Ensure that filesystem metadata contained within persistent snapshots

is always kept consistent.

Suggested by:	Jeff Roberson

7 files changed, 74 insertions, 42 deletions

diff --git a/sys/ufs/ffs/ffs_alloc.c b/sys/ufs/ffs/ffs_alloc.c
index f1db84d..7d7866c 100644
--- a/sys/ufs/ffs/ffs_alloc.c
+++ b/sys/ufs/ffs/ffs_alloc.c
@@ -390,7 +390,7 @@ retry:
 		bp->b_blkno = fsbtodb(fs, bno);
 		if (!DOINGSOFTDEP(vp))
 			ffs_blkfree(ump, fs, ip->i_devvp, bprev, (long)osize,
-			    ip->i_number, NULL);
+			    ip->i_number, vp->v_type, NULL);
 		delta = btodb(nsize - osize);
 		DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
 		if (flags & IO_EXT)
@@ -670,7 +670,7 @@ ffs_reallocblks_ufs1(ap)
 		if (!DOINGSOFTDEP(vp))
 			ffs_blkfree(ump, fs, ip->i_devvp,
 			    dbtofsb(fs, buflist->bs_children[i]->b_blkno),
-			    fs->fs_bsize, ip->i_number, NULL);
+			    fs->fs_bsize, ip->i_number, vp->v_type, NULL);
 		buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
 #ifdef INVARIANTS
 		if (!ffs_checkblk(ip,
@@ -878,7 +878,7 @@ ffs_reallocblks_ufs2(ap)
 		if (!DOINGSOFTDEP(vp))
 			ffs_blkfree(ump, fs, ip->i_devvp,
 			    dbtofsb(fs, buflist->bs_children[i]->b_blkno),
-			    fs->fs_bsize, ip->i_number, NULL);
+			    fs->fs_bsize, ip->i_number, vp->v_type, NULL);
 		buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
 #ifdef INVARIANTS
 		if (!ffs_checkblk(ip,
@@ -1880,7 +1880,7 @@ ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd)
 		printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n",
 		    devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize,
 		    size, fs->fs_fsmnt);
-		panic("ffs_blkfree: bad size");
+		panic("ffs_blkfree_cg: bad size");
 	}
 #endif
 	if ((u_int)bno >= fs->fs_size) {
@@ -1914,7 +1914,7 @@ ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd)
 			}
 			printf("dev = %s, block = %jd, fs = %s\n",
 			    devtoname(dev), (intmax_t)bno, fs->fs_fsmnt);
-			panic("ffs_blkfree: freeing free block");
+			panic("ffs_blkfree_cg: freeing free block");
 		}
 		ffs_setblock(fs, blksfree, fragno);
 		ffs_clusteracct(fs, cgp, fragno, 1);
@@ -1937,7 +1937,7 @@ ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd)
 				printf("dev = %s, block = %jd, fs = %s\n",
 				    devtoname(dev), (intmax_t)(bno + i),
 				    fs->fs_fsmnt);
-				panic("ffs_blkfree: freeing free frag");
+				panic("ffs_blkfree_cg: freeing free frag");
 			}
 			setbit(blksfree, cgbno + i);
 		}
@@ -2013,13 +2013,14 @@ ffs_blkfree_trim_completed(bip)
 }
 
 void
-ffs_blkfree(ump, fs, devvp, bno, size, inum, dephd)
+ffs_blkfree(ump, fs, devvp, bno, size, inum, vtype, dephd)
 	struct ufsmount *ump;
 	struct fs *fs;
 	struct vnode *devvp;
 	ufs2_daddr_t bno;
 	long size;
 	ino_t inum;
+	enum vtype vtype;
 	struct workhead *dephd;
 {
 	struct mount *mp;
@@ -2034,7 +2035,7 @@ ffs_blkfree(ump, fs, devvp, bno, size, inum, dephd)
 	 */
 	if (devvp->v_type != VREG &&
 	    (devvp->v_vflag & VV_COPYONWRITE) &&
-	    ffs_snapblkfree(fs, devvp, bno, size, inum, dephd)) {
+	    ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, dephd)) {
 		return;
 	}
 	if (!ump->um_candelete) {
@@ -2571,7 +2572,7 @@ sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS)
 			if (blksize > blkcnt)
 				blksize = blkcnt;
 			ffs_blkfree(ump, fs, ump->um_devvp, blkno,
-			    blksize * fs->fs_fsize, ROOTINO, NULL);
+			    blksize * fs->fs_fsize, ROOTINO, VDIR, NULL);
 			blkno += blksize;
 			blkcnt -= blksize;
 			blksize = fs->fs_frag;
diff --git a/sys/ufs/ffs/ffs_balloc.c b/sys/ufs/ffs/ffs_balloc.c
index 4e8b76b..63a4eba 100644
--- a/sys/ufs/ffs/ffs_balloc.c
+++ b/sys/ufs/ffs/ffs_balloc.c
@@ -506,7 +506,7 @@ fail:
 	 */
 	for (blkp = allociblk; blkp < allocblk; blkp++) {
 		ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize,
-		    ip->i_number, NULL);
+		    ip->i_number, vp->v_type, NULL);
 	}
 	return (error);
 }
@@ -1052,7 +1052,7 @@ fail:
 	 */
 	for (blkp = allociblk; blkp < allocblk; blkp++) {
 		ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize,
-		    ip->i_number, NULL);
+		    ip->i_number, vp->v_type, NULL);
 	}
 	return (error);
 }
diff --git a/sys/ufs/ffs/ffs_extern.h b/sys/ufs/ffs/ffs_extern.h
index 58c72ee..fb02605 100644
--- a/sys/ufs/ffs/ffs_extern.h
+++ b/sys/ufs/ffs/ffs_extern.h
@@ -57,7 +57,7 @@ int	ffs_balloc_ufs2(struct vnode *a_vp, off_t a_startoffset, int a_size,
             struct ucred *a_cred, int a_flags, struct buf **a_bpp);
 int	ffs_blkatoff(struct vnode *, off_t, char **, struct buf **);
 void	ffs_blkfree(struct ufsmount *, struct fs *, struct vnode *,
-	    ufs2_daddr_t, long, ino_t, struct workhead *);
+	    ufs2_daddr_t, long, ino_t, enum vtype, struct workhead *);
 ufs2_daddr_t ffs_blkpref_ufs1(struct inode *, ufs_lbn_t, int, ufs1_daddr_t *);
 ufs2_daddr_t ffs_blkpref_ufs2(struct inode *, ufs_lbn_t, int, ufs2_daddr_t *);
 int	ffs_checkfreefile(struct fs *, struct vnode *, ino_t);
@@ -82,7 +82,7 @@ int	ffs_realloccg(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
 int	ffs_sbupdate(struct ufsmount *, int, int);
 void	ffs_setblock(struct fs *, u_char *, ufs1_daddr_t);
 int	ffs_snapblkfree(struct fs *, struct vnode *, ufs2_daddr_t, long, ino_t,
-	    struct workhead *);
+	    enum vtype, struct workhead *);
 void	ffs_snapremove(struct vnode *vp);
 int	ffs_snapshot(struct mount *mp, char *snapfile);
 void	ffs_snapshot_mount(struct mount *mp);
diff --git a/sys/ufs/ffs/ffs_inode.c b/sys/ufs/ffs/ffs_inode.c
index 67e91c7..a7b43e2 100644
--- a/sys/ufs/ffs/ffs_inode.c
+++ b/sys/ufs/ffs/ffs_inode.c
@@ -235,7 +235,8 @@ ffs_truncate(vp, length, flags, cred, td)
 				if (oldblks[i] == 0)
 					continue;
 				ffs_blkfree(ump, fs, ip->i_devvp, oldblks[i],
-				    sblksize(fs, osize, i), ip->i_number, NULL);
+				    sblksize(fs, osize, i), ip->i_number,
+				    vp->v_type, NULL);
 			}
 		}
 	}
@@ -435,7 +436,8 @@ ffs_truncate(vp, length, flags, cred, td)
 			if (lastiblock[level] < 0) {
 				DIP_SET(ip, i_ib[level], 0);
 				ffs_blkfree(ump, fs, ip->i_devvp, bn,
-				    fs->fs_bsize, ip->i_number, NULL);
+				    fs->fs_bsize, ip->i_number,
+				    vp->v_type, NULL);
 				blocksreleased += nblocks;
 			}
 		}
@@ -455,7 +457,7 @@ ffs_truncate(vp, length, flags, cred, td)
 		DIP_SET(ip, i_db[i], 0);
 		bsize = blksize(fs, ip, i);
 		ffs_blkfree(ump, fs, ip->i_devvp, bn, bsize, ip->i_number,
-		    NULL);
+		    vp->v_type, NULL);
 		blocksreleased += btodb(bsize);
 	}
 	if (lastblock < 0)
@@ -487,7 +489,7 @@ ffs_truncate(vp, length, flags, cred, td)
 			 */
 			bn += numfrags(fs, newspace);
 			ffs_blkfree(ump, fs, ip->i_devvp, bn,
-			    oldspace - newspace, ip->i_number, NULL);
+			   oldspace - newspace, ip->i_number, vp->v_type, NULL);
 			blocksreleased += btodb(oldspace - newspace);
 		}
 	}
@@ -634,7 +636,7 @@ ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
 			blocksreleased += blkcount;
 		}
 		ffs_blkfree(ip->i_ump, fs, ip->i_devvp, nb, fs->fs_bsize,
-		    ip->i_number, NULL);
+		    ip->i_number, vp->v_type, NULL);
 		blocksreleased += nblocks;
 	}
 
diff --git a/sys/ufs/ffs/ffs_snapshot.c b/sys/ufs/ffs/ffs_snapshot.c
index 89fc596..8d236bd 100644
--- a/sys/ufs/ffs/ffs_snapshot.c
+++ b/sys/ufs/ffs/ffs_snapshot.c
@@ -81,12 +81,13 @@ ffs_snapshot(mp, snapfile)
 }
 
 int
-ffs_snapblkfree(fs, devvp, bno, size, inum, wkhd)
+ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd)
 	struct fs *fs;
 	struct vnode *devvp;
 	ufs2_daddr_t bno;
 	long size;
 	ino_t inum;
+	enum vtype vtype;
 	struct workhead *wkhd;
 {
 	return (EINVAL);
@@ -174,8 +175,10 @@ static int ffs_bp_snapblk(struct vnode *, struct buf *);
  * To ensure the consistency of snapshots across crashes, we must
  * synchronously write out copied blocks before allowing the
  * originals to be modified. Because of the rather severe speed
- * penalty that this imposes, the following flag allows this
- * crash persistence to be disabled.
+ * penalty that this imposes, the code normally only ensures
+ * persistence for the filesystem metadata contained within a
+ * snapshot. Setting the following flag allows this crash
+ * persistence to be enabled for file contents.
  */
 int dopersistence = 0;
 
@@ -582,7 +585,7 @@ loop:
 			if (len != 0 && len < fs->fs_bsize) {
 				ffs_blkfree(ump, copy_fs, vp,
 				    DIP(xp, i_db[loc]), len, xp->i_number,
-				    NULL);
+				    xvp->v_type, NULL);
 				blkno = DIP(xp, i_db[loc]);
 				DIP_SET(xp, i_db[loc], 0);
 			}
@@ -1245,7 +1248,8 @@ mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
 			*ip->i_snapblklist++ = lblkno;
 		if (blkno == BLK_SNAP)
 			blkno = blkstofrags(fs, lblkno);
-		ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum, NULL);
+		ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum,
+		    vp->v_type, NULL);
 	}
 	return (0);
 }
@@ -1528,7 +1532,8 @@ mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
 			*ip->i_snapblklist++ = lblkno;
 		if (blkno == BLK_SNAP)
 			blkno = blkstofrags(fs, lblkno);
-		ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum, NULL);
+		ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum,
+		    vp->v_type, NULL);
 	}
 	return (0);
 }
@@ -1633,7 +1638,7 @@ ffs_snapremove(vp)
 			DIP_SET(ip, i_db[blkno], 0);
 		else if ((dblk == blkstofrags(fs, blkno) &&
 		     ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
-		     ip->i_number, NULL))) {
+		     ip->i_number, vp->v_type, NULL))) {
 			DIP_SET(ip, i_blocks, DIP(ip, i_blocks) -
 			    btodb(fs->fs_bsize));
 			DIP_SET(ip, i_db[blkno], 0);
@@ -1658,7 +1663,8 @@ ffs_snapremove(vp)
 					((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
 				else if ((dblk == blkstofrags(fs, blkno) &&
 				     ffs_snapblkfree(fs, ip->i_devvp, dblk,
-				     fs->fs_bsize, ip->i_number, NULL))) {
+				     fs->fs_bsize, ip->i_number, vp->v_type,
+				     NULL))) {
 					ip->i_din1->di_blocks -=
 					    btodb(fs->fs_bsize);
 					((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
@@ -1672,7 +1678,7 @@ ffs_snapremove(vp)
 				((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
 			else if ((dblk == blkstofrags(fs, blkno) &&
 			     ffs_snapblkfree(fs, ip->i_devvp, dblk,
-			     fs->fs_bsize, ip->i_number, NULL))) {
+			     fs->fs_bsize, ip->i_number, vp->v_type, NULL))) {
 				ip->i_din2->di_blocks -= btodb(fs->fs_bsize);
 				((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
 			}
@@ -1720,12 +1726,13 @@ ffs_snapremove(vp)
  * must always have been allocated from a BLK_NOCOPY location.
  */
 int
-ffs_snapblkfree(fs, devvp, bno, size, inum, wkhd)
+ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd)
 	struct fs *fs;
 	struct vnode *devvp;
 	ufs2_daddr_t bno;
 	long size;
 	ino_t inum;
+	enum vtype vtype;
 	struct workhead *wkhd;
 {
 	struct buf *ibp, *cbp, *savedcbp = 0;
@@ -1874,12 +1881,16 @@ retry:
 		 * simply copy them to the new block. Note that we need
 		 * to synchronously write snapshots that have not been
 		 * unlinked, and hence will be visible after a crash,
-		 * to ensure their integrity.
+		 * to ensure their integrity. At a minimum we ensure the
+		 * integrity of the filesystem metadata, but use the
+		 * dopersistence sysctl-setable flag to decide on the
+		 * persistence needed for file content data.
 		 */
 		if (savedcbp != 0) {
 			bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
 			bawrite(cbp);
-			if (dopersistence && ip->i_effnlink > 0)
+			if ((vtype == VDIR || dopersistence) &&
+			    ip->i_effnlink > 0)
 				(void) ffs_syncvnode(vp, MNT_WAIT);
 			continue;
 		}
@@ -1889,7 +1900,8 @@ retry:
 		if ((error = readblock(vp, cbp, lbn)) != 0) {
 			bzero(cbp->b_data, fs->fs_bsize);
 			bawrite(cbp);
-			if (dopersistence && ip->i_effnlink > 0)
+			if ((vtype == VDIR || dopersistence) &&
+			    ip->i_effnlink > 0)
 				(void) ffs_syncvnode(vp, MNT_WAIT);
 			break;
 		}
@@ -1898,12 +1910,15 @@ retry:
 	/*
 	 * Note that we need to synchronously write snapshots that
 	 * have not been unlinked, and hence will be visible after
-	 * a crash, to ensure their integrity.
+	 * a crash, to ensure their integrity. At a minimum we
+	 * ensure the integrity of the filesystem metadata, but
+	 * use the dopersistence sysctl-setable flag to decide on
+	 * the persistence needed for file content data.
 	 */
 	if (savedcbp) {
 		vp = savedcbp->b_vp;
 		bawrite(savedcbp);
-		if (dopersistence && VTOI(vp)->i_effnlink > 0)
+		if ((vtype == VDIR || dopersistence) && ip->i_effnlink > 0)
 			(void) ffs_syncvnode(vp, MNT_WAIT);
 	}
 	/*
@@ -2358,12 +2373,16 @@ ffs_copyonwrite(devvp, bp)
 		 * simply copy them to the new block. Note that we need
 		 * to synchronously write snapshots that have not been
 		 * unlinked, and hence will be visible after a crash,
-		 * to ensure their integrity.
+		 * to ensure their integrity. At a minimum we ensure the
+		 * integrity of the filesystem metadata, but use the
+		 * dopersistence sysctl-setable flag to decide on the
+		 * persistence needed for file content data.
 		 */
 		if (savedcbp != 0) {
 			bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
 			bawrite(cbp);
-			if (dopersistence && ip->i_effnlink > 0)
+			if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
+			    dopersistence) && ip->i_effnlink > 0)
 				(void) ffs_syncvnode(vp, MNT_WAIT);
 			else
 				launched_async_io = 1;
@@ -2375,7 +2394,8 @@ ffs_copyonwrite(devvp, bp)
 		if ((error = readblock(vp, cbp, lbn)) != 0) {
 			bzero(cbp->b_data, fs->fs_bsize);
 			bawrite(cbp);
-			if (dopersistence && ip->i_effnlink > 0)
+			if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
+			    dopersistence) && ip->i_effnlink > 0)
 				(void) ffs_syncvnode(vp, MNT_WAIT);
 			else
 				launched_async_io = 1;
@@ -2386,12 +2406,16 @@ ffs_copyonwrite(devvp, bp)
 	/*
 	 * Note that we need to synchronously write snapshots that
 	 * have not been unlinked, and hence will be visible after
-	 * a crash, to ensure their integrity.
+	 * a crash, to ensure their integrity. At a minimum we
+	 * ensure the integrity of the filesystem metadata, but
+	 * use the dopersistence sysctl-setable flag to decide on
+	 * the persistence needed for file content data.
 	 */
 	if (savedcbp) {
 		vp = savedcbp->b_vp;
 		bawrite(savedcbp);
-		if (dopersistence && VTOI(vp)->i_effnlink > 0)
+		if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
+		    dopersistence) && VTOI(vp)->i_effnlink > 0)
 			(void) ffs_syncvnode(vp, MNT_WAIT);
 		else
 			launched_async_io = 1;
diff --git a/sys/ufs/ffs/ffs_softdep.c b/sys/ufs/ffs/ffs_softdep.c
index 6730dae..3734a5d 100644
--- a/sys/ufs/ffs/ffs_softdep.c
+++ b/sys/ufs/ffs/ffs_softdep.c
@@ -5172,6 +5172,7 @@ newfreefrag(ip, blkno, size, lbn)
 	freefrag->ff_state = ATTACHED;
 	LIST_INIT(&freefrag->ff_jwork);
 	freefrag->ff_inum = ip->i_number;
+	freefrag->ff_vtype = ITOV(ip)->v_type;
 	freefrag->ff_blkno = blkno;
 	freefrag->ff_fragsize = size;
 
@@ -5216,7 +5217,7 @@ handle_workitem_freefrag(freefrag)
 	}
 	FREE_LOCK(&lk);
 	ffs_blkfree(ump, ump->um_fs, ump->um_devvp, freefrag->ff_blkno,
-	    freefrag->ff_fragsize, freefrag->ff_inum, &wkhd);
+	   freefrag->ff_fragsize, freefrag->ff_inum, freefrag->ff_vtype, &wkhd);
 	ACQUIRE_LOCK(&lk);
 	WORKITEM_FREE(freefrag, D_FREEFRAG);
 	FREE_LOCK(&lk);
@@ -5724,6 +5725,7 @@ newfreeblks(mp, ip)
 	freeblks->fb_state = ATTACHED;
 	freeblks->fb_uid = ip->i_uid;
 	freeblks->fb_inum = ip->i_number;
+	freeblks->fb_vtype = ITOV(ip)->v_type;
 	freeblks->fb_modrev = DIP(ip, i_modrev);
 	freeblks->fb_devvp = ip->i_devvp;
 	freeblks->fb_chkcnt = 0;
@@ -7263,7 +7265,7 @@ freework_freeblock(freework)
 	freeblks->fb_freecnt += btodb(bsize);
 	FREE_LOCK(&lk);
 	ffs_blkfree(ump, fs, freeblks->fb_devvp, freework->fw_blkno, bsize,
-	    freeblks->fb_inum, &wkhd);
+	    freeblks->fb_inum, freeblks->fb_vtype, &wkhd);
 	ACQUIRE_LOCK(&lk);
 	/*
 	 * The jnewblk will be discarded and the bits in the map never
@@ -7669,7 +7671,8 @@ indir_trunc(freework, dbn, lbn)
 				freedeps++;
 			}
 			ffs_blkfree(ump, fs, freeblks->fb_devvp, nb,
-			    fs->fs_bsize, freeblks->fb_inum, &wkhd);
+			    fs->fs_bsize, freeblks->fb_inum,
+			    freeblks->fb_vtype, &wkhd);
 		}
 	}
 	if (goingaway) {
@@ -7702,7 +7705,7 @@ indir_trunc(freework, dbn, lbn)
 	fs_pendingblocks += nblocks;
 	dbn = dbtofsb(fs, dbn);
 	ffs_blkfree(ump, fs, freeblks->fb_devvp, dbn, fs->fs_bsize,
-	    freeblks->fb_inum, NULL);
+	    freeblks->fb_inum, freeblks->fb_vtype, NULL);
 	/* Non SUJ softdep does single-threaded truncations. */
 	freeblks->fb_freecnt += fs_pendingblocks;
 	if (freework->fw_blkno == dbn) {
diff --git a/sys/ufs/ffs/softdep.h b/sys/ufs/ffs/softdep.h
index 9be175c..80c7315 100644
--- a/sys/ufs/ffs/softdep.h
+++ b/sys/ufs/ffs/softdep.h
@@ -511,6 +511,7 @@ struct freefrag {
 	ufs2_daddr_t ff_blkno;		/* fragment physical block number */
 	long	ff_fragsize;		/* size of fragment being deleted */
 	ino_t	ff_inum;		/* owning inode number */
+	enum	vtype ff_vtype;		/* owning inode's file type */
 };
 
 /*
@@ -538,6 +539,7 @@ struct freeblks {
 	ufs2_daddr_t fb_chkcnt;		/* Expected blks released. */
 	ufs2_daddr_t fb_freecnt;	/* Actual blocks released. */
 	ino_t	fb_inum;		/* inode owner of blocks */
+	enum	vtype fb_vtype;		/* inode owner's file type */
 	uid_t	fb_uid;			/* uid of previous owner of blocks */
 	int	fb_ref;			/* Children outstanding. */
 	int	fb_cgwait;		/* cg writes outstanding. */