Add linked createdump support for NativeAOT executables - #132572
Draft
agocke wants to merge 4 commits into
Draft
Conversation
Implement a minimal ELF core dump writer in pure C for NativeAOT. This provides crash dump generation without depending on the C++ runtime, which NativeAOT cannot use. The implementation reads process state via /proc and ptrace, then writes a standard ELF core dump containing: - NT_PRPSINFO, NT_AUXV, NT_FILE notes - Per-thread NT_PRSTATUS, NT_FPREGSET, NT_SIGINFO notes - All readable memory regions (PT_LOAD segments) Supports x86_64 and aarch64. Builds as enabled/disabled static library variants following the established NativeAOT pattern (EventPipe, VxSort). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Integrate the linked-in createdump library into the NativeAOT build and runtime startup: - Bootstrap main.cpp: Check for GUID sentinel argument before runtime initialization. When detected, redirect into the dump writer instead of normal startup. - PalCreateDump.cpp: When the enabled library is linked, use self-restart mode (fork + re-exec self with sentinel) instead of looking for an external createdump binary. Falls back to external binary if self-path resolution fails. - MSBuild: Select enabled/disabled variant via CreatedumpSupportName property, tied to EventSourceSupport on Linux. - SDK packaging: Register both library variants in the platform manifest. - Smoke test: Self-invoking test that crashes a child process and validates the generated ELF core dump. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Add an optimized dump mode that excludes shared library code/rodata, reducing dump size from ~75MB to ~32MB for typical NativeAOT apps while preserving full GDB debugging capability. The new GetDumpSize() function determines per-region inclusion: - Full mode (--full / DbgMiniDumpType=4): all readable memory - Heap mode (default): writable memory, anonymous regions, main executable (including RELRO pages with r_debug for GDB shared library resolution), and the first page of each shared library ELF header. Shared library .text/.rodata is excluded since debuggers load it from disk via NT_FILE. Update the smoke test to exercise the heap dump path (type 2) and verify the linked-in createdump path is used via the [createdump] stderr marker. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Limit linked createdump dispatch to explicitly opted-in Linux NativeAOT executables while preserving external-createdump behavior for unsupported requests and output shapes. Correct ELF capture, mapping selection, error handling, and security checks, and strengthen end-to-end validation. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 49628d5c-675b-40b0-96d5-9ea25a1ea213
Contributor
|
Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib |
Contributor
There was a problem hiding this comment.
Pull request overview
Adds “linked-in createdump” support for Linux NativeAOT executables by introducing a minimal in-process (self re-exec) createdump implementation, wiring it into the NativeAOT bootstrap / PalCreateDump dispatch, and validating it with a new NativeAOT smoke test.
Changes:
- Add a new
nativeaot-createdump-enabledstatic library (Linux, supported arch) containing a minimal/proc+ptracereader and ELF core writer, plus a disabled stub variant. - Update NativeAOT bootstrap +
PalCreateDumpto support a self-reexec dispatch using a GUID sentinel and/proc/self/exe(while retaining external-helper fallback paths). - Add a Linux NativeAOT smoke test and integrate the new libraries into build/packaging plumbing.
Show a summary per file
| File | Description |
|---|---|
| src/tests/nativeaot/SmokeTests/CreatedumpLinkedIn/CreatedumpLinkedIn.csproj | New Linux NativeAOT smoke-test project enabling CreatedumpSupport. |
| src/tests/nativeaot/SmokeTests/CreatedumpLinkedIn/CreatedumpLinkedIn.cs | End-to-end validation for linked dump generation + external-helper fallback behavior. |
| src/installer/pkg/sfx/Microsoft.NETCore.App/Directory.Build.props | Adds NativeAOT createdump static libs to the platform manifest entries. |
| src/coreclr/nativeaot/Runtime/unix/PalCreateDump.cpp | Adds self-reexec mode selection and sentinel insertion for linked createdump dispatch. |
| src/coreclr/nativeaot/Runtime/createdump/process_reader.h | New minimal process snapshot types/APIs for dump generation. |
| src/coreclr/nativeaot/Runtime/createdump/process_reader.c | Implements /proc parsing + ptrace thread attach/register capture. |
| src/coreclr/nativeaot/Runtime/createdump/nativeaot_createdump_main.c | New linked createdump entrypoint invoked via sentinel re-exec. |
| src/coreclr/nativeaot/Runtime/createdump/elf_dump_writer.h | Declares minimal ELF core dump writer API. |
| src/coreclr/nativeaot/Runtime/createdump/elf_dump_writer.c | Implements ELF core file construction, notes, and filtered memory emission. |
| src/coreclr/nativeaot/Runtime/createdump/createdump_sentinel.h | Defines the GUID sentinel argument for createdump self-reexec dispatch. |
| src/coreclr/nativeaot/Runtime/createdump/createdump_disabled.c | Disabled stub implementation for unsupported/disabled configurations. |
| src/coreclr/nativeaot/Runtime/createdump/CMakeLists.txt | Adds build/install rules for enabled/disabled createdump libraries. |
| src/coreclr/nativeaot/Runtime/CMakeLists.txt | Adds the createdump subdirectory to the NativeAOT runtime build (non-Windows). |
| src/coreclr/nativeaot/BuildIntegration/Microsoft.NETCore.Native.Unix.targets | Adds CreatedumpSupportName selection and links the chosen library into NativeAOT Unix builds. |
| src/coreclr/nativeaot/BuildIntegration/Microsoft.NETCore.Native.targets | Introduces CreatedumpSupport MSBuild property defaulting to false. |
| src/coreclr/nativeaot/Bootstrap/main.cpp | Adds sentinel detection/dispatch to nativeaot_createdump_main before runtime initialization. |
Review details
Suppressed comments (2)
src/coreclr/nativeaot/Runtime/createdump/elf_dump_writer.c:187
CalculateNotesSizeaccumulates note sizes with unchecked arithmetic (includingauxv.count * sizeof(Elf64_auxv_t)and repeatedtotal += ...). If this overflows,notesSizecan become too small and the file layout computations inWriteElfCoreDumpcan produce overlapping sections.
// NT_PRPSINFO
total += NoteSize(CORE_NOTE_NAME_SIZE, sizeof(struct elf_prpsinfo));
// NT_AUXV
total += NoteSize(CORE_NOTE_NAME_SIZE, info->auxv.count * sizeof(Elf64_auxv_t));
src/coreclr/nativeaot/Runtime/createdump/elf_dump_writer.c:199
- Per-thread note size accumulation in
CalculateNotesSizealso needs overflow checks. Without them, a process with many threads could overflowtotaland causenotesSizeto under-report.
// NT_PRSTATUS
total += NoteSize(CORE_NOTE_NAME_SIZE, sizeof(struct elf_prstatus));
// NT_FPREGSET
total += NoteSize(CORE_NOTE_NAME_SIZE, sizeof(fp_regs_t));
- Files reviewed: 16/16 changed files
- Comments generated: 4
- Review effort level: Lite
Comment on lines
13
to
14
| #include <unistd.h> | ||
| #define __STDC_FORMAT_MACROS | ||
| #include <inttypes.h> |
Comment on lines
+156
to
+176
| static size_t NtFileDataSize(const ProcessInfo* info) | ||
| { | ||
| // Header: count (8 bytes) + page_size (8 bytes) | ||
| size_t size = 2 * sizeof(uint64_t); | ||
|
|
||
| // Per-file entry: start (8) + end (8) + offset (8) | ||
| size_t fileCount = CountFileRegions(info); | ||
| size += fileCount * 3 * sizeof(uint64_t); | ||
|
|
||
| // File name strings (null-terminated) | ||
| for (size_t i = 0; i < info->regions.count; i++) | ||
| { | ||
| if (info->regions.items[i].fileName[0] != '\0' && | ||
| info->regions.items[i].fileName[0] != '[') | ||
| { | ||
| size += strlen(info->regions.items[i].fileName) + 1; | ||
| } | ||
| } | ||
|
|
||
| return size; | ||
| } |
|
|
||
| # Enabled variant: full ELF core dump writer (Linux only). | ||
| # macOS support requires Mach-O dump format and Mach APIs — deferred to future work. | ||
| if(CLR_CMAKE_TARGET_LINUX) |
|
|
||
| <CreatedumpSupportName>libnativeaot-createdump-disabled</CreatedumpSupportName> | ||
| <!-- Linked createdump requires the standard executable bootstrap to handle self-restart. --> | ||
| <CreatedumpSupportName Condition="'$(CreatedumpSupport)' == 'true' and '$(_targetOS)' == 'linux' and '$(IsNativeExecutable)' == 'true' and '$(CustomNativeMain)' != 'true'">libnativeaot-createdump-enabled</CreatedumpSupportName> |
Member
|
cc @dotnet/dotnet-diag-contrib |
Member
Author
|
FYI this is PR is an experiment -- it's nowhere near ready to be merged and haven't looked at all the code. @eduardo-vp this is the PR that I mentioned I was experimenting with I think the work to shrink NAOT dumps is one of the most interesting parts and could even make it into createdump |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR is an experiment I've been working on, only sharing as draft so other people can look at it
Summary
Linked support is enabled with
CreatedumpSupport=trueand is intentionally scoped to standard Linux NativeAOT executables. Native libraries and custom native mains remain on the existing external-createdump path.Validation
CreatedumpLinkedInNativeAOT smoke test through the NativeAOT launcherPT_LOAD,NT_FILE,NT_PRSTATUS, andNT_SIGINFOcontents and runtime page-size metadataAn end-to-end seven-run comparison against the original proposal reduced median dump time from 443.90 ms to 427.37 ms and dump size from 38.95 MiB to 31.44 MiB.
Full-tree validation with the current SDK is blocked by pre-existing incompatibilities on this older branch, including NativeAOT task/target skew and runtime-pack manifest churn; targeted builds and tests pass.