xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision 7899f917b1c0ea178f1d2be0cfb452086d079d23)
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 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/namei.h>
41 #include <sys/priv.h>
42 #include <sys/proc.h>
43 #include <sys/kernel.h>
44 #include <sys/vnode.h>
45 #include <sys/mount.h>
46 #include <sys/bio.h>
47 #include <sys/buf.h>
48 #include <sys/conf.h>
49 #include <sys/endian.h>
50 #include <sys/fcntl.h>
51 #include <sys/malloc.h>
52 #include <sys/sdt.h>
53 #include <sys/stat.h>
54 #include <sys/mutex.h>
55 
56 #include <geom/geom.h>
57 #include <geom/geom_vfs.h>
58 
59 #include <fs/ext2fs/fs.h>
60 #include <fs/ext2fs/ext2_mount.h>
61 #include <fs/ext2fs/inode.h>
62 
63 #include <fs/ext2fs/ext2fs.h>
64 #include <fs/ext2fs/ext2_dinode.h>
65 #include <fs/ext2fs/ext2_extern.h>
66 #include <fs/ext2fs/ext2_extents.h>
67 
68 SDT_PROVIDER_DECLARE(ext2fs);
69 /*
70  * ext2fs trace probe:
71  * arg0: verbosity. Higher numbers give more verbose messages
72  * arg1: Textual message
73  */
74 SDT_PROBE_DEFINE2(ext2fs, , vfsops, trace, "int", "char*");
75 SDT_PROBE_DEFINE2(ext2fs, , vfsops, ext2_cg_validate_error, "char*", "int");
76 SDT_PROBE_DEFINE1(ext2fs, , vfsops, ext2_compute_sb_data_error, "char*");
77 
78 static int	ext2_flushfiles(struct mount *mp, int flags, struct thread *td);
79 static int	ext2_mountfs(struct vnode *, struct mount *);
80 static int	ext2_reload(struct mount *mp, struct thread *td);
81 static int	ext2_sbupdate(struct ext2mount *, int);
82 static int	ext2_cgupdate(struct ext2mount *, int);
83 static vfs_unmount_t		ext2_unmount;
84 static vfs_root_t		ext2_root;
85 static vfs_statfs_t		ext2_statfs;
86 static vfs_sync_t		ext2_sync;
87 static vfs_vget_t		ext2_vget;
88 static vfs_fhtovp_t		ext2_fhtovp;
89 static vfs_mount_t		ext2_mount;
90 
91 MALLOC_DEFINE(M_EXT2NODE, "ext2_node", "EXT2 vnode private part");
92 static MALLOC_DEFINE(M_EXT2MNT, "ext2_mount", "EXT2 mount structure");
93 
94 static struct vfsops ext2fs_vfsops = {
95 	.vfs_fhtovp =		ext2_fhtovp,
96 	.vfs_mount =		ext2_mount,
97 	.vfs_root =		ext2_root,	/* root inode via vget */
98 	.vfs_statfs =		ext2_statfs,
99 	.vfs_sync =		ext2_sync,
100 	.vfs_unmount =		ext2_unmount,
101 	.vfs_vget =		ext2_vget,
102 };
103 
104 VFS_SET(ext2fs_vfsops, ext2fs, 0);
105 
106 static int	ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev,
107 		    int ronly);
108 static int	ext2_compute_sb_data(struct vnode * devvp,
109 		    struct ext2fs * es, struct m_ext2fs * fs);
110 
111 static const char *ext2_opts[] = { "acls", "async", "noatime", "noclusterr",
112     "noclusterw", "noexec", "export", "force", "from", "multilabel",
113     "suiddir", "nosymfollow", "sync", "union", NULL };
114 
115 /*
116  * VFS Operations.
117  *
118  * mount system call
119  */
120 static int
121 ext2_mount(struct mount *mp)
122 {
123 	struct vfsoptlist *opts;
124 	struct vnode *devvp;
125 	struct thread *td;
126 	struct ext2mount *ump = NULL;
127 	struct m_ext2fs *fs;
128 	struct nameidata nd, *ndp = &nd;
129 	accmode_t accmode;
130 	char *path, *fspec;
131 	int error, flags, len;
132 
133 	td = curthread;
134 	opts = mp->mnt_optnew;
135 
136 	if (vfs_filteropt(opts, ext2_opts))
137 		return (EINVAL);
138 
139 	vfs_getopt(opts, "fspath", (void **)&path, NULL);
140 	/* Double-check the length of path.. */
141 	if (strlen(path) >= MAXMNTLEN)
142 		return (ENAMETOOLONG);
143 
144 	fspec = NULL;
145 	error = vfs_getopt(opts, "from", (void **)&fspec, &len);
146 	if (!error && fspec[len - 1] != '\0')
147 		return (EINVAL);
148 
149 	/*
150 	 * If updating, check whether changing from read-only to
151 	 * read/write; if there is no device name, that's all we do.
152 	 */
153 	if (mp->mnt_flag & MNT_UPDATE) {
154 		ump = VFSTOEXT2(mp);
155 		fs = ump->um_e2fs;
156 		error = 0;
157 		if (fs->e2fs_ronly == 0 &&
158 		    vfs_flagopt(opts, "ro", NULL, 0)) {
159 			error = VFS_SYNC(mp, MNT_WAIT);
160 			if (error)
161 				return (error);
162 			flags = WRITECLOSE;
163 			if (mp->mnt_flag & MNT_FORCE)
164 				flags |= FORCECLOSE;
165 			error = ext2_flushfiles(mp, flags, td);
166 			if (error == 0 && fs->e2fs_wasvalid &&
167 			    ext2_cgupdate(ump, MNT_WAIT) == 0) {
168 				fs->e2fs->e2fs_state =
169 				    htole16((le16toh(fs->e2fs->e2fs_state) |
170 				    E2FS_ISCLEAN));
171 				ext2_sbupdate(ump, MNT_WAIT);
172 			}
173 			fs->e2fs_ronly = 1;
174 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
175 			g_topology_lock();
176 			g_access(ump->um_cp, 0, -1, 0);
177 			g_topology_unlock();
178 		}
179 		if (!error && (mp->mnt_flag & MNT_RELOAD))
180 			error = ext2_reload(mp, td);
181 		if (error)
182 			return (error);
183 		devvp = ump->um_devvp;
184 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
185 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
186 				return (EPERM);
187 
188 			/*
189 			 * If upgrade to read-write by non-root, then verify
190 			 * that user has necessary permissions on the device.
191 			 */
192 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
193 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
194 			    td->td_ucred, td);
195 			if (error)
196 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
197 			if (error) {
198 				VOP_UNLOCK(devvp);
199 				return (error);
200 			}
201 			VOP_UNLOCK(devvp);
202 			g_topology_lock();
203 			error = g_access(ump->um_cp, 0, 1, 0);
204 			g_topology_unlock();
205 			if (error)
206 				return (error);
207 
208 			if ((le16toh(fs->e2fs->e2fs_state) & E2FS_ISCLEAN) == 0 ||
209 			    (le16toh(fs->e2fs->e2fs_state) & E2FS_ERRORS)) {
210 				if (mp->mnt_flag & MNT_FORCE) {
211 					printf(
212 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
213 				} else {
214 					printf(
215 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
216 					    fs->e2fs_fsmnt);
217 					return (EPERM);
218 				}
219 			}
220 			fs->e2fs->e2fs_state =
221 			    htole16(le16toh(fs->e2fs->e2fs_state) & ~E2FS_ISCLEAN);
222 			(void)ext2_cgupdate(ump, MNT_WAIT);
223 			fs->e2fs_ronly = 0;
224 			MNT_ILOCK(mp);
225 			mp->mnt_flag &= ~MNT_RDONLY;
226 			MNT_IUNLOCK(mp);
227 		}
228 		if (vfs_flagopt(opts, "export", NULL, 0)) {
229 			/* Process export requests in vfs_mount.c. */
230 			return (error);
231 		}
232 	}
233 
234 	/*
235 	 * Not an update, or updating the name: look up the name
236 	 * and verify that it refers to a sensible disk device.
237 	 */
238 	if (fspec == NULL)
239 		return (EINVAL);
240 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec);
241 	if ((error = namei(ndp)) != 0)
242 		return (error);
243 	NDFREE_PNBUF(ndp);
244 	devvp = ndp->ni_vp;
245 
246 	if (!vn_isdisk_error(devvp, &error)) {
247 		vput(devvp);
248 		return (error);
249 	}
250 
251 	/*
252 	 * If mount by non-root, then verify that user has necessary
253 	 * permissions on the device.
254 	 *
255 	 * XXXRW: VOP_ACCESS() enough?
256 	 */
257 	accmode = VREAD;
258 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
259 		accmode |= VWRITE;
260 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
261 	if (error)
262 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
263 	if (error) {
264 		vput(devvp);
265 		return (error);
266 	}
267 
268 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
269 		error = ext2_mountfs(devvp, mp);
270 	} else {
271 		if (devvp != ump->um_devvp) {
272 			vput(devvp);
273 			return (EINVAL);	/* needs translation */
274 		} else
275 			vput(devvp);
276 	}
277 	if (error) {
278 		vrele(devvp);
279 		return (error);
280 	}
281 	ump = VFSTOEXT2(mp);
282 	fs = ump->um_e2fs;
283 
284 	/*
285 	 * Note that this strncpy() is ok because of a check at the start
286 	 * of ext2_mount().
287 	 */
288 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
289 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
290 	vfs_mountedfrom(mp, fspec);
291 	return (0);
292 }
293 
294 static int
295 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
296 {
297 	uint32_t i, mask;
298 
299 	if (le16toh(es->e2fs_magic) != E2FS_MAGIC) {
300 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
301 		    devtoname(dev), le16toh(es->e2fs_magic), E2FS_MAGIC);
302 		return (1);
303 	}
304 	if (le32toh(es->e2fs_rev) > E2FS_REV0) {
305 		mask = le32toh(es->e2fs_features_incompat) & ~(EXT2F_INCOMPAT_SUPP);
306 		if (mask) {
307 			printf("WARNING: mount of %s denied due to "
308 			    "unsupported optional features:\n", devtoname(dev));
309 			for (i = 0;
310 			    i < sizeof(incompat)/sizeof(struct ext2_feature);
311 			    i++)
312 				if (mask & incompat[i].mask)
313 					printf("%s ", incompat[i].name);
314 			printf("\n");
315 			return (1);
316 		}
317 		mask = le32toh(es->e2fs_features_rocompat) & ~EXT2F_ROCOMPAT_SUPP;
318 		if (!ronly && mask) {
319 			printf("WARNING: R/W mount of %s denied due to "
320 			    "unsupported optional features:\n", devtoname(dev));
321 			for (i = 0;
322 			    i < sizeof(ro_compat)/sizeof(struct ext2_feature);
323 			    i++)
324 				if (mask & ro_compat[i].mask)
325 					printf("%s ", ro_compat[i].name);
326 			printf("\n");
327 			return (1);
328 		}
329 	}
330 	return (0);
331 }
332 
333 static e4fs_daddr_t
334 ext2_cg_location(struct m_ext2fs *fs, int number)
335 {
336 	int cg, descpb, logical_sb, has_super = 0;
337 
338 	/*
339 	 * Adjust logical superblock block number.
340 	 * Godmar thinks: if the blocksize is greater than 1024, then
341 	 * the superblock is logically part of block zero.
342 	 */
343 	logical_sb = fs->e2fs_bsize > SBLOCKSIZE ? 0 : 1;
344 
345 	if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
346 	    number < le32toh(fs->e2fs->e3fs_first_meta_bg))
347 		return (logical_sb + number + 1);
348 
349 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT))
350 		descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
351 	else
352 		descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
353 
354 	cg = descpb * number;
355 
356 	if (ext2_cg_has_sb(fs, cg))
357 		has_super = 1;
358 
359 	return (has_super + cg * (e4fs_daddr_t)EXT2_BLOCKS_PER_GROUP(fs) +
360 	    le32toh(fs->e2fs->e2fs_first_dblock));
361 }
362 
363 static int
364 ext2_cg_validate(struct m_ext2fs *fs)
365 {
366 	uint64_t b_bitmap;
367 	uint64_t i_bitmap;
368 	uint64_t i_tables;
369 	uint64_t first_block, last_block, last_cg_block;
370 	struct ext2_gd *gd;
371 	unsigned int i, cg_count;
372 
373 	first_block = le32toh(fs->e2fs->e2fs_first_dblock);
374 	last_cg_block = ext2_cg_number_gdb(fs, 0);
375 	cg_count = fs->e2fs_gcount;
376 
377 	for (i = 0; i < fs->e2fs_gcount; i++) {
378 		gd = &fs->e2fs_gd[i];
379 
380 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
381 		    i == fs->e2fs_gcount - 1) {
382 			last_block = fs->e2fs_bcount - 1;
383 		} else {
384 			last_block = first_block +
385 			    (EXT2_BLOCKS_PER_GROUP(fs) - 1);
386 		}
387 
388 		if ((cg_count == fs->e2fs_gcount) &&
389 		    !(le16toh(gd->ext4bgd_flags) & EXT2_BG_INODE_ZEROED))
390 			cg_count = i;
391 
392 		b_bitmap = e2fs_gd_get_b_bitmap(gd);
393 		if (b_bitmap == 0) {
394 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
395 			    "block bitmap is zero", i);
396 			return (EINVAL);
397 		}
398 		if (b_bitmap <= last_cg_block) {
399 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
400 			    "block bitmap overlaps gds", i);
401 			return (EINVAL);
402 		}
403 		if (b_bitmap < first_block || b_bitmap > last_block) {
404 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
405 			    "block bitmap not in group", i);
406 			return (EINVAL);
407 		}
408 
409 		i_bitmap = e2fs_gd_get_i_bitmap(gd);
410 		if (i_bitmap == 0) {
411 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
412 			    "inode bitmap is zero", i);
413 			return (EINVAL);
414 		}
415 		if (i_bitmap <= last_cg_block) {
416 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
417 			    "inode bitmap overlaps gds", i);
418 			return (EINVAL);
419 		}
420 		if (i_bitmap < first_block || i_bitmap > last_block) {
421 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
422 			    "inode bitmap not in group blk", i);
423 			return (EINVAL);
424 		}
425 
426 		i_tables = e2fs_gd_get_i_tables(gd);
427 		if (i_tables == 0) {
428 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
429 			    "inode table is zero", i);
430 			return (EINVAL);
431 		}
432 		if (i_tables <= last_cg_block) {
433 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
434 			    "inode tables overlaps gds", i);
435 			return (EINVAL);
436 		}
437 		if (i_tables < first_block ||
438 		    i_tables + fs->e2fs_itpg - 1 > last_block) {
439 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
440 			    "inode tables not in group blk", i);
441 			return (EINVAL);
442 		}
443 
444 		if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG))
445 			first_block += EXT2_BLOCKS_PER_GROUP(fs);
446 	}
447 
448 	return (0);
449 }
450 
451 /*
452  * This computes the fields of the m_ext2fs structure from the
453  * data in the ext2fs structure read in.
454  */
455 static int
456 ext2_compute_sb_data(struct vnode *devvp, struct ext2fs *es,
457     struct m_ext2fs *fs)
458 {
459 	struct buf *bp;
460 	uint32_t e2fs_descpb, e2fs_gdbcount_alloc;
461 	int i, j;
462 	int g_count = 0;
463 	int error;
464 
465 	/* Check checksum features */
466 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) &&
467 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
468 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
469 		    "incorrect checksum features combination");
470 		return (EINVAL);
471 	}
472 
473 	/* Precompute checksum seed for all metadata */
474 	ext2_sb_csum_set_seed(fs);
475 
476 	/* Verify sb csum if possible */
477 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
478 		error = ext2_sb_csum_verify(fs);
479 		if (error) {
480 			return (error);
481 		}
482 	}
483 
484 	/* Check for block size = 1K|2K|4K */
485 	if (le32toh(es->e2fs_log_bsize) > 2) {
486 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
487 		    "bad block size");
488 		return (EINVAL);
489 	}
490 
491 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + le32toh(es->e2fs_log_bsize);
492 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
493 	fs->e2fs_fsbtodb = le32toh(es->e2fs_log_bsize) + 1;
494 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
495 
496 	/* Check for fragment size */
497 	if (le32toh(es->e2fs_log_fsize) >
498 	    (EXT2_MAX_FRAG_LOG_SIZE - EXT2_MIN_BLOCK_LOG_SIZE)) {
499 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
500 		    "invalid log cluster size");
501 		return (EINVAL);
502 	}
503 
504 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << le32toh(es->e2fs_log_fsize);
505 	if (fs->e2fs_fsize != fs->e2fs_bsize) {
506 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
507 		    "fragment size != block size");
508 		return (EINVAL);
509 	}
510 
511 	fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
512 
513 	/* Check reserved gdt blocks for future filesystem expansion */
514 	if (le16toh(es->e2fs_reserved_ngdb) > (fs->e2fs_bsize / 4)) {
515 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
516 		    "number of reserved GDT blocks too large");
517 		return (EINVAL);
518 	}
519 
520 	if (le32toh(es->e2fs_rev) == E2FS_REV0) {
521 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
522 	} else {
523 		fs->e2fs_isize = le16toh(es->e2fs_inode_size);
524 
525 		/*
526 		 * Check first ino.
527 		 */
528 		if (le32toh(es->e2fs_first_ino) < EXT2_FIRSTINO) {
529 			SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
530 			    "invalid first ino");
531 			return (EINVAL);
532 		}
533 
534 		/*
535 		 * Simple sanity check for superblock inode size value.
536 		 */
537 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
538 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
539 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
540 			SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
541 			    "invalid inode size");
542 			return (EINVAL);
543 		}
544 	}
545 
546 	/* Check group descriptors */
547 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT) &&
548 	    le16toh(es->e3fs_desc_size) != E2FS_64BIT_GD_SIZE) {
549 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
550 		    "unsupported 64bit descriptor size");
551 		return (EINVAL);
552 	}
553 
554 	fs->e2fs_bpg = le32toh(es->e2fs_bpg);
555 	fs->e2fs_fpg = le32toh(es->e2fs_fpg);
556 	if (fs->e2fs_bpg == 0 || fs->e2fs_fpg == 0) {
557 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
558 		    "zero blocks/fragments per group");
559 		return (EINVAL);
560 	} else if (fs->e2fs_bpg != fs->e2fs_fpg) {
561 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
562 		    "blocks per group not equal fragments per group");
563 		return (EINVAL);
564 	}
565 
566 	if (fs->e2fs_bpg != fs->e2fs_bsize * 8) {
567 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
568 		    "non-standard group size unsupported");
569 		return (EINVAL);
570 	}
571 
572 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
573 	if (fs->e2fs_ipb == 0 ||
574 	    fs->e2fs_ipb > fs->e2fs_bsize / E2FS_REV0_INODE_SIZE) {
575 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
576 		    "bad inodes per block size");
577 		return (EINVAL);
578 	}
579 
580 	fs->e2fs_ipg = le32toh(es->e2fs_ipg);
581 	if (fs->e2fs_ipg < fs->e2fs_ipb || fs->e2fs_ipg >  fs->e2fs_bsize * 8) {
582 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
583 		    "invalid inodes per group");
584 		return (EINVAL);
585 	}
586 
587 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
588 
589 	fs->e2fs_bcount = le32toh(es->e2fs_bcount);
590 	fs->e2fs_rbcount = le32toh(es->e2fs_rbcount);
591 	fs->e2fs_fbcount = le32toh(es->e2fs_fbcount);
592 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
593 		fs->e2fs_bcount |= (uint64_t)(le32toh(es->e4fs_bcount_hi)) << 32;
594 		fs->e2fs_rbcount |= (uint64_t)(le32toh(es->e4fs_rbcount_hi)) << 32;
595 		fs->e2fs_fbcount |= (uint64_t)(le32toh(es->e4fs_fbcount_hi)) << 32;
596 	}
597 	if (fs->e2fs_rbcount > fs->e2fs_bcount ||
598 	    fs->e2fs_fbcount > fs->e2fs_bcount) {
599 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
600 		    "invalid block count");
601 		return (EINVAL);
602 	}
603 
604 	fs->e2fs_ficount = le32toh(es->e2fs_ficount);
605 	if (fs->e2fs_ficount > le32toh(es->e2fs_icount)) {
606 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
607 		    "invalid number of free inodes");
608 		return (EINVAL);
609 	}
610 
611 	if (le32toh(es->e2fs_first_dblock) != (fs->e2fs_bsize > 1024 ? 0 : 1) ||
612 	    le32toh(es->e2fs_first_dblock) >= fs->e2fs_bcount) {
613 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
614 		    "first data block out of range");
615 		return (EINVAL);
616 	}
617 
618 	fs->e2fs_gcount = howmany(fs->e2fs_bcount -
619 	    le32toh(es->e2fs_first_dblock), EXT2_BLOCKS_PER_GROUP(fs));
620 	if (fs->e2fs_gcount > ((uint64_t)1 << 32) - EXT2_DESCS_PER_BLOCK(fs)) {
621 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
622 		    "groups count too large");
623 		return (EINVAL);
624 	}
625 
626 	/* Check for extra isize in big inodes. */
627 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
628 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
629 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
630 		    "no space for extra inode timestamps");
631 		return (EINVAL);
632 	}
633 
634 	/* s_resuid / s_resgid ? */
635 
636 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
637 		e2fs_descpb = fs->e2fs_bsize / E2FS_64BIT_GD_SIZE;
638 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount, e2fs_descpb);
639 	} else {
640 		e2fs_descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
641 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount,
642 		    fs->e2fs_bsize / sizeof(struct ext2_gd));
643 	}
644 	fs->e2fs_gdbcount = howmany(fs->e2fs_gcount, e2fs_descpb);
645 	fs->e2fs_gd = malloc(e2fs_gdbcount_alloc * fs->e2fs_bsize,
646 	    M_EXT2MNT, M_WAITOK | M_ZERO);
647 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
648 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
649 
650 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
651 		error = bread(devvp,
652 		    fsbtodb(fs, ext2_cg_location(fs, i)),
653 		    fs->e2fs_bsize, NOCRED, &bp);
654 		if (error) {
655 			/*
656 			 * fs->e2fs_gd and fs->e2fs_contigdirs
657 			 * will be freed later by the caller,
658 			 * because this function could be called from
659 			 * MNT_UPDATE path.
660 			 */
661 			return (error);
662 		}
663 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
664 			memcpy(&fs->e2fs_gd[
665 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
666 			    bp->b_data, fs->e2fs_bsize);
667 		} else {
668 			for (j = 0; j < e2fs_descpb &&
669 			    g_count < fs->e2fs_gcount; j++, g_count++)
670 				memcpy(&fs->e2fs_gd[g_count],
671 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
672 				    E2FS_REV0_GD_SIZE);
673 		}
674 		brelse(bp);
675 		bp = NULL;
676 	}
677 
678 	/* Validate cgs consistency */
679 	error = ext2_cg_validate(fs);
680 	if (error)
681 		return (error);
682 
683 	/* Verfy cgs csum */
684 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
685 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
686 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
687 		if (error)
688 			return (error);
689 	}
690 	/* Initialization for the ext2 Orlov allocator variant. */
691 	fs->e2fs_total_dir = 0;
692 	for (i = 0; i < fs->e2fs_gcount; i++)
693 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
694 
695 	if (le32toh(es->e2fs_rev) == E2FS_REV0 ||
696 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
697 		fs->e2fs_maxfilesize = 0x7fffffff;
698 	else {
699 		fs->e2fs_maxfilesize = 0xffffffffffff;
700 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
701 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
702 	}
703 	if (le32toh(es->e4fs_flags) & E2FS_UNSIGNED_HASH) {
704 		fs->e2fs_uhash = 3;
705 	} else if ((le32toh(es->e4fs_flags) & E2FS_SIGNED_HASH) == 0) {
706 #ifdef __CHAR_UNSIGNED__
707 		es->e4fs_flags = htole32(le32toh(es->e4fs_flags) | E2FS_UNSIGNED_HASH);
708 		fs->e2fs_uhash = 3;
709 #else
710 		es->e4fs_flags = htole32(le32toh(es->e4fs_flags) | E2FS_SIGNED_HASH);
711 #endif
712 	}
713 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
714 		error = ext2_sb_csum_verify(fs);
715 
716 	return (error);
717 }
718 
719 /*
720  * Reload all incore data for a filesystem (used after running fsck on
721  * the root filesystem and finding things to fix). The filesystem must
722  * be mounted read-only.
723  *
724  * Things to do to update the mount:
725  *	1) invalidate all cached meta-data.
726  *	2) re-read superblock from disk.
727  *	3) invalidate all cluster summary information.
728  *	4) invalidate all inactive vnodes.
729  *	5) invalidate all cached file data.
730  *	6) re-read inode data for all active vnodes.
731  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
732  */
733 static int
734 ext2_reload(struct mount *mp, struct thread *td)
735 {
736 	struct vnode *vp, *mvp, *devvp;
737 	struct inode *ip;
738 	struct buf *bp;
739 	struct ext2fs *es;
740 	struct m_ext2fs *fs;
741 	struct csum *sump;
742 	int error, i;
743 	int32_t *lp;
744 
745 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
746 		return (EINVAL);
747 	/*
748 	 * Step 1: invalidate all cached meta-data.
749 	 */
750 	devvp = VFSTOEXT2(mp)->um_devvp;
751 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
752 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
753 		panic("ext2_reload: dirty1");
754 	VOP_UNLOCK(devvp);
755 
756 	/*
757 	 * Step 2: re-read superblock from disk.
758 	 * constants have been adjusted for ext2
759 	 */
760 	if ((error = bread(devvp, SBLOCK, SBLOCKBLKSIZE, NOCRED, &bp)) != 0)
761 		return (error);
762 	es = (struct ext2fs *)((char *)bp->b_data + SBLOCKOFFSET);
763 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
764 		brelse(bp);
765 		return (EIO);		/* XXX needs translation */
766 	}
767 	fs = VFSTOEXT2(mp)->um_e2fs;
768 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
769 
770 	if ((error = ext2_compute_sb_data(devvp, es, fs)) != 0) {
771 		brelse(bp);
772 		return (error);
773 	}
774 
775 	brelse(bp);
776 
777 	/*
778 	 * Step 3: invalidate all cluster summary information.
779 	 */
780 	if (fs->e2fs_contigsumsize > 0) {
781 		lp = fs->e2fs_maxcluster;
782 		sump = fs->e2fs_clustersum;
783 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
784 			*lp++ = fs->e2fs_contigsumsize;
785 			sump->cs_init = 0;
786 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
787 		}
788 	}
789 
790 loop:
791 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
792 		/*
793 		 * Step 4: invalidate all cached file data.
794 		 */
795 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
796 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
797 			goto loop;
798 		}
799 		if (vinvalbuf(vp, 0, 0, 0))
800 			panic("ext2_reload: dirty2");
801 
802 		/*
803 		 * Step 5: re-read inode data for all active vnodes.
804 		 */
805 		ip = VTOI(vp);
806 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
807 		    (int)fs->e2fs_bsize, NOCRED, &bp);
808 		if (error) {
809 			vput(vp);
810 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
811 			return (error);
812 		}
813 
814 		error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
815 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
816 
817 		brelse(bp);
818 		vput(vp);
819 
820 		if (error) {
821 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
822 			return (error);
823 		}
824 	}
825 	return (0);
826 }
827 
828 /*
829  * Common code for mount and mountroot.
830  */
831 static int
832 ext2_mountfs(struct vnode *devvp, struct mount *mp)
833 {
834 	struct ext2mount *ump;
835 	struct buf *bp;
836 	struct m_ext2fs *fs;
837 	struct ext2fs *es;
838 	struct cdev *dev = devvp->v_rdev;
839 	struct g_consumer *cp;
840 	struct bufobj *bo;
841 	struct csum *sump;
842 	int error;
843 	int ronly;
844 	int i;
845 	u_long size;
846 	int32_t *lp;
847 	int32_t e2fs_maxcontig;
848 
849 	bp = NULL;
850 	ump = NULL;
851 
852 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
853 	/* XXX: use VOP_ACESS to check FS perms */
854 	g_topology_lock();
855 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
856 	g_topology_unlock();
857 	VOP_UNLOCK(devvp);
858 	if (error)
859 		return (error);
860 
861 	if (PAGE_SIZE != SBLOCKBLKSIZE) {
862 		printf("WARNING: Unsupported page size %d\n", PAGE_SIZE);
863 		error = EINVAL;
864 		goto out;
865 	}
866 	if (cp->provider->sectorsize > PAGE_SIZE) {
867 		printf("WARNING: Device sectorsize(%d) is more than %d\n",
868 		    cp->provider->sectorsize, PAGE_SIZE);
869 		error = EINVAL;
870 		goto out;
871 	}
872 
873 	bo = &devvp->v_bufobj;
874 	bo->bo_private = cp;
875 	bo->bo_ops = g_vfs_bufops;
876 	if (devvp->v_rdev->si_iosize_max != 0)
877 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
878 	if (mp->mnt_iosize_max > maxphys)
879 		mp->mnt_iosize_max = maxphys;
880 	if ((error = bread(devvp, SBLOCK, SBLOCKBLKSIZE, NOCRED, &bp)) != 0)
881 		goto out;
882 	es = (struct ext2fs *)((char *)bp->b_data + SBLOCKOFFSET);
883 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
884 		error = EINVAL;		/* XXX needs translation */
885 		goto out;
886 	}
887 	if ((le16toh(es->e2fs_state) & E2FS_ISCLEAN) == 0 ||
888 	    (le16toh(es->e2fs_state) & E2FS_ERRORS)) {
889 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
890 			printf(
891 "WARNING: Filesystem was not properly dismounted\n");
892 		} else {
893 			printf(
894 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
895 			error = EPERM;
896 			goto out;
897 		}
898 	}
899 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
900 
901 	/*
902 	 * I don't know whether this is the right strategy. Note that
903 	 * we dynamically allocate both an m_ext2fs and an ext2fs
904 	 * while Linux keeps the super block in a locked buffer.
905 	 */
906 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
907 	    M_EXT2MNT, M_WAITOK | M_ZERO);
908 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
909 	    M_EXT2MNT, M_WAITOK);
910 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
911 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
912 	if ((error = ext2_compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
913 		goto out;
914 
915 	/*
916 	 * Calculate the maximum contiguous blocks and size of cluster summary
917 	 * array.  In FFS this is done by newfs; however, the superblock
918 	 * in ext2fs doesn't have these variables, so we can calculate
919 	 * them here.
920 	 */
921 	e2fs_maxcontig = MAX(1, maxphys / ump->um_e2fs->e2fs_bsize);
922 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
923 	ump->um_e2fs->e2fs_maxsymlinklen = EXT2_MAXSYMLINKLEN;
924 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
925 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
926 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
927 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
928 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
929 		lp = ump->um_e2fs->e2fs_maxcluster;
930 		sump = ump->um_e2fs->e2fs_clustersum;
931 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
932 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
933 			sump->cs_init = 0;
934 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
935 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
936 		}
937 	}
938 
939 	brelse(bp);
940 	bp = NULL;
941 	fs = ump->um_e2fs;
942 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
943 
944 	/*
945 	 * If the fs is not mounted read-only, make sure the super block is
946 	 * always written back on a sync().
947 	 */
948 	fs->e2fs_wasvalid = le16toh(fs->e2fs->e2fs_state) & E2FS_ISCLEAN ? 1 : 0;
949 	if (ronly == 0) {
950 		fs->e2fs_fmod = 1;	/* mark it modified and set fs invalid */
951 		fs->e2fs->e2fs_state =
952 		    htole16(le16toh(fs->e2fs->e2fs_state) & ~E2FS_ISCLEAN);
953 	}
954 	mp->mnt_data = ump;
955 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
956 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
957 	MNT_ILOCK(mp);
958 	mp->mnt_flag |= MNT_LOCAL;
959 	MNT_IUNLOCK(mp);
960 	ump->um_mountp = mp;
961 	ump->um_dev = dev;
962 	ump->um_devvp = devvp;
963 	ump->um_bo = &devvp->v_bufobj;
964 	ump->um_cp = cp;
965 
966 	/*
967 	 * Setting those two parameters allowed us to use
968 	 * ufs_bmap w/o changse!
969 	 */
970 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
971 	ump->um_bptrtodb = le32toh(fs->e2fs->e2fs_log_bsize) + 1;
972 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
973 	if (ronly == 0)
974 		ext2_sbupdate(ump, MNT_WAIT);
975 	/*
976 	 * Initialize filesystem stat information in mount struct.
977 	 */
978 	MNT_ILOCK(mp);
979 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
980 	    MNTK_USES_BCACHE;
981 	MNT_IUNLOCK(mp);
982 	return (0);
983 out:
984 	if (bp)
985 		brelse(bp);
986 	if (cp != NULL) {
987 		g_topology_lock();
988 		g_vfs_close(cp);
989 		g_topology_unlock();
990 	}
991 	if (ump) {
992 		mtx_destroy(EXT2_MTX(ump));
993 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
994 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
995 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
996 		free(ump->um_e2fs, M_EXT2MNT);
997 		free(ump, M_EXT2MNT);
998 		mp->mnt_data = NULL;
999 	}
1000 	return (error);
1001 }
1002 
1003 /*
1004  * Unmount system call.
1005  */
1006 static int
1007 ext2_unmount(struct mount *mp, int mntflags)
1008 {
1009 	struct ext2mount *ump;
1010 	struct m_ext2fs *fs;
1011 	struct csum *sump;
1012 	int error, flags, i, ronly;
1013 
1014 	flags = 0;
1015 	if (mntflags & MNT_FORCE) {
1016 		if (mp->mnt_flag & MNT_ROOTFS)
1017 			return (EINVAL);
1018 		flags |= FORCECLOSE;
1019 	}
1020 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
1021 		return (error);
1022 	ump = VFSTOEXT2(mp);
1023 	fs = ump->um_e2fs;
1024 	ronly = fs->e2fs_ronly;
1025 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
1026 		if (fs->e2fs_wasvalid)
1027 			fs->e2fs->e2fs_state =
1028 			    htole16(le16toh(fs->e2fs->e2fs_state) | E2FS_ISCLEAN);
1029 		ext2_sbupdate(ump, MNT_WAIT);
1030 	}
1031 
1032 	g_topology_lock();
1033 	g_vfs_close(ump->um_cp);
1034 	g_topology_unlock();
1035 	vrele(ump->um_devvp);
1036 	sump = fs->e2fs_clustersum;
1037 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
1038 		free(sump->cs_sum, M_EXT2MNT);
1039 	free(fs->e2fs_clustersum, M_EXT2MNT);
1040 	free(fs->e2fs_maxcluster, M_EXT2MNT);
1041 	free(fs->e2fs_gd, M_EXT2MNT);
1042 	free(fs->e2fs_contigdirs, M_EXT2MNT);
1043 	free(fs->e2fs, M_EXT2MNT);
1044 	free(fs, M_EXT2MNT);
1045 	free(ump, M_EXT2MNT);
1046 	mp->mnt_data = NULL;
1047 	return (error);
1048 }
1049 
1050 /*
1051  * Flush out all the files in a filesystem.
1052  */
1053 static int
1054 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
1055 {
1056 	int error;
1057 
1058 	error = vflush(mp, 0, flags, td);
1059 	return (error);
1060 }
1061 
1062 /*
1063  * Get filesystem statistics.
1064  */
1065 int
1066 ext2_statfs(struct mount *mp, struct statfs *sbp)
1067 {
1068 	struct ext2mount *ump;
1069 	struct m_ext2fs *fs;
1070 	uint32_t overhead, overhead_per_group, ngdb;
1071 	int i, ngroups;
1072 
1073 	ump = VFSTOEXT2(mp);
1074 	fs = ump->um_e2fs;
1075 	if (le16toh(fs->e2fs->e2fs_magic) != E2FS_MAGIC)
1076 		panic("ext2_statfs");
1077 
1078 	/*
1079 	 * Compute the overhead (FS structures)
1080 	 */
1081 	overhead_per_group =
1082 	    1 /* block bitmap */ +
1083 	    1 /* inode bitmap */ +
1084 	    fs->e2fs_itpg;
1085 	overhead = le32toh(fs->e2fs->e2fs_first_dblock) +
1086 	    fs->e2fs_gcount * overhead_per_group;
1087 	if (le32toh(fs->e2fs->e2fs_rev) > E2FS_REV0 &&
1088 	    le32toh(fs->e2fs->e2fs_features_rocompat) & EXT2F_ROCOMPAT_SPARSESUPER) {
1089 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
1090 			if (ext2_cg_has_sb(fs, i))
1091 				ngroups++;
1092 		}
1093 	} else {
1094 		ngroups = fs->e2fs_gcount;
1095 	}
1096 	ngdb = fs->e2fs_gdbcount;
1097 	if (le32toh(fs->e2fs->e2fs_rev) > E2FS_REV0 &&
1098 	    le32toh(fs->e2fs->e2fs_features_compat) & EXT2F_COMPAT_RESIZE)
1099 		ngdb += le16toh(fs->e2fs->e2fs_reserved_ngdb);
1100 	overhead += ngroups * (1 /* superblock */ + ngdb);
1101 
1102 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
1103 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
1104 	sbp->f_blocks = fs->e2fs_bcount - overhead;
1105 	sbp->f_bfree = fs->e2fs_fbcount;
1106 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
1107 	sbp->f_files = le32toh(fs->e2fs->e2fs_icount);
1108 	sbp->f_ffree = fs->e2fs_ficount;
1109 	return (0);
1110 }
1111 
1112 /*
1113  * Go through the disk queues to initiate sandbagged IO;
1114  * go through the inodes to write those that have been modified;
1115  * initiate the writing of the super block if it has been modified.
1116  *
1117  * Note: we are always called with the filesystem marked `MPBUSY'.
1118  */
1119 static int
1120 ext2_sync(struct mount *mp, int waitfor)
1121 {
1122 	struct vnode *mvp, *vp;
1123 	struct thread *td;
1124 	struct inode *ip;
1125 	struct ext2mount *ump = VFSTOEXT2(mp);
1126 	struct m_ext2fs *fs;
1127 	int error, allerror = 0;
1128 
1129 	td = curthread;
1130 	fs = ump->um_e2fs;
1131 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
1132 		panic("ext2_sync: rofs mod fs=%s", fs->e2fs_fsmnt);
1133 	}
1134 
1135 	/*
1136 	 * Write back each (modified) inode.
1137 	 */
1138 loop:
1139 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1140 		if (vp->v_type == VNON) {
1141 			VI_UNLOCK(vp);
1142 			continue;
1143 		}
1144 		ip = VTOI(vp);
1145 		if ((ip->i_flag &
1146 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1147 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1148 		    waitfor == MNT_LAZY)) {
1149 			VI_UNLOCK(vp);
1150 			continue;
1151 		}
1152 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK);
1153 		if (error) {
1154 			if (error == ENOENT) {
1155 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1156 				goto loop;
1157 			}
1158 			continue;
1159 		}
1160 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
1161 			allerror = error;
1162 		vput(vp);
1163 	}
1164 
1165 	/*
1166 	 * Force stale filesystem control information to be flushed.
1167 	 */
1168 	if (waitfor != MNT_LAZY) {
1169 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1170 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
1171 			allerror = error;
1172 		VOP_UNLOCK(ump->um_devvp);
1173 	}
1174 
1175 	/*
1176 	 * Write back modified superblock.
1177 	 */
1178 	if (fs->e2fs_fmod != 0) {
1179 		fs->e2fs_fmod = 0;
1180 		fs->e2fs->e2fs_wtime = htole32(time_second);
1181 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
1182 			allerror = error;
1183 	}
1184 	return (allerror);
1185 }
1186 
1187 /*
1188  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
1189  * in from disk.  If it is in core, wait for the lock bit to clear, then
1190  * return the inode locked.  Detection and handling of mount points must be
1191  * done by the calling routine.
1192  */
1193 static int
1194 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
1195 {
1196 	struct m_ext2fs *fs;
1197 	struct inode *ip;
1198 	struct ext2mount *ump;
1199 	struct buf *bp;
1200 	struct vnode *vp;
1201 	struct thread *td;
1202 	unsigned int i, used_blocks;
1203 	int error;
1204 
1205 	td = curthread;
1206 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
1207 	if (error || *vpp != NULL)
1208 		return (error);
1209 
1210 	ump = VFSTOEXT2(mp);
1211 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
1212 
1213 	/* Allocate a new vnode/inode. */
1214 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
1215 		*vpp = NULL;
1216 		free(ip, M_EXT2NODE);
1217 		return (error);
1218 	}
1219 	vp->v_data = ip;
1220 	ip->i_vnode = vp;
1221 	ip->i_e2fs = fs = ump->um_e2fs;
1222 	ip->i_ump = ump;
1223 	ip->i_number = ino;
1224 	cluster_init_vn(&ip->i_clusterw);
1225 
1226 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1227 	error = insmntque(vp, mp);
1228 	if (error != 0) {
1229 		free(ip, M_EXT2NODE);
1230 		*vpp = NULL;
1231 		return (error);
1232 	}
1233 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
1234 	if (error || *vpp != NULL)
1235 		return (error);
1236 
1237 	/* Read in the disk contents for the inode, copy into the inode. */
1238 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1239 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
1240 		/*
1241 		 * The inode does not contain anything useful, so it would
1242 		 * be misleading to leave it on its hash chain. With mode
1243 		 * still zero, it will be unlinked and returned to the free
1244 		 * list by vput().
1245 		 */
1246 		brelse(bp);
1247 		vput(vp);
1248 		*vpp = NULL;
1249 		return (error);
1250 	}
1251 	/* convert ext2 inode to dinode */
1252 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1253 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1254 	if (error) {
1255 		brelse(bp);
1256 		vput(vp);
1257 		*vpp = NULL;
1258 		return (error);
1259 	}
1260 	ip->i_block_group = ino_to_cg(fs, ino);
1261 	ip->i_next_alloc_block = 0;
1262 	ip->i_next_alloc_goal = 0;
1263 
1264 	/*
1265 	 * Now we want to make sure that block pointers for unused
1266 	 * blocks are zeroed out - ext2_balloc depends on this
1267 	 * although for regular files and directories only
1268 	 *
1269 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1270 	 * out because we could corrupt the extent tree.
1271 	 */
1272 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1273 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1274 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1275 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1276 			ip->i_db[i] = 0;
1277 	}
1278 
1279 	bqrelse(bp);
1280 
1281 #ifdef EXT2FS_PRINT_EXTENTS
1282 	ext2_print_inode(ip);
1283 	error = ext4_ext_walk(ip);
1284 	if (error) {
1285 		vput(vp);
1286 		*vpp = NULL;
1287 		return (error);
1288 	}
1289 #endif
1290 
1291 	/*
1292 	 * Initialize the vnode from the inode, check for aliases.
1293 	 * Note that the underlying vnode may have changed.
1294 	 */
1295 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1296 		vput(vp);
1297 		*vpp = NULL;
1298 		return (error);
1299 	}
1300 
1301 	/*
1302 	 * Finish inode initialization.
1303 	 */
1304 
1305 	vn_set_state(vp, VSTATE_CONSTRUCTED);
1306 	*vpp = vp;
1307 	return (0);
1308 }
1309 
1310 /*
1311  * File handle to vnode
1312  *
1313  * Have to be really careful about stale file handles:
1314  * - check that the inode number is valid
1315  * - call ext2_vget() to get the locked inode
1316  * - check for an unallocated inode (i_mode == 0)
1317  * - check that the given client host has export rights and return
1318  *   those rights via. exflagsp and credanonp
1319  */
1320 static int
1321 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1322 {
1323 	struct inode *ip;
1324 	struct ufid *ufhp;
1325 	struct vnode *nvp;
1326 	struct m_ext2fs *fs;
1327 	int error;
1328 
1329 	ufhp = (struct ufid *)fhp;
1330 	fs = VFSTOEXT2(mp)->um_e2fs;
1331 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1332 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs_ipg)
1333 		return (ESTALE);
1334 
1335 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1336 	if (error) {
1337 		*vpp = NULLVP;
1338 		return (error);
1339 	}
1340 	ip = VTOI(nvp);
1341 	if (ip->i_mode == 0 ||
1342 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1343 		vput(nvp);
1344 		*vpp = NULLVP;
1345 		return (ESTALE);
1346 	}
1347 	*vpp = nvp;
1348 	vnode_create_vobject(*vpp, ip->i_size, curthread);
1349 	return (0);
1350 }
1351 
1352 /*
1353  * Write a superblock and associated information back to disk.
1354  */
1355 static int
1356 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1357 {
1358 	struct m_ext2fs *fs = mp->um_e2fs;
1359 	struct ext2fs *es = fs->e2fs;
1360 	struct buf *bp;
1361 	int error = 0;
1362 
1363 	es->e2fs_bcount = htole32(fs->e2fs_bcount & 0xffffffff);
1364 	es->e2fs_rbcount = htole32(fs->e2fs_rbcount & 0xffffffff);
1365 	es->e2fs_fbcount = htole32(fs->e2fs_fbcount & 0xffffffff);
1366 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1367 		es->e4fs_bcount_hi = htole32(fs->e2fs_bcount >> 32);
1368 		es->e4fs_rbcount_hi = htole32(fs->e2fs_rbcount >> 32);
1369 		es->e4fs_fbcount_hi = htole32(fs->e2fs_fbcount >> 32);
1370 	}
1371 
1372 	es->e2fs_ficount = htole32(fs->e2fs_ficount);
1373 
1374 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1375 		ext2_sb_csum_set(fs);
1376 
1377 	error = bread(mp->um_devvp, SBLOCK, SBLOCKBLKSIZE, NOCRED, &bp);
1378 	if (error != 0)
1379 		return (error);
1380 
1381 	memcpy((char *)bp->b_data + SBLOCKOFFSET, (caddr_t)es,
1382 	    (u_int)sizeof(struct ext2fs));
1383 	if (waitfor == MNT_WAIT)
1384 		error = bwrite(bp);
1385 	else
1386 		bawrite(bp);
1387 
1388 	/*
1389 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1390 	 * are not busy at this point and are (hopefully) written by the
1391 	 * usual sync mechanism. No need to write them here.
1392 	 */
1393 	return (error);
1394 }
1395 int
1396 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1397 {
1398 	struct m_ext2fs *fs = mp->um_e2fs;
1399 	struct buf *bp;
1400 	int i, j, g_count = 0, error = 0, allerror = 0;
1401 
1402 	allerror = ext2_sbupdate(mp, waitfor);
1403 
1404 	/* Update gd csums */
1405 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1406 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1407 		ext2_gd_csum_set(fs);
1408 
1409 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1410 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1411 		    ext2_cg_location(fs, i)),
1412 		    fs->e2fs_bsize, 0, 0, 0);
1413 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1414 			memcpy(bp->b_data, &fs->e2fs_gd[
1415 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1416 			    fs->e2fs_bsize);
1417 		} else {
1418 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1419 			    g_count < fs->e2fs_gcount; j++, g_count++)
1420 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1421 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1422 		}
1423 		if (waitfor == MNT_WAIT)
1424 			error = bwrite(bp);
1425 		else
1426 			bawrite(bp);
1427 	}
1428 
1429 	if (!allerror && error)
1430 		allerror = error;
1431 	return (allerror);
1432 }
1433 
1434 /*
1435  * Return the root of a filesystem.
1436  */
1437 static int
1438 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1439 {
1440 	struct vnode *nvp;
1441 	int error;
1442 
1443 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1444 	if (error)
1445 		return (error);
1446 	*vpp = nvp;
1447 	return (0);
1448 }
1449