daniel/MontaukOS
1
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects Releases Wiki Activity

fix: stream ELF program loading instead of buffering whole files

Browse Source 
ElfLoad read the entire binary into one kernel heap allocation before
copying segments out. The heap grows through ReallocConsecutive, so a
40 MB cc1plus required 10k physically contiguous pages - effectively
impossible after boot with a 368 MB ramdisk module resident, and the
failed spawn surfaced as posix_spawnp ENOENT in the gcc driver.

The loader now reads the ELF header and program header table (bounded
at 64 entries), then copies each PT_LOAD page and the PT_TLS template
directly from the VFS into freshly mapped process pages. Peak kernel
memory per load drops from fileSize to one page regardless of binary
size. Boot-smoke verified: all userspace loads through this path.

Co-Authored-By: Claude Fable 5 <[email protected]>
This commit is contained in:
daniel
2026-07-17 10:25:30 +02:00
co-authored by Claude Fable 5
parent 78ce551688
commit 2f1010b02e
2 changed files with 60 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files
kernel/src/Api/BuildNo.hpp
+1 -1
View File
@@ -12,4 +12,4 @@
#pragma once
#define MONTAUK_BUILD_NUMBER 22
#define MONTAUK_BUILD_NUMBER 23
kernel/src/Sched/ElfLoader.cpp
+59 -53
View File
@@ -433,33 +433,58 @@ namespace Sched {
return 0;
}
// Read entire file into a heap buffer
uint8_t* fileData = (uint8_t*)Memory::g_heap->Request(fileSize);
if (fileData == nullptr) {
Kt::KernelLogStream(Kt::ERROR, "ELF") << "Failed to allocate " << fileSize << " bytes for file";
// Stream the load: headers first, then each segment page copied
// straight from the file into freshly mapped process pages. The
// previous whole-file heap buffer needed fileSize bytes of
// physically contiguous kernel heap, which a 40 MB compiler
// binary cannot rely on after boot.
Elf64Header hdr{};
if (Fs::Vfs::ReadBackendFile(file, (uint8_t*)&hdr, 0, sizeof(hdr))
!= (int)sizeof(hdr) || !ValidateElfHeader(&hdr)) {
Fs::Vfs::CloseBackendFile(file);
return 0;
}
Fs::Vfs::ReadBackendFile(file, fileData, 0, fileSize);
Fs::Vfs::CloseBackendFile(file);
// Prevent the optimizer from reordering the VfsRead store past the
// header validation reads that follow.
asm volatile("" ::: "memory");
// Validate ELF header
Elf64Header* hdr = (Elf64Header*)fileData;
if (!ValidateElfHeader(hdr)) {
Memory::g_heap->Free(fileData);
if (hdr.e_phnum == 0 || hdr.e_phnum > 64 ||
hdr.e_phentsize != sizeof(Elf64ProgramHeader)) {
Kt::KernelLogStream(Kt::ERROR, "ELF")
<< "Unsupported program header table (" << (uint64_t)hdr.e_phnum
<< " entries of " << (uint64_t)hdr.e_phentsize << " bytes)";
Fs::Vfs::CloseBackendFile(file);
return 0;
}
// Process program headers
Elf64ProgramHeader phdrs[64];
if (Fs::Vfs::ReadBackendFile(file, (uint8_t*)phdrs, hdr.e_phoff,
(uint64_t)hdr.e_phnum * sizeof(Elf64ProgramHeader))
!= (int)((uint64_t)hdr.e_phnum * sizeof(Elf64ProgramHeader))) {
Fs::Vfs::CloseBackendFile(file);
return 0;
}
// Copy the file bytes that overlap [pageStart, pageStart+0x1000)
// for a segment whose file image spans [vaddr, vaddr+filesz) at
// file offset fileOff.
auto copyPageFromFile = [&](uint64_t physAddr, uint64_t pageStart,
uint64_t vaddr, uint64_t fileOff,
uint64_t filesz) -> bool {
uint64_t pageEnd = pageStart + 0x1000;
uint64_t dataEnd = vaddr + filesz;
uint64_t copyStart = (pageStart > vaddr) ? pageStart : vaddr;
uint64_t copyEnd = (pageEnd < dataEnd) ? pageEnd : dataEnd;
if (copyStart >= copyEnd) {
return true;
}
uint8_t* dst = (uint8_t*)Memory::HHDM(physAddr) + (copyStart - pageStart);
uint64_t srcOff = fileOff + (copyStart - vaddr);
uint64_t len = copyEnd - copyStart;
return Fs::Vfs::ReadBackendFile(file, dst, srcOff, len) == (int)len;
};
Elf64ProgramHeader* tlsPhdr = nullptr;
uint64_t maxVaddrEnd = 0;
for (uint16_t i = 0; i < hdr->e_phnum; i++) {
Elf64ProgramHeader* phdr = (Elf64ProgramHeader*)(fileData + hdr->e_phoff + i * hdr->e_phentsize);
for (uint16_t i = 0; i < hdr.e_phnum; i++) {
Elf64ProgramHeader* phdr = &phdrs[i];
if (phdr->p_type == PT_TLS && phdr->p_memsz > 0) {
tlsPhdr = phdr;
@@ -487,7 +512,7 @@ namespace Sched {
void* page = Memory::g_pfa->AllocateZeroed();
if (page == nullptr) {
Kt::KernelLogStream(Kt::ERROR, "ELF") << "Out of physical pages";
Memory::g_heap->Free(fileData);
Fs::Vfs::CloseBackendFile(file);
return 0;
}
@@ -497,28 +522,15 @@ namespace Sched {
// Map into the process's PML4 with User bit set
if (!Memory::VMM::Paging::MapUserIn(pml4Phys, physAddr, virtAddr)) {
Kt::KernelLogStream(Kt::ERROR, "ELF") << "Failed to map page";
Memory::g_heap->Free(fileData);
Fs::Vfs::CloseBackendFile(file);
return 0;
}
// Copy file data that overlaps this page (via HHDM)
uint64_t pageStart = virtAddr;
uint64_t pageEnd = virtAddr + 0x1000;
uint64_t segFileStart = phdr->p_vaddr;
uint64_t segFileEnd = phdr->p_vaddr + phdr->p_filesz;
uint64_t copyStart = (pageStart > segFileStart) ? pageStart : segFileStart;
uint64_t copyEnd = (pageEnd < segFileEnd) ? pageEnd : segFileEnd;
if (copyStart < copyEnd) {
uint64_t dstOffset = copyStart - pageStart;
uint64_t srcOffset = copyStart - phdr->p_vaddr + phdr->p_offset;
uint64_t copySize = copyEnd - copyStart;
uint8_t* dst = (uint8_t*)Memory::HHDM(physAddr) + dstOffset;
uint8_t* src = fileData + srcOffset;
memcpy(dst, src, copySize);
if (!copyPageFromFile(physAddr, virtAddr, phdr->p_vaddr,
phdr->p_offset, phdr->p_filesz)) {
Kt::KernelLogStream(Kt::ERROR, "ELF") << "Segment read failed";
Fs::Vfs::CloseBackendFile(file);
return 0;
}
}
}
@@ -550,21 +562,15 @@ namespace Sched {
}
// Copy the .tdata template bytes that land in this page.
uint64_t pageStart = virtAddr;
uint64_t pageEnd = virtAddr + 0x1000;
uint64_t dataStart = base;
uint64_t dataEnd = base + tlsPhdr->p_filesz;
uint64_t copyStart = (pageStart > dataStart) ? pageStart : dataStart;
uint64_t copyEnd = (pageEnd < dataEnd) ? pageEnd : dataEnd;
if (copyStart < copyEnd) {
memcpy((uint8_t*)Memory::HHDM(physAddr) + (copyStart - pageStart),
fileData + tlsPhdr->p_offset + (copyStart - dataStart),
copyEnd - copyStart);
if (!copyPageFromFile(physAddr, virtAddr, base,
tlsPhdr->p_offset, tlsPhdr->p_filesz)) {
ok = false;
break;
}
// Write the self-pointer if TP falls in this page.
if (tp >= pageStart && tp + 8 <= pageEnd) {
*(uint64_t*)((uint8_t*)Memory::HHDM(physAddr) + (tp - pageStart)) = tp;
if (tp >= virtAddr && tp + 8 <= virtAddr + 0x1000) {
*(uint64_t*)((uint8_t*)Memory::HHDM(physAddr) + (tp - virtAddr)) = tp;
}
}
@@ -576,13 +582,13 @@ namespace Sched {
outTls->align = align;
} else {
Kt::KernelLogStream(Kt::ERROR, "ELF")
<< "PT_TLS setup failed (out of pages?), "
<< "PT_TLS setup failed (out of pages?), "
<< "process will run without TLS";
}
}
uint64_t entryPoint = hdr->e_entry;
Memory::g_heap->Free(fileData);
uint64_t entryPoint = hdr.e_entry;
Fs::Vfs::CloseBackendFile(file);
return entryPoint;
}