/usr/lib/bup/cmd/bup-on--server is in bup 0.25-1.
This file is owned by root:root, with mode 0o755.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 | #! /usr/bin/python
import sys, os, struct
from bup import options, helpers
optspec = """
bup on--server
--
This command is run automatically by 'bup on'
"""
o = options.Options(optspec)
(opt, flags, extra) = o.parse(sys.argv[1:])
if extra:
o.fatal('no arguments expected')
# get the subcommand's argv.
# Normally we could just pass this on the command line, but since we'll often
# be getting called on the other end of an ssh pipe, which tends to mangle
# argv (by sending it via the shell), this way is much safer.
buf = sys.stdin.read(4)
sz = struct.unpack('!I', buf)[0]
assert(sz > 0)
assert(sz < 1000000)
buf = sys.stdin.read(sz)
assert(len(buf) == sz)
argv = buf.split('\0')
# stdin/stdout are supposedly connected to 'bup server' that the caller
# started for us (often on the other end of an ssh tunnel), so we don't want
# to misuse them. Move them out of the way, then replace stdout with
# a pointer to stderr in case our subcommand wants to do something with it.
#
# It might be nice to do the same with stdin, but my experiments showed that
# ssh seems to make its child's stderr a readable-but-never-reads-anything
# socket. They really should have used shutdown(SHUT_WR) on the other end
# of it, but probably didn't. Anyway, it's too messy, so let's just make sure
# anyone reading from stdin is disappointed.
#
# (You can't just leave stdin/stdout "not open" by closing the file
# descriptors. Then the next file that opens is automatically assigned 0 or 1,
# and people *trying* to read/write stdin/stdout get screwed.)
os.dup2(0, 3)
os.dup2(1, 4)
os.dup2(2, 1)
fd = os.open('/dev/null', os.O_RDONLY)
os.dup2(fd, 0)
os.close(fd)
os.environ['BUP_SERVER_REVERSE'] = helpers.hostname()
os.execvp(argv[0], argv)
sys.exit(99)
|