xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision 8e9b3e707151d136ec95e7f1d37556e39c1e228c)
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/fs.h>
62 #include <fs/ext2fs/ext2_mount.h>
63 #include <fs/ext2fs/inode.h>
64 
65 #include <fs/ext2fs/ext2fs.h>
66 #include <fs/ext2fs/ext2_dinode.h>
67 #include <fs/ext2fs/ext2_extern.h>
68 #include <fs/ext2fs/ext2_extents.h>
69 
70 
71 static int	ext2_flushfiles(struct mount *mp, int flags, struct thread *td);
72 static int	ext2_mountfs(struct vnode *, struct mount *);
73 static int	ext2_reload(struct mount *mp, struct thread *td);
74 static int	ext2_sbupdate(struct ext2mount *, int);
75 static int	ext2_cgupdate(struct ext2mount *, int);
76 static vfs_unmount_t		ext2_unmount;
77 static vfs_root_t		ext2_root;
78 static vfs_statfs_t		ext2_statfs;
79 static vfs_sync_t		ext2_sync;
80 static vfs_vget_t		ext2_vget;
81 static vfs_fhtovp_t		ext2_fhtovp;
82 static vfs_mount_t		ext2_mount;
83 
84 MALLOC_DEFINE(M_EXT2NODE, "ext2_node", "EXT2 vnode private part");
85 static MALLOC_DEFINE(M_EXT2MNT, "ext2_mount", "EXT2 mount structure");
86 
87 static struct vfsops ext2fs_vfsops = {
88 	.vfs_fhtovp =		ext2_fhtovp,
89 	.vfs_mount =		ext2_mount,
90 	.vfs_root =		ext2_root,	/* root inode via vget */
91 	.vfs_statfs =		ext2_statfs,
92 	.vfs_sync =		ext2_sync,
93 	.vfs_unmount =		ext2_unmount,
94 	.vfs_vget =		ext2_vget,
95 };
96 
97 VFS_SET(ext2fs_vfsops, ext2fs, 0);
98 
99 static int	ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev,
100 		    int ronly);
101 static int	compute_sb_data(struct vnode * devvp,
102 		    struct ext2fs * es, struct m_ext2fs * fs);
103 
104 static const char *ext2_opts[] = { "acls", "async", "noatime", "noclusterr",
105     "noclusterw", "noexec", "export", "force", "from", "multilabel",
106     "suiddir", "nosymfollow", "sync", "union", NULL };
107 
108 /*
109  * VFS Operations.
110  *
111  * mount system call
112  */
113 static int
114 ext2_mount(struct mount *mp)
115 {
116 	struct vfsoptlist *opts;
117 	struct vnode *devvp;
118 	struct thread *td;
119 	struct ext2mount *ump = NULL;
120 	struct m_ext2fs *fs;
121 	struct nameidata nd, *ndp = &nd;
122 	accmode_t accmode;
123 	char *path, *fspec;
124 	int error, flags, len;
125 
126 	td = curthread;
127 	opts = mp->mnt_optnew;
128 
129 	if (vfs_filteropt(opts, ext2_opts))
130 		return (EINVAL);
131 
132 	vfs_getopt(opts, "fspath", (void **)&path, NULL);
133 	/* Double-check the length of path.. */
134 	if (strlen(path) >= MAXMNTLEN)
135 		return (ENAMETOOLONG);
136 
137 	fspec = NULL;
138 	error = vfs_getopt(opts, "from", (void **)&fspec, &len);
139 	if (!error && fspec[len - 1] != '\0')
140 		return (EINVAL);
141 
142 	/*
143 	 * If updating, check whether changing from read-only to
144 	 * read/write; if there is no device name, that's all we do.
145 	 */
146 	if (mp->mnt_flag & MNT_UPDATE) {
147 		ump = VFSTOEXT2(mp);
148 		fs = ump->um_e2fs;
149 		error = 0;
150 		if (fs->e2fs_ronly == 0 &&
151 		    vfs_flagopt(opts, "ro", NULL, 0)) {
152 			error = VFS_SYNC(mp, MNT_WAIT);
153 			if (error)
154 				return (error);
155 			flags = WRITECLOSE;
156 			if (mp->mnt_flag & MNT_FORCE)
157 				flags |= FORCECLOSE;
158 			error = ext2_flushfiles(mp, flags, td);
159 			if (error == 0 && fs->e2fs_wasvalid &&
160 			    ext2_cgupdate(ump, MNT_WAIT) == 0) {
161 				fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
162 				ext2_sbupdate(ump, MNT_WAIT);
163 			}
164 			fs->e2fs_ronly = 1;
165 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
166 			g_topology_lock();
167 			g_access(ump->um_cp, 0, -1, 0);
168 			g_topology_unlock();
169 		}
170 		if (!error && (mp->mnt_flag & MNT_RELOAD))
171 			error = ext2_reload(mp, td);
172 		if (error)
173 			return (error);
174 		devvp = ump->um_devvp;
175 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
176 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
177 				return (EPERM);
178 
179 			/*
180 			 * If upgrade to read-write by non-root, then verify
181 			 * that user has necessary permissions on the device.
182 			 */
183 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
184 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
185 			    td->td_ucred, td);
186 			if (error)
187 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
188 			if (error) {
189 				VOP_UNLOCK(devvp, 0);
190 				return (error);
191 			}
192 			VOP_UNLOCK(devvp, 0);
193 			g_topology_lock();
194 			error = g_access(ump->um_cp, 0, 1, 0);
195 			g_topology_unlock();
196 			if (error)
197 				return (error);
198 
199 			if ((fs->e2fs->e2fs_state & E2FS_ISCLEAN) == 0 ||
200 			    (fs->e2fs->e2fs_state & E2FS_ERRORS)) {
201 				if (mp->mnt_flag & MNT_FORCE) {
202 					printf(
203 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
204 				} else {
205 					printf(
206 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
207 					    fs->e2fs_fsmnt);
208 					return (EPERM);
209 				}
210 			}
211 			fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;
212 			(void)ext2_cgupdate(ump, MNT_WAIT);
213 			fs->e2fs_ronly = 0;
214 			MNT_ILOCK(mp);
215 			mp->mnt_flag &= ~MNT_RDONLY;
216 			MNT_IUNLOCK(mp);
217 		}
218 		if (vfs_flagopt(opts, "export", NULL, 0)) {
219 			/* Process export requests in vfs_mount.c. */
220 			return (error);
221 		}
222 	}
223 
224 	/*
225 	 * Not an update, or updating the name: look up the name
226 	 * and verify that it refers to a sensible disk device.
227 	 */
228 	if (fspec == NULL)
229 		return (EINVAL);
230 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
231 	if ((error = namei(ndp)) != 0)
232 		return (error);
233 	NDFREE(ndp, NDF_ONLY_PNBUF);
234 	devvp = ndp->ni_vp;
235 
236 	if (!vn_isdisk(devvp, &error)) {
237 		vput(devvp);
238 		return (error);
239 	}
240 
241 	/*
242 	 * If mount by non-root, then verify that user has necessary
243 	 * permissions on the device.
244 	 *
245 	 * XXXRW: VOP_ACCESS() enough?
246 	 */
247 	accmode = VREAD;
248 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
249 		accmode |= VWRITE;
250 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
251 	if (error)
252 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
253 	if (error) {
254 		vput(devvp);
255 		return (error);
256 	}
257 
258 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
259 		error = ext2_mountfs(devvp, mp);
260 	} else {
261 		if (devvp != ump->um_devvp) {
262 			vput(devvp);
263 			return (EINVAL);	/* needs translation */
264 		} else
265 			vput(devvp);
266 	}
267 	if (error) {
268 		vrele(devvp);
269 		return (error);
270 	}
271 	ump = VFSTOEXT2(mp);
272 	fs = ump->um_e2fs;
273 
274 	/*
275 	 * Note that this strncpy() is ok because of a check at the start
276 	 * of ext2_mount().
277 	 */
278 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
279 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
280 	vfs_mountedfrom(mp, fspec);
281 	return (0);
282 }
283 
284 static int
285 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
286 {
287 	uint32_t i, mask;
288 
289 	if (es->e2fs_magic != E2FS_MAGIC) {
290 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
291 		    devtoname(dev), es->e2fs_magic, E2FS_MAGIC);
292 		return (1);
293 	}
294 	if (es->e2fs_rev > E2FS_REV0) {
295 		mask = es->e2fs_features_incompat & ~(EXT2F_INCOMPAT_SUPP);
296 		if (mask) {
297 			printf("WARNING: mount of %s denied due to "
298 			    "unsupported optional features:\n", devtoname(dev));
299 			for (i = 0;
300 			    i < sizeof(incompat)/sizeof(struct ext2_feature);
301 			    i++)
302 				if (mask & incompat[i].mask)
303 					printf("%s ", incompat[i].name);
304 			printf("\n");
305 			return (1);
306 		}
307 		mask = es->e2fs_features_rocompat & ~EXT2F_ROCOMPAT_SUPP;
308 		if (!ronly && mask) {
309 			printf("WARNING: R/W mount of %s denied due to "
310 			    "unsupported optional features:\n", devtoname(dev));
311 			for (i = 0;
312 			    i < sizeof(ro_compat)/sizeof(struct ext2_feature);
313 			    i++)
314 				if (mask & ro_compat[i].mask)
315 					printf("%s ", ro_compat[i].name);
316 			printf("\n");
317 			return (1);
318 		}
319 	}
320 	return (0);
321 }
322 
323 static e4fs_daddr_t
324 cg_location(struct m_ext2fs *fs, int number)
325 {
326 	int cg, descpb, logical_sb, has_super = 0;
327 
328 	/*
329 	 * Adjust logical superblock block number.
330 	 * Godmar thinks: if the blocksize is greater than 1024, then
331 	 * the superblock is logically part of block zero.
332 	 */
333 	logical_sb = fs->e2fs_bsize > SBSIZE ? 0 : 1;
334 
335 	if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
336 	    number < fs->e2fs->e3fs_first_meta_bg)
337 		return (logical_sb + number + 1);
338 
339 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT))
340 		descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
341 	else
342 		descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
343 
344 	cg = descpb * number;
345 
346 	if (ext2_cg_has_sb(fs, cg))
347 		has_super = 1;
348 
349 	return (has_super + cg * (e4fs_daddr_t)EXT2_BLOCKS_PER_GROUP(fs) +
350 	    fs->e2fs->e2fs_first_dblock);
351 }
352 
353 /*
354  * This computes the fields of the m_ext2fs structure from the
355  * data in the ext2fs structure read in.
356  */
357 static int
358 compute_sb_data(struct vnode *devvp, struct ext2fs *es,
359     struct m_ext2fs *fs)
360 {
361 	int g_count = 0, error;
362 	int i, j;
363 	struct buf *bp;
364 	uint32_t e2fs_descpb, e2fs_gdbcount_alloc;
365 
366 	fs->e2fs_bcount = es->e2fs_bcount;
367 	fs->e2fs_rbcount = es->e2fs_rbcount;
368 	fs->e2fs_fbcount = es->e2fs_fbcount;
369 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
370 		fs->e2fs_bcount |= (uint64_t)(es->e4fs_bcount_hi) << 32;
371 		fs->e2fs_rbcount |= (uint64_t)(es->e4fs_rbcount_hi) << 32;
372 		fs->e2fs_fbcount |= (uint64_t)(es->e4fs_fbcount_hi) << 32;
373 	}
374 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
375 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
376 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
377 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
378 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
379 	if (fs->e2fs_fsize)
380 		fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
381 	fs->e2fs_bpg = es->e2fs_bpg;
382 	fs->e2fs_fpg = es->e2fs_fpg;
383 	fs->e2fs_ipg = es->e2fs_ipg;
384 	if (es->e2fs_rev == E2FS_REV0) {
385 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
386 	} else {
387 		fs->e2fs_isize = es->e2fs_inode_size;
388 
389 		/*
390 		 * Simple sanity check for superblock inode size value.
391 		 */
392 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
393 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
394 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
395 			printf("ext2fs: invalid inode size %d\n",
396 			    fs->e2fs_isize);
397 			return (EIO);
398 		}
399 	}
400 	/* Check for extra isize in big inodes. */
401 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
402 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
403 		printf("ext2fs: no space for extra inode timestamps\n");
404 		return (EINVAL);
405 	}
406 	/* Check checksum features */
407 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) &&
408 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
409 		printf("ext2fs: incorrect checksum features combination\n");
410 		return (EINVAL);
411 	}
412 	/* Check for group descriptor size */
413 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT) &&
414 	    (es->e3fs_desc_size != sizeof(struct ext2_gd))) {
415 		printf("ext2fs: group descriptor size unsupported %d\n",
416 		    es->e3fs_desc_size);
417 		return (EINVAL);
418 	}
419 	/* Check for group size */
420 	if (fs->e2fs_bpg != fs->e2fs_bsize * 8) {
421 		printf("ext2fs: non-standard group size unsupported %d\n",
422 		    fs->e2fs_bpg);
423 		return (EINVAL);
424 	}
425 
426 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
427 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
428 	/* s_resuid / s_resgid ? */
429 	fs->e2fs_gcount = howmany(fs->e2fs_bcount - es->e2fs_first_dblock,
430 	    EXT2_BLOCKS_PER_GROUP(fs));
431 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
432 		e2fs_descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
433 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount, e2fs_descpb);
434 	} else {
435 		e2fs_descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
436 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount,
437 		    fs->e2fs_bsize / sizeof(struct ext2_gd));
438 	}
439 	fs->e2fs_gdbcount = howmany(fs->e2fs_gcount, e2fs_descpb);
440 	fs->e2fs_gd = malloc(e2fs_gdbcount_alloc * fs->e2fs_bsize,
441 	    M_EXT2MNT, M_WAITOK | M_ZERO);
442 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
443 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
444 
445 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
446 		error = bread(devvp,
447 		    fsbtodb(fs, cg_location(fs, i)),
448 		    fs->e2fs_bsize, NOCRED, &bp);
449 		if (error) {
450 			free(fs->e2fs_contigdirs, M_EXT2MNT);
451 			free(fs->e2fs_gd, M_EXT2MNT);
452 			brelse(bp);
453 			return (error);
454 		}
455 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
456 			memcpy(&fs->e2fs_gd[
457 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
458 			    bp->b_data, fs->e2fs_bsize);
459 		} else {
460 			for (j = 0; j < e2fs_descpb &&
461 			    g_count < fs->e2fs_gcount; j++, g_count++)
462 				memcpy(&fs->e2fs_gd[g_count],
463 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
464 				    E2FS_REV0_GD_SIZE);
465 		}
466 		brelse(bp);
467 		bp = NULL;
468 	}
469 	/* Precompute checksum seed for all metadata */
470 	ext2_sb_csum_set_seed(fs);
471 	/* Verfy cg csum */
472 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
473 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
474 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
475 		if (error)
476 			return (error);
477 	}
478 	/* Initialization for the ext2 Orlov allocator variant. */
479 	fs->e2fs_total_dir = 0;
480 	for (i = 0; i < fs->e2fs_gcount; i++)
481 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
482 
483 	if (es->e2fs_rev == E2FS_REV0 ||
484 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
485 		fs->e2fs_maxfilesize = 0x7fffffff;
486 	else {
487 		fs->e2fs_maxfilesize = 0xffffffffffff;
488 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
489 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
490 	}
491 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
492 		fs->e2fs_uhash = 3;
493 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
494 #ifdef __CHAR_UNSIGNED__
495 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
496 		fs->e2fs_uhash = 3;
497 #else
498 		es->e4fs_flags |= E2FS_SIGNED_HASH;
499 #endif
500 	}
501 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
502 		error = ext2_sb_csum_verify(fs);
503 
504 	return (error);
505 }
506 
507 /*
508  * Reload all incore data for a filesystem (used after running fsck on
509  * the root filesystem and finding things to fix). The filesystem must
510  * be mounted read-only.
511  *
512  * Things to do to update the mount:
513  *	1) invalidate all cached meta-data.
514  *	2) re-read superblock from disk.
515  *	3) invalidate all cluster summary information.
516  *	4) invalidate all inactive vnodes.
517  *	5) invalidate all cached file data.
518  *	6) re-read inode data for all active vnodes.
519  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
520  */
521 static int
522 ext2_reload(struct mount *mp, struct thread *td)
523 {
524 	struct vnode *vp, *mvp, *devvp;
525 	struct inode *ip;
526 	struct buf *bp;
527 	struct ext2fs *es;
528 	struct m_ext2fs *fs;
529 	struct csum *sump;
530 	int error, i;
531 	int32_t *lp;
532 
533 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
534 		return (EINVAL);
535 	/*
536 	 * Step 1: invalidate all cached meta-data.
537 	 */
538 	devvp = VFSTOEXT2(mp)->um_devvp;
539 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
540 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
541 		panic("ext2_reload: dirty1");
542 	VOP_UNLOCK(devvp, 0);
543 
544 	/*
545 	 * Step 2: re-read superblock from disk.
546 	 * constants have been adjusted for ext2
547 	 */
548 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
549 		return (error);
550 	es = (struct ext2fs *)bp->b_data;
551 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
552 		brelse(bp);
553 		return (EIO);		/* XXX needs translation */
554 	}
555 	fs = VFSTOEXT2(mp)->um_e2fs;
556 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
557 
558 	if ((error = compute_sb_data(devvp, es, fs)) != 0) {
559 		brelse(bp);
560 		return (error);
561 	}
562 #ifdef UNKLAR
563 	if (fs->fs_sbsize < SBSIZE)
564 		bp->b_flags |= B_INVAL;
565 #endif
566 	brelse(bp);
567 
568 	/*
569 	 * Step 3: invalidate all cluster summary information.
570 	 */
571 	if (fs->e2fs_contigsumsize > 0) {
572 		lp = fs->e2fs_maxcluster;
573 		sump = fs->e2fs_clustersum;
574 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
575 			*lp++ = fs->e2fs_contigsumsize;
576 			sump->cs_init = 0;
577 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
578 		}
579 	}
580 
581 loop:
582 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
583 		/*
584 		 * Step 4: invalidate all cached file data.
585 		 */
586 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
587 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
588 			goto loop;
589 		}
590 		if (vinvalbuf(vp, 0, 0, 0))
591 			panic("ext2_reload: dirty2");
592 
593 		/*
594 		 * Step 5: re-read inode data for all active vnodes.
595 		 */
596 		ip = VTOI(vp);
597 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
598 		    (int)fs->e2fs_bsize, NOCRED, &bp);
599 		if (error) {
600 			VOP_UNLOCK(vp, 0);
601 			vrele(vp);
602 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
603 			return (error);
604 		}
605 		ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
606 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
607 		brelse(bp);
608 		VOP_UNLOCK(vp, 0);
609 		vrele(vp);
610 	}
611 	return (0);
612 }
613 
614 /*
615  * Common code for mount and mountroot.
616  */
617 static int
618 ext2_mountfs(struct vnode *devvp, struct mount *mp)
619 {
620 	struct ext2mount *ump;
621 	struct buf *bp;
622 	struct m_ext2fs *fs;
623 	struct ext2fs *es;
624 	struct cdev *dev = devvp->v_rdev;
625 	struct g_consumer *cp;
626 	struct bufobj *bo;
627 	struct csum *sump;
628 	int error;
629 	int ronly;
630 	int i;
631 	u_long size;
632 	int32_t *lp;
633 	int32_t e2fs_maxcontig;
634 
635 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
636 	/* XXX: use VOP_ACESS to check FS perms */
637 	g_topology_lock();
638 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
639 	g_topology_unlock();
640 	VOP_UNLOCK(devvp, 0);
641 	if (error)
642 		return (error);
643 
644 	/* XXX: should we check for some sectorsize or 512 instead? */
645 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
646 	    (SBSIZE < cp->provider->sectorsize)) {
647 		g_topology_lock();
648 		g_vfs_close(cp);
649 		g_topology_unlock();
650 		return (EINVAL);
651 	}
652 
653 	bo = &devvp->v_bufobj;
654 	bo->bo_private = cp;
655 	bo->bo_ops = g_vfs_bufops;
656 	if (devvp->v_rdev->si_iosize_max != 0)
657 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
658 	if (mp->mnt_iosize_max > MAXPHYS)
659 		mp->mnt_iosize_max = MAXPHYS;
660 
661 	bp = NULL;
662 	ump = NULL;
663 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
664 		goto out;
665 	es = (struct ext2fs *)bp->b_data;
666 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
667 		error = EINVAL;		/* XXX needs translation */
668 		goto out;
669 	}
670 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
671 	    (es->e2fs_state & E2FS_ERRORS)) {
672 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
673 			printf(
674 "WARNING: Filesystem was not properly dismounted\n");
675 		} else {
676 			printf(
677 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
678 			error = EPERM;
679 			goto out;
680 		}
681 	}
682 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
683 
684 	/*
685 	 * I don't know whether this is the right strategy. Note that
686 	 * we dynamically allocate both an m_ext2fs and an ext2fs
687 	 * while Linux keeps the super block in a locked buffer.
688 	 */
689 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
690 	    M_EXT2MNT, M_WAITOK | M_ZERO);
691 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
692 	    M_EXT2MNT, M_WAITOK);
693 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
694 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
695 	if ((error = compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
696 		goto out;
697 
698 	/*
699 	 * Calculate the maximum contiguous blocks and size of cluster summary
700 	 * array.  In FFS this is done by newfs; however, the superblock
701 	 * in ext2fs doesn't have these variables, so we can calculate
702 	 * them here.
703 	 */
704 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
705 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
706 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
707 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
708 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
709 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
710 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
711 		lp = ump->um_e2fs->e2fs_maxcluster;
712 		sump = ump->um_e2fs->e2fs_clustersum;
713 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
714 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
715 			sump->cs_init = 0;
716 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
717 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
718 		}
719 	}
720 
721 	brelse(bp);
722 	bp = NULL;
723 	fs = ump->um_e2fs;
724 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
725 
726 	/*
727 	 * If the fs is not mounted read-only, make sure the super block is
728 	 * always written back on a sync().
729 	 */
730 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
731 	if (ronly == 0) {
732 		fs->e2fs_fmod = 1;	/* mark it modified */
733 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
734 	}
735 	mp->mnt_data = ump;
736 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
737 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
738 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
739 	MNT_ILOCK(mp);
740 	mp->mnt_flag |= MNT_LOCAL;
741 	MNT_IUNLOCK(mp);
742 	ump->um_mountp = mp;
743 	ump->um_dev = dev;
744 	ump->um_devvp = devvp;
745 	ump->um_bo = &devvp->v_bufobj;
746 	ump->um_cp = cp;
747 
748 	/*
749 	 * Setting those two parameters allowed us to use
750 	 * ufs_bmap w/o changse!
751 	 */
752 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
753 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
754 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
755 	if (ronly == 0)
756 		ext2_sbupdate(ump, MNT_WAIT);
757 	/*
758 	 * Initialize filesystem stat information in mount struct.
759 	 */
760 	MNT_ILOCK(mp);
761 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
762 	    MNTK_USES_BCACHE;
763 	MNT_IUNLOCK(mp);
764 	return (0);
765 out:
766 	if (bp)
767 		brelse(bp);
768 	if (cp != NULL) {
769 		g_topology_lock();
770 		g_vfs_close(cp);
771 		g_topology_unlock();
772 	}
773 	if (ump) {
774 		mtx_destroy(EXT2_MTX(ump));
775 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
776 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
777 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
778 		free(ump->um_e2fs, M_EXT2MNT);
779 		free(ump, M_EXT2MNT);
780 		mp->mnt_data = NULL;
781 	}
782 	return (error);
783 }
784 
785 /*
786  * Unmount system call.
787  */
788 static int
789 ext2_unmount(struct mount *mp, int mntflags)
790 {
791 	struct ext2mount *ump;
792 	struct m_ext2fs *fs;
793 	struct csum *sump;
794 	int error, flags, i, ronly;
795 
796 	flags = 0;
797 	if (mntflags & MNT_FORCE) {
798 		if (mp->mnt_flag & MNT_ROOTFS)
799 			return (EINVAL);
800 		flags |= FORCECLOSE;
801 	}
802 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
803 		return (error);
804 	ump = VFSTOEXT2(mp);
805 	fs = ump->um_e2fs;
806 	ronly = fs->e2fs_ronly;
807 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
808 		if (fs->e2fs_wasvalid)
809 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
810 		ext2_sbupdate(ump, MNT_WAIT);
811 	}
812 
813 	g_topology_lock();
814 	g_vfs_close(ump->um_cp);
815 	g_topology_unlock();
816 	vrele(ump->um_devvp);
817 	sump = fs->e2fs_clustersum;
818 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
819 		free(sump->cs_sum, M_EXT2MNT);
820 	free(fs->e2fs_clustersum, M_EXT2MNT);
821 	free(fs->e2fs_maxcluster, M_EXT2MNT);
822 	free(fs->e2fs_gd, M_EXT2MNT);
823 	free(fs->e2fs_contigdirs, M_EXT2MNT);
824 	free(fs->e2fs, M_EXT2MNT);
825 	free(fs, M_EXT2MNT);
826 	free(ump, M_EXT2MNT);
827 	mp->mnt_data = NULL;
828 	MNT_ILOCK(mp);
829 	mp->mnt_flag &= ~MNT_LOCAL;
830 	MNT_IUNLOCK(mp);
831 	return (error);
832 }
833 
834 /*
835  * Flush out all the files in a filesystem.
836  */
837 static int
838 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
839 {
840 	int error;
841 
842 	error = vflush(mp, 0, flags, td);
843 	return (error);
844 }
845 
846 /*
847  * Get filesystem statistics.
848  */
849 int
850 ext2_statfs(struct mount *mp, struct statfs *sbp)
851 {
852 	struct ext2mount *ump;
853 	struct m_ext2fs *fs;
854 	uint32_t overhead, overhead_per_group, ngdb;
855 	int i, ngroups;
856 
857 	ump = VFSTOEXT2(mp);
858 	fs = ump->um_e2fs;
859 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
860 		panic("ext2_statfs");
861 
862 	/*
863 	 * Compute the overhead (FS structures)
864 	 */
865 	overhead_per_group =
866 	    1 /* block bitmap */ +
867 	    1 /* inode bitmap */ +
868 	    fs->e2fs_itpg;
869 	overhead = fs->e2fs->e2fs_first_dblock +
870 	    fs->e2fs_gcount * overhead_per_group;
871 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
872 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
873 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
874 			if (ext2_cg_has_sb(fs, i))
875 				ngroups++;
876 		}
877 	} else {
878 		ngroups = fs->e2fs_gcount;
879 	}
880 	ngdb = fs->e2fs_gdbcount;
881 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
882 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
883 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
884 	overhead += ngroups * (1 /* superblock */ + ngdb);
885 
886 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
887 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
888 	sbp->f_blocks = fs->e2fs_bcount - overhead;
889 	sbp->f_bfree = fs->e2fs_fbcount;
890 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
891 	sbp->f_files = fs->e2fs->e2fs_icount;
892 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
893 	return (0);
894 }
895 
896 /*
897  * Go through the disk queues to initiate sandbagged IO;
898  * go through the inodes to write those that have been modified;
899  * initiate the writing of the super block if it has been modified.
900  *
901  * Note: we are always called with the filesystem marked `MPBUSY'.
902  */
903 static int
904 ext2_sync(struct mount *mp, int waitfor)
905 {
906 	struct vnode *mvp, *vp;
907 	struct thread *td;
908 	struct inode *ip;
909 	struct ext2mount *ump = VFSTOEXT2(mp);
910 	struct m_ext2fs *fs;
911 	int error, allerror = 0;
912 
913 	td = curthread;
914 	fs = ump->um_e2fs;
915 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
916 		printf("fs = %s\n", fs->e2fs_fsmnt);
917 		panic("ext2_sync: rofs mod");
918 	}
919 
920 	/*
921 	 * Write back each (modified) inode.
922 	 */
923 loop:
924 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
925 		if (vp->v_type == VNON) {
926 			VI_UNLOCK(vp);
927 			continue;
928 		}
929 		ip = VTOI(vp);
930 		if ((ip->i_flag &
931 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
932 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
933 		    waitfor == MNT_LAZY)) {
934 			VI_UNLOCK(vp);
935 			continue;
936 		}
937 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
938 		if (error) {
939 			if (error == ENOENT) {
940 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
941 				goto loop;
942 			}
943 			continue;
944 		}
945 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
946 			allerror = error;
947 		VOP_UNLOCK(vp, 0);
948 		vrele(vp);
949 	}
950 
951 	/*
952 	 * Force stale filesystem control information to be flushed.
953 	 */
954 	if (waitfor != MNT_LAZY) {
955 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
956 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
957 			allerror = error;
958 		VOP_UNLOCK(ump->um_devvp, 0);
959 	}
960 
961 	/*
962 	 * Write back modified superblock.
963 	 */
964 	if (fs->e2fs_fmod != 0) {
965 		fs->e2fs_fmod = 0;
966 		fs->e2fs->e2fs_wtime = time_second;
967 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
968 			allerror = error;
969 	}
970 	return (allerror);
971 }
972 
973 /*
974  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
975  * in from disk.  If it is in core, wait for the lock bit to clear, then
976  * return the inode locked.  Detection and handling of mount points must be
977  * done by the calling routine.
978  */
979 static int
980 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
981 {
982 	struct m_ext2fs *fs;
983 	struct inode *ip;
984 	struct ext2mount *ump;
985 	struct buf *bp;
986 	struct vnode *vp;
987 	struct thread *td;
988 	int i, error;
989 	int used_blocks;
990 
991 	td = curthread;
992 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
993 	if (error || *vpp != NULL)
994 		return (error);
995 
996 	ump = VFSTOEXT2(mp);
997 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
998 
999 	/* Allocate a new vnode/inode. */
1000 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
1001 		*vpp = NULL;
1002 		free(ip, M_EXT2NODE);
1003 		return (error);
1004 	}
1005 	vp->v_data = ip;
1006 	ip->i_vnode = vp;
1007 	ip->i_e2fs = fs = ump->um_e2fs;
1008 	ip->i_ump = ump;
1009 	ip->i_number = ino;
1010 
1011 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1012 	error = insmntque(vp, mp);
1013 	if (error != 0) {
1014 		free(ip, M_EXT2NODE);
1015 		*vpp = NULL;
1016 		return (error);
1017 	}
1018 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
1019 	if (error || *vpp != NULL)
1020 		return (error);
1021 
1022 	/* Read in the disk contents for the inode, copy into the inode. */
1023 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1024 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
1025 		/*
1026 		 * The inode does not contain anything useful, so it would
1027 		 * be misleading to leave it on its hash chain. With mode
1028 		 * still zero, it will be unlinked and returned to the free
1029 		 * list by vput().
1030 		 */
1031 		brelse(bp);
1032 		vput(vp);
1033 		*vpp = NULL;
1034 		return (error);
1035 	}
1036 	/* convert ext2 inode to dinode */
1037 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1038 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1039 	if (error) {
1040 		printf("ext2fs: Bad inode %lu csum - run fsck\n",
1041 		    (unsigned long)ino);
1042 		brelse(bp);
1043 		vput(vp);
1044 		*vpp = NULL;
1045 		return (error);
1046 	}
1047 	ip->i_block_group = ino_to_cg(fs, ino);
1048 	ip->i_next_alloc_block = 0;
1049 	ip->i_next_alloc_goal = 0;
1050 
1051 	/*
1052 	 * Now we want to make sure that block pointers for unused
1053 	 * blocks are zeroed out - ext2_balloc depends on this
1054 	 * although for regular files and directories only
1055 	 *
1056 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1057 	 * out because we could corrupt the extent tree.
1058 	 */
1059 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1060 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1061 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1062 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1063 			ip->i_db[i] = 0;
1064 	}
1065 #ifdef EXT2FS_DEBUG
1066 	ext2_print_inode(ip);
1067 	ext4_ext_print_extent_tree_status(ip);
1068 #endif
1069 	bqrelse(bp);
1070 
1071 	/*
1072 	 * Initialize the vnode from the inode, check for aliases.
1073 	 * Note that the underlying vnode may have changed.
1074 	 */
1075 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1076 		vput(vp);
1077 		*vpp = NULL;
1078 		return (error);
1079 	}
1080 
1081 	/*
1082 	 * Finish inode initialization.
1083 	 */
1084 
1085 	*vpp = vp;
1086 	return (0);
1087 }
1088 
1089 /*
1090  * File handle to vnode
1091  *
1092  * Have to be really careful about stale file handles:
1093  * - check that the inode number is valid
1094  * - call ext2_vget() to get the locked inode
1095  * - check for an unallocated inode (i_mode == 0)
1096  * - check that the given client host has export rights and return
1097  *   those rights via. exflagsp and credanonp
1098  */
1099 static int
1100 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1101 {
1102 	struct inode *ip;
1103 	struct ufid *ufhp;
1104 	struct vnode *nvp;
1105 	struct m_ext2fs *fs;
1106 	int error;
1107 
1108 	ufhp = (struct ufid *)fhp;
1109 	fs = VFSTOEXT2(mp)->um_e2fs;
1110 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1111 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1112 		return (ESTALE);
1113 
1114 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1115 	if (error) {
1116 		*vpp = NULLVP;
1117 		return (error);
1118 	}
1119 	ip = VTOI(nvp);
1120 	if (ip->i_mode == 0 ||
1121 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1122 		vput(nvp);
1123 		*vpp = NULLVP;
1124 		return (ESTALE);
1125 	}
1126 	*vpp = nvp;
1127 	vnode_create_vobject(*vpp, 0, curthread);
1128 	return (0);
1129 }
1130 
1131 /*
1132  * Write a superblock and associated information back to disk.
1133  */
1134 static int
1135 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1136 {
1137 	struct m_ext2fs *fs = mp->um_e2fs;
1138 	struct ext2fs *es = fs->e2fs;
1139 	struct buf *bp;
1140 	int error = 0;
1141 
1142 	es->e2fs_bcount = fs->e2fs_bcount & 0xffffffff;
1143 	es->e2fs_rbcount = fs->e2fs_rbcount & 0xffffffff;
1144 	es->e2fs_fbcount = fs->e2fs_fbcount & 0xffffffff;
1145 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1146 		es->e4fs_bcount_hi = fs->e2fs_bcount >> 32;
1147 		es->e4fs_rbcount_hi = fs->e2fs_rbcount >> 32;
1148 		es->e4fs_fbcount_hi = fs->e2fs_fbcount >> 32;
1149 	}
1150 
1151 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1152 		ext2_sb_csum_set(fs);
1153 
1154 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1155 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1156 	if (waitfor == MNT_WAIT)
1157 		error = bwrite(bp);
1158 	else
1159 		bawrite(bp);
1160 
1161 	/*
1162 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1163 	 * are not busy at this point and are (hopefully) written by the
1164 	 * usual sync mechanism. No need to write them here.
1165 	 */
1166 	return (error);
1167 }
1168 int
1169 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1170 {
1171 	struct m_ext2fs *fs = mp->um_e2fs;
1172 	struct buf *bp;
1173 	int i, j, g_count = 0, error = 0, allerror = 0;
1174 
1175 	allerror = ext2_sbupdate(mp, waitfor);
1176 
1177 	/* Update gd csums */
1178 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1179 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1180 		ext2_gd_csum_set(fs);
1181 
1182 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1183 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1184 		    cg_location(fs, i)),
1185 		    fs->e2fs_bsize, 0, 0, 0);
1186 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1187 			memcpy(bp->b_data, &fs->e2fs_gd[
1188 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1189 			    fs->e2fs_bsize);
1190 		} else {
1191 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1192 			    g_count < fs->e2fs_gcount; j++, g_count++)
1193 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1194 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1195 		}
1196 		if (waitfor == MNT_WAIT)
1197 			error = bwrite(bp);
1198 		else
1199 			bawrite(bp);
1200 	}
1201 
1202 	if (!allerror && error)
1203 		allerror = error;
1204 	return (allerror);
1205 }
1206 
1207 /*
1208  * Return the root of a filesystem.
1209  */
1210 static int
1211 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1212 {
1213 	struct vnode *nvp;
1214 	int error;
1215 
1216 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1217 	if (error)
1218 		return (error);
1219 	*vpp = nvp;
1220 	return (0);
1221 }
1222