Introduce tests for sysctl(3), and sysarch(2) for AMD64, i386, ARM & MIPS architectures

[aniketp]

Summary

This revision introduces a test-program for miscellaneous (or 'other') audit class.
The syscalls for which tests are introduced, are:

• sysctl(3)
• sysarch(2)

Note: Since sysarch(3) is dependent on machine architecture, tests will be generated
on the basis of allowed archs.

For example, on my machine, both AMD64 and i386 are supported, thus

▶ kyua list miscellaneous
miscellaneous:sysarch_amd64_success
miscellaneous:sysarch_failure
miscellaneous:sysarch_i386_success
miscellaneous:sysctl_failure
miscellaneous:sysctl_success

Test Plan

Execute make && make install from test/sys/audit.
Execute kyua test from /usr/tests/sys/audit. All testcases should succeed.

[aniketp]

• Updates in Copyright

[asomers]

How about audit(2)?

[aniketp]

• Add test for audit(2) in failure case

[asomers]

The failure test case looks ok, but I think we should wait for the success test case before committing.

[asomers]

tests/sys/audit/miscellaneous.c

260 ↗

(On Diff #44856)

I think you should rename the sysarch tests. We aren't actually testing AMD64_GET_FSBASE; we don't care what it does. We're just testing the architecture-independent audit(4) connection of sysarch(2). Therefore I think you should have just a single test case called sysarch_success. Use conditional compilation to call atf_skip on unsupported architectures.

[aniketp]

Create a single success mode test case for sysarch(2) with sysnum being dependent on the system architecture

[asomers]

tests/sys/audit/miscellaneous.c

99 ↗

(On Diff #45280)

I'm curious. Why is sysarch unsafe on sparc?

[aniketp]

tests/sys/audit/miscellaneous.c

99 ↗

(On Diff #45280)

Well. TBH, I don't know. The only allowed syscalls were SPARC_UTRAP_INSTALL and SPARC_SIGTRAMP_INSTALL. I don't know what they do and not sure if I should add them to the tests.

Though if you think they are safe enough, I can add their tests no issue.

[asomers]

tests/sys/audit/miscellaneous.c

99 ↗

(On Diff #45280)

It looks fine. According to an old Solaris man page for that function, you can use it to install no handler for a certain trap. That should be a nop. https://docs.oracle.com/cd/E19455-01/806-0626/6j9vgh69s/index.html

[aniketp]

Include Sparc64 in the possible architectures for sysarch(2)

[aniketp]

Add the missed \#endif statement

[aniketp]

• Add the missed sysnum for sparc64 tests

[asomers]

tests/sys/audit/miscellaneous.c

123 ↗

(On Diff #45288)

This isn't safe. For one thing, sysarg is uninitialized. For another, the kernel may be interpreting it differently for different syscalls. It might be expecting a pointer to a structure containing other pointers, for example.

[aniketp]

• Use respective args instead of a common void *sysarg

[asomers]

tests/sys/audit/miscellaneous.c

93 ↗

(On Diff #45396)

Technically this is an input/output argument, so you still need to initialize all fields or tools like Coverity will complain.

97 ↗

(On Diff #45396)

This should be a register_t, not a void*.

103 ↗

(On Diff #45396)

Did you mean to change the value of sysnum?

133 ↗

(On Diff #45396)

May as well inline sysnum here instead of setting it as a variable above.

[aniketp]

• Changes as suggested in the previous review

[asomers]

The build fails on arm.

/home/asomers/freebsd/base/head/tests/sys/audit/miscellaneous.c:99:39: error: expected ';' at end of declaration
        struct arm_sync_icache_args armsysarg {

[aniketp]

tests/sys/audit/miscellaneous.c

102 ↗

(On Diff #45407)

@asomers, I have the semicolon here. Why would the build fail otherwise?

[aniketp]

• Replace ARM_SYNC_ICACHE with ARM_DRAIN_WRITEBUF as the syscall argument

ARM_SYNC_ICACHE for some reason breaks build on arm as pointed out by @asomers. It can't find the declaration of the corresponding struct and hence, assumes that the test-program is declaring the struct instead of defining it.

[asomers]

In D16116#346010, @aniketp wrote:

• Replace ARM_SYNC_ICACHE with ARM_DRAIN_WRITEBUF as the syscall argument

ARM_SYNC_ICACHE for some reason breaks build on arm as pointed out by @asomers. It can't find the declaration of the corresponding struct and hence, assumes that the test-program is declaring the struct instead of defining it.

You should really try to figure out why it didn't compile with ARM_SYNC_ICACHE. It would be a good exercise for you.

[aniketp]

• Fix build for ARM and Sparc64 architectures

[asomers]

Build fails on sparc.

/home/asomers/freebsd/base/head/tests/sys/audit/miscellaneous.c: In function 'atfu_sysarch_success_body':
/home/asomers/freebsd/base/head/tests/sys/audit/miscellaneous.c:106: warning: initialization makes pointer from integer without a cast
/home/asomers/freebsd/base/head/tests/sys/audit/miscellaneous.c:107: warning: initialization makes pointer from integer without a cast

[aniketp]

• Fix 'pointer from integer without a cast' build failure on sparc64