NativeObjectSecurity is implemented.
I've done it in such a way that, for OSes that do not have a unified API for all ACLs,
overrides can be done in subclasses such as FileSystemSecurity without trouble.
ObjectSecurity, CommonObjectSecurity, and DirectoryObjectSecurity all implemented.
MutexSecurity, PipeSecurity, etc. all just add type specialization and nothing else. Implemented all of these as well.
Access and Audit rules now correctly use NTAccount in their string overloads (see unit tests).
The constructors all provide correct AccessMasks now. In other words, these classes are all now un-broken/no longer stubs.
More unit tests for CommonSecurityDescriptor and fixes to pass them.
A few had checks that were redundant (implemented by (Authorization|Audit|Access)Rule) as well. These have been removed.
The Allow Everyone Full Access default for null DiscretionaryAcls in CommonSecurityDescriptor is, I think,
properly implemented. This required some changes to GenericSecurityDescriptor as well, but now Mono matches
MS.NET on GetSddl/GetBinaryForm for null DiscretionaryAcl including roundtrip.
What's still missing:
(1) CommonAcl's RemoveAccess/RemoveAudit. RemoveAll and RemoveSpecific work so it's not a big issue.
(2) CommonAcl merging will not yet try to merge a CommonAce together with an ObjectAce.
(3) The various *Security constructors. Get/SetAccessControl work on files and directories.
However, you still have to create and then set. I am not too familiar with Mono runtime internal
calls, but to someone who is: If the SECURITY_ATTRIBUTES parameter to Create*'s lpSecurityDescriptor
pointed to the output of GenericSecurityDescriptor's GetBinaryForm, that would be a complete
implementation.
Unit tests included - for SystemAcl they are based on DiscretionaryAcl's mostly,
with one extra and one somewhat changed since AuditFlags merge together unlike Allow/Deny.
Anyway, it should be practical to implement CommonSecurityDescriptor and friends with these classes working.
Also fixed:
CommonAcl's merging now filters out AceFlags.InheritedFlags when !IsContainer.
For matching AccessMask, CommonAcl will now correctly merge ContainerInherit and ObjectInherit.
GenericAcl's SyncRoot is now virtual as per spec.
SecurityIdentifier.CompareTo throws ArgumentNullException, not NullReferenceException, on null parameter.
Miscellaneous unit tests:
Verify that RawAcl throws for negative capacity.
and SystemAcl need work, but CommonAcl can be manipulated using RawAcls nicely
now and will test canonicity, merge ACEs, remove meaningless ACEs, sort explicit
deny/allow ACEs by SecurityIdentifier... all the nice things CommonAcl is supposed
to do.
Anyway, this patch... Off the top of my head...
Fixed CommonAce and ObjectAce deserialization/roundtripping with opaque data.
Fixed ObjectAce having the wrong BinaryLength depending on ObjectAceFlags.
GenericAce had two protected static SDDL methods which should be internal.
Added some unit tests for the above. One of the tests compares the binary
output with a blob I produced using the *same unit test* on MS.NET (with a
File.WriteAllText creating the byte array to paste in, naturally), and
happily, it produces identical binary output. Very nice.
I determined the behavior of CompareTo by first writing unit tests for
CommonAcl (instantiated as DiscretionaryAcl but using RawAcl instead of
DiscretionaryAcl methods to do the manipulation to keep DiscretionaryAcl
out of the loop). I developed and ran these unit tests on MS.NET using
MonoDevelop...
Anyway, CompareTo appears to compare authority, then the *number* of
subauthorities, and lastly the subauthorities. I'd be curious if there
is more detail, but so far this seems to cover it. It's definitely
*not* just a comparison of the string.
This patch also modifies CommonAcl to take advantage of this new sorting,
so now it should sort groups of explicit ACEs (access deny, access allow)
in canonical ACLs correctly.
With that down, I've got two unit tests that still fail on Mono.
Now, accessMask matches the MutexRules passed in. The MutexRules enum is a bitmask
corresponding to Win32 ACE accessMasks and the two match.
I verified that accessMask corresponds directly to the MutexRules passed in the
constructor on Win32 using Reflection. It stands to reason, and it turns out to be
the case. System.Security.AccessControl is still mostly broken but this at least
corrects MutexAccessRule.
Also added constructors to ObjectSecurity<T>.
It's supposed to have them (they just pass to NativeObjectSecurity),
but the only user in Mono is MemoryMappedFileSecurity so this is
just for completeness's sake. Also, this lets the test depend on neither
System.Core nor internal methods.
Many derived classes of AuthorizationRule have string overloads which pass
'new NTAccount (identity)'. The exception docs state "can be cast as a
SecurityIdentifier" not "is a SecurityIdentifier". Anyway, this fixes all the
string constructors in System.Security.AccessControl.
Also, I added checking for inheritanceFlags and propagationFlags parameters.
While building the unit test, I found accessMask == 0 throws ArgumentException,
not ArgumentOutOfRangeException, on MS.NET 4.0 (contrary to docs). So,
I've made this match.
Details worth being aware of, that aren't well documented in MS.NET:
(1) The Rights property is just a casted AccessMask.
I've checked this by testing with MS.NET -- if you use AccessRule<Test>
with an arbitrary struct Test { }, MS.NET has an InvalidCastException
on the Rights getter if you create with ObjectSecurity's factory method
(which is passed an int), and in the constructor if you use AccessRule's
constructor.
(2) The string constructor overloads call 'new NTAccount(identity)'.
What these methods do for AccessRule is not documented by MS.NET,
but this *is* documented for SemaphoreAccessRule, and my testing
with C# shows AccessRule does the same thing. So, I've implemented
this behavior to match.
I will submit a separate patch afterwards to make AuthorizationRule's
accept NTAccount, which is required for the string overloads to work.
Contrary to the MS.NET *documentation* (which Mono follows), AuthorizationRule
actually accepts NTAccount as well as (the documented) SecurityIdentifier.
I verified this by deriving a class from AuthorizationRule.
* EventWaitHandle.cs:
* Mutex.cs: Implement OpenExisting
* NativeEventCalls.cs: Add OpenEvent icall for OpenExisting in
2.0. Add a "created" boolean out parameter to CreateEvent icall.
* ManualResetEvent.cs:
* AutoResetEvent.cs: Update CreateEvent icall signature (now has
"created" boolean out parameter.)
2005-12-23 Dick Porter <[email protected]>
* Semaphore.cs: Implement OpenExisting
2005-12-23 Dick Porter <[email protected]>
* MutexRights.cs: New for 2.0 profile
2005-12-23 Dick Porter <[email protected]>
* SemaphoreRights.cs: Make the [Flags] enum more obvious
2005-12-23 Dick Porter <[email protected]>
* SemaphoreTest.cs: Enable another test
2005-12-23 Dick Porter <[email protected]>
* corlib.dll.sources: Added System.Security.AccessControl.MutexRights
svn path=/trunk/mcs/; revision=54801
* AccessControlSections.cs: New (2.0). Enum required for DirectoryInfo
* DirectorySecurity.cs: New (2.0). Partial stub for DirectoryInfo
svn path=/trunk/mcs/; revision=54392
* CryptoKeySecurity.cs: Class declaration to complete CspParameters
and CspKeyContainerInfo in System.Security.Cryptography.
svn path=/trunk/mcs/; revision=38665
* AccessControlActions.cs: New enum in Fx 2.0. Required for new stuff
in System.Security.Permissions. Will also "creates" the namespace for
"corcompare".
svn path=/trunk/mcs/; revision=30849