9.4.2.2. Serialization and Deserialization¶
9.4.2.2.1. [Critical] DataReader on a Topic using an appendable type may receive samples with incorrect value¶
A DataReader subscribing to a Topic on an appendable type may have received incorrect samples from a matching DataWriter.
The problem only occurred when the DataWriter published a type with fewer members than the DataReader type. For example, consider a DataWriter on FooBase and a DataReader on FooDerived:
@appendable struct FooBase {
sequence<uint8,1024>base_value;
};
@appendable struct FooDerived {
sequence<uint8,1024> base_value;
@default(12) uint8 derived_value;
};
In this case, the serialized sample stream would be padded with extra
bytes to align the stream to 4 bytes as required by the OMG Extensible
and Dynamic Topic Types for DDS
specification, version 1.3. However, the additional padding bytes were
incorrectly interpreted as part of the data and derived_value
may
have been set to a random value.
For example, in the case above, when the DataWriter published a sample
with type FooBase, in some cases the DataReader received a sample in
which the field derived_value
was set to 0 instead of 12.
Note
Connext Micro does not support the @default
annotation.
[RTI Issue ID MICRO-6402]
9.4.2.2.2. [Critical] Malformed samples with invalid strings not dropped by DataReader¶
A DataReader may have provided the application a malformed sample containing an invalid value (not Null-terminated) for a string member. The string member may not have been Null-terminated, resulting in undefined behavior if the application tried to access it.
Now, the DataReader will not deserialize the sample and the sample will not be provided to the application.
[RTI Issue ID MICRO-3039]
9.4.2.2.3. [Major] Float and double ranges may not have been enforced correctly¶
Float and double ranges may not have been enforced correctly. Float and double member values that should not have passed the check ended up passing it.
This issue only occurred under any of the following conditions:
For “float”:
When @min was set to -3.4E38 for a member, a value smaller than @min passed the check when it should not have.
When @max was set to 3.4E38 for a member, a value greater than @max passed the check when it should not have.
For “double”:
When @min was set to -1.7E+308 for a member, a value smaller than @min passed the check when it should not have.
When @max was set to 1.7E+308 for a member, a value greater than @max passed the check when it should not have.
For “float” and “double”:
When the member value was set to INFINITY, samples passed the range check when they should not have.
When the member value was set to NaN, samples passed the range check when they should not have.
[RTI Issue ID MICRO-3280]
9.4.2.2.4. [Major] Deserialization of tampered/corrupted samples may have unexpectedly succeeded¶
A DataReader may not have detected that a truncated sample due to corruption or tampering was invalid. As a result, the application may have received samples with invalid content.
Now, the deserialization of corrupted samples fails, and they are not provided to the application.
[RTI Issue ID MICRO-3057]
9.4.2.2.5. [Major] Invalid serialization of samples with types containing nested structures with primitive members that require padding¶
In Connext DDS 6.0.1 and earlier, the serialization of samples with a type containing two or more levels of nested complex types, where the nested types have primitive members that require padding, may have failed. This means that a DataReader may have received an invalid value for a sample. Example:
// Level-2 Nested type
struct Struct1 {
uint8 m1;
uint8 m2;
int32 m3;
};
// Level-1 Nested type
struct Struct2 {
int32 m1;
int32 m2;
uint8 m3;
uint8 m4;
Struct1 m5;
};
struct Struct3 {
Struct2 m1;
};
In the above example, Struct2 and Struct1 are nested, and there is padding between Struct1::m2 (1-byte aligned) and Struct1::m3 (4-byte aligned) of 2 bytes.
This issue only applied to nested types that are appendable or final for XCDR1 data representation or final for XCDR2 data representation.
This problem affected DynamicData and the generated code for the following languages: C, C++, C++03, and C++11.
For generated code, a potential workaround to this problem was to generate code with a value of 1 or 0 for the -optimization, but this may have had performance implications.
[RTI Issue ID MICRO-2744]
9.4.2.2.6. [Minor] Serialization of string members did not check for null-terminated strings in C, traditional C++, and modern C++¶
The code executed by a DataWriter that serializes string members in a Topic type did not check that the strings are null-terminated. This may have led to undefined behavior, because the serialization code calls strlen.
This problem has been fixed. The serialization code now checks for null-terminated strings with the maximum allowed length and reports the following error if the string is not well-formed:
RTIXCdrInterpreter_serializeString:StrStruct:member2 serialization error. String length (at least 6) is larger than maximum 5
[RTI Issue ID MICRO-3040]