xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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  * 4. 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/fcntl.h>
51 #include <sys/malloc.h>
52 #include <sys/stat.h>
53 #include <sys/mutex.h>
54 
55 #include <geom/geom.h>
56 #include <geom/geom_vfs.h>
57 
58 #include <fs/ext2fs/ext2_mount.h>
59 #include <fs/ext2fs/inode.h>
60 
61 #include <fs/ext2fs/fs.h>
62 #include <fs/ext2fs/ext2fs.h>
63 #include <fs/ext2fs/ext2_dinode.h>
64 #include <fs/ext2fs/ext2_extern.h>
65 
66 static int	ext2_flushfiles(struct mount *mp, int flags, struct thread *td);
67 static int	ext2_mountfs(struct vnode *, struct mount *);
68 static int	ext2_reload(struct mount *mp, struct thread *td);
69 static int	ext2_sbupdate(struct ext2mount *, int);
70 static int	ext2_cgupdate(struct ext2mount *, int);
71 static vfs_unmount_t		ext2_unmount;
72 static vfs_root_t		ext2_root;
73 static vfs_statfs_t		ext2_statfs;
74 static vfs_sync_t		ext2_sync;
75 static vfs_vget_t		ext2_vget;
76 static vfs_fhtovp_t		ext2_fhtovp;
77 static vfs_mount_t		ext2_mount;
78 
79 MALLOC_DEFINE(M_EXT2NODE, "ext2_node", "EXT2 vnode private part");
80 static MALLOC_DEFINE(M_EXT2MNT, "ext2_mount", "EXT2 mount structure");
81 
82 static struct vfsops ext2fs_vfsops = {
83 	.vfs_fhtovp =		ext2_fhtovp,
84 	.vfs_mount =		ext2_mount,
85 	.vfs_root =		ext2_root,	/* root inode via vget */
86 	.vfs_statfs =		ext2_statfs,
87 	.vfs_sync =		ext2_sync,
88 	.vfs_unmount =		ext2_unmount,
89 	.vfs_vget =		ext2_vget,
90 };
91 
92 VFS_SET(ext2fs_vfsops, ext2fs, 0);
93 
94 static int	ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev,
95 		    int ronly);
96 static int	compute_sb_data(struct vnode * devvp,
97 		    struct ext2fs * es, struct m_ext2fs * fs);
98 
99 static const char *ext2_opts[] = { "acls", "async", "noatime", "noclusterr",
100     "noclusterw", "noexec", "export", "force", "from", "multilabel",
101     "suiddir", "nosymfollow", "sync", "union", NULL };
102 
103 /*
104  * VFS Operations.
105  *
106  * mount system call
107  */
108 static int
109 ext2_mount(struct mount *mp)
110 {
111 	struct vfsoptlist *opts;
112 	struct vnode *devvp;
113 	struct thread *td;
114 	struct ext2mount *ump = 0;
115 	struct m_ext2fs *fs;
116 	struct nameidata nd, *ndp = &nd;
117 	accmode_t accmode;
118 	char *path, *fspec;
119 	int error, flags, len;
120 
121 	td = curthread;
122 	opts = mp->mnt_optnew;
123 
124 	if (vfs_filteropt(opts, ext2_opts))
125 		return (EINVAL);
126 
127 	vfs_getopt(opts, "fspath", (void **)&path, NULL);
128 	/* Double-check the length of path.. */
129 	if (strlen(path) >= MAXMNTLEN - 1)
130 		return (ENAMETOOLONG);
131 
132 	fspec = NULL;
133 	error = vfs_getopt(opts, "from", (void **)&fspec, &len);
134 	if (!error && fspec[len - 1] != '\0')
135 		return (EINVAL);
136 
137 	/*
138 	 * If updating, check whether changing from read-only to
139 	 * read/write; if there is no device name, that's all we do.
140 	 */
141 	if (mp->mnt_flag & MNT_UPDATE) {
142 		ump = VFSTOEXT2(mp);
143 		fs = ump->um_e2fs;
144 		error = 0;
145 		if (fs->e2fs_ronly == 0 &&
146 		    vfs_flagopt(opts, "ro", NULL, 0)) {
147 			error = VFS_SYNC(mp, MNT_WAIT);
148 			if (error)
149 				return (error);
150 			flags = WRITECLOSE;
151 			if (mp->mnt_flag & MNT_FORCE)
152 				flags |= FORCECLOSE;
153 			error = ext2_flushfiles(mp, flags, td);
154 			if ( error == 0 && fs->e2fs_wasvalid && ext2_cgupdate(ump, MNT_WAIT) == 0) {
155 				fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
156 				ext2_sbupdate(ump, MNT_WAIT);
157 			}
158 			fs->e2fs_ronly = 1;
159 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
160 			DROP_GIANT();
161 			g_topology_lock();
162 			g_access(ump->um_cp, 0, -1, 0);
163 			g_topology_unlock();
164 			PICKUP_GIANT();
165 		}
166 		if (!error && (mp->mnt_flag & MNT_RELOAD))
167 			error = ext2_reload(mp, td);
168 		if (error)
169 			return (error);
170 		devvp = ump->um_devvp;
171 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
172 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
173 				return (EPERM);
174 
175 			/*
176 			 * If upgrade to read-write by non-root, then verify
177 			 * that user has necessary permissions on the device.
178 			 */
179 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
180 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
181 			    td->td_ucred, td);
182 			if (error)
183 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
184 			if (error) {
185 				VOP_UNLOCK(devvp, 0);
186 				return (error);
187 			}
188 			VOP_UNLOCK(devvp, 0);
189 			DROP_GIANT();
190 			g_topology_lock();
191 			error = g_access(ump->um_cp, 0, 1, 0);
192 			g_topology_unlock();
193 			PICKUP_GIANT();
194 			if (error)
195 				return (error);
196 
197 			if ((fs->e2fs->e2fs_state & E2FS_ISCLEAN) == 0 ||
198 			    (fs->e2fs->e2fs_state & E2FS_ERRORS)) {
199 				if (mp->mnt_flag & MNT_FORCE) {
200 					printf(
201 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
202 				} else {
203 					printf(
204 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
205 					    fs->e2fs_fsmnt);
206 					return (EPERM);
207 				}
208 			}
209 			fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;
210 			(void)ext2_cgupdate(ump, MNT_WAIT);
211 			fs->e2fs_ronly = 0;
212 			MNT_ILOCK(mp);
213 			mp->mnt_flag &= ~MNT_RDONLY;
214 			MNT_IUNLOCK(mp);
215 		}
216 		if (vfs_flagopt(opts, "export", NULL, 0)) {
217 			/* Process export requests in vfs_mount.c. */
218 			return (error);
219 		}
220 	}
221 
222 	/*
223 	 * Not an update, or updating the name: look up the name
224 	 * and verify that it refers to a sensible disk device.
225 	 */
226 	if (fspec == NULL)
227 		return (EINVAL);
228 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
229 	if ((error = namei(ndp)) != 0)
230 		return (error);
231 	NDFREE(ndp, NDF_ONLY_PNBUF);
232 	devvp = ndp->ni_vp;
233 
234 	if (!vn_isdisk(devvp, &error)) {
235 		vput(devvp);
236 		return (error);
237 	}
238 
239 	/*
240 	 * If mount by non-root, then verify that user has necessary
241 	 * permissions on the device.
242 	 *
243 	 * XXXRW: VOP_ACCESS() enough?
244 	 */
245 	accmode = VREAD;
246 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
247 		accmode |= VWRITE;
248 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
249 	if (error)
250 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
251 	if (error) {
252 		vput(devvp);
253 		return (error);
254 	}
255 
256 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
257 		error = ext2_mountfs(devvp, mp);
258 	} else {
259 		if (devvp != ump->um_devvp) {
260 			vput(devvp);
261 			return (EINVAL);	/* needs translation */
262 		} else
263 			vput(devvp);
264 	}
265 	if (error) {
266 		vrele(devvp);
267 		return (error);
268 	}
269 	ump = VFSTOEXT2(mp);
270 	fs = ump->um_e2fs;
271 
272 	/*
273 	 * Note that this strncpy() is ok because of a check at the start
274 	 * of ext2_mount().
275 	 */
276 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
277 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
278 	vfs_mountedfrom(mp, fspec);
279 	return (0);
280 }
281 
282 static int
283 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
284 {
285 
286 	if (es->e2fs_magic != E2FS_MAGIC) {
287 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
288 		    devtoname(dev), es->e2fs_magic, E2FS_MAGIC);
289 		return (1);
290 	}
291 	if (es->e2fs_rev > E2FS_REV0) {
292 		if (es->e2fs_features_incompat & ~EXT2F_INCOMPAT_SUPP) {
293 			printf(
294 "WARNING: mount of %s denied due to unsupported optional features\n",
295 			    devtoname(dev));
296 			return (1);
297 		}
298 		if (!ronly &&
299 		    (es->e2fs_features_rocompat & ~EXT2F_ROCOMPAT_SUPP)) {
300 			printf("WARNING: R/W mount of %s denied due to "
301 			    "unsupported optional features\n", devtoname(dev));
302 			return (1);
303 		}
304 	}
305 	return (0);
306 }
307 
308 /*
309  * This computes the fields of the  ext2_sb_info structure from the
310  * data in the ext2_super_block structure read in.
311  */
312 static int
313 compute_sb_data(struct vnode *devvp, struct ext2fs *es,
314     struct m_ext2fs *fs)
315 {
316 	int db_count, error;
317 	int i;
318 	int logic_sb_block = 1;	/* XXX for now */
319 	struct buf *bp;
320 
321 	fs->e2fs_bsize = EXT2_MIN_BLOCK_SIZE << es->e2fs_log_bsize;
322 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
323 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
324 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
325 	fs->e2fs_blocksize_bits = es->e2fs_log_bsize + 10;
326 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
327 	if (fs->e2fs_fsize)
328 		fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
329 	fs->e2fs_bpg = es->e2fs_bpg;
330 	fs->e2fs_fpg = es->e2fs_fpg;
331 	fs->e2fs_ipg = es->e2fs_ipg;
332 	if (es->e2fs_rev == E2FS_REV0) {
333 		fs->e2fs_first_inode = EXT2_FIRSTINO;
334 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE ;
335 	} else {
336 		fs->e2fs_first_inode = es->e2fs_first_ino;
337 		fs->e2fs_isize = es->e2fs_inode_size;
338 
339 		/*
340 		 * Simple sanity check for superblock inode size value.
341 		 */
342 		if (fs->e2fs_isize < E2FS_REV0_INODE_SIZE  ||
343 		    fs->e2fs_isize > fs->e2fs_bsize ||
344 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
345 			printf("EXT2-fs: invalid inode size %d\n",
346 			    fs->e2fs_isize);
347 			return (EIO);
348 		}
349 	}
350 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
351 	fs->e2fs_itpg = fs->e2fs_ipg /fs->e2fs_ipb;
352 	fs->e2fs_descpb = fs->e2fs_bsize / sizeof (struct ext2_gd);
353 	/* s_resuid / s_resgid ? */
354 	fs->e2fs_gcount = (es->e2fs_bcount - es->e2fs_first_dblock +
355 	    EXT2_BLOCKS_PER_GROUP(fs) - 1) / EXT2_BLOCKS_PER_GROUP(fs);
356 	db_count = (fs->e2fs_gcount + EXT2_DESC_PER_BLOCK(fs) - 1) /
357 	    EXT2_DESC_PER_BLOCK(fs);
358 	fs->e2fs_gdbcount = db_count;
359 	fs->e2fs_gd = malloc(db_count * fs->e2fs_bsize,
360 	    M_EXT2MNT, M_WAITOK);
361 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount * sizeof(*fs->e2fs_contigdirs),
362 	    M_EXT2MNT, M_WAITOK);
363 
364 	/*
365 	 * Adjust logic_sb_block.
366 	 * Godmar thinks: if the blocksize is greater than 1024, then
367 	 * the superblock is logically part of block zero.
368 	 */
369 	if(fs->e2fs_bsize > SBSIZE)
370 		logic_sb_block = 0;
371 	for (i = 0; i < db_count; i++) {
372 		error = bread(devvp ,
373 			 fsbtodb(fs, logic_sb_block + i + 1 ),
374 			fs->e2fs_bsize, NOCRED, &bp);
375 		if (error) {
376 			free(fs->e2fs_gd, M_EXT2MNT);
377 			brelse(bp);
378 			return (error);
379 		}
380 		e2fs_cgload((struct ext2_gd *)bp->b_data,
381 		    &fs->e2fs_gd[
382 			i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
383 		    fs->e2fs_bsize);
384 		brelse(bp);
385 		bp = NULL;
386 	}
387 	fs->e2fs_total_dir = 0;
388 	for (i=0; i < fs->e2fs_gcount; i++){
389 		fs->e2fs_total_dir += fs->e2fs_gd[i].ext2bgd_ndirs;
390 		fs->e2fs_contigdirs[i] = 0;
391 	}
392 	if (es->e2fs_rev == E2FS_REV0 ||
393 	    (es->e2fs_features_rocompat & EXT2F_ROCOMPAT_LARGEFILE) == 0)
394 		fs->e2fs_maxfilesize = 0x7fffffff;
395 	else
396 		fs->e2fs_maxfilesize = 0x7fffffffffffffff;
397 	return (0);
398 }
399 
400 /*
401  * Reload all incore data for a filesystem (used after running fsck on
402  * the root filesystem and finding things to fix). The filesystem must
403  * be mounted read-only.
404  *
405  * Things to do to update the mount:
406  *	1) invalidate all cached meta-data.
407  *	2) re-read superblock from disk.
408  *	3) re-read summary information from disk.
409  *	4) invalidate all inactive vnodes.
410  *	5) invalidate all cached file data.
411  *	6) re-read inode data for all active vnodes.
412  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
413  */
414 static int
415 ext2_reload(struct mount *mp, struct thread *td)
416 {
417 	struct vnode *vp, *mvp, *devvp;
418 	struct inode *ip;
419 	struct buf *bp;
420 	struct ext2fs *es;
421 	struct m_ext2fs *fs;
422 	int error;
423 
424 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
425 		return (EINVAL);
426 	/*
427 	 * Step 1: invalidate all cached meta-data.
428 	 */
429 	devvp = VFSTOEXT2(mp)->um_devvp;
430 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
431 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
432 		panic("ext2_reload: dirty1");
433 	VOP_UNLOCK(devvp, 0);
434 
435 	/*
436 	 * Step 2: re-read superblock from disk.
437 	 * constants have been adjusted for ext2
438 	 */
439 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
440 		return (error);
441 	es = (struct ext2fs *)bp->b_data;
442 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
443 		brelse(bp);
444 		return (EIO);		/* XXX needs translation */
445 	}
446 	fs = VFSTOEXT2(mp)->um_e2fs;
447 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
448 
449 	if((error = compute_sb_data(devvp, es, fs)) != 0) {
450 		brelse(bp);
451 		return (error);
452 	}
453 #ifdef UNKLAR
454 	if (fs->fs_sbsize < SBSIZE)
455 		bp->b_flags |= B_INVAL;
456 #endif
457 	brelse(bp);
458 
459 loop:
460 	MNT_ILOCK(mp);
461 	MNT_VNODE_FOREACH(vp, mp, mvp) {
462 		VI_LOCK(vp);
463 		if (vp->v_iflag & VI_DOOMED) {
464 			VI_UNLOCK(vp);
465 			continue;
466 		}
467 		MNT_IUNLOCK(mp);
468 		/*
469 		 * Step 4: invalidate all cached file data.
470 		 */
471 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
472 			MNT_VNODE_FOREACH_ABORT(mp, mvp);
473 			goto loop;
474 		}
475 		if (vinvalbuf(vp, 0, 0, 0))
476 			panic("ext2_reload: dirty2");
477 
478 		/*
479 		 * Step 5: re-read inode data for all active vnodes.
480 		 */
481 		ip = VTOI(vp);
482 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
483 		    (int)fs->e2fs_bsize, NOCRED, &bp);
484 		if (error) {
485 			VOP_UNLOCK(vp, 0);
486 			vrele(vp);
487 			MNT_VNODE_FOREACH_ABORT(mp, mvp);
488 			return (error);
489 		}
490 		ext2_ei2i((struct ext2fs_dinode *) ((char *)bp->b_data +
491 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
492 		brelse(bp);
493 		VOP_UNLOCK(vp, 0);
494 		vrele(vp);
495 		MNT_ILOCK(mp);
496 	}
497 	MNT_IUNLOCK(mp);
498 	return (0);
499 }
500 
501 /*
502  * Common code for mount and mountroot.
503  */
504 static int
505 ext2_mountfs(struct vnode *devvp, struct mount *mp)
506 {
507 	struct ext2mount *ump;
508 	struct buf *bp;
509 	struct m_ext2fs *fs;
510 	struct ext2fs *es;
511 	struct cdev *dev = devvp->v_rdev;
512 	struct g_consumer *cp;
513 	struct bufobj *bo;
514 	int error;
515 	int ronly;
516 
517 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
518 	/* XXX: use VOP_ACESS to check FS perms */
519 	DROP_GIANT();
520 	g_topology_lock();
521 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
522 	g_topology_unlock();
523 	PICKUP_GIANT();
524 	VOP_UNLOCK(devvp, 0);
525 	if (error)
526 		return (error);
527 
528 	/* XXX: should we check for some sectorsize or 512 instead? */
529 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
530 	    (SBSIZE < cp->provider->sectorsize)) {
531 		DROP_GIANT();
532 		g_topology_lock();
533 		g_vfs_close(cp);
534 		g_topology_unlock();
535 		PICKUP_GIANT();
536 		return (EINVAL);
537 	}
538 
539 	bo = &devvp->v_bufobj;
540 	bo->bo_private = cp;
541 	bo->bo_ops = g_vfs_bufops;
542 	if (devvp->v_rdev->si_iosize_max != 0)
543 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
544 	if (mp->mnt_iosize_max > MAXPHYS)
545 		mp->mnt_iosize_max = MAXPHYS;
546 
547 	bp = NULL;
548 	ump = NULL;
549 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
550 		goto out;
551 	es = (struct ext2fs *)bp->b_data;
552 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
553 		error = EINVAL;		/* XXX needs translation */
554 		goto out;
555 	}
556 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
557 	    (es->e2fs_state & E2FS_ERRORS)) {
558 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
559 			printf(
560 "WARNING: Filesystem was not properly dismounted\n");
561 		} else {
562 			printf(
563 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
564 			error = EPERM;
565 			goto out;
566 		}
567 	}
568 	ump = malloc(sizeof *ump, M_EXT2MNT, M_WAITOK);
569 	bzero((caddr_t)ump, sizeof *ump);
570 
571 	/*
572 	 * I don't know whether this is the right strategy. Note that
573 	 * we dynamically allocate both an ext2_sb_info and an ext2_super_block
574 	 * while Linux keeps the super block in a locked buffer.
575 	 */
576 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
577 		M_EXT2MNT, M_WAITOK);
578 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
579 		M_EXT2MNT, M_WAITOK);
580 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
581 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
582 	if ((error = compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
583 		goto out;
584 
585 	brelse(bp);
586 	bp = NULL;
587 	fs = ump->um_e2fs;
588 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
589 
590 	/*
591 	 * If the fs is not mounted read-only, make sure the super block is
592 	 * always written back on a sync().
593 	 */
594 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
595 	if (ronly == 0) {
596 		fs->e2fs_fmod = 1;		/* mark it modified */
597 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
598 	}
599 	mp->mnt_data = ump;
600 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
601 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
602 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
603 	MNT_ILOCK(mp);
604 	mp->mnt_flag |= MNT_LOCAL;
605 	MNT_IUNLOCK(mp);
606 	ump->um_mountp = mp;
607 	ump->um_dev = dev;
608 	ump->um_devvp = devvp;
609 	ump->um_bo = &devvp->v_bufobj;
610 	ump->um_cp = cp;
611 
612 	/*
613 	 * Setting those two parameters allowed us to use
614 	 * ufs_bmap w/o changse!
615 	 */
616 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
617 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
618 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
619 	if (ronly == 0)
620 		ext2_sbupdate(ump, MNT_WAIT);
621 	/*
622 	 * Initialize filesystem stat information in mount struct.
623 	 */
624 	MNT_ILOCK(mp);
625  	mp->mnt_kern_flag |= MNTK_MPSAFE | MNTK_LOOKUP_SHARED |
626             MNTK_EXTENDED_SHARED;
627 	MNT_IUNLOCK(mp);
628 	return (0);
629 out:
630 	if (bp)
631 		brelse(bp);
632 	if (cp != NULL) {
633 		DROP_GIANT();
634 		g_topology_lock();
635 		g_vfs_close(cp);
636 		g_topology_unlock();
637 		PICKUP_GIANT();
638 	}
639 	if (ump) {
640 	  	mtx_destroy(EXT2_MTX(ump));
641 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
642 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
643 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
644 		free(ump->um_e2fs, M_EXT2MNT);
645 		free(ump, M_EXT2MNT);
646 		mp->mnt_data = NULL;
647 	}
648 	return (error);
649 }
650 
651 /*
652  * Unmount system call.
653  */
654 static int
655 ext2_unmount(struct mount *mp, int mntflags)
656 {
657 	struct ext2mount *ump;
658 	struct m_ext2fs *fs;
659 	int error, flags, ronly;
660 
661 	flags = 0;
662 	if (mntflags & MNT_FORCE) {
663 		if (mp->mnt_flag & MNT_ROOTFS)
664 			return (EINVAL);
665 		flags |= FORCECLOSE;
666 	}
667 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
668 		return (error);
669 	ump = VFSTOEXT2(mp);
670 	fs = ump->um_e2fs;
671 	ronly = fs->e2fs_ronly;
672 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
673 		if (fs->e2fs_wasvalid)
674  			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
675  		ext2_sbupdate(ump, MNT_WAIT);
676 	}
677 
678 	DROP_GIANT();
679 	g_topology_lock();
680 	g_vfs_close(ump->um_cp);
681 	g_topology_unlock();
682 	PICKUP_GIANT();
683 	vrele(ump->um_devvp);
684 	free(fs->e2fs_gd, M_EXT2MNT);
685 	free(fs->e2fs_contigdirs, M_EXT2MNT);
686 	free(fs->e2fs, M_EXT2MNT);
687 	free(fs, M_EXT2MNT);
688 	free(ump, M_EXT2MNT);
689 	mp->mnt_data = NULL;
690 	MNT_ILOCK(mp);
691 	mp->mnt_flag &= ~MNT_LOCAL;
692 	MNT_IUNLOCK(mp);
693 	return (error);
694 }
695 
696 /*
697  * Flush out all the files in a filesystem.
698  */
699 static int
700 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
701 {
702 	int error;
703 
704 	error = vflush(mp, 0, flags, td);
705 	return (error);
706 }
707 /*
708  * Get file system statistics.
709  */
710 int
711 ext2_statfs(struct mount *mp, struct statfs *sbp)
712 {
713 	struct ext2mount *ump;
714 	struct m_ext2fs *fs;
715 	uint32_t overhead, overhead_per_group, ngdb;
716 	int i, ngroups;
717 
718 	ump = VFSTOEXT2(mp);
719 	fs = ump->um_e2fs;
720 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
721 		panic("ext2fs_statvfs");
722 
723 	/*
724 	 * Compute the overhead (FS structures)
725 	 */
726 	overhead_per_group =
727 	    1 /* block bitmap */ +
728 	    1 /* inode bitmap */ +
729 	    fs->e2fs_itpg;
730 	overhead = fs->e2fs->e2fs_first_dblock +
731 	    fs->e2fs_gcount * overhead_per_group;
732 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
733 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
734 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
735 			if (cg_has_sb(i))
736 				ngroups++;
737 		}
738 	} else {
739 		ngroups = fs->e2fs_gcount;
740 	}
741 	ngdb = fs->e2fs_gdbcount;
742 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
743 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
744 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
745 	overhead += ngroups * (1 /* superblock */ + ngdb);
746 
747 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
748 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
749 	sbp->f_blocks = fs->e2fs->e2fs_bcount - overhead;
750 	sbp->f_bfree = fs->e2fs->e2fs_fbcount;
751 	sbp->f_bavail = sbp->f_bfree - fs->e2fs->e2fs_rbcount;
752 	sbp->f_files = fs->e2fs->e2fs_icount;
753 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
754 	return (0);
755 }
756 
757 /*
758  * Go through the disk queues to initiate sandbagged IO;
759  * go through the inodes to write those that have been modified;
760  * initiate the writing of the super block if it has been modified.
761  *
762  * Note: we are always called with the filesystem marked `MPBUSY'.
763  */
764 static int
765 ext2_sync(struct mount *mp, int waitfor)
766 {
767 	struct vnode *mvp, *vp;
768 	struct thread *td;
769 	struct inode *ip;
770 	struct ext2mount *ump = VFSTOEXT2(mp);
771 	struct m_ext2fs *fs;
772 	int error, allerror = 0;
773 
774 	td = curthread;
775 	fs = ump->um_e2fs;
776 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
777 		printf("fs = %s\n", fs->e2fs_fsmnt);
778 		panic("ext2_sync: rofs mod");
779 	}
780 
781 	/*
782 	 * Write back each (modified) inode.
783 	 */
784 	MNT_ILOCK(mp);
785 loop:
786 	MNT_VNODE_FOREACH(vp, mp, mvp) {
787 		VI_LOCK(vp);
788 		if (vp->v_type == VNON || (vp->v_iflag & VI_DOOMED)) {
789 			VI_UNLOCK(vp);
790 			continue;
791 		}
792 		MNT_IUNLOCK(mp);
793 		ip = VTOI(vp);
794 		if ((ip->i_flag &
795 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
796 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
797 		    waitfor == MNT_LAZY)) {
798 			VI_UNLOCK(vp);
799 			MNT_ILOCK(mp);
800 			continue;
801 		}
802 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
803 		if (error) {
804 			MNT_ILOCK(mp);
805 			if (error == ENOENT) {
806 				MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
807 				goto loop;
808 			}
809 			continue;
810 		}
811 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
812 			allerror = error;
813 		VOP_UNLOCK(vp, 0);
814 		vrele(vp);
815 		MNT_ILOCK(mp);
816 	}
817 	MNT_IUNLOCK(mp);
818 
819 	/*
820 	 * Force stale file system control information to be flushed.
821 	 */
822 	if (waitfor != MNT_LAZY) {
823 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
824 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
825 			allerror = error;
826 		VOP_UNLOCK(ump->um_devvp, 0);
827 	}
828 
829 	/*
830 	 * Write back modified superblock.
831 	 */
832 	if (fs->e2fs_fmod != 0) {
833 		fs->e2fs_fmod = 0;
834 		fs->e2fs->e2fs_wtime = time_second;
835 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
836 			allerror = error;
837 	}
838 	return (allerror);
839 }
840 
841 /*
842  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
843  * in from disk.  If it is in core, wait for the lock bit to clear, then
844  * return the inode locked.  Detection and handling of mount points must be
845  * done by the calling routine.
846  */
847 static int
848 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
849 {
850 	struct m_ext2fs *fs;
851 	struct inode *ip;
852 	struct ext2mount *ump;
853 	struct buf *bp;
854 	struct vnode *vp;
855 	struct cdev *dev;
856 	struct thread *td;
857 	int i, error;
858 	int used_blocks;
859 
860 	td = curthread;
861 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
862 	if (error || *vpp != NULL)
863 		return (error);
864 
865 	ump = VFSTOEXT2(mp);
866 	dev = ump->um_dev;
867 
868 	/*
869 	 * If this malloc() is performed after the getnewvnode()
870 	 * it might block, leaving a vnode with a NULL v_data to be
871 	 * found by ext2_sync() if a sync happens to fire right then,
872 	 * which will cause a panic because ext2_sync() blindly
873 	 * dereferences vp->v_data (as well it should).
874 	 */
875 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
876 
877 	/* Allocate a new vnode/inode. */
878 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
879 		*vpp = NULL;
880 		free(ip, M_EXT2NODE);
881 		return (error);
882 	}
883 	vp->v_data = ip;
884 	ip->i_vnode = vp;
885 	ip->i_e2fs = fs = ump->um_e2fs;
886 	ip->i_ump  = ump;
887 	ip->i_number = ino;
888 
889 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
890 	error = insmntque(vp, mp);
891 	if (error != 0) {
892 		free(ip, M_EXT2NODE);
893 		*vpp = NULL;
894 		return (error);
895 	}
896 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
897 	if (error || *vpp != NULL)
898 		return (error);
899 
900 	/* Read in the disk contents for the inode, copy into the inode. */
901 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
902 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
903 		/*
904 		 * The inode does not contain anything useful, so it would
905 		 * be misleading to leave it on its hash chain. With mode
906 		 * still zero, it will be unlinked and returned to the free
907 		 * list by vput().
908 		 */
909 		brelse(bp);
910 		vput(vp);
911 		*vpp = NULL;
912 		return (error);
913 	}
914 	/* convert ext2 inode to dinode */
915 	ext2_ei2i((struct ext2fs_dinode *) ((char *)bp->b_data + EXT2_INODE_SIZE(fs) *
916 			ino_to_fsbo(fs, ino)), ip);
917 	ip->i_block_group = ino_to_cg(fs, ino);
918 	ip->i_next_alloc_block = 0;
919 	ip->i_next_alloc_goal = 0;
920 	ip->i_prealloc_count = 0;
921 	ip->i_prealloc_block = 0;
922 
923 	/*
924 	 * Now we want to make sure that block pointers for unused
925 	 * blocks are zeroed out - ext2_balloc depends on this
926 	 * although for regular files and directories only
927 	 */
928 	if(S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode)) {
929 		used_blocks = (ip->i_size+fs->e2fs_bsize-1) / fs->e2fs_bsize;
930 		for(i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
931 			ip->i_db[i] = 0;
932 	}
933 /*
934 	ext2_print_inode(ip);
935 */
936 	bqrelse(bp);
937 
938 	/*
939 	 * Initialize the vnode from the inode, check for aliases.
940 	 * Note that the underlying vnode may have changed.
941 	 */
942 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
943 		vput(vp);
944 		*vpp = NULL;
945 		return (error);
946 	}
947 
948 	/*
949 	 * Finish inode initialization.
950 	 */
951 
952 	/*
953 	 * Set up a generation number for this inode if it does not
954 	 * already have one. This should only happen on old filesystems.
955 	 */
956 	if (ip->i_gen == 0) {
957 		ip->i_gen = random() / 2 + 1;
958 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
959 			ip->i_flag |= IN_MODIFIED;
960 	}
961 	*vpp = vp;
962 	return (0);
963 }
964 
965 /*
966  * File handle to vnode
967  *
968  * Have to be really careful about stale file handles:
969  * - check that the inode number is valid
970  * - call ext2_vget() to get the locked inode
971  * - check for an unallocated inode (i_mode == 0)
972  * - check that the given client host has export rights and return
973  *   those rights via. exflagsp and credanonp
974  */
975 static int
976 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
977 {
978 	struct inode *ip;
979 	struct ufid *ufhp;
980 	struct vnode *nvp;
981 	struct m_ext2fs *fs;
982 	int error;
983 
984 	ufhp = (struct ufid *)fhp;
985 	fs = VFSTOEXT2(mp)->um_e2fs;
986 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
987 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
988 		return (ESTALE);
989 
990 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
991 	if (error) {
992 		*vpp = NULLVP;
993 		return (error);
994 	}
995 	ip = VTOI(nvp);
996 	if (ip->i_mode == 0 ||
997 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
998 		vput(nvp);
999 		*vpp = NULLVP;
1000 		return (ESTALE);
1001 	}
1002 	*vpp = nvp;
1003 	vnode_create_vobject(*vpp, 0, curthread);
1004 	return (0);
1005 }
1006 
1007 /*
1008  * Write a superblock and associated information back to disk.
1009  */
1010 static int
1011 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1012 {
1013 	struct m_ext2fs *fs = mp->um_e2fs;
1014 	struct ext2fs *es = fs->e2fs;
1015 	struct buf *bp;
1016 	int error = 0;
1017 
1018 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1019 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1020 	if (waitfor == MNT_WAIT)
1021 		error = bwrite(bp);
1022 	else
1023 		bawrite(bp);
1024 
1025 	/*
1026 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1027 	 * are not busy at this point and are (hopefully) written by the
1028 	 * usual sync mechanism. No need to write them here.
1029 	 */
1030 	return (error);
1031 }
1032 int
1033 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1034 {
1035 	struct m_ext2fs *fs = mp->um_e2fs;
1036 	struct buf *bp;
1037 	int i, error = 0, allerror = 0;
1038 
1039 	allerror = ext2_sbupdate(mp, waitfor);
1040 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1041 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1042 		    fs->e2fs->e2fs_first_dblock +
1043 		    1 /* superblock */ + i), fs->e2fs_bsize, 0, 0, 0);
1044 		e2fs_cgsave(&fs->e2fs_gd[
1045 		    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1046 		    (struct ext2_gd *)bp->b_data, fs->e2fs_bsize);
1047 		if (waitfor == MNT_WAIT)
1048 			error = bwrite(bp);
1049 		else
1050 			bawrite(bp);
1051 	}
1052 
1053 	if (!allerror && error)
1054 		allerror = error;
1055 	return (allerror);
1056 }
1057 /*
1058  * Return the root of a filesystem.
1059  */
1060 static int
1061 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1062 {
1063 	struct vnode *nvp;
1064 	int error;
1065 
1066 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1067 	if (error)
1068 		return (error);
1069 	*vpp = nvp;
1070 	return (0);
1071 }
1072