• Home
  • History
  • Annotate
  • only in /system/extras/tests/net_test/
NameDateSize

..20-Dec-20164 KiB

all_tests.sh20-Dec-20161.2 KiB

anycast_test.py20-Dec-20163 KiB

csocket.py20-Dec-20165.5 KiB

cstruct.py20-Dec-20166.1 KiB

cstruct_test.py20-Dec-20161.8 KiB

forwarding_test.py20-Dec-20163.7 KiB

iproute.py20-Dec-201617.8 KiB

multinetwork_base.py20-Dec-201621.6 KiB

multinetwork_test.py20-Dec-201634.3 KiB

neighbour_test.py20-Dec-201610.3 KiB

net_test.py20-Dec-201610.9 KiB

net_test.sh20-Dec-2016524

netlink.py20-Dec-20167.8 KiB

packets.py20-Dec-20166.9 KiB

ping6_test.py20-Dec-201625.7 KiB

ping6_test.sh20-Dec-2016266

README20-Dec-20162.6 KiB

run_net_test.sh20-Dec-20164.2 KiB

sock_diag.py20-Dec-201611 KiB

sock_diag_test.py20-Dec-201621.1 KiB

srcaddr_selection_test.py20-Dec-201613.9 KiB

tcp_nuke_addr_test.py20-Dec-20167.6 KiB

tcp_test.py20-Dec-20164 KiB

README

1                                net_test v0.1
2                                =============
3
4A simple framework for blackbox testing of kernel networking code.
5
6
7Why use it?
8===========
9
10- Fast test / boot cycle.
11- Access to host filesystem and networking via L2 bridging.
12- Full Linux userland including Python, etc.
13- Kernel bugs don't crash the system.
14
15
16How to use it
17=============
18
19cd <kerneldir>
20path/to/net_test/run_net_test.sh <test>
21
22where <test> is the name of a test binary in the net_test directory. This can
23be an x86 binary, a shell script, a Python script. etc.
24
25
26How it works
27============
28
29net_test compiles the kernel to a user-mode linux binary, which runs as a
30process on the host machine. It runs the binary to start a Linux "virtual
31machine" whose root filesystem is the supplied Debian disk image. The machine
32boots, mounts the root filesystem read-only, runs the specified test from init, and then drops to a shell.
33
34
35Access to host filesystem
36=========================
37
38The VM mounts the host filesystem at /host, so the test can be modified and
39re-run without rebooting the VM.
40
41
42Access to host networking
43=========================
44
45Access to host networking is provided by tap interfaces. On the host, the
46interfaces are named <user>TAP0, <user>TAP1, etc., where <user> is the first
4710 characters of the username running net_test. (10 characters because
48IFNAMSIZ = 16). On the guest, they are named eth0, eth1, etc.
49
50net_test does not do any networking setup beyond creating the tap interfaces.
51IP connectivity can be provided on the host side by setting up a DHCP server
52and NAT, sending IPv6 router advertisements, etc. By default, the VM has IPv6
53privacy addresses disabled, so its IPv6 addresses can be predicted using a tool
54such as ipv6calc.
55
56The provided filesystem contains a DHCPv4 client and simple networking
57utilities such as ping[6], traceroute[6], and wget.
58
59The number of tap interfaces is currently hardcoded to two. To change this
60number, modify run_net_test.sh.
61
62
63Logging into the VM, installing packages, etc.
64==============================================
65
66net_test mounts the root filesystem read-only, and runs the test from init, but
67since the filesystem contains a full Linux userland, it's possible to boot into
68userland and modify the filesystem, for example to install packages using
69apt-get install. Log in as root with no password. By default, the filesystem is
70configured to perform DHCPv4 on eth0 and listen to RAs.
71
72
73Bugs
74====
75
76Since the test mounts the filesystem read-only, tests cannot modify
77/etc/resolv.conf and the system resolver is hardcoded to 8.8.8.8.
78