afr: write zeros to sink for non-sparse files

Backport of http://review.gluster.org/#/c/12371/
Problem: If a file is created with zeroes ('dd', 'fallocate' etc.) when
a brick is down, the self-heal does not write the zeroes to the sink
after it comes up. Consequenty, there is a mismatch in disk-usage
amongst the bricks of the replica.

Fix: If we definitely know that the file is not sparse, then write the
zeroes to the sink even if the checksums match.

Change-Id: Ic739b3da5dbf47d99801c0e1743bb13aeb3af864
BUG: 1275921
Signed-off-by: Ravishankar N <ravishankar@redhat.com>
Reviewed-on: http://review.gluster.org/12436
Tested-by: NetBSD Build System <jenkins@build.gluster.org>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>

2 files changed, 43 insertions, 16 deletions

diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c
index 279f2faaaaf..6a3d6e13b1b 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -33,13 +33,18 @@ __checksum_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 		dict_t *xdata)
 {
 	afr_local_t *local = NULL;
+        struct afr_reply *replies = NULL;
 	int i = (long) cookie;
 
 	local = frame->local;
-
-	local->replies[i].valid = 1;
-	local->replies[i].op_ret = op_ret;
-	local->replies[i].op_errno = op_errno;
+        replies = local->replies;
+
+	replies[i].valid = 1;
+	replies[i].op_ret = op_ret;
+	replies[i].op_errno = op_errno;
+        if (xdata)
+                replies[i].buf_has_zeroes = dict_get_str_boolean (xdata,
+                                                   "buf-has-zeroes", _gf_false);
 	if (strong)
 		memcpy (local->replies[i].checksum, strong, MD5_DIGEST_LENGTH);
 
@@ -75,19 +80,23 @@ attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
 
 static gf_boolean_t
-__afr_selfheal_data_checksums_match (call_frame_t *frame, xlator_t *this,
-				     fd_t *fd, int source,
-				     unsigned char *healed_sinks,
-				     off_t offset, size_t size)
+__afr_can_skip_data_block_heal (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                                int source, unsigned char *healed_sinks,
+				off_t offset, size_t size,
+                                struct iatt *poststat)
 {
 	afr_private_t *priv = NULL;
 	afr_local_t *local = NULL;
 	unsigned char *wind_subvols = NULL;
+        gf_boolean_t checksum_match = _gf_true;
+        dict_t *xdata = NULL;
 	int i = 0;
 
 	priv = this->private;
 	local = frame->local;
-
+        xdata = dict_new();
+        if (xdata)
+                i = dict_set_int32 (xdata, "check-zero-filled", 1);
 	wind_subvols = alloca0 (priv->child_count);
 	for (i = 0; i < priv->child_count; i++) {
 		if (i == source || healed_sinks[i])
@@ -95,7 +104,9 @@ __afr_selfheal_data_checksums_match (call_frame_t *frame, xlator_t *this,
 	}
 
 	AFR_ONLIST (wind_subvols, frame, __checksum_cbk, rchecksum, fd,
-		    offset, size, NULL);
+		    offset, size, xdata);
+        if (xdata)
+                dict_unref (xdata);
 
 	if (!local->replies[source].valid || local->replies[source].op_ret != 0)
 		return _gf_false;
@@ -106,12 +117,26 @@ __afr_selfheal_data_checksums_match (call_frame_t *frame, xlator_t *this,
                 if (local->replies[i].valid) {
                         if (memcmp (local->replies[source].checksum,
                                     local->replies[i].checksum,
-                                    MD5_DIGEST_LENGTH))
-                                return _gf_false;
+                                    MD5_DIGEST_LENGTH)) {
+                                checksum_match = _gf_false;
+                                break;
+                        }
                 }
 	}
 
-	return _gf_true;
+        if (checksum_match) {
+                if (HAS_HOLES (poststat))
+                        return _gf_true;
+
+                /* For non-sparse files, we might be better off writing the
+                 * zeroes to sinks to avoid mismatch of disk-usage in bricks. */
+                if (local->replies[source].buf_has_zeroes)
+                        return _gf_false;
+                else
+                        return _gf_true;
+        }
+
+        return _gf_false;
 }
 
 
@@ -225,7 +250,6 @@ __afr_selfheal_data_read_write (call_frame_t *frame, xlator_t *this, fd_t *fd,
 	return ret;
 }
 
-
 static int
 afr_selfheal_data_block (call_frame_t *frame, xlator_t *this, fd_t *fd,
 			 int source, unsigned char *healed_sinks, off_t offset,
@@ -249,8 +273,9 @@ afr_selfheal_data_block (call_frame_t *frame, xlator_t *this, fd_t *fd,
 		}
 
 		if (type == AFR_SELFHEAL_DATA_DIFF &&
-		    __afr_selfheal_data_checksums_match (frame, this, fd, source,
-							 healed_sinks, offset, size)) {
+		    __afr_can_skip_data_block_heal (frame, this, fd, source,
+					            healed_sinks, offset, size,
+                                                   &replies[source].poststat)) {
 			ret = 0;
 			goto unlock;
 		}
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 9ee5ae36df0..c2fd1166d96 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -270,7 +270,9 @@ struct afr_reply {
 	struct iatt preparent;
 	struct iatt preparent2;
 	struct iatt postparent2;
+        /* For rchecksum */
 	uint8_t checksum[MD5_DIGEST_LENGTH];
+        gf_boolean_t buf_has_zeroes;
 };
 
 typedef enum {