xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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/kernel.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/ioccom.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
55 #include <sys/rwlock.h>
56 
57 #include <security/mac/mac_framework.h>
58 
59 #include <ufs/ufs/extattr.h>
60 #include <ufs/ufs/gjournal.h>
61 #include <ufs/ufs/quota.h>
62 #include <ufs/ufs/ufsmount.h>
63 #include <ufs/ufs/inode.h>
64 #include <ufs/ufs/ufs_extern.h>
65 
66 #include <ufs/ffs/fs.h>
67 #include <ufs/ffs/ffs_extern.h>
68 
69 #include <vm/vm.h>
70 #include <vm/uma.h>
71 #include <vm/vm_page.h>
72 
73 #include <geom/geom.h>
74 #include <geom/geom_vfs.h>
75 
76 #include <ddb/ddb.h>
77 
78 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
79 
80 static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
81 static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
82 		    ufs2_daddr_t);
83 static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
84 static int	ffs_sync_lazy(struct mount *mp);
85 
86 static vfs_init_t ffs_init;
87 static vfs_uninit_t ffs_uninit;
88 static vfs_extattrctl_t ffs_extattrctl;
89 static vfs_cmount_t ffs_cmount;
90 static vfs_unmount_t ffs_unmount;
91 static vfs_mount_t ffs_mount;
92 static vfs_statfs_t ffs_statfs;
93 static vfs_fhtovp_t ffs_fhtovp;
94 static vfs_sync_t ffs_sync;
95 
96 static struct vfsops ufs_vfsops = {
97 	.vfs_extattrctl =	ffs_extattrctl,
98 	.vfs_fhtovp =		ffs_fhtovp,
99 	.vfs_init =		ffs_init,
100 	.vfs_mount =		ffs_mount,
101 	.vfs_cmount =		ffs_cmount,
102 	.vfs_quotactl =		ufs_quotactl,
103 	.vfs_root =		ufs_root,
104 	.vfs_statfs =		ffs_statfs,
105 	.vfs_sync =		ffs_sync,
106 	.vfs_uninit =		ffs_uninit,
107 	.vfs_unmount =		ffs_unmount,
108 	.vfs_vget =		ffs_vget,
109 	.vfs_susp_clean =	process_deferred_inactive,
110 };
111 
112 VFS_SET(ufs_vfsops, ufs, 0);
113 MODULE_VERSION(ufs, 1);
114 
115 static b_strategy_t ffs_geom_strategy;
116 static b_write_t ffs_bufwrite;
117 
118 static struct buf_ops ffs_ops = {
119 	.bop_name =	"FFS",
120 	.bop_write =	ffs_bufwrite,
121 	.bop_strategy =	ffs_geom_strategy,
122 	.bop_sync =	bufsync,
123 #ifdef NO_FFS_SNAPSHOT
124 	.bop_bdflush =	bufbdflush,
125 #else
126 	.bop_bdflush =	ffs_bdflush,
127 #endif
128 };
129 
130 /*
131  * Note that userquota and groupquota options are not currently used
132  * by UFS/FFS code and generally mount(8) does not pass those options
133  * from userland, but they can be passed by loader(8) via
134  * vfs.root.mountfrom.options.
135  */
136 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
137     "noclusterw", "noexec", "export", "force", "from", "groupquota",
138     "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
139     "nosymfollow", "sync", "union", "userquota", NULL };
140 
141 static int
142 ffs_mount(struct mount *mp)
143 {
144 	struct vnode *devvp;
145 	struct thread *td;
146 	struct ufsmount *ump = NULL;
147 	struct fs *fs;
148 	pid_t fsckpid = 0;
149 	int error, flags;
150 	uint64_t mntorflags;
151 	accmode_t accmode;
152 	struct nameidata ndp;
153 	char *fspec;
154 
155 	td = curthread;
156 	if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
157 		return (EINVAL);
158 	if (uma_inode == NULL) {
159 		uma_inode = uma_zcreate("FFS inode",
160 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
161 		    UMA_ALIGN_PTR, 0);
162 		uma_ufs1 = uma_zcreate("FFS1 dinode",
163 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
164 		    UMA_ALIGN_PTR, 0);
165 		uma_ufs2 = uma_zcreate("FFS2 dinode",
166 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
167 		    UMA_ALIGN_PTR, 0);
168 	}
169 
170 	vfs_deleteopt(mp->mnt_optnew, "groupquota");
171 	vfs_deleteopt(mp->mnt_optnew, "userquota");
172 
173 	fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
174 	if (error)
175 		return (error);
176 
177 	mntorflags = 0;
178 	if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
179 		mntorflags |= MNT_ACLS;
180 
181 	if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
182 		mntorflags |= MNT_SNAPSHOT;
183 		/*
184 		 * Once we have set the MNT_SNAPSHOT flag, do not
185 		 * persist "snapshot" in the options list.
186 		 */
187 		vfs_deleteopt(mp->mnt_optnew, "snapshot");
188 		vfs_deleteopt(mp->mnt_opt, "snapshot");
189 	}
190 
191 	if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
192 	    vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
193 		/*
194 		 * Once we have set the restricted PID, do not
195 		 * persist "fsckpid" in the options list.
196 		 */
197 		vfs_deleteopt(mp->mnt_optnew, "fsckpid");
198 		vfs_deleteopt(mp->mnt_opt, "fsckpid");
199 		if (mp->mnt_flag & MNT_UPDATE) {
200 			if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
201 			     vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
202 				vfs_mount_error(mp,
203 				    "Checker enable: Must be read-only");
204 				return (EINVAL);
205 			}
206 		} else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
207 			vfs_mount_error(mp,
208 			    "Checker enable: Must be read-only");
209 			return (EINVAL);
210 		}
211 		/* Set to -1 if we are done */
212 		if (fsckpid == 0)
213 			fsckpid = -1;
214 	}
215 
216 	if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
217 		if (mntorflags & MNT_ACLS) {
218 			vfs_mount_error(mp,
219 			    "\"acls\" and \"nfsv4acls\" options "
220 			    "are mutually exclusive");
221 			return (EINVAL);
222 		}
223 		mntorflags |= MNT_NFS4ACLS;
224 	}
225 
226 	MNT_ILOCK(mp);
227 	mp->mnt_flag |= mntorflags;
228 	MNT_IUNLOCK(mp);
229 	/*
230 	 * If updating, check whether changing from read-only to
231 	 * read/write; if there is no device name, that's all we do.
232 	 */
233 	if (mp->mnt_flag & MNT_UPDATE) {
234 		ump = VFSTOUFS(mp);
235 		fs = ump->um_fs;
236 		devvp = ump->um_devvp;
237 		if (fsckpid == -1 && ump->um_fsckpid > 0) {
238 			if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
239 			    (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
240 				return (error);
241 			DROP_GIANT();
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 			PICKUP_GIANT();
249 			ump->um_fsckpid = 0;
250 		}
251 		if (fs->fs_ronly == 0 &&
252 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
253 			/*
254 			 * Flush any dirty data and suspend filesystem.
255 			 */
256 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
257 				return (error);
258 			error = vfs_write_suspend_umnt(mp);
259 			if (error != 0)
260 				return (error);
261 			/*
262 			 * Check for and optionally get rid of files open
263 			 * for writing.
264 			 */
265 			flags = WRITECLOSE;
266 			if (mp->mnt_flag & MNT_FORCE)
267 				flags |= FORCECLOSE;
268 			if (MOUNTEDSOFTDEP(mp)) {
269 				error = softdep_flushfiles(mp, flags, td);
270 			} else {
271 				error = ffs_flushfiles(mp, flags, td);
272 			}
273 			if (error) {
274 				vfs_write_resume(mp, 0);
275 				return (error);
276 			}
277 			if (fs->fs_pendingblocks != 0 ||
278 			    fs->fs_pendinginodes != 0) {
279 				printf("WARNING: %s Update error: blocks %jd "
280 				    "files %d\n", fs->fs_fsmnt,
281 				    (intmax_t)fs->fs_pendingblocks,
282 				    fs->fs_pendinginodes);
283 				fs->fs_pendingblocks = 0;
284 				fs->fs_pendinginodes = 0;
285 			}
286 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
287 				fs->fs_clean = 1;
288 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
289 				fs->fs_ronly = 0;
290 				fs->fs_clean = 0;
291 				vfs_write_resume(mp, 0);
292 				return (error);
293 			}
294 			if (MOUNTEDSOFTDEP(mp))
295 				softdep_unmount(mp);
296 			DROP_GIANT();
297 			g_topology_lock();
298 			/*
299 			 * Drop our write and exclusive access.
300 			 */
301 			g_access(ump->um_cp, 0, -1, -1);
302 			g_topology_unlock();
303 			PICKUP_GIANT();
304 			fs->fs_ronly = 1;
305 			MNT_ILOCK(mp);
306 			mp->mnt_flag |= MNT_RDONLY;
307 			MNT_IUNLOCK(mp);
308 			/*
309 			 * Allow the writers to note that filesystem
310 			 * is ro now.
311 			 */
312 			vfs_write_resume(mp, 0);
313 		}
314 		if ((mp->mnt_flag & MNT_RELOAD) &&
315 		    (error = ffs_reload(mp, td, 0)) != 0)
316 			return (error);
317 		if (fs->fs_ronly &&
318 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
319 			/*
320 			 * If we are running a checker, do not allow upgrade.
321 			 */
322 			if (ump->um_fsckpid > 0) {
323 				vfs_mount_error(mp,
324 				    "Active checker, cannot upgrade to write");
325 				return (EINVAL);
326 			}
327 			/*
328 			 * If upgrade to read-write by non-root, then verify
329 			 * that user has necessary permissions on the device.
330 			 */
331 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
332 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
333 			    td->td_ucred, td);
334 			if (error)
335 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
336 			if (error) {
337 				VOP_UNLOCK(devvp, 0);
338 				return (error);
339 			}
340 			VOP_UNLOCK(devvp, 0);
341 			fs->fs_flags &= ~FS_UNCLEAN;
342 			if (fs->fs_clean == 0) {
343 				fs->fs_flags |= FS_UNCLEAN;
344 				if ((mp->mnt_flag & MNT_FORCE) ||
345 				    ((fs->fs_flags &
346 				     (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
347 				     (fs->fs_flags & FS_DOSOFTDEP))) {
348 					printf("WARNING: %s was not properly "
349 					   "dismounted\n", fs->fs_fsmnt);
350 				} else {
351 					vfs_mount_error(mp,
352 					   "R/W mount of %s denied. %s.%s",
353 					   fs->fs_fsmnt,
354 					   "Filesystem is not clean - run fsck",
355 					   (fs->fs_flags & FS_SUJ) == 0 ? "" :
356 					   " Forced mount will invalidate"
357 					   " journal contents");
358 					return (EPERM);
359 				}
360 			}
361 			DROP_GIANT();
362 			g_topology_lock();
363 			/*
364 			 * Request exclusive write access.
365 			 */
366 			error = g_access(ump->um_cp, 0, 1, 1);
367 			g_topology_unlock();
368 			PICKUP_GIANT();
369 			if (error)
370 				return (error);
371 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
372 				return (error);
373 			fs->fs_ronly = 0;
374 			MNT_ILOCK(mp);
375 			mp->mnt_flag &= ~MNT_RDONLY;
376 			MNT_IUNLOCK(mp);
377 			fs->fs_mtime = time_second;
378 			/* check to see if we need to start softdep */
379 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
380 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
381 				vn_finished_write(mp);
382 				return (error);
383 			}
384 			fs->fs_clean = 0;
385 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
386 				vn_finished_write(mp);
387 				return (error);
388 			}
389 			if (fs->fs_snapinum[0] != 0)
390 				ffs_snapshot_mount(mp);
391 			vn_finished_write(mp);
392 		}
393 		/*
394 		 * Soft updates is incompatible with "async",
395 		 * so if we are doing softupdates stop the user
396 		 * from setting the async flag in an update.
397 		 * Softdep_mount() clears it in an initial mount
398 		 * or ro->rw remount.
399 		 */
400 		if (MOUNTEDSOFTDEP(mp)) {
401 			/* XXX: Reset too late ? */
402 			MNT_ILOCK(mp);
403 			mp->mnt_flag &= ~MNT_ASYNC;
404 			MNT_IUNLOCK(mp);
405 		}
406 		/*
407 		 * Keep MNT_ACLS flag if it is stored in superblock.
408 		 */
409 		if ((fs->fs_flags & FS_ACLS) != 0) {
410 			/* XXX: Set too late ? */
411 			MNT_ILOCK(mp);
412 			mp->mnt_flag |= MNT_ACLS;
413 			MNT_IUNLOCK(mp);
414 		}
415 
416 		if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
417 			/* XXX: Set too late ? */
418 			MNT_ILOCK(mp);
419 			mp->mnt_flag |= MNT_NFS4ACLS;
420 			MNT_IUNLOCK(mp);
421 		}
422 		/*
423 		 * If this is a request from fsck to clean up the filesystem,
424 		 * then allow the specified pid to proceed.
425 		 */
426 		if (fsckpid > 0) {
427 			if (ump->um_fsckpid != 0) {
428 				vfs_mount_error(mp,
429 				    "Active checker already running on %s",
430 				    fs->fs_fsmnt);
431 				return (EINVAL);
432 			}
433 			KASSERT(MOUNTEDSOFTDEP(mp) == 0,
434 			    ("soft updates enabled on read-only file system"));
435 			DROP_GIANT();
436 			g_topology_lock();
437 			/*
438 			 * Request write access.
439 			 */
440 			error = g_access(ump->um_cp, 0, 1, 0);
441 			g_topology_unlock();
442 			PICKUP_GIANT();
443 			if (error) {
444 				vfs_mount_error(mp,
445 				    "Checker activation failed on %s",
446 				    fs->fs_fsmnt);
447 				return (error);
448 			}
449 			ump->um_fsckpid = fsckpid;
450 			if (fs->fs_snapinum[0] != 0)
451 				ffs_snapshot_mount(mp);
452 			fs->fs_mtime = time_second;
453 			fs->fs_fmod = 1;
454 			fs->fs_clean = 0;
455 			(void) ffs_sbupdate(ump, MNT_WAIT, 0);
456 		}
457 
458 		/*
459 		 * If this is a snapshot request, take the snapshot.
460 		 */
461 		if (mp->mnt_flag & MNT_SNAPSHOT)
462 			return (ffs_snapshot(mp, fspec));
463 	}
464 
465 	/*
466 	 * Not an update, or updating the name: look up the name
467 	 * and verify that it refers to a sensible disk device.
468 	 */
469 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
470 	if ((error = namei(&ndp)) != 0)
471 		return (error);
472 	NDFREE(&ndp, NDF_ONLY_PNBUF);
473 	devvp = ndp.ni_vp;
474 	if (!vn_isdisk(devvp, &error)) {
475 		vput(devvp);
476 		return (error);
477 	}
478 
479 	/*
480 	 * If mount by non-root, then verify that user has necessary
481 	 * permissions on the device.
482 	 */
483 	accmode = VREAD;
484 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
485 		accmode |= VWRITE;
486 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
487 	if (error)
488 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
489 	if (error) {
490 		vput(devvp);
491 		return (error);
492 	}
493 
494 	if (mp->mnt_flag & MNT_UPDATE) {
495 		/*
496 		 * Update only
497 		 *
498 		 * If it's not the same vnode, or at least the same device
499 		 * then it's not correct.
500 		 */
501 
502 		if (devvp->v_rdev != ump->um_devvp->v_rdev)
503 			error = EINVAL;	/* needs translation */
504 		vput(devvp);
505 		if (error)
506 			return (error);
507 	} else {
508 		/*
509 		 * New mount
510 		 *
511 		 * We need the name for the mount point (also used for
512 		 * "last mounted on") copied in. If an error occurs,
513 		 * the mount point is discarded by the upper level code.
514 		 * Note that vfs_mount() populates f_mntonname for us.
515 		 */
516 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
517 			vrele(devvp);
518 			return (error);
519 		}
520 		if (fsckpid > 0) {
521 			KASSERT(MOUNTEDSOFTDEP(mp) == 0,
522 			    ("soft updates enabled on read-only file system"));
523 			ump = VFSTOUFS(mp);
524 			fs = ump->um_fs;
525 			DROP_GIANT();
526 			g_topology_lock();
527 			/*
528 			 * Request write access.
529 			 */
530 			error = g_access(ump->um_cp, 0, 1, 0);
531 			g_topology_unlock();
532 			PICKUP_GIANT();
533 			if (error) {
534 				printf("WARNING: %s: Checker activation "
535 				    "failed\n", fs->fs_fsmnt);
536 			} else {
537 				ump->um_fsckpid = fsckpid;
538 				if (fs->fs_snapinum[0] != 0)
539 					ffs_snapshot_mount(mp);
540 				fs->fs_mtime = time_second;
541 				fs->fs_clean = 0;
542 				(void) ffs_sbupdate(ump, MNT_WAIT, 0);
543 			}
544 		}
545 	}
546 	vfs_mountedfrom(mp, fspec);
547 	return (0);
548 }
549 
550 /*
551  * Compatibility with old mount system call.
552  */
553 
554 static int
555 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
556 {
557 	struct ufs_args args;
558 	struct export_args exp;
559 	int error;
560 
561 	if (data == NULL)
562 		return (EINVAL);
563 	error = copyin(data, &args, sizeof args);
564 	if (error)
565 		return (error);
566 	vfs_oexport_conv(&args.export, &exp);
567 
568 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
569 	ma = mount_arg(ma, "export", &exp, sizeof(exp));
570 	error = kernel_mount(ma, flags);
571 
572 	return (error);
573 }
574 
575 /*
576  * Reload all incore data for a filesystem (used after running fsck on
577  * the root filesystem and finding things to fix). If the 'force' flag
578  * is 0, the filesystem must be mounted read-only.
579  *
580  * Things to do to update the mount:
581  *	1) invalidate all cached meta-data.
582  *	2) re-read superblock from disk.
583  *	3) re-read summary information from disk.
584  *	4) invalidate all inactive vnodes.
585  *	5) invalidate all cached file data.
586  *	6) re-read inode data for all active vnodes.
587  */
588 int
589 ffs_reload(struct mount *mp, struct thread *td, int force)
590 {
591 	struct vnode *vp, *mvp, *devvp;
592 	struct inode *ip;
593 	void *space;
594 	struct buf *bp;
595 	struct fs *fs, *newfs;
596 	struct ufsmount *ump;
597 	ufs2_daddr_t sblockloc;
598 	int i, blks, size, error;
599 	int32_t *lp;
600 
601 	ump = VFSTOUFS(mp);
602 
603 	MNT_ILOCK(mp);
604 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && force == 0) {
605 		MNT_IUNLOCK(mp);
606 		return (EINVAL);
607 	}
608 	MNT_IUNLOCK(mp);
609 
610 	/*
611 	 * Step 1: invalidate all cached meta-data.
612 	 */
613 	devvp = VFSTOUFS(mp)->um_devvp;
614 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
615 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
616 		panic("ffs_reload: dirty1");
617 	VOP_UNLOCK(devvp, 0);
618 
619 	/*
620 	 * Step 2: re-read superblock from disk.
621 	 */
622 	fs = VFSTOUFS(mp)->um_fs;
623 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
624 	    NOCRED, &bp)) != 0)
625 		return (error);
626 	newfs = (struct fs *)bp->b_data;
627 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
628 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
629 	    newfs->fs_bsize > MAXBSIZE ||
630 	    newfs->fs_bsize < sizeof(struct fs)) {
631 			brelse(bp);
632 			return (EIO);		/* XXX needs translation */
633 	}
634 	/*
635 	 * Copy pointer fields back into superblock before copying in	XXX
636 	 * new superblock. These should really be in the ufsmount.	XXX
637 	 * Note that important parameters (eg fs_ncg) are unchanged.
638 	 */
639 	newfs->fs_csp = fs->fs_csp;
640 	newfs->fs_maxcluster = fs->fs_maxcluster;
641 	newfs->fs_contigdirs = fs->fs_contigdirs;
642 	newfs->fs_active = fs->fs_active;
643 	newfs->fs_ronly = fs->fs_ronly;
644 	sblockloc = fs->fs_sblockloc;
645 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
646 	brelse(bp);
647 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
648 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
649 	UFS_LOCK(ump);
650 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
651 		printf("WARNING: %s: reload pending error: blocks %jd "
652 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
653 		    fs->fs_pendinginodes);
654 		fs->fs_pendingblocks = 0;
655 		fs->fs_pendinginodes = 0;
656 	}
657 	UFS_UNLOCK(ump);
658 
659 	/*
660 	 * Step 3: re-read summary information from disk.
661 	 */
662 	size = fs->fs_cssize;
663 	blks = howmany(size, fs->fs_fsize);
664 	if (fs->fs_contigsumsize > 0)
665 		size += fs->fs_ncg * sizeof(int32_t);
666 	size += fs->fs_ncg * sizeof(u_int8_t);
667 	free(fs->fs_csp, M_UFSMNT);
668 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
669 	fs->fs_csp = space;
670 	for (i = 0; i < blks; i += fs->fs_frag) {
671 		size = fs->fs_bsize;
672 		if (i + fs->fs_frag > blks)
673 			size = (blks - i) * fs->fs_fsize;
674 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
675 		    NOCRED, &bp);
676 		if (error)
677 			return (error);
678 		bcopy(bp->b_data, space, (u_int)size);
679 		space = (char *)space + size;
680 		brelse(bp);
681 	}
682 	/*
683 	 * We no longer know anything about clusters per cylinder group.
684 	 */
685 	if (fs->fs_contigsumsize > 0) {
686 		fs->fs_maxcluster = lp = space;
687 		for (i = 0; i < fs->fs_ncg; i++)
688 			*lp++ = fs->fs_contigsumsize;
689 		space = lp;
690 	}
691 	size = fs->fs_ncg * sizeof(u_int8_t);
692 	fs->fs_contigdirs = (u_int8_t *)space;
693 	bzero(fs->fs_contigdirs, size);
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, size, ronly;
756 	int32_t *lp;
757 	struct ucred *cred;
758 	struct g_consumer *cp;
759 	struct mount *nmp;
760 
761 	bp = NULL;
762 	ump = NULL;
763 	cred = td ? td->td_ucred : NOCRED;
764 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
765 
766 	dev = devvp->v_rdev;
767 	dev_ref(dev);
768 	DROP_GIANT();
769 	g_topology_lock();
770 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
771 	g_topology_unlock();
772 	PICKUP_GIANT();
773 	VOP_UNLOCK(devvp, 0);
774 	if (error)
775 		goto out;
776 	if (devvp->v_rdev->si_iosize_max != 0)
777 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
778 	if (mp->mnt_iosize_max > MAXPHYS)
779 		mp->mnt_iosize_max = MAXPHYS;
780 
781 	devvp->v_bufobj.bo_ops = &ffs_ops;
782 
783 	fs = NULL;
784 	sblockloc = 0;
785 	/*
786 	 * Try reading the superblock in each of its possible locations.
787 	 */
788 	for (i = 0; sblock_try[i] != -1; i++) {
789 		if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
790 			error = EINVAL;
791 			vfs_mount_error(mp,
792 			    "Invalid sectorsize %d for superblock size %d",
793 			    cp->provider->sectorsize, SBLOCKSIZE);
794 			goto out;
795 		}
796 		if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
797 		    cred, &bp)) != 0)
798 			goto out;
799 		fs = (struct fs *)bp->b_data;
800 		sblockloc = sblock_try[i];
801 		if ((fs->fs_magic == FS_UFS1_MAGIC ||
802 		     (fs->fs_magic == FS_UFS2_MAGIC &&
803 		      (fs->fs_sblockloc == sblockloc ||
804 		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
805 		    fs->fs_bsize <= MAXBSIZE &&
806 		    fs->fs_bsize >= sizeof(struct fs))
807 			break;
808 		brelse(bp);
809 		bp = NULL;
810 	}
811 	if (sblock_try[i] == -1) {
812 		error = EINVAL;		/* XXX needs translation */
813 		goto out;
814 	}
815 	fs->fs_fmod = 0;
816 	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indicies */
817 	fs->fs_flags &= ~FS_UNCLEAN;
818 	if (fs->fs_clean == 0) {
819 		fs->fs_flags |= FS_UNCLEAN;
820 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
821 		    ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
822 		     (fs->fs_flags & FS_DOSOFTDEP))) {
823 			printf("WARNING: %s was not properly dismounted\n",
824 			    fs->fs_fsmnt);
825 		} else {
826 			vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
827 			    fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
828 			    (fs->fs_flags & FS_SUJ) == 0 ? "" :
829 			    " Forced mount will invalidate journal contents");
830 			error = EPERM;
831 			goto out;
832 		}
833 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
834 		    (mp->mnt_flag & MNT_FORCE)) {
835 			printf("WARNING: %s: lost blocks %jd files %d\n",
836 			    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
837 			    fs->fs_pendinginodes);
838 			fs->fs_pendingblocks = 0;
839 			fs->fs_pendinginodes = 0;
840 		}
841 	}
842 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
843 		printf("WARNING: %s: mount pending error: blocks %jd "
844 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
845 		    fs->fs_pendinginodes);
846 		fs->fs_pendingblocks = 0;
847 		fs->fs_pendinginodes = 0;
848 	}
849 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
850 #ifdef UFS_GJOURNAL
851 		/*
852 		 * Get journal provider name.
853 		 */
854 		size = 1024;
855 		mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
856 		if (g_io_getattr("GJOURNAL::provider", cp, &size,
857 		    mp->mnt_gjprovider) == 0) {
858 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
859 			    M_UFSMNT, M_WAITOK);
860 			MNT_ILOCK(mp);
861 			mp->mnt_flag |= MNT_GJOURNAL;
862 			MNT_IUNLOCK(mp);
863 		} else {
864 			printf("WARNING: %s: GJOURNAL flag on fs "
865 			    "but no gjournal provider below\n",
866 			    mp->mnt_stat.f_mntonname);
867 			free(mp->mnt_gjprovider, M_UFSMNT);
868 			mp->mnt_gjprovider = NULL;
869 		}
870 #else
871 		printf("WARNING: %s: GJOURNAL flag on fs but no "
872 		    "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
873 #endif
874 	} else {
875 		mp->mnt_gjprovider = NULL;
876 	}
877 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
878 	ump->um_cp = cp;
879 	ump->um_bo = &devvp->v_bufobj;
880 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
881 	if (fs->fs_magic == FS_UFS1_MAGIC) {
882 		ump->um_fstype = UFS1;
883 		ump->um_balloc = ffs_balloc_ufs1;
884 	} else {
885 		ump->um_fstype = UFS2;
886 		ump->um_balloc = ffs_balloc_ufs2;
887 	}
888 	ump->um_blkatoff = ffs_blkatoff;
889 	ump->um_truncate = ffs_truncate;
890 	ump->um_update = ffs_update;
891 	ump->um_valloc = ffs_valloc;
892 	ump->um_vfree = ffs_vfree;
893 	ump->um_ifree = ffs_ifree;
894 	ump->um_rdonly = ffs_rdonly;
895 	ump->um_snapgone = ffs_snapgone;
896 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
897 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
898 	if (fs->fs_sbsize < SBLOCKSIZE)
899 		bp->b_flags |= B_INVAL | B_NOCACHE;
900 	brelse(bp);
901 	bp = NULL;
902 	fs = ump->um_fs;
903 	ffs_oldfscompat_read(fs, ump, sblockloc);
904 	fs->fs_ronly = ronly;
905 	size = fs->fs_cssize;
906 	blks = howmany(size, fs->fs_fsize);
907 	if (fs->fs_contigsumsize > 0)
908 		size += fs->fs_ncg * sizeof(int32_t);
909 	size += fs->fs_ncg * sizeof(u_int8_t);
910 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
911 	fs->fs_csp = space;
912 	for (i = 0; i < blks; i += fs->fs_frag) {
913 		size = fs->fs_bsize;
914 		if (i + fs->fs_frag > blks)
915 			size = (blks - i) * fs->fs_fsize;
916 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
917 		    cred, &bp)) != 0) {
918 			free(fs->fs_csp, M_UFSMNT);
919 			goto out;
920 		}
921 		bcopy(bp->b_data, space, (u_int)size);
922 		space = (char *)space + size;
923 		brelse(bp);
924 		bp = NULL;
925 	}
926 	if (fs->fs_contigsumsize > 0) {
927 		fs->fs_maxcluster = lp = space;
928 		for (i = 0; i < fs->fs_ncg; i++)
929 			*lp++ = fs->fs_contigsumsize;
930 		space = lp;
931 	}
932 	size = fs->fs_ncg * sizeof(u_int8_t);
933 	fs->fs_contigdirs = (u_int8_t *)space;
934 	bzero(fs->fs_contigdirs, size);
935 	fs->fs_active = NULL;
936 	mp->mnt_data = ump;
937 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
938 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
939 	nmp = NULL;
940 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
941 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
942 		if (nmp)
943 			vfs_rel(nmp);
944 		vfs_getnewfsid(mp);
945 	}
946 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
947 	MNT_ILOCK(mp);
948 	mp->mnt_flag |= MNT_LOCAL;
949 	MNT_IUNLOCK(mp);
950 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
951 #ifdef MAC
952 		MNT_ILOCK(mp);
953 		mp->mnt_flag |= MNT_MULTILABEL;
954 		MNT_IUNLOCK(mp);
955 #else
956 		printf("WARNING: %s: multilabel flag on fs but "
957 		    "no MAC support\n", mp->mnt_stat.f_mntonname);
958 #endif
959 	}
960 	if ((fs->fs_flags & FS_ACLS) != 0) {
961 #ifdef UFS_ACL
962 		MNT_ILOCK(mp);
963 
964 		if (mp->mnt_flag & MNT_NFS4ACLS)
965 			printf("WARNING: %s: ACLs flag on fs conflicts with "
966 			    "\"nfsv4acls\" mount option; option ignored\n",
967 			    mp->mnt_stat.f_mntonname);
968 		mp->mnt_flag &= ~MNT_NFS4ACLS;
969 		mp->mnt_flag |= MNT_ACLS;
970 
971 		MNT_IUNLOCK(mp);
972 #else
973 		printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
974 		    mp->mnt_stat.f_mntonname);
975 #endif
976 	}
977 	if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
978 #ifdef UFS_ACL
979 		MNT_ILOCK(mp);
980 
981 		if (mp->mnt_flag & MNT_ACLS)
982 			printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
983 			    "with \"acls\" mount option; option ignored\n",
984 			    mp->mnt_stat.f_mntonname);
985 		mp->mnt_flag &= ~MNT_ACLS;
986 		mp->mnt_flag |= MNT_NFS4ACLS;
987 
988 		MNT_IUNLOCK(mp);
989 #else
990 		printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
991 		    "ACLs support\n", mp->mnt_stat.f_mntonname);
992 #endif
993 	}
994 	if ((fs->fs_flags & FS_TRIM) != 0) {
995 		size = sizeof(int);
996 		if (g_io_getattr("GEOM::candelete", cp, &size,
997 		    &ump->um_candelete) == 0) {
998 			if (!ump->um_candelete)
999 				printf("WARNING: %s: TRIM flag on fs but disk "
1000 				    "does not support TRIM\n",
1001 				    mp->mnt_stat.f_mntonname);
1002 		} else {
1003 			printf("WARNING: %s: TRIM flag on fs but disk does "
1004 			    "not confirm that it supports TRIM\n",
1005 			    mp->mnt_stat.f_mntonname);
1006 			ump->um_candelete = 0;
1007 		}
1008 	}
1009 
1010 	ump->um_mountp = mp;
1011 	ump->um_dev = dev;
1012 	ump->um_devvp = devvp;
1013 	ump->um_nindir = fs->fs_nindir;
1014 	ump->um_bptrtodb = fs->fs_fsbtodb;
1015 	ump->um_seqinc = fs->fs_frag;
1016 	for (i = 0; i < MAXQUOTAS; i++)
1017 		ump->um_quotas[i] = NULLVP;
1018 #ifdef UFS_EXTATTR
1019 	ufs_extattr_uepm_init(&ump->um_extattr);
1020 #endif
1021 	/*
1022 	 * Set FS local "last mounted on" information (NULL pad)
1023 	 */
1024 	bzero(fs->fs_fsmnt, MAXMNTLEN);
1025 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1026 	mp->mnt_stat.f_iosize = fs->fs_bsize;
1027 
1028 	if (mp->mnt_flag & MNT_ROOTFS) {
1029 		/*
1030 		 * Root mount; update timestamp in mount structure.
1031 		 * this will be used by the common root mount code
1032 		 * to update the system clock.
1033 		 */
1034 		mp->mnt_time = fs->fs_time;
1035 	}
1036 
1037 	if (ronly == 0) {
1038 		fs->fs_mtime = time_second;
1039 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
1040 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1041 			free(fs->fs_csp, M_UFSMNT);
1042 			ffs_flushfiles(mp, FORCECLOSE, td);
1043 			goto out;
1044 		}
1045 		if (devvp->v_type == VCHR && devvp->v_rdev != NULL)
1046 			devvp->v_rdev->si_mountpt = mp;
1047 		if (fs->fs_snapinum[0] != 0)
1048 			ffs_snapshot_mount(mp);
1049 		fs->fs_fmod = 1;
1050 		fs->fs_clean = 0;
1051 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
1052 	}
1053 	/*
1054 	 * Initialize filesystem stat information in mount struct.
1055 	 */
1056 	MNT_ILOCK(mp);
1057 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1058 	    MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_SUSPENDABLE;
1059 	MNT_IUNLOCK(mp);
1060 #ifdef UFS_EXTATTR
1061 #ifdef UFS_EXTATTR_AUTOSTART
1062 	/*
1063 	 *
1064 	 * Auto-starting does the following:
1065 	 *	- check for /.attribute in the fs, and extattr_start if so
1066 	 *	- for each file in .attribute, enable that file with
1067 	 * 	  an attribute of the same name.
1068 	 * Not clear how to report errors -- probably eat them.
1069 	 * This would all happen while the filesystem was busy/not
1070 	 * available, so would effectively be "atomic".
1071 	 */
1072 	(void) ufs_extattr_autostart(mp, td);
1073 #endif /* !UFS_EXTATTR_AUTOSTART */
1074 #endif /* !UFS_EXTATTR */
1075 	return (0);
1076 out:
1077 	if (bp)
1078 		brelse(bp);
1079 	if (cp != NULL) {
1080 		DROP_GIANT();
1081 		g_topology_lock();
1082 		g_vfs_close(cp);
1083 		g_topology_unlock();
1084 		PICKUP_GIANT();
1085 	}
1086 	if (ump) {
1087 		mtx_destroy(UFS_MTX(ump));
1088 		if (mp->mnt_gjprovider != NULL) {
1089 			free(mp->mnt_gjprovider, M_UFSMNT);
1090 			mp->mnt_gjprovider = NULL;
1091 		}
1092 		free(ump->um_fs, M_UFSMNT);
1093 		free(ump, M_UFSMNT);
1094 		mp->mnt_data = NULL;
1095 	}
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 	DROP_GIANT();
1264 	g_topology_lock();
1265 	if (ump->um_fsckpid > 0) {
1266 		/*
1267 		 * Return to normal read-only mode.
1268 		 */
1269 		error = g_access(ump->um_cp, 0, -1, 0);
1270 		ump->um_fsckpid = 0;
1271 	}
1272 	g_vfs_close(ump->um_cp);
1273 	g_topology_unlock();
1274 	PICKUP_GIANT();
1275 	if (ump->um_devvp->v_type == VCHR && ump->um_devvp->v_rdev != NULL)
1276 		ump->um_devvp->v_rdev->si_mountpt = NULL;
1277 	vrele(ump->um_devvp);
1278 	dev_rel(ump->um_dev);
1279 	mtx_destroy(UFS_MTX(ump));
1280 	if (mp->mnt_gjprovider != NULL) {
1281 		free(mp->mnt_gjprovider, M_UFSMNT);
1282 		mp->mnt_gjprovider = NULL;
1283 	}
1284 	free(fs->fs_csp, M_UFSMNT);
1285 	free(fs, M_UFSMNT);
1286 	free(ump, M_UFSMNT);
1287 	mp->mnt_data = NULL;
1288 	MNT_ILOCK(mp);
1289 	mp->mnt_flag &= ~MNT_LOCAL;
1290 	MNT_IUNLOCK(mp);
1291 	return (error);
1292 
1293 fail:
1294 	if (susp)
1295 		vfs_write_resume(mp, VR_START_WRITE);
1296 fail1:
1297 #ifdef UFS_EXTATTR
1298 	if (e_restart) {
1299 		ufs_extattr_uepm_init(&ump->um_extattr);
1300 #ifdef UFS_EXTATTR_AUTOSTART
1301 		(void) ufs_extattr_autostart(mp, td);
1302 #endif
1303 	}
1304 #endif
1305 
1306 	return (error);
1307 }
1308 
1309 /*
1310  * Flush out all the files in a filesystem.
1311  */
1312 int
1313 ffs_flushfiles(mp, flags, td)
1314 	struct mount *mp;
1315 	int flags;
1316 	struct thread *td;
1317 {
1318 	struct ufsmount *ump;
1319 	int qerror, error;
1320 
1321 	ump = VFSTOUFS(mp);
1322 	qerror = 0;
1323 #ifdef QUOTA
1324 	if (mp->mnt_flag & MNT_QUOTA) {
1325 		int i;
1326 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1327 		if (error)
1328 			return (error);
1329 		for (i = 0; i < MAXQUOTAS; i++) {
1330 			error = quotaoff(td, mp, i);
1331 			if (error != 0) {
1332 				if ((flags & EARLYFLUSH) == 0)
1333 					return (error);
1334 				else
1335 					qerror = error;
1336 			}
1337 		}
1338 
1339 		/*
1340 		 * Here we fall through to vflush again to ensure that
1341 		 * we have gotten rid of all the system vnodes, unless
1342 		 * quotas must not be closed.
1343 		 */
1344 	}
1345 #endif
1346 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1347 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1348 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1349 			return (error);
1350 		ffs_snapshot_unmount(mp);
1351 		flags |= FORCECLOSE;
1352 		/*
1353 		 * Here we fall through to vflush again to ensure
1354 		 * that we have gotten rid of all the system vnodes.
1355 		 */
1356 	}
1357 
1358 	/*
1359 	 * Do not close system files if quotas were not closed, to be
1360 	 * able to sync the remaining dquots.  The freeblks softupdate
1361 	 * workitems might hold a reference on a dquot, preventing
1362 	 * quotaoff() from completing.  Next round of
1363 	 * softdep_flushworklist() iteration should process the
1364 	 * blockers, allowing the next run of quotaoff() to finally
1365 	 * flush held dquots.
1366 	 *
1367 	 * Otherwise, flush all the files.
1368 	 */
1369 	if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1370 		return (error);
1371 
1372 	/*
1373 	 * Flush filesystem metadata.
1374 	 */
1375 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1376 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1377 	VOP_UNLOCK(ump->um_devvp, 0);
1378 	return (error);
1379 }
1380 
1381 /*
1382  * Get filesystem statistics.
1383  */
1384 static int
1385 ffs_statfs(mp, sbp)
1386 	struct mount *mp;
1387 	struct statfs *sbp;
1388 {
1389 	struct ufsmount *ump;
1390 	struct fs *fs;
1391 
1392 	ump = VFSTOUFS(mp);
1393 	fs = ump->um_fs;
1394 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1395 		panic("ffs_statfs");
1396 	sbp->f_version = STATFS_VERSION;
1397 	sbp->f_bsize = fs->fs_fsize;
1398 	sbp->f_iosize = fs->fs_bsize;
1399 	sbp->f_blocks = fs->fs_dsize;
1400 	UFS_LOCK(ump);
1401 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1402 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1403 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1404 	    dbtofsb(fs, fs->fs_pendingblocks);
1405 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1406 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1407 	UFS_UNLOCK(ump);
1408 	sbp->f_namemax = NAME_MAX;
1409 	return (0);
1410 }
1411 
1412 /*
1413  * For a lazy sync, we only care about access times, quotas and the
1414  * superblock.  Other filesystem changes are already converted to
1415  * cylinder group blocks or inode blocks updates and are written to
1416  * disk by syncer.
1417  */
1418 static int
1419 ffs_sync_lazy(mp)
1420      struct mount *mp;
1421 {
1422 	struct vnode *mvp, *vp;
1423 	struct inode *ip;
1424 	struct thread *td;
1425 	int allerror, error;
1426 
1427 	allerror = 0;
1428 	td = curthread;
1429 	if ((mp->mnt_flag & MNT_NOATIME) != 0)
1430 		goto qupdate;
1431 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1432 		if (vp->v_type == VNON) {
1433 			VI_UNLOCK(vp);
1434 			continue;
1435 		}
1436 		ip = VTOI(vp);
1437 
1438 		/*
1439 		 * The IN_ACCESS flag is converted to IN_MODIFIED by
1440 		 * ufs_close() and ufs_getattr() by the calls to
1441 		 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1442 		 * Test also all the other timestamp flags too, to pick up
1443 		 * any other cases that could be missed.
1444 		 */
1445 		if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1446 		    IN_UPDATE)) == 0) {
1447 			VI_UNLOCK(vp);
1448 			continue;
1449 		}
1450 		if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
1451 		    td)) != 0)
1452 			continue;
1453 		error = ffs_update(vp, 0);
1454 		if (error != 0)
1455 			allerror = error;
1456 		vput(vp);
1457 	}
1458 
1459 qupdate:
1460 #ifdef QUOTA
1461 	qsync(mp);
1462 #endif
1463 
1464 	if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1465 	    (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1466 		allerror = error;
1467 	return (allerror);
1468 }
1469 
1470 /*
1471  * Go through the disk queues to initiate sandbagged IO;
1472  * go through the inodes to write those that have been modified;
1473  * initiate the writing of the super block if it has been modified.
1474  *
1475  * Note: we are always called with the filesystem marked busy using
1476  * vfs_busy().
1477  */
1478 static int
1479 ffs_sync(mp, waitfor)
1480 	struct mount *mp;
1481 	int waitfor;
1482 {
1483 	struct vnode *mvp, *vp, *devvp;
1484 	struct thread *td;
1485 	struct inode *ip;
1486 	struct ufsmount *ump = VFSTOUFS(mp);
1487 	struct fs *fs;
1488 	int error, count, wait, lockreq, allerror = 0;
1489 	int suspend;
1490 	int suspended;
1491 	int secondary_writes;
1492 	int secondary_accwrites;
1493 	int softdep_deps;
1494 	int softdep_accdeps;
1495 	struct bufobj *bo;
1496 
1497 	wait = 0;
1498 	suspend = 0;
1499 	suspended = 0;
1500 	td = curthread;
1501 	fs = ump->um_fs;
1502 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1503 		panic("%s: ffs_sync: modification on read-only filesystem",
1504 		    fs->fs_fsmnt);
1505 	if (waitfor == MNT_LAZY)
1506 		return (ffs_sync_lazy(mp));
1507 
1508 	/*
1509 	 * Write back each (modified) inode.
1510 	 */
1511 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1512 	if (waitfor == MNT_SUSPEND) {
1513 		suspend = 1;
1514 		waitfor = MNT_WAIT;
1515 	}
1516 	if (waitfor == MNT_WAIT) {
1517 		wait = 1;
1518 		lockreq = LK_EXCLUSIVE;
1519 	}
1520 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1521 loop:
1522 	/* Grab snapshot of secondary write counts */
1523 	MNT_ILOCK(mp);
1524 	secondary_writes = mp->mnt_secondary_writes;
1525 	secondary_accwrites = mp->mnt_secondary_accwrites;
1526 	MNT_IUNLOCK(mp);
1527 
1528 	/* Grab snapshot of softdep dependency counts */
1529 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1530 
1531 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1532 		/*
1533 		 * Depend on the vnode interlock to keep things stable enough
1534 		 * for a quick test.  Since there might be hundreds of
1535 		 * thousands of vnodes, we cannot afford even a subroutine
1536 		 * call unless there's a good chance that we have work to do.
1537 		 */
1538 		if (vp->v_type == VNON) {
1539 			VI_UNLOCK(vp);
1540 			continue;
1541 		}
1542 		ip = VTOI(vp);
1543 		if ((ip->i_flag &
1544 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1545 		    vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1546 			VI_UNLOCK(vp);
1547 			continue;
1548 		}
1549 		if ((error = vget(vp, lockreq, td)) != 0) {
1550 			if (error == ENOENT || error == ENOLCK) {
1551 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1552 				goto loop;
1553 			}
1554 			continue;
1555 		}
1556 		if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0)
1557 			allerror = error;
1558 		vput(vp);
1559 	}
1560 	/*
1561 	 * Force stale filesystem control information to be flushed.
1562 	 */
1563 	if (waitfor == MNT_WAIT) {
1564 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1565 			allerror = error;
1566 		/* Flushed work items may create new vnodes to clean */
1567 		if (allerror == 0 && count)
1568 			goto loop;
1569 	}
1570 #ifdef QUOTA
1571 	qsync(mp);
1572 #endif
1573 
1574 	devvp = ump->um_devvp;
1575 	bo = &devvp->v_bufobj;
1576 	BO_LOCK(bo);
1577 	if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1578 		BO_UNLOCK(bo);
1579 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1580 		if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0)
1581 			allerror = error;
1582 		VOP_UNLOCK(devvp, 0);
1583 		if (allerror == 0 && waitfor == MNT_WAIT)
1584 			goto loop;
1585 	} else if (suspend != 0) {
1586 		if (softdep_check_suspend(mp,
1587 					  devvp,
1588 					  softdep_deps,
1589 					  softdep_accdeps,
1590 					  secondary_writes,
1591 					  secondary_accwrites) != 0) {
1592 			MNT_IUNLOCK(mp);
1593 			goto loop;	/* More work needed */
1594 		}
1595 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1596 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1597 		MNT_IUNLOCK(mp);
1598 		suspended = 1;
1599 	} else
1600 		BO_UNLOCK(bo);
1601 	/*
1602 	 * Write back modified superblock.
1603 	 */
1604 	if (fs->fs_fmod != 0 &&
1605 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1606 		allerror = error;
1607 	return (allerror);
1608 }
1609 
1610 int
1611 ffs_vget(mp, ino, flags, vpp)
1612 	struct mount *mp;
1613 	ino_t ino;
1614 	int flags;
1615 	struct vnode **vpp;
1616 {
1617 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1618 }
1619 
1620 int
1621 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1622 	struct mount *mp;
1623 	ino_t ino;
1624 	int flags;
1625 	struct vnode **vpp;
1626 	int ffs_flags;
1627 {
1628 	struct fs *fs;
1629 	struct inode *ip;
1630 	struct ufsmount *ump;
1631 	struct buf *bp;
1632 	struct vnode *vp;
1633 	struct cdev *dev;
1634 	int error;
1635 
1636 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1637 	if (error || *vpp != NULL)
1638 		return (error);
1639 
1640 	/*
1641 	 * We must promote to an exclusive lock for vnode creation.  This
1642 	 * can happen if lookup is passed LOCKSHARED.
1643 	 */
1644 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1645 		flags &= ~LK_TYPE_MASK;
1646 		flags |= LK_EXCLUSIVE;
1647 	}
1648 
1649 	/*
1650 	 * We do not lock vnode creation as it is believed to be too
1651 	 * expensive for such rare case as simultaneous creation of vnode
1652 	 * for same ino by different processes. We just allow them to race
1653 	 * and check later to decide who wins. Let the race begin!
1654 	 */
1655 
1656 	ump = VFSTOUFS(mp);
1657 	dev = ump->um_dev;
1658 	fs = ump->um_fs;
1659 	ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1660 
1661 	/* Allocate a new vnode/inode. */
1662 	if (fs->fs_magic == FS_UFS1_MAGIC)
1663 		error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp);
1664 	else
1665 		error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp);
1666 	if (error) {
1667 		*vpp = NULL;
1668 		uma_zfree(uma_inode, ip);
1669 		return (error);
1670 	}
1671 	/*
1672 	 * FFS supports recursive locking.
1673 	 */
1674 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1675 	VN_LOCK_AREC(vp);
1676 	vp->v_data = ip;
1677 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1678 	ip->i_vnode = vp;
1679 	ip->i_ump = ump;
1680 	ip->i_fs = fs;
1681 	ip->i_dev = dev;
1682 	ip->i_number = ino;
1683 	ip->i_ea_refs = 0;
1684 #ifdef QUOTA
1685 	{
1686 		int i;
1687 		for (i = 0; i < MAXQUOTAS; i++)
1688 			ip->i_dquot[i] = NODQUOT;
1689 	}
1690 #endif
1691 
1692 	if (ffs_flags & FFSV_FORCEINSMQ)
1693 		vp->v_vflag |= VV_FORCEINSMQ;
1694 	error = insmntque(vp, mp);
1695 	if (error != 0) {
1696 		uma_zfree(uma_inode, ip);
1697 		*vpp = NULL;
1698 		return (error);
1699 	}
1700 	vp->v_vflag &= ~VV_FORCEINSMQ;
1701 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1702 	if (error || *vpp != NULL)
1703 		return (error);
1704 
1705 	/* Read in the disk contents for the inode, copy into the inode. */
1706 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1707 	    (int)fs->fs_bsize, NOCRED, &bp);
1708 	if (error) {
1709 		/*
1710 		 * The inode does not contain anything useful, so it would
1711 		 * be misleading to leave it on its hash chain. With mode
1712 		 * still zero, it will be unlinked and returned to the free
1713 		 * list by vput().
1714 		 */
1715 		brelse(bp);
1716 		vput(vp);
1717 		*vpp = NULL;
1718 		return (error);
1719 	}
1720 	if (ip->i_ump->um_fstype == UFS1)
1721 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1722 	else
1723 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1724 	ffs_load_inode(bp, ip, fs, ino);
1725 	if (DOINGSOFTDEP(vp))
1726 		softdep_load_inodeblock(ip);
1727 	else
1728 		ip->i_effnlink = ip->i_nlink;
1729 	bqrelse(bp);
1730 
1731 	/*
1732 	 * Initialize the vnode from the inode, check for aliases.
1733 	 * Note that the underlying vnode may have changed.
1734 	 */
1735 	if (ip->i_ump->um_fstype == UFS1)
1736 		error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1737 	else
1738 		error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1739 	if (error) {
1740 		vput(vp);
1741 		*vpp = NULL;
1742 		return (error);
1743 	}
1744 
1745 	/*
1746 	 * Finish inode initialization.
1747 	 */
1748 	if (vp->v_type != VFIFO) {
1749 		/* FFS supports shared locking for all files except fifos. */
1750 		VN_LOCK_ASHARE(vp);
1751 	}
1752 
1753 	/*
1754 	 * Set up a generation number for this inode if it does not
1755 	 * already have one. This should only happen on old filesystems.
1756 	 */
1757 	if (ip->i_gen == 0) {
1758 		ip->i_gen = arc4random() / 2 + 1;
1759 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1760 			ip->i_flag |= IN_MODIFIED;
1761 			DIP_SET(ip, i_gen, ip->i_gen);
1762 		}
1763 	}
1764 #ifdef MAC
1765 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1766 		/*
1767 		 * If this vnode is already allocated, and we're running
1768 		 * multi-label, attempt to perform a label association
1769 		 * from the extended attributes on the inode.
1770 		 */
1771 		error = mac_vnode_associate_extattr(mp, vp);
1772 		if (error) {
1773 			/* ufs_inactive will release ip->i_devvp ref. */
1774 			vput(vp);
1775 			*vpp = NULL;
1776 			return (error);
1777 		}
1778 	}
1779 #endif
1780 
1781 	*vpp = vp;
1782 	return (0);
1783 }
1784 
1785 /*
1786  * File handle to vnode
1787  *
1788  * Have to be really careful about stale file handles:
1789  * - check that the inode number is valid
1790  * - call ffs_vget() to get the locked inode
1791  * - check for an unallocated inode (i_mode == 0)
1792  * - check that the given client host has export rights and return
1793  *   those rights via. exflagsp and credanonp
1794  */
1795 static int
1796 ffs_fhtovp(mp, fhp, flags, vpp)
1797 	struct mount *mp;
1798 	struct fid *fhp;
1799 	int flags;
1800 	struct vnode **vpp;
1801 {
1802 	struct ufid *ufhp;
1803 	struct fs *fs;
1804 
1805 	ufhp = (struct ufid *)fhp;
1806 	fs = VFSTOUFS(mp)->um_fs;
1807 	if (ufhp->ufid_ino < ROOTINO ||
1808 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1809 		return (ESTALE);
1810 	return (ufs_fhtovp(mp, ufhp, flags, vpp));
1811 }
1812 
1813 /*
1814  * Initialize the filesystem.
1815  */
1816 static int
1817 ffs_init(vfsp)
1818 	struct vfsconf *vfsp;
1819 {
1820 
1821 	ffs_susp_initialize();
1822 	softdep_initialize();
1823 	return (ufs_init(vfsp));
1824 }
1825 
1826 /*
1827  * Undo the work of ffs_init().
1828  */
1829 static int
1830 ffs_uninit(vfsp)
1831 	struct vfsconf *vfsp;
1832 {
1833 	int ret;
1834 
1835 	ret = ufs_uninit(vfsp);
1836 	softdep_uninitialize();
1837 	ffs_susp_uninitialize();
1838 	return (ret);
1839 }
1840 
1841 /*
1842  * Write a superblock and associated information back to disk.
1843  */
1844 int
1845 ffs_sbupdate(ump, waitfor, suspended)
1846 	struct ufsmount *ump;
1847 	int waitfor;
1848 	int suspended;
1849 {
1850 	struct fs *fs = ump->um_fs;
1851 	struct buf *sbbp;
1852 	struct buf *bp;
1853 	int blks;
1854 	void *space;
1855 	int i, size, error, allerror = 0;
1856 
1857 	if (fs->fs_ronly == 1 &&
1858 	    (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1859 	    (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
1860 		panic("ffs_sbupdate: write read-only filesystem");
1861 	/*
1862 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1863 	 */
1864 	sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
1865 	    (int)fs->fs_sbsize, 0, 0, 0);
1866 	/*
1867 	 * First write back the summary information.
1868 	 */
1869 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1870 	space = fs->fs_csp;
1871 	for (i = 0; i < blks; i += fs->fs_frag) {
1872 		size = fs->fs_bsize;
1873 		if (i + fs->fs_frag > blks)
1874 			size = (blks - i) * fs->fs_fsize;
1875 		bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1876 		    size, 0, 0, 0);
1877 		bcopy(space, bp->b_data, (u_int)size);
1878 		space = (char *)space + size;
1879 		if (suspended)
1880 			bp->b_flags |= B_VALIDSUSPWRT;
1881 		if (waitfor != MNT_WAIT)
1882 			bawrite(bp);
1883 		else if ((error = bwrite(bp)) != 0)
1884 			allerror = error;
1885 	}
1886 	/*
1887 	 * Now write back the superblock itself. If any errors occurred
1888 	 * up to this point, then fail so that the superblock avoids
1889 	 * being written out as clean.
1890 	 */
1891 	if (allerror) {
1892 		brelse(sbbp);
1893 		return (allerror);
1894 	}
1895 	bp = sbbp;
1896 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1897 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1898 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1899 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1900 		fs->fs_sblockloc = SBLOCK_UFS1;
1901 	}
1902 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1903 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1904 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1905 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1906 		fs->fs_sblockloc = SBLOCK_UFS2;
1907 	}
1908 	fs->fs_fmod = 0;
1909 	fs->fs_time = time_second;
1910 	if (MOUNTEDSOFTDEP(ump->um_mountp))
1911 		softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
1912 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1913 	ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
1914 	if (suspended)
1915 		bp->b_flags |= B_VALIDSUSPWRT;
1916 	if (waitfor != MNT_WAIT)
1917 		bawrite(bp);
1918 	else if ((error = bwrite(bp)) != 0)
1919 		allerror = error;
1920 	return (allerror);
1921 }
1922 
1923 static int
1924 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1925 	int attrnamespace, const char *attrname)
1926 {
1927 
1928 #ifdef UFS_EXTATTR
1929 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1930 	    attrname));
1931 #else
1932 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1933 	    attrname));
1934 #endif
1935 }
1936 
1937 static void
1938 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1939 {
1940 
1941 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1942 		uma_zfree(uma_ufs1, ip->i_din1);
1943 	else if (ip->i_din2 != NULL)
1944 		uma_zfree(uma_ufs2, ip->i_din2);
1945 	uma_zfree(uma_inode, ip);
1946 }
1947 
1948 static int dobkgrdwrite = 1;
1949 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1950     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1951 
1952 /*
1953  * Complete a background write started from bwrite.
1954  */
1955 static void
1956 ffs_backgroundwritedone(struct buf *bp)
1957 {
1958 	struct bufobj *bufobj;
1959 	struct buf *origbp;
1960 
1961 	/*
1962 	 * Find the original buffer that we are writing.
1963 	 */
1964 	bufobj = bp->b_bufobj;
1965 	BO_LOCK(bufobj);
1966 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
1967 		panic("backgroundwritedone: lost buffer");
1968 	BO_UNLOCK(bufobj);
1969 	/*
1970 	 * Process dependencies then return any unfinished ones.
1971 	 */
1972 	pbrelvp(bp);
1973 	if (!LIST_EMPTY(&bp->b_dep))
1974 		buf_complete(bp);
1975 #ifdef SOFTUPDATES
1976 	if (!LIST_EMPTY(&bp->b_dep))
1977 		softdep_move_dependencies(bp, origbp);
1978 #endif
1979 	/*
1980 	 * This buffer is marked B_NOCACHE so when it is released
1981 	 * by biodone it will be tossed.
1982 	 */
1983 	bp->b_flags |= B_NOCACHE;
1984 	bp->b_flags &= ~B_CACHE;
1985 	bufdone(bp);
1986 	BO_LOCK(bufobj);
1987 	/*
1988 	 * Clear the BV_BKGRDINPROG flag in the original buffer
1989 	 * and awaken it if it is waiting for the write to complete.
1990 	 * If BV_BKGRDINPROG is not set in the original buffer it must
1991 	 * have been released and re-instantiated - which is not legal.
1992 	 */
1993 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
1994 	    ("backgroundwritedone: lost buffer2"));
1995 	origbp->b_vflags &= ~BV_BKGRDINPROG;
1996 	if (origbp->b_vflags & BV_BKGRDWAIT) {
1997 		origbp->b_vflags &= ~BV_BKGRDWAIT;
1998 		wakeup(&origbp->b_xflags);
1999 	}
2000 	BO_UNLOCK(bufobj);
2001 }
2002 
2003 
2004 /*
2005  * Write, release buffer on completion.  (Done by iodone
2006  * if async).  Do not bother writing anything if the buffer
2007  * is invalid.
2008  *
2009  * Note that we set B_CACHE here, indicating that buffer is
2010  * fully valid and thus cacheable.  This is true even of NFS
2011  * now so we set it generally.  This could be set either here
2012  * or in biodone() since the I/O is synchronous.  We put it
2013  * here.
2014  */
2015 static int
2016 ffs_bufwrite(struct buf *bp)
2017 {
2018 	struct buf *newbp;
2019 	int oldflags;
2020 
2021 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2022 	if (bp->b_flags & B_INVAL) {
2023 		brelse(bp);
2024 		return (0);
2025 	}
2026 
2027 	oldflags = bp->b_flags;
2028 
2029 	if (!BUF_ISLOCKED(bp))
2030 		panic("bufwrite: buffer is not busy???");
2031 	/*
2032 	 * If a background write is already in progress, delay
2033 	 * writing this block if it is asynchronous. Otherwise
2034 	 * wait for the background write to complete.
2035 	 */
2036 	BO_LOCK(bp->b_bufobj);
2037 	if (bp->b_vflags & BV_BKGRDINPROG) {
2038 		if (bp->b_flags & B_ASYNC) {
2039 			BO_UNLOCK(bp->b_bufobj);
2040 			bdwrite(bp);
2041 			return (0);
2042 		}
2043 		bp->b_vflags |= BV_BKGRDWAIT;
2044 		msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2045 		    "bwrbg", 0);
2046 		if (bp->b_vflags & BV_BKGRDINPROG)
2047 			panic("bufwrite: still writing");
2048 	}
2049 	BO_UNLOCK(bp->b_bufobj);
2050 
2051 	/*
2052 	 * If this buffer is marked for background writing and we
2053 	 * do not have to wait for it, make a copy and write the
2054 	 * copy so as to leave this buffer ready for further use.
2055 	 *
2056 	 * This optimization eats a lot of memory.  If we have a page
2057 	 * or buffer shortfall we can't do it.
2058 	 */
2059 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2060 	    (bp->b_flags & B_ASYNC) &&
2061 	    !vm_page_count_severe() &&
2062 	    !buf_dirty_count_severe()) {
2063 		KASSERT(bp->b_iodone == NULL,
2064 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2065 
2066 		/* get a new block */
2067 		newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2068 		if (newbp == NULL)
2069 			goto normal_write;
2070 
2071 		KASSERT((bp->b_flags & B_UNMAPPED) == 0, ("Unmapped cg"));
2072 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2073 		BO_LOCK(bp->b_bufobj);
2074 		bp->b_vflags |= BV_BKGRDINPROG;
2075 		BO_UNLOCK(bp->b_bufobj);
2076 		newbp->b_xflags |= BX_BKGRDMARKER;
2077 		newbp->b_lblkno = bp->b_lblkno;
2078 		newbp->b_blkno = bp->b_blkno;
2079 		newbp->b_offset = bp->b_offset;
2080 		newbp->b_iodone = ffs_backgroundwritedone;
2081 		newbp->b_flags |= B_ASYNC;
2082 		newbp->b_flags &= ~B_INVAL;
2083 		pbgetvp(bp->b_vp, newbp);
2084 
2085 #ifdef SOFTUPDATES
2086 		/*
2087 		 * Move over the dependencies.  If there are rollbacks,
2088 		 * leave the parent buffer dirtied as it will need to
2089 		 * be written again.
2090 		 */
2091 		if (LIST_EMPTY(&bp->b_dep) ||
2092 		    softdep_move_dependencies(bp, newbp) == 0)
2093 			bundirty(bp);
2094 #else
2095 		bundirty(bp);
2096 #endif
2097 
2098 		/*
2099 		 * Initiate write on the copy, release the original.  The
2100 		 * BKGRDINPROG flag prevents it from going away until
2101 		 * the background write completes.
2102 		 */
2103 		bqrelse(bp);
2104 		bp = newbp;
2105 	} else
2106 		/* Mark the buffer clean */
2107 		bundirty(bp);
2108 
2109 
2110 	/* Let the normal bufwrite do the rest for us */
2111 normal_write:
2112 	return (bufwrite(bp));
2113 }
2114 
2115 
2116 static void
2117 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2118 {
2119 	struct vnode *vp;
2120 	int error;
2121 	struct buf *tbp;
2122 	int nocopy;
2123 
2124 	vp = bo->__bo_vnode;
2125 	if (bp->b_iocmd == BIO_WRITE) {
2126 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2127 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2128 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2129 			panic("ffs_geom_strategy: bad I/O");
2130 		nocopy = bp->b_flags & B_NOCOPY;
2131 		bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2132 		if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2133 		    vp->v_rdev->si_snapdata != NULL) {
2134 			if ((bp->b_flags & B_CLUSTER) != 0) {
2135 				runningbufwakeup(bp);
2136 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2137 					      b_cluster.cluster_entry) {
2138 					error = ffs_copyonwrite(vp, tbp);
2139 					if (error != 0 &&
2140 					    error != EOPNOTSUPP) {
2141 						bp->b_error = error;
2142 						bp->b_ioflags |= BIO_ERROR;
2143 						bufdone(bp);
2144 						return;
2145 					}
2146 				}
2147 				bp->b_runningbufspace = bp->b_bufsize;
2148 				atomic_add_long(&runningbufspace,
2149 					       bp->b_runningbufspace);
2150 			} else {
2151 				error = ffs_copyonwrite(vp, bp);
2152 				if (error != 0 && error != EOPNOTSUPP) {
2153 					bp->b_error = error;
2154 					bp->b_ioflags |= BIO_ERROR;
2155 					bufdone(bp);
2156 					return;
2157 				}
2158 			}
2159 		}
2160 #ifdef SOFTUPDATES
2161 		if ((bp->b_flags & B_CLUSTER) != 0) {
2162 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2163 				      b_cluster.cluster_entry) {
2164 				if (!LIST_EMPTY(&tbp->b_dep))
2165 					buf_start(tbp);
2166 			}
2167 		} else {
2168 			if (!LIST_EMPTY(&bp->b_dep))
2169 				buf_start(bp);
2170 		}
2171 
2172 #endif
2173 	}
2174 	g_vfs_strategy(bo, bp);
2175 }
2176 
2177 int
2178 ffs_own_mount(const struct mount *mp)
2179 {
2180 
2181 	if (mp->mnt_op == &ufs_vfsops)
2182 		return (1);
2183 	return (0);
2184 }
2185 
2186 #ifdef	DDB
2187 #ifdef SOFTUPDATES
2188 
2189 /* defined in ffs_softdep.c */
2190 extern void db_print_ffs(struct ufsmount *ump);
2191 
2192 DB_SHOW_COMMAND(ffs, db_show_ffs)
2193 {
2194 	struct mount *mp;
2195 	struct ufsmount *ump;
2196 
2197 	if (have_addr) {
2198 		ump = VFSTOUFS((struct mount *)addr);
2199 		db_print_ffs(ump);
2200 		return;
2201 	}
2202 
2203 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2204 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2205 			db_print_ffs(VFSTOUFS(mp));
2206 	}
2207 }
2208 
2209 #endif	/* SOFTUPDATES */
2210 #endif	/* DDB */
2211