xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision f699532fb5b2c7b2445404b62f87bcd436195e5a)
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) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary
577  *	   writers, if requested.
578  *	6) invalidate all cached file data.
579  *	7) re-read inode data for all active vnodes.
580  */
581 int
582 ffs_reload(struct mount *mp, struct thread *td, int flags)
583 {
584 	struct vnode *vp, *mvp, *devvp;
585 	struct inode *ip;
586 	void *space;
587 	struct buf *bp;
588 	struct fs *fs, *newfs;
589 	struct ufsmount *ump;
590 	ufs2_daddr_t sblockloc;
591 	int i, blks, error;
592 	u_long size;
593 	int32_t *lp;
594 
595 	ump = VFSTOUFS(mp);
596 
597 	MNT_ILOCK(mp);
598 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) {
599 		MNT_IUNLOCK(mp);
600 		return (EINVAL);
601 	}
602 	MNT_IUNLOCK(mp);
603 
604 	/*
605 	 * Step 1: invalidate all cached meta-data.
606 	 */
607 	devvp = VFSTOUFS(mp)->um_devvp;
608 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
609 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
610 		panic("ffs_reload: dirty1");
611 	VOP_UNLOCK(devvp, 0);
612 
613 	/*
614 	 * Step 2: re-read superblock from disk.
615 	 */
616 	fs = VFSTOUFS(mp)->um_fs;
617 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
618 	    NOCRED, &bp)) != 0)
619 		return (error);
620 	newfs = (struct fs *)bp->b_data;
621 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
622 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
623 	    newfs->fs_bsize > MAXBSIZE ||
624 	    newfs->fs_bsize < sizeof(struct fs)) {
625 			brelse(bp);
626 			return (EIO);		/* XXX needs translation */
627 	}
628 	/*
629 	 * Copy pointer fields back into superblock before copying in	XXX
630 	 * new superblock. These should really be in the ufsmount.	XXX
631 	 * Note that important parameters (eg fs_ncg) are unchanged.
632 	 */
633 	newfs->fs_csp = fs->fs_csp;
634 	newfs->fs_maxcluster = fs->fs_maxcluster;
635 	newfs->fs_contigdirs = fs->fs_contigdirs;
636 	newfs->fs_active = fs->fs_active;
637 	newfs->fs_ronly = fs->fs_ronly;
638 	sblockloc = fs->fs_sblockloc;
639 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
640 	brelse(bp);
641 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
642 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
643 	UFS_LOCK(ump);
644 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
645 		printf("WARNING: %s: reload pending error: blocks %jd "
646 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
647 		    fs->fs_pendinginodes);
648 		fs->fs_pendingblocks = 0;
649 		fs->fs_pendinginodes = 0;
650 	}
651 	UFS_UNLOCK(ump);
652 
653 	/*
654 	 * Step 3: re-read summary information from disk.
655 	 */
656 	size = fs->fs_cssize;
657 	blks = howmany(size, fs->fs_fsize);
658 	if (fs->fs_contigsumsize > 0)
659 		size += fs->fs_ncg * sizeof(int32_t);
660 	size += fs->fs_ncg * sizeof(u_int8_t);
661 	free(fs->fs_csp, M_UFSMNT);
662 	space = malloc(size, M_UFSMNT, M_WAITOK);
663 	fs->fs_csp = space;
664 	for (i = 0; i < blks; i += fs->fs_frag) {
665 		size = fs->fs_bsize;
666 		if (i + fs->fs_frag > blks)
667 			size = (blks - i) * fs->fs_fsize;
668 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
669 		    NOCRED, &bp);
670 		if (error)
671 			return (error);
672 		bcopy(bp->b_data, space, (u_int)size);
673 		space = (char *)space + size;
674 		brelse(bp);
675 	}
676 	/*
677 	 * We no longer know anything about clusters per cylinder group.
678 	 */
679 	if (fs->fs_contigsumsize > 0) {
680 		fs->fs_maxcluster = lp = space;
681 		for (i = 0; i < fs->fs_ncg; i++)
682 			*lp++ = fs->fs_contigsumsize;
683 		space = lp;
684 	}
685 	size = fs->fs_ncg * sizeof(u_int8_t);
686 	fs->fs_contigdirs = (u_int8_t *)space;
687 	bzero(fs->fs_contigdirs, size);
688 	if ((flags & FFSR_UNSUSPEND) != 0) {
689 		MNT_ILOCK(mp);
690 		mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
691 		wakeup(&mp->mnt_flag);
692 		MNT_IUNLOCK(mp);
693 	}
694 
695 loop:
696 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
697 		/*
698 		 * Skip syncer vnode.
699 		 */
700 		if (vp->v_type == VNON) {
701 			VI_UNLOCK(vp);
702 			continue;
703 		}
704 		/*
705 		 * Step 4: invalidate all cached file data.
706 		 */
707 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
708 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
709 			goto loop;
710 		}
711 		if (vinvalbuf(vp, 0, 0, 0))
712 			panic("ffs_reload: dirty2");
713 		/*
714 		 * Step 5: re-read inode data for all active vnodes.
715 		 */
716 		ip = VTOI(vp);
717 		error =
718 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
719 		    (int)fs->fs_bsize, NOCRED, &bp);
720 		if (error) {
721 			VOP_UNLOCK(vp, 0);
722 			vrele(vp);
723 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
724 			return (error);
725 		}
726 		ffs_load_inode(bp, ip, fs, ip->i_number);
727 		ip->i_effnlink = ip->i_nlink;
728 		brelse(bp);
729 		VOP_UNLOCK(vp, 0);
730 		vrele(vp);
731 	}
732 	return (0);
733 }
734 
735 /*
736  * Possible superblock locations ordered from most to least likely.
737  */
738 static int sblock_try[] = SBLOCKSEARCH;
739 
740 /*
741  * Common code for mount and mountroot
742  */
743 static int
744 ffs_mountfs(devvp, mp, td)
745 	struct vnode *devvp;
746 	struct mount *mp;
747 	struct thread *td;
748 {
749 	struct ufsmount *ump;
750 	struct buf *bp;
751 	struct fs *fs;
752 	struct cdev *dev;
753 	void *space;
754 	ufs2_daddr_t sblockloc;
755 	int error, i, blks, len, ronly;
756 	u_long size;
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 	KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
768 	dev = devvp->v_rdev;
769 	if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
770 	    (uintptr_t)mp) == 0) {
771 		VOP_UNLOCK(devvp, 0);
772 		return (EBUSY);
773 	}
774 	g_topology_lock();
775 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
776 	g_topology_unlock();
777 	if (error != 0) {
778 		atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
779 		VOP_UNLOCK(devvp, 0);
780 		return (error);
781 	}
782 	dev_ref(dev);
783 	devvp->v_bufobj.bo_ops = &ffs_ops;
784 	VOP_UNLOCK(devvp, 0);
785 	if (dev->si_iosize_max != 0)
786 		mp->mnt_iosize_max = dev->si_iosize_max;
787 	if (mp->mnt_iosize_max > MAXPHYS)
788 		mp->mnt_iosize_max = MAXPHYS;
789 
790 	fs = NULL;
791 	sblockloc = 0;
792 	/*
793 	 * Try reading the superblock in each of its possible locations.
794 	 */
795 	for (i = 0; sblock_try[i] != -1; i++) {
796 		if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
797 			error = EINVAL;
798 			vfs_mount_error(mp,
799 			    "Invalid sectorsize %d for superblock size %d",
800 			    cp->provider->sectorsize, SBLOCKSIZE);
801 			goto out;
802 		}
803 		if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
804 		    cred, &bp)) != 0)
805 			goto out;
806 		fs = (struct fs *)bp->b_data;
807 		sblockloc = sblock_try[i];
808 		if ((fs->fs_magic == FS_UFS1_MAGIC ||
809 		     (fs->fs_magic == FS_UFS2_MAGIC &&
810 		      (fs->fs_sblockloc == sblockloc ||
811 		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
812 		    fs->fs_bsize <= MAXBSIZE &&
813 		    fs->fs_bsize >= sizeof(struct fs))
814 			break;
815 		brelse(bp);
816 		bp = NULL;
817 	}
818 	if (sblock_try[i] == -1) {
819 		error = EINVAL;		/* XXX needs translation */
820 		goto out;
821 	}
822 	fs->fs_fmod = 0;
823 	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indices */
824 	fs->fs_flags &= ~FS_UNCLEAN;
825 	if (fs->fs_clean == 0) {
826 		fs->fs_flags |= FS_UNCLEAN;
827 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
828 		    ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
829 		     (fs->fs_flags & FS_DOSOFTDEP))) {
830 			printf("WARNING: %s was not properly dismounted\n",
831 			    fs->fs_fsmnt);
832 		} else {
833 			vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
834 			    fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
835 			    (fs->fs_flags & FS_SUJ) == 0 ? "" :
836 			    " Forced mount will invalidate journal contents");
837 			error = EPERM;
838 			goto out;
839 		}
840 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
841 		    (mp->mnt_flag & MNT_FORCE)) {
842 			printf("WARNING: %s: lost blocks %jd files %d\n",
843 			    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
844 			    fs->fs_pendinginodes);
845 			fs->fs_pendingblocks = 0;
846 			fs->fs_pendinginodes = 0;
847 		}
848 	}
849 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
850 		printf("WARNING: %s: mount pending error: blocks %jd "
851 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
852 		    fs->fs_pendinginodes);
853 		fs->fs_pendingblocks = 0;
854 		fs->fs_pendinginodes = 0;
855 	}
856 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
857 #ifdef UFS_GJOURNAL
858 		/*
859 		 * Get journal provider name.
860 		 */
861 		len = 1024;
862 		mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK);
863 		if (g_io_getattr("GJOURNAL::provider", cp, &len,
864 		    mp->mnt_gjprovider) == 0) {
865 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len,
866 			    M_UFSMNT, M_WAITOK);
867 			MNT_ILOCK(mp);
868 			mp->mnt_flag |= MNT_GJOURNAL;
869 			MNT_IUNLOCK(mp);
870 		} else {
871 			printf("WARNING: %s: GJOURNAL flag on fs "
872 			    "but no gjournal provider below\n",
873 			    mp->mnt_stat.f_mntonname);
874 			free(mp->mnt_gjprovider, M_UFSMNT);
875 			mp->mnt_gjprovider = NULL;
876 		}
877 #else
878 		printf("WARNING: %s: GJOURNAL flag on fs but no "
879 		    "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
880 #endif
881 	} else {
882 		mp->mnt_gjprovider = NULL;
883 	}
884 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
885 	ump->um_cp = cp;
886 	ump->um_bo = &devvp->v_bufobj;
887 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
888 	if (fs->fs_magic == FS_UFS1_MAGIC) {
889 		ump->um_fstype = UFS1;
890 		ump->um_balloc = ffs_balloc_ufs1;
891 	} else {
892 		ump->um_fstype = UFS2;
893 		ump->um_balloc = ffs_balloc_ufs2;
894 	}
895 	ump->um_blkatoff = ffs_blkatoff;
896 	ump->um_truncate = ffs_truncate;
897 	ump->um_update = ffs_update;
898 	ump->um_valloc = ffs_valloc;
899 	ump->um_vfree = ffs_vfree;
900 	ump->um_ifree = ffs_ifree;
901 	ump->um_rdonly = ffs_rdonly;
902 	ump->um_snapgone = ffs_snapgone;
903 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
904 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
905 	if (fs->fs_sbsize < SBLOCKSIZE)
906 		bp->b_flags |= B_INVAL | B_NOCACHE;
907 	brelse(bp);
908 	bp = NULL;
909 	fs = ump->um_fs;
910 	ffs_oldfscompat_read(fs, ump, sblockloc);
911 	fs->fs_ronly = ronly;
912 	size = fs->fs_cssize;
913 	blks = howmany(size, fs->fs_fsize);
914 	if (fs->fs_contigsumsize > 0)
915 		size += fs->fs_ncg * sizeof(int32_t);
916 	size += fs->fs_ncg * sizeof(u_int8_t);
917 	space = malloc(size, M_UFSMNT, M_WAITOK);
918 	fs->fs_csp = space;
919 	for (i = 0; i < blks; i += fs->fs_frag) {
920 		size = fs->fs_bsize;
921 		if (i + fs->fs_frag > blks)
922 			size = (blks - i) * fs->fs_fsize;
923 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
924 		    cred, &bp)) != 0) {
925 			free(fs->fs_csp, M_UFSMNT);
926 			goto out;
927 		}
928 		bcopy(bp->b_data, space, (u_int)size);
929 		space = (char *)space + size;
930 		brelse(bp);
931 		bp = NULL;
932 	}
933 	if (fs->fs_contigsumsize > 0) {
934 		fs->fs_maxcluster = lp = space;
935 		for (i = 0; i < fs->fs_ncg; i++)
936 			*lp++ = fs->fs_contigsumsize;
937 		space = lp;
938 	}
939 	size = fs->fs_ncg * sizeof(u_int8_t);
940 	fs->fs_contigdirs = (u_int8_t *)space;
941 	bzero(fs->fs_contigdirs, size);
942 	fs->fs_active = NULL;
943 	mp->mnt_data = ump;
944 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
945 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
946 	nmp = NULL;
947 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
948 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
949 		if (nmp)
950 			vfs_rel(nmp);
951 		vfs_getnewfsid(mp);
952 	}
953 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
954 	MNT_ILOCK(mp);
955 	mp->mnt_flag |= MNT_LOCAL;
956 	MNT_IUNLOCK(mp);
957 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
958 #ifdef MAC
959 		MNT_ILOCK(mp);
960 		mp->mnt_flag |= MNT_MULTILABEL;
961 		MNT_IUNLOCK(mp);
962 #else
963 		printf("WARNING: %s: multilabel flag on fs but "
964 		    "no MAC support\n", mp->mnt_stat.f_mntonname);
965 #endif
966 	}
967 	if ((fs->fs_flags & FS_ACLS) != 0) {
968 #ifdef UFS_ACL
969 		MNT_ILOCK(mp);
970 
971 		if (mp->mnt_flag & MNT_NFS4ACLS)
972 			printf("WARNING: %s: ACLs flag on fs conflicts with "
973 			    "\"nfsv4acls\" mount option; option ignored\n",
974 			    mp->mnt_stat.f_mntonname);
975 		mp->mnt_flag &= ~MNT_NFS4ACLS;
976 		mp->mnt_flag |= MNT_ACLS;
977 
978 		MNT_IUNLOCK(mp);
979 #else
980 		printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
981 		    mp->mnt_stat.f_mntonname);
982 #endif
983 	}
984 	if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
985 #ifdef UFS_ACL
986 		MNT_ILOCK(mp);
987 
988 		if (mp->mnt_flag & MNT_ACLS)
989 			printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
990 			    "with \"acls\" mount option; option ignored\n",
991 			    mp->mnt_stat.f_mntonname);
992 		mp->mnt_flag &= ~MNT_ACLS;
993 		mp->mnt_flag |= MNT_NFS4ACLS;
994 
995 		MNT_IUNLOCK(mp);
996 #else
997 		printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
998 		    "ACLs support\n", mp->mnt_stat.f_mntonname);
999 #endif
1000 	}
1001 	if ((fs->fs_flags & FS_TRIM) != 0) {
1002 		len = sizeof(int);
1003 		if (g_io_getattr("GEOM::candelete", cp, &len,
1004 		    &ump->um_candelete) == 0) {
1005 			if (!ump->um_candelete)
1006 				printf("WARNING: %s: TRIM flag on fs but disk "
1007 				    "does not support TRIM\n",
1008 				    mp->mnt_stat.f_mntonname);
1009 		} else {
1010 			printf("WARNING: %s: TRIM flag on fs but disk does "
1011 			    "not confirm that it supports TRIM\n",
1012 			    mp->mnt_stat.f_mntonname);
1013 			ump->um_candelete = 0;
1014 		}
1015 		if (ump->um_candelete) {
1016 			ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1017 			    taskqueue_thread_enqueue, &ump->um_trim_tq);
1018 			taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1019 			    "%s trim", mp->mnt_stat.f_mntonname);
1020 		}
1021 	}
1022 
1023 	ump->um_mountp = mp;
1024 	ump->um_dev = dev;
1025 	ump->um_devvp = devvp;
1026 	ump->um_nindir = fs->fs_nindir;
1027 	ump->um_bptrtodb = fs->fs_fsbtodb;
1028 	ump->um_seqinc = fs->fs_frag;
1029 	for (i = 0; i < MAXQUOTAS; i++)
1030 		ump->um_quotas[i] = NULLVP;
1031 #ifdef UFS_EXTATTR
1032 	ufs_extattr_uepm_init(&ump->um_extattr);
1033 #endif
1034 	/*
1035 	 * Set FS local "last mounted on" information (NULL pad)
1036 	 */
1037 	bzero(fs->fs_fsmnt, MAXMNTLEN);
1038 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1039 	mp->mnt_stat.f_iosize = fs->fs_bsize;
1040 
1041 	if (mp->mnt_flag & MNT_ROOTFS) {
1042 		/*
1043 		 * Root mount; update timestamp in mount structure.
1044 		 * this will be used by the common root mount code
1045 		 * to update the system clock.
1046 		 */
1047 		mp->mnt_time = fs->fs_time;
1048 	}
1049 
1050 	if (ronly == 0) {
1051 		fs->fs_mtime = time_second;
1052 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
1053 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1054 			free(fs->fs_csp, M_UFSMNT);
1055 			ffs_flushfiles(mp, FORCECLOSE, td);
1056 			goto out;
1057 		}
1058 		if (fs->fs_snapinum[0] != 0)
1059 			ffs_snapshot_mount(mp);
1060 		fs->fs_fmod = 1;
1061 		fs->fs_clean = 0;
1062 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
1063 	}
1064 	/*
1065 	 * Initialize filesystem state information in mount struct.
1066 	 */
1067 	MNT_ILOCK(mp);
1068 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1069 	    MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1070 	MNT_IUNLOCK(mp);
1071 #ifdef UFS_EXTATTR
1072 #ifdef UFS_EXTATTR_AUTOSTART
1073 	/*
1074 	 *
1075 	 * Auto-starting does the following:
1076 	 *	- check for /.attribute in the fs, and extattr_start if so
1077 	 *	- for each file in .attribute, enable that file with
1078 	 * 	  an attribute of the same name.
1079 	 * Not clear how to report errors -- probably eat them.
1080 	 * This would all happen while the filesystem was busy/not
1081 	 * available, so would effectively be "atomic".
1082 	 */
1083 	(void) ufs_extattr_autostart(mp, td);
1084 #endif /* !UFS_EXTATTR_AUTOSTART */
1085 #endif /* !UFS_EXTATTR */
1086 	return (0);
1087 out:
1088 	if (bp)
1089 		brelse(bp);
1090 	if (cp != NULL) {
1091 		g_topology_lock();
1092 		g_vfs_close(cp);
1093 		g_topology_unlock();
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 	atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1106 	dev_rel(dev);
1107 	return (error);
1108 }
1109 
1110 #include <sys/sysctl.h>
1111 static int bigcgs = 0;
1112 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1113 
1114 /*
1115  * Sanity checks for loading old filesystem superblocks.
1116  * See ffs_oldfscompat_write below for unwound actions.
1117  *
1118  * XXX - Parts get retired eventually.
1119  * Unfortunately new bits get added.
1120  */
1121 static void
1122 ffs_oldfscompat_read(fs, ump, sblockloc)
1123 	struct fs *fs;
1124 	struct ufsmount *ump;
1125 	ufs2_daddr_t sblockloc;
1126 {
1127 	off_t maxfilesize;
1128 
1129 	/*
1130 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
1131 	 */
1132 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1133 		fs->fs_flags = fs->fs_old_flags;
1134 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
1135 		fs->fs_sblockloc = sblockloc;
1136 	}
1137 	/*
1138 	 * If not yet done, update UFS1 superblock with new wider fields.
1139 	 */
1140 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1141 		fs->fs_maxbsize = fs->fs_bsize;
1142 		fs->fs_time = fs->fs_old_time;
1143 		fs->fs_size = fs->fs_old_size;
1144 		fs->fs_dsize = fs->fs_old_dsize;
1145 		fs->fs_csaddr = fs->fs_old_csaddr;
1146 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1147 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1148 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1149 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1150 	}
1151 	if (fs->fs_magic == FS_UFS1_MAGIC &&
1152 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
1153 		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1154 		fs->fs_qbmask = ~fs->fs_bmask;
1155 		fs->fs_qfmask = ~fs->fs_fmask;
1156 	}
1157 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1158 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1159 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1160 		if (fs->fs_maxfilesize > maxfilesize)
1161 			fs->fs_maxfilesize = maxfilesize;
1162 	}
1163 	/* Compatibility for old filesystems */
1164 	if (fs->fs_avgfilesize <= 0)
1165 		fs->fs_avgfilesize = AVFILESIZ;
1166 	if (fs->fs_avgfpdir <= 0)
1167 		fs->fs_avgfpdir = AFPDIR;
1168 	if (bigcgs) {
1169 		fs->fs_save_cgsize = fs->fs_cgsize;
1170 		fs->fs_cgsize = fs->fs_bsize;
1171 	}
1172 }
1173 
1174 /*
1175  * Unwinding superblock updates for old filesystems.
1176  * See ffs_oldfscompat_read above for details.
1177  *
1178  * XXX - Parts get retired eventually.
1179  * Unfortunately new bits get added.
1180  */
1181 void
1182 ffs_oldfscompat_write(fs, ump)
1183 	struct fs *fs;
1184 	struct ufsmount *ump;
1185 {
1186 
1187 	/*
1188 	 * Copy back UFS2 updated fields that UFS1 inspects.
1189 	 */
1190 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1191 		fs->fs_old_time = fs->fs_time;
1192 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1193 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1194 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1195 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1196 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1197 	}
1198 	if (bigcgs) {
1199 		fs->fs_cgsize = fs->fs_save_cgsize;
1200 		fs->fs_save_cgsize = 0;
1201 	}
1202 }
1203 
1204 /*
1205  * unmount system call
1206  */
1207 static int
1208 ffs_unmount(mp, mntflags)
1209 	struct mount *mp;
1210 	int mntflags;
1211 {
1212 	struct thread *td;
1213 	struct ufsmount *ump = VFSTOUFS(mp);
1214 	struct fs *fs;
1215 	int error, flags, susp;
1216 #ifdef UFS_EXTATTR
1217 	int e_restart;
1218 #endif
1219 
1220 	flags = 0;
1221 	td = curthread;
1222 	fs = ump->um_fs;
1223 	susp = 0;
1224 	if (mntflags & MNT_FORCE) {
1225 		flags |= FORCECLOSE;
1226 		susp = fs->fs_ronly == 0;
1227 	}
1228 #ifdef UFS_EXTATTR
1229 	if ((error = ufs_extattr_stop(mp, td))) {
1230 		if (error != EOPNOTSUPP)
1231 			printf("WARNING: unmount %s: ufs_extattr_stop "
1232 			    "returned errno %d\n", mp->mnt_stat.f_mntonname,
1233 			    error);
1234 		e_restart = 0;
1235 	} else {
1236 		ufs_extattr_uepm_destroy(&ump->um_extattr);
1237 		e_restart = 1;
1238 	}
1239 #endif
1240 	if (susp) {
1241 		error = vfs_write_suspend_umnt(mp);
1242 		if (error != 0)
1243 			goto fail1;
1244 	}
1245 	if (MOUNTEDSOFTDEP(mp))
1246 		error = softdep_flushfiles(mp, flags, td);
1247 	else
1248 		error = ffs_flushfiles(mp, flags, td);
1249 	if (error != 0 && error != ENXIO)
1250 		goto fail;
1251 
1252 	UFS_LOCK(ump);
1253 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1254 		printf("WARNING: unmount %s: pending error: blocks %jd "
1255 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1256 		    fs->fs_pendinginodes);
1257 		fs->fs_pendingblocks = 0;
1258 		fs->fs_pendinginodes = 0;
1259 	}
1260 	UFS_UNLOCK(ump);
1261 	if (MOUNTEDSOFTDEP(mp))
1262 		softdep_unmount(mp);
1263 	if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1264 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1265 		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1266 		if (error && error != ENXIO) {
1267 			fs->fs_clean = 0;
1268 			goto fail;
1269 		}
1270 	}
1271 	if (susp)
1272 		vfs_write_resume(mp, VR_START_WRITE);
1273 	if (ump->um_trim_tq != NULL) {
1274 		while (ump->um_trim_inflight != 0)
1275 			pause("ufsutr", hz);
1276 		taskqueue_drain_all(ump->um_trim_tq);
1277 		taskqueue_free(ump->um_trim_tq);
1278 	}
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 	atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1290 	vrele(ump->um_devvp);
1291 	dev_rel(ump->um_dev);
1292 	mtx_destroy(UFS_MTX(ump));
1293 	if (mp->mnt_gjprovider != NULL) {
1294 		free(mp->mnt_gjprovider, M_UFSMNT);
1295 		mp->mnt_gjprovider = NULL;
1296 	}
1297 	free(fs->fs_csp, M_UFSMNT);
1298 	free(fs, M_UFSMNT);
1299 	free(ump, M_UFSMNT);
1300 	mp->mnt_data = NULL;
1301 	MNT_ILOCK(mp);
1302 	mp->mnt_flag &= ~MNT_LOCAL;
1303 	MNT_IUNLOCK(mp);
1304 	return (error);
1305 
1306 fail:
1307 	if (susp)
1308 		vfs_write_resume(mp, VR_START_WRITE);
1309 fail1:
1310 #ifdef UFS_EXTATTR
1311 	if (e_restart) {
1312 		ufs_extattr_uepm_init(&ump->um_extattr);
1313 #ifdef UFS_EXTATTR_AUTOSTART
1314 		(void) ufs_extattr_autostart(mp, td);
1315 #endif
1316 	}
1317 #endif
1318 
1319 	return (error);
1320 }
1321 
1322 /*
1323  * Flush out all the files in a filesystem.
1324  */
1325 int
1326 ffs_flushfiles(mp, flags, td)
1327 	struct mount *mp;
1328 	int flags;
1329 	struct thread *td;
1330 {
1331 	struct ufsmount *ump;
1332 	int qerror, error;
1333 
1334 	ump = VFSTOUFS(mp);
1335 	qerror = 0;
1336 #ifdef QUOTA
1337 	if (mp->mnt_flag & MNT_QUOTA) {
1338 		int i;
1339 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1340 		if (error)
1341 			return (error);
1342 		for (i = 0; i < MAXQUOTAS; i++) {
1343 			error = quotaoff(td, mp, i);
1344 			if (error != 0) {
1345 				if ((flags & EARLYFLUSH) == 0)
1346 					return (error);
1347 				else
1348 					qerror = error;
1349 			}
1350 		}
1351 
1352 		/*
1353 		 * Here we fall through to vflush again to ensure that
1354 		 * we have gotten rid of all the system vnodes, unless
1355 		 * quotas must not be closed.
1356 		 */
1357 	}
1358 #endif
1359 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1360 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1361 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1362 			return (error);
1363 		ffs_snapshot_unmount(mp);
1364 		flags |= FORCECLOSE;
1365 		/*
1366 		 * Here we fall through to vflush again to ensure
1367 		 * that we have gotten rid of all the system vnodes.
1368 		 */
1369 	}
1370 
1371 	/*
1372 	 * Do not close system files if quotas were not closed, to be
1373 	 * able to sync the remaining dquots.  The freeblks softupdate
1374 	 * workitems might hold a reference on a dquot, preventing
1375 	 * quotaoff() from completing.  Next round of
1376 	 * softdep_flushworklist() iteration should process the
1377 	 * blockers, allowing the next run of quotaoff() to finally
1378 	 * flush held dquots.
1379 	 *
1380 	 * Otherwise, flush all the files.
1381 	 */
1382 	if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1383 		return (error);
1384 
1385 	/*
1386 	 * Flush filesystem metadata.
1387 	 */
1388 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1389 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1390 	VOP_UNLOCK(ump->um_devvp, 0);
1391 	return (error);
1392 }
1393 
1394 /*
1395  * Get filesystem statistics.
1396  */
1397 static int
1398 ffs_statfs(mp, sbp)
1399 	struct mount *mp;
1400 	struct statfs *sbp;
1401 {
1402 	struct ufsmount *ump;
1403 	struct fs *fs;
1404 
1405 	ump = VFSTOUFS(mp);
1406 	fs = ump->um_fs;
1407 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1408 		panic("ffs_statfs");
1409 	sbp->f_version = STATFS_VERSION;
1410 	sbp->f_bsize = fs->fs_fsize;
1411 	sbp->f_iosize = fs->fs_bsize;
1412 	sbp->f_blocks = fs->fs_dsize;
1413 	UFS_LOCK(ump);
1414 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1415 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1416 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1417 	    dbtofsb(fs, fs->fs_pendingblocks);
1418 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1419 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1420 	UFS_UNLOCK(ump);
1421 	sbp->f_namemax = NAME_MAX;
1422 	return (0);
1423 }
1424 
1425 static bool
1426 sync_doupdate(struct inode *ip)
1427 {
1428 
1429 	return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1430 	    IN_UPDATE)) != 0);
1431 }
1432 
1433 /*
1434  * For a lazy sync, we only care about access times, quotas and the
1435  * superblock.  Other filesystem changes are already converted to
1436  * cylinder group blocks or inode blocks updates and are written to
1437  * disk by syncer.
1438  */
1439 static int
1440 ffs_sync_lazy(mp)
1441      struct mount *mp;
1442 {
1443 	struct vnode *mvp, *vp;
1444 	struct inode *ip;
1445 	struct thread *td;
1446 	int allerror, error;
1447 
1448 	allerror = 0;
1449 	td = curthread;
1450 	if ((mp->mnt_flag & MNT_NOATIME) != 0)
1451 		goto qupdate;
1452 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1453 		if (vp->v_type == VNON) {
1454 			VI_UNLOCK(vp);
1455 			continue;
1456 		}
1457 		ip = VTOI(vp);
1458 
1459 		/*
1460 		 * The IN_ACCESS flag is converted to IN_MODIFIED by
1461 		 * ufs_close() and ufs_getattr() by the calls to
1462 		 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1463 		 * Test also all the other timestamp flags too, to pick up
1464 		 * any other cases that could be missed.
1465 		 */
1466 		if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1467 			VI_UNLOCK(vp);
1468 			continue;
1469 		}
1470 		if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
1471 		    td)) != 0)
1472 			continue;
1473 		if (sync_doupdate(ip))
1474 			error = ffs_update(vp, 0);
1475 		if (error != 0)
1476 			allerror = error;
1477 		vput(vp);
1478 	}
1479 
1480 qupdate:
1481 #ifdef QUOTA
1482 	qsync(mp);
1483 #endif
1484 
1485 	if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1486 	    (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1487 		allerror = error;
1488 	return (allerror);
1489 }
1490 
1491 /*
1492  * Go through the disk queues to initiate sandbagged IO;
1493  * go through the inodes to write those that have been modified;
1494  * initiate the writing of the super block if it has been modified.
1495  *
1496  * Note: we are always called with the filesystem marked busy using
1497  * vfs_busy().
1498  */
1499 static int
1500 ffs_sync(mp, waitfor)
1501 	struct mount *mp;
1502 	int waitfor;
1503 {
1504 	struct vnode *mvp, *vp, *devvp;
1505 	struct thread *td;
1506 	struct inode *ip;
1507 	struct ufsmount *ump = VFSTOUFS(mp);
1508 	struct fs *fs;
1509 	int error, count, lockreq, allerror = 0;
1510 	int suspend;
1511 	int suspended;
1512 	int secondary_writes;
1513 	int secondary_accwrites;
1514 	int softdep_deps;
1515 	int softdep_accdeps;
1516 	struct bufobj *bo;
1517 
1518 	suspend = 0;
1519 	suspended = 0;
1520 	td = curthread;
1521 	fs = ump->um_fs;
1522 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1523 		panic("%s: ffs_sync: modification on read-only filesystem",
1524 		    fs->fs_fsmnt);
1525 	if (waitfor == MNT_LAZY) {
1526 		if (!rebooting)
1527 			return (ffs_sync_lazy(mp));
1528 		waitfor = MNT_NOWAIT;
1529 	}
1530 
1531 	/*
1532 	 * Write back each (modified) inode.
1533 	 */
1534 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1535 	if (waitfor == MNT_SUSPEND) {
1536 		suspend = 1;
1537 		waitfor = MNT_WAIT;
1538 	}
1539 	if (waitfor == MNT_WAIT)
1540 		lockreq = LK_EXCLUSIVE;
1541 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1542 loop:
1543 	/* Grab snapshot of secondary write counts */
1544 	MNT_ILOCK(mp);
1545 	secondary_writes = mp->mnt_secondary_writes;
1546 	secondary_accwrites = mp->mnt_secondary_accwrites;
1547 	MNT_IUNLOCK(mp);
1548 
1549 	/* Grab snapshot of softdep dependency counts */
1550 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1551 
1552 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1553 		/*
1554 		 * Depend on the vnode interlock to keep things stable enough
1555 		 * for a quick test.  Since there might be hundreds of
1556 		 * thousands of vnodes, we cannot afford even a subroutine
1557 		 * call unless there's a good chance that we have work to do.
1558 		 */
1559 		if (vp->v_type == VNON) {
1560 			VI_UNLOCK(vp);
1561 			continue;
1562 		}
1563 		ip = VTOI(vp);
1564 		if ((ip->i_flag &
1565 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1566 		    vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1567 			VI_UNLOCK(vp);
1568 			continue;
1569 		}
1570 		if ((error = vget(vp, lockreq, td)) != 0) {
1571 			if (error == ENOENT || error == ENOLCK) {
1572 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1573 				goto loop;
1574 			}
1575 			continue;
1576 		}
1577 		if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0)
1578 			allerror = error;
1579 		vput(vp);
1580 	}
1581 	/*
1582 	 * Force stale filesystem control information to be flushed.
1583 	 */
1584 	if (waitfor == MNT_WAIT || rebooting) {
1585 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1586 			allerror = error;
1587 		/* Flushed work items may create new vnodes to clean */
1588 		if (allerror == 0 && count)
1589 			goto loop;
1590 	}
1591 #ifdef QUOTA
1592 	qsync(mp);
1593 #endif
1594 
1595 	devvp = ump->um_devvp;
1596 	bo = &devvp->v_bufobj;
1597 	BO_LOCK(bo);
1598 	if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1599 		BO_UNLOCK(bo);
1600 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1601 		error = VOP_FSYNC(devvp, waitfor, td);
1602 		VOP_UNLOCK(devvp, 0);
1603 		if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1604 			error = ffs_sbupdate(ump, waitfor, 0);
1605 		if (error != 0)
1606 			allerror = error;
1607 		if (allerror == 0 && waitfor == MNT_WAIT)
1608 			goto loop;
1609 	} else if (suspend != 0) {
1610 		if (softdep_check_suspend(mp,
1611 					  devvp,
1612 					  softdep_deps,
1613 					  softdep_accdeps,
1614 					  secondary_writes,
1615 					  secondary_accwrites) != 0) {
1616 			MNT_IUNLOCK(mp);
1617 			goto loop;	/* More work needed */
1618 		}
1619 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1620 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1621 		MNT_IUNLOCK(mp);
1622 		suspended = 1;
1623 	} else
1624 		BO_UNLOCK(bo);
1625 	/*
1626 	 * Write back modified superblock.
1627 	 */
1628 	if (fs->fs_fmod != 0 &&
1629 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1630 		allerror = error;
1631 	return (allerror);
1632 }
1633 
1634 int
1635 ffs_vget(mp, ino, flags, vpp)
1636 	struct mount *mp;
1637 	ino_t ino;
1638 	int flags;
1639 	struct vnode **vpp;
1640 {
1641 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1642 }
1643 
1644 int
1645 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1646 	struct mount *mp;
1647 	ino_t ino;
1648 	int flags;
1649 	struct vnode **vpp;
1650 	int ffs_flags;
1651 {
1652 	struct fs *fs;
1653 	struct inode *ip;
1654 	struct ufsmount *ump;
1655 	struct buf *bp;
1656 	struct vnode *vp;
1657 	int error;
1658 
1659 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1660 	if (error || *vpp != NULL)
1661 		return (error);
1662 
1663 	/*
1664 	 * We must promote to an exclusive lock for vnode creation.  This
1665 	 * can happen if lookup is passed LOCKSHARED.
1666 	 */
1667 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1668 		flags &= ~LK_TYPE_MASK;
1669 		flags |= LK_EXCLUSIVE;
1670 	}
1671 
1672 	/*
1673 	 * We do not lock vnode creation as it is believed to be too
1674 	 * expensive for such rare case as simultaneous creation of vnode
1675 	 * for same ino by different processes. We just allow them to race
1676 	 * and check later to decide who wins. Let the race begin!
1677 	 */
1678 
1679 	ump = VFSTOUFS(mp);
1680 	fs = ump->um_fs;
1681 	ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1682 
1683 	/* Allocate a new vnode/inode. */
1684 	error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1685 	    &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1686 	if (error) {
1687 		*vpp = NULL;
1688 		uma_zfree(uma_inode, ip);
1689 		return (error);
1690 	}
1691 	/*
1692 	 * FFS supports recursive locking.
1693 	 */
1694 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1695 	VN_LOCK_AREC(vp);
1696 	vp->v_data = ip;
1697 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1698 	ip->i_vnode = vp;
1699 	ip->i_ump = ump;
1700 	ip->i_number = ino;
1701 	ip->i_ea_refs = 0;
1702 	ip->i_nextclustercg = -1;
1703 	ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2;
1704 #ifdef QUOTA
1705 	{
1706 		int i;
1707 		for (i = 0; i < MAXQUOTAS; i++)
1708 			ip->i_dquot[i] = NODQUOT;
1709 	}
1710 #endif
1711 
1712 	if (ffs_flags & FFSV_FORCEINSMQ)
1713 		vp->v_vflag |= VV_FORCEINSMQ;
1714 	error = insmntque(vp, mp);
1715 	if (error != 0) {
1716 		uma_zfree(uma_inode, ip);
1717 		*vpp = NULL;
1718 		return (error);
1719 	}
1720 	vp->v_vflag &= ~VV_FORCEINSMQ;
1721 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1722 	if (error || *vpp != NULL)
1723 		return (error);
1724 
1725 	/* Read in the disk contents for the inode, copy into the inode. */
1726 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1727 	    (int)fs->fs_bsize, NOCRED, &bp);
1728 	if (error) {
1729 		/*
1730 		 * The inode does not contain anything useful, so it would
1731 		 * be misleading to leave it on its hash chain. With mode
1732 		 * still zero, it will be unlinked and returned to the free
1733 		 * list by vput().
1734 		 */
1735 		brelse(bp);
1736 		vput(vp);
1737 		*vpp = NULL;
1738 		return (error);
1739 	}
1740 	if (I_IS_UFS1(ip))
1741 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1742 	else
1743 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1744 	ffs_load_inode(bp, ip, fs, ino);
1745 	if (DOINGSOFTDEP(vp))
1746 		softdep_load_inodeblock(ip);
1747 	else
1748 		ip->i_effnlink = ip->i_nlink;
1749 	bqrelse(bp);
1750 
1751 	/*
1752 	 * Initialize the vnode from the inode, check for aliases.
1753 	 * Note that the underlying vnode may have changed.
1754 	 */
1755 	error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2,
1756 	    &vp);
1757 	if (error) {
1758 		vput(vp);
1759 		*vpp = NULL;
1760 		return (error);
1761 	}
1762 
1763 	/*
1764 	 * Finish inode initialization.
1765 	 */
1766 	if (vp->v_type != VFIFO) {
1767 		/* FFS supports shared locking for all files except fifos. */
1768 		VN_LOCK_ASHARE(vp);
1769 	}
1770 
1771 	/*
1772 	 * Set up a generation number for this inode if it does not
1773 	 * already have one. This should only happen on old filesystems.
1774 	 */
1775 	if (ip->i_gen == 0) {
1776 		while (ip->i_gen == 0)
1777 			ip->i_gen = arc4random();
1778 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1779 			ip->i_flag |= IN_MODIFIED;
1780 			DIP_SET(ip, i_gen, ip->i_gen);
1781 		}
1782 	}
1783 #ifdef MAC
1784 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1785 		/*
1786 		 * If this vnode is already allocated, and we're running
1787 		 * multi-label, attempt to perform a label association
1788 		 * from the extended attributes on the inode.
1789 		 */
1790 		error = mac_vnode_associate_extattr(mp, vp);
1791 		if (error) {
1792 			/* ufs_inactive will release ip->i_devvp ref. */
1793 			vput(vp);
1794 			*vpp = NULL;
1795 			return (error);
1796 		}
1797 	}
1798 #endif
1799 
1800 	*vpp = vp;
1801 	return (0);
1802 }
1803 
1804 /*
1805  * File handle to vnode
1806  *
1807  * Have to be really careful about stale file handles:
1808  * - check that the inode number is valid
1809  * - for UFS2 check that the inode number is initialized
1810  * - call ffs_vget() to get the locked inode
1811  * - check for an unallocated inode (i_mode == 0)
1812  * - check that the given client host has export rights and return
1813  *   those rights via. exflagsp and credanonp
1814  */
1815 static int
1816 ffs_fhtovp(mp, fhp, flags, vpp)
1817 	struct mount *mp;
1818 	struct fid *fhp;
1819 	int flags;
1820 	struct vnode **vpp;
1821 {
1822 	struct ufid *ufhp;
1823 	struct ufsmount *ump;
1824 	struct fs *fs;
1825 	struct cg *cgp;
1826 	struct buf *bp;
1827 	ino_t ino;
1828 	u_int cg;
1829 	int error;
1830 
1831 	ufhp = (struct ufid *)fhp;
1832 	ino = ufhp->ufid_ino;
1833 	ump = VFSTOUFS(mp);
1834 	fs = ump->um_fs;
1835 	if (ino < ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
1836 		return (ESTALE);
1837 	/*
1838 	 * Need to check if inode is initialized because UFS2 does lazy
1839 	 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
1840 	 */
1841 	if (fs->fs_magic != FS_UFS2_MAGIC)
1842 		return (ufs_fhtovp(mp, ufhp, flags, vpp));
1843 	cg = ino_to_cg(fs, ino);
1844 	error = bread(ump->um_devvp, fsbtodb(fs, cgtod(fs, cg)),
1845 		(int)fs->fs_cgsize, NOCRED, &bp);
1846 	if (error)
1847 		return (error);
1848 	cgp = (struct cg *)bp->b_data;
1849 	if (!cg_chkmagic(cgp) || ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
1850 		brelse(bp);
1851 		return (ESTALE);
1852 	}
1853 	brelse(bp);
1854 	return (ufs_fhtovp(mp, ufhp, flags, vpp));
1855 }
1856 
1857 /*
1858  * Initialize the filesystem.
1859  */
1860 static int
1861 ffs_init(vfsp)
1862 	struct vfsconf *vfsp;
1863 {
1864 
1865 	ffs_susp_initialize();
1866 	softdep_initialize();
1867 	return (ufs_init(vfsp));
1868 }
1869 
1870 /*
1871  * Undo the work of ffs_init().
1872  */
1873 static int
1874 ffs_uninit(vfsp)
1875 	struct vfsconf *vfsp;
1876 {
1877 	int ret;
1878 
1879 	ret = ufs_uninit(vfsp);
1880 	softdep_uninitialize();
1881 	ffs_susp_uninitialize();
1882 	return (ret);
1883 }
1884 
1885 /*
1886  * Write a superblock and associated information back to disk.
1887  */
1888 int
1889 ffs_sbupdate(ump, waitfor, suspended)
1890 	struct ufsmount *ump;
1891 	int waitfor;
1892 	int suspended;
1893 {
1894 	struct fs *fs = ump->um_fs;
1895 	struct buf *sbbp;
1896 	struct buf *bp;
1897 	int blks;
1898 	void *space;
1899 	int i, size, error, allerror = 0;
1900 
1901 	if (fs->fs_ronly == 1 &&
1902 	    (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1903 	    (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
1904 		panic("ffs_sbupdate: write read-only filesystem");
1905 	/*
1906 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1907 	 */
1908 	sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
1909 	    (int)fs->fs_sbsize, 0, 0, 0);
1910 	/*
1911 	 * First write back the summary information.
1912 	 */
1913 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1914 	space = fs->fs_csp;
1915 	for (i = 0; i < blks; i += fs->fs_frag) {
1916 		size = fs->fs_bsize;
1917 		if (i + fs->fs_frag > blks)
1918 			size = (blks - i) * fs->fs_fsize;
1919 		bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1920 		    size, 0, 0, 0);
1921 		bcopy(space, bp->b_data, (u_int)size);
1922 		space = (char *)space + size;
1923 		if (suspended)
1924 			bp->b_flags |= B_VALIDSUSPWRT;
1925 		if (waitfor != MNT_WAIT)
1926 			bawrite(bp);
1927 		else if ((error = bwrite(bp)) != 0)
1928 			allerror = error;
1929 	}
1930 	/*
1931 	 * Now write back the superblock itself. If any errors occurred
1932 	 * up to this point, then fail so that the superblock avoids
1933 	 * being written out as clean.
1934 	 */
1935 	if (allerror) {
1936 		brelse(sbbp);
1937 		return (allerror);
1938 	}
1939 	bp = sbbp;
1940 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1941 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1942 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1943 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1944 		fs->fs_sblockloc = SBLOCK_UFS1;
1945 	}
1946 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
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_UFS2);
1950 		fs->fs_sblockloc = SBLOCK_UFS2;
1951 	}
1952 	fs->fs_fmod = 0;
1953 	fs->fs_time = time_second;
1954 	if (MOUNTEDSOFTDEP(ump->um_mountp))
1955 		softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
1956 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1957 	ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
1958 	if (suspended)
1959 		bp->b_flags |= B_VALIDSUSPWRT;
1960 	if (waitfor != MNT_WAIT)
1961 		bawrite(bp);
1962 	else if ((error = bwrite(bp)) != 0)
1963 		allerror = error;
1964 	return (allerror);
1965 }
1966 
1967 static int
1968 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1969 	int attrnamespace, const char *attrname)
1970 {
1971 
1972 #ifdef UFS_EXTATTR
1973 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1974 	    attrname));
1975 #else
1976 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1977 	    attrname));
1978 #endif
1979 }
1980 
1981 static void
1982 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1983 {
1984 
1985 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1986 		uma_zfree(uma_ufs1, ip->i_din1);
1987 	else if (ip->i_din2 != NULL)
1988 		uma_zfree(uma_ufs2, ip->i_din2);
1989 	uma_zfree(uma_inode, ip);
1990 }
1991 
1992 static int dobkgrdwrite = 1;
1993 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1994     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1995 
1996 /*
1997  * Complete a background write started from bwrite.
1998  */
1999 static void
2000 ffs_backgroundwritedone(struct buf *bp)
2001 {
2002 	struct bufobj *bufobj;
2003 	struct buf *origbp;
2004 
2005 	/*
2006 	 * Find the original buffer that we are writing.
2007 	 */
2008 	bufobj = bp->b_bufobj;
2009 	BO_LOCK(bufobj);
2010 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2011 		panic("backgroundwritedone: lost buffer");
2012 
2013 	/*
2014 	 * We should mark the cylinder group buffer origbp as
2015 	 * dirty, to not loose the failed write.
2016 	 */
2017 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2018 		origbp->b_vflags |= BV_BKGRDERR;
2019 	BO_UNLOCK(bufobj);
2020 	/*
2021 	 * Process dependencies then return any unfinished ones.
2022 	 */
2023 	pbrelvp(bp);
2024 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2025 		buf_complete(bp);
2026 #ifdef SOFTUPDATES
2027 	if (!LIST_EMPTY(&bp->b_dep))
2028 		softdep_move_dependencies(bp, origbp);
2029 #endif
2030 	/*
2031 	 * This buffer is marked B_NOCACHE so when it is released
2032 	 * by biodone it will be tossed.
2033 	 */
2034 	bp->b_flags |= B_NOCACHE;
2035 	bp->b_flags &= ~B_CACHE;
2036 
2037 	/*
2038 	 * Prevent brelse() from trying to keep and re-dirtying bp on
2039 	 * errors. It causes b_bufobj dereference in
2040 	 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2041 	 * pbrelvp() above.
2042 	 */
2043 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2044 		bp->b_flags |= B_INVAL;
2045 	bufdone(bp);
2046 	BO_LOCK(bufobj);
2047 	/*
2048 	 * Clear the BV_BKGRDINPROG flag in the original buffer
2049 	 * and awaken it if it is waiting for the write to complete.
2050 	 * If BV_BKGRDINPROG is not set in the original buffer it must
2051 	 * have been released and re-instantiated - which is not legal.
2052 	 */
2053 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2054 	    ("backgroundwritedone: lost buffer2"));
2055 	origbp->b_vflags &= ~BV_BKGRDINPROG;
2056 	if (origbp->b_vflags & BV_BKGRDWAIT) {
2057 		origbp->b_vflags &= ~BV_BKGRDWAIT;
2058 		wakeup(&origbp->b_xflags);
2059 	}
2060 	BO_UNLOCK(bufobj);
2061 }
2062 
2063 
2064 /*
2065  * Write, release buffer on completion.  (Done by iodone
2066  * if async).  Do not bother writing anything if the buffer
2067  * is invalid.
2068  *
2069  * Note that we set B_CACHE here, indicating that buffer is
2070  * fully valid and thus cacheable.  This is true even of NFS
2071  * now so we set it generally.  This could be set either here
2072  * or in biodone() since the I/O is synchronous.  We put it
2073  * here.
2074  */
2075 static int
2076 ffs_bufwrite(struct buf *bp)
2077 {
2078 	struct buf *newbp;
2079 
2080 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2081 	if (bp->b_flags & B_INVAL) {
2082 		brelse(bp);
2083 		return (0);
2084 	}
2085 
2086 	if (!BUF_ISLOCKED(bp))
2087 		panic("bufwrite: buffer is not busy???");
2088 	/*
2089 	 * If a background write is already in progress, delay
2090 	 * writing this block if it is asynchronous. Otherwise
2091 	 * wait for the background write to complete.
2092 	 */
2093 	BO_LOCK(bp->b_bufobj);
2094 	if (bp->b_vflags & BV_BKGRDINPROG) {
2095 		if (bp->b_flags & B_ASYNC) {
2096 			BO_UNLOCK(bp->b_bufobj);
2097 			bdwrite(bp);
2098 			return (0);
2099 		}
2100 		bp->b_vflags |= BV_BKGRDWAIT;
2101 		msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2102 		    "bwrbg", 0);
2103 		if (bp->b_vflags & BV_BKGRDINPROG)
2104 			panic("bufwrite: still writing");
2105 	}
2106 	bp->b_vflags &= ~BV_BKGRDERR;
2107 	BO_UNLOCK(bp->b_bufobj);
2108 
2109 	/*
2110 	 * If this buffer is marked for background writing and we
2111 	 * do not have to wait for it, make a copy and write the
2112 	 * copy so as to leave this buffer ready for further use.
2113 	 *
2114 	 * This optimization eats a lot of memory.  If we have a page
2115 	 * or buffer shortfall we can't do it.
2116 	 */
2117 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2118 	    (bp->b_flags & B_ASYNC) &&
2119 	    !vm_page_count_severe() &&
2120 	    !buf_dirty_count_severe()) {
2121 		KASSERT(bp->b_iodone == NULL,
2122 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2123 
2124 		/* get a new block */
2125 		newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2126 		if (newbp == NULL)
2127 			goto normal_write;
2128 
2129 		KASSERT(buf_mapped(bp), ("Unmapped cg"));
2130 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2131 		BO_LOCK(bp->b_bufobj);
2132 		bp->b_vflags |= BV_BKGRDINPROG;
2133 		BO_UNLOCK(bp->b_bufobj);
2134 		newbp->b_xflags |= BX_BKGRDMARKER;
2135 		newbp->b_lblkno = bp->b_lblkno;
2136 		newbp->b_blkno = bp->b_blkno;
2137 		newbp->b_offset = bp->b_offset;
2138 		newbp->b_iodone = ffs_backgroundwritedone;
2139 		newbp->b_flags |= B_ASYNC;
2140 		newbp->b_flags &= ~B_INVAL;
2141 		pbgetvp(bp->b_vp, newbp);
2142 
2143 #ifdef SOFTUPDATES
2144 		/*
2145 		 * Move over the dependencies.  If there are rollbacks,
2146 		 * leave the parent buffer dirtied as it will need to
2147 		 * be written again.
2148 		 */
2149 		if (LIST_EMPTY(&bp->b_dep) ||
2150 		    softdep_move_dependencies(bp, newbp) == 0)
2151 			bundirty(bp);
2152 #else
2153 		bundirty(bp);
2154 #endif
2155 
2156 		/*
2157 		 * Initiate write on the copy, release the original.  The
2158 		 * BKGRDINPROG flag prevents it from going away until
2159 		 * the background write completes.
2160 		 */
2161 		bqrelse(bp);
2162 		bp = newbp;
2163 	} else
2164 		/* Mark the buffer clean */
2165 		bundirty(bp);
2166 
2167 
2168 	/* Let the normal bufwrite do the rest for us */
2169 normal_write:
2170 	return (bufwrite(bp));
2171 }
2172 
2173 
2174 static void
2175 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2176 {
2177 	struct vnode *vp;
2178 	int error;
2179 	struct buf *tbp;
2180 	int nocopy;
2181 
2182 	vp = bo2vnode(bo);
2183 	if (bp->b_iocmd == BIO_WRITE) {
2184 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2185 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2186 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2187 			panic("ffs_geom_strategy: bad I/O");
2188 		nocopy = bp->b_flags & B_NOCOPY;
2189 		bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2190 		if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2191 		    vp->v_rdev->si_snapdata != NULL) {
2192 			if ((bp->b_flags & B_CLUSTER) != 0) {
2193 				runningbufwakeup(bp);
2194 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2195 					      b_cluster.cluster_entry) {
2196 					error = ffs_copyonwrite(vp, tbp);
2197 					if (error != 0 &&
2198 					    error != EOPNOTSUPP) {
2199 						bp->b_error = error;
2200 						bp->b_ioflags |= BIO_ERROR;
2201 						bufdone(bp);
2202 						return;
2203 					}
2204 				}
2205 				bp->b_runningbufspace = bp->b_bufsize;
2206 				atomic_add_long(&runningbufspace,
2207 					       bp->b_runningbufspace);
2208 			} else {
2209 				error = ffs_copyonwrite(vp, bp);
2210 				if (error != 0 && error != EOPNOTSUPP) {
2211 					bp->b_error = error;
2212 					bp->b_ioflags |= BIO_ERROR;
2213 					bufdone(bp);
2214 					return;
2215 				}
2216 			}
2217 		}
2218 #ifdef SOFTUPDATES
2219 		if ((bp->b_flags & B_CLUSTER) != 0) {
2220 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2221 				      b_cluster.cluster_entry) {
2222 				if (!LIST_EMPTY(&tbp->b_dep))
2223 					buf_start(tbp);
2224 			}
2225 		} else {
2226 			if (!LIST_EMPTY(&bp->b_dep))
2227 				buf_start(bp);
2228 		}
2229 
2230 #endif
2231 	}
2232 	g_vfs_strategy(bo, bp);
2233 }
2234 
2235 int
2236 ffs_own_mount(const struct mount *mp)
2237 {
2238 
2239 	if (mp->mnt_op == &ufs_vfsops)
2240 		return (1);
2241 	return (0);
2242 }
2243 
2244 #ifdef	DDB
2245 #ifdef SOFTUPDATES
2246 
2247 /* defined in ffs_softdep.c */
2248 extern void db_print_ffs(struct ufsmount *ump);
2249 
2250 DB_SHOW_COMMAND(ffs, db_show_ffs)
2251 {
2252 	struct mount *mp;
2253 	struct ufsmount *ump;
2254 
2255 	if (have_addr) {
2256 		ump = VFSTOUFS((struct mount *)addr);
2257 		db_print_ffs(ump);
2258 		return;
2259 	}
2260 
2261 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2262 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2263 			db_print_ffs(VFSTOUFS(mp));
2264 	}
2265 }
2266 
2267 #endif	/* SOFTUPDATES */
2268 #endif	/* DDB */
2269