xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision 86aa9539fef591a363b06a0ebd3aa7a07f4c1579)
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 			brelse(bp);
655 			return (error);
656 		}
657 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
658 			memcpy(&fs->e2fs_gd[
659 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
660 			    bp->b_data, fs->e2fs_bsize);
661 		} else {
662 			for (j = 0; j < e2fs_descpb &&
663 			    g_count < fs->e2fs_gcount; j++, g_count++)
664 				memcpy(&fs->e2fs_gd[g_count],
665 				    bp->b_data + j * E2FS_REV0_GD_SIZE,
666 				    E2FS_REV0_GD_SIZE);
667 		}
668 		brelse(bp);
669 		bp = NULL;
670 	}
671 
672 	/* Validate cgs consistency */
673 	error = ext2_cg_validate(fs);
674 	if (error)
675 		return (error);
676 
677 	/* Verfy cgs csum */
678 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
679 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
680 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
681 		if (error)
682 			return (error);
683 	}
684 	/* Initialization for the ext2 Orlov allocator variant. */
685 	fs->e2fs_total_dir = 0;
686 	for (i = 0; i < fs->e2fs_gcount; i++)
687 		fs->e2fs_total_dir += e2fs_gd_get_ndirs(&fs->e2fs_gd[i]);
688 
689 	if (es->e2fs_rev == E2FS_REV0 ||
690 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
691 		fs->e2fs_maxfilesize = 0x7fffffff;
692 	else {
693 		fs->e2fs_maxfilesize = 0xffffffffffff;
694 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
695 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
696 	}
697 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
698 		fs->e2fs_uhash = 3;
699 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
700 #ifdef __CHAR_UNSIGNED__
701 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
702 		fs->e2fs_uhash = 3;
703 #else
704 		es->e4fs_flags |= E2FS_SIGNED_HASH;
705 #endif
706 	}
707 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
708 		error = ext2_sb_csum_verify(fs);
709 
710 	return (error);
711 }
712 
713 /*
714  * Reload all incore data for a filesystem (used after running fsck on
715  * the root filesystem and finding things to fix). The filesystem must
716  * be mounted read-only.
717  *
718  * Things to do to update the mount:
719  *	1) invalidate all cached meta-data.
720  *	2) re-read superblock from disk.
721  *	3) invalidate all cluster summary information.
722  *	4) invalidate all inactive vnodes.
723  *	5) invalidate all cached file data.
724  *	6) re-read inode data for all active vnodes.
725  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
726  */
727 static int
728 ext2_reload(struct mount *mp, struct thread *td)
729 {
730 	struct vnode *vp, *mvp, *devvp;
731 	struct inode *ip;
732 	struct buf *bp;
733 	struct ext2fs *es;
734 	struct m_ext2fs *fs;
735 	struct csum *sump;
736 	int error, i;
737 	int32_t *lp;
738 
739 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
740 		return (EINVAL);
741 	/*
742 	 * Step 1: invalidate all cached meta-data.
743 	 */
744 	devvp = VFSTOEXT2(mp)->um_devvp;
745 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
746 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
747 		panic("ext2_reload: dirty1");
748 	VOP_UNLOCK(devvp, 0);
749 
750 	/*
751 	 * Step 2: re-read superblock from disk.
752 	 * constants have been adjusted for ext2
753 	 */
754 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
755 		return (error);
756 	es = (struct ext2fs *)bp->b_data;
757 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
758 		brelse(bp);
759 		return (EIO);		/* XXX needs translation */
760 	}
761 	fs = VFSTOEXT2(mp)->um_e2fs;
762 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
763 
764 	if ((error = ext2_compute_sb_data(devvp, es, fs)) != 0) {
765 		brelse(bp);
766 		return (error);
767 	}
768 #ifdef UNKLAR
769 	if (fs->fs_sbsize < SBSIZE)
770 		bp->b_flags |= B_INVAL;
771 #endif
772 	brelse(bp);
773 
774 	/*
775 	 * Step 3: invalidate all cluster summary information.
776 	 */
777 	if (fs->e2fs_contigsumsize > 0) {
778 		lp = fs->e2fs_maxcluster;
779 		sump = fs->e2fs_clustersum;
780 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
781 			*lp++ = fs->e2fs_contigsumsize;
782 			sump->cs_init = 0;
783 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
784 		}
785 	}
786 
787 loop:
788 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
789 		/*
790 		 * Step 4: invalidate all cached file data.
791 		 */
792 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
793 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
794 			goto loop;
795 		}
796 		if (vinvalbuf(vp, 0, 0, 0))
797 			panic("ext2_reload: dirty2");
798 
799 		/*
800 		 * Step 5: re-read inode data for all active vnodes.
801 		 */
802 		ip = VTOI(vp);
803 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
804 		    (int)fs->e2fs_bsize, NOCRED, &bp);
805 		if (error) {
806 			VOP_UNLOCK(vp, 0);
807 			vrele(vp);
808 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
809 			return (error);
810 		}
811 
812 		error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
813 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
814 
815 		brelse(bp);
816 		VOP_UNLOCK(vp, 0);
817 		vrele(vp);
818 
819 		if (error) {
820 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
821 			return (error);
822 		}
823 	}
824 	return (0);
825 }
826 
827 /*
828  * Common code for mount and mountroot.
829  */
830 static int
831 ext2_mountfs(struct vnode *devvp, struct mount *mp)
832 {
833 	struct ext2mount *ump;
834 	struct buf *bp;
835 	struct m_ext2fs *fs;
836 	struct ext2fs *es;
837 	struct cdev *dev = devvp->v_rdev;
838 	struct g_consumer *cp;
839 	struct bufobj *bo;
840 	struct csum *sump;
841 	int error;
842 	int ronly;
843 	int i;
844 	u_long size;
845 	int32_t *lp;
846 	int32_t e2fs_maxcontig;
847 
848 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
849 	/* XXX: use VOP_ACESS to check FS perms */
850 	g_topology_lock();
851 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
852 	g_topology_unlock();
853 	VOP_UNLOCK(devvp, 0);
854 	if (error)
855 		return (error);
856 
857 	/* XXX: should we check for some sectorsize or 512 instead? */
858 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
859 	    (SBSIZE < cp->provider->sectorsize)) {
860 		g_topology_lock();
861 		g_vfs_close(cp);
862 		g_topology_unlock();
863 		return (EINVAL);
864 	}
865 
866 	bo = &devvp->v_bufobj;
867 	bo->bo_private = cp;
868 	bo->bo_ops = g_vfs_bufops;
869 	if (devvp->v_rdev->si_iosize_max != 0)
870 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
871 	if (mp->mnt_iosize_max > MAXPHYS)
872 		mp->mnt_iosize_max = MAXPHYS;
873 
874 	bp = NULL;
875 	ump = NULL;
876 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
877 		goto out;
878 	es = (struct ext2fs *)bp->b_data;
879 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
880 		error = EINVAL;		/* XXX needs translation */
881 		goto out;
882 	}
883 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
884 	    (es->e2fs_state & E2FS_ERRORS)) {
885 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
886 			printf(
887 "WARNING: Filesystem was not properly dismounted\n");
888 		} else {
889 			printf(
890 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
891 			error = EPERM;
892 			goto out;
893 		}
894 	}
895 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
896 
897 	/*
898 	 * I don't know whether this is the right strategy. Note that
899 	 * we dynamically allocate both an m_ext2fs and an ext2fs
900 	 * while Linux keeps the super block in a locked buffer.
901 	 */
902 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
903 	    M_EXT2MNT, M_WAITOK | M_ZERO);
904 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
905 	    M_EXT2MNT, M_WAITOK);
906 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
907 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
908 	if ((error = ext2_compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
909 		goto out;
910 
911 	/*
912 	 * Calculate the maximum contiguous blocks and size of cluster summary
913 	 * array.  In FFS this is done by newfs; however, the superblock
914 	 * in ext2fs doesn't have these variables, so we can calculate
915 	 * them here.
916 	 */
917 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
918 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
919 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
920 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
921 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
922 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
923 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
924 		lp = ump->um_e2fs->e2fs_maxcluster;
925 		sump = ump->um_e2fs->e2fs_clustersum;
926 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
927 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
928 			sump->cs_init = 0;
929 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
930 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
931 		}
932 	}
933 
934 	brelse(bp);
935 	bp = NULL;
936 	fs = ump->um_e2fs;
937 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
938 
939 	/*
940 	 * If the fs is not mounted read-only, make sure the super block is
941 	 * always written back on a sync().
942 	 */
943 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
944 	if (ronly == 0) {
945 		fs->e2fs_fmod = 1;	/* mark it modified */
946 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
947 	}
948 	mp->mnt_data = ump;
949 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
950 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
951 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
952 	MNT_ILOCK(mp);
953 	mp->mnt_flag |= MNT_LOCAL;
954 	MNT_IUNLOCK(mp);
955 	ump->um_mountp = mp;
956 	ump->um_dev = dev;
957 	ump->um_devvp = devvp;
958 	ump->um_bo = &devvp->v_bufobj;
959 	ump->um_cp = cp;
960 
961 	/*
962 	 * Setting those two parameters allowed us to use
963 	 * ufs_bmap w/o changse!
964 	 */
965 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
966 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
967 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
968 	if (ronly == 0)
969 		ext2_sbupdate(ump, MNT_WAIT);
970 	/*
971 	 * Initialize filesystem stat information in mount struct.
972 	 */
973 	MNT_ILOCK(mp);
974 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
975 	    MNTK_USES_BCACHE;
976 	MNT_IUNLOCK(mp);
977 	return (0);
978 out:
979 	if (bp)
980 		brelse(bp);
981 	if (cp != NULL) {
982 		g_topology_lock();
983 		g_vfs_close(cp);
984 		g_topology_unlock();
985 	}
986 	if (ump) {
987 		mtx_destroy(EXT2_MTX(ump));
988 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
989 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
990 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
991 		free(ump->um_e2fs, M_EXT2MNT);
992 		free(ump, M_EXT2MNT);
993 		mp->mnt_data = NULL;
994 	}
995 	return (error);
996 }
997 
998 /*
999  * Unmount system call.
1000  */
1001 static int
1002 ext2_unmount(struct mount *mp, int mntflags)
1003 {
1004 	struct ext2mount *ump;
1005 	struct m_ext2fs *fs;
1006 	struct csum *sump;
1007 	int error, flags, i, ronly;
1008 
1009 	flags = 0;
1010 	if (mntflags & MNT_FORCE) {
1011 		if (mp->mnt_flag & MNT_ROOTFS)
1012 			return (EINVAL);
1013 		flags |= FORCECLOSE;
1014 	}
1015 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
1016 		return (error);
1017 	ump = VFSTOEXT2(mp);
1018 	fs = ump->um_e2fs;
1019 	ronly = fs->e2fs_ronly;
1020 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
1021 		if (fs->e2fs_wasvalid)
1022 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
1023 		ext2_sbupdate(ump, MNT_WAIT);
1024 	}
1025 
1026 	g_topology_lock();
1027 	g_vfs_close(ump->um_cp);
1028 	g_topology_unlock();
1029 	vrele(ump->um_devvp);
1030 	sump = fs->e2fs_clustersum;
1031 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
1032 		free(sump->cs_sum, M_EXT2MNT);
1033 	free(fs->e2fs_clustersum, M_EXT2MNT);
1034 	free(fs->e2fs_maxcluster, M_EXT2MNT);
1035 	free(fs->e2fs_gd, M_EXT2MNT);
1036 	free(fs->e2fs_contigdirs, M_EXT2MNT);
1037 	free(fs->e2fs, M_EXT2MNT);
1038 	free(fs, M_EXT2MNT);
1039 	free(ump, M_EXT2MNT);
1040 	mp->mnt_data = NULL;
1041 	MNT_ILOCK(mp);
1042 	mp->mnt_flag &= ~MNT_LOCAL;
1043 	MNT_IUNLOCK(mp);
1044 	return (error);
1045 }
1046 
1047 /*
1048  * Flush out all the files in a filesystem.
1049  */
1050 static int
1051 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
1052 {
1053 	int error;
1054 
1055 	error = vflush(mp, 0, flags, td);
1056 	return (error);
1057 }
1058 
1059 /*
1060  * Get filesystem statistics.
1061  */
1062 int
1063 ext2_statfs(struct mount *mp, struct statfs *sbp)
1064 {
1065 	struct ext2mount *ump;
1066 	struct m_ext2fs *fs;
1067 	uint32_t overhead, overhead_per_group, ngdb;
1068 	int i, ngroups;
1069 
1070 	ump = VFSTOEXT2(mp);
1071 	fs = ump->um_e2fs;
1072 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
1073 		panic("ext2_statfs");
1074 
1075 	/*
1076 	 * Compute the overhead (FS structures)
1077 	 */
1078 	overhead_per_group =
1079 	    1 /* block bitmap */ +
1080 	    1 /* inode bitmap */ +
1081 	    fs->e2fs_itpg;
1082 	overhead = fs->e2fs->e2fs_first_dblock +
1083 	    fs->e2fs_gcount * overhead_per_group;
1084 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
1085 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
1086 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
1087 			if (ext2_cg_has_sb(fs, i))
1088 				ngroups++;
1089 		}
1090 	} else {
1091 		ngroups = fs->e2fs_gcount;
1092 	}
1093 	ngdb = fs->e2fs_gdbcount;
1094 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
1095 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
1096 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
1097 	overhead += ngroups * (1 /* superblock */ + ngdb);
1098 
1099 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
1100 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
1101 	sbp->f_blocks = fs->e2fs_bcount - overhead;
1102 	sbp->f_bfree = fs->e2fs_fbcount;
1103 	sbp->f_bavail = sbp->f_bfree - fs->e2fs_rbcount;
1104 	sbp->f_files = fs->e2fs->e2fs_icount;
1105 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
1106 	return (0);
1107 }
1108 
1109 /*
1110  * Go through the disk queues to initiate sandbagged IO;
1111  * go through the inodes to write those that have been modified;
1112  * initiate the writing of the super block if it has been modified.
1113  *
1114  * Note: we are always called with the filesystem marked `MPBUSY'.
1115  */
1116 static int
1117 ext2_sync(struct mount *mp, int waitfor)
1118 {
1119 	struct vnode *mvp, *vp;
1120 	struct thread *td;
1121 	struct inode *ip;
1122 	struct ext2mount *ump = VFSTOEXT2(mp);
1123 	struct m_ext2fs *fs;
1124 	int error, allerror = 0;
1125 
1126 	td = curthread;
1127 	fs = ump->um_e2fs;
1128 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
1129 		panic("ext2_sync: rofs mod fs=%s", fs->e2fs_fsmnt);
1130 	}
1131 
1132 	/*
1133 	 * Write back each (modified) inode.
1134 	 */
1135 loop:
1136 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1137 		if (vp->v_type == VNON) {
1138 			VI_UNLOCK(vp);
1139 			continue;
1140 		}
1141 		ip = VTOI(vp);
1142 		if ((ip->i_flag &
1143 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1144 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1145 		    waitfor == MNT_LAZY)) {
1146 			VI_UNLOCK(vp);
1147 			continue;
1148 		}
1149 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
1150 		if (error) {
1151 			if (error == ENOENT) {
1152 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1153 				goto loop;
1154 			}
1155 			continue;
1156 		}
1157 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
1158 			allerror = error;
1159 		VOP_UNLOCK(vp, 0);
1160 		vrele(vp);
1161 	}
1162 
1163 	/*
1164 	 * Force stale filesystem control information to be flushed.
1165 	 */
1166 	if (waitfor != MNT_LAZY) {
1167 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1168 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
1169 			allerror = error;
1170 		VOP_UNLOCK(ump->um_devvp, 0);
1171 	}
1172 
1173 	/*
1174 	 * Write back modified superblock.
1175 	 */
1176 	if (fs->e2fs_fmod != 0) {
1177 		fs->e2fs_fmod = 0;
1178 		fs->e2fs->e2fs_wtime = time_second;
1179 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
1180 			allerror = error;
1181 	}
1182 	return (allerror);
1183 }
1184 
1185 /*
1186  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
1187  * in from disk.  If it is in core, wait for the lock bit to clear, then
1188  * return the inode locked.  Detection and handling of mount points must be
1189  * done by the calling routine.
1190  */
1191 static int
1192 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
1193 {
1194 	struct m_ext2fs *fs;
1195 	struct inode *ip;
1196 	struct ext2mount *ump;
1197 	struct buf *bp;
1198 	struct vnode *vp;
1199 	struct thread *td;
1200 	unsigned int i, used_blocks;
1201 	int error;
1202 
1203 	td = curthread;
1204 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
1205 	if (error || *vpp != NULL)
1206 		return (error);
1207 
1208 	ump = VFSTOEXT2(mp);
1209 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
1210 
1211 	/* Allocate a new vnode/inode. */
1212 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
1213 		*vpp = NULL;
1214 		free(ip, M_EXT2NODE);
1215 		return (error);
1216 	}
1217 	vp->v_data = ip;
1218 	ip->i_vnode = vp;
1219 	ip->i_e2fs = fs = ump->um_e2fs;
1220 	ip->i_ump = ump;
1221 	ip->i_number = ino;
1222 
1223 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1224 	error = insmntque(vp, mp);
1225 	if (error != 0) {
1226 		free(ip, M_EXT2NODE);
1227 		*vpp = NULL;
1228 		return (error);
1229 	}
1230 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
1231 	if (error || *vpp != NULL)
1232 		return (error);
1233 
1234 	/* Read in the disk contents for the inode, copy into the inode. */
1235 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1236 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
1237 		/*
1238 		 * The inode does not contain anything useful, so it would
1239 		 * be misleading to leave it on its hash chain. With mode
1240 		 * still zero, it will be unlinked and returned to the free
1241 		 * list by vput().
1242 		 */
1243 		brelse(bp);
1244 		vput(vp);
1245 		*vpp = NULL;
1246 		return (error);
1247 	}
1248 	/* convert ext2 inode to dinode */
1249 	error = ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
1250 	    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ino)), ip);
1251 	if (error) {
1252 		brelse(bp);
1253 		vput(vp);
1254 		*vpp = NULL;
1255 		return (error);
1256 	}
1257 	ip->i_block_group = ino_to_cg(fs, ino);
1258 	ip->i_next_alloc_block = 0;
1259 	ip->i_next_alloc_goal = 0;
1260 
1261 	/*
1262 	 * Now we want to make sure that block pointers for unused
1263 	 * blocks are zeroed out - ext2_balloc depends on this
1264 	 * although for regular files and directories only
1265 	 *
1266 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
1267 	 * out because we could corrupt the extent tree.
1268 	 */
1269 	if (!(ip->i_flag & IN_E4EXTENTS) &&
1270 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
1271 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
1272 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
1273 			ip->i_db[i] = 0;
1274 	}
1275 #ifdef EXT2FS_PRINT_EXTENTS
1276 	ext2_print_inode(ip);
1277 	ext4_ext_print_extent_tree_status(ip);
1278 #endif
1279 	bqrelse(bp);
1280 
1281 	/*
1282 	 * Initialize the vnode from the inode, check for aliases.
1283 	 * Note that the underlying vnode may have changed.
1284 	 */
1285 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1286 		vput(vp);
1287 		*vpp = NULL;
1288 		return (error);
1289 	}
1290 
1291 	/*
1292 	 * Finish inode initialization.
1293 	 */
1294 
1295 	*vpp = vp;
1296 	return (0);
1297 }
1298 
1299 /*
1300  * File handle to vnode
1301  *
1302  * Have to be really careful about stale file handles:
1303  * - check that the inode number is valid
1304  * - call ext2_vget() to get the locked inode
1305  * - check for an unallocated inode (i_mode == 0)
1306  * - check that the given client host has export rights and return
1307  *   those rights via. exflagsp and credanonp
1308  */
1309 static int
1310 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1311 {
1312 	struct inode *ip;
1313 	struct ufid *ufhp;
1314 	struct vnode *nvp;
1315 	struct m_ext2fs *fs;
1316 	int error;
1317 
1318 	ufhp = (struct ufid *)fhp;
1319 	fs = VFSTOEXT2(mp)->um_e2fs;
1320 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1321 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1322 		return (ESTALE);
1323 
1324 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1325 	if (error) {
1326 		*vpp = NULLVP;
1327 		return (error);
1328 	}
1329 	ip = VTOI(nvp);
1330 	if (ip->i_mode == 0 ||
1331 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1332 		vput(nvp);
1333 		*vpp = NULLVP;
1334 		return (ESTALE);
1335 	}
1336 	*vpp = nvp;
1337 	vnode_create_vobject(*vpp, 0, curthread);
1338 	return (0);
1339 }
1340 
1341 /*
1342  * Write a superblock and associated information back to disk.
1343  */
1344 static int
1345 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1346 {
1347 	struct m_ext2fs *fs = mp->um_e2fs;
1348 	struct ext2fs *es = fs->e2fs;
1349 	struct buf *bp;
1350 	int error = 0;
1351 
1352 	es->e2fs_bcount = fs->e2fs_bcount & 0xffffffff;
1353 	es->e2fs_rbcount = fs->e2fs_rbcount & 0xffffffff;
1354 	es->e2fs_fbcount = fs->e2fs_fbcount & 0xffffffff;
1355 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1356 		es->e4fs_bcount_hi = fs->e2fs_bcount >> 32;
1357 		es->e4fs_rbcount_hi = fs->e2fs_rbcount >> 32;
1358 		es->e4fs_fbcount_hi = fs->e2fs_fbcount >> 32;
1359 	}
1360 
1361 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1362 		ext2_sb_csum_set(fs);
1363 
1364 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1365 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1366 	if (waitfor == MNT_WAIT)
1367 		error = bwrite(bp);
1368 	else
1369 		bawrite(bp);
1370 
1371 	/*
1372 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1373 	 * are not busy at this point and are (hopefully) written by the
1374 	 * usual sync mechanism. No need to write them here.
1375 	 */
1376 	return (error);
1377 }
1378 int
1379 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1380 {
1381 	struct m_ext2fs *fs = mp->um_e2fs;
1382 	struct buf *bp;
1383 	int i, j, g_count = 0, error = 0, allerror = 0;
1384 
1385 	allerror = ext2_sbupdate(mp, waitfor);
1386 
1387 	/* Update gd csums */
1388 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1389 	    EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM))
1390 		ext2_gd_csum_set(fs);
1391 
1392 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1393 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1394 		    ext2_cg_location(fs, i)),
1395 		    fs->e2fs_bsize, 0, 0, 0);
1396 		if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT)) {
1397 			memcpy(bp->b_data, &fs->e2fs_gd[
1398 			    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1399 			    fs->e2fs_bsize);
1400 		} else {
1401 			for (j = 0; j < fs->e2fs_bsize / E2FS_REV0_GD_SIZE &&
1402 			    g_count < fs->e2fs_gcount; j++, g_count++)
1403 				memcpy(bp->b_data + j * E2FS_REV0_GD_SIZE,
1404 				    &fs->e2fs_gd[g_count], E2FS_REV0_GD_SIZE);
1405 		}
1406 		if (waitfor == MNT_WAIT)
1407 			error = bwrite(bp);
1408 		else
1409 			bawrite(bp);
1410 	}
1411 
1412 	if (!allerror && error)
1413 		allerror = error;
1414 	return (allerror);
1415 }
1416 
1417 /*
1418  * Return the root of a filesystem.
1419  */
1420 static int
1421 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1422 {
1423 	struct vnode *nvp;
1424 	int error;
1425 
1426 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1427 	if (error)
1428 		return (error);
1429 	*vpp = nvp;
1430 	return (0);
1431 }
1432