xref: /netbsd-src/sys/ufs/lfs/lfs_rfw.c (revision a5847cc334d9a7029f6352b847e9e8d71a0f9e0c)
1 /*	$NetBSD: lfs_rfw.c,v 1.12 2009/02/22 20:28:07 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: lfs_rfw.c,v 1.12 2009/02/22 20:28:07 ad Exp $");
34 
35 #if defined(_KERNEL_OPT)
36 #include "opt_quota.h"
37 #endif
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/namei.h>
42 #include <sys/proc.h>
43 #include <sys/kernel.h>
44 #include <sys/vnode.h>
45 #include <sys/mount.h>
46 #include <sys/kthread.h>
47 #include <sys/buf.h>
48 #include <sys/device.h>
49 #include <sys/mbuf.h>
50 #include <sys/file.h>
51 #include <sys/disklabel.h>
52 #include <sys/ioctl.h>
53 #include <sys/errno.h>
54 #include <sys/malloc.h>
55 #include <sys/pool.h>
56 #include <sys/socket.h>
57 #include <sys/syslog.h>
58 #include <uvm/uvm_extern.h>
59 #include <sys/sysctl.h>
60 #include <sys/conf.h>
61 #include <sys/kauth.h>
62 
63 #include <miscfs/specfs/specdev.h>
64 
65 #include <ufs/ufs/quota.h>
66 #include <ufs/ufs/inode.h>
67 #include <ufs/ufs/ufsmount.h>
68 #include <ufs/ufs/ufs_extern.h>
69 
70 #include <uvm/uvm.h>
71 #include <uvm/uvm_stat.h>
72 #include <uvm/uvm_pager.h>
73 #include <uvm/uvm_pdaemon.h>
74 
75 #include <ufs/lfs/lfs.h>
76 #include <ufs/lfs/lfs_extern.h>
77 
78 #include <miscfs/genfs/genfs.h>
79 #include <miscfs/genfs/genfs_node.h>
80 
81 /*
82  * Roll-forward code.
83  */
84 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t,
85     kauth_cred_t, int, int *, struct lwp *);
86 
87 extern int lfs_do_rfw;
88 
89 /*
90  * Allocate a particular inode with a particular version number, freeing
91  * any previous versions of this inode that may have gone before.
92  * Used by the roll-forward code.
93  *
94  * XXX this function does not have appropriate locking to be used on a live fs;
95  * XXX but something similar could probably be used for an "undelete" call.
96  *
97  * Called with the Ifile inode locked.
98  */
99 int
100 lfs_rf_valloc(struct lfs *fs, ino_t ino, int vers, struct lwp *l,
101 	      struct vnode **vpp)
102 {
103 	IFILE *ifp;
104 	struct buf *bp, *cbp;
105 	struct vnode *vp;
106 	struct inode *ip;
107 	ino_t tino, oldnext;
108 	int error;
109 	CLEANERINFO *cip;
110 
111 	ASSERT_SEGLOCK(fs); /* XXX it doesn't, really */
112 
113 	/*
114 	 * First, just try a vget. If the version number is the one we want,
115 	 * we don't have to do anything else.  If the version number is wrong,
116 	 * take appropriate action.
117 	 */
118 	error = VFS_VGET(fs->lfs_ivnode->v_mount, ino, &vp);
119 	if (error == 0) {
120 		DLOG((DLOG_RF, "lfs_rf_valloc[1]: ino %d vp %p\n", ino, vp));
121 
122 		*vpp = vp;
123 		ip = VTOI(vp);
124 		if (ip->i_gen == vers)
125 			return 0;
126 		else if (ip->i_gen < vers) {
127 			lfs_truncate(vp, (off_t)0, 0, NOCRED);
128 			ip->i_gen = ip->i_ffs1_gen = vers;
129 			LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
130 			return 0;
131 		} else {
132 			DLOG((DLOG_RF, "ino %d: sought version %d, got %d\n",
133 			       ino, vers, ip->i_ffs1_gen));
134 			vput(vp);
135 			*vpp = NULLVP;
136 			return EEXIST;
137 		}
138 	}
139 
140 	/*
141 	 * The inode is not in use.  Find it on the free list.
142 	 */
143 	/* If the Ifile is too short to contain this inum, extend it */
144 	while (VTOI(fs->lfs_ivnode)->i_size <= (ino /
145 		fs->lfs_ifpb + fs->lfs_cleansz + fs->lfs_segtabsz)
146 		<< fs->lfs_bshift) {
147 		lfs_extend_ifile(fs, NOCRED);
148 	}
149 
150 	LFS_IENTRY(ifp, fs, ino, bp);
151 	oldnext = ifp->if_nextfree;
152 	ifp->if_version = vers;
153 	brelse(bp, 0);
154 
155 	LFS_GET_HEADFREE(fs, cip, cbp, &ino);
156 	if (ino) {
157 		LFS_PUT_HEADFREE(fs, cip, cbp, oldnext);
158 	} else {
159 		tino = ino;
160 		while (1) {
161 			LFS_IENTRY(ifp, fs, tino, bp);
162 			if (ifp->if_nextfree == ino ||
163 			    ifp->if_nextfree == LFS_UNUSED_INUM)
164 				break;
165 			tino = ifp->if_nextfree;
166 			brelse(bp, 0);
167 		}
168 		if (ifp->if_nextfree == LFS_UNUSED_INUM) {
169 			brelse(bp, 0);
170 			return ENOENT;
171 		}
172 		ifp->if_nextfree = oldnext;
173 		LFS_BWRITE_LOG(bp);
174 	}
175 
176 	error = lfs_ialloc(fs, fs->lfs_ivnode, ino, vers, &vp);
177 	if (error == 0) {
178 		/*
179 		 * Make it VREG so we can put blocks on it.  We will change
180 		 * this later if it turns out to be some other kind of file.
181 		 */
182 		ip = VTOI(vp);
183 		ip->i_mode = ip->i_ffs1_mode = IFREG;
184 		ip->i_nlink = ip->i_ffs1_nlink = 1;
185 		ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, &vp);
186 		ip = VTOI(vp);
187 
188 		DLOG((DLOG_RF, "lfs_rf_valloc: ino %d vp %p\n", ino, vp));
189 
190 		/* The dirop-nature of this vnode is past */
191 		lfs_unmark_vnode(vp);
192 		(void)lfs_vunref(vp);
193 		vp->v_uflag &= ~VU_DIROP;
194 		mutex_enter(&lfs_lock);
195 		--lfs_dirvcount;
196 		--fs->lfs_dirvcount;
197 		TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
198 		wakeup(&lfs_dirvcount);
199 		wakeup(&fs->lfs_dirvcount);
200 		mutex_exit(&lfs_lock);
201 	}
202 	*vpp = vp;
203 	return error;
204 }
205 
206 /*
207  * Load the appropriate indirect block, and change the appropriate pointer.
208  * Mark the block dirty.  Do segment and avail accounting.
209  */
210 static int
211 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn,
212 	    daddr_t ndaddr, size_t size, struct lwp *l)
213 {
214 	int error;
215 	struct vnode *vp;
216 	struct inode *ip;
217 #ifdef DEBUG
218 	daddr_t odaddr;
219 	struct indir a[NIADDR];
220 	int num;
221 	int i;
222 #endif /* DEBUG */
223 	struct buf *bp;
224 	SEGUSE *sup;
225 
226 	KASSERT(lbn >= 0);	/* no indirect blocks */
227 
228 	if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) {
229 		DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc"
230 		      " returned %d\n", ino, error));
231 		return error;
232 	}
233 
234 	if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size,
235 				NOCRED, 0, &bp)) != 0) {
236 		vput(vp);
237 		return (error);
238 	}
239 	/* No need to write, the block is already on disk */
240 	if (bp->b_oflags & BO_DELWRI) {
241 		LFS_UNLOCK_BUF(bp);
242 		fs->lfs_avail += btofsb(fs, bp->b_bcount);
243 	}
244 	brelse(bp, BC_INVAL);
245 
246 	/*
247 	 * Extend the file, if it is not large enough already.
248 	 * XXX this is not exactly right, we don't know how much of the
249 	 * XXX last block is actually used.  We hope that an inode will
250 	 * XXX appear later to give the correct size.
251 	 */
252 	ip = VTOI(vp);
253 	if (ip->i_size <= (lbn << fs->lfs_bshift)) {
254 		u_int64_t newsize;
255 
256 		if (lbn < NDADDR)
257 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) +
258 				(size - fs->lfs_fsize) + 1;
259 		else
260 			newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1;
261 
262 		if (ip->i_size < newsize) {
263 			ip->i_size = newsize;
264 			/*
265 			 * tell vm our new size for the case the inode won't
266 			 * appear later.
267 			 */
268 			uvm_vnp_setsize(vp, newsize);
269 		}
270 	}
271 
272 	lfs_update_single(fs, NULL, vp, lbn, ndaddr, size);
273 
274 	LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
275 	sup->su_nbytes += size;
276 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
277 
278 	/* differences here should be due to UNWRITTEN indirect blocks. */
279 	KASSERT((lblkno(fs, ip->i_size) > NDADDR &&
280 	    ip->i_lfs_effnblks == ip->i_ffs1_blocks) ||
281 	    ip->i_lfs_effnblks >= ip->i_ffs1_blocks);
282 
283 #ifdef DEBUG
284 	/* Now look again to make sure it worked */
285 	ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
286 	for (i = num; i > 0; i--) {
287 		if (!a[i].in_exists)
288 			panic("update_meta: absent %d lv indirect block", i);
289 	}
290 	if (dbtofsb(fs, odaddr) != ndaddr)
291 		DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %"
292 		      PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr));
293 #endif /* DEBUG */
294 	vput(vp);
295 	return 0;
296 }
297 
298 static int
299 update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred,
300 	      struct lwp *l)
301 {
302 	struct vnode *devvp, *vp;
303 	struct inode *ip;
304 	struct ufs1_dinode *dip;
305 	struct buf *dbp, *ibp;
306 	int error;
307 	daddr_t daddr;
308 	IFILE *ifp;
309 	SEGUSE *sup;
310 
311 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
312 
313 	/*
314 	 * Get the inode, update times and perms.
315 	 * DO NOT update disk blocks, we do that separately.
316 	 */
317 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize,
318 	    cred, 0, &dbp);
319 	if (error) {
320 		DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error));
321 		return error;
322 	}
323 	dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs);
324 	while (--dip >= (struct ufs1_dinode *)dbp->b_data) {
325 		if (dip->di_inumber > LFS_IFILE_INUM) {
326 			error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
327 					      l, &vp);
328 			if (error) {
329 				DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc"
330 				      " returned %d\n", error));
331 				continue;
332 			}
333 			ip = VTOI(vp);
334 			if (dip->di_size != ip->i_size)
335 				lfs_truncate(vp, dip->di_size, 0, NOCRED);
336 			/* Get mode, link count, size, and times */
337 			memcpy(ip->i_din.ffs1_din, dip,
338 			       offsetof(struct ufs1_dinode, di_db[0]));
339 
340 			/* Then the rest, except di_blocks */
341 			ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
342 			ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
343 			ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
344 			ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
345 
346 			ip->i_mode = ip->i_ffs1_mode;
347 			ip->i_nlink = ip->i_ffs1_nlink;
348 			ip->i_size = ip->i_ffs1_size;
349 
350 			LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
351 
352 			/* Re-initialize to get type right */
353 			ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
354 				  &vp);
355 			vput(vp);
356 
357 			/* Record change in location */
358 			LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
359 			daddr = ifp->if_daddr;
360 			ifp->if_daddr = dbtofsb(fs, dbp->b_blkno);
361 			error = LFS_BWRITE_LOG(ibp); /* Ifile */
362 			/* And do segment accounting */
363 			if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) {
364 				if (daddr > 0) {
365 					LFS_SEGENTRY(sup, fs, dtosn(fs, daddr),
366 						     ibp);
367 					sup->su_nbytes -= sizeof (struct ufs1_dinode);
368 					LFS_WRITESEGENTRY(sup, fs,
369 							  dtosn(fs, daddr),
370 							  ibp);
371 				}
372 				LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
373 					     ibp);
374 				sup->su_nbytes += sizeof (struct ufs1_dinode);
375 				LFS_WRITESEGENTRY(sup, fs,
376 						  dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
377 						  ibp);
378 			}
379 		}
380 	}
381 	brelse(dbp, BC_AGE);
382 
383 	return 0;
384 }
385 
386 #define CHECK_CKSUM   0x0001  /* Check the checksum to make sure it's valid */
387 #define CHECK_UPDATE  0x0002  /* Update Ifile for new data blocks / inodes */
388 
389 static daddr_t
390 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial,
391 	     kauth_cred_t cred, int flags, int *pseg_flags, struct lwp *l)
392 {
393 	struct vnode *devvp;
394 	struct buf *bp, *dbp;
395 	int error, nblocks = 0, ninos, i, j; /* XXX: gcc */
396 	SEGSUM *ssp;
397 	u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */
398 	daddr_t oldoffset;
399 	int32_t *iaddr;	/* XXX ondisk32 */
400 	FINFO *fip;
401 	SEGUSE *sup;
402 	size_t size;
403 
404 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
405 	/*
406 	 * If the segment has a superblock and we're at the top
407 	 * of the segment, skip the superblock.
408 	 */
409 	if (sntod(fs, dtosn(fs, offset)) == offset) {
410 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
411 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
412 			offset += btofsb(fs, LFS_SBPAD);
413 		brelse(bp, 0);
414 	}
415 
416 	/* Read in the segment summary */
417 	error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize,
418 	    cred, 0, &bp);
419 	if (error)
420 		return -1;
421 
422 	/* Check summary checksum */
423 	ssp = (SEGSUM *)bp->b_data;
424 	if (flags & CHECK_CKSUM) {
425 		if (ssp->ss_sumsum != cksum(&ssp->ss_datasum,
426 					   fs->lfs_sumsize -
427 					   sizeof(ssp->ss_sumsum))) {
428 			DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset));
429 			offset = -1;
430 			goto err1;
431 		}
432 		if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) {
433 			DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset));
434 			offset = -1;
435 			goto err1;
436 		}
437 		if (ssp->ss_create < fs->lfs_tstamp) {
438 			DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset));
439 			offset = -1;
440 			goto err1;
441 		}
442 	}
443 	if (fs->lfs_version > 1) {
444 		if (ssp->ss_serial != nextserial) {
445 			DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64
446 			      "\n", offset));
447 			offset = -1;
448 			goto err1;
449 		}
450 		if (ssp->ss_ident != fs->lfs_ident) {
451 			DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%"
452 			      PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset));
453 			offset = -1;
454 			goto err1;
455 		}
456 	}
457 	if (pseg_flags)
458 		*pseg_flags = ssp->ss_flags;
459 	oldoffset = offset;
460 	offset += btofsb(fs, fs->lfs_sumsize);
461 
462 	ninos = howmany(ssp->ss_ninos, INOPB(fs));
463 	/* XXX ondisk32 */
464 	iaddr = (int32_t *)((char*)bp->b_data + fs->lfs_sumsize - sizeof(int32_t));
465 	if (flags & CHECK_CKSUM) {
466 		/* Count blocks */
467 		nblocks = 0;
468 		fip = (FINFO *)((char*)bp->b_data + SEGSUM_SIZE(fs));
469 		for (i = 0; i < ssp->ss_nfinfo; ++i) {
470 			nblocks += fip->fi_nblocks;
471 			if (fip->fi_nblocks <= 0)
472 				break;
473 			/* XXX ondisk32 */
474 			fip = (FINFO *)(((char *)fip) + FINFOSIZE +
475 					(fip->fi_nblocks * sizeof(int32_t)));
476 		}
477 		nblocks += ninos;
478 		/* Create the sum array */
479 		datap = dp = (u_long *)malloc(nblocks * sizeof(u_long),
480 					      M_SEGMENT, M_WAITOK);
481 	}
482 
483 	/* Handle individual blocks */
484 	fip = (FINFO *)((char*)bp->b_data + SEGSUM_SIZE(fs));
485 	for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) {
486 		/* Inode block? */
487 		if (ninos && *iaddr == offset) {
488 			if (flags & CHECK_CKSUM) {
489 				/* Read in the head and add to the buffer */
490 				error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize,
491 					      cred, 0, &dbp);
492 				if (error) {
493 					offset = -1;
494 					goto err2;
495 				}
496 				(*dp++) = ((u_long *)(dbp->b_data))[0];
497 				brelse(dbp, BC_AGE);
498 			}
499 			if (flags & CHECK_UPDATE) {
500 				if ((error = update_inoblk(fs, offset, cred, l))
501 				    != 0) {
502 					offset = -1;
503 					goto err2;
504 				}
505 			}
506 			offset += btofsb(fs, fs->lfs_ibsize);
507 			--iaddr;
508 			--ninos;
509 			--i; /* compensate */
510 			continue;
511 		}
512 		size = fs->lfs_bsize;
513 		for (j = 0; j < fip->fi_nblocks; ++j) {
514 			if (j == fip->fi_nblocks - 1)
515 				size = fip->fi_lastlength;
516 			if (flags & CHECK_CKSUM) {
517 				error = bread(devvp, fsbtodb(fs, offset), size,
518 				    cred, 0, &dbp);
519 				if (error) {
520 					offset = -1;
521 					goto err2;
522 				}
523 				(*dp++) = ((u_long *)(dbp->b_data))[0];
524 				brelse(dbp, BC_AGE);
525 			}
526 			/* Account for and update any direct blocks */
527 			if ((flags & CHECK_UPDATE) &&
528 			   fip->fi_ino > LFS_IFILE_INUM &&
529 			   fip->fi_blocks[j] >= 0) {
530 				update_meta(fs, fip->fi_ino, fip->fi_version,
531 					    fip->fi_blocks[j], offset, size, l);
532 			}
533 			offset += btofsb(fs, size);
534 		}
535 		/* XXX ondisk32 */
536 		fip = (FINFO *)(((char *)fip) + FINFOSIZE
537 				+ fip->fi_nblocks * sizeof(int32_t));
538 	}
539 	/* Checksum the array, compare */
540 	if ((flags & CHECK_CKSUM) &&
541 	   ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long)))
542 	{
543 		DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64
544 		      " (wanted %x got %x)\n",
545 		      offset, ssp->ss_datasum, cksum(datap, nblocks *
546 						     sizeof(u_long))));
547 		offset = -1;
548 		goto err2;
549 	}
550 
551 	/* If we're at the end of the segment, move to the next */
552 	if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) !=
553 	   dtosn(fs, offset)) {
554 		if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) {
555 			offset = -1;
556 			goto err2;
557 		}
558 		offset = ssp->ss_next;
559 		DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64
560 		       " -> segment %d\n", offset, dtosn(fs,offset)));
561 	}
562 
563 	if (flags & CHECK_UPDATE) {
564 		fs->lfs_avail -= (offset - oldoffset);
565 		/* Don't clog the buffer queue */
566 		mutex_enter(&lfs_lock);
567 		if (locked_queue_count > LFS_MAX_BUFS ||
568 		    locked_queue_bytes > LFS_MAX_BYTES) {
569 			lfs_flush(fs, SEGM_CKP, 0);
570 		}
571 		mutex_exit(&lfs_lock);
572 	}
573 
574     err2:
575 	if (flags & CHECK_CKSUM)
576 		free(datap, M_SEGMENT);
577     err1:
578 	brelse(bp, BC_AGE);
579 
580 	/* XXX should we update the serial number even for bad psegs? */
581 	if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1)
582 		fs->lfs_serial = nextserial;
583 	return offset;
584 }
585 
586 void
587 lfs_roll_forward(struct lfs *fs, struct mount *mp, struct lwp *l)
588 {
589 	int flags, dirty;
590 	daddr_t offset, oldoffset, lastgoodpseg;
591 	int sn, curseg, do_rollforward;
592 	struct proc *p;
593 	kauth_cred_t cred;
594 	SEGUSE *sup;
595 	struct buf *bp;
596 
597 	p = l ? l->l_proc : NULL;
598 	cred = p ? p->p_cred : NOCRED;
599 
600 	/*
601 	 * Roll forward.
602 	 *
603 	 * We don't roll forward for v1 filesystems, because
604 	 * of the danger that the clock was turned back between the last
605 	 * checkpoint and crash.  This would roll forward garbage.
606 	 *
607 	 * v2 filesystems don't have this problem because they use a
608 	 * monotonically increasing serial number instead of a timestamp.
609 	 */
610 	do_rollforward = (!(fs->lfs_pflags & LFS_PF_CLEAN) &&
611 			  lfs_do_rfw && fs->lfs_version > 1 && p != NULL);
612 	if (do_rollforward) {
613 		u_int64_t nextserial;
614 		/*
615 		 * Phase I: Find the address of the last good partial
616 		 * segment that was written after the checkpoint.  Mark
617 		 * the segments in question dirty, so they won't be
618 		 * reallocated.
619 		 */
620 		lastgoodpseg = oldoffset = offset = fs->lfs_offset;
621 		flags = 0x0;
622 		DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%"
623 		      PRIx64 "\n", offset));
624 		LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
625 		if (!(sup->su_flags & SEGUSE_DIRTY))
626 			--fs->lfs_nclean;
627 		sup->su_flags |= SEGUSE_DIRTY;
628 		LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp);
629 		nextserial = fs->lfs_serial + 1;
630 		while ((offset = check_segsum(fs, offset, nextserial,
631 		    cred, CHECK_CKSUM, &flags, l)) > 0) {
632 			nextserial++;
633 			if (sntod(fs, oldoffset) != sntod(fs, offset)) {
634 				LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset),
635 					     bp);
636 				if (!(sup->su_flags & SEGUSE_DIRTY))
637 					--fs->lfs_nclean;
638 				sup->su_flags |= SEGUSE_DIRTY;
639 				LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset),
640 					     bp);
641 			}
642 
643 			DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%"
644 			      PRIx64 "\n", offset));
645 			if (flags & SS_DIROP) {
646 				DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%"
647 				      PRIx64 "\n", oldoffset));
648 				if (!(flags & SS_CONT)) {
649 				     DLOG((DLOG_RF, "lfs_mountfs: dirops end "
650 					   "at 0x%" PRIx64 "\n", oldoffset));
651 				}
652 			}
653 			if (!(flags & SS_CONT))
654 				lastgoodpseg = offset;
655 			oldoffset = offset;
656 		}
657 		if (flags & SS_CONT) {
658 			DLOG((DLOG_RF, "LFS roll forward: warning: incomplete "
659 			      "dirops discarded\n"));
660 		}
661 		DLOG((DLOG_RF, "LFS roll forward phase 1: completed: "
662 		      "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg));
663 		oldoffset = fs->lfs_offset;
664 		if (fs->lfs_offset != lastgoodpseg) {
665 			/* Don't overwrite what we're trying to preserve */
666 			offset = fs->lfs_offset;
667 			fs->lfs_offset = lastgoodpseg;
668 			fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset));
669 			for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) {
670 				sn = (sn + 1) % fs->lfs_nseg;
671 				if (sn == curseg)
672 					panic("lfs_mountfs: no clean segments");
673 				LFS_SEGENTRY(sup, fs, sn, bp);
674 				dirty = (sup->su_flags & SEGUSE_DIRTY);
675 				brelse(bp, 0);
676 				if (!dirty)
677 					break;
678 			}
679 			fs->lfs_nextseg = sntod(fs, sn);
680 
681 			/*
682 			 * Phase II: Roll forward from the first superblock.
683 			 */
684 			while (offset != lastgoodpseg) {
685 				DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%"
686 				      PRIx64 "\n", offset));
687 				offset = check_segsum(fs, offset,
688 				    fs->lfs_serial + 1, cred, CHECK_UPDATE,
689 				    NULL, l);
690 			}
691 
692 			/*
693 			 * Finish: flush our changes to disk.
694 			 */
695 			lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
696 			DLOG((DLOG_RF, "lfs_mountfs: roll forward ",
697 			      "recovered %lld blocks\n",
698 			      (long long)(lastgoodpseg - oldoffset)));
699 		}
700 		DLOG((DLOG_RF, "LFS roll forward complete\n"));
701 	}
702 }
703