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 * SPDX-License-Identifier: BSD-3-Clause
10 * Copyright (c) 1982, 1986, 1989, 1993
11 * The Regents of the University of California. All rights reserved.
12 * (c) UNIX System Laboratories, Inc.
13 * All or some portions of this file are derived from material licensed
14 * to the University of California by American Telephone and Telegraph
15 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
16 * the permission of UNIX System Laboratories, Inc.
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions
21 * 1. Redistributions of source code must retain the above copyright
22 * notice, this list of conditions and the following disclaimer.
23 * 2. Redistributions in binary form must reproduce the above copyright
24 * notice, this list of conditions and the following disclaimer in the
25 * documentation and/or other materials provided with the distribution.
26 * 3. Neither the name of the University nor the names of its contributors
27 * may be used to endorse or promote products derived from this software
28 * without specific prior written permission.
30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42 * @(#)ufs_vnops.c 8.7 (Berkeley) 2/3/94
43 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
47 #include "opt_suiddir.h"
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/kernel.h>
52 #include <sys/fcntl.h>
53 #include <sys/filio.h>
54 #include <sys/limits.h>
59 #include <sys/endian.h>
61 #include <sys/rwlock.h>
62 #include <sys/mount.h>
63 #include <sys/unistd.h>
65 #include <sys/vnode.h>
66 #include <sys/namei.h>
67 #include <sys/lockf.h>
68 #include <sys/event.h>
71 #include <sys/extattr.h>
72 #include <sys/vmmeter.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vnode_pager.h>
82 #include "opt_directio.h"
84 #include <ufs/ufs/dir.h>
86 #include <fs/ext2fs/fs.h>
87 #include <fs/ext2fs/inode.h>
88 #include <fs/ext2fs/ext2_acl.h>
89 #include <fs/ext2fs/ext2fs.h>
90 #include <fs/ext2fs/ext2_extern.h>
91 #include <fs/ext2fs/ext2_dinode.h>
92 #include <fs/ext2fs/ext2_dir.h>
93 #include <fs/ext2fs/ext2_mount.h>
94 #include <fs/ext2fs/ext2_extattr.h>
95 #include <fs/ext2fs/ext2_extents.h>
97 SDT_PROVIDER_DECLARE(ext2fs);
100 * arg0: verbosity. Higher numbers give more verbose messages
101 * arg1: Textual message
103 SDT_PROBE_DEFINE2(ext2fs, , vnops, trace, "int", "char*");
105 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
106 static void ext2_itimes_locked(struct vnode *);
108 static vop_access_t ext2_access;
109 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
110 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
112 static vop_close_t ext2_close;
113 static vop_create_t ext2_create;
114 static vop_fsync_t ext2_fsync;
115 static vop_getattr_t ext2_getattr;
116 static vop_ioctl_t ext2_ioctl;
117 static vop_link_t ext2_link;
118 static vop_mkdir_t ext2_mkdir;
119 static vop_mknod_t ext2_mknod;
120 static vop_open_t ext2_open;
121 static vop_pathconf_t ext2_pathconf;
122 static vop_print_t ext2_print;
123 static vop_read_t ext2_read;
124 static vop_readlink_t ext2_readlink;
125 static vop_remove_t ext2_remove;
126 static vop_rename_t ext2_rename;
127 static vop_rmdir_t ext2_rmdir;
128 static vop_setattr_t ext2_setattr;
129 static vop_strategy_t ext2_strategy;
130 static vop_symlink_t ext2_symlink;
131 static vop_write_t ext2_write;
132 static vop_deleteextattr_t ext2_deleteextattr;
133 static vop_getextattr_t ext2_getextattr;
134 static vop_listextattr_t ext2_listextattr;
135 static vop_setextattr_t ext2_setextattr;
136 static vop_vptofh_t ext2_vptofh;
137 static vop_close_t ext2fifo_close;
138 static vop_kqfilter_t ext2fifo_kqfilter;
140 /* Global vfs data structures for ext2. */
141 struct vop_vector ext2_vnodeops = {
142 .vop_default = &default_vnodeops,
143 .vop_access = ext2_access,
144 .vop_bmap = ext2_bmap,
145 .vop_cachedlookup = ext2_lookup,
146 .vop_close = ext2_close,
147 .vop_create = ext2_create,
148 .vop_fsync = ext2_fsync,
149 .vop_getpages = vnode_pager_local_getpages,
150 .vop_getpages_async = vnode_pager_local_getpages_async,
151 .vop_getattr = ext2_getattr,
152 .vop_inactive = ext2_inactive,
153 .vop_ioctl = ext2_ioctl,
154 .vop_link = ext2_link,
155 .vop_lookup = vfs_cache_lookup,
156 .vop_mkdir = ext2_mkdir,
157 .vop_mknod = ext2_mknod,
158 .vop_open = ext2_open,
159 .vop_pathconf = ext2_pathconf,
160 .vop_poll = vop_stdpoll,
161 .vop_print = ext2_print,
162 .vop_read = ext2_read,
163 .vop_readdir = ext2_readdir,
164 .vop_readlink = ext2_readlink,
165 .vop_reallocblks = ext2_reallocblks,
166 .vop_reclaim = ext2_reclaim,
167 .vop_remove = ext2_remove,
168 .vop_rename = ext2_rename,
169 .vop_rmdir = ext2_rmdir,
170 .vop_setattr = ext2_setattr,
171 .vop_strategy = ext2_strategy,
172 .vop_symlink = ext2_symlink,
173 .vop_write = ext2_write,
174 .vop_deleteextattr = ext2_deleteextattr,
175 .vop_getextattr = ext2_getextattr,
176 .vop_listextattr = ext2_listextattr,
177 .vop_setextattr = ext2_setextattr,
179 .vop_getacl = ext2_getacl,
180 .vop_setacl = ext2_setacl,
181 .vop_aclcheck = ext2_aclcheck,
183 .vop_vptofh = ext2_vptofh,
185 VFS_VOP_VECTOR_REGISTER(ext2_vnodeops);
187 struct vop_vector ext2_fifoops = {
188 .vop_default = &fifo_specops,
189 .vop_access = ext2_access,
190 .vop_close = ext2fifo_close,
191 .vop_fsync = ext2_fsync,
192 .vop_getattr = ext2_getattr,
193 .vop_inactive = ext2_inactive,
194 .vop_kqfilter = ext2fifo_kqfilter,
195 .vop_pathconf = ext2_pathconf,
196 .vop_print = ext2_print,
197 .vop_read = VOP_PANIC,
198 .vop_reclaim = ext2_reclaim,
199 .vop_setattr = ext2_setattr,
200 .vop_write = VOP_PANIC,
201 .vop_vptofh = ext2_vptofh,
203 VFS_VOP_VECTOR_REGISTER(ext2_fifoops);
206 * A virgin directory (no blushing please).
207 * Note that the type and namlen fields are reversed relative to ext2.
208 * Also, we don't use `struct odirtemplate', since it would just cause
209 * endianness problems.
211 static struct dirtemplate mastertemplate = {
212 0, htole16(12), 1, EXT2_FT_DIR, ".",
213 0, htole16(DIRBLKSIZ - 12), 2, EXT2_FT_DIR, ".."
215 static struct dirtemplate omastertemplate = {
216 0, htole16(12), 1, EXT2_FT_UNKNOWN, ".",
217 0, htole16(DIRBLKSIZ - 12), 2, EXT2_FT_UNKNOWN, ".."
221 ext2_itimes_locked(struct vnode *vp)
226 ASSERT_VI_LOCKED(vp, __func__);
229 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
231 if ((vp->v_type == VBLK || vp->v_type == VCHR))
232 ip->i_flag |= IN_LAZYMOD;
234 ip->i_flag |= IN_MODIFIED;
235 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
237 if (ip->i_flag & IN_ACCESS) {
238 ip->i_atime = ts.tv_sec;
239 ip->i_atimensec = ts.tv_nsec;
241 if (ip->i_flag & IN_UPDATE) {
242 ip->i_mtime = ts.tv_sec;
243 ip->i_mtimensec = ts.tv_nsec;
246 if (ip->i_flag & IN_CHANGE) {
247 ip->i_ctime = ts.tv_sec;
248 ip->i_ctimensec = ts.tv_nsec;
251 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
255 ext2_itimes(struct vnode *vp)
259 ext2_itimes_locked(vp);
264 * Create a regular file
267 ext2_create(struct vop_create_args *ap)
272 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
273 ap->a_dvp, ap->a_vpp, ap->a_cnp);
276 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
277 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
282 ext2_open(struct vop_open_args *ap)
285 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
289 * Files marked append-only must be opened for appending.
291 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
292 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
295 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
303 * Update the times on the inode.
306 ext2_close(struct vop_close_args *ap)
308 struct vnode *vp = ap->a_vp;
311 if (vp->v_usecount > 1)
312 ext2_itimes_locked(vp);
318 ext2_access(struct vop_access_args *ap)
320 struct vnode *vp = ap->a_vp;
321 struct inode *ip = VTOI(vp);
322 accmode_t accmode = ap->a_accmode;
326 * Disallow write attempts on read-only file systems;
327 * unless the file is a socket, fifo, or a block or
328 * character device resident on the file system.
330 if (accmode & VWRITE) {
331 switch (vp->v_type) {
335 if (vp->v_mount->mnt_flag & MNT_RDONLY)
343 /* If immutable bit set, nobody gets to write it. */
344 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
347 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
348 ap->a_accmode, ap->a_cred);
353 ext2_getattr(struct vop_getattr_args *ap)
355 struct vnode *vp = ap->a_vp;
356 struct inode *ip = VTOI(vp);
357 struct vattr *vap = ap->a_vap;
361 * Copy from inode table
363 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
364 vap->va_fileid = ip->i_number;
365 vap->va_mode = ip->i_mode & ~IFMT;
366 vap->va_nlink = ip->i_nlink;
367 vap->va_uid = ip->i_uid;
368 vap->va_gid = ip->i_gid;
369 vap->va_rdev = ip->i_rdev;
370 vap->va_size = ip->i_size;
371 vap->va_atime.tv_sec = ip->i_atime;
372 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
373 vap->va_mtime.tv_sec = ip->i_mtime;
374 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
375 vap->va_ctime.tv_sec = ip->i_ctime;
376 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
377 if (E2DI_HAS_XTIME(ip)) {
378 vap->va_birthtime.tv_sec = ip->i_birthtime;
379 vap->va_birthtime.tv_nsec = ip->i_birthnsec;
381 vap->va_flags = ip->i_flags;
382 vap->va_gen = ip->i_gen;
383 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
384 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
385 vap->va_type = IFTOVT(ip->i_mode);
386 vap->va_filerev = ip->i_modrev;
391 * Set attribute vnode op. called from several syscalls
394 ext2_setattr(struct vop_setattr_args *ap)
396 struct vattr *vap = ap->a_vap;
397 struct vnode *vp = ap->a_vp;
398 struct inode *ip = VTOI(vp);
399 struct ucred *cred = ap->a_cred;
400 struct thread *td = curthread;
404 * Check for unsettable attributes.
406 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
407 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
408 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
409 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
412 if (vap->va_flags != VNOVAL) {
413 /* Disallow flags not supported by ext2fs. */
414 if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
417 if (vp->v_mount->mnt_flag & MNT_RDONLY)
420 * Callers may only modify the file flags on objects they
421 * have VADMIN rights for.
423 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
426 * Unprivileged processes and privileged processes in
427 * jail() are not permitted to unset system flags, or
428 * modify flags if any system flags are set.
429 * Privileged non-jail processes may not modify system flags
430 * if securelevel > 0 and any existing system flags are set.
432 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
433 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
434 error = securelevel_gt(cred, 0);
439 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
440 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
443 ip->i_flags = vap->va_flags;
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 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
496 if (vap->va_atime.tv_sec != VNOVAL) {
497 ip->i_flag &= ~IN_ACCESS;
498 ip->i_atime = vap->va_atime.tv_sec;
499 ip->i_atimensec = vap->va_atime.tv_nsec;
501 if (vap->va_mtime.tv_sec != VNOVAL) {
502 ip->i_flag &= ~IN_UPDATE;
503 ip->i_mtime = vap->va_mtime.tv_sec;
504 ip->i_mtimensec = vap->va_mtime.tv_nsec;
506 if (E2DI_HAS_XTIME(ip) && vap->va_birthtime.tv_sec != VNOVAL) {
507 ip->i_birthtime = vap->va_birthtime.tv_sec;
508 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(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
530 struct inode *ip = VTOI(vp);
534 * To modify the permissions on a file, must possess VADMIN
537 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
540 * Privileged processes may set the sticky bit on non-directories,
541 * as well as set the setgid bit on a file with a group that the
542 * process is not a member of.
544 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
545 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE);
549 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
550 error = priv_check_cred(cred, PRIV_VFS_SETGID);
554 ip->i_mode &= ~ALLPERMS;
555 ip->i_mode |= (mode & ALLPERMS);
556 ip->i_flag |= IN_CHANGE;
561 * Perform chown operation on inode ip;
562 * inode must be locked prior to call.
565 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
568 struct inode *ip = VTOI(vp);
573 if (uid == (uid_t)VNOVAL)
575 if (gid == (gid_t)VNOVAL)
578 * To modify the ownership of a file, must possess VADMIN
581 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
584 * To change the owner of a file, or change the group of a file
585 * to a group of which we are not a member, the caller must
588 if (uid != ip->i_uid || (gid != ip->i_gid &&
589 !groupmember(gid, cred))) {
590 error = priv_check_cred(cred, PRIV_VFS_CHOWN);
598 ip->i_flag |= IN_CHANGE;
599 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
600 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID) != 0)
601 ip->i_mode &= ~(ISUID | ISGID);
607 * Synch an open file.
611 ext2_fsync(struct vop_fsync_args *ap)
614 * Flush all dirty buffers associated with a vnode.
619 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
623 ext2_check_mknod_limits(dev_t dev)
625 unsigned maj = major(dev);
626 unsigned min = minor(dev);
628 if (maj > EXT2_MAJOR_MAX || min > EXT2_MINOR_MAX)
639 ext2_mknod(struct vop_mknod_args *ap)
641 struct vattr *vap = ap->a_vap;
642 struct vnode **vpp = ap->a_vpp;
647 if (vap->va_rdev != VNOVAL) {
648 error = ext2_check_mknod_limits(vap->va_rdev);
653 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
654 ap->a_dvp, vpp, ap->a_cnp);
658 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
659 if (vap->va_rdev != VNOVAL)
660 ip->i_rdev = vap->va_rdev;
663 * Remove inode, then reload it through VFS_VGET so it is
664 * checked to see if it is an alias of an existing entry in
665 * the inode cache. XXX I don't believe this is necessary now.
667 (*vpp)->v_type = VNON;
668 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
671 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
680 ext2_remove(struct vop_remove_args *ap)
683 struct vnode *vp = ap->a_vp;
684 struct vnode *dvp = ap->a_dvp;
688 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
689 (VTOI(dvp)->i_flags & APPEND)) {
693 error = ext2_dirremove(dvp, ap->a_cnp);
696 ip->i_flag |= IN_CHANGE;
706 ext2_link(struct vop_link_args *ap)
708 struct vnode *vp = ap->a_vp;
709 struct vnode *tdvp = ap->a_tdvp;
710 struct componentname *cnp = ap->a_cnp;
715 if ((cnp->cn_flags & HASBUF) == 0)
716 panic("ext2_link: no name");
719 if ((nlink_t)ip->i_nlink >= EXT4_LINK_MAX) {
723 if (ip->i_flags & (IMMUTABLE | APPEND)) {
728 ip->i_flag |= IN_CHANGE;
729 error = ext2_update(vp, !DOINGASYNC(vp));
731 error = ext2_direnter(ip, tdvp, cnp);
734 ip->i_flag |= IN_CHANGE;
741 ext2_inc_nlink(struct inode *ip)
746 if (S_ISDIR(ip->i_mode) &&
747 EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK) &&
749 if (ip->i_nlink >= EXT4_LINK_MAX || ip->i_nlink == 2)
751 } else if (ip->i_nlink > EXT4_LINK_MAX) {
760 ext2_dec_nlink(struct inode *ip)
763 if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2)
768 * Rename system call.
769 * rename("foo", "bar");
772 * link("foo", "bar");
774 * but ``atomically''. Can't do full commit without saving state in the
775 * inode on disk which isn't feasible at this time. Best we can do is
776 * always guarantee the target exists.
778 * Basic algorithm is:
780 * 1) Bump link count on source while we're linking it to the
781 * target. This also ensure the inode won't be deleted out
782 * from underneath us while we work (it may be truncated by
783 * a concurrent `trunc' or `open' for creation).
784 * 2) Link source to destination. If destination already exists,
786 * 3) Unlink source reference to inode if still around. If a
787 * directory was moved and the parent of the destination
788 * is different from the source, patch the ".." entry in the
792 ext2_rename(struct vop_rename_args *ap)
794 struct vnode *tvp = ap->a_tvp;
795 struct vnode *tdvp = ap->a_tdvp;
796 struct vnode *fvp = ap->a_fvp;
797 struct vnode *fdvp = ap->a_fdvp;
798 struct componentname *tcnp = ap->a_tcnp;
799 struct componentname *fcnp = ap->a_fcnp;
800 struct inode *ip, *xp, *dp;
801 struct dirtemplate *dirbuf;
802 int doingdirectory = 0, oldparent = 0, newparent = 0;
807 if ((tcnp->cn_flags & HASBUF) == 0 ||
808 (fcnp->cn_flags & HASBUF) == 0)
809 panic("ext2_rename: no name");
812 * Check for cross-device rename.
814 if ((fvp->v_mount != tdvp->v_mount) ||
815 (tvp && (fvp->v_mount != tvp->v_mount))) {
829 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
830 (VTOI(tdvp)->i_flags & APPEND))) {
836 * Renaming a file to itself has no effect. The upper layers should
837 * not call us in that case. Temporarily just warn if they do.
840 SDT_PROBE2(ext2fs, , vnops, trace, 1,
841 "rename: fvp == tvp (can't happen)");
846 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
850 if (ip->i_nlink >= EXT4_LINK_MAX &&
851 !EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
856 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
857 || (dp->i_flags & APPEND)) {
862 if ((ip->i_mode & IFMT) == IFDIR) {
864 * Avoid ".", "..", and aliases of "." for obvious reasons.
866 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
867 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
868 (ip->i_flag & IN_RENAME)) {
873 ip->i_flag |= IN_RENAME;
874 oldparent = dp->i_number;
880 * When the target exists, both the directory
881 * and target vnodes are returned locked.
889 * 1) Bump link count while we're moving stuff
890 * around. If we crash somewhere before
891 * completing our work, the link count
892 * may be wrong, but correctable.
895 ip->i_flag |= IN_CHANGE;
896 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
902 * If ".." must be changed (ie the directory gets a new
903 * parent) then the source directory must not be in the
904 * directory hierarchy above the target, as this would
905 * orphan everything below the source directory. Also
906 * the user must have write permission in the source so
907 * as to be able to change "..". We must repeat the call
908 * to namei, as the parent directory is unlocked by the
909 * call to checkpath().
911 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
913 if (oldparent != dp->i_number)
914 newparent = dp->i_number;
915 if (doingdirectory && newparent) {
916 if (error) /* write access check above */
920 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
924 error = relookup(tdvp, &tvp, tcnp);
934 * 2) If target doesn't exist, link the target
935 * to the source and unlink the source.
936 * Otherwise, rewrite the target directory
937 * entry to reference the source inode and
938 * expunge the original entry's existence.
941 if (dp->i_devvp != ip->i_devvp)
942 panic("ext2_rename: EXDEV");
944 * Account for ".." in new directory.
945 * When source and destination have the same
946 * parent we don't fool with the link count.
948 if (doingdirectory && newparent) {
949 error = ext2_inc_nlink(dp);
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)) {
998 if (!doingdirectory) {
1003 } else if (doingdirectory) {
1007 error = ext2_dirrewrite(dp, ip, tcnp);
1011 * If the target directory is in the same
1012 * directory as the source directory,
1013 * decrement the link count on the parent
1014 * of the target directory.
1016 if (doingdirectory && !newparent) {
1018 dp->i_flag |= IN_CHANGE;
1022 * Adjust the link count of the target to
1023 * reflect the dirrewrite above. If this is
1024 * a directory it is empty and there are
1025 * no links to it, so we can squash the inode and
1026 * any space associated with it. We disallowed
1027 * renaming over top of a directory with links to
1028 * it above, as the remaining link would point to
1029 * a directory without "." or ".." entries.
1032 if (doingdirectory) {
1033 if (xp->i_nlink > 2)
1034 panic("ext2_rename: linked directory");
1035 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1036 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. IN_RENAME is not sufficient
1059 * to protect against directory races due to timing windows,
1060 * so we can't panic here.
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 * From name resolves to a different inode. IN_RENAME is
1078 * not sufficient protection against timing window races
1079 * so we can't panic here.
1083 * If the source is a directory with a
1084 * new parent, the link count of the old
1085 * parent directory must be decremented
1086 * and ".." set to point to the new parent.
1088 if (doingdirectory && newparent) {
1090 dp->i_flag |= IN_CHANGE;
1091 dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO);
1092 error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf,
1093 ip->i_e2fs->e2fs_bsize, (off_t)0,
1094 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1095 tcnp->cn_cred, NOCRED, NULL, NULL);
1097 /* Like ufs little-endian: */
1098 namlen = dirbuf->dotdot_type;
1100 dirbuf->dotdot_name[0] != '.' ||
1101 dirbuf->dotdot_name[1] != '.') {
1102 ext2_dirbad(xp, (doff_t)12,
1103 "rename: mangled dir");
1105 dirbuf->dotdot_ino = htole32(newparent);
1107 * dirblock 0 could be htree root,
1108 * try both csum update functions.
1110 ext2_dirent_csum_set(ip,
1111 (struct ext2fs_direct_2 *)dirbuf);
1112 ext2_dx_csum_set(ip,
1113 (struct ext2fs_direct_2 *)dirbuf);
1114 (void)vn_rdwr(UIO_WRITE, fvp,
1116 ip->i_e2fs->e2fs_bsize,
1117 (off_t)0, UIO_SYSSPACE,
1118 IO_NODELOCKED | IO_SYNC |
1119 IO_NOMACCHECK, tcnp->cn_cred,
1120 NOCRED, NULL, NULL);
1124 free(dirbuf, M_TEMP);
1126 error = ext2_dirremove(fdvp, fcnp);
1129 xp->i_flag |= IN_CHANGE;
1131 xp->i_flag &= ~IN_RENAME;
1146 ip->i_flag &= ~IN_RENAME;
1147 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1149 ip->i_flag |= IN_CHANGE;
1150 ip->i_flag &= ~IN_RENAME;
1159 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1160 mode_t dmode, struct ucred *cred, struct thread *td)
1163 struct inode *ip = VTOI(tvp);
1164 struct acl *dacl, *acl;
1166 acl = acl_alloc(M_WAITOK);
1167 dacl = acl_alloc(M_WAITOK);
1170 * Retrieve default ACL from parent, if any.
1172 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1176 * Retrieved a default ACL, so merge mode and ACL if
1177 * necessary. If the ACL is empty, fall through to
1178 * the "not defined or available" case.
1180 if (acl->acl_cnt != 0) {
1181 dmode = acl_posix1e_newfilemode(dmode, acl);
1184 ext2_sync_acl_from_inode(ip, acl);
1191 * Just use the mode as-is.
1201 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1203 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1210 * XXX: This should not happen, as EOPNOTSUPP above
1211 * was supposed to free acl.
1214 printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1230 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1231 mode_t mode, struct ucred *cred, struct thread *td)
1234 struct inode *ip = VTOI(tvp);
1237 acl = acl_alloc(M_WAITOK);
1240 * Retrieve default ACL for parent, if any.
1242 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1246 * Retrieved a default ACL, so merge mode and ACL if
1249 if (acl->acl_cnt != 0) {
1251 * Two possible ways for default ACL to not
1252 * be present. First, the EA can be
1253 * undefined, or second, the default ACL can
1254 * be blank. If it's blank, fall through to
1255 * the it's not defined case.
1257 mode = acl_posix1e_newfilemode(mode, acl);
1259 ext2_sync_acl_from_inode(ip, acl);
1266 * Just use the mode as-is.
1276 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1283 * XXX: This should not happen, as EOPNOTSUPP above was
1284 * supposed to free acl.
1286 printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1287 "but no VOP_SETACL()\n");
1288 /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1289 "but no VOP_SETACL()"); */
1302 #endif /* UFS_ACL */
1308 ext2_mkdir(struct vop_mkdir_args *ap)
1310 struct m_ext2fs *fs;
1311 struct vnode *dvp = ap->a_dvp;
1312 struct vattr *vap = ap->a_vap;
1313 struct componentname *cnp = ap->a_cnp;
1314 struct inode *ip, *dp;
1316 struct dirtemplate dirtemplate, *dtp;
1321 if ((cnp->cn_flags & HASBUF) == 0)
1322 panic("ext2_mkdir: no name");
1325 if ((nlink_t)dp->i_nlink >= EXT4_LINK_MAX &&
1326 !EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
1330 dmode = vap->va_mode & 0777;
1333 * Must simulate part of ext2_makeinode here to acquire the inode,
1334 * but not have it entered in the parent directory. The entry is
1335 * made later after writing "." and ".." entries.
1337 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1342 ip->i_gid = dp->i_gid;
1346 * if we are hacking owners here, (only do this where told to)
1347 * and we are not giving it TOO root, (would subvert quotas)
1348 * then go ahead and give it to the other user.
1349 * The new directory also inherits the SUID bit.
1350 * If user's UID and dir UID are the same,
1351 * 'give it away' so that the SUID is still forced on.
1353 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1354 (dp->i_mode & ISUID) && dp->i_uid) {
1356 ip->i_uid = dp->i_uid;
1358 ip->i_uid = cnp->cn_cred->cr_uid;
1362 ip->i_uid = cnp->cn_cred->cr_uid;
1364 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1366 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1368 if (cnp->cn_flags & ISWHITEOUT)
1369 ip->i_flags |= UF_OPAQUE;
1370 error = ext2_update(tvp, 1);
1373 * Bump link count in parent directory
1374 * to reflect work done below. Should
1375 * be done before reference is created
1376 * so reparation is possible if we crash.
1379 dp->i_flag |= IN_CHANGE;
1380 error = ext2_update(dvp, !DOINGASYNC(dvp));
1384 /* Initialize directory with "." and ".." from static template. */
1385 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1386 EXT2F_INCOMPAT_FTYPE))
1387 dtp = &mastertemplate;
1389 dtp = &omastertemplate;
1391 dirtemplate.dot_ino = htole32(ip->i_number);
1392 dirtemplate.dotdot_ino = htole32(dp->i_number);
1394 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1395 * just redefine it - for this function only
1398 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize
1399 dirtemplate.dotdot_reclen = htole16(DIRBLKSIZ - 12);
1400 buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO);
1401 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1402 dirtemplate.dotdot_reclen =
1403 htole16(le16toh(dirtemplate.dotdot_reclen) -
1404 sizeof(struct ext2fs_direct_tail));
1405 ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ));
1407 memcpy(buf, &dirtemplate, sizeof(dirtemplate));
1408 ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf);
1409 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf,
1410 DIRBLKSIZ, (off_t)0, UIO_SYSSPACE,
1411 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1415 dp->i_flag |= IN_CHANGE;
1418 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1419 /* XXX should grow with balloc() */
1420 panic("ext2_mkdir: blksize");
1422 ip->i_size = DIRBLKSIZ;
1423 ip->i_flag |= IN_CHANGE;
1427 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1428 error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1429 cnp->cn_cred, cnp->cn_thread);
1434 #endif /* UFS_ACL */
1436 /* Directory set up, now install its entry in the parent directory. */
1437 error = ext2_direnter(ip, dvp, cnp);
1440 dp->i_flag |= IN_CHANGE;
1444 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1445 * for us because we set the link count to 0.
1449 ip->i_flag |= IN_CHANGE;
1457 #define DIRBLKSIZ DEV_BSIZE
1461 * Rmdir system call.
1464 ext2_rmdir(struct vop_rmdir_args *ap)
1466 struct vnode *vp = ap->a_vp;
1467 struct vnode *dvp = ap->a_dvp;
1468 struct componentname *cnp = ap->a_cnp;
1469 struct inode *ip, *dp;
1476 * Verify the directory is empty (and valid).
1477 * (Rmdir ".." won't be valid since
1478 * ".." will contain a reference to
1479 * the current directory and thus be
1482 if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1486 if ((dp->i_flags & APPEND)
1487 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1492 * Delete reference to directory before purging
1493 * inode. If we crash in between, the directory
1494 * will be reattached to lost+found,
1496 error = ext2_dirremove(dvp, cnp);
1500 dp->i_flag |= IN_CHANGE;
1504 * Truncate inode. The only stuff left
1505 * in the directory is "." and "..".
1508 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1510 cache_purge(ITOV(ip));
1511 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1513 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1514 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1521 * symlink -- make a symbolic link
1524 ext2_symlink(struct vop_symlink_args *ap)
1526 struct vnode *vp, **vpp = ap->a_vpp;
1530 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1535 len = strlen(ap->a_target);
1536 if (len < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) {
1538 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1540 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1542 error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target),
1543 len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1544 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1551 * Return target name of a symbolic link
1554 ext2_readlink(struct vop_readlink_args *ap)
1556 struct vnode *vp = ap->a_vp;
1557 struct inode *ip = VTOI(vp);
1561 if (isize < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) {
1562 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1565 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1569 * Calculate the logical to physical mapping if not done already,
1570 * then call the device strategy routine.
1572 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1573 * deadlock on memory. See ext2_bmap() for details.
1576 ext2_strategy(struct vop_strategy_args *ap)
1578 struct buf *bp = ap->a_bp;
1579 struct vnode *vp = ap->a_vp;
1584 if (vp->v_type == VBLK || vp->v_type == VCHR)
1585 panic("ext2_strategy: spec");
1586 if (bp->b_blkno == bp->b_lblkno) {
1587 if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
1588 error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL);
1590 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1592 bp->b_blkno = blkno;
1594 bp->b_error = error;
1595 bp->b_ioflags |= BIO_ERROR;
1599 if ((long)bp->b_blkno == -1)
1602 if ((long)bp->b_blkno == -1) {
1606 bp->b_iooffset = dbtob(bp->b_blkno);
1607 bo = VFSTOEXT2(vp->v_mount)->um_bo;
1608 BO_STRATEGY(bo, bp);
1613 * Print out the contents of an inode.
1616 ext2_print(struct vop_print_args *ap)
1618 struct vnode *vp = ap->a_vp;
1619 struct inode *ip = VTOI(vp);
1621 vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1622 if (vp->v_type == VFIFO)
1629 * Close wrapper for fifos.
1631 * Update the times on the inode then do device close.
1634 ext2fifo_close(struct vop_close_args *ap)
1636 struct vnode *vp = ap->a_vp;
1639 if (vp->v_usecount > 1)
1640 ext2_itimes_locked(vp);
1642 return (fifo_specops.vop_close(ap));
1646 * Kqfilter wrapper for fifos.
1648 * Fall through to ext2 kqfilter routines if needed
1651 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1655 error = fifo_specops.vop_kqfilter(ap);
1657 error = vfs_kqfilter(ap);
1662 * Return POSIX pathconf information applicable to ext2 filesystems.
1665 ext2_pathconf(struct vop_pathconf_args *ap)
1669 switch (ap->a_name) {
1671 if (EXT2_HAS_RO_COMPAT_FEATURE(VTOI(ap->a_vp)->i_e2fs,
1672 EXT2F_ROCOMPAT_DIR_NLINK))
1673 *ap->a_retval = INT_MAX;
1675 *ap->a_retval = EXT4_LINK_MAX;
1678 *ap->a_retval = NAME_MAX;
1681 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
1682 *ap->a_retval = PIPE_BUF;
1686 case _PC_CHOWN_RESTRICTED:
1694 case _PC_ACL_EXTENDED:
1695 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1700 case _PC_ACL_PATH_MAX:
1701 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1702 *ap->a_retval = ACL_MAX_ENTRIES;
1706 #endif /* UFS_ACL */
1708 case _PC_MIN_HOLE_SIZE:
1709 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1717 case _PC_ALLOC_SIZE_MIN:
1718 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1720 case _PC_FILESIZEBITS:
1723 case _PC_REC_INCR_XFER_SIZE:
1724 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1726 case _PC_REC_MAX_XFER_SIZE:
1727 *ap->a_retval = -1; /* means ``unlimited'' */
1729 case _PC_REC_MIN_XFER_SIZE:
1730 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1732 case _PC_REC_XFER_ALIGN:
1733 *ap->a_retval = PAGE_SIZE;
1735 case _PC_SYMLINK_MAX:
1736 *ap->a_retval = MAXPATHLEN;
1740 error = vop_stdpathconf(ap);
1747 * Vnode operation to remove a named attribute.
1750 ext2_deleteextattr(struct vop_deleteextattr_args *ap)
1753 struct m_ext2fs *fs;
1756 ip = VTOI(ap->a_vp);
1759 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1760 return (EOPNOTSUPP);
1762 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1763 return (EOPNOTSUPP);
1765 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1766 ap->a_cred, ap->a_td, VWRITE);
1772 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1773 error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1774 if (error != ENOATTR)
1779 error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1785 * Vnode operation to retrieve a named extended attribute.
1788 ext2_getextattr(struct vop_getextattr_args *ap)
1791 struct m_ext2fs *fs;
1794 ip = VTOI(ap->a_vp);
1797 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1798 return (EOPNOTSUPP);
1800 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1801 return (EOPNOTSUPP);
1803 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1804 ap->a_cred, ap->a_td, VREAD);
1808 if (ap->a_size != NULL)
1813 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1814 error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1815 ap->a_name, ap->a_uio, ap->a_size);
1816 if (error != ENOATTR)
1821 error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1822 ap->a_name, ap->a_uio, ap->a_size);
1828 * Vnode operation to retrieve extended attributes on a vnode.
1831 ext2_listextattr(struct vop_listextattr_args *ap)
1834 struct m_ext2fs *fs;
1837 ip = VTOI(ap->a_vp);
1840 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1841 return (EOPNOTSUPP);
1843 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1844 return (EOPNOTSUPP);
1846 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1847 ap->a_cred, ap->a_td, VREAD);
1851 if (ap->a_size != NULL)
1854 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1855 error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1856 ap->a_uio, ap->a_size);
1862 error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1863 ap->a_uio, ap->a_size);
1869 * Vnode operation to set a named attribute.
1872 ext2_setextattr(struct vop_setextattr_args *ap)
1875 struct m_ext2fs *fs;
1878 ip = VTOI(ap->a_vp);
1881 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1882 return (EOPNOTSUPP);
1884 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1885 return (EOPNOTSUPP);
1887 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1888 ap->a_cred, ap->a_td, VWRITE);
1892 error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1896 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1897 error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1898 ap->a_name, ap->a_uio);
1899 if (error != ENOSPC)
1903 error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1904 ap->a_name, ap->a_uio);
1910 * Vnode pointer to File handle
1914 ext2_vptofh(struct vop_vptofh_args *ap)
1919 ip = VTOI(ap->a_vp);
1920 ufhp = (struct ufid *)ap->a_fhp;
1921 ufhp->ufid_len = sizeof(struct ufid);
1922 ufhp->ufid_ino = ip->i_number;
1923 ufhp->ufid_gen = ip->i_gen;
1928 * Initialize the vnode associated with a new inode, handle aliased
1932 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1939 vp->v_type = IFTOVT(ip->i_mode);
1941 * Only unallocated inodes should be of type VNON.
1943 if (ip->i_mode != 0 && vp->v_type == VNON)
1945 if (vp->v_type == VFIFO)
1948 if (ip->i_number == EXT2_ROOTINO)
1949 vp->v_vflag |= VV_ROOT;
1950 ip->i_modrev = init_va_filerev();
1956 * Allocate a new inode.
1959 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1960 struct componentname *cnp)
1962 struct inode *ip, *pdir;
1968 if ((cnp->cn_flags & HASBUF) == 0)
1969 panic("ext2_makeinode: no name");
1972 if ((mode & IFMT) == 0)
1975 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1980 ip->i_gid = pdir->i_gid;
1985 * not the owner of the directory,
1986 * and we are hacking owners here, (only do this where told to)
1987 * and we are not giving it TOO root, (would subvert quotas)
1988 * then go ahead and give it to the other user.
1989 * Note that this drops off the execute bits for security.
1991 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1992 (pdir->i_mode & ISUID) &&
1993 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1994 ip->i_uid = pdir->i_uid;
1997 ip->i_uid = cnp->cn_cred->cr_uid;
2001 ip->i_uid = cnp->cn_cred->cr_uid;
2003 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2005 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2007 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
2008 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID))
2009 ip->i_mode &= ~ISGID;
2012 if (cnp->cn_flags & ISWHITEOUT)
2013 ip->i_flags |= UF_OPAQUE;
2016 * Make sure inode goes to disk before directory entry.
2018 error = ext2_update(tvp, !DOINGASYNC(tvp));
2023 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2024 error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
2025 cnp->cn_cred, cnp->cn_thread);
2029 #endif /* UFS_ACL */
2031 error = ext2_direnter(ip, dvp, cnp);
2040 * Write error occurred trying to update the inode
2041 * or the directory so must deallocate the inode.
2044 ip->i_flag |= IN_CHANGE;
2050 * Vnode op for reading.
2053 ext2_read(struct vop_read_args *ap)
2058 struct m_ext2fs *fs;
2060 daddr_t lbn, nextlbn;
2062 long size, xfersize, blkoffset;
2063 int error, orig_resid, seqcount;
2068 ioflag = ap->a_ioflag;
2070 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
2074 if (uio->uio_rw != UIO_READ)
2075 panic("%s: mode", "ext2_read");
2077 if (vp->v_type == VLNK) {
2078 if ((int)ip->i_size <
2079 VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen)
2080 panic("%s: short symlink", "ext2_read");
2081 } else if (vp->v_type != VREG && vp->v_type != VDIR)
2082 panic("%s: type %d", "ext2_read", vp->v_type);
2084 orig_resid = uio->uio_resid;
2085 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
2086 if (orig_resid == 0)
2088 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
2090 if (uio->uio_offset < ip->i_size &&
2091 uio->uio_offset >= fs->e2fs_maxfilesize)
2094 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
2095 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
2097 lbn = lblkno(fs, uio->uio_offset);
2099 size = blksize(fs, ip, lbn);
2100 blkoffset = blkoff(fs, uio->uio_offset);
2102 xfersize = fs->e2fs_fsize - blkoffset;
2103 if (uio->uio_resid < xfersize)
2104 xfersize = uio->uio_resid;
2105 if (bytesinfile < xfersize)
2106 xfersize = bytesinfile;
2108 if (lblktosize(fs, nextlbn) >= ip->i_size)
2109 error = bread(vp, lbn, size, NOCRED, &bp);
2110 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
2111 error = cluster_read(vp, ip->i_size, lbn, size,
2112 NOCRED, blkoffset + uio->uio_resid, seqcount,
2114 } else if (seqcount > 1) {
2115 u_int nextsize = blksize(fs, ip, nextlbn);
2117 error = breadn(vp, lbn,
2118 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
2120 error = bread(vp, lbn, size, NOCRED, &bp);
2128 * We should only get non-zero b_resid when an I/O error
2129 * has occurred, which should cause us to break above.
2130 * However, if the short read did not cause an error,
2131 * then we want to ensure that we do not uiomove bad
2132 * or uninitialized data.
2134 size -= bp->b_resid;
2135 if (size < xfersize) {
2140 error = uiomove((char *)bp->b_data + blkoffset,
2141 (int)xfersize, uio);
2144 vfs_bio_brelse(bp, ioflag);
2148 * This can only happen in the case of an error because the loop
2149 * above resets bp to NULL on each iteration and on normal
2150 * completion has not set a new value into it. so it must have come
2151 * from a 'break' statement
2154 vfs_bio_brelse(bp, ioflag);
2156 if ((error == 0 || uio->uio_resid != orig_resid) &&
2157 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
2158 ip->i_flag |= IN_ACCESS;
2163 ext2_ioctl(struct vop_ioctl_args *ap)
2169 switch (ap->a_command) {
2171 if (!(VTOI(vp)->i_flag & IN_E4EXTENTS)) {
2172 error = vn_lock(vp, LK_SHARED);
2174 error = ext2_bmap_seekdata(vp,
2175 (off_t *)ap->a_data);
2182 return (vn_bmap_seekhole(vp, ap->a_command,
2183 (off_t *)ap->a_data, ap->a_cred));
2190 * Vnode op for writing.
2193 ext2_write(struct vop_write_args *ap)
2198 struct m_ext2fs *fs;
2202 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2204 ioflag = ap->a_ioflag;
2208 seqcount = ioflag >> IO_SEQSHIFT;
2212 if (uio->uio_rw != UIO_WRITE)
2213 panic("%s: mode", "ext2_write");
2216 switch (vp->v_type) {
2218 if (ioflag & IO_APPEND)
2219 uio->uio_offset = ip->i_size;
2220 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2226 /* XXX differs from ffs -- this is called from ext2_mkdir(). */
2227 if ((ioflag & IO_SYNC) == 0)
2228 panic("ext2_write: nonsync dir write");
2231 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2232 vp->v_type, (intmax_t)uio->uio_offset,
2233 (intmax_t)uio->uio_resid);
2236 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2237 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2239 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2242 * Maybe this should be above the vnode op call, but so long as
2243 * file servers have no limits, I don't think it matters.
2245 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2248 resid = uio->uio_resid;
2250 if (seqcount > BA_SEQMAX)
2251 flags = BA_SEQMAX << BA_SEQSHIFT;
2253 flags = seqcount << BA_SEQSHIFT;
2254 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2257 for (error = 0; uio->uio_resid > 0;) {
2258 lbn = lblkno(fs, uio->uio_offset);
2259 blkoffset = blkoff(fs, uio->uio_offset);
2260 xfersize = fs->e2fs_fsize - blkoffset;
2261 if (uio->uio_resid < xfersize)
2262 xfersize = uio->uio_resid;
2263 if (uio->uio_offset + xfersize > ip->i_size)
2264 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2267 * We must perform a read-before-write if the transfer size
2268 * does not cover the entire buffer.
2270 if (fs->e2fs_bsize > xfersize)
2273 flags &= ~BA_CLRBUF;
2274 error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2275 ap->a_cred, &bp, flags);
2279 if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2280 bp->b_flags |= B_NOCACHE;
2281 if (uio->uio_offset + xfersize > ip->i_size)
2282 ip->i_size = uio->uio_offset + xfersize;
2283 size = blksize(fs, ip, lbn) - bp->b_resid;
2284 if (size < xfersize)
2288 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2290 * If the buffer is not already filled and we encounter an
2291 * error while trying to fill it, we have to clear out any
2292 * garbage data from the pages instantiated for the buffer.
2293 * If we do not, a failed uiomove() during a write can leave
2294 * the prior contents of the pages exposed to a userland mmap.
2296 * Note that we need only clear buffers with a transfer size
2297 * equal to the block size because buffers with a shorter
2298 * transfer size were cleared above by the call to ext2_balloc()
2299 * with the BA_CLRBUF flag set.
2301 * If the source region for uiomove identically mmaps the
2302 * buffer, uiomove() performed the NOP copy, and the buffer
2303 * content remains valid because the page fault handler
2304 * validated the pages.
2306 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2307 fs->e2fs_bsize == xfersize)
2310 vfs_bio_set_flags(bp, ioflag);
2313 * If IO_SYNC each buffer is written synchronously. Otherwise
2314 * if we have a severe page deficiency write the buffer
2315 * asynchronously. Otherwise try to cluster, and if that
2316 * doesn't do it then either do an async write (if O_DIRECT),
2317 * or a delayed write (if not).
2319 if (ioflag & IO_SYNC) {
2321 } else if (vm_page_count_severe() ||
2322 buf_dirty_count_severe() ||
2323 (ioflag & IO_ASYNC)) {
2324 bp->b_flags |= B_CLUSTEROK;
2326 } else if (xfersize + blkoffset == fs->e2fs_fsize) {
2327 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2328 bp->b_flags |= B_CLUSTEROK;
2329 cluster_write(vp, &ip->i_clusterw, bp,
2330 ip->i_size, seqcount, 0);
2334 } else if (ioflag & IO_DIRECT) {
2335 bp->b_flags |= B_CLUSTEROK;
2338 bp->b_flags |= B_CLUSTEROK;
2341 if (error || xfersize == 0)
2345 * If we successfully wrote any data, and we are not the superuser
2346 * we clear the setuid and setgid bits as a precaution against
2349 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2351 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID))
2352 ip->i_mode &= ~(ISUID | ISGID);
2355 if (ioflag & IO_UNIT) {
2356 (void)ext2_truncate(vp, osize,
2357 ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2358 uio->uio_offset -= resid - uio->uio_resid;
2359 uio->uio_resid = resid;
2362 if (uio->uio_resid != resid) {
2363 ip->i_flag |= IN_CHANGE | IN_UPDATE;
2364 if (ioflag & IO_SYNC)
2365 error = ext2_update(vp, 1);