xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision ee2ea5ceafed78a5bd9810beb9e3ca927180c226)
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  * $FreeBSD$
35  */
36 
37 #include "opt_quota.h"
38 #include "opt_ufs.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
43 #include <sys/proc.h>
44 #include <sys/kernel.h>
45 #include <sys/vnode.h>
46 #include <sys/mount.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/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/vm_page.h>
66 
67 static MALLOC_DEFINE(M_FFSNODE, "FFS node", "FFS vnode private part");
68 
69 static int	ffs_sbupdate(struct ufsmount *, int);
70 int	ffs_reload(struct mount *,struct ucred *,struct thread *);
71 static int	ffs_oldfscompat(struct fs *);
72 static int	ffs_init(struct vfsconf *);
73 
74 static struct vfsops ufs_vfsops = {
75 	ffs_mount,
76 	ufs_start,
77 	ffs_unmount,
78 	ufs_root,
79 	ufs_quotactl,
80 	ffs_statfs,
81 	ffs_sync,
82 	ffs_vget,
83 	ffs_fhtovp,
84 	vfs_stdcheckexp,
85 	ffs_vptofh,
86 	ffs_init,
87 	vfs_stduninit,
88 #ifdef UFS_EXTATTR
89 	ufs_extattrctl,
90 #else
91 	vfs_stdextattrctl,
92 #endif
93 };
94 
95 VFS_SET(ufs_vfsops, ufs, 0);
96 
97 /*
98  * ffs_mount
99  *
100  * Called when mounting local physical media
101  *
102  * PARAMETERS:
103  *		mountroot
104  *			mp	mount point structure
105  *			path	NULL (flag for root mount!!!)
106  *			data	<unused>
107  *			ndp	<unused>
108  *			p	process (user credentials check [statfs])
109  *
110  *		mount
111  *			mp	mount point structure
112  *			path	path to mount point
113  *			data	pointer to argument struct in user space
114  *			ndp	mount point namei() return (used for
115  *				credentials on reload), reused to look
116  *				up block device.
117  *			p	process (user credentials check)
118  *
119  * RETURNS:	0	Success
120  *		!0	error number (errno.h)
121  *
122  * LOCK STATE:
123  *
124  *		ENTRY
125  *			mount point is locked
126  *		EXIT
127  *			mount point is locked
128  *
129  * NOTES:
130  *		A NULL path can be used for a flag since the mount
131  *		system call will fail with EFAULT in copyinstr in
132  *		namei() if it is a genuine NULL from the user.
133  */
134 int
135 ffs_mount(mp, path, data, ndp, td)
136         struct mount		*mp;	/* mount struct pointer*/
137         char			*path;	/* path to mount point*/
138         caddr_t			data;	/* arguments to FS specific mount*/
139         struct nameidata	*ndp;	/* mount point credentials*/
140         struct thread		*td;	/* process requesting mount*/
141 {
142 	size_t		size;
143 	struct vnode	*devvp;
144 	struct ufs_args args;
145 	struct ufsmount *ump = 0;
146 	register struct fs *fs;
147 	int error, flags;
148 	mode_t accessmode;
149 
150 	/*
151 	 * Use NULL path to indicate we are mounting the root file system.
152 	 */
153 	if (path == NULL) {
154 		if ((error = bdevvp(rootdev, &rootvp))) {
155 			printf("ffs_mountroot: can't find rootvp\n");
156 			return (error);
157 		}
158 
159 		if ((error = ffs_mountfs(rootvp, mp, td, M_FFSNODE)) != 0)
160 			return (error);
161 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
162 		return (0);
163 	}
164 
165 	/*
166 	 * Mounting non-root file system or updating a file system
167 	 */
168 	if ((error = copyin(data, (caddr_t)&args, sizeof(struct ufs_args)))!= 0)
169 		return (error);
170 
171 	/*
172 	 * If updating, check whether changing from read-only to
173 	 * read/write; if there is no device name, that's all we do.
174 	 */
175 	if (mp->mnt_flag & MNT_UPDATE) {
176 		ump = VFSTOUFS(mp);
177 		fs = ump->um_fs;
178 		devvp = ump->um_devvp;
179 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
180 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
181 				return (error);
182 			/*
183 			 * Flush any dirty data.
184 			 */
185 			VFS_SYNC(mp, MNT_WAIT, td->td_proc->p_ucred, td);
186 			/*
187 			 * Check for and optionally get rid of files open
188 			 * for writing.
189 			 */
190 			flags = WRITECLOSE;
191 			if (mp->mnt_flag & MNT_FORCE)
192 				flags |= FORCECLOSE;
193 			if (mp->mnt_flag & MNT_SOFTDEP) {
194 				error = softdep_flushfiles(mp, flags, td);
195 			} else {
196 				error = ffs_flushfiles(mp, flags, td);
197 			}
198 			if (error) {
199 				vn_finished_write(mp);
200 				return (error);
201 			}
202 			if (fs->fs_pendingblocks != 0 ||
203 			    fs->fs_pendinginodes != 0) {
204 				printf("%s: update error: blocks %d files %d\n",
205 				    fs->fs_fsmnt, fs->fs_pendingblocks,
206 				    fs->fs_pendinginodes);
207 				fs->fs_pendingblocks = 0;
208 				fs->fs_pendinginodes = 0;
209 			}
210 			fs->fs_ronly = 1;
211 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
212 				fs->fs_clean = 1;
213 			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
214 				fs->fs_ronly = 0;
215 				fs->fs_clean = 0;
216 				vn_finished_write(mp);
217 				return (error);
218 			}
219 			vn_finished_write(mp);
220 		}
221 		if ((mp->mnt_flag & MNT_RELOAD) &&
222 		    (error = ffs_reload(mp, ndp->ni_cnd.cn_cred, td)) != 0)
223 			return (error);
224 		if (fs->fs_ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
225 			/*
226 			 * If upgrade to read-write by non-root, then verify
227 			 * that user has necessary permissions on the device.
228 			 */
229 			if (suser(td)) {
230 				vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
231 				if ((error = VOP_ACCESS(devvp, VREAD | VWRITE,
232 				    td->td_ucred, td)) != 0) {
233 					VOP_UNLOCK(devvp, 0, td);
234 					return (error);
235 				}
236 				VOP_UNLOCK(devvp, 0, td);
237 			}
238 			fs->fs_flags &= ~FS_UNCLEAN;
239 			if (fs->fs_clean == 0) {
240 				fs->fs_flags |= FS_UNCLEAN;
241 				if ((mp->mnt_flag & MNT_FORCE) ||
242 				    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
243 				     (fs->fs_flags & FS_DOSOFTDEP))) {
244 					printf("WARNING: %s was not %s\n",
245 					   fs->fs_fsmnt, "properly dismounted");
246 				} else {
247 					printf(
248 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
249 					    fs->fs_fsmnt);
250 					return (EPERM);
251 				}
252 			}
253 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
254 				return (error);
255 			fs->fs_ronly = 0;
256 			fs->fs_clean = 0;
257 			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
258 				vn_finished_write(mp);
259 				return (error);
260 			}
261 			/* check to see if we need to start softdep */
262 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
263 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
264 				vn_finished_write(mp);
265 				return (error);
266 			}
267 			if (fs->fs_snapinum[0] != 0)
268 				ffs_snapshot_mount(mp);
269 			vn_finished_write(mp);
270 		}
271 		/*
272 		 * Soft updates is incompatible with "async",
273 		 * so if we are doing softupdates stop the user
274 		 * from setting the async flag in an update.
275 		 * Softdep_mount() clears it in an initial mount
276 		 * or ro->rw remount.
277 		 */
278 		if (mp->mnt_flag & MNT_SOFTDEP)
279 			mp->mnt_flag &= ~MNT_ASYNC;
280 		/*
281 		 * If not updating name, process export requests.
282 		 */
283 		if (args.fspec == 0)
284 			return (vfs_export(mp, &args.export));
285 		/*
286 		 * If this is a snapshot request, take the snapshot.
287 		 */
288 		if (mp->mnt_flag & MNT_SNAPSHOT)
289 			return (ffs_snapshot(mp, args.fspec));
290 	}
291 
292 	/*
293 	 * Not an update, or updating the name: look up the name
294 	 * and verify that it refers to a sensible block device.
295 	 */
296 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, td);
297 	if ((error = namei(ndp)) != 0)
298 		return (error);
299 	NDFREE(ndp, NDF_ONLY_PNBUF);
300 	devvp = ndp->ni_vp;
301 	if (!vn_isdisk(devvp, &error)) {
302 		vrele(devvp);
303 		return (error);
304 	}
305 
306 	/*
307 	 * If mount by non-root, then verify that user has necessary
308 	 * permissions on the device.
309 	 */
310 	if (suser(td)) {
311 		accessmode = VREAD;
312 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
313 			accessmode |= VWRITE;
314 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
315 		if ((error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td))!= 0){
316 			vput(devvp);
317 			return (error);
318 		}
319 		VOP_UNLOCK(devvp, 0, td);
320 	}
321 
322 	if (mp->mnt_flag & MNT_UPDATE) {
323 		/*
324 		 * Update only
325 		 *
326 		 * If it's not the same vnode, or at least the same device
327 		 * then it's not correct.
328 		 */
329 
330 		if (devvp != ump->um_devvp &&
331 		    devvp->v_rdev != ump->um_devvp->v_rdev)
332 			error = EINVAL;	/* needs translation */
333 		vrele(devvp);
334 		if (error)
335 			return (error);
336 	} else {
337 		/*
338 		 * New mount
339 		 *
340 		 * We need the name for the mount point (also used for
341 		 * "last mounted on") copied in. If an error occurs,
342 		 * the mount point is discarded by the upper level code.
343 		 * Note that vfs_mount() populates f_mntonname for us.
344 		 */
345 		if ((error = ffs_mountfs(devvp, mp, td, M_FFSNODE)) != 0) {
346 			vrele(devvp);
347 			return (error);
348 		}
349 	}
350 	/*
351 	 * Save "mounted from" device name info for mount point (NULL pad).
352 	 */
353 	copyinstr(args.fspec, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
354 	bzero( mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
355 	/*
356 	 * Initialize filesystem stat information in mount struct.
357 	 */
358 	(void)VFS_STATFS(mp, &mp->mnt_stat, td);
359 	return (0);
360 }
361 
362 /*
363  * Reload all incore data for a filesystem (used after running fsck on
364  * the root filesystem and finding things to fix). The filesystem must
365  * be mounted read-only.
366  *
367  * Things to do to update the mount:
368  *	1) invalidate all cached meta-data.
369  *	2) re-read superblock from disk.
370  *	3) re-read summary information from disk.
371  *	4) invalidate all inactive vnodes.
372  *	5) invalidate all cached file data.
373  *	6) re-read inode data for all active vnodes.
374  */
375 int
376 ffs_reload(mp, cred, td)
377 	register struct mount *mp;
378 	struct ucred *cred;
379 	struct thread *td;
380 {
381 	register struct vnode *vp, *nvp, *devvp;
382 	struct inode *ip;
383 	void *space;
384 	struct buf *bp;
385 	struct fs *fs, *newfs;
386 	dev_t dev;
387 	int i, blks, size, error;
388 	int32_t *lp;
389 	u_int sectorsize;
390 
391 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
392 		return (EINVAL);
393 	/*
394 	 * Step 1: invalidate all cached meta-data.
395 	 */
396 	devvp = VFSTOUFS(mp)->um_devvp;
397 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
398 	error = vinvalbuf(devvp, 0, cred, td, 0, 0);
399 	VOP_UNLOCK(devvp, 0, td);
400 	if (error)
401 		panic("ffs_reload: dirty1");
402 
403 	dev = devvp->v_rdev;
404 
405 	/*
406 	 * Only VMIO the backing device if the backing device is a real
407 	 * block device.
408 	 */
409 	if (vn_isdisk(devvp, NULL)) {
410 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
411 		vfs_object_create(devvp, td, td->td_ucred);
412 		mtx_lock(&devvp->v_interlock);
413 		VOP_UNLOCK(devvp, LK_INTERLOCK, td);
414 	}
415 
416 	/*
417 	 * Step 2: re-read superblock from disk.
418 	 */
419 	if (VOP_IOCTL(devvp, DIOCGSECTORSIZE, (caddr_t)&sectorsize,
420 	    FREAD, NOCRED, td) != 0)
421 		size = DEV_BSIZE;
422 	else
423 		size = sectorsize;
424 	if ((error = bread(devvp, (ufs_daddr_t)(SBOFF/size), SBSIZE, NOCRED,&bp)) != 0)
425 		return (error);
426 	newfs = (struct fs *)bp->b_data;
427 	if (newfs->fs_magic != FS_MAGIC || newfs->fs_bsize > MAXBSIZE ||
428 		newfs->fs_bsize < sizeof(struct fs)) {
429 			brelse(bp);
430 			return (EIO);		/* XXX needs translation */
431 	}
432 	fs = VFSTOUFS(mp)->um_fs;
433 	/*
434 	 * Copy pointer fields back into superblock before copying in	XXX
435 	 * new superblock. These should really be in the ufsmount.	XXX
436 	 * Note that important parameters (eg fs_ncg) are unchanged.
437 	 */
438 	newfs->fs_csp = fs->fs_csp;
439 	newfs->fs_maxcluster = fs->fs_maxcluster;
440 	newfs->fs_contigdirs = fs->fs_contigdirs;
441 	newfs->fs_active = fs->fs_active;
442 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
443 	if (fs->fs_sbsize < SBSIZE)
444 		bp->b_flags |= B_INVAL | B_NOCACHE;
445 	brelse(bp);
446 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
447 	ffs_oldfscompat(fs);
448 	/* An old fsck may have zeroed these fields, so recheck them. */
449 	if (fs->fs_avgfilesize <= 0)		/* XXX */
450 		fs->fs_avgfilesize = AVFILESIZ;	/* XXX */
451 	if (fs->fs_avgfpdir <= 0)		/* XXX */
452 		fs->fs_avgfpdir = AFPDIR;	/* XXX */
453 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
454 		printf("%s: reload pending error: blocks %d files %d\n",
455 		    fs->fs_fsmnt, fs->fs_pendingblocks, fs->fs_pendinginodes);
456 		fs->fs_pendingblocks = 0;
457 		fs->fs_pendinginodes = 0;
458 	}
459 
460 	/*
461 	 * Step 3: re-read summary information from disk.
462 	 */
463 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
464 	space = fs->fs_csp;
465 	for (i = 0; i < blks; i += fs->fs_frag) {
466 		size = fs->fs_bsize;
467 		if (i + fs->fs_frag > blks)
468 			size = (blks - i) * fs->fs_fsize;
469 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
470 		    NOCRED, &bp);
471 		if (error)
472 			return (error);
473 		bcopy(bp->b_data, space, (u_int)size);
474 		space = (char *)space + size;
475 		brelse(bp);
476 	}
477 	/*
478 	 * We no longer know anything about clusters per cylinder group.
479 	 */
480 	if (fs->fs_contigsumsize > 0) {
481 		lp = fs->fs_maxcluster;
482 		for (i = 0; i < fs->fs_ncg; i++)
483 			*lp++ = fs->fs_contigsumsize;
484 	}
485 
486 loop:
487 	mtx_lock(&mntvnode_mtx);
488 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
489 		if (vp->v_mount != mp) {
490 			mtx_unlock(&mntvnode_mtx);
491 			goto loop;
492 		}
493 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
494 		mtx_unlock(&mntvnode_mtx);
495 		/*
496 		 * Step 4: invalidate all inactive vnodes.
497 		 */
498 		if (vrecycle(vp, NULL, td))
499 			goto loop;
500 		/*
501 		 * Step 5: invalidate all cached file data.
502 		 */
503 		mtx_lock(&vp->v_interlock);
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 			vput(vp);
518 			return (error);
519 		}
520 		ip->i_din = *((struct dinode *)bp->b_data +
521 		    ino_to_fsbo(fs, ip->i_number));
522 		ip->i_effnlink = ip->i_nlink;
523 		brelse(bp);
524 		vput(vp);
525 		mtx_lock(&mntvnode_mtx);
526 	}
527 	mtx_unlock(&mntvnode_mtx);
528 	return (0);
529 }
530 
531 #include <sys/sysctl.h>
532 int bigcgs = 0;
533 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
534 
535 /*
536  * Common code for mount and mountroot
537  */
538 int
539 ffs_mountfs(devvp, mp, td, malloctype)
540 	register struct vnode *devvp;
541 	struct mount *mp;
542 	struct thread *td;
543 	struct malloc_type *malloctype;
544 {
545 	register struct ufsmount *ump;
546 	struct buf *bp;
547 	register struct fs *fs;
548 	dev_t dev;
549 	void *space;
550 	int error, i, blks, size, ronly;
551 	int32_t *lp;
552 	struct ucred *cred;
553 	u_int64_t maxfilesize;					/* XXX */
554 	size_t strsize;
555 	int ncount;
556 	u_int sectorsize;
557 
558 	dev = devvp->v_rdev;
559 	cred = td ? td->td_ucred : NOCRED;
560 	/*
561 	 * Disallow multiple mounts of the same device.
562 	 * Disallow mounting of a device that is currently in use
563 	 * (except for root, which might share swap device for miniroot).
564 	 * Flush out any old buffers remaining from a previous use.
565 	 */
566 	error = vfs_mountedon(devvp);
567 	if (error)
568 		return (error);
569 	ncount = vcount(devvp);
570 
571 	if (ncount > 1 && devvp != rootvp)
572 		return (EBUSY);
573 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
574 	error = vinvalbuf(devvp, V_SAVE, cred, td, 0, 0);
575 	VOP_UNLOCK(devvp, 0, td);
576 	if (error)
577 		return (error);
578 
579 	/*
580 	 * Only VMIO the backing device if the backing device is a real
581 	 * block device.
582 	 * Note that it is optional that the backing device be VMIOed.  This
583 	 * increases the opportunity for metadata caching.
584 	 */
585 	if (vn_isdisk(devvp, NULL)) {
586 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
587 		vfs_object_create(devvp, td, cred);
588 		mtx_lock(&devvp->v_interlock);
589 		VOP_UNLOCK(devvp, LK_INTERLOCK, td);
590 	}
591 
592 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
593 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
594 	/*
595 	 * XXX: We don't re-VOP_OPEN in FREAD|FWRITE mode if the filesystem
596 	 * XXX: is subsequently remounted, so open it FREAD|FWRITE from the
597 	 * XXX: start to avoid getting trashed later on.
598 	 */
599 #ifdef notyet
600 	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, td);
601 #else
602 	error = VOP_OPEN(devvp, FREAD|FWRITE, FSCRED, td);
603 #endif
604 	VOP_UNLOCK(devvp, 0, td);
605 	if (error)
606 		return (error);
607 	if (devvp->v_rdev->si_iosize_max != 0)
608 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
609 	if (mp->mnt_iosize_max > MAXPHYS)
610 		mp->mnt_iosize_max = MAXPHYS;
611 
612 	if (VOP_IOCTL(devvp, DIOCGSECTORSIZE, (caddr_t)&sectorsize,
613 	    FREAD, cred, td) != 0)
614 		size = DEV_BSIZE;
615 	else
616 		size = sectorsize;
617 
618 	bp = NULL;
619 	ump = NULL;
620 	if ((error = bread(devvp, SBLOCK, SBSIZE, cred, &bp)) != 0)
621 		goto out;
622 	fs = (struct fs *)bp->b_data;
623 	if (fs->fs_magic != FS_MAGIC || fs->fs_bsize > MAXBSIZE ||
624 	    fs->fs_bsize < sizeof(struct fs)) {
625 		error = EINVAL;		/* XXX needs translation */
626 		goto out;
627 	}
628 	fs->fs_fmod = 0;
629 	fs->fs_flags &= ~FS_UNCLEAN;
630 	if (fs->fs_clean == 0) {
631 		fs->fs_flags |= FS_UNCLEAN;
632 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
633 		    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
634 		     (fs->fs_flags & FS_DOSOFTDEP))) {
635 			printf(
636 "WARNING: %s was not properly dismounted\n",
637 			    fs->fs_fsmnt);
638 		} else {
639 			printf(
640 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
641 			    fs->fs_fsmnt);
642 			error = EPERM;
643 			goto out;
644 		}
645 		if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
646 			printf("%s: lost blocks %d files %d\n", fs->fs_fsmnt,
647 			    fs->fs_pendingblocks, fs->fs_pendinginodes);
648 			fs->fs_pendingblocks = 0;
649 			fs->fs_pendinginodes = 0;
650 		}
651 	}
652 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
653 		printf("%s: mount pending error: blocks %d files %d\n",
654 		    fs->fs_fsmnt, fs->fs_pendingblocks, fs->fs_pendinginodes);
655 		fs->fs_pendingblocks = 0;
656 		fs->fs_pendinginodes = 0;
657 	}
658 	/* XXX updating 4.2 FFS superblocks trashes rotational layout tables */
659 	if (fs->fs_postblformat == FS_42POSTBLFMT && !ronly) {
660 		error = EROFS;          /* needs translation */
661 		goto out;
662 	}
663 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
664 	ump->um_malloctype = malloctype;
665 	ump->um_i_effnlink_valid = 1;
666 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT,
667 	    M_WAITOK);
668 	ump->um_blkatoff = ffs_blkatoff;
669 	ump->um_truncate = ffs_truncate;
670 	ump->um_update = ffs_update;
671 	ump->um_valloc = ffs_valloc;
672 	ump->um_vfree = ffs_vfree;
673 	ump->um_balloc = ffs_balloc;
674 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
675 	if (fs->fs_sbsize < SBSIZE)
676 		bp->b_flags |= B_INVAL | B_NOCACHE;
677 	brelse(bp);
678 	bp = NULL;
679 	fs = ump->um_fs;
680 	fs->fs_ronly = ronly;
681 	size = fs->fs_cssize;
682 	blks = howmany(size, fs->fs_fsize);
683 	if (fs->fs_contigsumsize > 0)
684 		size += fs->fs_ncg * sizeof(int32_t);
685 	size += fs->fs_ncg * sizeof(u_int8_t);
686 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
687 	fs->fs_csp = space;
688 	for (i = 0; i < blks; i += fs->fs_frag) {
689 		size = fs->fs_bsize;
690 		if (i + fs->fs_frag > blks)
691 			size = (blks - i) * fs->fs_fsize;
692 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
693 		    cred, &bp)) != 0) {
694 			free(fs->fs_csp, M_UFSMNT);
695 			goto out;
696 		}
697 		bcopy(bp->b_data, space, (u_int)size);
698 		space = (char *)space + size;
699 		brelse(bp);
700 		bp = NULL;
701 	}
702 	if (fs->fs_contigsumsize > 0) {
703 		fs->fs_maxcluster = lp = space;
704 		for (i = 0; i < fs->fs_ncg; i++)
705 			*lp++ = fs->fs_contigsumsize;
706 		space = lp;
707 	}
708 	size = fs->fs_ncg * sizeof(u_int8_t);
709 	fs->fs_contigdirs = (u_int8_t *)space;
710 	bzero(fs->fs_contigdirs, size);
711 	fs->fs_active = NULL;
712 	/* Compatibility for old filesystems 	   XXX */
713 	if (fs->fs_avgfilesize <= 0)		/* XXX */
714 		fs->fs_avgfilesize = AVFILESIZ;	/* XXX */
715 	if (fs->fs_avgfpdir <= 0)		/* XXX */
716 		fs->fs_avgfpdir = AFPDIR;	/* XXX */
717 	mp->mnt_data = (qaddr_t)ump;
718 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
719 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
720 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
721 	    vfs_getvfs(&mp->mnt_stat.f_fsid))
722 		vfs_getnewfsid(mp);
723 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
724 	mp->mnt_flag |= MNT_LOCAL;
725 	ump->um_mountp = mp;
726 	ump->um_dev = dev;
727 	ump->um_devvp = devvp;
728 	ump->um_nindir = fs->fs_nindir;
729 	ump->um_bptrtodb = fs->fs_fsbtodb;
730 	ump->um_seqinc = fs->fs_frag;
731 	for (i = 0; i < MAXQUOTAS; i++)
732 		ump->um_quotas[i] = NULLVP;
733 #ifdef UFS_EXTATTR
734 	ufs_extattr_uepm_init(&ump->um_extattr);
735 #endif
736 	devvp->v_rdev->si_mountpoint = mp;
737 	ffs_oldfscompat(fs);
738 
739 	/*
740 	 * Set FS local "last mounted on" information (NULL pad)
741 	 */
742 	copystr(	mp->mnt_stat.f_mntonname,	/* mount point*/
743 			fs->fs_fsmnt,			/* copy area*/
744 			sizeof(fs->fs_fsmnt) - 1,	/* max size*/
745 			&strsize);			/* real size*/
746 	bzero( fs->fs_fsmnt + strsize, sizeof(fs->fs_fsmnt) - strsize);
747 
748 	if( mp->mnt_flag & MNT_ROOTFS) {
749 		/*
750 		 * Root mount; update timestamp in mount structure.
751 		 * this will be used by the common root mount code
752 		 * to update the system clock.
753 		 */
754 		mp->mnt_time = fs->fs_time;
755 	}
756 
757 	ump->um_savedmaxfilesize = fs->fs_maxfilesize;		/* XXX */
758 	maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1;	/* XXX */
759 	if (fs->fs_maxfilesize > maxfilesize)			/* XXX */
760 		fs->fs_maxfilesize = maxfilesize;		/* XXX */
761 	if (bigcgs) {
762 		if (fs->fs_sparecon[0] <= 0)
763 			fs->fs_sparecon[0] = fs->fs_cgsize;
764 		fs->fs_cgsize = fs->fs_bsize;
765 	}
766 	if (ronly == 0) {
767 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
768 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
769 			free(fs->fs_csp, M_UFSMNT);
770 			goto out;
771 		}
772 		if (fs->fs_snapinum[0] != 0)
773 			ffs_snapshot_mount(mp);
774 		fs->fs_fmod = 1;
775 		fs->fs_clean = 0;
776 		(void) ffs_sbupdate(ump, MNT_WAIT);
777 	}
778 #ifdef UFS_EXTATTR
779 #ifdef UFS_EXTATTR_AUTOSTART
780 	/*
781 	 *
782 	 * Auto-starting does the following:
783 	 *	- check for /.attribute in the fs, and extattr_start if so
784 	 *	- for each file in .attribute, enable that file with
785 	 * 	  an attribute of the same name.
786 	 * Not clear how to report errors -- probably eat them.
787 	 * This would all happen while the file system was busy/not
788 	 * available, so would effectively be "atomic".
789 	 */
790 	(void) ufs_extattr_autostart(mp, td);
791 #endif /* !UFS_EXTATTR_AUTOSTART */
792 #endif /* !UFS_EXTATTR */
793 	return (0);
794 out:
795 	devvp->v_rdev->si_mountpoint = NULL;
796 	if (bp)
797 		brelse(bp);
798 	/* XXX: see comment above VOP_OPEN */
799 #ifdef notyet
800 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, td);
801 #else
802 	(void)VOP_CLOSE(devvp, FREAD|FWRITE, cred, td);
803 #endif
804 	if (ump) {
805 		free(ump->um_fs, M_UFSMNT);
806 		free(ump, M_UFSMNT);
807 		mp->mnt_data = (qaddr_t)0;
808 	}
809 	return (error);
810 }
811 
812 /*
813  * Sanity checks for old file systems.
814  *
815  * XXX - goes away some day.
816  */
817 static int
818 ffs_oldfscompat(fs)
819 	struct fs *fs;
820 {
821 
822 	fs->fs_npsect = max(fs->fs_npsect, fs->fs_nsect);	/* XXX */
823 	fs->fs_interleave = max(fs->fs_interleave, 1);		/* XXX */
824 	if (fs->fs_postblformat == FS_42POSTBLFMT)		/* XXX */
825 		fs->fs_nrpos = 8;				/* XXX */
826 	if (fs->fs_inodefmt < FS_44INODEFMT) {			/* XXX */
827 #if 0
828 		int i;						/* XXX */
829 		u_int64_t sizepb = fs->fs_bsize;		/* XXX */
830 								/* XXX */
831 		fs->fs_maxfilesize = fs->fs_bsize * NDADDR - 1;	/* XXX */
832 		for (i = 0; i < NIADDR; i++) {			/* XXX */
833 			sizepb *= NINDIR(fs);			/* XXX */
834 			fs->fs_maxfilesize += sizepb;		/* XXX */
835 		}						/* XXX */
836 #endif
837 		fs->fs_maxfilesize = (u_quad_t) 1LL << 39;
838 		fs->fs_qbmask = ~fs->fs_bmask;			/* XXX */
839 		fs->fs_qfmask = ~fs->fs_fmask;			/* XXX */
840 	}							/* XXX */
841 	return (0);
842 }
843 
844 /*
845  * unmount system call
846  */
847 int
848 ffs_unmount(mp, mntflags, td)
849 	struct mount *mp;
850 	int mntflags;
851 	struct thread *td;
852 {
853 	register struct ufsmount *ump = VFSTOUFS(mp);
854 	register struct fs *fs;
855 	int error, flags;
856 
857 	flags = 0;
858 	if (mntflags & MNT_FORCE) {
859 		flags |= FORCECLOSE;
860 	}
861 #ifdef UFS_EXTATTR
862 	if ((error = ufs_extattr_stop(mp, td))) {
863 		if (error != EOPNOTSUPP)
864 			printf("ffs_unmount: ufs_extattr_stop returned %d\n",
865 			    error);
866 	} else {
867 		ufs_extattr_uepm_destroy(&ump->um_extattr);
868 	}
869 #endif
870 	if (mp->mnt_flag & MNT_SOFTDEP) {
871 		if ((error = softdep_flushfiles(mp, flags, td)) != 0)
872 			return (error);
873 	} else {
874 		if ((error = ffs_flushfiles(mp, flags, td)) != 0)
875 			return (error);
876 	}
877 	fs = ump->um_fs;
878 	if (bigcgs) {
879 		fs->fs_cgsize = fs->fs_sparecon[0];
880 		fs->fs_sparecon[0] = 0;
881 	}
882 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
883 		printf("%s: unmount pending error: blocks %d files %d\n",
884 		    fs->fs_fsmnt, fs->fs_pendingblocks, fs->fs_pendinginodes);
885 		fs->fs_pendingblocks = 0;
886 		fs->fs_pendinginodes = 0;
887 	}
888 	if (fs->fs_ronly == 0) {
889 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
890 		error = ffs_sbupdate(ump, MNT_WAIT);
891 		if (error) {
892 			fs->fs_clean = 0;
893 			return (error);
894 		}
895 	}
896 	ump->um_devvp->v_rdev->si_mountpoint = NULL;
897 
898 	vinvalbuf(ump->um_devvp, V_SAVE, NOCRED, td, 0, 0);
899 	/* XXX: see comment above VOP_OPEN */
900 #ifdef notyet
901 	error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD|FWRITE,
902 		NOCRED, td);
903 #else
904 	error = VOP_CLOSE(ump->um_devvp, FREAD|FWRITE, NOCRED, td);
905 #endif
906 
907 	vrele(ump->um_devvp);
908 
909 	free(fs->fs_csp, M_UFSMNT);
910 	free(fs, M_UFSMNT);
911 	free(ump, M_UFSMNT);
912 	mp->mnt_data = (qaddr_t)0;
913 	mp->mnt_flag &= ~MNT_LOCAL;
914 	return (error);
915 }
916 
917 /*
918  * Flush out all the files in a filesystem.
919  */
920 int
921 ffs_flushfiles(mp, flags, td)
922 	register struct mount *mp;
923 	int flags;
924 	struct thread *td;
925 {
926 	register struct ufsmount *ump;
927 	int error;
928 
929 	ump = VFSTOUFS(mp);
930 #ifdef QUOTA
931 	if (mp->mnt_flag & MNT_QUOTA) {
932 		int i;
933 		error = vflush(mp, 0, SKIPSYSTEM|flags);
934 		if (error)
935 			return (error);
936 		for (i = 0; i < MAXQUOTAS; i++) {
937 			if (ump->um_quotas[i] == NULLVP)
938 				continue;
939 			quotaoff(td, mp, i);
940 		}
941 		/*
942 		 * Here we fall through to vflush again to ensure
943 		 * that we have gotten rid of all the system vnodes.
944 		 */
945 	}
946 #endif
947 	if (ump->um_devvp->v_flag & VCOPYONWRITE) {
948 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags)) != 0)
949 			return (error);
950 		ffs_snapshot_unmount(mp);
951 		/*
952 		 * Here we fall through to vflush again to ensure
953 		 * that we have gotten rid of all the system vnodes.
954 		 */
955 	}
956         /*
957 	 * Flush all the files.
958 	 */
959 	if ((error = vflush(mp, 0, flags)) != 0)
960 		return (error);
961 	/*
962 	 * Flush filesystem metadata.
963 	 */
964 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td);
965 	error = VOP_FSYNC(ump->um_devvp, td->td_ucred, MNT_WAIT, td);
966 	VOP_UNLOCK(ump->um_devvp, 0, td);
967 	return (error);
968 }
969 
970 /*
971  * Get file system statistics.
972  */
973 int
974 ffs_statfs(mp, sbp, td)
975 	struct mount *mp;
976 	register struct statfs *sbp;
977 	struct thread *td;
978 {
979 	register struct ufsmount *ump;
980 	register struct fs *fs;
981 
982 	ump = VFSTOUFS(mp);
983 	fs = ump->um_fs;
984 	if (fs->fs_magic != FS_MAGIC)
985 		panic("ffs_statfs");
986 	sbp->f_bsize = fs->fs_fsize;
987 	sbp->f_iosize = fs->fs_bsize;
988 	sbp->f_blocks = fs->fs_dsize;
989 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
990 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
991 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
992 	    dbtofsb(fs, fs->fs_pendingblocks);
993 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
994 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
995 	if (sbp != &mp->mnt_stat) {
996 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
997 		bcopy((caddr_t)mp->mnt_stat.f_mntonname,
998 			(caddr_t)&sbp->f_mntonname[0], MNAMELEN);
999 		bcopy((caddr_t)mp->mnt_stat.f_mntfromname,
1000 			(caddr_t)&sbp->f_mntfromname[0], MNAMELEN);
1001 	}
1002 	return (0);
1003 }
1004 
1005 /*
1006  * Go through the disk queues to initiate sandbagged IO;
1007  * go through the inodes to write those that have been modified;
1008  * initiate the writing of the super block if it has been modified.
1009  *
1010  * Note: we are always called with the filesystem marked `MPBUSY'.
1011  */
1012 int
1013 ffs_sync(mp, waitfor, cred, td)
1014 	struct mount *mp;
1015 	int waitfor;
1016 	struct ucred *cred;
1017 	struct thread *td;
1018 {
1019 	struct vnode *nvp, *vp, *devvp;
1020 	struct inode *ip;
1021 	struct ufsmount *ump = VFSTOUFS(mp);
1022 	struct fs *fs;
1023 	int error, count, wait, lockreq, allerror = 0;
1024 
1025 	fs = ump->um_fs;
1026 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
1027 		printf("fs = %s\n", fs->fs_fsmnt);
1028 		panic("ffs_sync: rofs mod");
1029 	}
1030 	/*
1031 	 * Write back each (modified) inode.
1032 	 */
1033 	wait = 0;
1034 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1035 	if (waitfor == MNT_WAIT) {
1036 		wait = 1;
1037 		lockreq = LK_EXCLUSIVE;
1038 	}
1039 	mtx_lock(&mntvnode_mtx);
1040 loop:
1041 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
1042 		/*
1043 		 * If the vnode that we are about to sync is no longer
1044 		 * associated with this mount point, start over.
1045 		 */
1046 		if (vp->v_mount != mp)
1047 			goto loop;
1048 
1049 		/*
1050 		 * Depend on the mntvnode_slock to keep things stable enough
1051 		 * for a quick test.  Since there might be hundreds of
1052 		 * thousands of vnodes, we cannot afford even a subroutine
1053 		 * call unless there's a good chance that we have work to do.
1054 		 */
1055 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1056 		ip = VTOI(vp);
1057 		if (vp->v_type == VNON || ((ip->i_flag &
1058 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1059 		    TAILQ_EMPTY(&vp->v_dirtyblkhd))) {
1060 			continue;
1061 		}
1062 		if (vp->v_type != VCHR) {
1063 			mtx_unlock(&mntvnode_mtx);
1064 			if ((error = vget(vp, lockreq, td)) != 0) {
1065 				mtx_lock(&mntvnode_mtx);
1066 				if (error == ENOENT)
1067 					goto loop;
1068 			} else {
1069 				if ((error = VOP_FSYNC(vp, cred, waitfor, td)) != 0)
1070 					allerror = error;
1071 				VOP_UNLOCK(vp, 0, td);
1072 				vrele(vp);
1073 				mtx_lock(&mntvnode_mtx);
1074 			}
1075 		} else {
1076 			mtx_unlock(&mntvnode_mtx);
1077 			UFS_UPDATE(vp, wait);
1078 			mtx_lock(&mntvnode_mtx);
1079 		}
1080 		if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp)
1081 			goto loop;
1082 	}
1083 	mtx_unlock(&mntvnode_mtx);
1084 	/*
1085 	 * Force stale file system control information to be flushed.
1086 	 */
1087 	if (waitfor == MNT_WAIT) {
1088 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1089 			allerror = error;
1090 		/* Flushed work items may create new vnodes to clean */
1091 		if (count) {
1092 			mtx_lock(&mntvnode_mtx);
1093 			goto loop;
1094 		}
1095 	}
1096 #ifdef QUOTA
1097 	qsync(mp);
1098 #endif
1099 	devvp = ump->um_devvp;
1100 	mtx_lock(&devvp->v_interlock);
1101 	if (waitfor != MNT_LAZY &&
1102 	    (devvp->v_numoutput > 0 || TAILQ_FIRST(&devvp->v_dirtyblkhd))) {
1103 		mtx_unlock(&devvp->v_interlock);
1104 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
1105 		if ((error = VOP_FSYNC(devvp, cred, waitfor, td)) != 0)
1106 			allerror = error;
1107 		VOP_UNLOCK(devvp, 0, td);
1108 		if (waitfor == MNT_WAIT) {
1109 			mtx_lock(&mntvnode_mtx);
1110 			goto loop;
1111 		}
1112 	} else
1113 		mtx_unlock(&devvp->v_interlock);
1114 	/*
1115 	 * Write back modified superblock.
1116 	 */
1117 	if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor)) != 0)
1118 		allerror = error;
1119 	return (allerror);
1120 }
1121 
1122 /*
1123  * Look up a FFS dinode number to find its incore vnode, otherwise read it
1124  * in from disk.  If it is in core, wait for the lock bit to clear, then
1125  * return the inode locked.  Detection and handling of mount points must be
1126  * done by the calling routine.
1127  */
1128 static int ffs_inode_hash_lock;
1129 /*
1130  * ffs_inode_hash_lock is a variable to manage mutual exclusion
1131  * of vnode allocation and intertion to the hash, especially to
1132  * avoid holding more than one vnodes for the same inode in the
1133  * hash table. ffs_inode_hash_lock must hence be tested-and-set
1134  * or cleared atomically, accomplished by ffs_inode_hash_mtx.
1135  *
1136  * As vnode allocation may block during MALLOC() and zone
1137  * allocation, we should also do msleep() to give away the CPU
1138  * if anyone else is allocating a vnode. lockmgr is not suitable
1139  * here because someone else may insert to the hash table the
1140  * vnode we are trying to allocate during our sleep, in which
1141  * case the hash table needs to be examined once again after
1142  * waking up.
1143  */
1144 static struct mtx ffs_inode_hash_mtx;
1145 
1146 int
1147 ffs_vget(mp, ino, flags, vpp)
1148 	struct mount *mp;
1149 	ino_t ino;
1150 	int flags;
1151 	struct vnode **vpp;
1152 {
1153 	struct fs *fs;
1154 	struct inode *ip;
1155 	struct ufsmount *ump;
1156 	struct buf *bp;
1157 	struct vnode *vp;
1158 	dev_t dev;
1159 	int error, want_wakeup;
1160 
1161 	ump = VFSTOUFS(mp);
1162 	dev = ump->um_dev;
1163 restart:
1164 	if ((error = ufs_ihashget(dev, ino, flags, vpp)) != 0)
1165 		return (error);
1166 	if (*vpp != NULL)
1167 		return (0);
1168 
1169 	/*
1170 	 * Lock out the creation of new entries in the FFS hash table in
1171 	 * case getnewvnode() or MALLOC() blocks, otherwise a duplicate
1172 	 * may occur!
1173 	 */
1174 	mtx_lock(&ffs_inode_hash_mtx);
1175 	if (ffs_inode_hash_lock) {
1176 		while (ffs_inode_hash_lock) {
1177 			ffs_inode_hash_lock = -1;
1178 			msleep(&ffs_inode_hash_lock, &ffs_inode_hash_mtx, PVM, "ffsvgt", 0);
1179 		}
1180 		mtx_unlock(&ffs_inode_hash_mtx);
1181 		goto restart;
1182 	}
1183 	ffs_inode_hash_lock = 1;
1184 	mtx_unlock(&ffs_inode_hash_mtx);
1185 
1186 	/*
1187 	 * If this MALLOC() is performed after the getnewvnode()
1188 	 * it might block, leaving a vnode with a NULL v_data to be
1189 	 * found by ffs_sync() if a sync happens to fire right then,
1190 	 * which will cause a panic because ffs_sync() blindly
1191 	 * dereferences vp->v_data (as well it should).
1192 	 */
1193 	MALLOC(ip, struct inode *, sizeof(struct inode),
1194 	    ump->um_malloctype, M_WAITOK);
1195 
1196 	/* Allocate a new vnode/inode. */
1197 	error = getnewvnode(VT_UFS, mp, ffs_vnodeop_p, &vp);
1198 	if (error) {
1199 		/*
1200 		 * Do not wake up processes while holding the mutex,
1201 		 * otherwise the processes waken up immediately hit
1202 		 * themselves into the mutex.
1203 		 */
1204 		mtx_lock(&ffs_inode_hash_mtx);
1205 		want_wakeup = ffs_inode_hash_lock < 0;
1206 		ffs_inode_hash_lock = 0;
1207 		mtx_unlock(&ffs_inode_hash_mtx);
1208 		if (want_wakeup)
1209 			wakeup(&ffs_inode_hash_lock);
1210 		*vpp = NULL;
1211 		FREE(ip, ump->um_malloctype);
1212 		return (error);
1213 	}
1214 	bzero((caddr_t)ip, sizeof(struct inode));
1215 	/*
1216 	 * FFS supports lock sharing in the stack of vnodes
1217 	 */
1218 	vp->v_vnlock = &vp->v_lock;
1219 	lockinit(vp->v_vnlock, PINOD, "inode", VLKTIMEOUT, LK_CANRECURSE);
1220 	vp->v_data = ip;
1221 	ip->i_vnode = vp;
1222 	ip->i_fs = fs = ump->um_fs;
1223 	ip->i_dev = dev;
1224 	ip->i_number = ino;
1225 #ifdef QUOTA
1226 	{
1227 		int i;
1228 		for (i = 0; i < MAXQUOTAS; i++)
1229 			ip->i_dquot[i] = NODQUOT;
1230 	}
1231 #endif
1232 	/*
1233 	 * Put it onto its hash chain and lock it so that other requests for
1234 	 * this inode will block if they arrive while we are sleeping waiting
1235 	 * for old data structures to be purged or for the contents of the
1236 	 * disk portion of this inode to be read.
1237 	 */
1238 	ufs_ihashins(ip);
1239 
1240 	/*
1241 	 * Do not wake up processes while holding the mutex,
1242 	 * otherwise the processes waken up immediately hit
1243 	 * themselves into the mutex.
1244 	 */
1245 	mtx_lock(&ffs_inode_hash_mtx);
1246 	want_wakeup = ffs_inode_hash_lock < 0;
1247 	ffs_inode_hash_lock = 0;
1248 	mtx_unlock(&ffs_inode_hash_mtx);
1249 	if (want_wakeup)
1250 		wakeup(&ffs_inode_hash_lock);
1251 
1252 	/* Read in the disk contents for the inode, copy into the inode. */
1253 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1254 	    (int)fs->fs_bsize, NOCRED, &bp);
1255 	if (error) {
1256 		/*
1257 		 * The inode does not contain anything useful, so it would
1258 		 * be misleading to leave it on its hash chain. With mode
1259 		 * still zero, it will be unlinked and returned to the free
1260 		 * list by vput().
1261 		 */
1262 		brelse(bp);
1263 		vput(vp);
1264 		*vpp = NULL;
1265 		return (error);
1266 	}
1267 	ip->i_din = *((struct dinode *)bp->b_data + ino_to_fsbo(fs, ino));
1268 	if (DOINGSOFTDEP(vp))
1269 		softdep_load_inodeblock(ip);
1270 	else
1271 		ip->i_effnlink = ip->i_nlink;
1272 	bqrelse(bp);
1273 
1274 	/*
1275 	 * Initialize the vnode from the inode, check for aliases.
1276 	 * Note that the underlying vnode may have changed.
1277 	 */
1278 	error = ufs_vinit(mp, ffs_specop_p, ffs_fifoop_p, &vp);
1279 	if (error) {
1280 		vput(vp);
1281 		*vpp = NULL;
1282 		return (error);
1283 	}
1284 	/*
1285 	 * Finish inode initialization now that aliasing has been resolved.
1286 	 */
1287 	ip->i_devvp = ump->um_devvp;
1288 	VREF(ip->i_devvp);
1289 	/*
1290 	 * Set up a generation number for this inode if it does not
1291 	 * already have one. This should only happen on old filesystems.
1292 	 */
1293 	if (ip->i_gen == 0) {
1294 		ip->i_gen = random() / 2 + 1;
1295 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
1296 			ip->i_flag |= IN_MODIFIED;
1297 	}
1298 	/*
1299 	 * Ensure that uid and gid are correct. This is a temporary
1300 	 * fix until fsck has been changed to do the update.
1301 	 */
1302 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
1303 		ip->i_uid = ip->i_din.di_ouid;		/* XXX */
1304 		ip->i_gid = ip->i_din.di_ogid;		/* XXX */
1305 	}						/* XXX */
1306 
1307 	*vpp = vp;
1308 	return (0);
1309 }
1310 
1311 /*
1312  * File handle to vnode
1313  *
1314  * Have to be really careful about stale file handles:
1315  * - check that the inode number is valid
1316  * - call ffs_vget() to get the locked inode
1317  * - check for an unallocated inode (i_mode == 0)
1318  * - check that the given client host has export rights and return
1319  *   those rights via. exflagsp and credanonp
1320  */
1321 int
1322 ffs_fhtovp(mp, fhp, vpp)
1323 	register struct mount *mp;
1324 	struct fid *fhp;
1325 	struct vnode **vpp;
1326 {
1327 	register struct ufid *ufhp;
1328 	struct fs *fs;
1329 
1330 	ufhp = (struct ufid *)fhp;
1331 	fs = VFSTOUFS(mp)->um_fs;
1332 	if (ufhp->ufid_ino < ROOTINO ||
1333 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1334 		return (ESTALE);
1335 	return (ufs_fhtovp(mp, ufhp, vpp));
1336 }
1337 
1338 /*
1339  * Vnode pointer to File handle
1340  */
1341 /* ARGSUSED */
1342 int
1343 ffs_vptofh(vp, fhp)
1344 	struct vnode *vp;
1345 	struct fid *fhp;
1346 {
1347 	register struct inode *ip;
1348 	register struct ufid *ufhp;
1349 
1350 	ip = VTOI(vp);
1351 	ufhp = (struct ufid *)fhp;
1352 	ufhp->ufid_len = sizeof(struct ufid);
1353 	ufhp->ufid_ino = ip->i_number;
1354 	ufhp->ufid_gen = ip->i_gen;
1355 	return (0);
1356 }
1357 
1358 /*
1359  * Initialize the filesystem; just use ufs_init.
1360  */
1361 static int
1362 ffs_init(vfsp)
1363 	struct vfsconf *vfsp;
1364 {
1365 
1366 	softdep_initialize();
1367 	mtx_init(&ffs_inode_hash_mtx, "ifsvgt", NULL, MTX_DEF);
1368 	return (ufs_init(vfsp));
1369 }
1370 
1371 /*
1372  * Write a superblock and associated information back to disk.
1373  */
1374 static int
1375 ffs_sbupdate(mp, waitfor)
1376 	struct ufsmount *mp;
1377 	int waitfor;
1378 {
1379 	register struct fs *dfs, *fs = mp->um_fs;
1380 	register struct buf *bp;
1381 	int blks;
1382 	void *space;
1383 	int i, size, error, allerror = 0;
1384 
1385 	/*
1386 	 * First write back the summary information.
1387 	 */
1388 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1389 	space = fs->fs_csp;
1390 	for (i = 0; i < blks; i += fs->fs_frag) {
1391 		size = fs->fs_bsize;
1392 		if (i + fs->fs_frag > blks)
1393 			size = (blks - i) * fs->fs_fsize;
1394 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1395 		    size, 0, 0);
1396 		bcopy(space, bp->b_data, (u_int)size);
1397 		space = (char *)space + size;
1398 		if (waitfor != MNT_WAIT)
1399 			bawrite(bp);
1400 		else if ((error = bwrite(bp)) != 0)
1401 			allerror = error;
1402 	}
1403 	/*
1404 	 * Now write back the superblock itself. If any errors occurred
1405 	 * up to this point, then fail so that the superblock avoids
1406 	 * being written out as clean.
1407 	 */
1408 	if (allerror)
1409 		return (allerror);
1410 	bp = getblk(mp->um_devvp, SBLOCK, (int)fs->fs_sbsize, 0, 0);
1411 	fs->fs_fmod = 0;
1412 	fs->fs_time = time_second;
1413 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1414 	/* Restore compatibility to old file systems.		   XXX */
1415 	dfs = (struct fs *)bp->b_data;				/* XXX */
1416 	if (fs->fs_postblformat == FS_42POSTBLFMT)		/* XXX */
1417 		dfs->fs_nrpos = -1;				/* XXX */
1418 	if (fs->fs_inodefmt < FS_44INODEFMT) {			/* XXX */
1419 		int32_t *lp, tmp;				/* XXX */
1420 								/* XXX */
1421 		lp = (int32_t *)&dfs->fs_qbmask;		/* XXX */
1422 		tmp = lp[4];					/* XXX */
1423 		for (i = 4; i > 0; i--)				/* XXX */
1424 			lp[i] = lp[i-1];			/* XXX */
1425 		lp[0] = tmp;					/* XXX */
1426 	}							/* XXX */
1427 	dfs->fs_maxfilesize = mp->um_savedmaxfilesize;		/* XXX */
1428 	if (waitfor != MNT_WAIT)
1429 		bawrite(bp);
1430 	else if ((error = bwrite(bp)) != 0)
1431 		allerror = error;
1432 	return (allerror);
1433 }
1434