OpenZFS 6513 - partially filled holes lose birth time

Reviewed by: Matthew Ahrens <[email protected]>
Reviewed by: George Wilson <[email protected]>
Reviewed by: Boris Protopopov <[email protected]>
Approved by: Richard Lowe <[email protected]>a
Ported by: Boris Protopopov <[email protected]>
Signed-off-by: Boris Protopopov <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>

OpenZFS-issue: https://www.illumos.org/issues/6513
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/8df0bcf0

If a ZFS object contains a hole at level one, and then a data block is
created at level 0 underneath that l1 block, l0 holes will be created.
However, these l0 holes do not have the birth time property set; as a
result, incremental sends will not send those holes.

Fix is to modify the dbuf_read code to fill in birth time data.

1 files changed, 118 insertions, 7 deletions

diff --git a/module/zfs/dbuf.c b/module/zfs/dbuf.c
index fd51d59a8..126748994 100644
--- a/module/zfs/dbuf.c
+++ b/module/zfs/dbuf.c
@@ -543,13 +543,50 @@ dbuf_verify(dmu_buf_impl_t *db)
 		 * If the blkptr isn't set but they have nonzero data,
 		 * it had better be dirty, otherwise we'll lose that
 		 * data when we evict this buffer.
+		 *
+		 * There is an exception to this rule for indirect blocks; in
+		 * this case, if the indirect block is a hole, we fill in a few
+		 * fields on each of the child blocks (importantly, birth time)
+		 * to prevent hole birth times from being lost when you
+		 * partially fill in a hole.
 		 */
 		if (db->db_dirtycnt == 0) {
-			ASSERTV(uint64_t *buf = db->db.db_data);
-			int i;
+			if (db->db_level == 0) {
+				uint64_t *buf = db->db.db_data;
+				int i;
 
-			for (i = 0; i < db->db.db_size >> 3; i++) {
-				ASSERT(buf[i] == 0);
+				for (i = 0; i < db->db.db_size >> 3; i++) {
+					ASSERT(buf[i] == 0);
+				}
+			} else {
+				int i;
+				blkptr_t *bps = db->db.db_data;
+				ASSERT3U(1 << DB_DNODE(db)->dn_indblkshift, ==,
+				    db->db.db_size);
+				/*
+				 * We want to verify that all the blkptrs in the
+				 * indirect block are holes, but we may have
+				 * automatically set up a few fields for them.
+				 * We iterate through each blkptr and verify
+				 * they only have those fields set.
+				 */
+				for (i = 0;
+				    i < db->db.db_size / sizeof (blkptr_t);
+				    i++) {
+					blkptr_t *bp = &bps[i];
+					ASSERT(ZIO_CHECKSUM_IS_ZERO(
+					    &bp->blk_cksum));
+					ASSERT(
+					    DVA_IS_EMPTY(&bp->blk_dva[0]) &&
+					    DVA_IS_EMPTY(&bp->blk_dva[1]) &&
+					    DVA_IS_EMPTY(&bp->blk_dva[2]));
+					ASSERT0(bp->blk_fill);
+					ASSERT0(bp->blk_pad[0]);
+					ASSERT0(bp->blk_pad[1]);
+					ASSERT(!BP_IS_EMBEDDED(bp));
+					ASSERT(BP_IS_HOLE(bp));
+					ASSERT0(bp->blk_phys_birth);
+				}
 			}
 		}
 	}
@@ -718,10 +755,32 @@ dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags)
 	    BP_IS_HOLE(db->db_blkptr)))) {
 		arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
 
-		DB_DNODE_EXIT(db);
 		dbuf_set_data(db, arc_buf_alloc(db->db_objset->os_spa,
 		    db->db.db_size, db, type));
 		bzero(db->db.db_data, db->db.db_size);
+
+		if (db->db_blkptr != NULL && db->db_level > 0 &&
+		    BP_IS_HOLE(db->db_blkptr) &&
+		    db->db_blkptr->blk_birth != 0) {
+			blkptr_t *bps = db->db.db_data;
+			int i;
+			for (i = 0; i < ((1 <<
+			    DB_DNODE(db)->dn_indblkshift) / sizeof (blkptr_t));
+			    i++) {
+				blkptr_t *bp = &bps[i];
+				ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==,
+				    1 << dn->dn_indblkshift);
+				BP_SET_LSIZE(bp,
+				    BP_GET_LEVEL(db->db_blkptr) == 1 ?
+				    dn->dn_datablksz :
+				    BP_GET_LSIZE(db->db_blkptr));
+				BP_SET_TYPE(bp, BP_GET_TYPE(db->db_blkptr));
+				BP_SET_LEVEL(bp,
+				    BP_GET_LEVEL(db->db_blkptr) - 1);
+				BP_SET_BIRTH(bp, db->db_blkptr->blk_birth, 0);
+			}
+		}
+		DB_DNODE_EXIT(db);
 		db->db_state = DB_CACHED;
 		mutex_exit(&db->db_mtx);
 		return (0);
@@ -3094,6 +3153,45 @@ dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
 	rw_exit(&dn->dn_struct_rwlock);
 }
 
+/* ARGSUSED */
+/*
+ * This function gets called just prior to running through the compression
+ * stage of the zio pipeline. If we're an indirect block comprised of only
+ * holes, then we want this indirect to be compressed away to a hole. In
+ * order to do that we must zero out any information about the holes that
+ * this indirect points to prior to before we try to compress it.
+ */
+static void
+dbuf_write_children_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
+{
+	dmu_buf_impl_t *db = vdb;
+	dnode_t *dn;
+	blkptr_t *bp;
+	uint64_t i;
+	int epbs;
+
+	ASSERT3U(db->db_level, >, 0);
+	DB_DNODE_ENTER(db);
+	dn = DB_DNODE(db);
+	epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT;
+
+	/* Determine if all our children are holes */
+	for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) {
+		if (!BP_IS_HOLE(bp))
+			break;
+	}
+
+	/*
+	 * If all the children are holes, then zero them all out so that
+	 * we may get compressed away.
+	 */
+	if (i == 1 << epbs) {
+		/* didn't find any non-holes */
+		bzero(db->db.db_data, db->db.db_size);
+	}
+	DB_DNODE_EXIT(db);
+}
+
 /*
  * The SPA will call this callback several times for each zio - once
  * for every physical child i/o (zio->io_phys_children times).  This
@@ -3348,7 +3446,8 @@ dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx)
 
 		dr->dr_zio = zio_write(zio, os->os_spa, txg,
 		    &dr->dr_bp_copy, contents, db->db.db_size, &zp,
-		    dbuf_write_override_ready, NULL, dbuf_write_override_done,
+		    dbuf_write_override_ready, NULL, NULL,
+		    dbuf_write_override_done,
 		    dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
 		mutex_enter(&db->db_mtx);
 		dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
@@ -3359,14 +3458,26 @@ dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx)
 		ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF);
 		dr->dr_zio = zio_write(zio, os->os_spa, txg,
 		    &dr->dr_bp_copy, NULL, db->db.db_size, &zp,
-		    dbuf_write_nofill_ready, NULL, dbuf_write_nofill_done, db,
+		    dbuf_write_nofill_ready, NULL, NULL,
+		    dbuf_write_nofill_done, db,
 		    ZIO_PRIORITY_ASYNC_WRITE,
 		    ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_NODATA, &zb);
 	} else {
+		arc_done_func_t *children_ready_cb = NULL;
 		ASSERT(arc_released(data));
+
+		/*
+		 * For indirect blocks, we want to setup the children
+		 * ready callback so that we can properly handle an indirect
+		 * block that only contains holes.
+		 */
+		if (db->db_level != 0)
+			children_ready_cb = dbuf_write_children_ready;
+
 		dr->dr_zio = arc_write(zio, os->os_spa, txg,
 		    &dr->dr_bp_copy, data, DBUF_IS_L2CACHEABLE(db),
 		    DBUF_IS_L2COMPRESSIBLE(db), &zp, dbuf_write_ready,
+		    children_ready_cb,
 		    dbuf_write_physdone, dbuf_write_done, db,
 		    ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
 	}