LLDB `SBMemoryRegionInfo` Python API does not expose `type` of memory region (macOS)

[patryk4815]

On macOS, when querying memory regions via LLDB’s Python API, the type field returned by gdb-remote packets (e.g., type:malloc-metadata) is not exposed through lldb.SBMemoryRegionInfo.

There is no GetType() method, so scripts cannot access memory region types even though the underlying packets provide this information.

Here type: is parsed:

llvm-project/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationClient.cpp

Lines 1638 to 1645 in a87f776

	} else if (name == "type") {
	for (llvm::StringRef entry : llvm::split(value, ',')) {
	if (entry == "stack")
	region_info.SetIsStackMemory(MemoryRegionInfo::eYes);
	else if (entry == "heap")
	region_info.SetIsStackMemory(MemoryRegionInfo::eNo);
	}
	} else if (name == "error") {

Example

(lldb) b main
(lldb) log enable gdb-remote all
(lldb) script
>>> regions = lldb.process.GetMemoryRegions()
# Example packets show type field is present
<  87> read packet: $start:100020000;size:4000;permissions:r;dirty-pages:100020000;type:malloc-metadata;#00
<  88> read packet: $start:100024000;size:4000;permissions:rw;dirty-pages:100024000;type:malloc-metadata;#00

>>> region = lldb.SBMemoryRegionInfo()
>>> regions.GetMemoryRegionAtIndex(7, region)
True
>>> region
[0x0000000100024000-0x0000000100028000 RW-]
>>> region.GetType()
Traceback (most recent call last):
  File "<console>", line 1, in <module>
AttributeError: 'SBMemoryRegionInfo' object has no attribute 'GetType'

Expected behavior

• SBMemoryRegionInfo should expose the type field from gdb-remote packets

[llvmbot]

@llvm/issue-subscribers-lldb

Author: None (patryk4815)

On macOS, when querying memory regions via LLDB’s Python API, the `type` field returned by gdb-remote packets (e.g., `type:malloc-metadata`) is **not exposed** through `lldb.SBMemoryRegionInfo`.

There is no GetType() method, so scripts cannot access memory region types even though the underlying packets provide this information.

Here type: is parsed:

llvm-project/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationClient.cpp

Lines 1638 to 1645 in a87f776

	} else if (name == "type") {
	for (llvm::StringRef entry : llvm::split(value, ',')) {
	if (entry == "stack")
	region_info.SetIsStackMemory(MemoryRegionInfo::eYes);
	else if (entry == "heap")
	region_info.SetIsStackMemory(MemoryRegionInfo::eNo);
	}
	} else if (name == "error") {

Example

(lldb) b main
(lldb) log enable gdb-remote all
(lldb) script
>>> regions = lldb.process.GetMemoryRegions()
# Example packets show type field is present
&lt;  87&gt; read packet: $start:100020000;size:4000;permissions:r;dirty-pages:100020000;type:malloc-metadata;#<!-- -->00
&lt;  88&gt; read packet: $start:100024000;size:4000;permissions:rw;dirty-pages:100024000;type:malloc-metadata;#<!-- -->00

&gt;&gt;&gt; region = lldb.SBMemoryRegionInfo()
&gt;&gt;&gt; regions.GetMemoryRegionAtIndex(7, region)
True
&gt;&gt;&gt; region
[0x0000000100024000-0x0000000100028000 RW-]
&gt;&gt;&gt; region.GetType()
Traceback (most recent call last):
  File "&lt;console&gt;", line 1, in &lt;module&gt;
AttributeError: 'SBMemoryRegionInfo' object has no attribute 'GetType'

Expected behavior

• SBMemoryRegionInfo should expose the type field from gdb-remote packets

[Rajveer100]

cc @DavidSpickett

[patryk4815]

I added this patch into my lldb (I know it is not proper patch for upstream)

--- a/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationClient.cpp	(revision 87f0227cb60147a26a1eeb4fb06e3b505e9c7261)
+++ b/lldb/source/Plugins/Process/gdb-remote/GDBRemoteCommunicationClient.cpp	(date 1756334485170)
@@ -1641,6 +1641,10 @@
             }
           }
         } else if (name == "type") {
+          if (region_info.GetName() == nullptr) {
+            std::string name = "[" + std::string(value) + "]";
+            region_info.SetName(name.c_str());
+          }
           for (llvm::StringRef entry : llvm::split(value, ',')) {
             if (entry == "stack")
               region_info.SetIsStackMemory(MemoryRegionInfo::eYes);

And then in pwndbg-lldb, we can display nice memory mapping with names (types):

- we display `[anon_*]` when there is empty name/type in memory info

[DavidSpickett]

We use flags and type to set various parts of the MemoryRegionInfo:

  OptionalBool m_flash = eDontKnow;
  lldb::offset_t m_blocksize = 0;
  OptionalBool m_memory_tagged = eDontKnow;
  OptionalBool m_is_stack_memory = eDontKnow;
  OptionalBool m_is_shadow_stack = eDontKnow;

So getting all the type related stuff into one GetType() is awkward. We can be literal and say it's anything returned in "type" in the packet but then you're basically just implementing IsStackMemory.

Part of the reason this is awkward as is is that each of these region attribute things gets added for a specific use case rather than for a client that might want to show everything. So we have all these IsSpecificThing.

I won't have time to work on it myself, but adding IsStackMemory to SBMemoryRegionInfo is what I recommend that someone does.

(in fact, you can construct an SBMemoryRegionInfo and control whether it's stack memory, but not read it back)

If someone is more ambitious, we can talk about a general getter for these sorts of things. We need to see examples of how a client would use them though, to get the design right. E.g. some GetAttributes() that returns a bitmask enum and we let the clients figure out how to format what they find.

[DavidSpickett]

Also a hypothetical IsStackMemory will have to either be a yes/know/don't know return type, or be very clearly documented so that clients don't mark every memory region as heap if the method returns False.

(the internal type is a tri-state bool, but I don't know if we've ever returned one in the SBAPI)

Bit of a weird API in that sense. Programs generally have a stack and a heap then a bunch of other stuff. The other stuff is not stack or heap, so it couldn't be true or false. 🤔

Applies to the other attributes too. So we can't just return an array of things that are enabled for that region, sometimes we don't know. We could convert don't know into false and document that clients should only print things if it's present. Like lldb prints that a memory region is tagged if it is tagged, but does not print anything extra if it is not.

[patryk4815]

@DavidSpickett Or debugserver should implement name: attribute. In pwndbg we just want to display good memory page name for user

[DavidSpickett]

Surprised that it doesn't. Please open an issue requesting that and I'll tag the people who work on debugserver.