cluster/ec: create eager-lock option for non-regular files

A new option is added to allow independent configuration of eager
locking for regular files and non-regular files.

Change-Id: I8f80e46d36d8551011132b15c0fac549b7fb1c60
BUG: 1502610
Signed-off-by: Xavier Hernandez <jahernan@redhat.com>

3 files changed, 47 insertions, 13 deletions

diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c
index f3463ba4489..6a15223e0cc 100644
--- a/xlators/cluster/ec/src/ec-common.c
+++ b/xlators/cluster/ec/src/ec-common.c
@@ -2445,6 +2445,26 @@ void ec_flush_size_version(ec_fop_data_t * fop)
     ec_update_info(&fop->locks[0]);
 }
 
+static gf_boolean_t
+ec_use_eager_lock(ec_t *ec, ec_fop_data_t *fop)
+{
+        /* Fops with no locks at this point mean that they are sent as sub-fops
+         * of other higher level fops. In this case we simply assume that the
+         * parent fop will take correct care of the eager lock. */
+        if (fop->lock_count == 0) {
+                return _gf_true;
+        }
+
+        /* We may have more than one lock, but this only happens in the rename
+         * fop, and both locks will reference an inode of the same type (a
+         * directory in this case), so we only need to check the first lock. */
+        if (fop->locks[0].lock->loc.inode->ia_type == IA_IFREG) {
+                return ec->eager_lock;
+        }
+
+        return ec->other_eager_lock;
+}
+
 void ec_lock_reuse(ec_fop_data_t *fop)
 {
     ec_cbk_data_t *cbk;
@@ -2454,7 +2474,7 @@ void ec_lock_reuse(ec_fop_data_t *fop)
     ec = fop->xl->private;
     cbk = fop->answer;
 
-    if (ec->eager_lock && cbk != NULL) {
+    if (ec_use_eager_lock(ec, fop) && cbk != NULL) {
         if (cbk->xdata != NULL) {
             if ((dict_get_int32(cbk->xdata, GLUSTERFS_INODELK_COUNT,
                                 &count) == 0) && (count > 1)) {
diff --git a/xlators/cluster/ec/src/ec-types.h b/xlators/cluster/ec/src/ec-types.h
index 9aed7dc1d27..23b30548450 100644
--- a/xlators/cluster/ec/src/ec-types.h
+++ b/xlators/cluster/ec/src/ec-types.h
@@ -573,6 +573,7 @@ struct _ec {
     gf_timer_t        *timer;
     gf_boolean_t       shutdown;
     gf_boolean_t       eager_lock;
+    gf_boolean_t       other_eager_lock;
     gf_boolean_t       optimistic_changelog;
     gf_boolean_t       parallel_writes;
     uint32_t           background_heals;
diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c
index 09c5fa83eb9..9f361a54aa3 100644
--- a/xlators/cluster/ec/src/ec.c
+++ b/xlators/cluster/ec/src/ec.c
@@ -276,6 +276,8 @@ reconfigure (xlator_t *this, dict_t *options)
                           bool, failed);
         GF_OPTION_RECONF ("eager-lock", ec->eager_lock, options,
                           bool, failed);
+        GF_OPTION_RECONF ("other-eager-lock", ec->other_eager_lock, options,
+                          bool, failed);
         GF_OPTION_RECONF ("background-heals", background_heals, options,
                           uint32, failed);
         GF_OPTION_RECONF ("heal-wait-qlength", heal_wait_qlen, options,
@@ -654,6 +656,7 @@ init (xlator_t *this)
     GF_OPTION_INIT ("self-heal-daemon", ec->shd.enabled, bool, failed);
     GF_OPTION_INIT ("iam-self-heal-daemon", ec->shd.iamshd, bool, failed);
     GF_OPTION_INIT ("eager-lock", ec->eager_lock, bool, failed);
+    GF_OPTION_INIT ("other-eager-lock", ec->other_eager_lock, bool, failed);
     GF_OPTION_INIT ("background-heals", ec->background_heals, uint32, failed);
     GF_OPTION_INIT ("heal-wait-qlength", ec->heal_wait_qlen, uint32, failed);
     GF_OPTION_INIT ("self-heal-window-size", ec->self_heal_window_size, uint32,
@@ -1397,18 +1400,28 @@ struct volume_options options[] =
     { .key = {"eager-lock"},
       .type = GF_OPTION_TYPE_BOOL,
       .default_value = "on",
-      .description = "Enable/Disable eager lock for disperse volume. "
-                     "If a fop takes a lock and completes its operation, "
-                     "it waits for next 1 second before releasing the lock, "
-                     "to see if the lock can be reused for next fop from "
-                     "the same client. If ec finds any lock contention within "
-                     "1 second it releases the lock immediately before time "
-                     "expires. This improves the performance of file operations."
-                     "However, as it takes lock on first brick, for few operations "
-                     "like read, discovery of lock contention might take long time "
-                     "and can actually degrade the performance. "
-                     "If eager lock is disabled, lock will be released as soon as fop "
-                     "completes. "
+      .description = "Enable/Disable eager lock for regular files on a "
+                     "disperse volume. If a fop takes a lock and completes "
+                     "its operation, it waits for next 1 second before "
+                     "releasing the lock, to see if the lock can be reused "
+                     "for next fop from the same client. If ec finds any lock "
+                     "contention within 1 second it releases the lock "
+                     "immediately before time expires. This improves the "
+                     "performance of file operations. However, as it takes "
+                     "lock on first brick, for few operations like read, "
+                     "discovery of lock contention might take long time and "
+                     "can actually degrade the performance. If eager lock is "
+                     "disabled, lock will be released as soon as fop "
+                     "completes."
+    },
+    { .key = {"other-eager-lock"},
+      .type = GF_OPTION_TYPE_BOOL,
+      .default_value = "on",
+      .op_version = { GD_OP_VERSION_3_12_3 },
+      .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+      .tags = { "disperse" },
+      .description = "It's equivalent to the eager-lock option but for non "
+                     "regular files."
     },
     { .key = {"background-heals"},
       .type = GF_OPTION_TYPE_INT,