xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision 792bbaba989533a1fc93823df1720c8c4aaf0442)
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  * Copyright (c) 1989, 1991, 1993, 1994
9  *	The Regents of the University of California.  All rights reserved.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)ffs_vfsops.c	8.8 (Berkeley) 4/18/94
36  * $FreeBSD$
37  */
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/namei.h>
42 #include <sys/priv.h>
43 #include <sys/proc.h>
44 #include <sys/kernel.h>
45 #include <sys/vnode.h>
46 #include <sys/mount.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/conf.h>
50 #include <sys/endian.h>
51 #include <sys/fcntl.h>
52 #include <sys/malloc.h>
53 #include <sys/stat.h>
54 #include <sys/mutex.h>
55 
56 #include <geom/geom.h>
57 #include <geom/geom_vfs.h>
58 
59 #include <fs/ext2fs/ext2_mount.h>
60 #include <fs/ext2fs/inode.h>
61 
62 #include <fs/ext2fs/fs.h>
63 #include <fs/ext2fs/ext2fs.h>
64 #include <fs/ext2fs/ext2_dinode.h>
65 #include <fs/ext2fs/ext2_extern.h>
66 
67 static int	ext2_flushfiles(struct mount *mp, int flags, struct thread *td);
68 static int	ext2_mountfs(struct vnode *, struct mount *);
69 static int	ext2_reload(struct mount *mp, struct thread *td);
70 static int	ext2_sbupdate(struct ext2mount *, int);
71 static int	ext2_cgupdate(struct ext2mount *, int);
72 static vfs_unmount_t		ext2_unmount;
73 static vfs_root_t		ext2_root;
74 static vfs_statfs_t		ext2_statfs;
75 static vfs_sync_t		ext2_sync;
76 static vfs_vget_t		ext2_vget;
77 static vfs_fhtovp_t		ext2_fhtovp;
78 static vfs_mount_t		ext2_mount;
79 
80 MALLOC_DEFINE(M_EXT2NODE, "ext2_node", "EXT2 vnode private part");
81 static MALLOC_DEFINE(M_EXT2MNT, "ext2_mount", "EXT2 mount structure");
82 
83 static struct vfsops ext2fs_vfsops = {
84 	.vfs_fhtovp =		ext2_fhtovp,
85 	.vfs_mount =		ext2_mount,
86 	.vfs_root =		ext2_root,	/* root inode via vget */
87 	.vfs_statfs =		ext2_statfs,
88 	.vfs_sync =		ext2_sync,
89 	.vfs_unmount =		ext2_unmount,
90 	.vfs_vget =		ext2_vget,
91 };
92 
93 VFS_SET(ext2fs_vfsops, ext2fs, 0);
94 
95 static int	ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev,
96 		    int ronly);
97 static int	compute_sb_data(struct vnode * devvp,
98 		    struct ext2fs * es, struct m_ext2fs * fs);
99 
100 static const char *ext2_opts[] = { "acls", "async", "noatime", "noclusterr",
101     "noclusterw", "noexec", "export", "force", "from", "multilabel",
102     "suiddir", "nosymfollow", "sync", "union", NULL };
103 
104 /*
105  * VFS Operations.
106  *
107  * mount system call
108  */
109 static int
110 ext2_mount(struct mount *mp)
111 {
112 	struct vfsoptlist *opts;
113 	struct vnode *devvp;
114 	struct thread *td;
115 	struct ext2mount *ump = NULL;
116 	struct m_ext2fs *fs;
117 	struct nameidata nd, *ndp = &nd;
118 	accmode_t accmode;
119 	char *path, *fspec;
120 	int error, flags, len;
121 
122 	td = curthread;
123 	opts = mp->mnt_optnew;
124 
125 	if (vfs_filteropt(opts, ext2_opts))
126 		return (EINVAL);
127 
128 	vfs_getopt(opts, "fspath", (void **)&path, NULL);
129 	/* Double-check the length of path.. */
130 	if (strlen(path) >= MAXMNTLEN)
131 		return (ENAMETOOLONG);
132 
133 	fspec = NULL;
134 	error = vfs_getopt(opts, "from", (void **)&fspec, &len);
135 	if (!error && fspec[len - 1] != '\0')
136 		return (EINVAL);
137 
138 	/*
139 	 * If updating, check whether changing from read-only to
140 	 * read/write; if there is no device name, that's all we do.
141 	 */
142 	if (mp->mnt_flag & MNT_UPDATE) {
143 		ump = VFSTOEXT2(mp);
144 		fs = ump->um_e2fs;
145 		error = 0;
146 		if (fs->e2fs_ronly == 0 &&
147 		    vfs_flagopt(opts, "ro", NULL, 0)) {
148 			error = VFS_SYNC(mp, MNT_WAIT);
149 			if (error)
150 				return (error);
151 			flags = WRITECLOSE;
152 			if (mp->mnt_flag & MNT_FORCE)
153 				flags |= FORCECLOSE;
154 			error = ext2_flushfiles(mp, flags, td);
155 			if (error == 0 && fs->e2fs_wasvalid && ext2_cgupdate(ump, MNT_WAIT) == 0) {
156 				fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
157 				ext2_sbupdate(ump, MNT_WAIT);
158 			}
159 			fs->e2fs_ronly = 1;
160 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
161 			g_topology_lock();
162 			g_access(ump->um_cp, 0, -1, 0);
163 			g_topology_unlock();
164 		}
165 		if (!error && (mp->mnt_flag & MNT_RELOAD))
166 			error = ext2_reload(mp, td);
167 		if (error)
168 			return (error);
169 		devvp = ump->um_devvp;
170 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
171 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
172 				return (EPERM);
173 
174 			/*
175 			 * If upgrade to read-write by non-root, then verify
176 			 * that user has necessary permissions on the device.
177 			 */
178 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
179 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
180 			    td->td_ucred, td);
181 			if (error)
182 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
183 			if (error) {
184 				VOP_UNLOCK(devvp, 0);
185 				return (error);
186 			}
187 			VOP_UNLOCK(devvp, 0);
188 			g_topology_lock();
189 			error = g_access(ump->um_cp, 0, 1, 0);
190 			g_topology_unlock();
191 			if (error)
192 				return (error);
193 
194 			if ((fs->e2fs->e2fs_state & E2FS_ISCLEAN) == 0 ||
195 			    (fs->e2fs->e2fs_state & E2FS_ERRORS)) {
196 				if (mp->mnt_flag & MNT_FORCE) {
197 					printf(
198 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
199 				} else {
200 					printf(
201 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
202 					    fs->e2fs_fsmnt);
203 					return (EPERM);
204 				}
205 			}
206 			fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;
207 			(void)ext2_cgupdate(ump, MNT_WAIT);
208 			fs->e2fs_ronly = 0;
209 			MNT_ILOCK(mp);
210 			mp->mnt_flag &= ~MNT_RDONLY;
211 			MNT_IUNLOCK(mp);
212 		}
213 		if (vfs_flagopt(opts, "export", NULL, 0)) {
214 			/* Process export requests in vfs_mount.c. */
215 			return (error);
216 		}
217 	}
218 
219 	/*
220 	 * Not an update, or updating the name: look up the name
221 	 * and verify that it refers to a sensible disk device.
222 	 */
223 	if (fspec == NULL)
224 		return (EINVAL);
225 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
226 	if ((error = namei(ndp)) != 0)
227 		return (error);
228 	NDFREE(ndp, NDF_ONLY_PNBUF);
229 	devvp = ndp->ni_vp;
230 
231 	if (!vn_isdisk(devvp, &error)) {
232 		vput(devvp);
233 		return (error);
234 	}
235 
236 	/*
237 	 * If mount by non-root, then verify that user has necessary
238 	 * permissions on the device.
239 	 *
240 	 * XXXRW: VOP_ACCESS() enough?
241 	 */
242 	accmode = VREAD;
243 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
244 		accmode |= VWRITE;
245 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
246 	if (error)
247 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
248 	if (error) {
249 		vput(devvp);
250 		return (error);
251 	}
252 
253 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
254 		error = ext2_mountfs(devvp, mp);
255 	} else {
256 		if (devvp != ump->um_devvp) {
257 			vput(devvp);
258 			return (EINVAL);	/* needs translation */
259 		} else
260 			vput(devvp);
261 	}
262 	if (error) {
263 		vrele(devvp);
264 		return (error);
265 	}
266 	ump = VFSTOEXT2(mp);
267 	fs = ump->um_e2fs;
268 
269 	/*
270 	 * Note that this strncpy() is ok because of a check at the start
271 	 * of ext2_mount().
272 	 */
273 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
274 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
275 	vfs_mountedfrom(mp, fspec);
276 	return (0);
277 }
278 
279 static int
280 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
281 {
282 
283 	if (es->e2fs_magic != E2FS_MAGIC) {
284 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
285 		    devtoname(dev), es->e2fs_magic, E2FS_MAGIC);
286 		return (1);
287 	}
288 	if (es->e2fs_rev > E2FS_REV0) {
289 		if (es->e2fs_features_incompat & ~(EXT2F_INCOMPAT_SUPP |
290 						   EXT4F_RO_INCOMPAT_SUPP)) {
291 			printf(
292 "WARNING: mount of %s denied due to unsupported optional features\n",
293 			    devtoname(dev));
294 			return (1);
295 		}
296 		if (!ronly &&
297 		    (es->e2fs_features_rocompat & ~EXT2F_ROCOMPAT_SUPP)) {
298 			printf("WARNING: R/W mount of %s denied due to "
299 			    "unsupported optional features\n", devtoname(dev));
300 			return (1);
301 		}
302 	}
303 	return (0);
304 }
305 
306 /*
307  * This computes the fields of the m_ext2fs structure from the
308  * data in the ext2fs structure read in.
309  */
310 static int
311 compute_sb_data(struct vnode *devvp, struct ext2fs *es,
312     struct m_ext2fs *fs)
313 {
314 	int db_count, error;
315 	int i;
316 	int logic_sb_block = 1;	/* XXX for now */
317 	struct buf *bp;
318 	uint32_t e2fs_descpb;
319 
320 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
321 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
322 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
323 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
324 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
325 	if (fs->e2fs_fsize)
326 		fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
327 	fs->e2fs_bpg = es->e2fs_bpg;
328 	fs->e2fs_fpg = es->e2fs_fpg;
329 	fs->e2fs_ipg = es->e2fs_ipg;
330 	if (es->e2fs_rev == E2FS_REV0) {
331 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
332 	} else {
333 		fs->e2fs_isize = es->e2fs_inode_size;
334 
335 		/*
336 		 * Simple sanity check for superblock inode size value.
337 		 */
338 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
339 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
340 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
341 			printf("ext2fs: invalid inode size %d\n",
342 			    fs->e2fs_isize);
343 			return (EIO);
344 		}
345 	}
346 	/* Check for extra isize in big inodes. */
347 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
348 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
349 		printf("ext2fs: no space for extra inode timestamps\n");
350 		return (EINVAL);
351 	}
352 
353 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
354 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
355 	/* s_resuid / s_resgid ? */
356 	fs->e2fs_gcount = howmany(es->e2fs_bcount - es->e2fs_first_dblock,
357 	    EXT2_BLOCKS_PER_GROUP(fs));
358 	e2fs_descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
359 	db_count = howmany(fs->e2fs_gcount, e2fs_descpb);
360 	fs->e2fs_gdbcount = db_count;
361 	fs->e2fs_gd = malloc(db_count * fs->e2fs_bsize,
362 	    M_EXT2MNT, M_WAITOK);
363 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
364 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
365 
366 	/*
367 	 * Adjust logic_sb_block.
368 	 * Godmar thinks: if the blocksize is greater than 1024, then
369 	 * the superblock is logically part of block zero.
370 	 */
371 	if (fs->e2fs_bsize > SBSIZE)
372 		logic_sb_block = 0;
373 	for (i = 0; i < db_count; i++) {
374 		error = bread(devvp,
375 		    fsbtodb(fs, logic_sb_block + i + 1),
376 		    fs->e2fs_bsize, NOCRED, &bp);
377 		if (error) {
378 			free(fs->e2fs_contigdirs, M_EXT2MNT);
379 			free(fs->e2fs_gd, M_EXT2MNT);
380 			brelse(bp);
381 			return (error);
382 		}
383 		e2fs_cgload((struct ext2_gd *)bp->b_data,
384 		    &fs->e2fs_gd[
385 		    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
386 		    fs->e2fs_bsize);
387 		brelse(bp);
388 		bp = NULL;
389 	}
390 	/* Initialization for the ext2 Orlov allocator variant. */
391 	fs->e2fs_total_dir = 0;
392 	for (i = 0; i < fs->e2fs_gcount; i++)
393 		fs->e2fs_total_dir += fs->e2fs_gd[i].ext2bgd_ndirs;
394 
395 	if (es->e2fs_rev == E2FS_REV0 ||
396 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
397 		fs->e2fs_maxfilesize = 0x7fffffff;
398 	else {
399 		fs->e2fs_maxfilesize = 0xffffffffffff;
400 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
401 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
402 	}
403 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
404 		fs->e2fs_uhash = 3;
405 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
406 #ifdef __CHAR_UNSIGNED__
407 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
408 		fs->e2fs_uhash = 3;
409 #else
410 		es->e4fs_flags |= E2FS_SIGNED_HASH;
411 #endif
412 	}
413 
414 	return (0);
415 }
416 
417 /*
418  * Reload all incore data for a filesystem (used after running fsck on
419  * the root filesystem and finding things to fix). The filesystem must
420  * be mounted read-only.
421  *
422  * Things to do to update the mount:
423  *	1) invalidate all cached meta-data.
424  *	2) re-read superblock from disk.
425  *	3) invalidate all cluster summary information.
426  *	4) invalidate all inactive vnodes.
427  *	5) invalidate all cached file data.
428  *	6) re-read inode data for all active vnodes.
429  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
430  */
431 static int
432 ext2_reload(struct mount *mp, struct thread *td)
433 {
434 	struct vnode *vp, *mvp, *devvp;
435 	struct inode *ip;
436 	struct buf *bp;
437 	struct ext2fs *es;
438 	struct m_ext2fs *fs;
439 	struct csum *sump;
440 	int error, i;
441 	int32_t *lp;
442 
443 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
444 		return (EINVAL);
445 	/*
446 	 * Step 1: invalidate all cached meta-data.
447 	 */
448 	devvp = VFSTOEXT2(mp)->um_devvp;
449 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
450 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
451 		panic("ext2_reload: dirty1");
452 	VOP_UNLOCK(devvp, 0);
453 
454 	/*
455 	 * Step 2: re-read superblock from disk.
456 	 * constants have been adjusted for ext2
457 	 */
458 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
459 		return (error);
460 	es = (struct ext2fs *)bp->b_data;
461 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
462 		brelse(bp);
463 		return (EIO);		/* XXX needs translation */
464 	}
465 	fs = VFSTOEXT2(mp)->um_e2fs;
466 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
467 
468 	if ((error = compute_sb_data(devvp, es, fs)) != 0) {
469 		brelse(bp);
470 		return (error);
471 	}
472 #ifdef UNKLAR
473 	if (fs->fs_sbsize < SBSIZE)
474 		bp->b_flags |= B_INVAL;
475 #endif
476 	brelse(bp);
477 
478 	/*
479 	 * Step 3: invalidate all cluster summary information.
480 	 */
481 	if (fs->e2fs_contigsumsize > 0) {
482 		lp = fs->e2fs_maxcluster;
483 		sump = fs->e2fs_clustersum;
484 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
485 			*lp++ = fs->e2fs_contigsumsize;
486 			sump->cs_init = 0;
487 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
488 		}
489 	}
490 
491 loop:
492 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
493 		/*
494 		 * Step 4: invalidate all cached file data.
495 		 */
496 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
497 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
498 			goto loop;
499 		}
500 		if (vinvalbuf(vp, 0, 0, 0))
501 			panic("ext2_reload: dirty2");
502 
503 		/*
504 		 * Step 5: re-read inode data for all active vnodes.
505 		 */
506 		ip = VTOI(vp);
507 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
508 		    (int)fs->e2fs_bsize, NOCRED, &bp);
509 		if (error) {
510 			VOP_UNLOCK(vp, 0);
511 			vrele(vp);
512 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
513 			return (error);
514 		}
515 		ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
516 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
517 		brelse(bp);
518 		VOP_UNLOCK(vp, 0);
519 		vrele(vp);
520 	}
521 	return (0);
522 }
523 
524 /*
525  * Common code for mount and mountroot.
526  */
527 static int
528 ext2_mountfs(struct vnode *devvp, struct mount *mp)
529 {
530 	struct ext2mount *ump;
531 	struct buf *bp;
532 	struct m_ext2fs *fs;
533 	struct ext2fs *es;
534 	struct cdev *dev = devvp->v_rdev;
535 	struct g_consumer *cp;
536 	struct bufobj *bo;
537 	struct csum *sump;
538 	int error;
539 	int ronly;
540 	int i;
541 	u_long size;
542 	int32_t *lp;
543 	int32_t e2fs_maxcontig;
544 
545 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
546 	/* XXX: use VOP_ACESS to check FS perms */
547 	g_topology_lock();
548 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
549 	g_topology_unlock();
550 	VOP_UNLOCK(devvp, 0);
551 	if (error)
552 		return (error);
553 
554 	/* XXX: should we check for some sectorsize or 512 instead? */
555 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
556 	    (SBSIZE < cp->provider->sectorsize)) {
557 		g_topology_lock();
558 		g_vfs_close(cp);
559 		g_topology_unlock();
560 		return (EINVAL);
561 	}
562 
563 	bo = &devvp->v_bufobj;
564 	bo->bo_private = cp;
565 	bo->bo_ops = g_vfs_bufops;
566 	if (devvp->v_rdev->si_iosize_max != 0)
567 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
568 	if (mp->mnt_iosize_max > MAXPHYS)
569 		mp->mnt_iosize_max = MAXPHYS;
570 
571 	bp = NULL;
572 	ump = NULL;
573 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
574 		goto out;
575 	es = (struct ext2fs *)bp->b_data;
576 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
577 		error = EINVAL;		/* XXX needs translation */
578 		goto out;
579 	}
580 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
581 	    (es->e2fs_state & E2FS_ERRORS)) {
582 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
583 			printf(
584 "WARNING: Filesystem was not properly dismounted\n");
585 		} else {
586 			printf(
587 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
588 			error = EPERM;
589 			goto out;
590 		}
591 	}
592 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
593 
594 	/*
595 	 * I don't know whether this is the right strategy. Note that
596 	 * we dynamically allocate both an m_ext2fs and an ext2fs
597 	 * while Linux keeps the super block in a locked buffer.
598 	 */
599 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
600 	    M_EXT2MNT, M_WAITOK | M_ZERO);
601 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
602 	    M_EXT2MNT, M_WAITOK);
603 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
604 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
605 	if ((error = compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
606 		goto out;
607 
608 	/*
609 	 * Calculate the maximum contiguous blocks and size of cluster summary
610 	 * array.  In FFS this is done by newfs; however, the superblock
611 	 * in ext2fs doesn't have these variables, so we can calculate
612 	 * them here.
613 	 */
614 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
615 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
616 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
617 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
618 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
619 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
620 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
621 		lp = ump->um_e2fs->e2fs_maxcluster;
622 		sump = ump->um_e2fs->e2fs_clustersum;
623 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
624 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
625 			sump->cs_init = 0;
626 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
627 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
628 		}
629 	}
630 
631 	brelse(bp);
632 	bp = NULL;
633 	fs = ump->um_e2fs;
634 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
635 
636 	/*
637 	 * If the fs is not mounted read-only, make sure the super block is
638 	 * always written back on a sync().
639 	 */
640 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
641 	if (ronly == 0) {
642 		fs->e2fs_fmod = 1;	/* mark it modified */
643 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
644 	}
645 	mp->mnt_data = ump;
646 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
647 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
648 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
649 	MNT_ILOCK(mp);
650 	mp->mnt_flag |= MNT_LOCAL;
651 	MNT_IUNLOCK(mp);
652 	ump->um_mountp = mp;
653 	ump->um_dev = dev;
654 	ump->um_devvp = devvp;
655 	ump->um_bo = &devvp->v_bufobj;
656 	ump->um_cp = cp;
657 
658 	/*
659 	 * Setting those two parameters allowed us to use
660 	 * ufs_bmap w/o changse!
661 	 */
662 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
663 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
664 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
665 	if (ronly == 0)
666 		ext2_sbupdate(ump, MNT_WAIT);
667 	/*
668 	 * Initialize filesystem stat information in mount struct.
669 	 */
670 	MNT_ILOCK(mp);
671 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
672 	    MNTK_USES_BCACHE;
673 	MNT_IUNLOCK(mp);
674 	return (0);
675 out:
676 	if (bp)
677 		brelse(bp);
678 	if (cp != NULL) {
679 		g_topology_lock();
680 		g_vfs_close(cp);
681 		g_topology_unlock();
682 	}
683 	if (ump) {
684 		mtx_destroy(EXT2_MTX(ump));
685 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
686 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
687 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
688 		free(ump->um_e2fs, M_EXT2MNT);
689 		free(ump, M_EXT2MNT);
690 		mp->mnt_data = NULL;
691 	}
692 	return (error);
693 }
694 
695 /*
696  * Unmount system call.
697  */
698 static int
699 ext2_unmount(struct mount *mp, int mntflags)
700 {
701 	struct ext2mount *ump;
702 	struct m_ext2fs *fs;
703 	struct csum *sump;
704 	int error, flags, i, ronly;
705 
706 	flags = 0;
707 	if (mntflags & MNT_FORCE) {
708 		if (mp->mnt_flag & MNT_ROOTFS)
709 			return (EINVAL);
710 		flags |= FORCECLOSE;
711 	}
712 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
713 		return (error);
714 	ump = VFSTOEXT2(mp);
715 	fs = ump->um_e2fs;
716 	ronly = fs->e2fs_ronly;
717 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
718 		if (fs->e2fs_wasvalid)
719 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
720 		ext2_sbupdate(ump, MNT_WAIT);
721 	}
722 
723 	g_topology_lock();
724 	g_vfs_close(ump->um_cp);
725 	g_topology_unlock();
726 	vrele(ump->um_devvp);
727 	sump = fs->e2fs_clustersum;
728 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
729 		free(sump->cs_sum, M_EXT2MNT);
730 	free(fs->e2fs_clustersum, M_EXT2MNT);
731 	free(fs->e2fs_maxcluster, M_EXT2MNT);
732 	free(fs->e2fs_gd, M_EXT2MNT);
733 	free(fs->e2fs_contigdirs, M_EXT2MNT);
734 	free(fs->e2fs, M_EXT2MNT);
735 	free(fs, M_EXT2MNT);
736 	free(ump, M_EXT2MNT);
737 	mp->mnt_data = NULL;
738 	MNT_ILOCK(mp);
739 	mp->mnt_flag &= ~MNT_LOCAL;
740 	MNT_IUNLOCK(mp);
741 	return (error);
742 }
743 
744 /*
745  * Flush out all the files in a filesystem.
746  */
747 static int
748 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
749 {
750 	int error;
751 
752 	error = vflush(mp, 0, flags, td);
753 	return (error);
754 }
755 
756 /*
757  * Get filesystem statistics.
758  */
759 int
760 ext2_statfs(struct mount *mp, struct statfs *sbp)
761 {
762 	struct ext2mount *ump;
763 	struct m_ext2fs *fs;
764 	uint32_t overhead, overhead_per_group, ngdb;
765 	int i, ngroups;
766 
767 	ump = VFSTOEXT2(mp);
768 	fs = ump->um_e2fs;
769 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
770 		panic("ext2_statfs");
771 
772 	/*
773 	 * Compute the overhead (FS structures)
774 	 */
775 	overhead_per_group =
776 	    1 /* block bitmap */ +
777 	    1 /* inode bitmap */ +
778 	    fs->e2fs_itpg;
779 	overhead = fs->e2fs->e2fs_first_dblock +
780 	    fs->e2fs_gcount * overhead_per_group;
781 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
782 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
783 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
784 			if (cg_has_sb(i))
785 				ngroups++;
786 		}
787 	} else {
788 		ngroups = fs->e2fs_gcount;
789 	}
790 	ngdb = fs->e2fs_gdbcount;
791 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
792 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
793 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
794 	overhead += ngroups * (1 /* superblock */ + ngdb);
795 
796 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
797 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
798 	sbp->f_blocks = fs->e2fs->e2fs_bcount - overhead;
799 	sbp->f_bfree = fs->e2fs->e2fs_fbcount;
800 	sbp->f_bavail = sbp->f_bfree - fs->e2fs->e2fs_rbcount;
801 	sbp->f_files = fs->e2fs->e2fs_icount;
802 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
803 	return (0);
804 }
805 
806 /*
807  * Go through the disk queues to initiate sandbagged IO;
808  * go through the inodes to write those that have been modified;
809  * initiate the writing of the super block if it has been modified.
810  *
811  * Note: we are always called with the filesystem marked `MPBUSY'.
812  */
813 static int
814 ext2_sync(struct mount *mp, int waitfor)
815 {
816 	struct vnode *mvp, *vp;
817 	struct thread *td;
818 	struct inode *ip;
819 	struct ext2mount *ump = VFSTOEXT2(mp);
820 	struct m_ext2fs *fs;
821 	int error, allerror = 0;
822 
823 	td = curthread;
824 	fs = ump->um_e2fs;
825 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
826 		printf("fs = %s\n", fs->e2fs_fsmnt);
827 		panic("ext2_sync: rofs mod");
828 	}
829 
830 	/*
831 	 * Write back each (modified) inode.
832 	 */
833 loop:
834 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
835 		if (vp->v_type == VNON) {
836 			VI_UNLOCK(vp);
837 			continue;
838 		}
839 		ip = VTOI(vp);
840 		if ((ip->i_flag &
841 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
842 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
843 		    waitfor == MNT_LAZY)) {
844 			VI_UNLOCK(vp);
845 			continue;
846 		}
847 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
848 		if (error) {
849 			if (error == ENOENT) {
850 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
851 				goto loop;
852 			}
853 			continue;
854 		}
855 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
856 			allerror = error;
857 		VOP_UNLOCK(vp, 0);
858 		vrele(vp);
859 	}
860 
861 	/*
862 	 * Force stale filesystem control information to be flushed.
863 	 */
864 	if (waitfor != MNT_LAZY) {
865 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
866 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
867 			allerror = error;
868 		VOP_UNLOCK(ump->um_devvp, 0);
869 	}
870 
871 	/*
872 	 * Write back modified superblock.
873 	 */
874 	if (fs->e2fs_fmod != 0) {
875 		fs->e2fs_fmod = 0;
876 		fs->e2fs->e2fs_wtime = time_second;
877 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
878 			allerror = error;
879 	}
880 	return (allerror);
881 }
882 
883 /*
884  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
885  * in from disk.  If it is in core, wait for the lock bit to clear, then
886  * return the inode locked.  Detection and handling of mount points must be
887  * done by the calling routine.
888  */
889 static int
890 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
891 {
892 	struct m_ext2fs *fs;
893 	struct inode *ip;
894 	struct ext2mount *ump;
895 	struct buf *bp;
896 	struct vnode *vp;
897 	struct thread *td;
898 	int i, error;
899 	int used_blocks;
900 
901 	td = curthread;
902 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
903 	if (error || *vpp != NULL)
904 		return (error);
905 
906 	ump = VFSTOEXT2(mp);
907 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
908 
909 	/* Allocate a new vnode/inode. */
910 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
911 		*vpp = NULL;
912 		free(ip, M_EXT2NODE);
913 		return (error);
914 	}
915 	vp->v_data = ip;
916 	ip->i_vnode = vp;
917 	ip->i_e2fs = fs = ump->um_e2fs;
918 	ip->i_ump = ump;
919 	ip->i_number = ino;
920 
921 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
922 	error = insmntque(vp, mp);
923 	if (error != 0) {
924 		free(ip, M_EXT2NODE);
925 		*vpp = NULL;
926 		return (error);
927 	}
928 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
929 	if (error || *vpp != NULL)
930 		return (error);
931 
932 	/* Read in the disk contents for the inode, copy into the inode. */
933 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
934 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
935 		/*
936 		 * The inode does not contain anything useful, so it would
937 		 * be misleading to leave it on its hash chain. With mode
938 		 * still zero, it will be unlinked and returned to the free
939 		 * list by vput().
940 		 */
941 		brelse(bp);
942 		vput(vp);
943 		*vpp = NULL;
944 		return (error);
945 	}
946 	/* convert ext2 inode to dinode */
947 	ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data + EXT2_INODE_SIZE(fs) *
948 	    ino_to_fsbo(fs, ino)), ip);
949 	ip->i_block_group = ino_to_cg(fs, ino);
950 	ip->i_next_alloc_block = 0;
951 	ip->i_next_alloc_goal = 0;
952 
953 	/*
954 	 * Now we want to make sure that block pointers for unused
955 	 * blocks are zeroed out - ext2_balloc depends on this
956 	 * although for regular files and directories only
957 	 *
958 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
959 	 * out because we could corrupt the extent tree.
960 	 */
961 	if (!(ip->i_flag & IN_E4EXTENTS) &&
962 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
963 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
964 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
965 			ip->i_db[i] = 0;
966 	}
967 #ifdef EXT2FS_DEBUG
968 	ext2_print_inode(ip);
969 #endif
970 	bqrelse(bp);
971 
972 	/*
973 	 * Initialize the vnode from the inode, check for aliases.
974 	 * Note that the underlying vnode may have changed.
975 	 */
976 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
977 		vput(vp);
978 		*vpp = NULL;
979 		return (error);
980 	}
981 
982 	/*
983 	 * Finish inode initialization.
984 	 */
985 
986 	*vpp = vp;
987 	return (0);
988 }
989 
990 /*
991  * File handle to vnode
992  *
993  * Have to be really careful about stale file handles:
994  * - check that the inode number is valid
995  * - call ext2_vget() to get the locked inode
996  * - check for an unallocated inode (i_mode == 0)
997  * - check that the given client host has export rights and return
998  *   those rights via. exflagsp and credanonp
999  */
1000 static int
1001 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1002 {
1003 	struct inode *ip;
1004 	struct ufid *ufhp;
1005 	struct vnode *nvp;
1006 	struct m_ext2fs *fs;
1007 	int error;
1008 
1009 	ufhp = (struct ufid *)fhp;
1010 	fs = VFSTOEXT2(mp)->um_e2fs;
1011 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1012 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1013 		return (ESTALE);
1014 
1015 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1016 	if (error) {
1017 		*vpp = NULLVP;
1018 		return (error);
1019 	}
1020 	ip = VTOI(nvp);
1021 	if (ip->i_mode == 0 ||
1022 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1023 		vput(nvp);
1024 		*vpp = NULLVP;
1025 		return (ESTALE);
1026 	}
1027 	*vpp = nvp;
1028 	vnode_create_vobject(*vpp, 0, curthread);
1029 	return (0);
1030 }
1031 
1032 /*
1033  * Write a superblock and associated information back to disk.
1034  */
1035 static int
1036 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1037 {
1038 	struct m_ext2fs *fs = mp->um_e2fs;
1039 	struct ext2fs *es = fs->e2fs;
1040 	struct buf *bp;
1041 	int error = 0;
1042 
1043 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1044 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1045 	if (waitfor == MNT_WAIT)
1046 		error = bwrite(bp);
1047 	else
1048 		bawrite(bp);
1049 
1050 	/*
1051 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1052 	 * are not busy at this point and are (hopefully) written by the
1053 	 * usual sync mechanism. No need to write them here.
1054 	 */
1055 	return (error);
1056 }
1057 int
1058 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1059 {
1060 	struct m_ext2fs *fs = mp->um_e2fs;
1061 	struct buf *bp;
1062 	int i, error = 0, allerror = 0;
1063 
1064 	allerror = ext2_sbupdate(mp, waitfor);
1065 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1066 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1067 		    fs->e2fs->e2fs_first_dblock +
1068 		    1 /* superblock */ + i), fs->e2fs_bsize, 0, 0, 0);
1069 		e2fs_cgsave(&fs->e2fs_gd[
1070 		    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1071 		    (struct ext2_gd *)bp->b_data, fs->e2fs_bsize);
1072 		if (waitfor == MNT_WAIT)
1073 			error = bwrite(bp);
1074 		else
1075 			bawrite(bp);
1076 	}
1077 
1078 	if (!allerror && error)
1079 		allerror = error;
1080 	return (allerror);
1081 }
1082 
1083 /*
1084  * Return the root of a filesystem.
1085  */
1086 static int
1087 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1088 {
1089 	struct vnode *nvp;
1090 	int error;
1091 
1092 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1093 	if (error)
1094 		return (error);
1095 	*vpp = nvp;
1096 	return (0);
1097 }
1098