xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision b37f6c9805edb4b89f0a8c2b78f78a3dcfc0647b)
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  * SPDX-License-Identifier: BSD-3-Clause
9  *
10  * Copyright (c) 1989, 1991, 1993, 1994
11  *	The Regents of the University of California.  All rights reserved.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)ffs_vfsops.c	8.8 (Berkeley) 4/18/94
38  * $FreeBSD$
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/namei.h>
44 #include <sys/priv.h>
45 #include <sys/proc.h>
46 #include <sys/kernel.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/endian.h>
53 #include <sys/fcntl.h>
54 #include <sys/malloc.h>
55 #include <sys/stat.h>
56 #include <sys/mutex.h>
57 
58 #include <geom/geom.h>
59 #include <geom/geom_vfs.h>
60 
61 #include <fs/ext2fs/ext2_mount.h>
62 #include <fs/ext2fs/inode.h>
63 
64 #include <fs/ext2fs/fs.h>
65 #include <fs/ext2fs/ext2fs.h>
66 #include <fs/ext2fs/ext2_dinode.h>
67 #include <fs/ext2fs/ext2_extern.h>
68 
69 static int	ext2_flushfiles(struct mount *mp, int flags, struct thread *td);
70 static int	ext2_mountfs(struct vnode *, struct mount *);
71 static int	ext2_reload(struct mount *mp, struct thread *td);
72 static int	ext2_sbupdate(struct ext2mount *, int);
73 static int	ext2_cgupdate(struct ext2mount *, int);
74 static vfs_unmount_t		ext2_unmount;
75 static vfs_root_t		ext2_root;
76 static vfs_statfs_t		ext2_statfs;
77 static vfs_sync_t		ext2_sync;
78 static vfs_vget_t		ext2_vget;
79 static vfs_fhtovp_t		ext2_fhtovp;
80 static vfs_mount_t		ext2_mount;
81 
82 MALLOC_DEFINE(M_EXT2NODE, "ext2_node", "EXT2 vnode private part");
83 static MALLOC_DEFINE(M_EXT2MNT, "ext2_mount", "EXT2 mount structure");
84 
85 static struct vfsops ext2fs_vfsops = {
86 	.vfs_fhtovp =		ext2_fhtovp,
87 	.vfs_mount =		ext2_mount,
88 	.vfs_root =		ext2_root,	/* root inode via vget */
89 	.vfs_statfs =		ext2_statfs,
90 	.vfs_sync =		ext2_sync,
91 	.vfs_unmount =		ext2_unmount,
92 	.vfs_vget =		ext2_vget,
93 };
94 
95 VFS_SET(ext2fs_vfsops, ext2fs, 0);
96 
97 static int	ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev,
98 		    int ronly);
99 static int	compute_sb_data(struct vnode * devvp,
100 		    struct ext2fs * es, struct m_ext2fs * fs);
101 
102 static const char *ext2_opts[] = { "acls", "async", "noatime", "noclusterr",
103     "noclusterw", "noexec", "export", "force", "from", "multilabel",
104     "suiddir", "nosymfollow", "sync", "union", NULL };
105 
106 /*
107  * VFS Operations.
108  *
109  * mount system call
110  */
111 static int
112 ext2_mount(struct mount *mp)
113 {
114 	struct vfsoptlist *opts;
115 	struct vnode *devvp;
116 	struct thread *td;
117 	struct ext2mount *ump = NULL;
118 	struct m_ext2fs *fs;
119 	struct nameidata nd, *ndp = &nd;
120 	accmode_t accmode;
121 	char *path, *fspec;
122 	int error, flags, len;
123 
124 	td = curthread;
125 	opts = mp->mnt_optnew;
126 
127 	if (vfs_filteropt(opts, ext2_opts))
128 		return (EINVAL);
129 
130 	vfs_getopt(opts, "fspath", (void **)&path, NULL);
131 	/* Double-check the length of path.. */
132 	if (strlen(path) >= MAXMNTLEN)
133 		return (ENAMETOOLONG);
134 
135 	fspec = NULL;
136 	error = vfs_getopt(opts, "from", (void **)&fspec, &len);
137 	if (!error && fspec[len - 1] != '\0')
138 		return (EINVAL);
139 
140 	/*
141 	 * If updating, check whether changing from read-only to
142 	 * read/write; if there is no device name, that's all we do.
143 	 */
144 	if (mp->mnt_flag & MNT_UPDATE) {
145 		ump = VFSTOEXT2(mp);
146 		fs = ump->um_e2fs;
147 		error = 0;
148 		if (fs->e2fs_ronly == 0 &&
149 		    vfs_flagopt(opts, "ro", NULL, 0)) {
150 			error = VFS_SYNC(mp, MNT_WAIT);
151 			if (error)
152 				return (error);
153 			flags = WRITECLOSE;
154 			if (mp->mnt_flag & MNT_FORCE)
155 				flags |= FORCECLOSE;
156 			error = ext2_flushfiles(mp, flags, td);
157 			if (error == 0 && fs->e2fs_wasvalid && ext2_cgupdate(ump, MNT_WAIT) == 0) {
158 				fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
159 				ext2_sbupdate(ump, MNT_WAIT);
160 			}
161 			fs->e2fs_ronly = 1;
162 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
163 			g_topology_lock();
164 			g_access(ump->um_cp, 0, -1, 0);
165 			g_topology_unlock();
166 		}
167 		if (!error && (mp->mnt_flag & MNT_RELOAD))
168 			error = ext2_reload(mp, td);
169 		if (error)
170 			return (error);
171 		devvp = ump->um_devvp;
172 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
173 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
174 				return (EPERM);
175 
176 			/*
177 			 * If upgrade to read-write by non-root, then verify
178 			 * that user has necessary permissions on the device.
179 			 */
180 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
181 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
182 			    td->td_ucred, td);
183 			if (error)
184 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
185 			if (error) {
186 				VOP_UNLOCK(devvp, 0);
187 				return (error);
188 			}
189 			VOP_UNLOCK(devvp, 0);
190 			g_topology_lock();
191 			error = g_access(ump->um_cp, 0, 1, 0);
192 			g_topology_unlock();
193 			if (error)
194 				return (error);
195 
196 			if ((fs->e2fs->e2fs_state & E2FS_ISCLEAN) == 0 ||
197 			    (fs->e2fs->e2fs_state & E2FS_ERRORS)) {
198 				if (mp->mnt_flag & MNT_FORCE) {
199 					printf(
200 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
201 				} else {
202 					printf(
203 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
204 					    fs->e2fs_fsmnt);
205 					return (EPERM);
206 				}
207 			}
208 			fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;
209 			(void)ext2_cgupdate(ump, MNT_WAIT);
210 			fs->e2fs_ronly = 0;
211 			MNT_ILOCK(mp);
212 			mp->mnt_flag &= ~MNT_RDONLY;
213 			MNT_IUNLOCK(mp);
214 		}
215 		if (vfs_flagopt(opts, "export", NULL, 0)) {
216 			/* Process export requests in vfs_mount.c. */
217 			return (error);
218 		}
219 	}
220 
221 	/*
222 	 * Not an update, or updating the name: look up the name
223 	 * and verify that it refers to a sensible disk device.
224 	 */
225 	if (fspec == NULL)
226 		return (EINVAL);
227 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
228 	if ((error = namei(ndp)) != 0)
229 		return (error);
230 	NDFREE(ndp, NDF_ONLY_PNBUF);
231 	devvp = ndp->ni_vp;
232 
233 	if (!vn_isdisk(devvp, &error)) {
234 		vput(devvp);
235 		return (error);
236 	}
237 
238 	/*
239 	 * If mount by non-root, then verify that user has necessary
240 	 * permissions on the device.
241 	 *
242 	 * XXXRW: VOP_ACCESS() enough?
243 	 */
244 	accmode = VREAD;
245 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
246 		accmode |= VWRITE;
247 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
248 	if (error)
249 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
250 	if (error) {
251 		vput(devvp);
252 		return (error);
253 	}
254 
255 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
256 		error = ext2_mountfs(devvp, mp);
257 	} else {
258 		if (devvp != ump->um_devvp) {
259 			vput(devvp);
260 			return (EINVAL);	/* needs translation */
261 		} else
262 			vput(devvp);
263 	}
264 	if (error) {
265 		vrele(devvp);
266 		return (error);
267 	}
268 	ump = VFSTOEXT2(mp);
269 	fs = ump->um_e2fs;
270 
271 	/*
272 	 * Note that this strncpy() is ok because of a check at the start
273 	 * of ext2_mount().
274 	 */
275 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
276 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
277 	vfs_mountedfrom(mp, fspec);
278 	return (0);
279 }
280 
281 static int
282 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
283 {
284 	uint32_t i, mask;
285 
286 	if (es->e2fs_magic != E2FS_MAGIC) {
287 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
288 		    devtoname(dev), es->e2fs_magic, E2FS_MAGIC);
289 		return (1);
290 	}
291 	if (es->e2fs_rev > E2FS_REV0) {
292 		mask = es->e2fs_features_incompat & ~(EXT2F_INCOMPAT_SUPP |
293 		    EXT4F_RO_INCOMPAT_SUPP);
294 		if (mask) {
295 			printf("WARNING: mount of %s denied due to "
296 			    "unsupported optional features:\n", devtoname(dev));
297 			for (i = 0;
298 			    i < sizeof(incompat)/sizeof(struct ext2_feature);
299 			    i++)
300 				if (mask & incompat[i].mask)
301 					printf("%s ", incompat[i].name);
302 			printf("\n");
303 			return (1);
304 		}
305 		mask = es->e2fs_features_rocompat & ~EXT2F_ROCOMPAT_SUPP;
306 		if (!ronly && mask) {
307 			printf("WARNING: R/W mount of %s denied due to "
308 			    "unsupported optional features:\n", devtoname(dev));
309 			for (i = 0;
310 			    i < sizeof(ro_compat)/sizeof(struct ext2_feature);
311 			    i++)
312 				if (mask & ro_compat[i].mask)
313 					printf("%s ", ro_compat[i].name);
314 			printf("\n");
315 			return (1);
316 		}
317 	}
318 	return (0);
319 }
320 
321 /*
322  * This computes the fields of the m_ext2fs structure from the
323  * data in the ext2fs structure read in.
324  */
325 static int
326 compute_sb_data(struct vnode *devvp, struct ext2fs *es,
327     struct m_ext2fs *fs)
328 {
329 	int g_count = 0, error;
330 	int i, j;
331 	int logic_sb_block = 1;	/* XXX for now */
332 	struct buf *bp;
333 	uint32_t e2fs_descpb, e2fs_gdbcount_alloc;
334 
335 	fs->e2fs_bcount = es->e2fs_bcount;
336 	fs->e2fs_rbcount = es->e2fs_rbcount;
337 	fs->e2fs_fbcount = es->e2fs_fbcount;
338 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
339 		fs->e2fs_bcount |= (uint64_t)(es->e4fs_bcount_hi) << 32;
340 		fs->e2fs_rbcount |= (uint64_t)(es->e4fs_rbcount_hi) << 32;
341 		fs->e2fs_fbcount |= (uint64_t)(es->e4fs_fbcount_hi) << 32;
342 	}
343 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
344 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
345 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
346 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
347 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
348 	if (fs->e2fs_fsize)
349 		fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
350 	fs->e2fs_bpg = es->e2fs_bpg;
351 	fs->e2fs_fpg = es->e2fs_fpg;
352 	fs->e2fs_ipg = es->e2fs_ipg;
353 	if (es->e2fs_rev == E2FS_REV0) {
354 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
355 	} else {
356 		fs->e2fs_isize = es->e2fs_inode_size;
357 
358 		/*
359 		 * Simple sanity check for superblock inode size value.
360 		 */
361 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
362 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
363 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
364 			printf("ext2fs: invalid inode size %d\n",
365 			    fs->e2fs_isize);
366 			return (EIO);
367 		}
368 	}
369 	/* Check for extra isize in big inodes. */
370 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
371 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
372 		printf("ext2fs: no space for extra inode timestamps\n");
373 		return (EINVAL);
374 	}
375 	/* Check checksum features */
376 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) &&
377 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
378 		printf("ext2fs: incorrect checksum features combination\n");
379 		return (EINVAL);
380 	}
381 	/* Check for group descriptor size */
382 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT) &&
383 	    (es->e3fs_desc_size != sizeof(struct ext2_gd))) {
384 		printf("ext2fs: group descriptor size unsupported %d\n",
385 		    es->e3fs_desc_size);
386 		return (EINVAL);
387 	}
388 
389 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
390 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
391 	/* s_resuid / s_resgid ? */
392 	fs->e2fs_gcount = howmany(fs->e2fs_bcount - es->e2fs_first_dblock,
393 	    EXT2_BLOCKS_PER_GROUP(fs));
394 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
395 		e2fs_descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
396 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount, e2fs_descpb);
397 	} else {
398 		e2fs_descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
399 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount,
400 		    fs->e2fs_bsize / sizeof(struct ext2_gd));
401 	}
402 	fs->e2fs_gdbcount = howmany(fs->e2fs_gcount, e2fs_descpb);
403 	fs->e2fs_gd = mallocarray(e2fs_gdbcount_alloc, fs->e2fs_bsize,
404 	    M_EXT2MNT, M_WAITOK | M_ZERO);
405 	fs->e2fs_contigdirs = mallocarray(fs->e2fs_gcount,
406 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
407 
408 	/*
409 	 * Adjust logic_sb_block.
410 	 * Godmar thinks: if the blocksize is greater than 1024, then
411 	 * the superblock is logically part of block zero.
412 	 */
413 	if (fs->e2fs_bsize > SBSIZE)
414 		logic_sb_block = 0;
415 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
416 		error = bread(devvp,
417 		    fsbtodb(fs, logic_sb_block + i + 1),
418 		    fs->e2fs_bsize, NOCRED, &bp);
419 		if (error) {
420 			free(fs->e2fs_contigdirs, M_EXT2MNT);
421 			free(fs->e2fs_gd, M_EXT2MNT);
422 			brelse(bp);
423 			return (error);
424 		}
425 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
426 			memcpy(&fs->e2fs_gd[
427 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
428 			    bp->b_data, fs->e2fs_bsize);
429 		} else {
430 			for (j = 0; j < e2fs_descpb &&
431 			    g_count < fs->e2fs_gcount; j++, g_count++)
432 				memcpy(&fs->e2fs_gd[g_count],
433 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
434 				    E2FS_REV0_GD_SIZE);
435 		}
436 		brelse(bp);
437 		bp = NULL;
438 	}
439 	/* Precompute checksum seed for all metadata */
440 	ext2_sb_csum_set_seed(fs);
441 	/* Verfy cg csum */
442 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
443 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
444 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
445 		if (error)
446 			return (error);
447 	}
448 	/* Initialization for the ext2 Orlov allocator variant. */
449 	fs->e2fs_total_dir = 0;
450 	for (i = 0; i < fs->e2fs_gcount; i++)
451 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
452 
453 	if (es->e2fs_rev == E2FS_REV0 ||
454 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
455 		fs->e2fs_maxfilesize = 0x7fffffff;
456 	else {
457 		fs->e2fs_maxfilesize = 0xffffffffffff;
458 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
459 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
460 	}
461 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
462 		fs->e2fs_uhash = 3;
463 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
464 #ifdef __CHAR_UNSIGNED__
465 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
466 		fs->e2fs_uhash = 3;
467 #else
468 		es->e4fs_flags |= E2FS_SIGNED_HASH;
469 #endif
470 	}
471 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
472 		error = ext2_sb_csum_verify(fs);
473 
474 	return (error);
475 }
476 
477 /*
478  * Reload all incore data for a filesystem (used after running fsck on
479  * the root filesystem and finding things to fix). The filesystem must
480  * be mounted read-only.
481  *
482  * Things to do to update the mount:
483  *	1) invalidate all cached meta-data.
484  *	2) re-read superblock from disk.
485  *	3) invalidate all cluster summary information.
486  *	4) invalidate all inactive vnodes.
487  *	5) invalidate all cached file data.
488  *	6) re-read inode data for all active vnodes.
489  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
490  */
491 static int
492 ext2_reload(struct mount *mp, struct thread *td)
493 {
494 	struct vnode *vp, *mvp, *devvp;
495 	struct inode *ip;
496 	struct buf *bp;
497 	struct ext2fs *es;
498 	struct m_ext2fs *fs;
499 	struct csum *sump;
500 	int error, i;
501 	int32_t *lp;
502 
503 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
504 		return (EINVAL);
505 	/*
506 	 * Step 1: invalidate all cached meta-data.
507 	 */
508 	devvp = VFSTOEXT2(mp)->um_devvp;
509 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
510 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
511 		panic("ext2_reload: dirty1");
512 	VOP_UNLOCK(devvp, 0);
513 
514 	/*
515 	 * Step 2: re-read superblock from disk.
516 	 * constants have been adjusted for ext2
517 	 */
518 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
519 		return (error);
520 	es = (struct ext2fs *)bp->b_data;
521 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
522 		brelse(bp);
523 		return (EIO);		/* XXX needs translation */
524 	}
525 	fs = VFSTOEXT2(mp)->um_e2fs;
526 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
527 
528 	if ((error = compute_sb_data(devvp, es, fs)) != 0) {
529 		brelse(bp);
530 		return (error);
531 	}
532 #ifdef UNKLAR
533 	if (fs->fs_sbsize < SBSIZE)
534 		bp->b_flags |= B_INVAL;
535 #endif
536 	brelse(bp);
537 
538 	/*
539 	 * Step 3: invalidate all cluster summary information.
540 	 */
541 	if (fs->e2fs_contigsumsize > 0) {
542 		lp = fs->e2fs_maxcluster;
543 		sump = fs->e2fs_clustersum;
544 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
545 			*lp++ = fs->e2fs_contigsumsize;
546 			sump->cs_init = 0;
547 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
548 		}
549 	}
550 
551 loop:
552 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
553 		/*
554 		 * Step 4: invalidate all cached file data.
555 		 */
556 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
557 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
558 			goto loop;
559 		}
560 		if (vinvalbuf(vp, 0, 0, 0))
561 			panic("ext2_reload: dirty2");
562 
563 		/*
564 		 * Step 5: re-read inode data for all active vnodes.
565 		 */
566 		ip = VTOI(vp);
567 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
568 		    (int)fs->e2fs_bsize, NOCRED, &bp);
569 		if (error) {
570 			VOP_UNLOCK(vp, 0);
571 			vrele(vp);
572 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
573 			return (error);
574 		}
575 		ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
576 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
577 		brelse(bp);
578 		VOP_UNLOCK(vp, 0);
579 		vrele(vp);
580 	}
581 	return (0);
582 }
583 
584 /*
585  * Common code for mount and mountroot.
586  */
587 static int
588 ext2_mountfs(struct vnode *devvp, struct mount *mp)
589 {
590 	struct ext2mount *ump;
591 	struct buf *bp;
592 	struct m_ext2fs *fs;
593 	struct ext2fs *es;
594 	struct cdev *dev = devvp->v_rdev;
595 	struct g_consumer *cp;
596 	struct bufobj *bo;
597 	struct csum *sump;
598 	int error;
599 	int ronly;
600 	int i;
601 	u_long size;
602 	int32_t *lp;
603 	int32_t e2fs_maxcontig;
604 
605 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
606 	/* XXX: use VOP_ACESS to check FS perms */
607 	g_topology_lock();
608 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
609 	g_topology_unlock();
610 	VOP_UNLOCK(devvp, 0);
611 	if (error)
612 		return (error);
613 
614 	/* XXX: should we check for some sectorsize or 512 instead? */
615 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
616 	    (SBSIZE < cp->provider->sectorsize)) {
617 		g_topology_lock();
618 		g_vfs_close(cp);
619 		g_topology_unlock();
620 		return (EINVAL);
621 	}
622 
623 	bo = &devvp->v_bufobj;
624 	bo->bo_private = cp;
625 	bo->bo_ops = g_vfs_bufops;
626 	if (devvp->v_rdev->si_iosize_max != 0)
627 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
628 	if (mp->mnt_iosize_max > MAXPHYS)
629 		mp->mnt_iosize_max = MAXPHYS;
630 
631 	bp = NULL;
632 	ump = NULL;
633 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
634 		goto out;
635 	es = (struct ext2fs *)bp->b_data;
636 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
637 		error = EINVAL;		/* XXX needs translation */
638 		goto out;
639 	}
640 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
641 	    (es->e2fs_state & E2FS_ERRORS)) {
642 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
643 			printf(
644 "WARNING: Filesystem was not properly dismounted\n");
645 		} else {
646 			printf(
647 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
648 			error = EPERM;
649 			goto out;
650 		}
651 	}
652 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
653 
654 	/*
655 	 * I don't know whether this is the right strategy. Note that
656 	 * we dynamically allocate both an m_ext2fs and an ext2fs
657 	 * while Linux keeps the super block in a locked buffer.
658 	 */
659 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
660 	    M_EXT2MNT, M_WAITOK | M_ZERO);
661 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
662 	    M_EXT2MNT, M_WAITOK);
663 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
664 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
665 	if ((error = compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
666 		goto out;
667 
668 	/*
669 	 * Calculate the maximum contiguous blocks and size of cluster summary
670 	 * array.  In FFS this is done by newfs; however, the superblock
671 	 * in ext2fs doesn't have these variables, so we can calculate
672 	 * them here.
673 	 */
674 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
675 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
676 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
677 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
678 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
679 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
680 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
681 		lp = ump->um_e2fs->e2fs_maxcluster;
682 		sump = ump->um_e2fs->e2fs_clustersum;
683 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
684 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
685 			sump->cs_init = 0;
686 			sump->cs_sum = mallocarray(
687 			    ump->um_e2fs->e2fs_contigsumsize + 1,
688 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
689 		}
690 	}
691 
692 	brelse(bp);
693 	bp = NULL;
694 	fs = ump->um_e2fs;
695 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
696 
697 	/*
698 	 * If the fs is not mounted read-only, make sure the super block is
699 	 * always written back on a sync().
700 	 */
701 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
702 	if (ronly == 0) {
703 		fs->e2fs_fmod = 1;	/* mark it modified */
704 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
705 	}
706 	mp->mnt_data = ump;
707 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
708 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
709 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
710 	MNT_ILOCK(mp);
711 	mp->mnt_flag |= MNT_LOCAL;
712 	MNT_IUNLOCK(mp);
713 	ump->um_mountp = mp;
714 	ump->um_dev = dev;
715 	ump->um_devvp = devvp;
716 	ump->um_bo = &devvp->v_bufobj;
717 	ump->um_cp = cp;
718 
719 	/*
720 	 * Setting those two parameters allowed us to use
721 	 * ufs_bmap w/o changse!
722 	 */
723 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
724 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
725 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
726 	if (ronly == 0)
727 		ext2_sbupdate(ump, MNT_WAIT);
728 	/*
729 	 * Initialize filesystem stat information in mount struct.
730 	 */
731 	MNT_ILOCK(mp);
732 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
733 	    MNTK_USES_BCACHE;
734 	MNT_IUNLOCK(mp);
735 	return (0);
736 out:
737 	if (bp)
738 		brelse(bp);
739 	if (cp != NULL) {
740 		g_topology_lock();
741 		g_vfs_close(cp);
742 		g_topology_unlock();
743 	}
744 	if (ump) {
745 		mtx_destroy(EXT2_MTX(ump));
746 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
747 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
748 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
749 		free(ump->um_e2fs, M_EXT2MNT);
750 		free(ump, M_EXT2MNT);
751 		mp->mnt_data = NULL;
752 	}
753 	return (error);
754 }
755 
756 /*
757  * Unmount system call.
758  */
759 static int
760 ext2_unmount(struct mount *mp, int mntflags)
761 {
762 	struct ext2mount *ump;
763 	struct m_ext2fs *fs;
764 	struct csum *sump;
765 	int error, flags, i, ronly;
766 
767 	flags = 0;
768 	if (mntflags & MNT_FORCE) {
769 		if (mp->mnt_flag & MNT_ROOTFS)
770 			return (EINVAL);
771 		flags |= FORCECLOSE;
772 	}
773 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
774 		return (error);
775 	ump = VFSTOEXT2(mp);
776 	fs = ump->um_e2fs;
777 	ronly = fs->e2fs_ronly;
778 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
779 		if (fs->e2fs_wasvalid)
780 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
781 		ext2_sbupdate(ump, MNT_WAIT);
782 	}
783 
784 	g_topology_lock();
785 	g_vfs_close(ump->um_cp);
786 	g_topology_unlock();
787 	vrele(ump->um_devvp);
788 	sump = fs->e2fs_clustersum;
789 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
790 		free(sump->cs_sum, M_EXT2MNT);
791 	free(fs->e2fs_clustersum, M_EXT2MNT);
792 	free(fs->e2fs_maxcluster, M_EXT2MNT);
793 	free(fs->e2fs_gd, M_EXT2MNT);
794 	free(fs->e2fs_contigdirs, M_EXT2MNT);
795 	free(fs->e2fs, M_EXT2MNT);
796 	free(fs, M_EXT2MNT);
797 	free(ump, M_EXT2MNT);
798 	mp->mnt_data = NULL;
799 	MNT_ILOCK(mp);
800 	mp->mnt_flag &= ~MNT_LOCAL;
801 	MNT_IUNLOCK(mp);
802 	return (error);
803 }
804 
805 /*
806  * Flush out all the files in a filesystem.
807  */
808 static int
809 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
810 {
811 	int error;
812 
813 	error = vflush(mp, 0, flags, td);
814 	return (error);
815 }
816 
817 /*
818  * Get filesystem statistics.
819  */
820 int
821 ext2_statfs(struct mount *mp, struct statfs *sbp)
822 {
823 	struct ext2mount *ump;
824 	struct m_ext2fs *fs;
825 	uint32_t overhead, overhead_per_group, ngdb;
826 	int i, ngroups;
827 
828 	ump = VFSTOEXT2(mp);
829 	fs = ump->um_e2fs;
830 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
831 		panic("ext2_statfs");
832 
833 	/*
834 	 * Compute the overhead (FS structures)
835 	 */
836 	overhead_per_group =
837 	    1 /* block bitmap */ +
838 	    1 /* inode bitmap */ +
839 	    fs->e2fs_itpg;
840 	overhead = fs->e2fs->e2fs_first_dblock +
841 	    fs->e2fs_gcount * overhead_per_group;
842 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
843 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
844 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
845 			if (ext2_cg_has_sb(fs, i))
846 				ngroups++;
847 		}
848 	} else {
849 		ngroups = fs->e2fs_gcount;
850 	}
851 	ngdb = fs->e2fs_gdbcount;
852 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
853 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
854 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
855 	overhead += ngroups * (1 /* superblock */ + ngdb);
856 
857 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
858 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
859 	sbp->f_blocks = fs->e2fs_bcount - overhead;
860 	sbp->f_bfree = fs->e2fs_fbcount;
861 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
862 	sbp->f_files = fs->e2fs->e2fs_icount;
863 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
864 	return (0);
865 }
866 
867 /*
868  * Go through the disk queues to initiate sandbagged IO;
869  * go through the inodes to write those that have been modified;
870  * initiate the writing of the super block if it has been modified.
871  *
872  * Note: we are always called with the filesystem marked `MPBUSY'.
873  */
874 static int
875 ext2_sync(struct mount *mp, int waitfor)
876 {
877 	struct vnode *mvp, *vp;
878 	struct thread *td;
879 	struct inode *ip;
880 	struct ext2mount *ump = VFSTOEXT2(mp);
881 	struct m_ext2fs *fs;
882 	int error, allerror = 0;
883 
884 	td = curthread;
885 	fs = ump->um_e2fs;
886 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
887 		printf("fs = %s\n", fs->e2fs_fsmnt);
888 		panic("ext2_sync: rofs mod");
889 	}
890 
891 	/*
892 	 * Write back each (modified) inode.
893 	 */
894 loop:
895 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
896 		if (vp->v_type == VNON) {
897 			VI_UNLOCK(vp);
898 			continue;
899 		}
900 		ip = VTOI(vp);
901 		if ((ip->i_flag &
902 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
903 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
904 		    waitfor == MNT_LAZY)) {
905 			VI_UNLOCK(vp);
906 			continue;
907 		}
908 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
909 		if (error) {
910 			if (error == ENOENT) {
911 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
912 				goto loop;
913 			}
914 			continue;
915 		}
916 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
917 			allerror = error;
918 		VOP_UNLOCK(vp, 0);
919 		vrele(vp);
920 	}
921 
922 	/*
923 	 * Force stale filesystem control information to be flushed.
924 	 */
925 	if (waitfor != MNT_LAZY) {
926 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
927 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
928 			allerror = error;
929 		VOP_UNLOCK(ump->um_devvp, 0);
930 	}
931 
932 	/*
933 	 * Write back modified superblock.
934 	 */
935 	if (fs->e2fs_fmod != 0) {
936 		fs->e2fs_fmod = 0;
937 		fs->e2fs->e2fs_wtime = time_second;
938 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
939 			allerror = error;
940 	}
941 	return (allerror);
942 }
943 
944 /*
945  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
946  * in from disk.  If it is in core, wait for the lock bit to clear, then
947  * return the inode locked.  Detection and handling of mount points must be
948  * done by the calling routine.
949  */
950 static int
951 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
952 {
953 	struct m_ext2fs *fs;
954 	struct inode *ip;
955 	struct ext2mount *ump;
956 	struct buf *bp;
957 	struct vnode *vp;
958 	struct thread *td;
959 	int i, error;
960 	int used_blocks;
961 
962 	td = curthread;
963 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
964 	if (error || *vpp != NULL)
965 		return (error);
966 
967 	ump = VFSTOEXT2(mp);
968 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
969 
970 	/* Allocate a new vnode/inode. */
971 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
972 		*vpp = NULL;
973 		free(ip, M_EXT2NODE);
974 		return (error);
975 	}
976 	vp->v_data = ip;
977 	ip->i_vnode = vp;
978 	ip->i_e2fs = fs = ump->um_e2fs;
979 	ip->i_ump = ump;
980 	ip->i_number = ino;
981 
982 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
983 	error = insmntque(vp, mp);
984 	if (error != 0) {
985 		free(ip, M_EXT2NODE);
986 		*vpp = NULL;
987 		return (error);
988 	}
989 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
990 	if (error || *vpp != NULL)
991 		return (error);
992 
993 	/* Read in the disk contents for the inode, copy into the inode. */
994 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
995 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
996 		/*
997 		 * The inode does not contain anything useful, so it would
998 		 * be misleading to leave it on its hash chain. With mode
999 		 * still zero, it will be unlinked and returned to the free
1000 		 * list by vput().
1001 		 */
1002 		brelse(bp);
1003 		vput(vp);
1004 		*vpp = NULL;
1005 		return (error);
1006 	}
1007 	/* convert ext2 inode to dinode */
1008 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1009 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1010 	if (error) {
1011 		printf("ext2fs: Bad inode %lu csum - run fsck\n",
1012 		    (unsigned long)ino);
1013 		brelse(bp);
1014 		vput(vp);
1015 		*vpp = NULL;
1016 		return (error);
1017 	}
1018 	ip->i_block_group = ino_to_cg(fs, ino);
1019 	ip->i_next_alloc_block = 0;
1020 	ip->i_next_alloc_goal = 0;
1021 
1022 	/*
1023 	 * Now we want to make sure that block pointers for unused
1024 	 * blocks are zeroed out - ext2_balloc depends on this
1025 	 * although for regular files and directories only
1026 	 *
1027 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1028 	 * out because we could corrupt the extent tree.
1029 	 */
1030 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1031 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1032 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1033 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1034 			ip->i_db[i] = 0;
1035 	}
1036 #ifdef EXT2FS_DEBUG
1037 	ext2_print_inode(ip);
1038 	ext4_ext_print_extent_tree_status(ip);
1039 #endif
1040 	bqrelse(bp);
1041 
1042 	/*
1043 	 * Initialize the vnode from the inode, check for aliases.
1044 	 * Note that the underlying vnode may have changed.
1045 	 */
1046 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1047 		vput(vp);
1048 		*vpp = NULL;
1049 		return (error);
1050 	}
1051 
1052 	/*
1053 	 * Finish inode initialization.
1054 	 */
1055 
1056 	*vpp = vp;
1057 	return (0);
1058 }
1059 
1060 /*
1061  * File handle to vnode
1062  *
1063  * Have to be really careful about stale file handles:
1064  * - check that the inode number is valid
1065  * - call ext2_vget() to get the locked inode
1066  * - check for an unallocated inode (i_mode == 0)
1067  * - check that the given client host has export rights and return
1068  *   those rights via. exflagsp and credanonp
1069  */
1070 static int
1071 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1072 {
1073 	struct inode *ip;
1074 	struct ufid *ufhp;
1075 	struct vnode *nvp;
1076 	struct m_ext2fs *fs;
1077 	int error;
1078 
1079 	ufhp = (struct ufid *)fhp;
1080 	fs = VFSTOEXT2(mp)->um_e2fs;
1081 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1082 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1083 		return (ESTALE);
1084 
1085 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1086 	if (error) {
1087 		*vpp = NULLVP;
1088 		return (error);
1089 	}
1090 	ip = VTOI(nvp);
1091 	if (ip->i_mode == 0 ||
1092 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1093 		vput(nvp);
1094 		*vpp = NULLVP;
1095 		return (ESTALE);
1096 	}
1097 	*vpp = nvp;
1098 	vnode_create_vobject(*vpp, 0, curthread);
1099 	return (0);
1100 }
1101 
1102 /*
1103  * Write a superblock and associated information back to disk.
1104  */
1105 static int
1106 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1107 {
1108 	struct m_ext2fs *fs = mp->um_e2fs;
1109 	struct ext2fs *es = fs->e2fs;
1110 	struct buf *bp;
1111 	int error = 0;
1112 
1113 	es->e2fs_bcount = fs->e2fs_bcount & 0xffffffff;
1114 	es->e2fs_rbcount = fs->e2fs_rbcount & 0xffffffff;
1115 	es->e2fs_fbcount = fs->e2fs_fbcount & 0xffffffff;
1116 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1117 		es->e4fs_bcount_hi = fs->e2fs_bcount >> 32;
1118 		es->e4fs_rbcount_hi = fs->e2fs_rbcount >> 32;
1119 		es->e4fs_fbcount_hi = fs->e2fs_fbcount >> 32;
1120 	}
1121 
1122 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1123 		ext2_sb_csum_set(fs);
1124 
1125 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1126 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1127 	if (waitfor == MNT_WAIT)
1128 		error = bwrite(bp);
1129 	else
1130 		bawrite(bp);
1131 
1132 	/*
1133 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1134 	 * are not busy at this point and are (hopefully) written by the
1135 	 * usual sync mechanism. No need to write them here.
1136 	 */
1137 	return (error);
1138 }
1139 int
1140 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1141 {
1142 	struct m_ext2fs *fs = mp->um_e2fs;
1143 	struct buf *bp;
1144 	int i, j, g_count = 0, error = 0, allerror = 0;
1145 
1146 	allerror = ext2_sbupdate(mp, waitfor);
1147 
1148 	/* Update gd csums */
1149 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1150 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1151 		ext2_gd_csum_set(fs);
1152 
1153 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1154 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1155 		    fs->e2fs->e2fs_first_dblock +
1156 		    1 /* superblock */ + i), fs->e2fs_bsize, 0, 0, 0);
1157 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1158 			memcpy(bp->b_data, &fs->e2fs_gd[
1159 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1160 			    fs->e2fs_bsize);
1161 		} else {
1162 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1163 			    g_count < fs->e2fs_gcount; j++, g_count++)
1164 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1165 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1166 		}
1167 		if (waitfor == MNT_WAIT)
1168 			error = bwrite(bp);
1169 		else
1170 			bawrite(bp);
1171 	}
1172 
1173 	if (!allerror && error)
1174 		allerror = error;
1175 	return (allerror);
1176 }
1177 
1178 /*
1179  * Return the root of a filesystem.
1180  */
1181 static int
1182 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1183 {
1184 	struct vnode *nvp;
1185 	int error;
1186 
1187 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1188 	if (error)
1189 		return (error);
1190 	*vpp = nvp;
1191 	return (0);
1192 }
1193