Created
April 17, 2021 04:54
-
-
Save wenjianhn/e1d3258214576a5d0b963b12ecd51af2 to your computer and use it in GitHub Desktop.
A simple script that helps find the root cause of a memory leak issue
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#!/usr/bin/env python | |
from __future__ import print_function | |
import re | |
import shlex | |
import sys | |
import subprocess | |
from collections import Counter | |
# suggested by WWG | |
# def direct_ptov(p_addr): | |
# if p_addr >= phys_base: | |
# return p_addr - phys_base + __START_KERNEL_map | |
# else: | |
# return p_addr + PAGE_OFFSET | |
def crash_cmd(crash_process, cmd): | |
crash_process.stdin.write("%s\neval 1\n" % cmd) | |
crash_process.stdin.flush() | |
output = [] | |
while True: | |
line = crash_process.stdout.readline() | |
if line.endswith("%s\n" % '{:064}'.format(1)): | |
break | |
else: | |
output.append(line.rstrip()) | |
return output[:-3] | |
def get_slab_addr_objsize(crash_process, slab_name): | |
output = crash_cmd(crash_process, "kmem -s %s" % slab_name) | |
for line in output: | |
if line.startswith("CACHE"): | |
objsize_index = line.split().index("OBJSIZE") | |
elif line.startswith("ffff"): | |
fields = line.split() | |
addr = fields[0] | |
objsize = int(fields[objsize_index]) | |
return addr, objsize | |
else: | |
print("Failed to parse slab objsize", file=sys.stderr) | |
sys.exit(-1) | |
def page_address(page, vmemmap_base, page_offset_base): | |
return (((page - vmemmap_base)/64) << 12) + page_offset_base | |
def get_bases(crash_process): | |
vmemmap_base = 0xffffea0000000000 | |
page_offset_base = 0xffff880000000000 | |
output = crash_cmd(crash_process, "p -x vmemmap_base") | |
addr = output[0].split()[-1] | |
if addr.startswith("0xffff"): | |
vmemmap_base = long(addr, base=16) | |
output = crash_cmd(crash_process, "p -x page_offset_base") | |
addr = output[0].split()[-1] | |
page_offset_base = long(addr, base=16) | |
return vmemmap_base, page_offset_base | |
def get_release(crash_process): | |
cmd = "sys |grep RELEASE: | awk '{print $NF}'" | |
output = crash_cmd(crash_process, cmd) | |
return output[0] | |
def use_slub(crash_process): | |
if get_release(crash_process).startswith("2"): | |
return False | |
return True | |
def get_content(crash_process, slab_name): | |
vmemmap_base, page_offset_base = get_bases(crash_process) | |
slab_addr, objsize = get_slab_addr_objsize(crash_process, slab_name) | |
page_member = "slab_cache" if use_slub(crash_process) else "lru.next" | |
cmd = "kmem -m %s | grep %s | awk '{print $1}'" % (page_member, slab_addr) | |
output = crash_cmd(crash_process, cmd) | |
rd_cmds = [] | |
for line in output: | |
if line.startswith('ffff'): | |
addr = page_address(long(line, base=16), | |
vmemmap_base, page_offset_base) | |
rd_cmds.append("rd -x %x %d" % (addr, objsize/8)) | |
rd_cmds_output = [] | |
off = 0 | |
length = 1000 | |
while off < len(rd_cmds): | |
cmd = "\n".join(rd_cmds[off:off+length]) | |
output = crash_cmd(crash_process, cmd) | |
off += length | |
# # Only return contents that startswith ffff | |
# rd_cmds_output += [line for line in output if line.startswith('ffff')] | |
rd_cmds_output += output | |
return rd_cmds_output | |
def show(crash_process, top): | |
print('Content Count\tSymbol/Slab') | |
for v, c in top: | |
address = long(v, base=16) | |
symbol = "" | |
if 0xffff880000000000 < address < 0xffffc7ffffffffff: | |
output = crash_cmd(crash_process, "kmem %s" % v) | |
if output and output[0].startswith("CACHE"): | |
name_idx = output[0].split().index("NAME") | |
symbol += output[1].split()[name_idx] | |
elif 0xffffffff80000000 < address < 0xfffffffffeffffff: | |
cmd = "kmem %s |grep ^ffffffff | awk '{print $3,$4,$5}'" % v | |
output = crash_cmd(crash_process, cmd) | |
if output: | |
symbol += output[0] | |
print("%s %s\t%s" % (v, c, symbol)) | |
def show_usage(): | |
print("Usage:\n %s crash vmcore vmlinux slab_name [limit]" % sys.argv[0]) | |
print("Example:\n" | |
"Show top 10 content of the kmalloc-192 slab:\n" | |
" %s crash vmcore vmlinux kmalloc-192 10\n" | |
"Show top 20 content:\n" | |
" %s crash /var/crash/127.0.0.1-2020-04-16-23\:16\:56/vmcore " | |
"/usr/lib/debug/lib/modules/4.1.12-124.38.1.el7uek.x86_64/vmlinux " | |
"kmalloc-192" % (sys.argv[0], sys.argv[0])) | |
def open_crash_process(args): | |
p = subprocess.Popen(args, | |
stdin=subprocess.PIPE, | |
stdout=subprocess.PIPE, | |
stderr=sys.stderr) | |
crash_cmd(p, "eval 2") | |
return p | |
def show_slab_info(crash_process, slab_name): | |
output = crash_cmd(crash_process, "kmem -s %s" % slab_name) | |
print("\n".join(output)) | |
output = crash_cmd(crash_process, "p nr_online_nodes") | |
cnt = int(output[0].split()[-1]) | |
addr, _ = get_slab_addr_objsize(crash_process, slab_name) | |
output = crash_cmd(crash_process, "struct kmem_cache.node %s" % addr) | |
nodes = re.findall(r'0x\w+', output[0])[0:cnt] | |
print("") | |
for i, node in enumerate(nodes): | |
output = crash_cmd(crash_process, | |
"struct kmem_cache_node.nr_partial %s" % node) | |
nr_partial = int(output[0].split()[-1]) | |
print("Number of partially allocated slabs of %s on NUMA node %d: %d" % | |
(slab_name, i, nr_partial)) | |
print("") | |
def main(): | |
limit = 20 | |
if len(sys.argv) < 5 or len(sys.argv) > 6: | |
show_usage() | |
sys.exit(-1) | |
elif len(sys.argv) == 6: | |
limit = int(sys.argv[5]) | |
crash_process = open_crash_process(shlex.split(" ".join(sys.argv[1:4]))) | |
slab_name = sys.argv[4] | |
show_slab_info(crash_process, slab_name) | |
content = get_content(crash_process, slab_name) | |
counter = Counter() | |
for line in content: | |
for v in line.split()[1:]: | |
counter[v] += 1 | |
show(crash_process, counter.most_common(limit)) | |
crash_process.terminate() | |
if __name__ == '__main__': | |
main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment