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_getpages_t ext2_getpages;
101 static vop_getattr_t ext2_getattr;
102 static vop_ioctl_t ext2_ioctl;
103 static vop_link_t ext2_link;
104 static vop_mkdir_t ext2_mkdir;
105 static vop_mknod_t ext2_mknod;
106 static vop_open_t ext2_open;
107 static vop_pathconf_t ext2_pathconf;
108 static vop_print_t ext2_print;
109 static vop_read_t ext2_read;
110 static vop_readlink_t ext2_readlink;
111 static vop_remove_t ext2_remove;
112 static vop_rename_t ext2_rename;
113 static vop_rmdir_t ext2_rmdir;
114 static vop_setattr_t ext2_setattr;
115 static vop_strategy_t ext2_strategy;
116 static vop_symlink_t ext2_symlink;
117 static vop_write_t ext2_write;
118 static vop_vptofh_t ext2_vptofh;
119 static vop_close_t ext2fifo_close;
120 static vop_kqfilter_t ext2fifo_kqfilter;
122 /* Global vfs data structures for ext2. */
123 struct vop_vector ext2_vnodeops = {
124 .vop_default = &default_vnodeops,
125 .vop_access = ext2_access,
126 .vop_bmap = ext2_bmap,
127 .vop_cachedlookup = ext2_lookup,
128 .vop_close = ext2_close,
129 .vop_create = ext2_create,
130 .vop_fsync = ext2_fsync,
131 .vop_getpages = ext2_getpages,
132 .vop_getattr = ext2_getattr,
133 .vop_inactive = ext2_inactive,
134 .vop_ioctl = ext2_ioctl,
135 .vop_link = ext2_link,
136 .vop_lookup = vfs_cache_lookup,
137 .vop_mkdir = ext2_mkdir,
138 .vop_mknod = ext2_mknod,
139 .vop_open = ext2_open,
140 .vop_pathconf = ext2_pathconf,
141 .vop_poll = vop_stdpoll,
142 .vop_print = ext2_print,
143 .vop_read = ext2_read,
144 .vop_readdir = ext2_readdir,
145 .vop_readlink = ext2_readlink,
146 .vop_reallocblks = ext2_reallocblks,
147 .vop_reclaim = ext2_reclaim,
148 .vop_remove = ext2_remove,
149 .vop_rename = ext2_rename,
150 .vop_rmdir = ext2_rmdir,
151 .vop_setattr = ext2_setattr,
152 .vop_strategy = ext2_strategy,
153 .vop_symlink = ext2_symlink,
154 .vop_write = ext2_write,
155 .vop_vptofh = ext2_vptofh,
158 struct vop_vector ext2_fifoops = {
159 .vop_default = &fifo_specops,
160 .vop_access = ext2_access,
161 .vop_close = ext2fifo_close,
162 .vop_fsync = ext2_fsync,
163 .vop_getattr = ext2_getattr,
164 .vop_inactive = ext2_inactive,
165 .vop_kqfilter = ext2fifo_kqfilter,
166 .vop_print = ext2_print,
167 .vop_read = VOP_PANIC,
168 .vop_reclaim = ext2_reclaim,
169 .vop_setattr = ext2_setattr,
170 .vop_write = VOP_PANIC,
171 .vop_vptofh = ext2_vptofh,
175 * A virgin directory (no blushing please).
176 * Note that the type and namlen fields are reversed relative to ext2.
177 * Also, we don't use `struct odirtemplate', since it would just cause
178 * endianness problems.
180 static struct dirtemplate mastertemplate = {
181 0, 12, 1, EXT2_FT_DIR, ".",
182 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
184 static struct dirtemplate omastertemplate = {
185 0, 12, 1, EXT2_FT_UNKNOWN, ".",
186 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
190 ext2_itimes_locked(struct vnode *vp)
195 ASSERT_VI_LOCKED(vp, __func__);
198 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
200 if ((vp->v_type == VBLK || vp->v_type == VCHR))
201 ip->i_flag |= IN_LAZYMOD;
203 ip->i_flag |= IN_MODIFIED;
204 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
206 if (ip->i_flag & IN_ACCESS) {
207 ip->i_atime = ts.tv_sec;
208 ip->i_atimensec = ts.tv_nsec;
210 if (ip->i_flag & IN_UPDATE) {
211 ip->i_mtime = ts.tv_sec;
212 ip->i_mtimensec = ts.tv_nsec;
215 if (ip->i_flag & IN_CHANGE) {
216 ip->i_ctime = ts.tv_sec;
217 ip->i_ctimensec = ts.tv_nsec;
220 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
224 ext2_itimes(struct vnode *vp)
228 ext2_itimes_locked(vp);
233 * Create a regular file
236 ext2_create(struct vop_create_args *ap)
241 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
242 ap->a_dvp, ap->a_vpp, ap->a_cnp);
245 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
246 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
251 ext2_open(struct vop_open_args *ap)
254 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
258 * Files marked append-only must be opened for appending.
260 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
261 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
264 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
272 * Update the times on the inode.
275 ext2_close(struct vop_close_args *ap)
277 struct vnode *vp = ap->a_vp;
280 if (vp->v_usecount > 1)
281 ext2_itimes_locked(vp);
287 ext2_access(struct vop_access_args *ap)
289 struct vnode *vp = ap->a_vp;
290 struct inode *ip = VTOI(vp);
291 accmode_t accmode = ap->a_accmode;
294 if (vp->v_type == VBLK || vp->v_type == VCHR)
298 * Disallow write attempts on read-only file systems;
299 * unless the file is a socket, fifo, or a block or
300 * character device resident on the file system.
302 if (accmode & VWRITE) {
303 switch (vp->v_type) {
307 if (vp->v_mount->mnt_flag & MNT_RDONLY)
315 /* If immutable bit set, nobody gets to write it. */
316 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
319 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
320 ap->a_accmode, ap->a_cred, NULL);
325 ext2_getattr(struct vop_getattr_args *ap)
327 struct vnode *vp = ap->a_vp;
328 struct inode *ip = VTOI(vp);
329 struct vattr *vap = ap->a_vap;
333 * Copy from inode table
335 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
336 vap->va_fileid = ip->i_number;
337 vap->va_mode = ip->i_mode & ~IFMT;
338 vap->va_nlink = ip->i_nlink;
339 vap->va_uid = ip->i_uid;
340 vap->va_gid = ip->i_gid;
341 vap->va_rdev = ip->i_rdev;
342 vap->va_size = ip->i_size;
343 vap->va_atime.tv_sec = ip->i_atime;
344 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
345 vap->va_mtime.tv_sec = ip->i_mtime;
346 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
347 vap->va_ctime.tv_sec = ip->i_ctime;
348 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
349 if E2DI_HAS_XTIME(ip) {
350 vap->va_birthtime.tv_sec = ip->i_birthtime;
351 vap->va_birthtime.tv_nsec = ip->i_birthnsec;
353 vap->va_flags = ip->i_flags;
354 vap->va_gen = ip->i_gen;
355 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
356 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
357 vap->va_type = IFTOVT(ip->i_mode);
358 vap->va_filerev = ip->i_modrev;
363 * Set attribute vnode op. called from several syscalls
366 ext2_setattr(struct vop_setattr_args *ap)
368 struct vattr *vap = ap->a_vap;
369 struct vnode *vp = ap->a_vp;
370 struct inode *ip = VTOI(vp);
371 struct ucred *cred = ap->a_cred;
372 struct thread *td = curthread;
376 * Check for unsettable attributes.
378 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
379 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
380 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
381 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
384 if (vap->va_flags != VNOVAL) {
385 /* Disallow flags not supported by ext2fs. */
386 if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
389 if (vp->v_mount->mnt_flag & MNT_RDONLY)
392 * Callers may only modify the file flags on objects they
393 * have VADMIN rights for.
395 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
398 * Unprivileged processes and privileged processes in
399 * jail() are not permitted to unset system flags, or
400 * modify flags if any system flags are set.
401 * Privileged non-jail processes may not modify system flags
402 * if securelevel > 0 and any existing system flags are set.
404 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
405 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
406 error = securelevel_gt(cred, 0);
411 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
412 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
415 ip->i_flags = vap->va_flags;
416 ip->i_flag |= IN_CHANGE;
417 if (ip->i_flags & (IMMUTABLE | APPEND))
420 if (ip->i_flags & (IMMUTABLE | APPEND))
423 * Go through the fields and update iff not VNOVAL.
425 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
426 if (vp->v_mount->mnt_flag & MNT_RDONLY)
428 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
432 if (vap->va_size != VNOVAL) {
434 * Disallow write attempts on read-only file systems;
435 * unless the file is a socket, fifo, or a block or
436 * character device resident on the file system.
438 switch (vp->v_type) {
443 if (vp->v_mount->mnt_flag & MNT_RDONLY)
449 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
452 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
453 if (vp->v_mount->mnt_flag & MNT_RDONLY)
457 * If times is NULL, ... The caller must be the owner of
458 * the file, have permission to write the file, or be the
460 * If times is non-NULL, ... The caller must be the owner of
461 * the file or be the super-user.
463 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
464 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
465 (error = VOP_ACCESS(vp, VWRITE, cred, td))))
467 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
468 if (vap->va_atime.tv_sec != VNOVAL) {
469 ip->i_flag &= ~IN_ACCESS;
470 ip->i_atime = vap->va_atime.tv_sec;
471 ip->i_atimensec = vap->va_atime.tv_nsec;
473 if (vap->va_mtime.tv_sec != VNOVAL) {
474 ip->i_flag &= ~IN_UPDATE;
475 ip->i_mtime = vap->va_mtime.tv_sec;
476 ip->i_mtimensec = vap->va_mtime.tv_nsec;
478 ip->i_birthtime = vap->va_birthtime.tv_sec;
479 ip->i_birthnsec = vap->va_birthtime.tv_nsec;
480 error = ext2_update(vp, 0);
485 if (vap->va_mode != (mode_t)VNOVAL) {
486 if (vp->v_mount->mnt_flag & MNT_RDONLY)
488 error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
494 * Change the mode on a file.
495 * Inode must be locked before calling.
498 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
500 struct inode *ip = VTOI(vp);
504 * To modify the permissions on a file, must possess VADMIN
507 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
510 * Privileged processes may set the sticky bit on non-directories,
511 * as well as set the setgid bit on a file with a group that the
512 * process is not a member of.
514 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
515 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
519 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
520 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
524 ip->i_mode &= ~ALLPERMS;
525 ip->i_mode |= (mode & ALLPERMS);
526 ip->i_flag |= IN_CHANGE;
531 * Perform chown operation on inode ip;
532 * inode must be locked prior to call.
535 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
538 struct inode *ip = VTOI(vp);
543 if (uid == (uid_t)VNOVAL)
545 if (gid == (gid_t)VNOVAL)
548 * To modify the ownership of a file, must possess VADMIN
551 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
554 * To change the owner of a file, or change the group of a file
555 * to a group of which we are not a member, the caller must
558 if (uid != ip->i_uid || (gid != ip->i_gid &&
559 !groupmember(gid, cred))) {
560 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
568 ip->i_flag |= IN_CHANGE;
569 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
570 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
571 ip->i_mode &= ~(ISUID | ISGID);
577 * Synch an open file.
581 ext2_fsync(struct vop_fsync_args *ap)
584 * Flush all dirty buffers associated with a vnode.
589 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
597 ext2_mknod(struct vop_mknod_args *ap)
599 struct vattr *vap = ap->a_vap;
600 struct vnode **vpp = ap->a_vpp;
605 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
606 ap->a_dvp, vpp, ap->a_cnp);
610 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
611 if (vap->va_rdev != VNOVAL) {
613 * Want to be able to use this to make badblock
614 * inodes, so don't truncate the dev number.
616 ip->i_rdev = vap->va_rdev;
619 * Remove inode, then reload it through VFS_VGET so it is
620 * checked to see if it is an alias of an existing entry in
621 * the inode cache. XXX I don't believe this is necessary now.
623 (*vpp)->v_type = VNON;
624 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
627 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
636 ext2_remove(struct vop_remove_args *ap)
639 struct vnode *vp = ap->a_vp;
640 struct vnode *dvp = ap->a_dvp;
644 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
645 (VTOI(dvp)->i_flags & APPEND)) {
649 error = ext2_dirremove(dvp, ap->a_cnp);
652 ip->i_flag |= IN_CHANGE;
662 ext2_link(struct vop_link_args *ap)
664 struct vnode *vp = ap->a_vp;
665 struct vnode *tdvp = ap->a_tdvp;
666 struct componentname *cnp = ap->a_cnp;
671 if ((cnp->cn_flags & HASBUF) == 0)
672 panic("ext2_link: no name");
675 if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
679 if (ip->i_flags & (IMMUTABLE | APPEND)) {
684 ip->i_flag |= IN_CHANGE;
685 error = ext2_update(vp, !DOINGASYNC(vp));
687 error = ext2_direnter(ip, tdvp, cnp);
690 ip->i_flag |= IN_CHANGE;
697 * Rename system call.
698 * rename("foo", "bar");
701 * link("foo", "bar");
703 * but ``atomically''. Can't do full commit without saving state in the
704 * inode on disk which isn't feasible at this time. Best we can do is
705 * always guarantee the target exists.
707 * Basic algorithm is:
709 * 1) Bump link count on source while we're linking it to the
710 * target. This also ensure the inode won't be deleted out
711 * from underneath us while we work (it may be truncated by
712 * a concurrent `trunc' or `open' for creation).
713 * 2) Link source to destination. If destination already exists,
715 * 3) Unlink source reference to inode if still around. If a
716 * directory was moved and the parent of the destination
717 * is different from the source, patch the ".." entry in the
721 ext2_rename(struct vop_rename_args *ap)
723 struct vnode *tvp = ap->a_tvp;
724 struct vnode *tdvp = ap->a_tdvp;
725 struct vnode *fvp = ap->a_fvp;
726 struct vnode *fdvp = ap->a_fdvp;
727 struct componentname *tcnp = ap->a_tcnp;
728 struct componentname *fcnp = ap->a_fcnp;
729 struct inode *ip, *xp, *dp;
730 struct dirtemplate dirbuf;
731 int doingdirectory = 0, oldparent = 0, newparent = 0;
736 if ((tcnp->cn_flags & HASBUF) == 0 ||
737 (fcnp->cn_flags & HASBUF) == 0)
738 panic("ext2_rename: no name");
741 * Check for cross-device rename.
743 if ((fvp->v_mount != tdvp->v_mount) ||
744 (tvp && (fvp->v_mount != tvp->v_mount))) {
758 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
759 (VTOI(tdvp)->i_flags & APPEND))) {
765 * Renaming a file to itself has no effect. The upper layers should
766 * not call us in that case. Temporarily just warn if they do.
769 printf("ext2_rename: fvp == tvp (can't happen)\n");
774 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
778 if (ip->i_nlink >= EXT2_LINK_MAX) {
783 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
784 || (dp->i_flags & APPEND)) {
789 if ((ip->i_mode & IFMT) == IFDIR) {
791 * Avoid ".", "..", and aliases of "." for obvious reasons.
793 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
794 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
795 (ip->i_flag & IN_RENAME)) {
800 ip->i_flag |= IN_RENAME;
801 oldparent = dp->i_number;
807 * When the target exists, both the directory
808 * and target vnodes are returned locked.
816 * 1) Bump link count while we're moving stuff
817 * around. If we crash somewhere before
818 * completing our work, the link count
819 * may be wrong, but correctable.
822 ip->i_flag |= IN_CHANGE;
823 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
829 * If ".." must be changed (ie the directory gets a new
830 * parent) then the source directory must not be in the
831 * directory hierarchy above the target, as this would
832 * orphan everything below the source directory. Also
833 * the user must have write permission in the source so
834 * as to be able to change "..". We must repeat the call
835 * to namei, as the parent directory is unlocked by the
836 * call to checkpath().
838 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
840 if (oldparent != dp->i_number)
841 newparent = dp->i_number;
842 if (doingdirectory && newparent) {
843 if (error) /* write access check above */
847 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
851 error = relookup(tdvp, &tvp, tcnp);
861 * 2) If target doesn't exist, link the target
862 * to the source and unlink the source.
863 * Otherwise, rewrite the target directory
864 * entry to reference the source inode and
865 * expunge the original entry's existence.
868 if (dp->i_devvp != ip->i_devvp)
869 panic("ext2_rename: EXDEV");
871 * Account for ".." in new directory.
872 * When source and destination have the same
873 * parent we don't fool with the link count.
875 if (doingdirectory && newparent) {
876 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
881 dp->i_flag |= IN_CHANGE;
882 error = ext2_update(tdvp, !DOINGASYNC(tdvp));
886 error = ext2_direnter(ip, tdvp, tcnp);
888 if (doingdirectory && newparent) {
890 dp->i_flag |= IN_CHANGE;
891 (void)ext2_update(tdvp, 1);
897 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
898 panic("ext2_rename: EXDEV");
900 * Short circuit rename(foo, foo).
902 if (xp->i_number == ip->i_number)
903 panic("ext2_rename: same file");
905 * If the parent directory is "sticky", then the user must
906 * own the parent directory, or the destination of the rename,
907 * otherwise the destination may not be changed (except by
908 * root). This implements append-only directories.
910 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
911 tcnp->cn_cred->cr_uid != dp->i_uid &&
912 xp->i_uid != tcnp->cn_cred->cr_uid) {
917 * Target must be empty if a directory and have no links
918 * to it. Also, ensure source and target are compatible
919 * (both directories, or both not directories).
921 if ((xp->i_mode & IFMT) == IFDIR) {
922 if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
927 if (!doingdirectory) {
932 } else if (doingdirectory) {
936 error = ext2_dirrewrite(dp, ip, tcnp);
940 * If the target directory is in the same
941 * directory as the source directory,
942 * decrement the link count on the parent
943 * of the target directory.
945 if (doingdirectory && !newparent) {
947 dp->i_flag |= IN_CHANGE;
951 * Adjust the link count of the target to
952 * reflect the dirrewrite above. If this is
953 * a directory it is empty and there are
954 * no links to it, so we can squash the inode and
955 * any space associated with it. We disallowed
956 * renaming over top of a directory with links to
957 * it above, as the remaining link would point to
958 * a directory without "." or ".." entries.
961 if (doingdirectory) {
962 if (--xp->i_nlink != 0)
963 panic("ext2_rename: linked directory");
964 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
965 tcnp->cn_cred, tcnp->cn_thread);
967 xp->i_flag |= IN_CHANGE;
973 * 3) Unlink the source.
975 fcnp->cn_flags &= ~MODMASK;
976 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
978 error = relookup(fdvp, &fvp, fcnp);
986 * From name has disappeared. IN_RENAME is not sufficient
987 * to protect against directory races due to timing windows,
988 * so we can't panic here.
994 * Ensure that the directory entry still exists and has not
995 * changed while the new name has been entered. If the source is
996 * a file then the entry may have been unlinked or renamed. In
997 * either case there is no further work to be done. If the source
998 * is a directory then it cannot have been rmdir'ed; its link
999 * count of three would cause a rmdir to fail with ENOTEMPTY.
1000 * The IN_RENAME flag ensures that it cannot be moved by another
1005 * From name resolves to a different inode. IN_RENAME is
1006 * not sufficient protection against timing window races
1007 * so we can't panic here.
1011 * If the source is a directory with a
1012 * new parent, the link count of the old
1013 * parent directory must be decremented
1014 * and ".." set to point to the new parent.
1016 if (doingdirectory && newparent) {
1018 dp->i_flag |= IN_CHANGE;
1019 error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1020 sizeof(struct dirtemplate), (off_t)0,
1021 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1022 tcnp->cn_cred, NOCRED, NULL, NULL);
1024 /* Like ufs little-endian: */
1025 namlen = dirbuf.dotdot_type;
1027 dirbuf.dotdot_name[0] != '.' ||
1028 dirbuf.dotdot_name[1] != '.') {
1029 ext2_dirbad(xp, (doff_t)12,
1030 "rename: mangled dir");
1032 dirbuf.dotdot_ino = newparent;
1033 (void)vn_rdwr(UIO_WRITE, fvp,
1035 sizeof(struct dirtemplate),
1036 (off_t)0, UIO_SYSSPACE,
1037 IO_NODELOCKED | IO_SYNC |
1038 IO_NOMACCHECK, tcnp->cn_cred,
1039 NOCRED, NULL, NULL);
1044 error = ext2_dirremove(fdvp, fcnp);
1047 xp->i_flag |= IN_CHANGE;
1049 xp->i_flag &= ~IN_RENAME;
1064 ip->i_flag &= ~IN_RENAME;
1065 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1067 ip->i_flag |= IN_CHANGE;
1068 ip->i_flag &= ~IN_RENAME;
1079 ext2_mkdir(struct vop_mkdir_args *ap)
1081 struct vnode *dvp = ap->a_dvp;
1082 struct vattr *vap = ap->a_vap;
1083 struct componentname *cnp = ap->a_cnp;
1084 struct inode *ip, *dp;
1086 struct dirtemplate dirtemplate, *dtp;
1090 if ((cnp->cn_flags & HASBUF) == 0)
1091 panic("ext2_mkdir: no name");
1094 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1098 dmode = vap->va_mode & 0777;
1101 * Must simulate part of ext2_makeinode here to acquire the inode,
1102 * but not have it entered in the parent directory. The entry is
1103 * made later after writing "." and ".." entries.
1105 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1109 ip->i_gid = dp->i_gid;
1113 * if we are hacking owners here, (only do this where told to)
1114 * and we are not giving it TOO root, (would subvert quotas)
1115 * then go ahead and give it to the other user.
1116 * The new directory also inherits the SUID bit.
1117 * If user's UID and dir UID are the same,
1118 * 'give it away' so that the SUID is still forced on.
1120 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1121 (dp->i_mode & ISUID) && dp->i_uid) {
1123 ip->i_uid = dp->i_uid;
1125 ip->i_uid = cnp->cn_cred->cr_uid;
1129 ip->i_uid = cnp->cn_cred->cr_uid;
1131 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1133 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1135 if (cnp->cn_flags & ISWHITEOUT)
1136 ip->i_flags |= UF_OPAQUE;
1137 error = ext2_update(tvp, 1);
1140 * Bump link count in parent directory
1141 * to reflect work done below. Should
1142 * be done before reference is created
1143 * so reparation is possible if we crash.
1146 dp->i_flag |= IN_CHANGE;
1147 error = ext2_update(dvp, !DOINGASYNC(dvp));
1151 /* Initialize directory with "." and ".." from static template. */
1152 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1153 EXT2F_INCOMPAT_FTYPE))
1154 dtp = &mastertemplate;
1156 dtp = &omastertemplate;
1158 dirtemplate.dot_ino = ip->i_number;
1159 dirtemplate.dotdot_ino = dp->i_number;
1161 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1162 * just redefine it - for this function only
1165 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize
1166 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1167 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1168 sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1169 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1173 dp->i_flag |= IN_CHANGE;
1176 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1177 /* XXX should grow with balloc() */
1178 panic("ext2_mkdir: blksize");
1180 ip->i_size = DIRBLKSIZ;
1181 ip->i_flag |= IN_CHANGE;
1184 /* Directory set up, now install its entry in the parent directory. */
1185 error = ext2_direnter(ip, dvp, cnp);
1188 dp->i_flag |= IN_CHANGE;
1192 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1193 * for us because we set the link count to 0.
1197 ip->i_flag |= IN_CHANGE;
1204 #define DIRBLKSIZ DEV_BSIZE
1208 * Rmdir system call.
1211 ext2_rmdir(struct vop_rmdir_args *ap)
1213 struct vnode *vp = ap->a_vp;
1214 struct vnode *dvp = ap->a_dvp;
1215 struct componentname *cnp = ap->a_cnp;
1216 struct inode *ip, *dp;
1223 * Verify the directory is empty (and valid).
1224 * (Rmdir ".." won't be valid since
1225 * ".." will contain a reference to
1226 * the current directory and thus be
1229 if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1233 if ((dp->i_flags & APPEND)
1234 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1239 * Delete reference to directory before purging
1240 * inode. If we crash in between, the directory
1241 * will be reattached to lost+found,
1243 error = ext2_dirremove(dvp, cnp);
1247 dp->i_flag |= IN_CHANGE;
1251 * Truncate inode. The only stuff left
1252 * in the directory is "." and "..". The
1253 * "." reference is inconsequential since
1254 * we're quashing it. The ".." reference
1255 * has already been adjusted above. We've
1256 * removed the "." reference and the reference
1257 * in the parent directory, but there may be
1258 * other hard links so decrement by 2 and
1259 * worry about them later.
1262 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1264 cache_purge(ITOV(ip));
1265 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1267 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1268 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1275 * symlink -- make a symbolic link
1278 ext2_symlink(struct vop_symlink_args *ap)
1280 struct vnode *vp, **vpp = ap->a_vpp;
1284 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1289 len = strlen(ap->a_target);
1290 if (len < vp->v_mount->mnt_maxsymlinklen) {
1292 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1294 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1296 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1297 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1298 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1305 * Return target name of a symbolic link
1308 ext2_readlink(struct vop_readlink_args *ap)
1310 struct vnode *vp = ap->a_vp;
1311 struct inode *ip = VTOI(vp);
1315 if (isize < vp->v_mount->mnt_maxsymlinklen) {
1316 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1319 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1323 * Calculate the logical to physical mapping if not done already,
1324 * then call the device strategy routine.
1326 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1327 * deadlock on memory. See ext2_bmap() for details.
1330 ext2_strategy(struct vop_strategy_args *ap)
1332 struct buf *bp = ap->a_bp;
1333 struct vnode *vp = ap->a_vp;
1338 if (vp->v_type == VBLK || vp->v_type == VCHR)
1339 panic("ext2_strategy: spec");
1340 if (bp->b_blkno == bp->b_lblkno) {
1341 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1342 bp->b_blkno = blkno;
1344 bp->b_error = error;
1345 bp->b_ioflags |= BIO_ERROR;
1349 if ((long)bp->b_blkno == -1)
1352 if ((long)bp->b_blkno == -1) {
1356 bp->b_iooffset = dbtob(bp->b_blkno);
1357 bo = VFSTOEXT2(vp->v_mount)->um_bo;
1358 BO_STRATEGY(bo, bp);
1363 * Print out the contents of an inode.
1366 ext2_print(struct vop_print_args *ap)
1368 struct vnode *vp = ap->a_vp;
1369 struct inode *ip = VTOI(vp);
1371 vn_printf(ip->i_devvp, "\tino %lu", (u_long)ip->i_number);
1372 if (vp->v_type == VFIFO)
1379 * Close wrapper for fifos.
1381 * Update the times on the inode then do device close.
1384 ext2fifo_close(struct vop_close_args *ap)
1386 struct vnode *vp = ap->a_vp;
1389 if (vp->v_usecount > 1)
1390 ext2_itimes_locked(vp);
1392 return (fifo_specops.vop_close(ap));
1396 * Kqfilter wrapper for fifos.
1398 * Fall through to ext2 kqfilter routines if needed
1401 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1405 error = fifo_specops.vop_kqfilter(ap);
1407 error = vfs_kqfilter(ap);
1412 * Return POSIX pathconf information applicable to ext2 filesystems.
1415 ext2_pathconf(struct vop_pathconf_args *ap)
1419 switch (ap->a_name) {
1421 *ap->a_retval = EXT2_LINK_MAX;
1424 *ap->a_retval = NAME_MAX;
1427 *ap->a_retval = PATH_MAX;
1430 *ap->a_retval = PIPE_BUF;
1432 case _PC_CHOWN_RESTRICTED:
1438 case _PC_MIN_HOLE_SIZE:
1439 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1442 /* _PC_ASYNC_IO should have been handled by upper layers. */
1443 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1452 case _PC_ALLOC_SIZE_MIN:
1453 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1455 case _PC_FILESIZEBITS:
1458 case _PC_REC_INCR_XFER_SIZE:
1459 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1461 case _PC_REC_MAX_XFER_SIZE:
1462 *ap->a_retval = -1; /* means ``unlimited'' */
1464 case _PC_REC_MIN_XFER_SIZE:
1465 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1467 case _PC_REC_XFER_ALIGN:
1468 *ap->a_retval = PAGE_SIZE;
1470 case _PC_SYMLINK_MAX:
1471 *ap->a_retval = MAXPATHLEN;
1482 * Vnode pointer to File handle
1486 ext2_vptofh(struct vop_vptofh_args *ap)
1491 ip = VTOI(ap->a_vp);
1492 ufhp = (struct ufid *)ap->a_fhp;
1493 ufhp->ufid_len = sizeof(struct ufid);
1494 ufhp->ufid_ino = ip->i_number;
1495 ufhp->ufid_gen = ip->i_gen;
1500 * Initialize the vnode associated with a new inode, handle aliased
1504 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1511 vp->v_type = IFTOVT(ip->i_mode);
1512 if (vp->v_type == VFIFO)
1515 if (ip->i_number == EXT2_ROOTINO)
1516 vp->v_vflag |= VV_ROOT;
1517 ip->i_modrev = init_va_filerev();
1523 * Allocate a new inode.
1526 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1527 struct componentname *cnp)
1529 struct inode *ip, *pdir;
1535 if ((cnp->cn_flags & HASBUF) == 0)
1536 panic("ext2_makeinode: no name");
1539 if ((mode & IFMT) == 0)
1542 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1547 ip->i_gid = pdir->i_gid;
1552 * not the owner of the directory,
1553 * and we are hacking owners here, (only do this where told to)
1554 * and we are not giving it TOO root, (would subvert quotas)
1555 * then go ahead and give it to the other user.
1556 * Note that this drops off the execute bits for security.
1558 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1559 (pdir->i_mode & ISUID) &&
1560 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1561 ip->i_uid = pdir->i_uid;
1564 ip->i_uid = cnp->cn_cred->cr_uid;
1568 ip->i_uid = cnp->cn_cred->cr_uid;
1570 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1572 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
1574 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1575 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1576 ip->i_mode &= ~ISGID;
1579 if (cnp->cn_flags & ISWHITEOUT)
1580 ip->i_flags |= UF_OPAQUE;
1583 * Make sure inode goes to disk before directory entry.
1585 error = ext2_update(tvp, !DOINGASYNC(tvp));
1588 error = ext2_direnter(ip, dvp, cnp);
1597 * Write error occurred trying to update the inode
1598 * or the directory so must deallocate the inode.
1601 ip->i_flag |= IN_CHANGE;
1607 * Vnode op for reading.
1610 ext2_read(struct vop_read_args *ap)
1619 /* EXT4_EXT_LOCK(ip); */
1620 if (ip->i_flag & IN_E4EXTENTS)
1621 error = ext4_ext_read(ap);
1623 error = ext2_ind_read(ap);
1624 /* EXT4_EXT_UNLOCK(ip); */
1629 * Vnode op for reading.
1632 ext2_ind_read(struct vop_read_args *ap)
1637 struct m_ext2fs *fs;
1639 daddr_t lbn, nextlbn;
1641 long size, xfersize, blkoffset;
1642 int error, orig_resid, seqcount;
1647 ioflag = ap->a_ioflag;
1649 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1653 if (uio->uio_rw != UIO_READ)
1654 panic("%s: mode", "ext2_read");
1656 if (vp->v_type == VLNK) {
1657 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1658 panic("%s: short symlink", "ext2_read");
1659 } else if (vp->v_type != VREG && vp->v_type != VDIR)
1660 panic("%s: type %d", "ext2_read", vp->v_type);
1662 orig_resid = uio->uio_resid;
1663 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1664 if (orig_resid == 0)
1666 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1668 if (uio->uio_offset < ip->i_size &&
1669 uio->uio_offset >= fs->e2fs_maxfilesize)
1672 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1673 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1675 lbn = lblkno(fs, uio->uio_offset);
1677 size = blksize(fs, ip, lbn);
1678 blkoffset = blkoff(fs, uio->uio_offset);
1680 xfersize = fs->e2fs_fsize - blkoffset;
1681 if (uio->uio_resid < xfersize)
1682 xfersize = uio->uio_resid;
1683 if (bytesinfile < xfersize)
1684 xfersize = bytesinfile;
1686 if (lblktosize(fs, nextlbn) >= ip->i_size)
1687 error = bread(vp, lbn, size, NOCRED, &bp);
1688 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1689 error = cluster_read(vp, ip->i_size, lbn, size,
1690 NOCRED, blkoffset + uio->uio_resid, seqcount,
1692 } else if (seqcount > 1) {
1693 u_int nextsize = blksize(fs, ip, nextlbn);
1695 error = breadn(vp, lbn,
1696 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1698 error = bread(vp, lbn, size, NOCRED, &bp);
1706 * If IO_DIRECT then set B_DIRECT for the buffer. This
1707 * will cause us to attempt to release the buffer later on
1708 * and will cause the buffer cache to attempt to free the
1711 if (ioflag & IO_DIRECT)
1712 bp->b_flags |= B_DIRECT;
1715 * We should only get non-zero b_resid when an I/O error
1716 * has occurred, which should cause us to break above.
1717 * However, if the short read did not cause an error,
1718 * then we want to ensure that we do not uiomove bad
1719 * or uninitialized data.
1721 size -= bp->b_resid;
1722 if (size < xfersize) {
1727 error = uiomove((char *)bp->b_data + blkoffset,
1728 (int)xfersize, uio);
1732 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1734 * If it's VMIO or direct I/O, then we don't
1735 * need the buf, mark it available for
1736 * freeing. If it's non-direct VMIO, the VM has
1739 bp->b_flags |= B_RELBUF;
1743 * Otherwise let whoever
1744 * made the request take care of
1745 * freeing it. We just queue
1746 * it onto another list.
1753 * This can only happen in the case of an error because the loop
1754 * above resets bp to NULL on each iteration and on normal
1755 * completion has not set a new value into it. so it must have come
1756 * from a 'break' statement
1759 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1760 bp->b_flags |= B_RELBUF;
1767 if ((error == 0 || uio->uio_resid != orig_resid) &&
1768 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
1769 ip->i_flag |= IN_ACCESS;
1774 ext2_ioctl(struct vop_ioctl_args *ap)
1777 switch (ap->a_command) {
1780 return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1781 (off_t *)ap->a_data, ap->a_cred));
1788 * this function handles ext4 extents block mapping
1791 ext4_ext_read(struct vop_read_args *ap)
1796 struct m_ext2fs *fs;
1798 struct ext4_extent nex, *ep;
1799 struct ext4_extent_path path;
1800 daddr_t lbn, newblk;
1805 long size, xfersize, blkoffset;
1810 memset(&path, 0, sizeof(path));
1812 orig_resid = uio->uio_resid;
1813 KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__));
1814 if (orig_resid == 0)
1816 KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__));
1818 if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
1821 while (uio->uio_resid > 0) {
1822 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1824 lbn = lblkno(fs, uio->uio_offset);
1825 size = blksize(fs, ip, lbn);
1826 blkoffset = blkoff(fs, uio->uio_offset);
1828 xfersize = fs->e2fs_fsize - blkoffset;
1829 xfersize = MIN(xfersize, uio->uio_resid);
1830 xfersize = MIN(xfersize, bytesinfile);
1832 /* get block from ext4 extent cache */
1833 cache_type = ext4_ext_in_cache(ip, lbn, &nex);
1834 switch (cache_type) {
1835 case EXT4_EXT_CACHE_NO:
1836 ext4_ext_find_extent(fs, ip, lbn, &path);
1841 ext4_ext_put_cache(ip, ep, EXT4_EXT_CACHE_IN);
1843 newblk = lbn - ep->e_blk + (ep->e_start_lo |
1844 (daddr_t)ep->e_start_hi << 32);
1846 if (path.ep_bp != NULL) {
1852 case EXT4_EXT_CACHE_GAP:
1853 /* block has not been allocated yet */
1856 case EXT4_EXT_CACHE_IN:
1857 newblk = lbn - nex.e_blk + (nex.e_start_lo |
1858 (daddr_t)nex.e_start_hi << 32);
1862 panic("%s: invalid cache type", __func__);
1865 error = bread(ip->i_devvp, fsbtodb(fs, newblk), size, NOCRED, &bp);
1871 size -= bp->b_resid;
1872 if (size < xfersize) {
1879 error = uiomove(bp->b_data + blkoffset, (int)xfersize, uio);
1889 * Vnode op for writing.
1892 ext2_write(struct vop_write_args *ap)
1897 struct m_ext2fs *fs;
1901 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1903 ioflag = ap->a_ioflag;
1907 seqcount = ioflag >> IO_SEQSHIFT;
1911 if (uio->uio_rw != UIO_WRITE)
1912 panic("%s: mode", "ext2_write");
1915 switch (vp->v_type) {
1917 if (ioflag & IO_APPEND)
1918 uio->uio_offset = ip->i_size;
1919 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1925 /* XXX differs from ffs -- this is called from ext2_mkdir(). */
1926 if ((ioflag & IO_SYNC) == 0)
1927 panic("ext2_write: nonsync dir write");
1930 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
1931 vp->v_type, (intmax_t)uio->uio_offset,
1932 (intmax_t)uio->uio_resid);
1935 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
1936 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
1938 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
1941 * Maybe this should be above the vnode op call, but so long as
1942 * file servers have no limits, I don't think it matters.
1944 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
1947 resid = uio->uio_resid;
1949 if (seqcount > BA_SEQMAX)
1950 flags = BA_SEQMAX << BA_SEQSHIFT;
1952 flags = seqcount << BA_SEQSHIFT;
1953 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1956 for (error = 0; uio->uio_resid > 0;) {
1957 lbn = lblkno(fs, uio->uio_offset);
1958 blkoffset = blkoff(fs, uio->uio_offset);
1959 xfersize = fs->e2fs_fsize - blkoffset;
1960 if (uio->uio_resid < xfersize)
1961 xfersize = uio->uio_resid;
1962 if (uio->uio_offset + xfersize > ip->i_size)
1963 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
1966 * We must perform a read-before-write if the transfer size
1967 * does not cover the entire buffer.
1969 if (fs->e2fs_bsize > xfersize)
1972 flags &= ~BA_CLRBUF;
1973 error = ext2_balloc(ip, lbn, blkoffset + xfersize,
1974 ap->a_cred, &bp, flags);
1978 if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
1979 bp->b_flags |= B_NOCACHE;
1980 if (uio->uio_offset + xfersize > ip->i_size)
1981 ip->i_size = uio->uio_offset + xfersize;
1982 size = blksize(fs, ip, lbn) - bp->b_resid;
1983 if (size < xfersize)
1987 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1989 * If the buffer is not already filled and we encounter an
1990 * error while trying to fill it, we have to clear out any
1991 * garbage data from the pages instantiated for the buffer.
1992 * If we do not, a failed uiomove() during a write can leave
1993 * the prior contents of the pages exposed to a userland mmap.
1995 * Note that we need only clear buffers with a transfer size
1996 * equal to the block size because buffers with a shorter
1997 * transfer size were cleared above by the call to ext2_balloc()
1998 * with the BA_CLRBUF flag set.
2000 * If the source region for uiomove identically mmaps the
2001 * buffer, uiomove() performed the NOP copy, and the buffer
2002 * content remains valid because the page fault handler
2003 * validated the pages.
2005 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2006 fs->e2fs_bsize == xfersize)
2008 if (ioflag & (IO_VMIO|IO_DIRECT)) {
2009 bp->b_flags |= B_RELBUF;
2013 * If IO_SYNC each buffer is written synchronously. Otherwise
2014 * if we have a severe page deficiency write the buffer
2015 * asynchronously. Otherwise try to cluster, and if that
2016 * doesn't do it then either do an async write (if O_DIRECT),
2017 * or a delayed write (if not).
2019 if (ioflag & IO_SYNC) {
2021 } else if (vm_page_count_severe() ||
2022 buf_dirty_count_severe() ||
2023 (ioflag & IO_ASYNC)) {
2024 bp->b_flags |= B_CLUSTEROK;
2026 } else if (xfersize + blkoffset == fs->e2fs_fsize) {
2027 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2028 bp->b_flags |= B_CLUSTEROK;
2029 cluster_write(vp, bp, ip->i_size, seqcount, 0);
2033 } else if (ioflag & IO_DIRECT) {
2034 bp->b_flags |= B_CLUSTEROK;
2037 bp->b_flags |= B_CLUSTEROK;
2040 if (error || xfersize == 0)
2044 * If we successfully wrote any data, and we are not the superuser
2045 * we clear the setuid and setgid bits as a precaution against
2048 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2050 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2051 ip->i_mode &= ~(ISUID | ISGID);
2054 if (ioflag & IO_UNIT) {
2055 (void)ext2_truncate(vp, osize,
2056 ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2057 uio->uio_offset -= resid - uio->uio_resid;
2058 uio->uio_resid = resid;
2061 if (uio->uio_resid != resid) {
2062 ip->i_flag |= IN_CHANGE | IN_UPDATE;
2063 if (ioflag & IO_SYNC)
2064 error = ext2_update(vp, 1);
2073 ext2_getpages(struct vop_getpages_args *ap)
2079 pcount = round_page(ap->a_count) / PAGE_SIZE;
2080 mreq = ap->a_m[ap->a_reqpage];
2083 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
2084 * then the entire page is valid. Since the page may be mapped,
2085 * user programs might reference data beyond the actual end of file
2086 * occuring within the page. We have to zero that data.
2088 VM_OBJECT_WLOCK(mreq->object);
2090 if (mreq->valid != VM_PAGE_BITS_ALL)
2091 vm_page_zero_invalid(mreq, TRUE);
2092 for (i = 0; i < pcount; i++) {
2093 if (i != ap->a_reqpage) {
2094 vm_page_lock(ap->a_m[i]);
2095 vm_page_free(ap->a_m[i]);
2096 vm_page_unlock(ap->a_m[i]);
2099 VM_OBJECT_WUNLOCK(mreq->object);
2102 VM_OBJECT_WUNLOCK(mreq->object);
2104 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,