xref: /netbsd-src/sys/ufs/lfs/lfs_rfw.c (revision 7fa608457b817eca6e0977b37f758ae064f3c99c)
1 /*	$NetBSD: lfs_rfw.c,v 1.5 2007/10/10 20:42:35 ad 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 #ifdef LFS_KERNEL_RFW
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: lfs_rfw.c,v 1.5 2007/10/10 20:42:35 ad Exp $");
43 
44 #if defined(_KERNEL_OPT)
45 #include "opt_quota.h"
46 #endif
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/namei.h>
51 #include <sys/proc.h>
52 #include <sys/kernel.h>
53 #include <sys/vnode.h>
54 #include <sys/mount.h>
55 #include <sys/kthread.h>
56 #include <sys/buf.h>
57 #include <sys/device.h>
58 #include <sys/mbuf.h>
59 #include <sys/file.h>
60 #include <sys/disklabel.h>
61 #include <sys/ioctl.h>
62 #include <sys/errno.h>
63 #include <sys/malloc.h>
64 #include <sys/pool.h>
65 #include <sys/socket.h>
66 #include <sys/syslog.h>
67 #include <uvm/uvm_extern.h>
68 #include <sys/sysctl.h>
69 #include <sys/conf.h>
70 #include <sys/kauth.h>
71 
72 #include <miscfs/specfs/specdev.h>
73 
74 #include <ufs/ufs/quota.h>
75 #include <ufs/ufs/inode.h>
76 #include <ufs/ufs/ufsmount.h>
77 #include <ufs/ufs/ufs_extern.h>
78 
79 #include <uvm/uvm.h>
80 #include <uvm/uvm_stat.h>
81 #include <uvm/uvm_pager.h>
82 #include <uvm/uvm_pdaemon.h>
83 
84 #include <ufs/lfs/lfs.h>
85 #include <ufs/lfs/lfs_extern.h>
86 
87 #include <miscfs/genfs/genfs.h>
88 #include <miscfs/genfs/genfs_node.h>
89 
90 /*
91  * Roll-forward code.
92  */
93 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t,
94     kauth_cred_t, int, int *, struct lwp *);
95 
96 extern int lfs_do_rfw;
97 
98 /*
99  * Allocate a particular inode with a particular version number, freeing
100  * any previous versions of this inode that may have gone before.
101  * Used by the roll-forward code.
102  *
103  * XXX this function does not have appropriate locking to be used on a live fs;
104  * XXX but something similar could probably be used for an "undelete" call.
105  *
106  * Called with the Ifile inode locked.
107  */
108 int
109 lfs_rf_valloc(struct lfs *fs, ino_t ino, int vers, struct lwp *l,
110 	      struct vnode **vpp)
111 {
112 	IFILE *ifp;
113 	struct buf *bp, *cbp;
114 	struct vnode *vp;
115 	struct inode *ip;
116 	ino_t tino, oldnext;
117 	int error;
118 	CLEANERINFO *cip;
119 
120 	ASSERT_SEGLOCK(fs); /* XXX it doesn't, really */
121 
122 	/*
123 	 * First, just try a vget. If the version number is the one we want,
124 	 * we don't have to do anything else.  If the version number is wrong,
125 	 * take appropriate action.
126 	 */
127 	error = VFS_VGET(fs->lfs_ivnode->v_mount, ino, &vp);
128 	if (error == 0) {
129 		DLOG((DLOG_RF, "lfs_rf_valloc[1]: ino %d vp %p\n", ino, vp));
130 
131 		*vpp = vp;
132 		ip = VTOI(vp);
133 		if (ip->i_gen == vers)
134 			return 0;
135 		else if (ip->i_gen < vers) {
136 			lfs_truncate(vp, (off_t)0, 0, NOCRED, l);
137 			ip->i_gen = ip->i_ffs1_gen = vers;
138 			LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
139 			return 0;
140 		} else {
141 			DLOG((DLOG_RF, "ino %d: sought version %d, got %d\n",
142 			       ino, vers, ip->i_ffs1_gen));
143 			vput(vp);
144 			*vpp = NULLVP;
145 			return EEXIST;
146 		}
147 	}
148 
149 	/*
150 	 * The inode is not in use.  Find it on the free list.
151 	 */
152 	/* If the Ifile is too short to contain this inum, extend it */
153 	while (VTOI(fs->lfs_ivnode)->i_size <= (ino /
154 		fs->lfs_ifpb + fs->lfs_cleansz + fs->lfs_segtabsz)
155 		<< fs->lfs_bshift) {
156 		lfs_extend_ifile(fs, NOCRED);
157 	}
158 
159 	LFS_IENTRY(ifp, fs, ino, bp);
160 	oldnext = ifp->if_nextfree;
161 	ifp->if_version = vers;
162 	brelse(bp, 0);
163 
164 	LFS_GET_HEADFREE(fs, cip, cbp, &ino);
165 	if (ino) {
166 		LFS_PUT_HEADFREE(fs, cip, cbp, oldnext);
167 	} else {
168 		tino = ino;
169 		while (1) {
170 			LFS_IENTRY(ifp, fs, tino, bp);
171 			if (ifp->if_nextfree == ino ||
172 			    ifp->if_nextfree == LFS_UNUSED_INUM)
173 				break;
174 			tino = ifp->if_nextfree;
175 			brelse(bp, 0);
176 		}
177 		if (ifp->if_nextfree == LFS_UNUSED_INUM) {
178 			brelse(bp, 0);
179 			return ENOENT;
180 		}
181 		ifp->if_nextfree = oldnext;
182 		LFS_BWRITE_LOG(bp);
183 	}
184 
185 	error = lfs_ialloc(fs, fs->lfs_ivnode, ino, vers, &vp);
186 	if (error == 0) {
187 		/*
188 		 * Make it VREG so we can put blocks on it.  We will change
189 		 * this later if it turns out to be some other kind of file.
190 		 */
191 		ip = VTOI(vp);
192 		ip->i_mode = ip->i_ffs1_mode = IFREG;
193 		ip->i_nlink = ip->i_ffs1_nlink = 1;
194 		ip->i_ffs_effnlink = 1;
195 		ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, &vp);
196 		ip = VTOI(vp);
197 
198 		DLOG((DLOG_RF, "lfs_rf_valloc: ino %d vp %p\n", ino, vp));
199 
200 		/* The dirop-nature of this vnode is past */
201 		lfs_unmark_vnode(vp);
202 		(void)lfs_vunref(vp);
203 		vp->v_uflag &= ~VU_DIROP;
204 		simple_lock(&fs->lfs_interlock);
205 		simple_lock(&lfs_subsys_lock);
206 		--lfs_dirvcount;
207 		simple_unlock(&lfs_subsys_lock);
208 		--fs->lfs_dirvcount;
209 		TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
210 		wakeup(&lfs_dirvcount);
211 		wakeup(&fs->lfs_dirvcount);
212 		simple_unlock(&fs->lfs_interlock);
213 	}
214 	*vpp = vp;
215 	return error;
216 }
217 
218 /*
219  * Load the appropriate indirect block, and change the appropriate pointer.
220  * Mark the block dirty.  Do segment and avail accounting.
221  */
222 static int
223 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn,
224 	    daddr_t ndaddr, size_t size, struct lwp *l)
225 {
226 	int error;
227 	struct vnode *vp;
228 	struct inode *ip;
229 #ifdef DEBUG
230 	daddr_t odaddr;
231 	struct indir a[NIADDR];
232 	int num;
233 	int i;
234 #endif /* DEBUG */
235 	struct buf *bp;
236 	SEGUSE *sup;
237 
238 	KASSERT(lbn >= 0);	/* no indirect blocks */
239 
240 	if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) {
241 		DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc"
242 		      " returned %d\n", ino, error));
243 		return error;
244 	}
245 
246 	if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size,
247 				NOCRED, 0, &bp)) != 0) {
248 		vput(vp);
249 		return (error);
250 	}
251 	/* No need to write, the block is already on disk */
252 	if (bp->b_flags & B_DELWRI) {
253 		LFS_UNLOCK_BUF(bp);
254 		fs->lfs_avail += btofsb(fs, bp->b_bcount);
255 	}
256 	brelse(bp, BC_INVAL);
257 
258 	/*
259 	 * Extend the file, if it is not large enough already.
260 	 * XXX this is not exactly right, we don't know how much of the
261 	 * XXX last block is actually used.  We hope that an inode will
262 	 * XXX appear later to give the correct size.
263 	 */
264 	ip = VTOI(vp);
265 	if (ip->i_size <= (lbn << fs->lfs_bshift)) {
266 		u_int64_t newsize;
267 
268 		if (lbn < NDADDR)
269 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) +
270 				(size - fs->lfs_fsize) + 1;
271 		else
272 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1;
273 
274 		if (ip->i_size < newsize) {
275 			ip->i_size = newsize;
276 			/*
277 			 * tell vm our new size for the case the inode won't
278 			 * appear later.
279 			 */
280 			uvm_vnp_setsize(vp, newsize);
281 		}
282 	}
283 
284 	lfs_update_single(fs, NULL, vp, lbn, ndaddr, size);
285 
286 	LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
287 	sup->su_nbytes += size;
288 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
289 
290 	/* differences here should be due to UNWRITTEN indirect blocks. */
291 	KASSERT((lblkno(fs, ip->i_size) > NDADDR &&
292 	    ip->i_lfs_effnblks == ip->i_ffs1_blocks) ||
293 	    ip->i_lfs_effnblks >= ip->i_ffs1_blocks);
294 
295 #ifdef DEBUG
296 	/* Now look again to make sure it worked */
297 	ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
298 	for (i = num; i > 0; i--) {
299 		if (!a[i].in_exists)
300 			panic("update_meta: absent %d lv indirect block", i);
301 	}
302 	if (dbtofsb(fs, odaddr) != ndaddr)
303 		DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %"
304 		      PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr));
305 #endif /* DEBUG */
306 	vput(vp);
307 	return 0;
308 }
309 
310 static int
311 update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred,
312 	      struct lwp *l)
313 {
314 	struct vnode *devvp, *vp;
315 	struct inode *ip;
316 	struct ufs1_dinode *dip;
317 	struct buf *dbp, *ibp;
318 	int error;
319 	daddr_t daddr;
320 	IFILE *ifp;
321 	SEGUSE *sup;
322 
323 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
324 
325 	/*
326 	 * Get the inode, update times and perms.
327 	 * DO NOT update disk blocks, we do that separately.
328 	 */
329 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp);
330 	if (error) {
331 		DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error));
332 		return error;
333 	}
334 	dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs);
335 	while (--dip >= (struct ufs1_dinode *)dbp->b_data) {
336 		if (dip->di_inumber > LFS_IFILE_INUM) {
337 			error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
338 					      l, &vp);
339 			if (error) {
340 				DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc"
341 				      " returned %d\n", error));
342 				continue;
343 			}
344 			ip = VTOI(vp);
345 			if (dip->di_size != ip->i_size)
346 				lfs_truncate(vp, dip->di_size, 0, NOCRED, l);
347 			/* Get mode, link count, size, and times */
348 			memcpy(ip->i_din.ffs1_din, dip,
349 			       offsetof(struct ufs1_dinode, di_db[0]));
350 
351 			/* Then the rest, except di_blocks */
352 			ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
353 			ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
354 			ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
355 			ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
356 
357 			ip->i_mode = ip->i_ffs1_mode;
358 			ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink;
359 			ip->i_size = ip->i_ffs1_size;
360 
361 			LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
362 
363 			/* Re-initialize to get type right */
364 			ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
365 				  &vp);
366 			vput(vp);
367 
368 			/* Record change in location */
369 			LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
370 			daddr = ifp->if_daddr;
371 			ifp->if_daddr = dbtofsb(fs, dbp->b_blkno);
372 			error = LFS_BWRITE_LOG(ibp); /* Ifile */
373 			/* And do segment accounting */
374 			if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) {
375 				if (daddr > 0) {
376 					LFS_SEGENTRY(sup, fs, dtosn(fs, daddr),
377 						     ibp);
378 					sup->su_nbytes -= sizeof (struct ufs1_dinode);
379 					LFS_WRITESEGENTRY(sup, fs,
380 							  dtosn(fs, daddr),
381 							  ibp);
382 				}
383 				LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
384 					     ibp);
385 				sup->su_nbytes += sizeof (struct ufs1_dinode);
386 				LFS_WRITESEGENTRY(sup, fs,
387 						  dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
388 						  ibp);
389 			}
390 		}
391 	}
392 	brelse(dbp, BC_AGE);
393 
394 	return 0;
395 }
396 
397 #define CHECK_CKSUM   0x0001  /* Check the checksum to make sure it's valid */
398 #define CHECK_UPDATE  0x0002  /* Update Ifile for new data blocks / inodes */
399 
400 static daddr_t
401 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial,
402 	     kauth_cred_t cred, int flags, int *pseg_flags, struct lwp *l)
403 {
404 	struct vnode *devvp;
405 	struct buf *bp, *dbp;
406 	int error, nblocks = 0, ninos, i, j; /* XXX: gcc */
407 	SEGSUM *ssp;
408 	u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */
409 	daddr_t oldoffset;
410 	int32_t *iaddr;	/* XXX ondisk32 */
411 	FINFO *fip;
412 	SEGUSE *sup;
413 	size_t size;
414 
415 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
416 	/*
417 	 * If the segment has a superblock and we're at the top
418 	 * of the segment, skip the superblock.
419 	 */
420 	if (sntod(fs, dtosn(fs, offset)) == offset) {
421 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
422 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
423 			offset += btofsb(fs, LFS_SBPAD);
424 		brelse(bp, 0);
425 	}
426 
427 	/* Read in the segment summary */
428 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp);
429 	if (error)
430 		return -1;
431 
432 	/* Check summary checksum */
433 	ssp = (SEGSUM *)bp->b_data;
434 	if (flags & CHECK_CKSUM) {
435 		if (ssp->ss_sumsum != cksum(&ssp->ss_datasum,
436 					   fs->lfs_sumsize -
437 					   sizeof(ssp->ss_sumsum))) {
438 			DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset));
439 			offset = -1;
440 			goto err1;
441 		}
442 		if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) {
443 			DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset));
444 			offset = -1;
445 			goto err1;
446 		}
447 		if (ssp->ss_create < fs->lfs_tstamp) {
448 			DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset));
449 			offset = -1;
450 			goto err1;
451 		}
452 	}
453 	if (fs->lfs_version > 1) {
454 		if (ssp->ss_serial != nextserial) {
455 			DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64
456 			      "\n", offset));
457 			offset = -1;
458 			goto err1;
459 		}
460 		if (ssp->ss_ident != fs->lfs_ident) {
461 			DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%"
462 			      PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset));
463 			offset = -1;
464 			goto err1;
465 		}
466 	}
467 	if (pseg_flags)
468 		*pseg_flags = ssp->ss_flags;
469 	oldoffset = offset;
470 	offset += btofsb(fs, fs->lfs_sumsize);
471 
472 	ninos = howmany(ssp->ss_ninos, INOPB(fs));
473 	/* XXX ondisk32 */
474 	iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t));
475 	if (flags & CHECK_CKSUM) {
476 		/* Count blocks */
477 		nblocks = 0;
478 		fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
479 		for (i = 0; i < ssp->ss_nfinfo; ++i) {
480 			nblocks += fip->fi_nblocks;
481 			if (fip->fi_nblocks <= 0)
482 				break;
483 			/* XXX ondisk32 */
484 			fip = (FINFO *)(((char *)fip) + FINFOSIZE +
485 					(fip->fi_nblocks * sizeof(int32_t)));
486 		}
487 		nblocks += ninos;
488 		/* Create the sum array */
489 		datap = dp = (u_long *)malloc(nblocks * sizeof(u_long),
490 					      M_SEGMENT, M_WAITOK);
491 	}
492 
493 	/* Handle individual blocks */
494 	fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
495 	for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) {
496 		/* Inode block? */
497 		if (ninos && *iaddr == offset) {
498 			if (flags & CHECK_CKSUM) {
499 				/* Read in the head and add to the buffer */
500 				error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize,
501 					      cred, &dbp);
502 				if (error) {
503 					offset = -1;
504 					goto err2;
505 				}
506 				(*dp++) = ((u_long *)(dbp->b_data))[0];
507 				brelse(dbp, BC_AGE);
508 			}
509 			if (flags & CHECK_UPDATE) {
510 				if ((error = update_inoblk(fs, offset, cred, l))
511 				    != 0) {
512 					offset = -1;
513 					goto err2;
514 				}
515 			}
516 			offset += btofsb(fs, fs->lfs_ibsize);
517 			--iaddr;
518 			--ninos;
519 			--i; /* compensate */
520 			continue;
521 		}
522 		size = fs->lfs_bsize;
523 		for (j = 0; j < fip->fi_nblocks; ++j) {
524 			if (j == fip->fi_nblocks - 1)
525 				size = fip->fi_lastlength;
526 			if (flags & CHECK_CKSUM) {
527 				error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp);
528 				if (error) {
529 					offset = -1;
530 					goto err2;
531 				}
532 				(*dp++) = ((u_long *)(dbp->b_data))[0];
533 				brelse(dbp, BC_AGE);
534 			}
535 			/* Account for and update any direct blocks */
536 			if ((flags & CHECK_UPDATE) &&
537 			   fip->fi_ino > LFS_IFILE_INUM &&
538 			   fip->fi_blocks[j] >= 0) {
539 				update_meta(fs, fip->fi_ino, fip->fi_version,
540 					    fip->fi_blocks[j], offset, size, l);
541 			}
542 			offset += btofsb(fs, size);
543 		}
544 		/* XXX ondisk32 */
545 		fip = (FINFO *)(((char *)fip) + FINFOSIZE
546 				+ fip->fi_nblocks * sizeof(int32_t));
547 	}
548 	/* Checksum the array, compare */
549 	if ((flags & CHECK_CKSUM) &&
550 	   ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long)))
551 	{
552 		DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64
553 		      " (wanted %x got %x)\n",
554 		      offset, ssp->ss_datasum, cksum(datap, nblocks *
555 						     sizeof(u_long))));
556 		offset = -1;
557 		goto err2;
558 	}
559 
560 	/* If we're at the end of the segment, move to the next */
561 	if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) !=
562 	   dtosn(fs, offset)) {
563 		if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) {
564 			offset = -1;
565 			goto err2;
566 		}
567 		offset = ssp->ss_next;
568 		DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64
569 		       " -> segment %d\n", offset, dtosn(fs,offset)));
570 	}
571 
572 	if (flags & CHECK_UPDATE) {
573 		fs->lfs_avail -= (offset - oldoffset);
574 		/* Don't clog the buffer queue */
575 		simple_lock(&lfs_subsys_lock);
576 		if (locked_queue_count > LFS_MAX_BUFS ||
577 		    locked_queue_bytes > LFS_MAX_BYTES) {
578 			lfs_flush(fs, SEGM_CKP, 0);
579 		}
580 		simple_unlock(&lfs_subsys_lock);
581 	}
582 
583     err2:
584 	if (flags & CHECK_CKSUM)
585 		free(datap, M_SEGMENT);
586     err1:
587 	bp->b_flags |= B_AGE;
588 	brelse(bp);
589 
590 	/* XXX should we update the serial number even for bad psegs? */
591 	if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1)
592 		fs->lfs_serial = nextserial;
593 	return offset;
594 }
595 
596 void
597 lfs_roll_forward(struct lfs *fs, struct mount *mp, struct lwp *l)
598 {
599 	int flags, dirty;
600 	daddr_t offset, oldoffset, lastgoodpseg;
601 	int sn, curseg, do_rollforward;
602 	struct proc *p;
603 	kauth_cred_t cred;
604 	SEGUSE *sup;
605 	struct buf *bp;
606 
607 	p = l ? l->l_proc : NULL;
608 	cred = p ? p->p_cred : NOCRED;
609 
610 	/*
611 	 * Roll forward.
612 	 *
613 	 * We don't roll forward for v1 filesystems, because
614 	 * of the danger that the clock was turned back between the last
615 	 * checkpoint and crash.  This would roll forward garbage.
616 	 *
617 	 * v2 filesystems don't have this problem because they use a
618 	 * monotonically increasing serial number instead of a timestamp.
619 	 */
620 	do_rollforward = (!(fs->lfs_pflags & LFS_PF_CLEAN) &&
621 			  lfs_do_rfw && fs->lfs_version > 1 && p != NULL);
622 	if (do_rollforward) {
623 		u_int64_t nextserial;
624 		/*
625 		 * Phase I: Find the address of the last good partial
626 		 * segment that was written after the checkpoint.  Mark
627 		 * the segments in question dirty, so they won't be
628 		 * reallocated.
629 		 */
630 		lastgoodpseg = oldoffset = offset = fs->lfs_offset;
631 		flags = 0x0;
632 		DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%"
633 		      PRIx64 "\n", offset));
634 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
635 		if (!(sup->su_flags & SEGUSE_DIRTY))
636 			--fs->lfs_nclean;
637 		sup->su_flags |= SEGUSE_DIRTY;
638 		LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp);
639 		nextserial = fs->lfs_serial + 1;
640 		while ((offset = check_segsum(fs, offset, nextserial,
641 		    cred, CHECK_CKSUM, &flags, l)) > 0) {
642 			nextserial++;
643 			if (sntod(fs, oldoffset) != sntod(fs, offset)) {
644 				LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset),
645 					     bp);
646 				if (!(sup->su_flags & SEGUSE_DIRTY))
647 					--fs->lfs_nclean;
648 				sup->su_flags |= SEGUSE_DIRTY;
649 				LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset),
650 					     bp);
651 			}
652 
653 			DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%"
654 			      PRIx64 "\n", offset));
655 			if (flags & SS_DIROP) {
656 				DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%"
657 				      PRIx64 "\n", oldoffset));
658 				if (!(flags & SS_CONT))
659 				     DLOG((DLOG_RF, "lfs_mountfs: dirops end "
660 					   "at 0x%" PRIx64 "\n", oldoffset));
661 			}
662 			if (!(flags & SS_CONT))
663 				lastgoodpseg = offset;
664 			oldoffset = offset;
665 		}
666 		if (flags & SS_CONT) {
667 			DLOG((DLOG_RF, "LFS roll forward: warning: incomplete "
668 			      "dirops discarded\n"));
669 		}
670 		DLOG((DLOG_RF, "LFS roll forward phase 1: completed: "
671 		      "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg));
672 		oldoffset = fs->lfs_offset;
673 		if (fs->lfs_offset != lastgoodpseg) {
674 			/* Don't overwrite what we're trying to preserve */
675 			offset = fs->lfs_offset;
676 			fs->lfs_offset = lastgoodpseg;
677 			fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset));
678 			for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) {
679 				sn = (sn + 1) % fs->lfs_nseg;
680 				if (sn == curseg)
681 					panic("lfs_mountfs: no clean segments");
682 				LFS_SEGENTRY(sup, fs, sn, bp);
683 				dirty = (sup->su_flags & SEGUSE_DIRTY);
684 				brelse(bp, 0);
685 				if (!dirty)
686 					break;
687 			}
688 			fs->lfs_nextseg = sntod(fs, sn);
689 
690 			/*
691 			 * Phase II: Roll forward from the first superblock.
692 			 */
693 			while (offset != lastgoodpseg) {
694 				DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%"
695 				      PRIx64 "\n", offset));
696 				offset = check_segsum(fs, offset,
697 				    fs->lfs_serial + 1, cred, CHECK_UPDATE,
698 				    NULL, l);
699 			}
700 
701 			/*
702 			 * Finish: flush our changes to disk.
703 			 */
704 			lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
705 			DLOG((DLOG_RF, "lfs_mountfs: roll forward ",
706 			      "recovered %lld blocks\n",
707 			      (long long)(lastgoodpseg - oldoffset)));
708 		}
709 		DLOG((DLOG_RF, "LFS roll forward complete\n"));
710 	}
711 }
712 #endif /* LFS_KERNEL_RFW */
713