diff options
author | Itamar <itamar8910@gmail.com> | 2020-10-10 18:17:49 +0300 |
---|---|---|
committer | Andreas Kling <kling@serenityos.org> | 2020-12-14 23:05:53 +0100 |
commit | 07b49573612628cc6388e9fe30083dc9912b6a7b (patch) | |
tree | ba0722a6e8c74c9468a6de42a11b039fdf1ba9b5 /Libraries/LibELF | |
parent | 781aa424a9b34917ad4abb1753d7cf7809dea7be (diff) | |
download | serenity-07b49573612628cc6388e9fe30083dc9912b6a7b.zip |
Loader: Add dynamic loader program
The dynamic loader exists as /usr/lib/Loader.so and is loaded by the
kernel when ET_DYN programs are executed.
The dynamic loader is responsible for loading the dependencies of the
main program, allocating TLS storage, preparing all loaded objects for
execution and finally jumping to the entry of the main program.
Diffstat (limited to 'Libraries/LibELF')
-rw-r--r-- | Libraries/LibELF/DynamicLoader.cpp | 258 | ||||
-rw-r--r-- | Libraries/LibELF/DynamicLoader.h | 45 | ||||
-rw-r--r-- | Libraries/LibELF/DynamicObject.cpp | 36 | ||||
-rw-r--r-- | Libraries/LibELF/DynamicObject.h | 99 | ||||
-rw-r--r-- | Libraries/LibELF/Image.cpp | 5 | ||||
-rw-r--r-- | Libraries/LibELF/Image.h | 3 | ||||
-rw-r--r-- | Libraries/LibELF/exec_elf.h | 18 |
7 files changed, 366 insertions, 98 deletions
diff --git a/Libraries/LibELF/DynamicLoader.cpp b/Libraries/LibELF/DynamicLoader.cpp index 1a4b4a6d11..ad5c486bc0 100644 --- a/Libraries/LibELF/DynamicLoader.cpp +++ b/Libraries/LibELF/DynamicLoader.cpp @@ -1,5 +1,6 @@ /* * Copyright (c) 2019-2020, Andrew Kaster <andrewdkaster@gmail.com> + * Copyright (c) 2020, Itamar S. <itamar8910@gmail.com> * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -38,7 +39,8 @@ #ifndef DYNAMIC_LOAD_DEBUG # define DYNAMIC_LOAD_DEBUG #endif -//#define DYNAMIC_LOAD_VERBOSE + +// #define DYNAMIC_LOAD_VERBOSE #ifdef DYNAMIC_LOAD_VERBOSE # define VERBOSE(fmt, ...) dbgprintf(fmt, ##__VA_ARGS__) @@ -57,31 +59,67 @@ static void* mmap_with_name(void* addr, size_t length, int prot, int flags, int namespace ELF { -static bool s_always_bind_now = false; +static bool s_always_bind_now = true; NonnullRefPtr<DynamicLoader> DynamicLoader::construct(const char* filename, int fd, size_t size) { return adopt(*new DynamicLoader(filename, fd, size)); } +void* DynamicLoader::do_mmap(int fd, size_t size, const String& name) +{ + if (size < sizeof(Elf32_Ehdr)) + return MAP_FAILED; + + String file_mmap_name = String::format("ELF_DYN: %s", name.characters()); + return mmap_with_name(nullptr, size, PROT_READ, MAP_PRIVATE, fd, 0, file_mmap_name.characters()); +} + DynamicLoader::DynamicLoader(const char* filename, int fd, size_t size) : m_filename(filename) , m_file_size(size) , m_image_fd(fd) + , m_file_mapping(do_mmap(m_image_fd, m_file_size, m_filename)) + , m_elf_image((u8*)m_file_mapping, m_file_size) { - if (m_file_size < sizeof(Elf32_Ehdr)) { + + if (m_file_mapping == MAP_FAILED) { m_valid = false; return; } - String file_mmap_name = String::format("ELF_DYN: %s", m_filename.characters()); + m_tls_size = calculate_tls_size(); - m_file_mapping = mmap_with_name(nullptr, m_file_size, PROT_READ, MAP_PRIVATE, m_image_fd, 0, file_mmap_name.characters()); - if (MAP_FAILED == m_file_mapping) { - m_valid = false; - return; - } + m_valid = is_valid(); +} + +RefPtr<DynamicObject> DynamicLoader::dynamic_object_from_image() const +{ + VirtualAddress dynamic_section_address; + m_elf_image.for_each_program_header([&dynamic_section_address](auto program_header) { + if (program_header.type() == PT_DYNAMIC) { + dynamic_section_address = VirtualAddress(program_header.raw_data()); + } + }); + ASSERT(!dynamic_section_address.is_null()); + + return ELF::DynamicObject::construct(VirtualAddress(m_elf_image.base_address()), dynamic_section_address); +} + +size_t DynamicLoader::calculate_tls_size() const +{ + size_t tls_size = 0; + m_elf_image.for_each_program_header([&tls_size](auto program_header) { + if (program_header.type() == PT_TLS) { + tls_size = program_header.size_in_memory(); + } + }); + return tls_size; +} + +bool DynamicLoader::validate() +{ auto* elf_header = (Elf32_Ehdr*)m_file_mapping; if (!validate_elf_header(*elf_header, m_file_size) || !validate_program_headers(*elf_header, m_file_size, (u8*)m_file_mapping, m_file_size, &m_program_interpreter)) { @@ -105,35 +143,35 @@ void* DynamicLoader::symbol_for_name(const char* name) return m_dynamic_object->base_address().offset(symbol.value()).as_ptr(); } -bool DynamicLoader::load_from_image(unsigned flags) +RefPtr<DynamicObject> DynamicLoader::load_from_image(unsigned flags, size_t total_tls_size) { - Image elf_image((u8*)m_file_mapping, m_file_size); - m_valid = elf_image.is_valid() && elf_image.is_dynamic(); + m_valid = m_elf_image.is_valid() && m_elf_image.is_dynamic(); if (!m_valid) { - return false; + return nullptr; } #ifdef DYNAMIC_LOAD_VERBOSE - m_image->dump(); + // m_image->dump(); #endif - load_program_headers(elf_image); - - // Don't need this private mapping anymore - munmap(m_file_mapping, m_file_size); - m_file_mapping = MAP_FAILED; + load_program_headers(); - m_dynamic_object = AK::make<DynamicObject>(m_text_segment_load_address, m_dynamic_section_address); + m_dynamic_object = DynamicObject::construct(m_text_segment_load_address, m_dynamic_section_address); + m_dynamic_object->set_tls_offset(m_tls_offset); + m_dynamic_object->set_tls_size(m_tls_size); - return load_stage_2(flags); + auto rc = load_stage_2(flags, total_tls_size); + if (!rc) + return nullptr; + return m_dynamic_object; } -bool DynamicLoader::load_stage_2(unsigned flags) +bool DynamicLoader::load_stage_2(unsigned flags, size_t total_tls_size) { ASSERT(flags & RTLD_GLOBAL); - ASSERT(flags & RTLD_LAZY); + ASSERT(!(flags & RTLD_LAZY)); #ifdef DYNAMIC_LOAD_DEBUG m_dynamic_object->dump(); @@ -148,8 +186,12 @@ bool DynamicLoader::load_stage_2(unsigned flags) } } - do_relocations(); - setup_plt_trampoline(); + do_relocations(total_tls_size); + + if (flags & RTLD_LAZY) { + ASSERT_NOT_REACHED(); // TODO: Support lazy binding + setup_plt_trampoline(); + } // Clean up our setting of .text to PROT_READ | PROT_WRITE if (m_dynamic_object->has_text_relocations()) { @@ -167,7 +209,7 @@ bool DynamicLoader::load_stage_2(unsigned flags) return true; } -void DynamicLoader::load_program_headers(const Image& elf_image) +void DynamicLoader::load_program_headers() { Vector<ProgramHeaderRegion> program_headers; @@ -176,7 +218,7 @@ void DynamicLoader::load_program_headers(const Image& elf_image) ProgramHeaderRegion* tls_region_ptr = nullptr; VirtualAddress dynamic_region_desired_vaddr; - elf_image.for_each_program_header([&](const Image::ProgramHeader& program_header) { + m_elf_image.for_each_program_header([&](const Image::ProgramHeader& program_header) { ProgramHeaderRegion new_region; new_region.set_program_header(program_header.raw_header()); program_headers.append(move(new_region)); @@ -214,27 +256,19 @@ void DynamicLoader::load_program_headers(const Image& elf_image) VirtualAddress data_segment_actual_addr = region->desired_load_address().offset((FlatPtr)text_segment_begin); memcpy(data_segment_actual_addr.as_ptr(), (u8*)m_file_mapping + region->offset(), region->size_in_image()); - // FIXME: Do some kind of 'allocate TLS section' or some such from a per-application pool - if (tls_region_ptr) { - region = tls_region_ptr; - // FIXME: This can't be right either. TLS needs some real work i'd say :) - m_tls_segment_address = tls_region_ptr->desired_load_address(); - VirtualAddress tls_segment_actual_addr = region->desired_load_address().offset((FlatPtr)text_segment_begin); - memcpy(tls_segment_actual_addr.as_ptr(), (u8*)m_file_mapping + region->offset(), region->size_in_image()); - } + // FIXME: Initialize the values in the TLS section. Currently, it will always be zeroed. } -void DynamicLoader::do_relocations() +void DynamicLoader::do_relocations(size_t total_tls_size) { - FlatPtr load_base_address = m_dynamic_object->base_address().get(); - - // FIXME: We should really bail on undefined symbols here. - auto main_relocation_section = m_dynamic_object->relocation_section(); - - main_relocation_section.for_each_relocation([&](const DynamicObject::Relocation& relocation) { - VERBOSE("====== RELOCATION %d: offset 0x%08X, type %d, symidx %08X\n", relocation.offset_in_section() / main_relocation_section.entry_size(), relocation.offset(), relocation.type(), relocation.symbol_index()); - u32* patch_ptr = (u32*)(load_base_address + relocation.offset()); + main_relocation_section.for_each_relocation([&](ELF::DynamicObject::Relocation relocation) { + VERBOSE("Relocation symbol: %s, type: %d\n", relocation.symbol().name(), relocation.type()); + u32* patch_ptr = (u32*)(m_dynamic_object->base_address().as_ptr() + relocation.offset()); + // VERBOSE("dynamic object name: %s\n", dynamic_object.object_name()); + VERBOSE("dynamic object base address: %p\n", m_dynamic_object->base_address()); + VERBOSE("relocation offset: 0x%x\n", relocation.offset()); + VERBOSE("patch_ptr: %p\n", patch_ptr); switch (relocation.type()) { case R_386_NONE: // Apparently most loaders will just skip these? @@ -244,7 +278,9 @@ void DynamicLoader::do_relocations() case R_386_32: { auto symbol = relocation.symbol(); VERBOSE("Absolute relocation: name: '%s', value: %p\n", symbol.name(), symbol.value()); - u32 symbol_address = symbol.value() + load_base_address; + auto res = lookup_symbol(symbol); + ASSERT(res.found); + u32 symbol_address = res.address; *patch_ptr += symbol_address; VERBOSE(" Symbol address: %p\n", *patch_ptr); break; @@ -252,15 +288,34 @@ void DynamicLoader::do_relocations() case R_386_PC32: { auto symbol = relocation.symbol(); VERBOSE("PC-relative relocation: '%s', value: %p\n", symbol.name(), symbol.value()); - u32 relative_offset = (symbol.value() - relocation.offset()); + auto res = lookup_symbol(symbol); + ASSERT(res.found); + u32 relative_offset = (res.address - (FlatPtr)(m_dynamic_object->base_address().as_ptr() + relocation.offset())); *patch_ptr += relative_offset; VERBOSE(" Symbol address: %p\n", *patch_ptr); break; } case R_386_GLOB_DAT: { auto symbol = relocation.symbol(); + if (!strcmp(symbol.name(), "__deregister_frame_info") || !strcmp(symbol.name(), "_ITM_registerTMCloneTable") + || !strcmp(symbol.name(), "_ITM_deregisterTMCloneTable") || !strcmp(symbol.name(), "__register_frame_info")) { + // We do not support these + break; + } VERBOSE("Global data relocation: '%s', value: %p\n", symbol.name(), symbol.value()); - u32 symbol_location = load_base_address + symbol.value(); + auto res = lookup_symbol(symbol); + VERBOSE("was symbol found? %d, address: 0x%x\n", res.found, res.address); + + if (!res.found) { + // TODO this is a hack + ASSERT(!strcmp(symbol.name(), "__deregister_frame_info") || !strcmp(symbol.name(), "_ITM_registerTMCloneTable") + || !strcmp(symbol.name(), "_ITM_deregisterTMCloneTable") || !strcmp(symbol.name(), "__register_frame_info")); + ASSERT_NOT_REACHED(); + return IterationDecision::Continue; + } + // ASSERT(res.found); + u32 symbol_location = res.address; + ASSERT(symbol_location != (FlatPtr)m_dynamic_object->base_address().as_ptr()); *patch_ptr = symbol_location; VERBOSE(" Symbol address: %p\n", *patch_ptr); break; @@ -270,33 +325,58 @@ void DynamicLoader::do_relocations() // We could explicitly do them first using m_number_of_relocatoins from DT_RELCOUNT // However, our compiler is nice enough to put them at the front of the relocations for us :) VERBOSE("Load address relocation at offset %X\n", relocation.offset()); - VERBOSE(" patch ptr == %p, adding load base address (%p) to it and storing %p\n", *patch_ptr, load_base_address, *patch_ptr + load_base_address); - *patch_ptr += load_base_address; // + addend for RelA (addend for Rel is stored at addr) + VERBOSE(" patch ptr == %p, adding load base address (%p) to it and storing %p\n", *patch_ptr, m_dynamic_object->base_address().as_ptr(), *patch_ptr + m_dynamic_object->base_address().as_ptr()); + *patch_ptr += (FlatPtr)m_dynamic_object->base_address().as_ptr(); // + addend for RelA (addend for Rel is stored at addr) break; } + case R_386_TLS_TPOFF32: case R_386_TLS_TPOFF: { - VERBOSE("Relocation type: R_386_TLS_TPOFF at offset %X\n", relocation.offset()); - // FIXME: this can't be right? I have no idea what "negative offset into TLS storage" means... - // FIXME: Check m_has_static_tls and do something different for dynamic TLS - *patch_ptr = relocation.offset() - (FlatPtr)m_tls_segment_address.as_ptr() - *patch_ptr; + dbgprintf("Relocation type: R_386_TLS_TPOFF at offset %X\n", relocation.offset()); + auto symbol = relocation.symbol(); + // For some reason, LibC has a R_386_TLS_TPOFF that referes to the undefined symbol.. huh + if (relocation.symbol_index() == 0) + break; + dbgprintf("Symbol index: %d\n", symbol.index()); + dbgprintf("Symbol is_undefined?: %d\n", symbol.is_undefined()); + dbgprintf("TLS relocation: '%s', value: %p\n", symbol.name(), symbol.value()); + auto res = lookup_symbol(symbol); + if (!res.found) + break; + ASSERT(res.found); + u32 symbol_value = res.value; + dbgprintf("symbol value: %d\n", symbol_value); + const auto dynamic_object_of_symbol = res.dynamic_object; + ASSERT(dynamic_object_of_symbol); + size_t offset_of_tls_end = dynamic_object_of_symbol->tls_offset().value() + dynamic_object_of_symbol->tls_size().value(); + // size_t offset_of_tls_end = tls_offset() + tls_size(); + dbgprintf("patch ptr: 0x%x\n", patch_ptr); + dbgprintf("tls end offset: %d, total tls size: %d\n", offset_of_tls_end, total_tls_size); + *patch_ptr = (offset_of_tls_end - total_tls_size - symbol_value - sizeof(Elf32_Addr)); + dbgprintf("*patch ptr: %d\n", (i32)*patch_ptr); break; } default: // Raise the alarm! Someone needs to implement this relocation type - dbgprintf("Found a new exciting relocation type %d\n", relocation.type()); - printf("DynamicLoader: Found unknown relocation type %d\n", relocation.type()); + VERBOSE("Found a new exciting relocation type %d\n", relocation.type()); + // printf("DynamicLoader: Found unknown relocation type %d\n", relocation.type()); ASSERT_NOT_REACHED(); break; } return IterationDecision::Continue; }); + VERBOSE("plt relocations: 0x%x", m_dynamic_object->plt_relocation_section().address()); + VERBOSE("plt relocation count: 0x%x", m_dynamic_object->plt_relocation_section().address()); + VERBOSE("plt size: %d\n", m_dynamic_object->plt_relocation_section().size()); + VERBOSE("plt entry size: 0x%x\n", m_dynamic_object->plt_relocation_section().entry_size()); + // Handle PLT Global offset table relocations. m_dynamic_object->plt_relocation_section().for_each_relocation([&](const DynamicObject::Relocation& relocation) { // FIXME: Or BIND_NOW flag passed in? if (m_dynamic_object->must_bind_now() || s_always_bind_now) { // Eagerly BIND_NOW the PLT entries, doing all the symbol looking goodness // The patch method returns the address for the LAZY fixup path, but we don't need it here + VERBOSE("patching plt reloaction: 0x%x", relocation.offset_in_section()); (void)patch_plt_entry(relocation.offset_in_section()); } else { // LAZY-ily bind the PLT slots by just adding the base address to the offsets stored there @@ -305,7 +385,7 @@ void DynamicLoader::do_relocations() u8* relocation_address = relocation.address().as_ptr(); - *(u32*)relocation_address += load_base_address; + *(u32*)relocation_address += (FlatPtr)m_dynamic_object->base_address().as_ptr(); } return IterationDecision::Continue; }); @@ -349,7 +429,9 @@ Elf32_Addr DynamicLoader::patch_plt_entry(u32 relocation_offset) auto sym = relocation.symbol(); u8* relocation_address = relocation.address().as_ptr(); - u32 symbol_location = sym.address().get(); + auto res = lookup_symbol(sym); + ASSERT(res.found); + u32 symbol_location = res.address; VERBOSE("DynamicLoader: Jump slot relocation: putting %s (%p) into PLT at %p\n", sym.name(), symbol_location, relocation_address); @@ -360,28 +442,34 @@ Elf32_Addr DynamicLoader::patch_plt_entry(u32 relocation_offset) void DynamicLoader::call_object_init_functions() { + dbg() << "inside call_object_init_functions of " << m_filename; typedef void (*InitFunc)(); - auto init_function = (InitFunc)(m_dynamic_object->init_section().address().as_ptr()); -#ifdef DYNAMIC_LOAD_DEBUG - dbgprintf("Calling DT_INIT at %p\n", init_function); -#endif - (init_function)(); + if (m_dynamic_object->has_init_section()) { + auto init_function = (InitFunc)(m_dynamic_object->init_section().address().as_ptr()); - auto init_array_section = m_dynamic_object->init_array_section(); + // #ifdef DYNAMIC_LOAD_DEBUG + dbgprintf("Calling DT_INIT at %p\n", init_function); + // #endif + (init_function)(); + } - InitFunc* init_begin = (InitFunc*)(init_array_section.address().as_ptr()); - InitFunc* init_end = init_begin + init_array_section.entry_count(); - while (init_begin != init_end) { - // Android sources claim that these can be -1, to be ignored. - // 0 definitely shows up. Apparently 0/-1 are valid? Confusing. - if (!*init_begin || ((FlatPtr)*init_begin == (FlatPtr)-1)) - continue; -#ifdef DYNAMIC_LOAD_DEBUG - dbgprintf("Calling DT_INITARRAY entry at %p\n", *init_begin); -#endif - (*init_begin)(); - ++init_begin; + if (m_dynamic_object->has_init_array_section()) { + auto init_array_section = m_dynamic_object->init_array_section(); + + InitFunc* init_begin = (InitFunc*)(init_array_section.address().as_ptr()); + InitFunc* init_end = init_begin + init_array_section.entry_count(); + while (init_begin != init_end) { + // Android sources claim that these can be -1, to be ignored. + // 0 definitely shows up. Apparently 0/-1 are valid? Confusing. + if (!*init_begin || ((FlatPtr)*init_begin == (FlatPtr)-1)) + continue; + // #ifdef DYNAMIC_LOAD_DEBUG + dbgprintf("Calling DT_INITARRAY entry at %p\n", *init_begin); + // #endif + (*init_begin)(); + ++init_begin; + } } } @@ -394,4 +482,24 @@ u32 DynamicLoader::ProgramHeaderRegion::mmap_prot() const return prot; } +DynamicObject::SymbolLookupResult DynamicLoader::lookup_symbol(const ELF::DynamicObject::Symbol& symbol) const +{ + VERBOSE("looking up symbol: %s\n", symbol.name()); + if (!symbol.is_undefined()) { + VERBOSE("symbol is defiend in its object\n"); + return { true, symbol.value(), (FlatPtr)symbol.address().as_ptr(), &symbol.object() }; + } + ASSERT(m_global_symbol_lookup_func); + return m_global_symbol_lookup_func(symbol.name()); +} + +// Optional<DynamicLoader::SymbolLookupResult> DynamicLoader::lookup_symbol(const char* name) const +// { +// ASSERT(m_dynamic_object); +// auto res = m_dynamic_object->hash_section().lookup_symbol(name); +// if (res.is_undefined()) +// return {}; +// return SymbolLookupResult { true, res.value(), (FlatPtr)res.address().as_ptr(), m_dynamic_object.ptr() }; +// } + } // end namespace ELF diff --git a/Libraries/LibELF/DynamicLoader.h b/Libraries/LibELF/DynamicLoader.h index 7bb2f396eb..e812ab0d35 100644 --- a/Libraries/LibELF/DynamicLoader.h +++ b/Libraries/LibELF/DynamicLoader.h @@ -1,5 +1,6 @@ /* * Copyright (c) 2019-2020, Andrew Kaster <andrewdkaster@gmail.com> + * Copyright (c) 2020, Itamar S. <itamar8910@gmail.com> * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -49,12 +50,12 @@ public: // Load a full ELF image from file into the current process and create an DynamicObject // from the SHT_DYNAMIC in the file. - bool load_from_image(unsigned flags); + RefPtr<DynamicObject> load_from_image(unsigned flags, size_t total_tls_size); // Stage 2 of loading: relocations and init functions // Assumes that the program headers have been loaded and that m_dynamic_object is initialized // Splitting loading like this allows us to use the same code to relocate a main executable as an elf binary - bool load_stage_2(unsigned flags); + bool load_stage_2(unsigned flags, size_t total_tls_size); // Intended for use by dlsym or other internal methods void* symbol_for_name(const char*); @@ -67,6 +68,21 @@ public: // Requested program interpreter from program headers. May be empty string StringView program_interpreter() const { return m_program_interpreter; } + VirtualAddress text_segment_load_addresss() const { return m_text_segment_load_address; } + + void set_tls_offset(size_t offset) { m_tls_offset = offset; }; + size_t tls_size() const { return m_tls_size; } + size_t tls_offset() const { return m_tls_offset; } + const ELF::Image& image() const { return m_elf_image; } + + template<typename F> + void for_each_needed_library(F) const; + + using SymbolLookupFunction = DynamicObject::SymbolLookupResult (*)(const char*); + SymbolLookupFunction m_global_symbol_lookup_func { nullptr }; + + VirtualAddress text_segment_load_address() const { return m_text_segment_load_address; } + private: class ProgramHeaderRegion { public: @@ -94,31 +110,48 @@ private: Elf32_Phdr m_program_header; // Explicitly a copy of the PHDR in the image }; + static void* do_mmap(int fd, size_t size, const String& name); + RefPtr<DynamicObject> dynamic_object_from_image() const; + explicit DynamicLoader(const char* filename, int fd, size_t file_size); - explicit DynamicLoader(Elf32_Dyn* dynamic_location, Elf32_Addr load_address); // Stage 1 - void load_program_headers(const Image& elf_image); + void load_program_headers(); // Stage 2 - void do_relocations(); + void do_relocations(size_t total_tls_size); void setup_plt_trampoline(); void call_object_init_functions(); + bool validate(); + size_t calculate_tls_size() const; + + DynamicObject::SymbolLookupResult lookup_symbol(const ELF::DynamicObject::Symbol& symbol) const; + String m_filename; String m_program_interpreter; size_t m_file_size { 0 }; int m_image_fd { -1 }; void* m_file_mapping { MAP_FAILED }; + ELF::Image m_elf_image; bool m_valid { true }; - OwnPtr<DynamicObject> m_dynamic_object; + RefPtr<DynamicObject> m_dynamic_object; VirtualAddress m_text_segment_load_address; size_t m_text_segment_size { 0 }; VirtualAddress m_tls_segment_address; VirtualAddress m_dynamic_section_address; + + size_t m_tls_offset { 0 }; + size_t m_tls_size { 0 }; }; +template<typename F> +void DynamicLoader::for_each_needed_library(F func) const +{ + dynamic_object_from_image()->for_each_needed_library(move(func)); +} + } // end namespace ELF diff --git a/Libraries/LibELF/DynamicObject.cpp b/Libraries/LibELF/DynamicObject.cpp index 9f104981cf..bae2dd5024 100644 --- a/Libraries/LibELF/DynamicObject.cpp +++ b/Libraries/LibELF/DynamicObject.cpp @@ -1,5 +1,6 @@ /* * Copyright (c) 2019-2020, Andrew Kaster <andrewdkaster@gmail.com> + * Copyright (c) 2020, Itamar S. <itamar8910@gmail.com> * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -142,6 +143,13 @@ void DynamicObject::parse() m_soname_index = entry.val(); m_has_soname = true; break; + case DT_DEBUG: + break; + case DT_FLAGS_1: + break; + case DT_NEEDED: + // We handle these in for_each_needed_library + break; default: dbgprintf("DynamicObject: DYNAMIC tag handling not implemented for DT_%s\n", name_for_dtag(entry.tag())); printf("DynamicObject: DYNAMIC tag handling not implemented for DT_%s\n", name_for_dtag(entry.tag())); @@ -151,6 +159,13 @@ void DynamicObject::parse() return IterationDecision::Continue; }); + if (!m_size_of_relocation_entry) { + // TODO: FIXME, this shouldn't be hardcoded + // The reason we need this here is the for some reason, when there only PLT relocations, the compiler + // doesn't insert a 'PLTRELSZ' entry to the dynamic section + m_size_of_relocation_entry = sizeof(Elf32_Rel); + } + auto hash_section_address = hash_section().address().as_ptr(); auto num_hash_chains = ((u32*)hash_section_address)[1]; m_symbol_count = num_hash_chains; @@ -268,7 +283,7 @@ const DynamicObject::Symbol DynamicObject::HashSection::lookup_symbol(const char return symbol; } } - return m_dynamic.the_undefined_symbol(); + return Symbol::create_undefined(m_dynamic); } const char* DynamicObject::symbol_string_table_string(Elf32_Word index) const @@ -276,6 +291,12 @@ const char* DynamicObject::symbol_string_table_string(Elf32_Word index) const return (const char*)base_address().offset(m_string_table_offset + index).as_ptr(); } +DynamicObject::InitializationFunction DynamicObject::init_section_function() const +{ + ASSERT(has_init_section()); + return (InitializationFunction)init_section().address().as_ptr(); +} + static const char* name_for_dtag(Elf32_Sword d_tag) { switch (d_tag) { @@ -368,4 +389,17 @@ static const char* name_for_dtag(Elf32_Sword d_tag) } } +Optional<DynamicObject::SymbolLookupResult> DynamicObject::lookup_symbol(const char* name) const +{ + auto res = hash_section().lookup_symbol(name); + if (res.is_undefined()) + return {}; + return SymbolLookupResult { true, res.value(), (FlatPtr)res.address().as_ptr(), this }; +} + +NonnullRefPtr<DynamicObject> DynamicObject::construct(VirtualAddress base_address, VirtualAddress dynamic_section_address) +{ + return adopt(*new DynamicObject(base_address, dynamic_section_address)); +} + } // end namespace ELF diff --git a/Libraries/LibELF/DynamicObject.h b/Libraries/LibELF/DynamicObject.h index 0989c7d3ec..5541f1fa3e 100644 --- a/Libraries/LibELF/DynamicObject.h +++ b/Libraries/LibELF/DynamicObject.h @@ -1,5 +1,6 @@ /* * Copyright (c) 2019-2020, Andrew Kaster <andrewdkaster@gmail.com> + * Copyright (c) 2020, Itamar S. <itamar8910@gmail.com> * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -27,14 +28,16 @@ #pragma once #include <AK/Assertions.h> +#include <AK/RefCounted.h> #include <Kernel/VirtualAddress.h> #include <LibELF/exec_elf.h> namespace ELF { -class DynamicObject { +class DynamicObject : public RefCounted<DynamicObject> { public: - explicit DynamicObject(VirtualAddress base_address, VirtualAddress dynamic_section_address); + static NonnullRefPtr<DynamicObject> construct(VirtualAddress base_address, VirtualAddress dynamic_section_address); + ~DynamicObject(); void dump() const; @@ -69,6 +72,23 @@ public: , m_sym(sym) , m_index(index) { + if (section_index() == 0) + m_is_undefined = true; + } + + Symbol(const Symbol& other) + : m_dynamic(other.m_dynamic) + , m_sym(other.m_sym) + , m_index(other.m_index) + , m_is_undefined(other.m_is_undefined) + { + } + + static Symbol create_undefined(const DynamicObject& dynamic) + { + auto s = Symbol(dynamic, 0, {}); + s.m_is_undefined = true; + return s; } ~Symbol() { } @@ -80,13 +100,19 @@ public: unsigned index() const { return m_index; } unsigned type() const { return ELF32_ST_TYPE(m_sym.st_info); } unsigned bind() const { return ELF32_ST_BIND(m_sym.st_info); } - bool is_undefined() const { return this == &m_dynamic.the_undefined_symbol(); } + + bool is_undefined() const + { + return m_is_undefined; + } VirtualAddress address() const { return m_dynamic.base_address().offset(value()); } + const DynamicObject& object() const { return m_dynamic; } private: const DynamicObject& m_dynamic; const Elf32_Sym& m_sym; const unsigned m_index; + bool m_is_undefined { false }; }; class Section { @@ -105,7 +131,10 @@ public: unsigned offset() const { return m_section_offset; } unsigned size() const { return m_section_size_bytes; } unsigned entry_size() const { return m_entry_size; } - unsigned entry_count() const { return !entry_size() ? 0 : size() / entry_size(); } + unsigned entry_count() const + { + return !entry_size() ? 0 : size() / entry_size(); + } VirtualAddress address() const { return m_dynamic.base_address().offset(m_section_offset); } protected: @@ -191,9 +220,13 @@ public: unsigned symbol_count() const { return m_symbol_count; } const Symbol symbol(unsigned) const; - const Symbol& the_undefined_symbol() const { return m_the_undefined_symbol; } + typedef void (*InitializationFunction)(); + + bool has_init_section() const { return m_init_offset != 0; } + bool has_init_array_section() const { return m_init_array_offset != 0; } const Section init_section() const; + InitializationFunction init_section_function() const; const Section fini_section() const; const Section init_array_section() const; const Section fini_array_section() const; @@ -215,7 +248,28 @@ public: const char* soname() const { return m_has_soname ? symbol_string_table_string(m_soname_index) : nullptr; } + Optional<FlatPtr> tls_offset() const { return m_tls_offset; } + Optional<FlatPtr> tls_size() const { return m_tls_size; } + void set_tls_offset(FlatPtr offset) { m_tls_offset = offset; } + void set_tls_size(FlatPtr size) { m_tls_size = size; } + + template<typename F> + void for_each_needed_library(F) const; + + template<typename F> + void for_each_initialization_array_function(F f) const; + + struct SymbolLookupResult { + bool found { false }; + FlatPtr value { 0 }; + FlatPtr address { 0 }; + const ELF::DynamicObject* dynamic_object { nullptr }; // The object in which the symbol is defined + }; + Optional<SymbolLookupResult> lookup_symbol(const char* name) const; + private: + explicit DynamicObject(VirtualAddress base_address, VirtualAddress dynamic_section_address); + const char* symbol_string_table_string(Elf32_Word) const; void parse(); @@ -227,7 +281,6 @@ private: VirtualAddress m_base_address; VirtualAddress m_dynamic_address; - Symbol m_the_undefined_symbol { *this, 0, {} }; unsigned m_symbol_count { 0 }; @@ -264,6 +317,9 @@ private: bool m_has_soname { false }; Elf32_Word m_soname_index { 0 }; // Index into dynstr table for SONAME + + Optional<FlatPtr> m_tls_offset; + Optional<FlatPtr> m_tls_size; // End Section information from DT_* entries }; @@ -271,7 +327,10 @@ template<typename F> inline void DynamicObject::RelocationSection::for_each_relocation(F func) const { for (unsigned i = 0; i < relocation_count(); ++i) { - if (func(relocation(i)) == IterationDecision::Break) + const auto reloc = relocation(i); + if (reloc.type() == 0) + continue; + if (func(reloc) == IterationDecision::Break) break; } } @@ -290,6 +349,7 @@ inline void DynamicObject::for_each_dynamic_entry(F func) const { auto* dyns = reinterpret_cast<const Elf32_Dyn*>(m_dynamic_address.as_ptr()); for (unsigned i = 0;; ++i) { + // dbgprintf("%d\n", i); auto&& dyn = DynamicEntry(dyns[i]); if (dyn.tag() == DT_NULL) break; @@ -297,5 +357,30 @@ inline void DynamicObject::for_each_dynamic_entry(F func) const break; } } +template<typename F> +inline void DynamicObject::for_each_needed_library(F func) const +{ + for_each_dynamic_entry([func, this](auto entry) { + if (entry.tag() != DT_NEEDED) + return IterationDecision::Continue; + Elf32_Word offset = entry.val(); + const char* name = (const char*)(m_base_address.offset(m_string_table_offset).offset(offset)).as_ptr(); + if (func(StringView(name)) == IterationDecision::Break) + return IterationDecision::Break; + return IterationDecision::Continue; + }); +} + +template<typename F> +void DynamicObject::for_each_initialization_array_function(F f) const +{ + if (!has_init_array_section()) + return; + FlatPtr init_array = (FlatPtr)init_array_section().address().as_ptr(); + for (size_t i = 0; i < (m_init_array_size / sizeof(void*)); ++i) { + InitializationFunction current = ((InitializationFunction*)(init_array))[i]; + f(current); + } +} } // end namespace ELF diff --git a/Libraries/LibELF/Image.cpp b/Libraries/LibELF/Image.cpp index d2208234e8..8a5b7f56d8 100644 --- a/Libraries/LibELF/Image.cpp +++ b/Libraries/LibELF/Image.cpp @@ -259,6 +259,11 @@ const Image::ProgramHeader Image::program_header(unsigned index) const return ProgramHeader(*this, index); } +FlatPtr Image::program_header_table_offset() const +{ + return header().e_phoff; +} + const Image::Relocation Image::RelocationSection::relocation(unsigned index) const { ASSERT(index < relocation_count()); diff --git a/Libraries/LibELF/Image.h b/Libraries/LibELF/Image.h index 5f7540a3a6..5253ec8aa6 100644 --- a/Libraries/LibELF/Image.h +++ b/Libraries/LibELF/Image.h @@ -185,6 +185,7 @@ public: const Symbol symbol(unsigned) const; const Section section(unsigned) const; const ProgramHeader program_header(unsigned const) const; + FlatPtr program_header_table_offset() const; template<typename F> void for_each_section(F) const; @@ -204,6 +205,8 @@ public: bool is_dynamic() const { return header().e_type == ET_DYN; } VirtualAddress entry() const { return VirtualAddress(header().e_entry); } + FlatPtr base_address() const { return (FlatPtr)m_buffer; } + size_t size() const { return m_size; } private: const char* raw_data(unsigned offset) const; diff --git a/Libraries/LibELF/exec_elf.h b/Libraries/LibELF/exec_elf.h index 03871e6e26..45a0e9401f 100644 --- a/Libraries/LibELF/exec_elf.h +++ b/Libraries/LibELF/exec_elf.h @@ -778,15 +778,15 @@ struct elf_args { /* Relocation types */ #define R_386_NONE 0 -#define R_386_32 1 /* Symbol + Addend */ -#define R_386_PC32 2 /* Symbol + Addend - Section offset */ -#define R_386_GOT32 3 /* Used by build-time linker to create GOT entry */ -#define R_386_PLT32 4 /* Used by build-time linker to create PLT entry */ -#define R_386_COPY 5 /* https://docs.oracle.com/cd/E23824_01/html/819-0690/chapter4-10454.html#chapter4-84604 */ -#define R_386_GLOB_DAT 6 /* Relation b/w GOT entry and symbol */ -#define R_386_JMP_SLOT 7 /* Fixed up by dynamic loader */ -#define R_386_RELATIVE 8 /* Base address + Addned */ +#define R_386_32 1 /* Symbol + Addend */ +#define R_386_PC32 2 /* Symbol + Addend - Section offset */ +#define R_386_GOT32 3 /* Used by build-time linker to create GOT entry */ +#define R_386_PLT32 4 /* Used by build-time linker to create PLT entry */ +#define R_386_COPY 5 /* https://docs.oracle.com/cd/E23824_01/html/819-0690/chapter4-10454.html#chapter4-84604 */ +#define R_386_GLOB_DAT 6 /* Relation b/w GOT entry and symbol */ +#define R_386_JMP_SLOT 7 /* Fixed up by dynamic loader */ +#define R_386_RELATIVE 8 /* Base address + Addned */ #define R_386_TLS_TPOFF 14 /* Negative offset into the static TLS storage */ - +#define R_386_TLS_TPOFF32 37 #endif /* _SYS_EXEC_ELF_H_ */ |