Skip to content
This repository has been archived by the owner on May 21, 2024. It is now read-only.

Fix/ota 3057/metadata matching #1258

Merged
merged 11 commits into from
Jul 23, 2019
Merged

Conversation

pattivacek
Copy link
Collaborator

@pattivacek pattivacek commented Jul 17, 2019

Looks like a lot of changes, but almost all of it is just adapting existing tests to handle one big new change: hardware IDs must now match between Director and Image repo Target objects. This can be seen in tuf.cc.

Details: The Director sends a map of serials to HWIDs, but the Image repo sends us a vector of HWIDs. If the map and vector match, the comparison succeeds. Otherwise, we expect that every HWID in the map can be found in the vector. Added tests for all of these cases. Added explanations for the rest of the fields we use for why we don't check them. Added a requirement to aktualizr-repo that the "image" command must have a hardware ID so that we can properly populate the metadata.

This PR depends on advancedtelematic/tuf-test-vectors#56.

@pattivacek pattivacek force-pushed the fix/OTA-3057/metadata-matching branch 2 times, most recently from 644abf9 to 69c9791 Compare July 17, 2019 12:22
@codecov-io
Copy link

codecov-io commented Jul 17, 2019

Codecov Report

❗ No coverage uploaded for pull request base (master@383dfc3). Click here to learn what that means.
The diff coverage is 94.04%.

Impacted file tree graph

@@            Coverage Diff            @@
##             master    #1258   +/-   ##
=========================================
  Coverage          ?   79.38%           
=========================================
  Files             ?      171           
  Lines             ?    10211           
  Branches          ?        0           
=========================================
  Hits              ?     8106           
  Misses            ?     2105           
  Partials          ?        0
Impacted Files Coverage Δ
src/aktualizr_repo/image_repo.h 100% <ø> (ø)
src/aktualizr_repo/uptane_repo.h 100% <ø> (ø)
src/libaktualizr/primary/sotauptaneclient.h 100% <ø> (ø)
src/aktualizr_repo/director_repo.cc 91.17% <100%> (ø)
...libaktualizr/package_manager/packagemanagerfake.cc 89.79% <100%> (ø)
src/aktualizr_repo/uptane_repo.cc 96.55% <100%> (ø)
...tualizr/package_manager/packagemanagerinterface.cc 89.87% <100%> (ø)
src/aktualizr_repo/main.cc 81.7% <100%> (ø)
src/aktualizr_repo/image_repo.cc 93.69% <100%> (ø)
src/libaktualizr/storage/sqlstorage.cc 74.74% <100%> (ø)
... and 4 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 383dfc3...5320005. Read the comment docs.


for (auto it = t.ecus_.begin(); it != t.ecus_.end(); ++it) {
os << it->first;
for (auto it = t.ecus_.cbegin(); it != t.ecus_.cend(); ++it) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All these loops could use simpler for range loops:

for (const auto &ecu : t.ecus) {
  ...
}

if (filename_ != t2.filename_) {
return false;
}
if (length_ != t2.length_) {
return false;
}

// If the HWID vector and ECU->HWID map match, we're good. Otherwise, assume
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's maybe a naive question, but maybe it's the time where it's too much to have all this logic in an overloaded comparison operator and instead have a dedicated function? The potential for misuse and confusion seems big.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would you prefer to have operator== call that function, or to reserve operator== only for exact matches and use the new match function for Uptane purposes? I'm hoping the former, because otherwise I'd worry about accidentally doing the wrong thing (using equality instead of the matching function).

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I imagined that == would not call that but then you're probably right. In this case it would be better to delete the comparison operator and only keep the method.
But I'm ready to debate that.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Deleting the comparison operator is an interesting idea, but at that point, aren't we back to just having the same thing, just named differently? What's the benefit?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As a matter of style it can be surprising to have this complicated business logic inside this operator with a benign name. And the objects are not really "equal", they "match" in some application sense, so I think it's stretching the sense a bit. Also, it makes it hard to grep for usage of this comparison.

It's why I would prefer this to have a scary name but maybe it's not really feasible or we can argue that we contain the usage effectively.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see your points. I think if we delete the == operator then it's probably a reasonable way to go.

ecu_map.insert({Uptane::EcuSerial("serial"), hwid});
Uptane::Target target1("abc", generateDirectorTarget("hash_good", 739, ecu_map));
Uptane::Target target2("abc", generateImagesTarget("hash_good", 739, hardwareIds));
EXPECT_TRUE(target1 == target2);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also test target2 == target1?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Probably not a bad idea. They should be identical, but it would help prevent potential mistakes to future edits to the Target class.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes I would be more comfortable if all the tests checked the reflexivity.

@@ -216,7 +216,8 @@ void INvStorage::FSSToSQLS(FSStorageRead& fs_storage, SQLStorage& sql_storage) {
}

bool INvStorage::fsReadInstalledVersions(const boost::filesystem::path& filename,
std::vector<Uptane::Target>* installed_versions, size_t* current_version) {
std::vector<Uptane::Target>* installed_versions, size_t* current_version,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is it intentional that importInstalledVersions was not changed to adopt these changes? It's a code part that should get more hits than FSStorageRead::loadInstalledVersions which is only used for legacy migrations.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorta. This change was a bit of a hack to get tests to work. importInstalledVersions should only run on a primary, and there is already logic to assume it's for the primary when reading the installation versions in the function that converts FS to SQL storage. However, the tests had to do some fakery to test that effectively, so I had to force the serial/hwid. Would you prefer that importInstalledVersions also used it? In that case, I probably wouldn't need the parameters to be optional; I think I could make them a requirement.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's a bit annoying to have this almost dead code to make the tests work indeed. Then it should maybe explained somewhere with a good comment (maybe I've missed it?), as I found it a bit confusing in this state.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There is no comment at present. The other option is to do some more hackery in the test itself, which is maybe safer, but makes the test slightly less meaningful. It's kind of ugly no matter how you slice it, but I'm open to advice.

boost::filesystem::path repo_dir(path_ / "repo/image");

boost::filesystem::path targets_path =
delegation ? ((repo_dir / "delegations") / delegation.name).string() + ".json" : repo_dir / "targets.json";
Json::Value targets = Utils::parseJSONFile(targets_path)["signed"];
// TODO: support multiple hardware IDs.
target["custom"]["hardwareIds"][0] = hardware_id;
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am just wondering why it's in "custom" section which sounds like a bucket for optional params while hardware-ID looks like a mandatory param.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I couldn't really tell you, but that's how Uptane specifies it. See https://uptane.github.io/uptane-standard/uptane-standard.html#custom-metadata-about-images. I think the point is that it is not a hard requirement, but rather a SHOULD. If we were to want to change it, first we'd have to convince the backend team, but I think it's fine as is, even though it is admittedly confusing.

std::string hwid = vm["hwid"].as<std::string>();
std::string serial = vm["serial"].as<std::string>();
const std::string targetname = vm["targetname"].as<std::string>();
const std::string hwid = vm["hwid"].as<std::string>();
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like line #154 and #107 are identical

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Indeed. However, they are two different subcommands of aktualizr-repo, and other subcommands don't require it, so the redundancy is probably the cleanest solution.

@@ -26,7 +26,8 @@ TEST(PackageManagerFake, FinalizeAfterReboot) {

PackageManagerFake fakepm(config.pacman, storage, bootloader, nullptr);

Uptane::Target target("pkg", {Uptane::Hash(Uptane::Hash::Type::kSha256, "hash")}, 1, "");
std::map<Uptane::EcuSerial, Uptane::HardwareIdentifier> primary_ecu;
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's not exactly related to this specific PR rather to the code base in general - this map can be found in many places of the code and effectively represents an ID of ECU, just thinking if it makes sense to define corresponding types, e.g. class ECU { class/using/typedef ID ..... so it can be referenced as Uptane::ECU::ID ...

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not a bad thought. I'll look into it.

@@ -209,6 +215,61 @@ bool Target::IsOstree() const {
}
}

bool Target::operator==(const Target &t2) const {
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

At first glance, it's confusing that a single class represents two types of Targets that require different context/class members, as well as their comparison.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That is an interesting point. @lbonn what would you think of having two children classes, one for DirectorTarget and one for ImagesTarget? Almost all of the logic would still be in the parent class, but the ecu map and hwid vectors would be in the respective children, and the match function would probably have to be a static function outside of the classes. I think that would help make things much, much more explicit.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

On second thought, this is a pretty huge change and will require changing the API. I'm deep enough in this task already that I think I'd rather wait on that work and come back to it another time.

} else {
return false;
}
for (auto map_it = ecu_map->cbegin(); map_it != ecu_map->cend(); ++map_it) {
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am rather trying to understand a design & control flow here than comment the code. What's the point to specify ECU serial at all if the solution will update all ECUs that have HW-ID specified in the target metadata. At least it's how it works for secondaries, I ask to update just one ECU but the solution updates all ECUs of the given type/HW-ID.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That logic (of selecting all secondaries with the same HWID) is on the server-side. I'd prefer to not rely on that in libaktualizr. Besides, we still have to distinguish the primary. It's also a safety check: we need to confirm that the ECU has the HWID we expect it to (which is checked elsewhere).

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That makes sense, policies are defined at the backend and can be even account/user specific, a a client/aktualizr does exactly what Director says without applying any policies.

Json::Value toDebugJson() const;
friend std::ostream &operator<<(std::ostream &os, const Target &t);

private:
bool valid{true};
std::string filename_;
std::string type_;
std::map<EcuSerial, HardwareIdentifier> ecus_;
std::map<EcuSerial, HardwareIdentifier> ecus_; // Director only
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's what I meant in one of the earlier comment, a field is used just for a specific type of Target instances.

mike-sul
mike-sul previously approved these changes Jul 18, 2019
@pattivacek pattivacek force-pushed the fix/OTA-3057/metadata-matching branch 5 times, most recently from 907e488 to 3c3c64c Compare July 18, 2019 14:41
Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
Added a check for HW IDs. The Director sends a map of serials to HWIDs,
but the Image repo sends us a vector of HWIDs. If the map and vector
match, the comparison succeeds. Otherwise, we expect that every HWID in
the map can be found in the vector. Added tests for all of these cases.

Also added explanations for the rest of the fields we use for why we
don't check them.

Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
Mostly just better error handling and messaging for Target mismatches.

Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
@pattivacek pattivacek force-pushed the fix/OTA-3057/metadata-matching branch from 3c3c64c to 4620480 Compare July 19, 2019 09:01
This required adding functionality to aktualizr-repo to add HWIDs to the
image repo.

Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
Fixes OTA-2706 (Remove tests/test_data/fake_root). Also fixes the last
broken test due to the Target matching changes.

Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
@pattivacek pattivacek force-pushed the fix/OTA-3057/metadata-matching branch from 4620480 to 38537a4 Compare July 19, 2019 10:23
This should make Target matching more explicit (and easier to grep for).
The == operator is not usually meaningful or desirable for Target
objects.

Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
Signed-off-by: Patrick Vacek <patrickvacek@gmail.com>
@pattivacek
Copy link
Collaborator Author

I believe I've finally addressed all review comments except for making Director/Image-specific children of the Target class. I looked into it and it is way too much work to do as part of this PR. It's not a bad idea, but that's for some other time.

@pattivacek pattivacek merged commit 4348855 into master Jul 23, 2019
@pattivacek pattivacek deleted the fix/OTA-3057/metadata-matching branch July 23, 2019 13:48
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants