xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision f5147e312f43a9050468de539aeafa072caa1a60)
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 		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 static e4fs_daddr_t
322 cg_location(struct m_ext2fs *fs, int number)
323 {
324 	int cg, descpb, logical_sb, has_super = 0;
325 
326 	/*
327 	 * Adjust logical superblock block number.
328 	 * Godmar thinks: if the blocksize is greater than 1024, then
329 	 * the superblock is logically part of block zero.
330 	 */
331 	logical_sb = fs->e2fs_bsize > SBSIZE ? 0 : 1;
332 
333 	if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
334 	    number < fs->e2fs->e3fs_first_meta_bg)
335 		return (logical_sb + number + 1);
336 
337 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT))
338 		descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
339 	else
340 		descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
341 
342 	cg = descpb * number;
343 
344 	if (ext2_cg_has_sb(fs, cg))
345 		has_super = 1;
346 
347 	return (has_super + cg * (e4fs_daddr_t)EXT2_BLOCKS_PER_GROUP(fs) +
348 	    fs->e2fs->e2fs_first_dblock);
349 }
350 
351 /*
352  * This computes the fields of the m_ext2fs structure from the
353  * data in the ext2fs structure read in.
354  */
355 static int
356 compute_sb_data(struct vnode *devvp, struct ext2fs *es,
357     struct m_ext2fs *fs)
358 {
359 	int g_count = 0, error;
360 	int i, j;
361 	struct buf *bp;
362 	uint32_t e2fs_descpb, e2fs_gdbcount_alloc;
363 
364 	fs->e2fs_bcount = es->e2fs_bcount;
365 	fs->e2fs_rbcount = es->e2fs_rbcount;
366 	fs->e2fs_fbcount = es->e2fs_fbcount;
367 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
368 		fs->e2fs_bcount |= (uint64_t)(es->e4fs_bcount_hi) << 32;
369 		fs->e2fs_rbcount |= (uint64_t)(es->e4fs_rbcount_hi) << 32;
370 		fs->e2fs_fbcount |= (uint64_t)(es->e4fs_fbcount_hi) << 32;
371 	}
372 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
373 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
374 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
375 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
376 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
377 	if (fs->e2fs_fsize)
378 		fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
379 	fs->e2fs_bpg = es->e2fs_bpg;
380 	fs->e2fs_fpg = es->e2fs_fpg;
381 	fs->e2fs_ipg = es->e2fs_ipg;
382 	if (es->e2fs_rev == E2FS_REV0) {
383 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
384 	} else {
385 		fs->e2fs_isize = es->e2fs_inode_size;
386 
387 		/*
388 		 * Simple sanity check for superblock inode size value.
389 		 */
390 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
391 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
392 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
393 			printf("ext2fs: invalid inode size %d\n",
394 			    fs->e2fs_isize);
395 			return (EIO);
396 		}
397 	}
398 	/* Check for extra isize in big inodes. */
399 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
400 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
401 		printf("ext2fs: no space for extra inode timestamps\n");
402 		return (EINVAL);
403 	}
404 	/* Check checksum features */
405 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) &&
406 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
407 		printf("ext2fs: incorrect checksum features combination\n");
408 		return (EINVAL);
409 	}
410 	/* Check for group descriptor size */
411 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT) &&
412 	    (es->e3fs_desc_size != sizeof(struct ext2_gd))) {
413 		printf("ext2fs: group descriptor size unsupported %d\n",
414 		    es->e3fs_desc_size);
415 		return (EINVAL);
416 	}
417 	/* Check for group size */
418 	if (fs->e2fs_bpg != fs->e2fs_bsize * 8) {
419 		printf("ext2fs: non-standard group size unsupported %d\n",
420 		    fs->e2fs_bpg);
421 		return (EINVAL);
422 	}
423 
424 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
425 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
426 	/* s_resuid / s_resgid ? */
427 	fs->e2fs_gcount = howmany(fs->e2fs_bcount - es->e2fs_first_dblock,
428 	    EXT2_BLOCKS_PER_GROUP(fs));
429 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
430 		e2fs_descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
431 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount, e2fs_descpb);
432 	} else {
433 		e2fs_descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
434 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount,
435 		    fs->e2fs_bsize / sizeof(struct ext2_gd));
436 	}
437 	fs->e2fs_gdbcount = howmany(fs->e2fs_gcount, e2fs_descpb);
438 	fs->e2fs_gd = malloc(e2fs_gdbcount_alloc * fs->e2fs_bsize,
439 	    M_EXT2MNT, M_WAITOK | M_ZERO);
440 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
441 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
442 
443 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
444 		error = bread(devvp,
445 		    fsbtodb(fs, cg_location(fs, i)),
446 		    fs->e2fs_bsize, NOCRED, &bp);
447 		if (error) {
448 			free(fs->e2fs_contigdirs, M_EXT2MNT);
449 			free(fs->e2fs_gd, M_EXT2MNT);
450 			brelse(bp);
451 			return (error);
452 		}
453 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
454 			memcpy(&fs->e2fs_gd[
455 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
456 			    bp->b_data, fs->e2fs_bsize);
457 		} else {
458 			for (j = 0; j < e2fs_descpb &&
459 			    g_count < fs->e2fs_gcount; j++, g_count++)
460 				memcpy(&fs->e2fs_gd[g_count],
461 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
462 				    E2FS_REV0_GD_SIZE);
463 		}
464 		brelse(bp);
465 		bp = NULL;
466 	}
467 	/* Precompute checksum seed for all metadata */
468 	ext2_sb_csum_set_seed(fs);
469 	/* Verfy cg csum */
470 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
471 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
472 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
473 		if (error)
474 			return (error);
475 	}
476 	/* Initialization for the ext2 Orlov allocator variant. */
477 	fs->e2fs_total_dir = 0;
478 	for (i = 0; i < fs->e2fs_gcount; i++)
479 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
480 
481 	if (es->e2fs_rev == E2FS_REV0 ||
482 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
483 		fs->e2fs_maxfilesize = 0x7fffffff;
484 	else {
485 		fs->e2fs_maxfilesize = 0xffffffffffff;
486 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
487 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
488 	}
489 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
490 		fs->e2fs_uhash = 3;
491 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
492 #ifdef __CHAR_UNSIGNED__
493 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
494 		fs->e2fs_uhash = 3;
495 #else
496 		es->e4fs_flags |= E2FS_SIGNED_HASH;
497 #endif
498 	}
499 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
500 		error = ext2_sb_csum_verify(fs);
501 
502 	return (error);
503 }
504 
505 /*
506  * Reload all incore data for a filesystem (used after running fsck on
507  * the root filesystem and finding things to fix). The filesystem must
508  * be mounted read-only.
509  *
510  * Things to do to update the mount:
511  *	1) invalidate all cached meta-data.
512  *	2) re-read superblock from disk.
513  *	3) invalidate all cluster summary information.
514  *	4) invalidate all inactive vnodes.
515  *	5) invalidate all cached file data.
516  *	6) re-read inode data for all active vnodes.
517  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
518  */
519 static int
520 ext2_reload(struct mount *mp, struct thread *td)
521 {
522 	struct vnode *vp, *mvp, *devvp;
523 	struct inode *ip;
524 	struct buf *bp;
525 	struct ext2fs *es;
526 	struct m_ext2fs *fs;
527 	struct csum *sump;
528 	int error, i;
529 	int32_t *lp;
530 
531 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
532 		return (EINVAL);
533 	/*
534 	 * Step 1: invalidate all cached meta-data.
535 	 */
536 	devvp = VFSTOEXT2(mp)->um_devvp;
537 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
538 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
539 		panic("ext2_reload: dirty1");
540 	VOP_UNLOCK(devvp, 0);
541 
542 	/*
543 	 * Step 2: re-read superblock from disk.
544 	 * constants have been adjusted for ext2
545 	 */
546 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
547 		return (error);
548 	es = (struct ext2fs *)bp->b_data;
549 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
550 		brelse(bp);
551 		return (EIO);		/* XXX needs translation */
552 	}
553 	fs = VFSTOEXT2(mp)->um_e2fs;
554 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
555 
556 	if ((error = compute_sb_data(devvp, es, fs)) != 0) {
557 		brelse(bp);
558 		return (error);
559 	}
560 #ifdef UNKLAR
561 	if (fs->fs_sbsize < SBSIZE)
562 		bp->b_flags |= B_INVAL;
563 #endif
564 	brelse(bp);
565 
566 	/*
567 	 * Step 3: invalidate all cluster summary information.
568 	 */
569 	if (fs->e2fs_contigsumsize > 0) {
570 		lp = fs->e2fs_maxcluster;
571 		sump = fs->e2fs_clustersum;
572 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
573 			*lp++ = fs->e2fs_contigsumsize;
574 			sump->cs_init = 0;
575 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
576 		}
577 	}
578 
579 loop:
580 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
581 		/*
582 		 * Step 4: invalidate all cached file data.
583 		 */
584 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
585 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
586 			goto loop;
587 		}
588 		if (vinvalbuf(vp, 0, 0, 0))
589 			panic("ext2_reload: dirty2");
590 
591 		/*
592 		 * Step 5: re-read inode data for all active vnodes.
593 		 */
594 		ip = VTOI(vp);
595 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
596 		    (int)fs->e2fs_bsize, NOCRED, &bp);
597 		if (error) {
598 			VOP_UNLOCK(vp, 0);
599 			vrele(vp);
600 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
601 			return (error);
602 		}
603 		ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
604 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
605 		brelse(bp);
606 		VOP_UNLOCK(vp, 0);
607 		vrele(vp);
608 	}
609 	return (0);
610 }
611 
612 /*
613  * Common code for mount and mountroot.
614  */
615 static int
616 ext2_mountfs(struct vnode *devvp, struct mount *mp)
617 {
618 	struct ext2mount *ump;
619 	struct buf *bp;
620 	struct m_ext2fs *fs;
621 	struct ext2fs *es;
622 	struct cdev *dev = devvp->v_rdev;
623 	struct g_consumer *cp;
624 	struct bufobj *bo;
625 	struct csum *sump;
626 	int error;
627 	int ronly;
628 	int i;
629 	u_long size;
630 	int32_t *lp;
631 	int32_t e2fs_maxcontig;
632 
633 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
634 	/* XXX: use VOP_ACESS to check FS perms */
635 	g_topology_lock();
636 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
637 	g_topology_unlock();
638 	VOP_UNLOCK(devvp, 0);
639 	if (error)
640 		return (error);
641 
642 	/* XXX: should we check for some sectorsize or 512 instead? */
643 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
644 	    (SBSIZE < cp->provider->sectorsize)) {
645 		g_topology_lock();
646 		g_vfs_close(cp);
647 		g_topology_unlock();
648 		return (EINVAL);
649 	}
650 
651 	bo = &devvp->v_bufobj;
652 	bo->bo_private = cp;
653 	bo->bo_ops = g_vfs_bufops;
654 	if (devvp->v_rdev->si_iosize_max != 0)
655 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
656 	if (mp->mnt_iosize_max > MAXPHYS)
657 		mp->mnt_iosize_max = MAXPHYS;
658 
659 	bp = NULL;
660 	ump = NULL;
661 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
662 		goto out;
663 	es = (struct ext2fs *)bp->b_data;
664 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
665 		error = EINVAL;		/* XXX needs translation */
666 		goto out;
667 	}
668 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
669 	    (es->e2fs_state & E2FS_ERRORS)) {
670 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
671 			printf(
672 "WARNING: Filesystem was not properly dismounted\n");
673 		} else {
674 			printf(
675 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
676 			error = EPERM;
677 			goto out;
678 		}
679 	}
680 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
681 
682 	/*
683 	 * I don't know whether this is the right strategy. Note that
684 	 * we dynamically allocate both an m_ext2fs and an ext2fs
685 	 * while Linux keeps the super block in a locked buffer.
686 	 */
687 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
688 	    M_EXT2MNT, M_WAITOK | M_ZERO);
689 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
690 	    M_EXT2MNT, M_WAITOK);
691 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
692 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
693 	if ((error = compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
694 		goto out;
695 
696 	/*
697 	 * Calculate the maximum contiguous blocks and size of cluster summary
698 	 * array.  In FFS this is done by newfs; however, the superblock
699 	 * in ext2fs doesn't have these variables, so we can calculate
700 	 * them here.
701 	 */
702 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
703 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
704 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
705 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
706 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
707 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
708 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
709 		lp = ump->um_e2fs->e2fs_maxcluster;
710 		sump = ump->um_e2fs->e2fs_clustersum;
711 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
712 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
713 			sump->cs_init = 0;
714 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
715 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
716 		}
717 	}
718 
719 	brelse(bp);
720 	bp = NULL;
721 	fs = ump->um_e2fs;
722 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
723 
724 	/*
725 	 * If the fs is not mounted read-only, make sure the super block is
726 	 * always written back on a sync().
727 	 */
728 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
729 	if (ronly == 0) {
730 		fs->e2fs_fmod = 1;	/* mark it modified */
731 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
732 	}
733 	mp->mnt_data = ump;
734 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
735 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
736 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
737 	MNT_ILOCK(mp);
738 	mp->mnt_flag |= MNT_LOCAL;
739 	MNT_IUNLOCK(mp);
740 	ump->um_mountp = mp;
741 	ump->um_dev = dev;
742 	ump->um_devvp = devvp;
743 	ump->um_bo = &devvp->v_bufobj;
744 	ump->um_cp = cp;
745 
746 	/*
747 	 * Setting those two parameters allowed us to use
748 	 * ufs_bmap w/o changse!
749 	 */
750 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
751 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
752 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
753 	if (ronly == 0)
754 		ext2_sbupdate(ump, MNT_WAIT);
755 	/*
756 	 * Initialize filesystem stat information in mount struct.
757 	 */
758 	MNT_ILOCK(mp);
759 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
760 	    MNTK_USES_BCACHE;
761 	MNT_IUNLOCK(mp);
762 	return (0);
763 out:
764 	if (bp)
765 		brelse(bp);
766 	if (cp != NULL) {
767 		g_topology_lock();
768 		g_vfs_close(cp);
769 		g_topology_unlock();
770 	}
771 	if (ump) {
772 		mtx_destroy(EXT2_MTX(ump));
773 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
774 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
775 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
776 		free(ump->um_e2fs, M_EXT2MNT);
777 		free(ump, M_EXT2MNT);
778 		mp->mnt_data = NULL;
779 	}
780 	return (error);
781 }
782 
783 /*
784  * Unmount system call.
785  */
786 static int
787 ext2_unmount(struct mount *mp, int mntflags)
788 {
789 	struct ext2mount *ump;
790 	struct m_ext2fs *fs;
791 	struct csum *sump;
792 	int error, flags, i, ronly;
793 
794 	flags = 0;
795 	if (mntflags & MNT_FORCE) {
796 		if (mp->mnt_flag & MNT_ROOTFS)
797 			return (EINVAL);
798 		flags |= FORCECLOSE;
799 	}
800 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
801 		return (error);
802 	ump = VFSTOEXT2(mp);
803 	fs = ump->um_e2fs;
804 	ronly = fs->e2fs_ronly;
805 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
806 		if (fs->e2fs_wasvalid)
807 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
808 		ext2_sbupdate(ump, MNT_WAIT);
809 	}
810 
811 	g_topology_lock();
812 	g_vfs_close(ump->um_cp);
813 	g_topology_unlock();
814 	vrele(ump->um_devvp);
815 	sump = fs->e2fs_clustersum;
816 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
817 		free(sump->cs_sum, M_EXT2MNT);
818 	free(fs->e2fs_clustersum, M_EXT2MNT);
819 	free(fs->e2fs_maxcluster, M_EXT2MNT);
820 	free(fs->e2fs_gd, M_EXT2MNT);
821 	free(fs->e2fs_contigdirs, M_EXT2MNT);
822 	free(fs->e2fs, M_EXT2MNT);
823 	free(fs, M_EXT2MNT);
824 	free(ump, M_EXT2MNT);
825 	mp->mnt_data = NULL;
826 	MNT_ILOCK(mp);
827 	mp->mnt_flag &= ~MNT_LOCAL;
828 	MNT_IUNLOCK(mp);
829 	return (error);
830 }
831 
832 /*
833  * Flush out all the files in a filesystem.
834  */
835 static int
836 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
837 {
838 	int error;
839 
840 	error = vflush(mp, 0, flags, td);
841 	return (error);
842 }
843 
844 /*
845  * Get filesystem statistics.
846  */
847 int
848 ext2_statfs(struct mount *mp, struct statfs *sbp)
849 {
850 	struct ext2mount *ump;
851 	struct m_ext2fs *fs;
852 	uint32_t overhead, overhead_per_group, ngdb;
853 	int i, ngroups;
854 
855 	ump = VFSTOEXT2(mp);
856 	fs = ump->um_e2fs;
857 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
858 		panic("ext2_statfs");
859 
860 	/*
861 	 * Compute the overhead (FS structures)
862 	 */
863 	overhead_per_group =
864 	    1 /* block bitmap */ +
865 	    1 /* inode bitmap */ +
866 	    fs->e2fs_itpg;
867 	overhead = fs->e2fs->e2fs_first_dblock +
868 	    fs->e2fs_gcount * overhead_per_group;
869 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
870 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
871 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
872 			if (ext2_cg_has_sb(fs, i))
873 				ngroups++;
874 		}
875 	} else {
876 		ngroups = fs->e2fs_gcount;
877 	}
878 	ngdb = fs->e2fs_gdbcount;
879 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
880 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
881 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
882 	overhead += ngroups * (1 /* superblock */ + ngdb);
883 
884 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
885 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
886 	sbp->f_blocks = fs->e2fs_bcount - overhead;
887 	sbp->f_bfree = fs->e2fs_fbcount;
888 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
889 	sbp->f_files = fs->e2fs->e2fs_icount;
890 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
891 	return (0);
892 }
893 
894 /*
895  * Go through the disk queues to initiate sandbagged IO;
896  * go through the inodes to write those that have been modified;
897  * initiate the writing of the super block if it has been modified.
898  *
899  * Note: we are always called with the filesystem marked `MPBUSY'.
900  */
901 static int
902 ext2_sync(struct mount *mp, int waitfor)
903 {
904 	struct vnode *mvp, *vp;
905 	struct thread *td;
906 	struct inode *ip;
907 	struct ext2mount *ump = VFSTOEXT2(mp);
908 	struct m_ext2fs *fs;
909 	int error, allerror = 0;
910 
911 	td = curthread;
912 	fs = ump->um_e2fs;
913 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
914 		printf("fs = %s\n", fs->e2fs_fsmnt);
915 		panic("ext2_sync: rofs mod");
916 	}
917 
918 	/*
919 	 * Write back each (modified) inode.
920 	 */
921 loop:
922 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
923 		if (vp->v_type == VNON) {
924 			VI_UNLOCK(vp);
925 			continue;
926 		}
927 		ip = VTOI(vp);
928 		if ((ip->i_flag &
929 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
930 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
931 		    waitfor == MNT_LAZY)) {
932 			VI_UNLOCK(vp);
933 			continue;
934 		}
935 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
936 		if (error) {
937 			if (error == ENOENT) {
938 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
939 				goto loop;
940 			}
941 			continue;
942 		}
943 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
944 			allerror = error;
945 		VOP_UNLOCK(vp, 0);
946 		vrele(vp);
947 	}
948 
949 	/*
950 	 * Force stale filesystem control information to be flushed.
951 	 */
952 	if (waitfor != MNT_LAZY) {
953 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
954 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
955 			allerror = error;
956 		VOP_UNLOCK(ump->um_devvp, 0);
957 	}
958 
959 	/*
960 	 * Write back modified superblock.
961 	 */
962 	if (fs->e2fs_fmod != 0) {
963 		fs->e2fs_fmod = 0;
964 		fs->e2fs->e2fs_wtime = time_second;
965 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
966 			allerror = error;
967 	}
968 	return (allerror);
969 }
970 
971 /*
972  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
973  * in from disk.  If it is in core, wait for the lock bit to clear, then
974  * return the inode locked.  Detection and handling of mount points must be
975  * done by the calling routine.
976  */
977 static int
978 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
979 {
980 	struct m_ext2fs *fs;
981 	struct inode *ip;
982 	struct ext2mount *ump;
983 	struct buf *bp;
984 	struct vnode *vp;
985 	struct thread *td;
986 	int i, error;
987 	int used_blocks;
988 
989 	td = curthread;
990 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
991 	if (error || *vpp != NULL)
992 		return (error);
993 
994 	ump = VFSTOEXT2(mp);
995 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
996 
997 	/* Allocate a new vnode/inode. */
998 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
999 		*vpp = NULL;
1000 		free(ip, M_EXT2NODE);
1001 		return (error);
1002 	}
1003 	vp->v_data = ip;
1004 	ip->i_vnode = vp;
1005 	ip->i_e2fs = fs = ump->um_e2fs;
1006 	ip->i_ump = ump;
1007 	ip->i_number = ino;
1008 
1009 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1010 	error = insmntque(vp, mp);
1011 	if (error != 0) {
1012 		free(ip, M_EXT2NODE);
1013 		*vpp = NULL;
1014 		return (error);
1015 	}
1016 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
1017 	if (error || *vpp != NULL)
1018 		return (error);
1019 
1020 	/* Read in the disk contents for the inode, copy into the inode. */
1021 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1022 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
1023 		/*
1024 		 * The inode does not contain anything useful, so it would
1025 		 * be misleading to leave it on its hash chain. With mode
1026 		 * still zero, it will be unlinked and returned to the free
1027 		 * list by vput().
1028 		 */
1029 		brelse(bp);
1030 		vput(vp);
1031 		*vpp = NULL;
1032 		return (error);
1033 	}
1034 	/* convert ext2 inode to dinode */
1035 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1036 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1037 	if (error) {
1038 		printf("ext2fs: Bad inode %lu csum - run fsck\n",
1039 		    (unsigned long)ino);
1040 		brelse(bp);
1041 		vput(vp);
1042 		*vpp = NULL;
1043 		return (error);
1044 	}
1045 	ip->i_block_group = ino_to_cg(fs, ino);
1046 	ip->i_next_alloc_block = 0;
1047 	ip->i_next_alloc_goal = 0;
1048 
1049 	/*
1050 	 * Now we want to make sure that block pointers for unused
1051 	 * blocks are zeroed out - ext2_balloc depends on this
1052 	 * although for regular files and directories only
1053 	 *
1054 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1055 	 * out because we could corrupt the extent tree.
1056 	 */
1057 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1058 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1059 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1060 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1061 			ip->i_db[i] = 0;
1062 	}
1063 #ifdef EXT2FS_DEBUG
1064 	ext2_print_inode(ip);
1065 	ext4_ext_print_extent_tree_status(ip);
1066 #endif
1067 	bqrelse(bp);
1068 
1069 	/*
1070 	 * Initialize the vnode from the inode, check for aliases.
1071 	 * Note that the underlying vnode may have changed.
1072 	 */
1073 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1074 		vput(vp);
1075 		*vpp = NULL;
1076 		return (error);
1077 	}
1078 
1079 	/*
1080 	 * Finish inode initialization.
1081 	 */
1082 
1083 	*vpp = vp;
1084 	return (0);
1085 }
1086 
1087 /*
1088  * File handle to vnode
1089  *
1090  * Have to be really careful about stale file handles:
1091  * - check that the inode number is valid
1092  * - call ext2_vget() to get the locked inode
1093  * - check for an unallocated inode (i_mode == 0)
1094  * - check that the given client host has export rights and return
1095  *   those rights via. exflagsp and credanonp
1096  */
1097 static int
1098 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1099 {
1100 	struct inode *ip;
1101 	struct ufid *ufhp;
1102 	struct vnode *nvp;
1103 	struct m_ext2fs *fs;
1104 	int error;
1105 
1106 	ufhp = (struct ufid *)fhp;
1107 	fs = VFSTOEXT2(mp)->um_e2fs;
1108 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1109 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1110 		return (ESTALE);
1111 
1112 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1113 	if (error) {
1114 		*vpp = NULLVP;
1115 		return (error);
1116 	}
1117 	ip = VTOI(nvp);
1118 	if (ip->i_mode == 0 ||
1119 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1120 		vput(nvp);
1121 		*vpp = NULLVP;
1122 		return (ESTALE);
1123 	}
1124 	*vpp = nvp;
1125 	vnode_create_vobject(*vpp, 0, curthread);
1126 	return (0);
1127 }
1128 
1129 /*
1130  * Write a superblock and associated information back to disk.
1131  */
1132 static int
1133 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1134 {
1135 	struct m_ext2fs *fs = mp->um_e2fs;
1136 	struct ext2fs *es = fs->e2fs;
1137 	struct buf *bp;
1138 	int error = 0;
1139 
1140 	es->e2fs_bcount = fs->e2fs_bcount & 0xffffffff;
1141 	es->e2fs_rbcount = fs->e2fs_rbcount & 0xffffffff;
1142 	es->e2fs_fbcount = fs->e2fs_fbcount & 0xffffffff;
1143 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1144 		es->e4fs_bcount_hi = fs->e2fs_bcount >> 32;
1145 		es->e4fs_rbcount_hi = fs->e2fs_rbcount >> 32;
1146 		es->e4fs_fbcount_hi = fs->e2fs_fbcount >> 32;
1147 	}
1148 
1149 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1150 		ext2_sb_csum_set(fs);
1151 
1152 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1153 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1154 	if (waitfor == MNT_WAIT)
1155 		error = bwrite(bp);
1156 	else
1157 		bawrite(bp);
1158 
1159 	/*
1160 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1161 	 * are not busy at this point and are (hopefully) written by the
1162 	 * usual sync mechanism. No need to write them here.
1163 	 */
1164 	return (error);
1165 }
1166 int
1167 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1168 {
1169 	struct m_ext2fs *fs = mp->um_e2fs;
1170 	struct buf *bp;
1171 	int i, j, g_count = 0, error = 0, allerror = 0;
1172 
1173 	allerror = ext2_sbupdate(mp, waitfor);
1174 
1175 	/* Update gd csums */
1176 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1177 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1178 		ext2_gd_csum_set(fs);
1179 
1180 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1181 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1182 		    cg_location(fs, i)),
1183 		    fs->e2fs_bsize, 0, 0, 0);
1184 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1185 			memcpy(bp->b_data, &fs->e2fs_gd[
1186 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1187 			    fs->e2fs_bsize);
1188 		} else {
1189 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1190 			    g_count < fs->e2fs_gcount; j++, g_count++)
1191 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1192 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1193 		}
1194 		if (waitfor == MNT_WAIT)
1195 			error = bwrite(bp);
1196 		else
1197 			bawrite(bp);
1198 	}
1199 
1200 	if (!allerror && error)
1201 		allerror = error;
1202 	return (allerror);
1203 }
1204 
1205 /*
1206  * Return the root of a filesystem.
1207  */
1208 static int
1209 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1210 {
1211 	struct vnode *nvp;
1212 	int error;
1213 
1214 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1215 	if (error)
1216 		return (error);
1217 	*vpp = nvp;
1218 	return (0);
1219 }
1220