xref: /freebsd/sys/fs/ext2fs/ext2_vfsops.c (revision d06955f9bdb1416d9196043ed781f9b36dae9adc)
1 /*-
2  *  modified for EXT2FS support in Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
8  * SPDX-License-Identifier: BSD-3-Clause
9  *
10  * Copyright (c) 1989, 1991, 1993, 1994
11  *	The Regents of the University of California.  All rights reserved.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)ffs_vfsops.c	8.8 (Berkeley) 4/18/94
38  * $FreeBSD$
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/namei.h>
44 #include <sys/priv.h>
45 #include <sys/proc.h>
46 #include <sys/kernel.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/endian.h>
53 #include <sys/fcntl.h>
54 #include <sys/malloc.h>
55 #include <sys/stat.h>
56 #include <sys/mutex.h>
57 
58 #include <geom/geom.h>
59 #include <geom/geom_vfs.h>
60 
61 #include <fs/ext2fs/ext2_mount.h>
62 #include <fs/ext2fs/inode.h>
63 
64 #include <fs/ext2fs/fs.h>
65 #include <fs/ext2fs/ext2fs.h>
66 #include <fs/ext2fs/ext2_dinode.h>
67 #include <fs/ext2fs/ext2_extern.h>
68 
69 static int	ext2_flushfiles(struct mount *mp, int flags, struct thread *td);
70 static int	ext2_mountfs(struct vnode *, struct mount *);
71 static int	ext2_reload(struct mount *mp, struct thread *td);
72 static int	ext2_sbupdate(struct ext2mount *, int);
73 static int	ext2_cgupdate(struct ext2mount *, int);
74 static vfs_unmount_t		ext2_unmount;
75 static vfs_root_t		ext2_root;
76 static vfs_statfs_t		ext2_statfs;
77 static vfs_sync_t		ext2_sync;
78 static vfs_vget_t		ext2_vget;
79 static vfs_fhtovp_t		ext2_fhtovp;
80 static vfs_mount_t		ext2_mount;
81 
82 MALLOC_DEFINE(M_EXT2NODE, "ext2_node", "EXT2 vnode private part");
83 static MALLOC_DEFINE(M_EXT2MNT, "ext2_mount", "EXT2 mount structure");
84 
85 static struct vfsops ext2fs_vfsops = {
86 	.vfs_fhtovp =		ext2_fhtovp,
87 	.vfs_mount =		ext2_mount,
88 	.vfs_root =		ext2_root,	/* root inode via vget */
89 	.vfs_statfs =		ext2_statfs,
90 	.vfs_sync =		ext2_sync,
91 	.vfs_unmount =		ext2_unmount,
92 	.vfs_vget =		ext2_vget,
93 };
94 
95 VFS_SET(ext2fs_vfsops, ext2fs, 0);
96 
97 static int	ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev,
98 		    int ronly);
99 static int	compute_sb_data(struct vnode * devvp,
100 		    struct ext2fs * es, struct m_ext2fs * fs);
101 
102 static const char *ext2_opts[] = { "acls", "async", "noatime", "noclusterr",
103     "noclusterw", "noexec", "export", "force", "from", "multilabel",
104     "suiddir", "nosymfollow", "sync", "union", NULL };
105 
106 /*
107  * VFS Operations.
108  *
109  * mount system call
110  */
111 static int
112 ext2_mount(struct mount *mp)
113 {
114 	struct vfsoptlist *opts;
115 	struct vnode *devvp;
116 	struct thread *td;
117 	struct ext2mount *ump = NULL;
118 	struct m_ext2fs *fs;
119 	struct nameidata nd, *ndp = &nd;
120 	accmode_t accmode;
121 	char *path, *fspec;
122 	int error, flags, len;
123 
124 	td = curthread;
125 	opts = mp->mnt_optnew;
126 
127 	if (vfs_filteropt(opts, ext2_opts))
128 		return (EINVAL);
129 
130 	vfs_getopt(opts, "fspath", (void **)&path, NULL);
131 	/* Double-check the length of path.. */
132 	if (strlen(path) >= MAXMNTLEN)
133 		return (ENAMETOOLONG);
134 
135 	fspec = NULL;
136 	error = vfs_getopt(opts, "from", (void **)&fspec, &len);
137 	if (!error && fspec[len - 1] != '\0')
138 		return (EINVAL);
139 
140 	/*
141 	 * If updating, check whether changing from read-only to
142 	 * read/write; if there is no device name, that's all we do.
143 	 */
144 	if (mp->mnt_flag & MNT_UPDATE) {
145 		ump = VFSTOEXT2(mp);
146 		fs = ump->um_e2fs;
147 		error = 0;
148 		if (fs->e2fs_ronly == 0 &&
149 		    vfs_flagopt(opts, "ro", NULL, 0)) {
150 			error = VFS_SYNC(mp, MNT_WAIT);
151 			if (error)
152 				return (error);
153 			flags = WRITECLOSE;
154 			if (mp->mnt_flag & MNT_FORCE)
155 				flags |= FORCECLOSE;
156 			error = ext2_flushfiles(mp, flags, td);
157 			if (error == 0 && fs->e2fs_wasvalid && ext2_cgupdate(ump, MNT_WAIT) == 0) {
158 				fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
159 				ext2_sbupdate(ump, MNT_WAIT);
160 			}
161 			fs->e2fs_ronly = 1;
162 			vfs_flagopt(opts, "ro", &mp->mnt_flag, MNT_RDONLY);
163 			g_topology_lock();
164 			g_access(ump->um_cp, 0, -1, 0);
165 			g_topology_unlock();
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 			g_topology_lock();
191 			error = g_access(ump->um_cp, 0, 1, 0);
192 			g_topology_unlock();
193 			if (error)
194 				return (error);
195 
196 			if ((fs->e2fs->e2fs_state & E2FS_ISCLEAN) == 0 ||
197 			    (fs->e2fs->e2fs_state & E2FS_ERRORS)) {
198 				if (mp->mnt_flag & MNT_FORCE) {
199 					printf(
200 "WARNING: %s was not properly dismounted\n", fs->e2fs_fsmnt);
201 				} else {
202 					printf(
203 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
204 					    fs->e2fs_fsmnt);
205 					return (EPERM);
206 				}
207 			}
208 			fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;
209 			(void)ext2_cgupdate(ump, MNT_WAIT);
210 			fs->e2fs_ronly = 0;
211 			MNT_ILOCK(mp);
212 			mp->mnt_flag &= ~MNT_RDONLY;
213 			MNT_IUNLOCK(mp);
214 		}
215 		if (vfs_flagopt(opts, "export", NULL, 0)) {
216 			/* Process export requests in vfs_mount.c. */
217 			return (error);
218 		}
219 	}
220 
221 	/*
222 	 * Not an update, or updating the name: look up the name
223 	 * and verify that it refers to a sensible disk device.
224 	 */
225 	if (fspec == NULL)
226 		return (EINVAL);
227 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
228 	if ((error = namei(ndp)) != 0)
229 		return (error);
230 	NDFREE(ndp, NDF_ONLY_PNBUF);
231 	devvp = ndp->ni_vp;
232 
233 	if (!vn_isdisk(devvp, &error)) {
234 		vput(devvp);
235 		return (error);
236 	}
237 
238 	/*
239 	 * If mount by non-root, then verify that user has necessary
240 	 * permissions on the device.
241 	 *
242 	 * XXXRW: VOP_ACCESS() enough?
243 	 */
244 	accmode = VREAD;
245 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
246 		accmode |= VWRITE;
247 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
248 	if (error)
249 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
250 	if (error) {
251 		vput(devvp);
252 		return (error);
253 	}
254 
255 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
256 		error = ext2_mountfs(devvp, mp);
257 	} else {
258 		if (devvp != ump->um_devvp) {
259 			vput(devvp);
260 			return (EINVAL);	/* needs translation */
261 		} else
262 			vput(devvp);
263 	}
264 	if (error) {
265 		vrele(devvp);
266 		return (error);
267 	}
268 	ump = VFSTOEXT2(mp);
269 	fs = ump->um_e2fs;
270 
271 	/*
272 	 * Note that this strncpy() is ok because of a check at the start
273 	 * of ext2_mount().
274 	 */
275 	strncpy(fs->e2fs_fsmnt, path, MAXMNTLEN);
276 	fs->e2fs_fsmnt[MAXMNTLEN - 1] = '\0';
277 	vfs_mountedfrom(mp, fspec);
278 	return (0);
279 }
280 
281 static int
282 ext2_check_sb_compat(struct ext2fs *es, struct cdev *dev, int ronly)
283 {
284 	uint32_t i, mask;
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 		mask = es->e2fs_features_incompat & ~(EXT2F_INCOMPAT_SUPP |
293 		    EXT4F_RO_INCOMPAT_SUPP);
294 		if (mask) {
295 			printf("WARNING: mount of %s denied due to "
296 			    "unsupported optional features:\n", devtoname(dev));
297 			for (i = 0;
298 			    i < sizeof(incompat)/sizeof(struct ext2_feature);
299 			    i++)
300 				if (mask & incompat[i].mask)
301 					printf("%s ", incompat[i].name);
302 			printf("\n");
303 			return (1);
304 		}
305 		mask = es->e2fs_features_rocompat & ~EXT2F_ROCOMPAT_SUPP;
306 		if (!ronly && mask) {
307 			printf("WARNING: R/W mount of %s denied due to "
308 			    "unsupported optional features:\n", devtoname(dev));
309 			for (i = 0;
310 			    i < sizeof(ro_compat)/sizeof(struct ext2_feature);
311 			    i++)
312 				if (mask & ro_compat[i].mask)
313 					printf("%s ", ro_compat[i].name);
314 			printf("\n");
315 			return (1);
316 		}
317 	}
318 	return (0);
319 }
320 
321 /*
322  * This computes the fields of the m_ext2fs structure from the
323  * data in the ext2fs structure read in.
324  */
325 static int
326 compute_sb_data(struct vnode *devvp, struct ext2fs *es,
327     struct m_ext2fs *fs)
328 {
329 	int db_count, error;
330 	int i;
331 	int logic_sb_block = 1;	/* XXX for now */
332 	struct buf *bp;
333 	uint32_t e2fs_descpb;
334 
335 	fs->e2fs_bshift = EXT2_MIN_BLOCK_LOG_SIZE + es->e2fs_log_bsize;
336 	fs->e2fs_bsize = 1U << fs->e2fs_bshift;
337 	fs->e2fs_fsbtodb = es->e2fs_log_bsize + 1;
338 	fs->e2fs_qbmask = fs->e2fs_bsize - 1;
339 	fs->e2fs_fsize = EXT2_MIN_FRAG_SIZE << es->e2fs_log_fsize;
340 	if (fs->e2fs_fsize)
341 		fs->e2fs_fpb = fs->e2fs_bsize / fs->e2fs_fsize;
342 	fs->e2fs_bpg = es->e2fs_bpg;
343 	fs->e2fs_fpg = es->e2fs_fpg;
344 	fs->e2fs_ipg = es->e2fs_ipg;
345 	if (es->e2fs_rev == E2FS_REV0) {
346 		fs->e2fs_isize = E2FS_REV0_INODE_SIZE;
347 	} else {
348 		fs->e2fs_isize = es->e2fs_inode_size;
349 
350 		/*
351 		 * Simple sanity check for superblock inode size value.
352 		 */
353 		if (EXT2_INODE_SIZE(fs) < E2FS_REV0_INODE_SIZE ||
354 		    EXT2_INODE_SIZE(fs) > fs->e2fs_bsize ||
355 		    (fs->e2fs_isize & (fs->e2fs_isize - 1)) != 0) {
356 			printf("ext2fs: invalid inode size %d\n",
357 			    fs->e2fs_isize);
358 			return (EIO);
359 		}
360 	}
361 	/* Check for extra isize in big inodes. */
362 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_EXTRA_ISIZE) &&
363 	    EXT2_INODE_SIZE(fs) < sizeof(struct ext2fs_dinode)) {
364 		printf("ext2fs: no space for extra inode timestamps\n");
365 		return (EINVAL);
366 	}
367 	/* Check for group descriptor size */
368 	if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_64BIT) &&
369 	    (es->e3fs_desc_size != sizeof(struct ext2_gd))) {
370 		printf("ext2fs: group descriptor size unsupported %d\n",
371 		    es->e3fs_desc_size);
372 		return (EINVAL);
373 	}
374 
375 	fs->e2fs_ipb = fs->e2fs_bsize / EXT2_INODE_SIZE(fs);
376 	fs->e2fs_itpg = fs->e2fs_ipg / fs->e2fs_ipb;
377 	/* s_resuid / s_resgid ? */
378 	fs->e2fs_gcount = howmany(es->e2fs_bcount - es->e2fs_first_dblock,
379 	    EXT2_BLOCKS_PER_GROUP(fs));
380 	e2fs_descpb = fs->e2fs_bsize / sizeof(struct ext2_gd);
381 	db_count = howmany(fs->e2fs_gcount, e2fs_descpb);
382 	fs->e2fs_gdbcount = db_count;
383 	fs->e2fs_gd = malloc(db_count * fs->e2fs_bsize,
384 	    M_EXT2MNT, M_WAITOK);
385 	fs->e2fs_contigdirs = malloc(fs->e2fs_gcount *
386 	    sizeof(*fs->e2fs_contigdirs), M_EXT2MNT, M_WAITOK | M_ZERO);
387 
388 	/*
389 	 * Adjust logic_sb_block.
390 	 * Godmar thinks: if the blocksize is greater than 1024, then
391 	 * the superblock is logically part of block zero.
392 	 */
393 	if (fs->e2fs_bsize > SBSIZE)
394 		logic_sb_block = 0;
395 	for (i = 0; i < db_count; i++) {
396 		error = bread(devvp,
397 		    fsbtodb(fs, logic_sb_block + i + 1),
398 		    fs->e2fs_bsize, NOCRED, &bp);
399 		if (error) {
400 			free(fs->e2fs_contigdirs, M_EXT2MNT);
401 			free(fs->e2fs_gd, M_EXT2MNT);
402 			brelse(bp);
403 			return (error);
404 		}
405 		e2fs_cgload((struct ext2_gd *)bp->b_data,
406 		    &fs->e2fs_gd[
407 		    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
408 		    fs->e2fs_bsize);
409 		brelse(bp);
410 		bp = NULL;
411 	}
412 	/* Verify cg csum */
413 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM)) {
414 		error = ext2_gd_csum_verify(fs, devvp->v_rdev);
415 		if (error)
416 			return (error);
417 	}
418 	/* Initialization for the ext2 Orlov allocator variant. */
419 	fs->e2fs_total_dir = 0;
420 	for (i = 0; i < fs->e2fs_gcount; i++)
421 		fs->e2fs_total_dir += fs->e2fs_gd[i].ext2bgd_ndirs;
422 
423 	if (es->e2fs_rev == E2FS_REV0 ||
424 	    !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_LARGEFILE))
425 		fs->e2fs_maxfilesize = 0x7fffffff;
426 	else {
427 		fs->e2fs_maxfilesize = 0xffffffffffff;
428 		if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_HUGE_FILE))
429 			fs->e2fs_maxfilesize = 0x7fffffffffffffff;
430 	}
431 	if (es->e4fs_flags & E2FS_UNSIGNED_HASH) {
432 		fs->e2fs_uhash = 3;
433 	} else if ((es->e4fs_flags & E2FS_SIGNED_HASH) == 0) {
434 #ifdef __CHAR_UNSIGNED__
435 		es->e4fs_flags |= E2FS_UNSIGNED_HASH;
436 		fs->e2fs_uhash = 3;
437 #else
438 		es->e4fs_flags |= E2FS_SIGNED_HASH;
439 #endif
440 	}
441 
442 	return (0);
443 }
444 
445 /*
446  * Reload all incore data for a filesystem (used after running fsck on
447  * the root filesystem and finding things to fix). The filesystem must
448  * be mounted read-only.
449  *
450  * Things to do to update the mount:
451  *	1) invalidate all cached meta-data.
452  *	2) re-read superblock from disk.
453  *	3) invalidate all cluster summary information.
454  *	4) invalidate all inactive vnodes.
455  *	5) invalidate all cached file data.
456  *	6) re-read inode data for all active vnodes.
457  * XXX we are missing some steps, in particular # 3, this has to be reviewed.
458  */
459 static int
460 ext2_reload(struct mount *mp, struct thread *td)
461 {
462 	struct vnode *vp, *mvp, *devvp;
463 	struct inode *ip;
464 	struct buf *bp;
465 	struct ext2fs *es;
466 	struct m_ext2fs *fs;
467 	struct csum *sump;
468 	int error, i;
469 	int32_t *lp;
470 
471 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
472 		return (EINVAL);
473 	/*
474 	 * Step 1: invalidate all cached meta-data.
475 	 */
476 	devvp = VFSTOEXT2(mp)->um_devvp;
477 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
478 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
479 		panic("ext2_reload: dirty1");
480 	VOP_UNLOCK(devvp, 0);
481 
482 	/*
483 	 * Step 2: re-read superblock from disk.
484 	 * constants have been adjusted for ext2
485 	 */
486 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
487 		return (error);
488 	es = (struct ext2fs *)bp->b_data;
489 	if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
490 		brelse(bp);
491 		return (EIO);		/* XXX needs translation */
492 	}
493 	fs = VFSTOEXT2(mp)->um_e2fs;
494 	bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
495 
496 	if ((error = compute_sb_data(devvp, es, fs)) != 0) {
497 		brelse(bp);
498 		return (error);
499 	}
500 #ifdef UNKLAR
501 	if (fs->fs_sbsize < SBSIZE)
502 		bp->b_flags |= B_INVAL;
503 #endif
504 	brelse(bp);
505 
506 	/*
507 	 * Step 3: invalidate all cluster summary information.
508 	 */
509 	if (fs->e2fs_contigsumsize > 0) {
510 		lp = fs->e2fs_maxcluster;
511 		sump = fs->e2fs_clustersum;
512 		for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
513 			*lp++ = fs->e2fs_contigsumsize;
514 			sump->cs_init = 0;
515 			bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
516 		}
517 	}
518 
519 loop:
520 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
521 		/*
522 		 * Step 4: invalidate all cached file data.
523 		 */
524 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
525 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
526 			goto loop;
527 		}
528 		if (vinvalbuf(vp, 0, 0, 0))
529 			panic("ext2_reload: dirty2");
530 
531 		/*
532 		 * Step 5: re-read inode data for all active vnodes.
533 		 */
534 		ip = VTOI(vp);
535 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
536 		    (int)fs->e2fs_bsize, NOCRED, &bp);
537 		if (error) {
538 			VOP_UNLOCK(vp, 0);
539 			vrele(vp);
540 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
541 			return (error);
542 		}
543 		ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data +
544 		    EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
545 		brelse(bp);
546 		VOP_UNLOCK(vp, 0);
547 		vrele(vp);
548 	}
549 	return (0);
550 }
551 
552 /*
553  * Common code for mount and mountroot.
554  */
555 static int
556 ext2_mountfs(struct vnode *devvp, struct mount *mp)
557 {
558 	struct ext2mount *ump;
559 	struct buf *bp;
560 	struct m_ext2fs *fs;
561 	struct ext2fs *es;
562 	struct cdev *dev = devvp->v_rdev;
563 	struct g_consumer *cp;
564 	struct bufobj *bo;
565 	struct csum *sump;
566 	int error;
567 	int ronly;
568 	int i;
569 	u_long size;
570 	int32_t *lp;
571 	int32_t e2fs_maxcontig;
572 
573 	ronly = vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0);
574 	/* XXX: use VOP_ACESS to check FS perms */
575 	g_topology_lock();
576 	error = g_vfs_open(devvp, &cp, "ext2fs", ronly ? 0 : 1);
577 	g_topology_unlock();
578 	VOP_UNLOCK(devvp, 0);
579 	if (error)
580 		return (error);
581 
582 	/* XXX: should we check for some sectorsize or 512 instead? */
583 	if (((SBSIZE % cp->provider->sectorsize) != 0) ||
584 	    (SBSIZE < cp->provider->sectorsize)) {
585 		g_topology_lock();
586 		g_vfs_close(cp);
587 		g_topology_unlock();
588 		return (EINVAL);
589 	}
590 
591 	bo = &devvp->v_bufobj;
592 	bo->bo_private = cp;
593 	bo->bo_ops = g_vfs_bufops;
594 	if (devvp->v_rdev->si_iosize_max != 0)
595 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
596 	if (mp->mnt_iosize_max > MAXPHYS)
597 		mp->mnt_iosize_max = MAXPHYS;
598 
599 	bp = NULL;
600 	ump = NULL;
601 	if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
602 		goto out;
603 	es = (struct ext2fs *)bp->b_data;
604 	if (ext2_check_sb_compat(es, dev, ronly) != 0) {
605 		error = EINVAL;		/* XXX needs translation */
606 		goto out;
607 	}
608 	if ((es->e2fs_state & E2FS_ISCLEAN) == 0 ||
609 	    (es->e2fs_state & E2FS_ERRORS)) {
610 		if (ronly || (mp->mnt_flag & MNT_FORCE)) {
611 			printf(
612 "WARNING: Filesystem was not properly dismounted\n");
613 		} else {
614 			printf(
615 "WARNING: R/W mount denied.  Filesystem is not clean - run fsck\n");
616 			error = EPERM;
617 			goto out;
618 		}
619 	}
620 	ump = malloc(sizeof(*ump), M_EXT2MNT, M_WAITOK | M_ZERO);
621 
622 	/*
623 	 * I don't know whether this is the right strategy. Note that
624 	 * we dynamically allocate both an m_ext2fs and an ext2fs
625 	 * while Linux keeps the super block in a locked buffer.
626 	 */
627 	ump->um_e2fs = malloc(sizeof(struct m_ext2fs),
628 	    M_EXT2MNT, M_WAITOK | M_ZERO);
629 	ump->um_e2fs->e2fs = malloc(sizeof(struct ext2fs),
630 	    M_EXT2MNT, M_WAITOK);
631 	mtx_init(EXT2_MTX(ump), "EXT2FS", "EXT2FS Lock", MTX_DEF);
632 	bcopy(es, ump->um_e2fs->e2fs, (u_int)sizeof(struct ext2fs));
633 	if ((error = compute_sb_data(devvp, ump->um_e2fs->e2fs, ump->um_e2fs)))
634 		goto out;
635 
636 	/*
637 	 * Calculate the maximum contiguous blocks and size of cluster summary
638 	 * array.  In FFS this is done by newfs; however, the superblock
639 	 * in ext2fs doesn't have these variables, so we can calculate
640 	 * them here.
641 	 */
642 	e2fs_maxcontig = MAX(1, MAXPHYS / ump->um_e2fs->e2fs_bsize);
643 	ump->um_e2fs->e2fs_contigsumsize = MIN(e2fs_maxcontig, EXT2_MAXCONTIG);
644 	if (ump->um_e2fs->e2fs_contigsumsize > 0) {
645 		size = ump->um_e2fs->e2fs_gcount * sizeof(int32_t);
646 		ump->um_e2fs->e2fs_maxcluster = malloc(size, M_EXT2MNT, M_WAITOK);
647 		size = ump->um_e2fs->e2fs_gcount * sizeof(struct csum);
648 		ump->um_e2fs->e2fs_clustersum = malloc(size, M_EXT2MNT, M_WAITOK);
649 		lp = ump->um_e2fs->e2fs_maxcluster;
650 		sump = ump->um_e2fs->e2fs_clustersum;
651 		for (i = 0; i < ump->um_e2fs->e2fs_gcount; i++, sump++) {
652 			*lp++ = ump->um_e2fs->e2fs_contigsumsize;
653 			sump->cs_init = 0;
654 			sump->cs_sum = malloc((ump->um_e2fs->e2fs_contigsumsize + 1) *
655 			    sizeof(int32_t), M_EXT2MNT, M_WAITOK | M_ZERO);
656 		}
657 	}
658 
659 	brelse(bp);
660 	bp = NULL;
661 	fs = ump->um_e2fs;
662 	fs->e2fs_ronly = ronly;	/* ronly is set according to mnt_flags */
663 
664 	/*
665 	 * If the fs is not mounted read-only, make sure the super block is
666 	 * always written back on a sync().
667 	 */
668 	fs->e2fs_wasvalid = fs->e2fs->e2fs_state & E2FS_ISCLEAN ? 1 : 0;
669 	if (ronly == 0) {
670 		fs->e2fs_fmod = 1;	/* mark it modified */
671 		fs->e2fs->e2fs_state &= ~E2FS_ISCLEAN;	/* set fs invalid */
672 	}
673 	mp->mnt_data = ump;
674 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
675 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
676 	mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
677 	MNT_ILOCK(mp);
678 	mp->mnt_flag |= MNT_LOCAL;
679 	MNT_IUNLOCK(mp);
680 	ump->um_mountp = mp;
681 	ump->um_dev = dev;
682 	ump->um_devvp = devvp;
683 	ump->um_bo = &devvp->v_bufobj;
684 	ump->um_cp = cp;
685 
686 	/*
687 	 * Setting those two parameters allowed us to use
688 	 * ufs_bmap w/o changse!
689 	 */
690 	ump->um_nindir = EXT2_ADDR_PER_BLOCK(fs);
691 	ump->um_bptrtodb = fs->e2fs->e2fs_log_bsize + 1;
692 	ump->um_seqinc = EXT2_FRAGS_PER_BLOCK(fs);
693 	if (ronly == 0)
694 		ext2_sbupdate(ump, MNT_WAIT);
695 	/*
696 	 * Initialize filesystem stat information in mount struct.
697 	 */
698 	MNT_ILOCK(mp);
699 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
700 	    MNTK_USES_BCACHE;
701 	MNT_IUNLOCK(mp);
702 	return (0);
703 out:
704 	if (bp)
705 		brelse(bp);
706 	if (cp != NULL) {
707 		g_topology_lock();
708 		g_vfs_close(cp);
709 		g_topology_unlock();
710 	}
711 	if (ump) {
712 		mtx_destroy(EXT2_MTX(ump));
713 		free(ump->um_e2fs->e2fs_gd, M_EXT2MNT);
714 		free(ump->um_e2fs->e2fs_contigdirs, M_EXT2MNT);
715 		free(ump->um_e2fs->e2fs, M_EXT2MNT);
716 		free(ump->um_e2fs, M_EXT2MNT);
717 		free(ump, M_EXT2MNT);
718 		mp->mnt_data = NULL;
719 	}
720 	return (error);
721 }
722 
723 /*
724  * Unmount system call.
725  */
726 static int
727 ext2_unmount(struct mount *mp, int mntflags)
728 {
729 	struct ext2mount *ump;
730 	struct m_ext2fs *fs;
731 	struct csum *sump;
732 	int error, flags, i, ronly;
733 
734 	flags = 0;
735 	if (mntflags & MNT_FORCE) {
736 		if (mp->mnt_flag & MNT_ROOTFS)
737 			return (EINVAL);
738 		flags |= FORCECLOSE;
739 	}
740 	if ((error = ext2_flushfiles(mp, flags, curthread)) != 0)
741 		return (error);
742 	ump = VFSTOEXT2(mp);
743 	fs = ump->um_e2fs;
744 	ronly = fs->e2fs_ronly;
745 	if (ronly == 0 && ext2_cgupdate(ump, MNT_WAIT) == 0) {
746 		if (fs->e2fs_wasvalid)
747 			fs->e2fs->e2fs_state |= E2FS_ISCLEAN;
748 		ext2_sbupdate(ump, MNT_WAIT);
749 	}
750 
751 	g_topology_lock();
752 	g_vfs_close(ump->um_cp);
753 	g_topology_unlock();
754 	vrele(ump->um_devvp);
755 	sump = fs->e2fs_clustersum;
756 	for (i = 0; i < fs->e2fs_gcount; i++, sump++)
757 		free(sump->cs_sum, M_EXT2MNT);
758 	free(fs->e2fs_clustersum, M_EXT2MNT);
759 	free(fs->e2fs_maxcluster, M_EXT2MNT);
760 	free(fs->e2fs_gd, M_EXT2MNT);
761 	free(fs->e2fs_contigdirs, M_EXT2MNT);
762 	free(fs->e2fs, M_EXT2MNT);
763 	free(fs, M_EXT2MNT);
764 	free(ump, M_EXT2MNT);
765 	mp->mnt_data = NULL;
766 	MNT_ILOCK(mp);
767 	mp->mnt_flag &= ~MNT_LOCAL;
768 	MNT_IUNLOCK(mp);
769 	return (error);
770 }
771 
772 /*
773  * Flush out all the files in a filesystem.
774  */
775 static int
776 ext2_flushfiles(struct mount *mp, int flags, struct thread *td)
777 {
778 	int error;
779 
780 	error = vflush(mp, 0, flags, td);
781 	return (error);
782 }
783 
784 /*
785  * Get filesystem statistics.
786  */
787 int
788 ext2_statfs(struct mount *mp, struct statfs *sbp)
789 {
790 	struct ext2mount *ump;
791 	struct m_ext2fs *fs;
792 	uint32_t overhead, overhead_per_group, ngdb;
793 	int i, ngroups;
794 
795 	ump = VFSTOEXT2(mp);
796 	fs = ump->um_e2fs;
797 	if (fs->e2fs->e2fs_magic != E2FS_MAGIC)
798 		panic("ext2_statfs");
799 
800 	/*
801 	 * Compute the overhead (FS structures)
802 	 */
803 	overhead_per_group =
804 	    1 /* block bitmap */ +
805 	    1 /* inode bitmap */ +
806 	    fs->e2fs_itpg;
807 	overhead = fs->e2fs->e2fs_first_dblock +
808 	    fs->e2fs_gcount * overhead_per_group;
809 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
810 	    fs->e2fs->e2fs_features_rocompat & EXT2F_ROCOMPAT_SPARSESUPER) {
811 		for (i = 0, ngroups = 0; i < fs->e2fs_gcount; i++) {
812 			if (ext2_cg_has_sb(fs, i))
813 				ngroups++;
814 		}
815 	} else {
816 		ngroups = fs->e2fs_gcount;
817 	}
818 	ngdb = fs->e2fs_gdbcount;
819 	if (fs->e2fs->e2fs_rev > E2FS_REV0 &&
820 	    fs->e2fs->e2fs_features_compat & EXT2F_COMPAT_RESIZE)
821 		ngdb += fs->e2fs->e2fs_reserved_ngdb;
822 	overhead += ngroups * (1 /* superblock */ + ngdb);
823 
824 	sbp->f_bsize = EXT2_FRAG_SIZE(fs);
825 	sbp->f_iosize = EXT2_BLOCK_SIZE(fs);
826 	sbp->f_blocks = fs->e2fs->e2fs_bcount - overhead;
827 	sbp->f_bfree = fs->e2fs->e2fs_fbcount;
828 	sbp->f_bavail = sbp->f_bfree - fs->e2fs->e2fs_rbcount;
829 	sbp->f_files = fs->e2fs->e2fs_icount;
830 	sbp->f_ffree = fs->e2fs->e2fs_ficount;
831 	return (0);
832 }
833 
834 /*
835  * Go through the disk queues to initiate sandbagged IO;
836  * go through the inodes to write those that have been modified;
837  * initiate the writing of the super block if it has been modified.
838  *
839  * Note: we are always called with the filesystem marked `MPBUSY'.
840  */
841 static int
842 ext2_sync(struct mount *mp, int waitfor)
843 {
844 	struct vnode *mvp, *vp;
845 	struct thread *td;
846 	struct inode *ip;
847 	struct ext2mount *ump = VFSTOEXT2(mp);
848 	struct m_ext2fs *fs;
849 	int error, allerror = 0;
850 
851 	td = curthread;
852 	fs = ump->um_e2fs;
853 	if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) {		/* XXX */
854 		printf("fs = %s\n", fs->e2fs_fsmnt);
855 		panic("ext2_sync: rofs mod");
856 	}
857 
858 	/*
859 	 * Write back each (modified) inode.
860 	 */
861 loop:
862 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
863 		if (vp->v_type == VNON) {
864 			VI_UNLOCK(vp);
865 			continue;
866 		}
867 		ip = VTOI(vp);
868 		if ((ip->i_flag &
869 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
870 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
871 		    waitfor == MNT_LAZY)) {
872 			VI_UNLOCK(vp);
873 			continue;
874 		}
875 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
876 		if (error) {
877 			if (error == ENOENT) {
878 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
879 				goto loop;
880 			}
881 			continue;
882 		}
883 		if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
884 			allerror = error;
885 		VOP_UNLOCK(vp, 0);
886 		vrele(vp);
887 	}
888 
889 	/*
890 	 * Force stale filesystem control information to be flushed.
891 	 */
892 	if (waitfor != MNT_LAZY) {
893 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
894 		if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
895 			allerror = error;
896 		VOP_UNLOCK(ump->um_devvp, 0);
897 	}
898 
899 	/*
900 	 * Write back modified superblock.
901 	 */
902 	if (fs->e2fs_fmod != 0) {
903 		fs->e2fs_fmod = 0;
904 		fs->e2fs->e2fs_wtime = time_second;
905 		if ((error = ext2_cgupdate(ump, waitfor)) != 0)
906 			allerror = error;
907 	}
908 	return (allerror);
909 }
910 
911 /*
912  * Look up an EXT2FS dinode number to find its incore vnode, otherwise read it
913  * in from disk.  If it is in core, wait for the lock bit to clear, then
914  * return the inode locked.  Detection and handling of mount points must be
915  * done by the calling routine.
916  */
917 static int
918 ext2_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
919 {
920 	struct m_ext2fs *fs;
921 	struct inode *ip;
922 	struct ext2mount *ump;
923 	struct buf *bp;
924 	struct vnode *vp;
925 	struct thread *td;
926 	int i, error;
927 	int used_blocks;
928 
929 	td = curthread;
930 	error = vfs_hash_get(mp, ino, flags, td, vpp, NULL, NULL);
931 	if (error || *vpp != NULL)
932 		return (error);
933 
934 	ump = VFSTOEXT2(mp);
935 	ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
936 
937 	/* Allocate a new vnode/inode. */
938 	if ((error = getnewvnode("ext2fs", mp, &ext2_vnodeops, &vp)) != 0) {
939 		*vpp = NULL;
940 		free(ip, M_EXT2NODE);
941 		return (error);
942 	}
943 	vp->v_data = ip;
944 	ip->i_vnode = vp;
945 	ip->i_e2fs = fs = ump->um_e2fs;
946 	ip->i_ump = ump;
947 	ip->i_number = ino;
948 
949 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
950 	error = insmntque(vp, mp);
951 	if (error != 0) {
952 		free(ip, M_EXT2NODE);
953 		*vpp = NULL;
954 		return (error);
955 	}
956 	error = vfs_hash_insert(vp, ino, flags, td, vpp, NULL, NULL);
957 	if (error || *vpp != NULL)
958 		return (error);
959 
960 	/* Read in the disk contents for the inode, copy into the inode. */
961 	if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
962 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
963 		/*
964 		 * The inode does not contain anything useful, so it would
965 		 * be misleading to leave it on its hash chain. With mode
966 		 * still zero, it will be unlinked and returned to the free
967 		 * list by vput().
968 		 */
969 		brelse(bp);
970 		vput(vp);
971 		*vpp = NULL;
972 		return (error);
973 	}
974 	/* convert ext2 inode to dinode */
975 	ext2_ei2i((struct ext2fs_dinode *)((char *)bp->b_data + EXT2_INODE_SIZE(fs) *
976 	    ino_to_fsbo(fs, ino)), ip);
977 	ip->i_block_group = ino_to_cg(fs, ino);
978 	ip->i_next_alloc_block = 0;
979 	ip->i_next_alloc_goal = 0;
980 
981 	/*
982 	 * Now we want to make sure that block pointers for unused
983 	 * blocks are zeroed out - ext2_balloc depends on this
984 	 * although for regular files and directories only
985 	 *
986 	 * If IN_E4EXTENTS is enabled, unused blocks are not zeroed
987 	 * out because we could corrupt the extent tree.
988 	 */
989 	if (!(ip->i_flag & IN_E4EXTENTS) &&
990 	    (S_ISDIR(ip->i_mode) || S_ISREG(ip->i_mode))) {
991 		used_blocks = howmany(ip->i_size, fs->e2fs_bsize);
992 		for (i = used_blocks; i < EXT2_NDIR_BLOCKS; i++)
993 			ip->i_db[i] = 0;
994 	}
995 #ifdef EXT2FS_DEBUG
996 	ext2_print_inode(ip);
997 	ext4_ext_print_extent_tree_status(ip);
998 #endif
999 	bqrelse(bp);
1000 
1001 	/*
1002 	 * Initialize the vnode from the inode, check for aliases.
1003 	 * Note that the underlying vnode may have changed.
1004 	 */
1005 	if ((error = ext2_vinit(mp, &ext2_fifoops, &vp)) != 0) {
1006 		vput(vp);
1007 		*vpp = NULL;
1008 		return (error);
1009 	}
1010 
1011 	/*
1012 	 * Finish inode initialization.
1013 	 */
1014 
1015 	*vpp = vp;
1016 	return (0);
1017 }
1018 
1019 /*
1020  * File handle to vnode
1021  *
1022  * Have to be really careful about stale file handles:
1023  * - check that the inode number is valid
1024  * - call ext2_vget() to get the locked inode
1025  * - check for an unallocated inode (i_mode == 0)
1026  * - check that the given client host has export rights and return
1027  *   those rights via. exflagsp and credanonp
1028  */
1029 static int
1030 ext2_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1031 {
1032 	struct inode *ip;
1033 	struct ufid *ufhp;
1034 	struct vnode *nvp;
1035 	struct m_ext2fs *fs;
1036 	int error;
1037 
1038 	ufhp = (struct ufid *)fhp;
1039 	fs = VFSTOEXT2(mp)->um_e2fs;
1040 	if (ufhp->ufid_ino < EXT2_ROOTINO ||
1041 	    ufhp->ufid_ino > fs->e2fs_gcount * fs->e2fs->e2fs_ipg)
1042 		return (ESTALE);
1043 
1044 	error = VFS_VGET(mp, ufhp->ufid_ino, LK_EXCLUSIVE, &nvp);
1045 	if (error) {
1046 		*vpp = NULLVP;
1047 		return (error);
1048 	}
1049 	ip = VTOI(nvp);
1050 	if (ip->i_mode == 0 ||
1051 	    ip->i_gen != ufhp->ufid_gen || ip->i_nlink <= 0) {
1052 		vput(nvp);
1053 		*vpp = NULLVP;
1054 		return (ESTALE);
1055 	}
1056 	*vpp = nvp;
1057 	vnode_create_vobject(*vpp, 0, curthread);
1058 	return (0);
1059 }
1060 
1061 /*
1062  * Write a superblock and associated information back to disk.
1063  */
1064 static int
1065 ext2_sbupdate(struct ext2mount *mp, int waitfor)
1066 {
1067 	struct m_ext2fs *fs = mp->um_e2fs;
1068 	struct ext2fs *es = fs->e2fs;
1069 	struct buf *bp;
1070 	int error = 0;
1071 
1072 	bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
1073 	bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
1074 	if (waitfor == MNT_WAIT)
1075 		error = bwrite(bp);
1076 	else
1077 		bawrite(bp);
1078 
1079 	/*
1080 	 * The buffers for group descriptors, inode bitmaps and block bitmaps
1081 	 * are not busy at this point and are (hopefully) written by the
1082 	 * usual sync mechanism. No need to write them here.
1083 	 */
1084 	return (error);
1085 }
1086 int
1087 ext2_cgupdate(struct ext2mount *mp, int waitfor)
1088 {
1089 	struct m_ext2fs *fs = mp->um_e2fs;
1090 	struct buf *bp;
1091 	int i, error = 0, allerror = 0;
1092 
1093 	allerror = ext2_sbupdate(mp, waitfor);
1094 
1095 	/* Update gd csums */
1096 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM))
1097 		ext2_gd_csum_set(fs);
1098 
1099 	for (i = 0; i < fs->e2fs_gdbcount; i++) {
1100 		bp = getblk(mp->um_devvp, fsbtodb(fs,
1101 		    fs->e2fs->e2fs_first_dblock +
1102 		    1 /* superblock */ + i), fs->e2fs_bsize, 0, 0, 0);
1103 		e2fs_cgsave(&fs->e2fs_gd[
1104 		    i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
1105 		    (struct ext2_gd *)bp->b_data, fs->e2fs_bsize);
1106 		if (waitfor == MNT_WAIT)
1107 			error = bwrite(bp);
1108 		else
1109 			bawrite(bp);
1110 	}
1111 
1112 	if (!allerror && error)
1113 		allerror = error;
1114 	return (allerror);
1115 }
1116 
1117 /*
1118  * Return the root of a filesystem.
1119  */
1120 static int
1121 ext2_root(struct mount *mp, int flags, struct vnode **vpp)
1122 {
1123 	struct vnode *nvp;
1124 	int error;
1125 
1126 	error = VFS_VGET(mp, EXT2_ROOTINO, LK_EXCLUSIVE, &nvp);
1127 	if (error)
1128 		return (error);
1129 	*vpp = nvp;
1130 	return (0);
1131 }
1132