ffs_vfsops.c revision 121874
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 * 3. All advertising materials mentioning features or use of this software
14 *    must display the following acknowledgement:
15 *	This product includes software developed by the University of
16 *	California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 *    may be used to endorse or promote products derived from this software
19 *    without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
34 */
35
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD: head/sys/ufs/ffs/ffs_vfsops.c 121874 2003-11-02 04:52:53Z kan $");
38
39#include "opt_mac.h"
40#include "opt_quota.h"
41#include "opt_ufs.h"
42
43#include <sys/param.h>
44#include <sys/systm.h>
45#include <sys/namei.h>
46#include <sys/proc.h>
47#include <sys/kernel.h>
48#include <sys/mac.h>
49#include <sys/vnode.h>
50#include <sys/mount.h>
51#include <sys/bio.h>
52#include <sys/buf.h>
53#include <sys/conf.h>
54#include <sys/fcntl.h>
55#include <sys/disk.h>
56#include <sys/malloc.h>
57#include <sys/mutex.h>
58
59#include <ufs/ufs/extattr.h>
60#include <ufs/ufs/quota.h>
61#include <ufs/ufs/ufsmount.h>
62#include <ufs/ufs/inode.h>
63#include <ufs/ufs/ufs_extern.h>
64
65#include <ufs/ffs/fs.h>
66#include <ufs/ffs/ffs_extern.h>
67
68#include <vm/vm.h>
69#include <vm/uma.h>
70#include <vm/vm_page.h>
71
72uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
73
74static int	ffs_sbupdate(struct ufsmount *, int);
75       int	ffs_reload(struct mount *,struct ucred *,struct thread *);
76static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
77static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
78		    ufs2_daddr_t);
79static void	ffs_oldfscompat_write(struct fs *, struct ufsmount *);
80static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
81static vfs_init_t ffs_init;
82static vfs_uninit_t ffs_uninit;
83static vfs_extattrctl_t ffs_extattrctl;
84
85static struct vfsops ufs_vfsops = {
86	.vfs_extattrctl =	ffs_extattrctl,
87	.vfs_fhtovp =		ffs_fhtovp,
88	.vfs_init =		ffs_init,
89	.vfs_mount =		ffs_mount,
90	.vfs_quotactl =		ufs_quotactl,
91	.vfs_root =		ufs_root,
92	.vfs_start =		ufs_start,
93	.vfs_statfs =		ffs_statfs,
94	.vfs_sync =		ffs_sync,
95	.vfs_uninit =		ffs_uninit,
96	.vfs_unmount =		ffs_unmount,
97	.vfs_vget =		ffs_vget,
98	.vfs_vptofh =		ffs_vptofh,
99};
100
101VFS_SET(ufs_vfsops, ufs, 0);
102
103/*
104 * ffs_mount
105 *
106 * Called when mounting local physical media
107 *
108 * PARAMETERS:
109 *		mountroot
110 *			mp	mount point structure
111 *			path	NULL (flag for root mount!!!)
112 *			data	<unused>
113 *			ndp	<unused>
114 *			p	process (user credentials check [statfs])
115 *
116 *		mount
117 *			mp	mount point structure
118 *			path	path to mount point
119 *			data	pointer to argument struct in user space
120 *			ndp	mount point namei() return (used for
121 *				credentials on reload), reused to look
122 *				up block device.
123 *			p	process (user credentials check)
124 *
125 * RETURNS:	0	Success
126 *		!0	error number (errno.h)
127 *
128 * LOCK STATE:
129 *
130 *		ENTRY
131 *			mount point is locked
132 *		EXIT
133 *			mount point is locked
134 *
135 * NOTES:
136 *		A NULL path can be used for a flag since the mount
137 *		system call will fail with EFAULT in copyinstr in
138 *		namei() if it is a genuine NULL from the user.
139 */
140int
141ffs_mount(mp, path, data, ndp, td)
142        struct mount		*mp;	/* mount struct pointer*/
143        char			*path;	/* path to mount point*/
144        caddr_t			data;	/* arguments to FS specific mount*/
145        struct nameidata	*ndp;	/* mount point credentials*/
146        struct thread		*td;	/* process requesting mount*/
147{
148	size_t size;
149	struct vnode *devvp;
150	struct ufs_args args;
151	struct ufsmount *ump = 0;
152	struct fs *fs;
153	int error, flags;
154	mode_t accessmode;
155
156	if (uma_inode == NULL) {
157		uma_inode = uma_zcreate("FFS inode",
158		    sizeof(struct inode), NULL, NULL, NULL, NULL,
159		    UMA_ALIGN_PTR, 0);
160		uma_ufs1 = uma_zcreate("FFS1 dinode",
161		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
162		    UMA_ALIGN_PTR, 0);
163		uma_ufs2 = uma_zcreate("FFS2 dinode",
164		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
165		    UMA_ALIGN_PTR, 0);
166	}
167	/*
168	 * Use NULL path to indicate we are mounting the root filesystem.
169	 */
170	if (path == NULL) {
171		if ((error = bdevvp(rootdev, &rootvp))) {
172			printf("ffs_mountroot: can't find rootvp\n");
173			return (error);
174		}
175
176		if ((error = ffs_mountfs(rootvp, mp, td)) != 0)
177			return (error);
178		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
179		return (0);
180	}
181
182	/*
183	 * Mounting non-root filesystem or updating a filesystem
184	 */
185	if ((error = copyin(data, (caddr_t)&args, sizeof(struct ufs_args)))!= 0)
186		return (error);
187
188	/*
189	 * If updating, check whether changing from read-only to
190	 * read/write; if there is no device name, that's all we do.
191	 */
192	if (mp->mnt_flag & MNT_UPDATE) {
193		ump = VFSTOUFS(mp);
194		fs = ump->um_fs;
195		devvp = ump->um_devvp;
196		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
197			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
198				return (error);
199			/*
200			 * Flush any dirty data.
201			 */
202			if ((error = VFS_SYNC(mp, MNT_WAIT,
203			    td->td_ucred, td)) != 0) {
204				vn_finished_write(mp);
205				return (error);
206			}
207			/*
208			 * Check for and optionally get rid of files open
209			 * for writing.
210			 */
211			flags = WRITECLOSE;
212			if (mp->mnt_flag & MNT_FORCE)
213				flags |= FORCECLOSE;
214			if (mp->mnt_flag & MNT_SOFTDEP) {
215				error = softdep_flushfiles(mp, flags, td);
216			} else {
217				error = ffs_flushfiles(mp, flags, td);
218			}
219			if (error) {
220				vn_finished_write(mp);
221				return (error);
222			}
223			if (fs->fs_pendingblocks != 0 ||
224			    fs->fs_pendinginodes != 0) {
225				printf("%s: %s: blocks %jd files %d\n",
226				    fs->fs_fsmnt, "update error",
227				    (intmax_t)fs->fs_pendingblocks,
228				    fs->fs_pendinginodes);
229				fs->fs_pendingblocks = 0;
230				fs->fs_pendinginodes = 0;
231			}
232			fs->fs_ronly = 1;
233			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
234				fs->fs_clean = 1;
235			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
236				fs->fs_ronly = 0;
237				fs->fs_clean = 0;
238				vn_finished_write(mp);
239				return (error);
240			}
241			vn_finished_write(mp);
242		}
243		if ((mp->mnt_flag & MNT_RELOAD) &&
244		    (error = ffs_reload(mp, ndp->ni_cnd.cn_cred, td)) != 0)
245			return (error);
246		if (fs->fs_ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
247			/*
248			 * If upgrade to read-write by non-root, then verify
249			 * that user has necessary permissions on the device.
250			 */
251			if (suser(td)) {
252				vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
253				if ((error = VOP_ACCESS(devvp, VREAD | VWRITE,
254				    td->td_ucred, td)) != 0) {
255					VOP_UNLOCK(devvp, 0, td);
256					return (error);
257				}
258				VOP_UNLOCK(devvp, 0, td);
259			}
260			fs->fs_flags &= ~FS_UNCLEAN;
261			if (fs->fs_clean == 0) {
262				fs->fs_flags |= FS_UNCLEAN;
263				if ((mp->mnt_flag & MNT_FORCE) ||
264				    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
265				     (fs->fs_flags & FS_DOSOFTDEP))) {
266					printf("WARNING: %s was not %s\n",
267					   fs->fs_fsmnt, "properly dismounted");
268				} else {
269					printf(
270"WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
271					    fs->fs_fsmnt);
272					return (EPERM);
273				}
274			}
275			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
276				return (error);
277			fs->fs_ronly = 0;
278			fs->fs_clean = 0;
279			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
280				vn_finished_write(mp);
281				return (error);
282			}
283			/* check to see if we need to start softdep */
284			if ((fs->fs_flags & FS_DOSOFTDEP) &&
285			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
286				vn_finished_write(mp);
287				return (error);
288			}
289			if (fs->fs_snapinum[0] != 0)
290				ffs_snapshot_mount(mp);
291			vn_finished_write(mp);
292		}
293		/*
294		 * Soft updates is incompatible with "async",
295		 * so if we are doing softupdates stop the user
296		 * from setting the async flag in an update.
297		 * Softdep_mount() clears it in an initial mount
298		 * or ro->rw remount.
299		 */
300		if (mp->mnt_flag & MNT_SOFTDEP)
301			mp->mnt_flag &= ~MNT_ASYNC;
302		/*
303		 * If not updating name, process export requests.
304		 */
305		if (args.fspec == 0)
306			return (vfs_export(mp, &args.export));
307		/*
308		 * If this is a snapshot request, take the snapshot.
309		 */
310		if (mp->mnt_flag & MNT_SNAPSHOT)
311			return (ffs_snapshot(mp, args.fspec));
312	}
313
314	/*
315	 * Not an update, or updating the name: look up the name
316	 * and verify that it refers to a sensible block device.
317	 */
318	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, td);
319	if ((error = namei(ndp)) != 0)
320		return (error);
321	NDFREE(ndp, NDF_ONLY_PNBUF);
322	devvp = ndp->ni_vp;
323	if (!vn_isdisk(devvp, &error)) {
324		vrele(devvp);
325		return (error);
326	}
327
328	/*
329	 * If mount by non-root, then verify that user has necessary
330	 * permissions on the device.
331	 */
332	if (suser(td)) {
333		accessmode = VREAD;
334		if ((mp->mnt_flag & MNT_RDONLY) == 0)
335			accessmode |= VWRITE;
336		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
337		if ((error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td))!= 0){
338			vput(devvp);
339			return (error);
340		}
341		VOP_UNLOCK(devvp, 0, td);
342	}
343
344	if (mp->mnt_flag & MNT_UPDATE) {
345		/*
346		 * Update only
347		 *
348		 * If it's not the same vnode, or at least the same device
349		 * then it's not correct.
350		 */
351
352		if (devvp != ump->um_devvp &&
353		    devvp->v_rdev != ump->um_devvp->v_rdev)
354			error = EINVAL;	/* needs translation */
355		vrele(devvp);
356		if (error)
357			return (error);
358	} else {
359		/*
360		 * New mount
361		 *
362		 * We need the name for the mount point (also used for
363		 * "last mounted on") copied in. If an error occurs,
364		 * the mount point is discarded by the upper level code.
365		 * Note that vfs_mount() populates f_mntonname for us.
366		 */
367		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
368			vrele(devvp);
369			return (error);
370		}
371	}
372	/*
373	 * Save "mounted from" device name info for mount point (NULL pad).
374	 */
375	copyinstr(args.fspec, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
376	bzero( mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
377	/*
378	 * Initialize filesystem stat information in mount struct.
379	 */
380	(void)VFS_STATFS(mp, &mp->mnt_stat, td);
381	return (0);
382}
383
384/*
385 * Reload all incore data for a filesystem (used after running fsck on
386 * the root filesystem and finding things to fix). The filesystem must
387 * be mounted read-only.
388 *
389 * Things to do to update the mount:
390 *	1) invalidate all cached meta-data.
391 *	2) re-read superblock from disk.
392 *	3) re-read summary information from disk.
393 *	4) invalidate all inactive vnodes.
394 *	5) invalidate all cached file data.
395 *	6) re-read inode data for all active vnodes.
396 */
397int
398ffs_reload(mp, cred, td)
399	struct mount *mp;
400	struct ucred *cred;
401	struct thread *td;
402{
403	struct vnode *vp, *nvp, *devvp;
404	struct inode *ip;
405	void *space;
406	struct buf *bp;
407	struct fs *fs, *newfs;
408	ufs2_daddr_t sblockloc;
409	int i, blks, size, error;
410	int32_t *lp;
411
412	if ((mp->mnt_flag & MNT_RDONLY) == 0)
413		return (EINVAL);
414	/*
415	 * Step 1: invalidate all cached meta-data.
416	 */
417	devvp = VFSTOUFS(mp)->um_devvp;
418	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
419	error = vinvalbuf(devvp, 0, cred, td, 0, 0);
420	VOP_UNLOCK(devvp, 0, td);
421	if (error)
422		panic("ffs_reload: dirty1");
423
424	/*
425	 * Only VMIO the backing device if the backing device is a real
426	 * block device.
427	 */
428	if (vn_isdisk(devvp, NULL)) {
429		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
430		vfs_object_create(devvp, td, td->td_ucred);
431		VOP_UNLOCK(devvp, 0, td);
432	}
433
434	/*
435	 * Step 2: re-read superblock from disk.
436	 */
437	fs = VFSTOUFS(mp)->um_fs;
438	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
439	    NOCRED, &bp)) != 0)
440		return (error);
441	newfs = (struct fs *)bp->b_data;
442	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
443	     newfs->fs_magic != FS_UFS2_MAGIC) ||
444	    newfs->fs_bsize > MAXBSIZE ||
445	    newfs->fs_bsize < sizeof(struct fs)) {
446			brelse(bp);
447			return (EIO);		/* XXX needs translation */
448	}
449	/*
450	 * Copy pointer fields back into superblock before copying in	XXX
451	 * new superblock. These should really be in the ufsmount.	XXX
452	 * Note that important parameters (eg fs_ncg) are unchanged.
453	 */
454	newfs->fs_csp = fs->fs_csp;
455	newfs->fs_maxcluster = fs->fs_maxcluster;
456	newfs->fs_contigdirs = fs->fs_contigdirs;
457	newfs->fs_active = fs->fs_active;
458	sblockloc = fs->fs_sblockloc;
459	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
460	brelse(bp);
461	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
462	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
463	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
464		printf("%s: reload pending error: blocks %jd files %d\n",
465		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
466		    fs->fs_pendinginodes);
467		fs->fs_pendingblocks = 0;
468		fs->fs_pendinginodes = 0;
469	}
470
471	/*
472	 * Step 3: re-read summary information from disk.
473	 */
474	blks = howmany(fs->fs_cssize, fs->fs_fsize);
475	space = fs->fs_csp;
476	for (i = 0; i < blks; i += fs->fs_frag) {
477		size = fs->fs_bsize;
478		if (i + fs->fs_frag > blks)
479			size = (blks - i) * fs->fs_fsize;
480		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
481		    NOCRED, &bp);
482		if (error)
483			return (error);
484		bcopy(bp->b_data, space, (u_int)size);
485		space = (char *)space + size;
486		brelse(bp);
487	}
488	/*
489	 * We no longer know anything about clusters per cylinder group.
490	 */
491	if (fs->fs_contigsumsize > 0) {
492		lp = fs->fs_maxcluster;
493		for (i = 0; i < fs->fs_ncg; i++)
494			*lp++ = fs->fs_contigsumsize;
495	}
496
497loop:
498	mtx_lock(&mntvnode_mtx);
499	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
500		if (vp->v_mount != mp) {
501			mtx_unlock(&mntvnode_mtx);
502			goto loop;
503		}
504		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
505		VI_LOCK(vp);
506		if (vp->v_iflag & VI_XLOCK) {
507			VI_UNLOCK(vp);
508			continue;
509		}
510		mtx_unlock(&mntvnode_mtx);
511		/*
512		 * Step 4: invalidate all inactive vnodes.
513		 */
514		if (vp->v_usecount == 0) {
515			vgonel(vp, td);
516			goto loop;
517		}
518		/*
519		 * Step 5: invalidate all cached file data.
520		 */
521		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
522			goto loop;
523		}
524		if (vinvalbuf(vp, 0, cred, td, 0, 0))
525			panic("ffs_reload: dirty2");
526		/*
527		 * Step 6: re-read inode data for all active vnodes.
528		 */
529		ip = VTOI(vp);
530		error =
531		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
532		    (int)fs->fs_bsize, NOCRED, &bp);
533		if (error) {
534			vput(vp);
535			return (error);
536		}
537		ffs_load_inode(bp, ip, fs, ip->i_number);
538		ip->i_effnlink = ip->i_nlink;
539		brelse(bp);
540		VOP_UNLOCK(vp, 0, td);
541		vrele(vp);
542		mtx_lock(&mntvnode_mtx);
543	}
544	mtx_unlock(&mntvnode_mtx);
545	return (0);
546}
547
548/*
549 * Possible superblock locations ordered from most to least likely.
550 */
551static int sblock_try[] = SBLOCKSEARCH;
552
553/*
554 * Common code for mount and mountroot
555 */
556static int
557ffs_mountfs(devvp, mp, td)
558	struct vnode *devvp;
559	struct mount *mp;
560	struct thread *td;
561{
562	struct ufsmount *ump;
563	struct buf *bp;
564	struct fs *fs;
565	dev_t dev;
566	void *space;
567	ufs2_daddr_t sblockloc;
568	int error, i, blks, size, ronly;
569	int32_t *lp;
570	struct ucred *cred;
571	size_t strsize;
572	int ncount;
573
574	dev = devvp->v_rdev;
575	cred = td ? td->td_ucred : NOCRED;
576	/*
577	 * Disallow multiple mounts of the same device.
578	 * Disallow mounting of a device that is currently in use
579	 * (except for root, which might share swap device for miniroot).
580	 * Flush out any old buffers remaining from a previous use.
581	 */
582	error = vfs_mountedon(devvp);
583	if (error)
584		return (error);
585	ncount = vcount(devvp);
586
587	if (ncount > 1 && devvp != rootvp)
588		return (EBUSY);
589	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
590	error = vinvalbuf(devvp, V_SAVE, cred, td, 0, 0);
591	VOP_UNLOCK(devvp, 0, td);
592	if (error)
593		return (error);
594
595	/*
596	 * Only VMIO the backing device if the backing device is a real
597	 * block device.
598	 * Note that it is optional that the backing device be VMIOed.  This
599	 * increases the opportunity for metadata caching.
600	 */
601	if (vn_isdisk(devvp, NULL)) {
602		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
603		vfs_object_create(devvp, td, cred);
604		VOP_UNLOCK(devvp, 0, td);
605	}
606
607	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
608	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
609	/*
610	 * XXX: We don't re-VOP_OPEN in FREAD|FWRITE mode if the filesystem
611	 * XXX: is subsequently remounted, so open it FREAD|FWRITE from the
612	 * XXX: start to avoid getting trashed later on.
613	 */
614#ifdef notyet
615	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, td, -1);
616#else
617	error = VOP_OPEN(devvp, FREAD|FWRITE, FSCRED, td, -1);
618#endif
619	VOP_UNLOCK(devvp, 0, td);
620	if (error)
621		return (error);
622	if (devvp->v_rdev->si_iosize_max != 0)
623		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
624	if (mp->mnt_iosize_max > MAXPHYS)
625		mp->mnt_iosize_max = MAXPHYS;
626
627	bp = NULL;
628	ump = NULL;
629	fs = NULL;
630	sblockloc = 0;
631	/*
632	 * Try reading the superblock in each of its possible locations.
633	 */
634	for (i = 0; sblock_try[i] != -1; i++) {
635		if ((error = bread(devvp, sblock_try[i] / DEV_BSIZE, SBLOCKSIZE,
636		    cred, &bp)) != 0)
637			goto out;
638		fs = (struct fs *)bp->b_data;
639		sblockloc = sblock_try[i];
640		if ((fs->fs_magic == FS_UFS1_MAGIC ||
641		     (fs->fs_magic == FS_UFS2_MAGIC &&
642		      (fs->fs_sblockloc == sblockloc ||
643		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
644		    fs->fs_bsize <= MAXBSIZE &&
645		    fs->fs_bsize >= sizeof(struct fs))
646			break;
647		brelse(bp);
648		bp = NULL;
649	}
650	if (sblock_try[i] == -1) {
651		error = EINVAL;		/* XXX needs translation */
652		goto out;
653	}
654	fs->fs_fmod = 0;
655	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indicies */
656	fs->fs_flags &= ~FS_UNCLEAN;
657	if (fs->fs_clean == 0) {
658		fs->fs_flags |= FS_UNCLEAN;
659		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
660		    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
661		     (fs->fs_flags & FS_DOSOFTDEP))) {
662			printf(
663"WARNING: %s was not properly dismounted\n",
664			    fs->fs_fsmnt);
665		} else {
666			printf(
667"WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
668			    fs->fs_fsmnt);
669			error = EPERM;
670			goto out;
671		}
672		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
673		    (mp->mnt_flag & MNT_FORCE)) {
674			printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt,
675			    (intmax_t)fs->fs_pendingblocks,
676			    fs->fs_pendinginodes);
677			fs->fs_pendingblocks = 0;
678			fs->fs_pendinginodes = 0;
679		}
680	}
681	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
682		printf("%s: mount pending error: blocks %jd files %d\n",
683		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
684		    fs->fs_pendinginodes);
685		fs->fs_pendingblocks = 0;
686		fs->fs_pendinginodes = 0;
687	}
688	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
689	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT,
690	    M_WAITOK);
691	if (fs->fs_magic == FS_UFS1_MAGIC) {
692		ump->um_fstype = UFS1;
693		ump->um_balloc = ffs_balloc_ufs1;
694	} else {
695		ump->um_fstype = UFS2;
696		ump->um_balloc = ffs_balloc_ufs2;
697	}
698	ump->um_blkatoff = ffs_blkatoff;
699	ump->um_truncate = ffs_truncate;
700	ump->um_update = ffs_update;
701	ump->um_valloc = ffs_valloc;
702	ump->um_vfree = ffs_vfree;
703	ump->um_ifree = ffs_ifree;
704	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
705	if (fs->fs_sbsize < SBLOCKSIZE)
706		bp->b_flags |= B_INVAL | B_NOCACHE;
707	brelse(bp);
708	bp = NULL;
709	fs = ump->um_fs;
710	ffs_oldfscompat_read(fs, ump, sblockloc);
711	fs->fs_ronly = ronly;
712	size = fs->fs_cssize;
713	blks = howmany(size, fs->fs_fsize);
714	if (fs->fs_contigsumsize > 0)
715		size += fs->fs_ncg * sizeof(int32_t);
716	size += fs->fs_ncg * sizeof(u_int8_t);
717	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
718	fs->fs_csp = space;
719	for (i = 0; i < blks; i += fs->fs_frag) {
720		size = fs->fs_bsize;
721		if (i + fs->fs_frag > blks)
722			size = (blks - i) * fs->fs_fsize;
723		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
724		    cred, &bp)) != 0) {
725			free(fs->fs_csp, M_UFSMNT);
726			goto out;
727		}
728		bcopy(bp->b_data, space, (u_int)size);
729		space = (char *)space + size;
730		brelse(bp);
731		bp = NULL;
732	}
733	if (fs->fs_contigsumsize > 0) {
734		fs->fs_maxcluster = lp = space;
735		for (i = 0; i < fs->fs_ncg; i++)
736			*lp++ = fs->fs_contigsumsize;
737		space = lp;
738	}
739	size = fs->fs_ncg * sizeof(u_int8_t);
740	fs->fs_contigdirs = (u_int8_t *)space;
741	bzero(fs->fs_contigdirs, size);
742	fs->fs_active = NULL;
743	mp->mnt_data = (qaddr_t)ump;
744	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
745	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
746	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
747	    vfs_getvfs(&mp->mnt_stat.f_fsid))
748		vfs_getnewfsid(mp);
749	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
750	mp->mnt_flag |= MNT_LOCAL;
751	if ((fs->fs_flags & FS_MULTILABEL) != 0)
752#ifdef MAC
753		mp->mnt_flag |= MNT_MULTILABEL;
754#else
755		printf(
756"WARNING: %s: multilabel flag on fs but no MAC support\n",
757		    fs->fs_fsmnt);
758#endif
759	if ((fs->fs_flags & FS_ACLS) != 0)
760#ifdef UFS_ACL
761		mp->mnt_flag |= MNT_ACLS;
762#else
763		printf(
764"WARNING: %s: ACLs flag on fs but no ACLs support\n",
765		    fs->fs_fsmnt);
766#endif
767	ump->um_mountp = mp;
768	ump->um_dev = dev;
769	ump->um_devvp = devvp;
770	ump->um_nindir = fs->fs_nindir;
771	ump->um_bptrtodb = fs->fs_fsbtodb;
772	ump->um_seqinc = fs->fs_frag;
773	for (i = 0; i < MAXQUOTAS; i++)
774		ump->um_quotas[i] = NULLVP;
775#ifdef UFS_EXTATTR
776	ufs_extattr_uepm_init(&ump->um_extattr);
777#endif
778	devvp->v_rdev->si_mountpoint = mp;
779
780	/*
781	 * Set FS local "last mounted on" information (NULL pad)
782	 */
783	copystr(	mp->mnt_stat.f_mntonname,	/* mount point*/
784			fs->fs_fsmnt,			/* copy area*/
785			sizeof(fs->fs_fsmnt) - 1,	/* max size*/
786			&strsize);			/* real size*/
787	bzero( fs->fs_fsmnt + strsize, sizeof(fs->fs_fsmnt) - strsize);
788
789	if( mp->mnt_flag & MNT_ROOTFS) {
790		/*
791		 * Root mount; update timestamp in mount structure.
792		 * this will be used by the common root mount code
793		 * to update the system clock.
794		 */
795		mp->mnt_time = fs->fs_time;
796	}
797
798	if (ronly == 0) {
799		if ((fs->fs_flags & FS_DOSOFTDEP) &&
800		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
801			free(fs->fs_csp, M_UFSMNT);
802			goto out;
803		}
804		if (fs->fs_snapinum[0] != 0)
805			ffs_snapshot_mount(mp);
806		fs->fs_fmod = 1;
807		fs->fs_clean = 0;
808		(void) ffs_sbupdate(ump, MNT_WAIT);
809	}
810#ifdef UFS_EXTATTR
811#ifdef UFS_EXTATTR_AUTOSTART
812	/*
813	 *
814	 * Auto-starting does the following:
815	 *	- check for /.attribute in the fs, and extattr_start if so
816	 *	- for each file in .attribute, enable that file with
817	 * 	  an attribute of the same name.
818	 * Not clear how to report errors -- probably eat them.
819	 * This would all happen while the filesystem was busy/not
820	 * available, so would effectively be "atomic".
821	 */
822	(void) ufs_extattr_autostart(mp, td);
823#endif /* !UFS_EXTATTR_AUTOSTART */
824#endif /* !UFS_EXTATTR */
825	return (0);
826out:
827	devvp->v_rdev->si_mountpoint = NULL;
828	if (bp)
829		brelse(bp);
830	/* XXX: see comment above VOP_OPEN */
831#ifdef notyet
832	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, td);
833#else
834	(void)VOP_CLOSE(devvp, FREAD|FWRITE, cred, td);
835#endif
836	if (ump) {
837		free(ump->um_fs, M_UFSMNT);
838		free(ump, M_UFSMNT);
839		mp->mnt_data = (qaddr_t)0;
840	}
841	return (error);
842}
843
844#include <sys/sysctl.h>
845int bigcgs = 0;
846SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
847
848/*
849 * Sanity checks for loading old filesystem superblocks.
850 * See ffs_oldfscompat_write below for unwound actions.
851 *
852 * XXX - Parts get retired eventually.
853 * Unfortunately new bits get added.
854 */
855static void
856ffs_oldfscompat_read(fs, ump, sblockloc)
857	struct fs *fs;
858	struct ufsmount *ump;
859	ufs2_daddr_t sblockloc;
860{
861	off_t maxfilesize;
862
863	/*
864	 * If not yet done, update fs_flags location and value of fs_sblockloc.
865	 */
866	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
867		fs->fs_flags = fs->fs_old_flags;
868		fs->fs_old_flags |= FS_FLAGS_UPDATED;
869		fs->fs_sblockloc = sblockloc;
870	}
871	/*
872	 * If not yet done, update UFS1 superblock with new wider fields.
873	 */
874	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
875		fs->fs_maxbsize = fs->fs_bsize;
876		fs->fs_time = fs->fs_old_time;
877		fs->fs_size = fs->fs_old_size;
878		fs->fs_dsize = fs->fs_old_dsize;
879		fs->fs_csaddr = fs->fs_old_csaddr;
880		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
881		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
882		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
883		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
884	}
885	if (fs->fs_magic == FS_UFS1_MAGIC &&
886	    fs->fs_old_inodefmt < FS_44INODEFMT) {
887		fs->fs_maxfilesize = (u_quad_t) 1LL << 39;
888		fs->fs_qbmask = ~fs->fs_bmask;
889		fs->fs_qfmask = ~fs->fs_fmask;
890	}
891	if (fs->fs_magic == FS_UFS1_MAGIC) {
892		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
893		maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1;
894		if (fs->fs_maxfilesize > maxfilesize)
895			fs->fs_maxfilesize = maxfilesize;
896	}
897	/* Compatibility for old filesystems */
898	if (fs->fs_avgfilesize <= 0)
899		fs->fs_avgfilesize = AVFILESIZ;
900	if (fs->fs_avgfpdir <= 0)
901		fs->fs_avgfpdir = AFPDIR;
902	if (bigcgs) {
903		fs->fs_save_cgsize = fs->fs_cgsize;
904		fs->fs_cgsize = fs->fs_bsize;
905	}
906}
907
908/*
909 * Unwinding superblock updates for old filesystems.
910 * See ffs_oldfscompat_read above for details.
911 *
912 * XXX - Parts get retired eventually.
913 * Unfortunately new bits get added.
914 */
915static void
916ffs_oldfscompat_write(fs, ump)
917	struct fs *fs;
918	struct ufsmount *ump;
919{
920
921	/*
922	 * Copy back UFS2 updated fields that UFS1 inspects.
923	 */
924	if (fs->fs_magic == FS_UFS1_MAGIC) {
925		fs->fs_old_time = fs->fs_time;
926		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
927		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
928		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
929		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
930		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
931	}
932	if (bigcgs) {
933		fs->fs_cgsize = fs->fs_save_cgsize;
934		fs->fs_save_cgsize = 0;
935	}
936}
937
938/*
939 * unmount system call
940 */
941int
942ffs_unmount(mp, mntflags, td)
943	struct mount *mp;
944	int mntflags;
945	struct thread *td;
946{
947	struct ufsmount *ump = VFSTOUFS(mp);
948	struct fs *fs;
949	int error, flags;
950
951	flags = 0;
952	if (mntflags & MNT_FORCE) {
953		flags |= FORCECLOSE;
954	}
955#ifdef UFS_EXTATTR
956	if ((error = ufs_extattr_stop(mp, td))) {
957		if (error != EOPNOTSUPP)
958			printf("ffs_unmount: ufs_extattr_stop returned %d\n",
959			    error);
960	} else {
961		ufs_extattr_uepm_destroy(&ump->um_extattr);
962	}
963#endif
964	if (mp->mnt_flag & MNT_SOFTDEP) {
965		if ((error = softdep_flushfiles(mp, flags, td)) != 0)
966			return (error);
967	} else {
968		if ((error = ffs_flushfiles(mp, flags, td)) != 0)
969			return (error);
970	}
971	fs = ump->um_fs;
972	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
973		printf("%s: unmount pending error: blocks %jd files %d\n",
974		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
975		    fs->fs_pendinginodes);
976		fs->fs_pendingblocks = 0;
977		fs->fs_pendinginodes = 0;
978	}
979	if (fs->fs_ronly == 0) {
980		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
981		error = ffs_sbupdate(ump, MNT_WAIT);
982		if (error) {
983			fs->fs_clean = 0;
984			return (error);
985		}
986	}
987	ump->um_devvp->v_rdev->si_mountpoint = NULL;
988
989	vinvalbuf(ump->um_devvp, V_SAVE, NOCRED, td, 0, 0);
990	/* XXX: see comment above VOP_OPEN */
991#ifdef notyet
992	error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD|FWRITE,
993		NOCRED, td);
994#else
995	error = VOP_CLOSE(ump->um_devvp, FREAD|FWRITE, NOCRED, td);
996#endif
997
998	vrele(ump->um_devvp);
999
1000	free(fs->fs_csp, M_UFSMNT);
1001	free(fs, M_UFSMNT);
1002	free(ump, M_UFSMNT);
1003	mp->mnt_data = (qaddr_t)0;
1004	mp->mnt_flag &= ~MNT_LOCAL;
1005	return (error);
1006}
1007
1008/*
1009 * Flush out all the files in a filesystem.
1010 */
1011int
1012ffs_flushfiles(mp, flags, td)
1013	struct mount *mp;
1014	int flags;
1015	struct thread *td;
1016{
1017	struct ufsmount *ump;
1018	int error;
1019
1020	ump = VFSTOUFS(mp);
1021#ifdef QUOTA
1022	if (mp->mnt_flag & MNT_QUOTA) {
1023		int i;
1024		error = vflush(mp, 0, SKIPSYSTEM|flags);
1025		if (error)
1026			return (error);
1027		for (i = 0; i < MAXQUOTAS; i++) {
1028			if (ump->um_quotas[i] == NULLVP)
1029				continue;
1030			quotaoff(td, mp, i);
1031		}
1032		/*
1033		 * Here we fall through to vflush again to ensure
1034		 * that we have gotten rid of all the system vnodes.
1035		 */
1036	}
1037#endif
1038	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1039	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1040		if ((error = vflush(mp, 0, SKIPSYSTEM | flags)) != 0)
1041			return (error);
1042		ffs_snapshot_unmount(mp);
1043		/*
1044		 * Here we fall through to vflush again to ensure
1045		 * that we have gotten rid of all the system vnodes.
1046		 */
1047	}
1048        /*
1049	 * Flush all the files.
1050	 */
1051	if ((error = vflush(mp, 0, flags)) != 0)
1052		return (error);
1053	/*
1054	 * Flush filesystem metadata.
1055	 */
1056	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td);
1057	error = VOP_FSYNC(ump->um_devvp, td->td_ucred, MNT_WAIT, td);
1058	VOP_UNLOCK(ump->um_devvp, 0, td);
1059	return (error);
1060}
1061
1062/*
1063 * Get filesystem statistics.
1064 */
1065int
1066ffs_statfs(mp, sbp, td)
1067	struct mount *mp;
1068	struct statfs *sbp;
1069	struct thread *td;
1070{
1071	struct ufsmount *ump;
1072	struct fs *fs;
1073
1074	ump = VFSTOUFS(mp);
1075	fs = ump->um_fs;
1076	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1077		panic("ffs_statfs");
1078	sbp->f_bsize = fs->fs_fsize;
1079	sbp->f_iosize = fs->fs_bsize;
1080	sbp->f_blocks = fs->fs_dsize;
1081	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1082	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1083	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1084	    dbtofsb(fs, fs->fs_pendingblocks);
1085	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1086	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1087	if (sbp != &mp->mnt_stat) {
1088		sbp->f_type = mp->mnt_vfc->vfc_typenum;
1089		bcopy((caddr_t)mp->mnt_stat.f_mntonname,
1090			(caddr_t)&sbp->f_mntonname[0], MNAMELEN);
1091		bcopy((caddr_t)mp->mnt_stat.f_mntfromname,
1092			(caddr_t)&sbp->f_mntfromname[0], MNAMELEN);
1093	}
1094	return (0);
1095}
1096
1097/*
1098 * Go through the disk queues to initiate sandbagged IO;
1099 * go through the inodes to write those that have been modified;
1100 * initiate the writing of the super block if it has been modified.
1101 *
1102 * Note: we are always called with the filesystem marked `MPBUSY'.
1103 */
1104int
1105ffs_sync(mp, waitfor, cred, td)
1106	struct mount *mp;
1107	int waitfor;
1108	struct ucred *cred;
1109	struct thread *td;
1110{
1111	struct vnode *nvp, *vp, *devvp;
1112	struct inode *ip;
1113	struct ufsmount *ump = VFSTOUFS(mp);
1114	struct fs *fs;
1115	int error, count, wait, lockreq, allerror = 0;
1116
1117	fs = ump->um_fs;
1118	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
1119		printf("fs = %s\n", fs->fs_fsmnt);
1120		panic("ffs_sync: rofs mod");
1121	}
1122	/*
1123	 * Write back each (modified) inode.
1124	 */
1125	wait = 0;
1126	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1127	if (waitfor == MNT_WAIT) {
1128		wait = 1;
1129		lockreq = LK_EXCLUSIVE;
1130	}
1131	lockreq |= LK_INTERLOCK;
1132	mtx_lock(&mntvnode_mtx);
1133loop:
1134	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
1135		/*
1136		 * If the vnode that we are about to sync is no longer
1137		 * associated with this mount point, start over.
1138		 */
1139		if (vp->v_mount != mp)
1140			goto loop;
1141
1142		/*
1143		 * Depend on the mntvnode_slock to keep things stable enough
1144		 * for a quick test.  Since there might be hundreds of
1145		 * thousands of vnodes, we cannot afford even a subroutine
1146		 * call unless there's a good chance that we have work to do.
1147		 */
1148		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1149		VI_LOCK(vp);
1150		if (vp->v_iflag & VI_XLOCK) {
1151			VI_UNLOCK(vp);
1152			continue;
1153		}
1154		ip = VTOI(vp);
1155		if (vp->v_type == VNON || ((ip->i_flag &
1156		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1157		    TAILQ_EMPTY(&vp->v_dirtyblkhd))) {
1158			VI_UNLOCK(vp);
1159			continue;
1160		}
1161		mtx_unlock(&mntvnode_mtx);
1162		if ((error = vget(vp, lockreq, td)) != 0) {
1163			mtx_lock(&mntvnode_mtx);
1164			if (error == ENOENT)
1165				goto loop;
1166			continue;
1167		}
1168		if ((error = VOP_FSYNC(vp, cred, waitfor, td)) != 0)
1169			allerror = error;
1170		VOP_UNLOCK(vp, 0, td);
1171		vrele(vp);
1172		mtx_lock(&mntvnode_mtx);
1173		if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp)
1174			goto loop;
1175	}
1176	mtx_unlock(&mntvnode_mtx);
1177	/*
1178	 * Force stale filesystem control information to be flushed.
1179	 */
1180	if (waitfor == MNT_WAIT) {
1181		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1182			allerror = error;
1183		/* Flushed work items may create new vnodes to clean */
1184		if (allerror == 0 && count) {
1185			mtx_lock(&mntvnode_mtx);
1186			goto loop;
1187		}
1188	}
1189#ifdef QUOTA
1190	qsync(mp);
1191#endif
1192	devvp = ump->um_devvp;
1193	VI_LOCK(devvp);
1194	if (waitfor != MNT_LAZY &&
1195	    (devvp->v_numoutput > 0 || TAILQ_FIRST(&devvp->v_dirtyblkhd))) {
1196		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td);
1197		if ((error = VOP_FSYNC(devvp, cred, waitfor, td)) != 0)
1198			allerror = error;
1199		VOP_UNLOCK(devvp, 0, td);
1200		if (allerror == 0 && waitfor == MNT_WAIT) {
1201			mtx_lock(&mntvnode_mtx);
1202			goto loop;
1203		}
1204	} else
1205		VI_UNLOCK(devvp);
1206	/*
1207	 * Write back modified superblock.
1208	 */
1209	if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor)) != 0)
1210		allerror = error;
1211	return (allerror);
1212}
1213
1214int
1215ffs_vget(mp, ino, flags, vpp)
1216	struct mount *mp;
1217	ino_t ino;
1218	int flags;
1219	struct vnode **vpp;
1220{
1221	struct thread *td = curthread; 		/* XXX */
1222	struct fs *fs;
1223	struct inode *ip;
1224	struct ufsmount *ump;
1225	struct buf *bp;
1226	struct vnode *vp;
1227	dev_t dev;
1228	int error;
1229
1230	ump = VFSTOUFS(mp);
1231	dev = ump->um_dev;
1232
1233	/*
1234	 * We do not lock vnode creation as it is believed to be too
1235	 * expensive for such rare case as simultaneous creation of vnode
1236	 * for same ino by different processes. We just allow them to race
1237	 * and check later to decide who wins. Let the race begin!
1238	 */
1239	if ((error = ufs_ihashget(dev, ino, flags, vpp)) != 0)
1240		return (error);
1241	if (*vpp != NULL)
1242		return (0);
1243
1244	/*
1245	 * If this MALLOC() is performed after the getnewvnode()
1246	 * it might block, leaving a vnode with a NULL v_data to be
1247	 * found by ffs_sync() if a sync happens to fire right then,
1248	 * which will cause a panic because ffs_sync() blindly
1249	 * dereferences vp->v_data (as well it should).
1250	 */
1251	ip = uma_zalloc(uma_inode, M_WAITOK);
1252
1253	/* Allocate a new vnode/inode. */
1254	error = getnewvnode("ufs", mp, ffs_vnodeop_p, &vp);
1255	if (error) {
1256		*vpp = NULL;
1257		uma_zfree(uma_inode, ip);
1258		return (error);
1259	}
1260	bzero((caddr_t)ip, sizeof(struct inode));
1261	/*
1262	 * FFS supports recursive locking.
1263	 */
1264	vp->v_vnlock->lk_flags |= LK_CANRECURSE;
1265	vp->v_data = ip;
1266	ip->i_vnode = vp;
1267	ip->i_ump = ump;
1268	ip->i_fs = fs = ump->um_fs;
1269	ip->i_dev = dev;
1270	ip->i_number = ino;
1271#ifdef QUOTA
1272	{
1273		int i;
1274		for (i = 0; i < MAXQUOTAS; i++)
1275			ip->i_dquot[i] = NODQUOT;
1276	}
1277#endif
1278	/*
1279	 * Exclusively lock the vnode before adding to hash. Note, that we
1280	 * must not release nor downgrade the lock (despite flags argument
1281	 * says) till it is fully initialized.
1282	 */
1283	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, (struct mtx *)0, td);
1284
1285	/*
1286	 * Atomicaly (in terms of ufs_hash operations) check the hash for
1287	 * duplicate of vnode being created and add it to the hash. If a
1288	 * duplicate vnode was found, it will be vget()ed from hash for us.
1289	 */
1290	if ((error = ufs_ihashins(ip, flags, vpp)) != 0) {
1291		vput(vp);
1292		*vpp = NULL;
1293		return (error);
1294	}
1295
1296	/* We lost the race, then throw away our vnode and return existing */
1297	if (*vpp != NULL) {
1298		vput(vp);
1299		return (0);
1300	}
1301
1302	/* Read in the disk contents for the inode, copy into the inode. */
1303	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1304	    (int)fs->fs_bsize, NOCRED, &bp);
1305	if (error) {
1306		/*
1307		 * The inode does not contain anything useful, so it would
1308		 * be misleading to leave it on its hash chain. With mode
1309		 * still zero, it will be unlinked and returned to the free
1310		 * list by vput().
1311		 */
1312		brelse(bp);
1313		vput(vp);
1314		*vpp = NULL;
1315		return (error);
1316	}
1317	if (ip->i_ump->um_fstype == UFS1)
1318		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1319	else
1320		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1321	ffs_load_inode(bp, ip, fs, ino);
1322	if (DOINGSOFTDEP(vp))
1323		softdep_load_inodeblock(ip);
1324	else
1325		ip->i_effnlink = ip->i_nlink;
1326	bqrelse(bp);
1327
1328	/*
1329	 * Initialize the vnode from the inode, check for aliases.
1330	 * Note that the underlying vnode may have changed.
1331	 */
1332	error = ufs_vinit(mp, ffs_specop_p, ffs_fifoop_p, &vp);
1333	if (error) {
1334		vput(vp);
1335		*vpp = NULL;
1336		return (error);
1337	}
1338	/*
1339	 * Finish inode initialization.
1340	 */
1341	VREF(ip->i_devvp);
1342	/*
1343	 * Set up a generation number for this inode if it does not
1344	 * already have one. This should only happen on old filesystems.
1345	 */
1346	if (ip->i_gen == 0) {
1347		ip->i_gen = arc4random() / 2 + 1;
1348		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1349			ip->i_flag |= IN_MODIFIED;
1350			DIP(ip, i_gen) = ip->i_gen;
1351		}
1352	}
1353	/*
1354	 * Ensure that uid and gid are correct. This is a temporary
1355	 * fix until fsck has been changed to do the update.
1356	 */
1357	if (fs->fs_magic == FS_UFS1_MAGIC &&		/* XXX */
1358	    fs->fs_old_inodefmt < FS_44INODEFMT) {	/* XXX */
1359		ip->i_uid = ip->i_din1->di_ouid;	/* XXX */
1360		ip->i_gid = ip->i_din1->di_ogid;	/* XXX */
1361	}						/* XXX */
1362
1363#ifdef MAC
1364	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1365		/*
1366		 * If this vnode is already allocated, and we're running
1367		 * multi-label, attempt to perform a label association
1368		 * from the extended attributes on the inode.
1369		 */
1370		error = mac_associate_vnode_extattr(mp, vp);
1371		if (error) {
1372			/* ufs_inactive will release ip->i_devvp ref. */
1373			vput(vp);
1374			*vpp = NULL;
1375			return (error);
1376		}
1377	}
1378#endif
1379
1380	*vpp = vp;
1381	return (0);
1382}
1383
1384/*
1385 * File handle to vnode
1386 *
1387 * Have to be really careful about stale file handles:
1388 * - check that the inode number is valid
1389 * - call ffs_vget() to get the locked inode
1390 * - check for an unallocated inode (i_mode == 0)
1391 * - check that the given client host has export rights and return
1392 *   those rights via. exflagsp and credanonp
1393 */
1394int
1395ffs_fhtovp(mp, fhp, vpp)
1396	struct mount *mp;
1397	struct fid *fhp;
1398	struct vnode **vpp;
1399{
1400	struct ufid *ufhp;
1401	struct fs *fs;
1402
1403	ufhp = (struct ufid *)fhp;
1404	fs = VFSTOUFS(mp)->um_fs;
1405	if (ufhp->ufid_ino < ROOTINO ||
1406	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1407		return (ESTALE);
1408	return (ufs_fhtovp(mp, ufhp, vpp));
1409}
1410
1411/*
1412 * Vnode pointer to File handle
1413 */
1414/* ARGSUSED */
1415int
1416ffs_vptofh(vp, fhp)
1417	struct vnode *vp;
1418	struct fid *fhp;
1419{
1420	struct inode *ip;
1421	struct ufid *ufhp;
1422
1423	ip = VTOI(vp);
1424	ufhp = (struct ufid *)fhp;
1425	ufhp->ufid_len = sizeof(struct ufid);
1426	ufhp->ufid_ino = ip->i_number;
1427	ufhp->ufid_gen = ip->i_gen;
1428	return (0);
1429}
1430
1431/*
1432 * Initialize the filesystem.
1433 */
1434static int
1435ffs_init(vfsp)
1436	struct vfsconf *vfsp;
1437{
1438
1439	softdep_initialize();
1440	return (ufs_init(vfsp));
1441}
1442
1443/*
1444 * Undo the work of ffs_init().
1445 */
1446static int
1447ffs_uninit(vfsp)
1448	struct vfsconf *vfsp;
1449{
1450	int ret;
1451
1452	ret = ufs_uninit(vfsp);
1453	softdep_uninitialize();
1454	return (ret);
1455}
1456
1457/*
1458 * Write a superblock and associated information back to disk.
1459 */
1460static int
1461ffs_sbupdate(mp, waitfor)
1462	struct ufsmount *mp;
1463	int waitfor;
1464{
1465	struct fs *fs = mp->um_fs;
1466	struct buf *bp;
1467	int blks;
1468	void *space;
1469	int i, size, error, allerror = 0;
1470
1471	if (fs->fs_ronly == 1 &&
1472	    (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1473	    (MNT_RDONLY | MNT_UPDATE))
1474		panic("ffs_sbupdate: write read-only filesystem");
1475	/*
1476	 * First write back the summary information.
1477	 */
1478	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1479	space = fs->fs_csp;
1480	for (i = 0; i < blks; i += fs->fs_frag) {
1481		size = fs->fs_bsize;
1482		if (i + fs->fs_frag > blks)
1483			size = (blks - i) * fs->fs_fsize;
1484		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1485		    size, 0, 0, 0);
1486		bcopy(space, bp->b_data, (u_int)size);
1487		space = (char *)space + size;
1488		if (waitfor != MNT_WAIT)
1489			bawrite(bp);
1490		else if ((error = bwrite(bp)) != 0)
1491			allerror = error;
1492	}
1493	/*
1494	 * Now write back the superblock itself. If any errors occurred
1495	 * up to this point, then fail so that the superblock avoids
1496	 * being written out as clean.
1497	 */
1498	if (allerror)
1499		return (allerror);
1500	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1501	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1502		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1503		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1504		fs->fs_sblockloc = SBLOCK_UFS1;
1505	}
1506	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1507	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1508		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1509		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1510		fs->fs_sblockloc = SBLOCK_UFS2;
1511	}
1512	bp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize,
1513	    0, 0, 0);
1514	fs->fs_fmod = 0;
1515	fs->fs_time = time_second;
1516	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1517	ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
1518	if (waitfor != MNT_WAIT)
1519		bawrite(bp);
1520	else if ((error = bwrite(bp)) != 0)
1521		allerror = error;
1522	return (allerror);
1523}
1524
1525static int
1526ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1527	int attrnamespace, const char *attrname, struct thread *td)
1528{
1529
1530#ifdef UFS_EXTATTR
1531	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1532	    attrname, td));
1533#else
1534	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1535	    attrname, td));
1536#endif
1537}
1538
1539static void
1540ffs_ifree(struct ufsmount *ump, struct inode *ip)
1541{
1542
1543	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1544		uma_zfree(uma_ufs1, ip->i_din1);
1545	else if (ip->i_din2 != NULL)
1546		uma_zfree(uma_ufs2, ip->i_din2);
1547	uma_zfree(uma_inode, ip);
1548}
1549