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/rwlock.h>
58 #include <sys/mount.h>
59 #include <sys/unistd.h>
61 #include <sys/vnode.h>
62 #include <sys/namei.h>
63 #include <sys/lockf.h>
64 #include <sys/event.h>
69 #include <vm/vm_param.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_pager.h>
74 #include <vm/vnode_pager.h>
76 #include "opt_directio.h"
78 #include <ufs/ufs/dir.h>
80 #include <fs/ext2fs/fs.h>
81 #include <fs/ext2fs/inode.h>
82 #include <fs/ext2fs/ext2_extern.h>
83 #include <fs/ext2fs/ext2fs.h>
84 #include <fs/ext2fs/ext2_dinode.h>
85 #include <fs/ext2fs/ext2_dir.h>
86 #include <fs/ext2fs/ext2_mount.h>
88 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
89 static void ext2_itimes_locked(struct vnode *);
90 static int ext4_ext_read(struct vop_read_args *);
91 static int ext2_ind_read(struct vop_read_args *);
93 static vop_access_t ext2_access;
94 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
95 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
97 static vop_close_t ext2_close;
98 static vop_create_t ext2_create;
99 static vop_fsync_t ext2_fsync;
100 static vop_getattr_t ext2_getattr;
101 static vop_ioctl_t ext2_ioctl;
102 static vop_link_t ext2_link;
103 static vop_mkdir_t ext2_mkdir;
104 static vop_mknod_t ext2_mknod;
105 static vop_open_t ext2_open;
106 static vop_pathconf_t ext2_pathconf;
107 static vop_print_t ext2_print;
108 static vop_read_t ext2_read;
109 static vop_readlink_t ext2_readlink;
110 static vop_remove_t ext2_remove;
111 static vop_rename_t ext2_rename;
112 static vop_rmdir_t ext2_rmdir;
113 static vop_setattr_t ext2_setattr;
114 static vop_strategy_t ext2_strategy;
115 static vop_symlink_t ext2_symlink;
116 static vop_write_t ext2_write;
117 static vop_vptofh_t ext2_vptofh;
118 static vop_close_t ext2fifo_close;
119 static vop_kqfilter_t ext2fifo_kqfilter;
121 /* Global vfs data structures for ext2. */
122 struct vop_vector ext2_vnodeops = {
123 .vop_default = &default_vnodeops,
124 .vop_access = ext2_access,
125 .vop_bmap = ext2_bmap,
126 .vop_cachedlookup = ext2_lookup,
127 .vop_close = ext2_close,
128 .vop_create = ext2_create,
129 .vop_fsync = ext2_fsync,
130 .vop_getpages = vnode_pager_local_getpages,
131 .vop_getattr = ext2_getattr,
132 .vop_inactive = ext2_inactive,
133 .vop_ioctl = ext2_ioctl,
134 .vop_link = ext2_link,
135 .vop_lookup = vfs_cache_lookup,
136 .vop_mkdir = ext2_mkdir,
137 .vop_mknod = ext2_mknod,
138 .vop_open = ext2_open,
139 .vop_pathconf = ext2_pathconf,
140 .vop_poll = vop_stdpoll,
141 .vop_print = ext2_print,
142 .vop_read = ext2_read,
143 .vop_readdir = ext2_readdir,
144 .vop_readlink = ext2_readlink,
145 .vop_reallocblks = ext2_reallocblks,
146 .vop_reclaim = ext2_reclaim,
147 .vop_remove = ext2_remove,
148 .vop_rename = ext2_rename,
149 .vop_rmdir = ext2_rmdir,
150 .vop_setattr = ext2_setattr,
151 .vop_strategy = ext2_strategy,
152 .vop_symlink = ext2_symlink,
153 .vop_write = ext2_write,
154 .vop_vptofh = ext2_vptofh,
157 struct vop_vector ext2_fifoops = {
158 .vop_default = &fifo_specops,
159 .vop_access = ext2_access,
160 .vop_close = ext2fifo_close,
161 .vop_fsync = ext2_fsync,
162 .vop_getattr = ext2_getattr,
163 .vop_inactive = ext2_inactive,
164 .vop_kqfilter = ext2fifo_kqfilter,
165 .vop_print = ext2_print,
166 .vop_read = VOP_PANIC,
167 .vop_reclaim = ext2_reclaim,
168 .vop_setattr = ext2_setattr,
169 .vop_write = VOP_PANIC,
170 .vop_vptofh = ext2_vptofh,
174 * A virgin directory (no blushing please).
175 * Note that the type and namlen fields are reversed relative to ext2.
176 * Also, we don't use `struct odirtemplate', since it would just cause
177 * endianness problems.
179 static struct dirtemplate mastertemplate = {
180 0, 12, 1, EXT2_FT_DIR, ".",
181 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
183 static struct dirtemplate omastertemplate = {
184 0, 12, 1, EXT2_FT_UNKNOWN, ".",
185 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
189 ext2_itimes_locked(struct vnode *vp)
194 ASSERT_VI_LOCKED(vp, __func__);
197 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
199 if ((vp->v_type == VBLK || vp->v_type == VCHR))
200 ip->i_flag |= IN_LAZYMOD;
202 ip->i_flag |= IN_MODIFIED;
203 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
205 if (ip->i_flag & IN_ACCESS) {
206 ip->i_atime = ts.tv_sec;
207 ip->i_atimensec = ts.tv_nsec;
209 if (ip->i_flag & IN_UPDATE) {
210 ip->i_mtime = ts.tv_sec;
211 ip->i_mtimensec = ts.tv_nsec;
214 if (ip->i_flag & IN_CHANGE) {
215 ip->i_ctime = ts.tv_sec;
216 ip->i_ctimensec = ts.tv_nsec;
219 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
223 ext2_itimes(struct vnode *vp)
227 ext2_itimes_locked(vp);
232 * Create a regular file
235 ext2_create(struct vop_create_args *ap)
240 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
241 ap->a_dvp, ap->a_vpp, ap->a_cnp);
244 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
245 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
250 ext2_open(struct vop_open_args *ap)
253 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
257 * Files marked append-only must be opened for appending.
259 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
260 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
263 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
271 * Update the times on the inode.
274 ext2_close(struct vop_close_args *ap)
276 struct vnode *vp = ap->a_vp;
279 if (vp->v_usecount > 1)
280 ext2_itimes_locked(vp);
286 ext2_access(struct vop_access_args *ap)
288 struct vnode *vp = ap->a_vp;
289 struct inode *ip = VTOI(vp);
290 accmode_t accmode = ap->a_accmode;
293 if (vp->v_type == VBLK || vp->v_type == VCHR)
297 * Disallow write attempts on read-only file systems;
298 * unless the file is a socket, fifo, or a block or
299 * character device resident on the file system.
301 if (accmode & VWRITE) {
302 switch (vp->v_type) {
306 if (vp->v_mount->mnt_flag & MNT_RDONLY)
314 /* If immutable bit set, nobody gets to write it. */
315 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
318 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
319 ap->a_accmode, ap->a_cred, NULL);
324 ext2_getattr(struct vop_getattr_args *ap)
326 struct vnode *vp = ap->a_vp;
327 struct inode *ip = VTOI(vp);
328 struct vattr *vap = ap->a_vap;
332 * Copy from inode table
334 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
335 vap->va_fileid = ip->i_number;
336 vap->va_mode = ip->i_mode & ~IFMT;
337 vap->va_nlink = ip->i_nlink;
338 vap->va_uid = ip->i_uid;
339 vap->va_gid = ip->i_gid;
340 vap->va_rdev = ip->i_rdev;
341 vap->va_size = ip->i_size;
342 vap->va_atime.tv_sec = ip->i_atime;
343 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
344 vap->va_mtime.tv_sec = ip->i_mtime;
345 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
346 vap->va_ctime.tv_sec = ip->i_ctime;
347 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
348 if E2DI_HAS_XTIME(ip) {
349 vap->va_birthtime.tv_sec = ip->i_birthtime;
350 vap->va_birthtime.tv_nsec = ip->i_birthnsec;
352 vap->va_flags = ip->i_flags;
353 vap->va_gen = ip->i_gen;
354 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
355 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
356 vap->va_type = IFTOVT(ip->i_mode);
357 vap->va_filerev = ip->i_modrev;
362 * Set attribute vnode op. called from several syscalls
365 ext2_setattr(struct vop_setattr_args *ap)
367 struct vattr *vap = ap->a_vap;
368 struct vnode *vp = ap->a_vp;
369 struct inode *ip = VTOI(vp);
370 struct ucred *cred = ap->a_cred;
371 struct thread *td = curthread;
375 * Check for unsettable attributes.
377 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
378 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
379 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
380 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
383 if (vap->va_flags != VNOVAL) {
384 /* Disallow flags not supported by ext2fs. */
385 if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
388 if (vp->v_mount->mnt_flag & MNT_RDONLY)
391 * Callers may only modify the file flags on objects they
392 * have VADMIN rights for.
394 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
397 * Unprivileged processes and privileged processes in
398 * jail() are not permitted to unset system flags, or
399 * modify flags if any system flags are set.
400 * Privileged non-jail processes may not modify system flags
401 * if securelevel > 0 and any existing system flags are set.
403 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
404 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
405 error = securelevel_gt(cred, 0);
410 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
411 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
414 ip->i_flags = vap->va_flags;
415 ip->i_flag |= IN_CHANGE;
416 if (ip->i_flags & (IMMUTABLE | APPEND))
419 if (ip->i_flags & (IMMUTABLE | APPEND))
422 * Go through the fields and update iff not VNOVAL.
424 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
425 if (vp->v_mount->mnt_flag & MNT_RDONLY)
427 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
431 if (vap->va_size != VNOVAL) {
433 * Disallow write attempts on read-only file systems;
434 * unless the file is a socket, fifo, or a block or
435 * character device resident on the file system.
437 switch (vp->v_type) {
442 if (vp->v_mount->mnt_flag & MNT_RDONLY)
448 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
451 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
452 if (vp->v_mount->mnt_flag & MNT_RDONLY)
456 * If times is NULL, ... The caller must be the owner of
457 * the file, have permission to write the file, or be the
459 * If times is non-NULL, ... The caller must be the owner of
460 * the file or be the super-user.
462 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
463 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
464 (error = VOP_ACCESS(vp, VWRITE, cred, td))))
466 if (vap->va_atime.tv_sec != VNOVAL)
467 ip->i_flag |= IN_ACCESS;
468 if (vap->va_mtime.tv_sec != VNOVAL)
469 ip->i_flag |= IN_CHANGE | IN_UPDATE;
471 if (vap->va_atime.tv_sec != VNOVAL) {
472 ip->i_atime = vap->va_atime.tv_sec;
473 ip->i_atimensec = vap->va_atime.tv_nsec;
475 if (vap->va_mtime.tv_sec != VNOVAL) {
476 ip->i_mtime = vap->va_mtime.tv_sec;
477 ip->i_mtimensec = vap->va_mtime.tv_nsec;
479 ip->i_birthtime = vap->va_birthtime.tv_sec;
480 ip->i_birthnsec = vap->va_birthtime.tv_nsec;
481 error = ext2_update(vp, 0);
486 if (vap->va_mode != (mode_t)VNOVAL) {
487 if (vp->v_mount->mnt_flag & MNT_RDONLY)
489 error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
495 * Change the mode on a file.
496 * Inode must be locked before calling.
499 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
501 struct inode *ip = VTOI(vp);
505 * To modify the permissions on a file, must possess VADMIN
508 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
511 * Privileged processes may set the sticky bit on non-directories,
512 * as well as set the setgid bit on a file with a group that the
513 * process is not a member of.
515 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
516 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
520 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
521 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
525 ip->i_mode &= ~ALLPERMS;
526 ip->i_mode |= (mode & ALLPERMS);
527 ip->i_flag |= IN_CHANGE;
532 * Perform chown operation on inode ip;
533 * inode must be locked prior to call.
536 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
539 struct inode *ip = VTOI(vp);
544 if (uid == (uid_t)VNOVAL)
546 if (gid == (gid_t)VNOVAL)
549 * To modify the ownership of a file, must possess VADMIN
552 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
555 * To change the owner of a file, or change the group of a file
556 * to a group of which we are not a member, the caller must
559 if (uid != ip->i_uid || (gid != ip->i_gid &&
560 !groupmember(gid, cred))) {
561 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
569 ip->i_flag |= IN_CHANGE;
570 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
571 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
572 ip->i_mode &= ~(ISUID | ISGID);
578 * Synch an open file.
582 ext2_fsync(struct vop_fsync_args *ap)
585 * Flush all dirty buffers associated with a vnode.
590 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
598 ext2_mknod(struct vop_mknod_args *ap)
600 struct vattr *vap = ap->a_vap;
601 struct vnode **vpp = ap->a_vpp;
606 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
607 ap->a_dvp, vpp, ap->a_cnp);
611 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
612 if (vap->va_rdev != VNOVAL) {
614 * Want to be able to use this to make badblock
615 * inodes, so don't truncate the dev number.
617 ip->i_rdev = vap->va_rdev;
620 * Remove inode, then reload it through VFS_VGET so it is
621 * checked to see if it is an alias of an existing entry in
622 * the inode cache. XXX I don't believe this is necessary now.
624 (*vpp)->v_type = VNON;
625 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
628 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
637 ext2_remove(struct vop_remove_args *ap)
640 struct vnode *vp = ap->a_vp;
641 struct vnode *dvp = ap->a_dvp;
645 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
646 (VTOI(dvp)->i_flags & APPEND)) {
650 error = ext2_dirremove(dvp, ap->a_cnp);
653 ip->i_flag |= IN_CHANGE;
663 ext2_link(struct vop_link_args *ap)
665 struct vnode *vp = ap->a_vp;
666 struct vnode *tdvp = ap->a_tdvp;
667 struct componentname *cnp = ap->a_cnp;
672 if ((cnp->cn_flags & HASBUF) == 0)
673 panic("ext2_link: no name");
676 if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
680 if (ip->i_flags & (IMMUTABLE | APPEND)) {
685 ip->i_flag |= IN_CHANGE;
686 error = ext2_update(vp, !DOINGASYNC(vp));
688 error = ext2_direnter(ip, tdvp, cnp);
691 ip->i_flag |= IN_CHANGE;
698 * Rename system call.
699 * rename("foo", "bar");
702 * link("foo", "bar");
704 * but ``atomically''. Can't do full commit without saving state in the
705 * inode on disk which isn't feasible at this time. Best we can do is
706 * always guarantee the target exists.
708 * Basic algorithm is:
710 * 1) Bump link count on source while we're linking it to the
711 * target. This also ensure the inode won't be deleted out
712 * from underneath us while we work (it may be truncated by
713 * a concurrent `trunc' or `open' for creation).
714 * 2) Link source to destination. If destination already exists,
716 * 3) Unlink source reference to inode if still around. If a
717 * directory was moved and the parent of the destination
718 * is different from the source, patch the ".." entry in the
722 ext2_rename(struct vop_rename_args *ap)
724 struct vnode *tvp = ap->a_tvp;
725 struct vnode *tdvp = ap->a_tdvp;
726 struct vnode *fvp = ap->a_fvp;
727 struct vnode *fdvp = ap->a_fdvp;
728 struct componentname *tcnp = ap->a_tcnp;
729 struct componentname *fcnp = ap->a_fcnp;
730 struct inode *ip, *xp, *dp;
731 struct dirtemplate dirbuf;
732 int doingdirectory = 0, oldparent = 0, newparent = 0;
737 if ((tcnp->cn_flags & HASBUF) == 0 ||
738 (fcnp->cn_flags & HASBUF) == 0)
739 panic("ext2_rename: no name");
742 * Check for cross-device rename.
744 if ((fvp->v_mount != tdvp->v_mount) ||
745 (tvp && (fvp->v_mount != tvp->v_mount))) {
759 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
760 (VTOI(tdvp)->i_flags & APPEND))) {
766 * Renaming a file to itself has no effect. The upper layers should
767 * not call us in that case. Temporarily just warn if they do.
770 printf("ext2_rename: fvp == tvp (can't happen)\n");
775 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
779 if (ip->i_nlink >= EXT2_LINK_MAX) {
784 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
785 || (dp->i_flags & APPEND)) {
790 if ((ip->i_mode & IFMT) == IFDIR) {
792 * Avoid ".", "..", and aliases of "." for obvious reasons.
794 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
795 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
796 (ip->i_flag & IN_RENAME)) {
801 ip->i_flag |= IN_RENAME;
802 oldparent = dp->i_number;
808 * When the target exists, both the directory
809 * and target vnodes are returned locked.
817 * 1) Bump link count while we're moving stuff
818 * around. If we crash somewhere before
819 * completing our work, the link count
820 * may be wrong, but correctable.
823 ip->i_flag |= IN_CHANGE;
824 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
830 * If ".." must be changed (ie the directory gets a new
831 * parent) then the source directory must not be in the
832 * directory hierarchy above the target, as this would
833 * orphan everything below the source directory. Also
834 * the user must have write permission in the source so
835 * as to be able to change "..". We must repeat the call
836 * to namei, as the parent directory is unlocked by the
837 * call to checkpath().
839 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
841 if (oldparent != dp->i_number)
842 newparent = dp->i_number;
843 if (doingdirectory && newparent) {
844 if (error) /* write access check above */
848 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
852 error = relookup(tdvp, &tvp, tcnp);
862 * 2) If target doesn't exist, link the target
863 * to the source and unlink the source.
864 * Otherwise, rewrite the target directory
865 * entry to reference the source inode and
866 * expunge the original entry's existence.
869 if (dp->i_devvp != ip->i_devvp)
870 panic("ext2_rename: EXDEV");
872 * Account for ".." in new directory.
873 * When source and destination have the same
874 * parent we don't fool with the link count.
876 if (doingdirectory && newparent) {
877 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
882 dp->i_flag |= IN_CHANGE;
883 error = ext2_update(tdvp, !DOINGASYNC(tdvp));
887 error = ext2_direnter(ip, tdvp, tcnp);
889 if (doingdirectory && newparent) {
891 dp->i_flag |= IN_CHANGE;
892 (void)ext2_update(tdvp, 1);
898 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
899 panic("ext2_rename: EXDEV");
901 * Short circuit rename(foo, foo).
903 if (xp->i_number == ip->i_number)
904 panic("ext2_rename: same file");
906 * If the parent directory is "sticky", then the user must
907 * own the parent directory, or the destination of the rename,
908 * otherwise the destination may not be changed (except by
909 * root). This implements append-only directories.
911 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
912 tcnp->cn_cred->cr_uid != dp->i_uid &&
913 xp->i_uid != tcnp->cn_cred->cr_uid) {
918 * Target must be empty if a directory and have no links
919 * to it. Also, ensure source and target are compatible
920 * (both directories, or both not directories).
922 if ((xp->i_mode&IFMT) == IFDIR) {
923 if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
928 if (!doingdirectory) {
933 } else if (doingdirectory) {
937 error = ext2_dirrewrite(dp, ip, tcnp);
941 * If the target directory is in the same
942 * directory as the source directory,
943 * decrement the link count on the parent
944 * of the target directory.
946 if (doingdirectory && !newparent) {
948 dp->i_flag |= IN_CHANGE;
952 * Adjust the link count of the target to
953 * reflect the dirrewrite above. If this is
954 * a directory it is empty and there are
955 * no links to it, so we can squash the inode and
956 * any space associated with it. We disallowed
957 * renaming over top of a directory with links to
958 * it above, as the remaining link would point to
959 * a directory without "." or ".." entries.
962 if (doingdirectory) {
963 if (--xp->i_nlink != 0)
964 panic("ext2_rename: linked directory");
965 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
966 tcnp->cn_cred, tcnp->cn_thread);
968 xp->i_flag |= IN_CHANGE;
974 * 3) Unlink the source.
976 fcnp->cn_flags &= ~MODMASK;
977 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
979 error = relookup(fdvp, &fvp, fcnp);
987 * From name has disappeared.
990 panic("ext2_rename: lost dir entry");
995 * Ensure that the directory entry still exists and has not
996 * changed while the new name has been entered. If the source is
997 * a file then the entry may have been unlinked or renamed. In
998 * either case there is no further work to be done. If the source
999 * is a directory then it cannot have been rmdir'ed; its link
1000 * count of three would cause a rmdir to fail with ENOTEMPTY.
1001 * The IN_RENAME flag ensures that it cannot be moved by another
1006 panic("ext2_rename: lost dir entry");
1009 * If the source is a directory with a
1010 * new parent, the link count of the old
1011 * parent directory must be decremented
1012 * and ".." set to point to the new parent.
1014 if (doingdirectory && newparent) {
1016 dp->i_flag |= IN_CHANGE;
1017 error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1018 sizeof(struct dirtemplate), (off_t)0,
1019 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1020 tcnp->cn_cred, NOCRED, NULL, NULL);
1022 /* Like ufs little-endian: */
1023 namlen = dirbuf.dotdot_type;
1025 dirbuf.dotdot_name[0] != '.' ||
1026 dirbuf.dotdot_name[1] != '.') {
1027 ext2_dirbad(xp, (doff_t)12,
1028 "rename: mangled dir");
1030 dirbuf.dotdot_ino = newparent;
1031 (void) vn_rdwr(UIO_WRITE, fvp,
1033 sizeof(struct dirtemplate),
1034 (off_t)0, UIO_SYSSPACE,
1035 IO_NODELOCKED | IO_SYNC |
1036 IO_NOMACCHECK, tcnp->cn_cred,
1037 NOCRED, NULL, NULL);
1042 error = ext2_dirremove(fdvp, fcnp);
1045 xp->i_flag |= IN_CHANGE;
1047 xp->i_flag &= ~IN_RENAME;
1062 ip->i_flag &= ~IN_RENAME;
1063 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1065 ip->i_flag |= IN_CHANGE;
1066 ip->i_flag &= ~IN_RENAME;
1077 ext2_mkdir(struct vop_mkdir_args *ap)
1079 struct vnode *dvp = ap->a_dvp;
1080 struct vattr *vap = ap->a_vap;
1081 struct componentname *cnp = ap->a_cnp;
1082 struct inode *ip, *dp;
1084 struct dirtemplate dirtemplate, *dtp;
1088 if ((cnp->cn_flags & HASBUF) == 0)
1089 panic("ext2_mkdir: no name");
1092 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1096 dmode = vap->va_mode & 0777;
1099 * Must simulate part of ext2_makeinode here to acquire the inode,
1100 * but not have it entered in the parent directory. The entry is
1101 * made later after writing "." and ".." entries.
1103 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1107 ip->i_gid = dp->i_gid;
1111 * if we are hacking owners here, (only do this where told to)
1112 * and we are not giving it TOO root, (would subvert quotas)
1113 * then go ahead and give it to the other user.
1114 * The new directory also inherits the SUID bit.
1115 * If user's UID and dir UID are the same,
1116 * 'give it away' so that the SUID is still forced on.
1118 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1119 (dp->i_mode & ISUID) && dp->i_uid) {
1121 ip->i_uid = dp->i_uid;
1123 ip->i_uid = cnp->cn_cred->cr_uid;
1127 ip->i_uid = cnp->cn_cred->cr_uid;
1129 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1131 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1133 if (cnp->cn_flags & ISWHITEOUT)
1134 ip->i_flags |= UF_OPAQUE;
1135 error = ext2_update(tvp, 1);
1138 * Bump link count in parent directory
1139 * to reflect work done below. Should
1140 * be done before reference is created
1141 * so reparation is possible if we crash.
1144 dp->i_flag |= IN_CHANGE;
1145 error = ext2_update(dvp, !DOINGASYNC(dvp));
1149 /* Initialize directory with "." and ".." from static template. */
1150 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1151 EXT2F_INCOMPAT_FTYPE))
1152 dtp = &mastertemplate;
1154 dtp = &omastertemplate;
1156 dirtemplate.dot_ino = ip->i_number;
1157 dirtemplate.dotdot_ino = dp->i_number;
1158 /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1159 * so let's just redefine it - for this function only
1162 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize
1163 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1164 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1165 sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1166 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1170 dp->i_flag |= IN_CHANGE;
1173 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1174 /* XXX should grow with balloc() */
1175 panic("ext2_mkdir: blksize");
1177 ip->i_size = DIRBLKSIZ;
1178 ip->i_flag |= IN_CHANGE;
1181 /* Directory set up, now install its entry in the parent directory. */
1182 error = ext2_direnter(ip, dvp, cnp);
1185 dp->i_flag |= IN_CHANGE;
1189 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1190 * for us because we set the link count to 0.
1194 ip->i_flag |= IN_CHANGE;
1201 #define DIRBLKSIZ DEV_BSIZE
1205 * Rmdir system call.
1208 ext2_rmdir(struct vop_rmdir_args *ap)
1210 struct vnode *vp = ap->a_vp;
1211 struct vnode *dvp = ap->a_dvp;
1212 struct componentname *cnp = ap->a_cnp;
1213 struct inode *ip, *dp;
1220 * Verify the directory is empty (and valid).
1221 * (Rmdir ".." won't be valid since
1222 * ".." will contain a reference to
1223 * the current directory and thus be
1226 if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1230 if ((dp->i_flags & APPEND)
1231 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1236 * Delete reference to directory before purging
1237 * inode. If we crash in between, the directory
1238 * will be reattached to lost+found,
1240 error = ext2_dirremove(dvp, cnp);
1244 dp->i_flag |= IN_CHANGE;
1248 * Truncate inode. The only stuff left
1249 * in the directory is "." and "..". The
1250 * "." reference is inconsequential since
1251 * we're quashing it. The ".." reference
1252 * has already been adjusted above. We've
1253 * removed the "." reference and the reference
1254 * in the parent directory, but there may be
1255 * other hard links so decrement by 2 and
1256 * worry about them later.
1259 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1261 cache_purge(ITOV(ip));
1262 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1264 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1265 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1272 * symlink -- make a symbolic link
1275 ext2_symlink(struct vop_symlink_args *ap)
1277 struct vnode *vp, **vpp = ap->a_vpp;
1281 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1286 len = strlen(ap->a_target);
1287 if (len < vp->v_mount->mnt_maxsymlinklen) {
1289 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1291 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1293 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1294 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1295 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1302 * Return target name of a symbolic link
1305 ext2_readlink(struct vop_readlink_args *ap)
1307 struct vnode *vp = ap->a_vp;
1308 struct inode *ip = VTOI(vp);
1312 if (isize < vp->v_mount->mnt_maxsymlinklen) {
1313 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1316 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1320 * Calculate the logical to physical mapping if not done already,
1321 * then call the device strategy routine.
1323 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1324 * deadlock on memory. See ext2_bmap() for details.
1327 ext2_strategy(struct vop_strategy_args *ap)
1329 struct buf *bp = ap->a_bp;
1330 struct vnode *vp = ap->a_vp;
1335 if (vp->v_type == VBLK || vp->v_type == VCHR)
1336 panic("ext2_strategy: spec");
1337 if (bp->b_blkno == bp->b_lblkno) {
1338 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1339 bp->b_blkno = blkno;
1341 bp->b_error = error;
1342 bp->b_ioflags |= BIO_ERROR;
1346 if ((long)bp->b_blkno == -1)
1349 if ((long)bp->b_blkno == -1) {
1353 bp->b_iooffset = dbtob(bp->b_blkno);
1354 bo = VFSTOEXT2(vp->v_mount)->um_bo;
1355 BO_STRATEGY(bo, bp);
1360 * Print out the contents of an inode.
1363 ext2_print(struct vop_print_args *ap)
1365 struct vnode *vp = ap->a_vp;
1366 struct inode *ip = VTOI(vp);
1368 vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1369 if (vp->v_type == VFIFO)
1376 * Close wrapper for fifos.
1378 * Update the times on the inode then do device close.
1381 ext2fifo_close(struct vop_close_args *ap)
1383 struct vnode *vp = ap->a_vp;
1386 if (vp->v_usecount > 1)
1387 ext2_itimes_locked(vp);
1389 return (fifo_specops.vop_close(ap));
1393 * Kqfilter wrapper for fifos.
1395 * Fall through to ext2 kqfilter routines if needed
1398 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1402 error = fifo_specops.vop_kqfilter(ap);
1404 error = vfs_kqfilter(ap);
1409 * Return POSIX pathconf information applicable to ext2 filesystems.
1412 ext2_pathconf(struct vop_pathconf_args *ap)
1416 switch (ap->a_name) {
1418 *ap->a_retval = EXT2_LINK_MAX;
1421 *ap->a_retval = NAME_MAX;
1424 *ap->a_retval = PATH_MAX;
1427 *ap->a_retval = PIPE_BUF;
1429 case _PC_CHOWN_RESTRICTED:
1435 case _PC_MIN_HOLE_SIZE:
1436 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1439 /* _PC_ASYNC_IO should have been handled by upper layers. */
1440 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1449 case _PC_ALLOC_SIZE_MIN:
1450 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1452 case _PC_FILESIZEBITS:
1455 case _PC_REC_INCR_XFER_SIZE:
1456 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1458 case _PC_REC_MAX_XFER_SIZE:
1459 *ap->a_retval = -1; /* means ``unlimited'' */
1461 case _PC_REC_MIN_XFER_SIZE:
1462 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1464 case _PC_REC_XFER_ALIGN:
1465 *ap->a_retval = PAGE_SIZE;
1467 case _PC_SYMLINK_MAX:
1468 *ap->a_retval = MAXPATHLEN;
1479 * Vnode pointer to File handle
1483 ext2_vptofh(struct vop_vptofh_args *ap)
1488 ip = VTOI(ap->a_vp);
1489 ufhp = (struct ufid *)ap->a_fhp;
1490 ufhp->ufid_len = sizeof(struct ufid);
1491 ufhp->ufid_ino = ip->i_number;
1492 ufhp->ufid_gen = ip->i_gen;
1497 * Initialize the vnode associated with a new inode, handle aliased
1501 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1508 vp->v_type = IFTOVT(ip->i_mode);
1509 if (vp->v_type == VFIFO)
1512 if (ip->i_number == EXT2_ROOTINO)
1513 vp->v_vflag |= VV_ROOT;
1514 ip->i_modrev = init_va_filerev();
1520 * Allocate a new inode.
1523 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1524 struct componentname *cnp)
1526 struct inode *ip, *pdir;
1532 if ((cnp->cn_flags & HASBUF) == 0)
1533 panic("ext2_makeinode: no name");
1536 if ((mode & IFMT) == 0)
1539 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1544 ip->i_gid = pdir->i_gid;
1549 * not the owner of the directory,
1550 * and we are hacking owners here, (only do this where told to)
1551 * and we are not giving it TOO root, (would subvert quotas)
1552 * then go ahead and give it to the other user.
1553 * Note that this drops off the execute bits for security.
1555 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1556 (pdir->i_mode & ISUID) &&
1557 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1558 ip->i_uid = pdir->i_uid;
1561 ip->i_uid = cnp->cn_cred->cr_uid;
1565 ip->i_uid = cnp->cn_cred->cr_uid;
1567 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1569 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
1571 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1572 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1573 ip->i_mode &= ~ISGID;
1576 if (cnp->cn_flags & ISWHITEOUT)
1577 ip->i_flags |= UF_OPAQUE;
1580 * Make sure inode goes to disk before directory entry.
1582 error = ext2_update(tvp, !DOINGASYNC(tvp));
1585 error = ext2_direnter(ip, dvp, cnp);
1594 * Write error occurred trying to update the inode
1595 * or the directory so must deallocate the inode.
1598 ip->i_flag |= IN_CHANGE;
1604 * Vnode op for reading.
1607 ext2_read(struct vop_read_args *ap)
1616 /*EXT4_EXT_LOCK(ip);*/
1617 if (ip->i_flag & IN_E4EXTENTS)
1618 error = ext4_ext_read(ap);
1620 error = ext2_ind_read(ap);
1621 /*EXT4_EXT_UNLOCK(ip);*/
1626 * Vnode op for reading.
1629 ext2_ind_read(struct vop_read_args *ap)
1634 struct m_ext2fs *fs;
1636 daddr_t lbn, nextlbn;
1638 long size, xfersize, blkoffset;
1639 int error, orig_resid, seqcount;
1644 ioflag = ap->a_ioflag;
1646 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1650 if (uio->uio_rw != UIO_READ)
1651 panic("%s: mode", "ext2_read");
1653 if (vp->v_type == VLNK) {
1654 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1655 panic("%s: short symlink", "ext2_read");
1656 } else if (vp->v_type != VREG && vp->v_type != VDIR)
1657 panic("%s: type %d", "ext2_read", vp->v_type);
1659 orig_resid = uio->uio_resid;
1660 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1661 if (orig_resid == 0)
1663 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1665 if (uio->uio_offset < ip->i_size &&
1666 uio->uio_offset >= fs->e2fs_maxfilesize)
1669 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1670 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1672 lbn = lblkno(fs, uio->uio_offset);
1674 size = blksize(fs, ip, lbn);
1675 blkoffset = blkoff(fs, uio->uio_offset);
1677 xfersize = fs->e2fs_fsize - blkoffset;
1678 if (uio->uio_resid < xfersize)
1679 xfersize = uio->uio_resid;
1680 if (bytesinfile < xfersize)
1681 xfersize = bytesinfile;
1683 if (lblktosize(fs, nextlbn) >= ip->i_size)
1684 error = bread(vp, lbn, size, NOCRED, &bp);
1685 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1686 error = cluster_read(vp, ip->i_size, lbn, size,
1687 NOCRED, blkoffset + uio->uio_resid, seqcount,
1689 } else if (seqcount > 1) {
1690 u_int nextsize = blksize(fs, ip, nextlbn);
1691 error = breadn(vp, lbn,
1692 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1694 error = bread(vp, lbn, size, NOCRED, &bp);
1702 * If IO_DIRECT then set B_DIRECT for the buffer. This
1703 * will cause us to attempt to release the buffer later on
1704 * and will cause the buffer cache to attempt to free the
1707 if (ioflag & IO_DIRECT)
1708 bp->b_flags |= B_DIRECT;
1711 * We should only get non-zero b_resid when an I/O error
1712 * has occurred, which should cause us to break above.
1713 * However, if the short read did not cause an error,
1714 * then we want to ensure that we do not uiomove bad
1715 * or uninitialized data.
1717 size -= bp->b_resid;
1718 if (size < xfersize) {
1723 error = uiomove((char *)bp->b_data + blkoffset,
1724 (int)xfersize, uio);
1728 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1730 * If it's VMIO or direct I/O, then we don't
1731 * need the buf, mark it available for
1732 * freeing. If it's non-direct VMIO, the VM has
1735 bp->b_flags |= B_RELBUF;
1739 * Otherwise let whoever
1740 * made the request take care of
1741 * freeing it. We just queue
1742 * it onto another list.
1749 * This can only happen in the case of an error
1750 * because the loop above resets bp to NULL on each iteration
1751 * and on normal completion has not set a new value into it.
1752 * so it must have come from a 'break' statement
1755 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1756 bp->b_flags |= B_RELBUF;
1763 if ((error == 0 || uio->uio_resid != orig_resid) &&
1764 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
1765 ip->i_flag |= IN_ACCESS;
1770 ext2_ioctl(struct vop_ioctl_args *ap)
1773 switch (ap->a_command) {
1776 return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1777 (off_t *)ap->a_data, ap->a_cred));
1784 * this function handles ext4 extents block mapping
1787 ext4_ext_read(struct vop_read_args *ap)
1792 struct m_ext2fs *fs;
1794 struct ext4_extent nex, *ep;
1795 struct ext4_extent_path path;
1796 daddr_t lbn, newblk;
1801 long size, xfersize, blkoffset;
1806 memset(&path, 0, sizeof(path));
1808 orig_resid = uio->uio_resid;
1809 KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__));
1810 if (orig_resid == 0)
1812 KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__));
1814 if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
1817 while (uio->uio_resid > 0) {
1818 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1820 lbn = lblkno(fs, uio->uio_offset);
1821 size = blksize(fs, ip, lbn);
1822 blkoffset = blkoff(fs, uio->uio_offset);
1824 xfersize = fs->e2fs_fsize - blkoffset;
1825 xfersize = MIN(xfersize, uio->uio_resid);
1826 xfersize = MIN(xfersize, bytesinfile);
1828 /* get block from ext4 extent cache */
1829 cache_type = ext4_ext_in_cache(ip, lbn, &nex);
1830 switch (cache_type) {
1831 case EXT4_EXT_CACHE_NO:
1832 ext4_ext_find_extent(fs, ip, lbn, &path);
1837 ext4_ext_put_cache(ip, ep, EXT4_EXT_CACHE_IN);
1839 newblk = lbn - ep->e_blk + (ep->e_start_lo |
1840 (daddr_t)ep->e_start_hi << 32);
1842 if (path.ep_bp != NULL) {
1848 case EXT4_EXT_CACHE_GAP:
1849 /* block has not been allocated yet */
1852 case EXT4_EXT_CACHE_IN:
1853 newblk = lbn - nex.e_blk + (nex.e_start_lo |
1854 (daddr_t)nex.e_start_hi << 32);
1858 panic("%s: invalid cache type", __func__);
1861 error = bread(ip->i_devvp, fsbtodb(fs, newblk), size, NOCRED, &bp);
1867 size -= bp->b_resid;
1868 if (size < xfersize) {
1875 error = uiomove(bp->b_data + blkoffset, (int)xfersize, uio);
1885 * Vnode op for writing.
1888 ext2_write(struct vop_write_args *ap)
1893 struct m_ext2fs *fs;
1897 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1899 ioflag = ap->a_ioflag;
1903 seqcount = ioflag >> IO_SEQSHIFT;
1907 if (uio->uio_rw != UIO_WRITE)
1908 panic("%s: mode", "ext2_write");
1911 switch (vp->v_type) {
1913 if (ioflag & IO_APPEND)
1914 uio->uio_offset = ip->i_size;
1915 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1921 /* XXX differs from ffs -- this is called from ext2_mkdir(). */
1922 if ((ioflag & IO_SYNC) == 0)
1923 panic("ext2_write: nonsync dir write");
1926 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
1927 vp->v_type, (intmax_t)uio->uio_offset,
1928 (intmax_t)uio->uio_resid);
1931 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
1932 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
1934 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
1937 * Maybe this should be above the vnode op call, but so long as
1938 * file servers have no limits, I don't think it matters.
1940 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
1943 resid = uio->uio_resid;
1945 if (seqcount > BA_SEQMAX)
1946 flags = BA_SEQMAX << BA_SEQSHIFT;
1948 flags = seqcount << BA_SEQSHIFT;
1949 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1952 for (error = 0; uio->uio_resid > 0;) {
1953 lbn = lblkno(fs, uio->uio_offset);
1954 blkoffset = blkoff(fs, uio->uio_offset);
1955 xfersize = fs->e2fs_fsize - blkoffset;
1956 if (uio->uio_resid < xfersize)
1957 xfersize = uio->uio_resid;
1958 if (uio->uio_offset + xfersize > ip->i_size)
1959 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
1962 * We must perform a read-before-write if the transfer size
1963 * does not cover the entire buffer.
1965 if (fs->e2fs_bsize > xfersize)
1968 flags &= ~BA_CLRBUF;
1969 error = ext2_balloc(ip, lbn, blkoffset + xfersize,
1970 ap->a_cred, &bp, flags);
1974 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
1975 bp->b_flags |= B_NOCACHE;
1976 if (uio->uio_offset + xfersize > ip->i_size)
1977 ip->i_size = uio->uio_offset + xfersize;
1978 size = blksize(fs, ip, lbn) - bp->b_resid;
1979 if (size < xfersize)
1983 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1985 * If the buffer is not already filled and we encounter an
1986 * error while trying to fill it, we have to clear out any
1987 * garbage data from the pages instantiated for the buffer.
1988 * If we do not, a failed uiomove() during a write can leave
1989 * the prior contents of the pages exposed to a userland mmap.
1991 * Note that we need only clear buffers with a transfer size
1992 * equal to the block size because buffers with a shorter
1993 * transfer size were cleared above by the call to ext2_balloc()
1994 * with the BA_CLRBUF flag set.
1996 * If the source region for uiomove identically mmaps the
1997 * buffer, uiomove() performed the NOP copy, and the buffer
1998 * content remains valid because the page fault handler
1999 * validated the pages.
2001 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2002 fs->e2fs_bsize == xfersize)
2004 if (ioflag & (IO_VMIO|IO_DIRECT)) {
2005 bp->b_flags |= B_RELBUF;
2009 * If IO_SYNC each buffer is written synchronously. Otherwise
2010 * if we have a severe page deficiency write the buffer
2011 * asynchronously. Otherwise try to cluster, and if that
2012 * doesn't do it then either do an async write (if O_DIRECT),
2013 * or a delayed write (if not).
2015 if (ioflag & IO_SYNC) {
2017 } else if (vm_page_count_severe() ||
2018 buf_dirty_count_severe() ||
2019 (ioflag & IO_ASYNC)) {
2020 bp->b_flags |= B_CLUSTEROK;
2022 } else if (xfersize + blkoffset == fs->e2fs_fsize) {
2023 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2024 bp->b_flags |= B_CLUSTEROK;
2025 cluster_write(vp, bp, ip->i_size, seqcount, 0);
2029 } else if (ioflag & IO_DIRECT) {
2030 bp->b_flags |= B_CLUSTEROK;
2033 bp->b_flags |= B_CLUSTEROK;
2036 if (error || xfersize == 0)
2040 * If we successfully wrote any data, and we are not the superuser
2041 * we clear the setuid and setgid bits as a precaution against
2044 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2046 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2047 ip->i_mode &= ~(ISUID | ISGID);
2050 if (ioflag & IO_UNIT) {
2051 (void)ext2_truncate(vp, osize,
2052 ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2053 uio->uio_offset -= resid - uio->uio_resid;
2054 uio->uio_resid = resid;
2057 if (uio->uio_resid != resid) {
2058 ip->i_flag |= IN_CHANGE | IN_UPDATE;
2059 if (ioflag & IO_SYNC)
2060 error = ext2_update(vp, 1);