Add command-line support (but no doc) for enforce-quorum option.

Change-Id: Ia52ddb551e24c27969f7f5fa0f94c1044789731f
BUG: 3823
Reviewed-on: http://review.gluster.com/743
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vijay@gluster.com>

7 files changed, 120 insertions, 69 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index 9512f6a54..209024bbe 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -3893,31 +3893,36 @@ out:
 gf_boolean_t
 afr_have_quorum (char *logname, afr_private_t *priv)
 {
-	unsigned int	quorum = 0;
-
-	GF_VALIDATE_OR_GOTO(logname,priv,out);
-
-	quorum = priv->child_count / 2 + 1;
-	if (priv->up_count >= (priv->down_count + quorum)) {
-		return _gf_true;
-	}
-
-	/*
-	 * Special case for even numbers of nodes: if we have exactly half
-	 * and that includes the first ("senior-most") node, then that counts
-	 * as quorum even if it wouldn't otherwise.  This supports e.g. N=2
-	 * while preserving the critical property that there can only be one
-	 * such group.
-	 */
-	if ((priv->child_count % 2) == 0) {
-		quorum = priv->child_count / 2;
-		if (priv->up_count >= (priv->down_count + quorum)) {
-			if (priv->child_up[0]) {
-				return _gf_true;
-			}
-		}
-	}
+        unsigned int        quorum = 0;
+
+        GF_VALIDATE_OR_GOTO(logname,priv,out);
+
+        quorum = priv->quorum_count;
+        if (quorum != AFR_QUORUM_AUTO) {
+                return (priv->up_count >= (priv->down_count + quorum));
+        }
+
+        quorum = priv->child_count / 2 + 1;
+        if (priv->up_count >= (priv->down_count + quorum)) {
+                return _gf_true;
+        }
+
+        /*
+         * Special case for even numbers of nodes: if we have exactly half
+         * and that includes the first ("senior-most") node, then that counts
+         * as quorum even if it wouldn't otherwise.  This supports e.g. N=2
+         * while preserving the critical property that there can only be one
+         * such group.
+         */
+        if ((priv->child_count % 2) == 0) {
+                quorum = priv->child_count / 2;
+                if (priv->up_count >= (priv->down_count + quorum)) {
+                        if (priv->child_up[0]) {
+                                return _gf_true;
+                        }
+                }
+        }
 
 out:
-	return _gf_false;
+        return _gf_false;
 }
diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c
index 11df550d5..7ed8c5205 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -279,7 +279,7 @@ afr_create (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(create,out);
+        QUORUM_CHECK(create,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -504,7 +504,7 @@ afr_mknod (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(mknod,out);
+        QUORUM_CHECK(mknod,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -730,7 +730,7 @@ afr_mkdir (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(mkdir,out);
+        QUORUM_CHECK(mkdir,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -955,7 +955,7 @@ afr_link (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(link,out);
+        QUORUM_CHECK(link,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -1179,7 +1179,7 @@ afr_symlink (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(symlink,out);
+        QUORUM_CHECK(symlink,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -1397,7 +1397,7 @@ afr_rename (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(rename,out);
+        QUORUM_CHECK(rename,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -1596,7 +1596,7 @@ afr_unlink (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(unlink,out);
+        QUORUM_CHECK(unlink,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -1792,7 +1792,7 @@ afr_rmdir (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(rmdir,out);
+        QUORUM_CHECK(rmdir,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index 4135ba947..2fd2538cd 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -449,7 +449,7 @@ afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
 
         priv = this->private;
 
-	QUORUM_CHECK(writev,out);
+        QUORUM_CHECK(writev,out);
 
         ALLOC_OR_GOTO (local, afr_local_t, out);
 
@@ -649,7 +649,7 @@ afr_truncate (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(truncate,out);
+        QUORUM_CHECK(truncate,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -900,7 +900,7 @@ afr_ftruncate (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(ftruncate,out);
+        QUORUM_CHECK(ftruncate,out);
 
         ALLOC_OR_GOTO (local, afr_local_t, out);
         ret = AFR_LOCAL_INIT (local, priv);
@@ -1099,7 +1099,7 @@ afr_setattr (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(setattr,out);
+        QUORUM_CHECK(setattr,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -1306,7 +1306,7 @@ afr_fsetattr (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(fsetattr,out);
+        QUORUM_CHECK(fsetattr,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
@@ -1497,7 +1497,7 @@ afr_setxattr (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(setxattr,out);
+        QUORUM_CHECK(setxattr,out);
 
         ALLOC_OR_GOTO (local, afr_local_t, out);
 
@@ -1683,7 +1683,7 @@ afr_removexattr (call_frame_t *frame, xlator_t *this,
 
         priv = this->private;
 
-	QUORUM_CHECK(removexattr,out);
+        QUORUM_CHECK(removexattr,out);
 
         transaction_frame = copy_frame (frame);
         if (!transaction_frame) {
diff --git a/xlators/cluster/afr/src/afr-open.c b/xlators/cluster/afr/src/afr-open.c
index 5031cfd5e..0b46f213f 100644
--- a/xlators/cluster/afr/src/afr-open.c
+++ b/xlators/cluster/afr/src/afr-open.c
@@ -214,9 +214,9 @@ afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
 
         priv = this->private;
 
-	if (flags & (O_CREAT|O_TRUNC)) {
-		QUORUM_CHECK(open,out);
-	}
+        if (flags & (O_CREAT|O_TRUNC)) {
+                QUORUM_CHECK(open,out);
+        }
 
         if (afr_is_split_brain (this, loc->inode)) {
                 /* self-heal failed */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index 7791ec86b..c66f386f2 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -97,14 +97,30 @@ xlator_subvolume_count (xlator_t *this)
         return i;
 }
 
+void
+fix_quorum_options (xlator_t *this, afr_private_t *priv, char *qtype)
+{
+        if (priv->quorum_count && strcmp(qtype,"fixed")) {
+                gf_log(this->name,GF_LOG_WARNING,
+                       "quorum-type %s overriding quorum-count %u",
+                       qtype, priv->quorum_count);
+        }
+        if (!strcmp(qtype,"none")) {
+                priv->quorum_count = 0;
+        }
+        else if (!strcmp(qtype,"auto")) {
+                priv->quorum_count = AFR_QUORUM_AUTO;
+        }
+}
 
 int
 reconfigure (xlator_t *this, dict_t *options)
 {
-        afr_private_t * priv        = NULL;
-        xlator_t      * read_subvol     = NULL;
-        int             ret = -1;
-        int             index = -1;
+        afr_private_t *priv        = NULL;
+        xlator_t      *read_subvol = NULL;
+        int            ret         = -1;
+        int            index       = -1;
+        char          *qtype       = NULL;
 
         priv = this->private;
 
@@ -154,6 +170,11 @@ reconfigure (xlator_t *this, dict_t *options)
                 priv->read_child = index;
         }
 
+        GF_OPTION_RECONF ("quorum-type", qtype, options, str, out);
+        GF_OPTION_RECONF ("quorum-count", priv->quorum_count, options,
+                          uint32, out);
+        fix_quorum_options(this,priv,qtype);
+
         ret = 0;
 out:
         return ret;
@@ -173,15 +194,15 @@ static const char *favorite_child_warning_str = "You have specified subvolume '%
 int32_t
 init (xlator_t *this)
 {
-        afr_private_t * priv        = NULL;
-        int             child_count = 0;
-        xlator_list_t * trav        = NULL;
-        int             i           = 0;
-        int             ret         = -1;
-        GF_UNUSED int   op_errno    = 0;
-        xlator_t * read_subvol     = NULL;
-        xlator_t * fav_child       = NULL;
-
+        afr_private_t *priv        = NULL;
+        int            child_count = 0;
+        xlator_list_t *trav        = NULL;
+        int            i           = 0;
+        int            ret         = -1;
+        GF_UNUSED int  op_errno    = 0;
+        xlator_t      *read_subvol = NULL;
+        xlator_t      *fav_child   = NULL;
+        char          *qtype       = NULL;
 
         if (!this->children) {
                 gf_log (this->name, GF_LOG_ERROR,
@@ -260,7 +281,9 @@ init (xlator_t *this)
 
         GF_OPTION_INIT ("strict-readdir", priv->strict_readdir, bool, out);
 
-        GF_OPTION_INIT ("enforce-quorum", priv->enforce_quorum, bool, out);
+        GF_OPTION_INIT ("quorum-type", qtype, str, out);
+        GF_OPTION_INIT ("quorum-count", priv->quorum_count, uint32, out);
+        fix_quorum_options(this,priv,qtype);
 
         priv->wait_count = 1;
 
@@ -492,9 +515,24 @@ struct volume_options options[] = {
           .type = GF_OPTION_TYPE_BOOL,
           .default_value = "off",
         },
-        { .key = {"enforce-quorum"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
+        { .key = {"quorum-type"},
+          .type = GF_OPTION_TYPE_STR,
+          .value = { "none", "auto", "fixed", "" },
+          .default_value = "none",
+          .description = "If value is \"fixed\" only allow writes if "
+                         "quorum-count bricks are present.  If value is "
+                         "\"auto\" only allow writes if more than half of "
+                         "bricks, or exactly half including the first, are "
+                         "present.",
+        },
+        { .key = {"quorum-count"},
+          .type = GF_OPTION_TYPE_INT,
+          .min = 1,
+          .max = INT_MAX,
+          .default_value = 0,
+          .description = "If quorum-type is \"fixed\" only allow writes if "
+                         "this many bricks or present.  Other quorum types "
+                         "will OVERWRITE this value.",
         },
         { .key  = {NULL} },
 };
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 0566495fe..4aea44c42 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -144,9 +144,9 @@ typedef struct _afr_private {
 
         pthread_mutex_t  mutex;
         struct list_head saved_fds;   /* list of fds on which locks have succeeded */
-        gf_boolean_t     optimistic_change_log;
-        gf_boolean_t     eager_lock;
-	gf_boolean_t     enforce_quorum;
+        gf_boolean_t      optimistic_change_log;
+        gf_boolean_t      eager_lock;
+        unsigned int      quorum_count;
 
         char                   vol_uuid[UUID_SIZE + 1];
         int32_t                *last_event;
@@ -1006,16 +1006,22 @@ gf_boolean_t
 afr_have_quorum (char *logname, afr_private_t *priv);
 
 /*
+ * Special value indicating we should use the "auto" quorum method instead of
+ * a fixed value (including zero to turn off quorum enforcement).
+ */
+#define AFR_QUORUM_AUTO INT_MAX
+
+/*
  * Having this as a macro will make debugging a bit weirder, but does reduce
  * the probability of functions handling this check inconsistently.
  */
 #define QUORUM_CHECK(_func,_label) do {                                  \
-	if (priv->enforce_quorum && !afr_have_quorum(this->name,priv)) { \
-		gf_log(this->name,GF_LOG_WARNING,                        \
-		       "failing "#_func" due to lack of quorum");        \
-		op_errno = EROFS;                                        \
-		goto _label;                                             \
-	}                                                                \
+        if (priv->quorum_count && !afr_have_quorum(this->name,priv)) { \
+                gf_log(this->name,GF_LOG_WARNING,                        \
+                       "failing "#_func" due to lack of quorum");        \
+                op_errno = EROFS;                                        \
+                goto _label;                                             \
+        }                                                                \
 } while (0);
 
 #endif /* __AFR_H__ */
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index 0fa7601f1..f83a052ed 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -125,6 +125,8 @@ static struct volopt_map_entry glusterd_volopt_map[] = {
         {"cluster.data-change-log",              "cluster/replicate",  NULL, NULL, NO_DOC, 0     },
         {"cluster.metadata-change-log",          "cluster/replicate",  NULL, NULL, NO_DOC, 0     },
         {"cluster.data-self-heal-algorithm",     "cluster/replicate",         "data-self-heal-algorithm", NULL,DOC, 0},
+        {"cluster.quorum-type",                  "cluster/replicate",  "quorum-type", NULL, NO_DOC, 0},
+        {"cluster.quorum-count",                 "cluster/replicate",  "quorum-count", NULL, NO_DOC, 0},
 
         {"cluster.stripe-block-size",            "cluster/stripe",            "block-size", NULL, DOC, 0},