xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision 7d536dc855c85c15bf45f033d108a61b1f3cecc3)
1 /*-
2  *  modified for EXT2FS support in Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
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.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
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.
27  *
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
38  * SUCH DAMAGE.
39  *
40  *	@(#)ufs_vnops.c	8.7 (Berkeley) 2/3/94
41  *	@(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
42  * $FreeBSD$
43  */
44 
45 #include "opt_suiddir.h"
46 
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>
52 #include <sys/stat.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/endian.h>
56 #include <sys/priv.h>
57 #include <sys/rwlock.h>
58 #include <sys/mount.h>
59 #include <sys/unistd.h>
60 #include <sys/time.h>
61 #include <sys/vnode.h>
62 #include <sys/namei.h>
63 #include <sys/lockf.h>
64 #include <sys/event.h>
65 #include <sys/conf.h>
66 #include <sys/file.h>
67 
68 #include <vm/vm.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>
75 
76 #include "opt_directio.h"
77 
78 #include <ufs/ufs/dir.h>
79 
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>
87 
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 *);
92 
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 *,
96     struct thread *);
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;
120 
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_getpages_async =	vnode_pager_local_getpages_async,
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,
156 };
157 
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,
172 };
173 
174 /*
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.
179  */
180 static struct dirtemplate mastertemplate = {
181 	0, 12, 1, EXT2_FT_DIR, ".",
182 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
183 };
184 static struct dirtemplate omastertemplate = {
185 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
186 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
187 };
188 
189 static void
190 ext2_itimes_locked(struct vnode *vp)
191 {
192 	struct inode *ip;
193 	struct timespec ts;
194 
195 	ASSERT_VI_LOCKED(vp, __func__);
196 
197 	ip = VTOI(vp);
198 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
199 		return;
200 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
201 		ip->i_flag |= IN_LAZYMOD;
202 	else
203 		ip->i_flag |= IN_MODIFIED;
204 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
205 		vfs_timestamp(&ts);
206 		if (ip->i_flag & IN_ACCESS) {
207 			ip->i_atime = ts.tv_sec;
208 			ip->i_atimensec = ts.tv_nsec;
209 		}
210 		if (ip->i_flag & IN_UPDATE) {
211 			ip->i_mtime = ts.tv_sec;
212 			ip->i_mtimensec = ts.tv_nsec;
213 			ip->i_modrev++;
214 		}
215 		if (ip->i_flag & IN_CHANGE) {
216 			ip->i_ctime = ts.tv_sec;
217 			ip->i_ctimensec = ts.tv_nsec;
218 		}
219 	}
220 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
221 }
222 
223 void
224 ext2_itimes(struct vnode *vp)
225 {
226 
227 	VI_LOCK(vp);
228 	ext2_itimes_locked(vp);
229 	VI_UNLOCK(vp);
230 }
231 
232 /*
233  * Create a regular file
234  */
235 static int
236 ext2_create(struct vop_create_args *ap)
237 {
238 	int error;
239 
240 	error =
241 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
242 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
243 	if (error != 0)
244 		return (error);
245 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
246 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
247 	return (0);
248 }
249 
250 static int
251 ext2_open(struct vop_open_args *ap)
252 {
253 
254 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
255 		return (EOPNOTSUPP);
256 
257 	/*
258 	 * Files marked append-only must be opened for appending.
259 	 */
260 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
261 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
262 		return (EPERM);
263 
264 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
265 
266 	return (0);
267 }
268 
269 /*
270  * Close called.
271  *
272  * Update the times on the inode.
273  */
274 static int
275 ext2_close(struct vop_close_args *ap)
276 {
277 	struct vnode *vp = ap->a_vp;
278 
279 	VI_LOCK(vp);
280 	if (vp->v_usecount > 1)
281 		ext2_itimes_locked(vp);
282 	VI_UNLOCK(vp);
283 	return (0);
284 }
285 
286 static int
287 ext2_access(struct vop_access_args *ap)
288 {
289 	struct vnode *vp = ap->a_vp;
290 	struct inode *ip = VTOI(vp);
291 	accmode_t accmode = ap->a_accmode;
292 	int error;
293 
294 	if (vp->v_type == VBLK || vp->v_type == VCHR)
295 		return (EOPNOTSUPP);
296 
297 	/*
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.
301 	 */
302 	if (accmode & VWRITE) {
303 		switch (vp->v_type) {
304 		case VDIR:
305 		case VLNK:
306 		case VREG:
307 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
308 				return (EROFS);
309 			break;
310 		default:
311 			break;
312 		}
313 	}
314 
315 	/* If immutable bit set, nobody gets to write it. */
316 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
317 		return (EPERM);
318 
319 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
320 	    ap->a_accmode, ap->a_cred, NULL);
321 	return (error);
322 }
323 
324 static int
325 ext2_getattr(struct vop_getattr_args *ap)
326 {
327 	struct vnode *vp = ap->a_vp;
328 	struct inode *ip = VTOI(vp);
329 	struct vattr *vap = ap->a_vap;
330 
331 	ext2_itimes(vp);
332 	/*
333 	 * Copy from inode table
334 	 */
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;
352 	}
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;
359 	return (0);
360 }
361 
362 /*
363  * Set attribute vnode op. called from several syscalls
364  */
365 static int
366 ext2_setattr(struct vop_setattr_args *ap)
367 {
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;
373 	int error;
374 
375 	/*
376 	 * Check for unsettable attributes.
377 	 */
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)) {
382 		return (EINVAL);
383 	}
384 	if (vap->va_flags != VNOVAL) {
385 		/* Disallow flags not supported by ext2fs. */
386 		if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
387 			return (EOPNOTSUPP);
388 
389 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
390 			return (EROFS);
391 		/*
392 		 * Callers may only modify the file flags on objects they
393 		 * have VADMIN rights for.
394 		 */
395 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
396 			return (error);
397 		/*
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.
403 		 */
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);
407 				if (error)
408 					return (error);
409 			}
410 		} else {
411 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
412 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
413 				return (EPERM);
414 		}
415 		ip->i_flags = vap->va_flags;
416 		ip->i_flag |= IN_CHANGE;
417 		if (ip->i_flags & (IMMUTABLE | APPEND))
418 			return (0);
419 	}
420 	if (ip->i_flags & (IMMUTABLE | APPEND))
421 		return (EPERM);
422 	/*
423 	 * Go through the fields and update iff not VNOVAL.
424 	 */
425 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
426 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
427 			return (EROFS);
428 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
429 		    td)) != 0)
430 			return (error);
431 	}
432 	if (vap->va_size != VNOVAL) {
433 		/*
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.
437 		 */
438 		switch (vp->v_type) {
439 		case VDIR:
440 			return (EISDIR);
441 		case VLNK:
442 		case VREG:
443 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
444 				return (EROFS);
445 			break;
446 		default:
447 			break;
448 		}
449 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
450 			return (error);
451 	}
452 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
453 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
454 			return (EROFS);
455 		/*
456 		 * From utimes(2):
457 		 * If times is NULL, ... The caller must be the owner of
458 		 * the file, have permission to write the file, or be the
459 		 * super-user.
460 		 * If times is non-NULL, ... The caller must be the owner of
461 		 * the file or be the super-user.
462 		 */
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))))
466 			return (error);
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;
472 		}
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;
477 		}
478 		ip->i_birthtime = vap->va_birthtime.tv_sec;
479 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
480 		error = ext2_update(vp, 0);
481 		if (error)
482 			return (error);
483 	}
484 	error = 0;
485 	if (vap->va_mode != (mode_t)VNOVAL) {
486 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
487 			return (EROFS);
488 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
489 	}
490 	return (error);
491 }
492 
493 /*
494  * Change the mode on a file.
495  * Inode must be locked before calling.
496  */
497 static int
498 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
499 {
500 	struct inode *ip = VTOI(vp);
501 	int error;
502 
503 	/*
504 	 * To modify the permissions on a file, must possess VADMIN
505 	 * for that file.
506 	 */
507 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
508 		return (error);
509 	/*
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.
513 	 */
514 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
515 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
516 		if (error)
517 			return (EFTYPE);
518 	}
519 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
520 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
521 		if (error)
522 			return (error);
523 	}
524 	ip->i_mode &= ~ALLPERMS;
525 	ip->i_mode |= (mode & ALLPERMS);
526 	ip->i_flag |= IN_CHANGE;
527 	return (0);
528 }
529 
530 /*
531  * Perform chown operation on inode ip;
532  * inode must be locked prior to call.
533  */
534 static int
535 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
536     struct thread *td)
537 {
538 	struct inode *ip = VTOI(vp);
539 	uid_t ouid;
540 	gid_t ogid;
541 	int error = 0;
542 
543 	if (uid == (uid_t)VNOVAL)
544 		uid = ip->i_uid;
545 	if (gid == (gid_t)VNOVAL)
546 		gid = ip->i_gid;
547 	/*
548 	 * To modify the ownership of a file, must possess VADMIN
549 	 * for that file.
550 	 */
551 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
552 		return (error);
553 	/*
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
556 	 * have privilege.
557 	 */
558 	if (uid != ip->i_uid || (gid != ip->i_gid &&
559 	    !groupmember(gid, cred))) {
560 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
561 		if (error)
562 			return (error);
563 	}
564 	ogid = ip->i_gid;
565 	ouid = ip->i_uid;
566 	ip->i_gid = gid;
567 	ip->i_uid = uid;
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);
572 	}
573 	return (0);
574 }
575 
576 /*
577  * Synch an open file.
578  */
579 /* ARGSUSED */
580 static int
581 ext2_fsync(struct vop_fsync_args *ap)
582 {
583 	/*
584 	 * Flush all dirty buffers associated with a vnode.
585 	 */
586 
587 	vop_stdfsync(ap);
588 
589 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
590 }
591 
592 /*
593  * Mknod vnode call
594  */
595 /* ARGSUSED */
596 static int
597 ext2_mknod(struct vop_mknod_args *ap)
598 {
599 	struct vattr *vap = ap->a_vap;
600 	struct vnode **vpp = ap->a_vpp;
601 	struct inode *ip;
602 	ino_t ino;
603 	int error;
604 
605 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
606 	    ap->a_dvp, vpp, ap->a_cnp);
607 	if (error)
608 		return (error);
609 	ip = VTOI(*vpp);
610 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
611 	if (vap->va_rdev != VNOVAL) {
612 		/*
613 		 * Want to be able to use this to make badblock
614 		 * inodes, so don't truncate the dev number.
615 		 */
616 		ip->i_rdev = vap->va_rdev;
617 	}
618 	/*
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.
622 	 */
623 	(*vpp)->v_type = VNON;
624 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
625 	vgone(*vpp);
626 	vput(*vpp);
627 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
628 	if (error) {
629 		*vpp = NULL;
630 		return (error);
631 	}
632 	return (0);
633 }
634 
635 static int
636 ext2_remove(struct vop_remove_args *ap)
637 {
638 	struct inode *ip;
639 	struct vnode *vp = ap->a_vp;
640 	struct vnode *dvp = ap->a_dvp;
641 	int error;
642 
643 	ip = VTOI(vp);
644 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
645 	    (VTOI(dvp)->i_flags & APPEND)) {
646 		error = EPERM;
647 		goto out;
648 	}
649 	error = ext2_dirremove(dvp, ap->a_cnp);
650 	if (error == 0) {
651 		ip->i_nlink--;
652 		ip->i_flag |= IN_CHANGE;
653 	}
654 out:
655 	return (error);
656 }
657 
658 /*
659  * link vnode call
660  */
661 static int
662 ext2_link(struct vop_link_args *ap)
663 {
664 	struct vnode *vp = ap->a_vp;
665 	struct vnode *tdvp = ap->a_tdvp;
666 	struct componentname *cnp = ap->a_cnp;
667 	struct inode *ip;
668 	int error;
669 
670 #ifdef INVARIANTS
671 	if ((cnp->cn_flags & HASBUF) == 0)
672 		panic("ext2_link: no name");
673 #endif
674 	ip = VTOI(vp);
675 	if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
676 		error = EMLINK;
677 		goto out;
678 	}
679 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
680 		error = EPERM;
681 		goto out;
682 	}
683 	ip->i_nlink++;
684 	ip->i_flag |= IN_CHANGE;
685 	error = ext2_update(vp, !DOINGASYNC(vp));
686 	if (!error)
687 		error = ext2_direnter(ip, tdvp, cnp);
688 	if (error) {
689 		ip->i_nlink--;
690 		ip->i_flag |= IN_CHANGE;
691 	}
692 out:
693 	return (error);
694 }
695 
696 /*
697  * Rename system call.
698  * 	rename("foo", "bar");
699  * is essentially
700  *	unlink("bar");
701  *	link("foo", "bar");
702  *	unlink("foo");
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.
706  *
707  * Basic algorithm is:
708  *
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,
714  *    delete it first.
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
718  *    directory.
719  */
720 static int
721 ext2_rename(struct vop_rename_args *ap)
722 {
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;
732 	int error = 0;
733 	u_char namlen;
734 
735 #ifdef INVARIANTS
736 	if ((tcnp->cn_flags & HASBUF) == 0 ||
737 	    (fcnp->cn_flags & HASBUF) == 0)
738 		panic("ext2_rename: no name");
739 #endif
740 	/*
741 	 * Check for cross-device rename.
742 	 */
743 	if ((fvp->v_mount != tdvp->v_mount) ||
744 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
745 		error = EXDEV;
746 abortit:
747 		if (tdvp == tvp)
748 			vrele(tdvp);
749 		else
750 			vput(tdvp);
751 		if (tvp)
752 			vput(tvp);
753 		vrele(fdvp);
754 		vrele(fvp);
755 		return (error);
756 	}
757 
758 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
759 	    (VTOI(tdvp)->i_flags & APPEND))) {
760 		error = EPERM;
761 		goto abortit;
762 	}
763 
764 	/*
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.
767 	 */
768 	if (fvp == tvp) {
769 		printf("ext2_rename: fvp == tvp (can't happen)\n");
770 		error = 0;
771 		goto abortit;
772 	}
773 
774 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
775 		goto abortit;
776 	dp = VTOI(fdvp);
777 	ip = VTOI(fvp);
778 	if (ip->i_nlink >= EXT2_LINK_MAX) {
779 		VOP_UNLOCK(fvp, 0);
780 		error = EMLINK;
781 		goto abortit;
782 	}
783 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
784 	    || (dp->i_flags & APPEND)) {
785 		VOP_UNLOCK(fvp, 0);
786 		error = EPERM;
787 		goto abortit;
788 	}
789 	if ((ip->i_mode & IFMT) == IFDIR) {
790 		/*
791 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
792 		 */
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)) {
796 			VOP_UNLOCK(fvp, 0);
797 			error = EINVAL;
798 			goto abortit;
799 		}
800 		ip->i_flag |= IN_RENAME;
801 		oldparent = dp->i_number;
802 		doingdirectory++;
803 	}
804 	vrele(fdvp);
805 
806 	/*
807 	 * When the target exists, both the directory
808 	 * and target vnodes are returned locked.
809 	 */
810 	dp = VTOI(tdvp);
811 	xp = NULL;
812 	if (tvp)
813 		xp = VTOI(tvp);
814 
815 	/*
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.
820 	 */
821 	ip->i_nlink++;
822 	ip->i_flag |= IN_CHANGE;
823 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
824 		VOP_UNLOCK(fvp, 0);
825 		goto bad;
826 	}
827 
828 	/*
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().
837 	 */
838 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
839 	VOP_UNLOCK(fvp, 0);
840 	if (oldparent != dp->i_number)
841 		newparent = dp->i_number;
842 	if (doingdirectory && newparent) {
843 		if (error)	/* write access check above */
844 			goto bad;
845 		if (xp != NULL)
846 			vput(tvp);
847 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
848 		if (error)
849 			goto out;
850 		VREF(tdvp);
851 		error = relookup(tdvp, &tvp, tcnp);
852 		if (error)
853 			goto out;
854 		vrele(tdvp);
855 		dp = VTOI(tdvp);
856 		xp = NULL;
857 		if (tvp)
858 			xp = VTOI(tvp);
859 	}
860 	/*
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.
866 	 */
867 	if (xp == NULL) {
868 		if (dp->i_devvp != ip->i_devvp)
869 			panic("ext2_rename: EXDEV");
870 		/*
871 		 * Account for ".." in new directory.
872 		 * When source and destination have the same
873 		 * parent we don't fool with the link count.
874 		 */
875 		if (doingdirectory && newparent) {
876 			if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
877 				error = EMLINK;
878 				goto bad;
879 			}
880 			dp->i_nlink++;
881 			dp->i_flag |= IN_CHANGE;
882 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
883 			if (error)
884 				goto bad;
885 		}
886 		error = ext2_direnter(ip, tdvp, tcnp);
887 		if (error) {
888 			if (doingdirectory && newparent) {
889 				dp->i_nlink--;
890 				dp->i_flag |= IN_CHANGE;
891 				(void)ext2_update(tdvp, 1);
892 			}
893 			goto bad;
894 		}
895 		vput(tdvp);
896 	} else {
897 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
898 		       panic("ext2_rename: EXDEV");
899 		/*
900 		 * Short circuit rename(foo, foo).
901 		 */
902 		if (xp->i_number == ip->i_number)
903 			panic("ext2_rename: same file");
904 		/*
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.
909 		 */
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) {
913 			error = EPERM;
914 			goto bad;
915 		}
916 		/*
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).
920 		 */
921 		if ((xp->i_mode&IFMT) == IFDIR) {
922 			if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
923 			    xp->i_nlink > 2) {
924 				error = ENOTEMPTY;
925 				goto bad;
926 			}
927 			if (!doingdirectory) {
928 				error = ENOTDIR;
929 				goto bad;
930 			}
931 			cache_purge(tdvp);
932 		} else if (doingdirectory) {
933 			error = EISDIR;
934 			goto bad;
935 		}
936 		error = ext2_dirrewrite(dp, ip, tcnp);
937 		if (error)
938 			goto bad;
939 		/*
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.
944 		 */
945 		if (doingdirectory && !newparent) {
946 			dp->i_nlink--;
947 			dp->i_flag |= IN_CHANGE;
948 		}
949 		vput(tdvp);
950 		/*
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.
959 		 */
960 		xp->i_nlink--;
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);
966 		}
967 		xp->i_flag |= IN_CHANGE;
968 		vput(tvp);
969 		xp = NULL;
970 	}
971 
972 	/*
973 	 * 3) Unlink the source.
974 	 */
975 	fcnp->cn_flags &= ~MODMASK;
976 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
977 	VREF(fdvp);
978 	error = relookup(fdvp, &fvp, fcnp);
979 	if (error == 0)
980 		vrele(fdvp);
981 	if (fvp != NULL) {
982 		xp = VTOI(fvp);
983 		dp = VTOI(fdvp);
984 	} else {
985 		/*
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.
989 		 */
990 		vrele(ap->a_fvp);
991 		return (0);
992 	}
993 	/*
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
1001 	 * rename.
1002 	 */
1003 	if (xp != ip) {
1004 		/*
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.
1008 		 */
1009 	} else {
1010 		/*
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.
1015 		 */
1016 		if (doingdirectory && newparent) {
1017 			dp->i_nlink--;
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);
1023 			if (error == 0) {
1024 				/* Like ufs little-endian: */
1025 				namlen = dirbuf.dotdot_type;
1026 				if (namlen != 2 ||
1027 				    dirbuf.dotdot_name[0] != '.' ||
1028 				    dirbuf.dotdot_name[1] != '.') {
1029 					ext2_dirbad(xp, (doff_t)12,
1030 					    "rename: mangled dir");
1031 				} else {
1032 					dirbuf.dotdot_ino = newparent;
1033 					(void) vn_rdwr(UIO_WRITE, fvp,
1034 					    (caddr_t)&dirbuf,
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);
1040 					cache_purge(fdvp);
1041 				}
1042 			}
1043 		}
1044 		error = ext2_dirremove(fdvp, fcnp);
1045 		if (!error) {
1046 			xp->i_nlink--;
1047 			xp->i_flag |= IN_CHANGE;
1048 		}
1049 		xp->i_flag &= ~IN_RENAME;
1050 	}
1051 	if (dp)
1052 		vput(fdvp);
1053 	if (xp)
1054 		vput(fvp);
1055 	vrele(ap->a_fvp);
1056 	return (error);
1057 
1058 bad:
1059 	if (xp)
1060 		vput(ITOV(xp));
1061 	vput(ITOV(dp));
1062 out:
1063 	if (doingdirectory)
1064 		ip->i_flag &= ~IN_RENAME;
1065 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1066 		ip->i_nlink--;
1067 		ip->i_flag |= IN_CHANGE;
1068 		ip->i_flag &= ~IN_RENAME;
1069 		vput(fvp);
1070 	} else
1071 		vrele(fvp);
1072 	return (error);
1073 }
1074 
1075 /*
1076  * Mkdir system call
1077  */
1078 static int
1079 ext2_mkdir(struct vop_mkdir_args *ap)
1080 {
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;
1085 	struct vnode *tvp;
1086 	struct dirtemplate dirtemplate, *dtp;
1087 	int error, dmode;
1088 
1089 #ifdef INVARIANTS
1090 	if ((cnp->cn_flags & HASBUF) == 0)
1091 		panic("ext2_mkdir: no name");
1092 #endif
1093 	dp = VTOI(dvp);
1094 	if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1095 		error = EMLINK;
1096 		goto out;
1097 	}
1098 	dmode = vap->va_mode & 0777;
1099 	dmode |= IFDIR;
1100 	/*
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.
1104 	 */
1105 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1106 	if (error)
1107 		goto out;
1108 	ip = VTOI(tvp);
1109 	ip->i_gid = dp->i_gid;
1110 #ifdef SUIDDIR
1111 	{
1112 		/*
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.
1119 		 */
1120 		if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1121 		   (dp->i_mode & ISUID) && dp->i_uid) {
1122 			dmode |= ISUID;
1123 			ip->i_uid = dp->i_uid;
1124 		} else {
1125 			ip->i_uid = cnp->cn_cred->cr_uid;
1126 		}
1127 	}
1128 #else
1129 	ip->i_uid = cnp->cn_cred->cr_uid;
1130 #endif
1131 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1132 	ip->i_mode = dmode;
1133 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1134 	ip->i_nlink = 2;
1135 	if (cnp->cn_flags & ISWHITEOUT)
1136 		ip->i_flags |= UF_OPAQUE;
1137 	error = ext2_update(tvp, 1);
1138 
1139 	/*
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.
1144 	 */
1145 	dp->i_nlink++;
1146 	dp->i_flag |= IN_CHANGE;
1147 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1148 	if (error)
1149 		goto bad;
1150 
1151 	/* Initialize directory with "." and ".." from static template. */
1152 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1153 	    EXT2F_INCOMPAT_FTYPE))
1154 		dtp = &mastertemplate;
1155 	else
1156 		dtp = &omastertemplate;
1157 	dirtemplate = *dtp;
1158 	dirtemplate.dot_ino = ip->i_number;
1159 	dirtemplate.dotdot_ino = dp->i_number;
1160 	/* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1161 	 * so let's just redefine it - for this function only
1162 	 */
1163 #undef  DIRBLKSIZ
1164 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1165 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1166 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1167 	    sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1168 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1169 	    NULL, NULL);
1170 	if (error) {
1171 		dp->i_nlink--;
1172 		dp->i_flag |= IN_CHANGE;
1173 		goto bad;
1174 	}
1175 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1176 		/* XXX should grow with balloc() */
1177 		panic("ext2_mkdir: blksize");
1178 	else {
1179 		ip->i_size = DIRBLKSIZ;
1180 		ip->i_flag |= IN_CHANGE;
1181 	}
1182 
1183 	/* Directory set up, now install its entry in the parent directory. */
1184 	error = ext2_direnter(ip, dvp, cnp);
1185 	if (error) {
1186 		dp->i_nlink--;
1187 		dp->i_flag |= IN_CHANGE;
1188 	}
1189 bad:
1190 	/*
1191 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1192 	 * for us because we set the link count to 0.
1193 	 */
1194 	if (error) {
1195 		ip->i_nlink = 0;
1196 		ip->i_flag |= IN_CHANGE;
1197 		vput(tvp);
1198 	} else
1199 		*ap->a_vpp = tvp;
1200 out:
1201 	return (error);
1202 #undef  DIRBLKSIZ
1203 #define DIRBLKSIZ  DEV_BSIZE
1204 }
1205 
1206 /*
1207  * Rmdir system call.
1208  */
1209 static int
1210 ext2_rmdir(struct vop_rmdir_args *ap)
1211 {
1212 	struct vnode *vp = ap->a_vp;
1213 	struct vnode *dvp = ap->a_dvp;
1214 	struct componentname *cnp = ap->a_cnp;
1215 	struct inode *ip, *dp;
1216 	int error;
1217 
1218 	ip = VTOI(vp);
1219 	dp = VTOI(dvp);
1220 
1221 	/*
1222 	 * Verify the directory is empty (and valid).
1223 	 * (Rmdir ".." won't be valid since
1224 	 *  ".." will contain a reference to
1225 	 *  the current directory and thus be
1226 	 *  non-empty.)
1227 	 */
1228 	if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1229 		error = ENOTEMPTY;
1230 		goto out;
1231 	}
1232 	if ((dp->i_flags & APPEND)
1233 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1234 		error = EPERM;
1235 		goto out;
1236 	}
1237 	/*
1238 	 * Delete reference to directory before purging
1239 	 * inode.  If we crash in between, the directory
1240 	 * will be reattached to lost+found,
1241 	 */
1242 	error = ext2_dirremove(dvp, cnp);
1243 	if (error)
1244 		goto out;
1245 	dp->i_nlink--;
1246 	dp->i_flag |= IN_CHANGE;
1247 	cache_purge(dvp);
1248 	VOP_UNLOCK(dvp, 0);
1249 	/*
1250 	 * Truncate inode.  The only stuff left
1251 	 * in the directory is "." and "..".  The
1252 	 * "." reference is inconsequential since
1253 	 * we're quashing it.  The ".." reference
1254 	 * has already been adjusted above.  We've
1255 	 * removed the "." reference and the reference
1256 	 * in the parent directory, but there may be
1257 	 * other hard links so decrement by 2 and
1258 	 * worry about them later.
1259 	 */
1260 	ip->i_nlink -= 2;
1261 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1262 	    cnp->cn_thread);
1263 	cache_purge(ITOV(ip));
1264 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1265 		VOP_UNLOCK(vp, 0);
1266 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1267 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1268 	}
1269 out:
1270 	return (error);
1271 }
1272 
1273 /*
1274  * symlink -- make a symbolic link
1275  */
1276 static int
1277 ext2_symlink(struct vop_symlink_args *ap)
1278 {
1279 	struct vnode *vp, **vpp = ap->a_vpp;
1280 	struct inode *ip;
1281 	int len, error;
1282 
1283 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1284 	    vpp, ap->a_cnp);
1285 	if (error)
1286 		return (error);
1287 	vp = *vpp;
1288 	len = strlen(ap->a_target);
1289 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1290 		ip = VTOI(vp);
1291 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1292 		ip->i_size = len;
1293 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1294 	} else
1295 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1296 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1297 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1298 	if (error)
1299 		vput(vp);
1300 	return (error);
1301 }
1302 
1303 /*
1304  * Return target name of a symbolic link
1305  */
1306 static int
1307 ext2_readlink(struct vop_readlink_args *ap)
1308 {
1309 	struct vnode *vp = ap->a_vp;
1310 	struct inode *ip = VTOI(vp);
1311 	int isize;
1312 
1313 	isize = ip->i_size;
1314 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1315 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1316 		return (0);
1317 	}
1318 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1319 }
1320 
1321 /*
1322  * Calculate the logical to physical mapping if not done already,
1323  * then call the device strategy routine.
1324  *
1325  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1326  * deadlock on memory.  See ext2_bmap() for details.
1327  */
1328 static int
1329 ext2_strategy(struct vop_strategy_args *ap)
1330 {
1331 	struct buf *bp = ap->a_bp;
1332 	struct vnode *vp = ap->a_vp;
1333 	struct bufobj *bo;
1334 	daddr_t blkno;
1335 	int error;
1336 
1337 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1338 		panic("ext2_strategy: spec");
1339 	if (bp->b_blkno == bp->b_lblkno) {
1340 		error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1341 		bp->b_blkno = blkno;
1342 		if (error) {
1343 			bp->b_error = error;
1344 			bp->b_ioflags |= BIO_ERROR;
1345 			bufdone(bp);
1346 			return (0);
1347 		}
1348 		if ((long)bp->b_blkno == -1)
1349 			vfs_bio_clrbuf(bp);
1350 	}
1351 	if ((long)bp->b_blkno == -1) {
1352 		bufdone(bp);
1353 		return (0);
1354 	}
1355 	bp->b_iooffset = dbtob(bp->b_blkno);
1356 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1357 	BO_STRATEGY(bo, bp);
1358 	return (0);
1359 }
1360 
1361 /*
1362  * Print out the contents of an inode.
1363  */
1364 static int
1365 ext2_print(struct vop_print_args *ap)
1366 {
1367 	struct vnode *vp = ap->a_vp;
1368 	struct inode *ip = VTOI(vp);
1369 
1370 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1371 	if (vp->v_type == VFIFO)
1372 		fifo_printinfo(vp);
1373 	printf("\n");
1374 	return (0);
1375 }
1376 
1377 /*
1378  * Close wrapper for fifos.
1379  *
1380  * Update the times on the inode then do device close.
1381  */
1382 static int
1383 ext2fifo_close(struct vop_close_args *ap)
1384 {
1385 	struct vnode *vp = ap->a_vp;
1386 
1387 	VI_LOCK(vp);
1388 	if (vp->v_usecount > 1)
1389 		ext2_itimes_locked(vp);
1390 	VI_UNLOCK(vp);
1391 	return (fifo_specops.vop_close(ap));
1392 }
1393 
1394 /*
1395  * Kqfilter wrapper for fifos.
1396  *
1397  * Fall through to ext2 kqfilter routines if needed
1398  */
1399 static int
1400 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1401 {
1402 	int error;
1403 
1404 	error = fifo_specops.vop_kqfilter(ap);
1405 	if (error)
1406 		error = vfs_kqfilter(ap);
1407 	return (error);
1408 }
1409 
1410 /*
1411  * Return POSIX pathconf information applicable to ext2 filesystems.
1412  */
1413 static int
1414 ext2_pathconf(struct vop_pathconf_args *ap)
1415 {
1416 	int error = 0;
1417 
1418 	switch (ap->a_name) {
1419 	case _PC_LINK_MAX:
1420 		*ap->a_retval = EXT2_LINK_MAX;
1421 		break;
1422 	case _PC_NAME_MAX:
1423 		*ap->a_retval = NAME_MAX;
1424 		break;
1425 	case _PC_PATH_MAX:
1426 		*ap->a_retval = PATH_MAX;
1427 		break;
1428 	case _PC_PIPE_BUF:
1429 		*ap->a_retval = PIPE_BUF;
1430 		break;
1431 	case _PC_CHOWN_RESTRICTED:
1432 		*ap->a_retval = 1;
1433 		break;
1434 	case _PC_NO_TRUNC:
1435 		*ap->a_retval = 1;
1436 		break;
1437 	case _PC_MIN_HOLE_SIZE:
1438 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1439 		break;
1440 	case _PC_ASYNC_IO:
1441 		/* _PC_ASYNC_IO should have been handled by upper layers. */
1442 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1443 		error = EINVAL;
1444 		break;
1445 	case _PC_PRIO_IO:
1446 		*ap->a_retval = 0;
1447 		break;
1448 	case _PC_SYNC_IO:
1449 		*ap->a_retval = 0;
1450 		break;
1451 	case _PC_ALLOC_SIZE_MIN:
1452 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1453 		break;
1454 	case _PC_FILESIZEBITS:
1455 		*ap->a_retval = 64;
1456 		break;
1457 	case _PC_REC_INCR_XFER_SIZE:
1458 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1459 		break;
1460 	case _PC_REC_MAX_XFER_SIZE:
1461 		*ap->a_retval = -1; /* means ``unlimited'' */
1462 		break;
1463 	case _PC_REC_MIN_XFER_SIZE:
1464 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1465 		break;
1466 	case _PC_REC_XFER_ALIGN:
1467 		*ap->a_retval = PAGE_SIZE;
1468 		break;
1469 	case _PC_SYMLINK_MAX:
1470 		*ap->a_retval = MAXPATHLEN;
1471 		break;
1472 
1473 	default:
1474 		error = EINVAL;
1475 		break;
1476 	}
1477 	return (error);
1478 }
1479 
1480 /*
1481  * Vnode pointer to File handle
1482  */
1483 /* ARGSUSED */
1484 static int
1485 ext2_vptofh(struct vop_vptofh_args *ap)
1486 {
1487 	struct inode *ip;
1488 	struct ufid *ufhp;
1489 
1490 	ip = VTOI(ap->a_vp);
1491 	ufhp = (struct ufid *)ap->a_fhp;
1492 	ufhp->ufid_len = sizeof(struct ufid);
1493 	ufhp->ufid_ino = ip->i_number;
1494 	ufhp->ufid_gen = ip->i_gen;
1495 	return (0);
1496 }
1497 
1498 /*
1499  * Initialize the vnode associated with a new inode, handle aliased
1500  * vnodes.
1501  */
1502 int
1503 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1504 {
1505 	struct inode *ip;
1506 	struct vnode *vp;
1507 
1508 	vp = *vpp;
1509 	ip = VTOI(vp);
1510 	vp->v_type = IFTOVT(ip->i_mode);
1511 	if (vp->v_type == VFIFO)
1512 		vp->v_op = fifoops;
1513 
1514 	if (ip->i_number == EXT2_ROOTINO)
1515 		vp->v_vflag |= VV_ROOT;
1516 	ip->i_modrev = init_va_filerev();
1517 	*vpp = vp;
1518 	return (0);
1519 }
1520 
1521 /*
1522  * Allocate a new inode.
1523  */
1524 static int
1525 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1526     struct componentname *cnp)
1527 {
1528 	struct inode *ip, *pdir;
1529 	struct vnode *tvp;
1530 	int error;
1531 
1532 	pdir = VTOI(dvp);
1533 #ifdef INVARIANTS
1534 	if ((cnp->cn_flags & HASBUF) == 0)
1535 		panic("ext2_makeinode: no name");
1536 #endif
1537 	*vpp = NULL;
1538 	if ((mode & IFMT) == 0)
1539 		mode |= IFREG;
1540 
1541 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1542 	if (error) {
1543 		return (error);
1544 	}
1545 	ip = VTOI(tvp);
1546 	ip->i_gid = pdir->i_gid;
1547 #ifdef SUIDDIR
1548 	{
1549 		/*
1550 		 * if we are
1551 		 * not the owner of the directory,
1552 		 * and we are hacking owners here, (only do this where told to)
1553 		 * and we are not giving it TOO root, (would subvert quotas)
1554 		 * then go ahead and give it to the other user.
1555 		 * Note that this drops off the execute bits for security.
1556 		 */
1557 		if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1558 		     (pdir->i_mode & ISUID) &&
1559 		     (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1560 			ip->i_uid = pdir->i_uid;
1561 			mode &= ~07111;
1562 		} else {
1563 			ip->i_uid = cnp->cn_cred->cr_uid;
1564 		}
1565 	}
1566 #else
1567 	ip->i_uid = cnp->cn_cred->cr_uid;
1568 #endif
1569 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1570 	ip->i_mode = mode;
1571 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1572 	ip->i_nlink = 1;
1573 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1574 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1575 			ip->i_mode &= ~ISGID;
1576 	}
1577 
1578 	if (cnp->cn_flags & ISWHITEOUT)
1579 		ip->i_flags |= UF_OPAQUE;
1580 
1581 	/*
1582 	 * Make sure inode goes to disk before directory entry.
1583 	 */
1584 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1585 	if (error)
1586 		goto bad;
1587 	error = ext2_direnter(ip, dvp, cnp);
1588 	if (error)
1589 		goto bad;
1590 
1591 	*vpp = tvp;
1592 	return (0);
1593 
1594 bad:
1595 	/*
1596 	 * Write error occurred trying to update the inode
1597 	 * or the directory so must deallocate the inode.
1598 	 */
1599 	ip->i_nlink = 0;
1600 	ip->i_flag |= IN_CHANGE;
1601 	vput(tvp);
1602 	return (error);
1603 }
1604 
1605 /*
1606  * Vnode op for reading.
1607  */
1608 static int
1609 ext2_read(struct vop_read_args *ap)
1610 {
1611 	struct vnode *vp;
1612 	struct inode *ip;
1613 	int error;
1614 
1615 	vp = ap->a_vp;
1616 	ip = VTOI(vp);
1617 
1618 	/*EXT4_EXT_LOCK(ip);*/
1619 	if (ip->i_flag & IN_E4EXTENTS)
1620 		error = ext4_ext_read(ap);
1621 	else
1622 		error = ext2_ind_read(ap);
1623 	/*EXT4_EXT_UNLOCK(ip);*/
1624 	return (error);
1625 }
1626 
1627 /*
1628  * Vnode op for reading.
1629  */
1630 static int
1631 ext2_ind_read(struct vop_read_args *ap)
1632 {
1633 	struct vnode *vp;
1634 	struct inode *ip;
1635 	struct uio *uio;
1636 	struct m_ext2fs *fs;
1637 	struct buf *bp;
1638 	daddr_t lbn, nextlbn;
1639 	off_t bytesinfile;
1640 	long size, xfersize, blkoffset;
1641 	int error, orig_resid, seqcount;
1642 	int ioflag;
1643 
1644 	vp = ap->a_vp;
1645 	uio = ap->a_uio;
1646 	ioflag = ap->a_ioflag;
1647 
1648 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1649 	ip = VTOI(vp);
1650 
1651 #ifdef INVARIANTS
1652 	if (uio->uio_rw != UIO_READ)
1653 		panic("%s: mode", "ext2_read");
1654 
1655 	if (vp->v_type == VLNK) {
1656 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1657 			panic("%s: short symlink", "ext2_read");
1658 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
1659 		panic("%s: type %d", "ext2_read", vp->v_type);
1660 #endif
1661 	orig_resid = uio->uio_resid;
1662 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1663 	if (orig_resid == 0)
1664 		return (0);
1665 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1666 	fs = ip->i_e2fs;
1667 	if (uio->uio_offset < ip->i_size &&
1668 	    uio->uio_offset >= fs->e2fs_maxfilesize)
1669 	    	return (EOVERFLOW);
1670 
1671 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1672 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1673 			break;
1674 		lbn = lblkno(fs, uio->uio_offset);
1675 		nextlbn = lbn + 1;
1676 		size = blksize(fs, ip, lbn);
1677 		blkoffset = blkoff(fs, uio->uio_offset);
1678 
1679 		xfersize = fs->e2fs_fsize - blkoffset;
1680 		if (uio->uio_resid < xfersize)
1681 			xfersize = uio->uio_resid;
1682 		if (bytesinfile < xfersize)
1683 			xfersize = bytesinfile;
1684 
1685 		if (lblktosize(fs, nextlbn) >= ip->i_size)
1686 			error = bread(vp, lbn, size, NOCRED, &bp);
1687 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1688 			error = cluster_read(vp, ip->i_size, lbn, size,
1689 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
1690 			    0, &bp);
1691 		} else if (seqcount > 1) {
1692 			u_int nextsize = blksize(fs, ip, nextlbn);
1693 			error = breadn(vp, lbn,
1694 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1695 		} else
1696 			error = bread(vp, lbn, size, NOCRED, &bp);
1697 		if (error) {
1698 			brelse(bp);
1699 			bp = NULL;
1700 			break;
1701 		}
1702 
1703 		/*
1704 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
1705 		 * will cause us to attempt to release the buffer later on
1706 		 * and will cause the buffer cache to attempt to free the
1707 		 * underlying pages.
1708 		 */
1709 		if (ioflag & IO_DIRECT)
1710 			bp->b_flags |= B_DIRECT;
1711 
1712 		/*
1713 		 * We should only get non-zero b_resid when an I/O error
1714 		 * has occurred, which should cause us to break above.
1715 		 * However, if the short read did not cause an error,
1716 		 * then we want to ensure that we do not uiomove bad
1717 		 * or uninitialized data.
1718 		 */
1719 		size -= bp->b_resid;
1720 		if (size < xfersize) {
1721 			if (size == 0)
1722 				break;
1723 			xfersize = size;
1724 		}
1725 		error = uiomove((char *)bp->b_data + blkoffset,
1726 			(int)xfersize, uio);
1727 		if (error)
1728 			break;
1729 
1730 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
1731 			/*
1732 			 * If it's VMIO or direct I/O, then we don't
1733 			 * need the buf, mark it available for
1734 			 * freeing. If it's non-direct VMIO, the VM has
1735 			 * the data.
1736 			 */
1737 			bp->b_flags |= B_RELBUF;
1738 			brelse(bp);
1739 		} else {
1740 			/*
1741 			 * Otherwise let whoever
1742 			 * made the request take care of
1743 			 * freeing it. We just queue
1744 			 * it onto another list.
1745 			 */
1746 			bqrelse(bp);
1747 		}
1748 	}
1749 
1750 	/*
1751 	 * This can only happen in the case of an error
1752 	 * because the loop above resets bp to NULL on each iteration
1753 	 * and on normal completion has not set a new value into it.
1754 	 * so it must have come from a 'break' statement
1755 	 */
1756 	if (bp != NULL) {
1757 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
1758 			bp->b_flags |= B_RELBUF;
1759 			brelse(bp);
1760 		} else {
1761 			bqrelse(bp);
1762 		}
1763 	}
1764 
1765 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1766 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
1767 		ip->i_flag |= IN_ACCESS;
1768 	return (error);
1769 }
1770 
1771 static int
1772 ext2_ioctl(struct vop_ioctl_args *ap)
1773 {
1774 
1775 	switch (ap->a_command) {
1776 	case FIOSEEKDATA:
1777 	case FIOSEEKHOLE:
1778 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1779 		    (off_t *)ap->a_data, ap->a_cred));
1780 	default:
1781 		return (ENOTTY);
1782 	}
1783 }
1784 
1785 /*
1786  * this function handles ext4 extents block mapping
1787  */
1788 static int
1789 ext4_ext_read(struct vop_read_args *ap)
1790 {
1791 	static unsigned char zeroes[EXT2_MAX_BLOCK_SIZE];
1792 	struct vnode *vp;
1793 	struct inode *ip;
1794 	struct uio *uio;
1795 	struct m_ext2fs *fs;
1796 	struct buf *bp;
1797 	struct ext4_extent nex, *ep;
1798 	struct ext4_extent_path path;
1799 	daddr_t lbn, newblk;
1800 	off_t bytesinfile;
1801 	int cache_type;
1802 	ssize_t orig_resid;
1803 	int error;
1804 	long size, xfersize, blkoffset;
1805 
1806 	vp = ap->a_vp;
1807 	ip = VTOI(vp);
1808 	uio = ap->a_uio;
1809 	memset(&path, 0, sizeof(path));
1810 
1811 	orig_resid = uio->uio_resid;
1812 	KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__));
1813 	if (orig_resid == 0)
1814 		return (0);
1815 	KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__));
1816 	fs = ip->i_e2fs;
1817 	if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
1818 		return (EOVERFLOW);
1819 
1820 	while (uio->uio_resid > 0) {
1821 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1822 			break;
1823 		lbn = lblkno(fs, uio->uio_offset);
1824 		size = blksize(fs, ip, lbn);
1825 		blkoffset = blkoff(fs, uio->uio_offset);
1826 
1827 		xfersize = fs->e2fs_fsize - blkoffset;
1828 		xfersize = MIN(xfersize, uio->uio_resid);
1829 		xfersize = MIN(xfersize, bytesinfile);
1830 
1831 		/* get block from ext4 extent cache */
1832 		cache_type = ext4_ext_in_cache(ip, lbn, &nex);
1833 		switch (cache_type) {
1834 		case EXT4_EXT_CACHE_NO:
1835 			ext4_ext_find_extent(fs, ip, lbn, &path);
1836 			if (path.ep_is_sparse)
1837 				ep = &path.ep_sparse_ext;
1838 			else
1839 				ep = path.ep_ext;
1840 			if (ep == NULL)
1841 				return (EIO);
1842 
1843 			ext4_ext_put_cache(ip, ep,
1844 			    path.ep_is_sparse ? EXT4_EXT_CACHE_GAP : EXT4_EXT_CACHE_IN);
1845 
1846 			newblk = lbn - ep->e_blk + (ep->e_start_lo |
1847 			    (daddr_t)ep->e_start_hi << 32);
1848 
1849 			if (path.ep_bp != NULL) {
1850 				brelse(path.ep_bp);
1851 				path.ep_bp = NULL;
1852 			}
1853 			break;
1854 
1855 		case EXT4_EXT_CACHE_GAP:
1856 			/* block has not been allocated yet */
1857 			break;
1858 
1859 		case EXT4_EXT_CACHE_IN:
1860 			newblk = lbn - nex.e_blk + (nex.e_start_lo |
1861 			    (daddr_t)nex.e_start_hi << 32);
1862 			break;
1863 
1864 		default:
1865 			panic("%s: invalid cache type", __func__);
1866 		}
1867 
1868 		if (cache_type == EXT4_EXT_CACHE_GAP ||
1869 		    (cache_type == EXT4_EXT_CACHE_NO && path.ep_is_sparse)) {
1870 			if (xfersize > sizeof(zeroes))
1871 				xfersize = sizeof(zeroes);
1872 			error = uiomove(zeroes, xfersize, uio);
1873 			if (error)
1874 				return (error);
1875 		} else {
1876 			error = bread(ip->i_devvp, fsbtodb(fs, newblk), size,
1877 			    NOCRED, &bp);
1878 			if (error) {
1879 				brelse(bp);
1880 				return (error);
1881 			}
1882 
1883 			size -= bp->b_resid;
1884 			if (size < xfersize) {
1885 				if (size == 0) {
1886 					bqrelse(bp);
1887 					break;
1888 				}
1889 				xfersize = size;
1890 			}
1891 			error = uiomove(bp->b_data + blkoffset, xfersize, uio);
1892 			bqrelse(bp);
1893 			if (error)
1894 				return (error);
1895 		}
1896 	}
1897 
1898 	return (0);
1899 }
1900 
1901 /*
1902  * Vnode op for writing.
1903  */
1904 static int
1905 ext2_write(struct vop_write_args *ap)
1906 {
1907 	struct vnode *vp;
1908 	struct uio *uio;
1909 	struct inode *ip;
1910 	struct m_ext2fs *fs;
1911 	struct buf *bp;
1912 	daddr_t lbn;
1913 	off_t osize;
1914 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1915 
1916 	ioflag = ap->a_ioflag;
1917 	uio = ap->a_uio;
1918 	vp = ap->a_vp;
1919 
1920 	seqcount = ioflag >> IO_SEQSHIFT;
1921 	ip = VTOI(vp);
1922 
1923 #ifdef INVARIANTS
1924 	if (uio->uio_rw != UIO_WRITE)
1925 		panic("%s: mode", "ext2_write");
1926 #endif
1927 
1928 	switch (vp->v_type) {
1929 	case VREG:
1930 		if (ioflag & IO_APPEND)
1931 			uio->uio_offset = ip->i_size;
1932 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1933 			return (EPERM);
1934 		/* FALLTHROUGH */
1935 	case VLNK:
1936 		break;
1937 	case VDIR:
1938 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
1939 		if ((ioflag & IO_SYNC) == 0)
1940 		panic("ext2_write: nonsync dir write");
1941 		break;
1942 	default:
1943 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
1944 		    vp->v_type, (intmax_t)uio->uio_offset,
1945 		    (intmax_t)uio->uio_resid);
1946 	}
1947 
1948 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
1949 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
1950 	fs = ip->i_e2fs;
1951 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
1952 		return (EFBIG);
1953 	/*
1954 	 * Maybe this should be above the vnode op call, but so long as
1955 	 * file servers have no limits, I don't think it matters.
1956 	 */
1957 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
1958 		return (EFBIG);
1959 
1960 	resid = uio->uio_resid;
1961 	osize = ip->i_size;
1962 	if (seqcount > BA_SEQMAX)
1963 		flags = BA_SEQMAX << BA_SEQSHIFT;
1964 	else
1965 		flags = seqcount << BA_SEQSHIFT;
1966 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1967 		flags |= IO_SYNC;
1968 
1969 	for (error = 0; uio->uio_resid > 0;) {
1970 		lbn = lblkno(fs, uio->uio_offset);
1971 		blkoffset = blkoff(fs, uio->uio_offset);
1972 		xfersize = fs->e2fs_fsize - blkoffset;
1973 		if (uio->uio_resid < xfersize)
1974 			xfersize = uio->uio_resid;
1975 		if (uio->uio_offset + xfersize > ip->i_size)
1976 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
1977 
1978 		/*
1979 		 * We must perform a read-before-write if the transfer size
1980 		 * does not cover the entire buffer.
1981 		 */
1982 		if (fs->e2fs_bsize > xfersize)
1983 			flags |= BA_CLRBUF;
1984 		else
1985 			flags &= ~BA_CLRBUF;
1986 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
1987 		    ap->a_cred, &bp, flags);
1988 		if (error != 0)
1989 			break;
1990 
1991 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
1992 			bp->b_flags |= B_NOCACHE;
1993 		if (uio->uio_offset + xfersize > ip->i_size)
1994 			ip->i_size = uio->uio_offset + xfersize;
1995 		size = blksize(fs, ip, lbn) - bp->b_resid;
1996 		if (size < xfersize)
1997 			xfersize = size;
1998 
1999 		error =
2000 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2001 		/*
2002 		 * If the buffer is not already filled and we encounter an
2003 		 * error while trying to fill it, we have to clear out any
2004 		 * garbage data from the pages instantiated for the buffer.
2005 		 * If we do not, a failed uiomove() during a write can leave
2006 		 * the prior contents of the pages exposed to a userland mmap.
2007 		 *
2008 		 * Note that we need only clear buffers with a transfer size
2009 		 * equal to the block size because buffers with a shorter
2010 		 * transfer size were cleared above by the call to ext2_balloc()
2011 		 * with the BA_CLRBUF flag set.
2012 		 *
2013 		 * If the source region for uiomove identically mmaps the
2014 		 * buffer, uiomove() performed the NOP copy, and the buffer
2015 		 * content remains valid because the page fault handler
2016 		 * validated the pages.
2017 		 */
2018 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2019 		    fs->e2fs_bsize == xfersize)
2020 			vfs_bio_clrbuf(bp);
2021 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
2022 			bp->b_flags |= B_RELBUF;
2023 		}
2024 
2025 		/*
2026 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2027 		 * if we have a severe page deficiency write the buffer
2028 		 * asynchronously.  Otherwise try to cluster, and if that
2029 		 * doesn't do it then either do an async write (if O_DIRECT),
2030 		 * or a delayed write (if not).
2031 		 */
2032 		if (ioflag & IO_SYNC) {
2033 			(void)bwrite(bp);
2034 		} else if (vm_page_count_severe() ||
2035 		    buf_dirty_count_severe() ||
2036 		    (ioflag & IO_ASYNC)) {
2037 			bp->b_flags |= B_CLUSTEROK;
2038 			bawrite(bp);
2039 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2040 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2041 				bp->b_flags |= B_CLUSTEROK;
2042 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2043 			} else {
2044 				bawrite(bp);
2045 			}
2046 		} else if (ioflag & IO_DIRECT) {
2047 			bp->b_flags |= B_CLUSTEROK;
2048 			bawrite(bp);
2049 		} else {
2050 			bp->b_flags |= B_CLUSTEROK;
2051 			bdwrite(bp);
2052 		}
2053 		if (error || xfersize == 0)
2054 			break;
2055 	}
2056 	/*
2057 	 * If we successfully wrote any data, and we are not the superuser
2058 	 * we clear the setuid and setgid bits as a precaution against
2059 	 * tampering.
2060 	 */
2061 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2062 	    ap->a_cred) {
2063 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2064 			ip->i_mode &= ~(ISUID | ISGID);
2065 	}
2066 	if (error) {
2067 		if (ioflag & IO_UNIT) {
2068 			(void)ext2_truncate(vp, osize,
2069 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2070 			uio->uio_offset -= resid - uio->uio_resid;
2071 			uio->uio_resid = resid;
2072 		}
2073 	}
2074 	if (uio->uio_resid != resid) {
2075 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2076 		if (ioflag & IO_SYNC)
2077 			error = ext2_update(vp, 1);
2078 	}
2079 	return (error);
2080 }
2081