xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision 10ff414c14eef433d8157f0c17904d740693933b)
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", "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  * On first ENXIO error, initiate an asynchronous forcible unmount.
285  * Used to unmount filesystems whose underlying media has gone away.
286  *
287  * Return true if a cleanup is in progress.
288  */
289 int
290 ffs_fsfail_cleanup(struct ufsmount *ump, int error)
291 {
292 	int retval;
293 
294 	UFS_LOCK(ump);
295 	retval = ffs_fsfail_cleanup_locked(ump, error);
296 	UFS_UNLOCK(ump);
297 	return (retval);
298 }
299 
300 int
301 ffs_fsfail_cleanup_locked(struct ufsmount *ump, int error)
302 {
303 	mtx_assert(UFS_MTX(ump), MA_OWNED);
304 	if (error == ENXIO && (ump->um_flags & UM_FSFAIL_CLEANUP) == 0) {
305 		ump->um_flags |= UM_FSFAIL_CLEANUP;
306 		/*
307 		 * Queue an async forced unmount.
308 		 */
309 		vfs_ref(ump->um_mountp);
310 		dounmount(ump->um_mountp,
311 		    MNT_FORCE | MNT_RECURSE | MNT_DEFERRED, curthread);
312 		printf("UFS: forcibly unmounting %s from %s\n",
313 		    ump->um_mountp->mnt_stat.f_mntfromname,
314 		    ump->um_mountp->mnt_stat.f_mntonname);
315 	}
316 	return ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0);
317 }
318 
319 /*
320  * Wrapper used during ENXIO cleanup to allocate empty buffers when
321  * the kernel is unable to read the real one. They are needed so that
322  * the soft updates code can use them to unwind its dependencies.
323  */
324 int
325 ffs_breadz(struct ufsmount *ump, struct vnode *vp, daddr_t lblkno,
326     daddr_t dblkno, int size, daddr_t *rablkno, int *rabsize, int cnt,
327     struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *),
328     struct buf **bpp)
329 {
330 	int error;
331 
332 	flags |= GB_CVTENXIO;
333 	error = breadn_flags(vp, lblkno, dblkno, size, rablkno, rabsize, cnt,
334 	    cred, flags, ckhashfunc, bpp);
335 	if (error != 0 && ffs_fsfail_cleanup(ump, error)) {
336 		error = getblkx(vp, lblkno, dblkno, size, 0, 0, flags, bpp);
337 		KASSERT(error == 0, ("getblkx failed"));
338 		vfs_bio_bzero_buf(*bpp, 0, size);
339 	}
340 	return (error);
341 }
342 
343 static int
344 ffs_mount(struct mount *mp)
345 {
346 	struct vnode *devvp, *odevvp;
347 	struct thread *td;
348 	struct ufsmount *ump = NULL;
349 	struct fs *fs;
350 	int error, error1, flags;
351 	uint64_t mntorflags, saved_mnt_flag;
352 	accmode_t accmode;
353 	struct nameidata ndp;
354 	char *fspec;
355 	bool mounted_softdep;
356 
357 	td = curthread;
358 	if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
359 		return (EINVAL);
360 	if (uma_inode == NULL) {
361 		uma_inode = uma_zcreate("FFS inode",
362 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
363 		    UMA_ALIGN_PTR, 0);
364 		uma_ufs1 = uma_zcreate("FFS1 dinode",
365 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
366 		    UMA_ALIGN_PTR, 0);
367 		uma_ufs2 = uma_zcreate("FFS2 dinode",
368 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
369 		    UMA_ALIGN_PTR, 0);
370 		VFS_SMR_ZONE_SET(uma_inode);
371 	}
372 
373 	vfs_deleteopt(mp->mnt_optnew, "groupquota");
374 	vfs_deleteopt(mp->mnt_optnew, "userquota");
375 
376 	fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
377 	if (error)
378 		return (error);
379 
380 	mntorflags = 0;
381 	if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0)
382 		mntorflags |= MNT_UNTRUSTED;
383 
384 	if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
385 		mntorflags |= MNT_ACLS;
386 
387 	if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
388 		mntorflags |= MNT_SNAPSHOT;
389 		/*
390 		 * Once we have set the MNT_SNAPSHOT flag, do not
391 		 * persist "snapshot" in the options list.
392 		 */
393 		vfs_deleteopt(mp->mnt_optnew, "snapshot");
394 		vfs_deleteopt(mp->mnt_opt, "snapshot");
395 	}
396 
397 	if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
398 		if (mntorflags & MNT_ACLS) {
399 			vfs_mount_error(mp,
400 			    "\"acls\" and \"nfsv4acls\" options "
401 			    "are mutually exclusive");
402 			return (EINVAL);
403 		}
404 		mntorflags |= MNT_NFS4ACLS;
405 	}
406 
407 	MNT_ILOCK(mp);
408 	mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
409 	mp->mnt_flag |= mntorflags;
410 	MNT_IUNLOCK(mp);
411 	/*
412 	 * If updating, check whether changing from read-only to
413 	 * read/write; if there is no device name, that's all we do.
414 	 */
415 	if (mp->mnt_flag & MNT_UPDATE) {
416 		ump = VFSTOUFS(mp);
417 		fs = ump->um_fs;
418 		odevvp = ump->um_odevvp;
419 		devvp = ump->um_devvp;
420 		if (fs->fs_ronly == 0 &&
421 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
422 			/*
423 			 * Flush any dirty data and suspend filesystem.
424 			 */
425 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
426 				return (error);
427 			error = vfs_write_suspend_umnt(mp);
428 			if (error != 0)
429 				return (error);
430 
431 			fs->fs_ronly = 1;
432 			if (MOUNTEDSOFTDEP(mp)) {
433 				MNT_ILOCK(mp);
434 				mp->mnt_flag &= ~MNT_SOFTDEP;
435 				MNT_IUNLOCK(mp);
436 				mounted_softdep = true;
437 			} else
438 				mounted_softdep = false;
439 
440 			/*
441 			 * Check for and optionally get rid of files open
442 			 * for writing.
443 			 */
444 			flags = WRITECLOSE;
445 			if (mp->mnt_flag & MNT_FORCE)
446 				flags |= FORCECLOSE;
447 			if (mounted_softdep) {
448 				error = softdep_flushfiles(mp, flags, td);
449 			} else {
450 				error = ffs_flushfiles(mp, flags, td);
451 			}
452 			if (error) {
453 				fs->fs_ronly = 0;
454 				if (mounted_softdep) {
455 					MNT_ILOCK(mp);
456 					mp->mnt_flag |= MNT_SOFTDEP;
457 					MNT_IUNLOCK(mp);
458 				}
459 				vfs_write_resume(mp, 0);
460 				return (error);
461 			}
462 
463 			if (fs->fs_pendingblocks != 0 ||
464 			    fs->fs_pendinginodes != 0) {
465 				printf("WARNING: %s Update error: blocks %jd "
466 				    "files %d\n", fs->fs_fsmnt,
467 				    (intmax_t)fs->fs_pendingblocks,
468 				    fs->fs_pendinginodes);
469 				fs->fs_pendingblocks = 0;
470 				fs->fs_pendinginodes = 0;
471 			}
472 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
473 				fs->fs_clean = 1;
474 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
475 				fs->fs_ronly = 0;
476 				fs->fs_clean = 0;
477 				if (mounted_softdep) {
478 					MNT_ILOCK(mp);
479 					mp->mnt_flag |= MNT_SOFTDEP;
480 					MNT_IUNLOCK(mp);
481 				}
482 				vfs_write_resume(mp, 0);
483 				return (error);
484 			}
485 			if (mounted_softdep)
486 				softdep_unmount(mp);
487 			g_topology_lock();
488 			/*
489 			 * Drop our write and exclusive access.
490 			 */
491 			g_access(ump->um_cp, 0, -1, -1);
492 			g_topology_unlock();
493 			MNT_ILOCK(mp);
494 			mp->mnt_flag |= MNT_RDONLY;
495 			MNT_IUNLOCK(mp);
496 			/*
497 			 * Allow the writers to note that filesystem
498 			 * is ro now.
499 			 */
500 			vfs_write_resume(mp, 0);
501 		}
502 		if ((mp->mnt_flag & MNT_RELOAD) &&
503 		    (error = ffs_reload(mp, 0)) != 0)
504 			return (error);
505 		if (fs->fs_ronly &&
506 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
507 			/*
508 			 * If upgrade to read-write by non-root, then verify
509 			 * that user has necessary permissions on the device.
510 			 */
511 			vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY);
512 			error = VOP_ACCESS(odevvp, VREAD | VWRITE,
513 			    td->td_ucred, td);
514 			if (error)
515 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
516 			VOP_UNLOCK(odevvp);
517 			if (error) {
518 				return (error);
519 			}
520 			fs->fs_flags &= ~FS_UNCLEAN;
521 			if (fs->fs_clean == 0) {
522 				fs->fs_flags |= FS_UNCLEAN;
523 				if ((mp->mnt_flag & MNT_FORCE) ||
524 				    ((fs->fs_flags &
525 				     (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
526 				     (fs->fs_flags & FS_DOSOFTDEP))) {
527 					printf("WARNING: %s was not properly "
528 					   "dismounted\n", fs->fs_fsmnt);
529 				} else {
530 					vfs_mount_error(mp,
531 					   "R/W mount of %s denied. %s.%s",
532 					   fs->fs_fsmnt,
533 					   "Filesystem is not clean - run fsck",
534 					   (fs->fs_flags & FS_SUJ) == 0 ? "" :
535 					   " Forced mount will invalidate"
536 					   " journal contents");
537 					return (EPERM);
538 				}
539 			}
540 			g_topology_lock();
541 			/*
542 			 * Request exclusive write access.
543 			 */
544 			error = g_access(ump->um_cp, 0, 1, 1);
545 			g_topology_unlock();
546 			if (error)
547 				return (error);
548 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
549 				return (error);
550 			error = vfs_write_suspend_umnt(mp);
551 			if (error != 0)
552 				return (error);
553 			fs->fs_ronly = 0;
554 			MNT_ILOCK(mp);
555 			saved_mnt_flag = MNT_RDONLY;
556 			if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag &
557 			    MNT_ASYNC) != 0)
558 				saved_mnt_flag |= MNT_ASYNC;
559 			mp->mnt_flag &= ~saved_mnt_flag;
560 			MNT_IUNLOCK(mp);
561 			fs->fs_mtime = time_second;
562 			/* check to see if we need to start softdep */
563 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
564 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
565 				fs->fs_ronly = 1;
566 				MNT_ILOCK(mp);
567 				mp->mnt_flag |= saved_mnt_flag;
568 				MNT_IUNLOCK(mp);
569 				vfs_write_resume(mp, 0);
570 				return (error);
571 			}
572 			fs->fs_clean = 0;
573 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
574 				fs->fs_ronly = 1;
575 				if ((fs->fs_flags & FS_DOSOFTDEP) != 0)
576 					softdep_unmount(mp);
577 				MNT_ILOCK(mp);
578 				mp->mnt_flag |= saved_mnt_flag;
579 				MNT_IUNLOCK(mp);
580 				vfs_write_resume(mp, 0);
581 				return (error);
582 			}
583 			if (fs->fs_snapinum[0] != 0)
584 				ffs_snapshot_mount(mp);
585 			vfs_write_resume(mp, 0);
586 		}
587 		/*
588 		 * Soft updates is incompatible with "async",
589 		 * so if we are doing softupdates stop the user
590 		 * from setting the async flag in an update.
591 		 * Softdep_mount() clears it in an initial mount
592 		 * or ro->rw remount.
593 		 */
594 		if (MOUNTEDSOFTDEP(mp)) {
595 			/* XXX: Reset too late ? */
596 			MNT_ILOCK(mp);
597 			mp->mnt_flag &= ~MNT_ASYNC;
598 			MNT_IUNLOCK(mp);
599 		}
600 		/*
601 		 * Keep MNT_ACLS flag if it is stored in superblock.
602 		 */
603 		if ((fs->fs_flags & FS_ACLS) != 0) {
604 			/* XXX: Set too late ? */
605 			MNT_ILOCK(mp);
606 			mp->mnt_flag |= MNT_ACLS;
607 			MNT_IUNLOCK(mp);
608 		}
609 
610 		if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
611 			/* XXX: Set too late ? */
612 			MNT_ILOCK(mp);
613 			mp->mnt_flag |= MNT_NFS4ACLS;
614 			MNT_IUNLOCK(mp);
615 		}
616 
617 		/*
618 		 * If this is a snapshot request, take the snapshot.
619 		 */
620 		if (mp->mnt_flag & MNT_SNAPSHOT)
621 			return (ffs_snapshot(mp, fspec));
622 
623 		/*
624 		 * Must not call namei() while owning busy ref.
625 		 */
626 		vfs_unbusy(mp);
627 	}
628 
629 	/*
630 	 * Not an update, or updating the name: look up the name
631 	 * and verify that it refers to a sensible disk device.
632 	 */
633 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
634 	error = namei(&ndp);
635 	if ((mp->mnt_flag & MNT_UPDATE) != 0) {
636 		/*
637 		 * Unmount does not start if MNT_UPDATE is set.  Mount
638 		 * update busies mp before setting MNT_UPDATE.  We
639 		 * must be able to retain our busy ref succesfully,
640 		 * without sleep.
641 		 */
642 		error1 = vfs_busy(mp, MBF_NOWAIT);
643 		MPASS(error1 == 0);
644 	}
645 	if (error != 0)
646 		return (error);
647 	NDFREE(&ndp, NDF_ONLY_PNBUF);
648 	devvp = ndp.ni_vp;
649 	if (!vn_isdisk_error(devvp, &error)) {
650 		vput(devvp);
651 		return (error);
652 	}
653 
654 	/*
655 	 * If mount by non-root, then verify that user has necessary
656 	 * permissions on the device.
657 	 */
658 	accmode = VREAD;
659 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
660 		accmode |= VWRITE;
661 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
662 	if (error)
663 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
664 	if (error) {
665 		vput(devvp);
666 		return (error);
667 	}
668 
669 	if (mp->mnt_flag & MNT_UPDATE) {
670 		/*
671 		 * Update only
672 		 *
673 		 * If it's not the same vnode, or at least the same device
674 		 * then it's not correct.
675 		 */
676 
677 		if (devvp->v_rdev != ump->um_devvp->v_rdev)
678 			error = EINVAL;	/* needs translation */
679 		vput(devvp);
680 		if (error)
681 			return (error);
682 	} else {
683 		/*
684 		 * New mount
685 		 *
686 		 * We need the name for the mount point (also used for
687 		 * "last mounted on") copied in. If an error occurs,
688 		 * the mount point is discarded by the upper level code.
689 		 * Note that vfs_mount_alloc() populates f_mntonname for us.
690 		 */
691 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
692 			vrele(devvp);
693 			return (error);
694 		}
695 	}
696 
697 	MNT_ILOCK(mp);
698 	/*
699 	 * This is racy versus lookup, see ufs_fplookup_vexec for details.
700 	 */
701 	if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) != 0)
702 		panic("MNTK_FPLOOKUP set on mount %p when it should not be", mp);
703 	if ((mp->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS | MNT_UNION)) == 0)
704 		mp->mnt_kern_flag |= MNTK_FPLOOKUP;
705 	MNT_IUNLOCK(mp);
706 
707 	vfs_mountedfrom(mp, fspec);
708 	return (0);
709 }
710 
711 /*
712  * Compatibility with old mount system call.
713  */
714 
715 static int
716 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
717 {
718 	struct ufs_args args;
719 	int error;
720 
721 	if (data == NULL)
722 		return (EINVAL);
723 	error = copyin(data, &args, sizeof args);
724 	if (error)
725 		return (error);
726 
727 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
728 	ma = mount_arg(ma, "export", &args.export, sizeof(args.export));
729 	error = kernel_mount(ma, flags);
730 
731 	return (error);
732 }
733 
734 /*
735  * Reload all incore data for a filesystem (used after running fsck on
736  * the root filesystem and finding things to fix). If the 'force' flag
737  * is 0, the filesystem must be mounted read-only.
738  *
739  * Things to do to update the mount:
740  *	1) invalidate all cached meta-data.
741  *	2) re-read superblock from disk.
742  *	3) re-read summary information from disk.
743  *	4) invalidate all inactive vnodes.
744  *	5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary
745  *	   writers, if requested.
746  *	6) invalidate all cached file data.
747  *	7) re-read inode data for all active vnodes.
748  */
749 int
750 ffs_reload(struct mount *mp, int flags)
751 {
752 	struct vnode *vp, *mvp, *devvp;
753 	struct inode *ip;
754 	void *space;
755 	struct buf *bp;
756 	struct fs *fs, *newfs;
757 	struct ufsmount *ump;
758 	ufs2_daddr_t sblockloc;
759 	int i, blks, error;
760 	u_long size;
761 	int32_t *lp;
762 
763 	ump = VFSTOUFS(mp);
764 
765 	MNT_ILOCK(mp);
766 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) {
767 		MNT_IUNLOCK(mp);
768 		return (EINVAL);
769 	}
770 	MNT_IUNLOCK(mp);
771 
772 	/*
773 	 * Step 1: invalidate all cached meta-data.
774 	 */
775 	devvp = VFSTOUFS(mp)->um_devvp;
776 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
777 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
778 		panic("ffs_reload: dirty1");
779 	VOP_UNLOCK(devvp);
780 
781 	/*
782 	 * Step 2: re-read superblock from disk.
783 	 */
784 	fs = VFSTOUFS(mp)->um_fs;
785 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
786 	    NOCRED, &bp)) != 0)
787 		return (error);
788 	newfs = (struct fs *)bp->b_data;
789 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
790 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
791 	    newfs->fs_bsize > MAXBSIZE ||
792 	    newfs->fs_bsize < sizeof(struct fs)) {
793 			brelse(bp);
794 			return (EIO);		/* XXX needs translation */
795 	}
796 	/*
797 	 * Preserve the summary information, read-only status, and
798 	 * superblock location by copying these fields into our new
799 	 * superblock before using it to update the existing superblock.
800 	 */
801 	newfs->fs_si = fs->fs_si;
802 	newfs->fs_ronly = fs->fs_ronly;
803 	sblockloc = fs->fs_sblockloc;
804 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
805 	brelse(bp);
806 	ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
807 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
808 	UFS_LOCK(ump);
809 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
810 		printf("WARNING: %s: reload pending error: blocks %jd "
811 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
812 		    fs->fs_pendinginodes);
813 		fs->fs_pendingblocks = 0;
814 		fs->fs_pendinginodes = 0;
815 	}
816 	UFS_UNLOCK(ump);
817 
818 	/*
819 	 * Step 3: re-read summary information from disk.
820 	 */
821 	size = fs->fs_cssize;
822 	blks = howmany(size, fs->fs_fsize);
823 	if (fs->fs_contigsumsize > 0)
824 		size += fs->fs_ncg * sizeof(int32_t);
825 	size += fs->fs_ncg * sizeof(u_int8_t);
826 	free(fs->fs_csp, M_UFSMNT);
827 	space = malloc(size, M_UFSMNT, M_WAITOK);
828 	fs->fs_csp = space;
829 	for (i = 0; i < blks; i += fs->fs_frag) {
830 		size = fs->fs_bsize;
831 		if (i + fs->fs_frag > blks)
832 			size = (blks - i) * fs->fs_fsize;
833 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
834 		    NOCRED, &bp);
835 		if (error)
836 			return (error);
837 		bcopy(bp->b_data, space, (u_int)size);
838 		space = (char *)space + size;
839 		brelse(bp);
840 	}
841 	/*
842 	 * We no longer know anything about clusters per cylinder group.
843 	 */
844 	if (fs->fs_contigsumsize > 0) {
845 		fs->fs_maxcluster = lp = space;
846 		for (i = 0; i < fs->fs_ncg; i++)
847 			*lp++ = fs->fs_contigsumsize;
848 		space = lp;
849 	}
850 	size = fs->fs_ncg * sizeof(u_int8_t);
851 	fs->fs_contigdirs = (u_int8_t *)space;
852 	bzero(fs->fs_contigdirs, size);
853 	if ((flags & FFSR_UNSUSPEND) != 0) {
854 		MNT_ILOCK(mp);
855 		mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
856 		wakeup(&mp->mnt_flag);
857 		MNT_IUNLOCK(mp);
858 	}
859 
860 loop:
861 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
862 		/*
863 		 * Skip syncer vnode.
864 		 */
865 		if (vp->v_type == VNON) {
866 			VI_UNLOCK(vp);
867 			continue;
868 		}
869 		/*
870 		 * Step 4: invalidate all cached file data.
871 		 */
872 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
873 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
874 			goto loop;
875 		}
876 		if (vinvalbuf(vp, 0, 0, 0))
877 			panic("ffs_reload: dirty2");
878 		/*
879 		 * Step 5: re-read inode data for all active vnodes.
880 		 */
881 		ip = VTOI(vp);
882 		error =
883 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
884 		    (int)fs->fs_bsize, NOCRED, &bp);
885 		if (error) {
886 			vput(vp);
887 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
888 			return (error);
889 		}
890 		if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) {
891 			brelse(bp);
892 			vput(vp);
893 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
894 			return (error);
895 		}
896 		ip->i_effnlink = ip->i_nlink;
897 		brelse(bp);
898 		vput(vp);
899 	}
900 	return (0);
901 }
902 
903 /*
904  * Common code for mount and mountroot
905  */
906 static int
907 ffs_mountfs(odevvp, mp, td)
908 	struct vnode *odevvp;
909 	struct mount *mp;
910 	struct thread *td;
911 {
912 	struct ufsmount *ump;
913 	struct fs *fs;
914 	struct cdev *dev;
915 	int error, i, len, ronly;
916 	struct ucred *cred;
917 	struct g_consumer *cp;
918 	struct mount *nmp;
919 	struct vnode *devvp;
920 	int candelete, canspeedup;
921 	off_t loc;
922 
923 	fs = NULL;
924 	ump = NULL;
925 	cred = td ? td->td_ucred : NOCRED;
926 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
927 
928 	devvp = mntfs_allocvp(mp, odevvp);
929 	VOP_UNLOCK(odevvp);
930 	KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
931 	dev = devvp->v_rdev;
932 	KASSERT(dev->si_snapdata == NULL, ("non-NULL snapshot data"));
933 	if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
934 	    (uintptr_t)mp) == 0) {
935 		mntfs_freevp(devvp);
936 		return (EBUSY);
937 	}
938 	g_topology_lock();
939 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
940 	g_topology_unlock();
941 	if (error != 0) {
942 		atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
943 		mntfs_freevp(devvp);
944 		return (error);
945 	}
946 	dev_ref(dev);
947 	devvp->v_bufobj.bo_ops = &ffs_ops;
948 	BO_LOCK(&odevvp->v_bufobj);
949 	odevvp->v_bufobj.bo_flag |= BO_NOBUFS;
950 	BO_UNLOCK(&odevvp->v_bufobj);
951 	if (dev->si_iosize_max != 0)
952 		mp->mnt_iosize_max = dev->si_iosize_max;
953 	if (mp->mnt_iosize_max > maxphys)
954 		mp->mnt_iosize_max = maxphys;
955 	if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
956 		error = EINVAL;
957 		vfs_mount_error(mp,
958 		    "Invalid sectorsize %d for superblock size %d",
959 		    cp->provider->sectorsize, SBLOCKSIZE);
960 		goto out;
961 	}
962 	/* fetch the superblock and summary information */
963 	loc = STDSB;
964 	if ((mp->mnt_flag & MNT_ROOTFS) != 0)
965 		loc = STDSB_NOHASHFAIL;
966 	if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0)
967 		goto out;
968 	fs->fs_flags &= ~FS_UNCLEAN;
969 	if (fs->fs_clean == 0) {
970 		fs->fs_flags |= FS_UNCLEAN;
971 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
972 		    ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
973 		     (fs->fs_flags & FS_DOSOFTDEP))) {
974 			printf("WARNING: %s was not properly dismounted\n",
975 			    fs->fs_fsmnt);
976 		} else {
977 			vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
978 			    fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
979 			    (fs->fs_flags & FS_SUJ) == 0 ? "" :
980 			    " Forced mount will invalidate journal contents");
981 			error = EPERM;
982 			goto out;
983 		}
984 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
985 		    (mp->mnt_flag & MNT_FORCE)) {
986 			printf("WARNING: %s: lost blocks %jd files %d\n",
987 			    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
988 			    fs->fs_pendinginodes);
989 			fs->fs_pendingblocks = 0;
990 			fs->fs_pendinginodes = 0;
991 		}
992 	}
993 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
994 		printf("WARNING: %s: mount pending error: blocks %jd "
995 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
996 		    fs->fs_pendinginodes);
997 		fs->fs_pendingblocks = 0;
998 		fs->fs_pendinginodes = 0;
999 	}
1000 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
1001 #ifdef UFS_GJOURNAL
1002 		/*
1003 		 * Get journal provider name.
1004 		 */
1005 		len = 1024;
1006 		mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK);
1007 		if (g_io_getattr("GJOURNAL::provider", cp, &len,
1008 		    mp->mnt_gjprovider) == 0) {
1009 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len,
1010 			    M_UFSMNT, M_WAITOK);
1011 			MNT_ILOCK(mp);
1012 			mp->mnt_flag |= MNT_GJOURNAL;
1013 			MNT_IUNLOCK(mp);
1014 		} else {
1015 			printf("WARNING: %s: GJOURNAL flag on fs "
1016 			    "but no gjournal provider below\n",
1017 			    mp->mnt_stat.f_mntonname);
1018 			free(mp->mnt_gjprovider, M_UFSMNT);
1019 			mp->mnt_gjprovider = NULL;
1020 		}
1021 #else
1022 		printf("WARNING: %s: GJOURNAL flag on fs but no "
1023 		    "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
1024 #endif
1025 	} else {
1026 		mp->mnt_gjprovider = NULL;
1027 	}
1028 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1029 	ump->um_cp = cp;
1030 	ump->um_bo = &devvp->v_bufobj;
1031 	ump->um_fs = fs;
1032 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1033 		ump->um_fstype = UFS1;
1034 		ump->um_balloc = ffs_balloc_ufs1;
1035 	} else {
1036 		ump->um_fstype = UFS2;
1037 		ump->um_balloc = ffs_balloc_ufs2;
1038 	}
1039 	ump->um_blkatoff = ffs_blkatoff;
1040 	ump->um_truncate = ffs_truncate;
1041 	ump->um_update = ffs_update;
1042 	ump->um_valloc = ffs_valloc;
1043 	ump->um_vfree = ffs_vfree;
1044 	ump->um_ifree = ffs_ifree;
1045 	ump->um_rdonly = ffs_rdonly;
1046 	ump->um_snapgone = ffs_snapgone;
1047 	if ((mp->mnt_flag & MNT_UNTRUSTED) != 0)
1048 		ump->um_check_blkno = ffs_check_blkno;
1049 	else
1050 		ump->um_check_blkno = NULL;
1051 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
1052 	sx_init(&ump->um_checkpath_lock, "uchpth");
1053 	ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc);
1054 	fs->fs_ronly = ronly;
1055 	fs->fs_active = NULL;
1056 	mp->mnt_data = ump;
1057 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
1058 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
1059 	nmp = NULL;
1060 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
1061 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
1062 		if (nmp)
1063 			vfs_rel(nmp);
1064 		vfs_getnewfsid(mp);
1065 	}
1066 	ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
1067 	MNT_ILOCK(mp);
1068 	mp->mnt_flag |= MNT_LOCAL;
1069 	MNT_IUNLOCK(mp);
1070 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
1071 #ifdef MAC
1072 		MNT_ILOCK(mp);
1073 		mp->mnt_flag |= MNT_MULTILABEL;
1074 		MNT_IUNLOCK(mp);
1075 #else
1076 		printf("WARNING: %s: multilabel flag on fs but "
1077 		    "no MAC support\n", mp->mnt_stat.f_mntonname);
1078 #endif
1079 	}
1080 	if ((fs->fs_flags & FS_ACLS) != 0) {
1081 #ifdef UFS_ACL
1082 		MNT_ILOCK(mp);
1083 
1084 		if (mp->mnt_flag & MNT_NFS4ACLS)
1085 			printf("WARNING: %s: ACLs flag on fs conflicts with "
1086 			    "\"nfsv4acls\" mount option; option ignored\n",
1087 			    mp->mnt_stat.f_mntonname);
1088 		mp->mnt_flag &= ~MNT_NFS4ACLS;
1089 		mp->mnt_flag |= MNT_ACLS;
1090 
1091 		MNT_IUNLOCK(mp);
1092 #else
1093 		printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
1094 		    mp->mnt_stat.f_mntonname);
1095 #endif
1096 	}
1097 	if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
1098 #ifdef UFS_ACL
1099 		MNT_ILOCK(mp);
1100 
1101 		if (mp->mnt_flag & MNT_ACLS)
1102 			printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
1103 			    "with \"acls\" mount option; option ignored\n",
1104 			    mp->mnt_stat.f_mntonname);
1105 		mp->mnt_flag &= ~MNT_ACLS;
1106 		mp->mnt_flag |= MNT_NFS4ACLS;
1107 
1108 		MNT_IUNLOCK(mp);
1109 #else
1110 		printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
1111 		    "ACLs support\n", mp->mnt_stat.f_mntonname);
1112 #endif
1113 	}
1114 	if ((fs->fs_flags & FS_TRIM) != 0) {
1115 		len = sizeof(int);
1116 		if (g_io_getattr("GEOM::candelete", cp, &len,
1117 		    &candelete) == 0) {
1118 			if (candelete)
1119 				ump->um_flags |= UM_CANDELETE;
1120 			else
1121 				printf("WARNING: %s: TRIM flag on fs but disk "
1122 				    "does not support TRIM\n",
1123 				    mp->mnt_stat.f_mntonname);
1124 		} else {
1125 			printf("WARNING: %s: TRIM flag on fs but disk does "
1126 			    "not confirm that it supports TRIM\n",
1127 			    mp->mnt_stat.f_mntonname);
1128 		}
1129 		if (((ump->um_flags) & UM_CANDELETE) != 0) {
1130 			ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1131 			    taskqueue_thread_enqueue, &ump->um_trim_tq);
1132 			taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1133 			    "%s trim", mp->mnt_stat.f_mntonname);
1134 			ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM,
1135 			    &ump->um_trimlisthashsize);
1136 		}
1137 	}
1138 
1139 	len = sizeof(int);
1140 	if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) {
1141 		if (canspeedup)
1142 			ump->um_flags |= UM_CANSPEEDUP;
1143 	}
1144 
1145 	ump->um_mountp = mp;
1146 	ump->um_dev = dev;
1147 	ump->um_devvp = devvp;
1148 	ump->um_odevvp = odevvp;
1149 	ump->um_nindir = fs->fs_nindir;
1150 	ump->um_bptrtodb = fs->fs_fsbtodb;
1151 	ump->um_seqinc = fs->fs_frag;
1152 	for (i = 0; i < MAXQUOTAS; i++)
1153 		ump->um_quotas[i] = NULLVP;
1154 #ifdef UFS_EXTATTR
1155 	ufs_extattr_uepm_init(&ump->um_extattr);
1156 #endif
1157 	/*
1158 	 * Set FS local "last mounted on" information (NULL pad)
1159 	 */
1160 	bzero(fs->fs_fsmnt, MAXMNTLEN);
1161 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1162 	mp->mnt_stat.f_iosize = fs->fs_bsize;
1163 
1164 	if (mp->mnt_flag & MNT_ROOTFS) {
1165 		/*
1166 		 * Root mount; update timestamp in mount structure.
1167 		 * this will be used by the common root mount code
1168 		 * to update the system clock.
1169 		 */
1170 		mp->mnt_time = fs->fs_time;
1171 	}
1172 
1173 	if (ronly == 0) {
1174 		fs->fs_mtime = time_second;
1175 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
1176 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1177 			ffs_flushfiles(mp, FORCECLOSE, td);
1178 			goto out;
1179 		}
1180 		if (fs->fs_snapinum[0] != 0)
1181 			ffs_snapshot_mount(mp);
1182 		fs->fs_fmod = 1;
1183 		fs->fs_clean = 0;
1184 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
1185 	}
1186 	/*
1187 	 * Initialize filesystem state information in mount struct.
1188 	 */
1189 	MNT_ILOCK(mp);
1190 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1191 	    MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1192 	MNT_IUNLOCK(mp);
1193 #ifdef UFS_EXTATTR
1194 #ifdef UFS_EXTATTR_AUTOSTART
1195 	/*
1196 	 *
1197 	 * Auto-starting does the following:
1198 	 *	- check for /.attribute in the fs, and extattr_start if so
1199 	 *	- for each file in .attribute, enable that file with
1200 	 * 	  an attribute of the same name.
1201 	 * Not clear how to report errors -- probably eat them.
1202 	 * This would all happen while the filesystem was busy/not
1203 	 * available, so would effectively be "atomic".
1204 	 */
1205 	(void) ufs_extattr_autostart(mp, td);
1206 #endif /* !UFS_EXTATTR_AUTOSTART */
1207 #endif /* !UFS_EXTATTR */
1208 	return (0);
1209 out:
1210 	if (fs != NULL) {
1211 		free(fs->fs_csp, M_UFSMNT);
1212 		free(fs->fs_si, M_UFSMNT);
1213 		free(fs, M_UFSMNT);
1214 	}
1215 	if (cp != NULL) {
1216 		g_topology_lock();
1217 		g_vfs_close(cp);
1218 		g_topology_unlock();
1219 	}
1220 	if (ump != NULL) {
1221 		mtx_destroy(UFS_MTX(ump));
1222 		sx_destroy(&ump->um_checkpath_lock);
1223 		if (mp->mnt_gjprovider != NULL) {
1224 			free(mp->mnt_gjprovider, M_UFSMNT);
1225 			mp->mnt_gjprovider = NULL;
1226 		}
1227 		MPASS(ump->um_softdep == NULL);
1228 		free(ump, M_UFSMNT);
1229 		mp->mnt_data = NULL;
1230 	}
1231 	BO_LOCK(&odevvp->v_bufobj);
1232 	odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1233 	BO_UNLOCK(&odevvp->v_bufobj);
1234 	atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1235 	mntfs_freevp(devvp);
1236 	dev_rel(dev);
1237 	return (error);
1238 }
1239 
1240 /*
1241  * A read function for use by filesystem-layer routines.
1242  */
1243 static int
1244 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size)
1245 {
1246 	struct buf *bp;
1247 	int error;
1248 
1249 	KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp));
1250 	*bufp = malloc(size, M_UFSMNT, M_WAITOK);
1251 	if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED,
1252 	    &bp)) != 0)
1253 		return (error);
1254 	bcopy(bp->b_data, *bufp, size);
1255 	bp->b_flags |= B_INVAL | B_NOCACHE;
1256 	brelse(bp);
1257 	return (0);
1258 }
1259 
1260 static int bigcgs = 0;
1261 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1262 
1263 /*
1264  * Sanity checks for loading old filesystem superblocks.
1265  * See ffs_oldfscompat_write below for unwound actions.
1266  *
1267  * XXX - Parts get retired eventually.
1268  * Unfortunately new bits get added.
1269  */
1270 static void
1271 ffs_oldfscompat_read(fs, ump, sblockloc)
1272 	struct fs *fs;
1273 	struct ufsmount *ump;
1274 	ufs2_daddr_t sblockloc;
1275 {
1276 	off_t maxfilesize;
1277 
1278 	/*
1279 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
1280 	 */
1281 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1282 		fs->fs_flags = fs->fs_old_flags;
1283 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
1284 		fs->fs_sblockloc = sblockloc;
1285 	}
1286 	/*
1287 	 * If not yet done, update UFS1 superblock with new wider fields.
1288 	 */
1289 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1290 		fs->fs_maxbsize = fs->fs_bsize;
1291 		fs->fs_time = fs->fs_old_time;
1292 		fs->fs_size = fs->fs_old_size;
1293 		fs->fs_dsize = fs->fs_old_dsize;
1294 		fs->fs_csaddr = fs->fs_old_csaddr;
1295 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1296 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1297 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1298 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1299 	}
1300 	if (fs->fs_magic == FS_UFS1_MAGIC &&
1301 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
1302 		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1303 		fs->fs_qbmask = ~fs->fs_bmask;
1304 		fs->fs_qfmask = ~fs->fs_fmask;
1305 	}
1306 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1307 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1308 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1309 		if (fs->fs_maxfilesize > maxfilesize)
1310 			fs->fs_maxfilesize = maxfilesize;
1311 	}
1312 	/* Compatibility for old filesystems */
1313 	if (fs->fs_avgfilesize <= 0)
1314 		fs->fs_avgfilesize = AVFILESIZ;
1315 	if (fs->fs_avgfpdir <= 0)
1316 		fs->fs_avgfpdir = AFPDIR;
1317 	if (bigcgs) {
1318 		fs->fs_save_cgsize = fs->fs_cgsize;
1319 		fs->fs_cgsize = fs->fs_bsize;
1320 	}
1321 }
1322 
1323 /*
1324  * Unwinding superblock updates for old filesystems.
1325  * See ffs_oldfscompat_read above for details.
1326  *
1327  * XXX - Parts get retired eventually.
1328  * Unfortunately new bits get added.
1329  */
1330 void
1331 ffs_oldfscompat_write(fs, ump)
1332 	struct fs *fs;
1333 	struct ufsmount *ump;
1334 {
1335 
1336 	/*
1337 	 * Copy back UFS2 updated fields that UFS1 inspects.
1338 	 */
1339 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1340 		fs->fs_old_time = fs->fs_time;
1341 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1342 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1343 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1344 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1345 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1346 	}
1347 	if (bigcgs) {
1348 		fs->fs_cgsize = fs->fs_save_cgsize;
1349 		fs->fs_save_cgsize = 0;
1350 	}
1351 }
1352 
1353 /*
1354  * unmount system call
1355  */
1356 static int
1357 ffs_unmount(mp, mntflags)
1358 	struct mount *mp;
1359 	int mntflags;
1360 {
1361 	struct thread *td;
1362 	struct ufsmount *ump = VFSTOUFS(mp);
1363 	struct fs *fs;
1364 	int error, flags, susp;
1365 #ifdef UFS_EXTATTR
1366 	int e_restart;
1367 #endif
1368 
1369 	flags = 0;
1370 	td = curthread;
1371 	fs = ump->um_fs;
1372 	if (mntflags & MNT_FORCE)
1373 		flags |= FORCECLOSE;
1374 	susp = fs->fs_ronly == 0;
1375 #ifdef UFS_EXTATTR
1376 	if ((error = ufs_extattr_stop(mp, td))) {
1377 		if (error != EOPNOTSUPP)
1378 			printf("WARNING: unmount %s: ufs_extattr_stop "
1379 			    "returned errno %d\n", mp->mnt_stat.f_mntonname,
1380 			    error);
1381 		e_restart = 0;
1382 	} else {
1383 		ufs_extattr_uepm_destroy(&ump->um_extattr);
1384 		e_restart = 1;
1385 	}
1386 #endif
1387 	if (susp) {
1388 		error = vfs_write_suspend_umnt(mp);
1389 		if (error != 0)
1390 			goto fail1;
1391 	}
1392 	if (MOUNTEDSOFTDEP(mp))
1393 		error = softdep_flushfiles(mp, flags, td);
1394 	else
1395 		error = ffs_flushfiles(mp, flags, td);
1396 	if (error != 0 && !ffs_fsfail_cleanup(ump, error))
1397 		goto fail;
1398 
1399 	UFS_LOCK(ump);
1400 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1401 		printf("WARNING: unmount %s: pending error: blocks %jd "
1402 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1403 		    fs->fs_pendinginodes);
1404 		fs->fs_pendingblocks = 0;
1405 		fs->fs_pendinginodes = 0;
1406 	}
1407 	UFS_UNLOCK(ump);
1408 	if (MOUNTEDSOFTDEP(mp))
1409 		softdep_unmount(mp);
1410 	MPASS(ump->um_softdep == NULL);
1411 	if (fs->fs_ronly == 0) {
1412 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1413 		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1414 		if (ffs_fsfail_cleanup(ump, error))
1415 			error = 0;
1416 		if (error != 0 && !ffs_fsfail_cleanup(ump, error)) {
1417 			fs->fs_clean = 0;
1418 			goto fail;
1419 		}
1420 	}
1421 	if (susp)
1422 		vfs_write_resume(mp, VR_START_WRITE);
1423 	if (ump->um_trim_tq != NULL) {
1424 		while (ump->um_trim_inflight != 0)
1425 			pause("ufsutr", hz);
1426 		taskqueue_drain_all(ump->um_trim_tq);
1427 		taskqueue_free(ump->um_trim_tq);
1428 		free (ump->um_trimhash, M_TRIM);
1429 	}
1430 	g_topology_lock();
1431 	g_vfs_close(ump->um_cp);
1432 	g_topology_unlock();
1433 	BO_LOCK(&ump->um_odevvp->v_bufobj);
1434 	ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1435 	BO_UNLOCK(&ump->um_odevvp->v_bufobj);
1436 	atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1437 	mntfs_freevp(ump->um_devvp);
1438 	vrele(ump->um_odevvp);
1439 	dev_rel(ump->um_dev);
1440 	mtx_destroy(UFS_MTX(ump));
1441 	sx_destroy(&ump->um_checkpath_lock);
1442 	if (mp->mnt_gjprovider != NULL) {
1443 		free(mp->mnt_gjprovider, M_UFSMNT);
1444 		mp->mnt_gjprovider = NULL;
1445 	}
1446 	free(fs->fs_csp, M_UFSMNT);
1447 	free(fs->fs_si, M_UFSMNT);
1448 	free(fs, M_UFSMNT);
1449 	free(ump, M_UFSMNT);
1450 	mp->mnt_data = NULL;
1451 	MNT_ILOCK(mp);
1452 	mp->mnt_flag &= ~MNT_LOCAL;
1453 	MNT_IUNLOCK(mp);
1454 	if (td->td_su == mp) {
1455 		td->td_su = NULL;
1456 		vfs_rel(mp);
1457 	}
1458 	return (error);
1459 
1460 fail:
1461 	if (susp)
1462 		vfs_write_resume(mp, VR_START_WRITE);
1463 fail1:
1464 #ifdef UFS_EXTATTR
1465 	if (e_restart) {
1466 		ufs_extattr_uepm_init(&ump->um_extattr);
1467 #ifdef UFS_EXTATTR_AUTOSTART
1468 		(void) ufs_extattr_autostart(mp, td);
1469 #endif
1470 	}
1471 #endif
1472 
1473 	return (error);
1474 }
1475 
1476 /*
1477  * Flush out all the files in a filesystem.
1478  */
1479 int
1480 ffs_flushfiles(mp, flags, td)
1481 	struct mount *mp;
1482 	int flags;
1483 	struct thread *td;
1484 {
1485 	struct ufsmount *ump;
1486 	int qerror, error;
1487 
1488 	ump = VFSTOUFS(mp);
1489 	qerror = 0;
1490 #ifdef QUOTA
1491 	if (mp->mnt_flag & MNT_QUOTA) {
1492 		int i;
1493 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1494 		if (error)
1495 			return (error);
1496 		for (i = 0; i < MAXQUOTAS; i++) {
1497 			error = quotaoff(td, mp, i);
1498 			if (error != 0) {
1499 				if ((flags & EARLYFLUSH) == 0)
1500 					return (error);
1501 				else
1502 					qerror = error;
1503 			}
1504 		}
1505 
1506 		/*
1507 		 * Here we fall through to vflush again to ensure that
1508 		 * we have gotten rid of all the system vnodes, unless
1509 		 * quotas must not be closed.
1510 		 */
1511 	}
1512 #endif
1513 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1514 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1515 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1516 			return (error);
1517 		ffs_snapshot_unmount(mp);
1518 		flags |= FORCECLOSE;
1519 		/*
1520 		 * Here we fall through to vflush again to ensure
1521 		 * that we have gotten rid of all the system vnodes.
1522 		 */
1523 	}
1524 
1525 	/*
1526 	 * Do not close system files if quotas were not closed, to be
1527 	 * able to sync the remaining dquots.  The freeblks softupdate
1528 	 * workitems might hold a reference on a dquot, preventing
1529 	 * quotaoff() from completing.  Next round of
1530 	 * softdep_flushworklist() iteration should process the
1531 	 * blockers, allowing the next run of quotaoff() to finally
1532 	 * flush held dquots.
1533 	 *
1534 	 * Otherwise, flush all the files.
1535 	 */
1536 	if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1537 		return (error);
1538 
1539 	/*
1540 	 * Flush filesystem metadata.
1541 	 */
1542 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1543 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1544 	VOP_UNLOCK(ump->um_devvp);
1545 	return (error);
1546 }
1547 
1548 /*
1549  * Get filesystem statistics.
1550  */
1551 static int
1552 ffs_statfs(mp, sbp)
1553 	struct mount *mp;
1554 	struct statfs *sbp;
1555 {
1556 	struct ufsmount *ump;
1557 	struct fs *fs;
1558 
1559 	ump = VFSTOUFS(mp);
1560 	fs = ump->um_fs;
1561 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1562 		panic("ffs_statfs");
1563 	sbp->f_version = STATFS_VERSION;
1564 	sbp->f_bsize = fs->fs_fsize;
1565 	sbp->f_iosize = fs->fs_bsize;
1566 	sbp->f_blocks = fs->fs_dsize;
1567 	UFS_LOCK(ump);
1568 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1569 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1570 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1571 	    dbtofsb(fs, fs->fs_pendingblocks);
1572 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO;
1573 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1574 	UFS_UNLOCK(ump);
1575 	sbp->f_namemax = UFS_MAXNAMLEN;
1576 	return (0);
1577 }
1578 
1579 static bool
1580 sync_doupdate(struct inode *ip)
1581 {
1582 
1583 	return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1584 	    IN_UPDATE)) != 0);
1585 }
1586 
1587 static int
1588 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused)
1589 {
1590 	struct inode *ip;
1591 
1592 	/*
1593 	 * Flags are safe to access because ->v_data invalidation
1594 	 * is held off by listmtx.
1595 	 */
1596 	if (vp->v_type == VNON)
1597 		return (false);
1598 	ip = VTOI(vp);
1599 	if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0)
1600 		return (false);
1601 	return (true);
1602 }
1603 
1604 /*
1605  * For a lazy sync, we only care about access times, quotas and the
1606  * superblock.  Other filesystem changes are already converted to
1607  * cylinder group blocks or inode blocks updates and are written to
1608  * disk by syncer.
1609  */
1610 static int
1611 ffs_sync_lazy(mp)
1612      struct mount *mp;
1613 {
1614 	struct vnode *mvp, *vp;
1615 	struct inode *ip;
1616 	struct thread *td;
1617 	int allerror, error;
1618 
1619 	allerror = 0;
1620 	td = curthread;
1621 	if ((mp->mnt_flag & MNT_NOATIME) != 0) {
1622 #ifdef QUOTA
1623 		qsync(mp);
1624 #endif
1625 		goto sbupdate;
1626 	}
1627 	MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) {
1628 		if (vp->v_type == VNON) {
1629 			VI_UNLOCK(vp);
1630 			continue;
1631 		}
1632 		ip = VTOI(vp);
1633 
1634 		/*
1635 		 * The IN_ACCESS flag is converted to IN_MODIFIED by
1636 		 * ufs_close() and ufs_getattr() by the calls to
1637 		 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1638 		 * Test also all the other timestamp flags too, to pick up
1639 		 * any other cases that could be missed.
1640 		 */
1641 		if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1642 			VI_UNLOCK(vp);
1643 			continue;
1644 		}
1645 		if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0)
1646 			continue;
1647 #ifdef QUOTA
1648 		qsyncvp(vp);
1649 #endif
1650 		if (sync_doupdate(ip))
1651 			error = ffs_update(vp, 0);
1652 		if (error != 0)
1653 			allerror = error;
1654 		vput(vp);
1655 	}
1656 sbupdate:
1657 	if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1658 	    (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1659 		allerror = error;
1660 	return (allerror);
1661 }
1662 
1663 /*
1664  * Go through the disk queues to initiate sandbagged IO;
1665  * go through the inodes to write those that have been modified;
1666  * initiate the writing of the super block if it has been modified.
1667  *
1668  * Note: we are always called with the filesystem marked busy using
1669  * vfs_busy().
1670  */
1671 static int
1672 ffs_sync(mp, waitfor)
1673 	struct mount *mp;
1674 	int waitfor;
1675 {
1676 	struct vnode *mvp, *vp, *devvp;
1677 	struct thread *td;
1678 	struct inode *ip;
1679 	struct ufsmount *ump = VFSTOUFS(mp);
1680 	struct fs *fs;
1681 	int error, count, lockreq, allerror = 0;
1682 	int suspend;
1683 	int suspended;
1684 	int secondary_writes;
1685 	int secondary_accwrites;
1686 	int softdep_deps;
1687 	int softdep_accdeps;
1688 	struct bufobj *bo;
1689 
1690 	suspend = 0;
1691 	suspended = 0;
1692 	td = curthread;
1693 	fs = ump->um_fs;
1694 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0)
1695 		panic("%s: ffs_sync: modification on read-only filesystem",
1696 		    fs->fs_fsmnt);
1697 	if (waitfor == MNT_LAZY) {
1698 		if (!rebooting)
1699 			return (ffs_sync_lazy(mp));
1700 		waitfor = MNT_NOWAIT;
1701 	}
1702 
1703 	/*
1704 	 * Write back each (modified) inode.
1705 	 */
1706 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1707 	if (waitfor == MNT_SUSPEND) {
1708 		suspend = 1;
1709 		waitfor = MNT_WAIT;
1710 	}
1711 	if (waitfor == MNT_WAIT)
1712 		lockreq = LK_EXCLUSIVE;
1713 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1714 loop:
1715 	/* Grab snapshot of secondary write counts */
1716 	MNT_ILOCK(mp);
1717 	secondary_writes = mp->mnt_secondary_writes;
1718 	secondary_accwrites = mp->mnt_secondary_accwrites;
1719 	MNT_IUNLOCK(mp);
1720 
1721 	/* Grab snapshot of softdep dependency counts */
1722 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1723 
1724 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1725 		/*
1726 		 * Depend on the vnode interlock to keep things stable enough
1727 		 * for a quick test.  Since there might be hundreds of
1728 		 * thousands of vnodes, we cannot afford even a subroutine
1729 		 * call unless there's a good chance that we have work to do.
1730 		 */
1731 		if (vp->v_type == VNON) {
1732 			VI_UNLOCK(vp);
1733 			continue;
1734 		}
1735 		ip = VTOI(vp);
1736 		if ((ip->i_flag &
1737 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1738 		    vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1739 			VI_UNLOCK(vp);
1740 			continue;
1741 		}
1742 		if ((error = vget(vp, lockreq)) != 0) {
1743 			if (error == ENOENT || error == ENOLCK) {
1744 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1745 				goto loop;
1746 			}
1747 			continue;
1748 		}
1749 #ifdef QUOTA
1750 		qsyncvp(vp);
1751 #endif
1752 		for (;;) {
1753 			error = ffs_syncvnode(vp, waitfor, 0);
1754 			if (error == ERELOOKUP)
1755 				continue;
1756 			if (error != 0)
1757 				allerror = error;
1758 			break;
1759 		}
1760 		vput(vp);
1761 	}
1762 	/*
1763 	 * Force stale filesystem control information to be flushed.
1764 	 */
1765 	if (waitfor == MNT_WAIT || rebooting) {
1766 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1767 			allerror = error;
1768 		if (ffs_fsfail_cleanup(ump, allerror))
1769 			allerror = 0;
1770 		/* Flushed work items may create new vnodes to clean */
1771 		if (allerror == 0 && count)
1772 			goto loop;
1773 	}
1774 
1775 	devvp = ump->um_devvp;
1776 	bo = &devvp->v_bufobj;
1777 	BO_LOCK(bo);
1778 	if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1779 		BO_UNLOCK(bo);
1780 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1781 		error = VOP_FSYNC(devvp, waitfor, td);
1782 		VOP_UNLOCK(devvp);
1783 		if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1784 			error = ffs_sbupdate(ump, waitfor, 0);
1785 		if (error != 0)
1786 			allerror = error;
1787 		if (ffs_fsfail_cleanup(ump, allerror))
1788 			allerror = 0;
1789 		if (allerror == 0 && waitfor == MNT_WAIT)
1790 			goto loop;
1791 	} else if (suspend != 0) {
1792 		if (softdep_check_suspend(mp,
1793 					  devvp,
1794 					  softdep_deps,
1795 					  softdep_accdeps,
1796 					  secondary_writes,
1797 					  secondary_accwrites) != 0) {
1798 			MNT_IUNLOCK(mp);
1799 			goto loop;	/* More work needed */
1800 		}
1801 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1802 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1803 		MNT_IUNLOCK(mp);
1804 		suspended = 1;
1805 	} else
1806 		BO_UNLOCK(bo);
1807 	/*
1808 	 * Write back modified superblock.
1809 	 */
1810 	if (fs->fs_fmod != 0 &&
1811 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1812 		allerror = error;
1813 	if (ffs_fsfail_cleanup(ump, allerror))
1814 		allerror = 0;
1815 	return (allerror);
1816 }
1817 
1818 int
1819 ffs_vget(mp, ino, flags, vpp)
1820 	struct mount *mp;
1821 	ino_t ino;
1822 	int flags;
1823 	struct vnode **vpp;
1824 {
1825 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1826 }
1827 
1828 int
1829 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1830 	struct mount *mp;
1831 	ino_t ino;
1832 	int flags;
1833 	struct vnode **vpp;
1834 	int ffs_flags;
1835 {
1836 	struct fs *fs;
1837 	struct inode *ip;
1838 	struct ufsmount *ump;
1839 	struct buf *bp;
1840 	struct vnode *vp;
1841 	daddr_t dbn;
1842 	int error;
1843 
1844 	MPASS((ffs_flags & (FFSV_REPLACE | FFSV_REPLACE_DOOMED)) == 0 ||
1845 	    (flags & LK_EXCLUSIVE) != 0);
1846 
1847 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1848 	if (error != 0)
1849 		return (error);
1850 	if (*vpp != NULL) {
1851 		if ((ffs_flags & FFSV_REPLACE) == 0 ||
1852 		    ((ffs_flags & FFSV_REPLACE_DOOMED) == 0 ||
1853 		    !VN_IS_DOOMED(*vpp)))
1854 			return (0);
1855 		vgone(*vpp);
1856 		vput(*vpp);
1857 	}
1858 
1859 	/*
1860 	 * We must promote to an exclusive lock for vnode creation.  This
1861 	 * can happen if lookup is passed LOCKSHARED.
1862 	 */
1863 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1864 		flags &= ~LK_TYPE_MASK;
1865 		flags |= LK_EXCLUSIVE;
1866 	}
1867 
1868 	/*
1869 	 * We do not lock vnode creation as it is believed to be too
1870 	 * expensive for such rare case as simultaneous creation of vnode
1871 	 * for same ino by different processes. We just allow them to race
1872 	 * and check later to decide who wins. Let the race begin!
1873 	 */
1874 
1875 	ump = VFSTOUFS(mp);
1876 	fs = ump->um_fs;
1877 	ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO);
1878 
1879 	/* Allocate a new vnode/inode. */
1880 	error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1881 	    &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1882 	if (error) {
1883 		*vpp = NULL;
1884 		uma_zfree_smr(uma_inode, ip);
1885 		return (error);
1886 	}
1887 	/*
1888 	 * FFS supports recursive locking.
1889 	 */
1890 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1891 	VN_LOCK_AREC(vp);
1892 	vp->v_data = ip;
1893 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1894 	ip->i_vnode = vp;
1895 	ip->i_ump = ump;
1896 	ip->i_number = ino;
1897 	ip->i_ea_refs = 0;
1898 	ip->i_nextclustercg = -1;
1899 	ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2;
1900 	ip->i_mode = 0; /* ensure error cases below throw away vnode */
1901 	cluster_init_vn(&ip->i_clusterw);
1902 #ifdef DIAGNOSTIC
1903 	ufs_init_trackers(ip);
1904 #endif
1905 #ifdef QUOTA
1906 	{
1907 		int i;
1908 		for (i = 0; i < MAXQUOTAS; i++)
1909 			ip->i_dquot[i] = NODQUOT;
1910 	}
1911 #endif
1912 
1913 	if (ffs_flags & FFSV_FORCEINSMQ)
1914 		vp->v_vflag |= VV_FORCEINSMQ;
1915 	error = insmntque(vp, mp);
1916 	if (error != 0) {
1917 		uma_zfree_smr(uma_inode, ip);
1918 		*vpp = NULL;
1919 		return (error);
1920 	}
1921 	vp->v_vflag &= ~VV_FORCEINSMQ;
1922 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1923 	if (error != 0)
1924 		return (error);
1925 	if (*vpp != NULL) {
1926 		/*
1927 		 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set)
1928 		 * operate on empty inode, which must not be found by
1929 		 * other threads until fully filled.  Vnode for empty
1930 		 * inode must be not re-inserted on the hash by other
1931 		 * thread, after removal by us at the beginning.
1932 		 */
1933 		MPASS((ffs_flags & FFSV_REPLACE) == 0);
1934 		return (0);
1935 	}
1936 
1937 	/* Read in the disk contents for the inode, copy into the inode. */
1938 	dbn = fsbtodb(fs, ino_to_fsba(fs, ino));
1939 	error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize,
1940 	    NULL, NULL, 0, NOCRED, 0, NULL, &bp);
1941 	if (error != 0) {
1942 		/*
1943 		 * The inode does not contain anything useful, so it would
1944 		 * be misleading to leave it on its hash chain. With mode
1945 		 * still zero, it will be unlinked and returned to the free
1946 		 * list by vput().
1947 		 */
1948 		vgone(vp);
1949 		vput(vp);
1950 		*vpp = NULL;
1951 		return (error);
1952 	}
1953 	if (I_IS_UFS1(ip))
1954 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1955 	else
1956 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1957 	if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) {
1958 		bqrelse(bp);
1959 		vgone(vp);
1960 		vput(vp);
1961 		*vpp = NULL;
1962 		return (error);
1963 	}
1964 	if (DOINGSOFTDEP(vp) && (!fs->fs_ronly ||
1965 	    (ffs_flags & FFSV_FORCEINODEDEP) != 0))
1966 		softdep_load_inodeblock(ip);
1967 	else
1968 		ip->i_effnlink = ip->i_nlink;
1969 	bqrelse(bp);
1970 
1971 	/*
1972 	 * Initialize the vnode from the inode, check for aliases.
1973 	 * Note that the underlying vnode may have changed.
1974 	 */
1975 	error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2,
1976 	    &vp);
1977 	if (error) {
1978 		vgone(vp);
1979 		vput(vp);
1980 		*vpp = NULL;
1981 		return (error);
1982 	}
1983 
1984 	/*
1985 	 * Finish inode initialization.
1986 	 */
1987 	if (vp->v_type != VFIFO) {
1988 		/* FFS supports shared locking for all files except fifos. */
1989 		VN_LOCK_ASHARE(vp);
1990 	}
1991 
1992 	/*
1993 	 * Set up a generation number for this inode if it does not
1994 	 * already have one. This should only happen on old filesystems.
1995 	 */
1996 	if (ip->i_gen == 0) {
1997 		while (ip->i_gen == 0)
1998 			ip->i_gen = arc4random();
1999 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
2000 			UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
2001 			DIP_SET(ip, i_gen, ip->i_gen);
2002 		}
2003 	}
2004 #ifdef MAC
2005 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
2006 		/*
2007 		 * If this vnode is already allocated, and we're running
2008 		 * multi-label, attempt to perform a label association
2009 		 * from the extended attributes on the inode.
2010 		 */
2011 		error = mac_vnode_associate_extattr(mp, vp);
2012 		if (error) {
2013 			/* ufs_inactive will release ip->i_devvp ref. */
2014 			vgone(vp);
2015 			vput(vp);
2016 			*vpp = NULL;
2017 			return (error);
2018 		}
2019 	}
2020 #endif
2021 
2022 	*vpp = vp;
2023 	return (0);
2024 }
2025 
2026 /*
2027  * File handle to vnode
2028  *
2029  * Have to be really careful about stale file handles:
2030  * - check that the inode number is valid
2031  * - for UFS2 check that the inode number is initialized
2032  * - call ffs_vget() to get the locked inode
2033  * - check for an unallocated inode (i_mode == 0)
2034  * - check that the given client host has export rights and return
2035  *   those rights via. exflagsp and credanonp
2036  */
2037 static int
2038 ffs_fhtovp(mp, fhp, flags, vpp)
2039 	struct mount *mp;
2040 	struct fid *fhp;
2041 	int flags;
2042 	struct vnode **vpp;
2043 {
2044 	struct ufid *ufhp;
2045 
2046 	ufhp = (struct ufid *)fhp;
2047 	return (ffs_inotovp(mp, ufhp->ufid_ino, ufhp->ufid_gen, flags,
2048 	    vpp, 0));
2049 }
2050 
2051 int
2052 ffs_inotovp(mp, ino, gen, lflags, vpp, ffs_flags)
2053 	struct mount *mp;
2054 	ino_t ino;
2055 	u_int64_t gen;
2056 	int lflags;
2057 	struct vnode **vpp;
2058 	int ffs_flags;
2059 {
2060 	struct ufsmount *ump;
2061 	struct vnode *nvp;
2062 	struct inode *ip;
2063 	struct fs *fs;
2064 	struct cg *cgp;
2065 	struct buf *bp;
2066 	u_int cg;
2067 	int error;
2068 
2069 	ump = VFSTOUFS(mp);
2070 	fs = ump->um_fs;
2071 	*vpp = NULL;
2072 
2073 	if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
2074 		return (ESTALE);
2075 
2076 	/*
2077 	 * Need to check if inode is initialized because UFS2 does lazy
2078 	 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
2079 	 */
2080 	if (fs->fs_magic == FS_UFS2_MAGIC) {
2081 		cg = ino_to_cg(fs, ino);
2082 		error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp);
2083 		if (error != 0)
2084 			return (error);
2085 		if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
2086 			brelse(bp);
2087 			return (ESTALE);
2088 		}
2089 		brelse(bp);
2090 	}
2091 
2092 	error = ffs_vgetf(mp, ino, lflags, &nvp, ffs_flags);
2093 	if (error != 0)
2094 		return (error);
2095 
2096 	ip = VTOI(nvp);
2097 	if (ip->i_mode == 0 || ip->i_gen != gen || ip->i_effnlink <= 0) {
2098 		if (ip->i_mode == 0)
2099 			vgone(nvp);
2100 		vput(nvp);
2101 		return (ESTALE);
2102 	}
2103 
2104 	vnode_create_vobject(nvp, DIP(ip, i_size), curthread);
2105 	*vpp = nvp;
2106 	return (0);
2107 }
2108 
2109 /*
2110  * Initialize the filesystem.
2111  */
2112 static int
2113 ffs_init(vfsp)
2114 	struct vfsconf *vfsp;
2115 {
2116 
2117 	ffs_susp_initialize();
2118 	softdep_initialize();
2119 	return (ufs_init(vfsp));
2120 }
2121 
2122 /*
2123  * Undo the work of ffs_init().
2124  */
2125 static int
2126 ffs_uninit(vfsp)
2127 	struct vfsconf *vfsp;
2128 {
2129 	int ret;
2130 
2131 	ret = ufs_uninit(vfsp);
2132 	softdep_uninitialize();
2133 	ffs_susp_uninitialize();
2134 	taskqueue_drain_all(taskqueue_thread);
2135 	return (ret);
2136 }
2137 
2138 /*
2139  * Structure used to pass information from ffs_sbupdate to its
2140  * helper routine ffs_use_bwrite.
2141  */
2142 struct devfd {
2143 	struct ufsmount	*ump;
2144 	struct buf	*sbbp;
2145 	int		 waitfor;
2146 	int		 suspended;
2147 	int		 error;
2148 };
2149 
2150 /*
2151  * Write a superblock and associated information back to disk.
2152  */
2153 int
2154 ffs_sbupdate(ump, waitfor, suspended)
2155 	struct ufsmount *ump;
2156 	int waitfor;
2157 	int suspended;
2158 {
2159 	struct fs *fs;
2160 	struct buf *sbbp;
2161 	struct devfd devfd;
2162 
2163 	fs = ump->um_fs;
2164 	if (fs->fs_ronly == 1 &&
2165 	    (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
2166 	    (MNT_RDONLY | MNT_UPDATE))
2167 		panic("ffs_sbupdate: write read-only filesystem");
2168 	/*
2169 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
2170 	 */
2171 	sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
2172 	    (int)fs->fs_sbsize, 0, 0, 0);
2173 	/*
2174 	 * Initialize info needed for write function.
2175 	 */
2176 	devfd.ump = ump;
2177 	devfd.sbbp = sbbp;
2178 	devfd.waitfor = waitfor;
2179 	devfd.suspended = suspended;
2180 	devfd.error = 0;
2181 	return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite));
2182 }
2183 
2184 /*
2185  * Write function for use by filesystem-layer routines.
2186  */
2187 static int
2188 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size)
2189 {
2190 	struct devfd *devfdp;
2191 	struct ufsmount *ump;
2192 	struct buf *bp;
2193 	struct fs *fs;
2194 	int error;
2195 
2196 	devfdp = devfd;
2197 	ump = devfdp->ump;
2198 	fs = ump->um_fs;
2199 	/*
2200 	 * Writing the superblock summary information.
2201 	 */
2202 	if (loc != fs->fs_sblockloc) {
2203 		bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0);
2204 		bcopy(buf, bp->b_data, (u_int)size);
2205 		if (devfdp->suspended)
2206 			bp->b_flags |= B_VALIDSUSPWRT;
2207 		if (devfdp->waitfor != MNT_WAIT)
2208 			bawrite(bp);
2209 		else if ((error = bwrite(bp)) != 0)
2210 			devfdp->error = error;
2211 		return (0);
2212 	}
2213 	/*
2214 	 * Writing the superblock itself. We need to do special checks for it.
2215 	 */
2216 	bp = devfdp->sbbp;
2217 	if (ffs_fsfail_cleanup(ump, devfdp->error))
2218 		devfdp->error = 0;
2219 	if (devfdp->error != 0) {
2220 		brelse(bp);
2221 		return (devfdp->error);
2222 	}
2223 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
2224 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2225 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2226 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
2227 		fs->fs_sblockloc = SBLOCK_UFS1;
2228 	}
2229 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
2230 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2231 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2232 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
2233 		fs->fs_sblockloc = SBLOCK_UFS2;
2234 	}
2235 	if (MOUNTEDSOFTDEP(ump->um_mountp))
2236 		softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
2237 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
2238 	fs = (struct fs *)bp->b_data;
2239 	ffs_oldfscompat_write(fs, ump);
2240 	fs->fs_si = NULL;
2241 	/* Recalculate the superblock hash */
2242 	fs->fs_ckhash = ffs_calc_sbhash(fs);
2243 	if (devfdp->suspended)
2244 		bp->b_flags |= B_VALIDSUSPWRT;
2245 	if (devfdp->waitfor != MNT_WAIT)
2246 		bawrite(bp);
2247 	else if ((error = bwrite(bp)) != 0)
2248 		devfdp->error = error;
2249 	return (devfdp->error);
2250 }
2251 
2252 static int
2253 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
2254 	int attrnamespace, const char *attrname)
2255 {
2256 
2257 #ifdef UFS_EXTATTR
2258 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
2259 	    attrname));
2260 #else
2261 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
2262 	    attrname));
2263 #endif
2264 }
2265 
2266 static void
2267 ffs_ifree(struct ufsmount *ump, struct inode *ip)
2268 {
2269 
2270 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
2271 		uma_zfree(uma_ufs1, ip->i_din1);
2272 	else if (ip->i_din2 != NULL)
2273 		uma_zfree(uma_ufs2, ip->i_din2);
2274 	uma_zfree_smr(uma_inode, ip);
2275 }
2276 
2277 static int dobkgrdwrite = 1;
2278 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
2279     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2280 
2281 /*
2282  * Complete a background write started from bwrite.
2283  */
2284 static void
2285 ffs_backgroundwritedone(struct buf *bp)
2286 {
2287 	struct bufobj *bufobj;
2288 	struct buf *origbp;
2289 
2290 #ifdef SOFTUPDATES
2291 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0)
2292 		softdep_handle_error(bp);
2293 #endif
2294 
2295 	/*
2296 	 * Find the original buffer that we are writing.
2297 	 */
2298 	bufobj = bp->b_bufobj;
2299 	BO_LOCK(bufobj);
2300 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2301 		panic("backgroundwritedone: lost buffer");
2302 
2303 	/*
2304 	 * We should mark the cylinder group buffer origbp as
2305 	 * dirty, to not lose the failed write.
2306 	 */
2307 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2308 		origbp->b_vflags |= BV_BKGRDERR;
2309 	BO_UNLOCK(bufobj);
2310 	/*
2311 	 * Process dependencies then return any unfinished ones.
2312 	 */
2313 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2314 		buf_complete(bp);
2315 #ifdef SOFTUPDATES
2316 	if (!LIST_EMPTY(&bp->b_dep))
2317 		softdep_move_dependencies(bp, origbp);
2318 #endif
2319 	/*
2320 	 * This buffer is marked B_NOCACHE so when it is released
2321 	 * by biodone it will be tossed.  Clear B_IOSTARTED in case of error.
2322 	 */
2323 	bp->b_flags |= B_NOCACHE;
2324 	bp->b_flags &= ~(B_CACHE | B_IOSTARTED);
2325 	pbrelvp(bp);
2326 
2327 	/*
2328 	 * Prevent brelse() from trying to keep and re-dirtying bp on
2329 	 * errors. It causes b_bufobj dereference in
2330 	 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2331 	 * pbrelvp() above.
2332 	 */
2333 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2334 		bp->b_flags |= B_INVAL;
2335 	bufdone(bp);
2336 	BO_LOCK(bufobj);
2337 	/*
2338 	 * Clear the BV_BKGRDINPROG flag in the original buffer
2339 	 * and awaken it if it is waiting for the write to complete.
2340 	 * If BV_BKGRDINPROG is not set in the original buffer it must
2341 	 * have been released and re-instantiated - which is not legal.
2342 	 */
2343 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2344 	    ("backgroundwritedone: lost buffer2"));
2345 	origbp->b_vflags &= ~BV_BKGRDINPROG;
2346 	if (origbp->b_vflags & BV_BKGRDWAIT) {
2347 		origbp->b_vflags &= ~BV_BKGRDWAIT;
2348 		wakeup(&origbp->b_xflags);
2349 	}
2350 	BO_UNLOCK(bufobj);
2351 }
2352 
2353 /*
2354  * Write, release buffer on completion.  (Done by iodone
2355  * if async).  Do not bother writing anything if the buffer
2356  * is invalid.
2357  *
2358  * Note that we set B_CACHE here, indicating that buffer is
2359  * fully valid and thus cacheable.  This is true even of NFS
2360  * now so we set it generally.  This could be set either here
2361  * or in biodone() since the I/O is synchronous.  We put it
2362  * here.
2363  */
2364 static int
2365 ffs_bufwrite(struct buf *bp)
2366 {
2367 	struct buf *newbp;
2368 	struct cg *cgp;
2369 
2370 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2371 	if (bp->b_flags & B_INVAL) {
2372 		brelse(bp);
2373 		return (0);
2374 	}
2375 
2376 	if (!BUF_ISLOCKED(bp))
2377 		panic("bufwrite: buffer is not busy???");
2378 	/*
2379 	 * If a background write is already in progress, delay
2380 	 * writing this block if it is asynchronous. Otherwise
2381 	 * wait for the background write to complete.
2382 	 */
2383 	BO_LOCK(bp->b_bufobj);
2384 	if (bp->b_vflags & BV_BKGRDINPROG) {
2385 		if (bp->b_flags & B_ASYNC) {
2386 			BO_UNLOCK(bp->b_bufobj);
2387 			bdwrite(bp);
2388 			return (0);
2389 		}
2390 		bp->b_vflags |= BV_BKGRDWAIT;
2391 		msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2392 		    "bwrbg", 0);
2393 		if (bp->b_vflags & BV_BKGRDINPROG)
2394 			panic("bufwrite: still writing");
2395 	}
2396 	bp->b_vflags &= ~BV_BKGRDERR;
2397 	BO_UNLOCK(bp->b_bufobj);
2398 
2399 	/*
2400 	 * If this buffer is marked for background writing and we
2401 	 * do not have to wait for it, make a copy and write the
2402 	 * copy so as to leave this buffer ready for further use.
2403 	 *
2404 	 * This optimization eats a lot of memory.  If we have a page
2405 	 * or buffer shortfall we can't do it.
2406 	 */
2407 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2408 	    (bp->b_flags & B_ASYNC) &&
2409 	    !vm_page_count_severe() &&
2410 	    !buf_dirty_count_severe()) {
2411 		KASSERT(bp->b_iodone == NULL,
2412 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2413 
2414 		/* get a new block */
2415 		newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2416 		if (newbp == NULL)
2417 			goto normal_write;
2418 
2419 		KASSERT(buf_mapped(bp), ("Unmapped cg"));
2420 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2421 		BO_LOCK(bp->b_bufobj);
2422 		bp->b_vflags |= BV_BKGRDINPROG;
2423 		BO_UNLOCK(bp->b_bufobj);
2424 		newbp->b_xflags |=
2425 		    (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER;
2426 		newbp->b_lblkno = bp->b_lblkno;
2427 		newbp->b_blkno = bp->b_blkno;
2428 		newbp->b_offset = bp->b_offset;
2429 		newbp->b_iodone = ffs_backgroundwritedone;
2430 		newbp->b_flags |= B_ASYNC;
2431 		newbp->b_flags &= ~B_INVAL;
2432 		pbgetvp(bp->b_vp, newbp);
2433 
2434 #ifdef SOFTUPDATES
2435 		/*
2436 		 * Move over the dependencies.  If there are rollbacks,
2437 		 * leave the parent buffer dirtied as it will need to
2438 		 * be written again.
2439 		 */
2440 		if (LIST_EMPTY(&bp->b_dep) ||
2441 		    softdep_move_dependencies(bp, newbp) == 0)
2442 			bundirty(bp);
2443 #else
2444 		bundirty(bp);
2445 #endif
2446 
2447 		/*
2448 		 * Initiate write on the copy, release the original.  The
2449 		 * BKGRDINPROG flag prevents it from going away until
2450 		 * the background write completes. We have to recalculate
2451 		 * its check hash in case the buffer gets freed and then
2452 		 * reconstituted from the buffer cache during a later read.
2453 		 */
2454 		if ((bp->b_xflags & BX_CYLGRP) != 0) {
2455 			cgp = (struct cg *)bp->b_data;
2456 			cgp->cg_ckhash = 0;
2457 			cgp->cg_ckhash =
2458 			    calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2459 		}
2460 		bqrelse(bp);
2461 		bp = newbp;
2462 	} else
2463 		/* Mark the buffer clean */
2464 		bundirty(bp);
2465 
2466 	/* Let the normal bufwrite do the rest for us */
2467 normal_write:
2468 	/*
2469 	 * If we are writing a cylinder group, update its time.
2470 	 */
2471 	if ((bp->b_xflags & BX_CYLGRP) != 0) {
2472 		cgp = (struct cg *)bp->b_data;
2473 		cgp->cg_old_time = cgp->cg_time = time_second;
2474 	}
2475 	return (bufwrite(bp));
2476 }
2477 
2478 static void
2479 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2480 {
2481 	struct vnode *vp;
2482 	struct buf *tbp;
2483 	int error, nocopy;
2484 
2485 	/*
2486 	 * This is the bufobj strategy for the private VCHR vnodes
2487 	 * used by FFS to access the underlying storage device.
2488 	 * We override the default bufobj strategy and thus bypass
2489 	 * VOP_STRATEGY() for these vnodes.
2490 	 */
2491 	vp = bo2vnode(bo);
2492 	KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR ||
2493 	    bp->b_vp->v_rdev == NULL ||
2494 	    bp->b_vp->v_rdev->si_mountpt == NULL ||
2495 	    VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL ||
2496 	    vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp,
2497 	    ("ffs_geom_strategy() with wrong vp"));
2498 	if (bp->b_iocmd == BIO_WRITE) {
2499 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2500 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2501 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2502 			panic("ffs_geom_strategy: bad I/O");
2503 		nocopy = bp->b_flags & B_NOCOPY;
2504 		bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2505 		if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2506 		    vp->v_rdev->si_snapdata != NULL) {
2507 			if ((bp->b_flags & B_CLUSTER) != 0) {
2508 				runningbufwakeup(bp);
2509 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2510 					      b_cluster.cluster_entry) {
2511 					error = ffs_copyonwrite(vp, tbp);
2512 					if (error != 0 &&
2513 					    error != EOPNOTSUPP) {
2514 						bp->b_error = error;
2515 						bp->b_ioflags |= BIO_ERROR;
2516 						bp->b_flags &= ~B_BARRIER;
2517 						bufdone(bp);
2518 						return;
2519 					}
2520 				}
2521 				bp->b_runningbufspace = bp->b_bufsize;
2522 				atomic_add_long(&runningbufspace,
2523 					       bp->b_runningbufspace);
2524 			} else {
2525 				error = ffs_copyonwrite(vp, bp);
2526 				if (error != 0 && error != EOPNOTSUPP) {
2527 					bp->b_error = error;
2528 					bp->b_ioflags |= BIO_ERROR;
2529 					bp->b_flags &= ~B_BARRIER;
2530 					bufdone(bp);
2531 					return;
2532 				}
2533 			}
2534 		}
2535 #ifdef SOFTUPDATES
2536 		if ((bp->b_flags & B_CLUSTER) != 0) {
2537 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2538 				      b_cluster.cluster_entry) {
2539 				if (!LIST_EMPTY(&tbp->b_dep))
2540 					buf_start(tbp);
2541 			}
2542 		} else {
2543 			if (!LIST_EMPTY(&bp->b_dep))
2544 				buf_start(bp);
2545 		}
2546 
2547 #endif
2548 		/*
2549 		 * Check for metadata that needs check-hashes and update them.
2550 		 */
2551 		switch (bp->b_xflags & BX_FSPRIV) {
2552 		case BX_CYLGRP:
2553 			((struct cg *)bp->b_data)->cg_ckhash = 0;
2554 			((struct cg *)bp->b_data)->cg_ckhash =
2555 			    calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2556 			break;
2557 
2558 		case BX_SUPERBLOCK:
2559 		case BX_INODE:
2560 		case BX_INDIR:
2561 		case BX_DIR:
2562 			printf("Check-hash write is unimplemented!!!\n");
2563 			break;
2564 
2565 		case 0:
2566 			break;
2567 
2568 		default:
2569 			printf("multiple buffer types 0x%b\n",
2570 			    (u_int)(bp->b_xflags & BX_FSPRIV),
2571 			    PRINT_UFS_BUF_XFLAGS);
2572 			break;
2573 		}
2574 	}
2575 	if (bp->b_iocmd != BIO_READ && ffs_enxio_enable)
2576 		bp->b_xflags |= BX_CVTENXIO;
2577 	g_vfs_strategy(bo, bp);
2578 }
2579 
2580 int
2581 ffs_own_mount(const struct mount *mp)
2582 {
2583 
2584 	if (mp->mnt_op == &ufs_vfsops)
2585 		return (1);
2586 	return (0);
2587 }
2588 
2589 #ifdef	DDB
2590 #ifdef SOFTUPDATES
2591 
2592 /* defined in ffs_softdep.c */
2593 extern void db_print_ffs(struct ufsmount *ump);
2594 
2595 DB_SHOW_COMMAND(ffs, db_show_ffs)
2596 {
2597 	struct mount *mp;
2598 	struct ufsmount *ump;
2599 
2600 	if (have_addr) {
2601 		ump = VFSTOUFS((struct mount *)addr);
2602 		db_print_ffs(ump);
2603 		return;
2604 	}
2605 
2606 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2607 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2608 			db_print_ffs(VFSTOUFS(mp));
2609 	}
2610 }
2611 
2612 #endif	/* SOFTUPDATES */
2613 #endif	/* DDB */
2614