xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision c203bd70b5957f85616424b6fa374479372d06e3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1991, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_quota.h"
38 #include "opt_ufs.h"
39 #include "opt_ffs.h"
40 #include "opt_ddb.h"
41 
42 #include <sys/param.h>
43 #include <sys/gsb_crc32.h>
44 #include <sys/systm.h>
45 #include <sys/namei.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/kernel.h>
50 #include <sys/ktr.h>
51 #include <sys/vnode.h>
52 #include <sys/mount.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/conf.h>
56 #include <sys/fcntl.h>
57 #include <sys/ioccom.h>
58 #include <sys/malloc.h>
59 #include <sys/mutex.h>
60 #include <sys/rwlock.h>
61 #include <sys/sysctl.h>
62 #include <sys/vmmeter.h>
63 
64 #include <security/mac/mac_framework.h>
65 
66 #include <ufs/ufs/dir.h>
67 #include <ufs/ufs/extattr.h>
68 #include <ufs/ufs/gjournal.h>
69 #include <ufs/ufs/quota.h>
70 #include <ufs/ufs/ufsmount.h>
71 #include <ufs/ufs/inode.h>
72 #include <ufs/ufs/ufs_extern.h>
73 
74 #include <ufs/ffs/fs.h>
75 #include <ufs/ffs/ffs_extern.h>
76 
77 #include <vm/vm.h>
78 #include <vm/uma.h>
79 #include <vm/vm_page.h>
80 
81 #include <geom/geom.h>
82 #include <geom/geom_vfs.h>
83 
84 #include <ddb/ddb.h>
85 
86 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
87 VFS_SMR_DECLARE;
88 
89 static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
90 static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
91 		    ufs2_daddr_t);
92 static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
93 static int	ffs_sync_lazy(struct mount *mp);
94 static int	ffs_use_bread(void *devfd, off_t loc, void **bufp, int size);
95 static int	ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size);
96 
97 static vfs_init_t ffs_init;
98 static vfs_uninit_t ffs_uninit;
99 static vfs_extattrctl_t ffs_extattrctl;
100 static vfs_cmount_t ffs_cmount;
101 static vfs_unmount_t ffs_unmount;
102 static vfs_mount_t ffs_mount;
103 static vfs_statfs_t ffs_statfs;
104 static vfs_fhtovp_t ffs_fhtovp;
105 static vfs_sync_t ffs_sync;
106 
107 static struct vfsops ufs_vfsops = {
108 	.vfs_extattrctl =	ffs_extattrctl,
109 	.vfs_fhtovp =		ffs_fhtovp,
110 	.vfs_init =		ffs_init,
111 	.vfs_mount =		ffs_mount,
112 	.vfs_cmount =		ffs_cmount,
113 	.vfs_quotactl =		ufs_quotactl,
114 	.vfs_root =		vfs_cache_root,
115 	.vfs_cachedroot =	ufs_root,
116 	.vfs_statfs =		ffs_statfs,
117 	.vfs_sync =		ffs_sync,
118 	.vfs_uninit =		ffs_uninit,
119 	.vfs_unmount =		ffs_unmount,
120 	.vfs_vget =		ffs_vget,
121 	.vfs_susp_clean =	process_deferred_inactive,
122 };
123 
124 VFS_SET(ufs_vfsops, ufs, 0);
125 MODULE_VERSION(ufs, 1);
126 
127 static b_strategy_t ffs_geom_strategy;
128 static b_write_t ffs_bufwrite;
129 
130 static struct buf_ops ffs_ops = {
131 	.bop_name =	"FFS",
132 	.bop_write =	ffs_bufwrite,
133 	.bop_strategy =	ffs_geom_strategy,
134 	.bop_sync =	bufsync,
135 #ifdef NO_FFS_SNAPSHOT
136 	.bop_bdflush =	bufbdflush,
137 #else
138 	.bop_bdflush =	ffs_bdflush,
139 #endif
140 };
141 
142 /*
143  * Note that userquota and groupquota options are not currently used
144  * by UFS/FFS code and generally mount(8) does not pass those options
145  * from userland, but they can be passed by loader(8) via
146  * vfs.root.mountfrom.options.
147  */
148 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
149     "noclusterw", "noexec", "export", "force", "from", "groupquota",
150     "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
151     "nosymfollow", "sync", "union", "userquota", "untrusted", NULL };
152 
153 static int ffs_enxio_enable = 1;
154 SYSCTL_DECL(_vfs_ffs);
155 SYSCTL_INT(_vfs_ffs, OID_AUTO, enxio_enable, CTLFLAG_RWTUN,
156     &ffs_enxio_enable, 0,
157     "enable mapping of other disk I/O errors to ENXIO");
158 
159 /*
160  * Return buffer with the contents of block "offset" from the beginning of
161  * directory "ip".  If "res" is non-zero, fill it in with a pointer to the
162  * remaining space in the directory.
163  */
164 static int
165 ffs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp)
166 {
167 	struct inode *ip;
168 	struct fs *fs;
169 	struct buf *bp;
170 	ufs_lbn_t lbn;
171 	int bsize, error;
172 
173 	ip = VTOI(vp);
174 	fs = ITOFS(ip);
175 	lbn = lblkno(fs, offset);
176 	bsize = blksize(fs, ip, lbn);
177 
178 	*bpp = NULL;
179 	error = bread(vp, lbn, bsize, NOCRED, &bp);
180 	if (error) {
181 		return (error);
182 	}
183 	if (res)
184 		*res = (char *)bp->b_data + blkoff(fs, offset);
185 	*bpp = bp;
186 	return (0);
187 }
188 
189 /*
190  * Load up the contents of an inode and copy the appropriate pieces
191  * to the incore copy.
192  */
193 static int
194 ffs_load_inode(struct buf *bp, struct inode *ip, struct fs *fs, ino_t ino)
195 {
196 	struct ufs1_dinode *dip1;
197 	struct ufs2_dinode *dip2;
198 	int error;
199 
200 	if (I_IS_UFS1(ip)) {
201 		dip1 = ip->i_din1;
202 		*dip1 =
203 		    *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
204 		ip->i_mode = dip1->di_mode;
205 		ip->i_nlink = dip1->di_nlink;
206 		ip->i_effnlink = dip1->di_nlink;
207 		ip->i_size = dip1->di_size;
208 		ip->i_flags = dip1->di_flags;
209 		ip->i_gen = dip1->di_gen;
210 		ip->i_uid = dip1->di_uid;
211 		ip->i_gid = dip1->di_gid;
212 		return (0);
213 	}
214 	dip2 = ((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
215 	if ((error = ffs_verify_dinode_ckhash(fs, dip2)) != 0 &&
216 	    !ffs_fsfail_cleanup(ITOUMP(ip), error)) {
217 		printf("%s: inode %jd: check-hash failed\n", fs->fs_fsmnt,
218 		    (intmax_t)ino);
219 		return (error);
220 	}
221 	*ip->i_din2 = *dip2;
222 	dip2 = ip->i_din2;
223 	ip->i_mode = dip2->di_mode;
224 	ip->i_nlink = dip2->di_nlink;
225 	ip->i_effnlink = dip2->di_nlink;
226 	ip->i_size = dip2->di_size;
227 	ip->i_flags = dip2->di_flags;
228 	ip->i_gen = dip2->di_gen;
229 	ip->i_uid = dip2->di_uid;
230 	ip->i_gid = dip2->di_gid;
231 	return (0);
232 }
233 
234 /*
235  * Verify that a filesystem block number is a valid data block.
236  * This routine is only called on untrusted filesystems.
237  */
238 static int
239 ffs_check_blkno(struct mount *mp, ino_t inum, ufs2_daddr_t daddr, int blksize)
240 {
241 	struct fs *fs;
242 	struct ufsmount *ump;
243 	ufs2_daddr_t end_daddr;
244 	int cg, havemtx;
245 
246 	KASSERT((mp->mnt_flag & MNT_UNTRUSTED) != 0,
247 	    ("ffs_check_blkno called on a trusted file system"));
248 	ump = VFSTOUFS(mp);
249 	fs = ump->um_fs;
250 	cg = dtog(fs, daddr);
251 	end_daddr = daddr + numfrags(fs, blksize);
252 	/*
253 	 * Verify that the block number is a valid data block. Also check
254 	 * that it does not point to an inode block or a superblock. Accept
255 	 * blocks that are unalloacted (0) or part of snapshot metadata
256 	 * (BLK_NOCOPY or BLK_SNAP).
257 	 *
258 	 * Thus, the block must be in a valid range for the filesystem and
259 	 * either in the space before a backup superblock (except the first
260 	 * cylinder group where that space is used by the bootstrap code) or
261 	 * after the inode blocks and before the end of the cylinder group.
262 	 */
263 	if ((uint64_t)daddr <= BLK_SNAP ||
264 	    ((uint64_t)end_daddr <= fs->fs_size &&
265 	    ((cg > 0 && end_daddr <= cgsblock(fs, cg)) ||
266 	    (daddr >= cgdmin(fs, cg) &&
267 	    end_daddr <= cgbase(fs, cg) + fs->fs_fpg))))
268 		return (0);
269 	if ((havemtx = mtx_owned(UFS_MTX(ump))) == 0)
270 		UFS_LOCK(ump);
271 	if (ppsratecheck(&ump->um_last_integritymsg,
272 	    &ump->um_secs_integritymsg, 1)) {
273 		UFS_UNLOCK(ump);
274 		uprintf("\n%s: inode %jd, out-of-range indirect block "
275 		    "number %jd\n", mp->mnt_stat.f_mntonname, inum, daddr);
276 		if (havemtx)
277 			UFS_LOCK(ump);
278 	} else if (!havemtx)
279 		UFS_UNLOCK(ump);
280 	return (EINTEGRITY);
281 }
282 
283 /*
284  * Initiate a forcible unmount.
285  * Used to unmount filesystems whose underlying media has gone away.
286  */
287 static void
288 ffs_fsfail_unmount(void *v, int pending)
289 {
290 	struct fsfail_task *etp;
291 	struct mount *mp;
292 
293 	etp = v;
294 
295 	/*
296 	 * Find our mount and get a ref on it, then try to unmount.
297 	 */
298 	mp = vfs_getvfs(&etp->fsid);
299 	if (mp != NULL)
300 		dounmount(mp, MNT_FORCE, curthread);
301 	free(etp, M_UFSMNT);
302 }
303 
304 /*
305  * On first ENXIO error, start a task that forcibly unmounts the filesystem.
306  *
307  * Return true if a cleanup is in progress.
308  */
309 int
310 ffs_fsfail_cleanup(struct ufsmount *ump, int error)
311 {
312 	int retval;
313 
314 	UFS_LOCK(ump);
315 	retval = ffs_fsfail_cleanup_locked(ump, error);
316 	UFS_UNLOCK(ump);
317 	return (retval);
318 }
319 
320 int
321 ffs_fsfail_cleanup_locked(struct ufsmount *ump, int error)
322 {
323 	struct fsfail_task *etp;
324 	struct task *tp;
325 
326 	mtx_assert(UFS_MTX(ump), MA_OWNED);
327 	if (error == ENXIO && (ump->um_flags & UM_FSFAIL_CLEANUP) == 0) {
328 		ump->um_flags |= UM_FSFAIL_CLEANUP;
329 		/*
330 		 * Queue an async forced unmount.
331 		 */
332 		etp = ump->um_fsfail_task;
333 		ump->um_fsfail_task = NULL;
334 		if (etp != NULL) {
335 			tp = &etp->task;
336 			TASK_INIT(tp, 0, ffs_fsfail_unmount, etp);
337 			taskqueue_enqueue(taskqueue_thread, tp);
338 			printf("UFS: forcibly unmounting %s from %s\n",
339 			    ump->um_mountp->mnt_stat.f_mntfromname,
340 			    ump->um_mountp->mnt_stat.f_mntonname);
341 		}
342 	}
343 	return ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0);
344 }
345 
346 /*
347  * Wrapper used during ENXIO cleanup to allocate empty buffers when
348  * the kernel is unable to read the real one. They are needed so that
349  * the soft updates code can use them to unwind its dependencies.
350  */
351 int
352 ffs_breadz(struct ufsmount *ump, struct vnode *vp, daddr_t lblkno,
353     daddr_t dblkno, int size, daddr_t *rablkno, int *rabsize, int cnt,
354     struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *),
355     struct buf **bpp)
356 {
357 	int error;
358 
359 	flags |= GB_CVTENXIO;
360 	error = breadn_flags(vp, lblkno, dblkno, size, rablkno, rabsize, cnt,
361 	    cred, flags, ckhashfunc, bpp);
362 	if (error != 0 && ffs_fsfail_cleanup(ump, error)) {
363 		error = getblkx(vp, lblkno, dblkno, size, 0, 0, flags, bpp);
364 		KASSERT(error == 0, ("getblkx failed"));
365 		vfs_bio_bzero_buf(*bpp, 0, size);
366 	}
367 	return (error);
368 }
369 
370 static int
371 ffs_mount(struct mount *mp)
372 {
373 	struct vnode *devvp, *odevvp;
374 	struct thread *td;
375 	struct ufsmount *ump = NULL;
376 	struct fs *fs;
377 	pid_t fsckpid = 0;
378 	int error, error1, flags;
379 	uint64_t mntorflags, saved_mnt_flag;
380 	accmode_t accmode;
381 	struct nameidata ndp;
382 	char *fspec;
383 
384 	td = curthread;
385 	if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
386 		return (EINVAL);
387 	if (uma_inode == NULL) {
388 		uma_inode = uma_zcreate("FFS inode",
389 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
390 		    UMA_ALIGN_PTR, 0);
391 		uma_ufs1 = uma_zcreate("FFS1 dinode",
392 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
393 		    UMA_ALIGN_PTR, 0);
394 		uma_ufs2 = uma_zcreate("FFS2 dinode",
395 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
396 		    UMA_ALIGN_PTR, 0);
397 		VFS_SMR_ZONE_SET(uma_inode);
398 	}
399 
400 	vfs_deleteopt(mp->mnt_optnew, "groupquota");
401 	vfs_deleteopt(mp->mnt_optnew, "userquota");
402 
403 	fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
404 	if (error)
405 		return (error);
406 
407 	mntorflags = 0;
408 	if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0)
409 		mntorflags |= MNT_UNTRUSTED;
410 
411 	if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
412 		mntorflags |= MNT_ACLS;
413 
414 	if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
415 		mntorflags |= MNT_SNAPSHOT;
416 		/*
417 		 * Once we have set the MNT_SNAPSHOT flag, do not
418 		 * persist "snapshot" in the options list.
419 		 */
420 		vfs_deleteopt(mp->mnt_optnew, "snapshot");
421 		vfs_deleteopt(mp->mnt_opt, "snapshot");
422 	}
423 
424 	if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
425 	    vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
426 		/*
427 		 * Once we have set the restricted PID, do not
428 		 * persist "fsckpid" in the options list.
429 		 */
430 		vfs_deleteopt(mp->mnt_optnew, "fsckpid");
431 		vfs_deleteopt(mp->mnt_opt, "fsckpid");
432 		if (mp->mnt_flag & MNT_UPDATE) {
433 			if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
434 			     vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
435 				vfs_mount_error(mp,
436 				    "Checker enable: Must be read-only");
437 				return (EINVAL);
438 			}
439 		} else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
440 			vfs_mount_error(mp,
441 			    "Checker enable: Must be read-only");
442 			return (EINVAL);
443 		}
444 		/* Set to -1 if we are done */
445 		if (fsckpid == 0)
446 			fsckpid = -1;
447 	}
448 
449 	if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
450 		if (mntorflags & MNT_ACLS) {
451 			vfs_mount_error(mp,
452 			    "\"acls\" and \"nfsv4acls\" options "
453 			    "are mutually exclusive");
454 			return (EINVAL);
455 		}
456 		mntorflags |= MNT_NFS4ACLS;
457 	}
458 
459 	MNT_ILOCK(mp);
460 	mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
461 	mp->mnt_flag |= mntorflags;
462 	MNT_IUNLOCK(mp);
463 	/*
464 	 * If updating, check whether changing from read-only to
465 	 * read/write; if there is no device name, that's all we do.
466 	 */
467 	if (mp->mnt_flag & MNT_UPDATE) {
468 		ump = VFSTOUFS(mp);
469 		fs = ump->um_fs;
470 		odevvp = ump->um_odevvp;
471 		devvp = ump->um_devvp;
472 		if (fsckpid == -1 && ump->um_fsckpid > 0) {
473 			if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
474 			    (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
475 				return (error);
476 			g_topology_lock();
477 			/*
478 			 * Return to normal read-only mode.
479 			 */
480 			error = g_access(ump->um_cp, 0, -1, 0);
481 			g_topology_unlock();
482 			ump->um_fsckpid = 0;
483 		}
484 		if (fs->fs_ronly == 0 &&
485 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
486 			/*
487 			 * Flush any dirty data and suspend filesystem.
488 			 */
489 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
490 				return (error);
491 			error = vfs_write_suspend_umnt(mp);
492 			if (error != 0)
493 				return (error);
494 			/*
495 			 * Check for and optionally get rid of files open
496 			 * for writing.
497 			 */
498 			flags = WRITECLOSE;
499 			if (mp->mnt_flag & MNT_FORCE)
500 				flags |= FORCECLOSE;
501 			if (MOUNTEDSOFTDEP(mp)) {
502 				error = softdep_flushfiles(mp, flags, td);
503 			} else {
504 				error = ffs_flushfiles(mp, flags, td);
505 			}
506 			if (error) {
507 				vfs_write_resume(mp, 0);
508 				return (error);
509 			}
510 			if (fs->fs_pendingblocks != 0 ||
511 			    fs->fs_pendinginodes != 0) {
512 				printf("WARNING: %s Update error: blocks %jd "
513 				    "files %d\n", fs->fs_fsmnt,
514 				    (intmax_t)fs->fs_pendingblocks,
515 				    fs->fs_pendinginodes);
516 				fs->fs_pendingblocks = 0;
517 				fs->fs_pendinginodes = 0;
518 			}
519 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
520 				fs->fs_clean = 1;
521 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
522 				fs->fs_ronly = 0;
523 				fs->fs_clean = 0;
524 				vfs_write_resume(mp, 0);
525 				return (error);
526 			}
527 			if (MOUNTEDSOFTDEP(mp))
528 				softdep_unmount(mp);
529 			g_topology_lock();
530 			/*
531 			 * Drop our write and exclusive access.
532 			 */
533 			g_access(ump->um_cp, 0, -1, -1);
534 			g_topology_unlock();
535 			fs->fs_ronly = 1;
536 			MNT_ILOCK(mp);
537 			mp->mnt_flag |= MNT_RDONLY;
538 			MNT_IUNLOCK(mp);
539 			/*
540 			 * Allow the writers to note that filesystem
541 			 * is ro now.
542 			 */
543 			vfs_write_resume(mp, 0);
544 		}
545 		if ((mp->mnt_flag & MNT_RELOAD) &&
546 		    (error = ffs_reload(mp, td, 0)) != 0)
547 			return (error);
548 		if (fs->fs_ronly &&
549 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
550 			/*
551 			 * If we are running a checker, do not allow upgrade.
552 			 */
553 			if (ump->um_fsckpid > 0) {
554 				vfs_mount_error(mp,
555 				    "Active checker, cannot upgrade to write");
556 				return (EINVAL);
557 			}
558 			/*
559 			 * If upgrade to read-write by non-root, then verify
560 			 * that user has necessary permissions on the device.
561 			 */
562 			vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY);
563 			error = VOP_ACCESS(odevvp, VREAD | VWRITE,
564 			    td->td_ucred, td);
565 			if (error)
566 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
567 			VOP_UNLOCK(odevvp);
568 			if (error) {
569 				return (error);
570 			}
571 			fs->fs_flags &= ~FS_UNCLEAN;
572 			if (fs->fs_clean == 0) {
573 				fs->fs_flags |= FS_UNCLEAN;
574 				if ((mp->mnt_flag & MNT_FORCE) ||
575 				    ((fs->fs_flags &
576 				     (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
577 				     (fs->fs_flags & FS_DOSOFTDEP))) {
578 					printf("WARNING: %s was not properly "
579 					   "dismounted\n", fs->fs_fsmnt);
580 				} else {
581 					vfs_mount_error(mp,
582 					   "R/W mount of %s denied. %s.%s",
583 					   fs->fs_fsmnt,
584 					   "Filesystem is not clean - run fsck",
585 					   (fs->fs_flags & FS_SUJ) == 0 ? "" :
586 					   " Forced mount will invalidate"
587 					   " journal contents");
588 					return (EPERM);
589 				}
590 			}
591 			g_topology_lock();
592 			/*
593 			 * Request exclusive write access.
594 			 */
595 			error = g_access(ump->um_cp, 0, 1, 1);
596 			g_topology_unlock();
597 			if (error)
598 				return (error);
599 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
600 				return (error);
601 			error = vfs_write_suspend_umnt(mp);
602 			if (error != 0)
603 				return (error);
604 			fs->fs_ronly = 0;
605 			MNT_ILOCK(mp);
606 			saved_mnt_flag = MNT_RDONLY;
607 			if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag &
608 			    MNT_ASYNC) != 0)
609 				saved_mnt_flag |= MNT_ASYNC;
610 			mp->mnt_flag &= ~saved_mnt_flag;
611 			MNT_IUNLOCK(mp);
612 			fs->fs_mtime = time_second;
613 			/* check to see if we need to start softdep */
614 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
615 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
616 				fs->fs_ronly = 1;
617 				MNT_ILOCK(mp);
618 				mp->mnt_flag |= saved_mnt_flag;
619 				MNT_IUNLOCK(mp);
620 				vfs_write_resume(mp, 0);
621 				return (error);
622 			}
623 			fs->fs_clean = 0;
624 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
625 				fs->fs_ronly = 1;
626 				MNT_ILOCK(mp);
627 				mp->mnt_flag |= saved_mnt_flag;
628 				MNT_IUNLOCK(mp);
629 				vfs_write_resume(mp, 0);
630 				return (error);
631 			}
632 			if (fs->fs_snapinum[0] != 0)
633 				ffs_snapshot_mount(mp);
634 			vfs_write_resume(mp, 0);
635 		}
636 		/*
637 		 * Soft updates is incompatible with "async",
638 		 * so if we are doing softupdates stop the user
639 		 * from setting the async flag in an update.
640 		 * Softdep_mount() clears it in an initial mount
641 		 * or ro->rw remount.
642 		 */
643 		if (MOUNTEDSOFTDEP(mp)) {
644 			/* XXX: Reset too late ? */
645 			MNT_ILOCK(mp);
646 			mp->mnt_flag &= ~MNT_ASYNC;
647 			MNT_IUNLOCK(mp);
648 		}
649 		/*
650 		 * Keep MNT_ACLS flag if it is stored in superblock.
651 		 */
652 		if ((fs->fs_flags & FS_ACLS) != 0) {
653 			/* XXX: Set too late ? */
654 			MNT_ILOCK(mp);
655 			mp->mnt_flag |= MNT_ACLS;
656 			MNT_IUNLOCK(mp);
657 		}
658 
659 		if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
660 			/* XXX: Set too late ? */
661 			MNT_ILOCK(mp);
662 			mp->mnt_flag |= MNT_NFS4ACLS;
663 			MNT_IUNLOCK(mp);
664 		}
665 		/*
666 		 * If this is a request from fsck to clean up the filesystem,
667 		 * then allow the specified pid to proceed.
668 		 */
669 		if (fsckpid > 0) {
670 			if (ump->um_fsckpid != 0) {
671 				vfs_mount_error(mp,
672 				    "Active checker already running on %s",
673 				    fs->fs_fsmnt);
674 				return (EINVAL);
675 			}
676 			KASSERT(MOUNTEDSOFTDEP(mp) == 0,
677 			    ("soft updates enabled on read-only file system"));
678 			g_topology_lock();
679 			/*
680 			 * Request write access.
681 			 */
682 			error = g_access(ump->um_cp, 0, 1, 0);
683 			g_topology_unlock();
684 			if (error) {
685 				vfs_mount_error(mp,
686 				    "Checker activation failed on %s",
687 				    fs->fs_fsmnt);
688 				return (error);
689 			}
690 			ump->um_fsckpid = fsckpid;
691 			if (fs->fs_snapinum[0] != 0)
692 				ffs_snapshot_mount(mp);
693 			fs->fs_mtime = time_second;
694 			fs->fs_fmod = 1;
695 			fs->fs_clean = 0;
696 			(void) ffs_sbupdate(ump, MNT_WAIT, 0);
697 		}
698 
699 		/*
700 		 * If this is a snapshot request, take the snapshot.
701 		 */
702 		if (mp->mnt_flag & MNT_SNAPSHOT)
703 			return (ffs_snapshot(mp, fspec));
704 
705 		/*
706 		 * Must not call namei() while owning busy ref.
707 		 */
708 		vfs_unbusy(mp);
709 	}
710 
711 	/*
712 	 * Not an update, or updating the name: look up the name
713 	 * and verify that it refers to a sensible disk device.
714 	 */
715 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
716 	error = namei(&ndp);
717 	if ((mp->mnt_flag & MNT_UPDATE) != 0) {
718 		/*
719 		 * Unmount does not start if MNT_UPDATE is set.  Mount
720 		 * update busies mp before setting MNT_UPDATE.  We
721 		 * must be able to retain our busy ref succesfully,
722 		 * without sleep.
723 		 */
724 		error1 = vfs_busy(mp, MBF_NOWAIT);
725 		MPASS(error1 == 0);
726 	}
727 	if (error != 0)
728 		return (error);
729 	NDFREE(&ndp, NDF_ONLY_PNBUF);
730 	devvp = ndp.ni_vp;
731 	if (!vn_isdisk_error(devvp, &error)) {
732 		vput(devvp);
733 		return (error);
734 	}
735 
736 	/*
737 	 * If mount by non-root, then verify that user has necessary
738 	 * permissions on the device.
739 	 */
740 	accmode = VREAD;
741 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
742 		accmode |= VWRITE;
743 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
744 	if (error)
745 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
746 	if (error) {
747 		vput(devvp);
748 		return (error);
749 	}
750 
751 	if (mp->mnt_flag & MNT_UPDATE) {
752 		/*
753 		 * Update only
754 		 *
755 		 * If it's not the same vnode, or at least the same device
756 		 * then it's not correct.
757 		 */
758 
759 		if (devvp->v_rdev != ump->um_devvp->v_rdev)
760 			error = EINVAL;	/* needs translation */
761 		vput(devvp);
762 		if (error)
763 			return (error);
764 	} else {
765 		/*
766 		 * New mount
767 		 *
768 		 * We need the name for the mount point (also used for
769 		 * "last mounted on") copied in. If an error occurs,
770 		 * the mount point is discarded by the upper level code.
771 		 * Note that vfs_mount_alloc() populates f_mntonname for us.
772 		 */
773 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
774 			vrele(devvp);
775 			return (error);
776 		}
777 		if (fsckpid > 0) {
778 			KASSERT(MOUNTEDSOFTDEP(mp) == 0,
779 			    ("soft updates enabled on read-only file system"));
780 			ump = VFSTOUFS(mp);
781 			fs = ump->um_fs;
782 			g_topology_lock();
783 			/*
784 			 * Request write access.
785 			 */
786 			error = g_access(ump->um_cp, 0, 1, 0);
787 			g_topology_unlock();
788 			if (error) {
789 				printf("WARNING: %s: Checker activation "
790 				    "failed\n", fs->fs_fsmnt);
791 			} else {
792 				ump->um_fsckpid = fsckpid;
793 				if (fs->fs_snapinum[0] != 0)
794 					ffs_snapshot_mount(mp);
795 				fs->fs_mtime = time_second;
796 				fs->fs_clean = 0;
797 				(void) ffs_sbupdate(ump, MNT_WAIT, 0);
798 			}
799 		}
800 	}
801 
802 	MNT_ILOCK(mp);
803 	/*
804 	 * This is racy versus lookup, see ufs_fplookup_vexec for details.
805 	 */
806 	if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) != 0)
807 		panic("MNTK_FPLOOKUP set on mount %p when it should not be", mp);
808 	if ((mp->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS | MNT_UNION)) == 0)
809 		mp->mnt_kern_flag |= MNTK_FPLOOKUP;
810 	MNT_IUNLOCK(mp);
811 
812 	vfs_mountedfrom(mp, fspec);
813 	return (0);
814 }
815 
816 /*
817  * Compatibility with old mount system call.
818  */
819 
820 static int
821 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
822 {
823 	struct ufs_args args;
824 	int error;
825 
826 	if (data == NULL)
827 		return (EINVAL);
828 	error = copyin(data, &args, sizeof args);
829 	if (error)
830 		return (error);
831 
832 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
833 	ma = mount_arg(ma, "export", &args.export, sizeof(args.export));
834 	error = kernel_mount(ma, flags);
835 
836 	return (error);
837 }
838 
839 /*
840  * Reload all incore data for a filesystem (used after running fsck on
841  * the root filesystem and finding things to fix). If the 'force' flag
842  * is 0, the filesystem must be mounted read-only.
843  *
844  * Things to do to update the mount:
845  *	1) invalidate all cached meta-data.
846  *	2) re-read superblock from disk.
847  *	3) re-read summary information from disk.
848  *	4) invalidate all inactive vnodes.
849  *	5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary
850  *	   writers, if requested.
851  *	6) invalidate all cached file data.
852  *	7) re-read inode data for all active vnodes.
853  */
854 int
855 ffs_reload(struct mount *mp, struct thread *td, int flags)
856 {
857 	struct vnode *vp, *mvp, *devvp;
858 	struct inode *ip;
859 	void *space;
860 	struct buf *bp;
861 	struct fs *fs, *newfs;
862 	struct ufsmount *ump;
863 	ufs2_daddr_t sblockloc;
864 	int i, blks, error;
865 	u_long size;
866 	int32_t *lp;
867 
868 	ump = VFSTOUFS(mp);
869 
870 	MNT_ILOCK(mp);
871 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) {
872 		MNT_IUNLOCK(mp);
873 		return (EINVAL);
874 	}
875 	MNT_IUNLOCK(mp);
876 
877 	/*
878 	 * Step 1: invalidate all cached meta-data.
879 	 */
880 	devvp = VFSTOUFS(mp)->um_devvp;
881 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
882 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
883 		panic("ffs_reload: dirty1");
884 	VOP_UNLOCK(devvp);
885 
886 	/*
887 	 * Step 2: re-read superblock from disk.
888 	 */
889 	fs = VFSTOUFS(mp)->um_fs;
890 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
891 	    NOCRED, &bp)) != 0)
892 		return (error);
893 	newfs = (struct fs *)bp->b_data;
894 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
895 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
896 	    newfs->fs_bsize > MAXBSIZE ||
897 	    newfs->fs_bsize < sizeof(struct fs)) {
898 			brelse(bp);
899 			return (EIO);		/* XXX needs translation */
900 	}
901 	/*
902 	 * Preserve the summary information, read-only status, and
903 	 * superblock location by copying these fields into our new
904 	 * superblock before using it to update the existing superblock.
905 	 */
906 	newfs->fs_si = fs->fs_si;
907 	newfs->fs_ronly = fs->fs_ronly;
908 	sblockloc = fs->fs_sblockloc;
909 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
910 	brelse(bp);
911 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
912 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
913 	UFS_LOCK(ump);
914 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
915 		printf("WARNING: %s: reload pending error: blocks %jd "
916 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
917 		    fs->fs_pendinginodes);
918 		fs->fs_pendingblocks = 0;
919 		fs->fs_pendinginodes = 0;
920 	}
921 	UFS_UNLOCK(ump);
922 
923 	/*
924 	 * Step 3: re-read summary information from disk.
925 	 */
926 	size = fs->fs_cssize;
927 	blks = howmany(size, fs->fs_fsize);
928 	if (fs->fs_contigsumsize > 0)
929 		size += fs->fs_ncg * sizeof(int32_t);
930 	size += fs->fs_ncg * sizeof(u_int8_t);
931 	free(fs->fs_csp, M_UFSMNT);
932 	space = malloc(size, M_UFSMNT, M_WAITOK);
933 	fs->fs_csp = space;
934 	for (i = 0; i < blks; i += fs->fs_frag) {
935 		size = fs->fs_bsize;
936 		if (i + fs->fs_frag > blks)
937 			size = (blks - i) * fs->fs_fsize;
938 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
939 		    NOCRED, &bp);
940 		if (error)
941 			return (error);
942 		bcopy(bp->b_data, space, (u_int)size);
943 		space = (char *)space + size;
944 		brelse(bp);
945 	}
946 	/*
947 	 * We no longer know anything about clusters per cylinder group.
948 	 */
949 	if (fs->fs_contigsumsize > 0) {
950 		fs->fs_maxcluster = lp = space;
951 		for (i = 0; i < fs->fs_ncg; i++)
952 			*lp++ = fs->fs_contigsumsize;
953 		space = lp;
954 	}
955 	size = fs->fs_ncg * sizeof(u_int8_t);
956 	fs->fs_contigdirs = (u_int8_t *)space;
957 	bzero(fs->fs_contigdirs, size);
958 	if ((flags & FFSR_UNSUSPEND) != 0) {
959 		MNT_ILOCK(mp);
960 		mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
961 		wakeup(&mp->mnt_flag);
962 		MNT_IUNLOCK(mp);
963 	}
964 
965 loop:
966 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
967 		/*
968 		 * Skip syncer vnode.
969 		 */
970 		if (vp->v_type == VNON) {
971 			VI_UNLOCK(vp);
972 			continue;
973 		}
974 		/*
975 		 * Step 4: invalidate all cached file data.
976 		 */
977 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
978 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
979 			goto loop;
980 		}
981 		if (vinvalbuf(vp, 0, 0, 0))
982 			panic("ffs_reload: dirty2");
983 		/*
984 		 * Step 5: re-read inode data for all active vnodes.
985 		 */
986 		ip = VTOI(vp);
987 		error =
988 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
989 		    (int)fs->fs_bsize, NOCRED, &bp);
990 		if (error) {
991 			vput(vp);
992 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
993 			return (error);
994 		}
995 		if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) {
996 			brelse(bp);
997 			vput(vp);
998 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
999 			return (error);
1000 		}
1001 		ip->i_effnlink = ip->i_nlink;
1002 		brelse(bp);
1003 		vput(vp);
1004 	}
1005 	return (0);
1006 }
1007 
1008 /*
1009  * Common code for mount and mountroot
1010  */
1011 static int
1012 ffs_mountfs(odevvp, mp, td)
1013 	struct vnode *odevvp;
1014 	struct mount *mp;
1015 	struct thread *td;
1016 {
1017 	struct ufsmount *ump;
1018 	struct fs *fs;
1019 	struct cdev *dev;
1020 	int error, i, len, ronly;
1021 	struct ucred *cred;
1022 	struct g_consumer *cp;
1023 	struct mount *nmp;
1024 	struct vnode *devvp;
1025 	struct fsfail_task *etp;
1026 	int candelete, canspeedup;
1027 	off_t loc;
1028 
1029 	fs = NULL;
1030 	ump = NULL;
1031 	cred = td ? td->td_ucred : NOCRED;
1032 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
1033 
1034 	devvp = mntfs_allocvp(mp, odevvp);
1035 	VOP_UNLOCK(odevvp);
1036 	KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
1037 	dev = devvp->v_rdev;
1038 	KASSERT(dev->si_snapdata == NULL, ("non-NULL snapshot data"));
1039 	if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
1040 	    (uintptr_t)mp) == 0) {
1041 		mntfs_freevp(devvp);
1042 		return (EBUSY);
1043 	}
1044 	g_topology_lock();
1045 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
1046 	g_topology_unlock();
1047 	if (error != 0) {
1048 		atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1049 		mntfs_freevp(devvp);
1050 		return (error);
1051 	}
1052 	dev_ref(dev);
1053 	devvp->v_bufobj.bo_ops = &ffs_ops;
1054 	BO_LOCK(&odevvp->v_bufobj);
1055 	odevvp->v_bufobj.bo_flag |= BO_NOBUFS;
1056 	BO_UNLOCK(&odevvp->v_bufobj);
1057 	if (dev->si_iosize_max != 0)
1058 		mp->mnt_iosize_max = dev->si_iosize_max;
1059 	if (mp->mnt_iosize_max > maxphys)
1060 		mp->mnt_iosize_max = maxphys;
1061 	if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
1062 		error = EINVAL;
1063 		vfs_mount_error(mp,
1064 		    "Invalid sectorsize %d for superblock size %d",
1065 		    cp->provider->sectorsize, SBLOCKSIZE);
1066 		goto out;
1067 	}
1068 	/* fetch the superblock and summary information */
1069 	loc = STDSB;
1070 	if ((mp->mnt_flag & MNT_ROOTFS) != 0)
1071 		loc = STDSB_NOHASHFAIL;
1072 	if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0)
1073 		goto out;
1074 	fs->fs_flags &= ~FS_UNCLEAN;
1075 	if (fs->fs_clean == 0) {
1076 		fs->fs_flags |= FS_UNCLEAN;
1077 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
1078 		    ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
1079 		     (fs->fs_flags & FS_DOSOFTDEP))) {
1080 			printf("WARNING: %s was not properly dismounted\n",
1081 			    fs->fs_fsmnt);
1082 		} else {
1083 			vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
1084 			    fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
1085 			    (fs->fs_flags & FS_SUJ) == 0 ? "" :
1086 			    " Forced mount will invalidate journal contents");
1087 			error = EPERM;
1088 			goto out;
1089 		}
1090 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
1091 		    (mp->mnt_flag & MNT_FORCE)) {
1092 			printf("WARNING: %s: lost blocks %jd files %d\n",
1093 			    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1094 			    fs->fs_pendinginodes);
1095 			fs->fs_pendingblocks = 0;
1096 			fs->fs_pendinginodes = 0;
1097 		}
1098 	}
1099 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1100 		printf("WARNING: %s: mount pending error: blocks %jd "
1101 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1102 		    fs->fs_pendinginodes);
1103 		fs->fs_pendingblocks = 0;
1104 		fs->fs_pendinginodes = 0;
1105 	}
1106 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
1107 #ifdef UFS_GJOURNAL
1108 		/*
1109 		 * Get journal provider name.
1110 		 */
1111 		len = 1024;
1112 		mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK);
1113 		if (g_io_getattr("GJOURNAL::provider", cp, &len,
1114 		    mp->mnt_gjprovider) == 0) {
1115 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len,
1116 			    M_UFSMNT, M_WAITOK);
1117 			MNT_ILOCK(mp);
1118 			mp->mnt_flag |= MNT_GJOURNAL;
1119 			MNT_IUNLOCK(mp);
1120 		} else {
1121 			printf("WARNING: %s: GJOURNAL flag on fs "
1122 			    "but no gjournal provider below\n",
1123 			    mp->mnt_stat.f_mntonname);
1124 			free(mp->mnt_gjprovider, M_UFSMNT);
1125 			mp->mnt_gjprovider = NULL;
1126 		}
1127 #else
1128 		printf("WARNING: %s: GJOURNAL flag on fs but no "
1129 		    "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
1130 #endif
1131 	} else {
1132 		mp->mnt_gjprovider = NULL;
1133 	}
1134 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1135 	ump->um_cp = cp;
1136 	ump->um_bo = &devvp->v_bufobj;
1137 	ump->um_fs = fs;
1138 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1139 		ump->um_fstype = UFS1;
1140 		ump->um_balloc = ffs_balloc_ufs1;
1141 	} else {
1142 		ump->um_fstype = UFS2;
1143 		ump->um_balloc = ffs_balloc_ufs2;
1144 	}
1145 	ump->um_blkatoff = ffs_blkatoff;
1146 	ump->um_truncate = ffs_truncate;
1147 	ump->um_update = ffs_update;
1148 	ump->um_valloc = ffs_valloc;
1149 	ump->um_vfree = ffs_vfree;
1150 	ump->um_ifree = ffs_ifree;
1151 	ump->um_rdonly = ffs_rdonly;
1152 	ump->um_snapgone = ffs_snapgone;
1153 	if ((mp->mnt_flag & MNT_UNTRUSTED) != 0)
1154 		ump->um_check_blkno = ffs_check_blkno;
1155 	else
1156 		ump->um_check_blkno = NULL;
1157 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
1158 	ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc);
1159 	fs->fs_ronly = ronly;
1160 	fs->fs_active = NULL;
1161 	mp->mnt_data = ump;
1162 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
1163 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
1164 	nmp = NULL;
1165 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
1166 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
1167 		if (nmp)
1168 			vfs_rel(nmp);
1169 		vfs_getnewfsid(mp);
1170 	}
1171 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
1172 	MNT_ILOCK(mp);
1173 	mp->mnt_flag |= MNT_LOCAL;
1174 	MNT_IUNLOCK(mp);
1175 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
1176 #ifdef MAC
1177 		MNT_ILOCK(mp);
1178 		mp->mnt_flag |= MNT_MULTILABEL;
1179 		MNT_IUNLOCK(mp);
1180 #else
1181 		printf("WARNING: %s: multilabel flag on fs but "
1182 		    "no MAC support\n", mp->mnt_stat.f_mntonname);
1183 #endif
1184 	}
1185 	if ((fs->fs_flags & FS_ACLS) != 0) {
1186 #ifdef UFS_ACL
1187 		MNT_ILOCK(mp);
1188 
1189 		if (mp->mnt_flag & MNT_NFS4ACLS)
1190 			printf("WARNING: %s: ACLs flag on fs conflicts with "
1191 			    "\"nfsv4acls\" mount option; option ignored\n",
1192 			    mp->mnt_stat.f_mntonname);
1193 		mp->mnt_flag &= ~MNT_NFS4ACLS;
1194 		mp->mnt_flag |= MNT_ACLS;
1195 
1196 		MNT_IUNLOCK(mp);
1197 #else
1198 		printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
1199 		    mp->mnt_stat.f_mntonname);
1200 #endif
1201 	}
1202 	if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
1203 #ifdef UFS_ACL
1204 		MNT_ILOCK(mp);
1205 
1206 		if (mp->mnt_flag & MNT_ACLS)
1207 			printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
1208 			    "with \"acls\" mount option; option ignored\n",
1209 			    mp->mnt_stat.f_mntonname);
1210 		mp->mnt_flag &= ~MNT_ACLS;
1211 		mp->mnt_flag |= MNT_NFS4ACLS;
1212 
1213 		MNT_IUNLOCK(mp);
1214 #else
1215 		printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
1216 		    "ACLs support\n", mp->mnt_stat.f_mntonname);
1217 #endif
1218 	}
1219 	if ((fs->fs_flags & FS_TRIM) != 0) {
1220 		len = sizeof(int);
1221 		if (g_io_getattr("GEOM::candelete", cp, &len,
1222 		    &candelete) == 0) {
1223 			if (candelete)
1224 				ump->um_flags |= UM_CANDELETE;
1225 			else
1226 				printf("WARNING: %s: TRIM flag on fs but disk "
1227 				    "does not support TRIM\n",
1228 				    mp->mnt_stat.f_mntonname);
1229 		} else {
1230 			printf("WARNING: %s: TRIM flag on fs but disk does "
1231 			    "not confirm that it supports TRIM\n",
1232 			    mp->mnt_stat.f_mntonname);
1233 		}
1234 		if (((ump->um_flags) & UM_CANDELETE) != 0) {
1235 			ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1236 			    taskqueue_thread_enqueue, &ump->um_trim_tq);
1237 			taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1238 			    "%s trim", mp->mnt_stat.f_mntonname);
1239 			ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM,
1240 			    &ump->um_trimlisthashsize);
1241 		}
1242 	}
1243 
1244 	len = sizeof(int);
1245 	if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) {
1246 		if (canspeedup)
1247 			ump->um_flags |= UM_CANSPEEDUP;
1248 	}
1249 
1250 	ump->um_mountp = mp;
1251 	ump->um_dev = dev;
1252 	ump->um_devvp = devvp;
1253 	ump->um_odevvp = odevvp;
1254 	ump->um_nindir = fs->fs_nindir;
1255 	ump->um_bptrtodb = fs->fs_fsbtodb;
1256 	ump->um_seqinc = fs->fs_frag;
1257 	for (i = 0; i < MAXQUOTAS; i++)
1258 		ump->um_quotas[i] = NULLVP;
1259 #ifdef UFS_EXTATTR
1260 	ufs_extattr_uepm_init(&ump->um_extattr);
1261 #endif
1262 	/*
1263 	 * Set FS local "last mounted on" information (NULL pad)
1264 	 */
1265 	bzero(fs->fs_fsmnt, MAXMNTLEN);
1266 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1267 	mp->mnt_stat.f_iosize = fs->fs_bsize;
1268 
1269 	if (mp->mnt_flag & MNT_ROOTFS) {
1270 		/*
1271 		 * Root mount; update timestamp in mount structure.
1272 		 * this will be used by the common root mount code
1273 		 * to update the system clock.
1274 		 */
1275 		mp->mnt_time = fs->fs_time;
1276 	}
1277 
1278 	if (ronly == 0) {
1279 		fs->fs_mtime = time_second;
1280 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
1281 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1282 			ffs_flushfiles(mp, FORCECLOSE, td);
1283 			goto out;
1284 		}
1285 		if (fs->fs_snapinum[0] != 0)
1286 			ffs_snapshot_mount(mp);
1287 		fs->fs_fmod = 1;
1288 		fs->fs_clean = 0;
1289 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
1290 	}
1291 	/*
1292 	 * Initialize filesystem state information in mount struct.
1293 	 */
1294 	MNT_ILOCK(mp);
1295 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1296 	    MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1297 	MNT_IUNLOCK(mp);
1298 #ifdef UFS_EXTATTR
1299 #ifdef UFS_EXTATTR_AUTOSTART
1300 	/*
1301 	 *
1302 	 * Auto-starting does the following:
1303 	 *	- check for /.attribute in the fs, and extattr_start if so
1304 	 *	- for each file in .attribute, enable that file with
1305 	 * 	  an attribute of the same name.
1306 	 * Not clear how to report errors -- probably eat them.
1307 	 * This would all happen while the filesystem was busy/not
1308 	 * available, so would effectively be "atomic".
1309 	 */
1310 	(void) ufs_extattr_autostart(mp, td);
1311 #endif /* !UFS_EXTATTR_AUTOSTART */
1312 #endif /* !UFS_EXTATTR */
1313 	etp = malloc(sizeof *ump->um_fsfail_task, M_UFSMNT, M_WAITOK | M_ZERO);
1314 	etp->fsid = mp->mnt_stat.f_fsid;
1315 	ump->um_fsfail_task = etp;
1316 	return (0);
1317 out:
1318 	if (fs != NULL) {
1319 		free(fs->fs_csp, M_UFSMNT);
1320 		free(fs->fs_si, M_UFSMNT);
1321 		free(fs, M_UFSMNT);
1322 	}
1323 	if (cp != NULL) {
1324 		g_topology_lock();
1325 		g_vfs_close(cp);
1326 		g_topology_unlock();
1327 	}
1328 	if (ump) {
1329 		mtx_destroy(UFS_MTX(ump));
1330 		if (mp->mnt_gjprovider != NULL) {
1331 			free(mp->mnt_gjprovider, M_UFSMNT);
1332 			mp->mnt_gjprovider = NULL;
1333 		}
1334 		free(ump, M_UFSMNT);
1335 		mp->mnt_data = NULL;
1336 	}
1337 	BO_LOCK(&odevvp->v_bufobj);
1338 	odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1339 	BO_UNLOCK(&odevvp->v_bufobj);
1340 	atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1341 	mntfs_freevp(devvp);
1342 	dev_rel(dev);
1343 	return (error);
1344 }
1345 
1346 /*
1347  * A read function for use by filesystem-layer routines.
1348  */
1349 static int
1350 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size)
1351 {
1352 	struct buf *bp;
1353 	int error;
1354 
1355 	KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp));
1356 	*bufp = malloc(size, M_UFSMNT, M_WAITOK);
1357 	if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED,
1358 	    &bp)) != 0)
1359 		return (error);
1360 	bcopy(bp->b_data, *bufp, size);
1361 	bp->b_flags |= B_INVAL | B_NOCACHE;
1362 	brelse(bp);
1363 	return (0);
1364 }
1365 
1366 static int bigcgs = 0;
1367 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1368 
1369 /*
1370  * Sanity checks for loading old filesystem superblocks.
1371  * See ffs_oldfscompat_write below for unwound actions.
1372  *
1373  * XXX - Parts get retired eventually.
1374  * Unfortunately new bits get added.
1375  */
1376 static void
1377 ffs_oldfscompat_read(fs, ump, sblockloc)
1378 	struct fs *fs;
1379 	struct ufsmount *ump;
1380 	ufs2_daddr_t sblockloc;
1381 {
1382 	off_t maxfilesize;
1383 
1384 	/*
1385 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
1386 	 */
1387 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1388 		fs->fs_flags = fs->fs_old_flags;
1389 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
1390 		fs->fs_sblockloc = sblockloc;
1391 	}
1392 	/*
1393 	 * If not yet done, update UFS1 superblock with new wider fields.
1394 	 */
1395 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1396 		fs->fs_maxbsize = fs->fs_bsize;
1397 		fs->fs_time = fs->fs_old_time;
1398 		fs->fs_size = fs->fs_old_size;
1399 		fs->fs_dsize = fs->fs_old_dsize;
1400 		fs->fs_csaddr = fs->fs_old_csaddr;
1401 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1402 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1403 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1404 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1405 	}
1406 	if (fs->fs_magic == FS_UFS1_MAGIC &&
1407 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
1408 		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1409 		fs->fs_qbmask = ~fs->fs_bmask;
1410 		fs->fs_qfmask = ~fs->fs_fmask;
1411 	}
1412 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1413 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1414 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1415 		if (fs->fs_maxfilesize > maxfilesize)
1416 			fs->fs_maxfilesize = maxfilesize;
1417 	}
1418 	/* Compatibility for old filesystems */
1419 	if (fs->fs_avgfilesize <= 0)
1420 		fs->fs_avgfilesize = AVFILESIZ;
1421 	if (fs->fs_avgfpdir <= 0)
1422 		fs->fs_avgfpdir = AFPDIR;
1423 	if (bigcgs) {
1424 		fs->fs_save_cgsize = fs->fs_cgsize;
1425 		fs->fs_cgsize = fs->fs_bsize;
1426 	}
1427 }
1428 
1429 /*
1430  * Unwinding superblock updates for old filesystems.
1431  * See ffs_oldfscompat_read above for details.
1432  *
1433  * XXX - Parts get retired eventually.
1434  * Unfortunately new bits get added.
1435  */
1436 void
1437 ffs_oldfscompat_write(fs, ump)
1438 	struct fs *fs;
1439 	struct ufsmount *ump;
1440 {
1441 
1442 	/*
1443 	 * Copy back UFS2 updated fields that UFS1 inspects.
1444 	 */
1445 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1446 		fs->fs_old_time = fs->fs_time;
1447 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1448 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1449 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1450 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1451 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1452 	}
1453 	if (bigcgs) {
1454 		fs->fs_cgsize = fs->fs_save_cgsize;
1455 		fs->fs_save_cgsize = 0;
1456 	}
1457 }
1458 
1459 /*
1460  * unmount system call
1461  */
1462 static int
1463 ffs_unmount(mp, mntflags)
1464 	struct mount *mp;
1465 	int mntflags;
1466 {
1467 	struct thread *td;
1468 	struct ufsmount *ump = VFSTOUFS(mp);
1469 	struct fs *fs;
1470 	int error, flags, susp;
1471 #ifdef UFS_EXTATTR
1472 	int e_restart;
1473 #endif
1474 
1475 	flags = 0;
1476 	td = curthread;
1477 	fs = ump->um_fs;
1478 	if (mntflags & MNT_FORCE)
1479 		flags |= FORCECLOSE;
1480 	susp = fs->fs_ronly == 0;
1481 #ifdef UFS_EXTATTR
1482 	if ((error = ufs_extattr_stop(mp, td))) {
1483 		if (error != EOPNOTSUPP)
1484 			printf("WARNING: unmount %s: ufs_extattr_stop "
1485 			    "returned errno %d\n", mp->mnt_stat.f_mntonname,
1486 			    error);
1487 		e_restart = 0;
1488 	} else {
1489 		ufs_extattr_uepm_destroy(&ump->um_extattr);
1490 		e_restart = 1;
1491 	}
1492 #endif
1493 	if (susp) {
1494 		error = vfs_write_suspend_umnt(mp);
1495 		if (error != 0)
1496 			goto fail1;
1497 	}
1498 	if (MOUNTEDSOFTDEP(mp))
1499 		error = softdep_flushfiles(mp, flags, td);
1500 	else
1501 		error = ffs_flushfiles(mp, flags, td);
1502 	if (error != 0 && !ffs_fsfail_cleanup(ump, error))
1503 		goto fail;
1504 
1505 	UFS_LOCK(ump);
1506 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1507 		printf("WARNING: unmount %s: pending error: blocks %jd "
1508 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1509 		    fs->fs_pendinginodes);
1510 		fs->fs_pendingblocks = 0;
1511 		fs->fs_pendinginodes = 0;
1512 	}
1513 	UFS_UNLOCK(ump);
1514 	if (MOUNTEDSOFTDEP(mp))
1515 		softdep_unmount(mp);
1516 	if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1517 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1518 		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1519 		if (ffs_fsfail_cleanup(ump, error))
1520 			error = 0;
1521 		if (error != 0 && !ffs_fsfail_cleanup(ump, error)) {
1522 			fs->fs_clean = 0;
1523 			goto fail;
1524 		}
1525 	}
1526 	if (susp)
1527 		vfs_write_resume(mp, VR_START_WRITE);
1528 	if (ump->um_trim_tq != NULL) {
1529 		while (ump->um_trim_inflight != 0)
1530 			pause("ufsutr", hz);
1531 		taskqueue_drain_all(ump->um_trim_tq);
1532 		taskqueue_free(ump->um_trim_tq);
1533 		free (ump->um_trimhash, M_TRIM);
1534 	}
1535 	g_topology_lock();
1536 	if (ump->um_fsckpid > 0) {
1537 		/*
1538 		 * Return to normal read-only mode.
1539 		 */
1540 		error = g_access(ump->um_cp, 0, -1, 0);
1541 		ump->um_fsckpid = 0;
1542 	}
1543 	g_vfs_close(ump->um_cp);
1544 	g_topology_unlock();
1545 	BO_LOCK(&ump->um_odevvp->v_bufobj);
1546 	ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1547 	BO_UNLOCK(&ump->um_odevvp->v_bufobj);
1548 	atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1549 	mntfs_freevp(ump->um_devvp);
1550 	vrele(ump->um_odevvp);
1551 	dev_rel(ump->um_dev);
1552 	mtx_destroy(UFS_MTX(ump));
1553 	if (mp->mnt_gjprovider != NULL) {
1554 		free(mp->mnt_gjprovider, M_UFSMNT);
1555 		mp->mnt_gjprovider = NULL;
1556 	}
1557 	free(fs->fs_csp, M_UFSMNT);
1558 	free(fs->fs_si, M_UFSMNT);
1559 	free(fs, M_UFSMNT);
1560 	if (ump->um_fsfail_task != NULL)
1561 		free(ump->um_fsfail_task, M_UFSMNT);
1562 	free(ump, M_UFSMNT);
1563 	mp->mnt_data = NULL;
1564 	MNT_ILOCK(mp);
1565 	mp->mnt_flag &= ~MNT_LOCAL;
1566 	MNT_IUNLOCK(mp);
1567 	if (td->td_su == mp) {
1568 		td->td_su = NULL;
1569 		vfs_rel(mp);
1570 	}
1571 	return (error);
1572 
1573 fail:
1574 	if (susp)
1575 		vfs_write_resume(mp, VR_START_WRITE);
1576 fail1:
1577 #ifdef UFS_EXTATTR
1578 	if (e_restart) {
1579 		ufs_extattr_uepm_init(&ump->um_extattr);
1580 #ifdef UFS_EXTATTR_AUTOSTART
1581 		(void) ufs_extattr_autostart(mp, td);
1582 #endif
1583 	}
1584 #endif
1585 
1586 	return (error);
1587 }
1588 
1589 /*
1590  * Flush out all the files in a filesystem.
1591  */
1592 int
1593 ffs_flushfiles(mp, flags, td)
1594 	struct mount *mp;
1595 	int flags;
1596 	struct thread *td;
1597 {
1598 	struct ufsmount *ump;
1599 	int qerror, error;
1600 
1601 	ump = VFSTOUFS(mp);
1602 	qerror = 0;
1603 #ifdef QUOTA
1604 	if (mp->mnt_flag & MNT_QUOTA) {
1605 		int i;
1606 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1607 		if (error)
1608 			return (error);
1609 		for (i = 0; i < MAXQUOTAS; i++) {
1610 			error = quotaoff(td, mp, i);
1611 			if (error != 0) {
1612 				if ((flags & EARLYFLUSH) == 0)
1613 					return (error);
1614 				else
1615 					qerror = error;
1616 			}
1617 		}
1618 
1619 		/*
1620 		 * Here we fall through to vflush again to ensure that
1621 		 * we have gotten rid of all the system vnodes, unless
1622 		 * quotas must not be closed.
1623 		 */
1624 	}
1625 #endif
1626 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1627 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1628 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1629 			return (error);
1630 		ffs_snapshot_unmount(mp);
1631 		flags |= FORCECLOSE;
1632 		/*
1633 		 * Here we fall through to vflush again to ensure
1634 		 * that we have gotten rid of all the system vnodes.
1635 		 */
1636 	}
1637 
1638 	/*
1639 	 * Do not close system files if quotas were not closed, to be
1640 	 * able to sync the remaining dquots.  The freeblks softupdate
1641 	 * workitems might hold a reference on a dquot, preventing
1642 	 * quotaoff() from completing.  Next round of
1643 	 * softdep_flushworklist() iteration should process the
1644 	 * blockers, allowing the next run of quotaoff() to finally
1645 	 * flush held dquots.
1646 	 *
1647 	 * Otherwise, flush all the files.
1648 	 */
1649 	if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1650 		return (error);
1651 
1652 	/*
1653 	 * Flush filesystem metadata.
1654 	 */
1655 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1656 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1657 	VOP_UNLOCK(ump->um_devvp);
1658 	return (error);
1659 }
1660 
1661 /*
1662  * Get filesystem statistics.
1663  */
1664 static int
1665 ffs_statfs(mp, sbp)
1666 	struct mount *mp;
1667 	struct statfs *sbp;
1668 {
1669 	struct ufsmount *ump;
1670 	struct fs *fs;
1671 
1672 	ump = VFSTOUFS(mp);
1673 	fs = ump->um_fs;
1674 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1675 		panic("ffs_statfs");
1676 	sbp->f_version = STATFS_VERSION;
1677 	sbp->f_bsize = fs->fs_fsize;
1678 	sbp->f_iosize = fs->fs_bsize;
1679 	sbp->f_blocks = fs->fs_dsize;
1680 	UFS_LOCK(ump);
1681 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1682 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1683 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1684 	    dbtofsb(fs, fs->fs_pendingblocks);
1685 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO;
1686 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1687 	UFS_UNLOCK(ump);
1688 	sbp->f_namemax = UFS_MAXNAMLEN;
1689 	return (0);
1690 }
1691 
1692 static bool
1693 sync_doupdate(struct inode *ip)
1694 {
1695 
1696 	return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1697 	    IN_UPDATE)) != 0);
1698 }
1699 
1700 static int
1701 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused)
1702 {
1703 	struct inode *ip;
1704 
1705 	/*
1706 	 * Flags are safe to access because ->v_data invalidation
1707 	 * is held off by listmtx.
1708 	 */
1709 	if (vp->v_type == VNON)
1710 		return (false);
1711 	ip = VTOI(vp);
1712 	if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0)
1713 		return (false);
1714 	return (true);
1715 }
1716 
1717 /*
1718  * For a lazy sync, we only care about access times, quotas and the
1719  * superblock.  Other filesystem changes are already converted to
1720  * cylinder group blocks or inode blocks updates and are written to
1721  * disk by syncer.
1722  */
1723 static int
1724 ffs_sync_lazy(mp)
1725      struct mount *mp;
1726 {
1727 	struct vnode *mvp, *vp;
1728 	struct inode *ip;
1729 	struct thread *td;
1730 	int allerror, error;
1731 
1732 	allerror = 0;
1733 	td = curthread;
1734 	if ((mp->mnt_flag & MNT_NOATIME) != 0) {
1735 #ifdef QUOTA
1736 		qsync(mp);
1737 #endif
1738 		goto sbupdate;
1739 	}
1740 	MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) {
1741 		if (vp->v_type == VNON) {
1742 			VI_UNLOCK(vp);
1743 			continue;
1744 		}
1745 		ip = VTOI(vp);
1746 
1747 		/*
1748 		 * The IN_ACCESS flag is converted to IN_MODIFIED by
1749 		 * ufs_close() and ufs_getattr() by the calls to
1750 		 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1751 		 * Test also all the other timestamp flags too, to pick up
1752 		 * any other cases that could be missed.
1753 		 */
1754 		if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1755 			VI_UNLOCK(vp);
1756 			continue;
1757 		}
1758 		if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0)
1759 			continue;
1760 #ifdef QUOTA
1761 		qsyncvp(vp);
1762 #endif
1763 		if (sync_doupdate(ip))
1764 			error = ffs_update(vp, 0);
1765 		if (error != 0)
1766 			allerror = error;
1767 		vput(vp);
1768 	}
1769 sbupdate:
1770 	if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1771 	    (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1772 		allerror = error;
1773 	return (allerror);
1774 }
1775 
1776 /*
1777  * Go through the disk queues to initiate sandbagged IO;
1778  * go through the inodes to write those that have been modified;
1779  * initiate the writing of the super block if it has been modified.
1780  *
1781  * Note: we are always called with the filesystem marked busy using
1782  * vfs_busy().
1783  */
1784 static int
1785 ffs_sync(mp, waitfor)
1786 	struct mount *mp;
1787 	int waitfor;
1788 {
1789 	struct vnode *mvp, *vp, *devvp;
1790 	struct thread *td;
1791 	struct inode *ip;
1792 	struct ufsmount *ump = VFSTOUFS(mp);
1793 	struct fs *fs;
1794 	int error, count, lockreq, allerror = 0;
1795 	int suspend;
1796 	int suspended;
1797 	int secondary_writes;
1798 	int secondary_accwrites;
1799 	int softdep_deps;
1800 	int softdep_accdeps;
1801 	struct bufobj *bo;
1802 
1803 	suspend = 0;
1804 	suspended = 0;
1805 	td = curthread;
1806 	fs = ump->um_fs;
1807 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1808 		panic("%s: ffs_sync: modification on read-only filesystem",
1809 		    fs->fs_fsmnt);
1810 	if (waitfor == MNT_LAZY) {
1811 		if (!rebooting)
1812 			return (ffs_sync_lazy(mp));
1813 		waitfor = MNT_NOWAIT;
1814 	}
1815 
1816 	/*
1817 	 * Write back each (modified) inode.
1818 	 */
1819 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1820 	if (waitfor == MNT_SUSPEND) {
1821 		suspend = 1;
1822 		waitfor = MNT_WAIT;
1823 	}
1824 	if (waitfor == MNT_WAIT)
1825 		lockreq = LK_EXCLUSIVE;
1826 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1827 loop:
1828 	/* Grab snapshot of secondary write counts */
1829 	MNT_ILOCK(mp);
1830 	secondary_writes = mp->mnt_secondary_writes;
1831 	secondary_accwrites = mp->mnt_secondary_accwrites;
1832 	MNT_IUNLOCK(mp);
1833 
1834 	/* Grab snapshot of softdep dependency counts */
1835 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1836 
1837 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1838 		/*
1839 		 * Depend on the vnode interlock to keep things stable enough
1840 		 * for a quick test.  Since there might be hundreds of
1841 		 * thousands of vnodes, we cannot afford even a subroutine
1842 		 * call unless there's a good chance that we have work to do.
1843 		 */
1844 		if (vp->v_type == VNON) {
1845 			VI_UNLOCK(vp);
1846 			continue;
1847 		}
1848 		ip = VTOI(vp);
1849 		if ((ip->i_flag &
1850 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1851 		    vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1852 			VI_UNLOCK(vp);
1853 			continue;
1854 		}
1855 		if ((error = vget(vp, lockreq)) != 0) {
1856 			if (error == ENOENT || error == ENOLCK) {
1857 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1858 				goto loop;
1859 			}
1860 			continue;
1861 		}
1862 #ifdef QUOTA
1863 		qsyncvp(vp);
1864 #endif
1865 		for (;;) {
1866 			error = ffs_syncvnode(vp, waitfor, 0);
1867 			if (error == ERELOOKUP)
1868 				continue;
1869 			if (error != 0)
1870 				allerror = error;
1871 			break;
1872 		}
1873 		vput(vp);
1874 	}
1875 	/*
1876 	 * Force stale filesystem control information to be flushed.
1877 	 */
1878 	if (waitfor == MNT_WAIT || rebooting) {
1879 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1880 			allerror = error;
1881 		if (ffs_fsfail_cleanup(ump, allerror))
1882 			allerror = 0;
1883 		/* Flushed work items may create new vnodes to clean */
1884 		if (allerror == 0 && count)
1885 			goto loop;
1886 	}
1887 
1888 	devvp = ump->um_devvp;
1889 	bo = &devvp->v_bufobj;
1890 	BO_LOCK(bo);
1891 	if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1892 		BO_UNLOCK(bo);
1893 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1894 		error = VOP_FSYNC(devvp, waitfor, td);
1895 		VOP_UNLOCK(devvp);
1896 		if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1897 			error = ffs_sbupdate(ump, waitfor, 0);
1898 		if (error != 0)
1899 			allerror = error;
1900 		if (ffs_fsfail_cleanup(ump, allerror))
1901 			allerror = 0;
1902 		if (allerror == 0 && waitfor == MNT_WAIT)
1903 			goto loop;
1904 	} else if (suspend != 0) {
1905 		if (softdep_check_suspend(mp,
1906 					  devvp,
1907 					  softdep_deps,
1908 					  softdep_accdeps,
1909 					  secondary_writes,
1910 					  secondary_accwrites) != 0) {
1911 			MNT_IUNLOCK(mp);
1912 			goto loop;	/* More work needed */
1913 		}
1914 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1915 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1916 		MNT_IUNLOCK(mp);
1917 		suspended = 1;
1918 	} else
1919 		BO_UNLOCK(bo);
1920 	/*
1921 	 * Write back modified superblock.
1922 	 */
1923 	if (fs->fs_fmod != 0 &&
1924 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1925 		allerror = error;
1926 	if (ffs_fsfail_cleanup(ump, allerror))
1927 		allerror = 0;
1928 	return (allerror);
1929 }
1930 
1931 int
1932 ffs_vget(mp, ino, flags, vpp)
1933 	struct mount *mp;
1934 	ino_t ino;
1935 	int flags;
1936 	struct vnode **vpp;
1937 {
1938 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1939 }
1940 
1941 int
1942 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1943 	struct mount *mp;
1944 	ino_t ino;
1945 	int flags;
1946 	struct vnode **vpp;
1947 	int ffs_flags;
1948 {
1949 	struct fs *fs;
1950 	struct inode *ip;
1951 	struct ufsmount *ump;
1952 	struct buf *bp;
1953 	struct vnode *vp;
1954 	daddr_t dbn;
1955 	int error;
1956 
1957 	MPASS((ffs_flags & FFSV_REPLACE) == 0 || (flags & LK_EXCLUSIVE) != 0);
1958 
1959 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1960 	if (error != 0)
1961 		return (error);
1962 	if (*vpp != NULL) {
1963 		if ((ffs_flags & FFSV_REPLACE) == 0)
1964 			return (0);
1965 		vgone(*vpp);
1966 		vput(*vpp);
1967 	}
1968 
1969 	/*
1970 	 * We must promote to an exclusive lock for vnode creation.  This
1971 	 * can happen if lookup is passed LOCKSHARED.
1972 	 */
1973 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1974 		flags &= ~LK_TYPE_MASK;
1975 		flags |= LK_EXCLUSIVE;
1976 	}
1977 
1978 	/*
1979 	 * We do not lock vnode creation as it is believed to be too
1980 	 * expensive for such rare case as simultaneous creation of vnode
1981 	 * for same ino by different processes. We just allow them to race
1982 	 * and check later to decide who wins. Let the race begin!
1983 	 */
1984 
1985 	ump = VFSTOUFS(mp);
1986 	fs = ump->um_fs;
1987 	ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO);
1988 
1989 	/* Allocate a new vnode/inode. */
1990 	error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1991 	    &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1992 	if (error) {
1993 		*vpp = NULL;
1994 		uma_zfree_smr(uma_inode, ip);
1995 		return (error);
1996 	}
1997 	/*
1998 	 * FFS supports recursive locking.
1999 	 */
2000 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
2001 	VN_LOCK_AREC(vp);
2002 	vp->v_data = ip;
2003 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
2004 	ip->i_vnode = vp;
2005 	ip->i_ump = ump;
2006 	ip->i_number = ino;
2007 	ip->i_ea_refs = 0;
2008 	ip->i_nextclustercg = -1;
2009 	ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2;
2010 	ip->i_mode = 0; /* ensure error cases below throw away vnode */
2011 #ifdef DIAGNOSTIC
2012 	ufs_init_trackers(ip);
2013 #endif
2014 #ifdef QUOTA
2015 	{
2016 		int i;
2017 		for (i = 0; i < MAXQUOTAS; i++)
2018 			ip->i_dquot[i] = NODQUOT;
2019 	}
2020 #endif
2021 
2022 	if (ffs_flags & FFSV_FORCEINSMQ)
2023 		vp->v_vflag |= VV_FORCEINSMQ;
2024 	error = insmntque(vp, mp);
2025 	if (error != 0) {
2026 		uma_zfree_smr(uma_inode, ip);
2027 		*vpp = NULL;
2028 		return (error);
2029 	}
2030 	vp->v_vflag &= ~VV_FORCEINSMQ;
2031 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
2032 	if (error != 0)
2033 		return (error);
2034 	if (*vpp != NULL) {
2035 		/*
2036 		 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set)
2037 		 * operate on empty inode, which must not be found by
2038 		 * other threads until fully filled.  Vnode for empty
2039 		 * inode must be not re-inserted on the hash by other
2040 		 * thread, after removal by us at the beginning.
2041 		 */
2042 		MPASS((ffs_flags & FFSV_REPLACE) == 0);
2043 		return (0);
2044 	}
2045 
2046 	/* Read in the disk contents for the inode, copy into the inode. */
2047 	dbn = fsbtodb(fs, ino_to_fsba(fs, ino));
2048 	error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize,
2049 	    NULL, NULL, 0, NOCRED, 0, NULL, &bp);
2050 	if (error != 0) {
2051 		/*
2052 		 * The inode does not contain anything useful, so it would
2053 		 * be misleading to leave it on its hash chain. With mode
2054 		 * still zero, it will be unlinked and returned to the free
2055 		 * list by vput().
2056 		 */
2057 		vgone(vp);
2058 		vput(vp);
2059 		*vpp = NULL;
2060 		return (error);
2061 	}
2062 	if (I_IS_UFS1(ip))
2063 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
2064 	else
2065 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
2066 	if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) {
2067 		bqrelse(bp);
2068 		vgone(vp);
2069 		vput(vp);
2070 		*vpp = NULL;
2071 		return (error);
2072 	}
2073 	if (DOINGSOFTDEP(vp))
2074 		softdep_load_inodeblock(ip);
2075 	else
2076 		ip->i_effnlink = ip->i_nlink;
2077 	bqrelse(bp);
2078 
2079 	/*
2080 	 * Initialize the vnode from the inode, check for aliases.
2081 	 * Note that the underlying vnode may have changed.
2082 	 */
2083 	error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2,
2084 	    &vp);
2085 	if (error) {
2086 		vgone(vp);
2087 		vput(vp);
2088 		*vpp = NULL;
2089 		return (error);
2090 	}
2091 
2092 	/*
2093 	 * Finish inode initialization.
2094 	 */
2095 	if (vp->v_type != VFIFO) {
2096 		/* FFS supports shared locking for all files except fifos. */
2097 		VN_LOCK_ASHARE(vp);
2098 	}
2099 
2100 	/*
2101 	 * Set up a generation number for this inode if it does not
2102 	 * already have one. This should only happen on old filesystems.
2103 	 */
2104 	if (ip->i_gen == 0) {
2105 		while (ip->i_gen == 0)
2106 			ip->i_gen = arc4random();
2107 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
2108 			UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
2109 			DIP_SET(ip, i_gen, ip->i_gen);
2110 		}
2111 	}
2112 #ifdef MAC
2113 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
2114 		/*
2115 		 * If this vnode is already allocated, and we're running
2116 		 * multi-label, attempt to perform a label association
2117 		 * from the extended attributes on the inode.
2118 		 */
2119 		error = mac_vnode_associate_extattr(mp, vp);
2120 		if (error) {
2121 			/* ufs_inactive will release ip->i_devvp ref. */
2122 			vgone(vp);
2123 			vput(vp);
2124 			*vpp = NULL;
2125 			return (error);
2126 		}
2127 	}
2128 #endif
2129 
2130 	*vpp = vp;
2131 	return (0);
2132 }
2133 
2134 /*
2135  * File handle to vnode
2136  *
2137  * Have to be really careful about stale file handles:
2138  * - check that the inode number is valid
2139  * - for UFS2 check that the inode number is initialized
2140  * - call ffs_vget() to get the locked inode
2141  * - check for an unallocated inode (i_mode == 0)
2142  * - check that the given client host has export rights and return
2143  *   those rights via. exflagsp and credanonp
2144  */
2145 static int
2146 ffs_fhtovp(mp, fhp, flags, vpp)
2147 	struct mount *mp;
2148 	struct fid *fhp;
2149 	int flags;
2150 	struct vnode **vpp;
2151 {
2152 	struct ufid *ufhp;
2153 	struct ufsmount *ump;
2154 	struct fs *fs;
2155 	struct cg *cgp;
2156 	struct buf *bp;
2157 	ino_t ino;
2158 	u_int cg;
2159 	int error;
2160 
2161 	ufhp = (struct ufid *)fhp;
2162 	ino = ufhp->ufid_ino;
2163 	ump = VFSTOUFS(mp);
2164 	fs = ump->um_fs;
2165 	if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
2166 		return (ESTALE);
2167 	/*
2168 	 * Need to check if inode is initialized because UFS2 does lazy
2169 	 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
2170 	 */
2171 	if (fs->fs_magic != FS_UFS2_MAGIC)
2172 		return (ufs_fhtovp(mp, ufhp, flags, vpp));
2173 	cg = ino_to_cg(fs, ino);
2174 	if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0)
2175 		return (error);
2176 	if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
2177 		brelse(bp);
2178 		return (ESTALE);
2179 	}
2180 	brelse(bp);
2181 	return (ufs_fhtovp(mp, ufhp, flags, vpp));
2182 }
2183 
2184 /*
2185  * Initialize the filesystem.
2186  */
2187 static int
2188 ffs_init(vfsp)
2189 	struct vfsconf *vfsp;
2190 {
2191 
2192 	ffs_susp_initialize();
2193 	softdep_initialize();
2194 	return (ufs_init(vfsp));
2195 }
2196 
2197 /*
2198  * Undo the work of ffs_init().
2199  */
2200 static int
2201 ffs_uninit(vfsp)
2202 	struct vfsconf *vfsp;
2203 {
2204 	int ret;
2205 
2206 	ret = ufs_uninit(vfsp);
2207 	softdep_uninitialize();
2208 	ffs_susp_uninitialize();
2209 	taskqueue_drain_all(taskqueue_thread);
2210 	return (ret);
2211 }
2212 
2213 /*
2214  * Structure used to pass information from ffs_sbupdate to its
2215  * helper routine ffs_use_bwrite.
2216  */
2217 struct devfd {
2218 	struct ufsmount	*ump;
2219 	struct buf	*sbbp;
2220 	int		 waitfor;
2221 	int		 suspended;
2222 	int		 error;
2223 };
2224 
2225 /*
2226  * Write a superblock and associated information back to disk.
2227  */
2228 int
2229 ffs_sbupdate(ump, waitfor, suspended)
2230 	struct ufsmount *ump;
2231 	int waitfor;
2232 	int suspended;
2233 {
2234 	struct fs *fs;
2235 	struct buf *sbbp;
2236 	struct devfd devfd;
2237 
2238 	fs = ump->um_fs;
2239 	if (fs->fs_ronly == 1 &&
2240 	    (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
2241 	    (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
2242 		panic("ffs_sbupdate: write read-only filesystem");
2243 	/*
2244 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
2245 	 */
2246 	sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
2247 	    (int)fs->fs_sbsize, 0, 0, 0);
2248 	/*
2249 	 * Initialize info needed for write function.
2250 	 */
2251 	devfd.ump = ump;
2252 	devfd.sbbp = sbbp;
2253 	devfd.waitfor = waitfor;
2254 	devfd.suspended = suspended;
2255 	devfd.error = 0;
2256 	return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite));
2257 }
2258 
2259 /*
2260  * Write function for use by filesystem-layer routines.
2261  */
2262 static int
2263 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size)
2264 {
2265 	struct devfd *devfdp;
2266 	struct ufsmount *ump;
2267 	struct buf *bp;
2268 	struct fs *fs;
2269 	int error;
2270 
2271 	devfdp = devfd;
2272 	ump = devfdp->ump;
2273 	fs = ump->um_fs;
2274 	/*
2275 	 * Writing the superblock summary information.
2276 	 */
2277 	if (loc != fs->fs_sblockloc) {
2278 		bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0);
2279 		bcopy(buf, bp->b_data, (u_int)size);
2280 		if (devfdp->suspended)
2281 			bp->b_flags |= B_VALIDSUSPWRT;
2282 		if (devfdp->waitfor != MNT_WAIT)
2283 			bawrite(bp);
2284 		else if ((error = bwrite(bp)) != 0)
2285 			devfdp->error = error;
2286 		return (0);
2287 	}
2288 	/*
2289 	 * Writing the superblock itself. We need to do special checks for it.
2290 	 */
2291 	bp = devfdp->sbbp;
2292 	if (ffs_fsfail_cleanup(ump, devfdp->error))
2293 		devfdp->error = 0;
2294 	if (devfdp->error != 0) {
2295 		brelse(bp);
2296 		return (devfdp->error);
2297 	}
2298 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
2299 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2300 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2301 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
2302 		fs->fs_sblockloc = SBLOCK_UFS1;
2303 	}
2304 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
2305 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2306 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2307 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
2308 		fs->fs_sblockloc = SBLOCK_UFS2;
2309 	}
2310 	if (MOUNTEDSOFTDEP(ump->um_mountp))
2311 		softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
2312 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
2313 	fs = (struct fs *)bp->b_data;
2314 	ffs_oldfscompat_write(fs, ump);
2315 	fs->fs_si = NULL;
2316 	/* Recalculate the superblock hash */
2317 	fs->fs_ckhash = ffs_calc_sbhash(fs);
2318 	if (devfdp->suspended)
2319 		bp->b_flags |= B_VALIDSUSPWRT;
2320 	if (devfdp->waitfor != MNT_WAIT)
2321 		bawrite(bp);
2322 	else if ((error = bwrite(bp)) != 0)
2323 		devfdp->error = error;
2324 	return (devfdp->error);
2325 }
2326 
2327 static int
2328 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
2329 	int attrnamespace, const char *attrname)
2330 {
2331 
2332 #ifdef UFS_EXTATTR
2333 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
2334 	    attrname));
2335 #else
2336 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
2337 	    attrname));
2338 #endif
2339 }
2340 
2341 static void
2342 ffs_ifree(struct ufsmount *ump, struct inode *ip)
2343 {
2344 
2345 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
2346 		uma_zfree(uma_ufs1, ip->i_din1);
2347 	else if (ip->i_din2 != NULL)
2348 		uma_zfree(uma_ufs2, ip->i_din2);
2349 	uma_zfree_smr(uma_inode, ip);
2350 }
2351 
2352 static int dobkgrdwrite = 1;
2353 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
2354     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2355 
2356 /*
2357  * Complete a background write started from bwrite.
2358  */
2359 static void
2360 ffs_backgroundwritedone(struct buf *bp)
2361 {
2362 	struct bufobj *bufobj;
2363 	struct buf *origbp;
2364 
2365 #ifdef SOFTUPDATES
2366 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0)
2367 		softdep_handle_error(bp);
2368 #endif
2369 
2370 	/*
2371 	 * Find the original buffer that we are writing.
2372 	 */
2373 	bufobj = bp->b_bufobj;
2374 	BO_LOCK(bufobj);
2375 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2376 		panic("backgroundwritedone: lost buffer");
2377 
2378 	/*
2379 	 * We should mark the cylinder group buffer origbp as
2380 	 * dirty, to not lose the failed write.
2381 	 */
2382 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2383 		origbp->b_vflags |= BV_BKGRDERR;
2384 	BO_UNLOCK(bufobj);
2385 	/*
2386 	 * Process dependencies then return any unfinished ones.
2387 	 */
2388 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2389 		buf_complete(bp);
2390 #ifdef SOFTUPDATES
2391 	if (!LIST_EMPTY(&bp->b_dep))
2392 		softdep_move_dependencies(bp, origbp);
2393 #endif
2394 	/*
2395 	 * This buffer is marked B_NOCACHE so when it is released
2396 	 * by biodone it will be tossed.
2397 	 */
2398 	bp->b_flags |= B_NOCACHE;
2399 	bp->b_flags &= ~B_CACHE;
2400 	pbrelvp(bp);
2401 
2402 	/*
2403 	 * Prevent brelse() from trying to keep and re-dirtying bp on
2404 	 * errors. It causes b_bufobj dereference in
2405 	 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2406 	 * pbrelvp() above.
2407 	 */
2408 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2409 		bp->b_flags |= B_INVAL;
2410 	bufdone(bp);
2411 	BO_LOCK(bufobj);
2412 	/*
2413 	 * Clear the BV_BKGRDINPROG flag in the original buffer
2414 	 * and awaken it if it is waiting for the write to complete.
2415 	 * If BV_BKGRDINPROG is not set in the original buffer it must
2416 	 * have been released and re-instantiated - which is not legal.
2417 	 */
2418 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2419 	    ("backgroundwritedone: lost buffer2"));
2420 	origbp->b_vflags &= ~BV_BKGRDINPROG;
2421 	if (origbp->b_vflags & BV_BKGRDWAIT) {
2422 		origbp->b_vflags &= ~BV_BKGRDWAIT;
2423 		wakeup(&origbp->b_xflags);
2424 	}
2425 	BO_UNLOCK(bufobj);
2426 }
2427 
2428 /*
2429  * Write, release buffer on completion.  (Done by iodone
2430  * if async).  Do not bother writing anything if the buffer
2431  * is invalid.
2432  *
2433  * Note that we set B_CACHE here, indicating that buffer is
2434  * fully valid and thus cacheable.  This is true even of NFS
2435  * now so we set it generally.  This could be set either here
2436  * or in biodone() since the I/O is synchronous.  We put it
2437  * here.
2438  */
2439 static int
2440 ffs_bufwrite(struct buf *bp)
2441 {
2442 	struct buf *newbp;
2443 	struct cg *cgp;
2444 
2445 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2446 	if (bp->b_flags & B_INVAL) {
2447 		brelse(bp);
2448 		return (0);
2449 	}
2450 
2451 	if (!BUF_ISLOCKED(bp))
2452 		panic("bufwrite: buffer is not busy???");
2453 	/*
2454 	 * If a background write is already in progress, delay
2455 	 * writing this block if it is asynchronous. Otherwise
2456 	 * wait for the background write to complete.
2457 	 */
2458 	BO_LOCK(bp->b_bufobj);
2459 	if (bp->b_vflags & BV_BKGRDINPROG) {
2460 		if (bp->b_flags & B_ASYNC) {
2461 			BO_UNLOCK(bp->b_bufobj);
2462 			bdwrite(bp);
2463 			return (0);
2464 		}
2465 		bp->b_vflags |= BV_BKGRDWAIT;
2466 		msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2467 		    "bwrbg", 0);
2468 		if (bp->b_vflags & BV_BKGRDINPROG)
2469 			panic("bufwrite: still writing");
2470 	}
2471 	bp->b_vflags &= ~BV_BKGRDERR;
2472 	BO_UNLOCK(bp->b_bufobj);
2473 
2474 	/*
2475 	 * If this buffer is marked for background writing and we
2476 	 * do not have to wait for it, make a copy and write the
2477 	 * copy so as to leave this buffer ready for further use.
2478 	 *
2479 	 * This optimization eats a lot of memory.  If we have a page
2480 	 * or buffer shortfall we can't do it.
2481 	 */
2482 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2483 	    (bp->b_flags & B_ASYNC) &&
2484 	    !vm_page_count_severe() &&
2485 	    !buf_dirty_count_severe()) {
2486 		KASSERT(bp->b_iodone == NULL,
2487 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2488 
2489 		/* get a new block */
2490 		newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2491 		if (newbp == NULL)
2492 			goto normal_write;
2493 
2494 		KASSERT(buf_mapped(bp), ("Unmapped cg"));
2495 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2496 		BO_LOCK(bp->b_bufobj);
2497 		bp->b_vflags |= BV_BKGRDINPROG;
2498 		BO_UNLOCK(bp->b_bufobj);
2499 		newbp->b_xflags |=
2500 		    (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER;
2501 		newbp->b_lblkno = bp->b_lblkno;
2502 		newbp->b_blkno = bp->b_blkno;
2503 		newbp->b_offset = bp->b_offset;
2504 		newbp->b_iodone = ffs_backgroundwritedone;
2505 		newbp->b_flags |= B_ASYNC;
2506 		newbp->b_flags &= ~B_INVAL;
2507 		pbgetvp(bp->b_vp, newbp);
2508 
2509 #ifdef SOFTUPDATES
2510 		/*
2511 		 * Move over the dependencies.  If there are rollbacks,
2512 		 * leave the parent buffer dirtied as it will need to
2513 		 * be written again.
2514 		 */
2515 		if (LIST_EMPTY(&bp->b_dep) ||
2516 		    softdep_move_dependencies(bp, newbp) == 0)
2517 			bundirty(bp);
2518 #else
2519 		bundirty(bp);
2520 #endif
2521 
2522 		/*
2523 		 * Initiate write on the copy, release the original.  The
2524 		 * BKGRDINPROG flag prevents it from going away until
2525 		 * the background write completes. We have to recalculate
2526 		 * its check hash in case the buffer gets freed and then
2527 		 * reconstituted from the buffer cache during a later read.
2528 		 */
2529 		if ((bp->b_xflags & BX_CYLGRP) != 0) {
2530 			cgp = (struct cg *)bp->b_data;
2531 			cgp->cg_ckhash = 0;
2532 			cgp->cg_ckhash =
2533 			    calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2534 		}
2535 		bqrelse(bp);
2536 		bp = newbp;
2537 	} else
2538 		/* Mark the buffer clean */
2539 		bundirty(bp);
2540 
2541 	/* Let the normal bufwrite do the rest for us */
2542 normal_write:
2543 	/*
2544 	 * If we are writing a cylinder group, update its time.
2545 	 */
2546 	if ((bp->b_xflags & BX_CYLGRP) != 0) {
2547 		cgp = (struct cg *)bp->b_data;
2548 		cgp->cg_old_time = cgp->cg_time = time_second;
2549 	}
2550 	return (bufwrite(bp));
2551 }
2552 
2553 static void
2554 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2555 {
2556 	struct vnode *vp;
2557 	struct buf *tbp;
2558 	int error, nocopy;
2559 
2560 	/*
2561 	 * This is the bufobj strategy for the private VCHR vnodes
2562 	 * used by FFS to access the underlying storage device.
2563 	 * We override the default bufobj strategy and thus bypass
2564 	 * VOP_STRATEGY() for these vnodes.
2565 	 */
2566 	vp = bo2vnode(bo);
2567 	KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR ||
2568 	    bp->b_vp->v_rdev == NULL ||
2569 	    bp->b_vp->v_rdev->si_mountpt == NULL ||
2570 	    VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL ||
2571 	    vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp,
2572 	    ("ffs_geom_strategy() with wrong vp"));
2573 	if (bp->b_iocmd == BIO_WRITE) {
2574 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2575 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2576 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2577 			panic("ffs_geom_strategy: bad I/O");
2578 		nocopy = bp->b_flags & B_NOCOPY;
2579 		bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2580 		if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2581 		    vp->v_rdev->si_snapdata != NULL) {
2582 			if ((bp->b_flags & B_CLUSTER) != 0) {
2583 				runningbufwakeup(bp);
2584 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2585 					      b_cluster.cluster_entry) {
2586 					error = ffs_copyonwrite(vp, tbp);
2587 					if (error != 0 &&
2588 					    error != EOPNOTSUPP) {
2589 						bp->b_error = error;
2590 						bp->b_ioflags |= BIO_ERROR;
2591 						bp->b_flags &= ~B_BARRIER;
2592 						bufdone(bp);
2593 						return;
2594 					}
2595 				}
2596 				bp->b_runningbufspace = bp->b_bufsize;
2597 				atomic_add_long(&runningbufspace,
2598 					       bp->b_runningbufspace);
2599 			} else {
2600 				error = ffs_copyonwrite(vp, bp);
2601 				if (error != 0 && error != EOPNOTSUPP) {
2602 					bp->b_error = error;
2603 					bp->b_ioflags |= BIO_ERROR;
2604 					bp->b_flags &= ~B_BARRIER;
2605 					bufdone(bp);
2606 					return;
2607 				}
2608 			}
2609 		}
2610 #ifdef SOFTUPDATES
2611 		if ((bp->b_flags & B_CLUSTER) != 0) {
2612 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2613 				      b_cluster.cluster_entry) {
2614 				if (!LIST_EMPTY(&tbp->b_dep))
2615 					buf_start(tbp);
2616 			}
2617 		} else {
2618 			if (!LIST_EMPTY(&bp->b_dep))
2619 				buf_start(bp);
2620 		}
2621 
2622 #endif
2623 		/*
2624 		 * Check for metadata that needs check-hashes and update them.
2625 		 */
2626 		switch (bp->b_xflags & BX_FSPRIV) {
2627 		case BX_CYLGRP:
2628 			((struct cg *)bp->b_data)->cg_ckhash = 0;
2629 			((struct cg *)bp->b_data)->cg_ckhash =
2630 			    calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2631 			break;
2632 
2633 		case BX_SUPERBLOCK:
2634 		case BX_INODE:
2635 		case BX_INDIR:
2636 		case BX_DIR:
2637 			printf("Check-hash write is unimplemented!!!\n");
2638 			break;
2639 
2640 		case 0:
2641 			break;
2642 
2643 		default:
2644 			printf("multiple buffer types 0x%b\n",
2645 			    (u_int)(bp->b_xflags & BX_FSPRIV),
2646 			    PRINT_UFS_BUF_XFLAGS);
2647 			break;
2648 		}
2649 	}
2650 	if (bp->b_iocmd != BIO_READ && ffs_enxio_enable)
2651 		bp->b_xflags |= BX_CVTENXIO;
2652 	g_vfs_strategy(bo, bp);
2653 }
2654 
2655 int
2656 ffs_own_mount(const struct mount *mp)
2657 {
2658 
2659 	if (mp->mnt_op == &ufs_vfsops)
2660 		return (1);
2661 	return (0);
2662 }
2663 
2664 #ifdef	DDB
2665 #ifdef SOFTUPDATES
2666 
2667 /* defined in ffs_softdep.c */
2668 extern void db_print_ffs(struct ufsmount *ump);
2669 
2670 DB_SHOW_COMMAND(ffs, db_show_ffs)
2671 {
2672 	struct mount *mp;
2673 	struct ufsmount *ump;
2674 
2675 	if (have_addr) {
2676 		ump = VFSTOUFS((struct mount *)addr);
2677 		db_print_ffs(ump);
2678 		return;
2679 	}
2680 
2681 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2682 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2683 			db_print_ffs(VFSTOUFS(mp));
2684 	}
2685 }
2686 
2687 #endif	/* SOFTUPDATES */
2688 #endif	/* DDB */
2689