xfsdump: use the full 32-bit generation number

xfsdump historically has truncated the inode generation number to
the low 12 bits when writing out directory entries. This makes it
possible for xfsrestore to mistakingly think 2 directory entries
refer to the same inode when dealing with incremental or resumed
dumps. A message such as this is an indication of this problem:

    WARNING: unable to unlink current file prior to restore

This patch changes xfsdump to use the full 32-bit inode generation
number. A change to part of the dump format (direnthdr_t) was
required, so the dump format version has been bumped to 3. xfsdump
also required changes to its inode-to-generation cache. This map is
not persistent though, so no compatibility or version changes were
required there.

xfsdump can still generate a format 2 dump using the new -K option.
This is useful when moving a filesystem to a system with an older
version of xfsrestore.

xfsrestore has been changed to support the old and new dump formats.
This required a change to its persistent data structures (for
cumulative restores), so the housekeeping version number was bumped
as well.

When restoring a series of incremental/resumed dumps, if the oldest
restore used 12-bit generation numbers then they will be used
throughout the restore series to avoid mass confusion.

In the rare case that a cumulative restore is done using a format 3
dump followed by a format 2 dump, the user must specifically tell
xfsrestore to use format 2 generation numbers throughout the restore
series by using the -K option on the first restore.

It's recommended that users do a level 0 backup of their filesystems
with the new xfsdump so that future incremental restores can take
advantage of the full 32-bit generation number.

This patch also fixes a couple of instances where the dump format
was being displayed with an incorrect value.

Signed-off-by: Bill Kendall <wkendall@sgi.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

16 files changed, 335 insertions, 120 deletions

diff --git a/common/arch_xlate.c b/common/arch_xlate.c
index 1c7e8803..c1563133 100644
--- a/common/arch_xlate.c
+++ b/common/arch_xlate.c
@@ -438,8 +438,8 @@ xlate_direnthdr(direnthdr_t *dh1, direnthdr_t *dh2, int dir)
 
 	IXLATE(dh1, dh2, dh_ino);
 	IXLATE(dh1, dh2, dh_gen);
-	IXLATE(dh1, dh2, dh_sz);
 	IXLATE(dh1, dh2, dh_checksum);
+	IXLATE(dh1, dh2, dh_sz);
 
 	if (dir < 0) {
 		ptr1 = dh2;
@@ -450,7 +450,53 @@ xlate_direnthdr(direnthdr_t *dh1, direnthdr_t *dh2, int dir)
 
 	mlog(MLOG_NITTY, "xlate_direnthdr: pre-xlate\n"
 	     "\tdh_ino %llu\n"
-	     "\tdh_gen %d\n"
+	     "\tdh_gen %u\n"
+	     "\tdh_checksum %d\n"
+	     "\tdh_sz %d\n"
+	     "\tdh_name %.8s\n",
+	     ptr1->dh_ino,
+	     ptr1->dh_gen,
+	     ptr1->dh_checksum,
+	     ptr1->dh_sz,
+	     ptr1->dh_name );
+
+	mlog(MLOG_NITTY, "xlate_direnthdr: post-xlate\n"
+	     "\tdh_ino %llu\n"
+	     "\tdh_gen %u\n"
+	     "\tdh_checksum %d\n"
+	     "\tdh_sz %d\n"
+	     "\tdh_name %.8s\n",
+	     ptr2->dh_ino,
+	     ptr2->dh_gen,
+	     ptr2->dh_checksum,
+	     ptr2->dh_sz,
+	     ptr2->dh_name );
+}
+
+/*
+ * xlate_direnthdr_v1 - endian convert struct direnthdr_v1
+ */
+void
+xlate_direnthdr_v1(direnthdr_v1_t *dh1, direnthdr_v1_t *dh2, int dir)
+{
+	direnthdr_v1_t *ptr1 = dh1;
+	direnthdr_v1_t *ptr2 = dh2;
+
+	IXLATE(dh1, dh2, dh_ino);
+	IXLATE(dh1, dh2, dh_gen);
+	IXLATE(dh1, dh2, dh_sz);
+	IXLATE(dh1, dh2, dh_checksum);
+
+	if (dir < 0) {
+		ptr1 = dh2;
+		ptr2 = dh1;
+	}
+
+	BXLATE(dh_name);
+
+	mlog(MLOG_NITTY, "xlate_direnthdr_v1: pre-xlate\n"
+	     "\tdh_ino %llu\n"
+	     "\tdh_gen %u\n"
 	     "\tdh_sz %d\n"
 	     "\tdh_checksum %d\n"
 	     "\tdh_name %.8s\n",
@@ -460,9 +506,9 @@ xlate_direnthdr(direnthdr_t *dh1, direnthdr_t *dh2, int dir)
 	     ptr1->dh_checksum,
 	     ptr1->dh_name );
 
-	mlog(MLOG_NITTY, "xlate_direnthdr: post-xlate\n"
+	mlog(MLOG_NITTY, "xlate_direnthdr_v1: post-xlate\n"
 	     "\tdh_ino %llu\n"
-	     "\tdh_gen %d\n"
+	     "\tdh_gen %u\n"
 	     "\tdh_sz %d\n"
 	     "\tdh_checksum %d\n"
 	     "\tdh_name %.8s\n",
diff --git a/common/arch_xlate.h b/common/arch_xlate.h
index 3ad3c970..35333c68 100644
--- a/common/arch_xlate.h
+++ b/common/arch_xlate.h
@@ -98,6 +98,11 @@ void xlate_extenthdr(extenthdr_t *eh1, extenthdr_t *eh2, int dir);
 void xlate_direnthdr(direnthdr_t *dh1, direnthdr_t *dh2, int dir);
 
 /*
+ * xlate_direnthdr_v1 - endian convert struct direnthdr_v1
+ */
+void xlate_direnthdr_v1(direnthdr_v1_t *dh1, direnthdr_v1_t *dh2, int dir);
+
+/*
  * xlate_extattrhdr - endian convert struct extattrhdr
  */
 void xlate_extattrhdr(extattrhdr_t *eh1, extattrhdr_t *eh2, int dir);
diff --git a/common/content_inode.h b/common/content_inode.h
index 67c4f6d2..a25b66ea 100644
--- a/common/content_inode.h
+++ b/common/content_inode.h
@@ -284,26 +284,40 @@ typedef struct extenthdr extenthdr_t;
  * a sequence of directory entries is always terminated with a null direnthdr_t.
  * this is detected by looking for a zero ino.
  */
+typedef u_int32_t gen_t;
+
 #define DIRENTHDR_ALIGN	8
 
 #define DIRENTHDR_SZ	24
 
 struct direnthdr {
 	xfs_ino_t dh_ino;
+	gen_t dh_gen;
+	u_int32_t dh_checksum;
+	u_int16_t dh_sz; /* overall size of record */
+	char dh_name[ 6 ];
+};
+
+typedef struct direnthdr direnthdr_t;
+
+/* the old direnthdr truncated the inode generation number
+ * to the low 12 bits.
+ */
+
+struct direnthdr_v1 {
+	xfs_ino_t dh_ino;
 	u_int16_t dh_gen; /* generation count & DENTGENMASK of ref'ed inode */
 	u_int16_t dh_sz; /* overall size of record */
 	u_int32_t dh_checksum;
 	char dh_name[ 8 ];
 };
 
-typedef struct direnthdr direnthdr_t;
+typedef struct direnthdr_v1 direnthdr_v1_t;
 
 /* truncated generation count
  */
 #define DENTGENSZ		12	/* leave 4 bits for future flags */
 #define DENTGENMASK		(( 1 << DENTGENSZ ) - 1 )
-typedef u_int16_t gen_t;
-#define GEN_NULL		( ( gen_t )UINT16MAX )
 #define BIGGEN2GEN( bg )	( ( gen_t )( bg & DENTGENMASK ))
 
 
diff --git a/common/global.c b/common/global.c
index 737b7317..8e49d8b7 100644
--- a/common/global.c
+++ b/common/global.c
@@ -191,6 +191,10 @@ global_hdr_alloc( intgen_t argc, char *argv[ ] )
 			}
 			ghdrp->gh_timestamp = statb.st_mtime;
 			break;
+
+		case GETOPT_FMT2COMPAT:
+			ghdrp->gh_version = GLOBAL_HDR_VERSION_2;
+			break;
 #endif /* DUMP */
 		}
 	}
@@ -276,6 +280,7 @@ global_version_check( u_int32_t version )
 		case GLOBAL_HDR_VERSION_0:
 		case GLOBAL_HDR_VERSION_1:
 		case GLOBAL_HDR_VERSION_2:
+		case GLOBAL_HDR_VERSION_3:
 			return BOOL_TRUE;
 		default:
 			return BOOL_FALSE;
diff --git a/common/global.h b/common/global.h
index ea2b732a..6556a68e 100644
--- a/common/global.h
+++ b/common/global.h
@@ -27,13 +27,14 @@
 #define GLOBAL_HDR_VERSION_0	0
 #define GLOBAL_HDR_VERSION_1	1
 #define GLOBAL_HDR_VERSION_2	2
-	/* version 2 adds encoding of holes and a change to on-tape inventory format.
+#define GLOBAL_HDR_VERSION_3	3
+	/* version 3 uses the full 32-bit inode generation number in direnthdr_t.
+	 * version 2 adds encoding of holes and a change to on-tape inventory format.
 	 * version 1 adds extended file attribute dumping.
 	 * version 0 xfsrestore can't handle media produced
 	 * by version 1 xfsdump. 
 	 */
-#define GLOBAL_HDR_VERSION	GLOBAL_HDR_VERSION_2
-#define GLOBAL_HDR_VERSION_PREV	1
+#define GLOBAL_HDR_VERSION	GLOBAL_HDR_VERSION_3
 
 #define GLOBAL_HDR_STRING_SZ	0x100
 #define GLOBAL_HDR_TIME_SZ	4
diff --git a/common/main.c b/common/main.c
index c9a311bc..8e7451fb 100644
--- a/common/main.c
+++ b/common/main.c
@@ -103,8 +103,6 @@ static char *strpbrkquotes( char *p, const char *sep );
 
 /* definition of locally defined global variables ****************************/
 
-intgen_t version = 3;
-intgen_t subversion = 0;
 char *progname = 0;			/* used in all error output */
 char *homedir = 0;			/* directory invoked from */
 bool_t pipeline = BOOL_FALSE;
@@ -400,10 +398,8 @@ main( int argc, char *argv[] )
 	 */
 	if ( infoonly ) {
 		mlog( MLOG_NORMAL,
-		      _("version %s (dump format %d.%d)\n"),
-		      VERSION,
-		      version,
-		      subversion );
+		      _("version %s (dump format %d.0)\n"),
+		      VERSION, GLOBAL_HDR_VERSION );
 		usage( );
 		return mlog_exit(EXIT_NORMAL, RV_OK); /* normal termination */
 	}
@@ -475,10 +471,8 @@ main( int argc, char *argv[] )
 	 */
 	sistr = sigintstr( );
 	mlog( MLOG_VERBOSE,
-	      _("version %s (dump format %d.%d)"),
-	      VERSION,
-	      version,
-	      subversion );
+	      _("version %s (dump format %d.0)"),
+	      VERSION, GLOBAL_HDR_VERSION );
 	if ( ! pipeline && ! stdoutpiped && sistr && dlog_allowed( )) {
 		mlog( MLOG_VERBOSE | MLOG_BARE, _(
 		      " - "
@@ -933,6 +927,7 @@ usage( void )
 #endif /* REVEAL */
 	ULO(_("(display dump inventory)"),		GETOPT_INVPRINT );
 	ULO(_("(inhibit inventory update)"),		GETOPT_NOINVUPDATE );
+	ULO(_("(generate format 2 dump)"),		GETOPT_FMT2COMPAT );
 	ULO(_("<session label>"),			GETOPT_DUMPLABEL );
 	ULO(_("<media label> ..."),			GETOPT_MEDIALABEL );
 #ifdef REVEAL
@@ -981,6 +976,7 @@ usage( void )
 	ULO(_("(don't prompt)"),			GETOPT_FORCE );
 	ULO(_("(display dump inventory)"),		GETOPT_INVPRINT );
 	ULO(_("(inhibit inventory update)"),		GETOPT_NOINVUPDATE );
+	ULO(_("(force use of format 2 generation numbers)"),GETOPT_FMT2COMPAT );
 	ULO(_("<session label>"),			GETOPT_DUMPLABEL );
 #ifdef REVEAL
 	ULO(_("(timestamp messages)"),			GETOPT_TIMESTAMP );
diff --git a/dump/content.c b/dump/content.c
index 3a7f5089..78b303f0 100644
--- a/dump/content.c
+++ b/dump/content.c
@@ -290,7 +290,7 @@ static rv_t dump_dirent( drive_t *drivep,
 			 context_t *contextp,
 			 xfs_bstat_t *,
 			 xfs_ino_t,
-			 u_int32_t,
+			 gen_t,
 			 char *,
 			 size_t );
 static rv_t init_extent_group_context( jdm_fshandle_t *,
@@ -477,6 +477,10 @@ static bool_t sc_dumpextattrpr = BOOL_TRUE;
 static bool_t sc_dumpasoffline = BOOL_FALSE;
 	/* dump dual-residency HSM files as offline
 	 */
+static bool_t sc_use_old_direntpr = BOOL_FALSE;
+	/* dump dirents as dirent_v1_t instead of dirent_t
+	 * (for compat with dump format 2)
+	 */
 
 static bool_t sc_savequotas = BOOL_TRUE;
 /* save quota information in dump
@@ -560,6 +564,7 @@ content_init( intgen_t argc,
 	ASSERT( sizeof( filehdr_t ) == FILEHDR_SZ );
 	ASSERT( sizeof( extenthdr_t ) == EXTENTHDR_SZ );
 	ASSERT( sizeof( direnthdr_t ) == DIRENTHDR_SZ );
+	ASSERT( sizeof( direnthdr_v1_t ) == DIRENTHDR_SZ );
 	ASSERT( DIRENTHDR_SZ % DIRENTHDR_ALIGN == 0 );
 	ASSERT( sizeofmember( content_hdr_t, ch_specific )
 		>=
@@ -573,6 +578,10 @@ content_init( intgen_t argc,
 	cwhdrtemplatep = ( content_hdr_t * )mwhdrtemplatep->mh_upper;
 	scwhdrtemplatep = ( content_inode_hdr_t * ) cwhdrtemplatep->ch_specific;
 
+	if ( gwhdrtemplatep->gh_version < GLOBAL_HDR_VERSION_3 ) {
+		sc_use_old_direntpr = BOOL_TRUE;
+	}
+
 	/* process command line args
 	 */
 	optind = 1;
@@ -2902,7 +2911,7 @@ dump_dir( ix_t strmix,
 	struct dirent *gdp = ( struct dirent *)contextp->cc_getdentsbufp;
 	size_t gdsz = contextp->cc_getdentsbufsz;
 	intgen_t gdcnt;
-	u_int32_t gen;
+	gen_t gen;
 	rv_t rv;
 
 	/* no way this can be non-dir, but check anyway
@@ -3073,8 +3082,7 @@ dump_dir( ix_t strmix,
 			/* lookup the gen number in the ino-to-gen map.
 			 * if it's not there, we have to get it the slow way.
 			 */
-			gen = inomap_get_gen( NULL, p->d_ino );
-			if (gen == GEN_NULL) {
+			if ( inomap_get_gen( NULL, p->d_ino, &gen) ) {
 				xfs_bstat_t statbuf;
 				intgen_t scrval;
 				
@@ -5045,19 +5053,25 @@ dump_dirent( drive_t *drivep,
 	     context_t *contextp,
 	     xfs_bstat_t *statp,
 	     xfs_ino_t ino,
-	     u_int32_t gen,
+	     gen_t gen,
 	     char *name,
 	     size_t namelen )
 {
 	drive_ops_t *dop = drivep->d_opsp;
-	direnthdr_t *dhdrp = ( direnthdr_t * )contextp->cc_mdirentbufp;
-	direnthdr_t *tmpdhdrp;
+	char *outbufp;
 	size_t direntbufsz = contextp->cc_mdirentbufsz;
 	size_t sz;
+	size_t name_offset;
 	intgen_t rval;
 	rv_t rv;
 
-	sz = offsetofmember( direnthdr_t, dh_name )
+	if ( sc_use_old_direntpr ) {
+		name_offset = offsetofmember( direnthdr_v1_t, dh_name );
+	} else {
+		name_offset = offsetofmember( direnthdr_t, dh_name );
+	}
+
+	sz = name_offset
 	     +
 	     namelen
 	     +
@@ -5081,28 +5095,52 @@ dump_dirent( drive_t *drivep,
 	ASSERT( sz <= UINT16MAX );
 	ASSERT( sz >= DIRENTHDR_SZ );
 
-	memset( ( void * )dhdrp, 0, sz );
-	dhdrp->dh_ino = ino;
-	dhdrp->dh_sz = ( u_int16_t )sz;
-	dhdrp->dh_gen = ( u_int16_t )( gen & DENTGENMASK );
+	outbufp = malloc(sz);
 
-	if ( name ) {
-		strcpy( dhdrp->dh_name, name );
-	}
+	if ( sc_use_old_direntpr ) {
+		direnthdr_v1_t *dhdrp = ( direnthdr_v1_t * )contextp->cc_mdirentbufp;
+		direnthdr_v1_t *tmpdhdrp = ( direnthdr_v1_t * )outbufp;
+
+		memset( ( void * )dhdrp, 0, sz );
+		dhdrp->dh_ino = ino;
+		dhdrp->dh_sz = ( u_int16_t )sz;
+		dhdrp->dh_gen = ( u_int16_t )( gen & DENTGENMASK );
+		if ( name ) {
+			strcpy( dhdrp->dh_name, name );
+		}
 
-	dhdrp->dh_checksum = calc_checksum( dhdrp, DIRENTHDR_SZ );
+		dhdrp->dh_checksum = calc_checksum( dhdrp, DIRENTHDR_SZ );
 
-	tmpdhdrp = malloc(sz);
-	xlate_direnthdr(dhdrp, tmpdhdrp, 1);
-	if ( name ) {
-		strcpy( tmpdhdrp->dh_name, name );
+		xlate_direnthdr_v1( dhdrp, tmpdhdrp, 1 );
+		if ( name ) {
+			strcpy( tmpdhdrp->dh_name, name );
+		}
+	} else {
+		direnthdr_t *dhdrp = ( direnthdr_t * )contextp->cc_mdirentbufp;
+		direnthdr_t *tmpdhdrp = ( direnthdr_t * )outbufp;
+
+		memset( ( void * )dhdrp, 0, sz );
+		dhdrp->dh_ino = ino;
+		dhdrp->dh_gen = gen;
+		dhdrp->dh_sz = ( u_int16_t )sz;
+		if ( name ) {
+			strcpy( dhdrp->dh_name, name );
+		}
+
+		dhdrp->dh_checksum = calc_checksum( dhdrp, DIRENTHDR_SZ );
+
+		xlate_direnthdr( dhdrp, tmpdhdrp, 1 );
+		if ( name ) {
+			strcpy( tmpdhdrp->dh_name, name );
+		}
 	}
-	rval = write_buf( ( char * )tmpdhdrp,
+
+	rval = write_buf( outbufp,
 			  sz,
 			  ( void * )drivep,
 			  ( gwbfp_t )dop->do_get_write_buf,
 			  ( wfp_t )dop->do_write );
-	free(tmpdhdrp);
+	free(outbufp);
 	switch ( rval ) {
 	case 0:
 		rv = RV_OK;
diff --git a/dump/getopt.h b/dump/getopt.h
index ba26c936..3bab87ae 100644
--- a/dump/getopt.h
+++ b/dump/getopt.h
@@ -27,7 +27,7 @@
  * facilitating easy changes.
  */
 
-#define GETOPT_CMDSTRING	"ab:c:d:ef:hl:mop:qs:t:v:z:AB:CDEFG:H:I:JL:M:NO:PRSTUVWY:"
+#define GETOPT_CMDSTRING	"ab:c:d:ef:hl:mop:qs:t:v:z:AB:CDEFG:H:I:JKL:M:NO:PRSTUVWY:"
 
 #define GETOPT_DUMPASOFFLINE	'a'	/* dump DMF dualstate files as offline */
 #define	GETOPT_BLOCKSIZE	'b'	/* blocksize for rmt */
@@ -65,7 +65,7 @@
 #define GETOPT_MAXSTACKSZ	'H'	/* maximum stack size (bytes) */
 #define GETOPT_INVPRINT         'I'     /* just display the inventory */
 #define	GETOPT_NOINVUPDATE	'J'	/* do not update the dump inventory */
-/*				'K'	*/
+#define GETOPT_FMT2COMPAT	'K'	/* use dump format 2 for compat with old restore */
 #define	GETOPT_DUMPLABEL	'L'	/* dump session label (global.c) */
 #define	GETOPT_MEDIALABEL	'M'	/* media object label (media.c) */
 #define	GETOPT_TIMESTAMP	'N'	/* show timestamps in log msgs */
diff --git a/dump/inomap.c b/dump/inomap.c
index aa4f59d3..9b385ec0 100644
--- a/dump/inomap.c
+++ b/dump/inomap.c
@@ -940,18 +940,11 @@ cb_startpt( void *arg1,
 /* map context and operators
  */
 
-/* define structure for ino to gen mapping. Allocate 12 bits for the gen
- * instead of the 32-bit gen that XFS uses, as xfsdump currently truncates
- * the gen to 12 bits.
+/* define structure for ino to gen mapping.
  */
-#if DENTGENSZ != 12
-#error DENTGENSZ has changed. i2gseg_t and its users must be updated.
-#endif
-
 struct i2gseg {
 	u_int64_t s_valid;
-	u_char_t s_lower[ INOPERSEG ];
-	u_char_t s_upper[ INOPERSEG / 2 ];
+	gen_t s_gen[ INOPERSEG ];
 };
 typedef struct i2gseg i2gseg_t;
 
@@ -1382,51 +1375,31 @@ inomap_set_gen(void *contextp, xfs_ino_t ino, gen_t gen)
 
 	relino = ino - segp->base;
 	i2gsegp->s_valid |= (u_int64_t)1 << relino;
-	i2gsegp->s_lower[ relino ] = ( u_char_t )( gen & 0xff );
-	if ( relino & 1 ) {
-		/* odd, goes in high nibble */
-		i2gsegp->s_upper[relino / 2] &= ( u_char_t )( 0x0f );
-		i2gsegp->s_upper[relino / 2] |=
-			( u_char_t )( ( gen >> 4 ) & 0xf0 );
-	} else {
-		/* even, goes in low nibble */
-		i2gsegp->s_upper[ relino / 2 ] &= ( u_char_t )( 0xf0 );
-		i2gsegp->s_upper[ relino / 2 ] |=
-			( u_char_t )( ( gen >> 8 ) & 0x0f );
-	}
+	i2gsegp->s_gen[relino] = gen;
 }
 
-gen_t
-inomap_get_gen( void *contextp, xfs_ino_t ino )
+intgen_t
+inomap_get_gen( void *contextp, xfs_ino_t ino, gen_t *gen )
 {
 	seg_addr_t *addrp;
 	seg_addr_t addr;
 	seg_t *segp;
 	i2gseg_t *i2gsegp;
 	xfs_ino_t relino;
-	gen_t gen;
 
 	addrp = contextp ? (seg_addr_t *)contextp : &addr;
 	if ( !inomap_find_seg( addrp, ino ) )
-		return GEN_NULL;
+		return 1;
 
 	segp = inomap_addr2seg( addrp );
 	i2gsegp = &inomap.i2gmap[inomap_addr2segix( addrp )];
 
 	relino = ino - segp->base;
 	if ( ! (i2gsegp->s_valid & ((u_int64_t)1 << relino)) )
-		return GEN_NULL;
-
-	gen = i2gsegp->s_lower[relino];
-	if (relino & 1) {
-		/* odd, rest of gen in high nibble */
-		gen |= ( (gen_t)i2gsegp->s_upper[relino / 2] & 0xf0 ) << 4;
-	} else {
-		/* even, rest of gen in low nibble */
-		gen |= ( (gen_t)i2gsegp->s_upper[relino / 2] & 0x0f ) << 8;
-	}
+		return 1;
 
-	return gen;
+	*gen = i2gsegp->s_gen[relino];
+	return 0;
 }
 
 void
diff --git a/dump/inomap.h b/dump/inomap.h
index 16f2efb1..7d1db1fe 100644
--- a/dump/inomap.h
+++ b/dump/inomap.h
@@ -132,7 +132,7 @@ extern void *inomap_alloc_context( void );
 extern void inomap_reset_context( void *contextp );
 extern void inomap_free_context( void *contextp );
 extern intgen_t inomap_get_state( void *contextp, xfs_ino_t ino );
-extern gen_t inomap_get_gen( void *contextp, xfs_ino_t ino );
+extern intgen_t inomap_get_gen( void *contextp, xfs_ino_t ino, gen_t *gen );
 
 
 /* generators returning the next dir or non-dir ino selected in this dump.
diff --git a/man/man8/xfsdump.8 b/man/man8/xfsdump.8
index fb47f7b5..7d261ed6 100644
--- a/man/man8/xfsdump.8
+++ b/man/man8/xfsdump.8
@@ -333,6 +333,13 @@ Inhibits the normal update of the inventory.
 This is useful when the media being dumped to
 will be discarded or overwritten.
 .TP 5
+.B \-K
+Generate a format 2 dump instead of the current format. This is useful
+if the dump will be restored on a system with an older
+.I xfsrestore
+which does not understand the current dump format. Use of this option
+is otherwise not recommended.
+.TP 5
 \f3\-L\f1 \f2session_label\f1
 Specifies a label for the dump session.
 It can be any arbitrary string up to 255 characters long.
diff --git a/man/man8/xfsrestore.8 b/man/man8/xfsrestore.8
index aad97fae..60e43098 100644
--- a/man/man8/xfsrestore.8
+++ b/man/man8/xfsrestore.8
@@ -339,6 +339,15 @@ when it encounters an on-media session inventory,
 but only if run with an effective user id of root
 and only if this option is not given.
 .TP 5
+.B \-K
+Force
+.I xfsrestore
+to use dump format 2 generation numbers. Normally the need for this is
+determined automatically, but this option is required on the first
+.I xfsrestore
+invocation in the rare case that a cumulative restore begins
+with a format 3 (or newer) dump and will be followed by a format 2 dump.
+.TP 5
 \f3\-L\f1 \f2session_label\f1
 Specifies the label
 of the dump session to be restored.
diff --git a/restore/content.c b/restore/content.c
index a9e0b20a..9aa85818 100644
--- a/restore/content.c
+++ b/restore/content.c
@@ -73,8 +73,10 @@
 #define HOUSEKEEPING_MAGIC	0x686b6d61
 	/* "hkma" - see the housekeeping_magic field of pers_t below.
 	 */
-#define HOUSEKEEPING_VERSION	1
+#define HOUSEKEEPING_VERSION	2
 	/* see the housekeeping_version field of pers_t below.
+	 * version 2 changed the size of a gen_t, which caused node_t
+	 * to change in size. also p_truncategenpr was added to treepers_t.
 	 */
 
 #define WRITE_TRIES_MAX	3
@@ -629,6 +631,9 @@ struct tran {
 	size64_t t_dirdumps;
 		/* bitset of streams which contain a directory dump
 		 */
+	bool_t t_truncategenpr;
+		/* force use of truncated generation numbers
+		 */
 	sync_t t_sync1;
 		/* to single-thread attempt to validate command line
 		 * selection of dump with online inventory
@@ -1165,6 +1170,9 @@ content_init( intgen_t argc, char *argv[ ], size64_t vmsz )
 		case GETOPT_ROOTPERM:
 			restore_rootdir_permissions = BOOL_TRUE;
 			break;
+		case GETOPT_FMT2COMPAT:
+			tranp->t_truncategenpr = BOOL_TRUE;
+			break;
 		}
 	}
 
@@ -1473,6 +1481,13 @@ content_init( intgen_t argc, char *argv[ ], size64_t vmsz )
 			      GETOPT_NOSUBTREE );
 			return BOOL_FALSE;
 		}
+		if ( tranp->t_truncategenpr ) {
+			mlog( MLOG_NORMAL | MLOG_ERROR, _(
+			      "-%c valid only when initiating "
+			      "cumulative restore\n"),
+			      GETOPT_FMT2COMPAT );
+			return BOOL_FALSE;
+		}
 	} else {
 		if ( ! resumepr && ! sesscpltpr ) {
 			mlog( MLOG_NORMAL | MLOG_ERROR, _(
@@ -1534,6 +1549,12 @@ content_init( intgen_t argc, char *argv[ ], size64_t vmsz )
 			      GETOPT_NOSUBTREE );
 			return BOOL_FALSE;
 		}
+		if ( tranp->t_truncategenpr ) {
+			mlog( MLOG_NORMAL | MLOG_ERROR, _(
+			      "-%c valid only when initiating restore\n"),
+			      GETOPT_FMT2COMPAT );
+			return BOOL_FALSE;
+		}
 	}
 
 	if ( persp->a.valpr ) {
@@ -2328,12 +2349,21 @@ content_stream_restore( ix_t thrdix )
 					tranp->t_vmsz,
 					fullpr,
 					persp->a.restoredmpr,
-					persp->a.dstdirisxfspr );
+					persp->a.dstdirisxfspr,
+					grhdrp->gh_version,
+					tranp->t_truncategenpr );
 			if ( ! ok ) {
 				Media_end( Mediap );
 				return mlog_exit(EXIT_ERROR, RV_ERROR);
 			}
 			tranp->t_treeinitdonepr = BOOL_TRUE;
+
+		} else {
+			ok = tree_check_dump_format( grhdrp->gh_version );
+			if ( ! ok ) {
+				Media_end( Mediap );
+				return mlog_exit(EXIT_ERROR, RV_ERROR);
+			}
 		}
 
 		/* commit the session and accumulative state
@@ -3071,7 +3101,7 @@ applydirdump( drive_t *drivep,
 				 */
 				rv = tree_addent( dirh,
 					     dhdrp->dh_ino,
-					     ( size_t )dhdrp->dh_gen,
+					     dhdrp->dh_gen,
 					     dhdrp->dh_name,
 					     namelen );
 				if ( rv != RV_OK ) {
@@ -8109,23 +8139,26 @@ read_dirent( drive_t *drivep,
 	     size_t direntbufsz,
 	     bool_t dhcs )
 {
+	global_hdr_t *grhdrp = drivep->d_greadhdrp;
 	drive_ops_t *dop = drivep->d_opsp;
 	/* REFERENCED */
 	intgen_t nread;
 	intgen_t rval;
 	direnthdr_t tmpdh;
+	char *namep;    // beginning of name following the direnthdr_t
+
+	ASSERT( sizeof( direnthdr_t ) == DIRENTHDR_SZ );
+	ASSERT( sizeof( direnthdr_v1_t ) == DIRENTHDR_SZ );
 
 	/* read the head of the dirent
 	 */
 	nread = read_buf( ( char * )&tmpdh,
-			  sizeof( direnthdr_t ),
+			  DIRENTHDR_SZ,
 			  ( void * )drivep,
 			  ( rfp_t )dop->do_read,
 			  ( rrbfp_t )
 			  dop->do_return_read_buf,
 			  &rval );
-	xlate_direnthdr(&tmpdh, dhdrp, 1);
-
 	switch( rval ) {
 	case 0:
 		break;
@@ -8142,27 +8175,46 @@ read_dirent( drive_t *drivep,
 	default:
 		return RV_CORE;
 	}
-	ASSERT( ( size_t )nread == sizeof( direnthdr_t ));
+	ASSERT( ( size_t )nread == DIRENTHDR_SZ );
 
-	mlog( MLOG_NITTY,
-	      "read dirent hdr ino %llu gen %u size %u\n",
-	      dhdrp->dh_ino,
-	      ( size_t )dhdrp->dh_gen,
-	      ( size_t )dhdrp->dh_sz );
+	if ( grhdrp->gh_version >= GLOBAL_HDR_VERSION_3 ) {
+		xlate_direnthdr(&tmpdh, dhdrp, 1);
+		namep = dhdrp->dh_name + sizeof(dhdrp->dh_name);
 
-	if ( dhcs ) {
-		if ( dhdrp->dh_sz == 0 ) {
+		if ( dhcs && !is_checksum_valid( dhdrp, DIRENTHDR_SZ )) {
 			mlog( MLOG_NORMAL | MLOG_WARNING, _(
-			      "corrupt directory entry header\n") );
+				"bad directory entry header checksum\n") );
 			return RV_CORRUPT;
 		}
-		if ( !is_checksum_valid( dhdrp, DIRENTHDR_SZ )) {
+	} else {
+		direnthdr_v1_t dhdr_v1;
+		xlate_direnthdr_v1((direnthdr_v1_t *)&tmpdh, &dhdr_v1, 1);
+		dhdrp->dh_ino = dhdr_v1.dh_ino;
+		dhdrp->dh_gen = BIGGEN2GEN(dhdr_v1.dh_gen);
+		dhdrp->dh_checksum = dhdr_v1.dh_checksum;
+		dhdrp->dh_sz = dhdr_v1.dh_sz;
+		memcpy(dhdrp->dh_name, dhdr_v1.dh_name, sizeof(dhdr_v1.dh_name));
+		namep = dhdrp->dh_name + sizeof(dhdr_v1.dh_name);
+
+		if ( dhcs && !is_checksum_valid( &dhdr_v1, DIRENTHDR_SZ )) {
 			mlog( MLOG_NORMAL | MLOG_WARNING, _(
-			      "bad directory entry header checksum\n") );
+				"bad directory entry header checksum\n") );
 			return RV_CORRUPT;
 		}
 	}
 
+	mlog( MLOG_NITTY,
+	      "read dirent hdr ino %llu gen %u size %u\n",
+	      dhdrp->dh_ino,
+	      ( size_t )dhdrp->dh_gen,
+	      ( size_t )dhdrp->dh_sz );
+
+	if ( dhdrp->dh_sz == 0 ) {
+		mlog( MLOG_NORMAL | MLOG_WARNING, _(
+			"corrupt directory entry header\n") );
+		return RV_CORRUPT;
+	}
+
 	/* if null, return
 	 */
 	if ( dhdrp->dh_ino == 0 ) {
@@ -8177,7 +8229,7 @@ read_dirent( drive_t *drivep,
 	ASSERT( ! ( ( size_t )dhdrp->dh_sz & ( DIRENTHDR_ALIGN - 1 )));
 	if ( ( size_t )dhdrp->dh_sz > sizeof( direnthdr_t )) {
 		size_t remsz = ( size_t )dhdrp->dh_sz - sizeof( direnthdr_t );
-		nread = read_buf( ( char * )( dhdrp + 1 ),
+		nread = read_buf( namep,
 				  remsz,
 				  ( void * )drivep,
 				  ( rfp_t )dop->do_read,
diff --git a/restore/getopt.h b/restore/getopt.h
index 63568de9..361bc619 100644
--- a/restore/getopt.h
+++ b/restore/getopt.h
@@ -26,7 +26,7 @@
  * purpose is to contain that command string.
  */
 
-#define GETOPT_CMDSTRING	"a:b:c:def:himn:op:qrs:tv:wABCDEFG:H:I:JL:M:NO:PQRS:TUVWX:Y:"
+#define GETOPT_CMDSTRING	"a:b:c:def:himn:op:qrs:tv:wABCDEFG:H:I:JKL:M:NO:PQRS:TUVWX:Y:"
 
 #define GETOPT_WORKSPACE	'a'	/* workspace dir (content.c) */
 #define GETOPT_BLOCKSIZE        'b'     /* blocksize for rmt */
@@ -64,7 +64,7 @@
 #define GETOPT_MAXSTACKSZ	'H'	/* maximum stack size (bytes) */
 #define GETOPT_INVPRINT         'I'     /* just display the inventory */
 #define	GETOPT_NOINVUPDATE	'J'	/* do not update the dump inventory */
-/*				'K' */
+#define GETOPT_FMT2COMPAT	'K'	/* force use format 2 gen numbers */
 #define	GETOPT_DUMPLABEL	'L'	/* dump session label (global.c) */
 #define	GETOPT_MEDIALABEL	'M'	/* media object label (media.c) */
 #define	GETOPT_TIMESTAMP	'N'	/* show timestamps in log msgs */
diff --git a/restore/tree.c b/restore/tree.c
index 05e06281..6b4c8313 100644
--- a/restore/tree.c
+++ b/restore/tree.c
@@ -102,6 +102,10 @@ struct treePersStorage {
 	bool_t p_restoredmpr;
 		/* restore DMI event settings
 		 */
+	bool_t p_truncategenpr;
+		/* truncate inode generation number (for compatibility
+		 * with xfsdump format 2 and earlier)
+		 */
 };
 
 typedef struct treePersStorage treepers_t;
@@ -163,7 +167,7 @@ typedef struct tran tran_t;
 
 /* node structure. each node represents a directory entry
  */
-#define NODESZ	48
+#define NODESZ	56
 
 struct node {
 	xfs_ino_t n_ino;	/* 8  8 ino */
@@ -175,9 +179,10 @@ struct node {
 	nh_t n_sibprevh;	/* 4 36 prev sibling list - dbl link list */
 	nh_t n_cldh;		/* 4 40 children list */
 	nh_t n_lnkh;		/* 4 44 hard link list */
-	gen_t n_gen;		/* 2 46 generation count mod 0x10000 */
-	u_char_t n_flags;	/* 1 47 action and state flags */
-	u_char_t n_nodehkbyte;	/* 1 48 given to node abstraction */
+	gen_t n_gen;		/* 4 48 generation count mod 0x10000 */
+	u_char_t n_flags;	/* 1 49 action and state flags */
+	u_char_t n_nodehkbyte;	/* 1 50 given to node abstraction */
+	char n_pad[6];		/* 6 56 */
 };
 
 typedef struct node node_t;
@@ -335,7 +340,9 @@ tree_init( char *hkdir,
 	   size64_t vmsz,
 	   bool_t fullpr,
 	   bool_t restoredmpr,
-	   bool_t dstdirisxfspr )
+	   bool_t dstdirisxfspr,
+	   u_int32_t dumpformat,
+	   bool_t truncategenpr )
 {
 	off64_t nodeoff;
 	char *perspath;
@@ -496,6 +503,21 @@ tree_init( char *hkdir,
 	 */
 	persp->p_restoredmpr = restoredmpr;
 
+	/* record if truncated generation numbers are required
+	 */
+	if ( dumpformat < GLOBAL_HDR_VERSION_3 ) {
+		persp->p_truncategenpr = BOOL_TRUE;
+		mlog( MLOG_NORMAL | MLOG_DEBUG | MLOG_TREE, _(
+		      "dump format version %u used truncated inode generation numbers\n"),
+			dumpformat );
+	} else if ( truncategenpr ) {
+		persp->p_truncategenpr = BOOL_TRUE;
+		mlog( MLOG_NORMAL | MLOG_DEBUG | MLOG_TREE, _(
+		      "forcing use of truncated inode generation numbers\n"));
+	} else {
+		persp->p_truncategenpr = BOOL_FALSE;
+	}
+
 	return BOOL_TRUE;
 }
 
@@ -596,6 +618,15 @@ tree_sync( char *hkdir,
 	 */
 	persp->p_fullpr = fullpr;
 
+	/* regardless of the format of this dump, if the previously applied
+	 * dump used truncated generation numbers, then we need to as well.
+	 */
+	if ( persp->p_truncategenpr ) {
+		mlog( MLOG_NORMAL | MLOG_DEBUG | MLOG_TREE, _(
+		      "using truncated inode generation numbers for "
+		      "compatibility with previously applied restore\n") );
+	}
+
 	/* rsynchronize with the hash abstraction. it will map more of the
 	 * persistent state file.
 	 */
@@ -621,6 +652,24 @@ tree_sync( char *hkdir,
 	return BOOL_TRUE;
 }
 
+bool_t
+tree_check_dump_format( u_int32_t dumpformat )
+{
+	if ( dumpformat < GLOBAL_HDR_VERSION_3 && !persp->p_truncategenpr ) {
+		mlog( MLOG_NORMAL | MLOG_ERROR | MLOG_TREE, _(
+		      "encountered dump format %d after a "
+		      "restore of format %d or newer\n"),
+			dumpformat, GLOBAL_HDR_VERSION_3 );
+		mlog( MLOG_NORMAL | MLOG_ERROR | MLOG_TREE, _(
+		      "to restore this series of dumps, use the -%c "
+		      "option on the first restore\n"),
+			GETOPT_FMT2COMPAT );
+		return BOOL_FALSE;
+	}
+
+	return BOOL_TRUE;
+}
+
 /* recursively descend the tree clearing REFED and DIRDUMPED and NEWORPH
  * flags. force the orphanage to be refed and dumped, so we won't try
  * to orphan it, and so things added to it won't look like they are
@@ -682,10 +731,13 @@ tree_begindir( filehdr_t *fhdrp, dah_t *dahp )
 {
 	nh_t hardh;
 	xfs_ino_t ino = fhdrp->fh_stat.bs_ino;
-	u_int32_t biggen = fhdrp->fh_stat.bs_gen;
-	gen_t gen = BIGGEN2GEN( biggen );
+	gen_t gen = fhdrp->fh_stat.bs_gen;
 	dah_t dah;
 
+	if ( persp->p_truncategenpr ) {
+		gen = BIGGEN2GEN( gen );
+	}
+
 	/* sanity check - orphino is supposed to be an unused ino!
 	 */
 	ASSERT( ino != orphino );
@@ -708,7 +760,7 @@ tree_begindir( filehdr_t *fhdrp, dah_t *dahp )
 			      "upgrading to dir\n",
 			      ino,
 			      gen,
-			      biggen );
+			      fhdrp->fh_stat.bs_gen );
 			if ( ! tranp->t_toconlypr ) {
 				ASSERT( hardp->n_dah == DAH_NULL );
 				hardp->n_dah = dirattr_add( fhdrp );
@@ -721,7 +773,7 @@ tree_begindir( filehdr_t *fhdrp, dah_t *dahp )
 			      "updating\n",
 			      ino,
 			      gen,
-			      biggen );
+			      fhdrp->fh_stat.bs_gen );
 			hardp->n_dah = dirattr_add( fhdrp );
 		} else {
 			/* case 3: already has dirattr; must be restart
@@ -731,7 +783,7 @@ tree_begindir( filehdr_t *fhdrp, dah_t *dahp )
 			      "retaining\n",
 			      ino,
 			      gen,
-			      biggen );
+			      fhdrp->fh_stat.bs_gen );
 		}
 		hardp->n_flags |= NF_ISDIR;
 		hardp->n_flags |= NF_DUMPEDDIR;
@@ -745,7 +797,7 @@ tree_begindir( filehdr_t *fhdrp, dah_t *dahp )
 		      "new\n",
 		      ino,
 		      gen,
-		      biggen );
+		      fhdrp->fh_stat.bs_gen );
 		if ( ! tranp->t_toconlypr ) {
 			dah = dirattr_add( fhdrp );
 		} else {
@@ -767,11 +819,14 @@ tree_begindir( filehdr_t *fhdrp, dah_t *dahp )
 }
 
 rv_t
-tree_addent( nh_t parh, xfs_ino_t ino, size_t g, char *name, size_t namelen )
+tree_addent( nh_t parh, xfs_ino_t ino, gen_t gen, char *name, size_t namelen )
 {
-	gen_t gen = BIGGEN2GEN( g );
 	nh_t hardh;
 
+	if ( persp->p_truncategenpr ) {
+		gen = BIGGEN2GEN( gen );
+	}
+
 	/* sanity check - orphino is supposed to be an unused ino!
 	 */
 	ASSERT( ino != orphino );
@@ -1677,7 +1732,7 @@ rename_dirs( nh_t cldh,
  */
 rv_t
 tree_cb_links( xfs_ino_t ino,
-	       u_int32_t biggen,
+	       gen_t gen,
 	       int32_t ctime,
 	       int32_t mtime,
 	       bool_t ( * funcp )( void *contextp,
@@ -1688,13 +1743,16 @@ tree_cb_links( xfs_ino_t ino,
 	       char *path1,
 	       char *path2 )
 {
-	gen_t gen = BIGGEN2GEN( biggen );
 	nh_t hardh;
 	nh_t nh;
 	char *path;
 	bool_t ok;
 	int  rval;
 
+	if ( persp->p_truncategenpr ) {
+		gen = BIGGEN2GEN( gen );
+	}
+
 	/* find the hardhead
 	 */
 	hardh = link_hardh( ino, gen );
@@ -1887,7 +1945,7 @@ tree_cb_links( xfs_ino_t ino,
 			      "ino %llu gen %u not referenced: "
 			      "placing in orphanage\n"),
 			      ino,
-			      biggen );
+			      gen );
 			nh = Node_alloc( ino,
 					 gen,
 					 NRH_NULL,
@@ -3357,6 +3415,7 @@ Node_alloc( xfs_ino_t ino, gen_t gen, nrh_t nrh, dah_t dah, size_t flags )
 	np->n_lnkh = NH_NULL;
 	np->n_gen = gen;
 	np->n_flags = ( u_char_t )flags;
+	memset(np->n_pad, 0, sizeof(np->n_pad));
 	Node_unmap( nh, &np  );
 	return nh;
 }
diff --git a/restore/tree.h b/restore/tree.h
index 93621c73..7b1a76a3 100644
--- a/restore/tree.h
+++ b/restore/tree.h
@@ -32,7 +32,9 @@ extern bool_t tree_init( char *hkdir,
 			 size64_t vmsz,
 			 bool_t fullpr,
 			 bool_t restoredmpr,
-			 bool_t dstdirisxfspr );
+			 bool_t dstdirisxfspr,
+			 u_int32_t dumpformat,
+			 bool_t truncategenpr );
 
 /* tree_sync - synchronizes with an existing tree abstraction
  */
@@ -42,6 +44,14 @@ extern bool_t tree_sync( char *hkdir,
 			 bool_t fullpr,
 			 bool_t dstdirisxfspr );
 
+/* tree_check_dump_format - detect the rare case where a
+ * cumulative restore begins with a format 3 (or newer)
+ * dump, and a later restore in the series encounters
+ * a format 2 dump. the restore will fail unless the
+ * original restore was told to use format 2 gen numbers.
+ */
+extern bool_t tree_check_dump_format( u_int32_t dumpformat );
+
 
 /* tree_begindir - begins application of dumped directory to tree.
  * returns handle to dir node. returns by reference the dirattr
@@ -53,7 +63,7 @@ extern nh_t tree_begindir( filehdr_t *fhdrp, dah_t *dahp );
  */
 extern rv_t tree_addent( nh_t dirh,
 			 xfs_ino_t ino,
-			 size_t gen,
+			 gen_t gen,
 			 char *name,
 			 size_t namelen );
 
@@ -84,7 +94,7 @@ extern bool_t tree_subtree_parse( bool_t sensepr, char *path );
 extern bool_t tree_post( char *path1, char *path2 );
 
 extern rv_t tree_cb_links( xfs_ino_t ino,
-			   u_int32_t biggen,
+			   gen_t gen,
 			   int32_t ctime,
 			   int32_t mtime,
 			   bool_t ( * funcp )( void *contextp,