xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision 68e709cb296e23b3c827e705d6328cd2261a1d71)
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 &&
158 			    ext2_cgupdate(ump, MNT_WAIT) == 0) {
159 				fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
160 				ext2_sbupdate(ump, MNT_WAIT);
161 			}
162 			fs->e2fs_ronly = 1;
163 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
164 			g_topology_lock();
165 			g_access(ump->um_cp, 0, -1, 0);
166 			g_topology_unlock();
167 		}
168 		if (!error && (mp->mnt_flag & MNT_RELOAD))
169 			error = ext2_reload(mp, td);
170 		if (error)
171 			return (error);
172 		devvp = ump->um_devvp;
173 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
174 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
175 				return (EPERM);
176 
177 			/*
178 			 * If upgrade to read-write by non-root, then verify
179 			 * that user has necessary permissions on the device.
180 			 */
181 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
182 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
183 			    td->td_ucred, td);
184 			if (error)
185 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
186 			if (error) {
187 				VOP_UNLOCK(devvp, 0);
188 				return (error);
189 			}
190 			VOP_UNLOCK(devvp, 0);
191 			g_topology_lock();
192 			error = g_access(ump->um_cp, 0, 1, 0);
193 			g_topology_unlock();
194 			if (error)
195 				return (error);
196 
197 			if ((fs->e2fs->e2fs_state & E2FS_ISCLEAN) == 0 ||
198 			    (fs->e2fs->e2fs_state & E2FS_ERRORS)) {
199 				if (mp->mnt_flag & MNT_FORCE) {
200 					printf(
201 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
202 				} else {
203 					printf(
204 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
205 					    fs->e2fs_fsmnt);
206 					return (EPERM);
207 				}
208 			}
209 			fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;
210 			(void)ext2_cgupdate(ump, MNT_WAIT);
211 			fs->e2fs_ronly = 0;
212 			MNT_ILOCK(mp);
213 			mp->mnt_flag &= ~MNT_RDONLY;
214 			MNT_IUNLOCK(mp);
215 		}
216 		if (vfs_flagopt(opts, "export", NULL, 0)) {
217 			/* Process export requests in vfs_mount.c. */
218 			return (error);
219 		}
220 	}
221 
222 	/*
223 	 * Not an update, or updating the name: look up the name
224 	 * and verify that it refers to a sensible disk device.
225 	 */
226 	if (fspec == NULL)
227 		return (EINVAL);
228 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
229 	if ((error = namei(ndp)) != 0)
230 		return (error);
231 	NDFREE(ndp, NDF_ONLY_PNBUF);
232 	devvp = ndp->ni_vp;
233 
234 	if (!vn_isdisk(devvp, &error)) {
235 		vput(devvp);
236 		return (error);
237 	}
238 
239 	/*
240 	 * If mount by non-root, then verify that user has necessary
241 	 * permissions on the device.
242 	 *
243 	 * XXXRW: VOP_ACCESS() enough?
244 	 */
245 	accmode = VREAD;
246 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
247 		accmode |= VWRITE;
248 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
249 	if (error)
250 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
251 	if (error) {
252 		vput(devvp);
253 		return (error);
254 	}
255 
256 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
257 		error = ext2_mountfs(devvp, mp);
258 	} else {
259 		if (devvp != ump->um_devvp) {
260 			vput(devvp);
261 			return (EINVAL);	/* needs translation */
262 		} else
263 			vput(devvp);
264 	}
265 	if (error) {
266 		vrele(devvp);
267 		return (error);
268 	}
269 	ump = VFSTOEXT2(mp);
270 	fs = ump->um_e2fs;
271 
272 	/*
273 	 * Note that this strncpy() is ok because of a check at the start
274 	 * of ext2_mount().
275 	 */
276 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
277 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
278 	vfs_mountedfrom(mp, fspec);
279 	return (0);
280 }
281 
282 static int
283 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
284 {
285 	uint32_t i, mask;
286 
287 	if (es->e2fs_magic != E2FS_MAGIC) {
288 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
289 		    devtoname(dev), es->e2fs_magic, E2FS_MAGIC);
290 		return (1);
291 	}
292 	if (es->e2fs_rev > E2FS_REV0) {
293 		mask = es->e2fs_features_incompat & ~(EXT2F_INCOMPAT_SUPP |
294 		    EXT4F_RO_INCOMPAT_SUPP);
295 		if (mask) {
296 			printf("WARNING: mount of %s denied due to "
297 			    "unsupported optional features:\n", devtoname(dev));
298 			for (i = 0;
299 			    i < sizeof(incompat)/sizeof(struct ext2_feature);
300 			    i++)
301 				if (mask & incompat[i].mask)
302 					printf("%s ", incompat[i].name);
303 			printf("\n");
304 			return (1);
305 		}
306 		mask = es->e2fs_features_rocompat & ~EXT2F_ROCOMPAT_SUPP;
307 		if (!ronly && mask) {
308 			printf("WARNING: R/W mount of %s denied due to "
309 			    "unsupported optional features:\n", devtoname(dev));
310 			for (i = 0;
311 			    i < sizeof(ro_compat)/sizeof(struct ext2_feature);
312 			    i++)
313 				if (mask & ro_compat[i].mask)
314 					printf("%s ", ro_compat[i].name);
315 			printf("\n");
316 			return (1);
317 		}
318 	}
319 	return (0);
320 }
321 
322 static e4fs_daddr_t
323 cg_location(struct m_ext2fs *fs, int number)
324 {
325 	int cg, descpb, logical_sb, has_super = 0;
326 
327 	/*
328 	 * Adjust logical superblock block number.
329 	 * Godmar thinks: if the blocksize is greater than 1024, then
330 	 * the superblock is logically part of block zero.
331 	 */
332 	logical_sb = fs->e2fs_bsize > SBSIZE ? 0 : 1;
333 
334 	if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
335 	    number < fs->e2fs->e3fs_first_meta_bg)
336 		return (logical_sb + number + 1);
337 
338 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT))
339 		descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
340 	else
341 		descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
342 
343 	cg = descpb * number;
344 
345 	if (ext2_cg_has_sb(fs, cg))
346 		has_super = 1;
347 
348 	return (has_super + cg * (e4fs_daddr_t)EXT2_BLOCKS_PER_GROUP(fs) +
349 	    fs->e2fs->e2fs_first_dblock);
350 }
351 
352 /*
353  * This computes the fields of the m_ext2fs structure from the
354  * data in the ext2fs structure read in.
355  */
356 static int
357 compute_sb_data(struct vnode *devvp, struct ext2fs *es,
358     struct m_ext2fs *fs)
359 {
360 	int g_count = 0, error;
361 	int i, j;
362 	struct buf *bp;
363 	uint32_t e2fs_descpb, e2fs_gdbcount_alloc;
364 
365 	fs->e2fs_bcount = es->e2fs_bcount;
366 	fs->e2fs_rbcount = es->e2fs_rbcount;
367 	fs->e2fs_fbcount = es->e2fs_fbcount;
368 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
369 		fs->e2fs_bcount |= (uint64_t)(es->e4fs_bcount_hi) << 32;
370 		fs->e2fs_rbcount |= (uint64_t)(es->e4fs_rbcount_hi) << 32;
371 		fs->e2fs_fbcount |= (uint64_t)(es->e4fs_fbcount_hi) << 32;
372 	}
373 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
374 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
375 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
376 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
377 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
378 	if (fs->e2fs_fsize)
379 		fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
380 	fs->e2fs_bpg = es->e2fs_bpg;
381 	fs->e2fs_fpg = es->e2fs_fpg;
382 	fs->e2fs_ipg = es->e2fs_ipg;
383 	if (es->e2fs_rev == E2FS_REV0) {
384 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
385 	} else {
386 		fs->e2fs_isize = es->e2fs_inode_size;
387 
388 		/*
389 		 * Simple sanity check for superblock inode size value.
390 		 */
391 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
392 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
393 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
394 			printf("ext2fs: invalid inode size %d\n",
395 			    fs->e2fs_isize);
396 			return (EIO);
397 		}
398 	}
399 	/* Check for extra isize in big inodes. */
400 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
401 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
402 		printf("ext2fs: no space for extra inode timestamps\n");
403 		return (EINVAL);
404 	}
405 	/* Check checksum features */
406 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) &&
407 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
408 		printf("ext2fs: incorrect checksum features combination\n");
409 		return (EINVAL);
410 	}
411 	/* Check for group descriptor size */
412 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT) &&
413 	    (es->e3fs_desc_size != sizeof(struct ext2_gd))) {
414 		printf("ext2fs: group descriptor size unsupported %d\n",
415 		    es->e3fs_desc_size);
416 		return (EINVAL);
417 	}
418 	/* Check for group size */
419 	if (fs->e2fs_bpg != fs->e2fs_bsize * 8) {
420 		printf("ext2fs: non-standard group size unsupported %d\n",
421 		    fs->e2fs_bpg);
422 		return (EINVAL);
423 	}
424 
425 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
426 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
427 	/* s_resuid / s_resgid ? */
428 	fs->e2fs_gcount = howmany(fs->e2fs_bcount - es->e2fs_first_dblock,
429 	    EXT2_BLOCKS_PER_GROUP(fs));
430 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
431 		e2fs_descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
432 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount, e2fs_descpb);
433 	} else {
434 		e2fs_descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
435 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount,
436 		    fs->e2fs_bsize / sizeof(struct ext2_gd));
437 	}
438 	fs->e2fs_gdbcount = howmany(fs->e2fs_gcount, e2fs_descpb);
439 	fs->e2fs_gd = malloc(e2fs_gdbcount_alloc * fs->e2fs_bsize,
440 	    M_EXT2MNT, M_WAITOK | M_ZERO);
441 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
442 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
443 
444 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
445 		error = bread(devvp,
446 		    fsbtodb(fs, cg_location(fs, i)),
447 		    fs->e2fs_bsize, NOCRED, &bp);
448 		if (error) {
449 			free(fs->e2fs_contigdirs, M_EXT2MNT);
450 			free(fs->e2fs_gd, M_EXT2MNT);
451 			brelse(bp);
452 			return (error);
453 		}
454 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
455 			memcpy(&fs->e2fs_gd[
456 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
457 			    bp->b_data, fs->e2fs_bsize);
458 		} else {
459 			for (j = 0; j < e2fs_descpb &&
460 			    g_count < fs->e2fs_gcount; j++, g_count++)
461 				memcpy(&fs->e2fs_gd[g_count],
462 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
463 				    E2FS_REV0_GD_SIZE);
464 		}
465 		brelse(bp);
466 		bp = NULL;
467 	}
468 	/* Precompute checksum seed for all metadata */
469 	ext2_sb_csum_set_seed(fs);
470 	/* Verfy cg csum */
471 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
472 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
473 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
474 		if (error)
475 			return (error);
476 	}
477 	/* Initialization for the ext2 Orlov allocator variant. */
478 	fs->e2fs_total_dir = 0;
479 	for (i = 0; i < fs->e2fs_gcount; i++)
480 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
481 
482 	if (es->e2fs_rev == E2FS_REV0 ||
483 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
484 		fs->e2fs_maxfilesize = 0x7fffffff;
485 	else {
486 		fs->e2fs_maxfilesize = 0xffffffffffff;
487 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
488 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
489 	}
490 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
491 		fs->e2fs_uhash = 3;
492 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
493 #ifdef __CHAR_UNSIGNED__
494 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
495 		fs->e2fs_uhash = 3;
496 #else
497 		es->e4fs_flags |= E2FS_SIGNED_HASH;
498 #endif
499 	}
500 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
501 		error = ext2_sb_csum_verify(fs);
502 
503 	return (error);
504 }
505 
506 /*
507  * Reload all incore data for a filesystem (used after running fsck on
508  * the root filesystem and finding things to fix). The filesystem must
509  * be mounted read-only.
510  *
511  * Things to do to update the mount:
512  *	1) invalidate all cached meta-data.
513  *	2) re-read superblock from disk.
514  *	3) invalidate all cluster summary information.
515  *	4) invalidate all inactive vnodes.
516  *	5) invalidate all cached file data.
517  *	6) re-read inode data for all active vnodes.
518  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
519  */
520 static int
521 ext2_reload(struct mount *mp, struct thread *td)
522 {
523 	struct vnode *vp, *mvp, *devvp;
524 	struct inode *ip;
525 	struct buf *bp;
526 	struct ext2fs *es;
527 	struct m_ext2fs *fs;
528 	struct csum *sump;
529 	int error, i;
530 	int32_t *lp;
531 
532 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
533 		return (EINVAL);
534 	/*
535 	 * Step 1: invalidate all cached meta-data.
536 	 */
537 	devvp = VFSTOEXT2(mp)->um_devvp;
538 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
539 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
540 		panic("ext2_reload: dirty1");
541 	VOP_UNLOCK(devvp, 0);
542 
543 	/*
544 	 * Step 2: re-read superblock from disk.
545 	 * constants have been adjusted for ext2
546 	 */
547 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
548 		return (error);
549 	es = (struct ext2fs *)bp->b_data;
550 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
551 		brelse(bp);
552 		return (EIO);		/* XXX needs translation */
553 	}
554 	fs = VFSTOEXT2(mp)->um_e2fs;
555 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
556 
557 	if ((error = compute_sb_data(devvp, es, fs)) != 0) {
558 		brelse(bp);
559 		return (error);
560 	}
561 #ifdef UNKLAR
562 	if (fs->fs_sbsize < SBSIZE)
563 		bp->b_flags |= B_INVAL;
564 #endif
565 	brelse(bp);
566 
567 	/*
568 	 * Step 3: invalidate all cluster summary information.
569 	 */
570 	if (fs->e2fs_contigsumsize > 0) {
571 		lp = fs->e2fs_maxcluster;
572 		sump = fs->e2fs_clustersum;
573 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
574 			*lp++ = fs->e2fs_contigsumsize;
575 			sump->cs_init = 0;
576 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
577 		}
578 	}
579 
580 loop:
581 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
582 		/*
583 		 * Step 4: invalidate all cached file data.
584 		 */
585 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
586 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
587 			goto loop;
588 		}
589 		if (vinvalbuf(vp, 0, 0, 0))
590 			panic("ext2_reload: dirty2");
591 
592 		/*
593 		 * Step 5: re-read inode data for all active vnodes.
594 		 */
595 		ip = VTOI(vp);
596 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
597 		    (int)fs->e2fs_bsize, NOCRED, &bp);
598 		if (error) {
599 			VOP_UNLOCK(vp, 0);
600 			vrele(vp);
601 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
602 			return (error);
603 		}
604 		ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
605 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
606 		brelse(bp);
607 		VOP_UNLOCK(vp, 0);
608 		vrele(vp);
609 	}
610 	return (0);
611 }
612 
613 /*
614  * Common code for mount and mountroot.
615  */
616 static int
617 ext2_mountfs(struct vnode *devvp, struct mount *mp)
618 {
619 	struct ext2mount *ump;
620 	struct buf *bp;
621 	struct m_ext2fs *fs;
622 	struct ext2fs *es;
623 	struct cdev *dev = devvp->v_rdev;
624 	struct g_consumer *cp;
625 	struct bufobj *bo;
626 	struct csum *sump;
627 	int error;
628 	int ronly;
629 	int i;
630 	u_long size;
631 	int32_t *lp;
632 	int32_t e2fs_maxcontig;
633 
634 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
635 	/* XXX: use VOP_ACESS to check FS perms */
636 	g_topology_lock();
637 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
638 	g_topology_unlock();
639 	VOP_UNLOCK(devvp, 0);
640 	if (error)
641 		return (error);
642 
643 	/* XXX: should we check for some sectorsize or 512 instead? */
644 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
645 	    (SBSIZE < cp->provider->sectorsize)) {
646 		g_topology_lock();
647 		g_vfs_close(cp);
648 		g_topology_unlock();
649 		return (EINVAL);
650 	}
651 
652 	bo = &devvp->v_bufobj;
653 	bo->bo_private = cp;
654 	bo->bo_ops = g_vfs_bufops;
655 	if (devvp->v_rdev->si_iosize_max != 0)
656 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
657 	if (mp->mnt_iosize_max > MAXPHYS)
658 		mp->mnt_iosize_max = MAXPHYS;
659 
660 	bp = NULL;
661 	ump = NULL;
662 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
663 		goto out;
664 	es = (struct ext2fs *)bp->b_data;
665 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
666 		error = EINVAL;		/* XXX needs translation */
667 		goto out;
668 	}
669 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
670 	    (es->e2fs_state & E2FS_ERRORS)) {
671 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
672 			printf(
673 "WARNING: Filesystem was not properly dismounted\n");
674 		} else {
675 			printf(
676 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
677 			error = EPERM;
678 			goto out;
679 		}
680 	}
681 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
682 
683 	/*
684 	 * I don't know whether this is the right strategy. Note that
685 	 * we dynamically allocate both an m_ext2fs and an ext2fs
686 	 * while Linux keeps the super block in a locked buffer.
687 	 */
688 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
689 	    M_EXT2MNT, M_WAITOK | M_ZERO);
690 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
691 	    M_EXT2MNT, M_WAITOK);
692 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
693 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
694 	if ((error = compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
695 		goto out;
696 
697 	/*
698 	 * Calculate the maximum contiguous blocks and size of cluster summary
699 	 * array.  In FFS this is done by newfs; however, the superblock
700 	 * in ext2fs doesn't have these variables, so we can calculate
701 	 * them here.
702 	 */
703 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
704 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
705 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
706 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
707 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
708 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
709 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
710 		lp = ump->um_e2fs->e2fs_maxcluster;
711 		sump = ump->um_e2fs->e2fs_clustersum;
712 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
713 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
714 			sump->cs_init = 0;
715 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
716 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
717 		}
718 	}
719 
720 	brelse(bp);
721 	bp = NULL;
722 	fs = ump->um_e2fs;
723 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
724 
725 	/*
726 	 * If the fs is not mounted read-only, make sure the super block is
727 	 * always written back on a sync().
728 	 */
729 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
730 	if (ronly == 0) {
731 		fs->e2fs_fmod = 1;	/* mark it modified */
732 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
733 	}
734 	mp->mnt_data = ump;
735 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
736 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
737 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
738 	MNT_ILOCK(mp);
739 	mp->mnt_flag |= MNT_LOCAL;
740 	MNT_IUNLOCK(mp);
741 	ump->um_mountp = mp;
742 	ump->um_dev = dev;
743 	ump->um_devvp = devvp;
744 	ump->um_bo = &devvp->v_bufobj;
745 	ump->um_cp = cp;
746 
747 	/*
748 	 * Setting those two parameters allowed us to use
749 	 * ufs_bmap w/o changse!
750 	 */
751 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
752 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
753 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
754 	if (ronly == 0)
755 		ext2_sbupdate(ump, MNT_WAIT);
756 	/*
757 	 * Initialize filesystem stat information in mount struct.
758 	 */
759 	MNT_ILOCK(mp);
760 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
761 	    MNTK_USES_BCACHE;
762 	MNT_IUNLOCK(mp);
763 	return (0);
764 out:
765 	if (bp)
766 		brelse(bp);
767 	if (cp != NULL) {
768 		g_topology_lock();
769 		g_vfs_close(cp);
770 		g_topology_unlock();
771 	}
772 	if (ump) {
773 		mtx_destroy(EXT2_MTX(ump));
774 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
775 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
776 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
777 		free(ump->um_e2fs, M_EXT2MNT);
778 		free(ump, M_EXT2MNT);
779 		mp->mnt_data = NULL;
780 	}
781 	return (error);
782 }
783 
784 /*
785  * Unmount system call.
786  */
787 static int
788 ext2_unmount(struct mount *mp, int mntflags)
789 {
790 	struct ext2mount *ump;
791 	struct m_ext2fs *fs;
792 	struct csum *sump;
793 	int error, flags, i, ronly;
794 
795 	flags = 0;
796 	if (mntflags & MNT_FORCE) {
797 		if (mp->mnt_flag & MNT_ROOTFS)
798 			return (EINVAL);
799 		flags |= FORCECLOSE;
800 	}
801 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
802 		return (error);
803 	ump = VFSTOEXT2(mp);
804 	fs = ump->um_e2fs;
805 	ronly = fs->e2fs_ronly;
806 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
807 		if (fs->e2fs_wasvalid)
808 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
809 		ext2_sbupdate(ump, MNT_WAIT);
810 	}
811 
812 	g_topology_lock();
813 	g_vfs_close(ump->um_cp);
814 	g_topology_unlock();
815 	vrele(ump->um_devvp);
816 	sump = fs->e2fs_clustersum;
817 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
818 		free(sump->cs_sum, M_EXT2MNT);
819 	free(fs->e2fs_clustersum, M_EXT2MNT);
820 	free(fs->e2fs_maxcluster, M_EXT2MNT);
821 	free(fs->e2fs_gd, M_EXT2MNT);
822 	free(fs->e2fs_contigdirs, M_EXT2MNT);
823 	free(fs->e2fs, M_EXT2MNT);
824 	free(fs, M_EXT2MNT);
825 	free(ump, M_EXT2MNT);
826 	mp->mnt_data = NULL;
827 	MNT_ILOCK(mp);
828 	mp->mnt_flag &= ~MNT_LOCAL;
829 	MNT_IUNLOCK(mp);
830 	return (error);
831 }
832 
833 /*
834  * Flush out all the files in a filesystem.
835  */
836 static int
837 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
838 {
839 	int error;
840 
841 	error = vflush(mp, 0, flags, td);
842 	return (error);
843 }
844 
845 /*
846  * Get filesystem statistics.
847  */
848 int
849 ext2_statfs(struct mount *mp, struct statfs *sbp)
850 {
851 	struct ext2mount *ump;
852 	struct m_ext2fs *fs;
853 	uint32_t overhead, overhead_per_group, ngdb;
854 	int i, ngroups;
855 
856 	ump = VFSTOEXT2(mp);
857 	fs = ump->um_e2fs;
858 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
859 		panic("ext2_statfs");
860 
861 	/*
862 	 * Compute the overhead (FS structures)
863 	 */
864 	overhead_per_group =
865 	    1 /* block bitmap */ +
866 	    1 /* inode bitmap */ +
867 	    fs->e2fs_itpg;
868 	overhead = fs->e2fs->e2fs_first_dblock +
869 	    fs->e2fs_gcount * overhead_per_group;
870 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
871 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
872 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
873 			if (ext2_cg_has_sb(fs, i))
874 				ngroups++;
875 		}
876 	} else {
877 		ngroups = fs->e2fs_gcount;
878 	}
879 	ngdb = fs->e2fs_gdbcount;
880 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
881 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
882 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
883 	overhead += ngroups * (1 /* superblock */ + ngdb);
884 
885 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
886 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
887 	sbp->f_blocks = fs->e2fs_bcount - overhead;
888 	sbp->f_bfree = fs->e2fs_fbcount;
889 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
890 	sbp->f_files = fs->e2fs->e2fs_icount;
891 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
892 	return (0);
893 }
894 
895 /*
896  * Go through the disk queues to initiate sandbagged IO;
897  * go through the inodes to write those that have been modified;
898  * initiate the writing of the super block if it has been modified.
899  *
900  * Note: we are always called with the filesystem marked `MPBUSY'.
901  */
902 static int
903 ext2_sync(struct mount *mp, int waitfor)
904 {
905 	struct vnode *mvp, *vp;
906 	struct thread *td;
907 	struct inode *ip;
908 	struct ext2mount *ump = VFSTOEXT2(mp);
909 	struct m_ext2fs *fs;
910 	int error, allerror = 0;
911 
912 	td = curthread;
913 	fs = ump->um_e2fs;
914 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
915 		printf("fs = %s\n", fs->e2fs_fsmnt);
916 		panic("ext2_sync: rofs mod");
917 	}
918 
919 	/*
920 	 * Write back each (modified) inode.
921 	 */
922 loop:
923 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
924 		if (vp->v_type == VNON) {
925 			VI_UNLOCK(vp);
926 			continue;
927 		}
928 		ip = VTOI(vp);
929 		if ((ip->i_flag &
930 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
931 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
932 		    waitfor == MNT_LAZY)) {
933 			VI_UNLOCK(vp);
934 			continue;
935 		}
936 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
937 		if (error) {
938 			if (error == ENOENT) {
939 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
940 				goto loop;
941 			}
942 			continue;
943 		}
944 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
945 			allerror = error;
946 		VOP_UNLOCK(vp, 0);
947 		vrele(vp);
948 	}
949 
950 	/*
951 	 * Force stale filesystem control information to be flushed.
952 	 */
953 	if (waitfor != MNT_LAZY) {
954 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
955 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
956 			allerror = error;
957 		VOP_UNLOCK(ump->um_devvp, 0);
958 	}
959 
960 	/*
961 	 * Write back modified superblock.
962 	 */
963 	if (fs->e2fs_fmod != 0) {
964 		fs->e2fs_fmod = 0;
965 		fs->e2fs->e2fs_wtime = time_second;
966 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
967 			allerror = error;
968 	}
969 	return (allerror);
970 }
971 
972 /*
973  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
974  * in from disk.  If it is in core, wait for the lock bit to clear, then
975  * return the inode locked.  Detection and handling of mount points must be
976  * done by the calling routine.
977  */
978 static int
979 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
980 {
981 	struct m_ext2fs *fs;
982 	struct inode *ip;
983 	struct ext2mount *ump;
984 	struct buf *bp;
985 	struct vnode *vp;
986 	struct thread *td;
987 	int i, error;
988 	int used_blocks;
989 
990 	td = curthread;
991 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
992 	if (error || *vpp != NULL)
993 		return (error);
994 
995 	ump = VFSTOEXT2(mp);
996 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
997 
998 	/* Allocate a new vnode/inode. */
999 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
1000 		*vpp = NULL;
1001 		free(ip, M_EXT2NODE);
1002 		return (error);
1003 	}
1004 	vp->v_data = ip;
1005 	ip->i_vnode = vp;
1006 	ip->i_e2fs = fs = ump->um_e2fs;
1007 	ip->i_ump = ump;
1008 	ip->i_number = ino;
1009 
1010 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1011 	error = insmntque(vp, mp);
1012 	if (error != 0) {
1013 		free(ip, M_EXT2NODE);
1014 		*vpp = NULL;
1015 		return (error);
1016 	}
1017 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
1018 	if (error || *vpp != NULL)
1019 		return (error);
1020 
1021 	/* Read in the disk contents for the inode, copy into the inode. */
1022 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1023 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
1024 		/*
1025 		 * The inode does not contain anything useful, so it would
1026 		 * be misleading to leave it on its hash chain. With mode
1027 		 * still zero, it will be unlinked and returned to the free
1028 		 * list by vput().
1029 		 */
1030 		brelse(bp);
1031 		vput(vp);
1032 		*vpp = NULL;
1033 		return (error);
1034 	}
1035 	/* convert ext2 inode to dinode */
1036 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1037 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1038 	if (error) {
1039 		printf("ext2fs: Bad inode %lu csum - run fsck\n",
1040 		    (unsigned long)ino);
1041 		brelse(bp);
1042 		vput(vp);
1043 		*vpp = NULL;
1044 		return (error);
1045 	}
1046 	ip->i_block_group = ino_to_cg(fs, ino);
1047 	ip->i_next_alloc_block = 0;
1048 	ip->i_next_alloc_goal = 0;
1049 
1050 	/*
1051 	 * Now we want to make sure that block pointers for unused
1052 	 * blocks are zeroed out - ext2_balloc depends on this
1053 	 * although for regular files and directories only
1054 	 *
1055 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1056 	 * out because we could corrupt the extent tree.
1057 	 */
1058 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1059 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1060 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1061 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1062 			ip->i_db[i] = 0;
1063 	}
1064 #ifdef EXT2FS_DEBUG
1065 	ext2_print_inode(ip);
1066 	ext4_ext_print_extent_tree_status(ip);
1067 #endif
1068 	bqrelse(bp);
1069 
1070 	/*
1071 	 * Initialize the vnode from the inode, check for aliases.
1072 	 * Note that the underlying vnode may have changed.
1073 	 */
1074 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1075 		vput(vp);
1076 		*vpp = NULL;
1077 		return (error);
1078 	}
1079 
1080 	/*
1081 	 * Finish inode initialization.
1082 	 */
1083 
1084 	*vpp = vp;
1085 	return (0);
1086 }
1087 
1088 /*
1089  * File handle to vnode
1090  *
1091  * Have to be really careful about stale file handles:
1092  * - check that the inode number is valid
1093  * - call ext2_vget() to get the locked inode
1094  * - check for an unallocated inode (i_mode == 0)
1095  * - check that the given client host has export rights and return
1096  *   those rights via. exflagsp and credanonp
1097  */
1098 static int
1099 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1100 {
1101 	struct inode *ip;
1102 	struct ufid *ufhp;
1103 	struct vnode *nvp;
1104 	struct m_ext2fs *fs;
1105 	int error;
1106 
1107 	ufhp = (struct ufid *)fhp;
1108 	fs = VFSTOEXT2(mp)->um_e2fs;
1109 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1110 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1111 		return (ESTALE);
1112 
1113 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1114 	if (error) {
1115 		*vpp = NULLVP;
1116 		return (error);
1117 	}
1118 	ip = VTOI(nvp);
1119 	if (ip->i_mode == 0 ||
1120 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1121 		vput(nvp);
1122 		*vpp = NULLVP;
1123 		return (ESTALE);
1124 	}
1125 	*vpp = nvp;
1126 	vnode_create_vobject(*vpp, 0, curthread);
1127 	return (0);
1128 }
1129 
1130 /*
1131  * Write a superblock and associated information back to disk.
1132  */
1133 static int
1134 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1135 {
1136 	struct m_ext2fs *fs = mp->um_e2fs;
1137 	struct ext2fs *es = fs->e2fs;
1138 	struct buf *bp;
1139 	int error = 0;
1140 
1141 	es->e2fs_bcount = fs->e2fs_bcount & 0xffffffff;
1142 	es->e2fs_rbcount = fs->e2fs_rbcount & 0xffffffff;
1143 	es->e2fs_fbcount = fs->e2fs_fbcount & 0xffffffff;
1144 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1145 		es->e4fs_bcount_hi = fs->e2fs_bcount >> 32;
1146 		es->e4fs_rbcount_hi = fs->e2fs_rbcount >> 32;
1147 		es->e4fs_fbcount_hi = fs->e2fs_fbcount >> 32;
1148 	}
1149 
1150 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1151 		ext2_sb_csum_set(fs);
1152 
1153 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1154 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1155 	if (waitfor == MNT_WAIT)
1156 		error = bwrite(bp);
1157 	else
1158 		bawrite(bp);
1159 
1160 	/*
1161 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1162 	 * are not busy at this point and are (hopefully) written by the
1163 	 * usual sync mechanism. No need to write them here.
1164 	 */
1165 	return (error);
1166 }
1167 int
1168 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1169 {
1170 	struct m_ext2fs *fs = mp->um_e2fs;
1171 	struct buf *bp;
1172 	int i, j, g_count = 0, error = 0, allerror = 0;
1173 
1174 	allerror = ext2_sbupdate(mp, waitfor);
1175 
1176 	/* Update gd csums */
1177 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1178 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1179 		ext2_gd_csum_set(fs);
1180 
1181 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1182 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1183 		    cg_location(fs, i)),
1184 		    fs->e2fs_bsize, 0, 0, 0);
1185 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1186 			memcpy(bp->b_data, &fs->e2fs_gd[
1187 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1188 			    fs->e2fs_bsize);
1189 		} else {
1190 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1191 			    g_count < fs->e2fs_gcount; j++, g_count++)
1192 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1193 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1194 		}
1195 		if (waitfor == MNT_WAIT)
1196 			error = bwrite(bp);
1197 		else
1198 			bawrite(bp);
1199 	}
1200 
1201 	if (!allerror && error)
1202 		allerror = error;
1203 	return (allerror);
1204 }
1205 
1206 /*
1207  * Return the root of a filesystem.
1208  */
1209 static int
1210 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1211 {
1212 	struct vnode *nvp;
1213 	int error;
1214 
1215 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1216 	if (error)
1217 		return (error);
1218 	*vpp = nvp;
1219 	return (0);
1220 }
1221