xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision 7b01d357a29d6aa0dcafb04a9a370332e33eaae1)
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 	} else if (fs->e2fs_bpg != fs->e2fs_fpg) {
563 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
564 		    "blocks per group not equal fragments per group");
565 		return (EINVAL);
566 	}
567 
568 	if (fs->e2fs_bpg != fs->e2fs_bsize * 8) {
569 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
570 		    "non-standard group size unsupported");
571 		return (EINVAL);
572 	}
573 
574 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
575 	if (fs->e2fs_ipb == 0 ||
576 	    fs->e2fs_ipb > fs->e2fs_bsize / E2FS_REV0_INODE_SIZE) {
577 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
578 		    "bad inodes per block size");
579 		return (EINVAL);
580 	}
581 
582 	fs->e2fs_ipg = es->e2fs_ipg;
583 	if (fs->e2fs_ipg < fs->e2fs_ipb || fs->e2fs_ipg >  fs->e2fs_bsize * 8) {
584 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
585 		    "invalid inodes per group");
586 		return (EINVAL);
587 	}
588 
589 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
590 
591 	fs->e2fs_bcount = es->e2fs_bcount;
592 	fs->e2fs_rbcount = es->e2fs_rbcount;
593 	fs->e2fs_fbcount = es->e2fs_fbcount;
594 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
595 		fs->e2fs_bcount |= (uint64_t)(es->e4fs_bcount_hi) << 32;
596 		fs->e2fs_rbcount |= (uint64_t)(es->e4fs_rbcount_hi) << 32;
597 		fs->e2fs_fbcount |= (uint64_t)(es->e4fs_fbcount_hi) << 32;
598 	}
599 	if (fs->e2fs_rbcount > fs->e2fs_bcount ||
600 	    fs->e2fs_fbcount > fs->e2fs_bcount) {
601 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
602 		    "invalid block count");
603 		return (EINVAL);
604 	}
605 	if (es->e2fs_first_dblock >= fs->e2fs_bcount) {
606 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
607 		    "first data block out of range");
608 		return (EINVAL);
609 	}
610 
611 	fs->e2fs_gcount = howmany(fs->e2fs_bcount - es->e2fs_first_dblock,
612 	    EXT2_BLOCKS_PER_GROUP(fs));
613 	if (fs->e2fs_gcount > ((uint64_t)1 << 32) - EXT2_DESCS_PER_BLOCK(fs)) {
614 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
615 		    "groups count too large");
616 		return (EINVAL);
617 	}
618 
619 	/* Check for extra isize in big inodes. */
620 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
621 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
622 		SDT_PROBE1(ext2fs, , vfsops, ext2_compute_sb_data_error,
623 		    "no space for extra inode timestamps");
624 		return (EINVAL);
625 	}
626 
627 	/* s_resuid / s_resgid ? */
628 
629 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
630 		e2fs_descpb = fs->e2fs_bsize / E2FS_64BIT_GD_SIZE;
631 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount, e2fs_descpb);
632 	} else {
633 		e2fs_descpb = fs->e2fs_bsize / E2FS_REV0_GD_SIZE;
634 		e2fs_gdbcount_alloc = howmany(fs->e2fs_gcount,
635 		    fs->e2fs_bsize / sizeof(struct ext2_gd));
636 	}
637 	fs->e2fs_gdbcount = howmany(fs->e2fs_gcount, e2fs_descpb);
638 	fs->e2fs_gd = malloc(e2fs_gdbcount_alloc * fs->e2fs_bsize,
639 	    M_EXT2MNT, M_WAITOK | M_ZERO);
640 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
641 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
642 
643 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
644 		error = bread(devvp,
645 		    fsbtodb(fs, ext2_cg_location(fs, i)),
646 		    fs->e2fs_bsize, NOCRED, &bp);
647 		if (error) {
648 			/*
649 			 * fs->e2fs_gd and fs->e2fs_contigdirs
650 			 * will be freed later by the caller,
651 			 * because this function could be called from
652 			 * MNT_UPDATE path.
653 			 */
654 			return (error);
655 		}
656 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
657 			memcpy(&fs->e2fs_gd[
658 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
659 			    bp->b_data, fs->e2fs_bsize);
660 		} else {
661 			for (j = 0; j < e2fs_descpb &&
662 			    g_count < fs->e2fs_gcount; j++, g_count++)
663 				memcpy(&fs->e2fs_gd[g_count],
664 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
665 				    E2FS_REV0_GD_SIZE);
666 		}
667 		brelse(bp);
668 		bp = NULL;
669 	}
670 
671 	/* Validate cgs consistency */
672 	error = ext2_cg_validate(fs);
673 	if (error)
674 		return (error);
675 
676 	/* Verfy cgs csum */
677 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
678 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
679 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
680 		if (error)
681 			return (error);
682 	}
683 	/* Initialization for the ext2 Orlov allocator variant. */
684 	fs->e2fs_total_dir = 0;
685 	for (i = 0; i < fs->e2fs_gcount; i++)
686 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
687 
688 	if (es->e2fs_rev == E2FS_REV0 ||
689 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
690 		fs->e2fs_maxfilesize = 0x7fffffff;
691 	else {
692 		fs->e2fs_maxfilesize = 0xffffffffffff;
693 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
694 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
695 	}
696 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
697 		fs->e2fs_uhash = 3;
698 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
699 #ifdef __CHAR_UNSIGNED__
700 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
701 		fs->e2fs_uhash = 3;
702 #else
703 		es->e4fs_flags |= E2FS_SIGNED_HASH;
704 #endif
705 	}
706 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
707 		error = ext2_sb_csum_verify(fs);
708 
709 	return (error);
710 }
711 
712 /*
713  * Reload all incore data for a filesystem (used after running fsck on
714  * the root filesystem and finding things to fix). The filesystem must
715  * be mounted read-only.
716  *
717  * Things to do to update the mount:
718  *	1) invalidate all cached meta-data.
719  *	2) re-read superblock from disk.
720  *	3) invalidate all cluster summary information.
721  *	4) invalidate all inactive vnodes.
722  *	5) invalidate all cached file data.
723  *	6) re-read inode data for all active vnodes.
724  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
725  */
726 static int
727 ext2_reload(struct mount *mp, struct thread *td)
728 {
729 	struct vnode *vp, *mvp, *devvp;
730 	struct inode *ip;
731 	struct buf *bp;
732 	struct ext2fs *es;
733 	struct m_ext2fs *fs;
734 	struct csum *sump;
735 	int error, i;
736 	int32_t *lp;
737 
738 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
739 		return (EINVAL);
740 	/*
741 	 * Step 1: invalidate all cached meta-data.
742 	 */
743 	devvp = VFSTOEXT2(mp)->um_devvp;
744 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
745 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
746 		panic("ext2_reload: dirty1");
747 	VOP_UNLOCK(devvp, 0);
748 
749 	/*
750 	 * Step 2: re-read superblock from disk.
751 	 * constants have been adjusted for ext2
752 	 */
753 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
754 		return (error);
755 	es = (struct ext2fs *)bp->b_data;
756 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
757 		brelse(bp);
758 		return (EIO);		/* XXX needs translation */
759 	}
760 	fs = VFSTOEXT2(mp)->um_e2fs;
761 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
762 
763 	if ((error = ext2_compute_sb_data(devvp, es, fs)) != 0) {
764 		brelse(bp);
765 		return (error);
766 	}
767 #ifdef UNKLAR
768 	if (fs->fs_sbsize < SBSIZE)
769 		bp->b_flags |= B_INVAL;
770 #endif
771 	brelse(bp);
772 
773 	/*
774 	 * Step 3: invalidate all cluster summary information.
775 	 */
776 	if (fs->e2fs_contigsumsize > 0) {
777 		lp = fs->e2fs_maxcluster;
778 		sump = fs->e2fs_clustersum;
779 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
780 			*lp++ = fs->e2fs_contigsumsize;
781 			sump->cs_init = 0;
782 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
783 		}
784 	}
785 
786 loop:
787 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
788 		/*
789 		 * Step 4: invalidate all cached file data.
790 		 */
791 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
792 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
793 			goto loop;
794 		}
795 		if (vinvalbuf(vp, 0, 0, 0))
796 			panic("ext2_reload: dirty2");
797 
798 		/*
799 		 * Step 5: re-read inode data for all active vnodes.
800 		 */
801 		ip = VTOI(vp);
802 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
803 		    (int)fs->e2fs_bsize, NOCRED, &bp);
804 		if (error) {
805 			VOP_UNLOCK(vp, 0);
806 			vrele(vp);
807 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
808 			return (error);
809 		}
810 
811 		error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
812 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
813 
814 		brelse(bp);
815 		VOP_UNLOCK(vp, 0);
816 		vrele(vp);
817 
818 		if (error) {
819 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
820 			return (error);
821 		}
822 	}
823 	return (0);
824 }
825 
826 /*
827  * Common code for mount and mountroot.
828  */
829 static int
830 ext2_mountfs(struct vnode *devvp, struct mount *mp)
831 {
832 	struct ext2mount *ump;
833 	struct buf *bp;
834 	struct m_ext2fs *fs;
835 	struct ext2fs *es;
836 	struct cdev *dev = devvp->v_rdev;
837 	struct g_consumer *cp;
838 	struct bufobj *bo;
839 	struct csum *sump;
840 	int error;
841 	int ronly;
842 	int i;
843 	u_long size;
844 	int32_t *lp;
845 	int32_t e2fs_maxcontig;
846 
847 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
848 	/* XXX: use VOP_ACESS to check FS perms */
849 	g_topology_lock();
850 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
851 	g_topology_unlock();
852 	VOP_UNLOCK(devvp, 0);
853 	if (error)
854 		return (error);
855 
856 	/* XXX: should we check for some sectorsize or 512 instead? */
857 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
858 	    (SBSIZE < cp->provider->sectorsize)) {
859 		g_topology_lock();
860 		g_vfs_close(cp);
861 		g_topology_unlock();
862 		return (EINVAL);
863 	}
864 
865 	bo = &devvp->v_bufobj;
866 	bo->bo_private = cp;
867 	bo->bo_ops = g_vfs_bufops;
868 	if (devvp->v_rdev->si_iosize_max != 0)
869 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
870 	if (mp->mnt_iosize_max > MAXPHYS)
871 		mp->mnt_iosize_max = MAXPHYS;
872 
873 	bp = NULL;
874 	ump = NULL;
875 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
876 		goto out;
877 	es = (struct ext2fs *)bp->b_data;
878 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
879 		error = EINVAL;		/* XXX needs translation */
880 		goto out;
881 	}
882 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
883 	    (es->e2fs_state & E2FS_ERRORS)) {
884 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
885 			printf(
886 "WARNING: Filesystem was not properly dismounted\n");
887 		} else {
888 			printf(
889 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
890 			error = EPERM;
891 			goto out;
892 		}
893 	}
894 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
895 
896 	/*
897 	 * I don't know whether this is the right strategy. Note that
898 	 * we dynamically allocate both an m_ext2fs and an ext2fs
899 	 * while Linux keeps the super block in a locked buffer.
900 	 */
901 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
902 	    M_EXT2MNT, M_WAITOK | M_ZERO);
903 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
904 	    M_EXT2MNT, M_WAITOK);
905 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
906 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
907 	if ((error = ext2_compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
908 		goto out;
909 
910 	/*
911 	 * Calculate the maximum contiguous blocks and size of cluster summary
912 	 * array.  In FFS this is done by newfs; however, the superblock
913 	 * in ext2fs doesn't have these variables, so we can calculate
914 	 * them here.
915 	 */
916 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
917 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
918 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
919 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
920 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
921 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
922 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
923 		lp = ump->um_e2fs->e2fs_maxcluster;
924 		sump = ump->um_e2fs->e2fs_clustersum;
925 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
926 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
927 			sump->cs_init = 0;
928 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
929 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
930 		}
931 	}
932 
933 	brelse(bp);
934 	bp = NULL;
935 	fs = ump->um_e2fs;
936 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
937 
938 	/*
939 	 * If the fs is not mounted read-only, make sure the super block is
940 	 * always written back on a sync().
941 	 */
942 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
943 	if (ronly == 0) {
944 		fs->e2fs_fmod = 1;	/* mark it modified */
945 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
946 	}
947 	mp->mnt_data = ump;
948 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
949 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
950 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
951 	MNT_ILOCK(mp);
952 	mp->mnt_flag |= MNT_LOCAL;
953 	MNT_IUNLOCK(mp);
954 	ump->um_mountp = mp;
955 	ump->um_dev = dev;
956 	ump->um_devvp = devvp;
957 	ump->um_bo = &devvp->v_bufobj;
958 	ump->um_cp = cp;
959 
960 	/*
961 	 * Setting those two parameters allowed us to use
962 	 * ufs_bmap w/o changse!
963 	 */
964 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
965 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
966 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
967 	if (ronly == 0)
968 		ext2_sbupdate(ump, MNT_WAIT);
969 	/*
970 	 * Initialize filesystem stat information in mount struct.
971 	 */
972 	MNT_ILOCK(mp);
973 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
974 	    MNTK_USES_BCACHE;
975 	MNT_IUNLOCK(mp);
976 	return (0);
977 out:
978 	if (bp)
979 		brelse(bp);
980 	if (cp != NULL) {
981 		g_topology_lock();
982 		g_vfs_close(cp);
983 		g_topology_unlock();
984 	}
985 	if (ump) {
986 		mtx_destroy(EXT2_MTX(ump));
987 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
988 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
989 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
990 		free(ump->um_e2fs, M_EXT2MNT);
991 		free(ump, M_EXT2MNT);
992 		mp->mnt_data = NULL;
993 	}
994 	return (error);
995 }
996 
997 /*
998  * Unmount system call.
999  */
1000 static int
1001 ext2_unmount(struct mount *mp, int mntflags)
1002 {
1003 	struct ext2mount *ump;
1004 	struct m_ext2fs *fs;
1005 	struct csum *sump;
1006 	int error, flags, i, ronly;
1007 
1008 	flags = 0;
1009 	if (mntflags & MNT_FORCE) {
1010 		if (mp->mnt_flag & MNT_ROOTFS)
1011 			return (EINVAL);
1012 		flags |= FORCECLOSE;
1013 	}
1014 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
1015 		return (error);
1016 	ump = VFSTOEXT2(mp);
1017 	fs = ump->um_e2fs;
1018 	ronly = fs->e2fs_ronly;
1019 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
1020 		if (fs->e2fs_wasvalid)
1021 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
1022 		ext2_sbupdate(ump, MNT_WAIT);
1023 	}
1024 
1025 	g_topology_lock();
1026 	g_vfs_close(ump->um_cp);
1027 	g_topology_unlock();
1028 	vrele(ump->um_devvp);
1029 	sump = fs->e2fs_clustersum;
1030 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
1031 		free(sump->cs_sum, M_EXT2MNT);
1032 	free(fs->e2fs_clustersum, M_EXT2MNT);
1033 	free(fs->e2fs_maxcluster, M_EXT2MNT);
1034 	free(fs->e2fs_gd, M_EXT2MNT);
1035 	free(fs->e2fs_contigdirs, M_EXT2MNT);
1036 	free(fs->e2fs, M_EXT2MNT);
1037 	free(fs, M_EXT2MNT);
1038 	free(ump, M_EXT2MNT);
1039 	mp->mnt_data = NULL;
1040 	MNT_ILOCK(mp);
1041 	mp->mnt_flag &= ~MNT_LOCAL;
1042 	MNT_IUNLOCK(mp);
1043 	return (error);
1044 }
1045 
1046 /*
1047  * Flush out all the files in a filesystem.
1048  */
1049 static int
1050 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
1051 {
1052 	int error;
1053 
1054 	error = vflush(mp, 0, flags, td);
1055 	return (error);
1056 }
1057 
1058 /*
1059  * Get filesystem statistics.
1060  */
1061 int
1062 ext2_statfs(struct mount *mp, struct statfs *sbp)
1063 {
1064 	struct ext2mount *ump;
1065 	struct m_ext2fs *fs;
1066 	uint32_t overhead, overhead_per_group, ngdb;
1067 	int i, ngroups;
1068 
1069 	ump = VFSTOEXT2(mp);
1070 	fs = ump->um_e2fs;
1071 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
1072 		panic("ext2_statfs");
1073 
1074 	/*
1075 	 * Compute the overhead (FS structures)
1076 	 */
1077 	overhead_per_group =
1078 	    1 /* block bitmap */ +
1079 	    1 /* inode bitmap */ +
1080 	    fs->e2fs_itpg;
1081 	overhead = fs->e2fs->e2fs_first_dblock +
1082 	    fs->e2fs_gcount * overhead_per_group;
1083 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
1084 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
1085 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
1086 			if (ext2_cg_has_sb(fs, i))
1087 				ngroups++;
1088 		}
1089 	} else {
1090 		ngroups = fs->e2fs_gcount;
1091 	}
1092 	ngdb = fs->e2fs_gdbcount;
1093 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
1094 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
1095 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
1096 	overhead += ngroups * (1 /* superblock */ + ngdb);
1097 
1098 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
1099 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
1100 	sbp->f_blocks = fs->e2fs_bcount - overhead;
1101 	sbp->f_bfree = fs->e2fs_fbcount;
1102 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
1103 	sbp->f_files = fs->e2fs->e2fs_icount;
1104 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
1105 	return (0);
1106 }
1107 
1108 /*
1109  * Go through the disk queues to initiate sandbagged IO;
1110  * go through the inodes to write those that have been modified;
1111  * initiate the writing of the super block if it has been modified.
1112  *
1113  * Note: we are always called with the filesystem marked `MPBUSY'.
1114  */
1115 static int
1116 ext2_sync(struct mount *mp, int waitfor)
1117 {
1118 	struct vnode *mvp, *vp;
1119 	struct thread *td;
1120 	struct inode *ip;
1121 	struct ext2mount *ump = VFSTOEXT2(mp);
1122 	struct m_ext2fs *fs;
1123 	int error, allerror = 0;
1124 
1125 	td = curthread;
1126 	fs = ump->um_e2fs;
1127 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
1128 		panic("ext2_sync: rofs mod fs=%s", fs->e2fs_fsmnt);
1129 	}
1130 
1131 	/*
1132 	 * Write back each (modified) inode.
1133 	 */
1134 loop:
1135 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1136 		if (vp->v_type == VNON) {
1137 			VI_UNLOCK(vp);
1138 			continue;
1139 		}
1140 		ip = VTOI(vp);
1141 		if ((ip->i_flag &
1142 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1143 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1144 		    waitfor == MNT_LAZY)) {
1145 			VI_UNLOCK(vp);
1146 			continue;
1147 		}
1148 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
1149 		if (error) {
1150 			if (error == ENOENT) {
1151 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1152 				goto loop;
1153 			}
1154 			continue;
1155 		}
1156 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
1157 			allerror = error;
1158 		VOP_UNLOCK(vp, 0);
1159 		vrele(vp);
1160 	}
1161 
1162 	/*
1163 	 * Force stale filesystem control information to be flushed.
1164 	 */
1165 	if (waitfor != MNT_LAZY) {
1166 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1167 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
1168 			allerror = error;
1169 		VOP_UNLOCK(ump->um_devvp, 0);
1170 	}
1171 
1172 	/*
1173 	 * Write back modified superblock.
1174 	 */
1175 	if (fs->e2fs_fmod != 0) {
1176 		fs->e2fs_fmod = 0;
1177 		fs->e2fs->e2fs_wtime = time_second;
1178 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
1179 			allerror = error;
1180 	}
1181 	return (allerror);
1182 }
1183 
1184 /*
1185  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
1186  * in from disk.  If it is in core, wait for the lock bit to clear, then
1187  * return the inode locked.  Detection and handling of mount points must be
1188  * done by the calling routine.
1189  */
1190 static int
1191 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
1192 {
1193 	struct m_ext2fs *fs;
1194 	struct inode *ip;
1195 	struct ext2mount *ump;
1196 	struct buf *bp;
1197 	struct vnode *vp;
1198 	struct thread *td;
1199 	unsigned int i, used_blocks;
1200 	int error;
1201 
1202 	td = curthread;
1203 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
1204 	if (error || *vpp != NULL)
1205 		return (error);
1206 
1207 	ump = VFSTOEXT2(mp);
1208 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
1209 
1210 	/* Allocate a new vnode/inode. */
1211 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
1212 		*vpp = NULL;
1213 		free(ip, M_EXT2NODE);
1214 		return (error);
1215 	}
1216 	vp->v_data = ip;
1217 	ip->i_vnode = vp;
1218 	ip->i_e2fs = fs = ump->um_e2fs;
1219 	ip->i_ump = ump;
1220 	ip->i_number = ino;
1221 
1222 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1223 	error = insmntque(vp, mp);
1224 	if (error != 0) {
1225 		free(ip, M_EXT2NODE);
1226 		*vpp = NULL;
1227 		return (error);
1228 	}
1229 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
1230 	if (error || *vpp != NULL)
1231 		return (error);
1232 
1233 	/* Read in the disk contents for the inode, copy into the inode. */
1234 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1235 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
1236 		/*
1237 		 * The inode does not contain anything useful, so it would
1238 		 * be misleading to leave it on its hash chain. With mode
1239 		 * still zero, it will be unlinked and returned to the free
1240 		 * list by vput().
1241 		 */
1242 		brelse(bp);
1243 		vput(vp);
1244 		*vpp = NULL;
1245 		return (error);
1246 	}
1247 	/* convert ext2 inode to dinode */
1248 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1249 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1250 	if (error) {
1251 		brelse(bp);
1252 		vput(vp);
1253 		*vpp = NULL;
1254 		return (error);
1255 	}
1256 	ip->i_block_group = ino_to_cg(fs, ino);
1257 	ip->i_next_alloc_block = 0;
1258 	ip->i_next_alloc_goal = 0;
1259 
1260 	/*
1261 	 * Now we want to make sure that block pointers for unused
1262 	 * blocks are zeroed out - ext2_balloc depends on this
1263 	 * although for regular files and directories only
1264 	 *
1265 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1266 	 * out because we could corrupt the extent tree.
1267 	 */
1268 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1269 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1270 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1271 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1272 			ip->i_db[i] = 0;
1273 	}
1274 #ifdef EXT2FS_PRINT_EXTENTS
1275 	ext2_print_inode(ip);
1276 	ext4_ext_print_extent_tree_status(ip);
1277 #endif
1278 	bqrelse(bp);
1279 
1280 	/*
1281 	 * Initialize the vnode from the inode, check for aliases.
1282 	 * Note that the underlying vnode may have changed.
1283 	 */
1284 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1285 		vput(vp);
1286 		*vpp = NULL;
1287 		return (error);
1288 	}
1289 
1290 	/*
1291 	 * Finish inode initialization.
1292 	 */
1293 
1294 	*vpp = vp;
1295 	return (0);
1296 }
1297 
1298 /*
1299  * File handle to vnode
1300  *
1301  * Have to be really careful about stale file handles:
1302  * - check that the inode number is valid
1303  * - call ext2_vget() to get the locked inode
1304  * - check for an unallocated inode (i_mode == 0)
1305  * - check that the given client host has export rights and return
1306  *   those rights via. exflagsp and credanonp
1307  */
1308 static int
1309 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1310 {
1311 	struct inode *ip;
1312 	struct ufid *ufhp;
1313 	struct vnode *nvp;
1314 	struct m_ext2fs *fs;
1315 	int error;
1316 
1317 	ufhp = (struct ufid *)fhp;
1318 	fs = VFSTOEXT2(mp)->um_e2fs;
1319 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1320 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1321 		return (ESTALE);
1322 
1323 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1324 	if (error) {
1325 		*vpp = NULLVP;
1326 		return (error);
1327 	}
1328 	ip = VTOI(nvp);
1329 	if (ip->i_mode == 0 ||
1330 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1331 		vput(nvp);
1332 		*vpp = NULLVP;
1333 		return (ESTALE);
1334 	}
1335 	*vpp = nvp;
1336 	vnode_create_vobject(*vpp, 0, curthread);
1337 	return (0);
1338 }
1339 
1340 /*
1341  * Write a superblock and associated information back to disk.
1342  */
1343 static int
1344 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1345 {
1346 	struct m_ext2fs *fs = mp->um_e2fs;
1347 	struct ext2fs *es = fs->e2fs;
1348 	struct buf *bp;
1349 	int error = 0;
1350 
1351 	es->e2fs_bcount = fs->e2fs_bcount & 0xffffffff;
1352 	es->e2fs_rbcount = fs->e2fs_rbcount & 0xffffffff;
1353 	es->e2fs_fbcount = fs->e2fs_fbcount & 0xffffffff;
1354 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1355 		es->e4fs_bcount_hi = fs->e2fs_bcount >> 32;
1356 		es->e4fs_rbcount_hi = fs->e2fs_rbcount >> 32;
1357 		es->e4fs_fbcount_hi = fs->e2fs_fbcount >> 32;
1358 	}
1359 
1360 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1361 		ext2_sb_csum_set(fs);
1362 
1363 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1364 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1365 	if (waitfor == MNT_WAIT)
1366 		error = bwrite(bp);
1367 	else
1368 		bawrite(bp);
1369 
1370 	/*
1371 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1372 	 * are not busy at this point and are (hopefully) written by the
1373 	 * usual sync mechanism. No need to write them here.
1374 	 */
1375 	return (error);
1376 }
1377 int
1378 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1379 {
1380 	struct m_ext2fs *fs = mp->um_e2fs;
1381 	struct buf *bp;
1382 	int i, j, g_count = 0, error = 0, allerror = 0;
1383 
1384 	allerror = ext2_sbupdate(mp, waitfor);
1385 
1386 	/* Update gd csums */
1387 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1388 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1389 		ext2_gd_csum_set(fs);
1390 
1391 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1392 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1393 		    ext2_cg_location(fs, i)),
1394 		    fs->e2fs_bsize, 0, 0, 0);
1395 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1396 			memcpy(bp->b_data, &fs->e2fs_gd[
1397 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1398 			    fs->e2fs_bsize);
1399 		} else {
1400 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1401 			    g_count < fs->e2fs_gcount; j++, g_count++)
1402 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1403 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1404 		}
1405 		if (waitfor == MNT_WAIT)
1406 			error = bwrite(bp);
1407 		else
1408 			bawrite(bp);
1409 	}
1410 
1411 	if (!allerror && error)
1412 		allerror = error;
1413 	return (allerror);
1414 }
1415 
1416 /*
1417  * Return the root of a filesystem.
1418  */
1419 static int
1420 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1421 {
1422 	struct vnode *nvp;
1423 	int error;
1424 
1425 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1426 	if (error)
1427 		return (error);
1428 	*vpp = nvp;
1429 	return (0);
1430 }
1431