ocfs2_fallocate_bug_plain_write.py

#!/usr/bin/python3

# Simulate a specific write pattern that triggers data corruption on OCFS2 volume.
#
# Usage: /path/to/ocfs2_fallocate_bug_plain_write.py [MAX_TIME_SECONDS]
#
# How it works:
# 1. Opens the file for appending
# 2. Write N random bytes, where N is a random integer in range [1..256]
# 3. With 50% probability, wait M centiseconds (=0.01 of a second), where M is in range [1..100]
# 4. If the time elapsed since step 1 is less than MAX_TIME_SECONDS, repeat from step 2
#
# Default MAX_TIME_SECONDS is 20
#
# How to use:
# * Mount the same shared OCFS2 volume on two hosts in the cluster
# * On first host, change current directory to the directory where the OCFS2 volume is mounted
# * Run this script, notice the MD5 hash in its output
# * Calculate MD5 hash of "output.bin" file. It should match the one from the script output
# * Calculate MD5 hash of the same file from the other host
# If data on disk is corrupted, the hashes calculated on two hosts will differ.
# The reason is that the cache on the first host contains correct data, so the hash is correct,
# while the second host has no cached data for this file,
# so it reads the corrupted data directly from the block device.

import os, hashlib, random, time, sys

FILE = 'output.bin'     # file name to write
MAX_CHUNKS = 100000
TIME_LIMIT_SECONDS = 20

FALLOC_FL_KEEP_SIZE = 1
FALLOC_FL_PUNCH_HOLE = 2

if os.path.exists(FILE):
    os.unlink(FILE)

hash_sum = hashlib.md5()

written = 0
time_start = time.time()
time_end = time_start

if len(sys.argv) > 1:
    TIME_LIMIT_SECONDS = float(sys.argv[1])

last_progress = ''
with open(FILE, 'ab') as fd:
    fileno = fd.fileno()

    for chunk_idx in range(MAX_CHUNKS):
        chunk_size = random.randint(1, 256)
        with open('/dev/urandom', 'rb') as fd:
            chunk = fd.read(chunk_size)
            hash_sum.update(chunk)

        pos = os.lseek(fileno, 0, 1)
        os.lseek(fileno, pos, 0)
        written += os.write(fileno, chunk)

        if last_progress:
            print('\015' * len(last_progress),  end='', file=sys.stderr)

        last_progress = '{:8d}/{}'.format(chunk_idx + 1, MAX_CHUNKS)
        print(last_progress,  end='', file=sys.stderr)

        time_end = time.time()
        if time_end - time_start >= TIME_LIMIT_SECONDS:
            break

        delay = random.randint(1, 200)
        if delay <= 100:
            time.sleep(delay / 100)

    if last_progress:
        print('\015' * len(last_progress),  end='', file=sys.stderr)

print("Wrote {} bytes, {:.3f} seconds elapsed, MD5 = {}".format(written, time_end - time_start, hash_sum.hexdigest()))