xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision 994297b01b98816bea1abf45ae4bac1bc69ee7a0)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1991, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_quota.h"
38 #include "opt_ufs.h"
39 #include "opt_ffs.h"
40 #include "opt_ddb.h"
41 
42 #include <sys/param.h>
43 #include <sys/gsb_crc32.h>
44 #include <sys/systm.h>
45 #include <sys/namei.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/kernel.h>
50 #include <sys/ktr.h>
51 #include <sys/vnode.h>
52 #include <sys/mount.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/conf.h>
56 #include <sys/fcntl.h>
57 #include <sys/ioccom.h>
58 #include <sys/malloc.h>
59 #include <sys/mutex.h>
60 #include <sys/rwlock.h>
61 #include <sys/sysctl.h>
62 #include <sys/vmmeter.h>
63 
64 #include <security/mac/mac_framework.h>
65 
66 #include <ufs/ufs/dir.h>
67 #include <ufs/ufs/extattr.h>
68 #include <ufs/ufs/gjournal.h>
69 #include <ufs/ufs/quota.h>
70 #include <ufs/ufs/ufsmount.h>
71 #include <ufs/ufs/inode.h>
72 #include <ufs/ufs/ufs_extern.h>
73 
74 #include <ufs/ffs/fs.h>
75 #include <ufs/ffs/ffs_extern.h>
76 
77 #include <vm/vm.h>
78 #include <vm/uma.h>
79 #include <vm/vm_page.h>
80 
81 #include <geom/geom.h>
82 #include <geom/geom_vfs.h>
83 
84 #include <ddb/ddb.h>
85 
86 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
87 VFS_SMR_DECLARE;
88 
89 static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
90 static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
91 		    ufs2_daddr_t);
92 static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
93 static int	ffs_sync_lazy(struct mount *mp);
94 static int	ffs_use_bread(void *devfd, off_t loc, void **bufp, int size);
95 static int	ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size);
96 
97 static vfs_init_t ffs_init;
98 static vfs_uninit_t ffs_uninit;
99 static vfs_extattrctl_t ffs_extattrctl;
100 static vfs_cmount_t ffs_cmount;
101 static vfs_unmount_t ffs_unmount;
102 static vfs_mount_t ffs_mount;
103 static vfs_statfs_t ffs_statfs;
104 static vfs_fhtovp_t ffs_fhtovp;
105 static vfs_sync_t ffs_sync;
106 
107 static struct vfsops ufs_vfsops = {
108 	.vfs_extattrctl =	ffs_extattrctl,
109 	.vfs_fhtovp =		ffs_fhtovp,
110 	.vfs_init =		ffs_init,
111 	.vfs_mount =		ffs_mount,
112 	.vfs_cmount =		ffs_cmount,
113 	.vfs_quotactl =		ufs_quotactl,
114 	.vfs_root =		vfs_cache_root,
115 	.vfs_cachedroot =	ufs_root,
116 	.vfs_statfs =		ffs_statfs,
117 	.vfs_sync =		ffs_sync,
118 	.vfs_uninit =		ffs_uninit,
119 	.vfs_unmount =		ffs_unmount,
120 	.vfs_vget =		ffs_vget,
121 	.vfs_susp_clean =	process_deferred_inactive,
122 };
123 
124 VFS_SET(ufs_vfsops, ufs, 0);
125 MODULE_VERSION(ufs, 1);
126 
127 static b_strategy_t ffs_geom_strategy;
128 static b_write_t ffs_bufwrite;
129 
130 static struct buf_ops ffs_ops = {
131 	.bop_name =	"FFS",
132 	.bop_write =	ffs_bufwrite,
133 	.bop_strategy =	ffs_geom_strategy,
134 	.bop_sync =	bufsync,
135 #ifdef NO_FFS_SNAPSHOT
136 	.bop_bdflush =	bufbdflush,
137 #else
138 	.bop_bdflush =	ffs_bdflush,
139 #endif
140 };
141 
142 /*
143  * Note that userquota and groupquota options are not currently used
144  * by UFS/FFS code and generally mount(8) does not pass those options
145  * from userland, but they can be passed by loader(8) via
146  * vfs.root.mountfrom.options.
147  */
148 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
149     "noclusterw", "noexec", "export", "force", "from", "groupquota",
150     "multilabel", "nfsv4acls", "snapshot", "nosuid", "suiddir",
151     "nosymfollow", "sync", "union", "userquota", "untrusted", NULL };
152 
153 static int ffs_enxio_enable = 1;
154 SYSCTL_DECL(_vfs_ffs);
155 SYSCTL_INT(_vfs_ffs, OID_AUTO, enxio_enable, CTLFLAG_RWTUN,
156     &ffs_enxio_enable, 0,
157     "enable mapping of other disk I/O errors to ENXIO");
158 
159 /*
160  * Return buffer with the contents of block "offset" from the beginning of
161  * directory "ip".  If "res" is non-zero, fill it in with a pointer to the
162  * remaining space in the directory.
163  */
164 static int
165 ffs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp)
166 {
167 	struct inode *ip;
168 	struct fs *fs;
169 	struct buf *bp;
170 	ufs_lbn_t lbn;
171 	int bsize, error;
172 
173 	ip = VTOI(vp);
174 	fs = ITOFS(ip);
175 	lbn = lblkno(fs, offset);
176 	bsize = blksize(fs, ip, lbn);
177 
178 	*bpp = NULL;
179 	error = bread(vp, lbn, bsize, NOCRED, &bp);
180 	if (error) {
181 		return (error);
182 	}
183 	if (res)
184 		*res = (char *)bp->b_data + blkoff(fs, offset);
185 	*bpp = bp;
186 	return (0);
187 }
188 
189 /*
190  * Load up the contents of an inode and copy the appropriate pieces
191  * to the incore copy.
192  */
193 static int
194 ffs_load_inode(struct buf *bp, struct inode *ip, struct fs *fs, ino_t ino)
195 {
196 	struct ufs1_dinode *dip1;
197 	struct ufs2_dinode *dip2;
198 	int error;
199 
200 	if (I_IS_UFS1(ip)) {
201 		dip1 = ip->i_din1;
202 		*dip1 =
203 		    *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
204 		ip->i_mode = dip1->di_mode;
205 		ip->i_nlink = dip1->di_nlink;
206 		ip->i_effnlink = dip1->di_nlink;
207 		ip->i_size = dip1->di_size;
208 		ip->i_flags = dip1->di_flags;
209 		ip->i_gen = dip1->di_gen;
210 		ip->i_uid = dip1->di_uid;
211 		ip->i_gid = dip1->di_gid;
212 		return (0);
213 	}
214 	dip2 = ((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
215 	if ((error = ffs_verify_dinode_ckhash(fs, dip2)) != 0 &&
216 	    !ffs_fsfail_cleanup(ITOUMP(ip), error)) {
217 		printf("%s: inode %jd: check-hash failed\n", fs->fs_fsmnt,
218 		    (intmax_t)ino);
219 		return (error);
220 	}
221 	*ip->i_din2 = *dip2;
222 	dip2 = ip->i_din2;
223 	ip->i_mode = dip2->di_mode;
224 	ip->i_nlink = dip2->di_nlink;
225 	ip->i_effnlink = dip2->di_nlink;
226 	ip->i_size = dip2->di_size;
227 	ip->i_flags = dip2->di_flags;
228 	ip->i_gen = dip2->di_gen;
229 	ip->i_uid = dip2->di_uid;
230 	ip->i_gid = dip2->di_gid;
231 	return (0);
232 }
233 
234 /*
235  * Verify that a filesystem block number is a valid data block.
236  * This routine is only called on untrusted filesystems.
237  */
238 static int
239 ffs_check_blkno(struct mount *mp, ino_t inum, ufs2_daddr_t daddr, int blksize)
240 {
241 	struct fs *fs;
242 	struct ufsmount *ump;
243 	ufs2_daddr_t end_daddr;
244 	int cg, havemtx;
245 
246 	KASSERT((mp->mnt_flag & MNT_UNTRUSTED) != 0,
247 	    ("ffs_check_blkno called on a trusted file system"));
248 	ump = VFSTOUFS(mp);
249 	fs = ump->um_fs;
250 	cg = dtog(fs, daddr);
251 	end_daddr = daddr + numfrags(fs, blksize);
252 	/*
253 	 * Verify that the block number is a valid data block. Also check
254 	 * that it does not point to an inode block or a superblock. Accept
255 	 * blocks that are unalloacted (0) or part of snapshot metadata
256 	 * (BLK_NOCOPY or BLK_SNAP).
257 	 *
258 	 * Thus, the block must be in a valid range for the filesystem and
259 	 * either in the space before a backup superblock (except the first
260 	 * cylinder group where that space is used by the bootstrap code) or
261 	 * after the inode blocks and before the end of the cylinder group.
262 	 */
263 	if ((uint64_t)daddr <= BLK_SNAP ||
264 	    ((uint64_t)end_daddr <= fs->fs_size &&
265 	    ((cg > 0 && end_daddr <= cgsblock(fs, cg)) ||
266 	    (daddr >= cgdmin(fs, cg) &&
267 	    end_daddr <= cgbase(fs, cg) + fs->fs_fpg))))
268 		return (0);
269 	if ((havemtx = mtx_owned(UFS_MTX(ump))) == 0)
270 		UFS_LOCK(ump);
271 	if (ppsratecheck(&ump->um_last_integritymsg,
272 	    &ump->um_secs_integritymsg, 1)) {
273 		UFS_UNLOCK(ump);
274 		uprintf("\n%s: inode %jd, out-of-range indirect block "
275 		    "number %jd\n", mp->mnt_stat.f_mntonname, inum, daddr);
276 		if (havemtx)
277 			UFS_LOCK(ump);
278 	} else if (!havemtx)
279 		UFS_UNLOCK(ump);
280 	return (EINTEGRITY);
281 }
282 
283 /*
284  * On first ENXIO error, initiate an asynchronous forcible unmount.
285  * Used to unmount filesystems whose underlying media has gone away.
286  *
287  * Return true if a cleanup is in progress.
288  */
289 int
290 ffs_fsfail_cleanup(struct ufsmount *ump, int error)
291 {
292 	int retval;
293 
294 	UFS_LOCK(ump);
295 	retval = ffs_fsfail_cleanup_locked(ump, error);
296 	UFS_UNLOCK(ump);
297 	return (retval);
298 }
299 
300 int
301 ffs_fsfail_cleanup_locked(struct ufsmount *ump, int error)
302 {
303 	mtx_assert(UFS_MTX(ump), MA_OWNED);
304 	if (error == ENXIO && (ump->um_flags & UM_FSFAIL_CLEANUP) == 0) {
305 		ump->um_flags |= UM_FSFAIL_CLEANUP;
306 		/*
307 		 * Queue an async forced unmount.
308 		 */
309 		vfs_ref(ump->um_mountp);
310 		dounmount(ump->um_mountp,
311 		    MNT_FORCE | MNT_RECURSE | MNT_DEFERRED, curthread);
312 		printf("UFS: forcibly unmounting %s from %s\n",
313 		    ump->um_mountp->mnt_stat.f_mntfromname,
314 		    ump->um_mountp->mnt_stat.f_mntonname);
315 	}
316 	return ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0);
317 }
318 
319 /*
320  * Wrapper used during ENXIO cleanup to allocate empty buffers when
321  * the kernel is unable to read the real one. They are needed so that
322  * the soft updates code can use them to unwind its dependencies.
323  */
324 int
325 ffs_breadz(struct ufsmount *ump, struct vnode *vp, daddr_t lblkno,
326     daddr_t dblkno, int size, daddr_t *rablkno, int *rabsize, int cnt,
327     struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *),
328     struct buf **bpp)
329 {
330 	int error;
331 
332 	flags |= GB_CVTENXIO;
333 	error = breadn_flags(vp, lblkno, dblkno, size, rablkno, rabsize, cnt,
334 	    cred, flags, ckhashfunc, bpp);
335 	if (error != 0 && ffs_fsfail_cleanup(ump, error)) {
336 		error = getblkx(vp, lblkno, dblkno, size, 0, 0, flags, bpp);
337 		KASSERT(error == 0, ("getblkx failed"));
338 		vfs_bio_bzero_buf(*bpp, 0, size);
339 	}
340 	return (error);
341 }
342 
343 static int
344 ffs_mount(struct mount *mp)
345 {
346 	struct vnode *devvp, *odevvp;
347 	struct thread *td;
348 	struct ufsmount *ump = NULL;
349 	struct fs *fs;
350 	int error, flags;
351 	int error1 __diagused;
352 	uint64_t mntorflags, saved_mnt_flag;
353 	accmode_t accmode;
354 	struct nameidata ndp;
355 	char *fspec;
356 	bool mounted_softdep;
357 
358 	td = curthread;
359 	if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
360 		return (EINVAL);
361 	if (uma_inode == NULL) {
362 		uma_inode = uma_zcreate("FFS inode",
363 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
364 		    UMA_ALIGN_PTR, 0);
365 		uma_ufs1 = uma_zcreate("FFS1 dinode",
366 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
367 		    UMA_ALIGN_PTR, 0);
368 		uma_ufs2 = uma_zcreate("FFS2 dinode",
369 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
370 		    UMA_ALIGN_PTR, 0);
371 		VFS_SMR_ZONE_SET(uma_inode);
372 	}
373 
374 	vfs_deleteopt(mp->mnt_optnew, "groupquota");
375 	vfs_deleteopt(mp->mnt_optnew, "userquota");
376 
377 	fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
378 	if (error)
379 		return (error);
380 
381 	mntorflags = 0;
382 	if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0)
383 		mntorflags |= MNT_UNTRUSTED;
384 
385 	if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
386 		mntorflags |= MNT_ACLS;
387 
388 	if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
389 		mntorflags |= MNT_SNAPSHOT;
390 		/*
391 		 * Once we have set the MNT_SNAPSHOT flag, do not
392 		 * persist "snapshot" in the options list.
393 		 */
394 		vfs_deleteopt(mp->mnt_optnew, "snapshot");
395 		vfs_deleteopt(mp->mnt_opt, "snapshot");
396 	}
397 
398 	if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
399 		if (mntorflags & MNT_ACLS) {
400 			vfs_mount_error(mp,
401 			    "\"acls\" and \"nfsv4acls\" options "
402 			    "are mutually exclusive");
403 			return (EINVAL);
404 		}
405 		mntorflags |= MNT_NFS4ACLS;
406 	}
407 
408 	MNT_ILOCK(mp);
409 	mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
410 	mp->mnt_flag |= mntorflags;
411 	MNT_IUNLOCK(mp);
412 	/*
413 	 * If updating, check whether changing from read-only to
414 	 * read/write; if there is no device name, that's all we do.
415 	 */
416 	if (mp->mnt_flag & MNT_UPDATE) {
417 		ump = VFSTOUFS(mp);
418 		fs = ump->um_fs;
419 		odevvp = ump->um_odevvp;
420 		devvp = ump->um_devvp;
421 		if (fs->fs_ronly == 0 &&
422 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
423 			/*
424 			 * Flush any dirty data and suspend filesystem.
425 			 */
426 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
427 				return (error);
428 			error = vfs_write_suspend_umnt(mp);
429 			if (error != 0)
430 				return (error);
431 
432 			fs->fs_ronly = 1;
433 			if (MOUNTEDSOFTDEP(mp)) {
434 				MNT_ILOCK(mp);
435 				mp->mnt_flag &= ~MNT_SOFTDEP;
436 				MNT_IUNLOCK(mp);
437 				mounted_softdep = true;
438 			} else
439 				mounted_softdep = false;
440 
441 			/*
442 			 * Check for and optionally get rid of files open
443 			 * for writing.
444 			 */
445 			flags = WRITECLOSE;
446 			if (mp->mnt_flag & MNT_FORCE)
447 				flags |= FORCECLOSE;
448 			if (mounted_softdep) {
449 				error = softdep_flushfiles(mp, flags, td);
450 			} else {
451 				error = ffs_flushfiles(mp, flags, td);
452 			}
453 			if (error) {
454 				fs->fs_ronly = 0;
455 				if (mounted_softdep) {
456 					MNT_ILOCK(mp);
457 					mp->mnt_flag |= MNT_SOFTDEP;
458 					MNT_IUNLOCK(mp);
459 				}
460 				vfs_write_resume(mp, 0);
461 				return (error);
462 			}
463 
464 			if (fs->fs_pendingblocks != 0 ||
465 			    fs->fs_pendinginodes != 0) {
466 				printf("WARNING: %s Update error: blocks %jd "
467 				    "files %d\n", fs->fs_fsmnt,
468 				    (intmax_t)fs->fs_pendingblocks,
469 				    fs->fs_pendinginodes);
470 				fs->fs_pendingblocks = 0;
471 				fs->fs_pendinginodes = 0;
472 			}
473 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
474 				fs->fs_clean = 1;
475 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
476 				fs->fs_ronly = 0;
477 				fs->fs_clean = 0;
478 				if (mounted_softdep) {
479 					MNT_ILOCK(mp);
480 					mp->mnt_flag |= MNT_SOFTDEP;
481 					MNT_IUNLOCK(mp);
482 				}
483 				vfs_write_resume(mp, 0);
484 				return (error);
485 			}
486 			if (mounted_softdep)
487 				softdep_unmount(mp);
488 			g_topology_lock();
489 			/*
490 			 * Drop our write and exclusive access.
491 			 */
492 			g_access(ump->um_cp, 0, -1, -1);
493 			g_topology_unlock();
494 			MNT_ILOCK(mp);
495 			mp->mnt_flag |= MNT_RDONLY;
496 			MNT_IUNLOCK(mp);
497 			/*
498 			 * Allow the writers to note that filesystem
499 			 * is ro now.
500 			 */
501 			vfs_write_resume(mp, 0);
502 		}
503 		if ((mp->mnt_flag & MNT_RELOAD) &&
504 		    (error = ffs_reload(mp, 0)) != 0)
505 			return (error);
506 		if (fs->fs_ronly &&
507 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
508 			/*
509 			 * If upgrade to read-write by non-root, then verify
510 			 * that user has necessary permissions on the device.
511 			 */
512 			vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY);
513 			error = VOP_ACCESS(odevvp, VREAD | VWRITE,
514 			    td->td_ucred, td);
515 			if (error)
516 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
517 			VOP_UNLOCK(odevvp);
518 			if (error) {
519 				return (error);
520 			}
521 			fs->fs_flags &= ~FS_UNCLEAN;
522 			if (fs->fs_clean == 0) {
523 				fs->fs_flags |= FS_UNCLEAN;
524 				if ((mp->mnt_flag & MNT_FORCE) ||
525 				    ((fs->fs_flags &
526 				     (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
527 				     (fs->fs_flags & FS_DOSOFTDEP))) {
528 					printf("WARNING: %s was not properly "
529 					   "dismounted\n", fs->fs_fsmnt);
530 				} else {
531 					vfs_mount_error(mp,
532 					   "R/W mount of %s denied. %s.%s",
533 					   fs->fs_fsmnt,
534 					   "Filesystem is not clean - run fsck",
535 					   (fs->fs_flags & FS_SUJ) == 0 ? "" :
536 					   " Forced mount will invalidate"
537 					   " journal contents");
538 					return (EPERM);
539 				}
540 			}
541 			g_topology_lock();
542 			/*
543 			 * Request exclusive write access.
544 			 */
545 			error = g_access(ump->um_cp, 0, 1, 1);
546 			g_topology_unlock();
547 			if (error)
548 				return (error);
549 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
550 				return (error);
551 			error = vfs_write_suspend_umnt(mp);
552 			if (error != 0)
553 				return (error);
554 			fs->fs_ronly = 0;
555 			MNT_ILOCK(mp);
556 			saved_mnt_flag = MNT_RDONLY;
557 			if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag &
558 			    MNT_ASYNC) != 0)
559 				saved_mnt_flag |= MNT_ASYNC;
560 			mp->mnt_flag &= ~saved_mnt_flag;
561 			MNT_IUNLOCK(mp);
562 			fs->fs_mtime = time_second;
563 			/* check to see if we need to start softdep */
564 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
565 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
566 				fs->fs_ronly = 1;
567 				MNT_ILOCK(mp);
568 				mp->mnt_flag |= saved_mnt_flag;
569 				MNT_IUNLOCK(mp);
570 				vfs_write_resume(mp, 0);
571 				return (error);
572 			}
573 			fs->fs_clean = 0;
574 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
575 				fs->fs_ronly = 1;
576 				if ((fs->fs_flags & FS_DOSOFTDEP) != 0)
577 					softdep_unmount(mp);
578 				MNT_ILOCK(mp);
579 				mp->mnt_flag |= saved_mnt_flag;
580 				MNT_IUNLOCK(mp);
581 				vfs_write_resume(mp, 0);
582 				return (error);
583 			}
584 			if (fs->fs_snapinum[0] != 0)
585 				ffs_snapshot_mount(mp);
586 			vfs_write_resume(mp, 0);
587 		}
588 		/*
589 		 * Soft updates is incompatible with "async",
590 		 * so if we are doing softupdates stop the user
591 		 * from setting the async flag in an update.
592 		 * Softdep_mount() clears it in an initial mount
593 		 * or ro->rw remount.
594 		 */
595 		if (MOUNTEDSOFTDEP(mp)) {
596 			/* XXX: Reset too late ? */
597 			MNT_ILOCK(mp);
598 			mp->mnt_flag &= ~MNT_ASYNC;
599 			MNT_IUNLOCK(mp);
600 		}
601 		/*
602 		 * Keep MNT_ACLS flag if it is stored in superblock.
603 		 */
604 		if ((fs->fs_flags & FS_ACLS) != 0) {
605 			/* XXX: Set too late ? */
606 			MNT_ILOCK(mp);
607 			mp->mnt_flag |= MNT_ACLS;
608 			MNT_IUNLOCK(mp);
609 		}
610 
611 		if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
612 			/* XXX: Set too late ? */
613 			MNT_ILOCK(mp);
614 			mp->mnt_flag |= MNT_NFS4ACLS;
615 			MNT_IUNLOCK(mp);
616 		}
617 
618 		/*
619 		 * If this is a snapshot request, take the snapshot.
620 		 */
621 		if (mp->mnt_flag & MNT_SNAPSHOT)
622 			return (ffs_snapshot(mp, fspec));
623 
624 		/*
625 		 * Must not call namei() while owning busy ref.
626 		 */
627 		vfs_unbusy(mp);
628 	}
629 
630 	/*
631 	 * Not an update, or updating the name: look up the name
632 	 * and verify that it refers to a sensible disk device.
633 	 */
634 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec);
635 	error = namei(&ndp);
636 	if ((mp->mnt_flag & MNT_UPDATE) != 0) {
637 		/*
638 		 * Unmount does not start if MNT_UPDATE is set.  Mount
639 		 * update busies mp before setting MNT_UPDATE.  We
640 		 * must be able to retain our busy ref succesfully,
641 		 * without sleep.
642 		 */
643 		error1 = vfs_busy(mp, MBF_NOWAIT);
644 		MPASS(error1 == 0);
645 	}
646 	if (error != 0)
647 		return (error);
648 	NDFREE(&ndp, NDF_ONLY_PNBUF);
649 	devvp = ndp.ni_vp;
650 	if (!vn_isdisk_error(devvp, &error)) {
651 		vput(devvp);
652 		return (error);
653 	}
654 
655 	/*
656 	 * If mount by non-root, then verify that user has necessary
657 	 * permissions on the device.
658 	 */
659 	accmode = VREAD;
660 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
661 		accmode |= VWRITE;
662 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
663 	if (error)
664 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
665 	if (error) {
666 		vput(devvp);
667 		return (error);
668 	}
669 
670 	if (mp->mnt_flag & MNT_UPDATE) {
671 		/*
672 		 * Update only
673 		 *
674 		 * If it's not the same vnode, or at least the same device
675 		 * then it's not correct.
676 		 */
677 
678 		if (devvp->v_rdev != ump->um_devvp->v_rdev)
679 			error = EINVAL;	/* needs translation */
680 		vput(devvp);
681 		if (error)
682 			return (error);
683 	} else {
684 		/*
685 		 * New mount
686 		 *
687 		 * We need the name for the mount point (also used for
688 		 * "last mounted on") copied in. If an error occurs,
689 		 * the mount point is discarded by the upper level code.
690 		 * Note that vfs_mount_alloc() populates f_mntonname for us.
691 		 */
692 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
693 			vrele(devvp);
694 			return (error);
695 		}
696 	}
697 
698 	MNT_ILOCK(mp);
699 	/*
700 	 * This is racy versus lookup, see ufs_fplookup_vexec for details.
701 	 */
702 	if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) != 0)
703 		panic("MNTK_FPLOOKUP set on mount %p when it should not be", mp);
704 	if ((mp->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS | MNT_UNION)) == 0)
705 		mp->mnt_kern_flag |= MNTK_FPLOOKUP;
706 	MNT_IUNLOCK(mp);
707 
708 	vfs_mountedfrom(mp, fspec);
709 	return (0);
710 }
711 
712 /*
713  * Compatibility with old mount system call.
714  */
715 
716 static int
717 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
718 {
719 	struct ufs_args args;
720 	int error;
721 
722 	if (data == NULL)
723 		return (EINVAL);
724 	error = copyin(data, &args, sizeof args);
725 	if (error)
726 		return (error);
727 
728 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
729 	ma = mount_arg(ma, "export", &args.export, sizeof(args.export));
730 	error = kernel_mount(ma, flags);
731 
732 	return (error);
733 }
734 
735 /*
736  * Reload all incore data for a filesystem (used after running fsck on
737  * the root filesystem and finding things to fix). If the 'force' flag
738  * is 0, the filesystem must be mounted read-only.
739  *
740  * Things to do to update the mount:
741  *	1) invalidate all cached meta-data.
742  *	2) re-read superblock from disk.
743  *	3) re-read summary information from disk.
744  *	4) invalidate all inactive vnodes.
745  *	5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary
746  *	   writers, if requested.
747  *	6) invalidate all cached file data.
748  *	7) re-read inode data for all active vnodes.
749  */
750 int
751 ffs_reload(struct mount *mp, int flags)
752 {
753 	struct vnode *vp, *mvp, *devvp;
754 	struct inode *ip;
755 	void *space;
756 	struct buf *bp;
757 	struct fs *fs, *newfs;
758 	struct ufsmount *ump;
759 	ufs2_daddr_t sblockloc;
760 	int i, blks, error;
761 	u_long size;
762 	int32_t *lp;
763 
764 	ump = VFSTOUFS(mp);
765 
766 	MNT_ILOCK(mp);
767 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) {
768 		MNT_IUNLOCK(mp);
769 		return (EINVAL);
770 	}
771 	MNT_IUNLOCK(mp);
772 
773 	/*
774 	 * Step 1: invalidate all cached meta-data.
775 	 */
776 	devvp = VFSTOUFS(mp)->um_devvp;
777 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
778 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
779 		panic("ffs_reload: dirty1");
780 	VOP_UNLOCK(devvp);
781 
782 	/*
783 	 * Step 2: re-read superblock from disk.
784 	 */
785 	fs = VFSTOUFS(mp)->um_fs;
786 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
787 	    NOCRED, &bp)) != 0)
788 		return (error);
789 	newfs = (struct fs *)bp->b_data;
790 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
791 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
792 	    newfs->fs_bsize > MAXBSIZE ||
793 	    newfs->fs_bsize < sizeof(struct fs)) {
794 			brelse(bp);
795 			return (EIO);		/* XXX needs translation */
796 	}
797 	/*
798 	 * Preserve the summary information, read-only status, and
799 	 * superblock location by copying these fields into our new
800 	 * superblock before using it to update the existing superblock.
801 	 */
802 	newfs->fs_si = fs->fs_si;
803 	newfs->fs_ronly = fs->fs_ronly;
804 	sblockloc = fs->fs_sblockloc;
805 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
806 	brelse(bp);
807 	ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
808 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
809 	UFS_LOCK(ump);
810 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
811 		printf("WARNING: %s: reload pending error: blocks %jd "
812 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
813 		    fs->fs_pendinginodes);
814 		fs->fs_pendingblocks = 0;
815 		fs->fs_pendinginodes = 0;
816 	}
817 	UFS_UNLOCK(ump);
818 
819 	/*
820 	 * Step 3: re-read summary information from disk.
821 	 */
822 	size = fs->fs_cssize;
823 	blks = howmany(size, fs->fs_fsize);
824 	if (fs->fs_contigsumsize > 0)
825 		size += fs->fs_ncg * sizeof(int32_t);
826 	size += fs->fs_ncg * sizeof(u_int8_t);
827 	free(fs->fs_csp, M_UFSMNT);
828 	space = malloc(size, M_UFSMNT, M_WAITOK);
829 	fs->fs_csp = space;
830 	for (i = 0; i < blks; i += fs->fs_frag) {
831 		size = fs->fs_bsize;
832 		if (i + fs->fs_frag > blks)
833 			size = (blks - i) * fs->fs_fsize;
834 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
835 		    NOCRED, &bp);
836 		if (error)
837 			return (error);
838 		bcopy(bp->b_data, space, (u_int)size);
839 		space = (char *)space + size;
840 		brelse(bp);
841 	}
842 	/*
843 	 * We no longer know anything about clusters per cylinder group.
844 	 */
845 	if (fs->fs_contigsumsize > 0) {
846 		fs->fs_maxcluster = lp = space;
847 		for (i = 0; i < fs->fs_ncg; i++)
848 			*lp++ = fs->fs_contigsumsize;
849 		space = lp;
850 	}
851 	size = fs->fs_ncg * sizeof(u_int8_t);
852 	fs->fs_contigdirs = (u_int8_t *)space;
853 	bzero(fs->fs_contigdirs, size);
854 	if ((flags & FFSR_UNSUSPEND) != 0) {
855 		MNT_ILOCK(mp);
856 		mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
857 		wakeup(&mp->mnt_flag);
858 		MNT_IUNLOCK(mp);
859 	}
860 
861 loop:
862 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
863 		/*
864 		 * Skip syncer vnode.
865 		 */
866 		if (vp->v_type == VNON) {
867 			VI_UNLOCK(vp);
868 			continue;
869 		}
870 		/*
871 		 * Step 4: invalidate all cached file data.
872 		 */
873 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
874 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
875 			goto loop;
876 		}
877 		if (vinvalbuf(vp, 0, 0, 0))
878 			panic("ffs_reload: dirty2");
879 		/*
880 		 * Step 5: re-read inode data for all active vnodes.
881 		 */
882 		ip = VTOI(vp);
883 		error =
884 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
885 		    (int)fs->fs_bsize, NOCRED, &bp);
886 		if (error) {
887 			vput(vp);
888 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
889 			return (error);
890 		}
891 		if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) {
892 			brelse(bp);
893 			vput(vp);
894 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
895 			return (error);
896 		}
897 		ip->i_effnlink = ip->i_nlink;
898 		brelse(bp);
899 		vput(vp);
900 	}
901 	return (0);
902 }
903 
904 /*
905  * Common code for mount and mountroot
906  */
907 static int
908 ffs_mountfs(odevvp, mp, td)
909 	struct vnode *odevvp;
910 	struct mount *mp;
911 	struct thread *td;
912 {
913 	struct ufsmount *ump;
914 	struct fs *fs;
915 	struct cdev *dev;
916 	int error, i, len, ronly;
917 	struct ucred *cred;
918 	struct g_consumer *cp;
919 	struct mount *nmp;
920 	struct vnode *devvp;
921 	int candelete, canspeedup;
922 	off_t loc;
923 
924 	fs = NULL;
925 	ump = NULL;
926 	cred = td ? td->td_ucred : NOCRED;
927 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
928 
929 	devvp = mntfs_allocvp(mp, odevvp);
930 	VOP_UNLOCK(odevvp);
931 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
932 	KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
933 	dev = devvp->v_rdev;
934 	KASSERT(dev->si_snapdata == NULL, ("non-NULL snapshot data"));
935 	if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
936 	    (uintptr_t)mp) == 0) {
937 		mntfs_freevp(devvp);
938 		return (EBUSY);
939 	}
940 	g_topology_lock();
941 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
942 	g_topology_unlock();
943 	if (error != 0) {
944 		atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
945 		mntfs_freevp(devvp);
946 		return (error);
947 	}
948 	dev_ref(dev);
949 	devvp->v_bufobj.bo_ops = &ffs_ops;
950 	BO_LOCK(&odevvp->v_bufobj);
951 	odevvp->v_bufobj.bo_flag |= BO_NOBUFS;
952 	BO_UNLOCK(&odevvp->v_bufobj);
953 	VOP_UNLOCK(devvp);
954 	if (dev->si_iosize_max != 0)
955 		mp->mnt_iosize_max = dev->si_iosize_max;
956 	if (mp->mnt_iosize_max > maxphys)
957 		mp->mnt_iosize_max = maxphys;
958 	if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
959 		error = EINVAL;
960 		vfs_mount_error(mp,
961 		    "Invalid sectorsize %d for superblock size %d",
962 		    cp->provider->sectorsize, SBLOCKSIZE);
963 		goto out;
964 	}
965 	/* fetch the superblock and summary information */
966 	loc = STDSB;
967 	if ((mp->mnt_flag & (MNT_ROOTFS | MNT_FORCE)) != 0)
968 		loc = STDSB_NOHASHFAIL;
969 	if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0)
970 		goto out;
971 	fs->fs_flags &= ~FS_UNCLEAN;
972 	if (fs->fs_clean == 0) {
973 		fs->fs_flags |= FS_UNCLEAN;
974 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
975 		    ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
976 		     (fs->fs_flags & FS_DOSOFTDEP))) {
977 			printf("WARNING: %s was not properly dismounted\n",
978 			    fs->fs_fsmnt);
979 		} else {
980 			vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
981 			    fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
982 			    (fs->fs_flags & FS_SUJ) == 0 ? "" :
983 			    " Forced mount will invalidate journal contents");
984 			error = EPERM;
985 			goto out;
986 		}
987 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
988 		    (mp->mnt_flag & MNT_FORCE)) {
989 			printf("WARNING: %s: lost blocks %jd files %d\n",
990 			    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
991 			    fs->fs_pendinginodes);
992 			fs->fs_pendingblocks = 0;
993 			fs->fs_pendinginodes = 0;
994 		}
995 	}
996 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
997 		printf("WARNING: %s: mount pending error: blocks %jd "
998 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
999 		    fs->fs_pendinginodes);
1000 		fs->fs_pendingblocks = 0;
1001 		fs->fs_pendinginodes = 0;
1002 	}
1003 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
1004 #ifdef UFS_GJOURNAL
1005 		/*
1006 		 * Get journal provider name.
1007 		 */
1008 		len = 1024;
1009 		mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK);
1010 		if (g_io_getattr("GJOURNAL::provider", cp, &len,
1011 		    mp->mnt_gjprovider) == 0) {
1012 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len,
1013 			    M_UFSMNT, M_WAITOK);
1014 			MNT_ILOCK(mp);
1015 			mp->mnt_flag |= MNT_GJOURNAL;
1016 			MNT_IUNLOCK(mp);
1017 		} else {
1018 			if ((mp->mnt_flag & MNT_RDONLY) == 0)
1019 				printf("WARNING: %s: GJOURNAL flag on fs "
1020 				    "but no gjournal provider below\n",
1021 				    mp->mnt_stat.f_mntonname);
1022 			free(mp->mnt_gjprovider, M_UFSMNT);
1023 			mp->mnt_gjprovider = NULL;
1024 		}
1025 #else
1026 		printf("WARNING: %s: GJOURNAL flag on fs but no "
1027 		    "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
1028 #endif
1029 	} else {
1030 		mp->mnt_gjprovider = NULL;
1031 	}
1032 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1033 	ump->um_cp = cp;
1034 	ump->um_bo = &devvp->v_bufobj;
1035 	ump->um_fs = fs;
1036 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1037 		ump->um_fstype = UFS1;
1038 		ump->um_balloc = ffs_balloc_ufs1;
1039 	} else {
1040 		ump->um_fstype = UFS2;
1041 		ump->um_balloc = ffs_balloc_ufs2;
1042 	}
1043 	ump->um_blkatoff = ffs_blkatoff;
1044 	ump->um_truncate = ffs_truncate;
1045 	ump->um_update = ffs_update;
1046 	ump->um_valloc = ffs_valloc;
1047 	ump->um_vfree = ffs_vfree;
1048 	ump->um_ifree = ffs_ifree;
1049 	ump->um_rdonly = ffs_rdonly;
1050 	ump->um_snapgone = ffs_snapgone;
1051 	if ((mp->mnt_flag & MNT_UNTRUSTED) != 0)
1052 		ump->um_check_blkno = ffs_check_blkno;
1053 	else
1054 		ump->um_check_blkno = NULL;
1055 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
1056 	sx_init(&ump->um_checkpath_lock, "uchpth");
1057 	ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc);
1058 	fs->fs_ronly = ronly;
1059 	fs->fs_active = NULL;
1060 	mp->mnt_data = ump;
1061 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
1062 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
1063 	nmp = NULL;
1064 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
1065 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
1066 		if (nmp)
1067 			vfs_rel(nmp);
1068 		vfs_getnewfsid(mp);
1069 	}
1070 	ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
1071 	MNT_ILOCK(mp);
1072 	mp->mnt_flag |= MNT_LOCAL;
1073 	MNT_IUNLOCK(mp);
1074 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
1075 #ifdef MAC
1076 		MNT_ILOCK(mp);
1077 		mp->mnt_flag |= MNT_MULTILABEL;
1078 		MNT_IUNLOCK(mp);
1079 #else
1080 		printf("WARNING: %s: multilabel flag on fs but "
1081 		    "no MAC support\n", mp->mnt_stat.f_mntonname);
1082 #endif
1083 	}
1084 	if ((fs->fs_flags & FS_ACLS) != 0) {
1085 #ifdef UFS_ACL
1086 		MNT_ILOCK(mp);
1087 
1088 		if (mp->mnt_flag & MNT_NFS4ACLS)
1089 			printf("WARNING: %s: ACLs flag on fs conflicts with "
1090 			    "\"nfsv4acls\" mount option; option ignored\n",
1091 			    mp->mnt_stat.f_mntonname);
1092 		mp->mnt_flag &= ~MNT_NFS4ACLS;
1093 		mp->mnt_flag |= MNT_ACLS;
1094 
1095 		MNT_IUNLOCK(mp);
1096 #else
1097 		printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
1098 		    mp->mnt_stat.f_mntonname);
1099 #endif
1100 	}
1101 	if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
1102 #ifdef UFS_ACL
1103 		MNT_ILOCK(mp);
1104 
1105 		if (mp->mnt_flag & MNT_ACLS)
1106 			printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
1107 			    "with \"acls\" mount option; option ignored\n",
1108 			    mp->mnt_stat.f_mntonname);
1109 		mp->mnt_flag &= ~MNT_ACLS;
1110 		mp->mnt_flag |= MNT_NFS4ACLS;
1111 
1112 		MNT_IUNLOCK(mp);
1113 #else
1114 		printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
1115 		    "ACLs support\n", mp->mnt_stat.f_mntonname);
1116 #endif
1117 	}
1118 	if ((fs->fs_flags & FS_TRIM) != 0) {
1119 		len = sizeof(int);
1120 		if (g_io_getattr("GEOM::candelete", cp, &len,
1121 		    &candelete) == 0) {
1122 			if (candelete)
1123 				ump->um_flags |= UM_CANDELETE;
1124 			else
1125 				printf("WARNING: %s: TRIM flag on fs but disk "
1126 				    "does not support TRIM\n",
1127 				    mp->mnt_stat.f_mntonname);
1128 		} else {
1129 			printf("WARNING: %s: TRIM flag on fs but disk does "
1130 			    "not confirm that it supports TRIM\n",
1131 			    mp->mnt_stat.f_mntonname);
1132 		}
1133 		if (((ump->um_flags) & UM_CANDELETE) != 0) {
1134 			ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1135 			    taskqueue_thread_enqueue, &ump->um_trim_tq);
1136 			taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1137 			    "%s trim", mp->mnt_stat.f_mntonname);
1138 			ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM,
1139 			    &ump->um_trimlisthashsize);
1140 		}
1141 	}
1142 
1143 	len = sizeof(int);
1144 	if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) {
1145 		if (canspeedup)
1146 			ump->um_flags |= UM_CANSPEEDUP;
1147 	}
1148 
1149 	ump->um_mountp = mp;
1150 	ump->um_dev = dev;
1151 	ump->um_devvp = devvp;
1152 	ump->um_odevvp = odevvp;
1153 	ump->um_nindir = fs->fs_nindir;
1154 	ump->um_bptrtodb = fs->fs_fsbtodb;
1155 	ump->um_seqinc = fs->fs_frag;
1156 	for (i = 0; i < MAXQUOTAS; i++)
1157 		ump->um_quotas[i] = NULLVP;
1158 #ifdef UFS_EXTATTR
1159 	ufs_extattr_uepm_init(&ump->um_extattr);
1160 #endif
1161 	/*
1162 	 * Set FS local "last mounted on" information (NULL pad)
1163 	 */
1164 	bzero(fs->fs_fsmnt, MAXMNTLEN);
1165 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1166 	mp->mnt_stat.f_iosize = fs->fs_bsize;
1167 
1168 	if (mp->mnt_flag & MNT_ROOTFS) {
1169 		/*
1170 		 * Root mount; update timestamp in mount structure.
1171 		 * this will be used by the common root mount code
1172 		 * to update the system clock.
1173 		 */
1174 		mp->mnt_time = fs->fs_time;
1175 	}
1176 
1177 	if (ronly == 0) {
1178 		fs->fs_mtime = time_second;
1179 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
1180 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1181 			ffs_flushfiles(mp, FORCECLOSE, td);
1182 			goto out;
1183 		}
1184 		if (fs->fs_snapinum[0] != 0)
1185 			ffs_snapshot_mount(mp);
1186 		fs->fs_fmod = 1;
1187 		fs->fs_clean = 0;
1188 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
1189 	}
1190 	/*
1191 	 * Initialize filesystem state information in mount struct.
1192 	 */
1193 	MNT_ILOCK(mp);
1194 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1195 	    MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1196 	MNT_IUNLOCK(mp);
1197 #ifdef UFS_EXTATTR
1198 #ifdef UFS_EXTATTR_AUTOSTART
1199 	/*
1200 	 *
1201 	 * Auto-starting does the following:
1202 	 *	- check for /.attribute in the fs, and extattr_start if so
1203 	 *	- for each file in .attribute, enable that file with
1204 	 * 	  an attribute of the same name.
1205 	 * Not clear how to report errors -- probably eat them.
1206 	 * This would all happen while the filesystem was busy/not
1207 	 * available, so would effectively be "atomic".
1208 	 */
1209 	(void) ufs_extattr_autostart(mp, td);
1210 #endif /* !UFS_EXTATTR_AUTOSTART */
1211 #endif /* !UFS_EXTATTR */
1212 	return (0);
1213 out:
1214 	if (fs != NULL) {
1215 		free(fs->fs_csp, M_UFSMNT);
1216 		free(fs->fs_si, M_UFSMNT);
1217 		free(fs, M_UFSMNT);
1218 	}
1219 	if (cp != NULL) {
1220 		g_topology_lock();
1221 		g_vfs_close(cp);
1222 		g_topology_unlock();
1223 	}
1224 	if (ump != NULL) {
1225 		mtx_destroy(UFS_MTX(ump));
1226 		sx_destroy(&ump->um_checkpath_lock);
1227 		if (mp->mnt_gjprovider != NULL) {
1228 			free(mp->mnt_gjprovider, M_UFSMNT);
1229 			mp->mnt_gjprovider = NULL;
1230 		}
1231 		MPASS(ump->um_softdep == NULL);
1232 		free(ump, M_UFSMNT);
1233 		mp->mnt_data = NULL;
1234 	}
1235 	BO_LOCK(&odevvp->v_bufobj);
1236 	odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1237 	BO_UNLOCK(&odevvp->v_bufobj);
1238 	atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1239 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1240 	mntfs_freevp(devvp);
1241 	dev_rel(dev);
1242 	return (error);
1243 }
1244 
1245 /*
1246  * A read function for use by filesystem-layer routines.
1247  */
1248 static int
1249 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size)
1250 {
1251 	struct buf *bp;
1252 	int error;
1253 
1254 	KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp));
1255 	*bufp = malloc(size, M_UFSMNT, M_WAITOK);
1256 	if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED,
1257 	    &bp)) != 0)
1258 		return (error);
1259 	bcopy(bp->b_data, *bufp, size);
1260 	bp->b_flags |= B_INVAL | B_NOCACHE;
1261 	brelse(bp);
1262 	return (0);
1263 }
1264 
1265 static int bigcgs = 0;
1266 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1267 
1268 /*
1269  * Sanity checks for loading old filesystem superblocks.
1270  * See ffs_oldfscompat_write below for unwound actions.
1271  *
1272  * XXX - Parts get retired eventually.
1273  * Unfortunately new bits get added.
1274  */
1275 static void
1276 ffs_oldfscompat_read(fs, ump, sblockloc)
1277 	struct fs *fs;
1278 	struct ufsmount *ump;
1279 	ufs2_daddr_t sblockloc;
1280 {
1281 	off_t maxfilesize;
1282 
1283 	/*
1284 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
1285 	 */
1286 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1287 		fs->fs_flags = fs->fs_old_flags;
1288 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
1289 		fs->fs_sblockloc = sblockloc;
1290 	}
1291 	/*
1292 	 * If not yet done, update UFS1 superblock with new wider fields.
1293 	 */
1294 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1295 		fs->fs_maxbsize = fs->fs_bsize;
1296 		fs->fs_time = fs->fs_old_time;
1297 		fs->fs_size = fs->fs_old_size;
1298 		fs->fs_dsize = fs->fs_old_dsize;
1299 		fs->fs_csaddr = fs->fs_old_csaddr;
1300 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1301 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1302 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1303 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1304 	}
1305 	if (fs->fs_magic == FS_UFS1_MAGIC &&
1306 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
1307 		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1308 		fs->fs_qbmask = ~fs->fs_bmask;
1309 		fs->fs_qfmask = ~fs->fs_fmask;
1310 	}
1311 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1312 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1313 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1314 		if (fs->fs_maxfilesize > maxfilesize)
1315 			fs->fs_maxfilesize = maxfilesize;
1316 	}
1317 	/* Compatibility for old filesystems */
1318 	if (fs->fs_avgfilesize <= 0)
1319 		fs->fs_avgfilesize = AVFILESIZ;
1320 	if (fs->fs_avgfpdir <= 0)
1321 		fs->fs_avgfpdir = AFPDIR;
1322 	if (bigcgs) {
1323 		fs->fs_save_cgsize = fs->fs_cgsize;
1324 		fs->fs_cgsize = fs->fs_bsize;
1325 	}
1326 }
1327 
1328 /*
1329  * Unwinding superblock updates for old filesystems.
1330  * See ffs_oldfscompat_read above for details.
1331  *
1332  * XXX - Parts get retired eventually.
1333  * Unfortunately new bits get added.
1334  */
1335 void
1336 ffs_oldfscompat_write(fs, ump)
1337 	struct fs *fs;
1338 	struct ufsmount *ump;
1339 {
1340 
1341 	/*
1342 	 * Copy back UFS2 updated fields that UFS1 inspects.
1343 	 */
1344 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1345 		fs->fs_old_time = fs->fs_time;
1346 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1347 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1348 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1349 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1350 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1351 	}
1352 	if (bigcgs) {
1353 		fs->fs_cgsize = fs->fs_save_cgsize;
1354 		fs->fs_save_cgsize = 0;
1355 	}
1356 }
1357 
1358 /*
1359  * unmount system call
1360  */
1361 static int
1362 ffs_unmount(mp, mntflags)
1363 	struct mount *mp;
1364 	int mntflags;
1365 {
1366 	struct thread *td;
1367 	struct ufsmount *ump = VFSTOUFS(mp);
1368 	struct fs *fs;
1369 	int error, flags, susp;
1370 #ifdef UFS_EXTATTR
1371 	int e_restart;
1372 #endif
1373 
1374 	flags = 0;
1375 	td = curthread;
1376 	fs = ump->um_fs;
1377 	if (mntflags & MNT_FORCE)
1378 		flags |= FORCECLOSE;
1379 	susp = fs->fs_ronly == 0;
1380 #ifdef UFS_EXTATTR
1381 	if ((error = ufs_extattr_stop(mp, td))) {
1382 		if (error != EOPNOTSUPP)
1383 			printf("WARNING: unmount %s: ufs_extattr_stop "
1384 			    "returned errno %d\n", mp->mnt_stat.f_mntonname,
1385 			    error);
1386 		e_restart = 0;
1387 	} else {
1388 		ufs_extattr_uepm_destroy(&ump->um_extattr);
1389 		e_restart = 1;
1390 	}
1391 #endif
1392 	if (susp) {
1393 		error = vfs_write_suspend_umnt(mp);
1394 		if (error != 0)
1395 			goto fail1;
1396 	}
1397 	if (MOUNTEDSOFTDEP(mp))
1398 		error = softdep_flushfiles(mp, flags, td);
1399 	else
1400 		error = ffs_flushfiles(mp, flags, td);
1401 	if (error != 0 && !ffs_fsfail_cleanup(ump, error))
1402 		goto fail;
1403 
1404 	UFS_LOCK(ump);
1405 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1406 		printf("WARNING: unmount %s: pending error: blocks %jd "
1407 		    "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1408 		    fs->fs_pendinginodes);
1409 		fs->fs_pendingblocks = 0;
1410 		fs->fs_pendinginodes = 0;
1411 	}
1412 	UFS_UNLOCK(ump);
1413 	if (MOUNTEDSOFTDEP(mp))
1414 		softdep_unmount(mp);
1415 	MPASS(ump->um_softdep == NULL);
1416 	if (fs->fs_ronly == 0) {
1417 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1418 		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1419 		if (ffs_fsfail_cleanup(ump, error))
1420 			error = 0;
1421 		if (error != 0 && !ffs_fsfail_cleanup(ump, error)) {
1422 			fs->fs_clean = 0;
1423 			goto fail;
1424 		}
1425 	}
1426 	if (susp)
1427 		vfs_write_resume(mp, VR_START_WRITE);
1428 	if (ump->um_trim_tq != NULL) {
1429 		while (ump->um_trim_inflight != 0)
1430 			pause("ufsutr", hz);
1431 		taskqueue_drain_all(ump->um_trim_tq);
1432 		taskqueue_free(ump->um_trim_tq);
1433 		free (ump->um_trimhash, M_TRIM);
1434 	}
1435 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1436 	g_topology_lock();
1437 	g_vfs_close(ump->um_cp);
1438 	g_topology_unlock();
1439 	BO_LOCK(&ump->um_odevvp->v_bufobj);
1440 	ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1441 	BO_UNLOCK(&ump->um_odevvp->v_bufobj);
1442 	atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1443 	mntfs_freevp(ump->um_devvp);
1444 	vrele(ump->um_odevvp);
1445 	dev_rel(ump->um_dev);
1446 	mtx_destroy(UFS_MTX(ump));
1447 	sx_destroy(&ump->um_checkpath_lock);
1448 	if (mp->mnt_gjprovider != NULL) {
1449 		free(mp->mnt_gjprovider, M_UFSMNT);
1450 		mp->mnt_gjprovider = NULL;
1451 	}
1452 	free(fs->fs_csp, M_UFSMNT);
1453 	free(fs->fs_si, M_UFSMNT);
1454 	free(fs, M_UFSMNT);
1455 	free(ump, M_UFSMNT);
1456 	mp->mnt_data = NULL;
1457 	MNT_ILOCK(mp);
1458 	mp->mnt_flag &= ~MNT_LOCAL;
1459 	MNT_IUNLOCK(mp);
1460 	if (td->td_su == mp) {
1461 		td->td_su = NULL;
1462 		vfs_rel(mp);
1463 	}
1464 	return (error);
1465 
1466 fail:
1467 	if (susp)
1468 		vfs_write_resume(mp, VR_START_WRITE);
1469 fail1:
1470 #ifdef UFS_EXTATTR
1471 	if (e_restart) {
1472 		ufs_extattr_uepm_init(&ump->um_extattr);
1473 #ifdef UFS_EXTATTR_AUTOSTART
1474 		(void) ufs_extattr_autostart(mp, td);
1475 #endif
1476 	}
1477 #endif
1478 
1479 	return (error);
1480 }
1481 
1482 /*
1483  * Flush out all the files in a filesystem.
1484  */
1485 int
1486 ffs_flushfiles(mp, flags, td)
1487 	struct mount *mp;
1488 	int flags;
1489 	struct thread *td;
1490 {
1491 	struct ufsmount *ump;
1492 	int qerror, error;
1493 
1494 	ump = VFSTOUFS(mp);
1495 	qerror = 0;
1496 #ifdef QUOTA
1497 	if (mp->mnt_flag & MNT_QUOTA) {
1498 		int i;
1499 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1500 		if (error)
1501 			return (error);
1502 		for (i = 0; i < MAXQUOTAS; i++) {
1503 			error = quotaoff(td, mp, i);
1504 			if (error != 0) {
1505 				if ((flags & EARLYFLUSH) == 0)
1506 					return (error);
1507 				else
1508 					qerror = error;
1509 			}
1510 		}
1511 
1512 		/*
1513 		 * Here we fall through to vflush again to ensure that
1514 		 * we have gotten rid of all the system vnodes, unless
1515 		 * quotas must not be closed.
1516 		 */
1517 	}
1518 #endif
1519 	/* devvp is not locked there */
1520 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1521 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1522 			return (error);
1523 		ffs_snapshot_unmount(mp);
1524 		flags |= FORCECLOSE;
1525 		/*
1526 		 * Here we fall through to vflush again to ensure
1527 		 * that we have gotten rid of all the system vnodes.
1528 		 */
1529 	}
1530 
1531 	/*
1532 	 * Do not close system files if quotas were not closed, to be
1533 	 * able to sync the remaining dquots.  The freeblks softupdate
1534 	 * workitems might hold a reference on a dquot, preventing
1535 	 * quotaoff() from completing.  Next round of
1536 	 * softdep_flushworklist() iteration should process the
1537 	 * blockers, allowing the next run of quotaoff() to finally
1538 	 * flush held dquots.
1539 	 *
1540 	 * Otherwise, flush all the files.
1541 	 */
1542 	if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1543 		return (error);
1544 
1545 	/*
1546 	 * Flush filesystem metadata.
1547 	 */
1548 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1549 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1550 	VOP_UNLOCK(ump->um_devvp);
1551 	return (error);
1552 }
1553 
1554 /*
1555  * Get filesystem statistics.
1556  */
1557 static int
1558 ffs_statfs(mp, sbp)
1559 	struct mount *mp;
1560 	struct statfs *sbp;
1561 {
1562 	struct ufsmount *ump;
1563 	struct fs *fs;
1564 
1565 	ump = VFSTOUFS(mp);
1566 	fs = ump->um_fs;
1567 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1568 		panic("ffs_statfs");
1569 	sbp->f_version = STATFS_VERSION;
1570 	sbp->f_bsize = fs->fs_fsize;
1571 	sbp->f_iosize = fs->fs_bsize;
1572 	sbp->f_blocks = fs->fs_dsize;
1573 	UFS_LOCK(ump);
1574 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1575 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1576 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1577 	    dbtofsb(fs, fs->fs_pendingblocks);
1578 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO;
1579 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1580 	UFS_UNLOCK(ump);
1581 	sbp->f_namemax = UFS_MAXNAMLEN;
1582 	return (0);
1583 }
1584 
1585 static bool
1586 sync_doupdate(struct inode *ip)
1587 {
1588 
1589 	return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1590 	    IN_UPDATE)) != 0);
1591 }
1592 
1593 static int
1594 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused)
1595 {
1596 	struct inode *ip;
1597 
1598 	/*
1599 	 * Flags are safe to access because ->v_data invalidation
1600 	 * is held off by listmtx.
1601 	 */
1602 	if (vp->v_type == VNON)
1603 		return (false);
1604 	ip = VTOI(vp);
1605 	if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0)
1606 		return (false);
1607 	return (true);
1608 }
1609 
1610 /*
1611  * For a lazy sync, we only care about access times, quotas and the
1612  * superblock.  Other filesystem changes are already converted to
1613  * cylinder group blocks or inode blocks updates and are written to
1614  * disk by syncer.
1615  */
1616 static int
1617 ffs_sync_lazy(mp)
1618      struct mount *mp;
1619 {
1620 	struct vnode *mvp, *vp;
1621 	struct inode *ip;
1622 	int allerror, error;
1623 
1624 	allerror = 0;
1625 	if ((mp->mnt_flag & MNT_NOATIME) != 0) {
1626 #ifdef QUOTA
1627 		qsync(mp);
1628 #endif
1629 		goto sbupdate;
1630 	}
1631 	MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) {
1632 		if (vp->v_type == VNON) {
1633 			VI_UNLOCK(vp);
1634 			continue;
1635 		}
1636 		ip = VTOI(vp);
1637 
1638 		/*
1639 		 * The IN_ACCESS flag is converted to IN_MODIFIED by
1640 		 * ufs_close() and ufs_getattr() by the calls to
1641 		 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1642 		 * Test also all the other timestamp flags too, to pick up
1643 		 * any other cases that could be missed.
1644 		 */
1645 		if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1646 			VI_UNLOCK(vp);
1647 			continue;
1648 		}
1649 		if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0)
1650 			continue;
1651 #ifdef QUOTA
1652 		qsyncvp(vp);
1653 #endif
1654 		if (sync_doupdate(ip))
1655 			error = ffs_update(vp, 0);
1656 		if (error != 0)
1657 			allerror = error;
1658 		vput(vp);
1659 	}
1660 sbupdate:
1661 	if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1662 	    (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1663 		allerror = error;
1664 	return (allerror);
1665 }
1666 
1667 /*
1668  * Go through the disk queues to initiate sandbagged IO;
1669  * go through the inodes to write those that have been modified;
1670  * initiate the writing of the super block if it has been modified.
1671  *
1672  * Note: we are always called with the filesystem marked busy using
1673  * vfs_busy().
1674  */
1675 static int
1676 ffs_sync(mp, waitfor)
1677 	struct mount *mp;
1678 	int waitfor;
1679 {
1680 	struct vnode *mvp, *vp, *devvp;
1681 	struct thread *td;
1682 	struct inode *ip;
1683 	struct ufsmount *ump = VFSTOUFS(mp);
1684 	struct fs *fs;
1685 	int error, count, lockreq, allerror = 0;
1686 	int suspend;
1687 	int suspended;
1688 	int secondary_writes;
1689 	int secondary_accwrites;
1690 	int softdep_deps;
1691 	int softdep_accdeps;
1692 	struct bufobj *bo;
1693 
1694 	suspend = 0;
1695 	suspended = 0;
1696 	td = curthread;
1697 	fs = ump->um_fs;
1698 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0)
1699 		panic("%s: ffs_sync: modification on read-only filesystem",
1700 		    fs->fs_fsmnt);
1701 	if (waitfor == MNT_LAZY) {
1702 		if (!rebooting)
1703 			return (ffs_sync_lazy(mp));
1704 		waitfor = MNT_NOWAIT;
1705 	}
1706 
1707 	/*
1708 	 * Write back each (modified) inode.
1709 	 */
1710 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1711 	if (waitfor == MNT_SUSPEND) {
1712 		suspend = 1;
1713 		waitfor = MNT_WAIT;
1714 	}
1715 	if (waitfor == MNT_WAIT)
1716 		lockreq = LK_EXCLUSIVE;
1717 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1718 loop:
1719 	/* Grab snapshot of secondary write counts */
1720 	MNT_ILOCK(mp);
1721 	secondary_writes = mp->mnt_secondary_writes;
1722 	secondary_accwrites = mp->mnt_secondary_accwrites;
1723 	MNT_IUNLOCK(mp);
1724 
1725 	/* Grab snapshot of softdep dependency counts */
1726 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1727 
1728 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1729 		/*
1730 		 * Depend on the vnode interlock to keep things stable enough
1731 		 * for a quick test.  Since there might be hundreds of
1732 		 * thousands of vnodes, we cannot afford even a subroutine
1733 		 * call unless there's a good chance that we have work to do.
1734 		 */
1735 		if (vp->v_type == VNON) {
1736 			VI_UNLOCK(vp);
1737 			continue;
1738 		}
1739 		ip = VTOI(vp);
1740 		if ((ip->i_flag &
1741 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1742 		    vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1743 			VI_UNLOCK(vp);
1744 			continue;
1745 		}
1746 		if ((error = vget(vp, lockreq)) != 0) {
1747 			if (error == ENOENT || error == ENOLCK) {
1748 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1749 				goto loop;
1750 			}
1751 			continue;
1752 		}
1753 #ifdef QUOTA
1754 		qsyncvp(vp);
1755 #endif
1756 		for (;;) {
1757 			error = ffs_syncvnode(vp, waitfor, 0);
1758 			if (error == ERELOOKUP)
1759 				continue;
1760 			if (error != 0)
1761 				allerror = error;
1762 			break;
1763 		}
1764 		vput(vp);
1765 	}
1766 	/*
1767 	 * Force stale filesystem control information to be flushed.
1768 	 */
1769 	if (waitfor == MNT_WAIT || rebooting) {
1770 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1771 			allerror = error;
1772 		if (ffs_fsfail_cleanup(ump, allerror))
1773 			allerror = 0;
1774 		/* Flushed work items may create new vnodes to clean */
1775 		if (allerror == 0 && count)
1776 			goto loop;
1777 	}
1778 
1779 	devvp = ump->um_devvp;
1780 	bo = &devvp->v_bufobj;
1781 	BO_LOCK(bo);
1782 	if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1783 		BO_UNLOCK(bo);
1784 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1785 		error = VOP_FSYNC(devvp, waitfor, td);
1786 		VOP_UNLOCK(devvp);
1787 		if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1788 			error = ffs_sbupdate(ump, waitfor, 0);
1789 		if (error != 0)
1790 			allerror = error;
1791 		if (ffs_fsfail_cleanup(ump, allerror))
1792 			allerror = 0;
1793 		if (allerror == 0 && waitfor == MNT_WAIT)
1794 			goto loop;
1795 	} else if (suspend != 0) {
1796 		if (softdep_check_suspend(mp,
1797 					  devvp,
1798 					  softdep_deps,
1799 					  softdep_accdeps,
1800 					  secondary_writes,
1801 					  secondary_accwrites) != 0) {
1802 			MNT_IUNLOCK(mp);
1803 			goto loop;	/* More work needed */
1804 		}
1805 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1806 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1807 		MNT_IUNLOCK(mp);
1808 		suspended = 1;
1809 	} else
1810 		BO_UNLOCK(bo);
1811 	/*
1812 	 * Write back modified superblock.
1813 	 */
1814 	if (fs->fs_fmod != 0 &&
1815 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1816 		allerror = error;
1817 	if (ffs_fsfail_cleanup(ump, allerror))
1818 		allerror = 0;
1819 	return (allerror);
1820 }
1821 
1822 int
1823 ffs_vget(mp, ino, flags, vpp)
1824 	struct mount *mp;
1825 	ino_t ino;
1826 	int flags;
1827 	struct vnode **vpp;
1828 {
1829 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1830 }
1831 
1832 int
1833 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1834 	struct mount *mp;
1835 	ino_t ino;
1836 	int flags;
1837 	struct vnode **vpp;
1838 	int ffs_flags;
1839 {
1840 	struct fs *fs;
1841 	struct inode *ip;
1842 	struct ufsmount *ump;
1843 	struct buf *bp;
1844 	struct vnode *vp;
1845 	daddr_t dbn;
1846 	int error;
1847 
1848 	MPASS((ffs_flags & (FFSV_REPLACE | FFSV_REPLACE_DOOMED)) == 0 ||
1849 	    (flags & LK_EXCLUSIVE) != 0);
1850 
1851 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1852 	if (error != 0)
1853 		return (error);
1854 	if (*vpp != NULL) {
1855 		if ((ffs_flags & FFSV_REPLACE) == 0 ||
1856 		    ((ffs_flags & FFSV_REPLACE_DOOMED) == 0 ||
1857 		    !VN_IS_DOOMED(*vpp)))
1858 			return (0);
1859 		vgone(*vpp);
1860 		vput(*vpp);
1861 	}
1862 
1863 	/*
1864 	 * We must promote to an exclusive lock for vnode creation.  This
1865 	 * can happen if lookup is passed LOCKSHARED.
1866 	 */
1867 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1868 		flags &= ~LK_TYPE_MASK;
1869 		flags |= LK_EXCLUSIVE;
1870 	}
1871 
1872 	/*
1873 	 * We do not lock vnode creation as it is believed to be too
1874 	 * expensive for such rare case as simultaneous creation of vnode
1875 	 * for same ino by different processes. We just allow them to race
1876 	 * and check later to decide who wins. Let the race begin!
1877 	 */
1878 
1879 	ump = VFSTOUFS(mp);
1880 	fs = ump->um_fs;
1881 	ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO);
1882 
1883 	/* Allocate a new vnode/inode. */
1884 	error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1885 	    &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1886 	if (error) {
1887 		*vpp = NULL;
1888 		uma_zfree_smr(uma_inode, ip);
1889 		return (error);
1890 	}
1891 	/*
1892 	 * FFS supports recursive locking.
1893 	 */
1894 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
1895 	VN_LOCK_AREC(vp);
1896 	vp->v_data = ip;
1897 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1898 	ip->i_vnode = vp;
1899 	ip->i_ump = ump;
1900 	ip->i_number = ino;
1901 	ip->i_ea_refs = 0;
1902 	ip->i_nextclustercg = -1;
1903 	ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2;
1904 	ip->i_mode = 0; /* ensure error cases below throw away vnode */
1905 	cluster_init_vn(&ip->i_clusterw);
1906 #ifdef DIAGNOSTIC
1907 	ufs_init_trackers(ip);
1908 #endif
1909 #ifdef QUOTA
1910 	{
1911 		int i;
1912 		for (i = 0; i < MAXQUOTAS; i++)
1913 			ip->i_dquot[i] = NODQUOT;
1914 	}
1915 #endif
1916 
1917 	if (ffs_flags & FFSV_FORCEINSMQ)
1918 		vp->v_vflag |= VV_FORCEINSMQ;
1919 	error = insmntque(vp, mp);
1920 	if (error != 0) {
1921 		uma_zfree_smr(uma_inode, ip);
1922 		*vpp = NULL;
1923 		return (error);
1924 	}
1925 	vp->v_vflag &= ~VV_FORCEINSMQ;
1926 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1927 	if (error != 0)
1928 		return (error);
1929 	if (*vpp != NULL) {
1930 		/*
1931 		 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set)
1932 		 * operate on empty inode, which must not be found by
1933 		 * other threads until fully filled.  Vnode for empty
1934 		 * inode must be not re-inserted on the hash by other
1935 		 * thread, after removal by us at the beginning.
1936 		 */
1937 		MPASS((ffs_flags & FFSV_REPLACE) == 0);
1938 		return (0);
1939 	}
1940 
1941 	/* Read in the disk contents for the inode, copy into the inode. */
1942 	dbn = fsbtodb(fs, ino_to_fsba(fs, ino));
1943 	error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize,
1944 	    NULL, NULL, 0, NOCRED, 0, NULL, &bp);
1945 	if (error != 0) {
1946 		/*
1947 		 * The inode does not contain anything useful, so it would
1948 		 * be misleading to leave it on its hash chain. With mode
1949 		 * still zero, it will be unlinked and returned to the free
1950 		 * list by vput().
1951 		 */
1952 		vgone(vp);
1953 		vput(vp);
1954 		*vpp = NULL;
1955 		return (error);
1956 	}
1957 	if (I_IS_UFS1(ip))
1958 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1959 	else
1960 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1961 	if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) {
1962 		bqrelse(bp);
1963 		vgone(vp);
1964 		vput(vp);
1965 		*vpp = NULL;
1966 		return (error);
1967 	}
1968 	if (DOINGSOFTDEP(vp) && (!fs->fs_ronly ||
1969 	    (ffs_flags & FFSV_FORCEINODEDEP) != 0))
1970 		softdep_load_inodeblock(ip);
1971 	else
1972 		ip->i_effnlink = ip->i_nlink;
1973 	bqrelse(bp);
1974 
1975 	/*
1976 	 * Initialize the vnode from the inode, check for aliases.
1977 	 * Note that the underlying vnode may have changed.
1978 	 */
1979 	error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2,
1980 	    &vp);
1981 	if (error) {
1982 		vgone(vp);
1983 		vput(vp);
1984 		*vpp = NULL;
1985 		return (error);
1986 	}
1987 
1988 	/*
1989 	 * Finish inode initialization.
1990 	 */
1991 	if (vp->v_type != VFIFO) {
1992 		/* FFS supports shared locking for all files except fifos. */
1993 		VN_LOCK_ASHARE(vp);
1994 	}
1995 
1996 	/*
1997 	 * Set up a generation number for this inode if it does not
1998 	 * already have one. This should only happen on old filesystems.
1999 	 */
2000 	if (ip->i_gen == 0) {
2001 		while (ip->i_gen == 0)
2002 			ip->i_gen = arc4random();
2003 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
2004 			UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
2005 			DIP_SET(ip, i_gen, ip->i_gen);
2006 		}
2007 	}
2008 #ifdef MAC
2009 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
2010 		/*
2011 		 * If this vnode is already allocated, and we're running
2012 		 * multi-label, attempt to perform a label association
2013 		 * from the extended attributes on the inode.
2014 		 */
2015 		error = mac_vnode_associate_extattr(mp, vp);
2016 		if (error) {
2017 			/* ufs_inactive will release ip->i_devvp ref. */
2018 			vgone(vp);
2019 			vput(vp);
2020 			*vpp = NULL;
2021 			return (error);
2022 		}
2023 	}
2024 #endif
2025 
2026 	*vpp = vp;
2027 	return (0);
2028 }
2029 
2030 /*
2031  * File handle to vnode
2032  *
2033  * Have to be really careful about stale file handles:
2034  * - check that the inode number is valid
2035  * - for UFS2 check that the inode number is initialized
2036  * - call ffs_vget() to get the locked inode
2037  * - check for an unallocated inode (i_mode == 0)
2038  * - check that the given client host has export rights and return
2039  *   those rights via. exflagsp and credanonp
2040  */
2041 static int
2042 ffs_fhtovp(mp, fhp, flags, vpp)
2043 	struct mount *mp;
2044 	struct fid *fhp;
2045 	int flags;
2046 	struct vnode **vpp;
2047 {
2048 	struct ufid *ufhp;
2049 
2050 	ufhp = (struct ufid *)fhp;
2051 	return (ffs_inotovp(mp, ufhp->ufid_ino, ufhp->ufid_gen, flags,
2052 	    vpp, 0));
2053 }
2054 
2055 int
2056 ffs_inotovp(mp, ino, gen, lflags, vpp, ffs_flags)
2057 	struct mount *mp;
2058 	ino_t ino;
2059 	u_int64_t gen;
2060 	int lflags;
2061 	struct vnode **vpp;
2062 	int ffs_flags;
2063 {
2064 	struct ufsmount *ump;
2065 	struct vnode *nvp;
2066 	struct inode *ip;
2067 	struct fs *fs;
2068 	struct cg *cgp;
2069 	struct buf *bp;
2070 	u_int cg;
2071 	int error;
2072 
2073 	ump = VFSTOUFS(mp);
2074 	fs = ump->um_fs;
2075 	*vpp = NULL;
2076 
2077 	if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
2078 		return (ESTALE);
2079 
2080 	/*
2081 	 * Need to check if inode is initialized because UFS2 does lazy
2082 	 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
2083 	 */
2084 	if (fs->fs_magic == FS_UFS2_MAGIC) {
2085 		cg = ino_to_cg(fs, ino);
2086 		error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp);
2087 		if (error != 0)
2088 			return (error);
2089 		if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
2090 			brelse(bp);
2091 			return (ESTALE);
2092 		}
2093 		brelse(bp);
2094 	}
2095 
2096 	error = ffs_vgetf(mp, ino, lflags, &nvp, ffs_flags);
2097 	if (error != 0)
2098 		return (error);
2099 
2100 	ip = VTOI(nvp);
2101 	if (ip->i_mode == 0 || ip->i_gen != gen || ip->i_effnlink <= 0) {
2102 		if (ip->i_mode == 0)
2103 			vgone(nvp);
2104 		vput(nvp);
2105 		return (ESTALE);
2106 	}
2107 
2108 	vnode_create_vobject(nvp, DIP(ip, i_size), curthread);
2109 	*vpp = nvp;
2110 	return (0);
2111 }
2112 
2113 /*
2114  * Initialize the filesystem.
2115  */
2116 static int
2117 ffs_init(vfsp)
2118 	struct vfsconf *vfsp;
2119 {
2120 
2121 	ffs_susp_initialize();
2122 	softdep_initialize();
2123 	return (ufs_init(vfsp));
2124 }
2125 
2126 /*
2127  * Undo the work of ffs_init().
2128  */
2129 static int
2130 ffs_uninit(vfsp)
2131 	struct vfsconf *vfsp;
2132 {
2133 	int ret;
2134 
2135 	ret = ufs_uninit(vfsp);
2136 	softdep_uninitialize();
2137 	ffs_susp_uninitialize();
2138 	taskqueue_drain_all(taskqueue_thread);
2139 	return (ret);
2140 }
2141 
2142 /*
2143  * Structure used to pass information from ffs_sbupdate to its
2144  * helper routine ffs_use_bwrite.
2145  */
2146 struct devfd {
2147 	struct ufsmount	*ump;
2148 	struct buf	*sbbp;
2149 	int		 waitfor;
2150 	int		 suspended;
2151 	int		 error;
2152 };
2153 
2154 /*
2155  * Write a superblock and associated information back to disk.
2156  */
2157 int
2158 ffs_sbupdate(ump, waitfor, suspended)
2159 	struct ufsmount *ump;
2160 	int waitfor;
2161 	int suspended;
2162 {
2163 	struct fs *fs;
2164 	struct buf *sbbp;
2165 	struct devfd devfd;
2166 
2167 	fs = ump->um_fs;
2168 	if (fs->fs_ronly == 1 &&
2169 	    (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
2170 	    (MNT_RDONLY | MNT_UPDATE))
2171 		panic("ffs_sbupdate: write read-only filesystem");
2172 	/*
2173 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
2174 	 */
2175 	sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
2176 	    (int)fs->fs_sbsize, 0, 0, 0);
2177 	/*
2178 	 * Initialize info needed for write function.
2179 	 */
2180 	devfd.ump = ump;
2181 	devfd.sbbp = sbbp;
2182 	devfd.waitfor = waitfor;
2183 	devfd.suspended = suspended;
2184 	devfd.error = 0;
2185 	return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite));
2186 }
2187 
2188 /*
2189  * Write function for use by filesystem-layer routines.
2190  */
2191 static int
2192 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size)
2193 {
2194 	struct devfd *devfdp;
2195 	struct ufsmount *ump;
2196 	struct buf *bp;
2197 	struct fs *fs;
2198 	int error;
2199 
2200 	devfdp = devfd;
2201 	ump = devfdp->ump;
2202 	fs = ump->um_fs;
2203 	/*
2204 	 * Writing the superblock summary information.
2205 	 */
2206 	if (loc != fs->fs_sblockloc) {
2207 		bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0);
2208 		bcopy(buf, bp->b_data, (u_int)size);
2209 		if (devfdp->suspended)
2210 			bp->b_flags |= B_VALIDSUSPWRT;
2211 		if (devfdp->waitfor != MNT_WAIT)
2212 			bawrite(bp);
2213 		else if ((error = bwrite(bp)) != 0)
2214 			devfdp->error = error;
2215 		return (0);
2216 	}
2217 	/*
2218 	 * Writing the superblock itself. We need to do special checks for it.
2219 	 */
2220 	bp = devfdp->sbbp;
2221 	if (ffs_fsfail_cleanup(ump, devfdp->error))
2222 		devfdp->error = 0;
2223 	if (devfdp->error != 0) {
2224 		brelse(bp);
2225 		return (devfdp->error);
2226 	}
2227 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
2228 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2229 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2230 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
2231 		fs->fs_sblockloc = SBLOCK_UFS1;
2232 	}
2233 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
2234 	    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2235 		printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2236 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
2237 		fs->fs_sblockloc = SBLOCK_UFS2;
2238 	}
2239 	if (MOUNTEDSOFTDEP(ump->um_mountp))
2240 		softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
2241 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
2242 	fs = (struct fs *)bp->b_data;
2243 	ffs_oldfscompat_write(fs, ump);
2244 	fs->fs_si = NULL;
2245 	/* Recalculate the superblock hash */
2246 	fs->fs_ckhash = ffs_calc_sbhash(fs);
2247 	if (devfdp->suspended)
2248 		bp->b_flags |= B_VALIDSUSPWRT;
2249 	if (devfdp->waitfor != MNT_WAIT)
2250 		bawrite(bp);
2251 	else if ((error = bwrite(bp)) != 0)
2252 		devfdp->error = error;
2253 	return (devfdp->error);
2254 }
2255 
2256 static int
2257 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
2258 	int attrnamespace, const char *attrname)
2259 {
2260 
2261 #ifdef UFS_EXTATTR
2262 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
2263 	    attrname));
2264 #else
2265 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
2266 	    attrname));
2267 #endif
2268 }
2269 
2270 static void
2271 ffs_ifree(struct ufsmount *ump, struct inode *ip)
2272 {
2273 
2274 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
2275 		uma_zfree(uma_ufs1, ip->i_din1);
2276 	else if (ip->i_din2 != NULL)
2277 		uma_zfree(uma_ufs2, ip->i_din2);
2278 	uma_zfree_smr(uma_inode, ip);
2279 }
2280 
2281 static int dobkgrdwrite = 1;
2282 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
2283     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2284 
2285 /*
2286  * Complete a background write started from bwrite.
2287  */
2288 static void
2289 ffs_backgroundwritedone(struct buf *bp)
2290 {
2291 	struct bufobj *bufobj;
2292 	struct buf *origbp;
2293 
2294 #ifdef SOFTUPDATES
2295 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0)
2296 		softdep_handle_error(bp);
2297 #endif
2298 
2299 	/*
2300 	 * Find the original buffer that we are writing.
2301 	 */
2302 	bufobj = bp->b_bufobj;
2303 	BO_LOCK(bufobj);
2304 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2305 		panic("backgroundwritedone: lost buffer");
2306 
2307 	/*
2308 	 * We should mark the cylinder group buffer origbp as
2309 	 * dirty, to not lose the failed write.
2310 	 */
2311 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2312 		origbp->b_vflags |= BV_BKGRDERR;
2313 	BO_UNLOCK(bufobj);
2314 	/*
2315 	 * Process dependencies then return any unfinished ones.
2316 	 */
2317 	if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2318 		buf_complete(bp);
2319 #ifdef SOFTUPDATES
2320 	if (!LIST_EMPTY(&bp->b_dep))
2321 		softdep_move_dependencies(bp, origbp);
2322 #endif
2323 	/*
2324 	 * This buffer is marked B_NOCACHE so when it is released
2325 	 * by biodone it will be tossed.  Clear B_IOSTARTED in case of error.
2326 	 */
2327 	bp->b_flags |= B_NOCACHE;
2328 	bp->b_flags &= ~(B_CACHE | B_IOSTARTED);
2329 	pbrelvp(bp);
2330 
2331 	/*
2332 	 * Prevent brelse() from trying to keep and re-dirtying bp on
2333 	 * errors. It causes b_bufobj dereference in
2334 	 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2335 	 * pbrelvp() above.
2336 	 */
2337 	if ((bp->b_ioflags & BIO_ERROR) != 0)
2338 		bp->b_flags |= B_INVAL;
2339 	bufdone(bp);
2340 	BO_LOCK(bufobj);
2341 	/*
2342 	 * Clear the BV_BKGRDINPROG flag in the original buffer
2343 	 * and awaken it if it is waiting for the write to complete.
2344 	 * If BV_BKGRDINPROG is not set in the original buffer it must
2345 	 * have been released and re-instantiated - which is not legal.
2346 	 */
2347 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2348 	    ("backgroundwritedone: lost buffer2"));
2349 	origbp->b_vflags &= ~BV_BKGRDINPROG;
2350 	if (origbp->b_vflags & BV_BKGRDWAIT) {
2351 		origbp->b_vflags &= ~BV_BKGRDWAIT;
2352 		wakeup(&origbp->b_xflags);
2353 	}
2354 	BO_UNLOCK(bufobj);
2355 }
2356 
2357 /*
2358  * Write, release buffer on completion.  (Done by iodone
2359  * if async).  Do not bother writing anything if the buffer
2360  * is invalid.
2361  *
2362  * Note that we set B_CACHE here, indicating that buffer is
2363  * fully valid and thus cacheable.  This is true even of NFS
2364  * now so we set it generally.  This could be set either here
2365  * or in biodone() since the I/O is synchronous.  We put it
2366  * here.
2367  */
2368 static int
2369 ffs_bufwrite(struct buf *bp)
2370 {
2371 	struct buf *newbp;
2372 	struct cg *cgp;
2373 
2374 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2375 	if (bp->b_flags & B_INVAL) {
2376 		brelse(bp);
2377 		return (0);
2378 	}
2379 
2380 	if (!BUF_ISLOCKED(bp))
2381 		panic("bufwrite: buffer is not busy???");
2382 	/*
2383 	 * If a background write is already in progress, delay
2384 	 * writing this block if it is asynchronous. Otherwise
2385 	 * wait for the background write to complete.
2386 	 */
2387 	BO_LOCK(bp->b_bufobj);
2388 	if (bp->b_vflags & BV_BKGRDINPROG) {
2389 		if (bp->b_flags & B_ASYNC) {
2390 			BO_UNLOCK(bp->b_bufobj);
2391 			bdwrite(bp);
2392 			return (0);
2393 		}
2394 		bp->b_vflags |= BV_BKGRDWAIT;
2395 		msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2396 		    "bwrbg", 0);
2397 		if (bp->b_vflags & BV_BKGRDINPROG)
2398 			panic("bufwrite: still writing");
2399 	}
2400 	bp->b_vflags &= ~BV_BKGRDERR;
2401 	BO_UNLOCK(bp->b_bufobj);
2402 
2403 	/*
2404 	 * If this buffer is marked for background writing and we
2405 	 * do not have to wait for it, make a copy and write the
2406 	 * copy so as to leave this buffer ready for further use.
2407 	 *
2408 	 * This optimization eats a lot of memory.  If we have a page
2409 	 * or buffer shortfall we can't do it.
2410 	 */
2411 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2412 	    (bp->b_flags & B_ASYNC) &&
2413 	    !vm_page_count_severe() &&
2414 	    !buf_dirty_count_severe()) {
2415 		KASSERT(bp->b_iodone == NULL,
2416 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2417 
2418 		/* get a new block */
2419 		newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2420 		if (newbp == NULL)
2421 			goto normal_write;
2422 
2423 		KASSERT(buf_mapped(bp), ("Unmapped cg"));
2424 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2425 		BO_LOCK(bp->b_bufobj);
2426 		bp->b_vflags |= BV_BKGRDINPROG;
2427 		BO_UNLOCK(bp->b_bufobj);
2428 		newbp->b_xflags |=
2429 		    (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER;
2430 		newbp->b_lblkno = bp->b_lblkno;
2431 		newbp->b_blkno = bp->b_blkno;
2432 		newbp->b_offset = bp->b_offset;
2433 		newbp->b_iodone = ffs_backgroundwritedone;
2434 		newbp->b_flags |= B_ASYNC;
2435 		newbp->b_flags &= ~B_INVAL;
2436 		pbgetvp(bp->b_vp, newbp);
2437 
2438 #ifdef SOFTUPDATES
2439 		/*
2440 		 * Move over the dependencies.  If there are rollbacks,
2441 		 * leave the parent buffer dirtied as it will need to
2442 		 * be written again.
2443 		 */
2444 		if (LIST_EMPTY(&bp->b_dep) ||
2445 		    softdep_move_dependencies(bp, newbp) == 0)
2446 			bundirty(bp);
2447 #else
2448 		bundirty(bp);
2449 #endif
2450 
2451 		/*
2452 		 * Initiate write on the copy, release the original.  The
2453 		 * BKGRDINPROG flag prevents it from going away until
2454 		 * the background write completes. We have to recalculate
2455 		 * its check hash in case the buffer gets freed and then
2456 		 * reconstituted from the buffer cache during a later read.
2457 		 */
2458 		if ((bp->b_xflags & BX_CYLGRP) != 0) {
2459 			cgp = (struct cg *)bp->b_data;
2460 			cgp->cg_ckhash = 0;
2461 			cgp->cg_ckhash =
2462 			    calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2463 		}
2464 		bqrelse(bp);
2465 		bp = newbp;
2466 	} else
2467 		/* Mark the buffer clean */
2468 		bundirty(bp);
2469 
2470 	/* Let the normal bufwrite do the rest for us */
2471 normal_write:
2472 	/*
2473 	 * If we are writing a cylinder group, update its time.
2474 	 */
2475 	if ((bp->b_xflags & BX_CYLGRP) != 0) {
2476 		cgp = (struct cg *)bp->b_data;
2477 		cgp->cg_old_time = cgp->cg_time = time_second;
2478 	}
2479 	return (bufwrite(bp));
2480 }
2481 
2482 static void
2483 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2484 {
2485 	struct vnode *vp;
2486 	struct buf *tbp;
2487 	int error, nocopy;
2488 
2489 	/*
2490 	 * This is the bufobj strategy for the private VCHR vnodes
2491 	 * used by FFS to access the underlying storage device.
2492 	 * We override the default bufobj strategy and thus bypass
2493 	 * VOP_STRATEGY() for these vnodes.
2494 	 */
2495 	vp = bo2vnode(bo);
2496 	KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR ||
2497 	    bp->b_vp->v_rdev == NULL ||
2498 	    bp->b_vp->v_rdev->si_mountpt == NULL ||
2499 	    VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL ||
2500 	    vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp,
2501 	    ("ffs_geom_strategy() with wrong vp"));
2502 	if (bp->b_iocmd == BIO_WRITE) {
2503 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2504 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2505 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2506 			panic("ffs_geom_strategy: bad I/O");
2507 		nocopy = bp->b_flags & B_NOCOPY;
2508 		bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2509 		if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2510 		    vp->v_rdev->si_snapdata != NULL) {
2511 			if ((bp->b_flags & B_CLUSTER) != 0) {
2512 				runningbufwakeup(bp);
2513 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2514 					      b_cluster.cluster_entry) {
2515 					error = ffs_copyonwrite(vp, tbp);
2516 					if (error != 0 &&
2517 					    error != EOPNOTSUPP) {
2518 						bp->b_error = error;
2519 						bp->b_ioflags |= BIO_ERROR;
2520 						bp->b_flags &= ~B_BARRIER;
2521 						bufdone(bp);
2522 						return;
2523 					}
2524 				}
2525 				bp->b_runningbufspace = bp->b_bufsize;
2526 				atomic_add_long(&runningbufspace,
2527 					       bp->b_runningbufspace);
2528 			} else {
2529 				error = ffs_copyonwrite(vp, bp);
2530 				if (error != 0 && error != EOPNOTSUPP) {
2531 					bp->b_error = error;
2532 					bp->b_ioflags |= BIO_ERROR;
2533 					bp->b_flags &= ~B_BARRIER;
2534 					bufdone(bp);
2535 					return;
2536 				}
2537 			}
2538 		}
2539 #ifdef SOFTUPDATES
2540 		if ((bp->b_flags & B_CLUSTER) != 0) {
2541 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2542 				      b_cluster.cluster_entry) {
2543 				if (!LIST_EMPTY(&tbp->b_dep))
2544 					buf_start(tbp);
2545 			}
2546 		} else {
2547 			if (!LIST_EMPTY(&bp->b_dep))
2548 				buf_start(bp);
2549 		}
2550 
2551 #endif
2552 		/*
2553 		 * Check for metadata that needs check-hashes and update them.
2554 		 */
2555 		switch (bp->b_xflags & BX_FSPRIV) {
2556 		case BX_CYLGRP:
2557 			((struct cg *)bp->b_data)->cg_ckhash = 0;
2558 			((struct cg *)bp->b_data)->cg_ckhash =
2559 			    calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2560 			break;
2561 
2562 		case BX_SUPERBLOCK:
2563 		case BX_INODE:
2564 		case BX_INDIR:
2565 		case BX_DIR:
2566 			printf("Check-hash write is unimplemented!!!\n");
2567 			break;
2568 
2569 		case 0:
2570 			break;
2571 
2572 		default:
2573 			printf("multiple buffer types 0x%b\n",
2574 			    (u_int)(bp->b_xflags & BX_FSPRIV),
2575 			    PRINT_UFS_BUF_XFLAGS);
2576 			break;
2577 		}
2578 	}
2579 	if (bp->b_iocmd != BIO_READ && ffs_enxio_enable)
2580 		bp->b_xflags |= BX_CVTENXIO;
2581 	g_vfs_strategy(bo, bp);
2582 }
2583 
2584 int
2585 ffs_own_mount(const struct mount *mp)
2586 {
2587 
2588 	if (mp->mnt_op == &ufs_vfsops)
2589 		return (1);
2590 	return (0);
2591 }
2592 
2593 #ifdef	DDB
2594 #ifdef SOFTUPDATES
2595 
2596 /* defined in ffs_softdep.c */
2597 extern void db_print_ffs(struct ufsmount *ump);
2598 
2599 DB_SHOW_COMMAND(ffs, db_show_ffs)
2600 {
2601 	struct mount *mp;
2602 	struct ufsmount *ump;
2603 
2604 	if (have_addr) {
2605 		ump = VFSTOUFS((struct mount *)addr);
2606 		db_print_ffs(ump);
2607 		return;
2608 	}
2609 
2610 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2611 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2612 			db_print_ffs(VFSTOUFS(mp));
2613 	}
2614 }
2615 
2616 #endif	/* SOFTUPDATES */
2617 #endif	/* DDB */
2618