Package: Ledmon
Version: 0.95-2 & 0.97-1.1
Tags: patch
I have identified a memory leak in the ledmon package on servers with
the AMD SATA controller. In version 0.95 of ledmon, the amd_em_enabled
function in the amd.c file contains a pointer variable that is not
freed, leading to a significant memory leak. In version 0.97, there is
an additional memory leak in the vmdssd_check_slot_module function in
the vmdssd.c file along with the leak mentioned above, where another
pointer variable is not freed.
Here is a transcript:
/sbin/ledmon --foreground
pmap -p <ledmon pid> | grep -i total
anonymous memory: 132K
ps -eo pid,rss,comm | grep -i <ledmon pid>
<ledmon pid> 2252 ledmon
after a few minutes
pmap -p <ledmon pid> | grep -i total
anonymous memory: 232K
ps -eo pid,rss,comm | grep -i <ledmon pid>
<ledmon pid> 2656 ledmon
Here, we can identify a significant memory leak in the ledmon binary,
as it grows over time when the program runs for a few minutes.
Fix :
./src/amd.c:113
int amd_em_enabled(const char *path)
{
char *platform;
int rc;
/* Default to SGPIO interface */
amd_interface = AMD_INTF_SGPIO;
platform = get_text("/sys/class/dmi/id", "product_name");
if (!platform)
return 0;
/* Check IPMI platforms */
if (!strncmp(platform, "ETHANOL_X", 9)) {
amd_interface = AMD_INTF_IPMI;
amd_ipmi_platform = AMD_PLATFORM_ETHANOL_X;
} else if (!strncmp(platform, "DAYTONA_X", 9)) {
amd_interface = AMD_INTF_IPMI;
amd_ipmi_platform = AMD_PLATFORM_DAYTONA_X;
}
switch (amd_interface) {
case AMD_INTF_SGPIO:
rc = _amd_sgpio_em_enabled(path);
break;
case AMD_INTF_IPMI:
rc = _amd_ipmi_em_enabled(path);
break;
default:
log_error("Unknown interface for AMD %s platform\n",
platform);
rc = -EOPNOTSUPP;
break;
}
+ free(platform); // Free the allocated memory for platform
pointer variable
return rc;
}
/src/vmdssd.c:93
bool vmdssd_check_slot_module(const char *slot_path)
{
char *address;
struct cntrl_device *cntrl;
address = get_text(slot_path, "address");
if (address == NULL)
return false;
// Check if slot address contains vmd domain
list_for_each(sysfs_get_cntrl_devices(), cntrl) {
if (cntrl->cntrl_type == CNTRL_TYPE_VMD) {
if (cntrl->domain == NULL)
continue;
if (strstr(address, cntrl->domain) == NULL)
continue;
+ free(address); // Free memory before returning true
return true;
}
}
+ free(address); // Ensure memory is freed before returning false
return false;
}
I suggest that the proposed fix be implemented to address the memory
leak in servers with the AMD SATA controller. For reference, here is
the acknowledgment from the upstream developer:
https://github.com/intel/ledmon/issues/84#issuecomment-2391056141
I am using Debian GNU/Linux 11 (bullseye), kernel 5.10.191 and libc6 2.31-13.
