xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: lfs_vfsops.c,v 1.143 2004/01/28 20:57:15 he 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. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)lfs_vfsops.c	8.20 (Berkeley) 6/10/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.143 2004/01/28 20:57:15 he Exp $");
71 
72 #if defined(_KERNEL_OPT)
73 #include "opt_quota.h"
74 #endif
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/namei.h>
79 #include <sys/proc.h>
80 #include <sys/kernel.h>
81 #include <sys/vnode.h>
82 #include <sys/mount.h>
83 #include <sys/kthread.h>
84 #include <sys/buf.h>
85 #include <sys/device.h>
86 #include <sys/mbuf.h>
87 #include <sys/file.h>
88 #include <sys/disklabel.h>
89 #include <sys/ioctl.h>
90 #include <sys/errno.h>
91 #include <sys/malloc.h>
92 #include <sys/pool.h>
93 #include <sys/socket.h>
94 #include <uvm/uvm_extern.h>
95 #include <sys/sysctl.h>
96 #include <sys/conf.h>
97 
98 #include <miscfs/specfs/specdev.h>
99 
100 #include <ufs/ufs/quota.h>
101 #include <ufs/ufs/inode.h>
102 #include <ufs/ufs/ufsmount.h>
103 #include <ufs/ufs/ufs_extern.h>
104 
105 #include <uvm/uvm.h>
106 #include <uvm/uvm_stat.h>
107 #include <uvm/uvm_pager.h>
108 #include <uvm/uvm_pdaemon.h>
109 
110 #include <ufs/lfs/lfs.h>
111 #include <ufs/lfs/lfs_extern.h>
112 
113 #include <miscfs/genfs/genfs.h>
114 #include <miscfs/genfs/genfs_node.h>
115 
116 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
117 static boolean_t lfs_issequential_hole(const struct ufsmount *,
118     daddr_t, daddr_t);
119 
120 static int lfs_mountfs(struct vnode *, struct mount *, struct proc *);
121 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t,
122     struct ucred *, int, int *, struct proc *);
123 
124 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
125 extern const struct vnodeopv_desc lfs_specop_opv_desc;
126 extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
127 
128 pid_t lfs_writer_daemon = 0;
129 int lfs_do_flush = 0;
130 
131 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
132 	&lfs_vnodeop_opv_desc,
133 	&lfs_specop_opv_desc,
134 	&lfs_fifoop_opv_desc,
135 	NULL,
136 };
137 
138 struct vfsops lfs_vfsops = {
139 	MOUNT_LFS,
140 	lfs_mount,
141 	ufs_start,
142 	lfs_unmount,
143 	ufs_root,
144 	ufs_quotactl,
145 	lfs_statfs,
146 	lfs_sync,
147 	lfs_vget,
148 	lfs_fhtovp,
149 	lfs_vptofh,
150 	lfs_init,
151 	lfs_reinit,
152 	lfs_done,
153 	NULL,
154 	lfs_mountroot,
155 	ufs_check_export,
156 	lfs_vnodeopv_descs,
157 };
158 
159 struct genfs_ops lfs_genfsops = {
160 	lfs_gop_size,
161 	ufs_gop_alloc,
162 	lfs_gop_write,
163 };
164 
165 struct pool lfs_inode_pool;
166 struct pool lfs_dinode_pool;
167 struct pool lfs_inoext_pool;
168 
169 /*
170  * The writer daemon.  UVM keeps track of how many dirty pages we are holding
171  * in lfs_subsys_pages; the daemon flushes the filesystem when this value
172  * crosses the (user-defined) threshhold LFS_MAX_PAGES.
173  */
174 static void
175 lfs_writerd(void *arg)
176 {
177 #ifdef LFS_PD
178 	struct mount *mp, *nmp;
179 	struct lfs *fs;
180 #endif
181 
182 	lfs_writer_daemon = curproc->p_pid;
183 
184 	simple_lock(&lfs_subsys_lock);
185 	for (;;) {
186 		ltsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", 0,
187 		    &lfs_subsys_lock);
188 
189 #ifdef LFS_PD
190 		/*
191 		 * Look through the list of LFSs to see if any of them
192 		 * have requested pageouts.
193 		 */
194 		simple_lock(&mountlist_slock);
195 		for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
196 		     mp = nmp) {
197 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
198 				nmp = CIRCLEQ_NEXT(mp, mnt_list);
199 				continue;
200 			}
201 			if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS,
202 				    MFSNAMELEN) == 0) {
203 				fs = VFSTOUFS(mp)->um_lfs;
204 				if (fs->lfs_pdflush ||
205 				    !TAILQ_EMPTY(&fs->lfs_pchainhd)) {
206 					fs->lfs_pdflush = 0;
207 					lfs_flush_fs(fs, 0);
208 				}
209 			}
210 
211 			simple_lock(&mountlist_slock);
212 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
213 			vfs_unbusy(mp);
214 		}
215 		simple_unlock(&mountlist_slock);
216 #endif /* LFS_PD */
217 
218 		/*
219 		 * If global state wants a flush, flush everything.
220 		 */
221 		simple_lock(&lfs_subsys_lock);
222 		while (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
223 			locked_queue_bytes > LFS_MAX_BYTES ||
224 			lfs_subsys_pages > LFS_MAX_PAGES) {
225 
226 #ifdef DEBUG_LFS_FLUSH
227 			if (lfs_do_flush)
228 				printf("daemon: lfs_do_flush\n");
229 			if (locked_queue_count > LFS_MAX_BUFS)
230 				printf("daemon: lqc = %d, max %d\n",
231 					locked_queue_count, LFS_MAX_BUFS);
232 			if (locked_queue_bytes > LFS_MAX_BYTES)
233 				printf("daemon: lqb = %ld, max %ld\n",
234 					locked_queue_bytes, LFS_MAX_BYTES);
235 			if (lfs_subsys_pages > LFS_MAX_PAGES)
236 				printf("daemon: lssp = %d, max %d\n",
237 					lfs_subsys_pages, LFS_MAX_PAGES);
238 #endif /* DEBUG_LFS_FLUSH */
239 			lfs_flush(NULL, SEGM_WRITERD);
240 			lfs_do_flush = 0;
241 		}
242 	}
243 	/* NOTREACHED */
244 }
245 
246 /*
247  * Initialize the filesystem, most work done by ufs_init.
248  */
249 void
250 lfs_init()
251 {
252 	ufs_init();
253 
254 	/*
255 	 * XXX Same structure as FFS inodes?  Should we share a common pool?
256 	 */
257 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
258 		  "lfsinopl", &pool_allocator_nointr);
259 	pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
260 		  "lfsdinopl", &pool_allocator_nointr);
261 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
262 		  "lfsinoextpl", &pool_allocator_nointr);
263 #ifdef DEBUG
264 	memset(lfs_log, 0, sizeof(lfs_log));
265 #endif
266 	simple_lock_init(&lfs_subsys_lock);
267 }
268 
269 void
270 lfs_reinit()
271 {
272 	ufs_reinit();
273 }
274 
275 void
276 lfs_done()
277 {
278 	ufs_done();
279 	pool_destroy(&lfs_inode_pool);
280 	pool_destroy(&lfs_inoext_pool);
281 }
282 
283 /*
284  * Called by main() when ufs is going to be mounted as root.
285  */
286 int
287 lfs_mountroot()
288 {
289 	extern struct vnode *rootvp;
290 	struct mount *mp;
291 	struct proc *p = curproc;	/* XXX */
292 	int error;
293 
294 	if (root_device->dv_class != DV_DISK)
295 		return (ENODEV);
296 
297 	if (rootdev == NODEV)
298 		return (ENODEV);
299 	/*
300 	 * Get vnodes for swapdev and rootdev.
301 	 */
302 	if ((error = bdevvp(rootdev, &rootvp))) {
303 		printf("lfs_mountroot: can't setup bdevvp's");
304 		return (error);
305 	}
306 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
307 		vrele(rootvp);
308 		return (error);
309 	}
310 	if ((error = lfs_mountfs(rootvp, mp, p))) {
311 		mp->mnt_op->vfs_refcount--;
312 		vfs_unbusy(mp);
313 		free(mp, M_MOUNT);
314 		vrele(rootvp);
315 		return (error);
316 	}
317 	simple_lock(&mountlist_slock);
318 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
319 	simple_unlock(&mountlist_slock);
320 	(void)lfs_statfs(mp, &mp->mnt_stat, p);
321 	vfs_unbusy(mp);
322 	inittodr(VFSTOUFS(mp)->um_lfs->lfs_tstamp);
323 	return (0);
324 }
325 
326 /*
327  * VFS Operations.
328  *
329  * mount system call
330  */
331 int
332 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct proc *p)
333 {
334 	struct vnode *devvp;
335 	struct ufs_args args;
336 	struct ufsmount *ump = NULL;
337 	struct lfs *fs = NULL;				/* LFS */
338 	int error;
339 	mode_t accessmode;
340 
341 	if (mp->mnt_flag & MNT_GETARGS) {
342 		ump = VFSTOUFS(mp);
343 		if (ump == NULL)
344 			return EIO;
345 		args.fspec = NULL;
346 		vfs_showexport(mp, &args.export, &ump->um_export);
347 		return copyout(&args, data, sizeof(args));
348 	}
349 	error = copyin(data, &args, sizeof (struct ufs_args));
350 	if (error)
351 		return (error);
352 
353 	/*
354 	 * If updating, check whether changing from read-only to
355 	 * read/write; if there is no device name, that's all we do.
356 	 */
357 	if (mp->mnt_flag & MNT_UPDATE) {
358 		ump = VFSTOUFS(mp);
359 		fs = ump->um_lfs;
360 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
361 			/*
362 			 * If upgrade to read-write by non-root, then verify
363 			 * that user has necessary permissions on the device.
364 			 */
365 			if (p->p_ucred->cr_uid != 0) {
366 				vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
367 				error = VOP_ACCESS(ump->um_devvp, VREAD|VWRITE,
368 						   p->p_ucred, p);
369 				VOP_UNLOCK(ump->um_devvp, 0);
370 				if (error)
371 					return (error);
372 			}
373 			fs->lfs_ronly = 0;
374 		}
375 		if (args.fspec == 0) {
376 			/*
377 			 * Process export requests.
378 			 */
379 			return (vfs_export(mp, &ump->um_export, &args.export));
380 		}
381 	}
382 	/*
383 	 * Not an update, or updating the name: look up the name
384 	 * and verify that it refers to a sensible block device.
385 	 */
386 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p);
387 	if ((error = namei(ndp)) != 0)
388 		return (error);
389 	devvp = ndp->ni_vp;
390 	if (devvp->v_type != VBLK) {
391 		vrele(devvp);
392 		return (ENOTBLK);
393 	}
394 	if (bdevsw_lookup(devvp->v_rdev) == NULL) {
395 		vrele(devvp);
396 		return (ENXIO);
397 	}
398 	/*
399 	 * If mount by non-root, then verify that user has necessary
400 	 * permissions on the device.
401 	 */
402 	if (p->p_ucred->cr_uid != 0) {
403 		accessmode = VREAD;
404 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
405 			accessmode |= VWRITE;
406 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
407 		error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p);
408 		if (error) {
409 			vput(devvp);
410 			return (error);
411 		}
412 		VOP_UNLOCK(devvp, 0);
413 	}
414 	if ((mp->mnt_flag & MNT_UPDATE) == 0)
415 		error = lfs_mountfs(devvp, mp, p);		/* LFS */
416 	else {
417 		if (devvp != ump->um_devvp)
418 			error = EINVAL;	/* needs translation */
419 		else
420 			vrele(devvp);
421 	}
422 	if (error) {
423 		vrele(devvp);
424 		return (error);
425 	}
426 	ump = VFSTOUFS(mp);
427 	fs = ump->um_lfs;					/* LFS */
428 	return set_statfs_info(path, UIO_USERSPACE, args.fspec,
429 	    UIO_USERSPACE, mp, p);
430 }
431 
432 /*
433  * Roll-forward code.
434  */
435 
436 /*
437  * Load the appropriate indirect block, and change the appropriate pointer.
438  * Mark the block dirty.  Do segment and avail accounting.
439  */
440 static int
441 update_meta(struct lfs *fs, ino_t ino, int version, daddr_t lbn,
442 	    daddr_t ndaddr, size_t size, struct proc *p)
443 {
444 	int error;
445 	struct vnode *vp;
446 	struct inode *ip;
447 #ifdef DEBUG_LFS_RFW
448 	daddr_t odaddr;
449 	struct indir a[NIADDR];
450 	int num;
451 	int i;
452 #endif /* DEBUG_LFS_RFW */
453 	struct buf *bp;
454 	SEGUSE *sup;
455 
456 	KASSERT(lbn >= 0);	/* no indirect blocks */
457 
458 	if ((error = lfs_rf_valloc(fs, ino, version, p, &vp)) != 0) {
459 #ifdef DEBUG_LFS_RFW
460 		printf("update_meta: ino %d: lfs_rf_valloc returned %d\n", ino,
461 		       error);
462 #endif /* DEBUG_LFS_RFW */
463 		return error;
464 	}
465 
466 	if ((error = VOP_BALLOC(vp, (lbn << fs->lfs_bshift), size,
467 				NOCRED, 0, &bp)) != 0) {
468 		vput(vp);
469 		return (error);
470 	}
471 	/* No need to write, the block is already on disk */
472 	if (bp->b_flags & B_DELWRI) {
473 		LFS_UNLOCK_BUF(bp);
474 		fs->lfs_avail += btofsb(fs, bp->b_bcount);
475 	}
476 	bp->b_flags |= B_INVAL;
477 	brelse(bp);
478 
479 	/*
480 	 * Extend the file, if it is not large enough already.
481 	 * XXX this is not exactly right, we don't know how much of the
482 	 * XXX last block is actually used.  We hope that an inode will
483 	 * XXX appear later to give the correct size.
484 	 */
485 	ip = VTOI(vp);
486 	if (ip->i_size <= (lbn << fs->lfs_bshift)) {
487 		u_int64_t newsize;
488 
489 		if (lbn < NDADDR)
490 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) +
491 				(size - fs->lfs_fsize) + 1;
492 		else
493 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1;
494 
495 		if (ip->i_size < newsize) {
496 			ip->i_size = newsize;
497 			/*
498 			 * tell vm our new size for the case the inode won't
499 			 * appear later.
500 			 */
501 			uvm_vnp_setsize(vp, newsize);
502 		}
503 	}
504 
505 	lfs_update_single(fs, NULL, vp, lbn, ndaddr, size);
506 
507 	LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
508 	sup->su_nbytes += size;
509 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
510 
511 	/* differences here should be due to UNWRITTEN indirect blocks. */
512 	KASSERT((lblkno(fs, ip->i_size) > NDADDR &&
513 	    ip->i_lfs_effnblks == ip->i_ffs1_blocks) ||
514 	    ip->i_lfs_effnblks >= ip->i_ffs1_blocks);
515 
516 #ifdef DEBUG_LFS_RFW
517 	/* Now look again to make sure it worked */
518 	ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
519 	for (i = num; i > 0; i--) {
520 		if (!a[i].in_exists)
521 			panic("update_meta: absent %d lv indirect block", i);
522 	}
523 	if (dbtofsb(fs, odaddr) != ndaddr)
524 		printf("update_meta: failed setting ino %d lbn %" PRId64
525 		    " to %" PRId64 "\n", ino, lbn, ndaddr);
526 #endif /* DEBUG_LFS_RFW */
527 	vput(vp);
528 	return 0;
529 }
530 
531 static int
532 update_inoblk(struct lfs *fs, daddr_t offset, struct ucred *cred,
533 	      struct proc *p)
534 {
535 	struct vnode *devvp, *vp;
536 	struct inode *ip;
537 	struct ufs1_dinode *dip;
538 	struct buf *dbp, *ibp;
539 	int error;
540 	daddr_t daddr;
541 	IFILE *ifp;
542 	SEGUSE *sup;
543 
544 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
545 
546 	/*
547 	 * Get the inode, update times and perms.
548 	 * DO NOT update disk blocks, we do that separately.
549 	 */
550 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp);
551 	if (error) {
552 #ifdef DEBUG_LFS_RFW
553 		printf("update_inoblk: bread returned %d\n", error);
554 #endif
555 		return error;
556 	}
557 	dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs);
558 	while (--dip >= (struct ufs1_dinode *)dbp->b_data) {
559 		if (dip->di_inumber > LFS_IFILE_INUM) {
560 			/* printf("ino %d version %d\n", dip->di_inumber,
561 			       dip->di_gen); */
562 			error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
563 					      p, &vp);
564 			if (error) {
565 #ifdef DEBUG_LFS_RFW
566 				printf("update_inoblk: lfs_rf_valloc returned %d\n", error);
567 #endif
568 				continue;
569 			}
570 			ip = VTOI(vp);
571 			if (dip->di_size != ip->i_size)
572 				VOP_TRUNCATE(vp, dip->di_size, 0, NOCRED, p);
573 			/* Get mode, link count, size, and times */
574 			memcpy(ip->i_din.ffs1_din, dip,
575 			       offsetof(struct ufs1_dinode, di_db[0]));
576 
577 			/* Then the rest, except di_blocks */
578 			ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
579 			ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
580 			ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
581 			ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
582 
583 			ip->i_mode = ip->i_ffs1_mode;
584 			ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink;
585 			ip->i_size = ip->i_ffs1_size;
586 
587 			LFS_SET_UINO(ip, IN_CHANGE | IN_MODIFIED | IN_UPDATE);
588 
589 			/* Re-initialize to get type right */
590 			ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
591 				  &vp);
592 			vput(vp);
593 
594 			/* Record change in location */
595 			LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
596 			daddr = ifp->if_daddr;
597 			ifp->if_daddr = dbtofsb(fs, dbp->b_blkno);
598 			error = LFS_BWRITE_LOG(ibp); /* Ifile */
599 			/* And do segment accounting */
600 			if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) {
601 				if (daddr > 0) {
602 					LFS_SEGENTRY(sup, fs, dtosn(fs, daddr),
603 						     ibp);
604 					sup->su_nbytes -= sizeof (struct ufs1_dinode);
605 					LFS_WRITESEGENTRY(sup, fs,
606 							  dtosn(fs, daddr),
607 							  ibp);
608 				}
609 				LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
610 					     ibp);
611 				sup->su_nbytes += sizeof (struct ufs1_dinode);
612 				LFS_WRITESEGENTRY(sup, fs,
613 						  dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
614 						  ibp);
615 			}
616 		}
617 	}
618 	dbp->b_flags |= B_AGE;
619 	brelse(dbp);
620 
621 	return 0;
622 }
623 
624 #define CHECK_CKSUM   0x0001  /* Check the checksum to make sure it's valid */
625 #define CHECK_UPDATE  0x0002  /* Update Ifile for new data blocks / inodes */
626 
627 static daddr_t
628 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial,
629 	     struct ucred *cred, int flags, int *pseg_flags, struct proc *p)
630 {
631 	struct vnode *devvp;
632 	struct buf *bp, *dbp;
633 	int error, nblocks = 0, ninos, i, j; /* XXX: gcc */
634 	SEGSUM *ssp;
635 	u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */
636 	daddr_t oldoffset;
637 	int32_t *iaddr;	/* XXX ondisk32 */
638 	FINFO *fip;
639 	SEGUSE *sup;
640 	size_t size;
641 
642 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
643 	/*
644 	 * If the segment has a superblock and we're at the top
645 	 * of the segment, skip the superblock.
646 	 */
647 	if (sntod(fs, dtosn(fs, offset)) == offset) {
648 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
649 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
650 			offset += btofsb(fs, LFS_SBPAD);
651 		brelse(bp);
652 	}
653 
654 	/* Read in the segment summary */
655 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp);
656 	if (error)
657 		return -1;
658 
659 	/* Check summary checksum */
660 	ssp = (SEGSUM *)bp->b_data;
661 	if (flags & CHECK_CKSUM) {
662 		if (ssp->ss_sumsum != cksum(&ssp->ss_datasum,
663 					   fs->lfs_sumsize -
664 					   sizeof(ssp->ss_sumsum))) {
665 #ifdef DEBUG_LFS_RFW
666 			printf("Sumsum error at 0x%" PRIx64 "\n", offset);
667 #endif
668 			offset = -1;
669 			goto err1;
670 		}
671 		if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) {
672 #ifdef DEBUG_LFS_RFW
673 			printf("Empty pseg at 0x%" PRIx64 "\n", offset);
674 #endif
675 			offset = -1;
676 			goto err1;
677 		}
678 		if (ssp->ss_create < fs->lfs_tstamp) {
679 #ifdef DEBUG_LFS_RFW
680 			printf("Old data at 0x%" PRIx64 "\n", offset);
681 #endif
682 			offset = -1;
683 			goto err1;
684 		}
685 	}
686 	if (fs->lfs_version > 1) {
687 		if (ssp->ss_serial != nextserial) {
688 #ifdef DEBUG_LFS_RFW
689 			printf("Unexpected serial number at 0x%" PRIx64
690 			    "\n", offset);
691 #endif
692 			offset = -1;
693 			goto err1;
694 		}
695 		if (ssp->ss_ident != fs->lfs_ident) {
696 #ifdef DEBUG_LFS_RFW
697 			printf("Incorrect fsid (0x%x vs 0x%x) at 0x%"
698 			    PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset);
699 #endif
700 			offset = -1;
701 			goto err1;
702 		}
703 	}
704 	if (pseg_flags)
705 		*pseg_flags = ssp->ss_flags;
706 	oldoffset = offset;
707 	offset += btofsb(fs, fs->lfs_sumsize);
708 
709 	ninos = howmany(ssp->ss_ninos, INOPB(fs));
710 	/* XXX ondisk32 */
711 	iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t));
712 	if (flags & CHECK_CKSUM) {
713 		/* Count blocks */
714 		nblocks = 0;
715 		fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
716 		for (i = 0; i < ssp->ss_nfinfo; ++i) {
717 			nblocks += fip->fi_nblocks;
718 			if (fip->fi_nblocks <= 0)
719 				break;
720 			/* XXX ondisk32 */
721 			fip = (FINFO *)(((char *)fip) + FINFOSIZE +
722 					(fip->fi_nblocks * sizeof(int32_t)));
723 		}
724 		nblocks += ninos;
725 		/* Create the sum array */
726 		datap = dp = (u_long *)malloc(nblocks * sizeof(u_long),
727 					      M_SEGMENT, M_WAITOK);
728 	}
729 
730 	/* Handle individual blocks */
731 	fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
732 	for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) {
733 		/* Inode block? */
734 		if (ninos && *iaddr == offset) {
735 			if (flags & CHECK_CKSUM) {
736 				/* Read in the head and add to the buffer */
737 				error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize,
738 					      cred, &dbp);
739 				if (error) {
740 					offset = -1;
741 					goto err2;
742 				}
743 				(*dp++) = ((u_long *)(dbp->b_data))[0];
744 				dbp->b_flags |= B_AGE;
745 				brelse(dbp);
746 			}
747 			if (flags & CHECK_UPDATE) {
748 				if ((error = update_inoblk(fs, offset, cred, p))
749 				    != 0) {
750 					offset = -1;
751 					goto err2;
752 				}
753 			}
754 			offset += btofsb(fs, fs->lfs_ibsize);
755 			--iaddr;
756 			--ninos;
757 			--i; /* compensate */
758 			continue;
759 		}
760 		/* printf("check: blocks from ino %d version %d\n",
761 		       fip->fi_ino, fip->fi_version); */
762 		size = fs->lfs_bsize;
763 		for (j = 0; j < fip->fi_nblocks; ++j) {
764 			if (j == fip->fi_nblocks - 1)
765 				size = fip->fi_lastlength;
766 			if (flags & CHECK_CKSUM) {
767 				error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp);
768 				if (error) {
769 					offset = -1;
770 					goto err2;
771 				}
772 				(*dp++) = ((u_long *)(dbp->b_data))[0];
773 				dbp->b_flags |= B_AGE;
774 				brelse(dbp);
775 			}
776 			/* Account for and update any direct blocks */
777 			if ((flags & CHECK_UPDATE) &&
778 			   fip->fi_ino > LFS_IFILE_INUM &&
779 			   fip->fi_blocks[j] >= 0) {
780 				update_meta(fs, fip->fi_ino, fip->fi_version,
781 					    fip->fi_blocks[j], offset, size, p);
782 			}
783 			offset += btofsb(fs, size);
784 		}
785 		/* XXX ondisk32 */
786 		fip = (FINFO *)(((char *)fip) + FINFOSIZE
787 				+ fip->fi_nblocks * sizeof(int32_t));
788 	}
789 	/* Checksum the array, compare */
790 	if ((flags & CHECK_CKSUM) &&
791 	   ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long)))
792 	{
793 #ifdef DEBUG_LFS_RFW
794 		printf("Datasum error at 0x%" PRIx64 " (wanted %x got %x)\n",
795 		    offset, ssp->ss_datasum, cksum(datap, nblocks *
796 					      sizeof(u_long)));
797 #endif
798 		offset = -1;
799 		goto err2;
800 	}
801 
802 	/* If we're at the end of the segment, move to the next */
803 	if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) !=
804 	   dtosn(fs, offset)) {
805 		if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) {
806 			offset = -1;
807 			goto err2;
808 		}
809 		offset = ssp->ss_next;
810 #ifdef DEBUG_LFS_RFW
811 		printf("LFS roll forward: moving on to offset 0x%" PRIx64
812 		       " -> segment %d\n", offset, dtosn(fs,offset));
813 #endif
814 	}
815 
816 	if (flags & CHECK_UPDATE) {
817 		fs->lfs_avail -= (offset - oldoffset);
818 		/* Don't clog the buffer queue */
819 		simple_lock(&lfs_subsys_lock);
820 		if (locked_queue_count > LFS_MAX_BUFS ||
821 		    locked_queue_bytes > LFS_MAX_BYTES) {
822 			lfs_flush(fs, SEGM_CKP);
823 		}
824 		simple_unlock(&lfs_subsys_lock);
825 	}
826 
827     err2:
828 	if (flags & CHECK_CKSUM)
829 		free(datap, M_SEGMENT);
830     err1:
831 	bp->b_flags |= B_AGE;
832 	brelse(bp);
833 
834 	/* XXX should we update the serial number even for bad psegs? */
835 	if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1)
836 		fs->lfs_serial = nextserial;
837 	return offset;
838 }
839 
840 /*
841  * Common code for mount and mountroot
842  * LFS specific
843  */
844 int
845 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct proc *p)
846 {
847 	extern struct vnode *rootvp;
848 	struct dlfs *tdfs, *dfs, *adfs;
849 	struct lfs *fs;
850 	struct ufsmount *ump;
851 	struct vnode *vp;
852 	struct buf *bp, *abp;
853 	struct partinfo dpart;
854 	dev_t dev;
855 	int error, i, ronly, secsize, fsbsize;
856 	struct ucred *cred;
857 	CLEANERINFO *cip;
858 	SEGUSE *sup;
859 	int flags, dirty, do_rollforward;
860 	daddr_t offset, oldoffset, lastgoodpseg, sb_addr;
861 	int sn, curseg;
862 
863 	cred = p ? p->p_ucred : NOCRED;
864 	/*
865 	 * Disallow multiple mounts of the same device.
866 	 * Disallow mounting of a device that is currently in use
867 	 * (except for root, which might share swap device for miniroot).
868 	 * Flush out any old buffers remaining from a previous use.
869 	 */
870 	if ((error = vfs_mountedon(devvp)) != 0)
871 		return (error);
872 	if (vcount(devvp) > 1 && devvp != rootvp)
873 		return (EBUSY);
874 	if ((error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0)) != 0)
875 		return (error);
876 
877 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
878 	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p);
879 	if (error)
880 		return (error);
881 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, p) != 0)
882 		secsize = DEV_BSIZE;
883 	else
884 		secsize = dpart.disklab->d_secsize;
885 
886 	/* Don't free random space on error. */
887 	bp = NULL;
888 	abp = NULL;
889 	ump = NULL;
890 
891 	sb_addr = LFS_LABELPAD / secsize;
892 	while (1) {
893 		/* Read in the superblock. */
894 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp);
895 		if (error)
896 			goto out;
897 		dfs = (struct dlfs *)bp->b_data;
898 
899 		/* Check the basics. */
900 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize >= MAXBSIZE ||
901 		    dfs->dlfs_version > LFS_VERSION ||
902 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
903 #ifdef DEBUG_LFS
904 			printf("lfs_mountfs: primary superblock sanity failed\n");
905 #endif
906 			error = EINVAL;		/* XXX needs translation */
907 			goto out;
908 		}
909 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT)
910 			printf("lfs_mountfs: warning: unknown inode format %d\n",
911 			       dfs->dlfs_inodefmt);
912 
913 		if (dfs->dlfs_version == 1)
914 			fsbsize = secsize;
915 		else {
916 			fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb +
917 				dfs->dlfs_fsbtodb);
918 			/*
919 			 * Could be, if the frag size is large enough, that we
920 			 * don't have the "real" primary superblock.  If that's
921 			 * the case, get the real one, and try again.
922 			 */
923 			if (sb_addr != dfs->dlfs_sboffs[0] <<
924 				       dfs->dlfs_fsbtodb) {
925 /* #ifdef DEBUG_LFS */
926 				printf("lfs_mountfs: sb daddr 0x%llx is not right, trying 0x%llx\n",
927 					(long long)sb_addr, (long long)(dfs->dlfs_sboffs[0] <<
928 						 dfs->dlfs_fsbtodb));
929 /* #endif */
930 				sb_addr = dfs->dlfs_sboffs[0] <<
931 					  dfs->dlfs_fsbtodb;
932 				brelse(bp);
933 				continue;
934 			}
935 		}
936 		break;
937 	}
938 
939 	/*
940 	 * Check the second superblock to see which is newer; then mount
941 	 * using the older of the two.	This is necessary to ensure that
942 	 * the filesystem is valid if it was not unmounted cleanly.
943 	 */
944 
945 	if (dfs->dlfs_sboffs[1] &&
946 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
947 	{
948 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize),
949 			LFS_SBPAD, cred, &abp);
950 		if (error)
951 			goto out;
952 		adfs = (struct dlfs *)abp->b_data;
953 
954 		if (dfs->dlfs_version == 1) {
955 			/* 1s resolution comparison */
956 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
957 				tdfs = adfs;
958 			else
959 				tdfs = dfs;
960 		} else {
961 			/* monotonic infinite-resolution comparison */
962 			if (adfs->dlfs_serial < dfs->dlfs_serial)
963 				tdfs = adfs;
964 			else
965 				tdfs = dfs;
966 		}
967 
968 		/* Check the basics. */
969 		if (tdfs->dlfs_magic != LFS_MAGIC ||
970 		    tdfs->dlfs_bsize > MAXBSIZE ||
971 		    tdfs->dlfs_version > LFS_VERSION ||
972 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
973 #ifdef DEBUG_LFS
974 			printf("lfs_mountfs: alt superblock sanity failed\n");
975 #endif
976 			error = EINVAL;		/* XXX needs translation */
977 			goto out;
978 		}
979 	} else {
980 #ifdef DEBUG_LFS
981 		printf("lfs_mountfs: invalid alt superblock daddr=0x%x\n",
982 			dfs->dlfs_sboffs[1]);
983 #endif
984 		error = EINVAL;
985 		goto out;
986 	}
987 
988 	/* Allocate the mount structure, copy the superblock into it. */
989 	fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
990 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
991 
992 	/* Compatibility */
993 	if (fs->lfs_version < 2) {
994 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
995 		fs->lfs_ibsize = fs->lfs_bsize;
996 		fs->lfs_start = fs->lfs_sboffs[0];
997 		fs->lfs_tstamp = fs->lfs_otstamp;
998 		fs->lfs_fsbtodb = 0;
999 	}
1000 
1001 	/* Before rolling forward, lock so vget will sleep for other procs */
1002 	fs->lfs_flags = LFS_NOTYET;
1003 	fs->lfs_rfpid = p->p_pid;
1004 
1005 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1006 	ump->um_lfs = fs;
1007 	ump->um_fstype = UFS1;
1008 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
1009 		bp->b_flags |= B_INVAL;
1010 		abp->b_flags |= B_INVAL;
1011 	}
1012 	brelse(bp);
1013 	bp = NULL;
1014 	brelse(abp);
1015 	abp = NULL;
1016 
1017 	/* Set up the I/O information */
1018 	fs->lfs_devbsize = secsize;
1019 	fs->lfs_iocount = 0;
1020 	fs->lfs_diropwait = 0;
1021 	fs->lfs_activesb = 0;
1022 	fs->lfs_uinodes = 0;
1023 	fs->lfs_ravail = 0;
1024 	fs->lfs_sbactive = 0;
1025 
1026 	/* Set up the ifile and lock aflags */
1027 	fs->lfs_doifile = 0;
1028 	fs->lfs_writer = 0;
1029 	fs->lfs_dirops = 0;
1030 	fs->lfs_nadirop = 0;
1031 	fs->lfs_seglock = 0;
1032 	fs->lfs_pdflush = 0;
1033 	fs->lfs_sleepers = 0;
1034 	simple_lock_init(&fs->lfs_interlock);
1035 	lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0);
1036 
1037 	/* Set the file system readonly/modify bits. */
1038 	fs->lfs_ronly = ronly;
1039 	if (ronly == 0)
1040 		fs->lfs_fmod = 1;
1041 
1042 	/* Initialize the mount structure. */
1043 	dev = devvp->v_rdev;
1044 	mp->mnt_data = ump;
1045 	mp->mnt_stat.f_fsid.val[0] = (long)dev;
1046 	mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_LFS);
1047 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
1048 	mp->mnt_maxsymlinklen = fs->lfs_maxsymlinklen;
1049 	mp->mnt_flag |= MNT_LOCAL;
1050 	mp->mnt_fs_bshift = fs->lfs_bshift;
1051 	ump->um_flags = 0;
1052 	ump->um_mountp = mp;
1053 	ump->um_dev = dev;
1054 	ump->um_devvp = devvp;
1055 	ump->um_bptrtodb = fs->lfs_fsbtodb;
1056 	ump->um_seqinc = fragstofsb(fs, fs->lfs_frag);
1057 	ump->um_nindir = fs->lfs_nindir;
1058 	ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
1059 	for (i = 0; i < MAXQUOTAS; i++)
1060 		ump->um_quotas[i] = NULLVP;
1061 	devvp->v_specmountpoint = mp;
1062 
1063 	/* Set up reserved memory for pageout */
1064 	lfs_setup_resblks(fs);
1065 	/* Set up vdirop tailq */
1066 	TAILQ_INIT(&fs->lfs_dchainhd);
1067 	/* and paging tailq */
1068 	TAILQ_INIT(&fs->lfs_pchainhd);
1069 
1070 	/*
1071 	 * We use the ifile vnode for almost every operation.  Instead of
1072 	 * retrieving it from the hash table each time we retrieve it here,
1073 	 * artificially increment the reference count and keep a pointer
1074 	 * to it in the incore copy of the superblock.
1075 	 */
1076 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1077 #ifdef DEBUG
1078 		printf("lfs_mountfs: ifile vget failed, error=%d\n", error);
1079 #endif
1080 		goto out;
1081 	}
1082 	fs->lfs_ivnode = vp;
1083 	VREF(vp);
1084 
1085 	/* Set up segment usage flags for the autocleaner. */
1086 	fs->lfs_nactive = 0;
1087 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
1088 						M_SEGMENT, M_WAITOK);
1089 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1090 						 M_SEGMENT, M_WAITOK);
1091 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1092 						 M_SEGMENT, M_WAITOK);
1093 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
1094 	for (i = 0; i < fs->lfs_nseg; i++) {
1095 		int changed;
1096 
1097 		LFS_SEGENTRY(sup, fs, i, bp);
1098 		changed = 0;
1099 		if (!ronly) {
1100 			if (sup->su_nbytes == 0 &&
1101 			    !(sup->su_flags & SEGUSE_EMPTY)) {
1102 				sup->su_flags |= SEGUSE_EMPTY;
1103 				++changed;
1104 			} else if (!(sup->su_nbytes == 0) &&
1105 				   (sup->su_flags & SEGUSE_EMPTY)) {
1106 				sup->su_flags &= ~SEGUSE_EMPTY;
1107 				++changed;
1108 			}
1109 			if (sup->su_flags & SEGUSE_ACTIVE) {
1110 				sup->su_flags &= ~SEGUSE_ACTIVE;
1111 				++changed;
1112 			}
1113 		}
1114 		fs->lfs_suflags[0][i] = sup->su_flags;
1115 		if (changed)
1116 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1117 		else
1118 			brelse(bp);
1119 	}
1120 
1121 	/*
1122 	 * Roll forward.
1123 	 *
1124 	 * We don't automatically roll forward for v1 filesystems, because
1125 	 * of the danger that the clock was turned back between the last
1126 	 * checkpoint and crash.  This would roll forward garbage.
1127 	 *
1128 	 * v2 filesystems don't have this problem because they use a
1129 	 * monotonically increasing serial number instead of a timestamp.
1130 	 */
1131 #ifdef LFS_DO_ROLLFORWARD
1132 	do_rollforward = !fs->lfs_ronly;
1133 #else
1134 	do_rollforward = (fs->lfs_version > 1 && !fs->lfs_ronly &&
1135 			  !(fs->lfs_pflags & LFS_PF_CLEAN));
1136 #endif
1137 	if (do_rollforward) {
1138 		u_int64_t nextserial;
1139 		/*
1140 		 * Phase I: Find the address of the last good partial
1141 		 * segment that was written after the checkpoint.  Mark
1142 		 * the segments in question dirty, so they won't be
1143 		 * reallocated.
1144 		 */
1145 		lastgoodpseg = oldoffset = offset = fs->lfs_offset;
1146 		flags = 0x0;
1147 #ifdef DEBUG_LFS_RFW
1148 		printf("LFS roll forward phase 1: starting at offset 0x%"
1149 		    PRIx64 "\n", offset);
1150 #endif
1151 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
1152 		if (!(sup->su_flags & SEGUSE_DIRTY))
1153 			--fs->lfs_nclean;
1154 		sup->su_flags |= SEGUSE_DIRTY;
1155 		LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp);
1156 		nextserial = fs->lfs_serial + 1;
1157 		while ((offset = check_segsum(fs, offset, nextserial,
1158 		    cred, CHECK_CKSUM, &flags, p)) > 0) {
1159 			nextserial++;
1160 			if (sntod(fs, oldoffset) != sntod(fs, offset)) {
1161 				LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset),
1162 					     bp);
1163 				if (!(sup->su_flags & SEGUSE_DIRTY))
1164 					--fs->lfs_nclean;
1165 				sup->su_flags |= SEGUSE_DIRTY;
1166 				LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset),
1167 					     bp);
1168 			}
1169 
1170 #ifdef DEBUG_LFS_RFW
1171 			printf("LFS roll forward phase 1: offset=0x%"
1172 			    PRIx64 "\n", offset);
1173 			if (flags & SS_DIROP) {
1174 				printf("lfs_mountfs: dirops at 0x%" PRIx64 "\n",
1175 				       oldoffset);
1176 				if (!(flags & SS_CONT))
1177 					printf("lfs_mountfs: dirops end "
1178 					       "at 0x%" PRIx64 "\n", oldoffset);
1179 			}
1180 #endif
1181 			if (!(flags & SS_CONT))
1182 				lastgoodpseg = offset;
1183 			oldoffset = offset;
1184 		}
1185 #ifdef DEBUG_LFS_RFW
1186 		if (flags & SS_CONT) {
1187 			printf("LFS roll forward: warning: incomplete "
1188 			       "dirops discarded\n");
1189 		}
1190 		printf("LFS roll forward phase 1: completed: "
1191 		       "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg);
1192 #endif
1193 		oldoffset = fs->lfs_offset;
1194 		if (fs->lfs_offset != lastgoodpseg) {
1195 			/* Don't overwrite what we're trying to preserve */
1196 			offset = fs->lfs_offset;
1197 			fs->lfs_offset = lastgoodpseg;
1198 			fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset));
1199 			for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) {
1200 				sn = (sn + 1) % fs->lfs_nseg;
1201 				if (sn == curseg)
1202 					panic("lfs_mountfs: no clean segments");
1203 				LFS_SEGENTRY(sup, fs, sn, bp);
1204 				dirty = (sup->su_flags & SEGUSE_DIRTY);
1205 				brelse(bp);
1206 				if (!dirty)
1207 					break;
1208 			}
1209 			fs->lfs_nextseg = sntod(fs, sn);
1210 
1211 			/*
1212 			 * Phase II: Roll forward from the first superblock.
1213 			 */
1214 			while (offset != lastgoodpseg) {
1215 #ifdef DEBUG_LFS_RFW
1216 				printf("LFS roll forward phase 2: 0x%"
1217 				    PRIx64 "\n", offset);
1218 #endif
1219 				offset = check_segsum(fs, offset,
1220 				    fs->lfs_serial + 1, cred, CHECK_UPDATE,
1221 				    NULL, p);
1222 			}
1223 
1224 			/*
1225 			 * Finish: flush our changes to disk.
1226 			 */
1227 			lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1228 			printf("lfs_mountfs: roll forward recovered %lld blocks\n",
1229 			       (long long)(lastgoodpseg - oldoffset));
1230 		}
1231 #ifdef DEBUG_LFS_RFW
1232 		printf("LFS roll forward complete\n");
1233 #endif
1234 	}
1235 	/* If writing, sb is not clean; record in case of immediate crash */
1236 	if (!fs->lfs_ronly) {
1237 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
1238 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
1239 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
1240 	}
1241 
1242 	/* Allow vget now that roll-forward is complete */
1243 	fs->lfs_flags &= ~(LFS_NOTYET);
1244 	wakeup(&fs->lfs_flags);
1245 
1246 	/*
1247 	 * Initialize the ifile cleaner info with information from
1248 	 * the superblock.
1249 	 */
1250 	LFS_CLEANERINFO(cip, fs, bp);
1251 	cip->clean = fs->lfs_nclean;
1252 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1253 	cip->avail = fs->lfs_avail;
1254 	cip->bfree = fs->lfs_bfree;
1255 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
1256 
1257 	/*
1258 	 * Mark the current segment as ACTIVE, since we're going to
1259 	 * be writing to it.
1260 	 */
1261 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
1262 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1263 	fs->lfs_nactive++;
1264 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
1265 
1266 	/* Now that roll-forward is done, unlock the Ifile */
1267 	vput(vp);
1268 
1269 	/* Comment on ifile size if it is too large */
1270 	if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS) {
1271 		fs->lfs_flags |= LFS_WARNED;
1272 		printf("lfs_mountfs: please consider increasing NBUF to at least %lld\n",
1273 			(long long)(fs->lfs_ivnode->v_size / fs->lfs_bsize) * (nbuf / LFS_MAX_BUFS));
1274 	}
1275 	if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES) {
1276 		fs->lfs_flags |= LFS_WARNED;
1277 		printf("lfs_mountfs: please consider increasing BUFPAGES to at least %lld\n",
1278 			(long long)(fs->lfs_ivnode->v_size * bufpages / LFS_MAX_BYTES));
1279 	}
1280 
1281 	return (0);
1282 out:
1283 	if (bp)
1284 		brelse(bp);
1285 	if (abp)
1286 		brelse(abp);
1287 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1288 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p);
1289 	VOP_UNLOCK(devvp, 0);
1290 	if (ump) {
1291 		free(ump->um_lfs, M_UFSMNT);
1292 		free(ump, M_UFSMNT);
1293 		mp->mnt_data = NULL;
1294 	}
1295 
1296 	/* Start the pagedaemon-anticipating daemon */
1297 	if (lfs_writer_daemon == 0 &&
1298 	    kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1299 		panic("fork lfs_writer");
1300 
1301 	return (error);
1302 }
1303 
1304 /*
1305  * unmount system call
1306  */
1307 int
1308 lfs_unmount(struct mount *mp, int mntflags, struct proc *p)
1309 {
1310 	struct ufsmount *ump;
1311 	struct lfs *fs;
1312 	int error, flags, ronly;
1313 	int s;
1314 
1315 	flags = 0;
1316 	if (mntflags & MNT_FORCE)
1317 		flags |= FORCECLOSE;
1318 
1319 	ump = VFSTOUFS(mp);
1320 	fs = ump->um_lfs;
1321 
1322 	/* wake up the cleaner so it can die */
1323 	wakeup(&fs->lfs_nextseg);
1324 	wakeup(&lfs_allclean_wakeup);
1325 	simple_lock(&fs->lfs_interlock);
1326 	while (fs->lfs_sleepers)
1327 		ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1328 			&fs->lfs_interlock);
1329 	simple_unlock(&fs->lfs_interlock);
1330 
1331 #ifdef QUOTA
1332 	if (mp->mnt_flag & MNT_QUOTA) {
1333 		int i;
1334 		error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags);
1335 		if (error)
1336 			return (error);
1337 		for (i = 0; i < MAXQUOTAS; i++) {
1338 			if (ump->um_quotas[i] == NULLVP)
1339 				continue;
1340 			quotaoff(p, mp, i);
1341 		}
1342 		/*
1343 		 * Here we fall through to vflush again to ensure
1344 		 * that we have gotten rid of all the system vnodes.
1345 		 */
1346 	}
1347 #endif
1348 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1349 		return (error);
1350 	if ((error = VFS_SYNC(mp, 1, p->p_ucred, p)) != 0)
1351 		return (error);
1352 	s = splbio();
1353 	if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd))
1354 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1355 	splx(s);
1356 
1357 	/* Comment on ifile size if it has become too large */
1358 	if (!(fs->lfs_flags & LFS_WARNED)) {
1359 		if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS)
1360 			printf("lfs_unmount: please consider increasing"
1361 				" NBUF to at least %lld\n",
1362 				(long long)(fs->lfs_ivnode->v_size /
1363 					    fs->lfs_bsize) *
1364 				(long long)(nbuf / LFS_MAX_BUFS));
1365 		if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES)
1366 			printf("lfs_unmount: please consider increasing"
1367 				" BUFPAGES to at least %lld\n",
1368 				(long long)(fs->lfs_ivnode->v_size *
1369 				bufpages / LFS_MAX_BYTES));
1370 	}
1371 
1372 	/* Explicitly write the superblock, to update serial and pflags */
1373 	fs->lfs_pflags |= LFS_PF_CLEAN;
1374 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
1375 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
1376 	while (fs->lfs_iocount)
1377 		tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0);
1378 
1379 	/* Finish with the Ifile, now that we're done with it */
1380 	vrele(fs->lfs_ivnode);
1381 	vgone(fs->lfs_ivnode);
1382 
1383 	ronly = !fs->lfs_ronly;
1384 	if (ump->um_devvp->v_type != VBAD)
1385 		ump->um_devvp->v_specmountpoint = NULL;
1386 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1387 	error = VOP_CLOSE(ump->um_devvp,
1388 	    ronly ? FREAD : FREAD|FWRITE, NOCRED, p);
1389 	vput(ump->um_devvp);
1390 
1391 	/* Free per-mount data structures */
1392 	free(fs->lfs_suflags[0], M_SEGMENT);
1393 	free(fs->lfs_suflags[1], M_SEGMENT);
1394 	free(fs->lfs_suflags, M_SEGMENT);
1395 	lfs_free_resblks(fs);
1396 	free(fs, M_UFSMNT);
1397 	free(ump, M_UFSMNT);
1398 
1399 	mp->mnt_data = NULL;
1400 	mp->mnt_flag &= ~MNT_LOCAL;
1401 	return (error);
1402 }
1403 
1404 /*
1405  * Get file system statistics.
1406  */
1407 int
1408 lfs_statfs(struct mount *mp, struct statfs *sbp, struct proc *p)
1409 {
1410 	struct lfs *fs;
1411 	struct ufsmount *ump;
1412 
1413 	ump = VFSTOUFS(mp);
1414 	fs = ump->um_lfs;
1415 	if (fs->lfs_magic != LFS_MAGIC)
1416 		panic("lfs_statfs: magic");
1417 
1418 	sbp->f_type = 0;
1419 	sbp->f_bsize = fs->lfs_fsize;
1420 	sbp->f_iosize = fs->lfs_bsize;
1421 	sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs));
1422 	sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs));
1423 	sbp->f_bavail = fsbtofrags(fs, (long)LFS_EST_BFREE(fs) -
1424 				  (long)LFS_EST_RSVD(fs));
1425 
1426 	sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1427 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1428 	copy_statfs_info(sbp, mp);
1429 	return (0);
1430 }
1431 
1432 /*
1433  * Go through the disk queues to initiate sandbagged IO;
1434  * go through the inodes to write those that have been modified;
1435  * initiate the writing of the super block if it has been modified.
1436  *
1437  * Note: we are always called with the filesystem marked `MPBUSY'.
1438  */
1439 int
1440 lfs_sync(struct mount *mp, int waitfor, struct ucred *cred, struct proc *p)
1441 {
1442 	int error;
1443 	struct lfs *fs;
1444 
1445 	fs = VFSTOUFS(mp)->um_lfs;
1446 	if (fs->lfs_ronly)
1447 		return 0;
1448 	lfs_writer_enter(fs, "lfs_dirops");
1449 
1450 	/* All syncs must be checkpoints until roll-forward is implemented. */
1451 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1452 	lfs_writer_leave(fs);
1453 #ifdef QUOTA
1454 	qsync(mp);
1455 #endif
1456 	return (error);
1457 }
1458 
1459 extern struct lock ufs_hashlock;
1460 
1461 /*
1462  * Look up an LFS dinode number to find its incore vnode.  If not already
1463  * in core, read it in from the specified device.  Return the inode locked.
1464  * Detection and handling of mount points must be done by the calling routine.
1465  */
1466 int
1467 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1468 {
1469 	struct lfs *fs;
1470 	struct ufs1_dinode *dip;
1471 	struct inode *ip;
1472 	struct buf *bp;
1473 	struct ifile *ifp;
1474 	struct vnode *vp;
1475 	struct ufsmount *ump;
1476 	daddr_t daddr;
1477 	dev_t dev;
1478 	int error, retries;
1479 	struct timespec ts;
1480 
1481 	ump = VFSTOUFS(mp);
1482 	dev = ump->um_dev;
1483 	fs = ump->um_lfs;
1484 
1485 	/*
1486 	 * If the filesystem is not completely mounted yet, suspend
1487 	 * any access requests (wait for roll-forward to complete).
1488 	 */
1489 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1490 		tsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0);
1491 
1492 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1493 		return (0);
1494 
1495 	if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) {
1496 		*vpp = NULL;
1497 		 return (error);
1498 	}
1499 
1500 	do {
1501 		if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) {
1502 			ungetnewvnode(vp);
1503 			return (0);
1504 		}
1505 	} while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0));
1506 
1507 	/* Translate the inode number to a disk address. */
1508 	if (ino == LFS_IFILE_INUM)
1509 		daddr = fs->lfs_idaddr;
1510 	else {
1511 		/* XXX bounds-check this too */
1512 		LFS_IENTRY(ifp, fs, ino, bp);
1513 		daddr = ifp->if_daddr;
1514 		if (fs->lfs_version > 1) {
1515 			ts.tv_sec = ifp->if_atime_sec;
1516 			ts.tv_nsec = ifp->if_atime_nsec;
1517 		}
1518 
1519 		brelse(bp);
1520 		if (daddr == LFS_UNUSED_DADDR) {
1521 			*vpp = NULLVP;
1522 			ungetnewvnode(vp);
1523 			lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1524 			return (ENOENT);
1525 		}
1526 	}
1527 
1528 	/* Allocate/init new vnode/inode. */
1529 	lfs_vcreate(mp, ino, vp);
1530 
1531 	/*
1532 	 * Put it onto its hash chain and lock it so that other requests for
1533 	 * this inode will block if they arrive while we are sleeping waiting
1534 	 * for old data structures to be purged or for the contents of the
1535 	 * disk portion of this inode to be read.
1536 	 */
1537 	ip = VTOI(vp);
1538 	ufs_ihashins(ip);
1539 	lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1540 
1541 	/*
1542 	 * XXX
1543 	 * This may not need to be here, logically it should go down with
1544 	 * the i_devvp initialization.
1545 	 * Ask Kirk.
1546 	 */
1547 	ip->i_lfs = ump->um_lfs;
1548 
1549 	/* Read in the disk contents for the inode, copy into the inode. */
1550 	retries = 0;
1551     again:
1552 	error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1553 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1554 		NOCRED, &bp);
1555 	if (error) {
1556 		/*
1557 		 * The inode does not contain anything useful, so it would
1558 		 * be misleading to leave it on its hash chain. With mode
1559 		 * still zero, it will be unlinked and returned to the free
1560 		 * list by vput().
1561 		 */
1562 		vput(vp);
1563 		brelse(bp);
1564 		*vpp = NULL;
1565 		return (error);
1566 	}
1567 
1568 	dip = lfs_ifind(fs, ino, bp);
1569 	if (dip == NULL) {
1570 		/* Assume write has not completed yet; try again */
1571 		bp->b_flags |= B_INVAL;
1572 		brelse(bp);
1573 		++retries;
1574 		if (retries > LFS_IFIND_RETRIES) {
1575 #ifdef DEBUG
1576 			/* If the seglock is held look at the bpp to see
1577 			   what is there anyway */
1578 			if (fs->lfs_seglock > 0) {
1579 				struct buf **bpp;
1580 				struct ufs1_dinode *dp;
1581 				int i;
1582 
1583 				for (bpp = fs->lfs_sp->bpp;
1584 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1585 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1586 					    bpp != fs->lfs_sp->bpp) {
1587 						/* Inode block */
1588 						printf("block 0x%" PRIx64 ": ",
1589 						    (*bpp)->b_blkno);
1590 						dp = (struct ufs1_dinode *)(*bpp)->b_data;
1591 						for (i = 0; i < INOPB(fs); i++)
1592 							if (dp[i].di_u.inumber)
1593 								printf("%d ", dp[i].di_u.inumber);
1594 						printf("\n");
1595 					}
1596 				}
1597 			}
1598 #endif
1599 			panic("lfs_vget: dinode not found");
1600 		}
1601 		printf("lfs_vget: dinode %d not found, retrying...\n", ino);
1602 		(void)tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs ifind", 1);
1603 		goto again;
1604 	}
1605 	*ip->i_din.ffs1_din = *dip;
1606 	brelse(bp);
1607 
1608 	if (fs->lfs_version > 1) {
1609 		ip->i_ffs1_atime = ts.tv_sec;
1610 		ip->i_ffs1_atimensec = ts.tv_nsec;
1611 	}
1612 
1613 	lfs_vinit(mp, &vp);
1614 
1615 	*vpp = vp;
1616 
1617 	KASSERT(VOP_ISLOCKED(vp));
1618 
1619 	return (0);
1620 }
1621 
1622 /*
1623  * File handle to vnode
1624  */
1625 int
1626 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1627 {
1628 	struct lfid *lfhp;
1629 	struct buf *bp;
1630 	IFILE *ifp;
1631 	int32_t daddr;
1632 	struct lfs *fs;
1633 
1634 	lfhp = (struct lfid *)fhp;
1635 	if (lfhp->lfid_ino < LFS_IFILE_INUM)
1636 		return ESTALE;
1637 
1638 	fs = VFSTOUFS(mp)->um_lfs;
1639 	if (lfhp->lfid_ident != fs->lfs_ident)
1640 		return ESTALE;
1641 
1642 	if (lfhp->lfid_ino >
1643 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1644 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1645 		return ESTALE;
1646 
1647 	if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfhp->lfid_ino) == NULLVP) {
1648 		LFS_IENTRY(ifp, fs, lfhp->lfid_ino, bp);
1649 		daddr = ifp->if_daddr;
1650 		brelse(bp);
1651 		if (daddr == LFS_UNUSED_DADDR)
1652 			return ESTALE;
1653 	}
1654 
1655 	return (ufs_fhtovp(mp, &lfhp->lfid_ufid, vpp));
1656 }
1657 
1658 /*
1659  * Vnode pointer to File handle
1660  */
1661 /* ARGSUSED */
1662 int
1663 lfs_vptofh(struct vnode *vp, struct fid *fhp)
1664 {
1665 	struct inode *ip;
1666 	struct lfid *lfhp;
1667 
1668 	ip = VTOI(vp);
1669 	lfhp = (struct lfid *)fhp;
1670 	lfhp->lfid_len = sizeof(struct lfid);
1671 	lfhp->lfid_ino = ip->i_number;
1672 	lfhp->lfid_gen = ip->i_gen;
1673 	lfhp->lfid_ident = ip->i_lfs->lfs_ident;
1674 	return (0);
1675 }
1676 
1677 static int
1678 sysctl_lfs_dostats(SYSCTLFN_ARGS)
1679 {
1680 	extern struct lfs_stats lfs_stats;
1681 	extern int lfs_dostats;
1682 	int error;
1683 
1684 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
1685 	if (error || newp == NULL)
1686 		return (error);
1687 
1688 	if (lfs_dostats == 0)
1689 		memset(&lfs_stats,0,sizeof(lfs_stats));
1690 
1691 	return (0);
1692 }
1693 
1694 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs setup")
1695 {
1696 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead;
1697 
1698 	sysctl_createv(SYSCTL_PERMANENT,
1699 		       CTLTYPE_NODE, "vfs", NULL,
1700 		       NULL, 0, NULL, 0,
1701 		       CTL_VFS, CTL_EOL);
1702 	sysctl_createv(SYSCTL_PERMANENT,
1703 		       CTLTYPE_NODE, "lfs", NULL,
1704 		       NULL, 0, NULL, 0,
1705 		       CTL_VFS, 5, CTL_EOL);
1706 	/*
1707 	 * XXX the "5" above could be dynamic, thereby eliminating one
1708 	 * more instance of the "number to vfs" mapping problem, but
1709 	 * "2" is the order as taken from sys/mount.h
1710 	 */
1711 
1712 	sysctl_createv(SYSCTL_PERMANENT|SYSCTL_READWRITE,
1713 		       CTLTYPE_INT, "flushindir", NULL,
1714 		       NULL, 0, &lfs_writeindir, 0,
1715 		       CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
1716 	sysctl_createv(SYSCTL_PERMANENT|SYSCTL_READWRITE,
1717 		       CTLTYPE_INT, "clean_vnhead", NULL,
1718 		       NULL, 0, &lfs_clean_vnhead, 0,
1719 		       CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
1720 	sysctl_createv(SYSCTL_PERMANENT|SYSCTL_READWRITE,
1721 		       CTLTYPE_INT, "dostats", NULL,
1722 		       sysctl_lfs_dostats, 0, &lfs_dostats, 0,
1723 		       CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
1724 }
1725 
1726 /*
1727  * ufs_bmaparray callback function for writing.
1728  *
1729  * Since blocks will be written to the new segment anyway,
1730  * we don't care about current daddr of them.
1731  */
1732 static boolean_t
1733 lfs_issequential_hole(const struct ufsmount *ump,
1734     daddr_t daddr0, daddr_t daddr1)
1735 {
1736 
1737 	KASSERT(daddr0 == UNWRITTEN ||
1738 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1739 	KASSERT(daddr1 == UNWRITTEN ||
1740 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1741 
1742 	/* NOTE: all we want to know here is 'hole or not'. */
1743 	/* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */
1744 
1745 	/*
1746 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1747 	 */
1748 	if (daddr0 != 0 && daddr1 != 0)
1749 		return TRUE;
1750 
1751 	/*
1752 	 * both are in hole?
1753 	 */
1754 	if (daddr0 == 0 && daddr1 == 0)
1755 		return TRUE; /* all holes are 'contiguous' for us. */
1756 
1757 	return FALSE;
1758 }
1759 
1760 /*
1761  * lfs_gop_write functions exactly like genfs_gop_write, except that
1762  * (1) it requires the seglock to be held by its caller, and sp->fip
1763  *     to be properly initialized (it will return without re-initializing
1764  *     sp->fip, and without calling lfs_writeseg).
1765  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1766  *     to determine how large a block it can write at once (though it does
1767  *     still use VOP_BMAP to find holes in the file);
1768  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1769  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1770  *     now have clusters of clusters, ick.)
1771  */
1772 static int
1773 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
1774 {
1775 	int i, s, error, run;
1776 	int fs_bshift;
1777 	vaddr_t kva;
1778 	off_t eof, offset, startoffset;
1779 	size_t bytes, iobytes, skipbytes;
1780 	daddr_t lbn, blkno;
1781 	struct vm_page *pg;
1782 	struct buf *mbp, *bp;
1783 	struct vnode *devvp = VTOI(vp)->i_devvp;
1784 	struct inode *ip = VTOI(vp);
1785 	struct lfs *fs = ip->i_lfs;
1786 	struct segment *sp = fs->lfs_sp;
1787 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1788 
1789 	/* The Ifile lives in the buffer cache */
1790 	if (vp == fs->lfs_ivnode)
1791 		return genfs_compat_gop_write(vp, pgs, npages, flags);
1792 
1793 	/*
1794 	 * Sometimes things slip past the filters in lfs_putpages,
1795 	 * and the pagedaemon tries to write pages---problem is
1796 	 * that the pagedaemon never acquires the segment lock.
1797 	 *
1798 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1799 	 * queue under the hypothesis that they couldn't have got here
1800 	 * unless they were modified *quite* recently.
1801 	 *
1802 	 * XXXUBC that last statement is an oversimplification of course.
1803 	 */
1804 	if (!(fs->lfs_seglock) || fs->lfs_lockpid != curproc->p_pid) {
1805 		simple_lock(&vp->v_interlock);
1806 #ifdef DEBUG
1807 		printf("lfs_gop_write: seglock not held\n");
1808 #endif
1809 		uvm_lock_pageq();
1810 		for (i = 0; i < npages; i++) {
1811 			pg = pgs[i];
1812 
1813 			if (pg->flags & PG_PAGEOUT)
1814 				uvmexp.paging--;
1815 			if (pg->flags & PG_DELWRI) {
1816 				uvm_pageunwire(pg);
1817 			}
1818 			uvm_pageactivate(pg);
1819 			pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
1820 #ifdef DEBUG_LFS
1821 			printf("pg[%d]->flags = %x\n", i, pg->flags);
1822 			printf("pg[%d]->pqflags = %x\n", i, pg->pqflags);
1823 			printf("pg[%d]->uanon = %p\n", i, pg->uanon);
1824 			printf("pg[%d]->uobject = %p\n", i, pg->uobject);
1825 			printf("pg[%d]->wire_count = %d\n", i, pg->wire_count);
1826 			printf("pg[%d]->loan_count = %d\n", i, pg->loan_count);
1827 #endif
1828 		}
1829 		/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
1830 		uvm_page_unbusy(pgs, npages);
1831 		uvm_unlock_pageq();
1832 		simple_unlock(&vp->v_interlock);
1833 		return EAGAIN;
1834 	}
1835 
1836 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1837 	    vp, pgs, npages, flags);
1838 
1839 	GOP_SIZE(vp, vp->v_size, &eof, GOP_SIZE_WRITE);
1840 
1841 	if (vp->v_type == VREG)
1842 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1843 	else
1844 		fs_bshift = DEV_BSHIFT;
1845 	error = 0;
1846 	pg = pgs[0];
1847 	startoffset = pg->offset;
1848 	bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1849 	skipbytes = 0;
1850 
1851 	/* KASSERT(bytes != 0); */
1852 	if (bytes == 0)
1853 		printf("ino %d bytes == 0 offset %" PRId64 "\n",
1854 			VTOI(vp)->i_number, pgs[0]->offset);
1855 
1856 	/* Swap PG_DELWRI for PG_PAGEOUT */
1857 	for (i = 0; i < npages; i++)
1858 		if (pgs[i]->flags & PG_DELWRI) {
1859 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1860 			pgs[i]->flags &= ~PG_DELWRI;
1861 			pgs[i]->flags |= PG_PAGEOUT;
1862 			uvmexp.paging++;
1863 			uvm_lock_pageq();
1864 			uvm_pageunwire(pgs[i]);
1865 			uvm_unlock_pageq();
1866 		}
1867 
1868 	/*
1869 	 * Check to make sure we're starting on a block boundary.
1870 	 * We'll check later to make sure we always write entire
1871 	 * blocks (or fragments).
1872 	 */
1873 	if (startoffset & fs->lfs_bmask)
1874 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1875 			startoffset, fs->lfs_bmask,
1876 			startoffset & fs->lfs_bmask);
1877 	KASSERT((startoffset & fs->lfs_bmask) == 0);
1878 	if (bytes & fs->lfs_ffmask) {
1879 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1880 		panic("lfs_gop_write: non-integer blocks");
1881 	}
1882 
1883 	kva = uvm_pagermapin(pgs, npages,
1884 	    UVMPAGER_MAPIN_WRITE | UVMPAGER_MAPIN_WAITOK);
1885 
1886 	s = splbio();
1887 	simple_lock(&global_v_numoutput_slock);
1888 	vp->v_numoutput += 2; /* one for biodone, one for aiodone */
1889 	simple_unlock(&global_v_numoutput_slock);
1890 	mbp = pool_get(&bufpool, PR_WAITOK);
1891 	splx(s);
1892 
1893 	memset(mbp, 0, sizeof(*bp));
1894 	BUF_INIT(mbp);
1895 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1896 	    vp, mbp, vp->v_numoutput, bytes);
1897 	mbp->b_bufsize = npages << PAGE_SHIFT;
1898 	mbp->b_data = (void *)kva;
1899 	mbp->b_resid = mbp->b_bcount = bytes;
1900 	mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL;
1901 	mbp->b_iodone = uvm_aio_biodone;
1902 	mbp->b_vp = vp;
1903 
1904 	bp = NULL;
1905 	for (offset = startoffset;
1906 	    bytes > 0;
1907 	    offset += iobytes, bytes -= iobytes) {
1908 		lbn = offset >> fs_bshift;
1909 		error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1910 		    lfs_issequential_hole);
1911 		if (error) {
1912 			UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
1913 			    error,0,0,0);
1914 			skipbytes += bytes;
1915 			bytes = 0;
1916 			break;
1917 		}
1918 
1919 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1920 		    bytes);
1921 		if (blkno == (daddr_t)-1) {
1922 			skipbytes += iobytes;
1923 			continue;
1924 		}
1925 
1926 		/*
1927 		 * Discover how much we can really pack into this buffer.
1928 		 */
1929 		/* If no room in the current segment, finish it up */
1930 		if (sp->sum_bytes_left < sizeof(int32_t) ||
1931 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1932 			int version;
1933 
1934 			lfs_updatemeta(sp);
1935 
1936 			version = sp->fip->fi_version;
1937 			(void) lfs_writeseg(fs, sp);
1938 
1939 			sp->fip->fi_version = version;
1940 			sp->fip->fi_ino = ip->i_number;
1941 			/* Add the current file to the segment summary. */
1942 			++((SEGSUM *)(sp->segsum))->ss_nfinfo;
1943 			sp->sum_bytes_left -= FINFOSIZE;
1944 		}
1945 		/* Check both for space in segment and space in segsum */
1946 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1947 					<< fs_bshift);
1948 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1949 				       << fs_bshift);
1950 		KASSERT(iobytes > 0);
1951 
1952 		/* if it's really one i/o, don't make a second buf */
1953 		if (offset == startoffset && iobytes == bytes) {
1954 			bp = mbp;
1955 			/* printf("bp is mbp\n"); */
1956 			/* correct overcount if there is no second buffer */
1957 			s = splbio();
1958 			simple_lock(&global_v_numoutput_slock);
1959 			--vp->v_numoutput;
1960 			simple_unlock(&global_v_numoutput_slock);
1961 			splx(s);
1962 		} else {
1963 			/* printf("bp is not mbp\n"); */
1964 			s = splbio();
1965 			bp = pool_get(&bufpool, PR_WAITOK);
1966 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1967 			    vp, bp, vp->v_numoutput, 0);
1968 			splx(s);
1969 			memset(bp, 0, sizeof(*bp));
1970 			BUF_INIT(bp);
1971 			bp->b_data = (char *)kva +
1972 			    (vaddr_t)(offset - pg->offset);
1973 			bp->b_resid = bp->b_bcount = iobytes;
1974 			bp->b_flags = B_BUSY|B_WRITE|B_CALL;
1975 			bp->b_iodone = uvm_aio_biodone1;
1976 		}
1977 
1978 		/* XXX This is silly ... is this necessary? */
1979 		bp->b_vp = NULL;
1980 		s = splbio();
1981 		bgetvp(vp, bp);
1982 		splx(s);
1983 
1984 		bp->b_lblkno = lblkno(fs, offset);
1985 		bp->b_private = mbp;
1986 		if (devvp->v_type == VBLK) {
1987 			bp->b_dev = devvp->v_rdev;
1988 		}
1989 		VOP_BWRITE(bp);
1990 		while (lfs_gatherblock(sp, bp, NULL))
1991 			continue;
1992 	}
1993 
1994 	if (skipbytes) {
1995 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
1996 		s = splbio();
1997 		if (error) {
1998 			mbp->b_flags |= B_ERROR;
1999 			mbp->b_error = error;
2000 		}
2001 		mbp->b_resid -= skipbytes;
2002 		if (mbp->b_resid == 0) {
2003 			biodone(mbp);
2004 		}
2005 		splx(s);
2006 	}
2007 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
2008 	return (0);
2009 }
2010 
2011 /*
2012  * finish vnode/inode initialization.
2013  * used by lfs_vget and lfs_fastvget.
2014  */
2015 void
2016 lfs_vinit(struct mount *mp, struct vnode **vpp)
2017 {
2018 	struct vnode *vp = *vpp;
2019 	struct inode *ip = VTOI(vp);
2020 	struct ufsmount *ump = VFSTOUFS(mp);
2021 	int i;
2022 
2023 	ip->i_mode = ip->i_ffs1_mode;
2024 	ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink;
2025 	ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
2026 	ip->i_flags = ip->i_ffs1_flags;
2027 	ip->i_gen = ip->i_ffs1_gen;
2028 	ip->i_uid = ip->i_ffs1_uid;
2029 	ip->i_gid = ip->i_ffs1_gid;
2030 
2031 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
2032 
2033 	/*
2034 	 * Initialize the vnode from the inode, check for aliases.  In all
2035 	 * cases re-init ip, the underlying vnode/inode may have changed.
2036 	 */
2037 	ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2038 
2039 	memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
2040 	if (vp->v_type != VLNK ||
2041 	    VTOI(vp)->i_size >= vp->v_mount->mnt_maxsymlinklen) {
2042 		struct lfs *fs = ump->um_lfs;
2043 #ifdef DEBUG
2044 		for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
2045 		    i < NDADDR; i++) {
2046 			if (ip->i_ffs1_db[i] != 0) {
2047 inconsistent:
2048 				lfs_dump_dinode(ip->i_din.ffs1_din);
2049 				panic("inconsistent inode");
2050 			}
2051 		}
2052 		for ( ; i < NDADDR + NIADDR; i++) {
2053 			if (ip->i_ffs1_ib[i - NDADDR] != 0) {
2054 				goto inconsistent;
2055 			}
2056 		}
2057 #endif /* DEBUG */
2058 		for (i = 0; i < NDADDR; i++)
2059 			if (ip->i_ffs1_db[i] != 0)
2060 				ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
2061 	}
2062 
2063 #ifdef DEBUG
2064 	if (vp->v_type == VNON) {
2065 		printf("lfs_vinit: ino %d is type VNON! (ifmt=%o)\n",
2066 		       ip->i_number, (ip->i_mode & IFMT) >> 12);
2067 		lfs_dump_dinode(ip->i_din.ffs1_din);
2068 #ifdef DDB
2069 		Debugger();
2070 #endif /* DDB */
2071 	}
2072 #endif /* DEBUG */
2073 
2074 	/*
2075 	 * Finish inode initialization now that aliasing has been resolved.
2076 	 */
2077 
2078 	ip->i_devvp = ump->um_devvp;
2079 	VREF(ip->i_devvp);
2080 	genfs_node_init(vp, &lfs_genfsops);
2081 	uvm_vnp_setsize(vp, ip->i_size);
2082 
2083 	*vpp = vp;
2084 }
2085