xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision d96700a6da2afa88607fbd7405ade439424d10d9)
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 			g_topology_lock();
243 			/*
244 			 * Return to normal read-only mode.
245 			 */
246 			error = g_access(ump->um_cp, 0, -1, 0);
247 			g_topology_unlock();
248 			ump->um_fsckpid = 0;
249 		}
250 		if (fs->fs_ronly == 0 &&
251 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
252 			/*
253 			 * Flush any dirty data and suspend filesystem.
254 			 */
255 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
256 				return (error);
257 			error = vfs_write_suspend_umnt(mp);
258 			if (error != 0)
259 				return (error);
260 			/*
261 			 * Check for and optionally get rid of files open
262 			 * for writing.
263 			 */
264 			flags = WRITECLOSE;
265 			if (mp->mnt_flag & MNT_FORCE)
266 				flags |= FORCECLOSE;
267 			if (MOUNTEDSOFTDEP(mp)) {
268 				error = softdep_flushfiles(mp, flags, td);
269 			} else {
270 				error = ffs_flushfiles(mp, flags, td);
271 			}
272 			if (error) {
273 				vfs_write_resume(mp, 0);
274 				return (error);
275 			}
276 			if (fs->fs_pendingblocks != 0 ||
277 			    fs->fs_pendinginodes != 0) {
278 				printf("WARNING: %s Update error: blocks %jd "
279 				    "files %d\n", fs->fs_fsmnt,
280 				    (intmax_t)fs->fs_pendingblocks,
281 				    fs->fs_pendinginodes);
282 				fs->fs_pendingblocks = 0;
283 				fs->fs_pendinginodes = 0;
284 			}
285 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
286 				fs->fs_clean = 1;
287 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
288 				fs->fs_ronly = 0;
289 				fs->fs_clean = 0;
290 				vfs_write_resume(mp, 0);
291 				return (error);
292 			}
293 			if (MOUNTEDSOFTDEP(mp))
294 				softdep_unmount(mp);
295 			g_topology_lock();
296 			/*
297 			 * Drop our write and exclusive access.
298 			 */
299 			g_access(ump->um_cp, 0, -1, -1);
300 			g_topology_unlock();
301 			fs->fs_ronly = 1;
302 			MNT_ILOCK(mp);
303 			mp->mnt_flag |= MNT_RDONLY;
304 			MNT_IUNLOCK(mp);
305 			/*
306 			 * Allow the writers to note that filesystem
307 			 * is ro now.
308 			 */
309 			vfs_write_resume(mp, 0);
310 		}
311 		if ((mp->mnt_flag & MNT_RELOAD) &&
312 		    (error = ffs_reload(mp, td, 0)) != 0)
313 			return (error);
314 		if (fs->fs_ronly &&
315 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
316 			/*
317 			 * If we are running a checker, do not allow upgrade.
318 			 */
319 			if (ump->um_fsckpid > 0) {
320 				vfs_mount_error(mp,
321 				    "Active checker, cannot upgrade to write");
322 				return (EINVAL);
323 			}
324 			/*
325 			 * If upgrade to read-write by non-root, then verify
326 			 * that user has necessary permissions on the device.
327 			 */
328 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
329 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
330 			    td->td_ucred, td);
331 			if (error)
332 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
333 			if (error) {
334 				VOP_UNLOCK(devvp, 0);
335 				return (error);
336 			}
337 			VOP_UNLOCK(devvp, 0);
338 			fs->fs_flags &= ~FS_UNCLEAN;
339 			if (fs->fs_clean == 0) {
340 				fs->fs_flags |= FS_UNCLEAN;
341 				if ((mp->mnt_flag & MNT_FORCE) ||
342 				    ((fs->fs_flags &
343 				     (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
344 				     (fs->fs_flags & FS_DOSOFTDEP))) {
345 					printf("WARNING: %s was not properly "
346 					   "dismounted\n", fs->fs_fsmnt);
347 				} else {
348 					vfs_mount_error(mp,
349 					   "R/W mount of %s denied. %s.%s",
350 					   fs->fs_fsmnt,
351 					   "Filesystem is not clean - run fsck",
352 					   (fs->fs_flags & FS_SUJ) == 0 ? "" :
353 					   " Forced mount will invalidate"
354 					   " journal contents");
355 					return (EPERM);
356 				}
357 			}
358 			g_topology_lock();
359 			/*
360 			 * Request exclusive write access.
361 			 */
362 			error = g_access(ump->um_cp, 0, 1, 1);
363 			g_topology_unlock();
364 			if (error)
365 				return (error);
366 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
367 				return (error);
368 			fs->fs_ronly = 0;
369 			MNT_ILOCK(mp);
370 			mp->mnt_flag &= ~MNT_RDONLY;
371 			MNT_IUNLOCK(mp);
372 			fs->fs_mtime = time_second;
373 			/* check to see if we need to start softdep */
374 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
375 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
376 				vn_finished_write(mp);
377 				return (error);
378 			}
379 			fs->fs_clean = 0;
380 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
381 				vn_finished_write(mp);
382 				return (error);
383 			}
384 			if (fs->fs_snapinum[0] != 0)
385 				ffs_snapshot_mount(mp);
386 			vn_finished_write(mp);
387 		}
388 		/*
389 		 * Soft updates is incompatible with "async",
390 		 * so if we are doing softupdates stop the user
391 		 * from setting the async flag in an update.
392 		 * Softdep_mount() clears it in an initial mount
393 		 * or ro->rw remount.
394 		 */
395 		if (MOUNTEDSOFTDEP(mp)) {
396 			/* XXX: Reset too late ? */
397 			MNT_ILOCK(mp);
398 			mp->mnt_flag &= ~MNT_ASYNC;
399 			MNT_IUNLOCK(mp);
400 		}
401 		/*
402 		 * Keep MNT_ACLS flag if it is stored in superblock.
403 		 */
404 		if ((fs->fs_flags & FS_ACLS) != 0) {
405 			/* XXX: Set too late ? */
406 			MNT_ILOCK(mp);
407 			mp->mnt_flag |= MNT_ACLS;
408 			MNT_IUNLOCK(mp);
409 		}
410 
411 		if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
412 			/* XXX: Set too late ? */
413 			MNT_ILOCK(mp);
414 			mp->mnt_flag |= MNT_NFS4ACLS;
415 			MNT_IUNLOCK(mp);
416 		}
417 		/*
418 		 * If this is a request from fsck to clean up the filesystem,
419 		 * then allow the specified pid to proceed.
420 		 */
421 		if (fsckpid > 0) {
422 			if (ump->um_fsckpid != 0) {
423 				vfs_mount_error(mp,
424 				    "Active checker already running on %s",
425 				    fs->fs_fsmnt);
426 				return (EINVAL);
427 			}
428 			KASSERT(MOUNTEDSOFTDEP(mp) == 0,
429 			    ("soft updates enabled on read-only file system"));
430 			g_topology_lock();
431 			/*
432 			 * Request write access.
433 			 */
434 			error = g_access(ump->um_cp, 0, 1, 0);
435 			g_topology_unlock();
436 			if (error) {
437 				vfs_mount_error(mp,
438 				    "Checker activation failed on %s",
439 				    fs->fs_fsmnt);
440 				return (error);
441 			}
442 			ump->um_fsckpid = fsckpid;
443 			if (fs->fs_snapinum[0] != 0)
444 				ffs_snapshot_mount(mp);
445 			fs->fs_mtime = time_second;
446 			fs->fs_fmod = 1;
447 			fs->fs_clean = 0;
448 			(void) ffs_sbupdate(ump, MNT_WAIT, 0);
449 		}
450 
451 		/*
452 		 * If this is a snapshot request, take the snapshot.
453 		 */
454 		if (mp->mnt_flag & MNT_SNAPSHOT)
455 			return (ffs_snapshot(mp, fspec));
456 	}
457 
458 	/*
459 	 * Not an update, or updating the name: look up the name
460 	 * and verify that it refers to a sensible disk device.
461 	 */
462 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
463 	if ((error = namei(&ndp)) != 0)
464 		return (error);
465 	NDFREE(&ndp, NDF_ONLY_PNBUF);
466 	devvp = ndp.ni_vp;
467 	if (!vn_isdisk(devvp, &error)) {
468 		vput(devvp);
469 		return (error);
470 	}
471 
472 	/*
473 	 * If mount by non-root, then verify that user has necessary
474 	 * permissions on the device.
475 	 */
476 	accmode = VREAD;
477 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
478 		accmode |= VWRITE;
479 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
480 	if (error)
481 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
482 	if (error) {
483 		vput(devvp);
484 		return (error);
485 	}
486 
487 	if (mp->mnt_flag & MNT_UPDATE) {
488 		/*
489 		 * Update only
490 		 *
491 		 * If it's not the same vnode, or at least the same device
492 		 * then it's not correct.
493 		 */
494 
495 		if (devvp->v_rdev != ump->um_devvp->v_rdev)
496 			error = EINVAL;	/* needs translation */
497 		vput(devvp);
498 		if (error)
499 			return (error);
500 	} else {
501 		/*
502 		 * New mount
503 		 *
504 		 * We need the name for the mount point (also used for
505 		 * "last mounted on") copied in. If an error occurs,
506 		 * the mount point is discarded by the upper level code.
507 		 * Note that vfs_mount_alloc() populates f_mntonname for us.
508 		 */
509 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
510 			vrele(devvp);
511 			return (error);
512 		}
513 		if (fsckpid > 0) {
514 			KASSERT(MOUNTEDSOFTDEP(mp) == 0,
515 			    ("soft updates enabled on read-only file system"));
516 			ump = VFSTOUFS(mp);
517 			fs = ump->um_fs;
518 			g_topology_lock();
519 			/*
520 			 * Request write access.
521 			 */
522 			error = g_access(ump->um_cp, 0, 1, 0);
523 			g_topology_unlock();
524 			if (error) {
525 				printf("WARNING: %s: Checker activation "
526 				    "failed\n", fs->fs_fsmnt);
527 			} else {
528 				ump->um_fsckpid = fsckpid;
529 				if (fs->fs_snapinum[0] != 0)
530 					ffs_snapshot_mount(mp);
531 				fs->fs_mtime = time_second;
532 				fs->fs_clean = 0;
533 				(void) ffs_sbupdate(ump, MNT_WAIT, 0);
534 			}
535 		}
536 	}
537 	vfs_mountedfrom(mp, fspec);
538 	return (0);
539 }
540 
541 /*
542  * Compatibility with old mount system call.
543  */
544 
545 static int
546 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
547 {
548 	struct ufs_args args;
549 	struct export_args exp;
550 	int error;
551 
552 	if (data == NULL)
553 		return (EINVAL);
554 	error = copyin(data, &args, sizeof args);
555 	if (error)
556 		return (error);
557 	vfs_oexport_conv(&args.export, &exp);
558 
559 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
560 	ma = mount_arg(ma, "export", &exp, sizeof(exp));
561 	error = kernel_mount(ma, flags);
562 
563 	return (error);
564 }
565 
566 /*
567  * Reload all incore data for a filesystem (used after running fsck on
568  * the root filesystem and finding things to fix). If the 'force' flag
569  * is 0, the filesystem must be mounted read-only.
570  *
571  * Things to do to update the mount:
572  *	1) invalidate all cached meta-data.
573  *	2) re-read superblock from disk.
574  *	3) re-read summary information from disk.
575  *	4) invalidate all inactive vnodes.
576  *	5) invalidate all cached file data.
577  *	6) re-read inode data for all active vnodes.
578  */
579 int
580 ffs_reload(struct mount *mp, struct thread *td, int force)
581 {
582 	struct vnode *vp, *mvp, *devvp;
583 	struct inode *ip;
584 	void *space;
585 	struct buf *bp;
586 	struct fs *fs, *newfs;
587 	struct ufsmount *ump;
588 	ufs2_daddr_t sblockloc;
589 	int i, blks, size, error;
590 	int32_t *lp;
591 
592 	ump = VFSTOUFS(mp);
593 
594 	MNT_ILOCK(mp);
595 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && force == 0) {
596 		MNT_IUNLOCK(mp);
597 		return (EINVAL);
598 	}
599 	MNT_IUNLOCK(mp);
600 
601 	/*
602 	 * Step 1: invalidate all cached meta-data.
603 	 */
604 	devvp = VFSTOUFS(mp)->um_devvp;
605 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
606 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
607 		panic("ffs_reload: dirty1");
608 	VOP_UNLOCK(devvp, 0);
609 
610 	/*
611 	 * Step 2: re-read superblock from disk.
612 	 */
613 	fs = VFSTOUFS(mp)->um_fs;
614 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
615 	    NOCRED, &bp)) != 0)
616 		return (error);
617 	newfs = (struct fs *)bp->b_data;
618 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
619 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
620 	    newfs->fs_bsize > MAXBSIZE ||
621 	    newfs->fs_bsize < sizeof(struct fs)) {
622 			brelse(bp);
623 			return (EIO);		/* XXX needs translation */
624 	}
625 	/*
626 	 * Copy pointer fields back into superblock before copying in	XXX
627 	 * new superblock. These should really be in the ufsmount.	XXX
628 	 * Note that important parameters (eg fs_ncg) are unchanged.
629 	 */
630 	newfs->fs_csp = fs->fs_csp;
631 	newfs->fs_maxcluster = fs->fs_maxcluster;
632 	newfs->fs_contigdirs = fs->fs_contigdirs;
633 	newfs->fs_active = fs->fs_active;
634 	newfs->fs_ronly = fs->fs_ronly;
635 	sblockloc = fs->fs_sblockloc;
636 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
637 	brelse(bp);
638 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
639 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
640 	UFS_LOCK(ump);
641 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
642 		printf("WARNING: %s: reload pending error: blocks %jd "
643 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
644 		    fs->fs_pendinginodes);
645 		fs->fs_pendingblocks = 0;
646 		fs->fs_pendinginodes = 0;
647 	}
648 	UFS_UNLOCK(ump);
649 
650 	/*
651 	 * Step 3: re-read summary information from disk.
652 	 */
653 	size = fs->fs_cssize;
654 	blks = howmany(size, fs->fs_fsize);
655 	if (fs->fs_contigsumsize > 0)
656 		size += fs->fs_ncg * sizeof(int32_t);
657 	size += fs->fs_ncg * sizeof(u_int8_t);
658 	free(fs->fs_csp, M_UFSMNT);
659 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
660 	fs->fs_csp = space;
661 	for (i = 0; i < blks; i += fs->fs_frag) {
662 		size = fs->fs_bsize;
663 		if (i + fs->fs_frag > blks)
664 			size = (blks - i) * fs->fs_fsize;
665 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
666 		    NOCRED, &bp);
667 		if (error)
668 			return (error);
669 		bcopy(bp->b_data, space, (u_int)size);
670 		space = (char *)space + size;
671 		brelse(bp);
672 	}
673 	/*
674 	 * We no longer know anything about clusters per cylinder group.
675 	 */
676 	if (fs->fs_contigsumsize > 0) {
677 		fs->fs_maxcluster = lp = space;
678 		for (i = 0; i < fs->fs_ncg; i++)
679 			*lp++ = fs->fs_contigsumsize;
680 		space = lp;
681 	}
682 	size = fs->fs_ncg * sizeof(u_int8_t);
683 	fs->fs_contigdirs = (u_int8_t *)space;
684 	bzero(fs->fs_contigdirs, size);
685 
686 loop:
687 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
688 		/*
689 		 * Skip syncer vnode.
690 		 */
691 		if (vp->v_type == VNON) {
692 			VI_UNLOCK(vp);
693 			continue;
694 		}
695 		/*
696 		 * Step 4: invalidate all cached file data.
697 		 */
698 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
699 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
700 			goto loop;
701 		}
702 		if (vinvalbuf(vp, 0, 0, 0))
703 			panic("ffs_reload: dirty2");
704 		/*
705 		 * Step 5: re-read inode data for all active vnodes.
706 		 */
707 		ip = VTOI(vp);
708 		error =
709 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
710 		    (int)fs->fs_bsize, NOCRED, &bp);
711 		if (error) {
712 			VOP_UNLOCK(vp, 0);
713 			vrele(vp);
714 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
715 			return (error);
716 		}
717 		ffs_load_inode(bp, ip, fs, ip->i_number);
718 		ip->i_effnlink = ip->i_nlink;
719 		brelse(bp);
720 		VOP_UNLOCK(vp, 0);
721 		vrele(vp);
722 	}
723 	return (0);
724 }
725 
726 /*
727  * Possible superblock locations ordered from most to least likely.
728  */
729 static int sblock_try[] = SBLOCKSEARCH;
730 
731 /*
732  * Common code for mount and mountroot
733  */
734 static int
735 ffs_mountfs(devvp, mp, td)
736 	struct vnode *devvp;
737 	struct mount *mp;
738 	struct thread *td;
739 {
740 	struct ufsmount *ump;
741 	struct buf *bp;
742 	struct fs *fs;
743 	struct cdev *dev;
744 	void *space;
745 	ufs2_daddr_t sblockloc;
746 	int error, i, blks, size, ronly;
747 	int32_t *lp;
748 	struct ucred *cred;
749 	struct g_consumer *cp;
750 	struct mount *nmp;
751 
752 	bp = NULL;
753 	ump = NULL;
754 	cred = td ? td->td_ucred : NOCRED;
755 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
756 
757 	KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
758 	dev = devvp->v_rdev;
759 	if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
760 	    (uintptr_t)mp) == 0) {
761 		VOP_UNLOCK(devvp, 0);
762 		return (EBUSY);
763 	}
764 	g_topology_lock();
765 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
766 	g_topology_unlock();
767 	if (error != 0) {
768 		atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
769 		VOP_UNLOCK(devvp, 0);
770 		return (error);
771 	}
772 	dev_ref(dev);
773 	devvp->v_bufobj.bo_ops = &ffs_ops;
774 	VOP_UNLOCK(devvp, 0);
775 	if (dev->si_iosize_max != 0)
776 		mp->mnt_iosize_max = dev->si_iosize_max;
777 	if (mp->mnt_iosize_max > MAXPHYS)
778 		mp->mnt_iosize_max = MAXPHYS;
779 
780 	fs = NULL;
781 	sblockloc = 0;
782 	/*
783 	 * Try reading the superblock in each of its possible locations.
784 	 */
785 	for (i = 0; sblock_try[i] != -1; i++) {
786 		if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
787 			error = EINVAL;
788 			vfs_mount_error(mp,
789 			    "Invalid sectorsize %d for superblock size %d",
790 			    cp->provider->sectorsize, SBLOCKSIZE);
791 			goto out;
792 		}
793 		if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
794 		    cred, &bp)) != 0)
795 			goto out;
796 		fs = (struct fs *)bp->b_data;
797 		sblockloc = sblock_try[i];
798 		if ((fs->fs_magic == FS_UFS1_MAGIC ||
799 		     (fs->fs_magic == FS_UFS2_MAGIC &&
800 		      (fs->fs_sblockloc == sblockloc ||
801 		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
802 		    fs->fs_bsize <= MAXBSIZE &&
803 		    fs->fs_bsize >= sizeof(struct fs))
804 			break;
805 		brelse(bp);
806 		bp = NULL;
807 	}
808 	if (sblock_try[i] == -1) {
809 		error = EINVAL;		/* XXX needs translation */
810 		goto out;
811 	}
812 	fs->fs_fmod = 0;
813 	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indices */
814 	fs->fs_flags &= ~FS_UNCLEAN;
815 	if (fs->fs_clean == 0) {
816 		fs->fs_flags |= FS_UNCLEAN;
817 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
818 		    ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
819 		     (fs->fs_flags & FS_DOSOFTDEP))) {
820 			printf("WARNING: %s was not properly dismounted\n",
821 			    fs->fs_fsmnt);
822 		} else {
823 			vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
824 			    fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
825 			    (fs->fs_flags & FS_SUJ) == 0 ? "" :
826 			    " Forced mount will invalidate journal contents");
827 			error = EPERM;
828 			goto out;
829 		}
830 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
831 		    (mp->mnt_flag & MNT_FORCE)) {
832 			printf("WARNING: %s: lost blocks %jd files %d\n",
833 			    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
834 			    fs->fs_pendinginodes);
835 			fs->fs_pendingblocks = 0;
836 			fs->fs_pendinginodes = 0;
837 		}
838 	}
839 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
840 		printf("WARNING: %s: mount pending error: blocks %jd "
841 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
842 		    fs->fs_pendinginodes);
843 		fs->fs_pendingblocks = 0;
844 		fs->fs_pendinginodes = 0;
845 	}
846 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
847 #ifdef UFS_GJOURNAL
848 		/*
849 		 * Get journal provider name.
850 		 */
851 		size = 1024;
852 		mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
853 		if (g_io_getattr("GJOURNAL::provider", cp, &size,
854 		    mp->mnt_gjprovider) == 0) {
855 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
856 			    M_UFSMNT, M_WAITOK);
857 			MNT_ILOCK(mp);
858 			mp->mnt_flag |= MNT_GJOURNAL;
859 			MNT_IUNLOCK(mp);
860 		} else {
861 			printf("WARNING: %s: GJOURNAL flag on fs "
862 			    "but no gjournal provider below\n",
863 			    mp->mnt_stat.f_mntonname);
864 			free(mp->mnt_gjprovider, M_UFSMNT);
865 			mp->mnt_gjprovider = NULL;
866 		}
867 #else
868 		printf("WARNING: %s: GJOURNAL flag on fs but no "
869 		    "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
870 #endif
871 	} else {
872 		mp->mnt_gjprovider = NULL;
873 	}
874 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
875 	ump->um_cp = cp;
876 	ump->um_bo = &devvp->v_bufobj;
877 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
878 	if (fs->fs_magic == FS_UFS1_MAGIC) {
879 		ump->um_fstype = UFS1;
880 		ump->um_balloc = ffs_balloc_ufs1;
881 	} else {
882 		ump->um_fstype = UFS2;
883 		ump->um_balloc = ffs_balloc_ufs2;
884 	}
885 	ump->um_blkatoff = ffs_blkatoff;
886 	ump->um_truncate = ffs_truncate;
887 	ump->um_update = ffs_update;
888 	ump->um_valloc = ffs_valloc;
889 	ump->um_vfree = ffs_vfree;
890 	ump->um_ifree = ffs_ifree;
891 	ump->um_rdonly = ffs_rdonly;
892 	ump->um_snapgone = ffs_snapgone;
893 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
894 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
895 	if (fs->fs_sbsize < SBLOCKSIZE)
896 		bp->b_flags |= B_INVAL | B_NOCACHE;
897 	brelse(bp);
898 	bp = NULL;
899 	fs = ump->um_fs;
900 	ffs_oldfscompat_read(fs, ump, sblockloc);
901 	fs->fs_ronly = ronly;
902 	size = fs->fs_cssize;
903 	blks = howmany(size, fs->fs_fsize);
904 	if (fs->fs_contigsumsize > 0)
905 		size += fs->fs_ncg * sizeof(int32_t);
906 	size += fs->fs_ncg * sizeof(u_int8_t);
907 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
908 	fs->fs_csp = space;
909 	for (i = 0; i < blks; i += fs->fs_frag) {
910 		size = fs->fs_bsize;
911 		if (i + fs->fs_frag > blks)
912 			size = (blks - i) * fs->fs_fsize;
913 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
914 		    cred, &bp)) != 0) {
915 			free(fs->fs_csp, M_UFSMNT);
916 			goto out;
917 		}
918 		bcopy(bp->b_data, space, (u_int)size);
919 		space = (char *)space + size;
920 		brelse(bp);
921 		bp = NULL;
922 	}
923 	if (fs->fs_contigsumsize > 0) {
924 		fs->fs_maxcluster = lp = space;
925 		for (i = 0; i < fs->fs_ncg; i++)
926 			*lp++ = fs->fs_contigsumsize;
927 		space = lp;
928 	}
929 	size = fs->fs_ncg * sizeof(u_int8_t);
930 	fs->fs_contigdirs = (u_int8_t *)space;
931 	bzero(fs->fs_contigdirs, size);
932 	fs->fs_active = NULL;
933 	mp->mnt_data = ump;
934 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
935 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
936 	nmp = NULL;
937 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
938 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
939 		if (nmp)
940 			vfs_rel(nmp);
941 		vfs_getnewfsid(mp);
942 	}
943 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
944 	MNT_ILOCK(mp);
945 	mp->mnt_flag |= MNT_LOCAL;
946 	MNT_IUNLOCK(mp);
947 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
948 #ifdef MAC
949 		MNT_ILOCK(mp);
950 		mp->mnt_flag |= MNT_MULTILABEL;
951 		MNT_IUNLOCK(mp);
952 #else
953 		printf("WARNING: %s: multilabel flag on fs but "
954 		    "no MAC support\n", mp->mnt_stat.f_mntonname);
955 #endif
956 	}
957 	if ((fs->fs_flags & FS_ACLS) != 0) {
958 #ifdef UFS_ACL
959 		MNT_ILOCK(mp);
960 
961 		if (mp->mnt_flag & MNT_NFS4ACLS)
962 			printf("WARNING: %s: ACLs flag on fs conflicts with "
963 			    "\"nfsv4acls\" mount option; option ignored\n",
964 			    mp->mnt_stat.f_mntonname);
965 		mp->mnt_flag &= ~MNT_NFS4ACLS;
966 		mp->mnt_flag |= MNT_ACLS;
967 
968 		MNT_IUNLOCK(mp);
969 #else
970 		printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
971 		    mp->mnt_stat.f_mntonname);
972 #endif
973 	}
974 	if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
975 #ifdef UFS_ACL
976 		MNT_ILOCK(mp);
977 
978 		if (mp->mnt_flag & MNT_ACLS)
979 			printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
980 			    "with \"acls\" mount option; option ignored\n",
981 			    mp->mnt_stat.f_mntonname);
982 		mp->mnt_flag &= ~MNT_ACLS;
983 		mp->mnt_flag |= MNT_NFS4ACLS;
984 
985 		MNT_IUNLOCK(mp);
986 #else
987 		printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
988 		    "ACLs support\n", mp->mnt_stat.f_mntonname);
989 #endif
990 	}
991 	if ((fs->fs_flags & FS_TRIM) != 0) {
992 		size = sizeof(int);
993 		if (g_io_getattr("GEOM::candelete", cp, &size,
994 		    &ump->um_candelete) == 0) {
995 			if (!ump->um_candelete)
996 				printf("WARNING: %s: TRIM flag on fs but disk "
997 				    "does not support TRIM\n",
998 				    mp->mnt_stat.f_mntonname);
999 		} else {
1000 			printf("WARNING: %s: TRIM flag on fs but disk does "
1001 			    "not confirm that it supports TRIM\n",
1002 			    mp->mnt_stat.f_mntonname);
1003 			ump->um_candelete = 0;
1004 		}
1005 		if (ump->um_candelete) {
1006 			ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1007 			    taskqueue_thread_enqueue, &ump->um_trim_tq);
1008 			taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1009 			    "%s trim", mp->mnt_stat.f_mntonname);
1010 		}
1011 	}
1012 
1013 	ump->um_mountp = mp;
1014 	ump->um_dev = dev;
1015 	ump->um_devvp = devvp;
1016 	ump->um_nindir = fs->fs_nindir;
1017 	ump->um_bptrtodb = fs->fs_fsbtodb;
1018 	ump->um_seqinc = fs->fs_frag;
1019 	for (i = 0; i < MAXQUOTAS; i++)
1020 		ump->um_quotas[i] = NULLVP;
1021 #ifdef UFS_EXTATTR
1022 	ufs_extattr_uepm_init(&ump->um_extattr);
1023 #endif
1024 	/*
1025 	 * Set FS local "last mounted on" information (NULL pad)
1026 	 */
1027 	bzero(fs->fs_fsmnt, MAXMNTLEN);
1028 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1029 	mp->mnt_stat.f_iosize = fs->fs_bsize;
1030 
1031 	if (mp->mnt_flag & MNT_ROOTFS) {
1032 		/*
1033 		 * Root mount; update timestamp in mount structure.
1034 		 * this will be used by the common root mount code
1035 		 * to update the system clock.
1036 		 */
1037 		mp->mnt_time = fs->fs_time;
1038 	}
1039 
1040 	if (ronly == 0) {
1041 		fs->fs_mtime = time_second;
1042 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
1043 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1044 			free(fs->fs_csp, M_UFSMNT);
1045 			ffs_flushfiles(mp, FORCECLOSE, td);
1046 			goto out;
1047 		}
1048 		if (fs->fs_snapinum[0] != 0)
1049 			ffs_snapshot_mount(mp);
1050 		fs->fs_fmod = 1;
1051 		fs->fs_clean = 0;
1052 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
1053 	}
1054 	/*
1055 	 * Initialize filesystem state information in mount struct.
1056 	 */
1057 	MNT_ILOCK(mp);
1058 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1059 	    MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1060 	MNT_IUNLOCK(mp);
1061 #ifdef UFS_EXTATTR
1062 #ifdef UFS_EXTATTR_AUTOSTART
1063 	/*
1064 	 *
1065 	 * Auto-starting does the following:
1066 	 *	- check for /.attribute in the fs, and extattr_start if so
1067 	 *	- for each file in .attribute, enable that file with
1068 	 * 	  an attribute of the same name.
1069 	 * Not clear how to report errors -- probably eat them.
1070 	 * This would all happen while the filesystem was busy/not
1071 	 * available, so would effectively be "atomic".
1072 	 */
1073 	(void) ufs_extattr_autostart(mp, td);
1074 #endif /* !UFS_EXTATTR_AUTOSTART */
1075 #endif /* !UFS_EXTATTR */
1076 	return (0);
1077 out:
1078 	if (bp)
1079 		brelse(bp);
1080 	if (cp != NULL) {
1081 		g_topology_lock();
1082 		g_vfs_close(cp);
1083 		g_topology_unlock();
1084 	}
1085 	if (ump) {
1086 		mtx_destroy(UFS_MTX(ump));
1087 		if (mp->mnt_gjprovider != NULL) {
1088 			free(mp->mnt_gjprovider, M_UFSMNT);
1089 			mp->mnt_gjprovider = NULL;
1090 		}
1091 		free(ump->um_fs, M_UFSMNT);
1092 		free(ump, M_UFSMNT);
1093 		mp->mnt_data = NULL;
1094 	}
1095 	atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1096 	dev_rel(dev);
1097 	return (error);
1098 }
1099 
1100 #include <sys/sysctl.h>
1101 static int bigcgs = 0;
1102 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1103 
1104 /*
1105  * Sanity checks for loading old filesystem superblocks.
1106  * See ffs_oldfscompat_write below for unwound actions.
1107  *
1108  * XXX - Parts get retired eventually.
1109  * Unfortunately new bits get added.
1110  */
1111 static void
1112 ffs_oldfscompat_read(fs, ump, sblockloc)
1113 	struct fs *fs;
1114 	struct ufsmount *ump;
1115 	ufs2_daddr_t sblockloc;
1116 {
1117 	off_t maxfilesize;
1118 
1119 	/*
1120 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
1121 	 */
1122 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1123 		fs->fs_flags = fs->fs_old_flags;
1124 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
1125 		fs->fs_sblockloc = sblockloc;
1126 	}
1127 	/*
1128 	 * If not yet done, update UFS1 superblock with new wider fields.
1129 	 */
1130 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1131 		fs->fs_maxbsize = fs->fs_bsize;
1132 		fs->fs_time = fs->fs_old_time;
1133 		fs->fs_size = fs->fs_old_size;
1134 		fs->fs_dsize = fs->fs_old_dsize;
1135 		fs->fs_csaddr = fs->fs_old_csaddr;
1136 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1137 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1138 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1139 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1140 	}
1141 	if (fs->fs_magic == FS_UFS1_MAGIC &&
1142 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
1143 		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1144 		fs->fs_qbmask = ~fs->fs_bmask;
1145 		fs->fs_qfmask = ~fs->fs_fmask;
1146 	}
1147 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1148 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1149 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1150 		if (fs->fs_maxfilesize > maxfilesize)
1151 			fs->fs_maxfilesize = maxfilesize;
1152 	}
1153 	/* Compatibility for old filesystems */
1154 	if (fs->fs_avgfilesize <= 0)
1155 		fs->fs_avgfilesize = AVFILESIZ;
1156 	if (fs->fs_avgfpdir <= 0)
1157 		fs->fs_avgfpdir = AFPDIR;
1158 	if (bigcgs) {
1159 		fs->fs_save_cgsize = fs->fs_cgsize;
1160 		fs->fs_cgsize = fs->fs_bsize;
1161 	}
1162 }
1163 
1164 /*
1165  * Unwinding superblock updates for old filesystems.
1166  * See ffs_oldfscompat_read above for details.
1167  *
1168  * XXX - Parts get retired eventually.
1169  * Unfortunately new bits get added.
1170  */
1171 void
1172 ffs_oldfscompat_write(fs, ump)
1173 	struct fs *fs;
1174 	struct ufsmount *ump;
1175 {
1176 
1177 	/*
1178 	 * Copy back UFS2 updated fields that UFS1 inspects.
1179 	 */
1180 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1181 		fs->fs_old_time = fs->fs_time;
1182 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1183 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1184 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1185 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1186 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1187 	}
1188 	if (bigcgs) {
1189 		fs->fs_cgsize = fs->fs_save_cgsize;
1190 		fs->fs_save_cgsize = 0;
1191 	}
1192 }
1193 
1194 /*
1195  * unmount system call
1196  */
1197 static int
1198 ffs_unmount(mp, mntflags)
1199 	struct mount *mp;
1200 	int mntflags;
1201 {
1202 	struct thread *td;
1203 	struct ufsmount *ump = VFSTOUFS(mp);
1204 	struct fs *fs;
1205 	int error, flags, susp;
1206 #ifdef UFS_EXTATTR
1207 	int e_restart;
1208 #endif
1209 
1210 	flags = 0;
1211 	td = curthread;
1212 	fs = ump->um_fs;
1213 	susp = 0;
1214 	if (mntflags & MNT_FORCE) {
1215 		flags |= FORCECLOSE;
1216 		susp = fs->fs_ronly == 0;
1217 	}
1218 #ifdef UFS_EXTATTR
1219 	if ((error = ufs_extattr_stop(mp, td))) {
1220 		if (error != EOPNOTSUPP)
1221 			printf("WARNING: unmount %s: ufs_extattr_stop "
1222 			    "returned errno %d\n", mp->mnt_stat.f_mntonname,
1223 			    error);
1224 		e_restart = 0;
1225 	} else {
1226 		ufs_extattr_uepm_destroy(&ump->um_extattr);
1227 		e_restart = 1;
1228 	}
1229 #endif
1230 	if (susp) {
1231 		error = vfs_write_suspend_umnt(mp);
1232 		if (error != 0)
1233 			goto fail1;
1234 	}
1235 	if (MOUNTEDSOFTDEP(mp))
1236 		error = softdep_flushfiles(mp, flags, td);
1237 	else
1238 		error = ffs_flushfiles(mp, flags, td);
1239 	if (error != 0 && error != ENXIO)
1240 		goto fail;
1241 
1242 	UFS_LOCK(ump);
1243 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1244 		printf("WARNING: unmount %s: pending error: blocks %jd "
1245 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1246 		    fs->fs_pendinginodes);
1247 		fs->fs_pendingblocks = 0;
1248 		fs->fs_pendinginodes = 0;
1249 	}
1250 	UFS_UNLOCK(ump);
1251 	if (MOUNTEDSOFTDEP(mp))
1252 		softdep_unmount(mp);
1253 	if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1254 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1255 		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1256 		if (error && error != ENXIO) {
1257 			fs->fs_clean = 0;
1258 			goto fail;
1259 		}
1260 	}
1261 	if (susp)
1262 		vfs_write_resume(mp, VR_START_WRITE);
1263 	if (ump->um_trim_tq != NULL) {
1264 		while (ump->um_trim_inflight != 0)
1265 			pause("ufsutr", hz);
1266 		taskqueue_drain_all(ump->um_trim_tq);
1267 		taskqueue_free(ump->um_trim_tq);
1268 	}
1269 	g_topology_lock();
1270 	if (ump->um_fsckpid > 0) {
1271 		/*
1272 		 * Return to normal read-only mode.
1273 		 */
1274 		error = g_access(ump->um_cp, 0, -1, 0);
1275 		ump->um_fsckpid = 0;
1276 	}
1277 	g_vfs_close(ump->um_cp);
1278 	g_topology_unlock();
1279 	atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1280 	vrele(ump->um_devvp);
1281 	dev_rel(ump->um_dev);
1282 	mtx_destroy(UFS_MTX(ump));
1283 	if (mp->mnt_gjprovider != NULL) {
1284 		free(mp->mnt_gjprovider, M_UFSMNT);
1285 		mp->mnt_gjprovider = NULL;
1286 	}
1287 	free(fs->fs_csp, M_UFSMNT);
1288 	free(fs, M_UFSMNT);
1289 	free(ump, M_UFSMNT);
1290 	mp->mnt_data = NULL;
1291 	MNT_ILOCK(mp);
1292 	mp->mnt_flag &= ~MNT_LOCAL;
1293 	MNT_IUNLOCK(mp);
1294 	return (error);
1295 
1296 fail:
1297 	if (susp)
1298 		vfs_write_resume(mp, VR_START_WRITE);
1299 fail1:
1300 #ifdef UFS_EXTATTR
1301 	if (e_restart) {
1302 		ufs_extattr_uepm_init(&ump->um_extattr);
1303 #ifdef UFS_EXTATTR_AUTOSTART
1304 		(void) ufs_extattr_autostart(mp, td);
1305 #endif
1306 	}
1307 #endif
1308 
1309 	return (error);
1310 }
1311 
1312 /*
1313  * Flush out all the files in a filesystem.
1314  */
1315 int
1316 ffs_flushfiles(mp, flags, td)
1317 	struct mount *mp;
1318 	int flags;
1319 	struct thread *td;
1320 {
1321 	struct ufsmount *ump;
1322 	int qerror, error;
1323 
1324 	ump = VFSTOUFS(mp);
1325 	qerror = 0;
1326 #ifdef QUOTA
1327 	if (mp->mnt_flag & MNT_QUOTA) {
1328 		int i;
1329 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1330 		if (error)
1331 			return (error);
1332 		for (i = 0; i < MAXQUOTAS; i++) {
1333 			error = quotaoff(td, mp, i);
1334 			if (error != 0) {
1335 				if ((flags & EARLYFLUSH) == 0)
1336 					return (error);
1337 				else
1338 					qerror = error;
1339 			}
1340 		}
1341 
1342 		/*
1343 		 * Here we fall through to vflush again to ensure that
1344 		 * we have gotten rid of all the system vnodes, unless
1345 		 * quotas must not be closed.
1346 		 */
1347 	}
1348 #endif
1349 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1350 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1351 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1352 			return (error);
1353 		ffs_snapshot_unmount(mp);
1354 		flags |= FORCECLOSE;
1355 		/*
1356 		 * Here we fall through to vflush again to ensure
1357 		 * that we have gotten rid of all the system vnodes.
1358 		 */
1359 	}
1360 
1361 	/*
1362 	 * Do not close system files if quotas were not closed, to be
1363 	 * able to sync the remaining dquots.  The freeblks softupdate
1364 	 * workitems might hold a reference on a dquot, preventing
1365 	 * quotaoff() from completing.  Next round of
1366 	 * softdep_flushworklist() iteration should process the
1367 	 * blockers, allowing the next run of quotaoff() to finally
1368 	 * flush held dquots.
1369 	 *
1370 	 * Otherwise, flush all the files.
1371 	 */
1372 	if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1373 		return (error);
1374 
1375 	/*
1376 	 * Flush filesystem metadata.
1377 	 */
1378 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1379 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1380 	VOP_UNLOCK(ump->um_devvp, 0);
1381 	return (error);
1382 }
1383 
1384 /*
1385  * Get filesystem statistics.
1386  */
1387 static int
1388 ffs_statfs(mp, sbp)
1389 	struct mount *mp;
1390 	struct statfs *sbp;
1391 {
1392 	struct ufsmount *ump;
1393 	struct fs *fs;
1394 
1395 	ump = VFSTOUFS(mp);
1396 	fs = ump->um_fs;
1397 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1398 		panic("ffs_statfs");
1399 	sbp->f_version = STATFS_VERSION;
1400 	sbp->f_bsize = fs->fs_fsize;
1401 	sbp->f_iosize = fs->fs_bsize;
1402 	sbp->f_blocks = fs->fs_dsize;
1403 	UFS_LOCK(ump);
1404 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1405 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1406 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1407 	    dbtofsb(fs, fs->fs_pendingblocks);
1408 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1409 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1410 	UFS_UNLOCK(ump);
1411 	sbp->f_namemax = NAME_MAX;
1412 	return (0);
1413 }
1414 
1415 static bool
1416 sync_doupdate(struct inode *ip)
1417 {
1418 
1419 	return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1420 	    IN_UPDATE)) != 0);
1421 }
1422 
1423 /*
1424  * For a lazy sync, we only care about access times, quotas and the
1425  * superblock.  Other filesystem changes are already converted to
1426  * cylinder group blocks or inode blocks updates and are written to
1427  * disk by syncer.
1428  */
1429 static int
1430 ffs_sync_lazy(mp)
1431      struct mount *mp;
1432 {
1433 	struct vnode *mvp, *vp;
1434 	struct inode *ip;
1435 	struct thread *td;
1436 	int allerror, error;
1437 
1438 	allerror = 0;
1439 	td = curthread;
1440 	if ((mp->mnt_flag & MNT_NOATIME) != 0)
1441 		goto qupdate;
1442 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1443 		if (vp->v_type == VNON) {
1444 			VI_UNLOCK(vp);
1445 			continue;
1446 		}
1447 		ip = VTOI(vp);
1448 
1449 		/*
1450 		 * The IN_ACCESS flag is converted to IN_MODIFIED by
1451 		 * ufs_close() and ufs_getattr() by the calls to
1452 		 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1453 		 * Test also all the other timestamp flags too, to pick up
1454 		 * any other cases that could be missed.
1455 		 */
1456 		if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1457 			VI_UNLOCK(vp);
1458 			continue;
1459 		}
1460 		if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
1461 		    td)) != 0)
1462 			continue;
1463 		if (sync_doupdate(ip))
1464 			error = ffs_update(vp, 0);
1465 		if (error != 0)
1466 			allerror = error;
1467 		vput(vp);
1468 	}
1469 
1470 qupdate:
1471 #ifdef QUOTA
1472 	qsync(mp);
1473 #endif
1474 
1475 	if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1476 	    (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1477 		allerror = error;
1478 	return (allerror);
1479 }
1480 
1481 /*
1482  * Go through the disk queues to initiate sandbagged IO;
1483  * go through the inodes to write those that have been modified;
1484  * initiate the writing of the super block if it has been modified.
1485  *
1486  * Note: we are always called with the filesystem marked busy using
1487  * vfs_busy().
1488  */
1489 static int
1490 ffs_sync(mp, waitfor)
1491 	struct mount *mp;
1492 	int waitfor;
1493 {
1494 	struct vnode *mvp, *vp, *devvp;
1495 	struct thread *td;
1496 	struct inode *ip;
1497 	struct ufsmount *ump = VFSTOUFS(mp);
1498 	struct fs *fs;
1499 	int error, count, lockreq, allerror = 0;
1500 	int suspend;
1501 	int suspended;
1502 	int secondary_writes;
1503 	int secondary_accwrites;
1504 	int softdep_deps;
1505 	int softdep_accdeps;
1506 	struct bufobj *bo;
1507 
1508 	suspend = 0;
1509 	suspended = 0;
1510 	td = curthread;
1511 	fs = ump->um_fs;
1512 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1513 		panic("%s: ffs_sync: modification on read-only filesystem",
1514 		    fs->fs_fsmnt);
1515 	if (waitfor == MNT_LAZY) {
1516 		if (!rebooting)
1517 			return (ffs_sync_lazy(mp));
1518 		waitfor = MNT_NOWAIT;
1519 	}
1520 
1521 	/*
1522 	 * Write back each (modified) inode.
1523 	 */
1524 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1525 	if (waitfor == MNT_SUSPEND) {
1526 		suspend = 1;
1527 		waitfor = MNT_WAIT;
1528 	}
1529 	if (waitfor == MNT_WAIT)
1530 		lockreq = LK_EXCLUSIVE;
1531 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1532 loop:
1533 	/* Grab snapshot of secondary write counts */
1534 	MNT_ILOCK(mp);
1535 	secondary_writes = mp->mnt_secondary_writes;
1536 	secondary_accwrites = mp->mnt_secondary_accwrites;
1537 	MNT_IUNLOCK(mp);
1538 
1539 	/* Grab snapshot of softdep dependency counts */
1540 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1541 
1542 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1543 		/*
1544 		 * Depend on the vnode interlock to keep things stable enough
1545 		 * for a quick test.  Since there might be hundreds of
1546 		 * thousands of vnodes, we cannot afford even a subroutine
1547 		 * call unless there's a good chance that we have work to do.
1548 		 */
1549 		if (vp->v_type == VNON) {
1550 			VI_UNLOCK(vp);
1551 			continue;
1552 		}
1553 		ip = VTOI(vp);
1554 		if ((ip->i_flag &
1555 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1556 		    vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1557 			VI_UNLOCK(vp);
1558 			continue;
1559 		}
1560 		if ((error = vget(vp, lockreq, td)) != 0) {
1561 			if (error == ENOENT || error == ENOLCK) {
1562 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1563 				goto loop;
1564 			}
1565 			continue;
1566 		}
1567 		if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0)
1568 			allerror = error;
1569 		vput(vp);
1570 	}
1571 	/*
1572 	 * Force stale filesystem control information to be flushed.
1573 	 */
1574 	if (waitfor == MNT_WAIT || rebooting) {
1575 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1576 			allerror = error;
1577 		/* Flushed work items may create new vnodes to clean */
1578 		if (allerror == 0 && count)
1579 			goto loop;
1580 	}
1581 #ifdef QUOTA
1582 	qsync(mp);
1583 #endif
1584 
1585 	devvp = ump->um_devvp;
1586 	bo = &devvp->v_bufobj;
1587 	BO_LOCK(bo);
1588 	if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1589 		BO_UNLOCK(bo);
1590 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1591 		error = VOP_FSYNC(devvp, waitfor, td);
1592 		VOP_UNLOCK(devvp, 0);
1593 		if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1594 			error = ffs_sbupdate(ump, waitfor, 0);
1595 		if (error != 0)
1596 			allerror = error;
1597 		if (allerror == 0 && waitfor == MNT_WAIT)
1598 			goto loop;
1599 	} else if (suspend != 0) {
1600 		if (softdep_check_suspend(mp,
1601 					  devvp,
1602 					  softdep_deps,
1603 					  softdep_accdeps,
1604 					  secondary_writes,
1605 					  secondary_accwrites) != 0) {
1606 			MNT_IUNLOCK(mp);
1607 			goto loop;	/* More work needed */
1608 		}
1609 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1610 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1611 		MNT_IUNLOCK(mp);
1612 		suspended = 1;
1613 	} else
1614 		BO_UNLOCK(bo);
1615 	/*
1616 	 * Write back modified superblock.
1617 	 */
1618 	if (fs->fs_fmod != 0 &&
1619 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1620 		allerror = error;
1621 	return (allerror);
1622 }
1623 
1624 int
1625 ffs_vget(mp, ino, flags, vpp)
1626 	struct mount *mp;
1627 	ino_t ino;
1628 	int flags;
1629 	struct vnode **vpp;
1630 {
1631 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1632 }
1633 
1634 int
1635 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1636 	struct mount *mp;
1637 	ino_t ino;
1638 	int flags;
1639 	struct vnode **vpp;
1640 	int ffs_flags;
1641 {
1642 	struct fs *fs;
1643 	struct inode *ip;
1644 	struct ufsmount *ump;
1645 	struct buf *bp;
1646 	struct vnode *vp;
1647 	struct cdev *dev;
1648 	int error;
1649 
1650 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1651 	if (error || *vpp != NULL)
1652 		return (error);
1653 
1654 	/*
1655 	 * We must promote to an exclusive lock for vnode creation.  This
1656 	 * can happen if lookup is passed LOCKSHARED.
1657 	 */
1658 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1659 		flags &= ~LK_TYPE_MASK;
1660 		flags |= LK_EXCLUSIVE;
1661 	}
1662 
1663 	/*
1664 	 * We do not lock vnode creation as it is believed to be too
1665 	 * expensive for such rare case as simultaneous creation of vnode
1666 	 * for same ino by different processes. We just allow them to race
1667 	 * and check later to decide who wins. Let the race begin!
1668 	 */
1669 
1670 	ump = VFSTOUFS(mp);
1671 	dev = ump->um_dev;
1672 	fs = ump->um_fs;
1673 	ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1674 
1675 	/* Allocate a new vnode/inode. */
1676 	error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1677 	    &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1678 	if (error) {
1679 		*vpp = NULL;
1680 		uma_zfree(uma_inode, ip);
1681 		return (error);
1682 	}
1683 	/*
1684 	 * FFS supports recursive locking.
1685 	 */
1686 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1687 	VN_LOCK_AREC(vp);
1688 	vp->v_data = ip;
1689 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1690 	ip->i_vnode = vp;
1691 	ip->i_ump = ump;
1692 	ip->i_fs = fs;
1693 	ip->i_dev = dev;
1694 	ip->i_number = ino;
1695 	ip->i_ea_refs = 0;
1696 	ip->i_nextclustercg = -1;
1697 #ifdef QUOTA
1698 	{
1699 		int i;
1700 		for (i = 0; i < MAXQUOTAS; i++)
1701 			ip->i_dquot[i] = NODQUOT;
1702 	}
1703 #endif
1704 
1705 	if (ffs_flags & FFSV_FORCEINSMQ)
1706 		vp->v_vflag |= VV_FORCEINSMQ;
1707 	error = insmntque(vp, mp);
1708 	if (error != 0) {
1709 		uma_zfree(uma_inode, ip);
1710 		*vpp = NULL;
1711 		return (error);
1712 	}
1713 	vp->v_vflag &= ~VV_FORCEINSMQ;
1714 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1715 	if (error || *vpp != NULL)
1716 		return (error);
1717 
1718 	/* Read in the disk contents for the inode, copy into the inode. */
1719 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1720 	    (int)fs->fs_bsize, NOCRED, &bp);
1721 	if (error) {
1722 		/*
1723 		 * The inode does not contain anything useful, so it would
1724 		 * be misleading to leave it on its hash chain. With mode
1725 		 * still zero, it will be unlinked and returned to the free
1726 		 * list by vput().
1727 		 */
1728 		brelse(bp);
1729 		vput(vp);
1730 		*vpp = NULL;
1731 		return (error);
1732 	}
1733 	if (ip->i_ump->um_fstype == UFS1)
1734 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1735 	else
1736 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1737 	ffs_load_inode(bp, ip, fs, ino);
1738 	if (DOINGSOFTDEP(vp))
1739 		softdep_load_inodeblock(ip);
1740 	else
1741 		ip->i_effnlink = ip->i_nlink;
1742 	bqrelse(bp);
1743 
1744 	/*
1745 	 * Initialize the vnode from the inode, check for aliases.
1746 	 * Note that the underlying vnode may have changed.
1747 	 */
1748 	if (ip->i_ump->um_fstype == UFS1)
1749 		error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1750 	else
1751 		error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1752 	if (error) {
1753 		vput(vp);
1754 		*vpp = NULL;
1755 		return (error);
1756 	}
1757 
1758 	/*
1759 	 * Finish inode initialization.
1760 	 */
1761 	if (vp->v_type != VFIFO) {
1762 		/* FFS supports shared locking for all files except fifos. */
1763 		VN_LOCK_ASHARE(vp);
1764 	}
1765 
1766 	/*
1767 	 * Set up a generation number for this inode if it does not
1768 	 * already have one. This should only happen on old filesystems.
1769 	 */
1770 	if (ip->i_gen == 0) {
1771 		while (ip->i_gen == 0)
1772 			ip->i_gen = arc4random();
1773 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1774 			ip->i_flag |= IN_MODIFIED;
1775 			DIP_SET(ip, i_gen, ip->i_gen);
1776 		}
1777 	}
1778 #ifdef MAC
1779 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1780 		/*
1781 		 * If this vnode is already allocated, and we're running
1782 		 * multi-label, attempt to perform a label association
1783 		 * from the extended attributes on the inode.
1784 		 */
1785 		error = mac_vnode_associate_extattr(mp, vp);
1786 		if (error) {
1787 			/* ufs_inactive will release ip->i_devvp ref. */
1788 			vput(vp);
1789 			*vpp = NULL;
1790 			return (error);
1791 		}
1792 	}
1793 #endif
1794 
1795 	*vpp = vp;
1796 	return (0);
1797 }
1798 
1799 /*
1800  * File handle to vnode
1801  *
1802  * Have to be really careful about stale file handles:
1803  * - check that the inode number is valid
1804  * - for UFS2 check that the inode number is initialized
1805  * - call ffs_vget() to get the locked inode
1806  * - check for an unallocated inode (i_mode == 0)
1807  * - check that the given client host has export rights and return
1808  *   those rights via. exflagsp and credanonp
1809  */
1810 static int
1811 ffs_fhtovp(mp, fhp, flags, vpp)
1812 	struct mount *mp;
1813 	struct fid *fhp;
1814 	int flags;
1815 	struct vnode **vpp;
1816 {
1817 	struct ufid *ufhp;
1818 	struct ufsmount *ump;
1819 	struct fs *fs;
1820 	struct cg *cgp;
1821 	struct buf *bp;
1822 	ino_t ino;
1823 	u_int cg;
1824 	int error;
1825 
1826 	ufhp = (struct ufid *)fhp;
1827 	ino = ufhp->ufid_ino;
1828 	ump = VFSTOUFS(mp);
1829 	fs = ump->um_fs;
1830 	if (ino < ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
1831 		return (ESTALE);
1832 	/*
1833 	 * Need to check if inode is initialized because UFS2 does lazy
1834 	 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
1835 	 */
1836 	if (fs->fs_magic != FS_UFS2_MAGIC)
1837 		return (ufs_fhtovp(mp, ufhp, flags, vpp));
1838 	cg = ino_to_cg(fs, ino);
1839 	error = bread(ump->um_devvp, fsbtodb(fs, cgtod(fs, cg)),
1840 		(int)fs->fs_cgsize, NOCRED, &bp);
1841 	if (error)
1842 		return (error);
1843 	cgp = (struct cg *)bp->b_data;
1844 	if (!cg_chkmagic(cgp) || ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
1845 		brelse(bp);
1846 		return (ESTALE);
1847 	}
1848 	brelse(bp);
1849 	return (ufs_fhtovp(mp, ufhp, flags, vpp));
1850 }
1851 
1852 /*
1853  * Initialize the filesystem.
1854  */
1855 static int
1856 ffs_init(vfsp)
1857 	struct vfsconf *vfsp;
1858 {
1859 
1860 	ffs_susp_initialize();
1861 	softdep_initialize();
1862 	return (ufs_init(vfsp));
1863 }
1864 
1865 /*
1866  * Undo the work of ffs_init().
1867  */
1868 static int
1869 ffs_uninit(vfsp)
1870 	struct vfsconf *vfsp;
1871 {
1872 	int ret;
1873 
1874 	ret = ufs_uninit(vfsp);
1875 	softdep_uninitialize();
1876 	ffs_susp_uninitialize();
1877 	return (ret);
1878 }
1879 
1880 /*
1881  * Write a superblock and associated information back to disk.
1882  */
1883 int
1884 ffs_sbupdate(ump, waitfor, suspended)
1885 	struct ufsmount *ump;
1886 	int waitfor;
1887 	int suspended;
1888 {
1889 	struct fs *fs = ump->um_fs;
1890 	struct buf *sbbp;
1891 	struct buf *bp;
1892 	int blks;
1893 	void *space;
1894 	int i, size, error, allerror = 0;
1895 
1896 	if (fs->fs_ronly == 1 &&
1897 	    (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1898 	    (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
1899 		panic("ffs_sbupdate: write read-only filesystem");
1900 	/*
1901 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1902 	 */
1903 	sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
1904 	    (int)fs->fs_sbsize, 0, 0, 0);
1905 	/*
1906 	 * First write back the summary information.
1907 	 */
1908 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1909 	space = fs->fs_csp;
1910 	for (i = 0; i < blks; i += fs->fs_frag) {
1911 		size = fs->fs_bsize;
1912 		if (i + fs->fs_frag > blks)
1913 			size = (blks - i) * fs->fs_fsize;
1914 		bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1915 		    size, 0, 0, 0);
1916 		bcopy(space, bp->b_data, (u_int)size);
1917 		space = (char *)space + size;
1918 		if (suspended)
1919 			bp->b_flags |= B_VALIDSUSPWRT;
1920 		if (waitfor != MNT_WAIT)
1921 			bawrite(bp);
1922 		else if ((error = bwrite(bp)) != 0)
1923 			allerror = error;
1924 	}
1925 	/*
1926 	 * Now write back the superblock itself. If any errors occurred
1927 	 * up to this point, then fail so that the superblock avoids
1928 	 * being written out as clean.
1929 	 */
1930 	if (allerror) {
1931 		brelse(sbbp);
1932 		return (allerror);
1933 	}
1934 	bp = sbbp;
1935 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1936 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1937 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1938 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1939 		fs->fs_sblockloc = SBLOCK_UFS1;
1940 	}
1941 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1942 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1943 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1944 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1945 		fs->fs_sblockloc = SBLOCK_UFS2;
1946 	}
1947 	fs->fs_fmod = 0;
1948 	fs->fs_time = time_second;
1949 	if (MOUNTEDSOFTDEP(ump->um_mountp))
1950 		softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
1951 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1952 	ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
1953 	if (suspended)
1954 		bp->b_flags |= B_VALIDSUSPWRT;
1955 	if (waitfor != MNT_WAIT)
1956 		bawrite(bp);
1957 	else if ((error = bwrite(bp)) != 0)
1958 		allerror = error;
1959 	return (allerror);
1960 }
1961 
1962 static int
1963 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1964 	int attrnamespace, const char *attrname)
1965 {
1966 
1967 #ifdef UFS_EXTATTR
1968 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1969 	    attrname));
1970 #else
1971 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1972 	    attrname));
1973 #endif
1974 }
1975 
1976 static void
1977 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1978 {
1979 
1980 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1981 		uma_zfree(uma_ufs1, ip->i_din1);
1982 	else if (ip->i_din2 != NULL)
1983 		uma_zfree(uma_ufs2, ip->i_din2);
1984 	uma_zfree(uma_inode, ip);
1985 }
1986 
1987 static int dobkgrdwrite = 1;
1988 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1989     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1990 
1991 /*
1992  * Complete a background write started from bwrite.
1993  */
1994 static void
1995 ffs_backgroundwritedone(struct buf *bp)
1996 {
1997 	struct bufobj *bufobj;
1998 	struct buf *origbp;
1999 
2000 	/*
2001 	 * Find the original buffer that we are writing.
2002 	 */
2003 	bufobj = bp->b_bufobj;
2004 	BO_LOCK(bufobj);
2005 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2006 		panic("backgroundwritedone: lost buffer");
2007 
2008 	/*
2009 	 * We should mark the cylinder group buffer origbp as
2010 	 * dirty, to not loose the failed write.
2011 	 */
2012 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2013 		origbp->b_vflags |= BV_BKGRDERR;
2014 	BO_UNLOCK(bufobj);
2015 	/*
2016 	 * Process dependencies then return any unfinished ones.
2017 	 */
2018 	pbrelvp(bp);
2019 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2020 		buf_complete(bp);
2021 #ifdef SOFTUPDATES
2022 	if (!LIST_EMPTY(&bp->b_dep))
2023 		softdep_move_dependencies(bp, origbp);
2024 #endif
2025 	/*
2026 	 * This buffer is marked B_NOCACHE so when it is released
2027 	 * by biodone it will be tossed.
2028 	 */
2029 	bp->b_flags |= B_NOCACHE;
2030 	bp->b_flags &= ~B_CACHE;
2031 
2032 	/*
2033 	 * Prevent brelse() from trying to keep and re-dirtying bp on
2034 	 * errors. It causes b_bufobj dereference in
2035 	 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2036 	 * pbrelvp() above.
2037 	 */
2038 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2039 		bp->b_flags |= B_INVAL;
2040 	bufdone(bp);
2041 	BO_LOCK(bufobj);
2042 	/*
2043 	 * Clear the BV_BKGRDINPROG flag in the original buffer
2044 	 * and awaken it if it is waiting for the write to complete.
2045 	 * If BV_BKGRDINPROG is not set in the original buffer it must
2046 	 * have been released and re-instantiated - which is not legal.
2047 	 */
2048 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2049 	    ("backgroundwritedone: lost buffer2"));
2050 	origbp->b_vflags &= ~BV_BKGRDINPROG;
2051 	if (origbp->b_vflags & BV_BKGRDWAIT) {
2052 		origbp->b_vflags &= ~BV_BKGRDWAIT;
2053 		wakeup(&origbp->b_xflags);
2054 	}
2055 	BO_UNLOCK(bufobj);
2056 }
2057 
2058 
2059 /*
2060  * Write, release buffer on completion.  (Done by iodone
2061  * if async).  Do not bother writing anything if the buffer
2062  * is invalid.
2063  *
2064  * Note that we set B_CACHE here, indicating that buffer is
2065  * fully valid and thus cacheable.  This is true even of NFS
2066  * now so we set it generally.  This could be set either here
2067  * or in biodone() since the I/O is synchronous.  We put it
2068  * here.
2069  */
2070 static int
2071 ffs_bufwrite(struct buf *bp)
2072 {
2073 	struct buf *newbp;
2074 
2075 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2076 	if (bp->b_flags & B_INVAL) {
2077 		brelse(bp);
2078 		return (0);
2079 	}
2080 
2081 	if (!BUF_ISLOCKED(bp))
2082 		panic("bufwrite: buffer is not busy???");
2083 	/*
2084 	 * If a background write is already in progress, delay
2085 	 * writing this block if it is asynchronous. Otherwise
2086 	 * wait for the background write to complete.
2087 	 */
2088 	BO_LOCK(bp->b_bufobj);
2089 	if (bp->b_vflags & BV_BKGRDINPROG) {
2090 		if (bp->b_flags & B_ASYNC) {
2091 			BO_UNLOCK(bp->b_bufobj);
2092 			bdwrite(bp);
2093 			return (0);
2094 		}
2095 		bp->b_vflags |= BV_BKGRDWAIT;
2096 		msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2097 		    "bwrbg", 0);
2098 		if (bp->b_vflags & BV_BKGRDINPROG)
2099 			panic("bufwrite: still writing");
2100 	}
2101 	bp->b_vflags &= ~BV_BKGRDERR;
2102 	BO_UNLOCK(bp->b_bufobj);
2103 
2104 	/*
2105 	 * If this buffer is marked for background writing and we
2106 	 * do not have to wait for it, make a copy and write the
2107 	 * copy so as to leave this buffer ready for further use.
2108 	 *
2109 	 * This optimization eats a lot of memory.  If we have a page
2110 	 * or buffer shortfall we can't do it.
2111 	 */
2112 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2113 	    (bp->b_flags & B_ASYNC) &&
2114 	    !vm_page_count_severe() &&
2115 	    !buf_dirty_count_severe()) {
2116 		KASSERT(bp->b_iodone == NULL,
2117 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2118 
2119 		/* get a new block */
2120 		newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2121 		if (newbp == NULL)
2122 			goto normal_write;
2123 
2124 		KASSERT(buf_mapped(bp), ("Unmapped cg"));
2125 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2126 		BO_LOCK(bp->b_bufobj);
2127 		bp->b_vflags |= BV_BKGRDINPROG;
2128 		BO_UNLOCK(bp->b_bufobj);
2129 		newbp->b_xflags |= BX_BKGRDMARKER;
2130 		newbp->b_lblkno = bp->b_lblkno;
2131 		newbp->b_blkno = bp->b_blkno;
2132 		newbp->b_offset = bp->b_offset;
2133 		newbp->b_iodone = ffs_backgroundwritedone;
2134 		newbp->b_flags |= B_ASYNC;
2135 		newbp->b_flags &= ~B_INVAL;
2136 		pbgetvp(bp->b_vp, newbp);
2137 
2138 #ifdef SOFTUPDATES
2139 		/*
2140 		 * Move over the dependencies.  If there are rollbacks,
2141 		 * leave the parent buffer dirtied as it will need to
2142 		 * be written again.
2143 		 */
2144 		if (LIST_EMPTY(&bp->b_dep) ||
2145 		    softdep_move_dependencies(bp, newbp) == 0)
2146 			bundirty(bp);
2147 #else
2148 		bundirty(bp);
2149 #endif
2150 
2151 		/*
2152 		 * Initiate write on the copy, release the original.  The
2153 		 * BKGRDINPROG flag prevents it from going away until
2154 		 * the background write completes.
2155 		 */
2156 		bqrelse(bp);
2157 		bp = newbp;
2158 	} else
2159 		/* Mark the buffer clean */
2160 		bundirty(bp);
2161 
2162 
2163 	/* Let the normal bufwrite do the rest for us */
2164 normal_write:
2165 	return (bufwrite(bp));
2166 }
2167 
2168 
2169 static void
2170 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2171 {
2172 	struct vnode *vp;
2173 	int error;
2174 	struct buf *tbp;
2175 	int nocopy;
2176 
2177 	vp = bo->__bo_vnode;
2178 	if (bp->b_iocmd == BIO_WRITE) {
2179 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2180 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2181 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2182 			panic("ffs_geom_strategy: bad I/O");
2183 		nocopy = bp->b_flags & B_NOCOPY;
2184 		bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2185 		if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2186 		    vp->v_rdev->si_snapdata != NULL) {
2187 			if ((bp->b_flags & B_CLUSTER) != 0) {
2188 				runningbufwakeup(bp);
2189 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2190 					      b_cluster.cluster_entry) {
2191 					error = ffs_copyonwrite(vp, tbp);
2192 					if (error != 0 &&
2193 					    error != EOPNOTSUPP) {
2194 						bp->b_error = error;
2195 						bp->b_ioflags |= BIO_ERROR;
2196 						bufdone(bp);
2197 						return;
2198 					}
2199 				}
2200 				bp->b_runningbufspace = bp->b_bufsize;
2201 				atomic_add_long(&runningbufspace,
2202 					       bp->b_runningbufspace);
2203 			} else {
2204 				error = ffs_copyonwrite(vp, bp);
2205 				if (error != 0 && error != EOPNOTSUPP) {
2206 					bp->b_error = error;
2207 					bp->b_ioflags |= BIO_ERROR;
2208 					bufdone(bp);
2209 					return;
2210 				}
2211 			}
2212 		}
2213 #ifdef SOFTUPDATES
2214 		if ((bp->b_flags & B_CLUSTER) != 0) {
2215 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2216 				      b_cluster.cluster_entry) {
2217 				if (!LIST_EMPTY(&tbp->b_dep))
2218 					buf_start(tbp);
2219 			}
2220 		} else {
2221 			if (!LIST_EMPTY(&bp->b_dep))
2222 				buf_start(bp);
2223 		}
2224 
2225 #endif
2226 	}
2227 	g_vfs_strategy(bo, bp);
2228 }
2229 
2230 int
2231 ffs_own_mount(const struct mount *mp)
2232 {
2233 
2234 	if (mp->mnt_op == &ufs_vfsops)
2235 		return (1);
2236 	return (0);
2237 }
2238 
2239 #ifdef	DDB
2240 #ifdef SOFTUPDATES
2241 
2242 /* defined in ffs_softdep.c */
2243 extern void db_print_ffs(struct ufsmount *ump);
2244 
2245 DB_SHOW_COMMAND(ffs, db_show_ffs)
2246 {
2247 	struct mount *mp;
2248 	struct ufsmount *ump;
2249 
2250 	if (have_addr) {
2251 		ump = VFSTOUFS((struct mount *)addr);
2252 		db_print_ffs(ump);
2253 		return;
2254 	}
2255 
2256 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2257 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2258 			db_print_ffs(VFSTOUFS(mp));
2259 	}
2260 }
2261 
2262 #endif	/* SOFTUPDATES */
2263 #endif	/* DDB */
2264