xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision f0157ce528a128e2abb181a5c766033a2ce49a5f)
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/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 			DROP_GIANT();
162 			g_topology_lock();
163 			g_access(ump->um_cp, 0, -1, 0);
164 			g_topology_unlock();
165 			PICKUP_GIANT();
166 		}
167 		if (!error && (mp->mnt_flag & MNT_RELOAD))
168 			error = ext2_reload(mp, td);
169 		if (error)
170 			return (error);
171 		devvp = ump->um_devvp;
172 		if (fs->e2fs_ronly && !vfs_flagopt(opts, "ro", NULL, 0)) {
173 			if (ext2_check_sb_compat(fs->e2fs, devvp->v_rdev, 0))
174 				return (EPERM);
175 
176 			/*
177 			 * If upgrade to read-write by non-root, then verify
178 			 * that user has necessary permissions on the device.
179 			 */
180 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
181 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
182 			    td->td_ucred, td);
183 			if (error)
184 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
185 			if (error) {
186 				VOP_UNLOCK(devvp, 0);
187 				return (error);
188 			}
189 			VOP_UNLOCK(devvp, 0);
190 			DROP_GIANT();
191 			g_topology_lock();
192 			error = g_access(ump->um_cp, 0, 1, 0);
193 			g_topology_unlock();
194 			PICKUP_GIANT();
195 			if (error)
196 				return (error);
197 
198 			if ((fs->e2fs->e2fs_state & E2FS_ISCLEAN) == 0 ||
199 			    (fs->e2fs->e2fs_state & E2FS_ERRORS)) {
200 				if (mp->mnt_flag & MNT_FORCE) {
201 					printf(
202 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
203 				} else {
204 					printf(
205 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
206 					    fs->e2fs_fsmnt);
207 					return (EPERM);
208 				}
209 			}
210 			fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;
211 			(void)ext2_cgupdate(ump, MNT_WAIT);
212 			fs->e2fs_ronly = 0;
213 			MNT_ILOCK(mp);
214 			mp->mnt_flag &= ~MNT_RDONLY;
215 			MNT_IUNLOCK(mp);
216 		}
217 		if (vfs_flagopt(opts, "export", NULL, 0)) {
218 			/* Process export requests in vfs_mount.c. */
219 			return (error);
220 		}
221 	}
222 
223 	/*
224 	 * Not an update, or updating the name: look up the name
225 	 * and verify that it refers to a sensible disk device.
226 	 */
227 	if (fspec == NULL)
228 		return (EINVAL);
229 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
230 	if ((error = namei(ndp)) != 0)
231 		return (error);
232 	NDFREE(ndp, NDF_ONLY_PNBUF);
233 	devvp = ndp->ni_vp;
234 
235 	if (!vn_isdisk(devvp, &error)) {
236 		vput(devvp);
237 		return (error);
238 	}
239 
240 	/*
241 	 * If mount by non-root, then verify that user has necessary
242 	 * permissions on the device.
243 	 *
244 	 * XXXRW: VOP_ACCESS() enough?
245 	 */
246 	accmode = VREAD;
247 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
248 		accmode |= VWRITE;
249 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
250 	if (error)
251 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
252 	if (error) {
253 		vput(devvp);
254 		return (error);
255 	}
256 
257 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
258 		error = ext2_mountfs(devvp, mp);
259 	} else {
260 		if (devvp != ump->um_devvp) {
261 			vput(devvp);
262 			return (EINVAL);	/* needs translation */
263 		} else
264 			vput(devvp);
265 	}
266 	if (error) {
267 		vrele(devvp);
268 		return (error);
269 	}
270 	ump = VFSTOEXT2(mp);
271 	fs = ump->um_e2fs;
272 
273 	/*
274 	 * Note that this strncpy() is ok because of a check at the start
275 	 * of ext2_mount().
276 	 */
277 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
278 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
279 	vfs_mountedfrom(mp, fspec);
280 	return (0);
281 }
282 
283 static int
284 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
285 {
286 
287 	if (es->e2fs_magic != E2FS_MAGIC) {
288 		printf("ext2fs: %s: wrong magic number %#x (expected %#x)\n",
289 		    devtoname(dev), es->e2fs_magic, E2FS_MAGIC);
290 		return (1);
291 	}
292 	if (es->e2fs_rev > E2FS_REV0) {
293 		if (es->e2fs_features_incompat & ~EXT2F_INCOMPAT_SUPP) {
294 			printf(
295 "WARNING: mount of %s denied due to unsupported optional features\n",
296 			    devtoname(dev));
297 			return (1);
298 		}
299 		if (!ronly &&
300 		    (es->e2fs_features_rocompat & ~EXT2F_ROCOMPAT_SUPP)) {
301 			printf("WARNING: R/W mount of %s denied due to "
302 			    "unsupported optional features\n", devtoname(dev));
303 			return (1);
304 		}
305 	}
306 	return (0);
307 }
308 
309 /*
310  * This computes the fields of the  ext2_sb_info structure from the
311  * data in the ext2_super_block structure read in.
312  */
313 static int
314 compute_sb_data(struct vnode *devvp, struct ext2fs *es,
315     struct m_ext2fs *fs)
316 {
317 	int db_count, error;
318 	int i;
319 	int logic_sb_block = 1;	/* XXX for now */
320 	struct buf *bp;
321 	uint32_t e2fs_descpb;
322 
323 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
324 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
325 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
326 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
327 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
328 	if (fs->e2fs_fsize)
329 		fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
330 	fs->e2fs_bpg = es->e2fs_bpg;
331 	fs->e2fs_fpg = es->e2fs_fpg;
332 	fs->e2fs_ipg = es->e2fs_ipg;
333 	if (es->e2fs_rev == E2FS_REV0) {
334 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE ;
335 	} else {
336 		fs->e2fs_isize = es->e2fs_inode_size;
337 
338 		/*
339 		 * Simple sanity check for superblock inode size value.
340 		 */
341 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
342 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
343 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
344 			printf("ext2fs: invalid inode size %d\n",
345 			    fs->e2fs_isize);
346 			return (EIO);
347 		}
348 	}
349 	/* Check for extra isize in big inodes. */
350 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
351 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
352 		printf("ext2fs: no space for extra inode timestamps\n");
353 		return (EINVAL);
354 	}
355 
356 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
357 	fs->e2fs_itpg = fs->e2fs_ipg /fs->e2fs_ipb;
358 	/* s_resuid / s_resgid ? */
359 	fs->e2fs_gcount = (es->e2fs_bcount - es->e2fs_first_dblock +
360 	    EXT2_BLOCKS_PER_GROUP(fs) - 1) / EXT2_BLOCKS_PER_GROUP(fs);
361 	e2fs_descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
362 	db_count = (fs->e2fs_gcount + e2fs_descpb - 1) / e2fs_descpb;
363 	fs->e2fs_gdbcount = db_count;
364 	fs->e2fs_gd = malloc(db_count * fs->e2fs_bsize,
365 	    M_EXT2MNT, M_WAITOK);
366 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
367 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK);
368 
369 	/*
370 	 * Adjust logic_sb_block.
371 	 * Godmar thinks: if the blocksize is greater than 1024, then
372 	 * the superblock is logically part of block zero.
373 	 */
374 	if(fs->e2fs_bsize > SBSIZE)
375 		logic_sb_block = 0;
376 	for (i = 0; i < db_count; i++) {
377 		error = bread(devvp ,
378 			 fsbtodb(fs, logic_sb_block + i + 1 ),
379 			fs->e2fs_bsize, NOCRED, &bp);
380 		if (error) {
381 			free(fs->e2fs_gd, M_EXT2MNT);
382 			brelse(bp);
383 			return (error);
384 		}
385 		e2fs_cgload((struct ext2_gd *)bp->b_data,
386 		    &fs->e2fs_gd[
387 			i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
388 		    fs->e2fs_bsize);
389 		brelse(bp);
390 		bp = NULL;
391 	}
392 	fs->e2fs_total_dir = 0;
393 	for (i=0; i < fs->e2fs_gcount; i++){
394 		fs->e2fs_total_dir += fs->e2fs_gd[i].ext2bgd_ndirs;
395 		fs->e2fs_contigdirs[i] = 0;
396 	}
397 	if (es->e2fs_rev == E2FS_REV0 ||
398 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
399 		fs->e2fs_maxfilesize = 0x7fffffff;
400 	else
401 		fs->e2fs_maxfilesize = 0x7fffffffffffffff;
402 	return (0);
403 }
404 
405 /*
406  * Reload all incore data for a filesystem (used after running fsck on
407  * the root filesystem and finding things to fix). The filesystem must
408  * be mounted read-only.
409  *
410  * Things to do to update the mount:
411  *	1) invalidate all cached meta-data.
412  *	2) re-read superblock from disk.
413  *	3) invalidate all cluster summary information.
414  *	4) invalidate all inactive vnodes.
415  *	5) invalidate all cached file data.
416  *	6) re-read inode data for all active vnodes.
417  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
418  */
419 static int
420 ext2_reload(struct mount *mp, struct thread *td)
421 {
422 	struct vnode *vp, *mvp, *devvp;
423 	struct inode *ip;
424 	struct buf *bp;
425 	struct ext2fs *es;
426 	struct m_ext2fs *fs;
427 	struct csum *sump;
428 	int error, i;
429 	int32_t *lp;
430 
431 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
432 		return (EINVAL);
433 	/*
434 	 * Step 1: invalidate all cached meta-data.
435 	 */
436 	devvp = VFSTOEXT2(mp)->um_devvp;
437 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
438 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
439 		panic("ext2_reload: dirty1");
440 	VOP_UNLOCK(devvp, 0);
441 
442 	/*
443 	 * Step 2: re-read superblock from disk.
444 	 * constants have been adjusted for ext2
445 	 */
446 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
447 		return (error);
448 	es = (struct ext2fs *)bp->b_data;
449 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
450 		brelse(bp);
451 		return (EIO);		/* XXX needs translation */
452 	}
453 	fs = VFSTOEXT2(mp)->um_e2fs;
454 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
455 
456 	if((error = compute_sb_data(devvp, es, fs)) != 0) {
457 		brelse(bp);
458 		return (error);
459 	}
460 #ifdef UNKLAR
461 	if (fs->fs_sbsize < SBSIZE)
462 		bp->b_flags |= B_INVAL;
463 #endif
464 	brelse(bp);
465 
466 	/*
467 	 * Step 3: invalidate all cluster summary information.
468 	 */
469 	if (fs->e2fs_contigsumsize > 0) {
470 		lp = fs->e2fs_maxcluster;
471 		sump = fs->e2fs_clustersum;
472 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
473 			*lp++ = fs->e2fs_contigsumsize;
474 			sump->cs_init = 0;
475 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
476 		}
477 	}
478 
479 loop:
480 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
481 		/*
482 		 * Step 4: invalidate all cached file data.
483 		 */
484 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
485 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
486 			goto loop;
487 		}
488 		if (vinvalbuf(vp, 0, 0, 0))
489 			panic("ext2_reload: dirty2");
490 
491 		/*
492 		 * Step 5: re-read inode data for all active vnodes.
493 		 */
494 		ip = VTOI(vp);
495 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
496 		    (int)fs->e2fs_bsize, NOCRED, &bp);
497 		if (error) {
498 			VOP_UNLOCK(vp, 0);
499 			vrele(vp);
500 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
501 			return (error);
502 		}
503 		ext2_ei2i((struct ext2fs_dinode *) ((char *)bp->b_data +
504 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
505 		brelse(bp);
506 		VOP_UNLOCK(vp, 0);
507 		vrele(vp);
508 	}
509 	return (0);
510 }
511 
512 /*
513  * Common code for mount and mountroot.
514  */
515 static int
516 ext2_mountfs(struct vnode *devvp, struct mount *mp)
517 {
518 	struct ext2mount *ump;
519 	struct buf *bp;
520 	struct m_ext2fs *fs;
521 	struct ext2fs *es;
522 	struct cdev *dev = devvp->v_rdev;
523 	struct g_consumer *cp;
524 	struct bufobj *bo;
525 	struct csum *sump;
526 	int error;
527 	int ronly;
528 	int i, size;
529 	int32_t *lp;
530 	int32_t e2fs_maxcontig;
531 
532 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
533 	/* XXX: use VOP_ACESS to check FS perms */
534 	DROP_GIANT();
535 	g_topology_lock();
536 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
537 	g_topology_unlock();
538 	PICKUP_GIANT();
539 	VOP_UNLOCK(devvp, 0);
540 	if (error)
541 		return (error);
542 
543 	/* XXX: should we check for some sectorsize or 512 instead? */
544 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
545 	    (SBSIZE < cp->provider->sectorsize)) {
546 		DROP_GIANT();
547 		g_topology_lock();
548 		g_vfs_close(cp);
549 		g_topology_unlock();
550 		PICKUP_GIANT();
551 		return (EINVAL);
552 	}
553 
554 	bo = &devvp->v_bufobj;
555 	bo->bo_private = cp;
556 	bo->bo_ops = g_vfs_bufops;
557 	if (devvp->v_rdev->si_iosize_max != 0)
558 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
559 	if (mp->mnt_iosize_max > MAXPHYS)
560 		mp->mnt_iosize_max = MAXPHYS;
561 
562 	bp = NULL;
563 	ump = NULL;
564 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
565 		goto out;
566 	es = (struct ext2fs *)bp->b_data;
567 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
568 		error = EINVAL;		/* XXX needs translation */
569 		goto out;
570 	}
571 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
572 	    (es->e2fs_state & E2FS_ERRORS)) {
573 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
574 			printf(
575 "WARNING: Filesystem was not properly dismounted\n");
576 		} else {
577 			printf(
578 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
579 			error = EPERM;
580 			goto out;
581 		}
582 	}
583 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
584 
585 	/*
586 	 * I don't know whether this is the right strategy. Note that
587 	 * we dynamically allocate both an ext2_sb_info and an ext2_super_block
588 	 * while Linux keeps the super block in a locked buffer.
589 	 */
590 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
591 		M_EXT2MNT, M_WAITOK);
592 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
593 		M_EXT2MNT, M_WAITOK);
594 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
595 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
596 	if ((error = compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
597 		goto out;
598 
599 	/*
600 	 * Calculate the maximum contiguous blocks and size of cluster summary
601 	 * array.  In FFS this is done by newfs; however, the superblock
602 	 * in ext2fs doesn't have these variables, so we can calculate
603 	 * them here.
604 	 */
605 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
606 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
607 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
608 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
609 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
610 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
611 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
612 		lp = ump->um_e2fs->e2fs_maxcluster;
613 		sump = ump->um_e2fs->e2fs_clustersum;
614 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
615 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
616 			sump->cs_init = 0;
617 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
618 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
619 		}
620 	}
621 
622 	brelse(bp);
623 	bp = NULL;
624 	fs = ump->um_e2fs;
625 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
626 
627 	/*
628 	 * If the fs is not mounted read-only, make sure the super block is
629 	 * always written back on a sync().
630 	 */
631 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
632 	if (ronly == 0) {
633 		fs->e2fs_fmod = 1;		/* mark it modified */
634 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
635 	}
636 	mp->mnt_data = ump;
637 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
638 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
639 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
640 	MNT_ILOCK(mp);
641 	mp->mnt_flag |= MNT_LOCAL;
642 	MNT_IUNLOCK(mp);
643 	ump->um_mountp = mp;
644 	ump->um_dev = dev;
645 	ump->um_devvp = devvp;
646 	ump->um_bo = &devvp->v_bufobj;
647 	ump->um_cp = cp;
648 
649 	/*
650 	 * Setting those two parameters allowed us to use
651 	 * ufs_bmap w/o changse!
652 	 */
653 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
654 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
655 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
656 	if (ronly == 0)
657 		ext2_sbupdate(ump, MNT_WAIT);
658 	/*
659 	 * Initialize filesystem stat information in mount struct.
660 	 */
661 	MNT_ILOCK(mp);
662  	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED;
663 	MNT_IUNLOCK(mp);
664 	return (0);
665 out:
666 	if (bp)
667 		brelse(bp);
668 	if (cp != NULL) {
669 		DROP_GIANT();
670 		g_topology_lock();
671 		g_vfs_close(cp);
672 		g_topology_unlock();
673 		PICKUP_GIANT();
674 	}
675 	if (ump) {
676 	  	mtx_destroy(EXT2_MTX(ump));
677 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
678 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
679 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
680 		free(ump->um_e2fs, M_EXT2MNT);
681 		free(ump, M_EXT2MNT);
682 		mp->mnt_data = NULL;
683 	}
684 	return (error);
685 }
686 
687 /*
688  * Unmount system call.
689  */
690 static int
691 ext2_unmount(struct mount *mp, int mntflags)
692 {
693 	struct ext2mount *ump;
694 	struct m_ext2fs *fs;
695 	struct csum *sump;
696 	int error, flags, i, ronly;
697 
698 	flags = 0;
699 	if (mntflags & MNT_FORCE) {
700 		if (mp->mnt_flag & MNT_ROOTFS)
701 			return (EINVAL);
702 		flags |= FORCECLOSE;
703 	}
704 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
705 		return (error);
706 	ump = VFSTOEXT2(mp);
707 	fs = ump->um_e2fs;
708 	ronly = fs->e2fs_ronly;
709 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
710 		if (fs->e2fs_wasvalid)
711  			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
712  		ext2_sbupdate(ump, MNT_WAIT);
713 	}
714 
715 	DROP_GIANT();
716 	g_topology_lock();
717 	g_vfs_close(ump->um_cp);
718 	g_topology_unlock();
719 	PICKUP_GIANT();
720 	vrele(ump->um_devvp);
721 	sump = fs->e2fs_clustersum;
722 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
723 		free(sump->cs_sum, M_EXT2MNT);
724 	free(fs->e2fs_clustersum, M_EXT2MNT);
725 	free(fs->e2fs_maxcluster, M_EXT2MNT);
726 	free(fs->e2fs_gd, M_EXT2MNT);
727 	free(fs->e2fs_contigdirs, M_EXT2MNT);
728 	free(fs->e2fs, M_EXT2MNT);
729 	free(fs, M_EXT2MNT);
730 	free(ump, M_EXT2MNT);
731 	mp->mnt_data = NULL;
732 	MNT_ILOCK(mp);
733 	mp->mnt_flag &= ~MNT_LOCAL;
734 	MNT_IUNLOCK(mp);
735 	return (error);
736 }
737 
738 /*
739  * Flush out all the files in a filesystem.
740  */
741 static int
742 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
743 {
744 	int error;
745 
746 	error = vflush(mp, 0, flags, td);
747 	return (error);
748 }
749 /*
750  * Get filesystem statistics.
751  */
752 int
753 ext2_statfs(struct mount *mp, struct statfs *sbp)
754 {
755 	struct ext2mount *ump;
756 	struct m_ext2fs *fs;
757 	uint32_t overhead, overhead_per_group, ngdb;
758 	int i, ngroups;
759 
760 	ump = VFSTOEXT2(mp);
761 	fs = ump->um_e2fs;
762 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
763 		panic("ext2_statfs");
764 
765 	/*
766 	 * Compute the overhead (FS structures)
767 	 */
768 	overhead_per_group =
769 	    1 /* block bitmap */ +
770 	    1 /* inode bitmap */ +
771 	    fs->e2fs_itpg;
772 	overhead = fs->e2fs->e2fs_first_dblock +
773 	    fs->e2fs_gcount * overhead_per_group;
774 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
775 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
776 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
777 			if (cg_has_sb(i))
778 				ngroups++;
779 		}
780 	} else {
781 		ngroups = fs->e2fs_gcount;
782 	}
783 	ngdb = fs->e2fs_gdbcount;
784 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
785 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
786 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
787 	overhead += ngroups * (1 /* superblock */ + ngdb);
788 
789 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
790 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
791 	sbp->f_blocks = fs->e2fs->e2fs_bcount - overhead;
792 	sbp->f_bfree = fs->e2fs->e2fs_fbcount;
793 	sbp->f_bavail = sbp->f_bfree - fs->e2fs->e2fs_rbcount;
794 	sbp->f_files = fs->e2fs->e2fs_icount;
795 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
796 	return (0);
797 }
798 
799 /*
800  * Go through the disk queues to initiate sandbagged IO;
801  * go through the inodes to write those that have been modified;
802  * initiate the writing of the super block if it has been modified.
803  *
804  * Note: we are always called with the filesystem marked `MPBUSY'.
805  */
806 static int
807 ext2_sync(struct mount *mp, int waitfor)
808 {
809 	struct vnode *mvp, *vp;
810 	struct thread *td;
811 	struct inode *ip;
812 	struct ext2mount *ump = VFSTOEXT2(mp);
813 	struct m_ext2fs *fs;
814 	int error, allerror = 0;
815 
816 	td = curthread;
817 	fs = ump->um_e2fs;
818 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
819 		printf("fs = %s\n", fs->e2fs_fsmnt);
820 		panic("ext2_sync: rofs mod");
821 	}
822 
823 	/*
824 	 * Write back each (modified) inode.
825 	 */
826 loop:
827 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
828 		if (vp->v_type == VNON) {
829 			VI_UNLOCK(vp);
830 			continue;
831 		}
832 		ip = VTOI(vp);
833 		if ((ip->i_flag &
834 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
835 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
836 		    waitfor == MNT_LAZY)) {
837 			VI_UNLOCK(vp);
838 			continue;
839 		}
840 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
841 		if (error) {
842 			if (error == ENOENT) {
843 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
844 				goto loop;
845 			}
846 			continue;
847 		}
848 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
849 			allerror = error;
850 		VOP_UNLOCK(vp, 0);
851 		vrele(vp);
852 	}
853 
854 	/*
855 	 * Force stale filesystem control information to be flushed.
856 	 */
857 	if (waitfor != MNT_LAZY) {
858 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
859 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
860 			allerror = error;
861 		VOP_UNLOCK(ump->um_devvp, 0);
862 	}
863 
864 	/*
865 	 * Write back modified superblock.
866 	 */
867 	if (fs->e2fs_fmod != 0) {
868 		fs->e2fs_fmod = 0;
869 		fs->e2fs->e2fs_wtime = time_second;
870 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
871 			allerror = error;
872 	}
873 	return (allerror);
874 }
875 
876 /*
877  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
878  * in from disk.  If it is in core, wait for the lock bit to clear, then
879  * return the inode locked.  Detection and handling of mount points must be
880  * done by the calling routine.
881  */
882 static int
883 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
884 {
885 	struct m_ext2fs *fs;
886 	struct inode *ip;
887 	struct ext2mount *ump;
888 	struct buf *bp;
889 	struct vnode *vp;
890 	struct cdev *dev;
891 	struct thread *td;
892 	int i, error;
893 	int used_blocks;
894 
895 	td = curthread;
896 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
897 	if (error || *vpp != NULL)
898 		return (error);
899 
900 	ump = VFSTOEXT2(mp);
901 	dev = ump->um_dev;
902 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
903 
904 	/* Allocate a new vnode/inode. */
905 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
906 		*vpp = NULL;
907 		free(ip, M_EXT2NODE);
908 		return (error);
909 	}
910 	vp->v_data = ip;
911 	ip->i_vnode = vp;
912 	ip->i_e2fs = fs = ump->um_e2fs;
913 	ip->i_ump  = ump;
914 	ip->i_number = ino;
915 
916 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
917 	error = insmntque(vp, mp);
918 	if (error != 0) {
919 		free(ip, M_EXT2NODE);
920 		*vpp = NULL;
921 		return (error);
922 	}
923 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
924 	if (error || *vpp != NULL)
925 		return (error);
926 
927 	/* Read in the disk contents for the inode, copy into the inode. */
928 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
929 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
930 		/*
931 		 * The inode does not contain anything useful, so it would
932 		 * be misleading to leave it on its hash chain. With mode
933 		 * still zero, it will be unlinked and returned to the free
934 		 * list by vput().
935 		 */
936 		brelse(bp);
937 		vput(vp);
938 		*vpp = NULL;
939 		return (error);
940 	}
941 	/* convert ext2 inode to dinode */
942 	ext2_ei2i((struct ext2fs_dinode *) ((char *)bp->b_data + EXT2_INODE_SIZE(fs) *
943 			ino_to_fsbo(fs, ino)), ip);
944 	ip->i_block_group = ino_to_cg(fs, ino);
945 	ip->i_next_alloc_block = 0;
946 	ip->i_next_alloc_goal = 0;
947 
948 	/*
949 	 * Now we want to make sure that block pointers for unused
950 	 * blocks are zeroed out - ext2_balloc depends on this
951 	 * although for regular files and directories only
952 	 */
953 	if(S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode)) {
954 		used_blocks = (ip->i_size+fs->e2fs_bsize-1) / fs->e2fs_bsize;
955 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
956 			ip->i_db[i] = 0;
957 	}
958 /*
959 	ext2_print_inode(ip);
960 */
961 	bqrelse(bp);
962 
963 	/*
964 	 * Initialize the vnode from the inode, check for aliases.
965 	 * Note that the underlying vnode may have changed.
966 	 */
967 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
968 		vput(vp);
969 		*vpp = NULL;
970 		return (error);
971 	}
972 
973 	/*
974 	 * Finish inode initialization.
975 	 */
976 
977 	/*
978 	 * Set up a generation number for this inode if it does not
979 	 * already have one. This should only happen on old filesystems.
980 	 */
981 	if (ip->i_gen == 0) {
982 		ip->i_gen = random() + 1;
983 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
984 			ip->i_flag |= IN_MODIFIED;
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  * Return the root of a filesystem.
1084  */
1085 static int
1086 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1087 {
1088 	struct vnode *nvp;
1089 	int error;
1090 
1091 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1092 	if (error)
1093 		return (error);
1094 	*vpp = nvp;
1095 	return (0);
1096 }
1097