Potential memory leak in hlindex_open when open_table_from_share fails

[UsedByFeiYang]

In sql/vidx/vidx_index.cc, the hlindex_open function allocates memory for hlindex with my_malloc() at line 980, and increments s->hlindex->ref_count() at lines 988-990 before calling open_table_from_share().

However, when open_table_from_share() fails (error != 0), the function jumps to error_end at line 1001 without:

Decrementing the ref_count that was previously incremented
Freeing the allocated hlindex memory
Code location :

// Line 980: Allocate hlindex
hlindex = (TABLE *)my_malloc(key_memory_TABLE, sizeof(*hlindex), MYF(MY_WME));

// Lines 987-990: Increment ref_count
if (s->hlindex->tmp_table == NO_TMP_TABLE) {
mysql_mutex_lock(&LOCK_open);
s->hlindex->increment_ref_count();
mysql_mutex_unlock(&LOCK_open);
}

// Lines 993-997: Call open_table_from_share
int error = open_table_from_share(in_use, s->hlindex, hlindex_name,
(uint)(HA_OPEN_KEYFILE | HA_OPEN_RNDFILE |
HA_GET_INDEX | HA_TRY_READ_ONLY),
EXTRA_RECORD, in_use->open_options,
hlindex, false, hlindex_dd);

// Lines 999-1001: Problem: When error != 0, no cleanup before goto error_end
if (error != 0 || hlindex == nullptr) {
error_message = "Failed to open_table_from_share.";
goto error_end; // Memory leak here!
}

Note: open_table_from_share() does NOT free the passed outparam pointer on failure (verified in sql/table.cc). It only cleans internal resources.

Suggested fix:

if (error != 0 || hlindex == nullptr) {
error_message = "Failed to open_table_from_share.";
if (hlindex != nullptr) {
if (s->hlindex->tmp_table == NO_TMP_TABLE) {
mysql_mutex_lock(&LOCK_open);
s->hlindex->decrement_ref_count();
mysql_mutex_unlock(&LOCK_open);
}
my_free(hlindex);
hlindex = nullptr;
}
goto error_end;
}

[JoeJRW]

Thanks for the detailed report and the suggested fix.

I have verified this case, and the issue does exist. The failure path is rarely triggered in practice, so the probability of hitting it should be very low, but it is still a real cleanup issue.

I will fix it in a future release. Thanks again for helping improve AliSQL.