xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision d710132b4b8ce7f7cccaaf660cb16aa16b4077a0)
1 /*	$NetBSD: lfs_vfsops.c,v 1.117 2003/05/18 12:59:06 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 /*-
39  * Copyright (c) 1989, 1991, 1993, 1994
40  *	The Regents of the University of California.  All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)lfs_vfsops.c	8.20 (Berkeley) 6/10/95
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.117 2003/05/18 12:59:06 yamt Exp $");
75 
76 #if defined(_KERNEL_OPT)
77 #include "opt_quota.h"
78 #endif
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/namei.h>
83 #include <sys/proc.h>
84 #include <sys/kernel.h>
85 #include <sys/vnode.h>
86 #include <sys/mount.h>
87 #include <sys/kthread.h>
88 #include <sys/buf.h>
89 #include <sys/device.h>
90 #include <sys/mbuf.h>
91 #include <sys/file.h>
92 #include <sys/disklabel.h>
93 #include <sys/ioctl.h>
94 #include <sys/errno.h>
95 #include <sys/malloc.h>
96 #include <sys/pool.h>
97 #include <sys/socket.h>
98 #include <uvm/uvm_extern.h>
99 #include <sys/sysctl.h>
100 #include <sys/conf.h>
101 
102 #include <miscfs/specfs/specdev.h>
103 
104 #include <ufs/ufs/quota.h>
105 #include <ufs/ufs/inode.h>
106 #include <ufs/ufs/ufsmount.h>
107 #include <ufs/ufs/ufs_extern.h>
108 
109 #include <uvm/uvm.h>
110 #include <uvm/uvm_stat.h>
111 #include <uvm/uvm_pager.h>
112 #include <uvm/uvm_pdaemon.h>
113 
114 #include <ufs/lfs/lfs.h>
115 #include <ufs/lfs/lfs_extern.h>
116 
117 #include <miscfs/genfs/genfs.h>
118 #include <miscfs/genfs/genfs_node.h>
119 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
120 static boolean_t lfs_issequential_hole(const struct ufsmount *,
121     daddr_t, daddr_t);
122 
123 static int lfs_mountfs(struct vnode *, struct mount *, struct proc *);
124 
125 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
126 extern const struct vnodeopv_desc lfs_specop_opv_desc;
127 extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
128 extern int lfs_subsys_pages;
129 extern int  locked_queue_count;
130 extern long locked_queue_bytes;
131 extern struct simplelock lfs_subsys_lock;
132 
133 pid_t lfs_writer_daemon = 0;
134 int lfs_do_flush = 0;
135 
136 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
137 	&lfs_vnodeop_opv_desc,
138 	&lfs_specop_opv_desc,
139 	&lfs_fifoop_opv_desc,
140 	NULL,
141 };
142 
143 struct vfsops lfs_vfsops = {
144 	MOUNT_LFS,
145 	lfs_mount,
146 	ufs_start,
147 	lfs_unmount,
148 	ufs_root,
149 	ufs_quotactl,
150 	lfs_statfs,
151 	lfs_sync,
152 	lfs_vget,
153 	lfs_fhtovp,
154 	lfs_vptofh,
155 	lfs_init,
156 	lfs_reinit,
157 	lfs_done,
158 	lfs_sysctl,
159 	lfs_mountroot,
160 	ufs_check_export,
161 	lfs_vnodeopv_descs,
162 };
163 
164 struct genfs_ops lfs_genfsops = {
165 	lfs_gop_size,
166 	ufs_gop_alloc,
167 	lfs_gop_write,
168 };
169 
170 struct pool lfs_inode_pool;
171 struct pool lfs_dinode_pool;
172 struct pool lfs_inoext_pool;
173 
174 /*
175  * The writer daemon.  UVM keeps track of how many dirty pages we are holding
176  * in lfs_subsys_pages; the daemon flushes the filesystem when this value
177  * crosses the (user-defined) threshhold LFS_MAX_PAGES.
178  */
179 static void
180 lfs_writerd(void *arg)
181 {
182 #ifdef LFS_PD
183 	struct mount *mp, *nmp;
184 	struct lfs *fs;
185 #endif
186 
187 	lfs_writer_daemon = curproc->p_pid;
188 
189 	for (;;) {
190 		tsleep(&lfs_writer_daemon, PVM, "lfswriter", 0);
191 
192 #ifdef LFS_PD
193 		/*
194 		 * Look through the list of LFSs to see if any of them
195 		 * have requested pageouts.
196 		 */
197 		simple_lock(&mountlist_slock);
198 		for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
199 		     mp = nmp) {
200 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
201 				nmp = mp->mnt_list.cqe_next;
202 				continue;
203 			}
204 			if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS,
205 				    MFSNAMELEN) == 0) {
206 				fs = ((struct ufsmount *)mp->mnt_data)->ufsmount_u.lfs;
207 				if (fs->lfs_pdflush ||
208 				    !TAILQ_EMPTY(&fs->lfs_pchainhd)) {
209 					fs->lfs_pdflush = 0;
210 					lfs_flush_fs(fs, 0);
211 				}
212 			}
213 
214 			simple_lock(&mountlist_slock);
215 			nmp = mp->mnt_list.cqe_next;
216 			vfs_unbusy(mp);
217 		}
218 		simple_unlock(&mountlist_slock);
219 #endif /* LFS_PD */
220 
221 		/*
222 		 * If global state wants a flush, flush everything.
223 		 */
224 		while (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
225 			locked_queue_bytes > LFS_MAX_BYTES ||
226 			lfs_subsys_pages > LFS_MAX_PAGES) {
227 
228 #ifdef DEBUG_LFS_FLUSH
229 			if (lfs_do_flush)
230 				printf("daemon: lfs_do_flush\n");
231 			if (locked_queue_count > LFS_MAX_BUFS)
232 				printf("daemon: lqc = %d, max %d\n",
233 					locked_queue_count, LFS_MAX_BUFS);
234 			if (locked_queue_bytes > LFS_MAX_BYTES)
235 				printf("daemon: lqb = %ld, max %d\n",
236 					locked_queue_bytes, LFS_MAX_BYTES);
237 			if (lfs_subsys_pages > LFS_MAX_PAGES)
238 				printf("daemon: lssp = %d, max %d\n",
239 					lfs_subsys_pages, LFS_MAX_PAGES);
240 #endif /* DEBUG_LFS_FLUSH */
241 			lfs_flush(NULL, SEGM_WRITERD);
242 			lfs_do_flush = 0;
243 		}
244 		wakeup(&lfs_subsys_pages);
245 	}
246 	/* NOTREACHED */
247 }
248 
249 /*
250  * Initialize the filesystem, most work done by ufs_init.
251  */
252 void
253 lfs_init()
254 {
255 	ufs_init();
256 
257 	/*
258 	 * XXX Same structure as FFS inodes?  Should we share a common pool?
259 	 */
260 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
261 		  "lfsinopl", &pool_allocator_nointr);
262 	pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
263 		  "lfsdinopl", &pool_allocator_nointr);
264 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
265 		  "lfsinoextpl", &pool_allocator_nointr);
266 #ifdef DEBUG
267 	memset(lfs_log, 0, sizeof(lfs_log));
268 #endif
269 	simple_lock_init(&lfs_subsys_lock);
270 }
271 
272 void
273 lfs_reinit()
274 {
275 	ufs_reinit();
276 }
277 
278 void
279 lfs_done()
280 {
281 	ufs_done();
282 	pool_destroy(&lfs_inode_pool);
283 	pool_destroy(&lfs_inoext_pool);
284 }
285 
286 /*
287  * Called by main() when ufs is going to be mounted as root.
288  */
289 int
290 lfs_mountroot()
291 {
292 	extern struct vnode *rootvp;
293 	struct mount *mp;
294 	struct proc *p = curproc;	/* XXX */
295 	int error;
296 
297 	if (root_device->dv_class != DV_DISK)
298 		return (ENODEV);
299 
300 	if (rootdev == NODEV)
301 		return (ENODEV);
302 	/*
303 	 * Get vnodes for swapdev and rootdev.
304 	 */
305 	if ((error = bdevvp(rootdev, &rootvp))) {
306 		printf("lfs_mountroot: can't setup bdevvp's");
307 		return (error);
308 	}
309 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
310 		vrele(rootvp);
311 		return (error);
312 	}
313 	if ((error = lfs_mountfs(rootvp, mp, p))) {
314 		mp->mnt_op->vfs_refcount--;
315 		vfs_unbusy(mp);
316 		free(mp, M_MOUNT);
317 		vrele(rootvp);
318 		return (error);
319 	}
320 	simple_lock(&mountlist_slock);
321 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
322 	simple_unlock(&mountlist_slock);
323 	(void)lfs_statfs(mp, &mp->mnt_stat, p);
324 	vfs_unbusy(mp);
325 	inittodr(VFSTOUFS(mp)->um_lfs->lfs_tstamp);
326 	return (0);
327 }
328 
329 /*
330  * VFS Operations.
331  *
332  * mount system call
333  */
334 int
335 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct proc *p)
336 {
337 	struct vnode *devvp;
338 	struct ufs_args args;
339 	struct ufsmount *ump = NULL;
340 	struct lfs *fs = NULL;				/* LFS */
341 	int error;
342 	mode_t accessmode;
343 
344 	if (mp->mnt_flag & MNT_GETARGS) {
345 		ump = VFSTOUFS(mp);
346 		if (ump == NULL)
347 			return EIO;
348 		args.fspec = NULL;
349 		vfs_showexport(mp, &args.export, &ump->um_export);
350 		return copyout(&args, data, sizeof(args));
351 	}
352 	error = copyin(data, &args, sizeof (struct ufs_args));
353 	if (error)
354 		return (error);
355 
356 	/*
357 	 * If updating, check whether changing from read-only to
358 	 * read/write; if there is no device name, that's all we do.
359 	 */
360 	if (mp->mnt_flag & MNT_UPDATE) {
361 		ump = VFSTOUFS(mp);
362 		fs = ump->um_lfs;
363 		if (fs->lfs_ronly && (mp->mnt_flag & MNT_WANTRDWR)) {
364 			/*
365 			 * If upgrade to read-write by non-root, then verify
366 			 * that user has necessary permissions on the device.
367 			 */
368 			if (p->p_ucred->cr_uid != 0) {
369 				vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
370 				error = VOP_ACCESS(ump->um_devvp, VREAD|VWRITE,
371 						   p->p_ucred, p);
372 				VOP_UNLOCK(ump->um_devvp, 0);
373 				if (error)
374 					return (error);
375 			}
376 			fs->lfs_ronly = 0;
377 		}
378 		if (args.fspec == 0) {
379 			/*
380 			 * Process export requests.
381 			 */
382 			return (vfs_export(mp, &ump->um_export, &args.export));
383 		}
384 	}
385 	/*
386 	 * Not an update, or updating the name: look up the name
387 	 * and verify that it refers to a sensible block device.
388 	 */
389 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p);
390 	if ((error = namei(ndp)) != 0)
391 		return (error);
392 	devvp = ndp->ni_vp;
393 	if (devvp->v_type != VBLK) {
394 		vrele(devvp);
395 		return (ENOTBLK);
396 	}
397 	if (bdevsw_lookup(devvp->v_rdev) == NULL) {
398 		vrele(devvp);
399 		return (ENXIO);
400 	}
401 	/*
402 	 * If mount by non-root, then verify that user has necessary
403 	 * permissions on the device.
404 	 */
405 	if (p->p_ucred->cr_uid != 0) {
406 		accessmode = VREAD;
407 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
408 			accessmode |= VWRITE;
409 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
410 		error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p);
411 		if (error) {
412 			vput(devvp);
413 			return (error);
414 		}
415 		VOP_UNLOCK(devvp, 0);
416 	}
417 	if ((mp->mnt_flag & MNT_UPDATE) == 0)
418 		error = lfs_mountfs(devvp, mp, p);		/* LFS */
419 	else {
420 		if (devvp != ump->um_devvp)
421 			error = EINVAL;	/* needs translation */
422 		else
423 			vrele(devvp);
424 	}
425 	if (error) {
426 		vrele(devvp);
427 		return (error);
428 	}
429 	ump = VFSTOUFS(mp);
430 	fs = ump->um_lfs;					/* LFS */
431 	return set_statfs_info(path, UIO_USERSPACE, args.fspec,
432 	    UIO_USERSPACE, mp, p);
433 }
434 
435 /*
436  * Roll-forward code.
437  */
438 
439 /*
440  * Load the appropriate indirect block, and change the appropriate pointer.
441  * Mark the block dirty.  Do segment and avail accounting.
442  */
443 static int
444 update_meta(struct lfs *fs, ino_t ino, int version, daddr_t lbn,
445 	    daddr_t ndaddr, size_t size, struct proc *p)
446 {
447 	int error;
448 	struct vnode *vp;
449 	struct inode *ip;
450 	daddr_t odaddr, ooff;
451 	struct indir a[NIADDR], *ap;
452 	struct buf *bp;
453 	SEGUSE *sup;
454 	int num;
455 
456 	if ((error = lfs_rf_valloc(fs, ino, version, p, &vp)) != 0) {
457 #ifdef DEBUG_LFS_RFW
458 		printf("update_meta: ino %d: lfs_rf_valloc returned %d\n", ino,
459 		       error);
460 #endif
461 		return error;
462 	}
463 
464 	if ((error = VOP_BALLOC(vp, (lbn << fs->lfs_bshift), size,
465 				NOCRED, 0, &bp)) != 0) {
466 		vput(vp);
467 		return (error);
468 	}
469 	/* No need to write, the block is already on disk */
470 	if (bp->b_flags & B_DELWRI) {
471 		LFS_UNLOCK_BUF(bp);
472 		fs->lfs_avail += btofsb(fs, bp->b_bcount);
473 	}
474 	bp->b_flags |= B_INVAL;
475 	brelse(bp);
476 
477 	/*
478 	 * Extend the file, if it is not large enough already.
479 	 * XXX this is not exactly right, we don't know how much of the
480 	 * XXX last block is actually used.  We hope that an inode will
481 	 * XXX appear later to give the correct size.
482 	 */
483 	ip = VTOI(vp);
484 	if (ip->i_size <= (lbn << fs->lfs_bshift)) {
485 		if (lbn < NDADDR)
486 			ip->i_size = ip->i_ffs1_size = (lbn << fs->lfs_bshift) +
487 				(size - fs->lfs_fsize) + 1;
488 		else
489 			ip->i_size = ip->i_ffs1_size =
490 			    (lbn << fs->lfs_bshift) + 1;
491 	}
492 
493 	error = ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
494 	if (error) {
495 #ifdef DEBUG_LFS_RFW
496 		printf("update_meta: ufs_bmaparray returned %d\n", error);
497 #endif
498 		vput(vp);
499 		return error;
500 	}
501 	switch (num) {
502 	    case 0:
503 		ooff = ip->i_ffs1_db[lbn];
504 		if (ooff == UNWRITTEN)
505 			ip->i_ffs1_blocks += btofsb(fs, size);
506 		/* XXX what about fragment extension? */
507 		ip->i_ffs1_db[lbn] = ndaddr;
508 		break;
509 	    case 1:
510 		ooff = ip->i_ffs1_ib[a[0].in_off];
511 		if (ooff == UNWRITTEN)
512 			ip->i_ffs1_blocks += btofsb(fs, size);
513 		ip->i_ffs1_ib[a[0].in_off] = ndaddr;
514 		break;
515 	    default:
516 		ap = &a[num - 1];
517 		if (bread(vp, ap->in_lbn, fs->lfs_bsize, NOCRED, &bp))
518 			panic("update_meta: bread bno %lld",
519 			    (long long)ap->in_lbn);
520 
521 		/* XXX ondisk32 */
522 		ooff = ((int32_t *)bp->b_data)[ap->in_off];
523 		if (ooff == UNWRITTEN)
524 			ip->i_ffs1_blocks += btofsb(fs, size);
525 		/* XXX ondisk32 */
526 		((int32_t *)bp->b_data)[ap->in_off] = ndaddr;
527 		(void) VOP_BWRITE(bp);
528 	}
529 	LFS_SET_UINO(ip, IN_CHANGE | IN_MODIFIED | IN_UPDATE);
530 
531 	/* Update segment usage information. */
532 	if (odaddr > 0) {
533 		LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, odaddr)), bp);
534 #ifdef DIAGNOSTIC
535 		if (sup->su_nbytes < size) {
536 			panic("update_meta: negative bytes "
537 			      "(segment %" PRIu32 " short by %ld)\n",
538 			      dtosn(fs, dbtofsb(fs, odaddr)), (long)size - sup->su_nbytes);
539 			sup->su_nbytes = size;
540 		}
541 #endif
542 		sup->su_nbytes -= size;
543 		LFS_WRITESEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, odaddr)), bp);
544 	}
545 	LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
546 	sup->su_nbytes += size;
547 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
548 
549 	/* Fix this so it can be released */
550 	/* ip->i_lfs_effnblks = ip->i_ffs1_blocks; */
551 
552 #ifdef DEBUG_LFS_RFW
553 	/* Now look again to make sure it worked */
554 	ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
555 	if (dbtofsb(fs, odaddr) != ndaddr)
556 		printf("update_meta: failed setting ino %d lbn %" PRId64
557 		    " to %" PRId64 "\n", ino, lbn, ndaddr);
558 #endif
559 	vput(vp);
560 	return 0;
561 }
562 
563 static int
564 update_inoblk(struct lfs *fs, daddr_t offset, struct ucred *cred,
565 	      struct proc *p)
566 {
567 	struct vnode *devvp, *vp;
568 	struct inode *ip;
569 	struct ufs1_dinode *dip;
570 	struct buf *dbp, *ibp;
571 	int error;
572 	daddr_t daddr;
573 	IFILE *ifp;
574 	SEGUSE *sup;
575 
576 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
577 
578 	/*
579 	 * Get the inode, update times and perms.
580 	 * DO NOT update disk blocks, we do that separately.
581 	 */
582 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp);
583 	if (error) {
584 #ifdef DEBUG_LFS_RFW
585 		printf("update_inoblk: bread returned %d\n", error);
586 #endif
587 		return error;
588 	}
589 	dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs);
590 	while (--dip >= (struct ufs1_dinode *)dbp->b_data) {
591 		if (dip->di_inumber > LFS_IFILE_INUM) {
592 			/* printf("ino %d version %d\n", dip->di_inumber,
593 			       dip->di_gen); */
594 			error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
595 					      p, &vp);
596 			if (error) {
597 #ifdef DEBUG_LFS_RFW
598 				printf("update_inoblk: lfs_rf_valloc returned %d\n", error);
599 #endif
600 				continue;
601 			}
602 			ip = VTOI(vp);
603 			if (dip->di_size != ip->i_size)
604 				VOP_TRUNCATE(vp, dip->di_size, 0, NOCRED, p);
605 			/* Get mode, link count, size, and times */
606 			memcpy(ip->i_din.ffs1_din, dip,
607 			       offsetof(struct ufs1_dinode, di_db[0]));
608 
609 			/* Then the rest, except di_blocks */
610 			ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
611 			ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
612 			ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
613 			ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
614 
615 			ip->i_mode = ip->i_ffs1_mode;
616 			ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink;
617 			ip->i_size = ip->i_ffs1_size;
618 
619 			LFS_SET_UINO(ip, IN_CHANGE | IN_MODIFIED | IN_UPDATE);
620 
621 			/* Re-initialize to get type right */
622 			ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
623 				  &vp);
624 			vput(vp);
625 
626 			/* Record change in location */
627 			LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
628 			daddr = ifp->if_daddr;
629 			ifp->if_daddr = dbtofsb(fs, dbp->b_blkno);
630 			error = LFS_BWRITE_LOG(ibp); /* Ifile */
631 			/* And do segment accounting */
632 			if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) {
633 				if (daddr > 0) {
634 					LFS_SEGENTRY(sup, fs, dtosn(fs, daddr),
635 						     ibp);
636 					sup->su_nbytes -= sizeof (struct ufs1_dinode);
637 					LFS_WRITESEGENTRY(sup, fs,
638 							  dtosn(fs, daddr),
639 							  ibp);
640 				}
641 				LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
642 					     ibp);
643 				sup->su_nbytes += sizeof (struct ufs1_dinode);
644 				LFS_WRITESEGENTRY(sup, fs,
645 						  dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
646 						  ibp);
647 			}
648 		}
649 	}
650 	dbp->b_flags |= B_AGE;
651 	brelse(dbp);
652 
653 	return 0;
654 }
655 
656 #define CHECK_CKSUM   0x0001  /* Check the checksum to make sure it's valid */
657 #define CHECK_UPDATE  0x0002  /* Update Ifile for new data blocks / inodes */
658 
659 static daddr_t
660 check_segsum(struct lfs *fs, daddr_t offset,
661 	     struct ucred *cred, int flags, int *pseg_flags, struct proc *p)
662 {
663 	struct vnode *devvp;
664 	struct buf *bp, *dbp;
665 	int error, nblocks, ninos, i, j;
666 	SEGSUM *ssp;
667 	u_long *dp, *datap; /* XXX u_int32_t */
668 	daddr_t oldoffset;
669 	int32_t *iaddr;	/* XXX ondisk32 */
670 	FINFO *fip;
671 	SEGUSE *sup;
672 	size_t size;
673 	u_int64_t serial;
674 
675 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
676 	/*
677 	 * If the segment has a superblock and we're at the top
678 	 * of the segment, skip the superblock.
679 	 */
680 	if (sntod(fs, dtosn(fs, offset)) == offset) {
681 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
682 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
683 			offset += btofsb(fs, LFS_SBPAD);
684 		brelse(bp);
685 	}
686 
687 	/* Read in the segment summary */
688 	error = bread(devvp, offset, fs->lfs_sumsize, cred, &bp);
689 	if (error)
690 		return -1;
691 
692 	/* Check summary checksum */
693 	ssp = (SEGSUM *)bp->b_data;
694 	if (flags & CHECK_CKSUM) {
695 		if (ssp->ss_sumsum != cksum(&ssp->ss_datasum,
696 					   fs->lfs_sumsize -
697 					   sizeof(ssp->ss_sumsum))) {
698 #ifdef DEBUG_LFS_RFW
699 			printf("Sumsum error at 0x%" PRIx64 "\n", offset);
700 #endif
701 			offset = -1;
702 			goto err1;
703 		}
704 		if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) {
705 #ifdef DEBUG_LFS_RFW
706 			printf("Empty pseg at 0x%" PRIx64 "\n", offset);
707 #endif
708 			offset = -1;
709 			goto err1;
710 		}
711 		if (ssp->ss_create < fs->lfs_tstamp) {
712 #ifdef DEBUG_LFS_RFW
713 			printf("Old data at 0x%" PRIx64 "\n", offset);
714 #endif
715 			offset = -1;
716 			goto err1;
717 		}
718 	}
719 	if (fs->lfs_version > 1) {
720 		serial = ssp->ss_serial;
721 		if (serial != fs->lfs_serial + 1) {
722 #ifdef DEBUG_LFS_RFW
723 			printf("Unexpected serial number at 0x%" PRIx64
724 			    "\n", offset);
725 #endif
726 			offset = -1;
727 			goto err1;
728 		}
729 		if (ssp->ss_ident != fs->lfs_ident) {
730 #ifdef DEBUG_LFS_RFW
731 			printf("Incorrect fsid (0x%x vs 0x%x) at 0x%"
732 			    PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset);
733 #endif
734 			offset = -1;
735 			goto err1;
736 		}
737 	}
738 	if (pseg_flags)
739 		*pseg_flags = ssp->ss_flags;
740 	oldoffset = offset;
741 	offset += btofsb(fs, fs->lfs_sumsize);
742 
743 	ninos = howmany(ssp->ss_ninos, INOPB(fs));
744 	/* XXX ondisk32 */
745 	iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t));
746 	if (flags & CHECK_CKSUM) {
747 		/* Count blocks */
748 		nblocks = 0;
749 		fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
750 		for (i = 0; i < ssp->ss_nfinfo; ++i) {
751 			nblocks += fip->fi_nblocks;
752 			if (fip->fi_nblocks <= 0)
753 				break;
754 			/* XXX ondisk32 */
755 			fip = (FINFO *)(((char *)fip) + FINFOSIZE +
756 					(fip->fi_nblocks * sizeof(int32_t)));
757 		}
758 		nblocks += ninos;
759 		/* Create the sum array */
760 		datap = dp = (u_long *)malloc(nblocks * sizeof(u_long),
761 					      M_SEGMENT, M_WAITOK);
762 	}
763 
764 	/* Handle individual blocks */
765 	fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
766 	for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) {
767 		/* Inode block? */
768 		if (ninos && *iaddr == offset) {
769 			if (flags & CHECK_CKSUM) {
770 				/* Read in the head and add to the buffer */
771 				error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize,
772 					      cred, &dbp);
773 				if (error) {
774 					offset = -1;
775 					goto err2;
776 				}
777 				(*dp++) = ((u_long *)(dbp->b_data))[0];
778 				dbp->b_flags |= B_AGE;
779 				brelse(dbp);
780 			}
781 			if (flags & CHECK_UPDATE) {
782 				if ((error = update_inoblk(fs, offset, cred, p))
783 				    != 0) {
784 					offset = -1;
785 					goto err2;
786 				}
787 			}
788 			offset += btofsb(fs, fs->lfs_ibsize);
789 			--iaddr;
790 			--ninos;
791 			--i; /* compensate */
792 			continue;
793 		}
794 		/* printf("check: blocks from ino %d version %d\n",
795 		       fip->fi_ino, fip->fi_version); */
796 		size = fs->lfs_bsize;
797 		for (j = 0; j < fip->fi_nblocks; ++j) {
798 			if (j == fip->fi_nblocks - 1)
799 				size = fip->fi_lastlength;
800 			if (flags & CHECK_CKSUM) {
801 				error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp);
802 				if (error) {
803 					offset = -1;
804 					goto err2;
805 				}
806 				(*dp++) = ((u_long *)(dbp->b_data))[0];
807 				dbp->b_flags |= B_AGE;
808 				brelse(dbp);
809 			}
810 			/* Account for and update any direct blocks */
811 			if ((flags & CHECK_UPDATE) &&
812 			   fip->fi_ino > LFS_IFILE_INUM &&
813 			   fip->fi_blocks[j] >= 0) {
814 				update_meta(fs, fip->fi_ino, fip->fi_version,
815 					    fip->fi_blocks[j], offset, size, p);
816 			}
817 			offset += btofsb(fs, size);
818 		}
819 		/* XXX ondisk32 */
820 		fip = (FINFO *)(((char *)fip) + FINFOSIZE
821 				+ fip->fi_nblocks * sizeof(int32_t));
822 	}
823 	/* Checksum the array, compare */
824 	if ((flags & CHECK_CKSUM) &&
825 	   ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long)))
826 	{
827 #ifdef DEBUG_LFS_RFW
828 		printf("Datasum error at 0x%" PRIx64 " (wanted %x got %x)\n",
829 		    offset, ssp->ss_datasum, cksum(datap, nblocks *
830 					      sizeof(u_long)));
831 #endif
832 		offset = -1;
833 		goto err2;
834 	}
835 
836 	/* If we're at the end of the segment, move to the next */
837 	if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) !=
838 	   dtosn(fs, offset)) {
839 		if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) {
840 			offset = -1;
841 			goto err2;
842 		}
843 		offset = ssp->ss_next;
844 #ifdef DEBUG_LFS_RFW
845 		printf("LFS roll forward: moving on to offset 0x%" PRIx64
846 		       " -> segment %d\n", offset, dtosn(fs,offset));
847 #endif
848 	}
849 
850 	if (flags & CHECK_UPDATE) {
851 		fs->lfs_avail -= (offset - oldoffset);
852 		/* Don't clog the buffer queue */
853 		if (locked_queue_count > LFS_MAX_BUFS ||
854 		    locked_queue_bytes > LFS_MAX_BYTES) {
855 			++fs->lfs_writer;
856 			lfs_flush(fs, SEGM_CKP);
857 			if (--fs->lfs_writer == 0)
858 				wakeup(&fs->lfs_dirops);
859 		}
860 	}
861 
862     err2:
863 	if (flags & CHECK_CKSUM)
864 		free(datap, M_SEGMENT);
865     err1:
866 	bp->b_flags |= B_AGE;
867 	brelse(bp);
868 
869 	/* XXX should we update the serial number even for bad psegs? */
870 	if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1)
871 		fs->lfs_serial = serial;
872 	return offset;
873 }
874 
875 /*
876  * Common code for mount and mountroot
877  * LFS specific
878  */
879 int
880 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct proc *p)
881 {
882 	extern struct vnode *rootvp;
883 	struct dlfs *tdfs, *dfs, *adfs;
884 	struct lfs *fs;
885 	struct ufsmount *ump;
886 	struct vnode *vp;
887 	struct buf *bp, *abp;
888 	struct partinfo dpart;
889 	dev_t dev;
890 	int error, i, ronly, secsize, fsbsize;
891 	struct ucred *cred;
892 	CLEANERINFO *cip;
893 	SEGUSE *sup;
894 	int flags, dirty, do_rollforward;
895 	daddr_t offset, oldoffset, lastgoodpseg, sb_addr;
896 	int sn, curseg;
897 
898 	cred = p ? p->p_ucred : NOCRED;
899 	/*
900 	 * Disallow multiple mounts of the same device.
901 	 * Disallow mounting of a device that is currently in use
902 	 * (except for root, which might share swap device for miniroot).
903 	 * Flush out any old buffers remaining from a previous use.
904 	 */
905 	if ((error = vfs_mountedon(devvp)) != 0)
906 		return (error);
907 	if (vcount(devvp) > 1 && devvp != rootvp)
908 		return (EBUSY);
909 	if ((error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0)) != 0)
910 		return (error);
911 
912 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
913 	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p);
914 	if (error)
915 		return (error);
916 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, p) != 0)
917 		secsize = DEV_BSIZE;
918 	else
919 		secsize = dpart.disklab->d_secsize;
920 
921 	/* Don't free random space on error. */
922 	bp = NULL;
923 	abp = NULL;
924 	ump = NULL;
925 
926 	sb_addr = LFS_LABELPAD / secsize;
927 	while (1) {
928 		/* Read in the superblock. */
929 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp);
930 		if (error)
931 			goto out;
932 		dfs = (struct dlfs *)bp->b_data;
933 
934 		/* Check the basics. */
935 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize >= MAXBSIZE ||
936 		    dfs->dlfs_version > LFS_VERSION ||
937 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
938 #ifdef DEBUG_LFS
939 			printf("lfs_mountfs: primary superblock sanity failed\n");
940 #endif
941 			error = EINVAL;		/* XXX needs translation */
942 			goto out;
943 		}
944 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT)
945 			printf("lfs_mountfs: warning: unknown inode format %d\n",
946 			       dfs->dlfs_inodefmt);
947 
948 		if (dfs->dlfs_version == 1)
949 			fsbsize = secsize;
950 		else {
951 			fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb +
952 				dfs->dlfs_fsbtodb);
953 			/*
954 			 * Could be, if the frag size is large enough, that we
955 			 * don't have the "real" primary superblock.  If that's
956 			 * the case, get the real one, and try again.
957 			 */
958 			if (sb_addr != dfs->dlfs_sboffs[0] <<
959 				       dfs->dlfs_fsbtodb) {
960 /* #ifdef DEBUG_LFS */
961 				printf("lfs_mountfs: sb daddr 0x%llx is not right, trying 0x%llx\n",
962 					(long long)sb_addr, (long long)(dfs->dlfs_sboffs[0] <<
963 						 dfs->dlfs_fsbtodb));
964 /* #endif */
965 				sb_addr = dfs->dlfs_sboffs[0] <<
966 					  dfs->dlfs_fsbtodb;
967 				brelse(bp);
968 				continue;
969 			}
970 		}
971 		break;
972 	}
973 
974 	/*
975 	 * Check the second superblock to see which is newer; then mount
976 	 * using the older of the two.	This is necessary to ensure that
977 	 * the filesystem is valid if it was not unmounted cleanly.
978 	 */
979 
980 	if (dfs->dlfs_sboffs[1] &&
981 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
982 	{
983 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize),
984 			LFS_SBPAD, cred, &abp);
985 		if (error)
986 			goto out;
987 		adfs = (struct dlfs *)abp->b_data;
988 
989 		if (dfs->dlfs_version == 1) {
990 			/* 1s resolution comparison */
991 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
992 				tdfs = adfs;
993 			else
994 				tdfs = dfs;
995 		} else {
996 			/* monotonic infinite-resolution comparison */
997 			if (adfs->dlfs_serial < dfs->dlfs_serial)
998 				tdfs = adfs;
999 			else
1000 				tdfs = dfs;
1001 		}
1002 
1003 		/* Check the basics. */
1004 		if (tdfs->dlfs_magic != LFS_MAGIC ||
1005 		    tdfs->dlfs_bsize > MAXBSIZE ||
1006 		    tdfs->dlfs_version > LFS_VERSION ||
1007 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
1008 #ifdef DEBUG_LFS
1009 			printf("lfs_mountfs: alt superblock sanity failed\n");
1010 #endif
1011 			error = EINVAL;		/* XXX needs translation */
1012 			goto out;
1013 		}
1014 	} else {
1015 #ifdef DEBUG_LFS
1016 		printf("lfs_mountfs: invalid alt superblock daddr=0x%x\n",
1017 			dfs->dlfs_sboffs[1]);
1018 #endif
1019 		error = EINVAL;
1020 		goto out;
1021 	}
1022 
1023 	/* Allocate the mount structure, copy the superblock into it. */
1024 	fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
1025 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
1026 
1027 	/* Compatibility */
1028 	if (fs->lfs_version < 2) {
1029 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
1030 		fs->lfs_ibsize = fs->lfs_bsize;
1031 		fs->lfs_start = fs->lfs_sboffs[0];
1032 		fs->lfs_tstamp = fs->lfs_otstamp;
1033 		fs->lfs_fsbtodb = 0;
1034 	}
1035 
1036 	/* Before rolling forward, lock so vget will sleep for other procs */
1037 	fs->lfs_flags = LFS_NOTYET;
1038 	fs->lfs_rfpid = p->p_pid;
1039 
1040 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1041 	ump->um_lfs = fs;
1042 	ump->um_fstype = UFS1;
1043 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
1044 		bp->b_flags |= B_INVAL;
1045 		abp->b_flags |= B_INVAL;
1046 	}
1047 	brelse(bp);
1048 	bp = NULL;
1049 	brelse(abp);
1050 	abp = NULL;
1051 
1052 	/* Set up the I/O information */
1053 	fs->lfs_devbsize = secsize;
1054 	fs->lfs_iocount = 0;
1055 	fs->lfs_diropwait = 0;
1056 	fs->lfs_activesb = 0;
1057 	fs->lfs_uinodes = 0;
1058 	fs->lfs_ravail = 0;
1059 	fs->lfs_sbactive = 0;
1060 
1061 	/* Set up the ifile and lock aflags */
1062 	fs->lfs_doifile = 0;
1063 	fs->lfs_writer = 0;
1064 	fs->lfs_dirops = 0;
1065 	fs->lfs_nadirop = 0;
1066 	fs->lfs_seglock = 0;
1067 	fs->lfs_pdflush = 0;
1068 	fs->lfs_sleepers = 0;
1069 	simple_lock_init(&fs->lfs_interlock);
1070 	lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0);
1071 
1072 	/* Set the file system readonly/modify bits. */
1073 	fs->lfs_ronly = ronly;
1074 	if (ronly == 0)
1075 		fs->lfs_fmod = 1;
1076 
1077 	/* Initialize the mount structure. */
1078 	dev = devvp->v_rdev;
1079 	mp->mnt_data = ump;
1080 	mp->mnt_stat.f_fsid.val[0] = (long)dev;
1081 	mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_LFS);
1082 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
1083 	mp->mnt_maxsymlinklen = fs->lfs_maxsymlinklen;
1084 	mp->mnt_flag |= MNT_LOCAL;
1085 	mp->mnt_fs_bshift = fs->lfs_bshift;
1086 	ump->um_flags = 0;
1087 	ump->um_mountp = mp;
1088 	ump->um_dev = dev;
1089 	ump->um_devvp = devvp;
1090 	ump->um_bptrtodb = fs->lfs_fsbtodb;
1091 	ump->um_seqinc = fragstofsb(fs, fs->lfs_frag);
1092 	ump->um_nindir = fs->lfs_nindir;
1093 	ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
1094 	for (i = 0; i < MAXQUOTAS; i++)
1095 		ump->um_quotas[i] = NULLVP;
1096 	devvp->v_specmountpoint = mp;
1097 
1098 	/* Set up reserved memory for pageout */
1099 	lfs_setup_resblks(fs);
1100 	/* Set up vdirop tailq */
1101 	TAILQ_INIT(&fs->lfs_dchainhd);
1102 	/* and paging tailq */
1103 	TAILQ_INIT(&fs->lfs_pchainhd);
1104 
1105 	/*
1106 	 * We use the ifile vnode for almost every operation.  Instead of
1107 	 * retrieving it from the hash table each time we retrieve it here,
1108 	 * artificially increment the reference count and keep a pointer
1109 	 * to it in the incore copy of the superblock.
1110 	 */
1111 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1112 #ifdef DEBUG
1113 		printf("lfs_mountfs: ifile vget failed, error=%d\n", error);
1114 #endif
1115 		goto out;
1116 	}
1117 	fs->lfs_ivnode = vp;
1118 	VREF(vp);
1119 
1120 	/* Set up segment usage flags for the autocleaner. */
1121 	fs->lfs_nactive = 0;
1122 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
1123 						M_SEGMENT, M_WAITOK);
1124 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1125 						 M_SEGMENT, M_WAITOK);
1126 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1127 						 M_SEGMENT, M_WAITOK);
1128 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
1129 	for (i = 0; i < fs->lfs_nseg; i++) {
1130 		int changed;
1131 
1132 		LFS_SEGENTRY(sup, fs, i, bp);
1133 		changed = 0;
1134 		if (!ronly) {
1135 			if (sup->su_nbytes == 0 &&
1136 			    !(sup->su_flags & SEGUSE_EMPTY)) {
1137 				sup->su_flags |= SEGUSE_EMPTY;
1138 				++changed;
1139 			} else if (!(sup->su_nbytes == 0) &&
1140 				   (sup->su_flags & SEGUSE_EMPTY)) {
1141 				sup->su_flags &= ~SEGUSE_EMPTY;
1142 				++changed;
1143 			}
1144 			if (sup->su_flags & SEGUSE_ACTIVE) {
1145 				sup->su_flags &= ~SEGUSE_ACTIVE;
1146 				++changed;
1147 			}
1148 		}
1149 		fs->lfs_suflags[0][i] = sup->su_flags;
1150 		if (changed)
1151 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1152 		else
1153 			brelse(bp);
1154 	}
1155 
1156 	/*
1157 	 * Roll forward.
1158 	 *
1159 	 * We don't automatically roll forward for v1 filesystems, because
1160 	 * of the danger that the clock was turned back between the last
1161 	 * checkpoint and crash.  This would roll forward garbage.
1162 	 *
1163 	 * v2 filesystems don't have this problem because they use a
1164 	 * monotonically increasing serial number instead of a timestamp.
1165 	 */
1166 #ifdef LFS_DO_ROLLFORWARD
1167 	do_rollforward = !fs->lfs_ronly;
1168 #else
1169 	do_rollforward = (fs->lfs_version > 1 && !fs->lfs_ronly &&
1170 			  !(fs->lfs_pflags & LFS_PF_CLEAN));
1171 #endif
1172 	if (do_rollforward) {
1173 		/*
1174 		 * Phase I: Find the address of the last good partial
1175 		 * segment that was written after the checkpoint.  Mark
1176 		 * the segments in question dirty, so they won't be
1177 		 * reallocated.
1178 		 */
1179 		lastgoodpseg = oldoffset = offset = fs->lfs_offset;
1180 		flags = 0x0;
1181 #ifdef DEBUG_LFS_RFW
1182 		printf("LFS roll forward phase 1: starting at offset 0x%"
1183 		    PRIx64 "\n", offset);
1184 #endif
1185 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
1186 		if (!(sup->su_flags & SEGUSE_DIRTY))
1187 			--fs->lfs_nclean;
1188 		sup->su_flags |= SEGUSE_DIRTY;
1189 		LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp);
1190 		while ((offset = check_segsum(fs, offset, cred, CHECK_CKSUM,
1191 					      &flags, p)) > 0)
1192 		{
1193 			if (sntod(fs, oldoffset) != sntod(fs, offset)) {
1194 				LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset),
1195 					     bp);
1196 				if (!(sup->su_flags & SEGUSE_DIRTY))
1197 					--fs->lfs_nclean;
1198 				sup->su_flags |= SEGUSE_DIRTY;
1199 				LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset),
1200 					     bp);
1201 			}
1202 
1203 #ifdef DEBUG_LFS_RFW
1204 			printf("LFS roll forward phase 1: offset=0x%"
1205 			    PRIx64 "\n", offset);
1206 			if (flags & SS_DIROP) {
1207 				printf("lfs_mountfs: dirops at 0x%" PRIx64 "\n",
1208 				       oldoffset);
1209 				if (!(flags & SS_CONT))
1210 					printf("lfs_mountfs: dirops end "
1211 					       "at 0x%" PRIx64 "\n", oldoffset);
1212 			}
1213 #endif
1214 			if (!(flags & SS_CONT))
1215 				lastgoodpseg = offset;
1216 			oldoffset = offset;
1217 		}
1218 #ifdef DEBUG_LFS_RFW
1219 		if (flags & SS_CONT) {
1220 			printf("LFS roll forward: warning: incomplete "
1221 			       "dirops discarded\n");
1222 		}
1223 		printf("LFS roll forward phase 1: completed: "
1224 		       "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg);
1225 #endif
1226 		oldoffset = fs->lfs_offset;
1227 		if (fs->lfs_offset != lastgoodpseg) {
1228 			/* Don't overwrite what we're trying to preserve */
1229 			offset = fs->lfs_offset;
1230 			fs->lfs_offset = lastgoodpseg;
1231 			fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset));
1232 			for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) {
1233 				sn = (sn + 1) % fs->lfs_nseg;
1234 				if (sn == curseg)
1235 					panic("lfs_mountfs: no clean segments");
1236 				LFS_SEGENTRY(sup, fs, sn, bp);
1237 				dirty = (sup->su_flags & SEGUSE_DIRTY);
1238 				brelse(bp);
1239 				if (!dirty)
1240 					break;
1241 			}
1242 			fs->lfs_nextseg = sntod(fs, sn);
1243 
1244 			/*
1245 			 * Phase II: Roll forward from the first superblock.
1246 			 */
1247 			while (offset != lastgoodpseg) {
1248 #ifdef DEBUG_LFS_RFW
1249 				printf("LFS roll forward phase 2: 0x%"
1250 				    PRIx64 "\n", offset);
1251 #endif
1252 				offset = check_segsum(fs, offset, cred,
1253 						      CHECK_UPDATE, NULL, p);
1254 			}
1255 
1256 			/*
1257 			 * Finish: flush our changes to disk.
1258 			 */
1259 			lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1260 			printf("lfs_mountfs: roll forward recovered %lld blocks\n",
1261 			       (long long)(lastgoodpseg - oldoffset));
1262 		}
1263 #ifdef DEBUG_LFS_RFW
1264 		printf("LFS roll forward complete\n");
1265 #endif
1266 	}
1267 	/* If writing, sb is not clean; record in case of immediate crash */
1268 	if (!fs->lfs_ronly) {
1269 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
1270 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
1271 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
1272 	}
1273 
1274 	/* Allow vget now that roll-forward is complete */
1275 	fs->lfs_flags &= ~(LFS_NOTYET);
1276 	wakeup(&fs->lfs_flags);
1277 
1278 	/*
1279 	 * Initialize the ifile cleaner info with information from
1280 	 * the superblock.
1281 	 */
1282 	LFS_CLEANERINFO(cip, fs, bp);
1283 	cip->clean = fs->lfs_nclean;
1284 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1285 	cip->avail = fs->lfs_avail;
1286 	cip->bfree = fs->lfs_bfree;
1287 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
1288 
1289 	/*
1290 	 * Mark the current segment as ACTIVE, since we're going to
1291 	 * be writing to it.
1292 	 */
1293 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
1294 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1295 	fs->lfs_nactive++;
1296 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
1297 
1298 	/* Now that roll-forward is done, unlock the Ifile */
1299 	vput(vp);
1300 
1301 	/* Comment on ifile size if it is too large */
1302 	if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS) {
1303 		fs->lfs_flags |= LFS_WARNED;
1304 		printf("lfs_mountfs: please consider increasing NBUF to at least %lld\n",
1305 			(long long)(fs->lfs_ivnode->v_size / fs->lfs_bsize) * (nbuf / LFS_MAX_BUFS));
1306 	}
1307 	if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES) {
1308 		fs->lfs_flags |= LFS_WARNED;
1309 		printf("lfs_mountfs: please consider increasing BUFPAGES to at least %lld\n",
1310 			(long long)fs->lfs_ivnode->v_size * bufpages / LFS_MAX_BYTES);
1311 	}
1312 
1313 	return (0);
1314 out:
1315 	if (bp)
1316 		brelse(bp);
1317 	if (abp)
1318 		brelse(abp);
1319 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1320 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p);
1321 	VOP_UNLOCK(devvp, 0);
1322 	if (ump) {
1323 		free(ump->um_lfs, M_UFSMNT);
1324 		free(ump, M_UFSMNT);
1325 		mp->mnt_data = NULL;
1326 	}
1327 
1328 	/* Start the pagedaemon-anticipating daemon */
1329 	if (lfs_writer_daemon == 0 &&
1330 	    kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1331 		panic("fork lfs_writer");
1332 
1333 	return (error);
1334 }
1335 
1336 /*
1337  * unmount system call
1338  */
1339 int
1340 lfs_unmount(struct mount *mp, int mntflags, struct proc *p)
1341 {
1342 	struct ufsmount *ump;
1343 	struct lfs *fs;
1344 	int error, flags, ronly;
1345 	int s;
1346 
1347 	flags = 0;
1348 	if (mntflags & MNT_FORCE)
1349 		flags |= FORCECLOSE;
1350 
1351 	ump = VFSTOUFS(mp);
1352 	fs = ump->um_lfs;
1353 
1354 	/* wake up the cleaner so it can die */
1355 	wakeup(&fs->lfs_nextseg);
1356 	wakeup(&lfs_allclean_wakeup);
1357 	simple_lock(&fs->lfs_interlock);
1358 	while (fs->lfs_sleepers)
1359 		ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1360 			&fs->lfs_interlock);
1361 	simple_unlock(&fs->lfs_interlock);
1362 
1363 #ifdef QUOTA
1364 	if (mp->mnt_flag & MNT_QUOTA) {
1365 		int i;
1366 		error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags);
1367 		if (error)
1368 			return (error);
1369 		for (i = 0; i < MAXQUOTAS; i++) {
1370 			if (ump->um_quotas[i] == NULLVP)
1371 				continue;
1372 			quotaoff(p, mp, i);
1373 		}
1374 		/*
1375 		 * Here we fall through to vflush again to ensure
1376 		 * that we have gotten rid of all the system vnodes.
1377 		 */
1378 	}
1379 #endif
1380 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1381 		return (error);
1382 	if ((error = VFS_SYNC(mp, 1, p->p_ucred, p)) != 0)
1383 		return (error);
1384 	s = splbio();
1385 	if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd))
1386 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1387 	splx(s);
1388 
1389 	/* Comment on ifile size if it has become too large */
1390 	if (!(fs->lfs_flags & LFS_WARNED)) {
1391 		if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS)
1392 			printf("lfs_unmount: please consider increasing"
1393 				" NBUF to at least %lld\n",
1394 				(long long)(fs->lfs_ivnode->v_size /
1395 					    fs->lfs_bsize) *
1396 				(long long)(nbuf / LFS_MAX_BUFS));
1397 		if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES)
1398 			printf("lfs_unmount: please consider increasing"
1399 				" BUFPAGES to at least %lld\n",
1400 				(long long)fs->lfs_ivnode->v_size *
1401 				bufpages / LFS_MAX_BYTES);
1402 	}
1403 
1404 	/* Explicitly write the superblock, to update serial and pflags */
1405 	fs->lfs_pflags |= LFS_PF_CLEAN;
1406 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
1407 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
1408 	while (fs->lfs_iocount)
1409 		tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0);
1410 
1411 	/* Finish with the Ifile, now that we're done with it */
1412 	vrele(fs->lfs_ivnode);
1413 	vgone(fs->lfs_ivnode);
1414 
1415 	ronly = !fs->lfs_ronly;
1416 	if (ump->um_devvp->v_type != VBAD)
1417 		ump->um_devvp->v_specmountpoint = NULL;
1418 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1419 	error = VOP_CLOSE(ump->um_devvp,
1420 	    ronly ? FREAD : FREAD|FWRITE, NOCRED, p);
1421 	vput(ump->um_devvp);
1422 
1423 	/* Free per-mount data structures */
1424 	free(fs->lfs_suflags[0], M_SEGMENT);
1425 	free(fs->lfs_suflags[1], M_SEGMENT);
1426 	free(fs->lfs_suflags, M_SEGMENT);
1427 	lfs_free_resblks(fs);
1428 	free(fs, M_UFSMNT);
1429 	free(ump, M_UFSMNT);
1430 
1431 	mp->mnt_data = NULL;
1432 	mp->mnt_flag &= ~MNT_LOCAL;
1433 	return (error);
1434 }
1435 
1436 /*
1437  * Get file system statistics.
1438  */
1439 int
1440 lfs_statfs(struct mount *mp, struct statfs *sbp, struct proc *p)
1441 {
1442 	struct lfs *fs;
1443 	struct ufsmount *ump;
1444 
1445 	ump = VFSTOUFS(mp);
1446 	fs = ump->um_lfs;
1447 	if (fs->lfs_magic != LFS_MAGIC)
1448 		panic("lfs_statfs: magic");
1449 
1450 	sbp->f_type = 0;
1451 	sbp->f_bsize = fs->lfs_fsize;
1452 	sbp->f_iosize = fs->lfs_bsize;
1453 	sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs));
1454 	sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs));
1455 	sbp->f_bavail = fsbtofrags(fs, (long)LFS_EST_BFREE(fs) -
1456 				  (long)LFS_EST_RSVD(fs));
1457 
1458 	sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1459 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1460 	copy_statfs_info(sbp, mp);
1461 	return (0);
1462 }
1463 
1464 /*
1465  * Go through the disk queues to initiate sandbagged IO;
1466  * go through the inodes to write those that have been modified;
1467  * initiate the writing of the super block if it has been modified.
1468  *
1469  * Note: we are always called with the filesystem marked `MPBUSY'.
1470  */
1471 int
1472 lfs_sync(struct mount *mp, int waitfor, struct ucred *cred, struct proc *p)
1473 {
1474 	int error;
1475 	struct lfs *fs;
1476 
1477 	fs = ((struct ufsmount *)mp->mnt_data)->ufsmount_u.lfs;
1478 	if (fs->lfs_ronly)
1479 		return 0;
1480 	while (fs->lfs_dirops)
1481 		error = tsleep(&fs->lfs_writer, PRIBIO + 1, "lfs_dirops", 0);
1482 	fs->lfs_writer++;
1483 
1484 	/* All syncs must be checkpoints until roll-forward is implemented. */
1485 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1486 	if (--fs->lfs_writer == 0)
1487 		wakeup(&fs->lfs_dirops);
1488 #ifdef QUOTA
1489 	qsync(mp);
1490 #endif
1491 	return (error);
1492 }
1493 
1494 extern struct lock ufs_hashlock;
1495 
1496 /*
1497  * Look up an LFS dinode number to find its incore vnode.  If not already
1498  * in core, read it in from the specified device.  Return the inode locked.
1499  * Detection and handling of mount points must be done by the calling routine.
1500  */
1501 int
1502 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1503 {
1504 	struct lfs *fs;
1505 	struct ufs1_dinode *dip;
1506 	struct inode *ip;
1507 	struct buf *bp;
1508 	struct ifile *ifp;
1509 	struct vnode *vp;
1510 	struct ufsmount *ump;
1511 	daddr_t daddr;
1512 	dev_t dev;
1513 	int error, retries;
1514 	struct timespec ts;
1515 
1516 	ump = VFSTOUFS(mp);
1517 	dev = ump->um_dev;
1518 	fs = ump->um_lfs;
1519 
1520 	/*
1521 	 * If the filesystem is not completely mounted yet, suspend
1522 	 * any access requests (wait for roll-forward to complete).
1523 	 */
1524 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1525 		tsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0);
1526 
1527 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1528 		return (0);
1529 
1530 	if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) {
1531 		*vpp = NULL;
1532 		 return (error);
1533 	}
1534 
1535 	do {
1536 		if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) {
1537 			ungetnewvnode(vp);
1538 			return (0);
1539 		}
1540 	} while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0));
1541 
1542 	/* Translate the inode number to a disk address. */
1543 	if (ino == LFS_IFILE_INUM)
1544 		daddr = fs->lfs_idaddr;
1545 	else {
1546 		/* XXX bounds-check this too */
1547 		LFS_IENTRY(ifp, fs, ino, bp);
1548 		daddr = ifp->if_daddr;
1549 		if (fs->lfs_version > 1) {
1550 			ts.tv_sec = ifp->if_atime_sec;
1551 			ts.tv_nsec = ifp->if_atime_nsec;
1552 		}
1553 
1554 		brelse(bp);
1555 		if (daddr == LFS_UNUSED_DADDR) {
1556 			*vpp = NULLVP;
1557 			ungetnewvnode(vp);
1558 			lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1559 			return (ENOENT);
1560 		}
1561 	}
1562 
1563 	/* Allocate/init new vnode/inode. */
1564 	lfs_vcreate(mp, ino, vp);
1565 
1566 	/*
1567 	 * Put it onto its hash chain and lock it so that other requests for
1568 	 * this inode will block if they arrive while we are sleeping waiting
1569 	 * for old data structures to be purged or for the contents of the
1570 	 * disk portion of this inode to be read.
1571 	 */
1572 	ip = VTOI(vp);
1573 	ufs_ihashins(ip);
1574 	lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1575 
1576 	/*
1577 	 * XXX
1578 	 * This may not need to be here, logically it should go down with
1579 	 * the i_devvp initialization.
1580 	 * Ask Kirk.
1581 	 */
1582 	ip->i_lfs = ump->um_lfs;
1583 
1584 	/* Read in the disk contents for the inode, copy into the inode. */
1585 	retries = 0;
1586     again:
1587 	error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1588 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1589 		NOCRED, &bp);
1590 	if (error) {
1591 		/*
1592 		 * The inode does not contain anything useful, so it would
1593 		 * be misleading to leave it on its hash chain. With mode
1594 		 * still zero, it will be unlinked and returned to the free
1595 		 * list by vput().
1596 		 */
1597 		vput(vp);
1598 		brelse(bp);
1599 		*vpp = NULL;
1600 		return (error);
1601 	}
1602 
1603 	dip = lfs_ifind(fs, ino, bp);
1604 	if (dip == NULL) {
1605 		/* Assume write has not completed yet; try again */
1606 		bp->b_flags |= B_INVAL;
1607 		brelse(bp);
1608 		++retries;
1609 		if (retries > LFS_IFIND_RETRIES) {
1610 #ifdef DEBUG
1611 			/* If the seglock is held look at the bpp to see
1612 			   what is there anyway */
1613 			if (fs->lfs_seglock > 0) {
1614 				struct buf **bpp;
1615 				struct ufs1_dinode *dp;
1616 				int i;
1617 
1618 				for (bpp = fs->lfs_sp->bpp;
1619 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1620 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1621 					    bpp != fs->lfs_sp->bpp) {
1622 						/* Inode block */
1623 						printf("block 0x%" PRIx64 ": ",
1624 						    (*bpp)->b_blkno);
1625 						dp = (struct ufs1_dinode *)(*bpp)->b_data;
1626 						for (i = 0; i < INOPB(fs); i++)
1627 							if (dp[i].di_u.inumber)
1628 								printf("%d ", dp[i].di_u.inumber);
1629 						printf("\n");
1630 					}
1631 				}
1632 			}
1633 #endif
1634 			panic("lfs_vget: dinode not found");
1635 		}
1636 		printf("lfs_vget: dinode %d not found, retrying...\n", ino);
1637 		(void)tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs ifind", 1);
1638 		goto again;
1639 	}
1640 	*ip->i_din.ffs1_din = *dip;
1641 	brelse(bp);
1642 
1643 	if (fs->lfs_version > 1) {
1644 		ip->i_ffs1_atime = ts.tv_sec;
1645 		ip->i_ffs1_atimensec = ts.tv_nsec;
1646 	}
1647 
1648 	lfs_vinit(mp, vp);
1649 
1650 	*vpp = vp;
1651 
1652 	KASSERT(VOP_ISLOCKED(vp));
1653 
1654 	return (0);
1655 }
1656 
1657 /*
1658  * File handle to vnode
1659  */
1660 int
1661 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1662 {
1663 	struct lfid *lfhp;
1664 	struct buf *bp;
1665 	IFILE *ifp;
1666 	int32_t daddr;
1667 	struct lfs *fs;
1668 
1669 	lfhp = (struct lfid *)fhp;
1670 	if (lfhp->lfid_ino < LFS_IFILE_INUM)
1671 		return ESTALE;
1672 
1673 	fs = VFSTOUFS(mp)->um_lfs;
1674 	if (lfhp->lfid_ident != fs->lfs_ident)
1675 		return ESTALE;
1676 
1677 	if (lfhp->lfid_ino >
1678 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1679 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1680 		return ESTALE;
1681 
1682 	if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfhp->lfid_ino) == NULLVP) {
1683 		LFS_IENTRY(ifp, fs, lfhp->lfid_ino, bp);
1684 		daddr = ifp->if_daddr;
1685 		brelse(bp);
1686 		if (daddr == LFS_UNUSED_DADDR)
1687 			return ESTALE;
1688 	}
1689 
1690 	return (ufs_fhtovp(mp, &lfhp->lfid_ufid, vpp));
1691 }
1692 
1693 /*
1694  * Vnode pointer to File handle
1695  */
1696 /* ARGSUSED */
1697 int
1698 lfs_vptofh(struct vnode *vp, struct fid *fhp)
1699 {
1700 	struct inode *ip;
1701 	struct lfid *lfhp;
1702 
1703 	ip = VTOI(vp);
1704 	lfhp = (struct lfid *)fhp;
1705 	lfhp->lfid_len = sizeof(struct lfid);
1706 	lfhp->lfid_ino = ip->i_number;
1707 	lfhp->lfid_gen = ip->i_gen;
1708 	lfhp->lfid_ident = ip->i_lfs->lfs_ident;
1709 	return (0);
1710 }
1711 
1712 int
1713 lfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp, size_t newlen, struct proc *p)
1714 {
1715 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead;
1716 	extern struct lfs_stats lfs_stats;
1717 	int error;
1718 
1719 	/* all sysctl names at this level are terminal */
1720 	if (namelen != 1)
1721 		return (ENOTDIR);
1722 
1723 	switch (name[0]) {
1724 	case LFS_WRITEINDIR:
1725 		return (sysctl_int(oldp, oldlenp, newp, newlen,
1726 				   &lfs_writeindir));
1727 	case LFS_CLEAN_VNHEAD:
1728 		return (sysctl_int(oldp, oldlenp, newp, newlen,
1729 				   &lfs_clean_vnhead));
1730 	case LFS_DOSTATS:
1731 		if ((error = sysctl_int(oldp, oldlenp, newp, newlen,
1732 				       &lfs_dostats)))
1733 			return error;
1734 		if (lfs_dostats == 0)
1735 			memset(&lfs_stats,0,sizeof(lfs_stats));
1736 		return 0;
1737 	default:
1738 		return (EOPNOTSUPP);
1739 	}
1740 	/* NOTREACHED */
1741 }
1742 
1743 static boolean_t
1744 lfs_issequential_hole(const struct ufsmount *ump,
1745     daddr_t daddr0, daddr_t daddr1)
1746 {
1747 
1748 	/* NOTE: all we want to know here is 'hole or not'. */
1749 
1750 	/*
1751 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1752 	 */
1753 	if (daddr0 != 0 && daddr1 != 0)
1754 		return TRUE;
1755 
1756 	/*
1757 	 * both are in hole?
1758 	 */
1759 	if (daddr0 == 0 && daddr1 == 0)
1760 		return TRUE; /* all holes are 'contiguous' for us. */
1761 
1762 	return FALSE;
1763 }
1764 
1765 /*
1766  * lfs_gop_write functions exactly like genfs_gop_write, except that
1767  * (1) it requires the seglock to be held by its caller, and sp->fip
1768  *     to be properly initialized (it will return without re-initializing
1769  *     sp->fip, and without calling lfs_writeseg).
1770  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1771  *     to determine how large a block it can write at once (though it does
1772  *     still use VOP_BMAP to find holes in the file);
1773  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1774  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1775  *     now have clusters of clusters, ick.)
1776  */
1777 static int
1778 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
1779 {
1780 	int i, s, error, run;
1781 	int fs_bshift, dev_bshift;
1782 	vaddr_t kva;
1783 	off_t eof, offset, startoffset;
1784 	size_t bytes, iobytes, skipbytes;
1785 	daddr_t lbn, blkno;
1786 	struct vm_page *pg;
1787 	struct buf *mbp, *bp;
1788 	struct vnode *devvp = VTOI(vp)->i_devvp;
1789 	struct inode *ip = VTOI(vp);
1790 	struct lfs *fs = ip->i_lfs;
1791 	struct segment *sp = fs->lfs_sp;
1792 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1793 
1794 	/* The Ifile lives in the buffer cache */
1795 	if (vp == fs->lfs_ivnode)
1796 		return genfs_compat_gop_write(vp, pgs, npages, flags);
1797 
1798 	/*
1799 	 * Sometimes things slip past the filters in lfs_putpages,
1800 	 * and the pagedaemon tries to write pages---problem is
1801 	 * that the pagedaemon never acquires the segment lock.
1802 	 *
1803 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1804 	 * queue under the hypothesis that they couldn't have got here
1805 	 * unless they were modified *quite* recently.
1806 	 *
1807 	 * XXXUBC that last statement is an oversimplification of course.
1808 	 */
1809 	if (!(fs->lfs_seglock) || fs->lfs_lockpid != curproc->p_pid) {
1810 		simple_lock(&vp->v_interlock);
1811 #ifdef DEBUG
1812 		printf("lfs_gop_write: seglock not held\n");
1813 #endif
1814 		uvm_lock_pageq();
1815 		for (i = 0; i < npages; i++) {
1816 			pg = pgs[i];
1817 
1818 			if (pg->flags & PG_PAGEOUT)
1819 				uvmexp.paging--;
1820 			if (pg->flags & PG_DELWRI) {
1821 				uvm_pageunwire(pg);
1822 			}
1823 			uvm_pageactivate(pg);
1824 			pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
1825 #ifdef DEBUG_LFS
1826 			printf("pg[%d]->flags = %x\n", i, pg->flags);
1827 			printf("pg[%d]->pqflags = %x\n", i, pg->pqflags);
1828 			printf("pg[%d]->uanon = %p\n", i, pg->uanon);
1829 			printf("pg[%d]->uobject = %p\n", i, pg->uobject);
1830 			printf("pg[%d]->wire_count = %d\n", i, pg->wire_count);
1831 			printf("pg[%d]->loan_count = %d\n", i, pg->loan_count);
1832 #endif
1833 		}
1834 		/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
1835 		uvm_page_unbusy(pgs, npages);
1836 		uvm_unlock_pageq();
1837 		simple_unlock(&vp->v_interlock);
1838 		return EAGAIN;
1839 	}
1840 
1841 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1842 	    vp, pgs, npages, flags);
1843 
1844 	GOP_SIZE(vp, vp->v_size, &eof, GOP_SIZE_WRITE);
1845 
1846 	if (vp->v_type == VREG) {
1847 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1848 		dev_bshift = vp->v_mount->mnt_dev_bshift;
1849 	} else {
1850 		fs_bshift = DEV_BSHIFT;
1851 		dev_bshift = DEV_BSHIFT;
1852 	}
1853 	error = 0;
1854 	pg = pgs[0];
1855 	startoffset = pg->offset;
1856 	bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1857 	skipbytes = 0;
1858 
1859 	/* KASSERT(bytes != 0); */
1860 	if (bytes == 0)
1861 		printf("ino %d bytes == 0 offset %" PRId64 "\n",
1862 			VTOI(vp)->i_number, pgs[0]->offset);
1863 
1864 	/* Swap PG_DELWRI for PG_PAGEOUT */
1865 	for (i = 0; i < npages; i++)
1866 		if (pgs[i]->flags & PG_DELWRI) {
1867 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1868 			pgs[i]->flags &= ~PG_DELWRI;
1869 			pgs[i]->flags |= PG_PAGEOUT;
1870 			uvmexp.paging++;
1871 			uvm_lock_pageq();
1872 			uvm_pageunwire(pgs[i]);
1873 			uvm_unlock_pageq();
1874 		}
1875 
1876 	/*
1877 	 * Check to make sure we're starting on a block boundary.
1878 	 * We'll check later to make sure we always write entire
1879 	 * blocks (or fragments).
1880 	 */
1881 	if (startoffset & fs->lfs_bmask)
1882 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1883 			startoffset, fs->lfs_bmask,
1884 			startoffset & fs->lfs_bmask);
1885 	KASSERT((startoffset & fs->lfs_bmask) == 0);
1886 	if (bytes & fs->lfs_ffmask) {
1887 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1888 		panic("lfs_gop_write: non-integer blocks");
1889 	}
1890 
1891 	kva = uvm_pagermapin(pgs, npages,
1892 	    UVMPAGER_MAPIN_WRITE | UVMPAGER_MAPIN_WAITOK);
1893 
1894 	s = splbio();
1895 	simple_lock(&global_v_numoutput_slock);
1896 	vp->v_numoutput += 2; /* one for biodone, one for aiodone */
1897 	simple_unlock(&global_v_numoutput_slock);
1898 	mbp = pool_get(&bufpool, PR_WAITOK);
1899 	splx(s);
1900 
1901 	memset(mbp, 0, sizeof(*bp));
1902 	BUF_INIT(mbp);
1903 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1904 	    vp, mbp, vp->v_numoutput, bytes);
1905 	mbp->b_bufsize = npages << PAGE_SHIFT;
1906 	mbp->b_data = (void *)kva;
1907 	mbp->b_resid = mbp->b_bcount = bytes;
1908 	mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL;
1909 	mbp->b_iodone = uvm_aio_biodone;
1910 	mbp->b_vp = vp;
1911 
1912 	bp = NULL;
1913 	for (offset = startoffset;
1914 	    bytes > 0;
1915 	    offset += iobytes, bytes -= iobytes) {
1916 		lbn = offset >> fs_bshift;
1917 		error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1918 		    lfs_issequential_hole);
1919 		if (error) {
1920 			UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
1921 			    error,0,0,0);
1922 			skipbytes += bytes;
1923 			bytes = 0;
1924 			break;
1925 		}
1926 
1927 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1928 		    bytes);
1929 		if (blkno == (daddr_t)-1) {
1930 			skipbytes += iobytes;
1931 			continue;
1932 		}
1933 
1934 		/*
1935 		 * Discover how much we can really pack into this buffer.
1936 		 */
1937 		/* If no room in the current segment, finish it up */
1938 		if (sp->sum_bytes_left < sizeof(int32_t) ||
1939 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1940 			int version;
1941 
1942 			lfs_updatemeta(sp);
1943 
1944 			version = sp->fip->fi_version;
1945 			(void) lfs_writeseg(fs, sp);
1946 
1947 			sp->fip->fi_version = version;
1948 			sp->fip->fi_ino = ip->i_number;
1949 			/* Add the current file to the segment summary. */
1950 			++((SEGSUM *)(sp->segsum))->ss_nfinfo;
1951 			sp->sum_bytes_left -= FINFOSIZE;
1952 		}
1953 		/* Check both for space in segment and space in segsum */
1954 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1955 					<< fs_bshift);
1956 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1957 				       << fs_bshift);
1958 		KASSERT(iobytes > 0);
1959 
1960 		/* if it's really one i/o, don't make a second buf */
1961 		if (offset == startoffset && iobytes == bytes) {
1962 			bp = mbp;
1963 			/* printf("bp is mbp\n"); */
1964 			/* correct overcount if there is no second buffer */
1965 			s = splbio();
1966 			simple_lock(&global_v_numoutput_slock);
1967 			--vp->v_numoutput;
1968 			simple_unlock(&global_v_numoutput_slock);
1969 			splx(s);
1970 		} else {
1971 			/* printf("bp is not mbp\n"); */
1972 			s = splbio();
1973 			bp = pool_get(&bufpool, PR_WAITOK);
1974 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1975 			    vp, bp, vp->v_numoutput, 0);
1976 			splx(s);
1977 			memset(bp, 0, sizeof(*bp));
1978 			BUF_INIT(bp);
1979 			bp->b_data = (char *)kva +
1980 			    (vaddr_t)(offset - pg->offset);
1981 			bp->b_resid = bp->b_bcount = iobytes;
1982 			bp->b_flags = B_BUSY|B_WRITE|B_CALL;
1983 			bp->b_iodone = uvm_aio_biodone1;
1984 		}
1985 
1986 		/* XXX This is silly ... is this necessary? */
1987 		bp->b_vp = NULL;
1988 		s = splbio();
1989 		bgetvp(vp, bp);
1990 		splx(s);
1991 
1992 		bp->b_lblkno = lblkno(fs, offset);
1993 		bp->b_private = mbp;
1994 		if (devvp->v_type == VBLK) {
1995 			bp->b_dev = devvp->v_rdev;
1996 		}
1997 		VOP_BWRITE(bp);
1998 		while (lfs_gatherblock(sp, bp, NULL))
1999 			continue;
2000 	}
2001 
2002 	if (skipbytes) {
2003 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
2004 		s = splbio();
2005 		if (error) {
2006 			mbp->b_flags |= B_ERROR;
2007 			mbp->b_error = error;
2008 		}
2009 		mbp->b_resid -= skipbytes;
2010 		if (mbp->b_resid == 0) {
2011 			biodone(mbp);
2012 		}
2013 		splx(s);
2014 	}
2015 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
2016 	return (0);
2017 }
2018 
2019 /*
2020  * finish vnode/inode initialization.
2021  * used by lfs_vget and lfs_fastvget.
2022  */
2023 void
2024 lfs_vinit(struct mount *mp, struct vnode *vp)
2025 {
2026 	struct inode *ip = VTOI(vp);
2027 	struct ufsmount *ump = VFSTOUFS(mp);
2028 	int i;
2029 
2030 	ip->i_mode = ip->i_ffs1_mode;
2031 	ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink;
2032 	ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
2033 	ip->i_flags = ip->i_ffs1_flags;
2034 	ip->i_gen = ip->i_ffs1_gen;
2035 	ip->i_uid = ip->i_ffs1_uid;
2036 	ip->i_gid = ip->i_ffs1_gid;
2037 
2038 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
2039 
2040 	/*
2041 	 * Initialize the vnode from the inode, check for aliases.  In all
2042 	 * cases re-init ip, the underlying vnode/inode may have changed.
2043 	 */
2044 	ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2045 
2046 	memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
2047 	if (vp->v_type != VLNK ||
2048 	    VTOI(vp)->i_size >= vp->v_mount->mnt_maxsymlinklen) {
2049 		struct lfs *fs = ump->um_lfs;
2050 #ifdef DEBUG
2051 		for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
2052 		    i < NDADDR; i++) {
2053 			if (ip->i_ffs1_db[i] != 0) {
2054 inconsistent:
2055 				lfs_dump_dinode(ip->i_din.ffs1_din);
2056 				panic("inconsistent inode");
2057 			}
2058 		}
2059 		for ( ; i < NDADDR + NIADDR; i++) {
2060 			if (ip->i_ffs1_ib[i - NDADDR] != 0) {
2061 				goto inconsistent;
2062 			}
2063 		}
2064 #endif /* DEBUG */
2065 		for (i = 0; i < NDADDR; i++)
2066 			if (ip->i_ffs1_db[i] != 0)
2067 				ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
2068 	}
2069 
2070 #ifdef DEBUG
2071 	if (vp->v_type == VNON) {
2072 		printf("lfs_vinit: ino %d is type VNON! (ifmt=%o)\n",
2073 		       ip->i_number, (ip->i_mode & IFMT) >> 12);
2074 		lfs_dump_dinode(ip->i_din.ffs1_din);
2075 #ifdef DDB
2076 		Debugger();
2077 #endif /* DDB */
2078 	}
2079 #endif /* DEBUG */
2080 
2081 	/*
2082 	 * Finish inode initialization now that aliasing has been resolved.
2083 	 */
2084 
2085 	ip->i_devvp = ump->um_devvp;
2086 	VREF(ip->i_devvp);
2087 	genfs_node_init(vp, &lfs_genfsops);
2088 	uvm_vnp_setsize(vp, ip->i_size);
2089 }
2090