xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision c98323078dede7579020518ec84cdcb478e5c142)
1 /*
2  * Copyright (c) 1989, 1991, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_mac.h"
36 #include "opt_quota.h"
37 #include "opt_ufs.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/namei.h>
42 #include <sys/proc.h>
43 #include <sys/kernel.h>
44 #include <sys/mac.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/disk.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
54 
55 #include <ufs/ufs/extattr.h>
56 #include <ufs/ufs/quota.h>
57 #include <ufs/ufs/ufsmount.h>
58 #include <ufs/ufs/inode.h>
59 #include <ufs/ufs/ufs_extern.h>
60 
61 #include <ufs/ffs/fs.h>
62 #include <ufs/ffs/ffs_extern.h>
63 
64 #include <vm/vm.h>
65 #include <vm/uma.h>
66 #include <vm/vm_page.h>
67 
68 uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
69 
70 static int	ffs_sbupdate(struct ufsmount *, int);
71        int	ffs_reload(struct mount *,struct ucred *,struct thread *);
72 static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
73 static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
74 		    ufs2_daddr_t);
75 static void	ffs_oldfscompat_write(struct fs *, struct ufsmount *);
76 static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
77 static vfs_init_t ffs_init;
78 static vfs_uninit_t ffs_uninit;
79 static vfs_extattrctl_t ffs_extattrctl;
80 
81 static struct vfsops ufs_vfsops = {
82 	.vfs_extattrctl =	ffs_extattrctl,
83 	.vfs_fhtovp =		ffs_fhtovp,
84 	.vfs_init =		ffs_init,
85 	.vfs_mount =		ffs_mount,
86 	.vfs_quotactl =		ufs_quotactl,
87 	.vfs_root =		ufs_root,
88 	.vfs_start =		ufs_start,
89 	.vfs_statfs =		ffs_statfs,
90 	.vfs_sync =		ffs_sync,
91 	.vfs_uninit =		ffs_uninit,
92 	.vfs_unmount =		ffs_unmount,
93 	.vfs_vget =		ffs_vget,
94 	.vfs_vptofh =		ffs_vptofh,
95 };
96 
97 VFS_SET(ufs_vfsops, ufs, 0);
98 
99 /*
100  * ffs_mount
101  *
102  * Called when mounting local physical media
103  *
104  * PARAMETERS:
105  *		mountroot
106  *			mp	mount point structure
107  *			path	NULL (flag for root mount!!!)
108  *			data	<unused>
109  *			ndp	<unused>
110  *			p	process (user credentials check [statfs])
111  *
112  *		mount
113  *			mp	mount point structure
114  *			path	path to mount point
115  *			data	pointer to argument struct in user space
116  *			ndp	mount point namei() return (used for
117  *				credentials on reload), reused to look
118  *				up block device.
119  *			p	process (user credentials check)
120  *
121  * RETURNS:	0	Success
122  *		!0	error number (errno.h)
123  *
124  * LOCK STATE:
125  *
126  *		ENTRY
127  *			mount point is locked
128  *		EXIT
129  *			mount point is locked
130  *
131  * NOTES:
132  *		A NULL path can be used for a flag since the mount
133  *		system call will fail with EFAULT in copyinstr in
134  *		namei() if it is a genuine NULL from the user.
135  */
136 int
137 ffs_mount(mp, path, data, ndp, td)
138         struct mount		*mp;	/* mount struct pointer*/
139         char			*path;	/* path to mount point*/
140         caddr_t			data;	/* arguments to FS specific mount*/
141         struct nameidata	*ndp;	/* mount point credentials*/
142         struct thread		*td;	/* process requesting mount*/
143 {
144 	size_t size;
145 	struct vnode *devvp;
146 	struct ufs_args args;
147 	struct ufsmount *ump = 0;
148 	struct fs *fs;
149 	int error, flags;
150 	mode_t accessmode;
151 
152 	if (uma_inode == NULL) {
153 		uma_inode = uma_zcreate("FFS inode",
154 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
155 		    UMA_ALIGN_PTR, 0);
156 		uma_ufs1 = uma_zcreate("FFS1 dinode",
157 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
158 		    UMA_ALIGN_PTR, 0);
159 		uma_ufs2 = uma_zcreate("FFS2 dinode",
160 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
161 		    UMA_ALIGN_PTR, 0);
162 	}
163 	/*
164 	 * Use NULL path to indicate we are mounting the root filesystem.
165 	 */
166 	if (path == NULL) {
167 		if ((error = bdevvp(rootdev, &rootvp))) {
168 			printf("ffs_mountroot: can't find rootvp\n");
169 			return (error);
170 		}
171 
172 		if ((error = ffs_mountfs(rootvp, mp, td)) != 0)
173 			return (error);
174 		return (0);
175 	}
176 
177 	/*
178 	 * Mounting non-root filesystem or updating a filesystem
179 	 */
180 	if ((error = copyin(data, (caddr_t)&args, sizeof(struct ufs_args)))!= 0)
181 		return (error);
182 
183 	/*
184 	 * If updating, check whether changing from read-only to
185 	 * read/write; if there is no device name, that's all we do.
186 	 */
187 	if (mp->mnt_flag & MNT_UPDATE) {
188 		ump = VFSTOUFS(mp);
189 		fs = ump->um_fs;
190 		devvp = ump->um_devvp;
191 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
192 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
193 				return (error);
194 			/*
195 			 * Flush any dirty data.
196 			 */
197 			if ((error = VFS_SYNC(mp, MNT_WAIT,
198 			    td->td_ucred, td)) != 0) {
199 				vn_finished_write(mp);
200 				return (error);
201 			}
202 			/*
203 			 * Check for and optionally get rid of files open
204 			 * for writing.
205 			 */
206 			flags = WRITECLOSE;
207 			if (mp->mnt_flag & MNT_FORCE)
208 				flags |= FORCECLOSE;
209 			if (mp->mnt_flag & MNT_SOFTDEP) {
210 				error = softdep_flushfiles(mp, flags, td);
211 			} else {
212 				error = ffs_flushfiles(mp, flags, td);
213 			}
214 			if (error) {
215 				vn_finished_write(mp);
216 				return (error);
217 			}
218 			if (fs->fs_pendingblocks != 0 ||
219 			    fs->fs_pendinginodes != 0) {
220 				printf("%s: %s: blocks %jd files %d\n",
221 				    fs->fs_fsmnt, "update error",
222 				    (intmax_t)fs->fs_pendingblocks,
223 				    fs->fs_pendinginodes);
224 				fs->fs_pendingblocks = 0;
225 				fs->fs_pendinginodes = 0;
226 			}
227 			fs->fs_ronly = 1;
228 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
229 				fs->fs_clean = 1;
230 			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
231 				fs->fs_ronly = 0;
232 				fs->fs_clean = 0;
233 				vn_finished_write(mp);
234 				return (error);
235 			}
236 			vn_finished_write(mp);
237 		}
238 		if ((mp->mnt_flag & MNT_RELOAD) &&
239 		    (error = ffs_reload(mp, ndp->ni_cnd.cn_cred, td)) != 0)
240 			return (error);
241 		if (fs->fs_ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
242 			/*
243 			 * If upgrade to read-write by non-root, then verify
244 			 * that user has necessary permissions on the device.
245 			 */
246 			if (suser(td)) {
247 				vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
248 				if ((error = VOP_ACCESS(devvp, VREAD | VWRITE,
249 				    td->td_ucred, td)) != 0) {
250 					VOP_UNLOCK(devvp, 0, td);
251 					return (error);
252 				}
253 				VOP_UNLOCK(devvp, 0, td);
254 			}
255 			fs->fs_flags &= ~FS_UNCLEAN;
256 			if (fs->fs_clean == 0) {
257 				fs->fs_flags |= FS_UNCLEAN;
258 				if ((mp->mnt_flag & MNT_FORCE) ||
259 				    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
260 				     (fs->fs_flags & FS_DOSOFTDEP))) {
261 					printf("WARNING: %s was not %s\n",
262 					   fs->fs_fsmnt, "properly dismounted");
263 				} else {
264 					printf(
265 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
266 					    fs->fs_fsmnt);
267 					return (EPERM);
268 				}
269 			}
270 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
271 				return (error);
272 			fs->fs_ronly = 0;
273 			fs->fs_clean = 0;
274 			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
275 				vn_finished_write(mp);
276 				return (error);
277 			}
278 			/* check to see if we need to start softdep */
279 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
280 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
281 				vn_finished_write(mp);
282 				return (error);
283 			}
284 			if (fs->fs_snapinum[0] != 0)
285 				ffs_snapshot_mount(mp);
286 			vn_finished_write(mp);
287 		}
288 		/*
289 		 * Soft updates is incompatible with "async",
290 		 * so if we are doing softupdates stop the user
291 		 * from setting the async flag in an update.
292 		 * Softdep_mount() clears it in an initial mount
293 		 * or ro->rw remount.
294 		 */
295 		if (mp->mnt_flag & MNT_SOFTDEP)
296 			mp->mnt_flag &= ~MNT_ASYNC;
297 		/*
298 		 * If not updating name, process export requests.
299 		 */
300 		if (args.fspec == 0)
301 			return (vfs_export(mp, &args.export));
302 		/*
303 		 * If this is a snapshot request, take the snapshot.
304 		 */
305 		if (mp->mnt_flag & MNT_SNAPSHOT)
306 			return (ffs_snapshot(mp, args.fspec));
307 	}
308 
309 	/*
310 	 * Not an update, or updating the name: look up the name
311 	 * and verify that it refers to a sensible disk device.
312 	 */
313 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, td);
314 	if ((error = namei(ndp)) != 0)
315 		return (error);
316 	NDFREE(ndp, NDF_ONLY_PNBUF);
317 	devvp = ndp->ni_vp;
318 	if (!vn_isdisk(devvp, &error)) {
319 		vrele(devvp);
320 		return (error);
321 	}
322 
323 	/*
324 	 * If mount by non-root, then verify that user has necessary
325 	 * permissions on the device.
326 	 */
327 	if (suser(td)) {
328 		accessmode = VREAD;
329 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
330 			accessmode |= VWRITE;
331 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
332 		if ((error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td))!= 0){
333 			vput(devvp);
334 			return (error);
335 		}
336 		VOP_UNLOCK(devvp, 0, td);
337 	}
338 
339 	if (mp->mnt_flag & MNT_UPDATE) {
340 		/*
341 		 * Update only
342 		 *
343 		 * If it's not the same vnode, or at least the same device
344 		 * then it's not correct.
345 		 */
346 
347 		if (devvp != ump->um_devvp &&
348 		    devvp->v_rdev != ump->um_devvp->v_rdev)
349 			error = EINVAL;	/* needs translation */
350 		vrele(devvp);
351 		if (error)
352 			return (error);
353 	} else {
354 		/*
355 		 * New mount
356 		 *
357 		 * We need the name for the mount point (also used for
358 		 * "last mounted on") copied in. If an error occurs,
359 		 * the mount point is discarded by the upper level code.
360 		 * Note that vfs_mount() populates f_mntonname for us.
361 		 */
362 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
363 			vrele(devvp);
364 			return (error);
365 		}
366 	}
367 	/*
368 	 * Save "mounted from" device name info for mount point (NULL pad).
369 	 */
370 	copyinstr(args.fspec, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
371 	bzero( mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
372 	return (0);
373 }
374 
375 /*
376  * Reload all incore data for a filesystem (used after running fsck on
377  * the root filesystem and finding things to fix). The filesystem must
378  * be mounted read-only.
379  *
380  * Things to do to update the mount:
381  *	1) invalidate all cached meta-data.
382  *	2) re-read superblock from disk.
383  *	3) re-read summary information from disk.
384  *	4) invalidate all inactive vnodes.
385  *	5) invalidate all cached file data.
386  *	6) re-read inode data for all active vnodes.
387  */
388 int
389 ffs_reload(mp, cred, td)
390 	struct mount *mp;
391 	struct ucred *cred;
392 	struct thread *td;
393 {
394 	struct vnode *vp, *nvp, *devvp;
395 	struct inode *ip;
396 	void *space;
397 	struct buf *bp;
398 	struct fs *fs, *newfs;
399 	ufs2_daddr_t sblockloc;
400 	int i, blks, size, error;
401 	int32_t *lp;
402 
403 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
404 		return (EINVAL);
405 	/*
406 	 * Step 1: invalidate all cached meta-data.
407 	 */
408 	devvp = VFSTOUFS(mp)->um_devvp;
409 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
410 	if (vinvalbuf(devvp, 0, cred, td, 0, 0) != 0)
411 		panic("ffs_reload: dirty1");
412 	/*
413 	 * Only VMIO the backing device if the backing device is a real
414 	 * disk device.  See ffs_mountfs() for more details.
415 	 */
416 	if (vn_isdisk(devvp, NULL))
417 		vfs_object_create(devvp, td, td->td_ucred);
418 	VOP_UNLOCK(devvp, 0, td);
419 
420 	/*
421 	 * Step 2: re-read superblock from disk.
422 	 */
423 	fs = VFSTOUFS(mp)->um_fs;
424 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
425 	    NOCRED, &bp)) != 0)
426 		return (error);
427 	newfs = (struct fs *)bp->b_data;
428 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
429 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
430 	    newfs->fs_bsize > MAXBSIZE ||
431 	    newfs->fs_bsize < sizeof(struct fs)) {
432 			brelse(bp);
433 			return (EIO);		/* XXX needs translation */
434 	}
435 	/*
436 	 * Copy pointer fields back into superblock before copying in	XXX
437 	 * new superblock. These should really be in the ufsmount.	XXX
438 	 * Note that important parameters (eg fs_ncg) are unchanged.
439 	 */
440 	newfs->fs_csp = fs->fs_csp;
441 	newfs->fs_maxcluster = fs->fs_maxcluster;
442 	newfs->fs_contigdirs = fs->fs_contigdirs;
443 	newfs->fs_active = fs->fs_active;
444 	/* The file system is still read-only. */
445 	newfs->fs_ronly = 1;
446 	sblockloc = fs->fs_sblockloc;
447 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
448 	brelse(bp);
449 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
450 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
451 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
452 		printf("%s: reload pending error: blocks %jd files %d\n",
453 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
454 		    fs->fs_pendinginodes);
455 		fs->fs_pendingblocks = 0;
456 		fs->fs_pendinginodes = 0;
457 	}
458 
459 	/*
460 	 * Step 3: re-read summary information from disk.
461 	 */
462 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
463 	space = fs->fs_csp;
464 	for (i = 0; i < blks; i += fs->fs_frag) {
465 		size = fs->fs_bsize;
466 		if (i + fs->fs_frag > blks)
467 			size = (blks - i) * fs->fs_fsize;
468 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
469 		    NOCRED, &bp);
470 		if (error)
471 			return (error);
472 		bcopy(bp->b_data, space, (u_int)size);
473 		space = (char *)space + size;
474 		brelse(bp);
475 	}
476 	/*
477 	 * We no longer know anything about clusters per cylinder group.
478 	 */
479 	if (fs->fs_contigsumsize > 0) {
480 		lp = fs->fs_maxcluster;
481 		for (i = 0; i < fs->fs_ncg; i++)
482 			*lp++ = fs->fs_contigsumsize;
483 	}
484 
485 loop:
486 	MNT_ILOCK(mp);
487 	MNT_VNODE_FOREACH(vp, mp, nvp) {
488 		VI_LOCK(vp);
489 		if (vp->v_iflag & VI_XLOCK) {
490 			VI_UNLOCK(vp);
491 			continue;
492 		}
493 		MNT_IUNLOCK(mp);
494 		/*
495 		 * Step 4: invalidate all inactive vnodes.
496 		 */
497 		if (vp->v_usecount == 0) {
498 			vgonel(vp, td);
499 			goto loop;
500 		}
501 		/*
502 		 * Step 5: invalidate all cached file data.
503 		 */
504 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
505 			goto loop;
506 		}
507 		if (vinvalbuf(vp, 0, cred, td, 0, 0))
508 			panic("ffs_reload: dirty2");
509 		/*
510 		 * Step 6: re-read inode data for all active vnodes.
511 		 */
512 		ip = VTOI(vp);
513 		error =
514 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
515 		    (int)fs->fs_bsize, NOCRED, &bp);
516 		if (error) {
517 			VOP_UNLOCK(vp, 0, td);
518 			vrele(vp);
519 			return (error);
520 		}
521 		ffs_load_inode(bp, ip, fs, ip->i_number);
522 		ip->i_effnlink = ip->i_nlink;
523 		brelse(bp);
524 		VOP_UNLOCK(vp, 0, td);
525 		vrele(vp);
526 		MNT_ILOCK(mp);
527 	}
528 	MNT_IUNLOCK(mp);
529 	return (0);
530 }
531 
532 /*
533  * Possible superblock locations ordered from most to least likely.
534  */
535 static int sblock_try[] = SBLOCKSEARCH;
536 
537 /*
538  * Common code for mount and mountroot
539  */
540 static int
541 ffs_mountfs(devvp, mp, td)
542 	struct vnode *devvp;
543 	struct mount *mp;
544 	struct thread *td;
545 {
546 	struct ufsmount *ump;
547 	struct buf *bp;
548 	struct fs *fs;
549 	struct cdev *dev;
550 	void *space;
551 	ufs2_daddr_t sblockloc;
552 	int error, i, blks, size, ronly;
553 	int32_t *lp;
554 	struct ucred *cred;
555 	size_t strsize;
556 
557 	dev = devvp->v_rdev;
558 	cred = td ? td->td_ucred : NOCRED;
559 	/*
560 	 * Disallow multiple mounts of the same device.
561 	 * Disallow mounting of a device that is currently in use
562 	 * (except for root, which might share swap device for miniroot).
563 	 * Flush out any old buffers remaining from a previous use.
564 	 */
565 	error = vfs_mountedon(devvp);
566 	if (error)
567 		return (error);
568 	if (vcount(devvp) > 1 && devvp != rootvp)
569 		return (EBUSY);
570 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
571 	error = vinvalbuf(devvp, V_SAVE, cred, td, 0, 0);
572 	if (error) {
573 		VOP_UNLOCK(devvp, 0, td);
574 		return (error);
575 	}
576 
577 	/*
578 	 * Only VMIO the backing device if the backing device is a real
579 	 * disk device.
580 	 * Note that it is optional that the backing device be VMIOed.  This
581 	 * increases the opportunity for metadata caching.
582 	 */
583 	if (vn_isdisk(devvp, NULL))
584 		vfs_object_create(devvp, td, cred);
585 
586 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
587 	/*
588 	 * XXX: open the device with read and write access even if only
589 	 * read access is needed now.  Write access is needed if the
590 	 * filesystem is ever mounted read/write, and we don't change the
591 	 * access mode for remounts.
592 	 */
593 #ifdef notyet
594 	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD | FWRITE, FSCRED, td, -1);
595 #else
596 	error = VOP_OPEN(devvp, FREAD | FWRITE, FSCRED, td, -1);
597 #endif
598 	VOP_UNLOCK(devvp, 0, td);
599 	if (error)
600 		return (error);
601 	if (devvp->v_rdev->si_iosize_max != 0)
602 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
603 	if (mp->mnt_iosize_max > MAXPHYS)
604 		mp->mnt_iosize_max = MAXPHYS;
605 
606 	bp = NULL;
607 	ump = NULL;
608 	fs = NULL;
609 	sblockloc = 0;
610 	/*
611 	 * Try reading the superblock in each of its possible locations.
612 	 */
613 	for (i = 0; sblock_try[i] != -1; i++) {
614 		if ((error = bread(devvp, sblock_try[i] / DEV_BSIZE, SBLOCKSIZE,
615 		    cred, &bp)) != 0)
616 			goto out;
617 		fs = (struct fs *)bp->b_data;
618 		sblockloc = sblock_try[i];
619 		if ((fs->fs_magic == FS_UFS1_MAGIC ||
620 		     (fs->fs_magic == FS_UFS2_MAGIC &&
621 		      (fs->fs_sblockloc == sblockloc ||
622 		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
623 		    fs->fs_bsize <= MAXBSIZE &&
624 		    fs->fs_bsize >= sizeof(struct fs))
625 			break;
626 		brelse(bp);
627 		bp = NULL;
628 	}
629 	if (sblock_try[i] == -1) {
630 		error = EINVAL;		/* XXX needs translation */
631 		goto out;
632 	}
633 	fs->fs_fmod = 0;
634 	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indicies */
635 	fs->fs_flags &= ~FS_UNCLEAN;
636 	if (fs->fs_clean == 0) {
637 		fs->fs_flags |= FS_UNCLEAN;
638 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
639 		    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
640 		     (fs->fs_flags & FS_DOSOFTDEP))) {
641 			printf(
642 "WARNING: %s was not properly dismounted\n",
643 			    fs->fs_fsmnt);
644 		} else {
645 			printf(
646 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
647 			    fs->fs_fsmnt);
648 			error = EPERM;
649 			goto out;
650 		}
651 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
652 		    (mp->mnt_flag & MNT_FORCE)) {
653 			printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt,
654 			    (intmax_t)fs->fs_pendingblocks,
655 			    fs->fs_pendinginodes);
656 			fs->fs_pendingblocks = 0;
657 			fs->fs_pendinginodes = 0;
658 		}
659 	}
660 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
661 		printf("%s: mount pending error: blocks %jd files %d\n",
662 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
663 		    fs->fs_pendinginodes);
664 		fs->fs_pendingblocks = 0;
665 		fs->fs_pendinginodes = 0;
666 	}
667 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
668 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT,
669 	    M_WAITOK);
670 	if (fs->fs_magic == FS_UFS1_MAGIC) {
671 		ump->um_fstype = UFS1;
672 		ump->um_balloc = ffs_balloc_ufs1;
673 	} else {
674 		ump->um_fstype = UFS2;
675 		ump->um_balloc = ffs_balloc_ufs2;
676 	}
677 	ump->um_blkatoff = ffs_blkatoff;
678 	ump->um_truncate = ffs_truncate;
679 	ump->um_update = ffs_update;
680 	ump->um_valloc = ffs_valloc;
681 	ump->um_vfree = ffs_vfree;
682 	ump->um_ifree = ffs_ifree;
683 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
684 	if (fs->fs_sbsize < SBLOCKSIZE)
685 		bp->b_flags |= B_INVAL | B_NOCACHE;
686 	brelse(bp);
687 	bp = NULL;
688 	fs = ump->um_fs;
689 	ffs_oldfscompat_read(fs, ump, sblockloc);
690 	fs->fs_ronly = ronly;
691 	size = fs->fs_cssize;
692 	blks = howmany(size, fs->fs_fsize);
693 	if (fs->fs_contigsumsize > 0)
694 		size += fs->fs_ncg * sizeof(int32_t);
695 	size += fs->fs_ncg * sizeof(u_int8_t);
696 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
697 	fs->fs_csp = space;
698 	for (i = 0; i < blks; i += fs->fs_frag) {
699 		size = fs->fs_bsize;
700 		if (i + fs->fs_frag > blks)
701 			size = (blks - i) * fs->fs_fsize;
702 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
703 		    cred, &bp)) != 0) {
704 			free(fs->fs_csp, M_UFSMNT);
705 			goto out;
706 		}
707 		bcopy(bp->b_data, space, (u_int)size);
708 		space = (char *)space + size;
709 		brelse(bp);
710 		bp = NULL;
711 	}
712 	if (fs->fs_contigsumsize > 0) {
713 		fs->fs_maxcluster = lp = space;
714 		for (i = 0; i < fs->fs_ncg; i++)
715 			*lp++ = fs->fs_contigsumsize;
716 		space = lp;
717 	}
718 	size = fs->fs_ncg * sizeof(u_int8_t);
719 	fs->fs_contigdirs = (u_int8_t *)space;
720 	bzero(fs->fs_contigdirs, size);
721 	fs->fs_active = NULL;
722 	mp->mnt_data = (qaddr_t)ump;
723 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
724 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
725 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
726 	    vfs_getvfs(&mp->mnt_stat.f_fsid))
727 		vfs_getnewfsid(mp);
728 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
729 	mp->mnt_flag |= MNT_LOCAL;
730 	if ((fs->fs_flags & FS_MULTILABEL) != 0)
731 #ifdef MAC
732 		mp->mnt_flag |= MNT_MULTILABEL;
733 #else
734 		printf(
735 "WARNING: %s: multilabel flag on fs but no MAC support\n",
736 		    fs->fs_fsmnt);
737 #endif
738 	if ((fs->fs_flags & FS_ACLS) != 0)
739 #ifdef UFS_ACL
740 		mp->mnt_flag |= MNT_ACLS;
741 #else
742 		printf(
743 "WARNING: %s: ACLs flag on fs but no ACLs support\n",
744 		    fs->fs_fsmnt);
745 #endif
746 	ump->um_mountp = mp;
747 	ump->um_dev = dev;
748 	ump->um_devvp = devvp;
749 	ump->um_nindir = fs->fs_nindir;
750 	ump->um_bptrtodb = fs->fs_fsbtodb;
751 	ump->um_seqinc = fs->fs_frag;
752 	for (i = 0; i < MAXQUOTAS; i++)
753 		ump->um_quotas[i] = NULLVP;
754 #ifdef UFS_EXTATTR
755 	ufs_extattr_uepm_init(&ump->um_extattr);
756 #endif
757 	devvp->v_rdev->si_mountpoint = mp;
758 
759 	/*
760 	 * Set FS local "last mounted on" information (NULL pad)
761 	 */
762 	copystr(	mp->mnt_stat.f_mntonname,	/* mount point*/
763 			fs->fs_fsmnt,			/* copy area*/
764 			sizeof(fs->fs_fsmnt) - 1,	/* max size*/
765 			&strsize);			/* real size*/
766 	bzero( fs->fs_fsmnt + strsize, sizeof(fs->fs_fsmnt) - strsize);
767 
768 	if( mp->mnt_flag & MNT_ROOTFS) {
769 		/*
770 		 * Root mount; update timestamp in mount structure.
771 		 * this will be used by the common root mount code
772 		 * to update the system clock.
773 		 */
774 		mp->mnt_time = fs->fs_time;
775 	}
776 
777 	if (ronly == 0) {
778 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
779 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
780 			free(fs->fs_csp, M_UFSMNT);
781 			goto out;
782 		}
783 		if (fs->fs_snapinum[0] != 0)
784 			ffs_snapshot_mount(mp);
785 		fs->fs_fmod = 1;
786 		fs->fs_clean = 0;
787 		(void) ffs_sbupdate(ump, MNT_WAIT);
788 	}
789 	/*
790 	 * Initialize filesystem stat information in mount struct.
791 	 */
792 	(void)VFS_STATFS(mp, &mp->mnt_stat, td);
793 #ifdef UFS_EXTATTR
794 #ifdef UFS_EXTATTR_AUTOSTART
795 	/*
796 	 *
797 	 * Auto-starting does the following:
798 	 *	- check for /.attribute in the fs, and extattr_start if so
799 	 *	- for each file in .attribute, enable that file with
800 	 * 	  an attribute of the same name.
801 	 * Not clear how to report errors -- probably eat them.
802 	 * This would all happen while the filesystem was busy/not
803 	 * available, so would effectively be "atomic".
804 	 */
805 	(void) ufs_extattr_autostart(mp, td);
806 #endif /* !UFS_EXTATTR_AUTOSTART */
807 #endif /* !UFS_EXTATTR */
808 	return (0);
809 out:
810 	devvp->v_rdev->si_mountpoint = NULL;
811 	if (bp)
812 		brelse(bp);
813 	/* XXX: see comment above VOP_OPEN. */
814 #ifdef notyet
815 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD | FWRITE, cred, td);
816 #else
817 	(void)VOP_CLOSE(devvp, FREAD | FWRITE, cred, td);
818 #endif
819 	if (ump) {
820 		free(ump->um_fs, M_UFSMNT);
821 		free(ump, M_UFSMNT);
822 		mp->mnt_data = (qaddr_t)0;
823 	}
824 	return (error);
825 }
826 
827 #include <sys/sysctl.h>
828 int bigcgs = 0;
829 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
830 
831 /*
832  * Sanity checks for loading old filesystem superblocks.
833  * See ffs_oldfscompat_write below for unwound actions.
834  *
835  * XXX - Parts get retired eventually.
836  * Unfortunately new bits get added.
837  */
838 static void
839 ffs_oldfscompat_read(fs, ump, sblockloc)
840 	struct fs *fs;
841 	struct ufsmount *ump;
842 	ufs2_daddr_t sblockloc;
843 {
844 	off_t maxfilesize;
845 
846 	/*
847 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
848 	 */
849 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
850 		fs->fs_flags = fs->fs_old_flags;
851 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
852 		fs->fs_sblockloc = sblockloc;
853 	}
854 	/*
855 	 * If not yet done, update UFS1 superblock with new wider fields.
856 	 */
857 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
858 		fs->fs_maxbsize = fs->fs_bsize;
859 		fs->fs_time = fs->fs_old_time;
860 		fs->fs_size = fs->fs_old_size;
861 		fs->fs_dsize = fs->fs_old_dsize;
862 		fs->fs_csaddr = fs->fs_old_csaddr;
863 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
864 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
865 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
866 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
867 	}
868 	if (fs->fs_magic == FS_UFS1_MAGIC &&
869 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
870 		fs->fs_maxfilesize = (u_quad_t) 1LL << 39;
871 		fs->fs_qbmask = ~fs->fs_bmask;
872 		fs->fs_qfmask = ~fs->fs_fmask;
873 	}
874 	if (fs->fs_magic == FS_UFS1_MAGIC) {
875 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
876 		maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1;
877 		if (fs->fs_maxfilesize > maxfilesize)
878 			fs->fs_maxfilesize = maxfilesize;
879 	}
880 	/* Compatibility for old filesystems */
881 	if (fs->fs_avgfilesize <= 0)
882 		fs->fs_avgfilesize = AVFILESIZ;
883 	if (fs->fs_avgfpdir <= 0)
884 		fs->fs_avgfpdir = AFPDIR;
885 	if (bigcgs) {
886 		fs->fs_save_cgsize = fs->fs_cgsize;
887 		fs->fs_cgsize = fs->fs_bsize;
888 	}
889 }
890 
891 /*
892  * Unwinding superblock updates for old filesystems.
893  * See ffs_oldfscompat_read above for details.
894  *
895  * XXX - Parts get retired eventually.
896  * Unfortunately new bits get added.
897  */
898 static void
899 ffs_oldfscompat_write(fs, ump)
900 	struct fs *fs;
901 	struct ufsmount *ump;
902 {
903 
904 	/*
905 	 * Copy back UFS2 updated fields that UFS1 inspects.
906 	 */
907 	if (fs->fs_magic == FS_UFS1_MAGIC) {
908 		fs->fs_old_time = fs->fs_time;
909 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
910 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
911 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
912 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
913 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
914 	}
915 	if (bigcgs) {
916 		fs->fs_cgsize = fs->fs_save_cgsize;
917 		fs->fs_save_cgsize = 0;
918 	}
919 }
920 
921 /*
922  * unmount system call
923  */
924 int
925 ffs_unmount(mp, mntflags, td)
926 	struct mount *mp;
927 	int mntflags;
928 	struct thread *td;
929 {
930 	struct ufsmount *ump = VFSTOUFS(mp);
931 	struct fs *fs;
932 	int error, flags;
933 
934 	flags = 0;
935 	if (mntflags & MNT_FORCE) {
936 		flags |= FORCECLOSE;
937 	}
938 #ifdef UFS_EXTATTR
939 	if ((error = ufs_extattr_stop(mp, td))) {
940 		if (error != EOPNOTSUPP)
941 			printf("ffs_unmount: ufs_extattr_stop returned %d\n",
942 			    error);
943 	} else {
944 		ufs_extattr_uepm_destroy(&ump->um_extattr);
945 	}
946 #endif
947 	if (mp->mnt_flag & MNT_SOFTDEP) {
948 		if ((error = softdep_flushfiles(mp, flags, td)) != 0)
949 			return (error);
950 	} else {
951 		if ((error = ffs_flushfiles(mp, flags, td)) != 0)
952 			return (error);
953 	}
954 	fs = ump->um_fs;
955 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
956 		printf("%s: unmount pending error: blocks %jd files %d\n",
957 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
958 		    fs->fs_pendinginodes);
959 		fs->fs_pendingblocks = 0;
960 		fs->fs_pendinginodes = 0;
961 	}
962 	if (fs->fs_ronly == 0) {
963 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
964 		error = ffs_sbupdate(ump, MNT_WAIT);
965 		if (error) {
966 			fs->fs_clean = 0;
967 			return (error);
968 		}
969 	}
970 	ump->um_devvp->v_rdev->si_mountpoint = NULL;
971 
972 	vinvalbuf(ump->um_devvp, V_SAVE, NOCRED, td, 0, 0);
973 	/* XXX: see comment above VOP_OPEN. */
974 #ifdef notyet
975 	error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD | FWRITE,
976 	    NOCRED, td);
977 #else
978 	error = VOP_CLOSE(ump->um_devvp, FREAD | FWRITE, NOCRED, td);
979 #endif
980 	vrele(ump->um_devvp);
981 	free(fs->fs_csp, M_UFSMNT);
982 	free(fs, M_UFSMNT);
983 	free(ump, M_UFSMNT);
984 	mp->mnt_data = (qaddr_t)0;
985 	mp->mnt_flag &= ~MNT_LOCAL;
986 	return (error);
987 }
988 
989 /*
990  * Flush out all the files in a filesystem.
991  */
992 int
993 ffs_flushfiles(mp, flags, td)
994 	struct mount *mp;
995 	int flags;
996 	struct thread *td;
997 {
998 	struct ufsmount *ump;
999 	int error;
1000 
1001 	ump = VFSTOUFS(mp);
1002 #ifdef QUOTA
1003 	if (mp->mnt_flag & MNT_QUOTA) {
1004 		int i;
1005 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1006 		if (error)
1007 			return (error);
1008 		for (i = 0; i < MAXQUOTAS; i++) {
1009 			if (ump->um_quotas[i] == NULLVP)
1010 				continue;
1011 			quotaoff(td, mp, i);
1012 		}
1013 		/*
1014 		 * Here we fall through to vflush again to ensure
1015 		 * that we have gotten rid of all the system vnodes.
1016 		 */
1017 	}
1018 #endif
1019 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1020 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1021 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1022 			return (error);
1023 		ffs_snapshot_unmount(mp);
1024 		/*
1025 		 * Here we fall through to vflush again to ensure
1026 		 * that we have gotten rid of all the system vnodes.
1027 		 */
1028 	}
1029         /*
1030 	 * Flush all the files.
1031 	 */
1032 	if ((error = vflush(mp, 0, flags, td)) != 0)
1033 		return (error);
1034 	/*
1035 	 * Flush filesystem metadata.
1036 	 */
1037 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td);
1038 	error = VOP_FSYNC(ump->um_devvp, td->td_ucred, MNT_WAIT, td);
1039 	VOP_UNLOCK(ump->um_devvp, 0, td);
1040 	return (error);
1041 }
1042 
1043 /*
1044  * Get filesystem statistics.
1045  */
1046 int
1047 ffs_statfs(mp, sbp, td)
1048 	struct mount *mp;
1049 	struct statfs *sbp;
1050 	struct thread *td;
1051 {
1052 	struct ufsmount *ump;
1053 	struct fs *fs;
1054 
1055 	ump = VFSTOUFS(mp);
1056 	fs = ump->um_fs;
1057 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1058 		panic("ffs_statfs");
1059 	sbp->f_version = STATFS_VERSION;
1060 	sbp->f_bsize = fs->fs_fsize;
1061 	sbp->f_iosize = fs->fs_bsize;
1062 	sbp->f_blocks = fs->fs_dsize;
1063 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1064 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1065 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1066 	    dbtofsb(fs, fs->fs_pendingblocks);
1067 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1068 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1069 	sbp->f_namemax = NAME_MAX;
1070 	if (sbp != &mp->mnt_stat) {
1071 		sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1072 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
1073 		sbp->f_syncwrites = mp->mnt_stat.f_syncwrites;
1074 		sbp->f_asyncwrites = mp->mnt_stat.f_asyncwrites;
1075 		sbp->f_syncreads = mp->mnt_stat.f_syncreads;
1076 		sbp->f_asyncreads = mp->mnt_stat.f_asyncreads;
1077 		sbp->f_owner = mp->mnt_stat.f_owner;
1078 		sbp->f_fsid = mp->mnt_stat.f_fsid;
1079 		bcopy((caddr_t)mp->mnt_stat.f_fstypename,
1080 			(caddr_t)&sbp->f_fstypename[0], MFSNAMELEN);
1081 		bcopy((caddr_t)mp->mnt_stat.f_mntonname,
1082 			(caddr_t)&sbp->f_mntonname[0], MNAMELEN);
1083 		bcopy((caddr_t)mp->mnt_stat.f_mntfromname,
1084 			(caddr_t)&sbp->f_mntfromname[0], MNAMELEN);
1085 	}
1086 	return (0);
1087 }
1088 
1089 /*
1090  * Go through the disk queues to initiate sandbagged IO;
1091  * go through the inodes to write those that have been modified;
1092  * initiate the writing of the super block if it has been modified.
1093  *
1094  * Note: we are always called with the filesystem marked `MPBUSY'.
1095  */
1096 int
1097 ffs_sync(mp, waitfor, cred, td)
1098 	struct mount *mp;
1099 	int waitfor;
1100 	struct ucred *cred;
1101 	struct thread *td;
1102 {
1103 	struct vnode *nvp, *vp, *devvp;
1104 	struct inode *ip;
1105 	struct ufsmount *ump = VFSTOUFS(mp);
1106 	struct fs *fs;
1107 	int error, count, wait, lockreq, allerror = 0;
1108 
1109 	fs = ump->um_fs;
1110 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
1111 		printf("fs = %s\n", fs->fs_fsmnt);
1112 		panic("ffs_sync: rofs mod");
1113 	}
1114 	/*
1115 	 * Write back each (modified) inode.
1116 	 */
1117 	wait = 0;
1118 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1119 	if (waitfor == MNT_WAIT) {
1120 		wait = 1;
1121 		lockreq = LK_EXCLUSIVE;
1122 	}
1123 	lockreq |= LK_INTERLOCK;
1124 	MNT_ILOCK(mp);
1125 loop:
1126 	MNT_VNODE_FOREACH(vp, mp, nvp) {
1127 		/*
1128 		 * Depend on the mntvnode_slock to keep things stable enough
1129 		 * for a quick test.  Since there might be hundreds of
1130 		 * thousands of vnodes, we cannot afford even a subroutine
1131 		 * call unless there's a good chance that we have work to do.
1132 		 */
1133 		VI_LOCK(vp);
1134 		if (vp->v_iflag & VI_XLOCK) {
1135 			VI_UNLOCK(vp);
1136 			continue;
1137 		}
1138 		ip = VTOI(vp);
1139 		if (vp->v_type == VNON || ((ip->i_flag &
1140 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1141 		    TAILQ_EMPTY(&vp->v_dirtyblkhd))) {
1142 			VI_UNLOCK(vp);
1143 			continue;
1144 		}
1145 		MNT_IUNLOCK(mp);
1146 		if ((error = vget(vp, lockreq, td)) != 0) {
1147 			MNT_ILOCK(mp);
1148 			if (error == ENOENT)
1149 				goto loop;
1150 			continue;
1151 		}
1152 		if ((error = VOP_FSYNC(vp, cred, waitfor, td)) != 0)
1153 			allerror = error;
1154 		VOP_UNLOCK(vp, 0, td);
1155 		vrele(vp);
1156 		MNT_ILOCK(mp);
1157 	}
1158 	MNT_IUNLOCK(mp);
1159 	/*
1160 	 * Force stale filesystem control information to be flushed.
1161 	 */
1162 	if (waitfor == MNT_WAIT) {
1163 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1164 			allerror = error;
1165 		/* Flushed work items may create new vnodes to clean */
1166 		if (allerror == 0 && count) {
1167 			MNT_ILOCK(mp);
1168 			goto loop;
1169 		}
1170 	}
1171 #ifdef QUOTA
1172 	qsync(mp);
1173 #endif
1174 	devvp = ump->um_devvp;
1175 	VI_LOCK(devvp);
1176 	if (waitfor != MNT_LAZY &&
1177 	    (devvp->v_numoutput > 0 || TAILQ_FIRST(&devvp->v_dirtyblkhd))) {
1178 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td);
1179 		if ((error = VOP_FSYNC(devvp, cred, waitfor, td)) != 0)
1180 			allerror = error;
1181 		VOP_UNLOCK(devvp, 0, td);
1182 		if (allerror == 0 && waitfor == MNT_WAIT) {
1183 			MNT_ILOCK(mp);
1184 			goto loop;
1185 		}
1186 	} else
1187 		VI_UNLOCK(devvp);
1188 	/*
1189 	 * Write back modified superblock.
1190 	 */
1191 	if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor)) != 0)
1192 		allerror = error;
1193 	return (allerror);
1194 }
1195 
1196 int
1197 ffs_vget(mp, ino, flags, vpp)
1198 	struct mount *mp;
1199 	ino_t ino;
1200 	int flags;
1201 	struct vnode **vpp;
1202 {
1203 	struct thread *td = curthread; 		/* XXX */
1204 	struct fs *fs;
1205 	struct inode *ip;
1206 	struct ufsmount *ump;
1207 	struct buf *bp;
1208 	struct vnode *vp;
1209 	struct cdev *dev;
1210 	int error;
1211 
1212 	ump = VFSTOUFS(mp);
1213 	dev = ump->um_dev;
1214 
1215 	/*
1216 	 * We do not lock vnode creation as it is believed to be too
1217 	 * expensive for such rare case as simultaneous creation of vnode
1218 	 * for same ino by different processes. We just allow them to race
1219 	 * and check later to decide who wins. Let the race begin!
1220 	 */
1221 	if ((error = ufs_ihashget(dev, ino, flags, vpp)) != 0)
1222 		return (error);
1223 	if (*vpp != NULL)
1224 		return (0);
1225 
1226 	/*
1227 	 * If this MALLOC() is performed after the getnewvnode()
1228 	 * it might block, leaving a vnode with a NULL v_data to be
1229 	 * found by ffs_sync() if a sync happens to fire right then,
1230 	 * which will cause a panic because ffs_sync() blindly
1231 	 * dereferences vp->v_data (as well it should).
1232 	 */
1233 	ip = uma_zalloc(uma_inode, M_WAITOK);
1234 
1235 	/* Allocate a new vnode/inode. */
1236 	error = getnewvnode("ufs", mp, ffs_vnodeop_p, &vp);
1237 	if (error) {
1238 		*vpp = NULL;
1239 		uma_zfree(uma_inode, ip);
1240 		return (error);
1241 	}
1242 	bzero((caddr_t)ip, sizeof(struct inode));
1243 	/*
1244 	 * FFS supports recursive locking.
1245 	 */
1246 	fs = ump->um_fs;
1247 	vp->v_vnlock->lk_flags |= LK_CANRECURSE;
1248 	vp->v_data = ip;
1249 	vp->v_bsize = fs->fs_bsize;
1250 	ip->i_vnode = vp;
1251 	ip->i_ump = ump;
1252 	ip->i_fs = fs;
1253 	ip->i_dev = dev;
1254 	ip->i_number = ino;
1255 #ifdef QUOTA
1256 	{
1257 		int i;
1258 		for (i = 0; i < MAXQUOTAS; i++)
1259 			ip->i_dquot[i] = NODQUOT;
1260 	}
1261 #endif
1262 	/*
1263 	 * Exclusively lock the vnode before adding to hash. Note, that we
1264 	 * must not release nor downgrade the lock (despite flags argument
1265 	 * says) till it is fully initialized.
1266 	 */
1267 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, (struct mtx *)0, td);
1268 
1269 	/*
1270 	 * Atomicaly (in terms of ufs_hash operations) check the hash for
1271 	 * duplicate of vnode being created and add it to the hash. If a
1272 	 * duplicate vnode was found, it will be vget()ed from hash for us.
1273 	 */
1274 	if ((error = ufs_ihashins(ip, flags, vpp)) != 0) {
1275 		vput(vp);
1276 		*vpp = NULL;
1277 		return (error);
1278 	}
1279 
1280 	/* We lost the race, then throw away our vnode and return existing */
1281 	if (*vpp != NULL) {
1282 		vput(vp);
1283 		return (0);
1284 	}
1285 
1286 	/* Read in the disk contents for the inode, copy into the inode. */
1287 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1288 	    (int)fs->fs_bsize, NOCRED, &bp);
1289 	if (error) {
1290 		/*
1291 		 * The inode does not contain anything useful, so it would
1292 		 * be misleading to leave it on its hash chain. With mode
1293 		 * still zero, it will be unlinked and returned to the free
1294 		 * list by vput().
1295 		 */
1296 		brelse(bp);
1297 		vput(vp);
1298 		*vpp = NULL;
1299 		return (error);
1300 	}
1301 	if (ip->i_ump->um_fstype == UFS1)
1302 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1303 	else
1304 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1305 	ffs_load_inode(bp, ip, fs, ino);
1306 	if (DOINGSOFTDEP(vp))
1307 		softdep_load_inodeblock(ip);
1308 	else
1309 		ip->i_effnlink = ip->i_nlink;
1310 	bqrelse(bp);
1311 
1312 	/*
1313 	 * Initialize the vnode from the inode, check for aliases.
1314 	 * Note that the underlying vnode may have changed.
1315 	 */
1316 	error = ufs_vinit(mp, ffs_specop_p, ffs_fifoop_p, &vp);
1317 	if (error) {
1318 		vput(vp);
1319 		*vpp = NULL;
1320 		return (error);
1321 	}
1322 	/*
1323 	 * Finish inode initialization.
1324 	 */
1325 	VREF(ip->i_devvp);
1326 	/*
1327 	 * Set up a generation number for this inode if it does not
1328 	 * already have one. This should only happen on old filesystems.
1329 	 */
1330 	if (ip->i_gen == 0) {
1331 		ip->i_gen = arc4random() / 2 + 1;
1332 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1333 			ip->i_flag |= IN_MODIFIED;
1334 			DIP(ip, i_gen) = ip->i_gen;
1335 		}
1336 	}
1337 	/*
1338 	 * Ensure that uid and gid are correct. This is a temporary
1339 	 * fix until fsck has been changed to do the update.
1340 	 */
1341 	if (fs->fs_magic == FS_UFS1_MAGIC &&		/* XXX */
1342 	    fs->fs_old_inodefmt < FS_44INODEFMT) {	/* XXX */
1343 		ip->i_uid = ip->i_din1->di_ouid;	/* XXX */
1344 		ip->i_gid = ip->i_din1->di_ogid;	/* XXX */
1345 	}						/* XXX */
1346 
1347 #ifdef MAC
1348 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1349 		/*
1350 		 * If this vnode is already allocated, and we're running
1351 		 * multi-label, attempt to perform a label association
1352 		 * from the extended attributes on the inode.
1353 		 */
1354 		error = mac_associate_vnode_extattr(mp, vp);
1355 		if (error) {
1356 			/* ufs_inactive will release ip->i_devvp ref. */
1357 			vput(vp);
1358 			*vpp = NULL;
1359 			return (error);
1360 		}
1361 	}
1362 #endif
1363 
1364 	*vpp = vp;
1365 	return (0);
1366 }
1367 
1368 /*
1369  * File handle to vnode
1370  *
1371  * Have to be really careful about stale file handles:
1372  * - check that the inode number is valid
1373  * - call ffs_vget() to get the locked inode
1374  * - check for an unallocated inode (i_mode == 0)
1375  * - check that the given client host has export rights and return
1376  *   those rights via. exflagsp and credanonp
1377  */
1378 int
1379 ffs_fhtovp(mp, fhp, vpp)
1380 	struct mount *mp;
1381 	struct fid *fhp;
1382 	struct vnode **vpp;
1383 {
1384 	struct ufid *ufhp;
1385 	struct fs *fs;
1386 
1387 	ufhp = (struct ufid *)fhp;
1388 	fs = VFSTOUFS(mp)->um_fs;
1389 	if (ufhp->ufid_ino < ROOTINO ||
1390 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1391 		return (ESTALE);
1392 	return (ufs_fhtovp(mp, ufhp, vpp));
1393 }
1394 
1395 /*
1396  * Vnode pointer to File handle
1397  */
1398 /* ARGSUSED */
1399 int
1400 ffs_vptofh(vp, fhp)
1401 	struct vnode *vp;
1402 	struct fid *fhp;
1403 {
1404 	struct inode *ip;
1405 	struct ufid *ufhp;
1406 
1407 	ip = VTOI(vp);
1408 	ufhp = (struct ufid *)fhp;
1409 	ufhp->ufid_len = sizeof(struct ufid);
1410 	ufhp->ufid_ino = ip->i_number;
1411 	ufhp->ufid_gen = ip->i_gen;
1412 	return (0);
1413 }
1414 
1415 /*
1416  * Initialize the filesystem.
1417  */
1418 static int
1419 ffs_init(vfsp)
1420 	struct vfsconf *vfsp;
1421 {
1422 
1423 	softdep_initialize();
1424 	return (ufs_init(vfsp));
1425 }
1426 
1427 /*
1428  * Undo the work of ffs_init().
1429  */
1430 static int
1431 ffs_uninit(vfsp)
1432 	struct vfsconf *vfsp;
1433 {
1434 	int ret;
1435 
1436 	ret = ufs_uninit(vfsp);
1437 	softdep_uninitialize();
1438 	return (ret);
1439 }
1440 
1441 /*
1442  * Write a superblock and associated information back to disk.
1443  */
1444 static int
1445 ffs_sbupdate(mp, waitfor)
1446 	struct ufsmount *mp;
1447 	int waitfor;
1448 {
1449 	struct fs *fs = mp->um_fs;
1450 	struct buf *bp;
1451 	int blks;
1452 	void *space;
1453 	int i, size, error, allerror = 0;
1454 
1455 	if (fs->fs_ronly == 1 &&
1456 	    (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1457 	    (MNT_RDONLY | MNT_UPDATE))
1458 		panic("ffs_sbupdate: write read-only filesystem");
1459 	/*
1460 	 * First write back the summary information.
1461 	 */
1462 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1463 	space = fs->fs_csp;
1464 	for (i = 0; i < blks; i += fs->fs_frag) {
1465 		size = fs->fs_bsize;
1466 		if (i + fs->fs_frag > blks)
1467 			size = (blks - i) * fs->fs_fsize;
1468 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1469 		    size, 0, 0, 0);
1470 		bcopy(space, bp->b_data, (u_int)size);
1471 		space = (char *)space + size;
1472 		if (waitfor != MNT_WAIT)
1473 			bawrite(bp);
1474 		else if ((error = bwrite(bp)) != 0)
1475 			allerror = error;
1476 	}
1477 	/*
1478 	 * Now write back the superblock itself. If any errors occurred
1479 	 * up to this point, then fail so that the superblock avoids
1480 	 * being written out as clean.
1481 	 */
1482 	if (allerror)
1483 		return (allerror);
1484 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1485 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1486 		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1487 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1488 		fs->fs_sblockloc = SBLOCK_UFS1;
1489 	}
1490 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1491 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1492 		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1493 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1494 		fs->fs_sblockloc = SBLOCK_UFS2;
1495 	}
1496 	bp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize,
1497 	    0, 0, 0);
1498 	fs->fs_fmod = 0;
1499 	fs->fs_time = time_second;
1500 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1501 	ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
1502 	if (waitfor != MNT_WAIT)
1503 		bawrite(bp);
1504 	else if ((error = bwrite(bp)) != 0)
1505 		allerror = error;
1506 	return (allerror);
1507 }
1508 
1509 static int
1510 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1511 	int attrnamespace, const char *attrname, struct thread *td)
1512 {
1513 
1514 #ifdef UFS_EXTATTR
1515 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1516 	    attrname, td));
1517 #else
1518 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1519 	    attrname, td));
1520 #endif
1521 }
1522 
1523 static void
1524 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1525 {
1526 
1527 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1528 		uma_zfree(uma_ufs1, ip->i_din1);
1529 	else if (ip->i_din2 != NULL)
1530 		uma_zfree(uma_ufs2, ip->i_din2);
1531 	uma_zfree(uma_inode, ip);
1532 }
1533