Opening a verity device is an expensive operation. The kernelspace operations
are mostly sequential with a global lock held regardless of which device
is being opened. In userspace jumps in and out of multiple libraries are
required. When signatures are used, there's the additional cryptographic
checks.
We know when two devices are identical: they have the same root hash.
If libcrypsetup returns EEXIST, double check that the hashes are really
the same, and that either both or none have a signature, and if everything
matches simply remount the already open device. The kernel will do
reference counting for us.
In order to quickly and reliably discover if a device is already open,
change the node naming scheme from '/dev/mapper/major:minor-verity' to
'/dev/mapper/$roothash-verity'.
Unfortunately libdevmapper is not 100% reliable, so in some case it
will say that the device already exists and it is active, but in
reality it is not usable. Fallback to an individually-activated
unique device name in those cases for robustness.
The kernel interface requires setting up read-only bind-mounts in
two steps, the bind first and then a read-only remount.
Fix nspawn-mount, and cover this case in the integration test.
Fixes#16484
When the RTC time at boot is off in the future by a few days, OnCalendar=
timers will be scheduled based on the time at boot. But if the time has been
adjusted since boot, the timers will end up scheduled way in the future, which
may cause them not to fire as shortly or often as expected.
Update the logic so that the time will be adjusted based on monotonic time.
We do that by calculating the adjusted manager startup realtime from the
monotonic time stored at that time, by comparing that time with the realtime
and monotonic time of the current time.
Added a test case to validate this works as expected. The test case creates a
QEMU virtual machine with the clock 3 days in the future. Then we adjust the
clock back 3 days, and test creating a timer with an OnCalendar= for every 15
minutes. We also check the manager startup timestamp from both `systemd-analyze
dump` and from D-Bus.
Test output without the corresponding code changes that fix the issue:
Timer elapse outside of the expected 20 minute window.
next_elapsed=1594686119
now=1594426921
time_delta=259198
With the code changes in, the test passes as expected.
For some reason the wait-online is failing intermittently; it's unclear
exactly why, but this hopefully avoids the failure for unrelated PR.
This is a workaround (not fix) for #16105
Several recent failed runs show that the test is still racy in two ways:
1) Sometimes it takes a while before the PID file is created, leading
to:
```
[ 10.950540] testsuite-47.sh[308]: ++ cat /leakedtestpid
[ 10.959712] testsuite-47.sh[308]: cat: /leakedtestpid: No such file or directory
[ 10.959824] testsuite-47.sh[298]: + leaked_pid=
```
2) Again, sometimes we check the leaked PID before the unit is actually
stopped, leading to a false negative:
```
[ 18.099599] testsuite-47.sh[346]: ++ cat /leakedtestpid
[ 18.116462] testsuite-47.sh[333]: + leaked_pid=342
[ 18.117101] testsuite-47.sh[333]: + systemctl stop testsuite-47-repro
...
[ 20.033907] testsuite-47.sh[333]: + ps -p 342
[ 20.080050] testsuite-47.sh[351]: PID TTY TIME CMD
[ 20.080050] testsuite-47.sh[351]: 342 ? 00:00:00 sleep
[ 20.082040] testsuite-47.sh[333]: + exit 42
```
Add support for creating a MACVLAN interface in "source" mode by
specifying Mode=source in the [MACVLAN] section of a .netdev file.
A list of allowed MAC addresses for the corresponding MACVLAN can also
be specified with the SourceMACAddress= option of the [MACVLAN] section.
An example .netdev file:
[NetDev]
Name=macvlan0
Kind=macvlan
MACAddress=02:DE:AD:BE:EF:00
[MACVLAN]
Mode=source
SourceMACAddress=02:AB:AB:AB:AB:01 02:CD:CD:CD:CD:01
SourceMACAddress=02:EF:EF:EF:EF:01
The same keys can also be specified in [MACVTAP] for MACVTAP kinds of
interfaces, with the same semantics.
The fuzzer test case has a giant line with ";;;;;;;;;;;..." which is turned into
a strv of empty strings. Unfortunately, when pushing each string, strv_push() needs
to walk the whole array, which leads to quadratic behaviour. So let's use
greedy_allocation here and also keep location in the string to avoid iterating.
build/fuzz-xdg-desktop test/fuzz/fuzz-xdg-desktop/oss-fuzz-22812 51.10s user 0.01s system 99% cpu 51.295 total
↓
build/fuzz-xdg-desktop test/fuzz/fuzz-xdg-desktop/oss-fuzz-22812 0.07s user 0.01s system 96% cpu 0.083 total
Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=22812.
Other minor changes:
- say "was already defined" instead of "defined multiple times" to make it
clear that we're ignoring this second definition, and not all definitions
of the key
- unescaping needs to be done also for the last entry
When a command asks to load a unit directly and it is in state
UNIT_NOT_FOUND, and the cache is outdated, we refresh it and
attempto to load again.
Use the same logic when building up a transaction and a dependency in
UNIT_NOT_FOUND state is encountered.
Update the unit test to exercise this code path.
SIG-prefixed signals for `kill` are not POSIX compliant, so on Ubuntu CI
(which defaults to dash instead of bash) the TEST-52 contains following
error:
[ 9693.549638] sh[51]: + systemctl poweroff --no-block
[ 9693.553130] systemd-logind[26]: System is powering down.
[ 9693.608911] sh[54]: /bin/sh: 1: kill: Illegal option -S
This can be reproduced manually as well, either by running dash, or bash
in POSIX mode:
$ dash -c 'kill -SIGKILL 123'
dash: 1: kill: Illegal option -S
$ bash --posix -c 'kill -SIGKILL 123'
bash: line 0: kill: SIGKILL: invalid signal specification
SR-IOV provides the ability to partition a single physical PCI
resource into virtual PCI functions which can then be injected in
to a VM. In the case of network VFs, SR-IOV improves north-south n
etwork performance (that is, traffic with endpoints outside the
host machine) by allowing traffic to bypass the host machine’s network stack.
When the system is under heavy load, it can happen that the unit cache
is refreshed for an unrelated reason (in the test I simulate this by
attempting to start a non-existing unit). The new unit is found and
accounted for in the cache, but it's ignored since we are loading
something else.
When we actually look for it, by attempting to start it, the cache is
up to date so no refresh happens, and starting fails although we have
it loaded in the cache.
When the unit state is set to UNIT_NOT_FOUND, mark the timestamp in
u->fragment_loadtime. Then when attempting to load again we can check
both if the cache itself needs a refresh, OR if it was refreshed AFTER
the last failed attempt that resulted in the state being
UNIT_NOT_FOUND.
Update the test so that this issue reproduces more often.
Since the hwdb update from a79be2f807
the systemd-hwdb-update service started timing out under ASan when
compiled with gcc, as we started tripping over the 3 minutes timeout.
This affects only gcc runs, since the current gcc on Arch still suffers
from the detect_stack_use_after_return performance penalty[0]. Until
the fixed gcc is present in the respective repositories, let's bump
the timeout to 4 minutes, as we might not be able to upgrade right
away, due to systemd/systemd#16199.
Before the hwdb update:
[ 7958.292540] systemd[63]: systemd-hwdb-update.service: Executing: /usr/bin/time systemd-hwdb update
[ 7958.304005] systemd[1]: systemd-journald.service: Got notification message from PID 44 (FDSTORE=1)
[ 7958.314434] systemd[1]: systemd-journald.service: Added fd 3 (n/a) to fd store.
[ 8008.520082] systemd[1]: systemd-journald.service: Got notification message from PID 44 (WATCHDOG=1)
[ 8068.520151] systemd[1]: systemd-journald.service: Got notification message from PID 44 (WATCHDOG=1)
[ 8125.682843] time[63]: 84.47user 82.92system 2:47.50elapsed 99%CPU (0avgtext+0avgdata 811512maxresident)k
[ 8125.682843] time[63]: 0inputs+19680outputs (0major+25000853minor)pagefaults 0swaps
After the hwdb update:
[ 6215.491958] systemd[63]: systemd-hwdb-update.service: Executing: /usr/bin/time systemd-hwdb update
[ 6215.503380] systemd[1]: systemd-journald.service: Got notification message from PID 44 (FDSTORE=1)
[ 6215.514172] systemd[1]: systemd-journald.service: Added fd 3 (n/a) to fd store.
[ 6329.392918] systemd[1]: systemd-journald.service: Got notification message from PID 44 (WATCHDOG=1)
[ 6394.920205] time[63]: 89.48user 89.98system 2:59.55elapsed 99%CPU (0avgtext+0avgdata 812764maxresident)k
[ 6394.920205] time[63]: 0inputs+20568outputs (0major+27318354minor)pagefaults 0swaps
[0] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94910
https://tools.ietf.org/html/draft-knodel-terminology-02https://lwn.net/Articles/823224/
This gets rid of most but not occasions of these loaded terms:
1. scsi_id and friends are something that is supposed to be removed from
our tree (see #7594)
2. The test suite defines an API used by the ubuntu CI. We can remove
this too later, but this needs to be done in sync with the ubuntu CI.
3. In some cases the terms are part of APIs we call or where we expose
concepts the kernel names the way it names them. (In particular all
remaining uses of the word "slave" in our codebase are like this,
it's used by the POSIX PTY layer, by the network subsystem, the mount
API and the block device subsystem). Getting rid of the term in these
contexts would mean doing some major fixes of the kernel ABI first.
Regarding the replacements: when whitelist/blacklist is used as noun we
replace with with allow list/deny list, and when used as verb with
allow-list/deny-list.
This combines set_ensure_allocated() with set_consume(). The cool thing is that
because we know the hash ops, we can correctly free the item if appropriate.
Similarly to set_consume(), the goal is to simplify handling of the case where
the item needs to be freed on error and if already present in the set.
We'd try to map a zero-byte buffer from a NULL pointer, which is undefined behaviour.
src/systemd/src/libsystemd/sd-bus/bus-message.c:3161:60: runtime error: applying zero offset to null pointer
#0 0x7f6ff064e691 in find_part /work/build/../../src/systemd/src/libsystemd/sd-bus/bus-message.c:3161:60
#1 0x7f6ff0640788 in message_peek_body /work/build/../../src/systemd/src/libsystemd/sd-bus/bus-message.c:3283:16
#2 0x7f6ff064e8db in enter_struct_or_dict_entry /work/build/../../src/systemd/src/libsystemd/sd-bus/bus-message.c:3967:21
#3 0x7f6ff06444ac in bus_message_enter_struct /work/build/../../src/systemd/src/libsystemd/sd-bus/bus-message.c:4009:13
#4 0x7f6ff0641dde in sd_bus_message_enter_container /work/build/../../src/systemd/src/libsystemd/sd-bus/bus-message.c:4136:21
#5 0x7f6ff0619874 in sd_bus_message_dump /work/build/../../src/systemd/src/libsystemd/sd-bus/bus-dump.c:178:29
#6 0x4293d9 in LLVMFuzzerTestOneInput /work/build/../../src/systemd/src/fuzz/fuzz-bus-message.c:39:9
#7 0x441986 in fuzzer::Fuzzer::ExecuteCallback(unsigned char const*, unsigned long) /src/libfuzzer/FuzzerLoop.cpp:558:15
#8 0x44121e in fuzzer::Fuzzer::RunOne(unsigned char const*, unsigned long, bool, fuzzer::InputInfo*, bool*) /src/libfuzzer/FuzzerLoop.cpp:470:3
#9 0x443164 in fuzzer::Fuzzer::ReadAndExecuteSeedCorpora(std::__1::vector<fuzzer::SizedFile, fuzzer::fuzzer_allocator<fuzzer::SizedFile> >&) /src/libfuzzer/FuzzerLoop.cpp:770:7
#10 0x4434bc in fuzzer::Fuzzer::Loop(std::__1::vector<fuzzer::SizedFile, fuzzer::fuzzer_allocator<fuzzer::SizedFile> >&) /src/libfuzzer/FuzzerLoop.cpp:799:3
#11 0x42d2bc in fuzzer::FuzzerDriver(int*, char***, int (*)(unsigned char const*, unsigned long)) /src/libfuzzer/FuzzerDriver.cpp:846:6
#12 0x42978a in main /src/libfuzzer/FuzzerMain.cpp:19:10
#13 0x7f6fef13c82f in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2082f)
#14 0x407808 in _start (out/fuzz-bus-message+0x407808)
This is an attempt to clean up the POP3/SMTP/LPR/… DHCP lease server
data logic in networkd. This reduces code duplication and fixes a number
of bugs.
This removes any support for collecting POP3/SMPT/LPR servers acquired
via local DHCP client releases since noone uses that, and given how old
these protocols are I doubt this will change. It keeps support for
configuring them for the dhcp server however.
The differences between the DNS/NTP/SIP/POP3/SMTP/LPR configuration
logics are minimized.
This removes the relevant symbols from sd-network.h (which is an
internal API only at this point after all).
This is unfortunately not well test, given the old code for this had
barely any tests. But the new code should not perform worse at least,
and allow us to release, since it corrects some interfaces visible in
the .network configuration format.
Fixes: #15943
This test is failing becuase the setup state isn't reaching 'configured'
for unknown reasons; ignore the setup state for now to prevent failures
of CI until the reason can be investigated.
Prompted by systemd/systemd#16111.
* check if /var is a mountpoint - if not, something went wrong. In case
of systemd/systemd#16111 the /failed file was created, because
systemd-cryptsetup failed, but it ended up being empty, making the result
check incorrectly pass
* forward journal messages to console - if we fail to mount /var,
journald won't flush logs to the persistent storage and we end up
empty handed and with no clue what went wrong
For example, without systemd/systemd#16111 and with this patch:
...
[FAILED] Failed to start systemd-cryptsetup@varcrypt.service.
See 'systemctl status systemd-cryptsetup@varcrypt.service' for details.
[DEPEND] Dependency failed for cryptsetup.target.
...
[ 3.882451] systemd-cryptsetup[581]: Key file /etc/varkey is world-readable. This is not a good idea!
[ 3.883946] systemd-cryptsetup[581]: WARNING: Locking directory /run/cryptsetup is missing!
[ 3.884846] systemd-cryptsetup[581]: Failed to load Bitlocker superblock on device /dev/disk/by-uuid/180ba5ef-873b-4018-9968-47c23431f71a: Invalid argument
...
[ 4.099451] sh[606]: + mountpoint /var
[ 4.100025] sh[603]: + systemctl poweroff --no-block
[ 4.101636] systemd[1]: Finished systemd-user-sessions.service.
[ 4.102598] sh[608]: /var is not a mountpoint
[FAILED] Failed to start testsuite-02.service.
dm-verity support in dissect-image at the moment is restricted to GPT
volumes.
If the image a single-filesystem type without a partition table (eg: squashfs)
and a roothash/verity file are passed, set the verity flag and mark as
read-only.
After today's Arch Linux image update, there seems to be a slight change
in delay reporting which breaks some asserts in
systemd-networkd-tests.py:
Expected:
limit 100 delay 50.0ms 10.0ms loss 20%
Current:
limit 100 delay 50ms 10ms loss 20%
Note that run_test() calls coredumpctl in a loop because in certain
environments (1 vCPU unaccelerated QEMU VM) it might take quite a
while to process the coredump.
This test runs fine locally (both on Qemu and nspawn) but sporadically fails on
autopkgtest for some reason.
Disable it while the issue is investigated to reduce noise.
The time-based cache allows starting a new unit without an expensive
daemon-reload, unless there was already a reference to it because of
a dependency or ordering from another unit.
If the cache is out of date, check again if we can load the
fragment.
In DHCPv6-PD environment, where WAN interface requests IPv6 via DHCPv6,
receives the address as well as delegated prefixes, with LAN interfaces
serving those delegated prefixes in their router advertisement messages.
The LAN interfaces on the router themselves do not have
the IPv6 addresses assigned by networkd from the prefix it
serves on that interface. Now this patch enables it.
Let's create new images public by default and then symlink/copy them
into the respective private directories afterwards, not the other way
around. This should fix a nasty race condition in parallel runs where
one tests attempts to copy the backing public image at the same moment
another test is already modifying it.
The test currently doesn't actually test configure-without-carrier since
it does have carrier for the entire test. It now forces carrier down
before starting the network portion of the test. Also, it tests to verify
the configuration is retained across future carrier losses/gains.
This allows users to configure a subnet id that should be used instead
of automatically (sequentially) assigned subnets. The previous attempt
had the downside that the subnet id would not be the same between
networkd restarts. In some setups it is desirable to have predictable
subnet ids across restarts of services and systems.
The code for the assignment had to be broken up into two pieces. One of
them is the old (sequential) assignment of prefixes and the other is the
new assignment based on configured subnet ids. The new assignment code
has to be executed first and has to be taken into account when (later
on) allocating the "old" subnets from the same pool.
Instead of having one iteration through the links we are now trying to
allocate a prefix for every link on every delegated prefix, unless they
received an assignment in a previous iteration.
Defines how link-local and autoconf addresses are generated.
0: generate address based on EUI64 (default)
1: do no generate a link-local address, use EUI64 for addresses generated
from autoconf
2: generate stable privacy addresses, using the secret from
stable_secret (RFC7217)
3: generate stable privacy addresses, using a random secret if unset
In the linked reproducer, m->fields_size == 0, and we calculate ri == -1, which
of course doesn't end well. Skip the whole calculation if m->fields_size == 0,
and also check that we don't go negative even if it is non-zero.
Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=19446 and #15583.
Support running tests in parallel by switching to copying of the
base image instead of symlinking it..
This still requires some setup steps, like running `make setup` on tests
which have unique $IMAGE_NAME beforehand (and sequentially), otherwise
they'll all try to create the same base image when started in parallel,
leading to nasty issues. However, as running the integration tests in
parallel is such an unusual use case it should be good enough, for now.
Fixes: oss-fuzz#22208
```
test/fuzz/fuzz-calendarspec/oss-fuzz-22208... ../src/shared/calendarspec.c:666:48: runtime error: signed integer overflow: 2147000000 + 1000000 cannot be represented in type 'int'
#0 0x7f0b9f6cc56a in prepend_component ../src/shared/calendarspec.c:666
#1 0x7f0b9f6cd03a in parse_chain ../src/shared/calendarspec.c:718
#2 0x7f0b9f6cea1c in parse_calendar_time ../src/shared/calendarspec.c:845
#3 0x7f0b9f6d1397 in calendar_spec_from_string ../src/shared/calendarspec.c:1084
#4 0x401570 in LLVMFuzzerTestOneInput ../src/fuzz/fuzz-calendarspec.c:17
#5 0x401ae0 in main ../src/fuzz/fuzz-main.c:39
#6 0x7f0b9e31b1a2 in __libc_start_main (/lib64/libc.so.6+0x271a2)
#7 0x40122d in _start (/home/fsumsal/repos/systemd/build/fuzz-calendarspec+0x40122d)
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior ../src/shared/calendarspec.c:666:48 in
```
Link groups are similar to port ranges found in managed switches.
You can add network interfaces to a numbered group and perform operations
on all the interfaces from that group at once.
Check:
- There is only 3 messages logged with type stdout
- Check all messages logged does not have new line: LINE_BREAK=eof
- Check that the 3 messages are logged from a different PID
- Check the 3 MESSAGE= content
As documented in systemd.path(5):
When a service unit triggered by a path unit terminates (regardless
whether it exited successfully or failed), monitored paths are
checked immediately again, and the service accordingly restarted
instantly.
This commit implements this behaviour for PathExists=, PathExistsGlob=,
and DirectoryNotEmpty=. These predicates are essentially
"level-triggered": the service should be activated whenever the
predicate is true. PathChanged= and PathModified=, on the other hand,
are "edge-triggered": the service should only be activated when the
predicate *becomes* true.
The behaviour has been broken since at least as far back as commit
8fca6944c2 ("path: stop watching path specs once we triggered the target
unit"). This commit had systemd stop monitoring inotify whenever the
triggered unit was activated. Unfortunately this meant it never updated
the ->inotify_triggered flag, so it never rechecked the path specs when
the triggered unit deactivated.
With this commit, systemd rechecks all paths specs whenever the
triggered unit deactivates. If any PathExists=, PathExistsGlob= or
DirectoryNotEmpty= predicate passes, the triggered unit is reactivated.
If the target unit is activated by something outside of the path unit,
the path unit immediately transitions to a running state. This ensures
the path unit stops monitoring inotify in this situation.
With this change in place, commit d7cf8c24d4 ("core/path: fix spurious
triggering of PathExists= on restart/reload") is no longer necessary.
The path unit (and its triggered unit) is now always active whenever
the PathExists= predicate passes, so there is no spurious restart when
systemd is reloaded or restarted.
Systems where a mount point is expected to be read-write needs a way to
fail mount units that fallback as read-only.
Add a property to allow setting the -w option when calling mount(8).
As Debian/Ubuntu use /lib/systemd instead of /usr/lib/systemd,
add systemd-journal-remote to the list of programs that test-functions
detects the correct path to, and replace its direct usage with
$SYSTEMD_JOURNAL_REMOTE
Also use $JOURNALCTL instead of journalctl.
Also minor correction in install_plymouth() to look in /lib/... as
well as /usr/lib/... and /etc/...
Remove the artifact files indicating test result (testok, failed, and
skipped) just before running the test so we always get the latest and
most relevant result instead of incorrectly consuming previous results.
Discovered in https://github.com/systemd/systemd/pull/15378#issuecomment-616801873
The disk attributes can take some time to update on certain filesystems,
so let's strip them from inputs of both `homectl` and `userdbctl` before
comparing them to avoid unexpected fails.
Also, switch from `cmp` to `diff` to make a potential test fail a bit more
debuggable.
Fixes: #14755
Let's append the date to the domain in the file name, to be able
to have multiple versions for the same domain.
There is no particular rhyme or reason to the domains being used:
I just pulled a few domains that happened to be present in issues reported
on github, even though the issues were not about pretty printing.
This doesn't really matter, since in non-/usr-merged systems plymouth
needs to be in /bin and on merged ones it doesn't matter, but it is
still prettier to insert the right path, and avoid /bin on merged
systems, since it's just a compat symlink.
Replaces: #15351
Give systemd a chance to process the stop event before checking if the
PID has indeed leaked. This should fix the intermittent test fails in CI
even with a fixed systemd version, like this one:
```
Apr 08 10:22:09 testsuite-47.sh[345]: ++ cat /leakedtestpid
Apr 08 10:22:09 testsuite-47.sh[334]: + leaked_pid=342
Apr 08 10:22:09 testsuite-47.sh[334]: + systemctl stop testsuite-47-repro
Apr 08 10:22:10 testsuite-47.sh[334]: + ps -p 342
Apr 08 10:22:10 testsuite-47.sh[348]: PID TTY TIME CMD
Apr 08 10:22:10 testsuite-47.sh[348]: 342 ? 00:00:00 sleep
Apr 08 10:22:10 testsuite-47.sh[334]: + exit 42
```
Followup to 197298ff9f
When doing 'make clean', we remove the cached image. So doing
'make -C TEST-NN-foo clean setup run clean-again' in a loop is very slow.
Let's filter out the 'clean' target (if specified), and do the cleaning
in the beginning, and then run other targets in a loop as before.
The test would fail when run again from the same image. So let's
rename the stuff we create to be more unique, and remove it before
running the test. (Removing it after would be more elegant, but it's
hard to make sure that everything is removed when things fail halfway.
Cleanup *before* tests is much more rebust.)
Using s-j-remote fixes the following issue: when coalescing files from multiple
inputs, simply copying all files with into the the same directory might
potentially mess things up, because a newer system.journal might overwrite an
older journal. This happens because we run multiple tests from the same image,
and need to clean out the directory after each run.
By using systemd-journal-remote, we nicely coalesce all files. This has the
advantage that if there aren't too many logs, we end up with just one journal
file.
ARTIFACT_DIRECTORY is for ubuntuautopackagetests, where the journal files are
copied to a separate directory to preserve after tests have been run. This
functionality can now be recreated by setting
ARTIFACT_DIRECTORY=$AUTOPKGTEST_ARTIFACTS.
This was done downstream in debian and ubuntu [1]. I want to change the
downstream file to use run-integration-tests so we can change the way tests
work more easily. Let's start moving downstream functionality upstream.
$ sudo BLACKLIST_MARKERS='blacklist-ubuntu-ci-arm64 blacklist-ubuntu-ci' \
BUILD_DIR=build test/run-integration-tests.sh
[1] https://salsa.debian.org/systemd-team/systemd/-/blob/debian/master/debian/tests/upstream
It is more trouble than it is worth. The setup is of a loopback device
is very quick, so it's better to always create it when needed and
immediately drop afterwards.
This causes the unprivileged-nspawn-root directory to be removed
after running one test. The advantage is that we reduce the maximum
disk-space use quite a bit (47*400 MB → about 18GB).
has-overflow was a temporary hack that was removed in
844da987ef (Oct. 2016). All the makefiles
can be the same, and all the targets can be handled identically.
Before, we'd copy the test tree into nspawn-root, and run the tests from there.
This is OK, and doesn't actually take much extra time. But it uses quite a lot
of extra disk space. So let's make things a bit more efficient by running
directly from the image file.
We still run the unprivileged nspawn tests from a copy. Once the kernel
implements fs shift, we can do away with that too.
Before, we'd create a separate image for each test, in
/var/tmp/systemd-test.XXXXX/rootdisk.img. Most of the images
where very similar, except that each one had some unit files installed
specifically for the test. The installation of those custom unit files
was removed in previous commits (all the unit files are always installed).
The new approach is to only create as few distinct images as possible.
We have:
default.img: the "normal" image suitable for almost all the tests
basic.img: the same as default image but doesn't mask any services
cryptsetup.img: p2 is used for encrypted /var
badid.img: /etc/machine-id is overwritten with stuff
selinux.img: with selinux added for fun and fun
and a few others:
ls -l build/test/*img
lrwxrwxrwx 1 root root 38 Mar 21 21:23 build/test/badid.img -> /var/tmp/systemd-test.PJFFeo/badid.img
lrwxrwxrwx 1 root root 38 Mar 21 21:17 build/test/basic.img -> /var/tmp/systemd-test.na0xOI/basic.img
lrwxrwxrwx 1 root root 43 Mar 21 21:18 build/test/cryptsetup.img -> /var/tmp/systemd-test.Tzjv06/cryptsetup.img
lrwxrwxrwx 1 root root 40 Mar 21 21:19 build/test/default.img -> /var/tmp/systemd-test.EscAsS/default.img
lrwxrwxrwx 1 root root 39 Mar 21 21:22 build/test/nspawn.img -> /var/tmp/systemd-test.HSebKo/nspawn.img
lrwxrwxrwx 1 root root 40 Mar 21 21:20 build/test/selinux.img -> /var/tmp/systemd-test.daBjbx/selinux.img
lrwxrwxrwx 1 root root 39 Mar 21 21:21 build/test/test08.img -> /var/tmp/systemd-test.OgnN8Z/test08.img
I considered trying to use the same image everywhere. It would probably be
possible, but it would be very brittle. By using separate images where it is
necessary we keep various orthogonal modifications independent.
The way that images are cached is complicated by the fact that we still
want to keep them in /var/tmp. Thus, an image is created on first use and
linked to from build/test/ so it can be found by other tests.
Tests cannot be run in parallel. I think that is an acceptable limitation.
Creation of the images was probably taking more resources then the actual
tests, so we should be better off anyway.
We had an fstab for the sole purpose of remounting "/" rw. Mounting root ro
is a pointless excercise in obsolete approaches. More importantly, the nspawn
image is now the same as the qemu one.
The two timezone files are now installed in the global setup. I am not too
happy about this, but it still seems better than to create a completely
separate image just for this.
I picked the list of zone files to install by grepping through the code. This
is is a bit brittle, but installing all of them takes a while, and more
importantly, writes a lot of lines to the log.
Sometimes the test would fail there, nondeterministically. I'm not sure why,
but relying on PID1 not caching the file is clearly very brittle. Let's instead
call daemon-reload.
Unfortunately meson does not install symlinks, but copies the symlink
destination instead. So symlinks need to be created by a script.
This commit adds both symlinks in test/testsuite-08.units/ and meson
scriptlet calls. Strictly speaking, the first is not necessary, since nothing
reads stuff directly from the source tree.
Yu Watanabe