xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision f4bf2442a03f9b72cfe6d051766b650a4721f3d8)
1 /*-
2  * Copyright (c) 1989, 1991, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_quota.h"
36 #include "opt_ufs.h"
37 #include "opt_ffs.h"
38 #include "opt_ddb.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/taskqueue.h>
46 #include <sys/kernel.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/fcntl.h>
53 #include <sys/ioccom.h>
54 #include <sys/malloc.h>
55 #include <sys/mutex.h>
56 #include <sys/rwlock.h>
57 
58 #include <security/mac/mac_framework.h>
59 
60 #include <ufs/ufs/extattr.h>
61 #include <ufs/ufs/gjournal.h>
62 #include <ufs/ufs/quota.h>
63 #include <ufs/ufs/ufsmount.h>
64 #include <ufs/ufs/inode.h>
65 #include <ufs/ufs/ufs_extern.h>
66 
67 #include <ufs/ffs/fs.h>
68 #include <ufs/ffs/ffs_extern.h>
69 
70 #include <vm/vm.h>
71 #include <vm/uma.h>
72 #include <vm/vm_page.h>
73 
74 #include <geom/geom.h>
75 #include <geom/geom_vfs.h>
76 
77 #include <ddb/ddb.h>
78 
79 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
80 
81 static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
82 static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
83 		    ufs2_daddr_t);
84 static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
85 static int	ffs_sync_lazy(struct mount *mp);
86 
87 static vfs_init_t ffs_init;
88 static vfs_uninit_t ffs_uninit;
89 static vfs_extattrctl_t ffs_extattrctl;
90 static vfs_cmount_t ffs_cmount;
91 static vfs_unmount_t ffs_unmount;
92 static vfs_mount_t ffs_mount;
93 static vfs_statfs_t ffs_statfs;
94 static vfs_fhtovp_t ffs_fhtovp;
95 static vfs_sync_t ffs_sync;
96 
97 static struct vfsops ufs_vfsops = {
98 	.vfs_extattrctl =	ffs_extattrctl,
99 	.vfs_fhtovp =		ffs_fhtovp,
100 	.vfs_init =		ffs_init,
101 	.vfs_mount =		ffs_mount,
102 	.vfs_cmount =		ffs_cmount,
103 	.vfs_quotactl =		ufs_quotactl,
104 	.vfs_root =		ufs_root,
105 	.vfs_statfs =		ffs_statfs,
106 	.vfs_sync =		ffs_sync,
107 	.vfs_uninit =		ffs_uninit,
108 	.vfs_unmount =		ffs_unmount,
109 	.vfs_vget =		ffs_vget,
110 	.vfs_susp_clean =	process_deferred_inactive,
111 };
112 
113 VFS_SET(ufs_vfsops, ufs, 0);
114 MODULE_VERSION(ufs, 1);
115 
116 static b_strategy_t ffs_geom_strategy;
117 static b_write_t ffs_bufwrite;
118 
119 static struct buf_ops ffs_ops = {
120 	.bop_name =	"FFS",
121 	.bop_write =	ffs_bufwrite,
122 	.bop_strategy =	ffs_geom_strategy,
123 	.bop_sync =	bufsync,
124 #ifdef NO_FFS_SNAPSHOT
125 	.bop_bdflush =	bufbdflush,
126 #else
127 	.bop_bdflush =	ffs_bdflush,
128 #endif
129 };
130 
131 /*
132  * Note that userquota and groupquota options are not currently used
133  * by UFS/FFS code and generally mount(8) does not pass those options
134  * from userland, but they can be passed by loader(8) via
135  * vfs.root.mountfrom.options.
136  */
137 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
138     "noclusterw", "noexec", "export", "force", "from", "groupquota",
139     "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
140     "nosymfollow", "sync", "union", "userquota", NULL };
141 
142 static int
143 ffs_mount(struct mount *mp)
144 {
145 	struct vnode *devvp;
146 	struct thread *td;
147 	struct ufsmount *ump = NULL;
148 	struct fs *fs;
149 	pid_t fsckpid = 0;
150 	int error, flags;
151 	uint64_t mntorflags;
152 	accmode_t accmode;
153 	struct nameidata ndp;
154 	char *fspec;
155 
156 	td = curthread;
157 	if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
158 		return (EINVAL);
159 	if (uma_inode == NULL) {
160 		uma_inode = uma_zcreate("FFS inode",
161 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
162 		    UMA_ALIGN_PTR, 0);
163 		uma_ufs1 = uma_zcreate("FFS1 dinode",
164 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
165 		    UMA_ALIGN_PTR, 0);
166 		uma_ufs2 = uma_zcreate("FFS2 dinode",
167 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
168 		    UMA_ALIGN_PTR, 0);
169 	}
170 
171 	vfs_deleteopt(mp->mnt_optnew, "groupquota");
172 	vfs_deleteopt(mp->mnt_optnew, "userquota");
173 
174 	fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
175 	if (error)
176 		return (error);
177 
178 	mntorflags = 0;
179 	if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
180 		mntorflags |= MNT_ACLS;
181 
182 	if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
183 		mntorflags |= MNT_SNAPSHOT;
184 		/*
185 		 * Once we have set the MNT_SNAPSHOT flag, do not
186 		 * persist "snapshot" in the options list.
187 		 */
188 		vfs_deleteopt(mp->mnt_optnew, "snapshot");
189 		vfs_deleteopt(mp->mnt_opt, "snapshot");
190 	}
191 
192 	if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
193 	    vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
194 		/*
195 		 * Once we have set the restricted PID, do not
196 		 * persist "fsckpid" in the options list.
197 		 */
198 		vfs_deleteopt(mp->mnt_optnew, "fsckpid");
199 		vfs_deleteopt(mp->mnt_opt, "fsckpid");
200 		if (mp->mnt_flag & MNT_UPDATE) {
201 			if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
202 			     vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
203 				vfs_mount_error(mp,
204 				    "Checker enable: Must be read-only");
205 				return (EINVAL);
206 			}
207 		} else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
208 			vfs_mount_error(mp,
209 			    "Checker enable: Must be read-only");
210 			return (EINVAL);
211 		}
212 		/* Set to -1 if we are done */
213 		if (fsckpid == 0)
214 			fsckpid = -1;
215 	}
216 
217 	if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
218 		if (mntorflags & MNT_ACLS) {
219 			vfs_mount_error(mp,
220 			    "\"acls\" and \"nfsv4acls\" options "
221 			    "are mutually exclusive");
222 			return (EINVAL);
223 		}
224 		mntorflags |= MNT_NFS4ACLS;
225 	}
226 
227 	MNT_ILOCK(mp);
228 	mp->mnt_flag |= mntorflags;
229 	MNT_IUNLOCK(mp);
230 	/*
231 	 * If updating, check whether changing from read-only to
232 	 * read/write; if there is no device name, that's all we do.
233 	 */
234 	if (mp->mnt_flag & MNT_UPDATE) {
235 		ump = VFSTOUFS(mp);
236 		fs = ump->um_fs;
237 		devvp = ump->um_devvp;
238 		if (fsckpid == -1 && ump->um_fsckpid > 0) {
239 			if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
240 			    (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
241 				return (error);
242 			DROP_GIANT();
243 			g_topology_lock();
244 			/*
245 			 * Return to normal read-only mode.
246 			 */
247 			error = g_access(ump->um_cp, 0, -1, 0);
248 			g_topology_unlock();
249 			PICKUP_GIANT();
250 			ump->um_fsckpid = 0;
251 		}
252 		if (fs->fs_ronly == 0 &&
253 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
254 			/*
255 			 * Flush any dirty data and suspend filesystem.
256 			 */
257 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
258 				return (error);
259 			error = vfs_write_suspend_umnt(mp);
260 			if (error != 0)
261 				return (error);
262 			/*
263 			 * Check for and optionally get rid of files open
264 			 * for writing.
265 			 */
266 			flags = WRITECLOSE;
267 			if (mp->mnt_flag & MNT_FORCE)
268 				flags |= FORCECLOSE;
269 			if (MOUNTEDSOFTDEP(mp)) {
270 				error = softdep_flushfiles(mp, flags, td);
271 			} else {
272 				error = ffs_flushfiles(mp, flags, td);
273 			}
274 			if (error) {
275 				vfs_write_resume(mp, 0);
276 				return (error);
277 			}
278 			if (fs->fs_pendingblocks != 0 ||
279 			    fs->fs_pendinginodes != 0) {
280 				printf("WARNING: %s Update error: blocks %jd "
281 				    "files %d\n", fs->fs_fsmnt,
282 				    (intmax_t)fs->fs_pendingblocks,
283 				    fs->fs_pendinginodes);
284 				fs->fs_pendingblocks = 0;
285 				fs->fs_pendinginodes = 0;
286 			}
287 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
288 				fs->fs_clean = 1;
289 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
290 				fs->fs_ronly = 0;
291 				fs->fs_clean = 0;
292 				vfs_write_resume(mp, 0);
293 				return (error);
294 			}
295 			if (MOUNTEDSOFTDEP(mp))
296 				softdep_unmount(mp);
297 			DROP_GIANT();
298 			g_topology_lock();
299 			/*
300 			 * Drop our write and exclusive access.
301 			 */
302 			g_access(ump->um_cp, 0, -1, -1);
303 			g_topology_unlock();
304 			PICKUP_GIANT();
305 			fs->fs_ronly = 1;
306 			MNT_ILOCK(mp);
307 			mp->mnt_flag |= MNT_RDONLY;
308 			MNT_IUNLOCK(mp);
309 			/*
310 			 * Allow the writers to note that filesystem
311 			 * is ro now.
312 			 */
313 			vfs_write_resume(mp, 0);
314 		}
315 		if ((mp->mnt_flag & MNT_RELOAD) &&
316 		    (error = ffs_reload(mp, td, 0)) != 0)
317 			return (error);
318 		if (fs->fs_ronly &&
319 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
320 			/*
321 			 * If we are running a checker, do not allow upgrade.
322 			 */
323 			if (ump->um_fsckpid > 0) {
324 				vfs_mount_error(mp,
325 				    "Active checker, cannot upgrade to write");
326 				return (EINVAL);
327 			}
328 			/*
329 			 * If upgrade to read-write by non-root, then verify
330 			 * that user has necessary permissions on the device.
331 			 */
332 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
333 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
334 			    td->td_ucred, td);
335 			if (error)
336 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
337 			if (error) {
338 				VOP_UNLOCK(devvp, 0);
339 				return (error);
340 			}
341 			VOP_UNLOCK(devvp, 0);
342 			fs->fs_flags &= ~FS_UNCLEAN;
343 			if (fs->fs_clean == 0) {
344 				fs->fs_flags |= FS_UNCLEAN;
345 				if ((mp->mnt_flag & MNT_FORCE) ||
346 				    ((fs->fs_flags &
347 				     (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
348 				     (fs->fs_flags & FS_DOSOFTDEP))) {
349 					printf("WARNING: %s was not properly "
350 					   "dismounted\n", fs->fs_fsmnt);
351 				} else {
352 					vfs_mount_error(mp,
353 					   "R/W mount of %s denied. %s.%s",
354 					   fs->fs_fsmnt,
355 					   "Filesystem is not clean - run fsck",
356 					   (fs->fs_flags & FS_SUJ) == 0 ? "" :
357 					   " Forced mount will invalidate"
358 					   " journal contents");
359 					return (EPERM);
360 				}
361 			}
362 			DROP_GIANT();
363 			g_topology_lock();
364 			/*
365 			 * Request exclusive write access.
366 			 */
367 			error = g_access(ump->um_cp, 0, 1, 1);
368 			g_topology_unlock();
369 			PICKUP_GIANT();
370 			if (error)
371 				return (error);
372 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
373 				return (error);
374 			fs->fs_ronly = 0;
375 			MNT_ILOCK(mp);
376 			mp->mnt_flag &= ~MNT_RDONLY;
377 			MNT_IUNLOCK(mp);
378 			fs->fs_mtime = time_second;
379 			/* check to see if we need to start softdep */
380 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
381 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
382 				vn_finished_write(mp);
383 				return (error);
384 			}
385 			fs->fs_clean = 0;
386 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
387 				vn_finished_write(mp);
388 				return (error);
389 			}
390 			if (fs->fs_snapinum[0] != 0)
391 				ffs_snapshot_mount(mp);
392 			vn_finished_write(mp);
393 		}
394 		/*
395 		 * Soft updates is incompatible with "async",
396 		 * so if we are doing softupdates stop the user
397 		 * from setting the async flag in an update.
398 		 * Softdep_mount() clears it in an initial mount
399 		 * or ro->rw remount.
400 		 */
401 		if (MOUNTEDSOFTDEP(mp)) {
402 			/* XXX: Reset too late ? */
403 			MNT_ILOCK(mp);
404 			mp->mnt_flag &= ~MNT_ASYNC;
405 			MNT_IUNLOCK(mp);
406 		}
407 		/*
408 		 * Keep MNT_ACLS flag if it is stored in superblock.
409 		 */
410 		if ((fs->fs_flags & FS_ACLS) != 0) {
411 			/* XXX: Set too late ? */
412 			MNT_ILOCK(mp);
413 			mp->mnt_flag |= MNT_ACLS;
414 			MNT_IUNLOCK(mp);
415 		}
416 
417 		if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
418 			/* XXX: Set too late ? */
419 			MNT_ILOCK(mp);
420 			mp->mnt_flag |= MNT_NFS4ACLS;
421 			MNT_IUNLOCK(mp);
422 		}
423 		/*
424 		 * If this is a request from fsck to clean up the filesystem,
425 		 * then allow the specified pid to proceed.
426 		 */
427 		if (fsckpid > 0) {
428 			if (ump->um_fsckpid != 0) {
429 				vfs_mount_error(mp,
430 				    "Active checker already running on %s",
431 				    fs->fs_fsmnt);
432 				return (EINVAL);
433 			}
434 			KASSERT(MOUNTEDSOFTDEP(mp) == 0,
435 			    ("soft updates enabled on read-only file system"));
436 			DROP_GIANT();
437 			g_topology_lock();
438 			/*
439 			 * Request write access.
440 			 */
441 			error = g_access(ump->um_cp, 0, 1, 0);
442 			g_topology_unlock();
443 			PICKUP_GIANT();
444 			if (error) {
445 				vfs_mount_error(mp,
446 				    "Checker activation failed on %s",
447 				    fs->fs_fsmnt);
448 				return (error);
449 			}
450 			ump->um_fsckpid = fsckpid;
451 			if (fs->fs_snapinum[0] != 0)
452 				ffs_snapshot_mount(mp);
453 			fs->fs_mtime = time_second;
454 			fs->fs_fmod = 1;
455 			fs->fs_clean = 0;
456 			(void) ffs_sbupdate(ump, MNT_WAIT, 0);
457 		}
458 
459 		/*
460 		 * If this is a snapshot request, take the snapshot.
461 		 */
462 		if (mp->mnt_flag & MNT_SNAPSHOT)
463 			return (ffs_snapshot(mp, fspec));
464 	}
465 
466 	/*
467 	 * Not an update, or updating the name: look up the name
468 	 * and verify that it refers to a sensible disk device.
469 	 */
470 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
471 	if ((error = namei(&ndp)) != 0)
472 		return (error);
473 	NDFREE(&ndp, NDF_ONLY_PNBUF);
474 	devvp = ndp.ni_vp;
475 	if (!vn_isdisk(devvp, &error)) {
476 		vput(devvp);
477 		return (error);
478 	}
479 
480 	/*
481 	 * If mount by non-root, then verify that user has necessary
482 	 * permissions on the device.
483 	 */
484 	accmode = VREAD;
485 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
486 		accmode |= VWRITE;
487 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
488 	if (error)
489 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
490 	if (error) {
491 		vput(devvp);
492 		return (error);
493 	}
494 
495 	if (mp->mnt_flag & MNT_UPDATE) {
496 		/*
497 		 * Update only
498 		 *
499 		 * If it's not the same vnode, or at least the same device
500 		 * then it's not correct.
501 		 */
502 
503 		if (devvp->v_rdev != ump->um_devvp->v_rdev)
504 			error = EINVAL;	/* needs translation */
505 		vput(devvp);
506 		if (error)
507 			return (error);
508 	} else {
509 		/*
510 		 * New mount
511 		 *
512 		 * We need the name for the mount point (also used for
513 		 * "last mounted on") copied in. If an error occurs,
514 		 * the mount point is discarded by the upper level code.
515 		 * Note that vfs_mount_alloc() populates f_mntonname for us.
516 		 */
517 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
518 			vrele(devvp);
519 			return (error);
520 		}
521 		if (fsckpid > 0) {
522 			KASSERT(MOUNTEDSOFTDEP(mp) == 0,
523 			    ("soft updates enabled on read-only file system"));
524 			ump = VFSTOUFS(mp);
525 			fs = ump->um_fs;
526 			DROP_GIANT();
527 			g_topology_lock();
528 			/*
529 			 * Request write access.
530 			 */
531 			error = g_access(ump->um_cp, 0, 1, 0);
532 			g_topology_unlock();
533 			PICKUP_GIANT();
534 			if (error) {
535 				printf("WARNING: %s: Checker activation "
536 				    "failed\n", fs->fs_fsmnt);
537 			} else {
538 				ump->um_fsckpid = fsckpid;
539 				if (fs->fs_snapinum[0] != 0)
540 					ffs_snapshot_mount(mp);
541 				fs->fs_mtime = time_second;
542 				fs->fs_clean = 0;
543 				(void) ffs_sbupdate(ump, MNT_WAIT, 0);
544 			}
545 		}
546 	}
547 	vfs_mountedfrom(mp, fspec);
548 	return (0);
549 }
550 
551 /*
552  * Compatibility with old mount system call.
553  */
554 
555 static int
556 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
557 {
558 	struct ufs_args args;
559 	struct export_args exp;
560 	int error;
561 
562 	if (data == NULL)
563 		return (EINVAL);
564 	error = copyin(data, &args, sizeof args);
565 	if (error)
566 		return (error);
567 	vfs_oexport_conv(&args.export, &exp);
568 
569 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
570 	ma = mount_arg(ma, "export", &exp, sizeof(exp));
571 	error = kernel_mount(ma, flags);
572 
573 	return (error);
574 }
575 
576 /*
577  * Reload all incore data for a filesystem (used after running fsck on
578  * the root filesystem and finding things to fix). If the 'force' flag
579  * is 0, the filesystem must be mounted read-only.
580  *
581  * Things to do to update the mount:
582  *	1) invalidate all cached meta-data.
583  *	2) re-read superblock from disk.
584  *	3) re-read summary information from disk.
585  *	4) invalidate all inactive vnodes.
586  *	5) invalidate all cached file data.
587  *	6) re-read inode data for all active vnodes.
588  */
589 int
590 ffs_reload(struct mount *mp, struct thread *td, int force)
591 {
592 	struct vnode *vp, *mvp, *devvp;
593 	struct inode *ip;
594 	void *space;
595 	struct buf *bp;
596 	struct fs *fs, *newfs;
597 	struct ufsmount *ump;
598 	ufs2_daddr_t sblockloc;
599 	int i, blks, size, error;
600 	int32_t *lp;
601 
602 	ump = VFSTOUFS(mp);
603 
604 	MNT_ILOCK(mp);
605 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && force == 0) {
606 		MNT_IUNLOCK(mp);
607 		return (EINVAL);
608 	}
609 	MNT_IUNLOCK(mp);
610 
611 	/*
612 	 * Step 1: invalidate all cached meta-data.
613 	 */
614 	devvp = VFSTOUFS(mp)->um_devvp;
615 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
616 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
617 		panic("ffs_reload: dirty1");
618 	VOP_UNLOCK(devvp, 0);
619 
620 	/*
621 	 * Step 2: re-read superblock from disk.
622 	 */
623 	fs = VFSTOUFS(mp)->um_fs;
624 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
625 	    NOCRED, &bp)) != 0)
626 		return (error);
627 	newfs = (struct fs *)bp->b_data;
628 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
629 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
630 	    newfs->fs_bsize > MAXBSIZE ||
631 	    newfs->fs_bsize < sizeof(struct fs)) {
632 			brelse(bp);
633 			return (EIO);		/* XXX needs translation */
634 	}
635 	/*
636 	 * Copy pointer fields back into superblock before copying in	XXX
637 	 * new superblock. These should really be in the ufsmount.	XXX
638 	 * Note that important parameters (eg fs_ncg) are unchanged.
639 	 */
640 	newfs->fs_csp = fs->fs_csp;
641 	newfs->fs_maxcluster = fs->fs_maxcluster;
642 	newfs->fs_contigdirs = fs->fs_contigdirs;
643 	newfs->fs_active = fs->fs_active;
644 	newfs->fs_ronly = fs->fs_ronly;
645 	sblockloc = fs->fs_sblockloc;
646 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
647 	brelse(bp);
648 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
649 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
650 	UFS_LOCK(ump);
651 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
652 		printf("WARNING: %s: reload pending error: blocks %jd "
653 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
654 		    fs->fs_pendinginodes);
655 		fs->fs_pendingblocks = 0;
656 		fs->fs_pendinginodes = 0;
657 	}
658 	UFS_UNLOCK(ump);
659 
660 	/*
661 	 * Step 3: re-read summary information from disk.
662 	 */
663 	size = fs->fs_cssize;
664 	blks = howmany(size, fs->fs_fsize);
665 	if (fs->fs_contigsumsize > 0)
666 		size += fs->fs_ncg * sizeof(int32_t);
667 	size += fs->fs_ncg * sizeof(u_int8_t);
668 	free(fs->fs_csp, M_UFSMNT);
669 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
670 	fs->fs_csp = space;
671 	for (i = 0; i < blks; i += fs->fs_frag) {
672 		size = fs->fs_bsize;
673 		if (i + fs->fs_frag > blks)
674 			size = (blks - i) * fs->fs_fsize;
675 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
676 		    NOCRED, &bp);
677 		if (error)
678 			return (error);
679 		bcopy(bp->b_data, space, (u_int)size);
680 		space = (char *)space + size;
681 		brelse(bp);
682 	}
683 	/*
684 	 * We no longer know anything about clusters per cylinder group.
685 	 */
686 	if (fs->fs_contigsumsize > 0) {
687 		fs->fs_maxcluster = lp = space;
688 		for (i = 0; i < fs->fs_ncg; i++)
689 			*lp++ = fs->fs_contigsumsize;
690 		space = lp;
691 	}
692 	size = fs->fs_ncg * sizeof(u_int8_t);
693 	fs->fs_contigdirs = (u_int8_t *)space;
694 	bzero(fs->fs_contigdirs, size);
695 
696 loop:
697 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
698 		/*
699 		 * Skip syncer vnode.
700 		 */
701 		if (vp->v_type == VNON) {
702 			VI_UNLOCK(vp);
703 			continue;
704 		}
705 		/*
706 		 * Step 4: invalidate all cached file data.
707 		 */
708 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
709 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
710 			goto loop;
711 		}
712 		if (vinvalbuf(vp, 0, 0, 0))
713 			panic("ffs_reload: dirty2");
714 		/*
715 		 * Step 5: re-read inode data for all active vnodes.
716 		 */
717 		ip = VTOI(vp);
718 		error =
719 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
720 		    (int)fs->fs_bsize, NOCRED, &bp);
721 		if (error) {
722 			VOP_UNLOCK(vp, 0);
723 			vrele(vp);
724 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
725 			return (error);
726 		}
727 		ffs_load_inode(bp, ip, fs, ip->i_number);
728 		ip->i_effnlink = ip->i_nlink;
729 		brelse(bp);
730 		VOP_UNLOCK(vp, 0);
731 		vrele(vp);
732 	}
733 	return (0);
734 }
735 
736 /*
737  * Possible superblock locations ordered from most to least likely.
738  */
739 static int sblock_try[] = SBLOCKSEARCH;
740 
741 /*
742  * Common code for mount and mountroot
743  */
744 static int
745 ffs_mountfs(devvp, mp, td)
746 	struct vnode *devvp;
747 	struct mount *mp;
748 	struct thread *td;
749 {
750 	struct ufsmount *ump;
751 	struct buf *bp;
752 	struct fs *fs;
753 	struct cdev *dev;
754 	void *space;
755 	ufs2_daddr_t sblockloc;
756 	int error, i, blks, size, ronly;
757 	int32_t *lp;
758 	struct ucred *cred;
759 	struct g_consumer *cp;
760 	struct mount *nmp;
761 
762 	bp = NULL;
763 	ump = NULL;
764 	cred = td ? td->td_ucred : NOCRED;
765 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
766 
767 	dev = devvp->v_rdev;
768 	dev_ref(dev);
769 	DROP_GIANT();
770 	g_topology_lock();
771 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
772 	g_topology_unlock();
773 	PICKUP_GIANT();
774 	VOP_UNLOCK(devvp, 0);
775 	if (error)
776 		goto out;
777 	if (devvp->v_rdev->si_iosize_max != 0)
778 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
779 	if (mp->mnt_iosize_max > MAXPHYS)
780 		mp->mnt_iosize_max = MAXPHYS;
781 
782 	devvp->v_bufobj.bo_ops = &ffs_ops;
783 	if (devvp->v_type == VCHR)
784 		devvp->v_rdev->si_mountpt = mp;
785 
786 	fs = NULL;
787 	sblockloc = 0;
788 	/*
789 	 * Try reading the superblock in each of its possible locations.
790 	 */
791 	for (i = 0; sblock_try[i] != -1; i++) {
792 		if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
793 			error = EINVAL;
794 			vfs_mount_error(mp,
795 			    "Invalid sectorsize %d for superblock size %d",
796 			    cp->provider->sectorsize, SBLOCKSIZE);
797 			goto out;
798 		}
799 		if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
800 		    cred, &bp)) != 0)
801 			goto out;
802 		fs = (struct fs *)bp->b_data;
803 		sblockloc = sblock_try[i];
804 		if ((fs->fs_magic == FS_UFS1_MAGIC ||
805 		     (fs->fs_magic == FS_UFS2_MAGIC &&
806 		      (fs->fs_sblockloc == sblockloc ||
807 		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
808 		    fs->fs_bsize <= MAXBSIZE &&
809 		    fs->fs_bsize >= sizeof(struct fs))
810 			break;
811 		brelse(bp);
812 		bp = NULL;
813 	}
814 	if (sblock_try[i] == -1) {
815 		error = EINVAL;		/* XXX needs translation */
816 		goto out;
817 	}
818 	fs->fs_fmod = 0;
819 	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indices */
820 	fs->fs_flags &= ~FS_UNCLEAN;
821 	if (fs->fs_clean == 0) {
822 		fs->fs_flags |= FS_UNCLEAN;
823 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
824 		    ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
825 		     (fs->fs_flags & FS_DOSOFTDEP))) {
826 			printf("WARNING: %s was not properly dismounted\n",
827 			    fs->fs_fsmnt);
828 		} else {
829 			vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
830 			    fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
831 			    (fs->fs_flags & FS_SUJ) == 0 ? "" :
832 			    " Forced mount will invalidate journal contents");
833 			error = EPERM;
834 			goto out;
835 		}
836 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
837 		    (mp->mnt_flag & MNT_FORCE)) {
838 			printf("WARNING: %s: lost blocks %jd files %d\n",
839 			    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
840 			    fs->fs_pendinginodes);
841 			fs->fs_pendingblocks = 0;
842 			fs->fs_pendinginodes = 0;
843 		}
844 	}
845 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
846 		printf("WARNING: %s: mount pending error: blocks %jd "
847 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
848 		    fs->fs_pendinginodes);
849 		fs->fs_pendingblocks = 0;
850 		fs->fs_pendinginodes = 0;
851 	}
852 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
853 #ifdef UFS_GJOURNAL
854 		/*
855 		 * Get journal provider name.
856 		 */
857 		size = 1024;
858 		mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
859 		if (g_io_getattr("GJOURNAL::provider", cp, &size,
860 		    mp->mnt_gjprovider) == 0) {
861 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
862 			    M_UFSMNT, M_WAITOK);
863 			MNT_ILOCK(mp);
864 			mp->mnt_flag |= MNT_GJOURNAL;
865 			MNT_IUNLOCK(mp);
866 		} else {
867 			printf("WARNING: %s: GJOURNAL flag on fs "
868 			    "but no gjournal provider below\n",
869 			    mp->mnt_stat.f_mntonname);
870 			free(mp->mnt_gjprovider, M_UFSMNT);
871 			mp->mnt_gjprovider = NULL;
872 		}
873 #else
874 		printf("WARNING: %s: GJOURNAL flag on fs but no "
875 		    "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
876 #endif
877 	} else {
878 		mp->mnt_gjprovider = NULL;
879 	}
880 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
881 	ump->um_cp = cp;
882 	ump->um_bo = &devvp->v_bufobj;
883 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
884 	if (fs->fs_magic == FS_UFS1_MAGIC) {
885 		ump->um_fstype = UFS1;
886 		ump->um_balloc = ffs_balloc_ufs1;
887 	} else {
888 		ump->um_fstype = UFS2;
889 		ump->um_balloc = ffs_balloc_ufs2;
890 	}
891 	ump->um_blkatoff = ffs_blkatoff;
892 	ump->um_truncate = ffs_truncate;
893 	ump->um_update = ffs_update;
894 	ump->um_valloc = ffs_valloc;
895 	ump->um_vfree = ffs_vfree;
896 	ump->um_ifree = ffs_ifree;
897 	ump->um_rdonly = ffs_rdonly;
898 	ump->um_snapgone = ffs_snapgone;
899 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
900 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
901 	if (fs->fs_sbsize < SBLOCKSIZE)
902 		bp->b_flags |= B_INVAL | B_NOCACHE;
903 	brelse(bp);
904 	bp = NULL;
905 	fs = ump->um_fs;
906 	ffs_oldfscompat_read(fs, ump, sblockloc);
907 	fs->fs_ronly = ronly;
908 	size = fs->fs_cssize;
909 	blks = howmany(size, fs->fs_fsize);
910 	if (fs->fs_contigsumsize > 0)
911 		size += fs->fs_ncg * sizeof(int32_t);
912 	size += fs->fs_ncg * sizeof(u_int8_t);
913 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
914 	fs->fs_csp = space;
915 	for (i = 0; i < blks; i += fs->fs_frag) {
916 		size = fs->fs_bsize;
917 		if (i + fs->fs_frag > blks)
918 			size = (blks - i) * fs->fs_fsize;
919 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
920 		    cred, &bp)) != 0) {
921 			free(fs->fs_csp, M_UFSMNT);
922 			goto out;
923 		}
924 		bcopy(bp->b_data, space, (u_int)size);
925 		space = (char *)space + size;
926 		brelse(bp);
927 		bp = NULL;
928 	}
929 	if (fs->fs_contigsumsize > 0) {
930 		fs->fs_maxcluster = lp = space;
931 		for (i = 0; i < fs->fs_ncg; i++)
932 			*lp++ = fs->fs_contigsumsize;
933 		space = lp;
934 	}
935 	size = fs->fs_ncg * sizeof(u_int8_t);
936 	fs->fs_contigdirs = (u_int8_t *)space;
937 	bzero(fs->fs_contigdirs, size);
938 	fs->fs_active = NULL;
939 	mp->mnt_data = ump;
940 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
941 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
942 	nmp = NULL;
943 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
944 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
945 		if (nmp)
946 			vfs_rel(nmp);
947 		vfs_getnewfsid(mp);
948 	}
949 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
950 	MNT_ILOCK(mp);
951 	mp->mnt_flag |= MNT_LOCAL;
952 	MNT_IUNLOCK(mp);
953 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
954 #ifdef MAC
955 		MNT_ILOCK(mp);
956 		mp->mnt_flag |= MNT_MULTILABEL;
957 		MNT_IUNLOCK(mp);
958 #else
959 		printf("WARNING: %s: multilabel flag on fs but "
960 		    "no MAC support\n", mp->mnt_stat.f_mntonname);
961 #endif
962 	}
963 	if ((fs->fs_flags & FS_ACLS) != 0) {
964 #ifdef UFS_ACL
965 		MNT_ILOCK(mp);
966 
967 		if (mp->mnt_flag & MNT_NFS4ACLS)
968 			printf("WARNING: %s: ACLs flag on fs conflicts with "
969 			    "\"nfsv4acls\" mount option; option ignored\n",
970 			    mp->mnt_stat.f_mntonname);
971 		mp->mnt_flag &= ~MNT_NFS4ACLS;
972 		mp->mnt_flag |= MNT_ACLS;
973 
974 		MNT_IUNLOCK(mp);
975 #else
976 		printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
977 		    mp->mnt_stat.f_mntonname);
978 #endif
979 	}
980 	if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
981 #ifdef UFS_ACL
982 		MNT_ILOCK(mp);
983 
984 		if (mp->mnt_flag & MNT_ACLS)
985 			printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
986 			    "with \"acls\" mount option; option ignored\n",
987 			    mp->mnt_stat.f_mntonname);
988 		mp->mnt_flag &= ~MNT_ACLS;
989 		mp->mnt_flag |= MNT_NFS4ACLS;
990 
991 		MNT_IUNLOCK(mp);
992 #else
993 		printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
994 		    "ACLs support\n", mp->mnt_stat.f_mntonname);
995 #endif
996 	}
997 	if ((fs->fs_flags & FS_TRIM) != 0) {
998 		size = sizeof(int);
999 		if (g_io_getattr("GEOM::candelete", cp, &size,
1000 		    &ump->um_candelete) == 0) {
1001 			if (!ump->um_candelete)
1002 				printf("WARNING: %s: TRIM flag on fs but disk "
1003 				    "does not support TRIM\n",
1004 				    mp->mnt_stat.f_mntonname);
1005 		} else {
1006 			printf("WARNING: %s: TRIM flag on fs but disk does "
1007 			    "not confirm that it supports TRIM\n",
1008 			    mp->mnt_stat.f_mntonname);
1009 			ump->um_candelete = 0;
1010 		}
1011 		if (ump->um_candelete) {
1012 			ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1013 			    taskqueue_thread_enqueue, &ump->um_trim_tq);
1014 			taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1015 			    "%s trim", mp->mnt_stat.f_mntonname);
1016 		}
1017 	}
1018 
1019 	ump->um_mountp = mp;
1020 	ump->um_dev = dev;
1021 	ump->um_devvp = devvp;
1022 	ump->um_nindir = fs->fs_nindir;
1023 	ump->um_bptrtodb = fs->fs_fsbtodb;
1024 	ump->um_seqinc = fs->fs_frag;
1025 	for (i = 0; i < MAXQUOTAS; i++)
1026 		ump->um_quotas[i] = NULLVP;
1027 #ifdef UFS_EXTATTR
1028 	ufs_extattr_uepm_init(&ump->um_extattr);
1029 #endif
1030 	/*
1031 	 * Set FS local "last mounted on" information (NULL pad)
1032 	 */
1033 	bzero(fs->fs_fsmnt, MAXMNTLEN);
1034 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1035 	mp->mnt_stat.f_iosize = fs->fs_bsize;
1036 
1037 	if (mp->mnt_flag & MNT_ROOTFS) {
1038 		/*
1039 		 * Root mount; update timestamp in mount structure.
1040 		 * this will be used by the common root mount code
1041 		 * to update the system clock.
1042 		 */
1043 		mp->mnt_time = fs->fs_time;
1044 	}
1045 
1046 	if (ronly == 0) {
1047 		fs->fs_mtime = time_second;
1048 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
1049 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1050 			free(fs->fs_csp, M_UFSMNT);
1051 			ffs_flushfiles(mp, FORCECLOSE, td);
1052 			goto out;
1053 		}
1054 		if (fs->fs_snapinum[0] != 0)
1055 			ffs_snapshot_mount(mp);
1056 		fs->fs_fmod = 1;
1057 		fs->fs_clean = 0;
1058 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
1059 	}
1060 	/*
1061 	 * Initialize filesystem state information in mount struct.
1062 	 */
1063 	MNT_ILOCK(mp);
1064 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1065 	    MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1066 	MNT_IUNLOCK(mp);
1067 #ifdef UFS_EXTATTR
1068 #ifdef UFS_EXTATTR_AUTOSTART
1069 	/*
1070 	 *
1071 	 * Auto-starting does the following:
1072 	 *	- check for /.attribute in the fs, and extattr_start if so
1073 	 *	- for each file in .attribute, enable that file with
1074 	 * 	  an attribute of the same name.
1075 	 * Not clear how to report errors -- probably eat them.
1076 	 * This would all happen while the filesystem was busy/not
1077 	 * available, so would effectively be "atomic".
1078 	 */
1079 	(void) ufs_extattr_autostart(mp, td);
1080 #endif /* !UFS_EXTATTR_AUTOSTART */
1081 #endif /* !UFS_EXTATTR */
1082 	return (0);
1083 out:
1084 	if (bp)
1085 		brelse(bp);
1086 	if (devvp->v_type == VCHR && devvp->v_rdev != NULL)
1087 		devvp->v_rdev->si_mountpt = NULL;
1088 	if (cp != NULL) {
1089 		DROP_GIANT();
1090 		g_topology_lock();
1091 		g_vfs_close(cp);
1092 		g_topology_unlock();
1093 		PICKUP_GIANT();
1094 	}
1095 	if (ump) {
1096 		mtx_destroy(UFS_MTX(ump));
1097 		if (mp->mnt_gjprovider != NULL) {
1098 			free(mp->mnt_gjprovider, M_UFSMNT);
1099 			mp->mnt_gjprovider = NULL;
1100 		}
1101 		free(ump->um_fs, M_UFSMNT);
1102 		free(ump, M_UFSMNT);
1103 		mp->mnt_data = NULL;
1104 	}
1105 	dev_rel(dev);
1106 	return (error);
1107 }
1108 
1109 #include <sys/sysctl.h>
1110 static int bigcgs = 0;
1111 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1112 
1113 /*
1114  * Sanity checks for loading old filesystem superblocks.
1115  * See ffs_oldfscompat_write below for unwound actions.
1116  *
1117  * XXX - Parts get retired eventually.
1118  * Unfortunately new bits get added.
1119  */
1120 static void
1121 ffs_oldfscompat_read(fs, ump, sblockloc)
1122 	struct fs *fs;
1123 	struct ufsmount *ump;
1124 	ufs2_daddr_t sblockloc;
1125 {
1126 	off_t maxfilesize;
1127 
1128 	/*
1129 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
1130 	 */
1131 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1132 		fs->fs_flags = fs->fs_old_flags;
1133 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
1134 		fs->fs_sblockloc = sblockloc;
1135 	}
1136 	/*
1137 	 * If not yet done, update UFS1 superblock with new wider fields.
1138 	 */
1139 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1140 		fs->fs_maxbsize = fs->fs_bsize;
1141 		fs->fs_time = fs->fs_old_time;
1142 		fs->fs_size = fs->fs_old_size;
1143 		fs->fs_dsize = fs->fs_old_dsize;
1144 		fs->fs_csaddr = fs->fs_old_csaddr;
1145 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1146 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1147 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1148 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1149 	}
1150 	if (fs->fs_magic == FS_UFS1_MAGIC &&
1151 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
1152 		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1153 		fs->fs_qbmask = ~fs->fs_bmask;
1154 		fs->fs_qfmask = ~fs->fs_fmask;
1155 	}
1156 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1157 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1158 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1159 		if (fs->fs_maxfilesize > maxfilesize)
1160 			fs->fs_maxfilesize = maxfilesize;
1161 	}
1162 	/* Compatibility for old filesystems */
1163 	if (fs->fs_avgfilesize <= 0)
1164 		fs->fs_avgfilesize = AVFILESIZ;
1165 	if (fs->fs_avgfpdir <= 0)
1166 		fs->fs_avgfpdir = AFPDIR;
1167 	if (bigcgs) {
1168 		fs->fs_save_cgsize = fs->fs_cgsize;
1169 		fs->fs_cgsize = fs->fs_bsize;
1170 	}
1171 }
1172 
1173 /*
1174  * Unwinding superblock updates for old filesystems.
1175  * See ffs_oldfscompat_read above for details.
1176  *
1177  * XXX - Parts get retired eventually.
1178  * Unfortunately new bits get added.
1179  */
1180 void
1181 ffs_oldfscompat_write(fs, ump)
1182 	struct fs *fs;
1183 	struct ufsmount *ump;
1184 {
1185 
1186 	/*
1187 	 * Copy back UFS2 updated fields that UFS1 inspects.
1188 	 */
1189 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1190 		fs->fs_old_time = fs->fs_time;
1191 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1192 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1193 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1194 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1195 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1196 	}
1197 	if (bigcgs) {
1198 		fs->fs_cgsize = fs->fs_save_cgsize;
1199 		fs->fs_save_cgsize = 0;
1200 	}
1201 }
1202 
1203 /*
1204  * unmount system call
1205  */
1206 static int
1207 ffs_unmount(mp, mntflags)
1208 	struct mount *mp;
1209 	int mntflags;
1210 {
1211 	struct thread *td;
1212 	struct ufsmount *ump = VFSTOUFS(mp);
1213 	struct fs *fs;
1214 	int error, flags, susp;
1215 #ifdef UFS_EXTATTR
1216 	int e_restart;
1217 #endif
1218 
1219 	flags = 0;
1220 	td = curthread;
1221 	fs = ump->um_fs;
1222 	susp = 0;
1223 	if (mntflags & MNT_FORCE) {
1224 		flags |= FORCECLOSE;
1225 		susp = fs->fs_ronly == 0;
1226 	}
1227 #ifdef UFS_EXTATTR
1228 	if ((error = ufs_extattr_stop(mp, td))) {
1229 		if (error != EOPNOTSUPP)
1230 			printf("WARNING: unmount %s: ufs_extattr_stop "
1231 			    "returned errno %d\n", mp->mnt_stat.f_mntonname,
1232 			    error);
1233 		e_restart = 0;
1234 	} else {
1235 		ufs_extattr_uepm_destroy(&ump->um_extattr);
1236 		e_restart = 1;
1237 	}
1238 #endif
1239 	if (susp) {
1240 		error = vfs_write_suspend_umnt(mp);
1241 		if (error != 0)
1242 			goto fail1;
1243 	}
1244 	if (MOUNTEDSOFTDEP(mp))
1245 		error = softdep_flushfiles(mp, flags, td);
1246 	else
1247 		error = ffs_flushfiles(mp, flags, td);
1248 	if (error != 0 && error != ENXIO)
1249 		goto fail;
1250 
1251 	UFS_LOCK(ump);
1252 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1253 		printf("WARNING: unmount %s: pending error: blocks %jd "
1254 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1255 		    fs->fs_pendinginodes);
1256 		fs->fs_pendingblocks = 0;
1257 		fs->fs_pendinginodes = 0;
1258 	}
1259 	UFS_UNLOCK(ump);
1260 	if (MOUNTEDSOFTDEP(mp))
1261 		softdep_unmount(mp);
1262 	if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1263 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1264 		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1265 		if (error && error != ENXIO) {
1266 			fs->fs_clean = 0;
1267 			goto fail;
1268 		}
1269 	}
1270 	if (susp)
1271 		vfs_write_resume(mp, VR_START_WRITE);
1272 	if (ump->um_trim_tq != NULL) {
1273 		while (ump->um_trim_inflight != 0)
1274 			pause("ufsutr", hz);
1275 		taskqueue_drain_all(ump->um_trim_tq);
1276 		taskqueue_free(ump->um_trim_tq);
1277 	}
1278 	DROP_GIANT();
1279 	g_topology_lock();
1280 	if (ump->um_fsckpid > 0) {
1281 		/*
1282 		 * Return to normal read-only mode.
1283 		 */
1284 		error = g_access(ump->um_cp, 0, -1, 0);
1285 		ump->um_fsckpid = 0;
1286 	}
1287 	g_vfs_close(ump->um_cp);
1288 	g_topology_unlock();
1289 	PICKUP_GIANT();
1290 	if (ump->um_devvp->v_type == VCHR && ump->um_devvp->v_rdev != NULL)
1291 		ump->um_devvp->v_rdev->si_mountpt = NULL;
1292 	vrele(ump->um_devvp);
1293 	dev_rel(ump->um_dev);
1294 	mtx_destroy(UFS_MTX(ump));
1295 	if (mp->mnt_gjprovider != NULL) {
1296 		free(mp->mnt_gjprovider, M_UFSMNT);
1297 		mp->mnt_gjprovider = NULL;
1298 	}
1299 	free(fs->fs_csp, M_UFSMNT);
1300 	free(fs, M_UFSMNT);
1301 	free(ump, M_UFSMNT);
1302 	mp->mnt_data = NULL;
1303 	MNT_ILOCK(mp);
1304 	mp->mnt_flag &= ~MNT_LOCAL;
1305 	MNT_IUNLOCK(mp);
1306 	return (error);
1307 
1308 fail:
1309 	if (susp)
1310 		vfs_write_resume(mp, VR_START_WRITE);
1311 fail1:
1312 #ifdef UFS_EXTATTR
1313 	if (e_restart) {
1314 		ufs_extattr_uepm_init(&ump->um_extattr);
1315 #ifdef UFS_EXTATTR_AUTOSTART
1316 		(void) ufs_extattr_autostart(mp, td);
1317 #endif
1318 	}
1319 #endif
1320 
1321 	return (error);
1322 }
1323 
1324 /*
1325  * Flush out all the files in a filesystem.
1326  */
1327 int
1328 ffs_flushfiles(mp, flags, td)
1329 	struct mount *mp;
1330 	int flags;
1331 	struct thread *td;
1332 {
1333 	struct ufsmount *ump;
1334 	int qerror, error;
1335 
1336 	ump = VFSTOUFS(mp);
1337 	qerror = 0;
1338 #ifdef QUOTA
1339 	if (mp->mnt_flag & MNT_QUOTA) {
1340 		int i;
1341 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1342 		if (error)
1343 			return (error);
1344 		for (i = 0; i < MAXQUOTAS; i++) {
1345 			error = quotaoff(td, mp, i);
1346 			if (error != 0) {
1347 				if ((flags & EARLYFLUSH) == 0)
1348 					return (error);
1349 				else
1350 					qerror = error;
1351 			}
1352 		}
1353 
1354 		/*
1355 		 * Here we fall through to vflush again to ensure that
1356 		 * we have gotten rid of all the system vnodes, unless
1357 		 * quotas must not be closed.
1358 		 */
1359 	}
1360 #endif
1361 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1362 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1363 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1364 			return (error);
1365 		ffs_snapshot_unmount(mp);
1366 		flags |= FORCECLOSE;
1367 		/*
1368 		 * Here we fall through to vflush again to ensure
1369 		 * that we have gotten rid of all the system vnodes.
1370 		 */
1371 	}
1372 
1373 	/*
1374 	 * Do not close system files if quotas were not closed, to be
1375 	 * able to sync the remaining dquots.  The freeblks softupdate
1376 	 * workitems might hold a reference on a dquot, preventing
1377 	 * quotaoff() from completing.  Next round of
1378 	 * softdep_flushworklist() iteration should process the
1379 	 * blockers, allowing the next run of quotaoff() to finally
1380 	 * flush held dquots.
1381 	 *
1382 	 * Otherwise, flush all the files.
1383 	 */
1384 	if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1385 		return (error);
1386 
1387 	/*
1388 	 * Flush filesystem metadata.
1389 	 */
1390 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1391 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1392 	VOP_UNLOCK(ump->um_devvp, 0);
1393 	return (error);
1394 }
1395 
1396 /*
1397  * Get filesystem statistics.
1398  */
1399 static int
1400 ffs_statfs(mp, sbp)
1401 	struct mount *mp;
1402 	struct statfs *sbp;
1403 {
1404 	struct ufsmount *ump;
1405 	struct fs *fs;
1406 
1407 	ump = VFSTOUFS(mp);
1408 	fs = ump->um_fs;
1409 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1410 		panic("ffs_statfs");
1411 	sbp->f_version = STATFS_VERSION;
1412 	sbp->f_bsize = fs->fs_fsize;
1413 	sbp->f_iosize = fs->fs_bsize;
1414 	sbp->f_blocks = fs->fs_dsize;
1415 	UFS_LOCK(ump);
1416 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1417 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1418 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1419 	    dbtofsb(fs, fs->fs_pendingblocks);
1420 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1421 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1422 	UFS_UNLOCK(ump);
1423 	sbp->f_namemax = NAME_MAX;
1424 	return (0);
1425 }
1426 
1427 static bool
1428 sync_doupdate(struct inode *ip)
1429 {
1430 
1431 	return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1432 	    IN_UPDATE)) != 0);
1433 }
1434 
1435 /*
1436  * For a lazy sync, we only care about access times, quotas and the
1437  * superblock.  Other filesystem changes are already converted to
1438  * cylinder group blocks or inode blocks updates and are written to
1439  * disk by syncer.
1440  */
1441 static int
1442 ffs_sync_lazy(mp)
1443      struct mount *mp;
1444 {
1445 	struct vnode *mvp, *vp;
1446 	struct inode *ip;
1447 	struct thread *td;
1448 	int allerror, error;
1449 
1450 	allerror = 0;
1451 	td = curthread;
1452 	if ((mp->mnt_flag & MNT_NOATIME) != 0)
1453 		goto qupdate;
1454 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1455 		if (vp->v_type == VNON) {
1456 			VI_UNLOCK(vp);
1457 			continue;
1458 		}
1459 		ip = VTOI(vp);
1460 
1461 		/*
1462 		 * The IN_ACCESS flag is converted to IN_MODIFIED by
1463 		 * ufs_close() and ufs_getattr() by the calls to
1464 		 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1465 		 * Test also all the other timestamp flags too, to pick up
1466 		 * any other cases that could be missed.
1467 		 */
1468 		if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1469 			VI_UNLOCK(vp);
1470 			continue;
1471 		}
1472 		if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
1473 		    td)) != 0)
1474 			continue;
1475 		if (sync_doupdate(ip))
1476 			error = ffs_update(vp, 0);
1477 		if (error != 0)
1478 			allerror = error;
1479 		vput(vp);
1480 	}
1481 
1482 qupdate:
1483 #ifdef QUOTA
1484 	qsync(mp);
1485 #endif
1486 
1487 	if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1488 	    (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1489 		allerror = error;
1490 	return (allerror);
1491 }
1492 
1493 /*
1494  * Go through the disk queues to initiate sandbagged IO;
1495  * go through the inodes to write those that have been modified;
1496  * initiate the writing of the super block if it has been modified.
1497  *
1498  * Note: we are always called with the filesystem marked busy using
1499  * vfs_busy().
1500  */
1501 static int
1502 ffs_sync(mp, waitfor)
1503 	struct mount *mp;
1504 	int waitfor;
1505 {
1506 	struct vnode *mvp, *vp, *devvp;
1507 	struct thread *td;
1508 	struct inode *ip;
1509 	struct ufsmount *ump = VFSTOUFS(mp);
1510 	struct fs *fs;
1511 	int error, count, lockreq, allerror = 0;
1512 	int suspend;
1513 	int suspended;
1514 	int secondary_writes;
1515 	int secondary_accwrites;
1516 	int softdep_deps;
1517 	int softdep_accdeps;
1518 	struct bufobj *bo;
1519 
1520 	suspend = 0;
1521 	suspended = 0;
1522 	td = curthread;
1523 	fs = ump->um_fs;
1524 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1525 		panic("%s: ffs_sync: modification on read-only filesystem",
1526 		    fs->fs_fsmnt);
1527 	if (waitfor == MNT_LAZY) {
1528 		if (!rebooting)
1529 			return (ffs_sync_lazy(mp));
1530 		waitfor = MNT_NOWAIT;
1531 	}
1532 
1533 	/*
1534 	 * Write back each (modified) inode.
1535 	 */
1536 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1537 	if (waitfor == MNT_SUSPEND) {
1538 		suspend = 1;
1539 		waitfor = MNT_WAIT;
1540 	}
1541 	if (waitfor == MNT_WAIT)
1542 		lockreq = LK_EXCLUSIVE;
1543 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1544 loop:
1545 	/* Grab snapshot of secondary write counts */
1546 	MNT_ILOCK(mp);
1547 	secondary_writes = mp->mnt_secondary_writes;
1548 	secondary_accwrites = mp->mnt_secondary_accwrites;
1549 	MNT_IUNLOCK(mp);
1550 
1551 	/* Grab snapshot of softdep dependency counts */
1552 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1553 
1554 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1555 		/*
1556 		 * Depend on the vnode interlock to keep things stable enough
1557 		 * for a quick test.  Since there might be hundreds of
1558 		 * thousands of vnodes, we cannot afford even a subroutine
1559 		 * call unless there's a good chance that we have work to do.
1560 		 */
1561 		if (vp->v_type == VNON) {
1562 			VI_UNLOCK(vp);
1563 			continue;
1564 		}
1565 		ip = VTOI(vp);
1566 		if ((ip->i_flag &
1567 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1568 		    vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1569 			VI_UNLOCK(vp);
1570 			continue;
1571 		}
1572 		if ((error = vget(vp, lockreq, td)) != 0) {
1573 			if (error == ENOENT || error == ENOLCK) {
1574 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1575 				goto loop;
1576 			}
1577 			continue;
1578 		}
1579 		if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0)
1580 			allerror = error;
1581 		vput(vp);
1582 	}
1583 	/*
1584 	 * Force stale filesystem control information to be flushed.
1585 	 */
1586 	if (waitfor == MNT_WAIT || rebooting) {
1587 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1588 			allerror = error;
1589 		/* Flushed work items may create new vnodes to clean */
1590 		if (allerror == 0 && count)
1591 			goto loop;
1592 	}
1593 #ifdef QUOTA
1594 	qsync(mp);
1595 #endif
1596 
1597 	devvp = ump->um_devvp;
1598 	bo = &devvp->v_bufobj;
1599 	BO_LOCK(bo);
1600 	if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1601 		BO_UNLOCK(bo);
1602 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1603 		error = VOP_FSYNC(devvp, waitfor, td);
1604 		VOP_UNLOCK(devvp, 0);
1605 		if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1606 			error = ffs_sbupdate(ump, waitfor, 0);
1607 		if (error != 0)
1608 			allerror = error;
1609 		if (allerror == 0 && waitfor == MNT_WAIT)
1610 			goto loop;
1611 	} else if (suspend != 0) {
1612 		if (softdep_check_suspend(mp,
1613 					  devvp,
1614 					  softdep_deps,
1615 					  softdep_accdeps,
1616 					  secondary_writes,
1617 					  secondary_accwrites) != 0) {
1618 			MNT_IUNLOCK(mp);
1619 			goto loop;	/* More work needed */
1620 		}
1621 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1622 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1623 		MNT_IUNLOCK(mp);
1624 		suspended = 1;
1625 	} else
1626 		BO_UNLOCK(bo);
1627 	/*
1628 	 * Write back modified superblock.
1629 	 */
1630 	if (fs->fs_fmod != 0 &&
1631 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1632 		allerror = error;
1633 	return (allerror);
1634 }
1635 
1636 int
1637 ffs_vget(mp, ino, flags, vpp)
1638 	struct mount *mp;
1639 	ino_t ino;
1640 	int flags;
1641 	struct vnode **vpp;
1642 {
1643 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1644 }
1645 
1646 int
1647 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1648 	struct mount *mp;
1649 	ino_t ino;
1650 	int flags;
1651 	struct vnode **vpp;
1652 	int ffs_flags;
1653 {
1654 	struct fs *fs;
1655 	struct inode *ip;
1656 	struct ufsmount *ump;
1657 	struct buf *bp;
1658 	struct vnode *vp;
1659 	struct cdev *dev;
1660 	int error;
1661 
1662 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1663 	if (error || *vpp != NULL)
1664 		return (error);
1665 
1666 	/*
1667 	 * We must promote to an exclusive lock for vnode creation.  This
1668 	 * can happen if lookup is passed LOCKSHARED.
1669 	 */
1670 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1671 		flags &= ~LK_TYPE_MASK;
1672 		flags |= LK_EXCLUSIVE;
1673 	}
1674 
1675 	/*
1676 	 * We do not lock vnode creation as it is believed to be too
1677 	 * expensive for such rare case as simultaneous creation of vnode
1678 	 * for same ino by different processes. We just allow them to race
1679 	 * and check later to decide who wins. Let the race begin!
1680 	 */
1681 
1682 	ump = VFSTOUFS(mp);
1683 	dev = ump->um_dev;
1684 	fs = ump->um_fs;
1685 	ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1686 
1687 	/* Allocate a new vnode/inode. */
1688 	error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1689 	    &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1690 	if (error) {
1691 		*vpp = NULL;
1692 		uma_zfree(uma_inode, ip);
1693 		return (error);
1694 	}
1695 	/*
1696 	 * FFS supports recursive locking.
1697 	 */
1698 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1699 	VN_LOCK_AREC(vp);
1700 	vp->v_data = ip;
1701 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1702 	ip->i_vnode = vp;
1703 	ip->i_ump = ump;
1704 	ip->i_fs = fs;
1705 	ip->i_dev = dev;
1706 	ip->i_number = ino;
1707 	ip->i_ea_refs = 0;
1708 	ip->i_nextclustercg = -1;
1709 #ifdef QUOTA
1710 	{
1711 		int i;
1712 		for (i = 0; i < MAXQUOTAS; i++)
1713 			ip->i_dquot[i] = NODQUOT;
1714 	}
1715 #endif
1716 
1717 	if (ffs_flags & FFSV_FORCEINSMQ)
1718 		vp->v_vflag |= VV_FORCEINSMQ;
1719 	error = insmntque(vp, mp);
1720 	if (error != 0) {
1721 		uma_zfree(uma_inode, ip);
1722 		*vpp = NULL;
1723 		return (error);
1724 	}
1725 	vp->v_vflag &= ~VV_FORCEINSMQ;
1726 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1727 	if (error || *vpp != NULL)
1728 		return (error);
1729 
1730 	/* Read in the disk contents for the inode, copy into the inode. */
1731 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1732 	    (int)fs->fs_bsize, NOCRED, &bp);
1733 	if (error) {
1734 		/*
1735 		 * The inode does not contain anything useful, so it would
1736 		 * be misleading to leave it on its hash chain. With mode
1737 		 * still zero, it will be unlinked and returned to the free
1738 		 * list by vput().
1739 		 */
1740 		brelse(bp);
1741 		vput(vp);
1742 		*vpp = NULL;
1743 		return (error);
1744 	}
1745 	if (ip->i_ump->um_fstype == UFS1)
1746 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1747 	else
1748 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1749 	ffs_load_inode(bp, ip, fs, ino);
1750 	if (DOINGSOFTDEP(vp))
1751 		softdep_load_inodeblock(ip);
1752 	else
1753 		ip->i_effnlink = ip->i_nlink;
1754 	bqrelse(bp);
1755 
1756 	/*
1757 	 * Initialize the vnode from the inode, check for aliases.
1758 	 * Note that the underlying vnode may have changed.
1759 	 */
1760 	if (ip->i_ump->um_fstype == UFS1)
1761 		error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1762 	else
1763 		error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1764 	if (error) {
1765 		vput(vp);
1766 		*vpp = NULL;
1767 		return (error);
1768 	}
1769 
1770 	/*
1771 	 * Finish inode initialization.
1772 	 */
1773 	if (vp->v_type != VFIFO) {
1774 		/* FFS supports shared locking for all files except fifos. */
1775 		VN_LOCK_ASHARE(vp);
1776 	}
1777 
1778 	/*
1779 	 * Set up a generation number for this inode if it does not
1780 	 * already have one. This should only happen on old filesystems.
1781 	 */
1782 	if (ip->i_gen == 0) {
1783 		ip->i_gen = arc4random() / 2 + 1;
1784 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1785 			ip->i_flag |= IN_MODIFIED;
1786 			DIP_SET(ip, i_gen, ip->i_gen);
1787 		}
1788 	}
1789 #ifdef MAC
1790 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1791 		/*
1792 		 * If this vnode is already allocated, and we're running
1793 		 * multi-label, attempt to perform a label association
1794 		 * from the extended attributes on the inode.
1795 		 */
1796 		error = mac_vnode_associate_extattr(mp, vp);
1797 		if (error) {
1798 			/* ufs_inactive will release ip->i_devvp ref. */
1799 			vput(vp);
1800 			*vpp = NULL;
1801 			return (error);
1802 		}
1803 	}
1804 #endif
1805 
1806 	*vpp = vp;
1807 	return (0);
1808 }
1809 
1810 /*
1811  * File handle to vnode
1812  *
1813  * Have to be really careful about stale file handles:
1814  * - check that the inode number is valid
1815  * - for UFS2 check that the inode number is initialized
1816  * - call ffs_vget() to get the locked inode
1817  * - check for an unallocated inode (i_mode == 0)
1818  * - check that the given client host has export rights and return
1819  *   those rights via. exflagsp and credanonp
1820  */
1821 static int
1822 ffs_fhtovp(mp, fhp, flags, vpp)
1823 	struct mount *mp;
1824 	struct fid *fhp;
1825 	int flags;
1826 	struct vnode **vpp;
1827 {
1828 	struct ufid *ufhp;
1829 	struct ufsmount *ump;
1830 	struct fs *fs;
1831 	struct cg *cgp;
1832 	struct buf *bp;
1833 	ino_t ino;
1834 	u_int cg;
1835 	int error;
1836 
1837 	ufhp = (struct ufid *)fhp;
1838 	ino = ufhp->ufid_ino;
1839 	ump = VFSTOUFS(mp);
1840 	fs = ump->um_fs;
1841 	if (ino < ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
1842 		return (ESTALE);
1843 	/*
1844 	 * Need to check if inode is initialized because UFS2 does lazy
1845 	 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
1846 	 */
1847 	if (fs->fs_magic != FS_UFS2_MAGIC)
1848 		return (ufs_fhtovp(mp, ufhp, flags, vpp));
1849 	cg = ino_to_cg(fs, ino);
1850 	error = bread(ump->um_devvp, fsbtodb(fs, cgtod(fs, cg)),
1851 		(int)fs->fs_cgsize, NOCRED, &bp);
1852 	if (error)
1853 		return (error);
1854 	cgp = (struct cg *)bp->b_data;
1855 	if (!cg_chkmagic(cgp) || ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
1856 		brelse(bp);
1857 		return (ESTALE);
1858 	}
1859 	brelse(bp);
1860 	return (ufs_fhtovp(mp, ufhp, flags, vpp));
1861 }
1862 
1863 /*
1864  * Initialize the filesystem.
1865  */
1866 static int
1867 ffs_init(vfsp)
1868 	struct vfsconf *vfsp;
1869 {
1870 
1871 	ffs_susp_initialize();
1872 	softdep_initialize();
1873 	return (ufs_init(vfsp));
1874 }
1875 
1876 /*
1877  * Undo the work of ffs_init().
1878  */
1879 static int
1880 ffs_uninit(vfsp)
1881 	struct vfsconf *vfsp;
1882 {
1883 	int ret;
1884 
1885 	ret = ufs_uninit(vfsp);
1886 	softdep_uninitialize();
1887 	ffs_susp_uninitialize();
1888 	return (ret);
1889 }
1890 
1891 /*
1892  * Write a superblock and associated information back to disk.
1893  */
1894 int
1895 ffs_sbupdate(ump, waitfor, suspended)
1896 	struct ufsmount *ump;
1897 	int waitfor;
1898 	int suspended;
1899 {
1900 	struct fs *fs = ump->um_fs;
1901 	struct buf *sbbp;
1902 	struct buf *bp;
1903 	int blks;
1904 	void *space;
1905 	int i, size, error, allerror = 0;
1906 
1907 	if (fs->fs_ronly == 1 &&
1908 	    (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1909 	    (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
1910 		panic("ffs_sbupdate: write read-only filesystem");
1911 	/*
1912 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1913 	 */
1914 	sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
1915 	    (int)fs->fs_sbsize, 0, 0, 0);
1916 	/*
1917 	 * First write back the summary information.
1918 	 */
1919 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1920 	space = fs->fs_csp;
1921 	for (i = 0; i < blks; i += fs->fs_frag) {
1922 		size = fs->fs_bsize;
1923 		if (i + fs->fs_frag > blks)
1924 			size = (blks - i) * fs->fs_fsize;
1925 		bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1926 		    size, 0, 0, 0);
1927 		bcopy(space, bp->b_data, (u_int)size);
1928 		space = (char *)space + size;
1929 		if (suspended)
1930 			bp->b_flags |= B_VALIDSUSPWRT;
1931 		if (waitfor != MNT_WAIT)
1932 			bawrite(bp);
1933 		else if ((error = bwrite(bp)) != 0)
1934 			allerror = error;
1935 	}
1936 	/*
1937 	 * Now write back the superblock itself. If any errors occurred
1938 	 * up to this point, then fail so that the superblock avoids
1939 	 * being written out as clean.
1940 	 */
1941 	if (allerror) {
1942 		brelse(sbbp);
1943 		return (allerror);
1944 	}
1945 	bp = sbbp;
1946 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1947 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1948 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1949 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1950 		fs->fs_sblockloc = SBLOCK_UFS1;
1951 	}
1952 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1953 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1954 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1955 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1956 		fs->fs_sblockloc = SBLOCK_UFS2;
1957 	}
1958 	fs->fs_fmod = 0;
1959 	fs->fs_time = time_second;
1960 	if (MOUNTEDSOFTDEP(ump->um_mountp))
1961 		softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
1962 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1963 	ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
1964 	if (suspended)
1965 		bp->b_flags |= B_VALIDSUSPWRT;
1966 	if (waitfor != MNT_WAIT)
1967 		bawrite(bp);
1968 	else if ((error = bwrite(bp)) != 0)
1969 		allerror = error;
1970 	return (allerror);
1971 }
1972 
1973 static int
1974 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1975 	int attrnamespace, const char *attrname)
1976 {
1977 
1978 #ifdef UFS_EXTATTR
1979 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1980 	    attrname));
1981 #else
1982 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1983 	    attrname));
1984 #endif
1985 }
1986 
1987 static void
1988 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1989 {
1990 
1991 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1992 		uma_zfree(uma_ufs1, ip->i_din1);
1993 	else if (ip->i_din2 != NULL)
1994 		uma_zfree(uma_ufs2, ip->i_din2);
1995 	uma_zfree(uma_inode, ip);
1996 }
1997 
1998 static int dobkgrdwrite = 1;
1999 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
2000     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2001 
2002 /*
2003  * Complete a background write started from bwrite.
2004  */
2005 static void
2006 ffs_backgroundwritedone(struct buf *bp)
2007 {
2008 	struct bufobj *bufobj;
2009 	struct buf *origbp;
2010 
2011 	/*
2012 	 * Find the original buffer that we are writing.
2013 	 */
2014 	bufobj = bp->b_bufobj;
2015 	BO_LOCK(bufobj);
2016 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2017 		panic("backgroundwritedone: lost buffer");
2018 
2019 	/*
2020 	 * We should mark the cylinder group buffer origbp as
2021 	 * dirty, to not loose the failed write.
2022 	 */
2023 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2024 		origbp->b_vflags |= BV_BKGRDERR;
2025 	BO_UNLOCK(bufobj);
2026 	/*
2027 	 * Process dependencies then return any unfinished ones.
2028 	 */
2029 	pbrelvp(bp);
2030 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2031 		buf_complete(bp);
2032 #ifdef SOFTUPDATES
2033 	if (!LIST_EMPTY(&bp->b_dep))
2034 		softdep_move_dependencies(bp, origbp);
2035 #endif
2036 	/*
2037 	 * This buffer is marked B_NOCACHE so when it is released
2038 	 * by biodone it will be tossed.
2039 	 */
2040 	bp->b_flags |= B_NOCACHE;
2041 	bp->b_flags &= ~B_CACHE;
2042 
2043 	/*
2044 	 * Prevent brelse() from trying to keep and re-dirtying bp on
2045 	 * errors. It causes b_bufobj dereference in
2046 	 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2047 	 * pbrelvp() above.
2048 	 */
2049 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2050 		bp->b_flags |= B_INVAL;
2051 	bufdone(bp);
2052 	BO_LOCK(bufobj);
2053 	/*
2054 	 * Clear the BV_BKGRDINPROG flag in the original buffer
2055 	 * and awaken it if it is waiting for the write to complete.
2056 	 * If BV_BKGRDINPROG is not set in the original buffer it must
2057 	 * have been released and re-instantiated - which is not legal.
2058 	 */
2059 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2060 	    ("backgroundwritedone: lost buffer2"));
2061 	origbp->b_vflags &= ~BV_BKGRDINPROG;
2062 	if (origbp->b_vflags & BV_BKGRDWAIT) {
2063 		origbp->b_vflags &= ~BV_BKGRDWAIT;
2064 		wakeup(&origbp->b_xflags);
2065 	}
2066 	BO_UNLOCK(bufobj);
2067 }
2068 
2069 
2070 /*
2071  * Write, release buffer on completion.  (Done by iodone
2072  * if async).  Do not bother writing anything if the buffer
2073  * is invalid.
2074  *
2075  * Note that we set B_CACHE here, indicating that buffer is
2076  * fully valid and thus cacheable.  This is true even of NFS
2077  * now so we set it generally.  This could be set either here
2078  * or in biodone() since the I/O is synchronous.  We put it
2079  * here.
2080  */
2081 static int
2082 ffs_bufwrite(struct buf *bp)
2083 {
2084 	struct buf *newbp;
2085 
2086 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2087 	if (bp->b_flags & B_INVAL) {
2088 		brelse(bp);
2089 		return (0);
2090 	}
2091 
2092 	if (!BUF_ISLOCKED(bp))
2093 		panic("bufwrite: buffer is not busy???");
2094 	/*
2095 	 * If a background write is already in progress, delay
2096 	 * writing this block if it is asynchronous. Otherwise
2097 	 * wait for the background write to complete.
2098 	 */
2099 	BO_LOCK(bp->b_bufobj);
2100 	if (bp->b_vflags & BV_BKGRDINPROG) {
2101 		if (bp->b_flags & B_ASYNC) {
2102 			BO_UNLOCK(bp->b_bufobj);
2103 			bdwrite(bp);
2104 			return (0);
2105 		}
2106 		bp->b_vflags |= BV_BKGRDWAIT;
2107 		msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2108 		    "bwrbg", 0);
2109 		if (bp->b_vflags & BV_BKGRDINPROG)
2110 			panic("bufwrite: still writing");
2111 	}
2112 	bp->b_vflags &= ~BV_BKGRDERR;
2113 	BO_UNLOCK(bp->b_bufobj);
2114 
2115 	/*
2116 	 * If this buffer is marked for background writing and we
2117 	 * do not have to wait for it, make a copy and write the
2118 	 * copy so as to leave this buffer ready for further use.
2119 	 *
2120 	 * This optimization eats a lot of memory.  If we have a page
2121 	 * or buffer shortfall we can't do it.
2122 	 */
2123 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2124 	    (bp->b_flags & B_ASYNC) &&
2125 	    !vm_page_count_severe() &&
2126 	    !buf_dirty_count_severe()) {
2127 		KASSERT(bp->b_iodone == NULL,
2128 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2129 
2130 		/* get a new block */
2131 		newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2132 		if (newbp == NULL)
2133 			goto normal_write;
2134 
2135 		KASSERT(buf_mapped(bp), ("Unmapped cg"));
2136 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2137 		BO_LOCK(bp->b_bufobj);
2138 		bp->b_vflags |= BV_BKGRDINPROG;
2139 		BO_UNLOCK(bp->b_bufobj);
2140 		newbp->b_xflags |= BX_BKGRDMARKER;
2141 		newbp->b_lblkno = bp->b_lblkno;
2142 		newbp->b_blkno = bp->b_blkno;
2143 		newbp->b_offset = bp->b_offset;
2144 		newbp->b_iodone = ffs_backgroundwritedone;
2145 		newbp->b_flags |= B_ASYNC;
2146 		newbp->b_flags &= ~B_INVAL;
2147 		pbgetvp(bp->b_vp, newbp);
2148 
2149 #ifdef SOFTUPDATES
2150 		/*
2151 		 * Move over the dependencies.  If there are rollbacks,
2152 		 * leave the parent buffer dirtied as it will need to
2153 		 * be written again.
2154 		 */
2155 		if (LIST_EMPTY(&bp->b_dep) ||
2156 		    softdep_move_dependencies(bp, newbp) == 0)
2157 			bundirty(bp);
2158 #else
2159 		bundirty(bp);
2160 #endif
2161 
2162 		/*
2163 		 * Initiate write on the copy, release the original.  The
2164 		 * BKGRDINPROG flag prevents it from going away until
2165 		 * the background write completes.
2166 		 */
2167 		bqrelse(bp);
2168 		bp = newbp;
2169 	} else
2170 		/* Mark the buffer clean */
2171 		bundirty(bp);
2172 
2173 
2174 	/* Let the normal bufwrite do the rest for us */
2175 normal_write:
2176 	return (bufwrite(bp));
2177 }
2178 
2179 
2180 static void
2181 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2182 {
2183 	struct vnode *vp;
2184 	int error;
2185 	struct buf *tbp;
2186 	int nocopy;
2187 
2188 	vp = bo->__bo_vnode;
2189 	if (bp->b_iocmd == BIO_WRITE) {
2190 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2191 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2192 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2193 			panic("ffs_geom_strategy: bad I/O");
2194 		nocopy = bp->b_flags & B_NOCOPY;
2195 		bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2196 		if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2197 		    vp->v_rdev->si_snapdata != NULL) {
2198 			if ((bp->b_flags & B_CLUSTER) != 0) {
2199 				runningbufwakeup(bp);
2200 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2201 					      b_cluster.cluster_entry) {
2202 					error = ffs_copyonwrite(vp, tbp);
2203 					if (error != 0 &&
2204 					    error != EOPNOTSUPP) {
2205 						bp->b_error = error;
2206 						bp->b_ioflags |= BIO_ERROR;
2207 						bufdone(bp);
2208 						return;
2209 					}
2210 				}
2211 				bp->b_runningbufspace = bp->b_bufsize;
2212 				atomic_add_long(&runningbufspace,
2213 					       bp->b_runningbufspace);
2214 			} else {
2215 				error = ffs_copyonwrite(vp, bp);
2216 				if (error != 0 && error != EOPNOTSUPP) {
2217 					bp->b_error = error;
2218 					bp->b_ioflags |= BIO_ERROR;
2219 					bufdone(bp);
2220 					return;
2221 				}
2222 			}
2223 		}
2224 #ifdef SOFTUPDATES
2225 		if ((bp->b_flags & B_CLUSTER) != 0) {
2226 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2227 				      b_cluster.cluster_entry) {
2228 				if (!LIST_EMPTY(&tbp->b_dep))
2229 					buf_start(tbp);
2230 			}
2231 		} else {
2232 			if (!LIST_EMPTY(&bp->b_dep))
2233 				buf_start(bp);
2234 		}
2235 
2236 #endif
2237 	}
2238 	g_vfs_strategy(bo, bp);
2239 }
2240 
2241 int
2242 ffs_own_mount(const struct mount *mp)
2243 {
2244 
2245 	if (mp->mnt_op == &ufs_vfsops)
2246 		return (1);
2247 	return (0);
2248 }
2249 
2250 #ifdef	DDB
2251 #ifdef SOFTUPDATES
2252 
2253 /* defined in ffs_softdep.c */
2254 extern void db_print_ffs(struct ufsmount *ump);
2255 
2256 DB_SHOW_COMMAND(ffs, db_show_ffs)
2257 {
2258 	struct mount *mp;
2259 	struct ufsmount *ump;
2260 
2261 	if (have_addr) {
2262 		ump = VFSTOUFS((struct mount *)addr);
2263 		db_print_ffs(ump);
2264 		return;
2265 	}
2266 
2267 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2268 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2269 			db_print_ffs(VFSTOUFS(mp));
2270 	}
2271 }
2272 
2273 #endif	/* SOFTUPDATES */
2274 #endif	/* DDB */
2275