Compulsorily do self heal if file sizes differ.

If file sizes differ, then compulsorily do self-heal. If no 'wise'
sources are found, then pick a 'fool' with the biggest file size.
If even 'fools' aren't found, pick the 'innocent' source with the
biggest file size.

Signed-off-by: Anand V. Avati <avati@amp.gluster.com>

6 files changed, 100 insertions, 23 deletions

diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index 58ea4fdb992..7c2e403c72a 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -337,7 +337,7 @@ afr_sh_wise_nodes_exist (afr_node_character *characters, int child_count)
 
 
 /*
- * The 'wisdom' of a wise node is 0 if any other wise node accuses to it.
+ * The 'wisdom' of a wise node is 0 if any other wise node accuses it.
  * It is 1 if no other wise node accuses it. 
  * Only wise nodes with wisdom 1 are sources.
  *
@@ -407,32 +407,97 @@ afr_sh_mark_wisest_as_sources (int sources[],
         return nsources;
 }
 
+
+static int
+afr_sh_mark_if_size_differs (afr_self_heal_t *sh, int child_count)
+{
+        int32_t ** pending_matrix;
+        int i, j;
+
+        int size_differs = 0;
+
+        pending_matrix = sh->pending_matrix;
+
+        for (i = 0; i < child_count; i++) {
+                for (j = 0; j < child_count; j++) {
+                        if (SIZE_DIFFERS (&sh->buf[i], &sh->buf[j])
+                            && (pending_matrix[i][j] == 0)
+                            && (pending_matrix[j][i] == 0)) {
+                                
+                                pending_matrix[i][j] = 1;
+                                pending_matrix[j][i] = 1;
+
+                                size_differs = 1;
+                        }
+                }
+        }
+
+        return size_differs;
+}
+
         
 static int
-afr_sh_mark_a_fool_as_source (int sources[], afr_node_character *characters, 
-                              int child_count)
+afr_sh_mark_biggest_fool_as_source (afr_self_heal_t *sh,
+                                    afr_node_character *characters, 
+                                    int child_count)
 {
         int i = 0;
-        
-        int nsources = 0;
+        int biggest = 0;
         
         for (i = 0; i < child_count; i++) {
                 if (characters[i].type == AFR_NODE_FOOL) {
-                        sources[i] = 1;
-                        nsources++;
+                        biggest = i;
                         break;
                 }
         }
 
-        return nsources;
+        for (i = 0; i < child_count; i++) {
+                if (characters[i].type != AFR_NODE_FOOL)
+                        continue;
+                
+                if (SIZE_GREATER (&sh->buf[i], &sh->buf[biggest])) {
+                        biggest = i;
+                }
+        }
+
+        sh->sources[biggest] = 1;
+
+        return 1;
 }
 
-        
+
+static int
+afr_sh_mark_biggest_as_source (afr_self_heal_t *sh, int child_count)
+{
+        int biggest = 0;
+        int i;
+
+        for (i = 0; i < child_count; i++) {
+                if (SIZE_GREATER (&sh->buf[i], &sh->buf[biggest])) {
+                        biggest = i;
+                }
+        }
+
+        sh->sources[biggest] = 1;
+
+        return 1;
+}
+
+
 int
-afr_sh_mark_sources (int32_t *pending_matrix[], int sources[], int child_count)
+afr_sh_mark_sources (afr_self_heal_t *sh, int child_count,
+                     afr_self_heal_type type)
 {
 	int i = 0;
 
+        int32_t ** pending_matrix;
+        int *      sources;
+
+        int size_differs = 0;
+
+        pending_matrix = sh->pending_matrix;
+        sources        = sh->sources;
+
 	int nsources = 0;
 
         /* stores the 'characters' (innocent, fool, wise) of the nodes */
@@ -463,9 +528,15 @@ afr_sh_mark_sources (int32_t *pending_matrix[], int sources[], int child_count)
                 }
         }
 
+        if (type == AFR_SELF_HEAL_DATA) {
+                size_differs = afr_sh_mark_if_size_differs (sh, child_count);
+        }
+
         if (afr_sh_all_nodes_innocent (characters, child_count)) {
-                /* no self-heal needed */
-                goto out;
+                if (size_differs) {
+                        nsources = afr_sh_mark_biggest_as_source (sh,
+                                                                  child_count);
+                }
 
         } else if (afr_sh_wise_nodes_exist (characters, child_count)) {
                 afr_sh_compute_wisdom (pending_matrix, characters, child_count);
@@ -482,8 +553,8 @@ afr_sh_mark_sources (int32_t *pending_matrix[], int sources[], int child_count)
                                                                   child_count);
                 }
         } else {
-                nsources = afr_sh_mark_a_fool_as_source (sources, characters, 
-                                                         child_count);
+                nsources = afr_sh_mark_biggest_fool_as_source (sh, characters,
+                                                               child_count);
         }
 
 out:
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.h b/xlators/cluster/afr/src/afr-self-heal-common.h
index df786e99924..86f155b6897 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.h
+++ b/xlators/cluster/afr/src/afr-self-heal-common.h
@@ -22,6 +22,12 @@
 
 #define FILE_HAS_HOLES(buf) (((buf)->st_size) > ((buf)->st_blocks * 512))
 
+typedef enum {
+        AFR_SELF_HEAL_ENTRY,
+        AFR_SELF_HEAL_METADATA,
+        AFR_SELF_HEAL_DATA,
+} afr_self_heal_type;
+
 int
 afr_sh_select_source (int sources[], int child_count);
 
@@ -52,8 +58,8 @@ afr_sh_pending_to_delta (dict_t **xattr, char *key, int32_t *delta_matrix[],
 			 int32_t success[], int child_count);
 
 int
-afr_sh_mark_sources (int32_t *pending_matrix[], int sources[],
-		     int child_count);
+afr_sh_mark_sources (afr_self_heal_t *sh, int child_count,
+                     afr_self_heal_type type);
 
 int
 afr_sh_delta_to_xattr (int32_t *delta_matrix[], dict_t *xattr[],
diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c
index 8dad39257eb..2c1ceafffa1 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -779,8 +779,8 @@ afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
 	afr_sh_print_pending_matrix (sh->pending_matrix, this);
 
 
-	nsources = afr_sh_mark_sources (sh->pending_matrix, sh->sources,
-                                        priv->child_count);
+	nsources = afr_sh_mark_sources (sh, priv->child_count,
+                                        AFR_SELF_HEAL_DATA);
 
 	afr_sh_supress_empty_children (sh->sources, sh->xattr, sh->buf,
 				       priv->child_count, AFR_DATA_PENDING);
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index 665dfb93628..3906b707e0c 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -1830,8 +1830,8 @@ afr_sh_entry_fix (call_frame_t *frame, xlator_t *this)
 	afr_sh_print_pending_matrix (sh->pending_matrix, this);
 
 
-	afr_sh_mark_sources (sh->pending_matrix, sh->sources, 
-			     priv->child_count);
+	afr_sh_mark_sources (sh, priv->child_count,
+			     AFR_SELF_HEAL_ENTRY);
 
 	afr_sh_supress_errenous_children (sh->sources, sh->child_errno,
 					  priv->child_count);
diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c
index 9ce42691a83..6dce5d3afb0 100644
--- a/xlators/cluster/afr/src/afr-self-heal-metadata.c
+++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c
@@ -520,8 +520,8 @@ afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this)
 
 	afr_sh_print_pending_matrix (sh->pending_matrix, this);
 
-	nsources = afr_sh_mark_sources (sh->pending_matrix, sh->sources,
-                                        priv->child_count);
+	nsources = afr_sh_mark_sources (sh, priv->child_count,
+                                        AFR_SELF_HEAL_METADATA);
 
 	afr_sh_supress_errenous_children (sh->sources, sh->child_errno,
 					  priv->child_count);
diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h
index c98831b8bca..73abd8fb96f 100644
--- a/xlators/cluster/afr/src/afr-self-heal.h
+++ b/xlators/cluster/afr/src/afr-self-heal.h
@@ -27,7 +27,7 @@
 #define OWNERSHIP_DIFFERS(buf1,buf2) ((((struct stat *)buf1)->st_uid) != (((struct stat *)buf2)->st_uid) || (((struct stat *)buf1)->st_gid != (((struct stat *)buf2)->st_gid)))
 #define SIZE_DIFFERS(buf1,buf2) ((((struct stat *)buf1)->st_size) != (((struct stat *)buf2)->st_size))
 
-
+#define SIZE_GREATER(buf1,buf2) ((((struct stat *)buf1)->st_size > (((struct stat *)buf2)->st_size)))
 
 int
 afr_sh_has_metadata_pending (dict_t *xattr, int child_count, xlator_t *this);