xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision 969da55f1ccd8f94fb0d823d2bcc57601e67e15f)
1 /*	$NetBSD: lfs_vfsops.c,v 1.154 2004/07/05 07:28:46 pk 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 1.154 2004/07/05 07:28:46 pk 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_statvfs,
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 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
157 	lfs_vnodeopv_descs,
158 };
159 
160 struct genfs_ops lfs_genfsops = {
161 	lfs_gop_size,
162 	ufs_gop_alloc,
163 	lfs_gop_write,
164 };
165 
166 /*
167  * XXX Same structure as FFS inodes?  Should we share a common pool?
168  */
169 POOL_INIT(lfs_inode_pool, sizeof(struct inode), 0, 0, 0, "lfsinopl",
170     &pool_allocator_nointr);
171 POOL_INIT(lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, "lfsdinopl",
172     &pool_allocator_nointr);
173 POOL_INIT(lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, "lfsinoextpl",
174     &pool_allocator_nointr);
175 
176 /*
177  * The writer daemon.  UVM keeps track of how many dirty pages we are holding
178  * in lfs_subsys_pages; the daemon flushes the filesystem when this value
179  * crosses the (user-defined) threshhold LFS_MAX_PAGES.
180  */
181 static void
182 lfs_writerd(void *arg)
183 {
184 #ifdef LFS_PD
185 	struct mount *mp, *nmp;
186 	struct lfs *fs;
187 #endif
188 
189 	lfs_writer_daemon = curproc->p_pid;
190 
191 	simple_lock(&lfs_subsys_lock);
192 	for (;;) {
193 		ltsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", 0,
194 		    &lfs_subsys_lock);
195 
196 #ifdef LFS_PD
197 		/*
198 		 * Look through the list of LFSs to see if any of them
199 		 * have requested pageouts.
200 		 */
201 		simple_lock(&mountlist_slock);
202 		for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
203 		     mp = nmp) {
204 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
205 				nmp = CIRCLEQ_NEXT(mp, mnt_list);
206 				continue;
207 			}
208 			if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS,
209 				    MFSNAMELEN) == 0) {
210 				fs = VFSTOUFS(mp)->um_lfs;
211 				if (fs->lfs_pdflush ||
212 				    !TAILQ_EMPTY(&fs->lfs_pchainhd)) {
213 					fs->lfs_pdflush = 0;
214 					lfs_flush_fs(fs, 0);
215 				}
216 			}
217 
218 			simple_lock(&mountlist_slock);
219 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
220 			vfs_unbusy(mp);
221 		}
222 		simple_unlock(&mountlist_slock);
223 #endif /* LFS_PD */
224 
225 		/*
226 		 * If global state wants a flush, flush everything.
227 		 */
228 		simple_lock(&lfs_subsys_lock);
229 		while (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
230 			locked_queue_bytes > LFS_MAX_BYTES ||
231 			lfs_subsys_pages > LFS_MAX_PAGES) {
232 
233 #ifdef DEBUG_LFS_FLUSH
234 			if (lfs_do_flush)
235 				printf("daemon: lfs_do_flush\n");
236 			if (locked_queue_count > LFS_MAX_BUFS)
237 				printf("daemon: lqc = %d, max %d\n",
238 					locked_queue_count, LFS_MAX_BUFS);
239 			if (locked_queue_bytes > LFS_MAX_BYTES)
240 				printf("daemon: lqb = %ld, max %ld\n",
241 					locked_queue_bytes, LFS_MAX_BYTES);
242 			if (lfs_subsys_pages > LFS_MAX_PAGES)
243 				printf("daemon: lssp = %d, max %d\n",
244 					lfs_subsys_pages, LFS_MAX_PAGES);
245 #endif /* DEBUG_LFS_FLUSH */
246 			lfs_flush(NULL, SEGM_WRITERD);
247 			lfs_do_flush = 0;
248 		}
249 	}
250 	/* NOTREACHED */
251 }
252 
253 /*
254  * Initialize the filesystem, most work done by ufs_init.
255  */
256 void
257 lfs_init()
258 {
259 #ifdef _LKM
260 	malloc_type_attach(M_SEGMENT);
261 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
262 	    "lfsinopl", &pool_allocator_nointr);
263 	pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
264 	    "lfsdinopl", &pool_allocator_nointr);
265 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
266 	    "lfsinoextpl", &pool_allocator_nointr);
267 #endif
268 	ufs_init();
269 
270 #ifdef DEBUG
271 	memset(lfs_log, 0, sizeof(lfs_log));
272 #endif
273 	simple_lock_init(&lfs_subsys_lock);
274 }
275 
276 void
277 lfs_reinit()
278 {
279 	ufs_reinit();
280 }
281 
282 void
283 lfs_done()
284 {
285 	ufs_done();
286 #ifdef _LKM
287 	pool_destroy(&lfs_inode_pool);
288 	pool_destroy(&lfs_dinode_pool);
289 	pool_destroy(&lfs_inoext_pool);
290 	malloc_type_detach(M_SEGMENT);
291 #endif
292 }
293 
294 /*
295  * Called by main() when ufs is going to be mounted as root.
296  */
297 int
298 lfs_mountroot()
299 {
300 	extern struct vnode *rootvp;
301 	struct mount *mp;
302 	struct proc *p = curproc;	/* XXX */
303 	int error;
304 
305 	if (root_device->dv_class != DV_DISK)
306 		return (ENODEV);
307 
308 	if (rootdev == NODEV)
309 		return (ENODEV);
310 	/*
311 	 * Get vnodes for swapdev and rootdev.
312 	 */
313 	if ((error = bdevvp(rootdev, &rootvp))) {
314 		printf("lfs_mountroot: can't setup bdevvp's");
315 		return (error);
316 	}
317 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
318 		vrele(rootvp);
319 		return (error);
320 	}
321 	if ((error = lfs_mountfs(rootvp, mp, p))) {
322 		mp->mnt_op->vfs_refcount--;
323 		vfs_unbusy(mp);
324 		free(mp, M_MOUNT);
325 		vrele(rootvp);
326 		return (error);
327 	}
328 	simple_lock(&mountlist_slock);
329 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
330 	simple_unlock(&mountlist_slock);
331 	(void)lfs_statvfs(mp, &mp->mnt_stat, p);
332 	vfs_unbusy(mp);
333 	setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp));
334 	return (0);
335 }
336 
337 /*
338  * VFS Operations.
339  *
340  * mount system call
341  */
342 int
343 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct proc *p)
344 {
345 	struct vnode *devvp;
346 	struct ufs_args args;
347 	struct ufsmount *ump = NULL;
348 	struct lfs *fs = NULL;				/* LFS */
349 	int error;
350 	mode_t accessmode;
351 
352 	if (mp->mnt_flag & MNT_GETARGS) {
353 		ump = VFSTOUFS(mp);
354 		if (ump == NULL)
355 			return EIO;
356 		args.fspec = NULL;
357 		vfs_showexport(mp, &args.export, &ump->um_export);
358 		return copyout(&args, data, sizeof(args));
359 	}
360 	error = copyin(data, &args, sizeof (struct ufs_args));
361 	if (error)
362 		return (error);
363 
364 	/*
365 	 * If updating, check whether changing from read-only to
366 	 * read/write; if there is no device name, that's all we do.
367 	 */
368 	if (mp->mnt_flag & MNT_UPDATE) {
369 		ump = VFSTOUFS(mp);
370 		fs = ump->um_lfs;
371 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
372 			/*
373 			 * If upgrade to read-write by non-root, then verify
374 			 * that user has necessary permissions on the device.
375 			 */
376 			if (p->p_ucred->cr_uid != 0) {
377 				vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
378 				error = VOP_ACCESS(ump->um_devvp, VREAD|VWRITE,
379 						   p->p_ucred, p);
380 				VOP_UNLOCK(ump->um_devvp, 0);
381 				if (error)
382 					return (error);
383 			}
384 			fs->lfs_ronly = 0;
385 		}
386 		if (args.fspec == 0) {
387 			/*
388 			 * Process export requests.
389 			 */
390 			return (vfs_export(mp, &ump->um_export, &args.export));
391 		}
392 	}
393 	/*
394 	 * Not an update, or updating the name: look up the name
395 	 * and verify that it refers to a sensible block device.
396 	 */
397 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p);
398 	if ((error = namei(ndp)) != 0)
399 		return (error);
400 	devvp = ndp->ni_vp;
401 	if (devvp->v_type != VBLK) {
402 		vrele(devvp);
403 		return (ENOTBLK);
404 	}
405 	if (bdevsw_lookup(devvp->v_rdev) == NULL) {
406 		vrele(devvp);
407 		return (ENXIO);
408 	}
409 	/*
410 	 * If mount by non-root, then verify that user has necessary
411 	 * permissions on the device.
412 	 */
413 	if (p->p_ucred->cr_uid != 0) {
414 		accessmode = VREAD;
415 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
416 			accessmode |= VWRITE;
417 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
418 		error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p);
419 		if (error) {
420 			vput(devvp);
421 			return (error);
422 		}
423 		VOP_UNLOCK(devvp, 0);
424 	}
425 	if ((mp->mnt_flag & MNT_UPDATE) == 0)
426 		error = lfs_mountfs(devvp, mp, p);		/* LFS */
427 	else {
428 		if (devvp != ump->um_devvp)
429 			error = EINVAL;	/* needs translation */
430 		else
431 			vrele(devvp);
432 	}
433 	if (error) {
434 		vrele(devvp);
435 		return (error);
436 	}
437 	ump = VFSTOUFS(mp);
438 	fs = ump->um_lfs;					/* LFS */
439 	return set_statvfs_info(path, UIO_USERSPACE, args.fspec,
440 	    UIO_USERSPACE, mp, p);
441 }
442 
443 /*
444  * Roll-forward code.
445  */
446 
447 /*
448  * Load the appropriate indirect block, and change the appropriate pointer.
449  * Mark the block dirty.  Do segment and avail accounting.
450  */
451 static int
452 update_meta(struct lfs *fs, ino_t ino, int version, daddr_t lbn,
453 	    daddr_t ndaddr, size_t size, struct proc *p)
454 {
455 	int error;
456 	struct vnode *vp;
457 	struct inode *ip;
458 #ifdef DEBUG_LFS_RFW
459 	daddr_t odaddr;
460 	struct indir a[NIADDR];
461 	int num;
462 	int i;
463 #endif /* DEBUG_LFS_RFW */
464 	struct buf *bp;
465 	SEGUSE *sup;
466 
467 	KASSERT(lbn >= 0);	/* no indirect blocks */
468 
469 	if ((error = lfs_rf_valloc(fs, ino, version, p, &vp)) != 0) {
470 #ifdef DEBUG_LFS_RFW
471 		printf("update_meta: ino %d: lfs_rf_valloc returned %d\n", ino,
472 		       error);
473 #endif /* DEBUG_LFS_RFW */
474 		return error;
475 	}
476 
477 	if ((error = VOP_BALLOC(vp, (lbn << fs->lfs_bshift), size,
478 				NOCRED, 0, &bp)) != 0) {
479 		vput(vp);
480 		return (error);
481 	}
482 	/* No need to write, the block is already on disk */
483 	if (bp->b_flags & B_DELWRI) {
484 		LFS_UNLOCK_BUF(bp);
485 		fs->lfs_avail += btofsb(fs, bp->b_bcount);
486 	}
487 	bp->b_flags |= B_INVAL;
488 	brelse(bp);
489 
490 	/*
491 	 * Extend the file, if it is not large enough already.
492 	 * XXX this is not exactly right, we don't know how much of the
493 	 * XXX last block is actually used.  We hope that an inode will
494 	 * XXX appear later to give the correct size.
495 	 */
496 	ip = VTOI(vp);
497 	if (ip->i_size <= (lbn << fs->lfs_bshift)) {
498 		u_int64_t newsize;
499 
500 		if (lbn < NDADDR)
501 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) +
502 				(size - fs->lfs_fsize) + 1;
503 		else
504 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1;
505 
506 		if (ip->i_size < newsize) {
507 			ip->i_size = newsize;
508 			/*
509 			 * tell vm our new size for the case the inode won't
510 			 * appear later.
511 			 */
512 			uvm_vnp_setsize(vp, newsize);
513 		}
514 	}
515 
516 	lfs_update_single(fs, NULL, vp, lbn, ndaddr, size);
517 
518 	LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
519 	sup->su_nbytes += size;
520 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
521 
522 	/* differences here should be due to UNWRITTEN indirect blocks. */
523 	KASSERT((lblkno(fs, ip->i_size) > NDADDR &&
524 	    ip->i_lfs_effnblks == ip->i_ffs1_blocks) ||
525 	    ip->i_lfs_effnblks >= ip->i_ffs1_blocks);
526 
527 #ifdef DEBUG_LFS_RFW
528 	/* Now look again to make sure it worked */
529 	ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
530 	for (i = num; i > 0; i--) {
531 		if (!a[i].in_exists)
532 			panic("update_meta: absent %d lv indirect block", i);
533 	}
534 	if (dbtofsb(fs, odaddr) != ndaddr)
535 		printf("update_meta: failed setting ino %d lbn %" PRId64
536 		    " to %" PRId64 "\n", ino, lbn, ndaddr);
537 #endif /* DEBUG_LFS_RFW */
538 	vput(vp);
539 	return 0;
540 }
541 
542 static int
543 update_inoblk(struct lfs *fs, daddr_t offset, struct ucred *cred,
544 	      struct proc *p)
545 {
546 	struct vnode *devvp, *vp;
547 	struct inode *ip;
548 	struct ufs1_dinode *dip;
549 	struct buf *dbp, *ibp;
550 	int error;
551 	daddr_t daddr;
552 	IFILE *ifp;
553 	SEGUSE *sup;
554 
555 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
556 
557 	/*
558 	 * Get the inode, update times and perms.
559 	 * DO NOT update disk blocks, we do that separately.
560 	 */
561 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp);
562 	if (error) {
563 #ifdef DEBUG_LFS_RFW
564 		printf("update_inoblk: bread returned %d\n", error);
565 #endif
566 		return error;
567 	}
568 	dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs);
569 	while (--dip >= (struct ufs1_dinode *)dbp->b_data) {
570 		if (dip->di_inumber > LFS_IFILE_INUM) {
571 			/* printf("ino %d version %d\n", dip->di_inumber,
572 			       dip->di_gen); */
573 			error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
574 					      p, &vp);
575 			if (error) {
576 #ifdef DEBUG_LFS_RFW
577 				printf("update_inoblk: lfs_rf_valloc returned %d\n", error);
578 #endif
579 				continue;
580 			}
581 			ip = VTOI(vp);
582 			if (dip->di_size != ip->i_size)
583 				VOP_TRUNCATE(vp, dip->di_size, 0, NOCRED, p);
584 			/* Get mode, link count, size, and times */
585 			memcpy(ip->i_din.ffs1_din, dip,
586 			       offsetof(struct ufs1_dinode, di_db[0]));
587 
588 			/* Then the rest, except di_blocks */
589 			ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
590 			ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
591 			ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
592 			ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
593 
594 			ip->i_mode = ip->i_ffs1_mode;
595 			ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink;
596 			ip->i_size = ip->i_ffs1_size;
597 
598 			LFS_SET_UINO(ip, IN_CHANGE | IN_MODIFIED | IN_UPDATE);
599 
600 			/* Re-initialize to get type right */
601 			ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
602 				  &vp);
603 			vput(vp);
604 
605 			/* Record change in location */
606 			LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
607 			daddr = ifp->if_daddr;
608 			ifp->if_daddr = dbtofsb(fs, dbp->b_blkno);
609 			error = LFS_BWRITE_LOG(ibp); /* Ifile */
610 			/* And do segment accounting */
611 			if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) {
612 				if (daddr > 0) {
613 					LFS_SEGENTRY(sup, fs, dtosn(fs, daddr),
614 						     ibp);
615 					sup->su_nbytes -= sizeof (struct ufs1_dinode);
616 					LFS_WRITESEGENTRY(sup, fs,
617 							  dtosn(fs, daddr),
618 							  ibp);
619 				}
620 				LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
621 					     ibp);
622 				sup->su_nbytes += sizeof (struct ufs1_dinode);
623 				LFS_WRITESEGENTRY(sup, fs,
624 						  dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
625 						  ibp);
626 			}
627 		}
628 	}
629 	dbp->b_flags |= B_AGE;
630 	brelse(dbp);
631 
632 	return 0;
633 }
634 
635 #define CHECK_CKSUM   0x0001  /* Check the checksum to make sure it's valid */
636 #define CHECK_UPDATE  0x0002  /* Update Ifile for new data blocks / inodes */
637 
638 static daddr_t
639 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial,
640 	     struct ucred *cred, int flags, int *pseg_flags, struct proc *p)
641 {
642 	struct vnode *devvp;
643 	struct buf *bp, *dbp;
644 	int error, nblocks = 0, ninos, i, j; /* XXX: gcc */
645 	SEGSUM *ssp;
646 	u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */
647 	daddr_t oldoffset;
648 	int32_t *iaddr;	/* XXX ondisk32 */
649 	FINFO *fip;
650 	SEGUSE *sup;
651 	size_t size;
652 
653 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
654 	/*
655 	 * If the segment has a superblock and we're at the top
656 	 * of the segment, skip the superblock.
657 	 */
658 	if (sntod(fs, dtosn(fs, offset)) == offset) {
659 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
660 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
661 			offset += btofsb(fs, LFS_SBPAD);
662 		brelse(bp);
663 	}
664 
665 	/* Read in the segment summary */
666 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp);
667 	if (error)
668 		return -1;
669 
670 	/* Check summary checksum */
671 	ssp = (SEGSUM *)bp->b_data;
672 	if (flags & CHECK_CKSUM) {
673 		if (ssp->ss_sumsum != cksum(&ssp->ss_datasum,
674 					   fs->lfs_sumsize -
675 					   sizeof(ssp->ss_sumsum))) {
676 #ifdef DEBUG_LFS_RFW
677 			printf("Sumsum error at 0x%" PRIx64 "\n", offset);
678 #endif
679 			offset = -1;
680 			goto err1;
681 		}
682 		if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) {
683 #ifdef DEBUG_LFS_RFW
684 			printf("Empty pseg at 0x%" PRIx64 "\n", offset);
685 #endif
686 			offset = -1;
687 			goto err1;
688 		}
689 		if (ssp->ss_create < fs->lfs_tstamp) {
690 #ifdef DEBUG_LFS_RFW
691 			printf("Old data at 0x%" PRIx64 "\n", offset);
692 #endif
693 			offset = -1;
694 			goto err1;
695 		}
696 	}
697 	if (fs->lfs_version > 1) {
698 		if (ssp->ss_serial != nextserial) {
699 #ifdef DEBUG_LFS_RFW
700 			printf("Unexpected serial number at 0x%" PRIx64
701 			    "\n", offset);
702 #endif
703 			offset = -1;
704 			goto err1;
705 		}
706 		if (ssp->ss_ident != fs->lfs_ident) {
707 #ifdef DEBUG_LFS_RFW
708 			printf("Incorrect fsid (0x%x vs 0x%x) at 0x%"
709 			    PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset);
710 #endif
711 			offset = -1;
712 			goto err1;
713 		}
714 	}
715 	if (pseg_flags)
716 		*pseg_flags = ssp->ss_flags;
717 	oldoffset = offset;
718 	offset += btofsb(fs, fs->lfs_sumsize);
719 
720 	ninos = howmany(ssp->ss_ninos, INOPB(fs));
721 	/* XXX ondisk32 */
722 	iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t));
723 	if (flags & CHECK_CKSUM) {
724 		/* Count blocks */
725 		nblocks = 0;
726 		fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
727 		for (i = 0; i < ssp->ss_nfinfo; ++i) {
728 			nblocks += fip->fi_nblocks;
729 			if (fip->fi_nblocks <= 0)
730 				break;
731 			/* XXX ondisk32 */
732 			fip = (FINFO *)(((char *)fip) + FINFOSIZE +
733 					(fip->fi_nblocks * sizeof(int32_t)));
734 		}
735 		nblocks += ninos;
736 		/* Create the sum array */
737 		datap = dp = (u_long *)malloc(nblocks * sizeof(u_long),
738 					      M_SEGMENT, M_WAITOK);
739 	}
740 
741 	/* Handle individual blocks */
742 	fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
743 	for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) {
744 		/* Inode block? */
745 		if (ninos && *iaddr == offset) {
746 			if (flags & CHECK_CKSUM) {
747 				/* Read in the head and add to the buffer */
748 				error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize,
749 					      cred, &dbp);
750 				if (error) {
751 					offset = -1;
752 					goto err2;
753 				}
754 				(*dp++) = ((u_long *)(dbp->b_data))[0];
755 				dbp->b_flags |= B_AGE;
756 				brelse(dbp);
757 			}
758 			if (flags & CHECK_UPDATE) {
759 				if ((error = update_inoblk(fs, offset, cred, p))
760 				    != 0) {
761 					offset = -1;
762 					goto err2;
763 				}
764 			}
765 			offset += btofsb(fs, fs->lfs_ibsize);
766 			--iaddr;
767 			--ninos;
768 			--i; /* compensate */
769 			continue;
770 		}
771 		/* printf("check: blocks from ino %d version %d\n",
772 		       fip->fi_ino, fip->fi_version); */
773 		size = fs->lfs_bsize;
774 		for (j = 0; j < fip->fi_nblocks; ++j) {
775 			if (j == fip->fi_nblocks - 1)
776 				size = fip->fi_lastlength;
777 			if (flags & CHECK_CKSUM) {
778 				error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp);
779 				if (error) {
780 					offset = -1;
781 					goto err2;
782 				}
783 				(*dp++) = ((u_long *)(dbp->b_data))[0];
784 				dbp->b_flags |= B_AGE;
785 				brelse(dbp);
786 			}
787 			/* Account for and update any direct blocks */
788 			if ((flags & CHECK_UPDATE) &&
789 			   fip->fi_ino > LFS_IFILE_INUM &&
790 			   fip->fi_blocks[j] >= 0) {
791 				update_meta(fs, fip->fi_ino, fip->fi_version,
792 					    fip->fi_blocks[j], offset, size, p);
793 			}
794 			offset += btofsb(fs, size);
795 		}
796 		/* XXX ondisk32 */
797 		fip = (FINFO *)(((char *)fip) + FINFOSIZE
798 				+ fip->fi_nblocks * sizeof(int32_t));
799 	}
800 	/* Checksum the array, compare */
801 	if ((flags & CHECK_CKSUM) &&
802 	   ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long)))
803 	{
804 #ifdef DEBUG_LFS_RFW
805 		printf("Datasum error at 0x%" PRIx64 " (wanted %x got %x)\n",
806 		    offset, ssp->ss_datasum, cksum(datap, nblocks *
807 					      sizeof(u_long)));
808 #endif
809 		offset = -1;
810 		goto err2;
811 	}
812 
813 	/* If we're at the end of the segment, move to the next */
814 	if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) !=
815 	   dtosn(fs, offset)) {
816 		if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) {
817 			offset = -1;
818 			goto err2;
819 		}
820 		offset = ssp->ss_next;
821 #ifdef DEBUG_LFS_RFW
822 		printf("LFS roll forward: moving on to offset 0x%" PRIx64
823 		       " -> segment %d\n", offset, dtosn(fs,offset));
824 #endif
825 	}
826 
827 	if (flags & CHECK_UPDATE) {
828 		fs->lfs_avail -= (offset - oldoffset);
829 		/* Don't clog the buffer queue */
830 		simple_lock(&lfs_subsys_lock);
831 		if (locked_queue_count > LFS_MAX_BUFS ||
832 		    locked_queue_bytes > LFS_MAX_BYTES) {
833 			lfs_flush(fs, SEGM_CKP);
834 		}
835 		simple_unlock(&lfs_subsys_lock);
836 	}
837 
838     err2:
839 	if (flags & CHECK_CKSUM)
840 		free(datap, M_SEGMENT);
841     err1:
842 	bp->b_flags |= B_AGE;
843 	brelse(bp);
844 
845 	/* XXX should we update the serial number even for bad psegs? */
846 	if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1)
847 		fs->lfs_serial = nextserial;
848 	return offset;
849 }
850 
851 /*
852  * Common code for mount and mountroot
853  * LFS specific
854  */
855 int
856 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct proc *p)
857 {
858 	extern struct vnode *rootvp;
859 	struct dlfs *tdfs, *dfs, *adfs;
860 	struct lfs *fs;
861 	struct ufsmount *ump;
862 	struct vnode *vp;
863 	struct buf *bp, *abp;
864 	struct partinfo dpart;
865 	dev_t dev;
866 	int error, i, ronly, secsize, fsbsize;
867 	struct ucred *cred;
868 	CLEANERINFO *cip;
869 	SEGUSE *sup;
870 	int flags, dirty, do_rollforward;
871 	daddr_t offset, oldoffset, lastgoodpseg, sb_addr;
872 	int sn, curseg;
873 
874 	cred = p ? p->p_ucred : NOCRED;
875 	/*
876 	 * Disallow multiple mounts of the same device.
877 	 * Disallow mounting of a device that is currently in use
878 	 * (except for root, which might share swap device for miniroot).
879 	 * Flush out any old buffers remaining from a previous use.
880 	 */
881 	if ((error = vfs_mountedon(devvp)) != 0)
882 		return (error);
883 	if (vcount(devvp) > 1 && devvp != rootvp)
884 		return (EBUSY);
885 	if ((error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0)) != 0)
886 		return (error);
887 
888 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
889 	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p);
890 	if (error)
891 		return (error);
892 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, p) != 0)
893 		secsize = DEV_BSIZE;
894 	else
895 		secsize = dpart.disklab->d_secsize;
896 
897 	/* Don't free random space on error. */
898 	bp = NULL;
899 	abp = NULL;
900 	ump = NULL;
901 
902 	sb_addr = LFS_LABELPAD / secsize;
903 	while (1) {
904 		/* Read in the superblock. */
905 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp);
906 		if (error)
907 			goto out;
908 		dfs = (struct dlfs *)bp->b_data;
909 
910 		/* Check the basics. */
911 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize >= MAXBSIZE ||
912 		    dfs->dlfs_version > LFS_VERSION ||
913 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
914 #ifdef DEBUG_LFS
915 			printf("lfs_mountfs: primary superblock sanity failed\n");
916 #endif
917 			error = EINVAL;		/* XXX needs translation */
918 			goto out;
919 		}
920 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT)
921 			printf("lfs_mountfs: warning: unknown inode format %d\n",
922 			       dfs->dlfs_inodefmt);
923 
924 		if (dfs->dlfs_version == 1)
925 			fsbsize = secsize;
926 		else {
927 			fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb +
928 				dfs->dlfs_fsbtodb);
929 			/*
930 			 * Could be, if the frag size is large enough, that we
931 			 * don't have the "real" primary superblock.  If that's
932 			 * the case, get the real one, and try again.
933 			 */
934 			if (sb_addr != dfs->dlfs_sboffs[0] <<
935 				       dfs->dlfs_fsbtodb) {
936 /* #ifdef DEBUG_LFS */
937 				printf("lfs_mountfs: sb daddr 0x%llx is not right, trying 0x%llx\n",
938 					(long long)sb_addr, (long long)(dfs->dlfs_sboffs[0] <<
939 						 dfs->dlfs_fsbtodb));
940 /* #endif */
941 				sb_addr = dfs->dlfs_sboffs[0] <<
942 					  dfs->dlfs_fsbtodb;
943 				brelse(bp);
944 				continue;
945 			}
946 		}
947 		break;
948 	}
949 
950 	/*
951 	 * Check the second superblock to see which is newer; then mount
952 	 * using the older of the two.	This is necessary to ensure that
953 	 * the filesystem is valid if it was not unmounted cleanly.
954 	 */
955 
956 	if (dfs->dlfs_sboffs[1] &&
957 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
958 	{
959 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize),
960 			LFS_SBPAD, cred, &abp);
961 		if (error)
962 			goto out;
963 		adfs = (struct dlfs *)abp->b_data;
964 
965 		if (dfs->dlfs_version == 1) {
966 			/* 1s resolution comparison */
967 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
968 				tdfs = adfs;
969 			else
970 				tdfs = dfs;
971 		} else {
972 			/* monotonic infinite-resolution comparison */
973 			if (adfs->dlfs_serial < dfs->dlfs_serial)
974 				tdfs = adfs;
975 			else
976 				tdfs = dfs;
977 		}
978 
979 		/* Check the basics. */
980 		if (tdfs->dlfs_magic != LFS_MAGIC ||
981 		    tdfs->dlfs_bsize > MAXBSIZE ||
982 		    tdfs->dlfs_version > LFS_VERSION ||
983 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
984 #ifdef DEBUG_LFS
985 			printf("lfs_mountfs: alt superblock sanity failed\n");
986 #endif
987 			error = EINVAL;		/* XXX needs translation */
988 			goto out;
989 		}
990 	} else {
991 #ifdef DEBUG_LFS
992 		printf("lfs_mountfs: invalid alt superblock daddr=0x%x\n",
993 			dfs->dlfs_sboffs[1]);
994 #endif
995 		error = EINVAL;
996 		goto out;
997 	}
998 
999 	/* Allocate the mount structure, copy the superblock into it. */
1000 	fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
1001 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
1002 
1003 	/* Compatibility */
1004 	if (fs->lfs_version < 2) {
1005 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
1006 		fs->lfs_ibsize = fs->lfs_bsize;
1007 		fs->lfs_start = fs->lfs_sboffs[0];
1008 		fs->lfs_tstamp = fs->lfs_otstamp;
1009 		fs->lfs_fsbtodb = 0;
1010 	}
1011 
1012 	/* Before rolling forward, lock so vget will sleep for other procs */
1013 	fs->lfs_flags = LFS_NOTYET;
1014 	fs->lfs_rfpid = p->p_pid;
1015 
1016 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1017 	ump->um_lfs = fs;
1018 	ump->um_fstype = UFS1;
1019 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
1020 		bp->b_flags |= B_INVAL;
1021 		abp->b_flags |= B_INVAL;
1022 	}
1023 	brelse(bp);
1024 	bp = NULL;
1025 	brelse(abp);
1026 	abp = NULL;
1027 
1028 	/* Set up the I/O information */
1029 	fs->lfs_devbsize = secsize;
1030 	fs->lfs_iocount = 0;
1031 	fs->lfs_diropwait = 0;
1032 	fs->lfs_activesb = 0;
1033 	fs->lfs_uinodes = 0;
1034 	fs->lfs_ravail = 0;
1035 	fs->lfs_sbactive = 0;
1036 
1037 	/* Set up the ifile and lock aflags */
1038 	fs->lfs_doifile = 0;
1039 	fs->lfs_writer = 0;
1040 	fs->lfs_dirops = 0;
1041 	fs->lfs_nadirop = 0;
1042 	fs->lfs_seglock = 0;
1043 	fs->lfs_pdflush = 0;
1044 	fs->lfs_sleepers = 0;
1045 	simple_lock_init(&fs->lfs_interlock);
1046 	lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0);
1047 
1048 	/* Set the file system readonly/modify bits. */
1049 	fs->lfs_ronly = ronly;
1050 	if (ronly == 0)
1051 		fs->lfs_fmod = 1;
1052 
1053 	/* Initialize the mount structure. */
1054 	dev = devvp->v_rdev;
1055 	mp->mnt_data = ump;
1056 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
1057 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
1058 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
1059 	mp->mnt_stat.f_namemax = MAXNAMLEN;
1060 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
1061 	mp->mnt_maxsymlinklen = fs->lfs_maxsymlinklen;
1062 	mp->mnt_flag |= MNT_LOCAL;
1063 	mp->mnt_fs_bshift = fs->lfs_bshift;
1064 	ump->um_flags = 0;
1065 	ump->um_mountp = mp;
1066 	ump->um_dev = dev;
1067 	ump->um_devvp = devvp;
1068 	ump->um_bptrtodb = fs->lfs_fsbtodb;
1069 	ump->um_seqinc = fragstofsb(fs, fs->lfs_frag);
1070 	ump->um_nindir = fs->lfs_nindir;
1071 	ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
1072 	for (i = 0; i < MAXQUOTAS; i++)
1073 		ump->um_quotas[i] = NULLVP;
1074 	devvp->v_specmountpoint = mp;
1075 
1076 	/* Set up reserved memory for pageout */
1077 	lfs_setup_resblks(fs);
1078 	/* Set up vdirop tailq */
1079 	TAILQ_INIT(&fs->lfs_dchainhd);
1080 	/* and paging tailq */
1081 	TAILQ_INIT(&fs->lfs_pchainhd);
1082 
1083 	/*
1084 	 * We use the ifile vnode for almost every operation.  Instead of
1085 	 * retrieving it from the hash table each time we retrieve it here,
1086 	 * artificially increment the reference count and keep a pointer
1087 	 * to it in the incore copy of the superblock.
1088 	 */
1089 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1090 #ifdef DEBUG
1091 		printf("lfs_mountfs: ifile vget failed, error=%d\n", error);
1092 #endif
1093 		goto out;
1094 	}
1095 	fs->lfs_ivnode = vp;
1096 	VREF(vp);
1097 
1098 	/* Set up segment usage flags for the autocleaner. */
1099 	fs->lfs_nactive = 0;
1100 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
1101 						M_SEGMENT, M_WAITOK);
1102 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1103 						 M_SEGMENT, M_WAITOK);
1104 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1105 						 M_SEGMENT, M_WAITOK);
1106 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
1107 	for (i = 0; i < fs->lfs_nseg; i++) {
1108 		int changed;
1109 
1110 		LFS_SEGENTRY(sup, fs, i, bp);
1111 		changed = 0;
1112 		if (!ronly) {
1113 			if (sup->su_nbytes == 0 &&
1114 			    !(sup->su_flags & SEGUSE_EMPTY)) {
1115 				sup->su_flags |= SEGUSE_EMPTY;
1116 				++changed;
1117 			} else if (!(sup->su_nbytes == 0) &&
1118 				   (sup->su_flags & SEGUSE_EMPTY)) {
1119 				sup->su_flags &= ~SEGUSE_EMPTY;
1120 				++changed;
1121 			}
1122 			if (sup->su_flags & SEGUSE_ACTIVE) {
1123 				sup->su_flags &= ~SEGUSE_ACTIVE;
1124 				++changed;
1125 			}
1126 		}
1127 		fs->lfs_suflags[0][i] = sup->su_flags;
1128 		if (changed)
1129 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1130 		else
1131 			brelse(bp);
1132 	}
1133 
1134 	/*
1135 	 * Roll forward.
1136 	 *
1137 	 * We don't automatically roll forward for v1 filesystems, because
1138 	 * of the danger that the clock was turned back between the last
1139 	 * checkpoint and crash.  This would roll forward garbage.
1140 	 *
1141 	 * v2 filesystems don't have this problem because they use a
1142 	 * monotonically increasing serial number instead of a timestamp.
1143 	 */
1144 #ifdef LFS_DO_ROLLFORWARD
1145 	do_rollforward = !fs->lfs_ronly;
1146 #else
1147 	do_rollforward = (fs->lfs_version > 1 && !fs->lfs_ronly &&
1148 			  !(fs->lfs_pflags & LFS_PF_CLEAN));
1149 #endif
1150 	if (do_rollforward) {
1151 		u_int64_t nextserial;
1152 		/*
1153 		 * Phase I: Find the address of the last good partial
1154 		 * segment that was written after the checkpoint.  Mark
1155 		 * the segments in question dirty, so they won't be
1156 		 * reallocated.
1157 		 */
1158 		lastgoodpseg = oldoffset = offset = fs->lfs_offset;
1159 		flags = 0x0;
1160 #ifdef DEBUG_LFS_RFW
1161 		printf("LFS roll forward phase 1: starting at offset 0x%"
1162 		    PRIx64 "\n", offset);
1163 #endif
1164 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
1165 		if (!(sup->su_flags & SEGUSE_DIRTY))
1166 			--fs->lfs_nclean;
1167 		sup->su_flags |= SEGUSE_DIRTY;
1168 		LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp);
1169 		nextserial = fs->lfs_serial + 1;
1170 		while ((offset = check_segsum(fs, offset, nextserial,
1171 		    cred, CHECK_CKSUM, &flags, p)) > 0) {
1172 			nextserial++;
1173 			if (sntod(fs, oldoffset) != sntod(fs, offset)) {
1174 				LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset),
1175 					     bp);
1176 				if (!(sup->su_flags & SEGUSE_DIRTY))
1177 					--fs->lfs_nclean;
1178 				sup->su_flags |= SEGUSE_DIRTY;
1179 				LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset),
1180 					     bp);
1181 			}
1182 
1183 #ifdef DEBUG_LFS_RFW
1184 			printf("LFS roll forward phase 1: offset=0x%"
1185 			    PRIx64 "\n", offset);
1186 			if (flags & SS_DIROP) {
1187 				printf("lfs_mountfs: dirops at 0x%" PRIx64 "\n",
1188 				       oldoffset);
1189 				if (!(flags & SS_CONT))
1190 					printf("lfs_mountfs: dirops end "
1191 					       "at 0x%" PRIx64 "\n", oldoffset);
1192 			}
1193 #endif
1194 			if (!(flags & SS_CONT))
1195 				lastgoodpseg = offset;
1196 			oldoffset = offset;
1197 		}
1198 #ifdef DEBUG_LFS_RFW
1199 		if (flags & SS_CONT) {
1200 			printf("LFS roll forward: warning: incomplete "
1201 			       "dirops discarded\n");
1202 		}
1203 		printf("LFS roll forward phase 1: completed: "
1204 		       "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg);
1205 #endif
1206 		oldoffset = fs->lfs_offset;
1207 		if (fs->lfs_offset != lastgoodpseg) {
1208 			/* Don't overwrite what we're trying to preserve */
1209 			offset = fs->lfs_offset;
1210 			fs->lfs_offset = lastgoodpseg;
1211 			fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset));
1212 			for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) {
1213 				sn = (sn + 1) % fs->lfs_nseg;
1214 				if (sn == curseg)
1215 					panic("lfs_mountfs: no clean segments");
1216 				LFS_SEGENTRY(sup, fs, sn, bp);
1217 				dirty = (sup->su_flags & SEGUSE_DIRTY);
1218 				brelse(bp);
1219 				if (!dirty)
1220 					break;
1221 			}
1222 			fs->lfs_nextseg = sntod(fs, sn);
1223 
1224 			/*
1225 			 * Phase II: Roll forward from the first superblock.
1226 			 */
1227 			while (offset != lastgoodpseg) {
1228 #ifdef DEBUG_LFS_RFW
1229 				printf("LFS roll forward phase 2: 0x%"
1230 				    PRIx64 "\n", offset);
1231 #endif
1232 				offset = check_segsum(fs, offset,
1233 				    fs->lfs_serial + 1, cred, CHECK_UPDATE,
1234 				    NULL, p);
1235 			}
1236 
1237 			/*
1238 			 * Finish: flush our changes to disk.
1239 			 */
1240 			lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1241 			printf("lfs_mountfs: roll forward recovered %lld blocks\n",
1242 			       (long long)(lastgoodpseg - oldoffset));
1243 		}
1244 #ifdef DEBUG_LFS_RFW
1245 		printf("LFS roll forward complete\n");
1246 #endif
1247 	}
1248 	/* If writing, sb is not clean; record in case of immediate crash */
1249 	if (!fs->lfs_ronly) {
1250 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
1251 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
1252 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
1253 	}
1254 
1255 	/* Allow vget now that roll-forward is complete */
1256 	fs->lfs_flags &= ~(LFS_NOTYET);
1257 	wakeup(&fs->lfs_flags);
1258 
1259 	/*
1260 	 * Initialize the ifile cleaner info with information from
1261 	 * the superblock.
1262 	 */
1263 	LFS_CLEANERINFO(cip, fs, bp);
1264 	cip->clean = fs->lfs_nclean;
1265 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1266 	cip->avail = fs->lfs_avail;
1267 	cip->bfree = fs->lfs_bfree;
1268 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
1269 
1270 	/*
1271 	 * Mark the current segment as ACTIVE, since we're going to
1272 	 * be writing to it.
1273 	 */
1274 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
1275 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1276 	fs->lfs_nactive++;
1277 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
1278 
1279 	/* Now that roll-forward is done, unlock the Ifile */
1280 	vput(vp);
1281 
1282 	/* Comment on ifile size if it is too large */
1283 	if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS) {
1284 		fs->lfs_flags |= LFS_WARNED;
1285 		printf("lfs_mountfs: please consider increasing NBUF to at least %lld\n",
1286 			(long long)(fs->lfs_ivnode->v_size / fs->lfs_bsize) * (nbuf / LFS_MAX_BUFS));
1287 	}
1288 	if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES) {
1289 		fs->lfs_flags |= LFS_WARNED;
1290 		printf("lfs_mountfs: please consider increasing BUFPAGES to at least %lld\n",
1291 			(long long)(fs->lfs_ivnode->v_size * bufpages / LFS_MAX_BYTES));
1292 	}
1293 
1294 	return (0);
1295 out:
1296 	if (bp)
1297 		brelse(bp);
1298 	if (abp)
1299 		brelse(abp);
1300 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1301 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p);
1302 	VOP_UNLOCK(devvp, 0);
1303 	if (ump) {
1304 		free(ump->um_lfs, M_UFSMNT);
1305 		free(ump, M_UFSMNT);
1306 		mp->mnt_data = NULL;
1307 	}
1308 
1309 	/* Start the pagedaemon-anticipating daemon */
1310 	if (lfs_writer_daemon == 0 &&
1311 	    kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1312 		panic("fork lfs_writer");
1313 
1314 	return (error);
1315 }
1316 
1317 /*
1318  * unmount system call
1319  */
1320 int
1321 lfs_unmount(struct mount *mp, int mntflags, struct proc *p)
1322 {
1323 	struct ufsmount *ump;
1324 	struct lfs *fs;
1325 	int error, flags, ronly;
1326 	int s;
1327 
1328 	flags = 0;
1329 	if (mntflags & MNT_FORCE)
1330 		flags |= FORCECLOSE;
1331 
1332 	ump = VFSTOUFS(mp);
1333 	fs = ump->um_lfs;
1334 
1335 	/* wake up the cleaner so it can die */
1336 	wakeup(&fs->lfs_nextseg);
1337 	wakeup(&lfs_allclean_wakeup);
1338 	simple_lock(&fs->lfs_interlock);
1339 	while (fs->lfs_sleepers)
1340 		ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1341 			&fs->lfs_interlock);
1342 	simple_unlock(&fs->lfs_interlock);
1343 
1344 #ifdef QUOTA
1345 	if (mp->mnt_flag & MNT_QUOTA) {
1346 		int i;
1347 		error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags);
1348 		if (error)
1349 			return (error);
1350 		for (i = 0; i < MAXQUOTAS; i++) {
1351 			if (ump->um_quotas[i] == NULLVP)
1352 				continue;
1353 			quotaoff(p, mp, i);
1354 		}
1355 		/*
1356 		 * Here we fall through to vflush again to ensure
1357 		 * that we have gotten rid of all the system vnodes.
1358 		 */
1359 	}
1360 #endif
1361 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1362 		return (error);
1363 	if ((error = VFS_SYNC(mp, 1, p->p_ucred, p)) != 0)
1364 		return (error);
1365 	s = splbio();
1366 	if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd))
1367 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1368 	splx(s);
1369 
1370 	/* Comment on ifile size if it has become too large */
1371 	if (!(fs->lfs_flags & LFS_WARNED)) {
1372 		if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS)
1373 			printf("lfs_unmount: please consider increasing"
1374 				" NBUF to at least %lld\n",
1375 				(long long)(fs->lfs_ivnode->v_size /
1376 					    fs->lfs_bsize) *
1377 				(long long)(nbuf / LFS_MAX_BUFS));
1378 		if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES)
1379 			printf("lfs_unmount: please consider increasing"
1380 				" BUFPAGES to at least %lld\n",
1381 				(long long)(fs->lfs_ivnode->v_size *
1382 				bufpages / LFS_MAX_BYTES));
1383 	}
1384 
1385 	/* Explicitly write the superblock, to update serial and pflags */
1386 	fs->lfs_pflags |= LFS_PF_CLEAN;
1387 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
1388 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
1389 	while (fs->lfs_iocount)
1390 		tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0);
1391 
1392 	/* Finish with the Ifile, now that we're done with it */
1393 	vrele(fs->lfs_ivnode);
1394 	vgone(fs->lfs_ivnode);
1395 
1396 	ronly = !fs->lfs_ronly;
1397 	if (ump->um_devvp->v_type != VBAD)
1398 		ump->um_devvp->v_specmountpoint = NULL;
1399 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1400 	error = VOP_CLOSE(ump->um_devvp,
1401 	    ronly ? FREAD : FREAD|FWRITE, NOCRED, p);
1402 	vput(ump->um_devvp);
1403 
1404 	/* Free per-mount data structures */
1405 	free(fs->lfs_suflags[0], M_SEGMENT);
1406 	free(fs->lfs_suflags[1], M_SEGMENT);
1407 	free(fs->lfs_suflags, M_SEGMENT);
1408 	lfs_free_resblks(fs);
1409 	free(fs, M_UFSMNT);
1410 	free(ump, M_UFSMNT);
1411 
1412 	mp->mnt_data = NULL;
1413 	mp->mnt_flag &= ~MNT_LOCAL;
1414 	return (error);
1415 }
1416 
1417 /*
1418  * Get file system statistics.
1419  */
1420 int
1421 lfs_statvfs(struct mount *mp, struct statvfs *sbp, struct proc *p)
1422 {
1423 	struct lfs *fs;
1424 	struct ufsmount *ump;
1425 
1426 	ump = VFSTOUFS(mp);
1427 	fs = ump->um_lfs;
1428 	if (fs->lfs_magic != LFS_MAGIC)
1429 		panic("lfs_statvfs: magic");
1430 
1431 	sbp->f_bsize = fs->lfs_bsize;
1432 	sbp->f_frsize = fs->lfs_fsize;
1433 	sbp->f_iosize = fs->lfs_bsize;
1434 	sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs));
1435 	sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs));
1436 	sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs));
1437 	if (sbp->f_bfree > sbp->f_bresvd)
1438 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1439 	else
1440 		sbp->f_bavail = 0;
1441 
1442 	sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1443 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1444 	sbp->f_favail = sbp->f_ffree;
1445 	sbp->f_fresvd = 0;
1446 	copy_statvfs_info(sbp, mp);
1447 	return (0);
1448 }
1449 
1450 /*
1451  * Go through the disk queues to initiate sandbagged IO;
1452  * go through the inodes to write those that have been modified;
1453  * initiate the writing of the super block if it has been modified.
1454  *
1455  * Note: we are always called with the filesystem marked `MPBUSY'.
1456  */
1457 int
1458 lfs_sync(struct mount *mp, int waitfor, struct ucred *cred, struct proc *p)
1459 {
1460 	int error;
1461 	struct lfs *fs;
1462 
1463 	fs = VFSTOUFS(mp)->um_lfs;
1464 	if (fs->lfs_ronly)
1465 		return 0;
1466 	lfs_writer_enter(fs, "lfs_dirops");
1467 
1468 	/* All syncs must be checkpoints until roll-forward is implemented. */
1469 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1470 	lfs_writer_leave(fs);
1471 #ifdef QUOTA
1472 	qsync(mp);
1473 #endif
1474 	return (error);
1475 }
1476 
1477 extern struct lock ufs_hashlock;
1478 
1479 /*
1480  * Look up an LFS dinode number to find its incore vnode.  If not already
1481  * in core, read it in from the specified device.  Return the inode locked.
1482  * Detection and handling of mount points must be done by the calling routine.
1483  */
1484 int
1485 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1486 {
1487 	struct lfs *fs;
1488 	struct ufs1_dinode *dip;
1489 	struct inode *ip;
1490 	struct buf *bp;
1491 	struct ifile *ifp;
1492 	struct vnode *vp;
1493 	struct ufsmount *ump;
1494 	daddr_t daddr;
1495 	dev_t dev;
1496 	int error, retries;
1497 	struct timespec ts;
1498 
1499 	ump = VFSTOUFS(mp);
1500 	dev = ump->um_dev;
1501 	fs = ump->um_lfs;
1502 
1503 	/*
1504 	 * If the filesystem is not completely mounted yet, suspend
1505 	 * any access requests (wait for roll-forward to complete).
1506 	 */
1507 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1508 		tsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0);
1509 
1510 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1511 		return (0);
1512 
1513 	if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) {
1514 		*vpp = NULL;
1515 		 return (error);
1516 	}
1517 
1518 	do {
1519 		if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) {
1520 			ungetnewvnode(vp);
1521 			return (0);
1522 		}
1523 	} while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0));
1524 
1525 	/* Translate the inode number to a disk address. */
1526 	if (ino == LFS_IFILE_INUM)
1527 		daddr = fs->lfs_idaddr;
1528 	else {
1529 		/* XXX bounds-check this too */
1530 		LFS_IENTRY(ifp, fs, ino, bp);
1531 		daddr = ifp->if_daddr;
1532 		if (fs->lfs_version > 1) {
1533 			ts.tv_sec = ifp->if_atime_sec;
1534 			ts.tv_nsec = ifp->if_atime_nsec;
1535 		}
1536 
1537 		brelse(bp);
1538 		if (daddr == LFS_UNUSED_DADDR) {
1539 			*vpp = NULLVP;
1540 			ungetnewvnode(vp);
1541 			lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1542 			return (ENOENT);
1543 		}
1544 	}
1545 
1546 	/* Allocate/init new vnode/inode. */
1547 	lfs_vcreate(mp, ino, vp);
1548 
1549 	/*
1550 	 * Put it onto its hash chain and lock it so that other requests for
1551 	 * this inode will block if they arrive while we are sleeping waiting
1552 	 * for old data structures to be purged or for the contents of the
1553 	 * disk portion of this inode to be read.
1554 	 */
1555 	ip = VTOI(vp);
1556 	ufs_ihashins(ip);
1557 	lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1558 
1559 	/*
1560 	 * XXX
1561 	 * This may not need to be here, logically it should go down with
1562 	 * the i_devvp initialization.
1563 	 * Ask Kirk.
1564 	 */
1565 	ip->i_lfs = ump->um_lfs;
1566 
1567 	/* Read in the disk contents for the inode, copy into the inode. */
1568 	retries = 0;
1569     again:
1570 	error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1571 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1572 		NOCRED, &bp);
1573 	if (error) {
1574 		/*
1575 		 * The inode does not contain anything useful, so it would
1576 		 * be misleading to leave it on its hash chain. With mode
1577 		 * still zero, it will be unlinked and returned to the free
1578 		 * list by vput().
1579 		 */
1580 		vput(vp);
1581 		brelse(bp);
1582 		*vpp = NULL;
1583 		return (error);
1584 	}
1585 
1586 	dip = lfs_ifind(fs, ino, bp);
1587 	if (dip == NULL) {
1588 		/* Assume write has not completed yet; try again */
1589 		bp->b_flags |= B_INVAL;
1590 		brelse(bp);
1591 		++retries;
1592 		if (retries > LFS_IFIND_RETRIES) {
1593 #ifdef DEBUG
1594 			/* If the seglock is held look at the bpp to see
1595 			   what is there anyway */
1596 			if (fs->lfs_seglock > 0) {
1597 				struct buf **bpp;
1598 				struct ufs1_dinode *dp;
1599 				int i;
1600 
1601 				for (bpp = fs->lfs_sp->bpp;
1602 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1603 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1604 					    bpp != fs->lfs_sp->bpp) {
1605 						/* Inode block */
1606 						printf("block 0x%" PRIx64 ": ",
1607 						    (*bpp)->b_blkno);
1608 						dp = (struct ufs1_dinode *)(*bpp)->b_data;
1609 						for (i = 0; i < INOPB(fs); i++)
1610 							if (dp[i].di_u.inumber)
1611 								printf("%d ", dp[i].di_u.inumber);
1612 						printf("\n");
1613 					}
1614 				}
1615 			}
1616 #endif
1617 			panic("lfs_vget: dinode not found");
1618 		}
1619 		printf("lfs_vget: dinode %d not found, retrying...\n", ino);
1620 		(void)tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs ifind", 1);
1621 		goto again;
1622 	}
1623 	*ip->i_din.ffs1_din = *dip;
1624 	brelse(bp);
1625 
1626 	if (fs->lfs_version > 1) {
1627 		ip->i_ffs1_atime = ts.tv_sec;
1628 		ip->i_ffs1_atimensec = ts.tv_nsec;
1629 	}
1630 
1631 	lfs_vinit(mp, &vp);
1632 
1633 	*vpp = vp;
1634 
1635 	KASSERT(VOP_ISLOCKED(vp));
1636 
1637 	return (0);
1638 }
1639 
1640 /*
1641  * File handle to vnode
1642  */
1643 int
1644 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1645 {
1646 	struct lfid *lfhp;
1647 	struct buf *bp;
1648 	IFILE *ifp;
1649 	int32_t daddr;
1650 	struct lfs *fs;
1651 
1652 	lfhp = (struct lfid *)fhp;
1653 	if (lfhp->lfid_ino < LFS_IFILE_INUM)
1654 		return ESTALE;
1655 
1656 	fs = VFSTOUFS(mp)->um_lfs;
1657 	if (lfhp->lfid_ident != fs->lfs_ident)
1658 		return ESTALE;
1659 
1660 	if (lfhp->lfid_ino >
1661 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1662 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1663 		return ESTALE;
1664 
1665 	if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfhp->lfid_ino) == NULLVP) {
1666 		LFS_IENTRY(ifp, fs, lfhp->lfid_ino, bp);
1667 		daddr = ifp->if_daddr;
1668 		brelse(bp);
1669 		if (daddr == LFS_UNUSED_DADDR)
1670 			return ESTALE;
1671 	}
1672 
1673 	return (ufs_fhtovp(mp, &lfhp->lfid_ufid, vpp));
1674 }
1675 
1676 /*
1677  * Vnode pointer to File handle
1678  */
1679 /* ARGSUSED */
1680 int
1681 lfs_vptofh(struct vnode *vp, struct fid *fhp)
1682 {
1683 	struct inode *ip;
1684 	struct lfid *lfhp;
1685 
1686 	ip = VTOI(vp);
1687 	lfhp = (struct lfid *)fhp;
1688 	lfhp->lfid_len = sizeof(struct lfid);
1689 	lfhp->lfid_ino = ip->i_number;
1690 	lfhp->lfid_gen = ip->i_gen;
1691 	lfhp->lfid_ident = ip->i_lfs->lfs_ident;
1692 	return (0);
1693 }
1694 
1695 static int
1696 sysctl_lfs_dostats(SYSCTLFN_ARGS)
1697 {
1698 	extern struct lfs_stats lfs_stats;
1699 	extern int lfs_dostats;
1700 	int error;
1701 
1702 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
1703 	if (error || newp == NULL)
1704 		return (error);
1705 
1706 	if (lfs_dostats == 0)
1707 		memset(&lfs_stats,0,sizeof(lfs_stats));
1708 
1709 	return (0);
1710 }
1711 
1712 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup")
1713 {
1714 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead;
1715 
1716 	sysctl_createv(clog, 0, NULL, NULL,
1717 		       CTLFLAG_PERMANENT,
1718 		       CTLTYPE_NODE, "vfs", NULL,
1719 		       NULL, 0, NULL, 0,
1720 		       CTL_VFS, CTL_EOL);
1721 	sysctl_createv(clog, 0, NULL, NULL,
1722 		       CTLFLAG_PERMANENT,
1723 		       CTLTYPE_NODE, "lfs",
1724 		       SYSCTL_DESCR("Log-structured file system"),
1725 		       NULL, 0, NULL, 0,
1726 		       CTL_VFS, 5, CTL_EOL);
1727 	/*
1728 	 * XXX the "5" above could be dynamic, thereby eliminating one
1729 	 * more instance of the "number to vfs" mapping problem, but
1730 	 * "2" is the order as taken from sys/mount.h
1731 	 */
1732 
1733 	sysctl_createv(clog, 0, NULL, NULL,
1734 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1735 		       CTLTYPE_INT, "flushindir", NULL,
1736 		       NULL, 0, &lfs_writeindir, 0,
1737 		       CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
1738 	sysctl_createv(clog, 0, NULL, NULL,
1739 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1740 		       CTLTYPE_INT, "clean_vnhead", NULL,
1741 		       NULL, 0, &lfs_clean_vnhead, 0,
1742 		       CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
1743 	sysctl_createv(clog, 0, NULL, NULL,
1744 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1745 		       CTLTYPE_INT, "dostats",
1746 		       SYSCTL_DESCR("Maintain statistics on LFS operations"),
1747 		       sysctl_lfs_dostats, 0, &lfs_dostats, 0,
1748 		       CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
1749 }
1750 
1751 /*
1752  * ufs_bmaparray callback function for writing.
1753  *
1754  * Since blocks will be written to the new segment anyway,
1755  * we don't care about current daddr of them.
1756  */
1757 static boolean_t
1758 lfs_issequential_hole(const struct ufsmount *ump,
1759     daddr_t daddr0, daddr_t daddr1)
1760 {
1761 
1762 	KASSERT(daddr0 == UNWRITTEN ||
1763 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1764 	KASSERT(daddr1 == UNWRITTEN ||
1765 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1766 
1767 	/* NOTE: all we want to know here is 'hole or not'. */
1768 	/* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */
1769 
1770 	/*
1771 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1772 	 */
1773 	if (daddr0 != 0 && daddr1 != 0)
1774 		return TRUE;
1775 
1776 	/*
1777 	 * both are in hole?
1778 	 */
1779 	if (daddr0 == 0 && daddr1 == 0)
1780 		return TRUE; /* all holes are 'contiguous' for us. */
1781 
1782 	return FALSE;
1783 }
1784 
1785 /*
1786  * lfs_gop_write functions exactly like genfs_gop_write, except that
1787  * (1) it requires the seglock to be held by its caller, and sp->fip
1788  *     to be properly initialized (it will return without re-initializing
1789  *     sp->fip, and without calling lfs_writeseg).
1790  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1791  *     to determine how large a block it can write at once (though it does
1792  *     still use VOP_BMAP to find holes in the file);
1793  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1794  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1795  *     now have clusters of clusters, ick.)
1796  */
1797 static int
1798 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
1799 {
1800 	int i, s, error, run;
1801 	int fs_bshift;
1802 	vaddr_t kva;
1803 	off_t eof, offset, startoffset;
1804 	size_t bytes, iobytes, skipbytes;
1805 	daddr_t lbn, blkno;
1806 	struct vm_page *pg;
1807 	struct buf *mbp, *bp;
1808 	struct vnode *devvp = VTOI(vp)->i_devvp;
1809 	struct inode *ip = VTOI(vp);
1810 	struct lfs *fs = ip->i_lfs;
1811 	struct segment *sp = fs->lfs_sp;
1812 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1813 
1814 	/* The Ifile lives in the buffer cache */
1815 	KASSERT(vp != fs->lfs_ivnode);
1816 
1817 	/*
1818 	 * Sometimes things slip past the filters in lfs_putpages,
1819 	 * and the pagedaemon tries to write pages---problem is
1820 	 * that the pagedaemon never acquires the segment lock.
1821 	 *
1822 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1823 	 * queue under the hypothesis that they couldn't have got here
1824 	 * unless they were modified *quite* recently.
1825 	 *
1826 	 * XXXUBC that last statement is an oversimplification of course.
1827 	 */
1828 	if (!(fs->lfs_seglock) || fs->lfs_lockpid != curproc->p_pid) {
1829 		simple_lock(&vp->v_interlock);
1830 #ifdef DEBUG
1831 		printf("lfs_gop_write: seglock not held\n");
1832 #endif
1833 		uvm_lock_pageq();
1834 		for (i = 0; i < npages; i++) {
1835 			pg = pgs[i];
1836 
1837 			if (pg->flags & PG_PAGEOUT)
1838 				uvmexp.paging--;
1839 			if (pg->flags & PG_DELWRI) {
1840 				uvm_pageunwire(pg);
1841 			}
1842 			uvm_pageactivate(pg);
1843 			pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
1844 #ifdef DEBUG_LFS
1845 			printf("pg[%d]->flags = %x\n", i, pg->flags);
1846 			printf("pg[%d]->pqflags = %x\n", i, pg->pqflags);
1847 			printf("pg[%d]->uanon = %p\n", i, pg->uanon);
1848 			printf("pg[%d]->uobject = %p\n", i, pg->uobject);
1849 			printf("pg[%d]->wire_count = %d\n", i, pg->wire_count);
1850 			printf("pg[%d]->loan_count = %d\n", i, pg->loan_count);
1851 #endif
1852 		}
1853 		/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
1854 		uvm_page_unbusy(pgs, npages);
1855 		uvm_unlock_pageq();
1856 		simple_unlock(&vp->v_interlock);
1857 		return EAGAIN;
1858 	}
1859 
1860 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1861 	    vp, pgs, npages, flags);
1862 
1863 	GOP_SIZE(vp, vp->v_size, &eof, GOP_SIZE_WRITE);
1864 
1865 	if (vp->v_type == VREG)
1866 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1867 	else
1868 		fs_bshift = DEV_BSHIFT;
1869 	error = 0;
1870 	pg = pgs[0];
1871 	startoffset = pg->offset;
1872 	bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1873 	skipbytes = 0;
1874 
1875 	/* KASSERT(bytes != 0); */
1876 	if (bytes == 0)
1877 		printf("ino %d bytes == 0 offset %" PRId64 "\n",
1878 			VTOI(vp)->i_number, pgs[0]->offset);
1879 
1880 	/* Swap PG_DELWRI for PG_PAGEOUT */
1881 	for (i = 0; i < npages; i++)
1882 		if (pgs[i]->flags & PG_DELWRI) {
1883 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1884 			pgs[i]->flags &= ~PG_DELWRI;
1885 			pgs[i]->flags |= PG_PAGEOUT;
1886 			uvmexp.paging++;
1887 			uvm_lock_pageq();
1888 			uvm_pageunwire(pgs[i]);
1889 			uvm_unlock_pageq();
1890 		}
1891 
1892 	/*
1893 	 * Check to make sure we're starting on a block boundary.
1894 	 * We'll check later to make sure we always write entire
1895 	 * blocks (or fragments).
1896 	 */
1897 	if (startoffset & fs->lfs_bmask)
1898 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1899 			startoffset, fs->lfs_bmask,
1900 			startoffset & fs->lfs_bmask);
1901 	KASSERT((startoffset & fs->lfs_bmask) == 0);
1902 	if (bytes & fs->lfs_ffmask) {
1903 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1904 		panic("lfs_gop_write: non-integer blocks");
1905 	}
1906 
1907 	kva = uvm_pagermapin(pgs, npages,
1908 	    UVMPAGER_MAPIN_WRITE | UVMPAGER_MAPIN_WAITOK);
1909 
1910 	s = splbio();
1911 	simple_lock(&global_v_numoutput_slock);
1912 	vp->v_numoutput += 2; /* one for biodone, one for aiodone */
1913 	simple_unlock(&global_v_numoutput_slock);
1914 	mbp = pool_get(&bufpool, PR_WAITOK);
1915 	splx(s);
1916 
1917 	memset(mbp, 0, sizeof(*bp));
1918 	BUF_INIT(mbp);
1919 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1920 	    vp, mbp, vp->v_numoutput, bytes);
1921 	mbp->b_bufsize = npages << PAGE_SHIFT;
1922 	mbp->b_data = (void *)kva;
1923 	mbp->b_resid = mbp->b_bcount = bytes;
1924 	mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL;
1925 	mbp->b_iodone = uvm_aio_biodone;
1926 	mbp->b_vp = vp;
1927 
1928 	bp = NULL;
1929 	for (offset = startoffset;
1930 	    bytes > 0;
1931 	    offset += iobytes, bytes -= iobytes) {
1932 		lbn = offset >> fs_bshift;
1933 		error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1934 		    lfs_issequential_hole);
1935 		if (error) {
1936 			UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
1937 			    error,0,0,0);
1938 			skipbytes += bytes;
1939 			bytes = 0;
1940 			break;
1941 		}
1942 
1943 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1944 		    bytes);
1945 		if (blkno == (daddr_t)-1) {
1946 			skipbytes += iobytes;
1947 			continue;
1948 		}
1949 
1950 		/*
1951 		 * Discover how much we can really pack into this buffer.
1952 		 */
1953 		/* If no room in the current segment, finish it up */
1954 		if (sp->sum_bytes_left < sizeof(int32_t) ||
1955 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1956 			int version;
1957 
1958 			lfs_updatemeta(sp);
1959 
1960 			version = sp->fip->fi_version;
1961 			(void) lfs_writeseg(fs, sp);
1962 
1963 			sp->fip->fi_version = version;
1964 			sp->fip->fi_ino = ip->i_number;
1965 			/* Add the current file to the segment summary. */
1966 			++((SEGSUM *)(sp->segsum))->ss_nfinfo;
1967 			sp->sum_bytes_left -= FINFOSIZE;
1968 		}
1969 		/* Check both for space in segment and space in segsum */
1970 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1971 					<< fs_bshift);
1972 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1973 				       << fs_bshift);
1974 		KASSERT(iobytes > 0);
1975 
1976 		/* if it's really one i/o, don't make a second buf */
1977 		if (offset == startoffset && iobytes == bytes) {
1978 			bp = mbp;
1979 			/* printf("bp is mbp\n"); */
1980 			/* correct overcount if there is no second buffer */
1981 			s = splbio();
1982 			simple_lock(&global_v_numoutput_slock);
1983 			--vp->v_numoutput;
1984 			simple_unlock(&global_v_numoutput_slock);
1985 			splx(s);
1986 		} else {
1987 			/* printf("bp is not mbp\n"); */
1988 			s = splbio();
1989 			bp = pool_get(&bufpool, PR_WAITOK);
1990 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1991 			    vp, bp, vp->v_numoutput, 0);
1992 			splx(s);
1993 			memset(bp, 0, sizeof(*bp));
1994 			BUF_INIT(bp);
1995 			bp->b_data = (char *)kva +
1996 			    (vaddr_t)(offset - pg->offset);
1997 			bp->b_resid = bp->b_bcount = iobytes;
1998 			bp->b_flags = B_BUSY|B_WRITE|B_CALL;
1999 			bp->b_iodone = uvm_aio_biodone1;
2000 		}
2001 
2002 		/* XXX This is silly ... is this necessary? */
2003 		bp->b_vp = NULL;
2004 		s = splbio();
2005 		bgetvp(vp, bp);
2006 		splx(s);
2007 
2008 		bp->b_lblkno = lblkno(fs, offset);
2009 		bp->b_private = mbp;
2010 		if (devvp->v_type == VBLK) {
2011 			bp->b_dev = devvp->v_rdev;
2012 		}
2013 		VOP_BWRITE(bp);
2014 		while (lfs_gatherblock(sp, bp, NULL))
2015 			continue;
2016 	}
2017 
2018 	if (skipbytes) {
2019 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
2020 		s = splbio();
2021 		if (error) {
2022 			mbp->b_flags |= B_ERROR;
2023 			mbp->b_error = error;
2024 		}
2025 		mbp->b_resid -= skipbytes;
2026 		if (mbp->b_resid == 0) {
2027 			biodone(mbp);
2028 		}
2029 		splx(s);
2030 	}
2031 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
2032 	return (0);
2033 }
2034 
2035 /*
2036  * finish vnode/inode initialization.
2037  * used by lfs_vget and lfs_fastvget.
2038  */
2039 void
2040 lfs_vinit(struct mount *mp, struct vnode **vpp)
2041 {
2042 	struct vnode *vp = *vpp;
2043 	struct inode *ip = VTOI(vp);
2044 	struct ufsmount *ump = VFSTOUFS(mp);
2045 	int i;
2046 
2047 	ip->i_mode = ip->i_ffs1_mode;
2048 	ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink;
2049 	ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
2050 	ip->i_flags = ip->i_ffs1_flags;
2051 	ip->i_gen = ip->i_ffs1_gen;
2052 	ip->i_uid = ip->i_ffs1_uid;
2053 	ip->i_gid = ip->i_ffs1_gid;
2054 
2055 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
2056 
2057 	/*
2058 	 * Initialize the vnode from the inode, check for aliases.  In all
2059 	 * cases re-init ip, the underlying vnode/inode may have changed.
2060 	 */
2061 	ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2062 
2063 	memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
2064 	if (vp->v_type != VLNK ||
2065 	    VTOI(vp)->i_size >= vp->v_mount->mnt_maxsymlinklen) {
2066 		struct lfs *fs = ump->um_lfs;
2067 #ifdef DEBUG
2068 		for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
2069 		    i < NDADDR; i++) {
2070 			if (ip->i_ffs1_db[i] != 0) {
2071 inconsistent:
2072 				lfs_dump_dinode(ip->i_din.ffs1_din);
2073 				panic("inconsistent inode");
2074 			}
2075 		}
2076 		for ( ; i < NDADDR + NIADDR; i++) {
2077 			if (ip->i_ffs1_ib[i - NDADDR] != 0) {
2078 				goto inconsistent;
2079 			}
2080 		}
2081 #endif /* DEBUG */
2082 		for (i = 0; i < NDADDR; i++)
2083 			if (ip->i_ffs1_db[i] != 0)
2084 				ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
2085 	}
2086 
2087 #ifdef DEBUG
2088 	if (vp->v_type == VNON) {
2089 		printf("lfs_vinit: ino %d is type VNON! (ifmt=%o)\n",
2090 		       ip->i_number, (ip->i_mode & IFMT) >> 12);
2091 		lfs_dump_dinode(ip->i_din.ffs1_din);
2092 #ifdef DDB
2093 		Debugger();
2094 #endif /* DDB */
2095 	}
2096 #endif /* DEBUG */
2097 
2098 	/*
2099 	 * Finish inode initialization now that aliasing has been resolved.
2100 	 */
2101 
2102 	ip->i_devvp = ump->um_devvp;
2103 	VREF(ip->i_devvp);
2104 	genfs_node_init(vp, &lfs_genfsops);
2105 	uvm_vnp_setsize(vp, ip->i_size);
2106 
2107 	*vpp = vp;
2108 }
2109