *tvp = ap->a_tvp;
796 struct vnode *tdvp = ap->a_tdvp;
797 struct vnode *fvp = ap->a_fvp;
798 struct vnode *fdvp = ap->a_fdvp;
799 struct componentname *tcnp = ap->a_tcnp;
800 struct componentname *fcnp = ap->a_fcnp;
801 struct inode *ip, *xp, *dp;
802 struct dirtemplate dirbuf;
803 int doingdirectory = 0, oldparent = 0, newparent = 0;
808 if ((tcnp->cn_flags & HASBUF) == 0 ||
809 (fcnp->cn_flags & HASBUF) == 0)
810 panic("ext2_rename: no name");
813 * Check for cross-device rename.
815 if ((fvp->v_mount != tdvp->v_mount) ||
816 (tvp && (fvp->v_mount != tvp->v_mount))) {
830 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
831 (VTOI(tdvp)->i_flags & APPEND))) {
837 * Renaming a file to itself has no effect. The upper layers should
838 * not call us in that case. Temporarily just warn if they do.
841 printf("ext2_rename: fvp == tvp (can't happen)\n");
846 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
850 if (ip->i_nlink >= LINK_MAX) {
855 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
856 || (dp->i_flags & APPEND)) {
861 if ((ip->i_mode & IFMT) == IFDIR) {
863 * Avoid ".", "..", and aliases of "." for obvious reasons.
865 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
866 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
867 (ip->i_flag & IN_RENAME)) {
872 ip->i_flag |= IN_RENAME;
873 oldparent = dp->i_number;
879 * When the target exists, both the directory
880 * and target vnodes are returned locked.
888 * 1) Bump link count while we're moving stuff
889 * around. If we crash somewhere before
890 * completing our work, the link count
891 * may be wrong, but correctable.
894 ip->i_flag |= IN_CHANGE;
895 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
901 * If ".." must be changed (ie the directory gets a new
902 * parent) then the source directory must not be in the
903 * directory hierarchy above the target, as this would
904 * orphan everything below the source directory. Also
905 * the user must have write permission in the source so
906 * as to be able to change "..". We must repeat the call
907 * to namei, as the parent directory is unlocked by the
908 * call to checkpath().
910 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
912 if (oldparent != dp->i_number)
913 newparent = dp->i_number;
914 if (doingdirectory && newparent) {
915 if (error) /* write access check above */
919 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
923 error = relookup(tdvp, &tvp, tcnp);
933 * 2) If target doesn't exist, link the target
934 * to the source and unlink the source.
935 * Otherwise, rewrite the target directory
936 * entry to reference the source inode and
937 * expunge the original entry's existence.
940 if (dp->i_devvp != ip->i_devvp)
941 panic("ext2_rename: EXDEV");
943 * Account for ".." in new directory.
944 * When source and destination have the same
945 * parent we don't fool with the link count.
947 if (doingdirectory && newparent) {
948 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
953 dp->i_flag |= IN_CHANGE;
954 error = ext2_update(tdvp, !DOINGASYNC(tdvp));
958 error = ext2_direnter(ip, tdvp, tcnp);
960 if (doingdirectory && newparent) {
962 dp->i_flag |= IN_CHANGE;
963 (void)ext2_update(tdvp, 1);
969 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
970 panic("ext2_rename: EXDEV");
972 * Short circuit rename(foo, foo).
974 if (xp->i_number == ip->i_number)
975 panic("ext2_rename: same file");
977 * If the parent directory is "sticky", then the user must
978 * own the parent directory, or the destination of the rename,
979 * otherwise the destination may not be changed (except by
980 * root). This implements append-only directories.
982 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
983 tcnp->cn_cred->cr_uid != dp->i_uid &&
984 xp->i_uid != tcnp->cn_cred->cr_uid) {
989 * Target must be empty if a directory and have no links
990 * to it. Also, ensure source and target are compatible
991 * (both directories, or both not directories).
993 if ((xp->i_mode&IFMT) == IFDIR) {
994 if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
999 if (!doingdirectory) {
1004 } else if (doingdirectory) {
1008 error = ext2_dirrewrite(dp, ip, tcnp);
1012 * If the target directory is in the same
1013 * directory as the source directory,
1014 * decrement the link count on the parent
1015 * of the target directory.
1017 if (doingdirectory && !newparent) {
1019 dp->i_flag |= IN_CHANGE;
1023 * Adjust the link count of the target to
1024 * reflect the dirrewrite above. If this is
1025 * a directory it is empty and there are
1026 * no links to it, so we can squash the inode and
1027 * any space associated with it. We disallowed
1028 * renaming over top of a directory with links to
1029 * it above, as the remaining link would point to
1030 * a directory without "." or ".." entries.
1033 if (doingdirectory) {
1034 if (--xp->i_nlink != 0)
1035 panic("ext2_rename: linked directory");
1036 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1037 tcnp->cn_cred, tcnp->cn_thread);
1039 xp->i_flag |= IN_CHANGE;
1045 * 3) Unlink the source.
1047 fcnp->cn_flags &= ~MODMASK;
1048 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1050 error = relookup(fdvp, &fvp, fcnp);
1058 * From name has disappeared.
1061 panic("ext2_rename: lost dir entry");
1066 * Ensure that the directory entry still exists and has not
1067 * changed while the new name has been entered. If the source is
1068 * a file then the entry may have been unlinked or renamed. In
1069 * either case there is no further work to be done. If the source
1070 * is a directory then it cannot have been rmdir'ed; its link
1071 * count of three would cause a rmdir to fail with ENOTEMPTY.
1072 * The IN_RENAME flag ensures that it cannot be moved by another
1077 panic("ext2_rename: lost dir entry");
1080 * If the source is a directory with a
1081 * new parent, the link count of the old
1082 * parent directory must be decremented
1083 * and ".." set to point to the new parent.
1085 if (doingdirectory && newparent) {
1087 dp->i_flag |= IN_CHANGE;
1088 error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1089 sizeof(struct dirtemplate), (off_t)0,
1090 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1091 tcnp->cn_cred, NOCRED, NULL, NULL);
1093 /* Like ufs little-endian: */
1094 namlen = dirbuf.dotdot_type;
1096 dirbuf.dotdot_name[0] != '.' ||
1097 dirbuf.dotdot_name[1] != '.') {
1098 ext2_dirbad(xp, (doff_t)12,
1099 "rename: mangled dir");
1101 dirbuf.dotdot_ino = newparent;
1102 (void) vn_rdwr(UIO_WRITE, fvp,
1104 sizeof(struct dirtemplate),
1105 (off_t)0, UIO_SYSSPACE,
1106 IO_NODELOCKED | IO_SYNC |
1107 IO_NOMACCHECK, tcnp->cn_cred,
1108 NOCRED, NULL, NULL);
1113 error = ext2_dirremove(fdvp, fcnp);
1116 xp->i_flag |= IN_CHANGE;
1118 xp->i_flag &= ~IN_RENAME;
1133 ip->i_flag &= ~IN_RENAME;
1134 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1136 ip->i_flag |= IN_CHANGE;
1137 ip->i_flag &= ~IN_RENAME;
1149 struct vop_mkdir_args /* {
1150 struct vnode *a_dvp;
1151 struct vnode **a_vpp;
1152 struct componentname *a_cnp;
1153 struct vattr *a_vap;
1156 struct vnode *dvp = ap->a_dvp;
1157 struct vattr *vap = ap->a_vap;
1158 struct componentname *cnp = ap->a_cnp;
1159 struct inode *ip, *dp;
1161 struct dirtemplate dirtemplate, *dtp;
1165 if ((cnp->cn_flags & HASBUF) == 0)
1166 panic("ext2_mkdir: no name");
1169 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1173 dmode = vap->va_mode & 0777;
1176 * Must simulate part of ext2_makeinode here to acquire the inode,
1177 * but not have it entered in the parent directory. The entry is
1178 * made later after writing "." and ".." entries.
1180 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1184 ip->i_gid = dp->i_gid;
1188 * if we are hacking owners here, (only do this where told to)
1189 * and we are not giving it TOO root, (would subvert quotas)
1190 * then go ahead and give it to the other user.
1191 * The new directory also inherits the SUID bit.
1192 * If user's UID and dir UID are the same,
1193 * 'give it away' so that the SUID is still forced on.
1195 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1196 (dp->i_mode & ISUID) && dp->i_uid) {
1198 ip->i_uid = dp->i_uid;
1200 ip->i_uid = cnp->cn_cred->cr_uid;
1204 ip->i_uid = cnp->cn_cred->cr_uid;
1206 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1208 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1210 if (cnp->cn_flags & ISWHITEOUT)
1211 ip->i_flags |= UF_OPAQUE;
1212 error = ext2_update(tvp, 1);
1215 * Bump link count in parent directory
1216 * to reflect work done below. Should
1217 * be done before reference is created
1218 * so reparation is possible if we crash.
1221 dp->i_flag |= IN_CHANGE;
1222 error = ext2_update(dvp, !DOINGASYNC(dvp));
1226 /* Initialize directory with "." and ".." from static template. */
1227 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1228 EXT2F_INCOMPAT_FTYPE))
1229 dtp = &mastertemplate;
1231 dtp = &omastertemplate;
1233 dirtemplate.dot_ino = ip->i_number;
1234 dirtemplate.dotdot_ino = dp->i_number;
1235 /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1236 * so let's just redefine it - for this function only
1239 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize
1240 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1241 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1242 sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1243 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1247 dp->i_flag |= IN_CHANGE;
1250 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1251 /* XXX should grow with balloc() */
1252 panic("ext2_mkdir: blksize");
1254 ip->i_size = DIRBLKSIZ;
1255 ip->i_flag |= IN_CHANGE;
1258 /* Directory set up, now install its entry in the parent directory. */
1259 error = ext2_direnter(ip, dvp, cnp);
1262 dp->i_flag |= IN_CHANGE;
1266 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1267 * for us because we set the link count to 0.
1271 ip->i_flag |= IN_CHANGE;
1278 #define DIRBLKSIZ DEV_BSIZE
1282 * Rmdir system call.
1286 struct vop_rmdir_args /* {
1287 struct vnode *a_dvp;
1289 struct componentname *a_cnp;
1292 struct vnode *vp = ap->a_vp;
1293 struct vnode *dvp = ap->a_dvp;
1294 struct componentname *cnp = ap->a_cnp;
1295 struct inode *ip, *dp;
1302 * Verify the directory is empty (and valid).
1303 * (Rmdir ".." won't be valid since
1304 * ".." will contain a reference to
1305 * the current directory and thus be
1309 if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1313 if ((dp->i_flags & APPEND)
1314 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1319 * Delete reference to directory before purging
1320 * inode. If we crash in between, the directory
1321 * will be reattached to lost+found,
1323 error = ext2_dirremove(dvp, cnp);
1327 dp->i_flag |= IN_CHANGE;
1331 * Truncate inode. The only stuff left
1332 * in the directory is "." and "..". The
1333 * "." reference is inconsequential since
1334 * we're quashing it. The ".." reference
1335 * has already been adjusted above. We've
1336 * removed the "." reference and the reference
1337 * in the parent directory, but there may be
1338 * other hard links so decrement by 2 and
1339 * worry about them later.
1342 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1344 cache_purge(ITOV(ip));
1345 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1347 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1348 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1355 * symlink -- make a symbolic link
1359 struct vop_symlink_args /* {
1360 struct vnode *a_dvp;
1361 struct vnode **a_vpp;
1362 struct componentname *a_cnp;
1363 struct vattr *a_vap;
1367 struct vnode *vp, **vpp = ap->a_vpp;
1371 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1376 len = strlen(ap->a_target);
1377 if (len < vp->v_mount->mnt_maxsymlinklen) {
1379 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1381 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1383 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1384 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1385 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1392 * Return target name of a symbolic link
1396 struct vop_readlink_args /* {
1399 struct ucred *a_cred;
1402 struct vnode *vp = ap->a_vp;
1403 struct inode *ip = VTOI(vp);
1407 if (isize < vp->v_mount->mnt_maxsymlinklen) {
1408 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1411 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1415 * Calculate the logical to physical mapping if not done already,
1416 * then call the device strategy routine.
1418 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1419 * deadlock on memory. See ext2_bmap() for details.
1423 struct vop_strategy_args /* {
1428 struct buf *bp = ap->a_bp;
1429 struct vnode *vp = ap->a_vp;
1436 if (vp->v_type == VBLK || vp->v_type == VCHR)
1437 panic("ext2_strategy: spec");
1438 if (bp->b_blkno == bp->b_lblkno) {
1439 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1440 bp->b_blkno = blkno;
1442 bp->b_error = error;
1443 bp->b_ioflags |= BIO_ERROR;
1447 if ((long)bp->b_blkno == -1)
1450 if ((long)bp->b_blkno == -1) {
1454 bp->b_iooffset = dbtob(bp->b_blkno);
1455 bo = VFSTOEXT2(vp->v_mount)->um_bo;
1456 BO_STRATEGY(bo, bp);
1461 * Print out the contents of an inode.
1465 struct vop_print_args /* {
1469 struct vnode *vp = ap->a_vp;
1470 struct inode *ip = VTOI(vp);
1472 vn_printf(ip->i_devvp, "\tino %lu", (u_long)ip->i_number);
1473 if (vp->v_type == VFIFO)
1480 * Close wrapper for fifos.
1482 * Update the times on the inode then do device close.
1486 struct vop_close_args /* {
1489 struct ucred *a_cred;
1490 struct thread *a_td;
1493 struct vnode *vp = ap->a_vp;
1496 if (vp->v_usecount > 1)
1497 ext2_itimes_locked(vp);
1499 return (fifo_specops.vop_close(ap));
1503 * Kqfilter wrapper for fifos.
1505 * Fall through to ext2 kqfilter routines if needed
1508 ext2fifo_kqfilter(ap)
1509 struct vop_kqfilter_args *ap;
1513 error = fifo_specops.vop_kqfilter(ap);
1515 error = vfs_kqfilter(ap);
1520 * Return POSIX pathconf information applicable to ext2 filesystems.
1524 struct vop_pathconf_args /* {
1531 switch (ap->a_name) {
1533 *ap->a_retval = LINK_MAX;
1536 *ap->a_retval = NAME_MAX;
1539 *ap->a_retval = PATH_MAX;
1542 *ap->a_retval = PIPE_BUF;
1544 case _PC_CHOWN_RESTRICTED:
1557 * Vnode pointer to File handle
1562 struct vop_vptofh_args /* {
1570 ip = VTOI(ap->a_vp);
1571 ufhp = (struct ufid *)ap->a_fhp;
1572 ufhp->ufid_len = sizeof(struct ufid);
1573 ufhp->ufid_ino = ip->i_number;
1574 ufhp->ufid_gen = ip->i_gen;
1579 * Initialize the vnode associated with a new inode, handle aliased
1583 ext2_vinit(mntp, fifoops, vpp)
1585 struct vop_vector *fifoops;
1593 vp->v_type = IFTOVT(ip->i_mode);
1594 if (vp->v_type == VFIFO)
1597 if (ip->i_number == EXT2_ROOTINO)
1598 vp->v_vflag |= VV_ROOT;
1599 ip->i_modrev = init_va_filerev();
1605 * Allocate a new inode.
1608 ext2_makeinode(mode, dvp, vpp, cnp)
1612 struct componentname *cnp;
1614 struct inode *ip, *pdir;
1620 if ((cnp->cn_flags & HASBUF) == 0)
1621 panic("ext2_makeinode: no name");
1624 if ((mode & IFMT) == 0)
1627 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1632 ip->i_gid = pdir->i_gid;
1637 * not the owner of the directory,
1638 * and we are hacking owners here, (only do this where told to)
1639 * and we are not giving it TOO root, (would subvert quotas)
1640 * then go ahead and give it to the other user.
1641 * Note that this drops off the execute bits for security.
1643 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1644 (pdir->i_mode & ISUID) &&
1645 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1646 ip->i_uid = pdir->i_uid;
1649 ip->i_uid = cnp->cn_cred->cr_uid;
1653 ip->i_uid = cnp->cn_cred->cr_uid;
1655 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1657 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
1659 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1660 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1661 ip->i_mode &= ~ISGID;
1664 if (cnp->cn_flags & ISWHITEOUT)
1665 ip->i_flags |= UF_OPAQUE;
1668 * Make sure inode goes to disk before directory entry.
1670 error = ext2_update(tvp, !DOINGASYNC(tvp));
1673 error = ext2_direnter(ip, dvp, cnp);
1682 * Write error occurred trying to update the inode
1683 * or the directory so must deallocate the inode.
1686 ip->i_flag |= IN_CHANGE;
1692 * Vnode op for reading.
1696 struct vop_read_args /* {
1700 struct ucred *a_cred;
1706 struct m_ext2fs *fs;
1708 daddr_t lbn, nextlbn;
1710 long size, xfersize, blkoffset;
1711 int error, orig_resid, seqcount;
1716 ioflag = ap->a_ioflag;
1718 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1722 if (uio->uio_rw != UIO_READ)
1723 panic("%s: mode", "ext2_read");
1725 if (vp->v_type == VLNK) {
1726 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1727 panic("%s: short symlink", "ext2_read");
1728 } else if (vp->v_type != VREG && vp->v_type != VDIR)
1729 panic("%s: type %d", "ext2_read", vp->v_type);
1731 orig_resid = uio->uio_resid;
1732 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1733 if (orig_resid == 0)
1735 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1737 if (uio->uio_offset < ip->i_size &&
1738 uio->uio_offset >= fs->e2fs_maxfilesize)
1741 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1742 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1744 lbn = lblkno(fs, uio->uio_offset);
1746 size = blksize(fs, ip, lbn);
1747 blkoffset = blkoff(fs, uio->uio_offset);
1749 xfersize = fs->e2fs_fsize - blkoffset;
1750 if (uio->uio_resid < xfersize)
1751 xfersize = uio->uio_resid;
1752 if (bytesinfile < xfersize)
1753 xfersize = bytesinfile;
1755 if (lblktosize(fs, nextlbn) >= ip->i_size)
1756 error = bread(vp, lbn, size, NOCRED, &bp);
1757 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0)
1758 error = cluster_read(vp, ip->i_size, lbn, size,
1759 NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
1760 else if (seqcount > 1) {
1761 int nextsize = blksize(fs, ip, nextlbn);
1762 error = breadn(vp, lbn,
1763 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1765 error = bread(vp, lbn, size, NOCRED, &bp);
1773 * If IO_DIRECT then set B_DIRECT for the buffer. This
1774 * will cause us to attempt to release the buffer later on
1775 * and will cause the buffer cache to attempt to free the
1778 if (ioflag & IO_DIRECT)
1779 bp->b_flags |= B_DIRECT;
1782 * We should only get non-zero b_resid when an I/O error
1783 * has occurred, which should cause us to break above.
1784 * However, if the short read did not cause an error,
1785 * then we want to ensure that we do not uiomove bad
1786 * or uninitialized data.
1788 size -= bp->b_resid;
1789 if (size < xfersize) {
1794 error = uiomove((char *)bp->b_data + blkoffset,
1795 (int)xfersize, uio);
1799 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1801 * If it's VMIO or direct I/O, then we don't
1802 * need the buf, mark it available for
1803 * freeing. If it's non-direct VMIO, the VM has
1806 bp->b_flags |= B_RELBUF;
1810 * Otherwise let whoever
1811 * made the request take care of
1812 * freeing it. We just queue
1813 * it onto another list.
1820 * This can only happen in the case of an error
1821 * because the loop above resets bp to NULL on each iteration
1822 * and on normal completion has not set a new value into it.
1823 * so it must have come from a 'break' statement
1826 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1827 bp->b_flags |= B_RELBUF;
1834 if ((error == 0 || uio->uio_resid != orig_resid) &&
1835 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1836 ip->i_flag |= IN_ACCESS;
1841 * Vnode op for writing.
1845 struct vop_write_args /* {
1849 struct ucred *a_cred;
1855 struct m_ext2fs *fs;
1859 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1861 ioflag = ap->a_ioflag;
1865 seqcount = ioflag >> IO_SEQSHIFT;
1869 if (uio->uio_rw != UIO_WRITE)
1870 panic("%s: mode", "ext2_write");
1873 switch (vp->v_type) {
1875 if (ioflag & IO_APPEND)
1876 uio->uio_offset = ip->i_size;
1877 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1883 /* XXX differs from ffs -- this is called from ext2_mkdir(). */
1884 if ((ioflag & IO_SYNC) == 0)
1885 panic("ext2_write: nonsync dir write");
1888 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
1889 vp->v_type, (intmax_t)uio->uio_offset,
1890 (intmax_t)uio->uio_resid);
1893 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
1894 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
1896 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
1899 * Maybe this should be above the vnode op call, but so long as
1900 * file servers have no limits, I don't think it matters.
1902 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
1905 resid = uio->uio_resid;
1907 if (seqcount > BA_SEQMAX)
1908 flags = BA_SEQMAX << BA_SEQSHIFT;
1910 flags = seqcount << BA_SEQSHIFT;
1911 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1914 for (error = 0; uio->uio_resid > 0;) {
1915 lbn = lblkno(fs, uio->uio_offset);
1916 blkoffset = blkoff(fs, uio->uio_offset);
1917 xfersize = fs->e2fs_fsize - blkoffset;
1918 if (uio->uio_resid < xfersize)
1919 xfersize = uio->uio_resid;
1920 if (uio->uio_offset + xfersize > ip->i_size)
1921 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
1924 * We must perform a read-before-write if the transfer size
1925 * does not cover the entire buffer.
1927 if (fs->e2fs_bsize > xfersize)
1930 flags &= ~BA_CLRBUF;
1931 error = ext2_balloc(ip, lbn, blkoffset + xfersize,
1932 ap->a_cred, &bp, flags);
1937 * If the buffer is not valid and we did not clear garbage
1938 * out above, we have to do so here even though the write
1939 * covers the entire buffer in order to avoid a mmap()/write
1940 * race where another process may see the garbage prior to
1941 * the uiomove() for a write replacing it.
1943 if ((bp->b_flags & B_CACHE) == 0 && fs->e2fs_bsize <= xfersize)
1945 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
1946 bp->b_flags |= B_NOCACHE;
1947 if (uio->uio_offset + xfersize > ip->i_size)
1948 ip->i_size = uio->uio_offset + xfersize;
1949 size = blksize(fs, ip, lbn) - bp->b_resid;
1950 if (size < xfersize)
1954 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1955 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1956 bp->b_flags |= B_RELBUF;
1960 * If IO_SYNC each buffer is written synchronously. Otherwise
1961 * if we have a severe page deficiency write the buffer
1962 * asynchronously. Otherwise try to cluster, and if that
1963 * doesn't do it then either do an async write (if O_DIRECT),
1964 * or a delayed write (if not).
1966 if (ioflag & IO_SYNC) {
1968 } else if (vm_page_count_severe() ||
1969 buf_dirty_count_severe() ||
1970 (ioflag & IO_ASYNC)) {
1971 bp->b_flags |= B_CLUSTEROK;
1973 } else if (xfersize + blkoffset == fs->e2fs_fsize) {
1974 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
1975 bp->b_flags |= B_CLUSTEROK;
1976 cluster_write(vp, bp, ip->i_size, seqcount);
1980 } else if (ioflag & IO_DIRECT) {
1981 bp->b_flags |= B_CLUSTEROK;
1984 bp->b_flags |= B_CLUSTEROK;
1987 if (error || xfersize == 0)
1991 * If we successfully wrote any data, and we are not the superuser
1992 * we clear the setuid and setgid bits as a precaution against
1995 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
1997 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
1998 ip->i_mode &= ~(ISUID | ISGID);
2001 if (ioflag & IO_UNIT) {
2002 (void)ext2_truncate(vp, osize,
2003 ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2004 uio->uio_offset -= resid - uio->uio_resid;
2005 uio->uio_resid = resid;
2008 if (uio->uio_resid != resid) {
2009 ip->i_flag |= IN_CHANGE | IN_UPDATE;
2010 if (ioflag & IO_SYNC)
2011 error = ext2_update(vp, 1);