sdbusplus, branch master

Emit adding/removing interfaces for object server

2020-02-13T06:49:15+00:00

The object server currently either creats the objects and interfaces, or defer the signal by not adding objects. In practice, we have situations that the code would like to add interfaces to an existing object, and it's not supported, or needs tricky code to workaround. Exmaples: https://gerrit.openbmc-project.xyz/c/openbmc/phosphor-bmc-code-mgmt/+/5820 https://gerrit.openbmc-project.xyz/c/openbmc/openpower-pnor-code-mgmt/+/5346 This commit adds the support by: 1. Adding emit_added() in interface.hpp and the generated server.hpp 2. Adding a enum class in object's constructor to indicate which action to do, to create the object, or adding the interface, or defer signal as before. So the user of object<> could pass `action::emit_interface_added` to the constructor to tell the object server *only* emit interface added to DBus, without emitting object added. The previous code stays the same behavior: * If `true` is passed in object's constructor, it defers emitting object added signal; * If no extra parameter is passed in object's constructor, it emits object added signal as before. Tested: 1. Make sure the openbmc builds fine with https://gerrit.openbmc-project.xyz/c/openbmc/phosphor-logging/+/25089 because phosphor-logging uses its own server.hpp for interface, the above patch removes that. 2. Manually write a small service to verify the interfaces are added and removed by using the `emit_interface_added` action. 3. Added the unit test cases for object.hpp to check the ctor/dtor with different actions. Signed-off-by: Lei YU Change-Id: I178c5bed3c9ff39ee2ac8d143fbe9131b0753dfa

Fix an issue found by cppcheck

2020-02-11T03:29:31+00:00

cppcheck finds an issue: [sdbusplus/asio/object_server.hpp:46] -> [sdbusplus/asio/object_server.hpp:315]: (style) The function 'set' overrides a function in a base class but is not marked with a 'override' specifier. Fix it by adding the override keyword correctly. Signed-off-by: Lei YU Change-Id: Ie032229874bb416f7c6ec142d9bf38f8ffcff877

sdbus++: interface: fix indent of enum-string fn

2020-02-11T02:49:21+00:00

The convertFromString functions were indented incorrectly relative to the rest of the class. Signed-off-by: Patrick Williams Change-Id: Iabe9d30c33c8707bf2bf4c6a697e3d27d0a3000b

MAINTAINERS: add Patrick Williams

2020-02-11T02:44:02+00:00

Large portions of the codebase were originally written by me before maintainers existed. I am returning to the project and plan to do significant enhancements and cleanups to sdbusplus. Signed-off-by: Patrick Williams Change-Id: I3ac6f6597c78041785ded272d1e6ce8974d5b49a

Catch async errors so they do not throw

2020-02-10T19:13:10+00:00

When running an async operation, it is not a good idea to throw because the current running context is not the context that it was originally registered from, it is the io run-loop context. So any methods that *would* throw, must run in a try/catch and then set the error code properly so that the handler will be notified of the error. This was caught when a process (bmcweb) was used to invoke a call that had a bad path. That was not caught and it caused bmcweb to abort because of an unhandled exception. Change-Id: Ie66a029066a3f463759089b44368b7518d0de8ce Signed-off-by: Vernon Mauery

Make interface string public

2020-02-04T06:57:31+00:00

The interface string was private, and the users of sdbusplus have to define their own strings for the D-Bus interfaces. By making the interface string public, users do not have to define their own and the generated string could be used instead. E.g. one could directly use "Calculator::interface" instead of "net.poettering.Calculator" in the example. Signed-off-by: Lei YU Change-Id: I0846d4e0609c8588d81e8a19bdd69fd79df83848

maintainers: Add Lei YU as maintainer

2020-01-15T06:57:54+00:00

Lei has been an active reviewer for sdbusplus, and fixed the CI issue that has been there for quite a while. He is willing to become one of the maintainers for sdbusplus, which is good for speed up the review progress, and for redundancy. Signed-off-by: Lei YU Change-Id: I39ec5a1adc0d34c4ce2919a92bd7df463818b9d6

maintainers: Fix dkodihal's email address

2020-01-15T06:57:54+00:00

The email address of Deepak is incorrect that does match the one in gerrit. Fix it to use the correct one. Signed-off-by: Lei YU Change-Id: I0f2cc9bd89f0c2cf232d41018071aeea237cc1fe

Add valgrind suppression

2020-01-15T06:57:54+00:00

On latest OpenBMC, sdbusplus CI fails due to valgrind check like below on ppc64le systems: ==5290== Syscall param epoll_ctl(event) points to uninitialised byte(s) ==5290== at 0x4F2FB08: epoll_ctl (syscall-template.S:79) ==5290== by 0x493A8F7: UnknownInlinedFun (sd-event.c:961) ==5290== by 0x493A8F7: sd_event_add_time (sd-event.c:1019) ==5290== by 0x190BB3: phosphor::Timer::Timer(sd_event*, std::function) (timer.hpp:62) ==5290== by 0x192B93: TimerTest::TimerTest() (timer.cpp:25) ==5290== by 0x193A13: TimerTest_timerExpiresAfter2seconds_Test::TimerTest_timerExpiresAfter2seconds_Test() (timer.cpp:85) ==5290== by 0x19E11F: testing::internal::TestFactoryImpl::CreateTest() (gtest-internal.h:472) ==5290== by 0x4A52D3B: HandleSehExceptionsInMethodIfSupported (gtest.cc:2447) ==5290== by 0x4A52D3B: testing::Test* testing::internal::HandleExceptionsInMethodIfSupported(testing::internal::TestFactoryBase*, testing::Test * (testing::internal::TestFactoryBase::*)(), char const*) (gtest.cc:2483) ==5290== by 0x4A40BCB: Run (gtest.cc:2693) ==5290== by 0x4A40BCB: testing::TestInfo::Run() (gtest.cc:2676) ==5290== by 0x4A40DC3: Run (gtest.cc:2825) ==5290== by 0x4A40DC3: testing::TestCase::Run() (gtest.cc:2810) ==5290== by 0x4A414AF: testing::internal::UnitTestImpl::RunAllTests() (gtest.cc:5216) ==5290== by 0x4A5329B: HandleSehExceptionsInMethodIfSupported (gtest.cc:2447) ==5290== by 0x4A5329B: bool testing::internal::HandleExceptionsInMethodIfSupported(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl:: *)(), char const*) (gtest.cc:2483) ==5290== by 0x4A416AF: testing::UnitTest::Run() (gtest.cc:4824) ==5290== by 0x4A90917: RUN_ALL_TESTS (gtest.h:2370) ==5290== by 0x4A90917: main (gmock_main.cc:69) ==5290== Address 0x1fff00eafc is on thread 1's stack ==5290== in frame #0, created by epoll_ctl (syscall-template.S:78) ==5290== The root cause is: 1. GCC has a bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77992, that the padding bytes are not initialized. 2. In systemd, the code in [libsystemd/sd-event/sd-event.c][1] using epoll_event hit the above bug: typedef union epoll_data { void *ptr; int fd; uint32_t u32; uint64_t u64; } epoll_data_t; struct epoll_event { uint32_t events; /* Epoll events */ epoll_data_t data; /* User data variable */ } __EPOLL_PACKED; In glibc, on x86, `__EPOLL_PACKED` is defined as `__attribute__ ((__packed__))`[2], so it's packed and there are no internal padding bytes; On other architectures (e.g. ppc64le), __EPOLL_PACKED is not defined and thus there are 4 internal padding bytes between `events` and `data`, that are not initialized. 3. When epoll_ctl() is invoked, in [Linux kernel][3], it does a copy_from_user() to copy the whole struct into kernel space. That's why Valgrind reports "epoll_ctl(event) points to uninitialised byte(s)", only for non-x86 platforms. 4. The timer test in this repo invokes sd_event_add_time() and eventually hit the above code. The GCC bug is not resolved from 2016. The systemd code is actually correct. There is a [PR][4] sent to systemd trying to workaround the issue, and it is being discussed. Before GCC bug is resolved or systemd PR is accepted, suppress the valgrind error to make it pass the CI. [1]: https://github.com/systemd/systemd/blob/v242/src/libsystemd/sd-event/sd-event.c#L961 [2]: https://github.com/bminor/glibc/blob/f1a0eb5b6762b315517469da47735c51bde6f4ad/sysdeps/unix/sysv/linux/x86/bits/epoll.h#L29 [3]: https://github.com/torvalds/linux/blob/d1eef1c619749b2a57e514a3fa67d9a516ffa919/fs/eventpoll.c#L2095 [4]: https://github.com/systemd/systemd/pull/14353 Signed-off-by: Lei YU Change-Id: I3a29fd0e28e5c7b69037a7ef2ef9571f93d80df7

Fix vtable CI error

2020-01-15T06:57:52+00:00

The CI reports failure in VtableTest.SameContent, where the case is comparing the binary of example vtables constructed from sdbusplus and macros defined in systemd. The root cause has two parts. Part I: When systemd is updated, the members in struct sd_bus_vtable is updated, e.g. * From v239 to v242, sd_bus_vtable.x.start is updated from struct { size_t element_size; } start; to: struct { size_t element_size; uint64_t features; } start; and sd_bus_vtable.x.method is updated from struct { const char *member; const char *signature; const char *result; sd_bus_message_handler_t handler; size_t offset; } method; to: struct { const char *member; const char *signature; const char *result; sd_bus_message_handler_t handler; size_t offset; const char *names; } method; * From v242 to v243, sd_bus_vtable.x.start is updated to struct { size_t element_size; uint64_t features; const unsigned *vtable_format_reference; } start; The code in vtable.hpp only assign the members in v239, the new members are intialized with 0, so it differs from the vtable constructed from macros defined by systemd. The fix is to use macros from systemd in vtable.hpp as well, which is suggested by systemd: > Please do not initialize this structure directly, use the > macros below instead Part II: The `const char *names` in struct method and signal introduced between systemd v239 and v242 is a pointer to const strings. By default they are "empty" strings, but there is no guarantee that the compiler will use the same pointer for the same string. So the test case can not assume the binaries are the same for two vtables even they are constructed with the same parameters. Update the test case to use real `const char*` and handler/get/set functions and check each member of the struct instead of comparing the binary data by memcmp. Tested: Verify the CI passes. Change-Id: I9e94ff16075dd3f12d73e96438c0d864203bdcf4 Signed-off-by: Lei YU