xref: /minix/sys/ufs/lfs/lfs_rfw.c (revision 0a6a1f1d)
1 /*	$NetBSD: lfs_rfw.c,v 1.32 2015/10/03 08:27:55 dholland 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 1.32 2015/10/03 08:27:55 dholland 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/lfs/ulfs_quotacommon.h>
66 #include <ufs/lfs/ulfs_inode.h>
67 #include <ufs/lfs/ulfsmount.h>
68 #include <ufs/lfs/ulfs_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_accessors.h>
77 #include <ufs/lfs/lfs_kernel.h>
78 #include <ufs/lfs/lfs_extern.h>
79 
80 #include <miscfs/genfs/genfs.h>
81 #include <miscfs/genfs/genfs_node.h>
82 
83 /*
84  * Roll-forward code.
85  */
86 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t,
87     kauth_cred_t, int, int *, struct lwp *);
88 
89 extern int lfs_do_rfw;
90 
91 /*
92  * Allocate a particular inode with a particular version number, freeing
93  * any previous versions of this inode that may have gone before.
94  * Used by the roll-forward code.
95  *
96  * XXX this function does not have appropriate locking to be used on a live fs;
97  * XXX but something similar could probably be used for an "undelete" call.
98  *
99  * Called with the Ifile inode locked.
100  */
101 int
lfs_rf_valloc(struct lfs * fs,ino_t ino,int vers,struct lwp * l,struct vnode ** vpp)102 lfs_rf_valloc(struct lfs *fs, ino_t ino, int vers, struct lwp *l,
103 	      struct vnode **vpp)
104 {
105 	struct vattr va;
106 	struct vnode *vp;
107 	struct inode *ip;
108 	int error;
109 
110 	ASSERT_SEGLOCK(fs); /* XXX it doesn't, really */
111 
112 	/*
113 	 * First, just try a vget. If the version number is the one we want,
114 	 * we don't have to do anything else.  If the version number is wrong,
115 	 * take appropriate action.
116 	 */
117 	error = VFS_VGET(fs->lfs_ivnode->v_mount, ino, &vp);
118 	if (error == 0) {
119 		DLOG((DLOG_RF, "lfs_rf_valloc[1]: ino %d vp %p\n", ino, vp));
120 
121 		*vpp = vp;
122 		ip = VTOI(vp);
123 		if (ip->i_gen == vers)
124 			return 0;
125 		else if (ip->i_gen < vers) {
126 			lfs_truncate(vp, (off_t)0, 0, NOCRED);
127 			ip->i_gen = vers;
128 			lfs_dino_setgen(fs, ip->i_din, 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, lfs_dino_getgen(fs, ip->i_din)));
134 			vput(vp);
135 			*vpp = NULLVP;
136 			return EEXIST;
137 		}
138 	}
139 
140 	/* Not found, create as regular file. */
141 	vattr_null(&va);
142 	va.va_type = VREG;
143 	va.va_mode = 0;
144 	va.va_fileid = ino;
145 	va.va_gen = vers;
146 	error = vcache_new(fs->lfs_ivnode->v_mount, NULL, &va, NOCRED, &vp);
147 	if (error)
148 		return error;
149 	error = vn_lock(vp, LK_EXCLUSIVE);
150 	if (error) {
151 		vrele(vp);
152 		*vpp = NULLVP;
153 		return error;
154 	}
155 	ip = VTOI(vp);
156 	ip->i_nlink = 1;
157 	lfs_dino_setnlink(fs, ip->i_din, 1);
158 	*vpp = vp;
159 	return 0;
160 }
161 
162 /*
163  * Load the appropriate indirect block, and change the appropriate pointer.
164  * Mark the block dirty.  Do segment and avail accounting.
165  */
166 static int
update_meta(struct lfs * fs,ino_t ino,int vers,daddr_t lbn,daddr_t ndaddr,size_t size,struct lwp * l)167 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn,
168 	    daddr_t ndaddr, size_t size, struct lwp *l)
169 {
170 	int error;
171 	struct vnode *vp;
172 	struct inode *ip;
173 #ifdef DEBUG
174 	daddr_t odaddr;
175 	struct indir a[ULFS_NIADDR];
176 	int num;
177 	int i;
178 #endif /* DEBUG */
179 	struct buf *bp;
180 	SEGUSE *sup;
181 
182 	KASSERT(lbn >= 0);	/* no indirect blocks */
183 
184 	if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) {
185 		DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc"
186 		      " returned %d\n", ino, error));
187 		return error;
188 	}
189 
190 	if ((error = lfs_balloc(vp, (lbn << lfs_sb_getbshift(fs)), size,
191 				NOCRED, 0, &bp)) != 0) {
192 		vput(vp);
193 		return (error);
194 	}
195 	/* No need to write, the block is already on disk */
196 	if (bp->b_oflags & BO_DELWRI) {
197 		LFS_UNLOCK_BUF(bp);
198 		lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
199 		/* XXX should this wake up fs->lfs_availsleep? */
200 	}
201 	brelse(bp, BC_INVAL);
202 
203 	/*
204 	 * Extend the file, if it is not large enough already.
205 	 * XXX this is not exactly right, we don't know how much of the
206 	 * XXX last block is actually used.  We hope that an inode will
207 	 * XXX appear later to give the correct size.
208 	 */
209 	ip = VTOI(vp);
210 	if (ip->i_size <= (lbn << lfs_sb_getbshift(fs))) {
211 		u_int64_t newsize;
212 
213 		if (lbn < ULFS_NDADDR) {
214 			newsize = (lbn << lfs_sb_getbshift(fs)) +
215 				(size - lfs_sb_getfsize(fs)) + 1;
216 		} else {
217 			newsize = (lbn << lfs_sb_getbshift(fs)) + 1;
218 		}
219 		lfs_dino_setsize(fs, ip->i_din, newsize);
220 
221 		if (ip->i_size < newsize) {
222 			ip->i_size = newsize;
223 			/*
224 			 * tell vm our new size for the case the inode won't
225 			 * appear later.
226 			 */
227 			uvm_vnp_setsize(vp, newsize);
228 		}
229 	}
230 
231 	lfs_update_single(fs, NULL, vp, lbn, ndaddr, size);
232 
233 	LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, ndaddr), bp);
234 	sup->su_nbytes += size;
235 	LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, ndaddr), bp);
236 
237 	/* differences here should be due to UNWRITTEN indirect blocks. */
238 	KASSERT((lfs_lblkno(fs, ip->i_size) > ULFS_NDADDR &&
239 	    ip->i_lfs_effnblks == lfs_dino_getblocks(fs, ip->i_din)) ||
240 	    ip->i_lfs_effnblks >= lfs_dino_getblocks(fs, ip->i_din));
241 
242 #ifdef DEBUG
243 	/* Now look again to make sure it worked */
244 	ulfs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
245 	for (i = num; i > 0; i--) {
246 		if (!a[i].in_exists)
247 			panic("update_meta: absent %d lv indirect block", i);
248 	}
249 	if (LFS_DBTOFSB(fs, odaddr) != ndaddr)
250 		DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %"
251 		      PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr));
252 #endif /* DEBUG */
253 	vput(vp);
254 	return 0;
255 }
256 
257 /*
258  * Copy some the fields of the dinode as needed by update_inoblk().
259  */
260 static void
update_inoblk_copy_dinode(struct lfs * fs,union lfs_dinode * dstu,const union lfs_dinode * srcu)261 update_inoblk_copy_dinode(struct lfs *fs,
262     union lfs_dinode *dstu, const union lfs_dinode *srcu)
263 {
264 	if (fs->lfs_is64) {
265 		struct lfs64_dinode *dst = &dstu->u_64;
266 		const struct lfs64_dinode *src = &srcu->u_64;
267 		unsigned i;
268 
269 		/*
270 		 * Copy everything but the block pointers and di_blocks.
271 		 * XXX what about di_extb?
272 		 */
273 		dst->di_mode = src->di_mode;
274 		dst->di_nlink = src->di_nlink;
275 		dst->di_uid = src->di_uid;
276 		dst->di_gid = src->di_gid;
277 		dst->di_blksize = src->di_blksize;
278 		dst->di_size = src->di_size;
279 		dst->di_atime = src->di_atime;
280 		dst->di_mtime = src->di_mtime;
281 		dst->di_ctime = src->di_ctime;
282 		dst->di_birthtime = src->di_birthtime;
283 		dst->di_mtimensec = src->di_mtimensec;
284 		dst->di_atimensec = src->di_atimensec;
285 		dst->di_ctimensec = src->di_ctimensec;
286 		dst->di_birthnsec = src->di_birthnsec;
287 		dst->di_gen = src->di_gen;
288 		dst->di_kernflags = src->di_kernflags;
289 		dst->di_flags = src->di_flags;
290 		dst->di_extsize = src->di_extsize;
291 		dst->di_modrev = src->di_modrev;
292 		dst->di_inumber = src->di_inumber;
293 		for (i = 0; i < __arraycount(src->di_spare); i++) {
294 			dst->di_spare[i] = src->di_spare[i];
295 		}
296 	} else {
297 		struct lfs32_dinode *dst = &dstu->u_32;
298 		const struct lfs32_dinode *src = &srcu->u_32;
299 
300 		/* Get mode, link count, size, and times */
301 		memcpy(dst, src, offsetof(struct lfs32_dinode, di_db[0]));
302 
303 		/* Then the rest, except di_blocks */
304 		dst->di_flags = src->di_flags;
305 		dst->di_gen = src->di_gen;
306 		dst->di_uid = src->di_uid;
307 		dst->di_gid = src->di_gid;
308 		dst->di_modrev = src->di_modrev;
309 	}
310 }
311 
312 static int
update_inoblk(struct lfs * fs,daddr_t offset,kauth_cred_t cred,struct lwp * l)313 update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred,
314 	      struct lwp *l)
315 {
316 	struct vnode *devvp, *vp;
317 	struct inode *ip;
318 	union lfs_dinode *dip;
319 	struct buf *dbp, *ibp;
320 	int error;
321 	daddr_t daddr;
322 	IFILE *ifp;
323 	SEGUSE *sup;
324 	unsigned i, num;
325 
326 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
327 
328 	/*
329 	 * Get the inode, update times and perms.
330 	 * DO NOT update disk blocks, we do that separately.
331 	 */
332 	error = bread(devvp, LFS_FSBTODB(fs, offset), lfs_sb_getibsize(fs),
333 	    0, &dbp);
334 	if (error) {
335 		DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error));
336 		return error;
337 	}
338 	num = LFS_INOPB(fs);
339 	for (i = num; i-- > 0; ) {
340 		dip = DINO_IN_BLOCK(fs, dbp->b_data, i);
341 		if (lfs_dino_getinumber(fs, dip) > LFS_IFILE_INUM) {
342 			error = lfs_rf_valloc(fs, lfs_dino_getinumber(fs, dip),
343 					      lfs_dino_getgen(fs, dip),
344 					      l, &vp);
345 			if (error) {
346 				DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc"
347 				      " returned %d\n", error));
348 				continue;
349 			}
350 			ip = VTOI(vp);
351 			if (lfs_dino_getsize(fs, dip) != ip->i_size)
352 				lfs_truncate(vp, lfs_dino_getsize(fs, dip), 0,
353 					     NOCRED);
354 			update_inoblk_copy_dinode(fs, ip->i_din, dip);
355 
356 			ip->i_flags = lfs_dino_getflags(fs, dip);
357 			ip->i_gen = lfs_dino_getgen(fs, dip);
358 			ip->i_uid = lfs_dino_getuid(fs, dip);
359 			ip->i_gid = lfs_dino_getgid(fs, dip);
360 
361 			ip->i_mode = lfs_dino_getmode(fs, dip);
362 			ip->i_nlink = lfs_dino_getnlink(fs, dip);
363 			ip->i_size = lfs_dino_getsize(fs, dip);
364 
365 			LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
366 
367 			/* Re-initialize to get type right */
368 			ulfs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
369 				  &vp);
370 			vput(vp);
371 
372 			/* Record change in location */
373 			LFS_IENTRY(ifp, fs, lfs_dino_getinumber(fs, dip), ibp);
374 			daddr = lfs_if_getdaddr(fs, ifp);
375 			lfs_if_setdaddr(fs, ifp, LFS_DBTOFSB(fs, dbp->b_blkno));
376 			error = LFS_BWRITE_LOG(ibp); /* Ifile */
377 			/* And do segment accounting */
378 			if (lfs_dtosn(fs, daddr) != lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno))) {
379 				if (daddr > 0) {
380 					LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, daddr),
381 						     ibp);
382 					sup->su_nbytes -= DINOSIZE(fs);
383 					LFS_WRITESEGENTRY(sup, fs,
384 							  lfs_dtosn(fs, daddr),
385 							  ibp);
386 				}
387 				LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno)),
388 					     ibp);
389 				sup->su_nbytes += DINOSIZE(fs);
390 				LFS_WRITESEGENTRY(sup, fs,
391 						  lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno)),
392 						  ibp);
393 			}
394 		}
395 	}
396 	brelse(dbp, BC_AGE);
397 
398 	return 0;
399 }
400 
401 #define CHECK_CKSUM   0x0001  /* Check the checksum to make sure it's valid */
402 #define CHECK_UPDATE  0x0002  /* Update Ifile for new data blocks / inodes */
403 
404 static daddr_t
check_segsum(struct lfs * fs,daddr_t offset,u_int64_t nextserial,kauth_cred_t cred,int flags,int * pseg_flags,struct lwp * l)405 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial,
406 	     kauth_cred_t cred, int flags, int *pseg_flags, struct lwp *l)
407 {
408 	struct vnode *devvp;
409 	struct buf *bp, *dbp;
410 	int error, nblocks = 0, ninos, i, j; /* XXX: gcc */
411 	SEGSUM *ssp;
412 	u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */
413 	daddr_t oldoffset;
414 	IINFO *iip;
415 	FINFO *fip;
416 	SEGUSE *sup;
417 	size_t size;
418 	uint32_t datasum, foundsum;
419 
420 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
421 	/*
422 	 * If the segment has a superblock and we're at the top
423 	 * of the segment, skip the superblock.
424 	 */
425 	if (lfs_sntod(fs, lfs_dtosn(fs, offset)) == offset) {
426 		LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp);
427 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
428 			offset += lfs_btofsb(fs, LFS_SBPAD);
429 		brelse(bp, 0);
430 	}
431 
432 	/* Read in the segment summary */
433 	error = bread(devvp, LFS_FSBTODB(fs, offset), lfs_sb_getsumsize(fs),
434 	    0, &bp);
435 	if (error)
436 		return -1;
437 
438 	/* Check summary checksum */
439 	ssp = (SEGSUM *)bp->b_data;
440 	if (flags & CHECK_CKSUM) {
441 		size_t sumstart;
442 
443 		sumstart = lfs_ss_getsumstart(fs);
444 		if (lfs_ss_getsumsum(fs, ssp) !=
445 		    cksum((char *)ssp + sumstart,
446 			  lfs_sb_getsumsize(fs) - sumstart)) {
447 			DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset));
448 			offset = -1;
449 			goto err1;
450 		}
451 		if (lfs_ss_getnfinfo(fs, ssp) == 0 &&
452 		    lfs_ss_getninos(fs, ssp) == 0) {
453 			DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset));
454 			offset = -1;
455 			goto err1;
456 		}
457 		if (lfs_ss_getcreate(fs, ssp) < lfs_sb_gettstamp(fs)) {
458 			DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset));
459 			offset = -1;
460 			goto err1;
461 		}
462 	}
463 	if (lfs_sb_getversion(fs) > 1) {
464 		if (lfs_ss_getserial(fs, ssp) != nextserial) {
465 			DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64
466 			      "\n", offset));
467 			offset = -1;
468 			goto err1;
469 		}
470 		if (lfs_ss_getident(fs, ssp) != lfs_sb_getident(fs)) {
471 			DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%"
472 			      PRIx64 "\n", lfs_ss_getident(fs, ssp),
473 			      lfs_sb_getident(fs), offset));
474 			offset = -1;
475 			goto err1;
476 		}
477 	}
478 	if (pseg_flags)
479 		*pseg_flags = lfs_ss_getflags(fs, ssp);
480 	oldoffset = offset;
481 	offset += lfs_btofsb(fs, lfs_sb_getsumsize(fs));
482 
483 	ninos = howmany(lfs_ss_getninos(fs, ssp), LFS_INOPB(fs));
484 	iip = SEGSUM_IINFOSTART(fs, bp->b_data);
485 	if (flags & CHECK_CKSUM) {
486 		/* Count blocks */
487 		nblocks = 0;
488 		fip = SEGSUM_FINFOBASE(fs, (SEGSUM *)bp->b_data);
489 		for (i = 0; i < lfs_ss_getnfinfo(fs, ssp); ++i) {
490 			nblocks += lfs_fi_getnblocks(fs, fip);
491 			if (lfs_fi_getnblocks(fs, fip) <= 0)
492 				break;
493 			fip = NEXT_FINFO(fs, fip);
494 		}
495 		nblocks += ninos;
496 		/* Create the sum array */
497 		datap = dp = malloc(nblocks * sizeof(u_long),
498 				    M_SEGMENT, M_WAITOK);
499 	}
500 
501 	/* Handle individual blocks */
502 	fip = SEGSUM_FINFOBASE(fs, (SEGSUM *)bp->b_data);
503 	for (i = 0; i < lfs_ss_getnfinfo(fs, ssp) || ninos; ++i) {
504 		/* Inode block? */
505 		if (ninos && lfs_ii_getblock(fs, iip) == offset) {
506 			if (flags & CHECK_CKSUM) {
507 				/* Read in the head and add to the buffer */
508 				error = bread(devvp, LFS_FSBTODB(fs, offset), lfs_sb_getbsize(fs),
509 					      0, &dbp);
510 				if (error) {
511 					offset = -1;
512 					goto err2;
513 				}
514 				/* XXX this can't be right, on-disk u_long? */
515 				(*dp++) = ((u_long *)(dbp->b_data))[0];
516 				brelse(dbp, BC_AGE);
517 			}
518 			if (flags & CHECK_UPDATE) {
519 				if ((error = update_inoblk(fs, offset, cred, l))
520 				    != 0) {
521 					offset = -1;
522 					goto err2;
523 				}
524 			}
525 			offset += lfs_btofsb(fs, lfs_sb_getibsize(fs));
526 			iip = NEXTLOWER_IINFO(fs, iip);
527 			--ninos;
528 			--i; /* compensate for ++i in loop header */
529 			continue;
530 		}
531 		size = lfs_sb_getbsize(fs);
532 		for (j = 0; j < lfs_fi_getnblocks(fs, fip); ++j) {
533 			if (j == lfs_fi_getnblocks(fs, fip) - 1)
534 				size = lfs_fi_getlastlength(fs, fip);
535 			if (flags & CHECK_CKSUM) {
536 				error = bread(devvp, LFS_FSBTODB(fs, offset), size,
537 				    0, &dbp);
538 				if (error) {
539 					offset = -1;
540 					goto err2;
541 				}
542 				(*dp++) = ((u_long *)(dbp->b_data))[0];
543 				brelse(dbp, BC_AGE);
544 			}
545 			/* Account for and update any direct blocks */
546 			if ((flags & CHECK_UPDATE) &&
547 			   lfs_fi_getino(fs, fip) > LFS_IFILE_INUM &&
548 			   lfs_fi_getblock(fs, fip, j) >= 0) {
549 				update_meta(fs, lfs_fi_getino(fs, fip),
550 					    lfs_fi_getversion(fs, fip),
551 					    lfs_fi_getblock(fs, fip, j),
552 					    offset, size, l);
553 			}
554 			offset += lfs_btofsb(fs, size);
555 		}
556 		fip = NEXT_FINFO(fs, fip);
557 	}
558 	/* Checksum the array, compare */
559 	datasum = lfs_ss_getdatasum(fs, ssp);
560 	foundsum = cksum(datap, nblocks * sizeof(u_long));
561 	if ((flags & CHECK_CKSUM) && datasum != foundsum) {
562 		DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64
563 		      " (wanted %x got %x)\n",
564 		      offset, datasum, foundsum));
565 		offset = -1;
566 		goto err2;
567 	}
568 
569 	/* If we're at the end of the segment, move to the next */
570 	if (lfs_dtosn(fs, offset + lfs_btofsb(fs, lfs_sb_getsumsize(fs) + lfs_sb_getbsize(fs))) !=
571 	   lfs_dtosn(fs, offset)) {
572 		if (lfs_dtosn(fs, offset) == lfs_dtosn(fs, lfs_ss_getnext(fs, ssp))) {
573 			offset = -1;
574 			goto err2;
575 		}
576 		offset = lfs_ss_getnext(fs, ssp);
577 		DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64
578 		       " -> segment %d\n", offset, lfs_dtosn(fs,offset)));
579 	}
580 
581 	if (flags & CHECK_UPDATE) {
582 		lfs_sb_subavail(fs, offset - oldoffset);
583 		/* Don't clog the buffer queue */
584 		mutex_enter(&lfs_lock);
585 		if (locked_queue_count > LFS_MAX_BUFS ||
586 		    locked_queue_bytes > LFS_MAX_BYTES) {
587 			lfs_flush(fs, SEGM_CKP, 0);
588 		}
589 		mutex_exit(&lfs_lock);
590 	}
591 
592     err2:
593 	if (flags & CHECK_CKSUM)
594 		free(datap, M_SEGMENT);
595     err1:
596 	brelse(bp, BC_AGE);
597 
598 	/* XXX should we update the serial number even for bad psegs? */
599 	if ((flags & CHECK_UPDATE) && offset > 0 && lfs_sb_getversion(fs) > 1)
600 		lfs_sb_setserial(fs, nextserial);
601 	return offset;
602 }
603 
604 void
lfs_roll_forward(struct lfs * fs,struct mount * mp,struct lwp * l)605 lfs_roll_forward(struct lfs *fs, struct mount *mp, struct lwp *l)
606 {
607 	int flags, dirty;
608 	daddr_t offset, oldoffset, lastgoodpseg;
609 	int sn, curseg, do_rollforward;
610 	struct proc *p;
611 	kauth_cred_t cred;
612 	SEGUSE *sup;
613 	struct buf *bp;
614 
615 	p = l ? l->l_proc : NULL;
616 	cred = p ? p->p_cred : NOCRED;
617 
618 	/*
619 	 * Roll forward.
620 	 *
621 	 * We don't roll forward for v1 filesystems, because
622 	 * of the danger that the clock was turned back between the last
623 	 * checkpoint and crash.  This would roll forward garbage.
624 	 *
625 	 * v2 filesystems don't have this problem because they use a
626 	 * monotonically increasing serial number instead of a timestamp.
627 	 */
628 	do_rollforward = (!(lfs_sb_getpflags(fs) & LFS_PF_CLEAN) &&
629 			  lfs_do_rfw && lfs_sb_getversion(fs) > 1 && p != NULL);
630 	if (do_rollforward) {
631 		u_int64_t nextserial;
632 		/*
633 		 * Phase I: Find the address of the last good partial
634 		 * segment that was written after the checkpoint.  Mark
635 		 * the segments in question dirty, so they won't be
636 		 * reallocated.
637 		 */
638 		lastgoodpseg = oldoffset = offset = lfs_sb_getoffset(fs);
639 		flags = 0x0;
640 		DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%"
641 		      PRIx64 "\n", offset));
642 		LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp);
643 		if (!(sup->su_flags & SEGUSE_DIRTY))
644 			lfs_sb_subnclean(fs, 1);
645 		sup->su_flags |= SEGUSE_DIRTY;
646 		LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp);
647 		nextserial = lfs_sb_getserial(fs) + 1;
648 		while ((offset = check_segsum(fs, offset, nextserial,
649 		    cred, CHECK_CKSUM, &flags, l)) > 0) {
650 			nextserial++;
651 			if (lfs_sntod(fs, oldoffset) != lfs_sntod(fs, offset)) {
652 				LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, oldoffset),
653 					     bp);
654 				if (!(sup->su_flags & SEGUSE_DIRTY))
655 					lfs_sb_subnclean(fs, 1);
656 				sup->su_flags |= SEGUSE_DIRTY;
657 				LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, oldoffset),
658 					     bp);
659 			}
660 
661 			DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%"
662 			      PRIx64 "\n", offset));
663 			if (flags & SS_DIROP) {
664 				DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%"
665 				      PRIx64 "\n", oldoffset));
666 				if (!(flags & SS_CONT)) {
667 				     DLOG((DLOG_RF, "lfs_mountfs: dirops end "
668 					   "at 0x%" PRIx64 "\n", oldoffset));
669 				}
670 			}
671 			if (!(flags & SS_CONT))
672 				lastgoodpseg = offset;
673 			oldoffset = offset;
674 		}
675 		if (flags & SS_CONT) {
676 			DLOG((DLOG_RF, "LFS roll forward: warning: incomplete "
677 			      "dirops discarded\n"));
678 		}
679 		DLOG((DLOG_RF, "LFS roll forward phase 1: completed: "
680 		      "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg));
681 		oldoffset = lfs_sb_getoffset(fs);
682 		if (lfs_sb_getoffset(fs) != lastgoodpseg) {
683 			/* Don't overwrite what we're trying to preserve */
684 			offset = lfs_sb_getoffset(fs);
685 			lfs_sb_setoffset(fs, lastgoodpseg);
686 			lfs_sb_setcurseg(fs, lfs_sntod(fs, lfs_dtosn(fs, lfs_sb_getoffset(fs))));
687 			for (sn = curseg = lfs_dtosn(fs, lfs_sb_getcurseg(fs));;) {
688 				sn = (sn + 1) % lfs_sb_getnseg(fs);
689 				if (sn == curseg)
690 					panic("lfs_mountfs: no clean segments");
691 				LFS_SEGENTRY(sup, fs, sn, bp);
692 				dirty = (sup->su_flags & SEGUSE_DIRTY);
693 				brelse(bp, 0);
694 				if (!dirty)
695 					break;
696 			}
697 			lfs_sb_setnextseg(fs, lfs_sntod(fs, sn));
698 
699 			/*
700 			 * Phase II: Roll forward from the first superblock.
701 			 */
702 			while (offset != lastgoodpseg) {
703 				DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%"
704 				      PRIx64 "\n", offset));
705 				offset = check_segsum(fs, offset,
706 				    lfs_sb_getserial(fs) + 1, cred, CHECK_UPDATE,
707 				    NULL, l);
708 			}
709 
710 			/*
711 			 * Finish: flush our changes to disk.
712 			 */
713 			lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
714 			DLOG((DLOG_RF, "lfs_mountfs: roll forward ",
715 			      "recovered %jd blocks\n",
716 			      (intmax_t)(lastgoodpseg - oldoffset)));
717 		}
718 		DLOG((DLOG_RF, "LFS roll forward complete\n"));
719 	}
720 }
721