Sasha Goldshtein | 4f1ea67 | 2016-02-07 01:57:42 -0800 | [diff] [blame^] | 1 | #!/usr/bin/env python |
| 2 | |
| 3 | from bcc import BPF |
| 4 | from time import sleep |
| 5 | import argparse |
| 6 | import subprocess |
| 7 | |
| 8 | examples = """ |
| 9 | EXAMPLES: |
| 10 | |
| 11 | memleak.py -p $(pidof allocs) |
| 12 | Trace allocations and display a summary of "leaked" (outstanding) |
| 13 | allocations every 5 seconds |
| 14 | memleak.py -p $(pidof allocs) -t |
| 15 | Trace allocations and display each individual call to malloc/free |
| 16 | memleak.py -p $(pidof allocs) -a -i 10 |
| 17 | Trace allocations and display allocated addresses, sizes, and stacks |
| 18 | every 10 seconds for outstanding allocations |
| 19 | memleak.py |
| 20 | Trace allocations in kernel mode and display a summary of outstanding |
| 21 | allocations every 5 seconds |
| 22 | """ |
| 23 | |
| 24 | description = """ |
| 25 | Trace outstanding memory allocations that weren't freed. |
| 26 | Supports both user-mode allocations made with malloc/free and kernel-mode |
| 27 | allocations made with kmalloc/kfree. |
| 28 | """ |
| 29 | |
| 30 | parser = argparse.ArgumentParser(description=description, formatter_class=argparse.RawDescriptionHelpFormatter, epilog=examples) |
| 31 | parser.add_argument("-p", "--pid", help="the PID to trace; if not specified, trace kernel allocs") |
| 32 | parser.add_argument("-t", "--trace", action="store_true", help="print trace messages for each alloc/free call") |
| 33 | parser.add_argument("-i", "--interval", default=5, help="interval in seconds to print outstanding allocations") |
| 34 | parser.add_argument("-a", "--show-allocs", default=False, action="store_true", help="show allocation addresses and sizes as well as call stacks") |
| 35 | |
| 36 | args = parser.parse_args() |
| 37 | |
| 38 | pid = -1 if args.pid is None else int(args.pid) |
| 39 | kernel_trace = (pid == -1) |
| 40 | trace_all = args.trace |
| 41 | interval = int(args.interval) |
| 42 | |
| 43 | bpf_source = open("memleak.c").read() |
| 44 | bpf_source = bpf_source.replace("SHOULD_PRINT", "1" if trace_all else "0") |
| 45 | |
| 46 | bpf_program = BPF(text=bpf_source) |
| 47 | |
| 48 | if not kernel_trace: |
| 49 | print("Attaching to malloc and free in pid %d, Ctrl+C to quit." % pid) |
| 50 | bpf_program.attach_uprobe(name="c", sym="malloc", fn_name="alloc_enter", pid=pid) |
| 51 | bpf_program.attach_uretprobe(name="c", sym="malloc", fn_name="alloc_exit", pid=pid) |
| 52 | bpf_program.attach_uprobe(name="c", sym="free", fn_name="free_enter", pid=pid) |
| 53 | else: |
| 54 | print("Attaching to kmalloc and kfree, Ctrl+C to quit.") |
| 55 | bpf_program.attach_kprobe(event="__kmalloc", fn_name="alloc_enter") |
| 56 | bpf_program.attach_kretprobe(event="__kmalloc", fn_name="alloc_exit") |
| 57 | bpf_program.attach_kprobe(event="kfree", fn_name="free_enter") |
| 58 | |
| 59 | def get_code_ranges(pid): |
| 60 | ranges = {} |
| 61 | raw_ranges = open("/proc/%d/maps" % pid).readlines() |
| 62 | for raw_range in raw_ranges: |
| 63 | parts = raw_range.split() |
| 64 | if len(parts) < 6 or parts[5][0] == '[' or not 'x' in parts[1]: |
| 65 | continue |
| 66 | binary = parts[5] |
| 67 | range_parts = parts[0].split('-') |
| 68 | addr_range = (int(range_parts[0], 16), int(range_parts[1], 16)) |
| 69 | ranges[binary] = addr_range |
| 70 | return ranges |
| 71 | |
| 72 | def run_command(command): |
| 73 | p = subprocess.Popen(command.split(), stdout=subprocess.PIPE, stderr=subprocess.STDOUT) |
| 74 | return iter(p.stdout.readline, b'') |
| 75 | |
| 76 | ranges_cache = {} |
| 77 | |
| 78 | def get_sym_ranges(binary): |
| 79 | if binary in ranges_cache: |
| 80 | return ranges_cache[binary] |
| 81 | sym_ranges = {} |
| 82 | raw_symbols = run_command("objdump -t %s" % binary) |
| 83 | for raw_symbol in raw_symbols: |
| 84 | parts = raw_symbol.split() |
| 85 | if len(parts) < 6 or parts[3] != ".text" or parts[2] != "F": |
| 86 | continue |
| 87 | sym_start = int(parts[0], 16) |
| 88 | sym_len = int(parts[4], 16) |
| 89 | sym_name = parts[5] |
| 90 | sym_ranges[sym_name] = (sym_start, sym_len) |
| 91 | ranges_cache[binary] = sym_ranges |
| 92 | return sym_ranges |
| 93 | |
| 94 | def decode_sym(binary, offset): |
| 95 | sym_ranges = get_sym_ranges(binary) |
| 96 | for name, (start, length) in sym_ranges.items(): |
| 97 | if offset >= start and offset <= (start + length): |
| 98 | return "%s+0x%x" % (name, offset - start) |
| 99 | return "%x" % offset |
| 100 | |
| 101 | def decode_addr(code_ranges, addr): |
| 102 | for binary, (start, end) in code_ranges.items(): |
| 103 | if addr >= start and addr <= end: |
| 104 | offset = addr - start if binary.endswith(".so") else addr |
| 105 | return "%s %s" % (binary, decode_sym(binary, offset)) |
| 106 | return "%x" % addr |
| 107 | |
| 108 | def decode_stack(info): |
| 109 | stack = "" |
| 110 | if info.num_frames <= 0: |
| 111 | return "???" |
| 112 | for i in range(0, info.num_frames): |
| 113 | addr = info.callstack[i] |
| 114 | if kernel_trace: |
| 115 | stack += " %s (%x) ;" % (bpf_program.ksym(addr), addr) |
| 116 | else: |
| 117 | stack += " %s (%x) ;" % (decode_addr(code_ranges, addr), addr) |
| 118 | return stack |
| 119 | |
| 120 | def print_outstanding(): |
| 121 | stacks = {} |
| 122 | print("*** Outstanding allocations:") |
| 123 | allocs = bpf_program.get_table("allocs") |
| 124 | for address, info in sorted(allocs.items(), key=lambda a: -a[1].size): |
| 125 | stack = decode_stack(info) |
| 126 | if stack in stacks: stacks[stack] += info.size |
| 127 | else: stacks[stack] = info.size |
| 128 | if args.show_allocs: |
| 129 | print("\taddr = %x size = %s" % (address.value, info.size)) |
| 130 | for stack, size in sorted(stacks.items(), key=lambda s: -s[1]): |
| 131 | print("\t%d bytes allocated from stack\n\t\t%s" % (size, stack.replace(";", "\n\t\t"))) |
| 132 | |
| 133 | while True: |
| 134 | if trace_all: |
| 135 | print bpf_program.trace_fields() |
| 136 | else: |
| 137 | try: |
| 138 | sleep(interval) |
| 139 | except KeyboardInterrupt: |
| 140 | exit() |
| 141 | if not kernel_trace: |
| 142 | code_ranges = get_code_ranges(pid) |
| 143 | print_outstanding() |
| 144 | |