xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision dbfb4063ae95b956a2b0021c37c9a8be4c2e4393)
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 
82 static int	ext2_flushfiles(struct mount *mp, int flags, struct thread *td);
83 static int	ext2_mountfs(struct vnode *, struct mount *);
84 static int	ext2_reload(struct mount *mp, struct thread *td);
85 static int	ext2_sbupdate(struct ext2mount *, int);
86 static int	ext2_cgupdate(struct ext2mount *, int);
87 static vfs_unmount_t		ext2_unmount;
88 static vfs_root_t		ext2_root;
89 static vfs_statfs_t		ext2_statfs;
90 static vfs_sync_t		ext2_sync;
91 static vfs_vget_t		ext2_vget;
92 static vfs_fhtovp_t		ext2_fhtovp;
93 static vfs_mount_t		ext2_mount;
94 
95 MALLOC_DEFINE(M_EXT2NODE, "ext2_node", "EXT2 vnode private part");
96 static MALLOC_DEFINE(M_EXT2MNT, "ext2_mount", "EXT2 mount structure");
97 
98 static struct vfsops ext2fs_vfsops = {
99 	.vfs_fhtovp =		ext2_fhtovp,
100 	.vfs_mount =		ext2_mount,
101 	.vfs_root =		ext2_root,	/* root inode via vget */
102 	.vfs_statfs =		ext2_statfs,
103 	.vfs_sync =		ext2_sync,
104 	.vfs_unmount =		ext2_unmount,
105 	.vfs_vget =		ext2_vget,
106 };
107 
108 VFS_SET(ext2fs_vfsops, ext2fs, 0);
109 
110 static int	ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev,
111 		    int ronly);
112 static int	ext2_compute_sb_data(struct vnode * devvp,
113 		    struct ext2fs * es, struct m_ext2fs * fs);
114 
115 static const char *ext2_opts[] = { "acls", "async", "noatime", "noclusterr",
116     "noclusterw", "noexec", "export", "force", "from", "multilabel",
117     "suiddir", "nosymfollow", "sync", "union", NULL };
118 
119 /*
120  * VFS Operations.
121  *
122  * mount system call
123  */
124 static int
125 ext2_mount(struct mount *mp)
126 {
127 	struct vfsoptlist *opts;
128 	struct vnode *devvp;
129 	struct thread *td;
130 	struct ext2mount *ump = NULL;
131 	struct m_ext2fs *fs;
132 	struct nameidata nd, *ndp = &nd;
133 	accmode_t accmode;
134 	char *path, *fspec;
135 	int error, flags, len;
136 
137 	td = curthread;
138 	opts = mp->mnt_optnew;
139 
140 	if (vfs_filteropt(opts, ext2_opts))
141 		return (EINVAL);
142 
143 	vfs_getopt(opts, "fspath", (void **)&path, NULL);
144 	/* Double-check the length of path.. */
145 	if (strlen(path) >= MAXMNTLEN)
146 		return (ENAMETOOLONG);
147 
148 	fspec = NULL;
149 	error = vfs_getopt(opts, "from", (void **)&fspec, &len);
150 	if (!error && fspec[len - 1] != '\0')
151 		return (EINVAL);
152 
153 	/*
154 	 * If updating, check whether changing from read-only to
155 	 * read/write; if there is no device name, that's all we do.
156 	 */
157 	if (mp->mnt_flag & MNT_UPDATE) {
158 		ump = VFSTOEXT2(mp);
159 		fs = ump->um_e2fs;
160 		error = 0;
161 		if (fs->e2fs_ronly == 0 &&
162 		    vfs_flagopt(opts, "ro", NULL, 0)) {
163 			error = VFS_SYNC(mp, MNT_WAIT);
164 			if (error)
165 				return (error);
166 			flags = WRITECLOSE;
167 			if (mp->mnt_flag & MNT_FORCE)
168 				flags |= FORCECLOSE;
169 			error = ext2_flushfiles(mp, flags, td);
170 			if (error == 0 && fs->e2fs_wasvalid &&
171 			    ext2_cgupdate(ump, MNT_WAIT) == 0) {
172 				fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
173 				ext2_sbupdate(ump, MNT_WAIT);
174 			}
175 			fs->e2fs_ronly = 1;
176 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
177 			g_topology_lock();
178 			g_access(ump->um_cp, 0, -1, 0);
179 			g_topology_unlock();
180 		}
181 		if (!error && (mp->mnt_flag & MNT_RELOAD))
182 			error = ext2_reload(mp, td);
183 		if (error)
184 			return (error);
185 		devvp = ump->um_devvp;
186 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
187 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
188 				return (EPERM);
189 
190 			/*
191 			 * If upgrade to read-write by non-root, then verify
192 			 * that user has necessary permissions on the device.
193 			 */
194 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
195 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
196 			    td->td_ucred, td);
197 			if (error)
198 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
199 			if (error) {
200 				VOP_UNLOCK(devvp, 0);
201 				return (error);
202 			}
203 			VOP_UNLOCK(devvp, 0);
204 			g_topology_lock();
205 			error = g_access(ump->um_cp, 0, 1, 0);
206 			g_topology_unlock();
207 			if (error)
208 				return (error);
209 
210 			if ((fs->e2fs->e2fs_state & E2FS_ISCLEAN) == 0 ||
211 			    (fs->e2fs->e2fs_state & E2FS_ERRORS)) {
212 				if (mp->mnt_flag & MNT_FORCE) {
213 					printf(
214 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
215 				} else {
216 					printf(
217 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
218 					    fs->e2fs_fsmnt);
219 					return (EPERM);
220 				}
221 			}
222 			fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;
223 			(void)ext2_cgupdate(ump, MNT_WAIT);
224 			fs->e2fs_ronly = 0;
225 			MNT_ILOCK(mp);
226 			mp->mnt_flag &= ~MNT_RDONLY;
227 			MNT_IUNLOCK(mp);
228 		}
229 		if (vfs_flagopt(opts, "export", NULL, 0)) {
230 			/* Process export requests in vfs_mount.c. */
231 			return (error);
232 		}
233 	}
234 
235 	/*
236 	 * Not an update, or updating the name: look up the name
237 	 * and verify that it refers to a sensible disk device.
238 	 */
239 	if (fspec == NULL)
240 		return (EINVAL);
241 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
242 	if ((error = namei(ndp)) != 0)
243 		return (error);
244 	NDFREE(ndp, NDF_ONLY_PNBUF);
245 	devvp = ndp->ni_vp;
246 
247 	if (!vn_isdisk(devvp, &error)) {
248 		vput(devvp);
249 		return (error);
250 	}
251 
252 	/*
253 	 * If mount by non-root, then verify that user has necessary
254 	 * permissions on the device.
255 	 *
256 	 * XXXRW: VOP_ACCESS() enough?
257 	 */
258 	accmode = VREAD;
259 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
260 		accmode |= VWRITE;
261 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
262 	if (error)
263 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
264 	if (error) {
265 		vput(devvp);
266 		return (error);
267 	}
268 
269 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
270 		error = ext2_mountfs(devvp, mp);
271 	} else {
272 		if (devvp != ump->um_devvp) {
273 			vput(devvp);
274 			return (EINVAL);	/* needs translation */
275 		} else
276 			vput(devvp);
277 	}
278 	if (error) {
279 		vrele(devvp);
280 		return (error);
281 	}
282 	ump = VFSTOEXT2(mp);
283 	fs = ump->um_e2fs;
284 
285 	/*
286 	 * Note that this strncpy() is ok because of a check at the start
287 	 * of ext2_mount().
288 	 */
289 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
290 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
291 	vfs_mountedfrom(mp, fspec);
292 	return (0);
293 }
294 
295 static int
296 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
297 {
298 	uint32_t i, mask;
299 
300 	if (es->e2fs_magic != E2FS_MAGIC) {
301 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
302 		    devtoname(dev), es->e2fs_magic, E2FS_MAGIC);
303 		return (1);
304 	}
305 	if (es->e2fs_rev > E2FS_REV0) {
306 		mask = es->e2fs_features_incompat & ~(EXT2F_INCOMPAT_SUPP);
307 		if (mask) {
308 			printf("WARNING: mount of %s denied due to "
309 			    "unsupported optional features:\n", devtoname(dev));
310 			for (i = 0;
311 			    i < sizeof(incompat)/sizeof(struct ext2_feature);
312 			    i++)
313 				if (mask & incompat[i].mask)
314 					printf("%s ", incompat[i].name);
315 			printf("\n");
316 			return (1);
317 		}
318 		mask = es->e2fs_features_rocompat & ~EXT2F_ROCOMPAT_SUPP;
319 		if (!ronly && mask) {
320 			printf("WARNING: R/W mount of %s denied due to "
321 			    "unsupported optional features:\n", devtoname(dev));
322 			for (i = 0;
323 			    i < sizeof(ro_compat)/sizeof(struct ext2_feature);
324 			    i++)
325 				if (mask & ro_compat[i].mask)
326 					printf("%s ", ro_compat[i].name);
327 			printf("\n");
328 			return (1);
329 		}
330 	}
331 	return (0);
332 }
333 
334 static e4fs_daddr_t
335 ext2_cg_location(struct m_ext2fs *fs, int number)
336 {
337 	int cg, descpb, logical_sb, has_super = 0;
338 
339 	/*
340 	 * Adjust logical superblock block number.
341 	 * Godmar thinks: if the blocksize is greater than 1024, then
342 	 * the superblock is logically part of block zero.
343 	 */
344 	logical_sb = fs->e2fs_bsize > SBSIZE ? 0 : 1;
345 
346 	if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
347 	    number < fs->e2fs->e3fs_first_meta_bg)
348 		return (logical_sb + number + 1);
349 
350 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT))
351 		descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
352 	else
353 		descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
354 
355 	cg = descpb * number;
356 
357 	if (ext2_cg_has_sb(fs, cg))
358 		has_super = 1;
359 
360 	return (has_super + cg * (e4fs_daddr_t)EXT2_BLOCKS_PER_GROUP(fs) +
361 	    fs->e2fs->e2fs_first_dblock);
362 }
363 
364 static int
365 ext2_cg_validate(struct m_ext2fs *fs)
366 {
367 	uint64_t b_bitmap;
368 	uint64_t i_bitmap;
369 	uint64_t i_tables;
370 	uint64_t first_block, last_block, last_cg_block;
371 	struct ext2_gd *gd;
372 	unsigned int i, cg_count;
373 
374 	first_block = fs->e2fs->e2fs_first_dblock;
375 	last_cg_block = ext2_cg_number_gdb(fs, 0);
376 	cg_count = fs->e2fs_gcount;
377 
378 	for (i = 0; i < fs->e2fs_gcount; i++) {
379 		gd = &fs->e2fs_gd[i];
380 
381 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
382 		    i == fs->e2fs_gcount - 1) {
383 			last_block = fs->e2fs_bcount - 1;
384 		} else {
385 			last_block = first_block +
386 			    (EXT2_BLOCKS_PER_GROUP(fs) - 1);
387 		}
388 
389 		if ((cg_count == fs->e2fs_gcount) &&
390 		    !(gd->ext4bgd_flags & EXT2_BG_INODE_ZEROED))
391 			cg_count = i;
392 
393 		b_bitmap = e2fs_gd_get_b_bitmap(gd);
394 		if (b_bitmap == 0) {
395 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
396 			    "block bitmap is zero", i);
397 			return (EINVAL);
398 
399 		}
400 		if (b_bitmap <= last_cg_block) {
401 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
402 			    "block bitmap overlaps gds", i);
403 			return (EINVAL);
404 		}
405 		if (b_bitmap < first_block || b_bitmap > last_block) {
406 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
407 			    "block bitmap not in group", i);
408 			return (EINVAL);
409 		}
410 
411 		i_bitmap = e2fs_gd_get_i_bitmap(gd);
412 		if (i_bitmap == 0) {
413 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
414 			    "inode bitmap is zero", i);
415 			return (EINVAL);
416 		}
417 		if (i_bitmap <= last_cg_block) {
418 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
419 			    "inode bitmap overlaps gds", i);
420 			return (EINVAL);
421 		}
422 		if (i_bitmap < first_block || i_bitmap > last_block) {
423 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
424 			    "inode bitmap not in group blk", i);
425 			return (EINVAL);
426 		}
427 
428 		i_tables = e2fs_gd_get_i_tables(gd);
429 		if (i_tables == 0) {
430 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
431 			    "inode table is zero", i);
432 			return (EINVAL);
433 		}
434 		if (i_tables <= last_cg_block) {
435 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
436 			    "inode talbes overlaps gds", i);
437 			return (EINVAL);
438 		}
439 		if (i_tables < first_block ||
440 		    i_tables + fs->e2fs_itpg - 1 > last_block) {
441 			SDT_PROBE2(ext2fs, , vfsops, ext2_cg_validate_error,
442 			    "inode tables not in group blk", i);
443 			return (EINVAL);
444 		}
445 
446 		if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG))
447 			first_block += EXT2_BLOCKS_PER_GROUP(fs);
448 	}
449 
450 	return (0);
451 }
452 
453 /*
454  * This computes the fields of the m_ext2fs structure from the
455  * data in the ext2fs structure read in.
456  */
457 static int
458 ext2_compute_sb_data(struct vnode *devvp, struct ext2fs *es,
459     struct m_ext2fs *fs)
460 {
461 	struct buf *bp;
462 	uint32_t e2fs_descpb, e2fs_gdbcount_alloc;
463 	int i, j;
464 	int g_count = 0;
465 	int error;
466 
467 	/* Check checksum features */
468 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) &&
469 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
470 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
471 		    "incorrect checksum features combination");
472 		return (EINVAL);
473 	}
474 
475 	/* Precompute checksum seed for all metadata */
476 	ext2_sb_csum_set_seed(fs);
477 
478 	/* Verify sb csum if possible */
479 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
480 		error = ext2_sb_csum_verify(fs);
481 		if (error) {
482 			return (error);
483 		}
484 	}
485 
486 	/* Check for block size = 1K|2K|4K */
487 	if (es->e2fs_log_bsize > 2) {
488 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
489 		    "bad block size");
490 		return (EINVAL);
491 	}
492 
493 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
494 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
495 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
496 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
497 
498 	/* Check for fragment size */
499 	if (es->e2fs_log_fsize >
500 	    (EXT2_MAX_FRAG_LOG_SIZE - EXT2_MIN_BLOCK_LOG_SIZE)) {
501 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
502 		    "invalid log cluster size");
503 		return (EINVAL);
504 	}
505 
506 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
507 	if (fs->e2fs_fsize != fs->e2fs_bsize) {
508 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
509 		    "fragment size != block size");
510 		return (EINVAL);
511 	}
512 
513 	fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
514 
515 	/* Check reserved gdt blocks for future filesystem expansion */
516 	if (es->e2fs_reserved_ngdb > (fs->e2fs_bsize / 4)) {
517 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
518 		    "number of reserved GDT blocks too large");
519 		return (EINVAL);
520 	}
521 
522 	if (es->e2fs_rev == E2FS_REV0) {
523 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
524 	} else {
525 		fs->e2fs_isize = es->e2fs_inode_size;
526 
527 		/*
528 		 * Check first ino.
529 		 */
530 		if (es->e2fs_first_ino < EXT2_FIRSTINO) {
531 			SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
532 			    "invalid first ino");
533 			return (EINVAL);
534 		}
535 
536 		/*
537 		 * Simple sanity check for superblock inode size value.
538 		 */
539 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
540 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
541 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
542 			SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
543 			    "invalid inode size");
544 			return (EINVAL);
545 		}
546 	}
547 
548 	/* Check group descriptors */
549 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT) &&
550 	    es->e3fs_desc_size != E2FS_64BIT_GD_SIZE) {
551 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
552 		    "unsupported 64bit descriptor size");
553 		return (EINVAL);
554 	}
555 
556 	fs->e2fs_bpg = es->e2fs_bpg;
557 	fs->e2fs_fpg = es->e2fs_fpg;
558 	if (fs->e2fs_bpg == 0 || fs->e2fs_fpg == 0) {
559 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
560 		    "zero blocks/fragments per group");
561 		return (EINVAL);
562 	}
563 	if (fs->e2fs_bpg != fs->e2fs_bsize * 8) {
564 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
565 		    "non-standard group size unsupported");
566 		return (EINVAL);
567 	}
568 
569 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
570 	if (fs->e2fs_ipb == 0 ||
571 	    fs->e2fs_ipb > fs->e2fs_bsize / E2FS_REV0_INODE_SIZE) {
572 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
573 		    "bad inodes per block size");
574 		return (EINVAL);
575 	}
576 
577 	fs->e2fs_ipg = es->e2fs_ipg;
578 	if (fs->e2fs_ipg < fs->e2fs_ipb || fs->e2fs_ipg >  fs->e2fs_bsize * 8) {
579 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
580 		    "invalid inodes per group");
581 		return (EINVAL);
582 	}
583 
584 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
585 
586 	fs->e2fs_bcount = es->e2fs_bcount;
587 	fs->e2fs_rbcount = es->e2fs_rbcount;
588 	fs->e2fs_fbcount = es->e2fs_fbcount;
589 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
590 		fs->e2fs_bcount |= (uint64_t)(es->e4fs_bcount_hi) << 32;
591 		fs->e2fs_rbcount |= (uint64_t)(es->e4fs_rbcount_hi) << 32;
592 		fs->e2fs_fbcount |= (uint64_t)(es->e4fs_fbcount_hi) << 32;
593 	}
594 	if (fs->e2fs_rbcount > fs->e2fs_bcount ||
595 	    fs->e2fs_fbcount > fs->e2fs_bcount) {
596 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
597 		    "invalid block count");
598 		return (EINVAL);
599 	}
600 	if (es->e2fs_first_dblock >= fs->e2fs_bcount) {
601 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
602 		    "first data block out of range");
603 		return (EINVAL);
604 	}
605 
606 	fs->e2fs_gcount = howmany(fs->e2fs_bcount - es->e2fs_first_dblock,
607 	    EXT2_BLOCKS_PER_GROUP(fs));
608 	if (fs->e2fs_gcount > ((uint64_t)1 << 32) - EXT2_DESCS_PER_BLOCK(fs)) {
609 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
610 		    "groups count too large");
611 		return (EINVAL);
612 	}
613 
614 	/* Check for extra isize in big inodes. */
615 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
616 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
617 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
618 		    "no space for extra inode timestamps");
619 		return (EINVAL);
620 	}
621 
622 	/* s_resuid / s_resgid ? */
623 
624 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
625 		e2fs_descpb = fs->e2fs_bsize / E2FS_64BIT_GD_SIZE;
626 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount, e2fs_descpb);
627 	} else {
628 		e2fs_descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
629 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount,
630 		    fs->e2fs_bsize / sizeof(struct ext2_gd));
631 	}
632 	fs->e2fs_gdbcount = howmany(fs->e2fs_gcount, e2fs_descpb);
633 	fs->e2fs_gd = malloc(e2fs_gdbcount_alloc * fs->e2fs_bsize,
634 	    M_EXT2MNT, M_WAITOK | M_ZERO);
635 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
636 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
637 
638 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
639 		error = bread(devvp,
640 		    fsbtodb(fs, ext2_cg_location(fs, i)),
641 		    fs->e2fs_bsize, NOCRED, &bp);
642 		if (error) {
643 			/*
644 			 * fs->e2fs_gd and fs->e2fs_contigdirs
645 			 * will be freed later by the caller,
646 			 * because this function could be called from
647 			 * MNT_UPDATE path.
648 			 */
649 			brelse(bp);
650 			return (error);
651 		}
652 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
653 			memcpy(&fs->e2fs_gd[
654 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
655 			    bp->b_data, fs->e2fs_bsize);
656 		} else {
657 			for (j = 0; j < e2fs_descpb &&
658 			    g_count < fs->e2fs_gcount; j++, g_count++)
659 				memcpy(&fs->e2fs_gd[g_count],
660 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
661 				    E2FS_REV0_GD_SIZE);
662 		}
663 		brelse(bp);
664 		bp = NULL;
665 	}
666 
667 	/* Validate cgs consistency */
668 	error = ext2_cg_validate(fs);
669 	if (error)
670 		return (error);
671 
672 	/* Verfy cgs csum */
673 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
674 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
675 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
676 		if (error)
677 			return (error);
678 	}
679 	/* Initialization for the ext2 Orlov allocator variant. */
680 	fs->e2fs_total_dir = 0;
681 	for (i = 0; i < fs->e2fs_gcount; i++)
682 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
683 
684 	if (es->e2fs_rev == E2FS_REV0 ||
685 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
686 		fs->e2fs_maxfilesize = 0x7fffffff;
687 	else {
688 		fs->e2fs_maxfilesize = 0xffffffffffff;
689 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
690 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
691 	}
692 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
693 		fs->e2fs_uhash = 3;
694 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
695 #ifdef __CHAR_UNSIGNED__
696 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
697 		fs->e2fs_uhash = 3;
698 #else
699 		es->e4fs_flags |= E2FS_SIGNED_HASH;
700 #endif
701 	}
702 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
703 		error = ext2_sb_csum_verify(fs);
704 
705 	return (error);
706 }
707 
708 /*
709  * Reload all incore data for a filesystem (used after running fsck on
710  * the root filesystem and finding things to fix). The filesystem must
711  * be mounted read-only.
712  *
713  * Things to do to update the mount:
714  *	1) invalidate all cached meta-data.
715  *	2) re-read superblock from disk.
716  *	3) invalidate all cluster summary information.
717  *	4) invalidate all inactive vnodes.
718  *	5) invalidate all cached file data.
719  *	6) re-read inode data for all active vnodes.
720  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
721  */
722 static int
723 ext2_reload(struct mount *mp, struct thread *td)
724 {
725 	struct vnode *vp, *mvp, *devvp;
726 	struct inode *ip;
727 	struct buf *bp;
728 	struct ext2fs *es;
729 	struct m_ext2fs *fs;
730 	struct csum *sump;
731 	int error, i;
732 	int32_t *lp;
733 
734 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
735 		return (EINVAL);
736 	/*
737 	 * Step 1: invalidate all cached meta-data.
738 	 */
739 	devvp = VFSTOEXT2(mp)->um_devvp;
740 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
741 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
742 		panic("ext2_reload: dirty1");
743 	VOP_UNLOCK(devvp, 0);
744 
745 	/*
746 	 * Step 2: re-read superblock from disk.
747 	 * constants have been adjusted for ext2
748 	 */
749 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
750 		return (error);
751 	es = (struct ext2fs *)bp->b_data;
752 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
753 		brelse(bp);
754 		return (EIO);		/* XXX needs translation */
755 	}
756 	fs = VFSTOEXT2(mp)->um_e2fs;
757 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
758 
759 	if ((error = ext2_compute_sb_data(devvp, es, fs)) != 0) {
760 		brelse(bp);
761 		return (error);
762 	}
763 #ifdef UNKLAR
764 	if (fs->fs_sbsize < SBSIZE)
765 		bp->b_flags |= B_INVAL;
766 #endif
767 	brelse(bp);
768 
769 	/*
770 	 * Step 3: invalidate all cluster summary information.
771 	 */
772 	if (fs->e2fs_contigsumsize > 0) {
773 		lp = fs->e2fs_maxcluster;
774 		sump = fs->e2fs_clustersum;
775 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
776 			*lp++ = fs->e2fs_contigsumsize;
777 			sump->cs_init = 0;
778 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
779 		}
780 	}
781 
782 loop:
783 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
784 		/*
785 		 * Step 4: invalidate all cached file data.
786 		 */
787 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
788 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
789 			goto loop;
790 		}
791 		if (vinvalbuf(vp, 0, 0, 0))
792 			panic("ext2_reload: dirty2");
793 
794 		/*
795 		 * Step 5: re-read inode data for all active vnodes.
796 		 */
797 		ip = VTOI(vp);
798 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
799 		    (int)fs->e2fs_bsize, NOCRED, &bp);
800 		if (error) {
801 			VOP_UNLOCK(vp, 0);
802 			vrele(vp);
803 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
804 			return (error);
805 		}
806 
807 		error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
808 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
809 
810 		brelse(bp);
811 		VOP_UNLOCK(vp, 0);
812 		vrele(vp);
813 
814 		if (error) {
815 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
816 			return (error);
817 		}
818 	}
819 	return (0);
820 }
821 
822 /*
823  * Common code for mount and mountroot.
824  */
825 static int
826 ext2_mountfs(struct vnode *devvp, struct mount *mp)
827 {
828 	struct ext2mount *ump;
829 	struct buf *bp;
830 	struct m_ext2fs *fs;
831 	struct ext2fs *es;
832 	struct cdev *dev = devvp->v_rdev;
833 	struct g_consumer *cp;
834 	struct bufobj *bo;
835 	struct csum *sump;
836 	int error;
837 	int ronly;
838 	int i;
839 	u_long size;
840 	int32_t *lp;
841 	int32_t e2fs_maxcontig;
842 
843 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
844 	/* XXX: use VOP_ACESS to check FS perms */
845 	g_topology_lock();
846 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
847 	g_topology_unlock();
848 	VOP_UNLOCK(devvp, 0);
849 	if (error)
850 		return (error);
851 
852 	/* XXX: should we check for some sectorsize or 512 instead? */
853 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
854 	    (SBSIZE < cp->provider->sectorsize)) {
855 		g_topology_lock();
856 		g_vfs_close(cp);
857 		g_topology_unlock();
858 		return (EINVAL);
859 	}
860 
861 	bo = &devvp->v_bufobj;
862 	bo->bo_private = cp;
863 	bo->bo_ops = g_vfs_bufops;
864 	if (devvp->v_rdev->si_iosize_max != 0)
865 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
866 	if (mp->mnt_iosize_max > MAXPHYS)
867 		mp->mnt_iosize_max = MAXPHYS;
868 
869 	bp = NULL;
870 	ump = NULL;
871 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
872 		goto out;
873 	es = (struct ext2fs *)bp->b_data;
874 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
875 		error = EINVAL;		/* XXX needs translation */
876 		goto out;
877 	}
878 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
879 	    (es->e2fs_state & E2FS_ERRORS)) {
880 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
881 			printf(
882 "WARNING: Filesystem was not properly dismounted\n");
883 		} else {
884 			printf(
885 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
886 			error = EPERM;
887 			goto out;
888 		}
889 	}
890 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
891 
892 	/*
893 	 * I don't know whether this is the right strategy. Note that
894 	 * we dynamically allocate both an m_ext2fs and an ext2fs
895 	 * while Linux keeps the super block in a locked buffer.
896 	 */
897 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
898 	    M_EXT2MNT, M_WAITOK | M_ZERO);
899 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
900 	    M_EXT2MNT, M_WAITOK);
901 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
902 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
903 	if ((error = ext2_compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
904 		goto out;
905 
906 	/*
907 	 * Calculate the maximum contiguous blocks and size of cluster summary
908 	 * array.  In FFS this is done by newfs; however, the superblock
909 	 * in ext2fs doesn't have these variables, so we can calculate
910 	 * them here.
911 	 */
912 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
913 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
914 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
915 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
916 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
917 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
918 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
919 		lp = ump->um_e2fs->e2fs_maxcluster;
920 		sump = ump->um_e2fs->e2fs_clustersum;
921 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
922 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
923 			sump->cs_init = 0;
924 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
925 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
926 		}
927 	}
928 
929 	brelse(bp);
930 	bp = NULL;
931 	fs = ump->um_e2fs;
932 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
933 
934 	/*
935 	 * If the fs is not mounted read-only, make sure the super block is
936 	 * always written back on a sync().
937 	 */
938 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
939 	if (ronly == 0) {
940 		fs->e2fs_fmod = 1;	/* mark it modified */
941 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
942 	}
943 	mp->mnt_data = ump;
944 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
945 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
946 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
947 	MNT_ILOCK(mp);
948 	mp->mnt_flag |= MNT_LOCAL;
949 	MNT_IUNLOCK(mp);
950 	ump->um_mountp = mp;
951 	ump->um_dev = dev;
952 	ump->um_devvp = devvp;
953 	ump->um_bo = &devvp->v_bufobj;
954 	ump->um_cp = cp;
955 
956 	/*
957 	 * Setting those two parameters allowed us to use
958 	 * ufs_bmap w/o changse!
959 	 */
960 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
961 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
962 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
963 	if (ronly == 0)
964 		ext2_sbupdate(ump, MNT_WAIT);
965 	/*
966 	 * Initialize filesystem stat information in mount struct.
967 	 */
968 	MNT_ILOCK(mp);
969 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
970 	    MNTK_USES_BCACHE;
971 	MNT_IUNLOCK(mp);
972 	return (0);
973 out:
974 	if (bp)
975 		brelse(bp);
976 	if (cp != NULL) {
977 		g_topology_lock();
978 		g_vfs_close(cp);
979 		g_topology_unlock();
980 	}
981 	if (ump) {
982 		mtx_destroy(EXT2_MTX(ump));
983 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
984 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
985 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
986 		free(ump->um_e2fs, M_EXT2MNT);
987 		free(ump, M_EXT2MNT);
988 		mp->mnt_data = NULL;
989 	}
990 	return (error);
991 }
992 
993 /*
994  * Unmount system call.
995  */
996 static int
997 ext2_unmount(struct mount *mp, int mntflags)
998 {
999 	struct ext2mount *ump;
1000 	struct m_ext2fs *fs;
1001 	struct csum *sump;
1002 	int error, flags, i, ronly;
1003 
1004 	flags = 0;
1005 	if (mntflags & MNT_FORCE) {
1006 		if (mp->mnt_flag & MNT_ROOTFS)
1007 			return (EINVAL);
1008 		flags |= FORCECLOSE;
1009 	}
1010 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
1011 		return (error);
1012 	ump = VFSTOEXT2(mp);
1013 	fs = ump->um_e2fs;
1014 	ronly = fs->e2fs_ronly;
1015 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
1016 		if (fs->e2fs_wasvalid)
1017 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
1018 		ext2_sbupdate(ump, MNT_WAIT);
1019 	}
1020 
1021 	g_topology_lock();
1022 	g_vfs_close(ump->um_cp);
1023 	g_topology_unlock();
1024 	vrele(ump->um_devvp);
1025 	sump = fs->e2fs_clustersum;
1026 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
1027 		free(sump->cs_sum, M_EXT2MNT);
1028 	free(fs->e2fs_clustersum, M_EXT2MNT);
1029 	free(fs->e2fs_maxcluster, M_EXT2MNT);
1030 	free(fs->e2fs_gd, M_EXT2MNT);
1031 	free(fs->e2fs_contigdirs, M_EXT2MNT);
1032 	free(fs->e2fs, M_EXT2MNT);
1033 	free(fs, M_EXT2MNT);
1034 	free(ump, M_EXT2MNT);
1035 	mp->mnt_data = NULL;
1036 	MNT_ILOCK(mp);
1037 	mp->mnt_flag &= ~MNT_LOCAL;
1038 	MNT_IUNLOCK(mp);
1039 	return (error);
1040 }
1041 
1042 /*
1043  * Flush out all the files in a filesystem.
1044  */
1045 static int
1046 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
1047 {
1048 	int error;
1049 
1050 	error = vflush(mp, 0, flags, td);
1051 	return (error);
1052 }
1053 
1054 /*
1055  * Get filesystem statistics.
1056  */
1057 int
1058 ext2_statfs(struct mount *mp, struct statfs *sbp)
1059 {
1060 	struct ext2mount *ump;
1061 	struct m_ext2fs *fs;
1062 	uint32_t overhead, overhead_per_group, ngdb;
1063 	int i, ngroups;
1064 
1065 	ump = VFSTOEXT2(mp);
1066 	fs = ump->um_e2fs;
1067 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
1068 		panic("ext2_statfs");
1069 
1070 	/*
1071 	 * Compute the overhead (FS structures)
1072 	 */
1073 	overhead_per_group =
1074 	    1 /* block bitmap */ +
1075 	    1 /* inode bitmap */ +
1076 	    fs->e2fs_itpg;
1077 	overhead = fs->e2fs->e2fs_first_dblock +
1078 	    fs->e2fs_gcount * overhead_per_group;
1079 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
1080 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
1081 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
1082 			if (ext2_cg_has_sb(fs, i))
1083 				ngroups++;
1084 		}
1085 	} else {
1086 		ngroups = fs->e2fs_gcount;
1087 	}
1088 	ngdb = fs->e2fs_gdbcount;
1089 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
1090 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
1091 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
1092 	overhead += ngroups * (1 /* superblock */ + ngdb);
1093 
1094 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
1095 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
1096 	sbp->f_blocks = fs->e2fs_bcount - overhead;
1097 	sbp->f_bfree = fs->e2fs_fbcount;
1098 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
1099 	sbp->f_files = fs->e2fs->e2fs_icount;
1100 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
1101 	return (0);
1102 }
1103 
1104 /*
1105  * Go through the disk queues to initiate sandbagged IO;
1106  * go through the inodes to write those that have been modified;
1107  * initiate the writing of the super block if it has been modified.
1108  *
1109  * Note: we are always called with the filesystem marked `MPBUSY'.
1110  */
1111 static int
1112 ext2_sync(struct mount *mp, int waitfor)
1113 {
1114 	struct vnode *mvp, *vp;
1115 	struct thread *td;
1116 	struct inode *ip;
1117 	struct ext2mount *ump = VFSTOEXT2(mp);
1118 	struct m_ext2fs *fs;
1119 	int error, allerror = 0;
1120 
1121 	td = curthread;
1122 	fs = ump->um_e2fs;
1123 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
1124 		panic("ext2_sync: rofs mod fs=%s", fs->e2fs_fsmnt);
1125 	}
1126 
1127 	/*
1128 	 * Write back each (modified) inode.
1129 	 */
1130 loop:
1131 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1132 		if (vp->v_type == VNON) {
1133 			VI_UNLOCK(vp);
1134 			continue;
1135 		}
1136 		ip = VTOI(vp);
1137 		if ((ip->i_flag &
1138 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1139 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1140 		    waitfor == MNT_LAZY)) {
1141 			VI_UNLOCK(vp);
1142 			continue;
1143 		}
1144 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
1145 		if (error) {
1146 			if (error == ENOENT) {
1147 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1148 				goto loop;
1149 			}
1150 			continue;
1151 		}
1152 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
1153 			allerror = error;
1154 		VOP_UNLOCK(vp, 0);
1155 		vrele(vp);
1156 	}
1157 
1158 	/*
1159 	 * Force stale filesystem control information to be flushed.
1160 	 */
1161 	if (waitfor != MNT_LAZY) {
1162 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1163 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
1164 			allerror = error;
1165 		VOP_UNLOCK(ump->um_devvp, 0);
1166 	}
1167 
1168 	/*
1169 	 * Write back modified superblock.
1170 	 */
1171 	if (fs->e2fs_fmod != 0) {
1172 		fs->e2fs_fmod = 0;
1173 		fs->e2fs->e2fs_wtime = time_second;
1174 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
1175 			allerror = error;
1176 	}
1177 	return (allerror);
1178 }
1179 
1180 /*
1181  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
1182  * in from disk.  If it is in core, wait for the lock bit to clear, then
1183  * return the inode locked.  Detection and handling of mount points must be
1184  * done by the calling routine.
1185  */
1186 static int
1187 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
1188 {
1189 	struct m_ext2fs *fs;
1190 	struct inode *ip;
1191 	struct ext2mount *ump;
1192 	struct buf *bp;
1193 	struct vnode *vp;
1194 	struct thread *td;
1195 	unsigned int i, used_blocks;
1196 	int error;
1197 
1198 	td = curthread;
1199 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
1200 	if (error || *vpp != NULL)
1201 		return (error);
1202 
1203 	ump = VFSTOEXT2(mp);
1204 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
1205 
1206 	/* Allocate a new vnode/inode. */
1207 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
1208 		*vpp = NULL;
1209 		free(ip, M_EXT2NODE);
1210 		return (error);
1211 	}
1212 	vp->v_data = ip;
1213 	ip->i_vnode = vp;
1214 	ip->i_e2fs = fs = ump->um_e2fs;
1215 	ip->i_ump = ump;
1216 	ip->i_number = ino;
1217 
1218 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1219 	error = insmntque(vp, mp);
1220 	if (error != 0) {
1221 		free(ip, M_EXT2NODE);
1222 		*vpp = NULL;
1223 		return (error);
1224 	}
1225 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
1226 	if (error || *vpp != NULL)
1227 		return (error);
1228 
1229 	/* Read in the disk contents for the inode, copy into the inode. */
1230 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1231 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
1232 		/*
1233 		 * The inode does not contain anything useful, so it would
1234 		 * be misleading to leave it on its hash chain. With mode
1235 		 * still zero, it will be unlinked and returned to the free
1236 		 * list by vput().
1237 		 */
1238 		brelse(bp);
1239 		vput(vp);
1240 		*vpp = NULL;
1241 		return (error);
1242 	}
1243 	/* convert ext2 inode to dinode */
1244 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1245 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1246 	if (error) {
1247 		brelse(bp);
1248 		vput(vp);
1249 		*vpp = NULL;
1250 		return (error);
1251 	}
1252 	ip->i_block_group = ino_to_cg(fs, ino);
1253 	ip->i_next_alloc_block = 0;
1254 	ip->i_next_alloc_goal = 0;
1255 
1256 	/*
1257 	 * Now we want to make sure that block pointers for unused
1258 	 * blocks are zeroed out - ext2_balloc depends on this
1259 	 * although for regular files and directories only
1260 	 *
1261 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1262 	 * out because we could corrupt the extent tree.
1263 	 */
1264 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1265 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1266 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1267 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1268 			ip->i_db[i] = 0;
1269 	}
1270 #ifdef EXT2FS_PRINT_EXTENTS
1271 	ext2_print_inode(ip);
1272 	ext4_ext_print_extent_tree_status(ip);
1273 #endif
1274 	bqrelse(bp);
1275 
1276 	/*
1277 	 * Initialize the vnode from the inode, check for aliases.
1278 	 * Note that the underlying vnode may have changed.
1279 	 */
1280 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1281 		vput(vp);
1282 		*vpp = NULL;
1283 		return (error);
1284 	}
1285 
1286 	/*
1287 	 * Finish inode initialization.
1288 	 */
1289 
1290 	*vpp = vp;
1291 	return (0);
1292 }
1293 
1294 /*
1295  * File handle to vnode
1296  *
1297  * Have to be really careful about stale file handles:
1298  * - check that the inode number is valid
1299  * - call ext2_vget() to get the locked inode
1300  * - check for an unallocated inode (i_mode == 0)
1301  * - check that the given client host has export rights and return
1302  *   those rights via. exflagsp and credanonp
1303  */
1304 static int
1305 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1306 {
1307 	struct inode *ip;
1308 	struct ufid *ufhp;
1309 	struct vnode *nvp;
1310 	struct m_ext2fs *fs;
1311 	int error;
1312 
1313 	ufhp = (struct ufid *)fhp;
1314 	fs = VFSTOEXT2(mp)->um_e2fs;
1315 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1316 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1317 		return (ESTALE);
1318 
1319 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1320 	if (error) {
1321 		*vpp = NULLVP;
1322 		return (error);
1323 	}
1324 	ip = VTOI(nvp);
1325 	if (ip->i_mode == 0 ||
1326 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1327 		vput(nvp);
1328 		*vpp = NULLVP;
1329 		return (ESTALE);
1330 	}
1331 	*vpp = nvp;
1332 	vnode_create_vobject(*vpp, 0, curthread);
1333 	return (0);
1334 }
1335 
1336 /*
1337  * Write a superblock and associated information back to disk.
1338  */
1339 static int
1340 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1341 {
1342 	struct m_ext2fs *fs = mp->um_e2fs;
1343 	struct ext2fs *es = fs->e2fs;
1344 	struct buf *bp;
1345 	int error = 0;
1346 
1347 	es->e2fs_bcount = fs->e2fs_bcount & 0xffffffff;
1348 	es->e2fs_rbcount = fs->e2fs_rbcount & 0xffffffff;
1349 	es->e2fs_fbcount = fs->e2fs_fbcount & 0xffffffff;
1350 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1351 		es->e4fs_bcount_hi = fs->e2fs_bcount >> 32;
1352 		es->e4fs_rbcount_hi = fs->e2fs_rbcount >> 32;
1353 		es->e4fs_fbcount_hi = fs->e2fs_fbcount >> 32;
1354 	}
1355 
1356 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1357 		ext2_sb_csum_set(fs);
1358 
1359 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1360 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1361 	if (waitfor == MNT_WAIT)
1362 		error = bwrite(bp);
1363 	else
1364 		bawrite(bp);
1365 
1366 	/*
1367 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1368 	 * are not busy at this point and are (hopefully) written by the
1369 	 * usual sync mechanism. No need to write them here.
1370 	 */
1371 	return (error);
1372 }
1373 int
1374 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1375 {
1376 	struct m_ext2fs *fs = mp->um_e2fs;
1377 	struct buf *bp;
1378 	int i, j, g_count = 0, error = 0, allerror = 0;
1379 
1380 	allerror = ext2_sbupdate(mp, waitfor);
1381 
1382 	/* Update gd csums */
1383 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1384 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1385 		ext2_gd_csum_set(fs);
1386 
1387 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1388 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1389 		    ext2_cg_location(fs, i)),
1390 		    fs->e2fs_bsize, 0, 0, 0);
1391 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1392 			memcpy(bp->b_data, &fs->e2fs_gd[
1393 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1394 			    fs->e2fs_bsize);
1395 		} else {
1396 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1397 			    g_count < fs->e2fs_gcount; j++, g_count++)
1398 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1399 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1400 		}
1401 		if (waitfor == MNT_WAIT)
1402 			error = bwrite(bp);
1403 		else
1404 			bawrite(bp);
1405 	}
1406 
1407 	if (!allerror && error)
1408 		allerror = error;
1409 	return (allerror);
1410 }
1411 
1412 /*
1413  * Return the root of a filesystem.
1414  */
1415 static int
1416 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1417 {
1418 	struct vnode *nvp;
1419 	int error;
1420 
1421 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1422 	if (error)
1423 		return (error);
1424 	*vpp = nvp;
1425 	return (0);
1426 }
1427