xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: lfs_vfsops.c,v 1.192 2006/01/14 17:41:17 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 /*-
39  * Copyright (c) 1989, 1991, 1993, 1994
40  *	The Regents of the University of California.  All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. 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 1.192 2006/01/14 17:41:17 yamt 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 <sys/syslog.h>
95 #include <uvm/uvm_extern.h>
96 #include <sys/sysctl.h>
97 #include <sys/conf.h>
98 
99 #include <miscfs/specfs/specdev.h>
100 
101 #include <ufs/ufs/quota.h>
102 #include <ufs/ufs/inode.h>
103 #include <ufs/ufs/ufsmount.h>
104 #include <ufs/ufs/ufs_extern.h>
105 
106 #include <uvm/uvm.h>
107 #include <uvm/uvm_stat.h>
108 #include <uvm/uvm_pager.h>
109 #include <uvm/uvm_pdaemon.h>
110 
111 #include <ufs/lfs/lfs.h>
112 #include <ufs/lfs/lfs_extern.h>
113 
114 #include <miscfs/genfs/genfs.h>
115 #include <miscfs/genfs/genfs_node.h>
116 
117 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
118 static boolean_t lfs_issequential_hole(const struct ufsmount *,
119     daddr_t, daddr_t);
120 
121 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *);
122 static void warn_ifile_size(struct lfs *);
123 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t,
124     struct ucred *, int, int *, struct lwp *);
125 
126 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
127 extern const struct vnodeopv_desc lfs_specop_opv_desc;
128 extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
129 
130 pid_t lfs_writer_daemon = 0;
131 int lfs_do_flush = 0;
132 int lfs_do_rfw = 0;
133 
134 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
135 	&lfs_vnodeop_opv_desc,
136 	&lfs_specop_opv_desc,
137 	&lfs_fifoop_opv_desc,
138 	NULL,
139 };
140 
141 struct vfsops lfs_vfsops = {
142 	MOUNT_LFS,
143 	lfs_mount,
144 	ufs_start,
145 	lfs_unmount,
146 	ufs_root,
147 	ufs_quotactl,
148 	lfs_statvfs,
149 	lfs_sync,
150 	lfs_vget,
151 	lfs_fhtovp,
152 	lfs_vptofh,
153 	lfs_init,
154 	lfs_reinit,
155 	lfs_done,
156 	lfs_mountroot,
157 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
158 	vfs_stdextattrctl,
159 	lfs_vnodeopv_descs,
160 };
161 VFS_ATTACH(lfs_vfsops);
162 
163 const struct genfs_ops lfs_genfsops = {
164 	.gop_size = lfs_gop_size,
165 	.gop_alloc = ufs_gop_alloc,
166 	.gop_write = lfs_gop_write,
167 	.gop_markupdate = ufs_gop_markupdate,
168 };
169 
170 static const struct ufs_ops lfs_ufsops = {
171 	.uo_itimes = NULL,
172 	.uo_update = lfs_update,
173 	.uo_truncate = lfs_truncate,
174 	.uo_valloc = lfs_valloc,
175 	.uo_vfree = lfs_vfree,
176 	.uo_balloc = lfs_balloc,
177 };
178 
179 /*
180  * XXX Same structure as FFS inodes?  Should we share a common pool?
181  */
182 POOL_INIT(lfs_inode_pool, sizeof(struct inode), 0, 0, 0, "lfsinopl",
183     &pool_allocator_nointr);
184 POOL_INIT(lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, "lfsdinopl",
185     &pool_allocator_nointr);
186 POOL_INIT(lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, "lfsinoextpl",
187     &pool_allocator_nointr);
188 POOL_INIT(lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, "lfslbnpool",
189     &pool_allocator_nointr);
190 
191 /*
192  * The writer daemon.  UVM keeps track of how many dirty pages we are holding
193  * in lfs_subsys_pages; the daemon flushes the filesystem when this value
194  * crosses the (user-defined) threshhold LFS_MAX_PAGES.
195  */
196 static void
197 lfs_writerd(void *arg)
198 {
199 	struct mount *mp, *nmp;
200 	struct lfs *fs;
201 	int loopcount;
202 
203 	lfs_writer_daemon = curproc->p_pid;
204 
205 	simple_lock(&lfs_subsys_lock);
206 	for (;;) {
207 		ltsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10,
208 		    &lfs_subsys_lock);
209 
210 		/*
211 		 * Look through the list of LFSs to see if any of them
212 		 * have requested pageouts.
213 		 */
214 		simple_lock(&mountlist_slock);
215 		for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
216 		     mp = nmp) {
217 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
218 				nmp = CIRCLEQ_NEXT(mp, mnt_list);
219 				continue;
220 			}
221 			if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS,
222 				    MFSNAMELEN) == 0) {
223 				fs = VFSTOUFS(mp)->um_lfs;
224 				simple_lock(&fs->lfs_interlock);
225 				if (fs->lfs_pdflush ||
226 				    !TAILQ_EMPTY(&fs->lfs_pchainhd)) {
227 					DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n"));
228 					fs->lfs_pdflush = 0;
229 					lfs_flush_fs(fs, 0);
230 				}
231 				simple_unlock(&fs->lfs_interlock);
232 			}
233 
234 			simple_lock(&mountlist_slock);
235 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
236 			vfs_unbusy(mp);
237 		}
238 		simple_unlock(&mountlist_slock);
239 
240 		/*
241 		 * If global state wants a flush, flush everything.
242 		 */
243 		simple_lock(&lfs_subsys_lock);
244 		loopcount = 0;
245 		if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
246 			locked_queue_bytes > LFS_MAX_BYTES ||
247 			lfs_subsys_pages > LFS_MAX_PAGES) {
248 
249 			if (lfs_do_flush)
250 				DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n"));
251 			if (locked_queue_count > LFS_MAX_BUFS)
252 				DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n",
253 				      locked_queue_count, LFS_MAX_BUFS));
254 			if (locked_queue_bytes > LFS_MAX_BYTES)
255 				DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n",
256 				      locked_queue_bytes, LFS_MAX_BYTES));
257 			if (lfs_subsys_pages > LFS_MAX_PAGES)
258 				DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n",
259 				      lfs_subsys_pages, LFS_MAX_PAGES));
260 
261 			lfs_flush(NULL, SEGM_WRITERD, 0);
262 			lfs_do_flush = 0;
263 		}
264 	}
265 	/* NOTREACHED */
266 }
267 
268 /*
269  * Initialize the filesystem, most work done by ufs_init.
270  */
271 void
272 lfs_init()
273 {
274 #ifdef _LKM
275 	malloc_type_attach(M_SEGMENT);
276 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
277 	    "lfsinopl", &pool_allocator_nointr);
278 	pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
279 	    "lfsdinopl", &pool_allocator_nointr);
280 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
281 	    "lfsinoextpl", &pool_allocator_nointr);
282 	pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0,
283 	    "lfslbnpool", &pool_allocator_nointr);
284 #endif
285 	ufs_init();
286 
287 #ifdef DEBUG
288 	memset(lfs_log, 0, sizeof(lfs_log));
289 #endif
290 	simple_lock_init(&lfs_subsys_lock);
291 }
292 
293 void
294 lfs_reinit()
295 {
296 	ufs_reinit();
297 }
298 
299 void
300 lfs_done()
301 {
302 	ufs_done();
303 #ifdef _LKM
304 	pool_destroy(&lfs_inode_pool);
305 	pool_destroy(&lfs_dinode_pool);
306 	pool_destroy(&lfs_inoext_pool);
307 	pool_destroy(&lfs_lbnentry_pool);
308 	malloc_type_detach(M_SEGMENT);
309 #endif
310 }
311 
312 /*
313  * Called by main() when ufs is going to be mounted as root.
314  */
315 int
316 lfs_mountroot()
317 {
318 	extern struct vnode *rootvp;
319 	struct mount *mp;
320 	struct lwp *l = curlwp;	/* XXX */
321 	int error;
322 
323 	if (root_device->dv_class != DV_DISK)
324 		return (ENODEV);
325 
326 	if (rootdev == NODEV)
327 		return (ENODEV);
328 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
329 		vrele(rootvp);
330 		return (error);
331 	}
332 	if ((error = lfs_mountfs(rootvp, mp, l))) {
333 		mp->mnt_op->vfs_refcount--;
334 		vfs_unbusy(mp);
335 		free(mp, M_MOUNT);
336 		return (error);
337 	}
338 	simple_lock(&mountlist_slock);
339 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
340 	simple_unlock(&mountlist_slock);
341 	(void)lfs_statvfs(mp, &mp->mnt_stat, l);
342 	vfs_unbusy(mp);
343 	setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp));
344 	return (0);
345 }
346 
347 /*
348  * VFS Operations.
349  *
350  * mount system call
351  */
352 int
353 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct lwp *l)
354 {
355 	struct vnode *devvp;
356 	struct ufs_args args;
357 	struct ufsmount *ump = NULL;
358 	struct lfs *fs = NULL;				/* LFS */
359 	struct proc *p;
360 	int error, update;
361 	mode_t accessmode;
362 
363 	p = l->l_proc;
364 	if (mp->mnt_flag & MNT_GETARGS) {
365 		ump = VFSTOUFS(mp);
366 		if (ump == NULL)
367 			return EIO;
368 		args.fspec = NULL;
369 		return copyout(&args, data, sizeof(args));
370 	}
371 	error = copyin(data, &args, sizeof (struct ufs_args));
372 	if (error)
373 		return (error);
374 
375 	update = mp->mnt_flag & MNT_UPDATE;
376 
377 	/* Check arguments */
378 	if (args.fspec != NULL) {
379 		/*
380 		 * Look up the name and verify that it's sane.
381 		 */
382 		NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, l);
383 		if ((error = namei(ndp)) != 0)
384 			return (error);
385 		devvp = ndp->ni_vp;
386 
387 		if (!update) {
388 			/*
389 			 * Be sure this is a valid block device
390 			 */
391 			if (devvp->v_type != VBLK)
392 				error = ENOTBLK;
393 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
394 				error = ENXIO;
395 		} else {
396 			/*
397 			 * Be sure we're still naming the same device
398 			 * used for our initial mount
399 			 */
400 			ump = VFSTOUFS(mp);
401 			if (devvp != ump->um_devvp)
402 				error = EINVAL;
403 		}
404 	} else {
405 		if (!update) {
406 			/* New mounts must have a filename for the device */
407 			return (EINVAL);
408 		} else {
409 			/* Use the extant mount */
410 			ump = VFSTOUFS(mp);
411 			devvp = ump->um_devvp;
412 			vref(devvp);
413 		}
414 	}
415 
416 
417 	/*
418 	 * If mount by non-root, then verify that user has necessary
419 	 * permissions on the device.
420 	 */
421 	if (error == 0 && p->p_ucred->cr_uid != 0) {
422 		accessmode = VREAD;
423 		if (update ?
424 		    (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
425 		    (mp->mnt_flag & MNT_RDONLY) == 0)
426 			accessmode |= VWRITE;
427 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
428 		error = VOP_ACCESS(devvp, accessmode, p->p_ucred, l);
429 		VOP_UNLOCK(devvp, 0);
430 	}
431 
432 	if (error) {
433 		vrele(devvp);
434 		return (error);
435 	}
436 
437 	if (!update) {
438 		int flags;
439 
440 		/*
441 		 * Disallow multiple mounts of the same device.
442 		 * Disallow mounting of a device that is currently in use
443 		 * (except for root, which might share swap device for
444 		 * miniroot).
445 		 */
446 		error = vfs_mountedon(devvp);
447 		if (error)
448 			goto fail;
449 		if (vcount(devvp) > 1 && devvp != rootvp) {
450 			error = EBUSY;
451 			goto fail;
452 		}
453 		if (mp->mnt_flag & MNT_RDONLY)
454 			flags = FREAD;
455 		else
456 			flags = FREAD|FWRITE;
457 		error = VOP_OPEN(devvp, flags, FSCRED, l);
458 		if (error)
459 			goto fail;
460 		error = lfs_mountfs(devvp, mp, l);		/* LFS */
461 		if (error) {
462 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
463 			(void)VOP_CLOSE(devvp, flags, NOCRED, l);
464 			VOP_UNLOCK(devvp, 0);
465 			goto fail;
466 		}
467 
468 		ump = VFSTOUFS(mp);
469 		fs = ump->um_lfs;
470 	} else {
471 		/*
472 		 * Update the mount.
473 		 */
474 
475 		/*
476 		 * The initial mount got a reference on this
477 		 * device, so drop the one obtained via
478 		 * namei(), above.
479 		 */
480 		vrele(devvp);
481 
482 		ump = VFSTOUFS(mp);
483 		fs = ump->um_lfs;
484 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
485 			/*
486 			 * Changing from read-only to read/write
487 			 */
488 			fs->lfs_ronly = 0;
489 		}
490 		if (args.fspec == NULL)
491 			return EINVAL;
492 	}
493 
494 	error = set_statvfs_info(path, UIO_USERSPACE, args.fspec,
495 	    UIO_USERSPACE, mp, l);
496 	if (error == 0)
497 		(void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname,
498 			      sizeof(fs->lfs_fsmnt));
499 	return error;
500 
501 fail:
502 	vrele(devvp);
503 	return (error);
504 }
505 
506 /*
507  * Roll-forward code.
508  */
509 
510 /*
511  * Load the appropriate indirect block, and change the appropriate pointer.
512  * Mark the block dirty.  Do segment and avail accounting.
513  */
514 static int
515 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn,
516 	    daddr_t ndaddr, size_t size, struct lwp *l)
517 {
518 	int error;
519 	struct vnode *vp;
520 	struct inode *ip;
521 #ifdef DEBUG
522 	daddr_t odaddr;
523 	struct indir a[NIADDR];
524 	int num;
525 	int i;
526 #endif /* DEBUG */
527 	struct buf *bp;
528 	SEGUSE *sup;
529 
530 	KASSERT(lbn >= 0);	/* no indirect blocks */
531 
532 	if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) {
533 		DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc"
534 		      " returned %d\n", ino, error));
535 		return error;
536 	}
537 
538 	if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size,
539 				NOCRED, 0, &bp)) != 0) {
540 		vput(vp);
541 		return (error);
542 	}
543 	/* No need to write, the block is already on disk */
544 	if (bp->b_flags & B_DELWRI) {
545 		LFS_UNLOCK_BUF(bp);
546 		fs->lfs_avail += btofsb(fs, bp->b_bcount);
547 	}
548 	bp->b_flags |= B_INVAL;
549 	brelse(bp);
550 
551 	/*
552 	 * Extend the file, if it is not large enough already.
553 	 * XXX this is not exactly right, we don't know how much of the
554 	 * XXX last block is actually used.  We hope that an inode will
555 	 * XXX appear later to give the correct size.
556 	 */
557 	ip = VTOI(vp);
558 	if (ip->i_size <= (lbn << fs->lfs_bshift)) {
559 		u_int64_t newsize;
560 
561 		if (lbn < NDADDR)
562 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) +
563 				(size - fs->lfs_fsize) + 1;
564 		else
565 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1;
566 
567 		if (ip->i_size < newsize) {
568 			ip->i_size = newsize;
569 			/*
570 			 * tell vm our new size for the case the inode won't
571 			 * appear later.
572 			 */
573 			uvm_vnp_setsize(vp, newsize);
574 		}
575 	}
576 
577 	lfs_update_single(fs, NULL, vp, lbn, ndaddr, size);
578 
579 	LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
580 	sup->su_nbytes += size;
581 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
582 
583 	/* differences here should be due to UNWRITTEN indirect blocks. */
584 	KASSERT((lblkno(fs, ip->i_size) > NDADDR &&
585 	    ip->i_lfs_effnblks == ip->i_ffs1_blocks) ||
586 	    ip->i_lfs_effnblks >= ip->i_ffs1_blocks);
587 
588 #ifdef DEBUG
589 	/* Now look again to make sure it worked */
590 	ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
591 	for (i = num; i > 0; i--) {
592 		if (!a[i].in_exists)
593 			panic("update_meta: absent %d lv indirect block", i);
594 	}
595 	if (dbtofsb(fs, odaddr) != ndaddr)
596 		DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %"
597 		      PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr));
598 #endif /* DEBUG */
599 	vput(vp);
600 	return 0;
601 }
602 
603 static int
604 update_inoblk(struct lfs *fs, daddr_t offset, struct ucred *cred,
605 	      struct lwp *l)
606 {
607 	struct vnode *devvp, *vp;
608 	struct inode *ip;
609 	struct ufs1_dinode *dip;
610 	struct buf *dbp, *ibp;
611 	int error;
612 	daddr_t daddr;
613 	IFILE *ifp;
614 	SEGUSE *sup;
615 
616 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
617 
618 	/*
619 	 * Get the inode, update times and perms.
620 	 * DO NOT update disk blocks, we do that separately.
621 	 */
622 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp);
623 	if (error) {
624 		DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error));
625 		return error;
626 	}
627 	dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs);
628 	while (--dip >= (struct ufs1_dinode *)dbp->b_data) {
629 		if (dip->di_inumber > LFS_IFILE_INUM) {
630 			error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
631 					      l, &vp);
632 			if (error) {
633 				DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc"
634 				      " returned %d\n", error));
635 				continue;
636 			}
637 			ip = VTOI(vp);
638 			if (dip->di_size != ip->i_size)
639 				lfs_truncate(vp, dip->di_size, 0, NOCRED, l);
640 			/* Get mode, link count, size, and times */
641 			memcpy(ip->i_din.ffs1_din, dip,
642 			       offsetof(struct ufs1_dinode, di_db[0]));
643 
644 			/* Then the rest, except di_blocks */
645 			ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
646 			ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
647 			ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
648 			ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
649 
650 			ip->i_mode = ip->i_ffs1_mode;
651 			ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink;
652 			ip->i_size = ip->i_ffs1_size;
653 
654 			LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
655 
656 			/* Re-initialize to get type right */
657 			ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
658 				  &vp);
659 			vput(vp);
660 
661 			/* Record change in location */
662 			LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
663 			daddr = ifp->if_daddr;
664 			ifp->if_daddr = dbtofsb(fs, dbp->b_blkno);
665 			error = LFS_BWRITE_LOG(ibp); /* Ifile */
666 			/* And do segment accounting */
667 			if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) {
668 				if (daddr > 0) {
669 					LFS_SEGENTRY(sup, fs, dtosn(fs, daddr),
670 						     ibp);
671 					sup->su_nbytes -= sizeof (struct ufs1_dinode);
672 					LFS_WRITESEGENTRY(sup, fs,
673 							  dtosn(fs, daddr),
674 							  ibp);
675 				}
676 				LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
677 					     ibp);
678 				sup->su_nbytes += sizeof (struct ufs1_dinode);
679 				LFS_WRITESEGENTRY(sup, fs,
680 						  dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
681 						  ibp);
682 			}
683 		}
684 	}
685 	dbp->b_flags |= B_AGE;
686 	brelse(dbp);
687 
688 	return 0;
689 }
690 
691 #define CHECK_CKSUM   0x0001  /* Check the checksum to make sure it's valid */
692 #define CHECK_UPDATE  0x0002  /* Update Ifile for new data blocks / inodes */
693 
694 static daddr_t
695 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial,
696 	     struct ucred *cred, int flags, int *pseg_flags, struct lwp *l)
697 {
698 	struct vnode *devvp;
699 	struct buf *bp, *dbp;
700 	int error, nblocks = 0, ninos, i, j; /* XXX: gcc */
701 	SEGSUM *ssp;
702 	u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */
703 	daddr_t oldoffset;
704 	int32_t *iaddr;	/* XXX ondisk32 */
705 	FINFO *fip;
706 	SEGUSE *sup;
707 	size_t size;
708 
709 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
710 	/*
711 	 * If the segment has a superblock and we're at the top
712 	 * of the segment, skip the superblock.
713 	 */
714 	if (sntod(fs, dtosn(fs, offset)) == offset) {
715 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
716 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
717 			offset += btofsb(fs, LFS_SBPAD);
718 		brelse(bp);
719 	}
720 
721 	/* Read in the segment summary */
722 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp);
723 	if (error)
724 		return -1;
725 
726 	/* Check summary checksum */
727 	ssp = (SEGSUM *)bp->b_data;
728 	if (flags & CHECK_CKSUM) {
729 		if (ssp->ss_sumsum != cksum(&ssp->ss_datasum,
730 					   fs->lfs_sumsize -
731 					   sizeof(ssp->ss_sumsum))) {
732 			DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset));
733 			offset = -1;
734 			goto err1;
735 		}
736 		if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) {
737 			DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset));
738 			offset = -1;
739 			goto err1;
740 		}
741 		if (ssp->ss_create < fs->lfs_tstamp) {
742 			DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset));
743 			offset = -1;
744 			goto err1;
745 		}
746 	}
747 	if (fs->lfs_version > 1) {
748 		if (ssp->ss_serial != nextserial) {
749 			DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64
750 			      "\n", offset));
751 			offset = -1;
752 			goto err1;
753 		}
754 		if (ssp->ss_ident != fs->lfs_ident) {
755 			DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%"
756 			      PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset));
757 			offset = -1;
758 			goto err1;
759 		}
760 	}
761 	if (pseg_flags)
762 		*pseg_flags = ssp->ss_flags;
763 	oldoffset = offset;
764 	offset += btofsb(fs, fs->lfs_sumsize);
765 
766 	ninos = howmany(ssp->ss_ninos, INOPB(fs));
767 	/* XXX ondisk32 */
768 	iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t));
769 	if (flags & CHECK_CKSUM) {
770 		/* Count blocks */
771 		nblocks = 0;
772 		fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
773 		for (i = 0; i < ssp->ss_nfinfo; ++i) {
774 			nblocks += fip->fi_nblocks;
775 			if (fip->fi_nblocks <= 0)
776 				break;
777 			/* XXX ondisk32 */
778 			fip = (FINFO *)(((char *)fip) + FINFOSIZE +
779 					(fip->fi_nblocks * sizeof(int32_t)));
780 		}
781 		nblocks += ninos;
782 		/* Create the sum array */
783 		datap = dp = (u_long *)malloc(nblocks * sizeof(u_long),
784 					      M_SEGMENT, M_WAITOK);
785 	}
786 
787 	/* Handle individual blocks */
788 	fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
789 	for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) {
790 		/* Inode block? */
791 		if (ninos && *iaddr == offset) {
792 			if (flags & CHECK_CKSUM) {
793 				/* Read in the head and add to the buffer */
794 				error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize,
795 					      cred, &dbp);
796 				if (error) {
797 					offset = -1;
798 					goto err2;
799 				}
800 				(*dp++) = ((u_long *)(dbp->b_data))[0];
801 				dbp->b_flags |= B_AGE;
802 				brelse(dbp);
803 			}
804 			if (flags & CHECK_UPDATE) {
805 				if ((error = update_inoblk(fs, offset, cred, l))
806 				    != 0) {
807 					offset = -1;
808 					goto err2;
809 				}
810 			}
811 			offset += btofsb(fs, fs->lfs_ibsize);
812 			--iaddr;
813 			--ninos;
814 			--i; /* compensate */
815 			continue;
816 		}
817 		size = fs->lfs_bsize;
818 		for (j = 0; j < fip->fi_nblocks; ++j) {
819 			if (j == fip->fi_nblocks - 1)
820 				size = fip->fi_lastlength;
821 			if (flags & CHECK_CKSUM) {
822 				error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp);
823 				if (error) {
824 					offset = -1;
825 					goto err2;
826 				}
827 				(*dp++) = ((u_long *)(dbp->b_data))[0];
828 				dbp->b_flags |= B_AGE;
829 				brelse(dbp);
830 			}
831 			/* Account for and update any direct blocks */
832 			if ((flags & CHECK_UPDATE) &&
833 			   fip->fi_ino > LFS_IFILE_INUM &&
834 			   fip->fi_blocks[j] >= 0) {
835 				update_meta(fs, fip->fi_ino, fip->fi_version,
836 					    fip->fi_blocks[j], offset, size, l);
837 			}
838 			offset += btofsb(fs, size);
839 		}
840 		/* XXX ondisk32 */
841 		fip = (FINFO *)(((char *)fip) + FINFOSIZE
842 				+ fip->fi_nblocks * sizeof(int32_t));
843 	}
844 	/* Checksum the array, compare */
845 	if ((flags & CHECK_CKSUM) &&
846 	   ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long)))
847 	{
848 		DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64
849 		      " (wanted %x got %x)\n",
850 		      offset, ssp->ss_datasum, cksum(datap, nblocks *
851 						     sizeof(u_long))));
852 		offset = -1;
853 		goto err2;
854 	}
855 
856 	/* If we're at the end of the segment, move to the next */
857 	if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) !=
858 	   dtosn(fs, offset)) {
859 		if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) {
860 			offset = -1;
861 			goto err2;
862 		}
863 		offset = ssp->ss_next;
864 		DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64
865 		       " -> segment %d\n", offset, dtosn(fs,offset)));
866 	}
867 
868 	if (flags & CHECK_UPDATE) {
869 		fs->lfs_avail -= (offset - oldoffset);
870 		/* Don't clog the buffer queue */
871 		simple_lock(&lfs_subsys_lock);
872 		if (locked_queue_count > LFS_MAX_BUFS ||
873 		    locked_queue_bytes > LFS_MAX_BYTES) {
874 			lfs_flush(fs, SEGM_CKP, 0);
875 		}
876 		simple_unlock(&lfs_subsys_lock);
877 	}
878 
879     err2:
880 	if (flags & CHECK_CKSUM)
881 		free(datap, M_SEGMENT);
882     err1:
883 	bp->b_flags |= B_AGE;
884 	brelse(bp);
885 
886 	/* XXX should we update the serial number even for bad psegs? */
887 	if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1)
888 		fs->lfs_serial = nextserial;
889 	return offset;
890 }
891 
892 /*
893  * Common code for mount and mountroot
894  * LFS specific
895  */
896 int
897 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
898 {
899 	struct dlfs *tdfs, *dfs, *adfs;
900 	struct lfs *fs;
901 	struct ufsmount *ump;
902 	struct vnode *vp;
903 	struct buf *bp, *abp;
904 	struct partinfo dpart;
905 	struct proc *p;
906 	dev_t dev;
907 	int error, i, ronly, secsize, fsbsize;
908 	struct ucred *cred;
909 	CLEANERINFO *cip;
910 	SEGUSE *sup;
911 	int flags, dirty, do_rollforward;
912 	daddr_t offset, oldoffset, lastgoodpseg, sb_addr;
913 	int sn, curseg;
914 
915 	p = l ? l->l_proc : NULL;
916 	cred = p ? p->p_ucred : NOCRED;
917 
918 	/*
919 	 * Flush out any old buffers remaining from a previous use.
920 	 */
921 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
922 	error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
923 	VOP_UNLOCK(devvp, 0);
924 	if (error)
925 		return (error);
926 
927 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
928 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, l) != 0)
929 		secsize = DEV_BSIZE;
930 	else
931 		secsize = dpart.disklab->d_secsize;
932 
933 	/* Don't free random space on error. */
934 	bp = NULL;
935 	abp = NULL;
936 	ump = NULL;
937 
938 	sb_addr = LFS_LABELPAD / secsize;
939 	while (1) {
940 		/* Read in the superblock. */
941 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp);
942 		if (error)
943 			goto out;
944 		dfs = (struct dlfs *)bp->b_data;
945 
946 		/* Check the basics. */
947 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE ||
948 		    dfs->dlfs_version > LFS_VERSION ||
949 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
950 			DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
951 			error = EINVAL;		/* XXX needs translation */
952 			goto out;
953 		}
954 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) {
955 			DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
956 			       dfs->dlfs_inodefmt));
957 			error = EINVAL;
958 			goto out;
959 		}
960 
961 		if (dfs->dlfs_version == 1)
962 			fsbsize = secsize;
963 		else {
964 			fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb +
965 				dfs->dlfs_fsbtodb);
966 			/*
967 			 * Could be, if the frag size is large enough, that we
968 			 * don't have the "real" primary superblock.  If that's
969 			 * the case, get the real one, and try again.
970 			 */
971 			if (sb_addr != dfs->dlfs_sboffs[0] <<
972 				       dfs->dlfs_fsbtodb) {
973 				DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
974 				      " 0x%llx is not right, trying 0x%llx\n",
975 				      (long long)sb_addr,
976 				      (long long)(dfs->dlfs_sboffs[0] <<
977 						  dfs->dlfs_fsbtodb)));
978 				sb_addr = dfs->dlfs_sboffs[0] <<
979 					  dfs->dlfs_fsbtodb;
980 				brelse(bp);
981 				continue;
982 			}
983 		}
984 		break;
985 	}
986 
987 	/*
988 	 * Check the second superblock to see which is newer; then mount
989 	 * using the older of the two.	This is necessary to ensure that
990 	 * the filesystem is valid if it was not unmounted cleanly.
991 	 */
992 
993 	if (dfs->dlfs_sboffs[1] &&
994 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
995 	{
996 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize),
997 			LFS_SBPAD, cred, &abp);
998 		if (error)
999 			goto out;
1000 		adfs = (struct dlfs *)abp->b_data;
1001 
1002 		if (dfs->dlfs_version == 1) {
1003 			/* 1s resolution comparison */
1004 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
1005 				tdfs = adfs;
1006 			else
1007 				tdfs = dfs;
1008 		} else {
1009 			/* monotonic infinite-resolution comparison */
1010 			if (adfs->dlfs_serial < dfs->dlfs_serial)
1011 				tdfs = adfs;
1012 			else
1013 				tdfs = dfs;
1014 		}
1015 
1016 		/* Check the basics. */
1017 		if (tdfs->dlfs_magic != LFS_MAGIC ||
1018 		    tdfs->dlfs_bsize > MAXBSIZE ||
1019 		    tdfs->dlfs_version > LFS_VERSION ||
1020 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
1021 			DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
1022 			      " sanity failed\n"));
1023 			error = EINVAL;		/* XXX needs translation */
1024 			goto out;
1025 		}
1026 	} else {
1027 		DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock"
1028 		      " daddr=0x%x\n", dfs->dlfs_sboffs[1]));
1029 		error = EINVAL;
1030 		goto out;
1031 	}
1032 
1033 	/* Allocate the mount structure, copy the superblock into it. */
1034 	fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
1035 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
1036 
1037 	/* Compatibility */
1038 	if (fs->lfs_version < 2) {
1039 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
1040 		fs->lfs_ibsize = fs->lfs_bsize;
1041 		fs->lfs_start = fs->lfs_sboffs[0];
1042 		fs->lfs_tstamp = fs->lfs_otstamp;
1043 		fs->lfs_fsbtodb = 0;
1044 	}
1045 
1046 	/*
1047 	 * If we aren't going to be able to write meaningfully to this
1048 	 * filesystem, and were not mounted readonly, bomb out now.
1049 	 */
1050 	if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
1051 		DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
1052 		      " we need BUFPAGES >= %lld\n",
1053 		      (long long)((bufmem_hiwater / bufmem_lowater) *
1054 				  LFS_INVERSE_MAX_BYTES(
1055 					  fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
1056 		free(fs, M_UFSMNT);
1057 		error = EFBIG; /* XXX needs translation */
1058 		goto out;
1059 	}
1060 
1061 	/* Before rolling forward, lock so vget will sleep for other procs */
1062 	fs->lfs_flags = LFS_NOTYET;
1063 	fs->lfs_rfpid = p->p_pid;
1064 
1065 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1066 	ump->um_lfs = fs;
1067 	ump->um_ops = &lfs_ufsops;
1068 	ump->um_fstype = UFS1;
1069 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
1070 		bp->b_flags |= B_INVAL;
1071 		abp->b_flags |= B_INVAL;
1072 	}
1073 	brelse(bp);
1074 	bp = NULL;
1075 	brelse(abp);
1076 	abp = NULL;
1077 
1078 	/* Set up the I/O information */
1079 	fs->lfs_devbsize = secsize;
1080 	fs->lfs_iocount = 0;
1081 	fs->lfs_diropwait = 0;
1082 	fs->lfs_activesb = 0;
1083 	fs->lfs_uinodes = 0;
1084 	fs->lfs_ravail = 0;
1085 	fs->lfs_favail = 0;
1086 	fs->lfs_sbactive = 0;
1087 
1088 	/* Set up the ifile and lock aflags */
1089 	fs->lfs_doifile = 0;
1090 	fs->lfs_writer = 0;
1091 	fs->lfs_dirops = 0;
1092 	fs->lfs_nadirop = 0;
1093 	fs->lfs_seglock = 0;
1094 	fs->lfs_pdflush = 0;
1095 	fs->lfs_sleepers = 0;
1096 	fs->lfs_pages = 0;
1097 	simple_lock_init(&fs->lfs_interlock);
1098 	lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0);
1099 	lockinit(&fs->lfs_iflock, PINOD, "lfs_iflock", 0, 0);
1100 
1101 	/* Set the file system readonly/modify bits. */
1102 	fs->lfs_ronly = ronly;
1103 	if (ronly == 0)
1104 		fs->lfs_fmod = 1;
1105 
1106 	/* Initialize the mount structure. */
1107 	dev = devvp->v_rdev;
1108 	mp->mnt_data = ump;
1109 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
1110 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
1111 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
1112 	mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
1113 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
1114 	mp->mnt_flag |= MNT_LOCAL;
1115 	mp->mnt_fs_bshift = fs->lfs_bshift;
1116 	ump->um_flags = 0;
1117 	ump->um_mountp = mp;
1118 	ump->um_dev = dev;
1119 	ump->um_devvp = devvp;
1120 	ump->um_bptrtodb = fs->lfs_fsbtodb;
1121 	ump->um_seqinc = fragstofsb(fs, fs->lfs_frag);
1122 	ump->um_nindir = fs->lfs_nindir;
1123 	ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
1124 	for (i = 0; i < MAXQUOTAS; i++)
1125 		ump->um_quotas[i] = NULLVP;
1126 	ump->um_maxsymlinklen = fs->lfs_maxsymlinklen;
1127 	ump->um_dirblksiz = DIRBLKSIZ;
1128 	ump->um_maxfilesize = fs->lfs_maxfilesize;
1129 	if (ump->um_maxsymlinklen > 0)
1130 		mp->mnt_iflag |= IMNT_DTYPE;
1131 	devvp->v_specmountpoint = mp;
1132 
1133 	/* Set up reserved memory for pageout */
1134 	lfs_setup_resblks(fs);
1135 	/* Set up vdirop tailq */
1136 	TAILQ_INIT(&fs->lfs_dchainhd);
1137 	/* and paging tailq */
1138 	TAILQ_INIT(&fs->lfs_pchainhd);
1139 
1140 	/*
1141 	 * We use the ifile vnode for almost every operation.  Instead of
1142 	 * retrieving it from the hash table each time we retrieve it here,
1143 	 * artificially increment the reference count and keep a pointer
1144 	 * to it in the incore copy of the superblock.
1145 	 */
1146 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1147 		DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
1148 		goto out;
1149 	}
1150 	fs->lfs_ivnode = vp;
1151 	VREF(vp);
1152 
1153 	/* Set up segment usage flags for the autocleaner. */
1154 	fs->lfs_nactive = 0;
1155 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
1156 						M_SEGMENT, M_WAITOK);
1157 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1158 						 M_SEGMENT, M_WAITOK);
1159 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1160 						 M_SEGMENT, M_WAITOK);
1161 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
1162 	for (i = 0; i < fs->lfs_nseg; i++) {
1163 		int changed;
1164 
1165 		LFS_SEGENTRY(sup, fs, i, bp);
1166 		changed = 0;
1167 		if (!ronly) {
1168 			if (sup->su_nbytes == 0 &&
1169 			    !(sup->su_flags & SEGUSE_EMPTY)) {
1170 				sup->su_flags |= SEGUSE_EMPTY;
1171 				++changed;
1172 			} else if (!(sup->su_nbytes == 0) &&
1173 				   (sup->su_flags & SEGUSE_EMPTY)) {
1174 				sup->su_flags &= ~SEGUSE_EMPTY;
1175 				++changed;
1176 			}
1177 			if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
1178 				sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
1179 				++changed;
1180 			}
1181 		}
1182 		fs->lfs_suflags[0][i] = sup->su_flags;
1183 		if (changed)
1184 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1185 		else
1186 			brelse(bp);
1187 	}
1188 
1189 	/*
1190 	 * Roll forward.
1191 	 *
1192 	 * We don't roll forward for v1 filesystems, because
1193 	 * of the danger that the clock was turned back between the last
1194 	 * checkpoint and crash.  This would roll forward garbage.
1195 	 *
1196 	 * v2 filesystems don't have this problem because they use a
1197 	 * monotonically increasing serial number instead of a timestamp.
1198 	 */
1199 	do_rollforward = (!(fs->lfs_pflags & LFS_PF_CLEAN) &&
1200 			  lfs_do_rfw && fs->lfs_version > 1);
1201 	if (do_rollforward) {
1202 		u_int64_t nextserial;
1203 		/*
1204 		 * Phase I: Find the address of the last good partial
1205 		 * segment that was written after the checkpoint.  Mark
1206 		 * the segments in question dirty, so they won't be
1207 		 * reallocated.
1208 		 */
1209 		lastgoodpseg = oldoffset = offset = fs->lfs_offset;
1210 		flags = 0x0;
1211 		DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%"
1212 		      PRIx64 "\n", offset));
1213 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
1214 		if (!(sup->su_flags & SEGUSE_DIRTY))
1215 			--fs->lfs_nclean;
1216 		sup->su_flags |= SEGUSE_DIRTY;
1217 		LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp);
1218 		nextserial = fs->lfs_serial + 1;
1219 		while ((offset = check_segsum(fs, offset, nextserial,
1220 		    cred, CHECK_CKSUM, &flags, l)) > 0) {
1221 			nextserial++;
1222 			if (sntod(fs, oldoffset) != sntod(fs, offset)) {
1223 				LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset),
1224 					     bp);
1225 				if (!(sup->su_flags & SEGUSE_DIRTY))
1226 					--fs->lfs_nclean;
1227 				sup->su_flags |= SEGUSE_DIRTY;
1228 				LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset),
1229 					     bp);
1230 			}
1231 
1232 			DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%"
1233 			      PRIx64 "\n", offset));
1234 			if (flags & SS_DIROP) {
1235 				DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%"
1236 				      PRIx64 "\n", oldoffset));
1237 				if (!(flags & SS_CONT))
1238 				     DLOG((DLOG_RF, "lfs_mountfs: dirops end "
1239 					   "at 0x%" PRIx64 "\n", oldoffset));
1240 			}
1241 			if (!(flags & SS_CONT))
1242 				lastgoodpseg = offset;
1243 			oldoffset = offset;
1244 		}
1245 		if (flags & SS_CONT) {
1246 			DLOG((DLOG_RF, "LFS roll forward: warning: incomplete "
1247 			      "dirops discarded\n"));
1248 		}
1249 		DLOG((DLOG_RF, "LFS roll forward phase 1: completed: "
1250 		      "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg));
1251 		oldoffset = fs->lfs_offset;
1252 		if (fs->lfs_offset != lastgoodpseg) {
1253 			/* Don't overwrite what we're trying to preserve */
1254 			offset = fs->lfs_offset;
1255 			fs->lfs_offset = lastgoodpseg;
1256 			fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset));
1257 			for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) {
1258 				sn = (sn + 1) % fs->lfs_nseg;
1259 				if (sn == curseg)
1260 					panic("lfs_mountfs: no clean segments");
1261 				LFS_SEGENTRY(sup, fs, sn, bp);
1262 				dirty = (sup->su_flags & SEGUSE_DIRTY);
1263 				brelse(bp);
1264 				if (!dirty)
1265 					break;
1266 			}
1267 			fs->lfs_nextseg = sntod(fs, sn);
1268 
1269 			/*
1270 			 * Phase II: Roll forward from the first superblock.
1271 			 */
1272 			while (offset != lastgoodpseg) {
1273 				DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%"
1274 				      PRIx64 "\n", offset));
1275 				offset = check_segsum(fs, offset,
1276 				    fs->lfs_serial + 1, cred, CHECK_UPDATE,
1277 				    NULL, l);
1278 			}
1279 
1280 			/*
1281 			 * Finish: flush our changes to disk.
1282 			 */
1283 			lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1284 			DLOG((DLOG_RF, "lfs_mountfs: roll forward ",
1285 			      "recovered %lld blocks\n",
1286 			      (long long)(lastgoodpseg - oldoffset)));
1287 		}
1288 		DLOG((DLOG_RF, "LFS roll forward complete\n"));
1289 	}
1290 	/* If writing, sb is not clean; record in case of immediate crash */
1291 	if (!fs->lfs_ronly) {
1292 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
1293 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
1294 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
1295 	}
1296 
1297 	/* Allow vget now that roll-forward is complete */
1298 	fs->lfs_flags &= ~(LFS_NOTYET);
1299 	wakeup(&fs->lfs_flags);
1300 
1301 	/*
1302 	 * Initialize the ifile cleaner info with information from
1303 	 * the superblock.
1304 	 */
1305 	LFS_CLEANERINFO(cip, fs, bp);
1306 	cip->clean = fs->lfs_nclean;
1307 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1308 	cip->avail = fs->lfs_avail;
1309 	cip->bfree = fs->lfs_bfree;
1310 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
1311 
1312 	/*
1313 	 * Mark the current segment as ACTIVE, since we're going to
1314 	 * be writing to it.
1315 	 */
1316 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
1317 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1318 	fs->lfs_nactive++;
1319 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
1320 
1321 	/* Now that roll-forward is done, unlock the Ifile */
1322 	vput(vp);
1323 
1324 	/* Comment on ifile size if it is too large */
1325 	warn_ifile_size(fs);
1326 
1327 	/* Start the pagedaemon-anticipating daemon */
1328 	if (lfs_writer_daemon == 0 &&
1329 	    kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1330 		panic("fork lfs_writer");
1331 
1332 	return (0);
1333 
1334 out:
1335 	if (bp)
1336 		brelse(bp);
1337 	if (abp)
1338 		brelse(abp);
1339 	if (ump) {
1340 		free(ump->um_lfs, M_UFSMNT);
1341 		free(ump, M_UFSMNT);
1342 		mp->mnt_data = NULL;
1343 	}
1344 
1345 	return (error);
1346 }
1347 
1348 /*
1349  * unmount system call
1350  */
1351 int
1352 lfs_unmount(struct mount *mp, int mntflags, struct lwp *l)
1353 {
1354 	struct ufsmount *ump;
1355 	struct lfs *fs;
1356 	int error, flags, ronly;
1357 	int s;
1358 
1359 	flags = 0;
1360 	if (mntflags & MNT_FORCE)
1361 		flags |= FORCECLOSE;
1362 
1363 	ump = VFSTOUFS(mp);
1364 	fs = ump->um_lfs;
1365 
1366 	/* wake up the cleaner so it can die */
1367 	wakeup(&fs->lfs_nextseg);
1368 	wakeup(&lfs_allclean_wakeup);
1369 	simple_lock(&fs->lfs_interlock);
1370 	while (fs->lfs_sleepers)
1371 		ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1372 			&fs->lfs_interlock);
1373 	simple_unlock(&fs->lfs_interlock);
1374 
1375 #ifdef QUOTA
1376 	if (mp->mnt_flag & MNT_QUOTA) {
1377 		int i;
1378 		error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags);
1379 		if (error)
1380 			return (error);
1381 		for (i = 0; i < MAXQUOTAS; i++) {
1382 			if (ump->um_quotas[i] == NULLVP)
1383 				continue;
1384 			quotaoff(l, mp, i);
1385 		}
1386 		/*
1387 		 * Here we fall through to vflush again to ensure
1388 		 * that we have gotten rid of all the system vnodes.
1389 		 */
1390 	}
1391 #endif
1392 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1393 		return (error);
1394 	if ((error = VFS_SYNC(mp, 1, l->l_proc->p_ucred, l)) != 0)
1395 		return (error);
1396 	s = splbio();
1397 	if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd))
1398 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1399 	splx(s);
1400 
1401 	/* Comment on ifile size if it has become too large */
1402 	if (!(fs->lfs_flags & LFS_WARNED))
1403 		warn_ifile_size(fs);
1404 
1405 	/* Explicitly write the superblock, to update serial and pflags */
1406 	fs->lfs_pflags |= LFS_PF_CLEAN;
1407 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
1408 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
1409 	simple_lock(&fs->lfs_interlock);
1410 	while (fs->lfs_iocount)
1411 		ltsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
1412 			&fs->lfs_interlock);
1413 	simple_unlock(&fs->lfs_interlock);
1414 
1415 	/* Finish with the Ifile, now that we're done with it */
1416 	vrele(fs->lfs_ivnode);
1417 	vgone(fs->lfs_ivnode);
1418 
1419 	ronly = !fs->lfs_ronly;
1420 	if (ump->um_devvp->v_type != VBAD)
1421 		ump->um_devvp->v_specmountpoint = NULL;
1422 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1423 	error = VOP_CLOSE(ump->um_devvp,
1424 	    ronly ? FREAD : FREAD|FWRITE, NOCRED, l);
1425 	vput(ump->um_devvp);
1426 
1427 	/* Complain about page leakage */
1428 	if (fs->lfs_pages > 0)
1429 		printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
1430 			fs->lfs_pages, lfs_subsys_pages);
1431 
1432 	/* Free per-mount data structures */
1433 	free(fs->lfs_suflags[0], M_SEGMENT);
1434 	free(fs->lfs_suflags[1], M_SEGMENT);
1435 	free(fs->lfs_suflags, M_SEGMENT);
1436 	lfs_free_resblks(fs);
1437 	free(fs, M_UFSMNT);
1438 	free(ump, M_UFSMNT);
1439 
1440 	mp->mnt_data = NULL;
1441 	mp->mnt_flag &= ~MNT_LOCAL;
1442 	return (error);
1443 }
1444 
1445 /*
1446  * Get file system statistics.
1447  *
1448  * NB: We don't lock to access the superblock here, because it's not
1449  * really that important if we get it wrong.
1450  */
1451 int
1452 lfs_statvfs(struct mount *mp, struct statvfs *sbp, struct lwp *l)
1453 {
1454 	struct lfs *fs;
1455 	struct ufsmount *ump;
1456 
1457 	ump = VFSTOUFS(mp);
1458 	fs = ump->um_lfs;
1459 	if (fs->lfs_magic != LFS_MAGIC)
1460 		panic("lfs_statvfs: magic");
1461 
1462 	sbp->f_bsize = fs->lfs_bsize;
1463 	sbp->f_frsize = fs->lfs_fsize;
1464 	sbp->f_iosize = fs->lfs_bsize;
1465 	sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs));
1466 
1467 	sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs));
1468 	KASSERT(sbp->f_bfree <= fs->lfs_dsize);
1469 	if (sbp->f_bfree < 0)
1470 		sbp->f_bfree = 0;
1471 
1472 	sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs));
1473 	if (sbp->f_bfree > sbp->f_bresvd)
1474 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1475 	else
1476 		sbp->f_bavail = 0;
1477 
1478 	sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1479 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1480 	sbp->f_favail = sbp->f_ffree;
1481 	sbp->f_fresvd = 0;
1482 	copy_statvfs_info(sbp, mp);
1483 	return (0);
1484 }
1485 
1486 /*
1487  * Go through the disk queues to initiate sandbagged IO;
1488  * go through the inodes to write those that have been modified;
1489  * initiate the writing of the super block if it has been modified.
1490  *
1491  * Note: we are always called with the filesystem marked `MPBUSY'.
1492  */
1493 int
1494 lfs_sync(struct mount *mp, int waitfor, struct ucred *cred, struct lwp *l)
1495 {
1496 	int error;
1497 	struct lfs *fs;
1498 
1499 	fs = VFSTOUFS(mp)->um_lfs;
1500 	if (fs->lfs_ronly)
1501 		return 0;
1502 	lfs_writer_enter(fs, "lfs_dirops");
1503 
1504 	/* All syncs must be checkpoints until roll-forward is implemented. */
1505 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1506 	lfs_writer_leave(fs);
1507 #ifdef QUOTA
1508 	qsync(mp);
1509 #endif
1510 	return (error);
1511 }
1512 
1513 extern struct lock ufs_hashlock;
1514 
1515 /*
1516  * Look up an LFS dinode number to find its incore vnode.  If not already
1517  * in core, read it in from the specified device.  Return the inode locked.
1518  * Detection and handling of mount points must be done by the calling routine.
1519  */
1520 int
1521 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1522 {
1523 	struct lfs *fs;
1524 	struct ufs1_dinode *dip;
1525 	struct inode *ip;
1526 	struct buf *bp;
1527 	struct ifile *ifp;
1528 	struct vnode *vp;
1529 	struct ufsmount *ump;
1530 	daddr_t daddr;
1531 	dev_t dev;
1532 	int error, retries;
1533 	struct timespec ts;
1534 
1535 	ump = VFSTOUFS(mp);
1536 	dev = ump->um_dev;
1537 	fs = ump->um_lfs;
1538 
1539 	/*
1540 	 * If the filesystem is not completely mounted yet, suspend
1541 	 * any access requests (wait for roll-forward to complete).
1542 	 */
1543 	simple_lock(&fs->lfs_interlock);
1544 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1545 		ltsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1546 			&fs->lfs_interlock);
1547 	simple_unlock(&fs->lfs_interlock);
1548 
1549 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1550 		return (0);
1551 
1552 	if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) {
1553 		*vpp = NULL;
1554 		 return (error);
1555 	}
1556 
1557 	do {
1558 		if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) {
1559 			ungetnewvnode(vp);
1560 			return (0);
1561 		}
1562 	} while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0));
1563 
1564 	/* Translate the inode number to a disk address. */
1565 	if (ino == LFS_IFILE_INUM)
1566 		daddr = fs->lfs_idaddr;
1567 	else {
1568 		/* XXX bounds-check this too */
1569 		LFS_IENTRY(ifp, fs, ino, bp);
1570 		daddr = ifp->if_daddr;
1571 		if (fs->lfs_version > 1) {
1572 			ts.tv_sec = ifp->if_atime_sec;
1573 			ts.tv_nsec = ifp->if_atime_nsec;
1574 		}
1575 
1576 		brelse(bp);
1577 		if (daddr == LFS_UNUSED_DADDR) {
1578 			*vpp = NULLVP;
1579 			ungetnewvnode(vp);
1580 			lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1581 			return (ENOENT);
1582 		}
1583 	}
1584 
1585 	/* Allocate/init new vnode/inode. */
1586 	lfs_vcreate(mp, ino, vp);
1587 
1588 	/*
1589 	 * Put it onto its hash chain and lock it so that other requests for
1590 	 * this inode will block if they arrive while we are sleeping waiting
1591 	 * for old data structures to be purged or for the contents of the
1592 	 * disk portion of this inode to be read.
1593 	 */
1594 	ip = VTOI(vp);
1595 	ufs_ihashins(ip);
1596 	lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1597 
1598 	/*
1599 	 * XXX
1600 	 * This may not need to be here, logically it should go down with
1601 	 * the i_devvp initialization.
1602 	 * Ask Kirk.
1603 	 */
1604 	ip->i_lfs = ump->um_lfs;
1605 
1606 	/* Read in the disk contents for the inode, copy into the inode. */
1607 	retries = 0;
1608     again:
1609 	error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1610 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1611 		NOCRED, &bp);
1612 	if (error) {
1613 		/*
1614 		 * The inode does not contain anything useful, so it would
1615 		 * be misleading to leave it on its hash chain. With mode
1616 		 * still zero, it will be unlinked and returned to the free
1617 		 * list by vput().
1618 		 */
1619 		vput(vp);
1620 		brelse(bp);
1621 		*vpp = NULL;
1622 		return (error);
1623 	}
1624 
1625 	dip = lfs_ifind(fs, ino, bp);
1626 	if (dip == NULL) {
1627 		/* Assume write has not completed yet; try again */
1628 		bp->b_flags |= B_INVAL;
1629 		brelse(bp);
1630 		++retries;
1631 		if (retries > LFS_IFIND_RETRIES) {
1632 #ifdef DEBUG
1633 			/* If the seglock is held look at the bpp to see
1634 			   what is there anyway */
1635 			simple_lock(&fs->lfs_interlock);
1636 			if (fs->lfs_seglock > 0) {
1637 				struct buf **bpp;
1638 				struct ufs1_dinode *dp;
1639 				int i;
1640 
1641 				for (bpp = fs->lfs_sp->bpp;
1642 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1643 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1644 					    bpp != fs->lfs_sp->bpp) {
1645 						/* Inode block */
1646 						printf("lfs_vget: block 0x%" PRIx64 ": ",
1647 						       (*bpp)->b_blkno);
1648 						dp = (struct ufs1_dinode *)(*bpp)->b_data;
1649 						for (i = 0; i < INOPB(fs); i++)
1650 							if (dp[i].di_u.inumber)
1651 								printf("%d ", dp[i].di_u.inumber);
1652 						printf("\n");
1653 					}
1654 				}
1655 			}
1656 			simple_unlock(&fs->lfs_interlock);
1657 #endif /* DEBUG */
1658 			panic("lfs_vget: dinode not found");
1659 		}
1660 		simple_lock(&fs->lfs_interlock);
1661 		if (fs->lfs_iocount) {
1662 			DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino));
1663 			(void)ltsleep(&fs->lfs_iocount, PRIBIO + 1,
1664 				      "lfs ifind", 1, &fs->lfs_interlock);
1665 		} else
1666 			retries = LFS_IFIND_RETRIES;
1667 		simple_unlock(&fs->lfs_interlock);
1668 		goto again;
1669 	}
1670 	*ip->i_din.ffs1_din = *dip;
1671 	brelse(bp);
1672 
1673 	if (fs->lfs_version > 1) {
1674 		ip->i_ffs1_atime = ts.tv_sec;
1675 		ip->i_ffs1_atimensec = ts.tv_nsec;
1676 	}
1677 
1678 	lfs_vinit(mp, &vp);
1679 
1680 	*vpp = vp;
1681 
1682 	KASSERT(VOP_ISLOCKED(vp));
1683 
1684 	return (0);
1685 }
1686 
1687 /*
1688  * File handle to vnode
1689  */
1690 int
1691 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1692 {
1693 	struct lfid *lfhp;
1694 	struct buf *bp;
1695 	IFILE *ifp;
1696 	int32_t daddr;
1697 	struct lfs *fs;
1698 
1699 	lfhp = (struct lfid *)fhp;
1700 	if (lfhp->lfid_ino < LFS_IFILE_INUM)
1701 		return ESTALE;
1702 
1703 	fs = VFSTOUFS(mp)->um_lfs;
1704 	if (lfhp->lfid_ident != fs->lfs_ident)
1705 		return ESTALE;
1706 
1707 	if (lfhp->lfid_ino >
1708 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1709 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1710 		return ESTALE;
1711 
1712 	if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfhp->lfid_ino) == NULLVP) {
1713 		LFS_IENTRY(ifp, fs, lfhp->lfid_ino, bp);
1714 		daddr = ifp->if_daddr;
1715 		brelse(bp);
1716 		if (daddr == LFS_UNUSED_DADDR)
1717 			return ESTALE;
1718 	}
1719 
1720 	return (ufs_fhtovp(mp, &lfhp->lfid_ufid, vpp));
1721 }
1722 
1723 /*
1724  * Vnode pointer to File handle
1725  */
1726 /* ARGSUSED */
1727 int
1728 lfs_vptofh(struct vnode *vp, struct fid *fhp)
1729 {
1730 	struct inode *ip;
1731 	struct lfid *lfhp;
1732 
1733 	ip = VTOI(vp);
1734 	lfhp = (struct lfid *)fhp;
1735 	lfhp->lfid_len = sizeof(struct lfid);
1736 	lfhp->lfid_ino = ip->i_number;
1737 	lfhp->lfid_gen = ip->i_gen;
1738 	lfhp->lfid_ident = ip->i_lfs->lfs_ident;
1739 	return (0);
1740 }
1741 
1742 static int
1743 sysctl_lfs_dostats(SYSCTLFN_ARGS)
1744 {
1745 	extern struct lfs_stats lfs_stats;
1746 	extern int lfs_dostats;
1747 	int error;
1748 
1749 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
1750 	if (error || newp == NULL)
1751 		return (error);
1752 
1753 	if (lfs_dostats == 0)
1754 		memset(&lfs_stats, 0, sizeof(lfs_stats));
1755 
1756 	return (0);
1757 }
1758 
1759 struct shortlong {
1760 	const char *sname;
1761 	const char *lname;
1762 };
1763 
1764 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup")
1765 {
1766 	int i;
1767 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead,
1768 		   lfs_fs_pagetrip;
1769 #ifdef DEBUG
1770 	extern int lfs_debug_log_subsys[DLOG_MAX];
1771 	struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */
1772 		{ "rollforward", "Debug roll-forward code" },
1773 		{ "alloc",	"Debug inode allocation and free list" },
1774 		{ "avail",	"Debug space-available-now accounting" },
1775 		{ "flush",	"Debug flush triggers" },
1776 		{ "lockedlist",	"Debug locked list accounting" },
1777 		{ "vnode_verbose", "Verbose per-vnode-written debugging" },
1778 		{ "vnode",	"Debug vnode use during segment write" },
1779 		{ "segment",	"Debug segment writing" },
1780 		{ "seguse",	"Debug segment used-bytes accounting" },
1781 		{ "cleaner",	"Debug cleaning routines" },
1782 		{ "mount",	"Debug mount/unmount routines" },
1783 		{ "pagecache",	"Debug UBC interactions" },
1784 		{ "dirop",	"Debug directory-operation accounting" },
1785 		{ "malloc",	"Debug private malloc accounting" },
1786 	};
1787 #endif /* DEBUG */
1788 	struct shortlong stat_names[] = { /* Must match lfs.h! */
1789 		{ "segsused",	    "Number of new segments allocated" },
1790 		{ "psegwrites",	    "Number of partial-segment writes" },
1791 		{ "psyncwrites",    "Number of synchronous partial-segment"
1792 				    " writes" },
1793 		{ "pcleanwrites",   "Number of partial-segment writes by the"
1794 				    " cleaner" },
1795 		{ "blocktot",       "Number of blocks written" },
1796 		{ "cleanblocks",    "Number of blocks written by the cleaner" },
1797 		{ "ncheckpoints",   "Number of checkpoints made" },
1798 		{ "nwrites",        "Number of whole writes" },
1799 		{ "nsync_writes",   "Number of synchronous writes" },
1800 		{ "wait_exceeded",  "Number of times writer waited for"
1801 				    " cleaner" },
1802 		{ "write_exceeded", "Number of times writer invoked flush" },
1803 		{ "flush_invoked",  "Number of times flush was invoked" },
1804 		{ "vflush_invoked", "Number of time vflush was called" },
1805 		{ "clean_inlocked", "Number of vnodes skipped for VXLOCK" },
1806 		{ "clean_vnlocked", "Number of vnodes skipped for vget failure" },
1807 		{ "segs_reclaimed", "Number of segments reclaimed" },
1808 	};
1809 
1810 	sysctl_createv(clog, 0, NULL, NULL,
1811 		       CTLFLAG_PERMANENT,
1812 		       CTLTYPE_NODE, "vfs", NULL,
1813 		       NULL, 0, NULL, 0,
1814 		       CTL_VFS, CTL_EOL);
1815 	sysctl_createv(clog, 0, NULL, NULL,
1816 		       CTLFLAG_PERMANENT,
1817 		       CTLTYPE_NODE, "lfs",
1818 		       SYSCTL_DESCR("Log-structured file system"),
1819 		       NULL, 0, NULL, 0,
1820 		       CTL_VFS, 5, CTL_EOL);
1821 	/*
1822 	 * XXX the "5" above could be dynamic, thereby eliminating one
1823 	 * more instance of the "number to vfs" mapping problem, but
1824 	 * "5" is the order as taken from sys/mount.h
1825 	 */
1826 
1827 	sysctl_createv(clog, 0, NULL, NULL,
1828 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1829 		       CTLTYPE_INT, "flushindir", NULL,
1830 		       NULL, 0, &lfs_writeindir, 0,
1831 		       CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
1832 	sysctl_createv(clog, 0, NULL, NULL,
1833 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1834 		       CTLTYPE_INT, "clean_vnhead", NULL,
1835 		       NULL, 0, &lfs_clean_vnhead, 0,
1836 		       CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
1837 	sysctl_createv(clog, 0, NULL, NULL,
1838 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1839 		       CTLTYPE_INT, "dostats",
1840 		       SYSCTL_DESCR("Maintain statistics on LFS operations"),
1841 		       sysctl_lfs_dostats, 0, &lfs_dostats, 0,
1842 		       CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
1843 	sysctl_createv(clog, 0, NULL, NULL,
1844 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1845 		       CTLTYPE_INT, "pagetrip",
1846 		       SYSCTL_DESCR("How many dirty pages in fs triggers"
1847 				    " a flush"),
1848 		       NULL, 0, &lfs_fs_pagetrip, 0,
1849 		       CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL);
1850 	sysctl_createv(clog, 0, NULL, NULL,
1851 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1852 		       CTLTYPE_INT, "rfw",
1853 		       SYSCTL_DESCR("Use in-kernel roll-forward on mount"),
1854 		       NULL, 0, &lfs_do_rfw, 0,
1855 		       CTL_VFS, 5, LFS_DO_RFW, CTL_EOL);
1856 
1857 	sysctl_createv(clog, 0, NULL, NULL,
1858 		       CTLFLAG_PERMANENT,
1859 		       CTLTYPE_NODE, "stats",
1860 		       SYSCTL_DESCR("Debugging options"),
1861 		       NULL, 0, NULL, 0,
1862 		       CTL_VFS, 5, LFS_STATS, CTL_EOL);
1863 	for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) {
1864 		sysctl_createv(clog, 0, NULL, NULL,
1865 			       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
1866 			       CTLTYPE_INT, stat_names[i].sname,
1867 			       SYSCTL_DESCR(stat_names[i].lname),
1868 			       NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]),
1869 			       0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL);
1870 	}
1871 
1872 #ifdef DEBUG
1873 	sysctl_createv(clog, 0, NULL, NULL,
1874 		       CTLFLAG_PERMANENT,
1875 		       CTLTYPE_NODE, "debug",
1876 		       SYSCTL_DESCR("Debugging options"),
1877 		       NULL, 0, NULL, 0,
1878 		       CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL);
1879 	for (i = 0; i < DLOG_MAX; i++) {
1880 		sysctl_createv(clog, 0, NULL, NULL,
1881 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1882 			       CTLTYPE_INT, dlog_names[i].sname,
1883 			       SYSCTL_DESCR(dlog_names[i].lname),
1884 			       NULL, 0, &(lfs_debug_log_subsys[i]), 0,
1885 			       CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL);
1886 	}
1887 #endif
1888 }
1889 
1890 /*
1891  * ufs_bmaparray callback function for writing.
1892  *
1893  * Since blocks will be written to the new segment anyway,
1894  * we don't care about current daddr of them.
1895  */
1896 static boolean_t
1897 lfs_issequential_hole(const struct ufsmount *ump,
1898     daddr_t daddr0, daddr_t daddr1)
1899 {
1900 	daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */
1901 	daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */
1902 
1903 	KASSERT(daddr0 == UNWRITTEN ||
1904 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1905 	KASSERT(daddr1 == UNWRITTEN ||
1906 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1907 
1908 	/* NOTE: all we want to know here is 'hole or not'. */
1909 	/* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */
1910 
1911 	/*
1912 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1913 	 */
1914 	if (daddr0 != 0 && daddr1 != 0)
1915 		return TRUE;
1916 
1917 	/*
1918 	 * both are in hole?
1919 	 */
1920 	if (daddr0 == 0 && daddr1 == 0)
1921 		return TRUE; /* all holes are 'contiguous' for us. */
1922 
1923 	return FALSE;
1924 }
1925 
1926 /*
1927  * lfs_gop_write functions exactly like genfs_gop_write, except that
1928  * (1) it requires the seglock to be held by its caller, and sp->fip
1929  *     to be properly initialized (it will return without re-initializing
1930  *     sp->fip, and without calling lfs_writeseg).
1931  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1932  *     to determine how large a block it can write at once (though it does
1933  *     still use VOP_BMAP to find holes in the file);
1934  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1935  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1936  *     now have clusters of clusters, ick.)
1937  */
1938 static int
1939 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
1940 {
1941 	int i, s, error, run;
1942 	int fs_bshift;
1943 	vaddr_t kva;
1944 	off_t eof, offset, startoffset = 0;
1945 	size_t bytes, iobytes, skipbytes;
1946 	daddr_t lbn, blkno;
1947 	struct vm_page *pg;
1948 	struct buf *mbp, *bp;
1949 	struct vnode *devvp = VTOI(vp)->i_devvp;
1950 	struct inode *ip = VTOI(vp);
1951 	struct lfs *fs = ip->i_lfs;
1952 	struct segment *sp = fs->lfs_sp;
1953 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1954 
1955 	ASSERT_SEGLOCK(fs);
1956 
1957 	/* The Ifile lives in the buffer cache */
1958 	KASSERT(vp != fs->lfs_ivnode);
1959 
1960 	/*
1961 	 * Sometimes things slip past the filters in lfs_putpages,
1962 	 * and the pagedaemon tries to write pages---problem is
1963 	 * that the pagedaemon never acquires the segment lock.
1964 	 *
1965 	 * Alternatively, pages that were clean when we called
1966 	 * genfs_putpages may have become dirty in the meantime.  In this
1967 	 * case the segment header is not properly set up for blocks
1968 	 * to be added to it.
1969 	 *
1970 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1971 	 * queue under the hypothesis that they couldn't have got here
1972 	 * unless they were modified *quite* recently.
1973 	 *
1974 	 * XXXUBC that last statement is an oversimplification of course.
1975 	 */
1976 	if (!LFS_SEGLOCK_HELD(fs) ||
1977 	    (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) ||
1978 	    (pgs[0]->offset & fs->lfs_bmask) != 0) {
1979 		goto tryagain;
1980 	}
1981 
1982 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1983 	    vp, pgs, npages, flags);
1984 
1985 	GOP_SIZE(vp, vp->v_size, &eof, GOP_SIZE_WRITE);
1986 
1987 	if (vp->v_type == VREG)
1988 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1989 	else
1990 		fs_bshift = DEV_BSHIFT;
1991 	error = 0;
1992 	pg = pgs[0];
1993 	startoffset = pg->offset;
1994 	if (startoffset >= eof) {
1995 		goto tryagain;
1996 	} else
1997 		bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1998 	skipbytes = 0;
1999 
2000 	KASSERT(bytes != 0);
2001 
2002 	/* Swap PG_DELWRI for PG_PAGEOUT */
2003 	for (i = 0; i < npages; i++)
2004 		if (pgs[i]->flags & PG_DELWRI) {
2005 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
2006 			pgs[i]->flags &= ~PG_DELWRI;
2007 			pgs[i]->flags |= PG_PAGEOUT;
2008 			uvmexp.paging++;
2009 			uvm_lock_pageq();
2010 			uvm_pageunwire(pgs[i]);
2011 			uvm_unlock_pageq();
2012 		}
2013 
2014 	/*
2015 	 * Check to make sure we're starting on a block boundary.
2016 	 * We'll check later to make sure we always write entire
2017 	 * blocks (or fragments).
2018 	 */
2019 	if (startoffset & fs->lfs_bmask)
2020 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
2021 		       startoffset, fs->lfs_bmask,
2022 		       startoffset & fs->lfs_bmask);
2023 	KASSERT((startoffset & fs->lfs_bmask) == 0);
2024 	if (bytes & fs->lfs_ffmask) {
2025 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
2026 		panic("lfs_gop_write: non-integer blocks");
2027 	}
2028 
2029 	/*
2030 	 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
2031 	 * If we would, write what we have and try again.  If we don't
2032 	 * have anything to write, we'll have to sleep.
2033 	 */
2034 	if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
2035 				      (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ?
2036 				       UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
2037 		int vers;
2038 
2039 		DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
2040 #if 0
2041 		      " with nfinfo=%d at offset 0x%x\n",
2042 		      (int)((SEGSUM *)(sp->segsum))->ss_nfinfo,
2043 		      (unsigned)fs->lfs_offset));
2044 #endif
2045 		if (sp->fip->fi_nblocks == 0) {
2046 			/* Don't write zero-length finfos */
2047 			--((SEGSUM *)(sp->segsum))->ss_nfinfo;
2048 			sp->sum_bytes_left += FINFOSIZE;
2049 		} else
2050 			lfs_updatemeta(sp);
2051 
2052 		vers = sp->fip->fi_version;
2053 		(void) lfs_writeseg(fs, sp);
2054 
2055 		sp->fip->fi_version = vers;
2056 		sp->fip->fi_ino = ip->i_number;
2057 		/* Add the current file to the segment summary. */
2058 		++((SEGSUM *)(sp->segsum))->ss_nfinfo;
2059 		sp->sum_bytes_left -= FINFOSIZE;
2060 
2061 		/*
2062 		 * Having given up all of the pager_map we were holding,
2063 		 * we can now wait for aiodoned to reclaim it for us
2064 		 * without fear of deadlock.
2065 		 */
2066 		kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
2067 				     UVMPAGER_MAPIN_WAITOK);
2068 	}
2069 
2070 	s = splbio();
2071 	simple_lock(&global_v_numoutput_slock);
2072 	vp->v_numoutput += 2; /* one for biodone, one for aiodone */
2073 	simple_unlock(&global_v_numoutput_slock);
2074 	splx(s);
2075 
2076 	mbp = getiobuf();
2077 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
2078 	    vp, mbp, vp->v_numoutput, bytes);
2079 	mbp->b_bufsize = npages << PAGE_SHIFT;
2080 	mbp->b_data = (void *)kva;
2081 	mbp->b_resid = mbp->b_bcount = bytes;
2082 	mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL;
2083 	mbp->b_iodone = uvm_aio_biodone;
2084 	mbp->b_vp = vp;
2085 
2086 	bp = NULL;
2087 	for (offset = startoffset;
2088 	    bytes > 0;
2089 	    offset += iobytes, bytes -= iobytes) {
2090 		lbn = offset >> fs_bshift;
2091 		error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
2092 		    lfs_issequential_hole);
2093 		if (error) {
2094 			UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
2095 			    error,0,0,0);
2096 			skipbytes += bytes;
2097 			bytes = 0;
2098 			break;
2099 		}
2100 
2101 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
2102 		    bytes);
2103 		if (blkno == (daddr_t)-1) {
2104 			skipbytes += iobytes;
2105 			continue;
2106 		}
2107 
2108 		/*
2109 		 * Discover how much we can really pack into this buffer.
2110 		 */
2111 		/* If no room in the current segment, finish it up */
2112 		if (sp->sum_bytes_left < sizeof(int32_t) ||
2113 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
2114 			int vers;
2115 
2116 			lfs_updatemeta(sp);
2117 
2118 			vers = sp->fip->fi_version;
2119 			(void) lfs_writeseg(fs, sp);
2120 
2121 			sp->fip->fi_version = vers;
2122 			sp->fip->fi_ino = ip->i_number;
2123 			/* Add the current file to the segment summary. */
2124 			++((SEGSUM *)(sp->segsum))->ss_nfinfo;
2125 			sp->sum_bytes_left -= FINFOSIZE;
2126 		}
2127 		/* Check both for space in segment and space in segsum */
2128 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
2129 					<< fs_bshift);
2130 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
2131 				       << fs_bshift);
2132 		KASSERT(iobytes > 0);
2133 
2134 		/* if it's really one i/o, don't make a second buf */
2135 		if (offset == startoffset && iobytes == bytes) {
2136 			bp = mbp;
2137 			/* correct overcount if there is no second buffer */
2138 			s = splbio();
2139 			simple_lock(&global_v_numoutput_slock);
2140 			--vp->v_numoutput;
2141 			simple_unlock(&global_v_numoutput_slock);
2142 			splx(s);
2143 		} else {
2144 			bp = getiobuf();
2145 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
2146 			    vp, bp, vp->v_numoutput, 0);
2147 			bp->b_data = (char *)kva +
2148 			    (vaddr_t)(offset - pg->offset);
2149 			bp->b_resid = bp->b_bcount = iobytes;
2150 			bp->b_flags = B_BUSY|B_WRITE|B_CALL;
2151 			bp->b_iodone = uvm_aio_biodone1;
2152 		}
2153 
2154 		/* XXX This is silly ... is this necessary? */
2155 		bp->b_vp = NULL;
2156 		s = splbio();
2157 		bgetvp(vp, bp);
2158 		splx(s);
2159 
2160 		bp->b_lblkno = lblkno(fs, offset);
2161 		bp->b_private = mbp;
2162 		if (devvp->v_type == VBLK) {
2163 			bp->b_dev = devvp->v_rdev;
2164 		}
2165 		VOP_BWRITE(bp);
2166 		while (lfs_gatherblock(sp, bp, NULL))
2167 			continue;
2168 	}
2169 
2170 	if (skipbytes) {
2171 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
2172 		s = splbio();
2173 		if (error) {
2174 			mbp->b_flags |= B_ERROR;
2175 			mbp->b_error = error;
2176 		}
2177 		mbp->b_resid -= skipbytes;
2178 		if (mbp->b_resid == 0) {
2179 			biodone(mbp);
2180 		}
2181 		splx(s);
2182 	}
2183 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
2184 	return (0);
2185 
2186     tryagain:
2187 	/*
2188 	 * We can't write the pages, for whatever reason.
2189 	 * Clean up after ourselves, and make the caller try again.
2190 	 */
2191 	simple_lock(&vp->v_interlock);
2192 
2193 	/* Tell why we're here, if we know */
2194 	if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE)
2195 		DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n"));
2196 	else if ((pgs[0]->offset & fs->lfs_bmask) != 0)
2197 		DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n"));
2198 	else if (startoffset >= eof)
2199 		DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64
2200 		      " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number,
2201 		      pgs[0]->offset, eof, npages));
2202 	else
2203 		DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n"));
2204 
2205 	uvm_lock_pageq();
2206 	for (i = 0; i < npages; i++) {
2207 		pg = pgs[i];
2208 
2209 		if (pg->flags & PG_PAGEOUT)
2210 			uvmexp.paging--;
2211 		if (pg->flags & PG_DELWRI) {
2212 			uvm_pageunwire(pg);
2213 		}
2214 		uvm_pageactivate(pg);
2215 		pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
2216 		DLOG((DLOG_PAGE, "pg[%d] = %p\n", i, pg));
2217 		DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags));
2218 		DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags));
2219 		DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon));
2220 		DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject));
2221 		DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i,
2222 		      pg->wire_count));
2223 		DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i,
2224 		      pg->loan_count));
2225 	}
2226 	/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
2227 	uvm_page_unbusy(pgs, npages);
2228 	uvm_unlock_pageq();
2229 	simple_unlock(&vp->v_interlock);
2230 	return EAGAIN;
2231 }
2232 
2233 /*
2234  * finish vnode/inode initialization.
2235  * used by lfs_vget and lfs_fastvget.
2236  */
2237 void
2238 lfs_vinit(struct mount *mp, struct vnode **vpp)
2239 {
2240 	struct vnode *vp = *vpp;
2241 	struct inode *ip = VTOI(vp);
2242 	struct ufsmount *ump = VFSTOUFS(mp);
2243 	int i;
2244 
2245 	ip->i_mode = ip->i_ffs1_mode;
2246 	ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink;
2247 	ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
2248 	ip->i_flags = ip->i_ffs1_flags;
2249 	ip->i_gen = ip->i_ffs1_gen;
2250 	ip->i_uid = ip->i_ffs1_uid;
2251 	ip->i_gid = ip->i_ffs1_gid;
2252 
2253 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
2254 
2255 	/*
2256 	 * Initialize the vnode from the inode, check for aliases.  In all
2257 	 * cases re-init ip, the underlying vnode/inode may have changed.
2258 	 */
2259 	ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2260 	ip = VTOI(vp);
2261 
2262 	memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
2263 	if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) {
2264 		struct lfs *fs = ump->um_lfs;
2265 #ifdef DEBUG
2266 		for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
2267 		    i < NDADDR; i++) {
2268 			if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
2269 			    i == 0)
2270 				continue;
2271 			if (ip->i_ffs1_db[i] != 0) {
2272 inconsistent:
2273 				lfs_dump_dinode(ip->i_din.ffs1_din);
2274 				panic("inconsistent inode");
2275 			}
2276 		}
2277 		for ( ; i < NDADDR + NIADDR; i++) {
2278 			if (ip->i_ffs1_ib[i - NDADDR] != 0) {
2279 				goto inconsistent;
2280 			}
2281 		}
2282 #endif /* DEBUG */
2283 		for (i = 0; i < NDADDR; i++)
2284 			if (ip->i_ffs1_db[i] != 0)
2285 				ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
2286 	}
2287 
2288 #ifdef DIAGNOSTIC
2289 	if (vp->v_type == VNON) {
2290 # ifdef DEBUG
2291 		lfs_dump_dinode(ip->i_din.ffs1_din);
2292 # endif
2293 		panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n",
2294 		      (unsigned long long)ip->i_number,
2295 		      (ip->i_mode & IFMT) >> 12);
2296 	}
2297 #endif /* DIAGNOSTIC */
2298 
2299 	/*
2300 	 * Finish inode initialization now that aliasing has been resolved.
2301 	 */
2302 
2303 	ip->i_devvp = ump->um_devvp;
2304 	VREF(ip->i_devvp);
2305 	genfs_node_init(vp, &lfs_genfsops);
2306 	uvm_vnp_setsize(vp, ip->i_size);
2307 
2308 	/* Initialize hiblk from file size */
2309 	ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1;
2310 
2311 	*vpp = vp;
2312 }
2313 
2314 /*
2315  * Warn if the inode portion of the Ifile is too large to be contained
2316  * in the buffer cache, according to LFS_MAX_BUFS / LFS_MAX_BYTES.
2317  * XXX the estimates don't take multiple LFSs into account.
2318  */
2319 static void
2320 warn_ifile_size(struct lfs *fs)
2321 {
2322 	KASSERT(LFS_MAX_BUFS > 0);
2323 	KASSERT(LFS_MAX_BYTES > 0);
2324 	if (((fs->lfs_ivnode->v_size >> fs->lfs_bshift) - fs->lfs_segtabsz) >
2325 	    LFS_MAX_BUFS) {
2326 		simple_lock(&fs->lfs_interlock);
2327 		fs->lfs_flags |= LFS_WARNED;
2328 		simple_unlock(&fs->lfs_interlock);
2329 		log(LOG_WARNING, "lfs_mountfs: inode part of ifile of length %"
2330 				 PRId64 " cannot fit in %d buffers\n",
2331 				 fs->lfs_ivnode->v_size -
2332 				 (fs->lfs_segtabsz << fs->lfs_bshift),
2333 				 LFS_MAX_BUFS);
2334 		log(LOG_WARNING, "lfs_mountfs: please consider increasing NBUF"
2335 				 " to at least %" PRId64 "\n",
2336 				 LFS_INVERSE_MAX_BUFS((fs->lfs_ivnode->v_size >>
2337 						       fs->lfs_bshift) -
2338 						      fs->lfs_segtabsz));
2339 	} else if ((fs->lfs_ivnode->v_size >> fs->lfs_bshift) > LFS_MAX_BUFS) {
2340 		/* Same thing but LOG_NOTICE */
2341 		simple_lock(&fs->lfs_interlock);
2342 		fs->lfs_flags |= LFS_WARNED;
2343 		simple_unlock(&fs->lfs_interlock);
2344 		log(LOG_NOTICE, "lfs_mountfs: entire ifile of length %"
2345 				PRId64 " cannot fit in %d buffers\n",
2346 				fs->lfs_ivnode->v_size, LFS_MAX_BUFS);
2347 		log(LOG_NOTICE, "lfs_mountfs: please consider increasing NBUF"
2348 				" to at least %" PRId64 "\n",
2349 				LFS_INVERSE_MAX_BUFS(fs->lfs_ivnode->v_size >>
2350 						     fs->lfs_bshift));
2351 	}
2352 
2353 	if (fs->lfs_ivnode->v_size - (fs->lfs_segtabsz << fs->lfs_bshift) >
2354 	    LFS_MAX_BYTES) {
2355 		simple_lock(&fs->lfs_interlock);
2356 		fs->lfs_flags |= LFS_WARNED;
2357 		simple_unlock(&fs->lfs_interlock);
2358 		log(LOG_WARNING, "lfs_mountfs: inode part of ifile of length %"
2359 				 PRId64 " cannot fit in %lu bytes\n",
2360 				 fs->lfs_ivnode->v_size - (fs->lfs_segtabsz <<
2361 							   fs->lfs_bshift),
2362 				 LFS_MAX_BYTES);
2363 		log(LOG_WARNING, "lfs_mountfs: please consider increasing"
2364 				 " BUFPAGES to at least %" PRId64 "\n",
2365 				 LFS_INVERSE_MAX_BYTES(fs->lfs_ivnode->v_size -
2366 						       (fs->lfs_segtabsz <<
2367 							fs->lfs_bshift)) >>
2368 				 PAGE_SHIFT);
2369 	} else if(fs->lfs_ivnode->v_size > LFS_MAX_BYTES) {
2370 		simple_lock(&fs->lfs_interlock);
2371 		fs->lfs_flags |= LFS_WARNED;
2372 		simple_unlock(&fs->lfs_interlock);
2373 		log(LOG_NOTICE, "lfs_mountfs: entire ifile of length %" PRId64
2374 				" cannot fit in %lu buffer bytes\n",
2375 				fs->lfs_ivnode->v_size, LFS_MAX_BYTES);
2376 		log(LOG_NOTICE, "lfs_mountfs: please consider increasing"
2377 				" BUFPAGES to at least %" PRId64 "\n",
2378 				LFS_INVERSE_MAX_BYTES(fs->lfs_ivnode->v_size -
2379 						      (fs->lfs_segtabsz <<
2380 						       fs->lfs_bshift)) >>
2381 				PAGE_SHIFT);
2382 	}
2383 }
2384 
2385 /*
2386  * Resize the filesystem to contain the specified number of segments.
2387  */
2388 int
2389 lfs_resize_fs(struct lfs *fs, int newnsegs)
2390 {
2391 	SEGUSE *sup;
2392 	struct buf *bp, *obp;
2393 	daddr_t olast, nlast, ilast, noff, start, end;
2394 	struct vnode *ivp;
2395 	struct inode *ip;
2396 	int error, badnews, inc, oldnsegs;
2397 	int sbbytes, csbbytes, gain, cgain;
2398 	int i;
2399 
2400 	/* Only support v2 and up */
2401 	if (fs->lfs_version < 2)
2402 		return EOPNOTSUPP;
2403 
2404 	/* If we're doing nothing, do it fast */
2405 	oldnsegs = fs->lfs_nseg;
2406 	if (newnsegs == oldnsegs)
2407 		return 0;
2408 
2409 	/* We always have to have two superblocks */
2410 	if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1]))
2411 		return EFBIG;
2412 
2413 	ivp = fs->lfs_ivnode;
2414 	ip = VTOI(ivp);
2415 	error = 0;
2416 
2417 	/* Take the segment lock so no one else calls lfs_newseg() */
2418 	lfs_seglock(fs, SEGM_PROT);
2419 
2420 	/*
2421 	 * Make sure the segments we're going to be losing, if any,
2422 	 * are in fact empty.  We hold the seglock, so their status
2423 	 * cannot change underneath us.  Count the superblocks we lose,
2424 	 * while we're at it.
2425 	 */
2426 	sbbytes = csbbytes = 0;
2427 	cgain = 0;
2428 	for (i = newnsegs; i < oldnsegs; i++) {
2429 		LFS_SEGENTRY(sup, fs, i, bp);
2430 		badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL);
2431 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
2432 			sbbytes += LFS_SBPAD;
2433 		if (!(sup->su_flags & SEGUSE_DIRTY)) {
2434 			++cgain;
2435 			if (sup->su_flags & SEGUSE_SUPERBLOCK)
2436 				csbbytes += LFS_SBPAD;
2437 		}
2438 		brelse(bp);
2439 		if (badnews) {
2440 			error = EBUSY;
2441 			goto out;
2442 		}
2443 	}
2444 
2445 	/* Note old and new segment table endpoints, and old ifile size */
2446 	olast = fs->lfs_cleansz + fs->lfs_segtabsz;
2447 	nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz;
2448 	ilast = ivp->v_size >> fs->lfs_bshift;
2449 	noff = nlast - olast;
2450 
2451 	/*
2452 	 * Make sure no one can use the Ifile while we change it around.
2453 	 * Even after taking the iflock we need to make sure no one still
2454 	 * is holding Ifile buffers, so we get each one, to drain them.
2455 	 * (XXX this could be done better.)
2456 	 */
2457 	simple_lock(&fs->lfs_interlock);
2458 	lockmgr(&fs->lfs_iflock, LK_EXCLUSIVE, &fs->lfs_interlock);
2459 	simple_unlock(&fs->lfs_interlock);
2460 	vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY);
2461 	for (i = 0; i < ilast; i++) {
2462 		bread(ivp, i, fs->lfs_bsize, NOCRED, &bp);
2463 		brelse(bp);
2464 	}
2465 
2466 	/* Allocate new Ifile blocks */
2467 	for (i = ilast; i < ilast + noff; i++) {
2468 		if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0,
2469 			       &bp) != 0)
2470 			panic("balloc extending ifile");
2471 		memset(bp->b_data, 0, fs->lfs_bsize);
2472 		VOP_BWRITE(bp);
2473 	}
2474 
2475 	/* Register new ifile size */
2476 	ip->i_size += noff * fs->lfs_bsize;
2477 	ip->i_ffs1_size = ip->i_size;
2478 	uvm_vnp_setsize(ivp, ip->i_size);
2479 
2480 	/* Copy the inode table to its new position */
2481 	if (noff != 0) {
2482 		if (noff < 0) {
2483 			start = nlast;
2484 			end = ilast + noff;
2485 			inc = 1;
2486 		} else {
2487 			start = ilast + noff - 1;
2488 			end = nlast - 1;
2489 			inc = -1;
2490 		}
2491 		for (i = start; i != end; i += inc) {
2492 			if (bread(ivp, i, fs->lfs_bsize, NOCRED, &bp) != 0)
2493 				panic("resize: bread dst blk failed");
2494 			if (bread(ivp, i - noff, fs->lfs_bsize, NOCRED, &obp))
2495 				panic("resize: bread src blk failed");
2496 			memcpy(bp->b_data, obp->b_data, fs->lfs_bsize);
2497 			VOP_BWRITE(bp);
2498 			brelse(obp);
2499 		}
2500 	}
2501 
2502 	/* If we are expanding, write the new empty SEGUSE entries */
2503 	if (newnsegs > oldnsegs) {
2504 		for (i = oldnsegs; i < newnsegs; i++) {
2505 			if ((error = bread(ivp, i / fs->lfs_sepb +
2506 					   fs->lfs_cleansz,
2507 					   fs->lfs_bsize, NOCRED, &bp)) != 0)
2508 				panic("lfs: ifile read: %d", error);
2509 			while ((i + 1) % fs->lfs_sepb && i < newnsegs) {
2510 				sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb];
2511 				memset(sup, 0, sizeof(*sup));
2512 				i++;
2513 			}
2514 			VOP_BWRITE(bp);
2515 		}
2516 	}
2517 
2518 	/* Zero out unused superblock offsets */
2519 	for (i = 2; i < LFS_MAXNUMSB; i++)
2520 		if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs)
2521 			fs->lfs_sboffs[i] = 0x0;
2522 
2523 	/*
2524 	 * Correct superblock entries that depend on fs size.
2525 	 * The computations of these are as follows:
2526 	 *
2527 	 * size  = segtod(fs, nseg)
2528 	 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD)
2529 	 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used
2530 	 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD)
2531 	 *         + (segtod(fs, 1) - (offset - curseg))
2532 	 *	   - segtod(fs, minfreeseg - (minfreeseg / 2))
2533 	 *
2534 	 * XXX - we should probably adjust minfreeseg as well.
2535 	 */
2536 	gain = (newnsegs - oldnsegs);
2537 	fs->lfs_nseg = newnsegs;
2538 	fs->lfs_segtabsz = nlast - fs->lfs_cleansz;
2539 	fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize);
2540 	fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes);
2541 	fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes)
2542 		       - gain * btofsb(fs, fs->lfs_bsize / 2);
2543 	if (gain > 0) {
2544 		fs->lfs_nclean += gain;
2545 		fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize);
2546 	} else {
2547 		fs->lfs_nclean -= cgain;
2548 		fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) -
2549 				 btofsb(fs, csbbytes);
2550 	}
2551 
2552 	/* Resize segment flag cache */
2553 	fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0],
2554 						  fs->lfs_nseg * sizeof(u_int32_t),
2555 						  M_SEGMENT, M_WAITOK);
2556 	fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[0],
2557 						  fs->lfs_nseg * sizeof(u_int32_t),
2558 						  M_SEGMENT, M_WAITOK);
2559 	for (i = oldnsegs; i < newnsegs; i++)
2560 		fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0;
2561 
2562 	/* Truncate Ifile if necessary */
2563 	if (noff < 0)
2564 		lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0,
2565 			     NOCRED, curlwp);
2566 
2567 	/* Update cleaner info so the cleaner can die */
2568 	bread(ivp, 0, fs->lfs_bsize, NOCRED, &bp);
2569 	((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean;
2570 	((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean;
2571 	VOP_BWRITE(bp);
2572 
2573 	/* Let Ifile accesses proceed */
2574 	VOP_UNLOCK(ivp, 0);
2575 	simple_lock(&fs->lfs_interlock);
2576 	lockmgr(&fs->lfs_iflock, LK_RELEASE, &fs->lfs_interlock);
2577 	simple_unlock(&fs->lfs_interlock);
2578 
2579     out:
2580 	lfs_segunlock(fs);
2581 	return error;
2582 }
2583