Function stat_update_check

src/borg/archive.py:233–256

this checks for some race conditions between the first filename-based stat() we did before dispatching to the (hopefully correct) file type backup handler and the (hopefully) fd-based fstat() we did in the handler. if there is a problematic difference (e.g. file type changed), we r

(st_old, st_curr)

def stat_update_check(st_old, st_curr):
    """
    this checks for some race conditions between the first filename-based stat()
    we did before dispatching to the (hopefully correct) file type backup handler
    and the (hopefully) fd-based fstat() we did in the handler.

    if there is a problematic difference (e.g. file type changed), we rather
    skip the file than being tricked into a security problem.

    such races should only happen if:
    - we are backing up a live filesystem (no snapshot, not inactive)
    - if files change due to normal fs activity at an unfortunate time
    - if somebody is doing an attack against us
    """
    # assuming that a file type change implicates a different inode change AND that inode numbers
    # are not duplicate in a short timeframe, this check is redundant and solved by the ino check:
    if stat.S_IFMT(st_old.st_mode) != stat.S_IFMT(st_curr.st_mode):
        # in this case, we dispatched to wrong handler - abort
        raise BackupRaceConditionError("file type changed (race condition), skipping file")
    if st_old.st_ino != st_curr.st_ino:
        # in this case, the hard-links-related code in create_helper has the wrong inode - abort!
        raise BackupRaceConditionError("file inode changed (race condition), skipping file")
    # looks ok, we are still dealing with the same thing - return current stat:
    return st_curr


@contextmanager