xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision 2e4311906d8c8dc7a7c726345268253bca6d4acc)
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/sdt.h>
56 #include <sys/stat.h>
57 #include <sys/mutex.h>
58 
59 #include <geom/geom.h>
60 #include <geom/geom_vfs.h>
61 
62 #include <fs/ext2fs/fs.h>
63 #include <fs/ext2fs/ext2_mount.h>
64 #include <fs/ext2fs/inode.h>
65 
66 #include <fs/ext2fs/ext2fs.h>
67 #include <fs/ext2fs/ext2_dinode.h>
68 #include <fs/ext2fs/ext2_extern.h>
69 #include <fs/ext2fs/ext2_extents.h>
70 
71 SDT_PROVIDER_DECLARE(ext2fs);
72 /*
73  * ext2fs trace probe:
74  * arg0: verbosity. Higher numbers give more verbose messages
75  * arg1: Textual message
76  */
77 SDT_PROBE_DEFINE2(ext2fs, , vfsops, trace, "int", "char*");
78 SDT_PROBE_DEFINE2(ext2fs, , vfsops, ext2_cg_validate_error, "char*", "int");
79 SDT_PROBE_DEFINE1(ext2fs, , vfsops, ext2_compute_sb_data_error, "char*");
80 
81 static int	ext2_flushfiles(struct mount *mp, int flags, struct thread *td);
82 static int	ext2_mountfs(struct vnode *, struct mount *);
83 static int	ext2_reload(struct mount *mp, struct thread *td);
84 static int	ext2_sbupdate(struct ext2mount *, int);
85 static int	ext2_cgupdate(struct ext2mount *, int);
86 static vfs_unmount_t		ext2_unmount;
87 static vfs_root_t		ext2_root;
88 static vfs_statfs_t		ext2_statfs;
89 static vfs_sync_t		ext2_sync;
90 static vfs_vget_t		ext2_vget;
91 static vfs_fhtovp_t		ext2_fhtovp;
92 static vfs_mount_t		ext2_mount;
93 
94 MALLOC_DEFINE(M_EXT2NODE, "ext2_node", "EXT2 vnode private part");
95 static MALLOC_DEFINE(M_EXT2MNT, "ext2_mount", "EXT2 mount structure");
96 
97 static struct vfsops ext2fs_vfsops = {
98 	.vfs_fhtovp =		ext2_fhtovp,
99 	.vfs_mount =		ext2_mount,
100 	.vfs_root =		ext2_root,	/* root inode via vget */
101 	.vfs_statfs =		ext2_statfs,
102 	.vfs_sync =		ext2_sync,
103 	.vfs_unmount =		ext2_unmount,
104 	.vfs_vget =		ext2_vget,
105 };
106 
107 VFS_SET(ext2fs_vfsops, ext2fs, 0);
108 
109 static int	ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev,
110 		    int ronly);
111 static int	ext2_compute_sb_data(struct vnode * devvp,
112 		    struct ext2fs * es, struct m_ext2fs * fs);
113 
114 static const char *ext2_opts[] = { "acls", "async", "noatime", "noclusterr",
115     "noclusterw", "noexec", "export", "force", "from", "multilabel",
116     "suiddir", "nosymfollow", "sync", "union", NULL };
117 
118 /*
119  * VFS Operations.
120  *
121  * mount system call
122  */
123 static int
124 ext2_mount(struct mount *mp)
125 {
126 	struct vfsoptlist *opts;
127 	struct vnode *devvp;
128 	struct thread *td;
129 	struct ext2mount *ump = NULL;
130 	struct m_ext2fs *fs;
131 	struct nameidata nd, *ndp = &nd;
132 	accmode_t accmode;
133 	char *path, *fspec;
134 	int error, flags, len;
135 
136 	td = curthread;
137 	opts = mp->mnt_optnew;
138 
139 	if (vfs_filteropt(opts, ext2_opts))
140 		return (EINVAL);
141 
142 	vfs_getopt(opts, "fspath", (void **)&path, NULL);
143 	/* Double-check the length of path.. */
144 	if (strlen(path) >= MAXMNTLEN)
145 		return (ENAMETOOLONG);
146 
147 	fspec = NULL;
148 	error = vfs_getopt(opts, "from", (void **)&fspec, &len);
149 	if (!error && fspec[len - 1] != '\0')
150 		return (EINVAL);
151 
152 	/*
153 	 * If updating, check whether changing from read-only to
154 	 * read/write; if there is no device name, that's all we do.
155 	 */
156 	if (mp->mnt_flag & MNT_UPDATE) {
157 		ump = VFSTOEXT2(mp);
158 		fs = ump->um_e2fs;
159 		error = 0;
160 		if (fs->e2fs_ronly == 0 &&
161 		    vfs_flagopt(opts, "ro", NULL, 0)) {
162 			error = VFS_SYNC(mp, MNT_WAIT);
163 			if (error)
164 				return (error);
165 			flags = WRITECLOSE;
166 			if (mp->mnt_flag & MNT_FORCE)
167 				flags |= FORCECLOSE;
168 			error = ext2_flushfiles(mp, flags, td);
169 			if (error == 0 && fs->e2fs_wasvalid &&
170 			    ext2_cgupdate(ump, MNT_WAIT) == 0) {
171 				fs->e2fs->e2fs_state =
172 				    htole16((le16toh(fs->e2fs->e2fs_state) |
173 				    E2FS_ISCLEAN));
174 				ext2_sbupdate(ump, MNT_WAIT);
175 			}
176 			fs->e2fs_ronly = 1;
177 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
178 			g_topology_lock();
179 			g_access(ump->um_cp, 0, -1, 0);
180 			g_topology_unlock();
181 		}
182 		if (!error && (mp->mnt_flag & MNT_RELOAD))
183 			error = ext2_reload(mp, td);
184 		if (error)
185 			return (error);
186 		devvp = ump->um_devvp;
187 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
188 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
189 				return (EPERM);
190 
191 			/*
192 			 * If upgrade to read-write by non-root, then verify
193 			 * that user has necessary permissions on the device.
194 			 */
195 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
196 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
197 			    td->td_ucred, td);
198 			if (error)
199 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
200 			if (error) {
201 				VOP_UNLOCK(devvp);
202 				return (error);
203 			}
204 			VOP_UNLOCK(devvp);
205 			g_topology_lock();
206 			error = g_access(ump->um_cp, 0, 1, 0);
207 			g_topology_unlock();
208 			if (error)
209 				return (error);
210 
211 			if ((le16toh(fs->e2fs->e2fs_state) & E2FS_ISCLEAN) == 0 ||
212 			    (le16toh(fs->e2fs->e2fs_state) & E2FS_ERRORS)) {
213 				if (mp->mnt_flag & MNT_FORCE) {
214 					printf(
215 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
216 				} else {
217 					printf(
218 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
219 					    fs->e2fs_fsmnt);
220 					return (EPERM);
221 				}
222 			}
223 			fs->e2fs->e2fs_state =
224 			    htole16(le16toh(fs->e2fs->e2fs_state) & ~E2FS_ISCLEAN);
225 			(void)ext2_cgupdate(ump, MNT_WAIT);
226 			fs->e2fs_ronly = 0;
227 			MNT_ILOCK(mp);
228 			mp->mnt_flag &= ~MNT_RDONLY;
229 			MNT_IUNLOCK(mp);
230 		}
231 		if (vfs_flagopt(opts, "export", NULL, 0)) {
232 			/* Process export requests in vfs_mount.c. */
233 			return (error);
234 		}
235 	}
236 
237 	/*
238 	 * Not an update, or updating the name: look up the name
239 	 * and verify that it refers to a sensible disk device.
240 	 */
241 	if (fspec == NULL)
242 		return (EINVAL);
243 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec);
244 	if ((error = namei(ndp)) != 0)
245 		return (error);
246 	NDFREE(ndp, NDF_ONLY_PNBUF);
247 	devvp = ndp->ni_vp;
248 
249 	if (!vn_isdisk_error(devvp, &error)) {
250 		vput(devvp);
251 		return (error);
252 	}
253 
254 	/*
255 	 * If mount by non-root, then verify that user has necessary
256 	 * permissions on the device.
257 	 *
258 	 * XXXRW: VOP_ACCESS() enough?
259 	 */
260 	accmode = VREAD;
261 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
262 		accmode |= VWRITE;
263 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
264 	if (error)
265 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
266 	if (error) {
267 		vput(devvp);
268 		return (error);
269 	}
270 
271 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
272 		error = ext2_mountfs(devvp, mp);
273 	} else {
274 		if (devvp != ump->um_devvp) {
275 			vput(devvp);
276 			return (EINVAL);	/* needs translation */
277 		} else
278 			vput(devvp);
279 	}
280 	if (error) {
281 		vrele(devvp);
282 		return (error);
283 	}
284 	ump = VFSTOEXT2(mp);
285 	fs = ump->um_e2fs;
286 
287 	/*
288 	 * Note that this strncpy() is ok because of a check at the start
289 	 * of ext2_mount().
290 	 */
291 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
292 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
293 	vfs_mountedfrom(mp, fspec);
294 	return (0);
295 }
296 
297 static int
298 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
299 {
300 	uint32_t i, mask;
301 
302 	if (le16toh(es->e2fs_magic) != E2FS_MAGIC) {
303 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
304 		    devtoname(dev), le16toh(es->e2fs_magic), E2FS_MAGIC);
305 		return (1);
306 	}
307 	if (le32toh(es->e2fs_rev) > E2FS_REV0) {
308 		mask = le32toh(es->e2fs_features_incompat) & ~(EXT2F_INCOMPAT_SUPP);
309 		if (mask) {
310 			printf("WARNING: mount of %s denied due to "
311 			    "unsupported optional features:\n", devtoname(dev));
312 			for (i = 0;
313 			    i < sizeof(incompat)/sizeof(struct ext2_feature);
314 			    i++)
315 				if (mask & incompat[i].mask)
316 					printf("%s ", incompat[i].name);
317 			printf("\n");
318 			return (1);
319 		}
320 		mask = le32toh(es->e2fs_features_rocompat) & ~EXT2F_ROCOMPAT_SUPP;
321 		if (!ronly && mask) {
322 			printf("WARNING: R/W mount of %s denied due to "
323 			    "unsupported optional features:\n", devtoname(dev));
324 			for (i = 0;
325 			    i < sizeof(ro_compat)/sizeof(struct ext2_feature);
326 			    i++)
327 				if (mask & ro_compat[i].mask)
328 					printf("%s ", ro_compat[i].name);
329 			printf("\n");
330 			return (1);
331 		}
332 	}
333 	return (0);
334 }
335 
336 static e4fs_daddr_t
337 ext2_cg_location(struct m_ext2fs *fs, int number)
338 {
339 	int cg, descpb, logical_sb, has_super = 0;
340 
341 	/*
342 	 * Adjust logical superblock block number.
343 	 * Godmar thinks: if the blocksize is greater than 1024, then
344 	 * the superblock is logically part of block zero.
345 	 */
346 	logical_sb = fs->e2fs_bsize > SBSIZE ? 0 : 1;
347 
348 	if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
349 	    number < le32toh(fs->e2fs->e3fs_first_meta_bg))
350 		return (logical_sb + number + 1);
351 
352 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT))
353 		descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
354 	else
355 		descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
356 
357 	cg = descpb * number;
358 
359 	if (ext2_cg_has_sb(fs, cg))
360 		has_super = 1;
361 
362 	return (has_super + cg * (e4fs_daddr_t)EXT2_BLOCKS_PER_GROUP(fs) +
363 	    le32toh(fs->e2fs->e2fs_first_dblock));
364 }
365 
366 static int
367 ext2_cg_validate(struct m_ext2fs *fs)
368 {
369 	uint64_t b_bitmap;
370 	uint64_t i_bitmap;
371 	uint64_t i_tables;
372 	uint64_t first_block, last_block, last_cg_block;
373 	struct ext2_gd *gd;
374 	unsigned int i, cg_count;
375 
376 	first_block = le32toh(fs->e2fs->e2fs_first_dblock);
377 	last_cg_block = ext2_cg_number_gdb(fs, 0);
378 	cg_count = fs->e2fs_gcount;
379 
380 	for (i = 0; i < fs->e2fs_gcount; i++) {
381 		gd = &fs->e2fs_gd[i];
382 
383 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
384 		    i == fs->e2fs_gcount - 1) {
385 			last_block = fs->e2fs_bcount - 1;
386 		} else {
387 			last_block = first_block +
388 			    (EXT2_BLOCKS_PER_GROUP(fs) - 1);
389 		}
390 
391 		if ((cg_count == fs->e2fs_gcount) &&
392 		    !(le16toh(gd->ext4bgd_flags) & EXT2_BG_INODE_ZEROED))
393 			cg_count = i;
394 
395 		b_bitmap = e2fs_gd_get_b_bitmap(gd);
396 		if (b_bitmap == 0) {
397 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
398 			    "block bitmap is zero", i);
399 			return (EINVAL);
400 		}
401 		if (b_bitmap <= last_cg_block) {
402 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
403 			    "block bitmap overlaps gds", i);
404 			return (EINVAL);
405 		}
406 		if (b_bitmap < first_block || b_bitmap > last_block) {
407 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
408 			    "block bitmap not in group", i);
409 			return (EINVAL);
410 		}
411 
412 		i_bitmap = e2fs_gd_get_i_bitmap(gd);
413 		if (i_bitmap == 0) {
414 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
415 			    "inode bitmap is zero", i);
416 			return (EINVAL);
417 		}
418 		if (i_bitmap <= last_cg_block) {
419 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
420 			    "inode bitmap overlaps gds", i);
421 			return (EINVAL);
422 		}
423 		if (i_bitmap < first_block || i_bitmap > last_block) {
424 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
425 			    "inode bitmap not in group blk", i);
426 			return (EINVAL);
427 		}
428 
429 		i_tables = e2fs_gd_get_i_tables(gd);
430 		if (i_tables == 0) {
431 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
432 			    "inode table is zero", i);
433 			return (EINVAL);
434 		}
435 		if (i_tables <= last_cg_block) {
436 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
437 			    "inode tables overlaps gds", i);
438 			return (EINVAL);
439 		}
440 		if (i_tables < first_block ||
441 		    i_tables + fs->e2fs_itpg - 1 > last_block) {
442 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
443 			    "inode tables not in group blk", i);
444 			return (EINVAL);
445 		}
446 
447 		if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG))
448 			first_block += EXT2_BLOCKS_PER_GROUP(fs);
449 	}
450 
451 	return (0);
452 }
453 
454 /*
455  * This computes the fields of the m_ext2fs structure from the
456  * data in the ext2fs structure read in.
457  */
458 static int
459 ext2_compute_sb_data(struct vnode *devvp, struct ext2fs *es,
460     struct m_ext2fs *fs)
461 {
462 	struct buf *bp;
463 	uint32_t e2fs_descpb, e2fs_gdbcount_alloc;
464 	int i, j;
465 	int g_count = 0;
466 	int error;
467 
468 	/* Check checksum features */
469 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) &&
470 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
471 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
472 		    "incorrect checksum features combination");
473 		return (EINVAL);
474 	}
475 
476 	/* Precompute checksum seed for all metadata */
477 	ext2_sb_csum_set_seed(fs);
478 
479 	/* Verify sb csum if possible */
480 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
481 		error = ext2_sb_csum_verify(fs);
482 		if (error) {
483 			return (error);
484 		}
485 	}
486 
487 	/* Check for block size = 1K|2K|4K */
488 	if (le32toh(es->e2fs_log_bsize) > 2) {
489 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
490 		    "bad block size");
491 		return (EINVAL);
492 	}
493 
494 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + le32toh(es->e2fs_log_bsize);
495 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
496 	fs->e2fs_fsbtodb = le32toh(es->e2fs_log_bsize) + 1;
497 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
498 
499 	/* Check for fragment size */
500 	if (le32toh(es->e2fs_log_fsize) >
501 	    (EXT2_MAX_FRAG_LOG_SIZE - EXT2_MIN_BLOCK_LOG_SIZE)) {
502 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
503 		    "invalid log cluster size");
504 		return (EINVAL);
505 	}
506 
507 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << le32toh(es->e2fs_log_fsize);
508 	if (fs->e2fs_fsize != fs->e2fs_bsize) {
509 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
510 		    "fragment size != block size");
511 		return (EINVAL);
512 	}
513 
514 	fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
515 
516 	/* Check reserved gdt blocks for future filesystem expansion */
517 	if (le16toh(es->e2fs_reserved_ngdb) > (fs->e2fs_bsize / 4)) {
518 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
519 		    "number of reserved GDT blocks too large");
520 		return (EINVAL);
521 	}
522 
523 	if (le32toh(es->e2fs_rev) == E2FS_REV0) {
524 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
525 	} else {
526 		fs->e2fs_isize = le16toh(es->e2fs_inode_size);
527 
528 		/*
529 		 * Check first ino.
530 		 */
531 		if (le32toh(es->e2fs_first_ino) < EXT2_FIRSTINO) {
532 			SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
533 			    "invalid first ino");
534 			return (EINVAL);
535 		}
536 
537 		/*
538 		 * Simple sanity check for superblock inode size value.
539 		 */
540 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
541 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
542 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
543 			SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
544 			    "invalid inode size");
545 			return (EINVAL);
546 		}
547 	}
548 
549 	/* Check group descriptors */
550 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT) &&
551 	    le16toh(es->e3fs_desc_size) != E2FS_64BIT_GD_SIZE) {
552 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
553 		    "unsupported 64bit descriptor size");
554 		return (EINVAL);
555 	}
556 
557 	fs->e2fs_bpg = le32toh(es->e2fs_bpg);
558 	fs->e2fs_fpg = le32toh(es->e2fs_fpg);
559 	if (fs->e2fs_bpg == 0 || fs->e2fs_fpg == 0) {
560 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
561 		    "zero blocks/fragments per group");
562 		return (EINVAL);
563 	} else if (fs->e2fs_bpg != fs->e2fs_fpg) {
564 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
565 		    "blocks per group not equal fragments per group");
566 		return (EINVAL);
567 	}
568 
569 	if (fs->e2fs_bpg != fs->e2fs_bsize * 8) {
570 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
571 		    "non-standard group size unsupported");
572 		return (EINVAL);
573 	}
574 
575 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
576 	if (fs->e2fs_ipb == 0 ||
577 	    fs->e2fs_ipb > fs->e2fs_bsize / E2FS_REV0_INODE_SIZE) {
578 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
579 		    "bad inodes per block size");
580 		return (EINVAL);
581 	}
582 
583 	fs->e2fs_ipg = le32toh(es->e2fs_ipg);
584 	if (fs->e2fs_ipg < fs->e2fs_ipb || fs->e2fs_ipg >  fs->e2fs_bsize * 8) {
585 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
586 		    "invalid inodes per group");
587 		return (EINVAL);
588 	}
589 
590 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
591 
592 	fs->e2fs_bcount = le32toh(es->e2fs_bcount);
593 	fs->e2fs_rbcount = le32toh(es->e2fs_rbcount);
594 	fs->e2fs_fbcount = le32toh(es->e2fs_fbcount);
595 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
596 		fs->e2fs_bcount |= (uint64_t)(le32toh(es->e4fs_bcount_hi)) << 32;
597 		fs->e2fs_rbcount |= (uint64_t)(le32toh(es->e4fs_rbcount_hi)) << 32;
598 		fs->e2fs_fbcount |= (uint64_t)(le32toh(es->e4fs_fbcount_hi)) << 32;
599 	}
600 	if (fs->e2fs_rbcount > fs->e2fs_bcount ||
601 	    fs->e2fs_fbcount > fs->e2fs_bcount) {
602 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
603 		    "invalid block count");
604 		return (EINVAL);
605 	}
606 
607 	fs->e2fs_ficount = le32toh(es->e2fs_ficount);
608 	if (fs->e2fs_ficount > le32toh(es->e2fs_icount)) {
609 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
610 		    "invalid number of free inodes");
611 		return (EINVAL);
612 	}
613 
614 	if (le32toh(es->e2fs_first_dblock) != (fs->e2fs_bsize > 1024 ? 0 : 1) ||
615 	    le32toh(es->e2fs_first_dblock) >= fs->e2fs_bcount) {
616 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
617 		    "first data block out of range");
618 		return (EINVAL);
619 	}
620 
621 	fs->e2fs_gcount = howmany(fs->e2fs_bcount -
622 	    le32toh(es->e2fs_first_dblock), EXT2_BLOCKS_PER_GROUP(fs));
623 	if (fs->e2fs_gcount > ((uint64_t)1 << 32) - EXT2_DESCS_PER_BLOCK(fs)) {
624 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
625 		    "groups count too large");
626 		return (EINVAL);
627 	}
628 
629 	/* Check for extra isize in big inodes. */
630 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
631 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
632 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
633 		    "no space for extra inode timestamps");
634 		return (EINVAL);
635 	}
636 
637 	/* s_resuid / s_resgid ? */
638 
639 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
640 		e2fs_descpb = fs->e2fs_bsize / E2FS_64BIT_GD_SIZE;
641 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount, e2fs_descpb);
642 	} else {
643 		e2fs_descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
644 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount,
645 		    fs->e2fs_bsize / sizeof(struct ext2_gd));
646 	}
647 	fs->e2fs_gdbcount = howmany(fs->e2fs_gcount, e2fs_descpb);
648 	fs->e2fs_gd = malloc(e2fs_gdbcount_alloc * fs->e2fs_bsize,
649 	    M_EXT2MNT, M_WAITOK | M_ZERO);
650 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
651 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
652 
653 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
654 		error = bread(devvp,
655 		    fsbtodb(fs, ext2_cg_location(fs, i)),
656 		    fs->e2fs_bsize, NOCRED, &bp);
657 		if (error) {
658 			/*
659 			 * fs->e2fs_gd and fs->e2fs_contigdirs
660 			 * will be freed later by the caller,
661 			 * because this function could be called from
662 			 * MNT_UPDATE path.
663 			 */
664 			return (error);
665 		}
666 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
667 			memcpy(&fs->e2fs_gd[
668 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
669 			    bp->b_data, fs->e2fs_bsize);
670 		} else {
671 			for (j = 0; j < e2fs_descpb &&
672 			    g_count < fs->e2fs_gcount; j++, g_count++)
673 				memcpy(&fs->e2fs_gd[g_count],
674 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
675 				    E2FS_REV0_GD_SIZE);
676 		}
677 		brelse(bp);
678 		bp = NULL;
679 	}
680 
681 	/* Validate cgs consistency */
682 	error = ext2_cg_validate(fs);
683 	if (error)
684 		return (error);
685 
686 	/* Verfy cgs csum */
687 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
688 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
689 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
690 		if (error)
691 			return (error);
692 	}
693 	/* Initialization for the ext2 Orlov allocator variant. */
694 	fs->e2fs_total_dir = 0;
695 	for (i = 0; i < fs->e2fs_gcount; i++)
696 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
697 
698 	if (le32toh(es->e2fs_rev) == E2FS_REV0 ||
699 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
700 		fs->e2fs_maxfilesize = 0x7fffffff;
701 	else {
702 		fs->e2fs_maxfilesize = 0xffffffffffff;
703 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
704 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
705 	}
706 	if (le32toh(es->e4fs_flags) & E2FS_UNSIGNED_HASH) {
707 		fs->e2fs_uhash = 3;
708 	} else if ((le32toh(es->e4fs_flags) & E2FS_SIGNED_HASH) == 0) {
709 #ifdef __CHAR_UNSIGNED__
710 		es->e4fs_flags = htole32(le32toh(es->e4fs_flags) | E2FS_UNSIGNED_HASH);
711 		fs->e2fs_uhash = 3;
712 #else
713 		es->e4fs_flags = htole32(le32toh(es->e4fs_flags) | E2FS_SIGNED_HASH);
714 #endif
715 	}
716 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
717 		error = ext2_sb_csum_verify(fs);
718 
719 	return (error);
720 }
721 
722 /*
723  * Reload all incore data for a filesystem (used after running fsck on
724  * the root filesystem and finding things to fix). The filesystem must
725  * be mounted read-only.
726  *
727  * Things to do to update the mount:
728  *	1) invalidate all cached meta-data.
729  *	2) re-read superblock from disk.
730  *	3) invalidate all cluster summary information.
731  *	4) invalidate all inactive vnodes.
732  *	5) invalidate all cached file data.
733  *	6) re-read inode data for all active vnodes.
734  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
735  */
736 static int
737 ext2_reload(struct mount *mp, struct thread *td)
738 {
739 	struct vnode *vp, *mvp, *devvp;
740 	struct inode *ip;
741 	struct buf *bp;
742 	struct ext2fs *es;
743 	struct m_ext2fs *fs;
744 	struct csum *sump;
745 	int error, i;
746 	int32_t *lp;
747 
748 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
749 		return (EINVAL);
750 	/*
751 	 * Step 1: invalidate all cached meta-data.
752 	 */
753 	devvp = VFSTOEXT2(mp)->um_devvp;
754 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
755 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
756 		panic("ext2_reload: dirty1");
757 	VOP_UNLOCK(devvp);
758 
759 	/*
760 	 * Step 2: re-read superblock from disk.
761 	 * constants have been adjusted for ext2
762 	 */
763 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
764 		return (error);
765 	es = (struct ext2fs *)bp->b_data;
766 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
767 		brelse(bp);
768 		return (EIO);		/* XXX needs translation */
769 	}
770 	fs = VFSTOEXT2(mp)->um_e2fs;
771 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
772 
773 	if ((error = ext2_compute_sb_data(devvp, es, fs)) != 0) {
774 		brelse(bp);
775 		return (error);
776 	}
777 #ifdef UNKLAR
778 	if (fs->fs_sbsize < SBSIZE)
779 		bp->b_flags |= B_INVAL;
780 #endif
781 	brelse(bp);
782 
783 	/*
784 	 * Step 3: invalidate all cluster summary information.
785 	 */
786 	if (fs->e2fs_contigsumsize > 0) {
787 		lp = fs->e2fs_maxcluster;
788 		sump = fs->e2fs_clustersum;
789 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
790 			*lp++ = fs->e2fs_contigsumsize;
791 			sump->cs_init = 0;
792 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
793 		}
794 	}
795 
796 loop:
797 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
798 		/*
799 		 * Step 4: invalidate all cached file data.
800 		 */
801 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
802 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
803 			goto loop;
804 		}
805 		if (vinvalbuf(vp, 0, 0, 0))
806 			panic("ext2_reload: dirty2");
807 
808 		/*
809 		 * Step 5: re-read inode data for all active vnodes.
810 		 */
811 		ip = VTOI(vp);
812 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
813 		    (int)fs->e2fs_bsize, NOCRED, &bp);
814 		if (error) {
815 			VOP_UNLOCK(vp);
816 			vrele(vp);
817 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
818 			return (error);
819 		}
820 
821 		error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
822 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
823 
824 		brelse(bp);
825 		VOP_UNLOCK(vp);
826 		vrele(vp);
827 
828 		if (error) {
829 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
830 			return (error);
831 		}
832 	}
833 	return (0);
834 }
835 
836 /*
837  * Common code for mount and mountroot.
838  */
839 static int
840 ext2_mountfs(struct vnode *devvp, struct mount *mp)
841 {
842 	struct ext2mount *ump;
843 	struct buf *bp;
844 	struct m_ext2fs *fs;
845 	struct ext2fs *es;
846 	struct cdev *dev = devvp->v_rdev;
847 	struct g_consumer *cp;
848 	struct bufobj *bo;
849 	struct csum *sump;
850 	int error;
851 	int ronly;
852 	int i;
853 	u_long size;
854 	int32_t *lp;
855 	int32_t e2fs_maxcontig;
856 
857 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
858 	/* XXX: use VOP_ACESS to check FS perms */
859 	g_topology_lock();
860 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
861 	g_topology_unlock();
862 	VOP_UNLOCK(devvp);
863 	if (error)
864 		return (error);
865 
866 	/* XXX: should we check for some sectorsize or 512 instead? */
867 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
868 	    (SBSIZE < cp->provider->sectorsize)) {
869 		g_topology_lock();
870 		g_vfs_close(cp);
871 		g_topology_unlock();
872 		return (EINVAL);
873 	}
874 
875 	bo = &devvp->v_bufobj;
876 	bo->bo_private = cp;
877 	bo->bo_ops = g_vfs_bufops;
878 	if (devvp->v_rdev->si_iosize_max != 0)
879 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
880 	if (mp->mnt_iosize_max > maxphys)
881 		mp->mnt_iosize_max = maxphys;
882 
883 	bp = NULL;
884 	ump = NULL;
885 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
886 		goto out;
887 	es = (struct ext2fs *)bp->b_data;
888 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
889 		error = EINVAL;		/* XXX needs translation */
890 		goto out;
891 	}
892 	if ((le16toh(es->e2fs_state) & E2FS_ISCLEAN) == 0 ||
893 	    (le16toh(es->e2fs_state) & E2FS_ERRORS)) {
894 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
895 			printf(
896 "WARNING: Filesystem was not properly dismounted\n");
897 		} else {
898 			printf(
899 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
900 			error = EPERM;
901 			goto out;
902 		}
903 	}
904 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
905 
906 	/*
907 	 * I don't know whether this is the right strategy. Note that
908 	 * we dynamically allocate both an m_ext2fs and an ext2fs
909 	 * while Linux keeps the super block in a locked buffer.
910 	 */
911 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
912 	    M_EXT2MNT, M_WAITOK | M_ZERO);
913 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
914 	    M_EXT2MNT, M_WAITOK);
915 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
916 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
917 	if ((error = ext2_compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
918 		goto out;
919 
920 	/*
921 	 * Calculate the maximum contiguous blocks and size of cluster summary
922 	 * array.  In FFS this is done by newfs; however, the superblock
923 	 * in ext2fs doesn't have these variables, so we can calculate
924 	 * them here.
925 	 */
926 	e2fs_maxcontig = MAX(1, maxphys / ump->um_e2fs->e2fs_bsize);
927 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
928 	ump->um_e2fs->e2fs_maxsymlinklen = EXT2_MAXSYMLINKLEN;
929 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
930 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
931 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
932 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
933 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
934 		lp = ump->um_e2fs->e2fs_maxcluster;
935 		sump = ump->um_e2fs->e2fs_clustersum;
936 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
937 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
938 			sump->cs_init = 0;
939 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
940 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
941 		}
942 	}
943 
944 	brelse(bp);
945 	bp = NULL;
946 	fs = ump->um_e2fs;
947 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
948 
949 	/*
950 	 * If the fs is not mounted read-only, make sure the super block is
951 	 * always written back on a sync().
952 	 */
953 	fs->e2fs_wasvalid = le16toh(fs->e2fs->e2fs_state) & E2FS_ISCLEAN ? 1 : 0;
954 	if (ronly == 0) {
955 		fs->e2fs_fmod = 1;	/* mark it modified and set fs invalid */
956 		fs->e2fs->e2fs_state =
957 		    htole16(le16toh(fs->e2fs->e2fs_state) & ~E2FS_ISCLEAN);
958 	}
959 	mp->mnt_data = ump;
960 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
961 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
962 	MNT_ILOCK(mp);
963 	mp->mnt_flag |= MNT_LOCAL;
964 	MNT_IUNLOCK(mp);
965 	ump->um_mountp = mp;
966 	ump->um_dev = dev;
967 	ump->um_devvp = devvp;
968 	ump->um_bo = &devvp->v_bufobj;
969 	ump->um_cp = cp;
970 
971 	/*
972 	 * Setting those two parameters allowed us to use
973 	 * ufs_bmap w/o changse!
974 	 */
975 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
976 	ump->um_bptrtodb = le32toh(fs->e2fs->e2fs_log_bsize) + 1;
977 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
978 	if (ronly == 0)
979 		ext2_sbupdate(ump, MNT_WAIT);
980 	/*
981 	 * Initialize filesystem stat information in mount struct.
982 	 */
983 	MNT_ILOCK(mp);
984 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
985 	    MNTK_USES_BCACHE;
986 	MNT_IUNLOCK(mp);
987 	return (0);
988 out:
989 	if (bp)
990 		brelse(bp);
991 	if (cp != NULL) {
992 		g_topology_lock();
993 		g_vfs_close(cp);
994 		g_topology_unlock();
995 	}
996 	if (ump) {
997 		mtx_destroy(EXT2_MTX(ump));
998 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
999 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
1000 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
1001 		free(ump->um_e2fs, M_EXT2MNT);
1002 		free(ump, M_EXT2MNT);
1003 		mp->mnt_data = NULL;
1004 	}
1005 	return (error);
1006 }
1007 
1008 /*
1009  * Unmount system call.
1010  */
1011 static int
1012 ext2_unmount(struct mount *mp, int mntflags)
1013 {
1014 	struct ext2mount *ump;
1015 	struct m_ext2fs *fs;
1016 	struct csum *sump;
1017 	int error, flags, i, ronly;
1018 
1019 	flags = 0;
1020 	if (mntflags & MNT_FORCE) {
1021 		if (mp->mnt_flag & MNT_ROOTFS)
1022 			return (EINVAL);
1023 		flags |= FORCECLOSE;
1024 	}
1025 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
1026 		return (error);
1027 	ump = VFSTOEXT2(mp);
1028 	fs = ump->um_e2fs;
1029 	ronly = fs->e2fs_ronly;
1030 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
1031 		if (fs->e2fs_wasvalid)
1032 			fs->e2fs->e2fs_state =
1033 			    htole16(le16toh(fs->e2fs->e2fs_state) | E2FS_ISCLEAN);
1034 		ext2_sbupdate(ump, MNT_WAIT);
1035 	}
1036 
1037 	g_topology_lock();
1038 	g_vfs_close(ump->um_cp);
1039 	g_topology_unlock();
1040 	vrele(ump->um_devvp);
1041 	sump = fs->e2fs_clustersum;
1042 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
1043 		free(sump->cs_sum, M_EXT2MNT);
1044 	free(fs->e2fs_clustersum, M_EXT2MNT);
1045 	free(fs->e2fs_maxcluster, M_EXT2MNT);
1046 	free(fs->e2fs_gd, M_EXT2MNT);
1047 	free(fs->e2fs_contigdirs, M_EXT2MNT);
1048 	free(fs->e2fs, M_EXT2MNT);
1049 	free(fs, M_EXT2MNT);
1050 	free(ump, M_EXT2MNT);
1051 	mp->mnt_data = NULL;
1052 	MNT_ILOCK(mp);
1053 	mp->mnt_flag &= ~MNT_LOCAL;
1054 	MNT_IUNLOCK(mp);
1055 	return (error);
1056 }
1057 
1058 /*
1059  * Flush out all the files in a filesystem.
1060  */
1061 static int
1062 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
1063 {
1064 	int error;
1065 
1066 	error = vflush(mp, 0, flags, td);
1067 	return (error);
1068 }
1069 
1070 /*
1071  * Get filesystem statistics.
1072  */
1073 int
1074 ext2_statfs(struct mount *mp, struct statfs *sbp)
1075 {
1076 	struct ext2mount *ump;
1077 	struct m_ext2fs *fs;
1078 	uint32_t overhead, overhead_per_group, ngdb;
1079 	int i, ngroups;
1080 
1081 	ump = VFSTOEXT2(mp);
1082 	fs = ump->um_e2fs;
1083 	if (le16toh(fs->e2fs->e2fs_magic) != E2FS_MAGIC)
1084 		panic("ext2_statfs");
1085 
1086 	/*
1087 	 * Compute the overhead (FS structures)
1088 	 */
1089 	overhead_per_group =
1090 	    1 /* block bitmap */ +
1091 	    1 /* inode bitmap */ +
1092 	    fs->e2fs_itpg;
1093 	overhead = le32toh(fs->e2fs->e2fs_first_dblock) +
1094 	    fs->e2fs_gcount * overhead_per_group;
1095 	if (le32toh(fs->e2fs->e2fs_rev) > E2FS_REV0 &&
1096 	    le32toh(fs->e2fs->e2fs_features_rocompat) & EXT2F_ROCOMPAT_SPARSESUPER) {
1097 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
1098 			if (ext2_cg_has_sb(fs, i))
1099 				ngroups++;
1100 		}
1101 	} else {
1102 		ngroups = fs->e2fs_gcount;
1103 	}
1104 	ngdb = fs->e2fs_gdbcount;
1105 	if (le32toh(fs->e2fs->e2fs_rev) > E2FS_REV0 &&
1106 	    le32toh(fs->e2fs->e2fs_features_compat) & EXT2F_COMPAT_RESIZE)
1107 		ngdb += le16toh(fs->e2fs->e2fs_reserved_ngdb);
1108 	overhead += ngroups * (1 /* superblock */ + ngdb);
1109 
1110 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
1111 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
1112 	sbp->f_blocks = fs->e2fs_bcount - overhead;
1113 	sbp->f_bfree = fs->e2fs_fbcount;
1114 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
1115 	sbp->f_files = le32toh(fs->e2fs->e2fs_icount);
1116 	sbp->f_ffree = fs->e2fs_ficount;
1117 	return (0);
1118 }
1119 
1120 /*
1121  * Go through the disk queues to initiate sandbagged IO;
1122  * go through the inodes to write those that have been modified;
1123  * initiate the writing of the super block if it has been modified.
1124  *
1125  * Note: we are always called with the filesystem marked `MPBUSY'.
1126  */
1127 static int
1128 ext2_sync(struct mount *mp, int waitfor)
1129 {
1130 	struct vnode *mvp, *vp;
1131 	struct thread *td;
1132 	struct inode *ip;
1133 	struct ext2mount *ump = VFSTOEXT2(mp);
1134 	struct m_ext2fs *fs;
1135 	int error, allerror = 0;
1136 
1137 	td = curthread;
1138 	fs = ump->um_e2fs;
1139 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
1140 		panic("ext2_sync: rofs mod fs=%s", fs->e2fs_fsmnt);
1141 	}
1142 
1143 	/*
1144 	 * Write back each (modified) inode.
1145 	 */
1146 loop:
1147 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1148 		if (vp->v_type == VNON) {
1149 			VI_UNLOCK(vp);
1150 			continue;
1151 		}
1152 		ip = VTOI(vp);
1153 		if ((ip->i_flag &
1154 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1155 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1156 		    waitfor == MNT_LAZY)) {
1157 			VI_UNLOCK(vp);
1158 			continue;
1159 		}
1160 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK);
1161 		if (error) {
1162 			if (error == ENOENT) {
1163 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1164 				goto loop;
1165 			}
1166 			continue;
1167 		}
1168 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
1169 			allerror = error;
1170 		VOP_UNLOCK(vp);
1171 		vrele(vp);
1172 	}
1173 
1174 	/*
1175 	 * Force stale filesystem control information to be flushed.
1176 	 */
1177 	if (waitfor != MNT_LAZY) {
1178 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1179 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
1180 			allerror = error;
1181 		VOP_UNLOCK(ump->um_devvp);
1182 	}
1183 
1184 	/*
1185 	 * Write back modified superblock.
1186 	 */
1187 	if (fs->e2fs_fmod != 0) {
1188 		fs->e2fs_fmod = 0;
1189 		fs->e2fs->e2fs_wtime = htole32(time_second);
1190 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
1191 			allerror = error;
1192 	}
1193 	return (allerror);
1194 }
1195 
1196 /*
1197  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
1198  * in from disk.  If it is in core, wait for the lock bit to clear, then
1199  * return the inode locked.  Detection and handling of mount points must be
1200  * done by the calling routine.
1201  */
1202 static int
1203 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
1204 {
1205 	struct m_ext2fs *fs;
1206 	struct inode *ip;
1207 	struct ext2mount *ump;
1208 	struct buf *bp;
1209 	struct vnode *vp;
1210 	struct thread *td;
1211 	unsigned int i, used_blocks;
1212 	int error;
1213 
1214 	td = curthread;
1215 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
1216 	if (error || *vpp != NULL)
1217 		return (error);
1218 
1219 	ump = VFSTOEXT2(mp);
1220 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
1221 
1222 	/* Allocate a new vnode/inode. */
1223 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
1224 		*vpp = NULL;
1225 		free(ip, M_EXT2NODE);
1226 		return (error);
1227 	}
1228 	vp->v_data = ip;
1229 	ip->i_vnode = vp;
1230 	ip->i_e2fs = fs = ump->um_e2fs;
1231 	ip->i_ump = ump;
1232 	ip->i_number = ino;
1233 	cluster_init_vn(&ip->i_clusterw);
1234 
1235 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1236 	error = insmntque(vp, mp);
1237 	if (error != 0) {
1238 		free(ip, M_EXT2NODE);
1239 		*vpp = NULL;
1240 		return (error);
1241 	}
1242 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
1243 	if (error || *vpp != NULL)
1244 		return (error);
1245 
1246 	/* Read in the disk contents for the inode, copy into the inode. */
1247 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1248 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
1249 		/*
1250 		 * The inode does not contain anything useful, so it would
1251 		 * be misleading to leave it on its hash chain. With mode
1252 		 * still zero, it will be unlinked and returned to the free
1253 		 * list by vput().
1254 		 */
1255 		brelse(bp);
1256 		vput(vp);
1257 		*vpp = NULL;
1258 		return (error);
1259 	}
1260 	/* convert ext2 inode to dinode */
1261 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1262 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1263 	if (error) {
1264 		brelse(bp);
1265 		vput(vp);
1266 		*vpp = NULL;
1267 		return (error);
1268 	}
1269 	ip->i_block_group = ino_to_cg(fs, ino);
1270 	ip->i_next_alloc_block = 0;
1271 	ip->i_next_alloc_goal = 0;
1272 
1273 	/*
1274 	 * Now we want to make sure that block pointers for unused
1275 	 * blocks are zeroed out - ext2_balloc depends on this
1276 	 * although for regular files and directories only
1277 	 *
1278 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1279 	 * out because we could corrupt the extent tree.
1280 	 */
1281 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1282 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1283 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1284 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1285 			ip->i_db[i] = 0;
1286 	}
1287 
1288 	bqrelse(bp);
1289 
1290 #ifdef EXT2FS_PRINT_EXTENTS
1291 	ext2_print_inode(ip);
1292 	error = ext4_ext_walk(ip);
1293 	if (error) {
1294 		vput(vp);
1295 		*vpp = NULL;
1296 		return (error);
1297 	}
1298 #endif
1299 
1300 	/*
1301 	 * Initialize the vnode from the inode, check for aliases.
1302 	 * Note that the underlying vnode may have changed.
1303 	 */
1304 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1305 		vput(vp);
1306 		*vpp = NULL;
1307 		return (error);
1308 	}
1309 
1310 	/*
1311 	 * Finish inode initialization.
1312 	 */
1313 
1314 	*vpp = vp;
1315 	return (0);
1316 }
1317 
1318 /*
1319  * File handle to vnode
1320  *
1321  * Have to be really careful about stale file handles:
1322  * - check that the inode number is valid
1323  * - call ext2_vget() to get the locked inode
1324  * - check for an unallocated inode (i_mode == 0)
1325  * - check that the given client host has export rights and return
1326  *   those rights via. exflagsp and credanonp
1327  */
1328 static int
1329 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1330 {
1331 	struct inode *ip;
1332 	struct ufid *ufhp;
1333 	struct vnode *nvp;
1334 	struct m_ext2fs *fs;
1335 	int error;
1336 
1337 	ufhp = (struct ufid *)fhp;
1338 	fs = VFSTOEXT2(mp)->um_e2fs;
1339 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1340 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs_ipg)
1341 		return (ESTALE);
1342 
1343 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1344 	if (error) {
1345 		*vpp = NULLVP;
1346 		return (error);
1347 	}
1348 	ip = VTOI(nvp);
1349 	if (ip->i_mode == 0 ||
1350 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1351 		vput(nvp);
1352 		*vpp = NULLVP;
1353 		return (ESTALE);
1354 	}
1355 	*vpp = nvp;
1356 	vnode_create_vobject(*vpp, 0, curthread);
1357 	return (0);
1358 }
1359 
1360 /*
1361  * Write a superblock and associated information back to disk.
1362  */
1363 static int
1364 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1365 {
1366 	struct m_ext2fs *fs = mp->um_e2fs;
1367 	struct ext2fs *es = fs->e2fs;
1368 	struct buf *bp;
1369 	int error = 0;
1370 
1371 	es->e2fs_bcount = htole32(fs->e2fs_bcount & 0xffffffff);
1372 	es->e2fs_rbcount = htole32(fs->e2fs_rbcount & 0xffffffff);
1373 	es->e2fs_fbcount = htole32(fs->e2fs_fbcount & 0xffffffff);
1374 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1375 		es->e4fs_bcount_hi = htole32(fs->e2fs_bcount >> 32);
1376 		es->e4fs_rbcount_hi = htole32(fs->e2fs_rbcount >> 32);
1377 		es->e4fs_fbcount_hi = htole32(fs->e2fs_fbcount >> 32);
1378 	}
1379 
1380 	es->e2fs_ficount = htole32(fs->e2fs_ficount);
1381 
1382 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1383 		ext2_sb_csum_set(fs);
1384 
1385 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1386 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1387 	if (waitfor == MNT_WAIT)
1388 		error = bwrite(bp);
1389 	else
1390 		bawrite(bp);
1391 
1392 	/*
1393 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1394 	 * are not busy at this point and are (hopefully) written by the
1395 	 * usual sync mechanism. No need to write them here.
1396 	 */
1397 	return (error);
1398 }
1399 int
1400 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1401 {
1402 	struct m_ext2fs *fs = mp->um_e2fs;
1403 	struct buf *bp;
1404 	int i, j, g_count = 0, error = 0, allerror = 0;
1405 
1406 	allerror = ext2_sbupdate(mp, waitfor);
1407 
1408 	/* Update gd csums */
1409 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1410 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1411 		ext2_gd_csum_set(fs);
1412 
1413 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1414 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1415 		    ext2_cg_location(fs, i)),
1416 		    fs->e2fs_bsize, 0, 0, 0);
1417 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1418 			memcpy(bp->b_data, &fs->e2fs_gd[
1419 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1420 			    fs->e2fs_bsize);
1421 		} else {
1422 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1423 			    g_count < fs->e2fs_gcount; j++, g_count++)
1424 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1425 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1426 		}
1427 		if (waitfor == MNT_WAIT)
1428 			error = bwrite(bp);
1429 		else
1430 			bawrite(bp);
1431 	}
1432 
1433 	if (!allerror && error)
1434 		allerror = error;
1435 	return (allerror);
1436 }
1437 
1438 /*
1439  * Return the root of a filesystem.
1440  */
1441 static int
1442 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1443 {
1444 	struct vnode *nvp;
1445 	int error;
1446 
1447 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1448 	if (error)
1449 		return (error);
1450 	*vpp = nvp;
1451 	return (0);
1452 }
1453