From b166826b283d9071532174ebbec857dea600064b Mon Sep 17 00:00:00 2001
From: Ashish Pandey <aspandey@redhat.com>
Date: Thu, 23 Jul 2020 11:07:32 +0530
Subject: [PATCH 460/465] cluster/ec: Remove stale entries from indices/xattrop
 folder

Problem:
If a gfid is present in indices/xattrop folder while
the file/dir is actaully healthy and all the xattrs are healthy,
it causes lot of lookups by shd on an entry which does not need
to be healed.
This whole process eats up lot of CPU usage without doing meaningful
work.

Solution:
Set trusted.ec.dirty xattr of the entry so that actual heal process
happens and at the end of it, during unset of dirty, gfid enrty from
indices/xattrop will be removed.

>Upstream patch : https://review.gluster.org/#/c/glusterfs/+/24765/
>Fixes: #1385

Change-Id: Ib1b9377d8dda384bba49523e9ff6ba9f0699cc1b
BUG: 1785714
Signed-off-by: Ashish Pandey <aspandey@redhat.com>
Reviewed-on: https://code.engineering.redhat.com/gerrit/208591
Tested-by: RHGS Build Bot <nigelb@redhat.com>
Reviewed-by: Sunil Kumar Heggodu Gopala Acharya <sheggodu@redhat.com>
---
 xlators/cluster/ec/src/ec-heal.c  | 73 ++++++++++++++++++++++++++++++++++++++-
 xlators/cluster/ec/src/ec-types.h |  7 +++-
 2 files changed, 78 insertions(+), 2 deletions(-)

diff --git a/xlators/cluster/ec/src/ec-heal.c b/xlators/cluster/ec/src/ec-heal.c
index e2de879..7d25853 100644
--- a/xlators/cluster/ec/src/ec-heal.c
+++ b/xlators/cluster/ec/src/ec-heal.c
@@ -2488,6 +2488,59 @@ out:
     return ret;
 }
 
+int
+ec_heal_set_dirty_without_lock(call_frame_t *frame, ec_t *ec, inode_t *inode)
+{
+    int i = 0;
+    int ret = 0;
+    dict_t **xattr = NULL;
+    loc_t loc = {0};
+    uint64_t dirty_xattr[EC_VERSION_SIZE] = {0};
+    unsigned char *on = NULL;
+    default_args_cbk_t *replies = NULL;
+    dict_t *dict = NULL;
+
+    /* Allocate the required memory */
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    on = alloca0(ec->nodes);
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    xattr = GF_CALLOC(ec->nodes, sizeof(*xattr), gf_common_mt_pointer);
+    if (!xattr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    dict = dict_new();
+    if (!dict) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    for (i = 0; i < ec->nodes; i++) {
+        xattr[i] = dict;
+        on[i] = 1;
+    }
+    dirty_xattr[EC_METADATA_TXN] = hton64(1);
+    ret = dict_set_static_bin(dict, EC_XATTR_DIRTY, dirty_xattr,
+                              (sizeof(*dirty_xattr) * EC_VERSION_SIZE));
+    if (ret < 0) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    PARALLEL_FOP_ONLIST(ec->xl_list, on, ec->nodes, replies, frame,
+                        ec_wind_xattrop_parallel, &loc, GF_XATTROP_ADD_ARRAY64,
+                        xattr, NULL);
+out:
+    if (dict) {
+        dict_unref(dict);
+    }
+    if (xattr) {
+        GF_FREE(xattr);
+    }
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
 void
 ec_heal_do(xlator_t *this, void *data, loc_t *loc, int32_t partial)
 {
@@ -2563,7 +2616,18 @@ ec_heal_do(xlator_t *this, void *data, loc_t *loc, int32_t partial)
         ec_heal_inspect(frame, ec, loc->inode, up_subvols, _gf_false, _gf_false,
                         &need_heal);
 
-        if (need_heal == EC_HEAL_NONEED) {
+        if (need_heal == EC_HEAL_PURGE_INDEX) {
+            gf_msg(ec->xl->name, GF_LOG_INFO, 0, EC_MSG_HEAL_FAIL,
+                   "Index entry needs to be purged for: %s ",
+                   uuid_utoa(loc->gfid));
+            /* We need to send xattrop to set dirty flag so that it can be
+             * healed and index entry could be removed. We need not to take lock
+             * on this entry to do so as we are just setting dirty flag which
+             * actually increases the trusted.ec.dirty count and does not set
+             * the new value.
+             * This will make sure that it is not interfering in other fops.*/
+            ec_heal_set_dirty_without_lock(frame, ec, loc->inode);
+        } else if (need_heal == EC_HEAL_NONEED) {
             gf_msg(ec->xl->name, GF_LOG_DEBUG, 0, EC_MSG_HEAL_FAIL,
                    "Heal is not required for : %s ", uuid_utoa(loc->gfid));
             goto out;
@@ -2958,6 +3022,13 @@ _need_heal_calculate(ec_t *ec, uint64_t *dirty, unsigned char *sources,
                     goto out;
                 }
             }
+            /* If lock count is 0, all dirty flags are 0 and all the
+             * versions are macthing then why are we here. It looks
+             * like something went wrong while removing the index entries
+             * after completing a successful heal or fop. In this case
+             * we need to remove this index entry to avoid triggering heal
+             * in a loop and causing lookups again and again*/
+            *need_heal = EC_HEAL_PURGE_INDEX;
         } else {
             for (i = 0; i < ec->nodes; i++) {
                 /* Since each lock can only increment the dirty
diff --git a/xlators/cluster/ec/src/ec-types.h b/xlators/cluster/ec/src/ec-types.h
index f15429d..700dc39 100644
--- a/xlators/cluster/ec/src/ec-types.h
+++ b/xlators/cluster/ec/src/ec-types.h
@@ -130,7 +130,12 @@ typedef void (*ec_resume_f)(ec_fop_data_t *, int32_t);
 
 enum _ec_read_policy { EC_ROUND_ROBIN, EC_GFID_HASH, EC_READ_POLICY_MAX };
 
-enum _ec_heal_need { EC_HEAL_NONEED, EC_HEAL_MAYBE, EC_HEAL_MUST };
+enum _ec_heal_need {
+    EC_HEAL_NONEED,
+    EC_HEAL_MAYBE,
+    EC_HEAL_MUST,
+    EC_HEAL_PURGE_INDEX
+};
 
 enum _ec_stripe_part { EC_STRIPE_HEAD, EC_STRIPE_TAIL };
 
-- 
1.8.3.1