| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Adding kernel command line option 'panic=1' causes a panic
to reboot the system after 1 second.
However the pre-existing '-no-reboot' flag causes a reboot
signal to terminate QEMU instead of rebooting it.
(and it doesn't hurt UML either)
This fixes a problem where on panic you have to 'CTRL+A exit' from qemu.
It does not yet get rid of the occasional panic instead of clean shutdown
problem that I have not been able to track down the cause of.
Tested: via 'echo c > /proc/sysrq-trigger' on both UML and QEMU.
Bug: 140201217
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I0e6646fb843300d2f714aac8b65159dd2001ef16
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We only actually care about the value not becoming negative,
older kernels (with fix) treated negative as minimum,
newer ones treat it as maximum (but positive).
As such a >= min works on both.
Tested: by running leak_test on 5.1.21
Bug: 140201217
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I4d8c76e09b072ce3beae7e0076ebbd9d3f69e96b
|
|
|
|
|
|
|
|
|
|
| |
(found while looking at apparently unrelated error from
https://android-build.googleplex.com/builds/submitted/5834368/kernel_test/latest/view/logs/build_error.log
which is a failure on android-mainline kernel, but one line earlier)
Test: manually
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I0ac5df02d298250f25d5aee74ccbbc283e2556f4
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Terminal auto wrap is sometimes disabled by qemu and makes using long
commandlines a pain in interactive shells like bash.
Per:
https://serverfault.com/questions/322445/how-do-i-restore-xtermgnome-terminal-wrapping-after-telnet-to-hp-equipment-ha/619127#619127
The VT100 code to turn autowrap back on is "<ESC>[?7h",
which you could send to the terminal with this command:
printf "\033[?7h"
However, this assumes that your terminal is VT100-compliant,
which, while a good bet, is not a certainty.
The more correct way to do it is to rely on your terminfo settings
and run this command:
tput smam
which generates the exact same output as the above printf command,
assuming a VT100-style terminal.
The terminfo capability "smam" is "Set Mode: Automatic Margins".
Its partner is "rmam": "Reset Mode: Automatic Margins".
Tested: fixes auto-wrap post qemu exit
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ib134788236e4bd52d9d6ec0c12c64b09e9f2f2ba
|
|
|
|
|
|
|
| |
It's useful for various testing purposes.
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I566732956531991790f600c56e4dc6245ed8ed63
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Apparently kernel's default stdin/out/err file descriptors come from
some fully internal fake filesystem (device '1', presumably rootfs
default internal ram disk) and not from the 'active' /dev.
This confuses the newer tty cli shipped in buster and it thinks it is
'not a tty'.
One could argue this is a bug in 'tty', since stdin/stdout do
respond to tty ioctls and are thus obviously ttys.
It should probably either respond with 'unknown tty' or just honour
readlink /proc/self/fd/{0,1,2} results...
Regardless, easy enough to work around.
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ib2c8f93d0cb338fbe3764ab67947ea6746203a48
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Due to mtk design of ipsec monitor, kernel will create
a new SA when running the testUpdateActiveSaMarks case,
so there are two SAs, but the expected number of SAs is one.
We need to avoid creating the extra SA.
Bug: 137509253
Test: run vts -m VtsKernelNetTest -t VtsKernelNetTest#testKernelNetworking
Change-Id: Ic5a82ac9f8531b75e61ce9188aa35ef9a0ca0619
|
|/
|
|
|
|
|
|
|
| |
The variable J is used as number of jobs to parallelize the kernel
build. In uninitialized, this patch sets J as the number of processing
units available in the current system, instead of a fixed number of 64.
Change-Id: I825f50f0b904474b699f883066249537162a15d0
Signed-off-by: Alessio Balsini <balsini@google.com>
|
|
|
|
|
|
|
| |
Bug: 137262547
Test: built all the images
Change-Id: I9f9dac5e42b71a08c99a0777297e6067fe88ab35
Signed-off-by: Maciej Żenczykowski <maze@google.com>
|
|
|
|
|
|
|
|
|
|
| |
Add support for building a Debian Buster (10.x) root filesystem. This is
mostly the same as stretch, but systemd-timesyncd needs an additional
workaround, and we can remove some of the stretch workarounds.
Bug: 137227555
Change-Id: Ieaa541d6a475d66c752917e0a6b40961a6dfeb9a
Signed-off-by: Alistair Delva <adelva@google.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When testing in QEMU mode with DEFCONFIG unset, the kernel will use the
x86_64_defconfig automatically. This config does not specify console=,
and the kernel will not set this automatically for serial ports, which
results in no log messages or any way to interact with the instance.
Fix this by explicitly specifying console=ttyS0 or console=ttyAMA0
(which are the implied default values when using QEMU's -serial stdio
feature) and update the comments.
This will also help with an upcoming change to the
x86_64_cuttlefish_defconfig in the Android Common Kernel which will
remove the CONFIG_CMDLINE hard-coding of 'console='.
Bug: 137224399
Change-Id: I2fc0b2fd542969ec65ac933be4846f0edcd18a90
Signed-off-by: Alistair Delva <adelva@google.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When using run_net_tests.sh against the x86_64_defconfig upstream
(4.19+) some tests will fail or some root filesystem functionality will
be broken. Ensure that the following options are always =y:
INET_DIAG - needed by INET_UDP_DIAG, needed by sockdiag tests
IKCONFIG, IKCONFIG_PROC - needed by recently changed tests
CRYPTO_DEV_VIRTIO - needed in qemu mode by virtio-rng
Bug: 137227864
Change-Id: Ida422c2475f232e31b27d4d73e5d8362771e64ab
Signed-off-by: Alistair Delva <adelva@google.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There is different between bionic (on-device tests) and glibc
(VM tests) around interpreting IPv4-mapped addresses. Bionic
don't convert mapped addresses into IPv4 addresses.
Bug: 130199266
Test: run_net_test.sh tcp_repair_test.py
Test: m kernel_net_tests && adb push &&
adb shell /data/nativetest64/kernel_net_tests/kernel_net_tests
Change-Id: I7b1f32b5c8a64974db6154b126648a7bb81d2c95
|
|\
| |
| |
| |
| |
| |
| | |
* changes:
net-test: leak_test fix a test name typo
net-test: add CONFIG_WARN_ALL_UNSEEDED_RANDOM
net-test: add CONFIG_PCI
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
without this we have 2 different definitions of testRcvBufForce()
Test: ran test suite on x86_64 qemu
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ibec5b2638f02aa33329044c6b2f8ce97c0dec5cd
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
It doesn't hurt and is helpful to see what's grabbing
random bytes too early in the boot process
Test: ran test suite on x86_64 qemu
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Iabe80f7e1254f5b4651892faba343857aea80f9a
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
It defaults to off on Linux 5.0 but is needed for virtio on qemu.
Test: ran test suite on x86_64 qemu
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I820fa224bdb22cf6c273208786a640816bd17ac2
|
|/
|
|
|
|
|
|
|
|
|
| |
This patch add tests to ensure tcp connection works correctly
in repair mode.
Bug: 114151147
Bug: 129463499
Test: run_net_test.sh tcp_repair_test.py
Change-Id: I91df2e4156002ea62073d7543071c02f11a1de60
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This resolves:
Running /host/all_tests.py
Traceback (most recent call last):
File "/host/all_tests.py", line 51, in <module>
import_module(name)
File "/usr/lib/python2.7/importlib/__init__.py", line 37, in import_module
__import__(name)
File "/host/bpf_test.py", line 26, in <module>
from bpf import * # pylint: disable=wildcard-import
File "/host/bpf.py", line 41, in <module>
}[os.uname()[4] + "-" + platform.architecture()[0]]
KeyError: 'x86_64-32bit'
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I64f246a4fd34c8a84e9c349957060d7d1be18175
|
|
|
|
|
|
|
|
|
|
|
|
| |
Google shell style guide: https://google.github.io/styleguide/shell.xml
says to always use "${x}" instead of $x, and in general to double quote
everything (fixes issues with whitespace in variable expansions).
Similarly [[ is bash builtin, unlike [ or test.
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I507ceb630e7dd7347783e499826dcce2c2c50ee5
|
|
|
|
|
|
|
|
|
|
| |
This is a fix for:
/sbin/net_test.sh: line 33: ./net_test.sh: No such file or directory
not resulting in a test error (android common 3.18/4.4 on stretch rootfs)
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ib6c9698e98cec39eadcb371b5241241541e276d9
|
|
|
|
|
|
| |
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I64cf43e0522515a915d8e6b95fc73e50518683a4
|
|
|
|
|
|
| |
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I63b719e252dde32be3996a6c9747d116650c6b7e
|
|
|
|
|
|
| |
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I8ebb926ab6ce8f09baeec65861a23c8c4a59acff
|
|
|
|
|
|
|
|
|
| |
Tested via:
ARCH=um SUBARCH=x86_64 WERROR=0 /aosp-tests/net/test/run_net_test.sh --builder all_tests.py
Bug: 117102691
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I694e361bc478cdb53da19b7669e3aa2a42793cd9
|
|
|
|
|
|
| |
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I5a748479cb48e7bef33717d82e79348be97e5cd2
|
|
|
|
|
|
| |
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I8b980a5830e85cd5c4fd8d123c9911cfdf4daea7
|
|\ |
|
| |
| |
| |
| |
| |
| | |
Bug: 126465735
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I72cffe73ea69226f316b735a629bb45e69895e0f
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This reverts commit dd2d31985f4ebabe78c9bb212fadd3b489c849fa.
Thus reapplying commit 4406bae0a2cb27d99b6fe5f9f751399d05e0bb49.
Original commit log:
Test Updating OUTPUT_MARK on Active SAs
Test that the xfrm output mark is update-able
on an Active SA, which means that we can dynamically
reroute traffic to new underlying networks after an
SA has been created.
Bug: 71645364
Change-Id: I78c67001c63e2c82bf1f6e87af4efce855d34269
Test: run_net_test.sh xfrm_test.py
|
|
|
|
|
|
| |
Bug: 123953923
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ib05019a1db10fbaae4bf4aa5000d8c617b9837d5
|
|
|
|
|
|
|
|
|
|
| |
This change adds tests to ensure backwards compatibility of floating
policies with no output mark set, ensuring that they are still routed
correctly (over the default network) in mark-based routing schemes.
Bug: 122236988
Test: Ran on common-4.9, passing
Change-Id: I49d291ad5b6a51bc4556643c96e37885c49dce5d
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
CloseFileDescriptorThread finished
Because there are some platforms that take more time than 0.5 seconds,
such as platforms that uses MPTCP implementation from https://www.multipath-tcp.org/,
we change the waiting time to 3 seconds.
For kernels that MPTCP is ported, it take more than 0.5 sec to close tun interface,
because of DAD procedure within MPTCP fullmesh module.
It seems to be the duplicate address-timer takes a refcount on the IPv6-address,
preventing it from getting closed.
Test: run vts-kernel -m VtsKernelNetTest with 4.9 and 4.14 kernel using MPTCP implementation
Without this change, VtsKernelNetTest was failed with MPTCP enabled kernel every time.
After this change all will pass in 5 iterations.
Bug: 120248546
Change-Id: I69093d5aa55d2796eac1f6db10d8ba649b63ae9a
Signed-off-by: Yunsik Lee <yunsik.lee@lge.com>
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The backport for 4.19 kernel sets the initial receive window size to a
new value greater then 64k. Use a connected TCP socket to check the
patch is properly applied.
Test: ./sock_diag_test.py
Bug: 118591209
Change-Id: If21c6e951aede70c22699d1794dbc09ce91ba5e4
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This makes Ctrl+C and Ctrl+Z work.
(while we're at it add some more debugging print out to ease
debugging and better explain how the transition works)
Change-Id: I5d6d82a53161f7ff1bc6f1b5329ba864d0759a48
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
By default it initializes 60 seconds after boot (if there's enough
entropy at that point in time), which is too late (we don't want
to simply wait 60 seconds, but we do want it initialized prior
to running any tests).
We also switch from hex (base 16) to base64 encoding of the entropy
string just to make the kernel command line shorter.
Change-Id: Ia4924576c550b2858a938873baef18893550fa04
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This reverts commit 805d82cce067183af3dec98b4b04f0824cd1562f.
Problem has been fixed in upstream Linux kernels, 4.4/4.9/4.14 and the
code has been merged via LTS into kernel/common.
Bug: 111396209
Change-Id: I397a3ca34ced1b8528fab6f795144f6d94fb8d7f
Signed-off-by: Alistair Strachan <astrachan@google.com>
|
|/
|
|
|
| |
Bug: 118651133
Change-Id: I00e0ab757d0776c425802b60f5d60d5195a3f2dc
|
|
|
|
|
|
|
|
|
|
|
|
| |
Using non-blocking sockets appears to be causing flakiness due to the
tests racing against the kernel's crypto/networking infrastructure.
This commit allows these sockets to block for up to 500ms to allow time
for the crypto and routing to complete before throwing errors.
Bug: 118753529
Test: Run 20x locally, via QEMU; passing.
Change-Id: I92a39f8890244ebe918f48bcf0971e85795b2a1e
|
|
|
|
|
|
|
|
|
|
| |
The testCgroupSocketCreateBlock test left some socket open when test
exist and might cause other net test flaky. This patch fix that before
the test exits.
Bug: 118753529
Test: ./all_test.sh
Change-Id: I88be0b49b2dfa001bfc4fea699f9bf714e4ec8e3
|
|
|
|
|
|
|
|
|
|
| |
Add a base test for cgroup socket filter to check the cgroup socket eBPF
program can actually block socket creation for INET socket. This feature
will be used in 4.14 and above kernel to replace paranoid network.
Test: ./bpf_test.py
Bug: 111560739
Change-Id: I10f5c0c6847ec033cf757b8ce9dfa1a6b80c50fb
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
See:
https://android-build.googleplex.com/builds/submitted/5097760/kernel_test/latest/view/logs/build.log
ie.:
##### ./forwarding_test.py (5/23)
E.
======================================================================
ERROR: testForwardingCrashTcp (__main__.ForwardingTest)
----------------------------------------------------------------------
Traceback (most recent call last):
File "./forwarding_test.py", line 169, in testForwardingCrashTcp
self.CheckForwardingByProto(IPPROTO_TCP)
File "./forwarding_test.py", line 161, in CheckForwardingByProto
self.CheckForwardingHandlerByProto(proto, netid, iface1, iface2)
File "./forwarding_test.py", line 150, in CheckForwardingHandlerByProto
self.CheckForwardingCrashTcp(netid, iif, oif)
File "./forwarding_test.py", line 124, in CheckForwardingCrashTcp
sockets = self.ReadProcNetSocket("tcp6")
File "/host/usr/local/google/buildbot/src/partner-android/common-android-4.19/kernel/tests/net/test/net_test.py", line 462, in ReadProcNetSocket
_, _, uid, _, _, refcnt, _, extra) = regexp.match(line).groups()
AttributeError: 'NoneType' object has no attribute 'groups'
----------------------------------------------------------------------
Ran 2 tests in 0.452s
FAILED (errors=1)
E.
...(and then it reruns and promptly fails again in the same way)...
'./forwarding_test.py' failed more than once, giving up
#####
I've had no luck reproducing the failure locally...
But I'm guessing one of the 8 hex digit regexps is not matching due to printing
out a negative 64 bit integer (which requires 16 hex digits).
I'm intentionally leaving this in a way it can still fail, so we get to the
bottom of this and actually fix the regexps.
(while we're at it remove a spurious second | operator)
Bug: 118651133
Change-Id: I6fbdc563cedbba3648d74b6bea1911b94086d0f3
|
|
|
|
| |
Change-Id: I7758590d0da24729d2d0eb205e97405b75fd1d30
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Per:
https://android-build.googleplex.com/builds/submitted/5097760/kernel_test/latest/view/logs/build.log
we appear to still not have quite enough:
...
random: fast init done
adding entropy from hex string [fdf7357dcff34af28ec2ef7907fda944a9adbbd62e6d4cc4a4f36b0771bc1d75]
...
##### Found 23 tests.
##### ./csocket_test.py (1/23)
random: csocket_test.py: uninitialized urandom read (16 bytes read)
...
##### ./bpf_test.py (10/23)
...
random: bpf_test.py: uninitialized urandom read (16 bytes read)
random: crng init done
(although I cannot reproduce it on my desktop)
Change-Id: I90a4d9f7a3433049a3a22f60ddb32f2875e250e4
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Make the accommodations necessary for arm64 support:
- The compressed kernel image is called something different;
- Switched from isa-serial to pci-serial, as the ARM virt model has no
support for the PC ISA bus, and it doesn't matter for x86_64;
- Enabled hardware accelerated virtualization only on x86_64.
To test, make sure you have the Android clang and gcc toolchains pathed
for AArch64. Then, run something like the following:
$ CROSS_COMPILE=aarch64-linux-androidkernel- ARCH=arm64 \
DEFCONFIG=cuttlefish_defconfig ROOTFS=net_test.rootfs.arm64.20181025 \
/path/to/run_net_test.sh all_tests.sh
All of the changes are keyed off of ARCH=arm64, to minimize the impact
this change has on ARCH=um or ARCH=x86_64.
Bug: 118446450
Change-Id: Id52c4eb061d6da4d3ec3485cd9ca62ff8ef028c2
Signed-off-by: Alistair Strachan <astrachan@google.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Support building 'foreign' architecture images for wheezy and stretch.
Because the other control files were already architecture agnostic, this
change is very simple.
Bug: 118446450
Test: ./build_rootfs.sh -a arm64
Change-Id: Ib62d1ed319224dd3d4861172fed998bb42ab7280
Signed-off-by: Alistair Strachan <astrachan@google.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In kernel_net_test/xfrm_tunnel_test.XfrmVtiTest
The testcases write packets to the tunnel, then check the count of rx/tx
packets number on the interface.
However, the packets are processed *asynchronously* in kernel.
So, receive the packets before check the rx count
Test: run vts/kernel_net_tests
Bug: 117532237
Signed-off-by: Xu YiPing <xuyiping@hisilicon.com>
|
|
|
|
|
|
| |
This resolves IPv6 ping problems on recent (for example 4.17) kernels.
Change-Id: I5310bad2f3148d1195367aade10b9761995134a4
|