xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision 4a5216a6dc0c3ce4cf5f2d3ee8af0c3ff3402c4f)
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 #include "opt_ffs.h"
39 #include "opt_ddb.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/namei.h>
44 #include <sys/priv.h>
45 #include <sys/proc.h>
46 #include <sys/kernel.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/fcntl.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
55 
56 #include <security/mac/mac_framework.h>
57 
58 #include <ufs/ufs/extattr.h>
59 #include <ufs/ufs/gjournal.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 
72 #include <geom/geom.h>
73 #include <geom/geom_vfs.h>
74 
75 #include <ddb/ddb.h>
76 
77 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
78 
79 static int	ffs_reload(struct mount *, struct thread *);
80 static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
81 static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
82 		    ufs2_daddr_t);
83 static void	ffs_oldfscompat_write(struct fs *, struct ufsmount *);
84 static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
85 static vfs_init_t ffs_init;
86 static vfs_uninit_t ffs_uninit;
87 static vfs_extattrctl_t ffs_extattrctl;
88 static vfs_cmount_t ffs_cmount;
89 static vfs_unmount_t ffs_unmount;
90 static vfs_mount_t ffs_mount;
91 static vfs_statfs_t ffs_statfs;
92 static vfs_fhtovp_t ffs_fhtovp;
93 static vfs_sync_t ffs_sync;
94 
95 static struct vfsops ufs_vfsops = {
96 	.vfs_extattrctl =	ffs_extattrctl,
97 	.vfs_fhtovp =		ffs_fhtovp,
98 	.vfs_init =		ffs_init,
99 	.vfs_mount =		ffs_mount,
100 	.vfs_cmount =		ffs_cmount,
101 	.vfs_quotactl =		ufs_quotactl,
102 	.vfs_root =		ufs_root,
103 	.vfs_statfs =		ffs_statfs,
104 	.vfs_sync =		ffs_sync,
105 	.vfs_uninit =		ffs_uninit,
106 	.vfs_unmount =		ffs_unmount,
107 	.vfs_vget =		ffs_vget,
108 	.vfs_susp_clean =	process_deferred_inactive,
109 };
110 
111 VFS_SET(ufs_vfsops, ufs, 0);
112 MODULE_VERSION(ufs, 1);
113 
114 static b_strategy_t ffs_geom_strategy;
115 static b_write_t ffs_bufwrite;
116 
117 static struct buf_ops ffs_ops = {
118 	.bop_name =	"FFS",
119 	.bop_write =	ffs_bufwrite,
120 	.bop_strategy =	ffs_geom_strategy,
121 	.bop_sync =	bufsync,
122 #ifdef NO_FFS_SNAPSHOT
123 	.bop_bdflush =	bufbdflush,
124 #else
125 	.bop_bdflush =	ffs_bdflush,
126 #endif
127 };
128 
129 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
130     "noclusterw", "noexec", "export", "force", "from", "multilabel",
131     "snapshot", "nosuid", "suiddir", "nosymfollow", "sync",
132     "union", NULL };
133 
134 static int
135 ffs_mount(struct mount *mp, struct thread *td)
136 {
137 	struct vnode *devvp;
138 	struct ufsmount *ump = 0;
139 	struct fs *fs;
140 	int error, flags;
141 	u_int mntorflags, mntandnotflags;
142 	mode_t accessmode;
143 	struct nameidata ndp;
144 	char *fspec;
145 
146 	if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
147 		return (EINVAL);
148 	if (uma_inode == NULL) {
149 		uma_inode = uma_zcreate("FFS inode",
150 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
151 		    UMA_ALIGN_PTR, 0);
152 		uma_ufs1 = uma_zcreate("FFS1 dinode",
153 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
154 		    UMA_ALIGN_PTR, 0);
155 		uma_ufs2 = uma_zcreate("FFS2 dinode",
156 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
157 		    UMA_ALIGN_PTR, 0);
158 	}
159 
160 	fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
161 	if (error)
162 		return (error);
163 
164 	mntorflags = 0;
165 	mntandnotflags = 0;
166 	if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
167 		mntorflags |= MNT_ACLS;
168 
169 	if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
170 		mntorflags |= MNT_SNAPSHOT;
171 		/*
172 		 * Once we have set the MNT_SNAPSHOT flag, do not
173 		 * persist "snapshot" in the options list.
174 		 */
175 		vfs_deleteopt(mp->mnt_optnew, "snapshot");
176 		vfs_deleteopt(mp->mnt_opt, "snapshot");
177 	}
178 
179 	MNT_ILOCK(mp);
180 	mp->mnt_flag = (mp->mnt_flag | mntorflags) & ~mntandnotflags;
181 	MNT_IUNLOCK(mp);
182 	/*
183 	 * If updating, check whether changing from read-only to
184 	 * read/write; if there is no device name, that's all we do.
185 	 */
186 	if (mp->mnt_flag & MNT_UPDATE) {
187 		ump = VFSTOUFS(mp);
188 		fs = ump->um_fs;
189 		devvp = ump->um_devvp;
190 		if (fs->fs_ronly == 0 &&
191 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
192 			/*
193 			 * Flush any dirty data and suspend filesystem.
194 			 */
195 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
196 				return (error);
197 			for (;;) {
198 				vn_finished_write(mp);
199 				if ((error = vfs_write_suspend(mp)) != 0)
200 					return (error);
201 				MNT_ILOCK(mp);
202 				if (mp->mnt_kern_flag & MNTK_SUSPENDED) {
203 					/*
204 					 * Allow the secondary writes
205 					 * to proceed.
206 					 */
207 					mp->mnt_kern_flag &= ~(MNTK_SUSPENDED |
208 					    MNTK_SUSPEND2);
209 					wakeup(&mp->mnt_flag);
210 					MNT_IUNLOCK(mp);
211 					/*
212 					 * Allow the curthread to
213 					 * ignore the suspension to
214 					 * synchronize on-disk state.
215 					 */
216 					curthread->td_pflags |= TDP_IGNSUSP;
217 					break;
218 				}
219 				MNT_IUNLOCK(mp);
220 				vn_start_write(NULL, &mp, V_WAIT);
221 			}
222 			/*
223 			 * Check for and optionally get rid of files open
224 			 * for writing.
225 			 */
226 			flags = WRITECLOSE;
227 			if (mp->mnt_flag & MNT_FORCE)
228 				flags |= FORCECLOSE;
229 			if (mp->mnt_flag & MNT_SOFTDEP) {
230 				error = softdep_flushfiles(mp, flags, td);
231 			} else {
232 				error = ffs_flushfiles(mp, flags, td);
233 			}
234 			if (error) {
235 				vfs_write_resume(mp);
236 				return (error);
237 			}
238 			if (fs->fs_pendingblocks != 0 ||
239 			    fs->fs_pendinginodes != 0) {
240 				printf("%s: %s: blocks %jd files %d\n",
241 				    fs->fs_fsmnt, "update error",
242 				    (intmax_t)fs->fs_pendingblocks,
243 				    fs->fs_pendinginodes);
244 				fs->fs_pendingblocks = 0;
245 				fs->fs_pendinginodes = 0;
246 			}
247 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
248 				fs->fs_clean = 1;
249 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
250 				fs->fs_ronly = 0;
251 				fs->fs_clean = 0;
252 				vfs_write_resume(mp);
253 				return (error);
254 			}
255 			DROP_GIANT();
256 			g_topology_lock();
257 			g_access(ump->um_cp, 0, -1, 0);
258 			g_topology_unlock();
259 			PICKUP_GIANT();
260 			fs->fs_ronly = 1;
261 			MNT_ILOCK(mp);
262 			mp->mnt_flag |= MNT_RDONLY;
263 			MNT_IUNLOCK(mp);
264 			/*
265 			 * Allow the writers to note that filesystem
266 			 * is ro now.
267 			 */
268 			vfs_write_resume(mp);
269 		}
270 		if ((mp->mnt_flag & MNT_RELOAD) &&
271 		    (error = ffs_reload(mp, td)) != 0)
272 			return (error);
273 		if (fs->fs_ronly &&
274 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
275 			/*
276 			 * If upgrade to read-write by non-root, then verify
277 			 * that user has necessary permissions on the device.
278 			 */
279 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
280 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
281 			    td->td_ucred, td);
282 			if (error)
283 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
284 			if (error) {
285 				VOP_UNLOCK(devvp, 0);
286 				return (error);
287 			}
288 			VOP_UNLOCK(devvp, 0);
289 			fs->fs_flags &= ~FS_UNCLEAN;
290 			if (fs->fs_clean == 0) {
291 				fs->fs_flags |= FS_UNCLEAN;
292 				if ((mp->mnt_flag & MNT_FORCE) ||
293 				    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
294 				     (fs->fs_flags & FS_DOSOFTDEP))) {
295 					printf("WARNING: %s was not %s\n",
296 					   fs->fs_fsmnt, "properly dismounted");
297 				} else {
298 					printf(
299 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
300 					    fs->fs_fsmnt);
301 					return (EPERM);
302 				}
303 			}
304 			DROP_GIANT();
305 			g_topology_lock();
306 			/*
307 			 * If we're the root device, we may not have an E count
308 			 * yet, get it now.
309 			 */
310 			if (ump->um_cp->ace == 0)
311 				error = g_access(ump->um_cp, 0, 1, 1);
312 			else
313 				error = g_access(ump->um_cp, 0, 1, 0);
314 			g_topology_unlock();
315 			PICKUP_GIANT();
316 			if (error)
317 				return (error);
318 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
319 				return (error);
320 			fs->fs_ronly = 0;
321 			MNT_ILOCK(mp);
322 			mp->mnt_flag &= ~MNT_RDONLY;
323 			MNT_IUNLOCK(mp);
324 			fs->fs_clean = 0;
325 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
326 				vn_finished_write(mp);
327 				return (error);
328 			}
329 			/* check to see if we need to start softdep */
330 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
331 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
332 				vn_finished_write(mp);
333 				return (error);
334 			}
335 			if (fs->fs_snapinum[0] != 0)
336 				ffs_snapshot_mount(mp);
337 			vn_finished_write(mp);
338 		}
339 		/*
340 		 * Soft updates is incompatible with "async",
341 		 * so if we are doing softupdates stop the user
342 		 * from setting the async flag in an update.
343 		 * Softdep_mount() clears it in an initial mount
344 		 * or ro->rw remount.
345 		 */
346 		if (mp->mnt_flag & MNT_SOFTDEP) {
347 			/* XXX: Reset too late ? */
348 			MNT_ILOCK(mp);
349 			mp->mnt_flag &= ~MNT_ASYNC;
350 			MNT_IUNLOCK(mp);
351 		}
352 		/*
353 		 * Keep MNT_ACLS flag if it is stored in superblock.
354 		 */
355 		if ((fs->fs_flags & FS_ACLS) != 0) {
356 			/* XXX: Set too late ? */
357 			MNT_ILOCK(mp);
358 			mp->mnt_flag |= MNT_ACLS;
359 			MNT_IUNLOCK(mp);
360 		}
361 
362 		/*
363 		 * If this is a snapshot request, take the snapshot.
364 		 */
365 		if (mp->mnt_flag & MNT_SNAPSHOT)
366 			return (ffs_snapshot(mp, fspec));
367 	}
368 
369 	/*
370 	 * Not an update, or updating the name: look up the name
371 	 * and verify that it refers to a sensible disk device.
372 	 */
373 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
374 	if ((error = namei(&ndp)) != 0)
375 		return (error);
376 	NDFREE(&ndp, NDF_ONLY_PNBUF);
377 	devvp = ndp.ni_vp;
378 	if (!vn_isdisk(devvp, &error)) {
379 		vput(devvp);
380 		return (error);
381 	}
382 
383 	/*
384 	 * If mount by non-root, then verify that user has necessary
385 	 * permissions on the device.
386 	 */
387 	accessmode = VREAD;
388 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
389 		accessmode |= VWRITE;
390 	error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td);
391 	if (error)
392 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
393 	if (error) {
394 		vput(devvp);
395 		return (error);
396 	}
397 
398 	if (mp->mnt_flag & MNT_UPDATE) {
399 		/*
400 		 * Update only
401 		 *
402 		 * If it's not the same vnode, or at least the same device
403 		 * then it's not correct.
404 		 */
405 
406 		if (devvp->v_rdev != ump->um_devvp->v_rdev)
407 			error = EINVAL;	/* needs translation */
408 		vput(devvp);
409 		if (error)
410 			return (error);
411 	} else {
412 		/*
413 		 * New mount
414 		 *
415 		 * We need the name for the mount point (also used for
416 		 * "last mounted on") copied in. If an error occurs,
417 		 * the mount point is discarded by the upper level code.
418 		 * Note that vfs_mount() populates f_mntonname for us.
419 		 */
420 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
421 			vrele(devvp);
422 			return (error);
423 		}
424 	}
425 	vfs_mountedfrom(mp, fspec);
426 	return (0);
427 }
428 
429 /*
430  * Compatibility with old mount system call.
431  */
432 
433 static int
434 ffs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td)
435 {
436 	struct ufs_args args;
437 	int error;
438 
439 	if (data == NULL)
440 		return (EINVAL);
441 	error = copyin(data, &args, sizeof args);
442 	if (error)
443 		return (error);
444 
445 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
446 	ma = mount_arg(ma, "export", &args.export, sizeof args.export);
447 	error = kernel_mount(ma, flags);
448 
449 	return (error);
450 }
451 
452 /*
453  * Reload all incore data for a filesystem (used after running fsck on
454  * the root filesystem and finding things to fix). The filesystem must
455  * be mounted read-only.
456  *
457  * Things to do to update the mount:
458  *	1) invalidate all cached meta-data.
459  *	2) re-read superblock from disk.
460  *	3) re-read summary information from disk.
461  *	4) invalidate all inactive vnodes.
462  *	5) invalidate all cached file data.
463  *	6) re-read inode data for all active vnodes.
464  */
465 static int
466 ffs_reload(struct mount *mp, struct thread *td)
467 {
468 	struct vnode *vp, *mvp, *devvp;
469 	struct inode *ip;
470 	void *space;
471 	struct buf *bp;
472 	struct fs *fs, *newfs;
473 	struct ufsmount *ump;
474 	ufs2_daddr_t sblockloc;
475 	int i, blks, size, error;
476 	int32_t *lp;
477 
478 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
479 		return (EINVAL);
480 	ump = VFSTOUFS(mp);
481 	/*
482 	 * Step 1: invalidate all cached meta-data.
483 	 */
484 	devvp = VFSTOUFS(mp)->um_devvp;
485 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
486 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
487 		panic("ffs_reload: dirty1");
488 	VOP_UNLOCK(devvp, 0);
489 
490 	/*
491 	 * Step 2: re-read superblock from disk.
492 	 */
493 	fs = VFSTOUFS(mp)->um_fs;
494 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
495 	    NOCRED, &bp)) != 0)
496 		return (error);
497 	newfs = (struct fs *)bp->b_data;
498 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
499 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
500 	    newfs->fs_bsize > MAXBSIZE ||
501 	    newfs->fs_bsize < sizeof(struct fs)) {
502 			brelse(bp);
503 			return (EIO);		/* XXX needs translation */
504 	}
505 	/*
506 	 * Copy pointer fields back into superblock before copying in	XXX
507 	 * new superblock. These should really be in the ufsmount.	XXX
508 	 * Note that important parameters (eg fs_ncg) are unchanged.
509 	 */
510 	newfs->fs_csp = fs->fs_csp;
511 	newfs->fs_maxcluster = fs->fs_maxcluster;
512 	newfs->fs_contigdirs = fs->fs_contigdirs;
513 	newfs->fs_active = fs->fs_active;
514 	/* The file system is still read-only. */
515 	newfs->fs_ronly = 1;
516 	sblockloc = fs->fs_sblockloc;
517 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
518 	brelse(bp);
519 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
520 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
521 	UFS_LOCK(ump);
522 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
523 		printf("%s: reload pending error: blocks %jd files %d\n",
524 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
525 		    fs->fs_pendinginodes);
526 		fs->fs_pendingblocks = 0;
527 		fs->fs_pendinginodes = 0;
528 	}
529 	UFS_UNLOCK(ump);
530 
531 	/*
532 	 * Step 3: re-read summary information from disk.
533 	 */
534 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
535 	space = fs->fs_csp;
536 	for (i = 0; i < blks; i += fs->fs_frag) {
537 		size = fs->fs_bsize;
538 		if (i + fs->fs_frag > blks)
539 			size = (blks - i) * fs->fs_fsize;
540 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
541 		    NOCRED, &bp);
542 		if (error)
543 			return (error);
544 		bcopy(bp->b_data, space, (u_int)size);
545 		space = (char *)space + size;
546 		brelse(bp);
547 	}
548 	/*
549 	 * We no longer know anything about clusters per cylinder group.
550 	 */
551 	if (fs->fs_contigsumsize > 0) {
552 		lp = fs->fs_maxcluster;
553 		for (i = 0; i < fs->fs_ncg; i++)
554 			*lp++ = fs->fs_contigsumsize;
555 	}
556 
557 loop:
558 	MNT_ILOCK(mp);
559 	MNT_VNODE_FOREACH(vp, mp, mvp) {
560 		VI_LOCK(vp);
561 		if (vp->v_iflag & VI_DOOMED) {
562 			VI_UNLOCK(vp);
563 			continue;
564 		}
565 		MNT_IUNLOCK(mp);
566 		/*
567 		 * Step 4: invalidate all cached file data.
568 		 */
569 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
570 			MNT_VNODE_FOREACH_ABORT(mp, mvp);
571 			goto loop;
572 		}
573 		if (vinvalbuf(vp, 0, 0, 0))
574 			panic("ffs_reload: dirty2");
575 		/*
576 		 * Step 5: re-read inode data for all active vnodes.
577 		 */
578 		ip = VTOI(vp);
579 		error =
580 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
581 		    (int)fs->fs_bsize, NOCRED, &bp);
582 		if (error) {
583 			VOP_UNLOCK(vp, 0);
584 			vrele(vp);
585 			MNT_VNODE_FOREACH_ABORT(mp, mvp);
586 			return (error);
587 		}
588 		ffs_load_inode(bp, ip, fs, ip->i_number);
589 		ip->i_effnlink = ip->i_nlink;
590 		brelse(bp);
591 		VOP_UNLOCK(vp, 0);
592 		vrele(vp);
593 		MNT_ILOCK(mp);
594 	}
595 	MNT_IUNLOCK(mp);
596 	return (0);
597 }
598 
599 /*
600  * Possible superblock locations ordered from most to least likely.
601  */
602 static int sblock_try[] = SBLOCKSEARCH;
603 
604 /*
605  * Common code for mount and mountroot
606  */
607 static int
608 ffs_mountfs(devvp, mp, td)
609 	struct vnode *devvp;
610 	struct mount *mp;
611 	struct thread *td;
612 {
613 	struct ufsmount *ump;
614 	struct buf *bp;
615 	struct fs *fs;
616 	struct cdev *dev;
617 	void *space;
618 	ufs2_daddr_t sblockloc;
619 	int error, i, blks, size, ronly;
620 	int32_t *lp;
621 	struct ucred *cred;
622 	struct g_consumer *cp;
623 	struct mount *nmp;
624 
625 	dev = devvp->v_rdev;
626 	cred = td ? td->td_ucred : NOCRED;
627 
628 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
629 	DROP_GIANT();
630 	g_topology_lock();
631 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
632 
633 	/*
634 	 * If we are a root mount, drop the E flag so fsck can do its magic.
635 	 * We will pick it up again when we remount R/W.
636 	 */
637 	if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS))
638 		error = g_access(cp, 0, 0, -1);
639 	g_topology_unlock();
640 	PICKUP_GIANT();
641 	VOP_UNLOCK(devvp, 0);
642 	if (error)
643 		return (error);
644 	if (devvp->v_rdev->si_iosize_max != 0)
645 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
646 	if (mp->mnt_iosize_max > MAXPHYS)
647 		mp->mnt_iosize_max = MAXPHYS;
648 
649 	devvp->v_bufobj.bo_private = cp;
650 	devvp->v_bufobj.bo_ops = &ffs_ops;
651 
652 	bp = NULL;
653 	ump = NULL;
654 	fs = NULL;
655 	sblockloc = 0;
656 	/*
657 	 * Try reading the superblock in each of its possible locations.
658 	 */
659 	for (i = 0; sblock_try[i] != -1; i++) {
660 		if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
661 			error = EINVAL;
662 			vfs_mount_error(mp,
663 			    "Invalid sectorsize %d for superblock size %d",
664 			    cp->provider->sectorsize, SBLOCKSIZE);
665 			goto out;
666 		}
667 		if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
668 		    cred, &bp)) != 0)
669 			goto out;
670 		fs = (struct fs *)bp->b_data;
671 		sblockloc = sblock_try[i];
672 		if ((fs->fs_magic == FS_UFS1_MAGIC ||
673 		     (fs->fs_magic == FS_UFS2_MAGIC &&
674 		      (fs->fs_sblockloc == sblockloc ||
675 		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
676 		    fs->fs_bsize <= MAXBSIZE &&
677 		    fs->fs_bsize >= sizeof(struct fs))
678 			break;
679 		brelse(bp);
680 		bp = NULL;
681 	}
682 	if (sblock_try[i] == -1) {
683 		error = EINVAL;		/* XXX needs translation */
684 		goto out;
685 	}
686 	fs->fs_fmod = 0;
687 	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indicies */
688 	fs->fs_flags &= ~FS_UNCLEAN;
689 	if (fs->fs_clean == 0) {
690 		fs->fs_flags |= FS_UNCLEAN;
691 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
692 		    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
693 		     (fs->fs_flags & FS_DOSOFTDEP))) {
694 			printf(
695 "WARNING: %s was not properly dismounted\n",
696 			    fs->fs_fsmnt);
697 		} else {
698 			printf(
699 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
700 			    fs->fs_fsmnt);
701 			error = EPERM;
702 			goto out;
703 		}
704 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
705 		    (mp->mnt_flag & MNT_FORCE)) {
706 			printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt,
707 			    (intmax_t)fs->fs_pendingblocks,
708 			    fs->fs_pendinginodes);
709 			fs->fs_pendingblocks = 0;
710 			fs->fs_pendinginodes = 0;
711 		}
712 	}
713 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
714 		printf("%s: mount pending error: blocks %jd files %d\n",
715 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
716 		    fs->fs_pendinginodes);
717 		fs->fs_pendingblocks = 0;
718 		fs->fs_pendinginodes = 0;
719 	}
720 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
721 #ifdef UFS_GJOURNAL
722 		/*
723 		 * Get journal provider name.
724 		 */
725 		size = 1024;
726 		mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
727 		if (g_io_getattr("GJOURNAL::provider", cp, &size,
728 		    mp->mnt_gjprovider) == 0) {
729 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
730 			    M_UFSMNT, M_WAITOK);
731 			MNT_ILOCK(mp);
732 			mp->mnt_flag |= MNT_GJOURNAL;
733 			MNT_IUNLOCK(mp);
734 		} else {
735 			printf(
736 "WARNING: %s: GJOURNAL flag on fs but no gjournal provider below\n",
737 			    mp->mnt_stat.f_mntonname);
738 			free(mp->mnt_gjprovider, M_UFSMNT);
739 			mp->mnt_gjprovider = NULL;
740 		}
741 #else
742 		printf(
743 "WARNING: %s: GJOURNAL flag on fs but no UFS_GJOURNAL support\n",
744 		    mp->mnt_stat.f_mntonname);
745 #endif
746 	} else {
747 		mp->mnt_gjprovider = NULL;
748 	}
749 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
750 	ump->um_cp = cp;
751 	ump->um_bo = &devvp->v_bufobj;
752 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
753 	if (fs->fs_magic == FS_UFS1_MAGIC) {
754 		ump->um_fstype = UFS1;
755 		ump->um_balloc = ffs_balloc_ufs1;
756 	} else {
757 		ump->um_fstype = UFS2;
758 		ump->um_balloc = ffs_balloc_ufs2;
759 	}
760 	ump->um_blkatoff = ffs_blkatoff;
761 	ump->um_truncate = ffs_truncate;
762 	ump->um_update = ffs_update;
763 	ump->um_valloc = ffs_valloc;
764 	ump->um_vfree = ffs_vfree;
765 	ump->um_ifree = ffs_ifree;
766 	ump->um_rdonly = ffs_rdonly;
767 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
768 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
769 	if (fs->fs_sbsize < SBLOCKSIZE)
770 		bp->b_flags |= B_INVAL | B_NOCACHE;
771 	brelse(bp);
772 	bp = NULL;
773 	fs = ump->um_fs;
774 	ffs_oldfscompat_read(fs, ump, sblockloc);
775 	fs->fs_ronly = ronly;
776 	size = fs->fs_cssize;
777 	blks = howmany(size, fs->fs_fsize);
778 	if (fs->fs_contigsumsize > 0)
779 		size += fs->fs_ncg * sizeof(int32_t);
780 	size += fs->fs_ncg * sizeof(u_int8_t);
781 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
782 	fs->fs_csp = space;
783 	for (i = 0; i < blks; i += fs->fs_frag) {
784 		size = fs->fs_bsize;
785 		if (i + fs->fs_frag > blks)
786 			size = (blks - i) * fs->fs_fsize;
787 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
788 		    cred, &bp)) != 0) {
789 			free(fs->fs_csp, M_UFSMNT);
790 			goto out;
791 		}
792 		bcopy(bp->b_data, space, (u_int)size);
793 		space = (char *)space + size;
794 		brelse(bp);
795 		bp = NULL;
796 	}
797 	if (fs->fs_contigsumsize > 0) {
798 		fs->fs_maxcluster = lp = space;
799 		for (i = 0; i < fs->fs_ncg; i++)
800 			*lp++ = fs->fs_contigsumsize;
801 		space = lp;
802 	}
803 	size = fs->fs_ncg * sizeof(u_int8_t);
804 	fs->fs_contigdirs = (u_int8_t *)space;
805 	bzero(fs->fs_contigdirs, size);
806 	fs->fs_active = NULL;
807 	mp->mnt_data = ump;
808 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
809 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
810 	nmp = NULL;
811 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
812 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
813 		if (nmp)
814 			vfs_rel(nmp);
815 		vfs_getnewfsid(mp);
816 	}
817 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
818 	MNT_ILOCK(mp);
819 	mp->mnt_flag |= MNT_LOCAL;
820 	MNT_IUNLOCK(mp);
821 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
822 #ifdef MAC
823 		MNT_ILOCK(mp);
824 		mp->mnt_flag |= MNT_MULTILABEL;
825 		MNT_IUNLOCK(mp);
826 #else
827 		printf(
828 "WARNING: %s: multilabel flag on fs but no MAC support\n",
829 		    mp->mnt_stat.f_mntonname);
830 #endif
831 	}
832 	if ((fs->fs_flags & FS_ACLS) != 0) {
833 #ifdef UFS_ACL
834 		MNT_ILOCK(mp);
835 		mp->mnt_flag |= MNT_ACLS;
836 		MNT_IUNLOCK(mp);
837 #else
838 		printf(
839 "WARNING: %s: ACLs flag on fs but no ACLs support\n",
840 		    mp->mnt_stat.f_mntonname);
841 #endif
842 	}
843 	ump->um_mountp = mp;
844 	ump->um_dev = dev;
845 	ump->um_devvp = devvp;
846 	ump->um_nindir = fs->fs_nindir;
847 	ump->um_bptrtodb = fs->fs_fsbtodb;
848 	ump->um_seqinc = fs->fs_frag;
849 	for (i = 0; i < MAXQUOTAS; i++)
850 		ump->um_quotas[i] = NULLVP;
851 #ifdef UFS_EXTATTR
852 	ufs_extattr_uepm_init(&ump->um_extattr);
853 #endif
854 	/*
855 	 * Set FS local "last mounted on" information (NULL pad)
856 	 */
857 	bzero(fs->fs_fsmnt, MAXMNTLEN);
858 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
859 
860 	if( mp->mnt_flag & MNT_ROOTFS) {
861 		/*
862 		 * Root mount; update timestamp in mount structure.
863 		 * this will be used by the common root mount code
864 		 * to update the system clock.
865 		 */
866 		mp->mnt_time = fs->fs_time;
867 	}
868 
869 	if (ronly == 0) {
870 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
871 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
872 			free(fs->fs_csp, M_UFSMNT);
873 			goto out;
874 		}
875 		if (fs->fs_snapinum[0] != 0)
876 			ffs_snapshot_mount(mp);
877 		fs->fs_fmod = 1;
878 		fs->fs_clean = 0;
879 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
880 	}
881 	/*
882 	 * Initialize filesystem stat information in mount struct.
883 	 */
884 	MNT_ILOCK(mp);
885 	mp->mnt_kern_flag |= MNTK_MPSAFE | MNTK_LOOKUP_SHARED;
886 	MNT_IUNLOCK(mp);
887 #ifdef UFS_EXTATTR
888 #ifdef UFS_EXTATTR_AUTOSTART
889 	/*
890 	 *
891 	 * Auto-starting does the following:
892 	 *	- check for /.attribute in the fs, and extattr_start if so
893 	 *	- for each file in .attribute, enable that file with
894 	 * 	  an attribute of the same name.
895 	 * Not clear how to report errors -- probably eat them.
896 	 * This would all happen while the filesystem was busy/not
897 	 * available, so would effectively be "atomic".
898 	 */
899 	mp->mnt_stat.f_iosize = fs->fs_bsize;
900 	(void) ufs_extattr_autostart(mp, td);
901 #endif /* !UFS_EXTATTR_AUTOSTART */
902 #endif /* !UFS_EXTATTR */
903 	return (0);
904 out:
905 	if (bp)
906 		brelse(bp);
907 	if (cp != NULL) {
908 		DROP_GIANT();
909 		g_topology_lock();
910 		g_vfs_close(cp);
911 		g_topology_unlock();
912 		PICKUP_GIANT();
913 	}
914 	if (ump) {
915 		mtx_destroy(UFS_MTX(ump));
916 		if (mp->mnt_gjprovider != NULL) {
917 			free(mp->mnt_gjprovider, M_UFSMNT);
918 			mp->mnt_gjprovider = NULL;
919 		}
920 		free(ump->um_fs, M_UFSMNT);
921 		free(ump, M_UFSMNT);
922 		mp->mnt_data = NULL;
923 	}
924 	return (error);
925 }
926 
927 #include <sys/sysctl.h>
928 static int bigcgs = 0;
929 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
930 
931 /*
932  * Sanity checks for loading old filesystem superblocks.
933  * See ffs_oldfscompat_write below for unwound actions.
934  *
935  * XXX - Parts get retired eventually.
936  * Unfortunately new bits get added.
937  */
938 static void
939 ffs_oldfscompat_read(fs, ump, sblockloc)
940 	struct fs *fs;
941 	struct ufsmount *ump;
942 	ufs2_daddr_t sblockloc;
943 {
944 	off_t maxfilesize;
945 
946 	/*
947 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
948 	 */
949 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
950 		fs->fs_flags = fs->fs_old_flags;
951 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
952 		fs->fs_sblockloc = sblockloc;
953 	}
954 	/*
955 	 * If not yet done, update UFS1 superblock with new wider fields.
956 	 */
957 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
958 		fs->fs_maxbsize = fs->fs_bsize;
959 		fs->fs_time = fs->fs_old_time;
960 		fs->fs_size = fs->fs_old_size;
961 		fs->fs_dsize = fs->fs_old_dsize;
962 		fs->fs_csaddr = fs->fs_old_csaddr;
963 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
964 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
965 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
966 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
967 	}
968 	if (fs->fs_magic == FS_UFS1_MAGIC &&
969 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
970 		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
971 		fs->fs_qbmask = ~fs->fs_bmask;
972 		fs->fs_qfmask = ~fs->fs_fmask;
973 	}
974 	if (fs->fs_magic == FS_UFS1_MAGIC) {
975 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
976 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
977 		if (fs->fs_maxfilesize > maxfilesize)
978 			fs->fs_maxfilesize = maxfilesize;
979 	}
980 	/* Compatibility for old filesystems */
981 	if (fs->fs_avgfilesize <= 0)
982 		fs->fs_avgfilesize = AVFILESIZ;
983 	if (fs->fs_avgfpdir <= 0)
984 		fs->fs_avgfpdir = AFPDIR;
985 	if (bigcgs) {
986 		fs->fs_save_cgsize = fs->fs_cgsize;
987 		fs->fs_cgsize = fs->fs_bsize;
988 	}
989 }
990 
991 /*
992  * Unwinding superblock updates for old filesystems.
993  * See ffs_oldfscompat_read above for details.
994  *
995  * XXX - Parts get retired eventually.
996  * Unfortunately new bits get added.
997  */
998 static void
999 ffs_oldfscompat_write(fs, ump)
1000 	struct fs *fs;
1001 	struct ufsmount *ump;
1002 {
1003 
1004 	/*
1005 	 * Copy back UFS2 updated fields that UFS1 inspects.
1006 	 */
1007 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1008 		fs->fs_old_time = fs->fs_time;
1009 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1010 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1011 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1012 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1013 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1014 	}
1015 	if (bigcgs) {
1016 		fs->fs_cgsize = fs->fs_save_cgsize;
1017 		fs->fs_save_cgsize = 0;
1018 	}
1019 }
1020 
1021 /*
1022  * unmount system call
1023  */
1024 static int
1025 ffs_unmount(mp, mntflags, td)
1026 	struct mount *mp;
1027 	int mntflags;
1028 	struct thread *td;
1029 {
1030 	struct ufsmount *ump = VFSTOUFS(mp);
1031 	struct fs *fs;
1032 	int error, flags, susp;
1033 
1034 	flags = 0;
1035 	fs = ump->um_fs;
1036 	if (mntflags & MNT_FORCE) {
1037 		flags |= FORCECLOSE;
1038 		susp = fs->fs_ronly != 0;
1039 	} else
1040 		susp = 0;
1041 #ifdef UFS_EXTATTR
1042 	if ((error = ufs_extattr_stop(mp, td))) {
1043 		if (error != EOPNOTSUPP)
1044 			printf("ffs_unmount: ufs_extattr_stop returned %d\n",
1045 			    error);
1046 	} else {
1047 		ufs_extattr_uepm_destroy(&ump->um_extattr);
1048 	}
1049 #endif
1050 	if (susp) {
1051 		/*
1052 		 * dounmount already called vn_start_write().
1053 		 */
1054 		for (;;) {
1055 			vn_finished_write(mp);
1056 			if ((error = vfs_write_suspend(mp)) != 0)
1057 				return (error);
1058 			MNT_ILOCK(mp);
1059 			if (mp->mnt_kern_flag & MNTK_SUSPENDED) {
1060 				mp->mnt_kern_flag &= ~(MNTK_SUSPENDED |
1061 				    MNTK_SUSPEND2);
1062 				wakeup(&mp->mnt_flag);
1063 				MNT_IUNLOCK(mp);
1064 				curthread->td_pflags |= TDP_IGNSUSP;
1065 				break;
1066 			}
1067 			MNT_IUNLOCK(mp);
1068 			vn_start_write(NULL, &mp, V_WAIT);
1069 		}
1070 	}
1071 	if (mp->mnt_flag & MNT_SOFTDEP) {
1072 		if ((error = softdep_flushfiles(mp, flags, td)) != 0)
1073 			goto fail;
1074 	} else {
1075 		if ((error = ffs_flushfiles(mp, flags, td)) != 0)
1076 			goto fail;
1077 	}
1078 	UFS_LOCK(ump);
1079 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1080 		printf("%s: unmount pending error: blocks %jd files %d\n",
1081 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1082 		    fs->fs_pendinginodes);
1083 		fs->fs_pendingblocks = 0;
1084 		fs->fs_pendinginodes = 0;
1085 	}
1086 	UFS_UNLOCK(ump);
1087 	if (fs->fs_ronly == 0) {
1088 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1089 		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1090 		if (error) {
1091 			fs->fs_clean = 0;
1092 			goto fail;
1093 		}
1094 	}
1095 	if (susp) {
1096 		vfs_write_resume(mp);
1097 		vn_start_write(NULL, &mp, V_WAIT);
1098 	}
1099 	DROP_GIANT();
1100 	g_topology_lock();
1101 	g_vfs_close(ump->um_cp);
1102 	g_topology_unlock();
1103 	PICKUP_GIANT();
1104 	vrele(ump->um_devvp);
1105 	mtx_destroy(UFS_MTX(ump));
1106 	if (mp->mnt_gjprovider != NULL) {
1107 		free(mp->mnt_gjprovider, M_UFSMNT);
1108 		mp->mnt_gjprovider = NULL;
1109 	}
1110 	free(fs->fs_csp, M_UFSMNT);
1111 	free(fs, M_UFSMNT);
1112 	free(ump, M_UFSMNT);
1113 	mp->mnt_data = NULL;
1114 	MNT_ILOCK(mp);
1115 	mp->mnt_flag &= ~MNT_LOCAL;
1116 	MNT_IUNLOCK(mp);
1117 	return (error);
1118 
1119 fail:
1120 	if (susp) {
1121 		vfs_write_resume(mp);
1122 		vn_start_write(NULL, &mp, V_WAIT);
1123 	}
1124 	return (error);
1125 }
1126 
1127 /*
1128  * Flush out all the files in a filesystem.
1129  */
1130 int
1131 ffs_flushfiles(mp, flags, td)
1132 	struct mount *mp;
1133 	int flags;
1134 	struct thread *td;
1135 {
1136 	struct ufsmount *ump;
1137 	int error;
1138 
1139 	ump = VFSTOUFS(mp);
1140 #ifdef QUOTA
1141 	if (mp->mnt_flag & MNT_QUOTA) {
1142 		int i;
1143 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1144 		if (error)
1145 			return (error);
1146 		for (i = 0; i < MAXQUOTAS; i++) {
1147 			quotaoff(td, mp, i);
1148 		}
1149 		/*
1150 		 * Here we fall through to vflush again to ensure
1151 		 * that we have gotten rid of all the system vnodes.
1152 		 */
1153 	}
1154 #endif
1155 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1156 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1157 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1158 			return (error);
1159 		ffs_snapshot_unmount(mp);
1160 		flags |= FORCECLOSE;
1161 		/*
1162 		 * Here we fall through to vflush again to ensure
1163 		 * that we have gotten rid of all the system vnodes.
1164 		 */
1165 	}
1166         /*
1167 	 * Flush all the files.
1168 	 */
1169 	if ((error = vflush(mp, 0, flags, td)) != 0)
1170 		return (error);
1171 	/*
1172 	 * Flush filesystem metadata.
1173 	 */
1174 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1175 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1176 	VOP_UNLOCK(ump->um_devvp, 0);
1177 	return (error);
1178 }
1179 
1180 /*
1181  * Get filesystem statistics.
1182  */
1183 static int
1184 ffs_statfs(mp, sbp, td)
1185 	struct mount *mp;
1186 	struct statfs *sbp;
1187 	struct thread *td;
1188 {
1189 	struct ufsmount *ump;
1190 	struct fs *fs;
1191 
1192 	ump = VFSTOUFS(mp);
1193 	fs = ump->um_fs;
1194 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1195 		panic("ffs_statfs");
1196 	sbp->f_version = STATFS_VERSION;
1197 	sbp->f_bsize = fs->fs_fsize;
1198 	sbp->f_iosize = fs->fs_bsize;
1199 	sbp->f_blocks = fs->fs_dsize;
1200 	UFS_LOCK(ump);
1201 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1202 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1203 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1204 	    dbtofsb(fs, fs->fs_pendingblocks);
1205 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1206 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1207 	UFS_UNLOCK(ump);
1208 	sbp->f_namemax = NAME_MAX;
1209 	return (0);
1210 }
1211 
1212 /*
1213  * Go through the disk queues to initiate sandbagged IO;
1214  * go through the inodes to write those that have been modified;
1215  * initiate the writing of the super block if it has been modified.
1216  *
1217  * Note: we are always called with the filesystem marked `MPBUSY'.
1218  */
1219 static int
1220 ffs_sync(mp, waitfor, td)
1221 	struct mount *mp;
1222 	int waitfor;
1223 	struct thread *td;
1224 {
1225 	struct vnode *mvp, *vp, *devvp;
1226 	struct inode *ip;
1227 	struct ufsmount *ump = VFSTOUFS(mp);
1228 	struct fs *fs;
1229 	int error, count, wait, lockreq, allerror = 0;
1230 	int suspend;
1231 	int suspended;
1232 	int secondary_writes;
1233 	int secondary_accwrites;
1234 	int softdep_deps;
1235 	int softdep_accdeps;
1236 	struct bufobj *bo;
1237 
1238 	fs = ump->um_fs;
1239 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
1240 		printf("fs = %s\n", fs->fs_fsmnt);
1241 		panic("ffs_sync: rofs mod");
1242 	}
1243 	/*
1244 	 * Write back each (modified) inode.
1245 	 */
1246 	wait = 0;
1247 	suspend = 0;
1248 	suspended = 0;
1249 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1250 	if (waitfor == MNT_SUSPEND) {
1251 		suspend = 1;
1252 		waitfor = MNT_WAIT;
1253 	}
1254 	if (waitfor == MNT_WAIT) {
1255 		wait = 1;
1256 		lockreq = LK_EXCLUSIVE;
1257 	}
1258 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1259 	MNT_ILOCK(mp);
1260 loop:
1261 	/* Grab snapshot of secondary write counts */
1262 	secondary_writes = mp->mnt_secondary_writes;
1263 	secondary_accwrites = mp->mnt_secondary_accwrites;
1264 
1265 	/* Grab snapshot of softdep dependency counts */
1266 	MNT_IUNLOCK(mp);
1267 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1268 	MNT_ILOCK(mp);
1269 
1270 	MNT_VNODE_FOREACH(vp, mp, mvp) {
1271 		/*
1272 		 * Depend on the mntvnode_slock to keep things stable enough
1273 		 * for a quick test.  Since there might be hundreds of
1274 		 * thousands of vnodes, we cannot afford even a subroutine
1275 		 * call unless there's a good chance that we have work to do.
1276 		 */
1277 		VI_LOCK(vp);
1278 		if (vp->v_iflag & VI_DOOMED) {
1279 			VI_UNLOCK(vp);
1280 			continue;
1281 		}
1282 		ip = VTOI(vp);
1283 		if (vp->v_type == VNON || ((ip->i_flag &
1284 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1285 		    vp->v_bufobj.bo_dirty.bv_cnt == 0)) {
1286 			VI_UNLOCK(vp);
1287 			continue;
1288 		}
1289 		MNT_IUNLOCK(mp);
1290 		if ((error = vget(vp, lockreq, td)) != 0) {
1291 			MNT_ILOCK(mp);
1292 			if (error == ENOENT || error == ENOLCK) {
1293 				MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
1294 				goto loop;
1295 			}
1296 			continue;
1297 		}
1298 		if ((error = ffs_syncvnode(vp, waitfor)) != 0)
1299 			allerror = error;
1300 		vput(vp);
1301 		MNT_ILOCK(mp);
1302 	}
1303 	MNT_IUNLOCK(mp);
1304 	/*
1305 	 * Force stale filesystem control information to be flushed.
1306 	 */
1307 	if (waitfor == MNT_WAIT) {
1308 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1309 			allerror = error;
1310 		/* Flushed work items may create new vnodes to clean */
1311 		if (allerror == 0 && count) {
1312 			MNT_ILOCK(mp);
1313 			goto loop;
1314 		}
1315 	}
1316 #ifdef QUOTA
1317 	qsync(mp);
1318 #endif
1319 	devvp = ump->um_devvp;
1320 	bo = &devvp->v_bufobj;
1321 	BO_LOCK(bo);
1322 	if (waitfor != MNT_LAZY &&
1323 	    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
1324 		BO_UNLOCK(bo);
1325 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1326 		if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0)
1327 			allerror = error;
1328 		VOP_UNLOCK(devvp, 0);
1329 		if (allerror == 0 && waitfor == MNT_WAIT) {
1330 			MNT_ILOCK(mp);
1331 			goto loop;
1332 		}
1333 	} else if (suspend != 0) {
1334 		if (softdep_check_suspend(mp,
1335 					  devvp,
1336 					  softdep_deps,
1337 					  softdep_accdeps,
1338 					  secondary_writes,
1339 					  secondary_accwrites) != 0)
1340 			goto loop;	/* More work needed */
1341 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1342 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1343 		MNT_IUNLOCK(mp);
1344 		suspended = 1;
1345 	} else
1346 		BO_UNLOCK(bo);
1347 	/*
1348 	 * Write back modified superblock.
1349 	 */
1350 	if (fs->fs_fmod != 0 &&
1351 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1352 		allerror = error;
1353 	return (allerror);
1354 }
1355 
1356 int
1357 ffs_vget(mp, ino, flags, vpp)
1358 	struct mount *mp;
1359 	ino_t ino;
1360 	int flags;
1361 	struct vnode **vpp;
1362 {
1363 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1364 }
1365 
1366 int
1367 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1368 	struct mount *mp;
1369 	ino_t ino;
1370 	int flags;
1371 	struct vnode **vpp;
1372 	int ffs_flags;
1373 {
1374 	struct fs *fs;
1375 	struct inode *ip;
1376 	struct ufsmount *ump;
1377 	struct buf *bp;
1378 	struct vnode *vp;
1379 	struct cdev *dev;
1380 	int error;
1381 
1382 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1383 	if (error || *vpp != NULL)
1384 		return (error);
1385 
1386 	/*
1387 	 * We must promote to an exclusive lock for vnode creation.  This
1388 	 * can happen if lookup is passed LOCKSHARED.
1389  	 */
1390 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1391 		flags &= ~LK_TYPE_MASK;
1392 		flags |= LK_EXCLUSIVE;
1393 	}
1394 
1395 	/*
1396 	 * We do not lock vnode creation as it is believed to be too
1397 	 * expensive for such rare case as simultaneous creation of vnode
1398 	 * for same ino by different processes. We just allow them to race
1399 	 * and check later to decide who wins. Let the race begin!
1400 	 */
1401 
1402 	ump = VFSTOUFS(mp);
1403 	dev = ump->um_dev;
1404 	fs = ump->um_fs;
1405 
1406 	/*
1407 	 * If this MALLOC() is performed after the getnewvnode()
1408 	 * it might block, leaving a vnode with a NULL v_data to be
1409 	 * found by ffs_sync() if a sync happens to fire right then,
1410 	 * which will cause a panic because ffs_sync() blindly
1411 	 * dereferences vp->v_data (as well it should).
1412 	 */
1413 	ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1414 
1415 	/* Allocate a new vnode/inode. */
1416 	if (fs->fs_magic == FS_UFS1_MAGIC)
1417 		error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp);
1418 	else
1419 		error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp);
1420 	if (error) {
1421 		*vpp = NULL;
1422 		uma_zfree(uma_inode, ip);
1423 		return (error);
1424 	}
1425 	/*
1426 	 * FFS supports recursive and shared locking.
1427 	 */
1428 	VN_LOCK_AREC(vp);
1429 	VN_LOCK_ASHARE(vp);
1430 	vp->v_data = ip;
1431 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1432 	ip->i_vnode = vp;
1433 	ip->i_ump = ump;
1434 	ip->i_fs = fs;
1435 	ip->i_dev = dev;
1436 	ip->i_number = ino;
1437 #ifdef QUOTA
1438 	{
1439 		int i;
1440 		for (i = 0; i < MAXQUOTAS; i++)
1441 			ip->i_dquot[i] = NODQUOT;
1442 	}
1443 #endif
1444 
1445 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1446 	if (ffs_flags & FFSV_FORCEINSMQ)
1447 		vp->v_vflag |= VV_FORCEINSMQ;
1448 	error = insmntque(vp, mp);
1449 	if (error != 0) {
1450 		uma_zfree(uma_inode, ip);
1451 		*vpp = NULL;
1452 		return (error);
1453 	}
1454 	vp->v_vflag &= ~VV_FORCEINSMQ;
1455 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1456 	if (error || *vpp != NULL)
1457 		return (error);
1458 
1459 	/* Read in the disk contents for the inode, copy into the inode. */
1460 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1461 	    (int)fs->fs_bsize, NOCRED, &bp);
1462 	if (error) {
1463 		/*
1464 		 * The inode does not contain anything useful, so it would
1465 		 * be misleading to leave it on its hash chain. With mode
1466 		 * still zero, it will be unlinked and returned to the free
1467 		 * list by vput().
1468 		 */
1469 		brelse(bp);
1470 		vput(vp);
1471 		*vpp = NULL;
1472 		return (error);
1473 	}
1474 	if (ip->i_ump->um_fstype == UFS1)
1475 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1476 	else
1477 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1478 	ffs_load_inode(bp, ip, fs, ino);
1479 	if (DOINGSOFTDEP(vp))
1480 		softdep_load_inodeblock(ip);
1481 	else
1482 		ip->i_effnlink = ip->i_nlink;
1483 	bqrelse(bp);
1484 
1485 	/*
1486 	 * Initialize the vnode from the inode, check for aliases.
1487 	 * Note that the underlying vnode may have changed.
1488 	 */
1489 	if (ip->i_ump->um_fstype == UFS1)
1490 		error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1491 	else
1492 		error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1493 	if (error) {
1494 		vput(vp);
1495 		*vpp = NULL;
1496 		return (error);
1497 	}
1498 
1499 	/*
1500 	 * Finish inode initialization.
1501 	 */
1502 
1503 	/*
1504 	 * Set up a generation number for this inode if it does not
1505 	 * already have one. This should only happen on old filesystems.
1506 	 */
1507 	if (ip->i_gen == 0) {
1508 		ip->i_gen = arc4random() / 2 + 1;
1509 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1510 			ip->i_flag |= IN_MODIFIED;
1511 			DIP_SET(ip, i_gen, ip->i_gen);
1512 		}
1513 	}
1514 	/*
1515 	 * Ensure that uid and gid are correct. This is a temporary
1516 	 * fix until fsck has been changed to do the update.
1517 	 */
1518 	if (fs->fs_magic == FS_UFS1_MAGIC &&		/* XXX */
1519 	    fs->fs_old_inodefmt < FS_44INODEFMT) {	/* XXX */
1520 		ip->i_uid = ip->i_din1->di_ouid;	/* XXX */
1521 		ip->i_gid = ip->i_din1->di_ogid;	/* XXX */
1522 	}						/* XXX */
1523 
1524 #ifdef MAC
1525 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1526 		/*
1527 		 * If this vnode is already allocated, and we're running
1528 		 * multi-label, attempt to perform a label association
1529 		 * from the extended attributes on the inode.
1530 		 */
1531 		error = mac_vnode_associate_extattr(mp, vp);
1532 		if (error) {
1533 			/* ufs_inactive will release ip->i_devvp ref. */
1534 			vput(vp);
1535 			*vpp = NULL;
1536 			return (error);
1537 		}
1538 	}
1539 #endif
1540 
1541 	*vpp = vp;
1542 	return (0);
1543 }
1544 
1545 /*
1546  * File handle to vnode
1547  *
1548  * Have to be really careful about stale file handles:
1549  * - check that the inode number is valid
1550  * - call ffs_vget() to get the locked inode
1551  * - check for an unallocated inode (i_mode == 0)
1552  * - check that the given client host has export rights and return
1553  *   those rights via. exflagsp and credanonp
1554  */
1555 static int
1556 ffs_fhtovp(mp, fhp, vpp)
1557 	struct mount *mp;
1558 	struct fid *fhp;
1559 	struct vnode **vpp;
1560 {
1561 	struct ufid *ufhp;
1562 	struct fs *fs;
1563 
1564 	ufhp = (struct ufid *)fhp;
1565 	fs = VFSTOUFS(mp)->um_fs;
1566 	if (ufhp->ufid_ino < ROOTINO ||
1567 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1568 		return (ESTALE);
1569 	return (ufs_fhtovp(mp, ufhp, vpp));
1570 }
1571 
1572 /*
1573  * Initialize the filesystem.
1574  */
1575 static int
1576 ffs_init(vfsp)
1577 	struct vfsconf *vfsp;
1578 {
1579 
1580 	softdep_initialize();
1581 	return (ufs_init(vfsp));
1582 }
1583 
1584 /*
1585  * Undo the work of ffs_init().
1586  */
1587 static int
1588 ffs_uninit(vfsp)
1589 	struct vfsconf *vfsp;
1590 {
1591 	int ret;
1592 
1593 	ret = ufs_uninit(vfsp);
1594 	softdep_uninitialize();
1595 	return (ret);
1596 }
1597 
1598 /*
1599  * Write a superblock and associated information back to disk.
1600  */
1601 int
1602 ffs_sbupdate(mp, waitfor, suspended)
1603 	struct ufsmount *mp;
1604 	int waitfor;
1605 	int suspended;
1606 {
1607 	struct fs *fs = mp->um_fs;
1608 	struct buf *sbbp;
1609 	struct buf *bp;
1610 	int blks;
1611 	void *space;
1612 	int i, size, error, allerror = 0;
1613 
1614 	if (fs->fs_ronly == 1 &&
1615 	    (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1616 	    (MNT_RDONLY | MNT_UPDATE))
1617 		panic("ffs_sbupdate: write read-only filesystem");
1618 	/*
1619 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1620 	 */
1621 	sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize,
1622 	    0, 0, 0);
1623 	/*
1624 	 * First write back the summary information.
1625 	 */
1626 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1627 	space = fs->fs_csp;
1628 	for (i = 0; i < blks; i += fs->fs_frag) {
1629 		size = fs->fs_bsize;
1630 		if (i + fs->fs_frag > blks)
1631 			size = (blks - i) * fs->fs_fsize;
1632 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1633 		    size, 0, 0, 0);
1634 		bcopy(space, bp->b_data, (u_int)size);
1635 		space = (char *)space + size;
1636 		if (suspended)
1637 			bp->b_flags |= B_VALIDSUSPWRT;
1638 		if (waitfor != MNT_WAIT)
1639 			bawrite(bp);
1640 		else if ((error = bwrite(bp)) != 0)
1641 			allerror = error;
1642 	}
1643 	/*
1644 	 * Now write back the superblock itself. If any errors occurred
1645 	 * up to this point, then fail so that the superblock avoids
1646 	 * being written out as clean.
1647 	 */
1648 	if (allerror) {
1649 		brelse(sbbp);
1650 		return (allerror);
1651 	}
1652 	bp = sbbp;
1653 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1654 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1655 		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1656 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1657 		fs->fs_sblockloc = SBLOCK_UFS1;
1658 	}
1659 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1660 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1661 		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1662 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1663 		fs->fs_sblockloc = SBLOCK_UFS2;
1664 	}
1665 	fs->fs_fmod = 0;
1666 	fs->fs_time = time_second;
1667 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1668 	ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
1669 	if (suspended)
1670 		bp->b_flags |= B_VALIDSUSPWRT;
1671 	if (waitfor != MNT_WAIT)
1672 		bawrite(bp);
1673 	else if ((error = bwrite(bp)) != 0)
1674 		allerror = error;
1675 	return (allerror);
1676 }
1677 
1678 static int
1679 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1680 	int attrnamespace, const char *attrname, struct thread *td)
1681 {
1682 
1683 #ifdef UFS_EXTATTR
1684 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1685 	    attrname, td));
1686 #else
1687 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1688 	    attrname, td));
1689 #endif
1690 }
1691 
1692 static void
1693 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1694 {
1695 
1696 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1697 		uma_zfree(uma_ufs1, ip->i_din1);
1698 	else if (ip->i_din2 != NULL)
1699 		uma_zfree(uma_ufs2, ip->i_din2);
1700 	uma_zfree(uma_inode, ip);
1701 }
1702 
1703 static int dobkgrdwrite = 1;
1704 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1705     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1706 
1707 /*
1708  * Complete a background write started from bwrite.
1709  */
1710 static void
1711 ffs_backgroundwritedone(struct buf *bp)
1712 {
1713 	struct bufobj *bufobj;
1714 	struct buf *origbp;
1715 
1716 	/*
1717 	 * Find the original buffer that we are writing.
1718 	 */
1719 	bufobj = bp->b_bufobj;
1720 	BO_LOCK(bufobj);
1721 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
1722 		panic("backgroundwritedone: lost buffer");
1723 	/* Grab an extra reference to be dropped by the bufdone() below. */
1724 	bufobj_wrefl(bufobj);
1725 	BO_UNLOCK(bufobj);
1726 	/*
1727 	 * Process dependencies then return any unfinished ones.
1728 	 */
1729 	if (!LIST_EMPTY(&bp->b_dep))
1730 		buf_complete(bp);
1731 #ifdef SOFTUPDATES
1732 	if (!LIST_EMPTY(&bp->b_dep))
1733 		softdep_move_dependencies(bp, origbp);
1734 #endif
1735 	/*
1736 	 * This buffer is marked B_NOCACHE so when it is released
1737 	 * by biodone it will be tossed.
1738 	 */
1739 	bp->b_flags |= B_NOCACHE;
1740 	bp->b_flags &= ~B_CACHE;
1741 	bufdone(bp);
1742 	BO_LOCK(bufobj);
1743 	/*
1744 	 * Clear the BV_BKGRDINPROG flag in the original buffer
1745 	 * and awaken it if it is waiting for the write to complete.
1746 	 * If BV_BKGRDINPROG is not set in the original buffer it must
1747 	 * have been released and re-instantiated - which is not legal.
1748 	 */
1749 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
1750 	    ("backgroundwritedone: lost buffer2"));
1751 	origbp->b_vflags &= ~BV_BKGRDINPROG;
1752 	if (origbp->b_vflags & BV_BKGRDWAIT) {
1753 		origbp->b_vflags &= ~BV_BKGRDWAIT;
1754 		wakeup(&origbp->b_xflags);
1755 	}
1756 	BO_UNLOCK(bufobj);
1757 }
1758 
1759 
1760 /*
1761  * Write, release buffer on completion.  (Done by iodone
1762  * if async).  Do not bother writing anything if the buffer
1763  * is invalid.
1764  *
1765  * Note that we set B_CACHE here, indicating that buffer is
1766  * fully valid and thus cacheable.  This is true even of NFS
1767  * now so we set it generally.  This could be set either here
1768  * or in biodone() since the I/O is synchronous.  We put it
1769  * here.
1770  */
1771 static int
1772 ffs_bufwrite(struct buf *bp)
1773 {
1774 	int oldflags, s;
1775 	struct buf *newbp;
1776 
1777 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1778 	if (bp->b_flags & B_INVAL) {
1779 		brelse(bp);
1780 		return (0);
1781 	}
1782 
1783 	oldflags = bp->b_flags;
1784 
1785 	if (!BUF_ISLOCKED(bp))
1786 		panic("bufwrite: buffer is not busy???");
1787 	s = splbio();
1788 	/*
1789 	 * If a background write is already in progress, delay
1790 	 * writing this block if it is asynchronous. Otherwise
1791 	 * wait for the background write to complete.
1792 	 */
1793 	BO_LOCK(bp->b_bufobj);
1794 	if (bp->b_vflags & BV_BKGRDINPROG) {
1795 		if (bp->b_flags & B_ASYNC) {
1796 			BO_UNLOCK(bp->b_bufobj);
1797 			splx(s);
1798 			bdwrite(bp);
1799 			return (0);
1800 		}
1801 		bp->b_vflags |= BV_BKGRDWAIT;
1802 		msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0);
1803 		if (bp->b_vflags & BV_BKGRDINPROG)
1804 			panic("bufwrite: still writing");
1805 	}
1806 	BO_UNLOCK(bp->b_bufobj);
1807 
1808 	/* Mark the buffer clean */
1809 	bundirty(bp);
1810 
1811 	/*
1812 	 * If this buffer is marked for background writing and we
1813 	 * do not have to wait for it, make a copy and write the
1814 	 * copy so as to leave this buffer ready for further use.
1815 	 *
1816 	 * This optimization eats a lot of memory.  If we have a page
1817 	 * or buffer shortfall we can't do it.
1818 	 */
1819 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
1820 	    (bp->b_flags & B_ASYNC) &&
1821 	    !vm_page_count_severe() &&
1822 	    !buf_dirty_count_severe()) {
1823 		KASSERT(bp->b_iodone == NULL,
1824 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
1825 
1826 		/* get a new block */
1827 		newbp = geteblk(bp->b_bufsize);
1828 
1829 		/*
1830 		 * set it to be identical to the old block.  We have to
1831 		 * set b_lblkno and BKGRDMARKER before calling bgetvp()
1832 		 * to avoid confusing the splay tree and gbincore().
1833 		 */
1834 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
1835 		newbp->b_lblkno = bp->b_lblkno;
1836 		newbp->b_xflags |= BX_BKGRDMARKER;
1837 		BO_LOCK(bp->b_bufobj);
1838 		bp->b_vflags |= BV_BKGRDINPROG;
1839 		bgetvp(bp->b_vp, newbp);
1840 		BO_UNLOCK(bp->b_bufobj);
1841 		newbp->b_bufobj = &bp->b_vp->v_bufobj;
1842 		newbp->b_blkno = bp->b_blkno;
1843 		newbp->b_offset = bp->b_offset;
1844 		newbp->b_iodone = ffs_backgroundwritedone;
1845 		newbp->b_flags |= B_ASYNC;
1846 		newbp->b_flags &= ~B_INVAL;
1847 
1848 #ifdef SOFTUPDATES
1849 		/* move over the dependencies */
1850 		if (!LIST_EMPTY(&bp->b_dep))
1851 			softdep_move_dependencies(bp, newbp);
1852 #endif
1853 
1854 		/*
1855 		 * Initiate write on the copy, release the original to
1856 		 * the B_LOCKED queue so that it cannot go away until
1857 		 * the background write completes. If not locked it could go
1858 		 * away and then be reconstituted while it was being written.
1859 		 * If the reconstituted buffer were written, we could end up
1860 		 * with two background copies being written at the same time.
1861 		 */
1862 		bqrelse(bp);
1863 		bp = newbp;
1864 	}
1865 
1866 	/* Let the normal bufwrite do the rest for us */
1867 	return (bufwrite(bp));
1868 }
1869 
1870 
1871 static void
1872 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
1873 {
1874 	struct vnode *vp;
1875 	int error;
1876 	struct buf *tbp;
1877 
1878 	vp = bo->__bo_vnode;
1879 	if (bp->b_iocmd == BIO_WRITE) {
1880 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
1881 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
1882 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1883 			panic("ffs_geom_strategy: bad I/O");
1884 		bp->b_flags &= ~B_VALIDSUSPWRT;
1885 		if ((vp->v_vflag & VV_COPYONWRITE) &&
1886 		    vp->v_rdev->si_snapdata != NULL) {
1887 			if ((bp->b_flags & B_CLUSTER) != 0) {
1888 				runningbufwakeup(bp);
1889 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
1890 					      b_cluster.cluster_entry) {
1891 					error = ffs_copyonwrite(vp, tbp);
1892 					if (error != 0 &&
1893 					    error != EOPNOTSUPP) {
1894 						bp->b_error = error;
1895 						bp->b_ioflags |= BIO_ERROR;
1896 						bufdone(bp);
1897 						return;
1898 					}
1899 				}
1900 				bp->b_runningbufspace = bp->b_bufsize;
1901 				atomic_add_int(&runningbufspace,
1902 					       bp->b_runningbufspace);
1903 			} else {
1904 				error = ffs_copyonwrite(vp, bp);
1905 				if (error != 0 && error != EOPNOTSUPP) {
1906 					bp->b_error = error;
1907 					bp->b_ioflags |= BIO_ERROR;
1908 					bufdone(bp);
1909 					return;
1910 				}
1911 			}
1912 		}
1913 #ifdef SOFTUPDATES
1914 		if ((bp->b_flags & B_CLUSTER) != 0) {
1915 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
1916 				      b_cluster.cluster_entry) {
1917 				if (!LIST_EMPTY(&tbp->b_dep))
1918 					buf_start(tbp);
1919 			}
1920 		} else {
1921 			if (!LIST_EMPTY(&bp->b_dep))
1922 				buf_start(bp);
1923 		}
1924 
1925 #endif
1926 	}
1927 	g_vfs_strategy(bo, bp);
1928 }
1929 
1930 #ifdef	DDB
1931 
1932 static void
1933 db_print_ffs(struct ufsmount *ump)
1934 {
1935 	db_printf("mp %p %s devvp %p fs %p su_wl %d su_wl_in %d su_deps %d "
1936 		  "su_req %d\n",
1937 	    ump->um_mountp, ump->um_mountp->mnt_stat.f_mntonname,
1938 	    ump->um_devvp, ump->um_fs, ump->softdep_on_worklist,
1939 	    ump->softdep_on_worklist_inprogress, ump->softdep_deps,
1940 	    ump->softdep_req);
1941 }
1942 
1943 DB_SHOW_COMMAND(ffs, db_show_ffs)
1944 {
1945 	struct mount *mp;
1946 	struct ufsmount *ump;
1947 
1948 	if (have_addr) {
1949 		ump = VFSTOUFS((struct mount *)addr);
1950 		db_print_ffs(ump);
1951 		return;
1952 	}
1953 
1954 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1955 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
1956 			db_print_ffs(VFSTOUFS(mp));
1957 	}
1958 }
1959 
1960 #endif	/* DDB */
1961