diff options
author | Matt A. Tobin <mattatobin@localhost.localdomain> | 2018-02-02 04:16:08 -0500 |
---|---|---|
committer | Matt A. Tobin <mattatobin@localhost.localdomain> | 2018-02-02 04:16:08 -0500 |
commit | 5f8de423f190bbb79a62f804151bc24824fa32d8 (patch) | |
tree | 10027f336435511475e392454359edea8e25895d /mozglue/linker | |
parent | 49ee0794b5d912db1f95dce6eb52d781dc210db5 (diff) | |
download | uxp-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.gz |
Add m-esr52 at 52.6.0
Diffstat (limited to 'mozglue/linker')
25 files changed, 7348 insertions, 0 deletions
diff --git a/mozglue/linker/BaseElf.cpp b/mozglue/linker/BaseElf.cpp new file mode 100644 index 0000000000..8a78285ef6 --- /dev/null +++ b/mozglue/linker/BaseElf.cpp @@ -0,0 +1,215 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "BaseElf.h" +#include "Elfxx.h" +#include "Logging.h" +#include "mozilla/RefPtr.h" + +using namespace Elf; +using namespace mozilla; + +unsigned long +BaseElf::Hash(const char *symbol) +{ + const unsigned char *sym = reinterpret_cast<const unsigned char *>(symbol); + unsigned long h = 0, g; + while (*sym) { + h = (h << 4) + *sym++; + g = h & 0xf0000000; + h ^= g; + h ^= g >> 24; + } + return h; +} + +void * +BaseElf::GetSymbolPtr(const char *symbol) const +{ + return GetSymbolPtr(symbol, Hash(symbol)); +} + +void * +BaseElf::GetSymbolPtr(const char *symbol, unsigned long hash) const +{ + const Sym *sym = GetSymbol(symbol, hash); + void *ptr = nullptr; + if (sym && sym->st_shndx != SHN_UNDEF) + ptr = GetPtr(sym->st_value); + DEBUG_LOG("BaseElf::GetSymbolPtr(%p [\"%s\"], \"%s\") = %p", + reinterpret_cast<const void *>(this), GetPath(), symbol, ptr); + return ptr; +} + +const Sym * +BaseElf::GetSymbol(const char *symbol, unsigned long hash) const +{ + /* Search symbol with the buckets and chains tables. + * The hash computed from the symbol name gives an index in the buckets + * table. The corresponding value in the bucket table is an index in the + * symbols table and in the chains table. + * If the corresponding symbol in the symbols table matches, we're done. + * Otherwise, the corresponding value in the chains table is a new index + * in both tables, which corresponding symbol is tested and so on and so + * forth */ + size_t bucket = hash % buckets.numElements(); + for (size_t y = buckets[bucket]; y != STN_UNDEF; y = chains[y]) { + if (strcmp(symbol, strtab.GetStringAt(symtab[y].st_name))) + continue; + return &symtab[y]; + } + return nullptr; +} + +bool +BaseElf::Contains(void *addr) const +{ + return base.Contains(addr); +} + +#ifdef __ARM_EABI__ +const void * +BaseElf::FindExidx(int *pcount) const +{ + if (arm_exidx) { + *pcount = arm_exidx.numElements(); + return arm_exidx; + } + *pcount = 0; + return nullptr; +} +#endif + +already_AddRefed<LibHandle> +LoadedElf::Create(const char *path, void *base_addr) +{ + DEBUG_LOG("LoadedElf::Create(\"%s\", %p) = ...", path, base_addr); + + uint8_t mapped; + /* If the page is not mapped, mincore returns an error. If base_addr is + * nullptr, as would happen if the corresponding binary is prelinked with + * the prelink look (but not with the android apriori tool), no page being + * mapped there (right?), mincore returns an error, too, which makes + * prelinked libraries on glibc unsupported. This is not an interesting + * use case for now, so don't try supporting that case. + */ + if (mincore(const_cast<void*>(base_addr), PageSize(), &mapped)) + return nullptr; + + RefPtr<LoadedElf> elf = new LoadedElf(path); + + const Ehdr *ehdr = Ehdr::validate(base_addr); + if (!ehdr) + return nullptr; + + Addr min_vaddr = (Addr) -1; // We want to find the lowest and biggest + Addr max_vaddr = 0; // virtual address used by this Elf. + const Phdr *dyn = nullptr; +#ifdef __ARM_EABI__ + const Phdr *arm_exidx_phdr = nullptr; +#endif + + Array<Phdr> phdrs(reinterpret_cast<const char *>(ehdr) + ehdr->e_phoff, + ehdr->e_phnum); + for (auto phdr = phdrs.begin(); phdr < phdrs.end(); ++phdr) { + switch (phdr->p_type) { + case PT_LOAD: + if (phdr->p_vaddr < min_vaddr) + min_vaddr = phdr->p_vaddr; + if (max_vaddr < phdr->p_vaddr + phdr->p_memsz) + max_vaddr = phdr->p_vaddr + phdr->p_memsz; + break; + case PT_DYNAMIC: + dyn = &*phdr; + break; +#ifdef __ARM_EABI__ + case PT_ARM_EXIDX: + /* We cannot initialize arm_exidx here + because we don't have a base yet */ + arm_exidx_phdr = &*phdr; + break; +#endif + } + } + + /* If the lowest PT_LOAD virtual address in headers is not 0, then the ELF + * is either prelinked or a non-PIE executable. The former case is not + * possible, because base_addr would be nullptr and the mincore test above + * would already have made us return. + * For a non-PIE executable, PT_LOADs contain absolute addresses, so in + * practice, this means min_vaddr should be equal to base_addr. max_vaddr + * can thus be adjusted accordingly. + */ + if (min_vaddr != 0) { + void *min_vaddr_ptr = reinterpret_cast<void *>( + static_cast<uintptr_t>(min_vaddr)); + if (min_vaddr_ptr != base_addr) { + LOG("%s: %p != %p", elf->GetPath(), min_vaddr_ptr, base_addr); + return nullptr; + } + max_vaddr -= min_vaddr; + } + if (!dyn) { + LOG("%s: No PT_DYNAMIC segment found", elf->GetPath()); + return nullptr; + } + + elf->base.Assign(base_addr, max_vaddr); + + if (!elf->InitDyn(dyn)) + return nullptr; + +#ifdef __ARM_EABI__ + if (arm_exidx_phdr) + elf->arm_exidx.InitSize(elf->GetPtr(arm_exidx_phdr->p_vaddr), + arm_exidx_phdr->p_memsz); +#endif + + DEBUG_LOG("LoadedElf::Create(\"%s\", %p) = %p", path, base_addr, + static_cast<void *>(elf)); + + ElfLoader::Singleton.Register(elf); + return elf.forget(); +} + +bool +LoadedElf::InitDyn(const Phdr *pt_dyn) +{ + Array<Dyn> dyns; + dyns.InitSize(GetPtr<Dyn>(pt_dyn->p_vaddr), pt_dyn->p_filesz); + + size_t symnum = 0; + for (auto dyn = dyns.begin(); dyn < dyns.end() && dyn->d_tag; ++dyn) { + switch (dyn->d_tag) { + case DT_HASH: + { + DEBUG_LOG("%s 0x%08" PRIxAddr, "DT_HASH", dyn->d_un.d_val); + const Elf::Word *hash_table_header = \ + GetPtr<Elf::Word>(dyn->d_un.d_ptr); + symnum = hash_table_header[1]; + buckets.Init(&hash_table_header[2], hash_table_header[0]); + chains.Init(&*buckets.end()); + } + break; + case DT_STRTAB: + DEBUG_LOG("%s 0x%08" PRIxAddr, "DT_STRTAB", dyn->d_un.d_val); + strtab.Init(GetPtr(dyn->d_un.d_ptr)); + break; + case DT_SYMTAB: + DEBUG_LOG("%s 0x%08" PRIxAddr, "DT_SYMTAB", dyn->d_un.d_val); + symtab.Init(GetPtr(dyn->d_un.d_ptr)); + break; + } + } + if (!buckets || !symnum) { + ERROR("%s: Missing or broken DT_HASH", GetPath()); + } else if (!strtab) { + ERROR("%s: Missing DT_STRTAB", GetPath()); + } else if (!symtab) { + ERROR("%s: Missing DT_SYMTAB", GetPath()); + } else { + return true; + } + return false; +} diff --git a/mozglue/linker/BaseElf.h b/mozglue/linker/BaseElf.h new file mode 100644 index 0000000000..19f9061bce --- /dev/null +++ b/mozglue/linker/BaseElf.h @@ -0,0 +1,141 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef BaseElf_h +#define BaseElf_h + +#include "ElfLoader.h" +#include "Elfxx.h" + + +/** + * Base class for ELF libraries. This class includes things that will be + * common between SystemElfs and CustomElfs. + */ +class BaseElf: public LibHandle +{ +public: + /** + * Hash function for symbol lookup, as defined in ELF standard for System V. + */ + static unsigned long Hash(const char *symbol); + + /** + * Returns the address corresponding to the given symbol name (with a + * pre-computed hash). + */ + void *GetSymbolPtr(const char *symbol, unsigned long hash) const; + + /** + * Returns a pointer to the Elf Symbol in the Dynamic Symbol table + * corresponding to the given symbol name (with a pre-computed hash). + */ + const Elf::Sym *GetSymbol(const char *symbol, unsigned long hash) const; + + BaseElf(const char *path, Mappable *mappable = nullptr) + : LibHandle(path) + , mappable(mappable) + { + } + +protected: + /** + * Inherited from LibHandle. Those are temporary and are not supposed to + * be used. + */ + virtual void *GetSymbolPtr(const char *symbol) const; + virtual bool Contains(void *addr) const; + virtual void *GetBase() const { return GetPtr(0); } + +#ifdef __ARM_EABI__ + virtual const void *FindExidx(int *pcount) const; +#endif + + virtual Mappable *GetMappable() const { return NULL; }; + +public: +/* private: */ + /** + * Returns a pointer relative to the base address where the library is + * loaded. + */ + void *GetPtr(const Elf::Addr offset) const + { + if (reinterpret_cast<void *>(offset) > base) + return reinterpret_cast<void *>(offset); + return base + offset; + } + + /** + * Like the above, but returns a typed (const) pointer + */ + template <typename T> + const T *GetPtr(const Elf::Addr offset) const + { + if (reinterpret_cast<void *>(offset) > base) + return reinterpret_cast<const T *>(offset); + return reinterpret_cast<const T *>(base + offset); + } + + /* Appropriated Mappable */ + /* /!\ we rely on this being nullptr for BaseElf instances, but not + * CustomElf instances. */ + RefPtr<Mappable> mappable; + + /* Base address where the library is loaded */ + MappedPtr base; + + /* Buckets and chains for the System V symbol hash table */ + Array<Elf::Word> buckets; + UnsizedArray<Elf::Word> chains; + +/* protected: */ + /* String table */ + Elf::Strtab strtab; + + /* Symbol table */ + UnsizedArray<Elf::Sym> symtab; + +#ifdef __ARM_EABI__ + /* ARM.exidx information used by FindExidx */ + Array<uint32_t[2]> arm_exidx; +#endif +}; + + +/** + * Class for ELF libraries that already loaded in memory. + */ +class LoadedElf: public BaseElf +{ +public: + /** + * Returns a LoadedElf corresponding to the already loaded ELF + * at the given base address. + */ + static already_AddRefed<LibHandle> Create(const char *path, + void *base_addr); + +private: + LoadedElf(const char *path) + : BaseElf(path) { } + + ~LoadedElf() + { + /* Avoid base's destructor unmapping something that doesn't actually + * belong to it. */ + base.release(); + ElfLoader::Singleton.Forget(this); + } + + /** + * Initializes the library according to information found in the given + * PT_DYNAMIC header. + * Returns whether this succeeded or failed. + */ + bool InitDyn(const Elf::Phdr *pt_dyn); +}; + + +#endif /* BaseElf_h */ diff --git a/mozglue/linker/CustomElf.cpp b/mozglue/linker/CustomElf.cpp new file mode 100644 index 0000000000..dbab0cf0d1 --- /dev/null +++ b/mozglue/linker/CustomElf.cpp @@ -0,0 +1,774 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include <cstring> +#include <sys/mman.h> +#include <vector> +#include <dlfcn.h> +#include <signal.h> +#include <string.h> +#include "CustomElf.h" +#include "BaseElf.h" +#include "Mappable.h" +#include "Logging.h" + +using namespace Elf; +using namespace mozilla; + +/* TODO: Fill ElfLoader::Singleton.lastError on errors. */ + +/* Function used to report library mappings from the custom linker to Gecko + * crash reporter */ +#ifdef ANDROID +extern "C" { + void report_mapping(char *name, void *base, uint32_t len, uint32_t offset); + void delete_mapping(const char *name); +} +#else +#define report_mapping(...) +#define delete_mapping(...) +#endif + +const Ehdr *Ehdr::validate(const void *buf) +{ + if (!buf || buf == MAP_FAILED) + return nullptr; + + const Ehdr *ehdr = reinterpret_cast<const Ehdr *>(buf); + + /* Only support ELF executables or libraries for the host system */ + if (memcmp(ELFMAG, &ehdr->e_ident, SELFMAG) || + ehdr->e_ident[EI_CLASS] != ELFCLASS || + ehdr->e_ident[EI_DATA] != ELFDATA || + ehdr->e_ident[EI_VERSION] != 1 || + (ehdr->e_ident[EI_OSABI] != ELFOSABI && ehdr->e_ident[EI_OSABI] != ELFOSABI_NONE) || +#ifdef EI_ABIVERSION + ehdr->e_ident[EI_ABIVERSION] != ELFABIVERSION || +#endif + (ehdr->e_type != ET_EXEC && ehdr->e_type != ET_DYN) || + ehdr->e_machine != ELFMACHINE || + ehdr->e_version != 1 || + ehdr->e_phentsize != sizeof(Phdr)) + return nullptr; + + return ehdr; +} + +namespace { + +void debug_phdr(const char *type, const Phdr *phdr) +{ + DEBUG_LOG("%s @0x%08" PRIxAddr " (" + "filesz: 0x%08" PRIxAddr ", " + "memsz: 0x%08" PRIxAddr ", " + "offset: 0x%08" PRIxAddr ", " + "flags: %c%c%c)", + type, phdr->p_vaddr, phdr->p_filesz, phdr->p_memsz, + phdr->p_offset, phdr->p_flags & PF_R ? 'r' : '-', + phdr->p_flags & PF_W ? 'w' : '-', phdr->p_flags & PF_X ? 'x' : '-'); +} + +static int p_flags_to_mprot(Word flags) +{ + return ((flags & PF_X) ? PROT_EXEC : 0) | + ((flags & PF_W) ? PROT_WRITE : 0) | + ((flags & PF_R) ? PROT_READ : 0); +} + +} /* anonymous namespace */ + +/** + * RAII wrapper for a mapping of the first page off a Mappable object. + * This calls Mappable::munmap instead of system munmap. + */ +class Mappable1stPagePtr: public GenericMappedPtr<Mappable1stPagePtr> { +public: + Mappable1stPagePtr(Mappable *mappable) + : GenericMappedPtr<Mappable1stPagePtr>( + mappable->mmap(nullptr, PageSize(), PROT_READ, MAP_PRIVATE, 0)) + , mappable(mappable) + { + /* Ensure the content of this page */ + mappable->ensure(*this); + } + +private: + friend class GenericMappedPtr<Mappable1stPagePtr>; + void munmap(void *buf, size_t length) { + mappable->munmap(buf, length); + } + + RefPtr<Mappable> mappable; +}; + + +already_AddRefed<LibHandle> +CustomElf::Load(Mappable *mappable, const char *path, int flags) +{ + DEBUG_LOG("CustomElf::Load(\"%s\", 0x%x) = ...", path, flags); + if (!mappable) + return nullptr; + /* Keeping a RefPtr of the CustomElf is going to free the appropriate + * resources when returning nullptr */ + RefPtr<CustomElf> elf = new CustomElf(mappable, path); + /* Map the first page of the Elf object to access Elf and program headers */ + Mappable1stPagePtr ehdr_raw(mappable); + if (ehdr_raw == MAP_FAILED) + return nullptr; + + const Ehdr *ehdr = Ehdr::validate(ehdr_raw); + if (!ehdr) + return nullptr; + + /* Scan Elf Program Headers and gather some information about them */ + std::vector<const Phdr *> pt_loads; + Addr min_vaddr = (Addr) -1; // We want to find the lowest and biggest + Addr max_vaddr = 0; // virtual address used by this Elf. + const Phdr *dyn = nullptr; + + const Phdr *first_phdr = reinterpret_cast<const Phdr *>( + reinterpret_cast<const char *>(ehdr) + ehdr->e_phoff); + const Phdr *end_phdr = &first_phdr[ehdr->e_phnum]; +#ifdef __ARM_EABI__ + const Phdr *arm_exidx_phdr = nullptr; +#endif + + for (const Phdr *phdr = first_phdr; phdr < end_phdr; phdr++) { + switch (phdr->p_type) { + case PT_LOAD: + debug_phdr("PT_LOAD", phdr); + pt_loads.push_back(phdr); + if (phdr->p_vaddr < min_vaddr) + min_vaddr = phdr->p_vaddr; + if (max_vaddr < phdr->p_vaddr + phdr->p_memsz) + max_vaddr = phdr->p_vaddr + phdr->p_memsz; + break; + case PT_DYNAMIC: + debug_phdr("PT_DYNAMIC", phdr); + if (!dyn) { + dyn = phdr; + } else { + ERROR("%s: Multiple PT_DYNAMIC segments detected", elf->GetPath()); + return nullptr; + } + break; + case PT_TLS: + debug_phdr("PT_TLS", phdr); + if (phdr->p_memsz) { + ERROR("%s: TLS is not supported", elf->GetPath()); + return nullptr; + } + break; + case PT_GNU_STACK: + debug_phdr("PT_GNU_STACK", phdr); +// Skip on Android until bug 706116 is fixed +#ifndef ANDROID + if (phdr->p_flags & PF_X) { + ERROR("%s: Executable stack is not supported", elf->GetPath()); + return nullptr; + } +#endif + break; +#ifdef __ARM_EABI__ + case PT_ARM_EXIDX: + /* We cannot initialize arm_exidx here + because we don't have a base yet */ + arm_exidx_phdr = phdr; + break; +#endif + default: + DEBUG_LOG("%s: Program header type #%d not handled", + elf->GetPath(), phdr->p_type); + } + } + + if (min_vaddr != 0) { + ERROR("%s: Unsupported minimal virtual address: 0x%08" PRIxAddr, + elf->GetPath(), min_vaddr); + return nullptr; + } + if (!dyn) { + ERROR("%s: No PT_DYNAMIC segment found", elf->GetPath()); + return nullptr; + } + + /* Reserve enough memory to map the complete virtual address space for this + * library. + * As we are using the base address from here to mmap something else with + * MAP_FIXED | MAP_SHARED, we need to make sure these mmaps will work. For + * instance, on armv6, MAP_SHARED mappings require a 16k alignment, but mmap + * MAP_PRIVATE only returns a 4k aligned address. So we first get a base + * address with MAP_SHARED, which guarantees the kernel returns an address + * that we'll be able to use with MAP_FIXED, and then remap MAP_PRIVATE at + * the same address, because of some bad side effects of keeping it as + * MAP_SHARED. */ + elf->base.Assign(MemoryRange::mmap(nullptr, max_vaddr, PROT_NONE, + MAP_SHARED | MAP_ANONYMOUS, -1, 0)); + if ((elf->base == MAP_FAILED) || + (mmap(elf->base, max_vaddr, PROT_NONE, + MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0) != elf->base)) { + ERROR("%s: Failed to mmap", elf->GetPath()); + return nullptr; + } + + /* Load and initialize library */ + for (std::vector<const Phdr *>::iterator it = pt_loads.begin(); + it < pt_loads.end(); ++it) + if (!elf->LoadSegment(*it)) + return nullptr; + + /* We're not going to mmap anymore */ + mappable->finalize(); + + report_mapping(const_cast<char *>(elf->GetName()), elf->base, + (max_vaddr + PAGE_SIZE - 1) & PAGE_MASK, 0); + + elf->l_addr = elf->base; + elf->l_name = elf->GetPath(); + elf->l_ld = elf->GetPtr<Dyn>(dyn->p_vaddr); + ElfLoader::Singleton.Register(elf); + + if (!elf->InitDyn(dyn)) + return nullptr; + + if (elf->has_text_relocs) { + for (std::vector<const Phdr *>::iterator it = pt_loads.begin(); + it < pt_loads.end(); ++it) + mprotect(PageAlignedPtr(elf->GetPtr((*it)->p_vaddr)), + PageAlignedEndPtr((*it)->p_memsz), + p_flags_to_mprot((*it)->p_flags) | PROT_WRITE); + } + + if (!elf->Relocate() || !elf->RelocateJumps()) + return nullptr; + + if (elf->has_text_relocs) { + for (std::vector<const Phdr *>::iterator it = pt_loads.begin(); + it < pt_loads.end(); ++it) + mprotect(PageAlignedPtr(elf->GetPtr((*it)->p_vaddr)), + PageAlignedEndPtr((*it)->p_memsz), + p_flags_to_mprot((*it)->p_flags)); + } + + if (!elf->CallInit()) + return nullptr; + +#ifdef __ARM_EABI__ + if (arm_exidx_phdr) + elf->arm_exidx.InitSize(elf->GetPtr(arm_exidx_phdr->p_vaddr), + arm_exidx_phdr->p_memsz); +#endif + + if (MOZ_UNLIKELY(Logging::isVerbose())) { + elf->stats("oneLibLoaded"); + } + DEBUG_LOG("CustomElf::Load(\"%s\", 0x%x) = %p", path, flags, + static_cast<void *>(elf)); + return elf.forget(); +} + +CustomElf::~CustomElf() +{ + DEBUG_LOG("CustomElf::~CustomElf(%p [\"%s\"])", + reinterpret_cast<void *>(this), GetPath()); + CallFini(); + /* Normally, __cxa_finalize is called by the .fini function. However, + * Android NDK before r6b doesn't do that. Our wrapped cxa_finalize only + * calls destructors once, so call it in all cases. */ + ElfLoader::__wrap_cxa_finalize(this); + ElfLoader::Singleton.Forget(this); + delete_mapping(GetName()); +} + +void * +CustomElf::GetSymbolPtrInDeps(const char *symbol) const +{ + /* Resolve dlopen and related functions to point to ours */ + if (symbol[0] == 'd' && symbol[1] == 'l') { + if (strcmp(symbol + 2, "open") == 0) + return FunctionPtr(__wrap_dlopen); + if (strcmp(symbol + 2, "error") == 0) + return FunctionPtr(__wrap_dlerror); + if (strcmp(symbol + 2, "close") == 0) + return FunctionPtr(__wrap_dlclose); + if (strcmp(symbol + 2, "sym") == 0) + return FunctionPtr(__wrap_dlsym); + if (strcmp(symbol + 2, "addr") == 0) + return FunctionPtr(__wrap_dladdr); + if (strcmp(symbol + 2, "_iterate_phdr") == 0) + return FunctionPtr(__wrap_dl_iterate_phdr); + } else if (symbol[0] == '_' && symbol[1] == '_') { + /* Resolve a few C++ ABI specific functions to point to ours */ +#ifdef __ARM_EABI__ + if (strcmp(symbol + 2, "aeabi_atexit") == 0) + return FunctionPtr(&ElfLoader::__wrap_aeabi_atexit); +#else + if (strcmp(symbol + 2, "cxa_atexit") == 0) + return FunctionPtr(&ElfLoader::__wrap_cxa_atexit); +#endif + if (strcmp(symbol + 2, "cxa_finalize") == 0) + return FunctionPtr(&ElfLoader::__wrap_cxa_finalize); + if (strcmp(symbol + 2, "dso_handle") == 0) + return const_cast<CustomElf *>(this); + if (strcmp(symbol + 2, "moz_linker_stats") == 0) + return FunctionPtr(&ElfLoader::stats); +#ifdef __ARM_EABI__ + if (strcmp(symbol + 2, "gnu_Unwind_Find_exidx") == 0) + return FunctionPtr(__wrap___gnu_Unwind_Find_exidx); +#endif + } else if (symbol[0] == 's' && symbol[1] == 'i') { + if (strcmp(symbol + 2, "gnal") == 0) + return FunctionPtr(signal); + if (strcmp(symbol + 2, "gaction") == 0) + return FunctionPtr(sigaction); + } + + void *sym; + + unsigned long hash = Hash(symbol); + + /* self_elf should never be NULL, but better safe than sorry. */ + if (ElfLoader::Singleton.self_elf) { + /* We consider the library containing this code a permanent LD_PRELOAD, + * so, check if the symbol exists here first. */ + sym = static_cast<BaseElf *>( + ElfLoader::Singleton.self_elf.get())->GetSymbolPtr(symbol, hash); + if (sym) + return sym; + } + + /* Then search the symbol in our dependencies. Since we already searched in + * libraries the system linker loaded, skip those (on glibc systems). We + * also assume the symbol is to be found in one of the dependent libraries + * directly, not in their own dependent libraries. Building libraries with + * --no-allow-shlib-undefined ensures such indirect symbol dependency don't + * happen. */ + for (std::vector<RefPtr<LibHandle> >::const_iterator it = dependencies.begin(); + it < dependencies.end(); ++it) { + /* Skip if it's the library containing this code, since we've already + * looked at it above. */ + if (*it == ElfLoader::Singleton.self_elf) + continue; + if (BaseElf *be = (*it)->AsBaseElf()) { + sym = be->GetSymbolPtr(symbol, hash); + } else { + sym = (*it)->GetSymbolPtr(symbol); + } + if (sym) + return sym; + } + return nullptr; +} + +void +CustomElf::stats(const char *when) const +{ + mappable->stats(when, GetPath()); +} + +bool +CustomElf::LoadSegment(const Phdr *pt_load) const +{ + if (pt_load->p_type != PT_LOAD) { + DEBUG_LOG("%s: Elf::LoadSegment only takes PT_LOAD program headers", GetPath()); + return false;; + } + + int prot = p_flags_to_mprot(pt_load->p_flags); + + /* Mmap at page boundary */ + Addr align = PageSize(); + Addr align_offset; + void *mapped, *where; + do { + align_offset = pt_load->p_vaddr - AlignedPtr(pt_load->p_vaddr, align); + where = GetPtr(pt_load->p_vaddr - align_offset); + DEBUG_LOG("%s: Loading segment @%p %c%c%c", GetPath(), where, + prot & PROT_READ ? 'r' : '-', + prot & PROT_WRITE ? 'w' : '-', + prot & PROT_EXEC ? 'x' : '-'); + mapped = mappable->mmap(where, pt_load->p_filesz + align_offset, + prot, MAP_PRIVATE | MAP_FIXED, + pt_load->p_offset - align_offset); + if ((mapped != MAP_FAILED) || (pt_load->p_vaddr == 0) || + (pt_load->p_align == align)) + break; + /* The virtual address space for the library is properly aligned at + * 16k on ARMv6 (see CustomElf::Load), and so is the first segment + * (p_vaddr == 0). But subsequent segments may not be 16k aligned + * and fail to mmap. In such case, try to mmap again at the p_align + * boundary instead of page boundary. */ + DEBUG_LOG("%s: Failed to mmap, retrying", GetPath()); + align = pt_load->p_align; + } while (1); + + if (mapped != where) { + if (mapped == MAP_FAILED) { + ERROR("%s: Failed to mmap", GetPath()); + } else { + ERROR("%s: Didn't map at the expected location (wanted: %p, got: %p)", + GetPath(), where, mapped); + } + return false; + } + + /* Ensure the availability of all pages within the mapping if on-demand + * decompression is disabled (MOZ_LINKER_ONDEMAND=0 or signal handler not + * registered). */ + const char *ondemand = getenv("MOZ_LINKER_ONDEMAND"); + if (!ElfLoader::Singleton.hasRegisteredHandler() || + (ondemand && !strncmp(ondemand, "0", 2 /* Including '\0' */))) { + for (Addr off = 0; off < pt_load->p_filesz + align_offset; + off += PageSize()) { + mappable->ensure(reinterpret_cast<char *>(mapped) + off); + } + } + /* When p_memsz is greater than p_filesz, we need to have nulled out memory + * after p_filesz and before p_memsz. + * Above the end of the last page, and up to p_memsz, we already have nulled + * out memory because we mapped anonymous memory on the whole library virtual + * address space. We just need to adjust this anonymous memory protection + * flags. */ + if (pt_load->p_memsz > pt_load->p_filesz) { + Addr file_end = pt_load->p_vaddr + pt_load->p_filesz; + Addr mem_end = pt_load->p_vaddr + pt_load->p_memsz; + Addr next_page = PageAlignedEndPtr(file_end); + if (next_page > file_end) { + /* The library is not registered at this point, so we can't rely on + * on-demand decompression to handle missing pages here. */ + void *ptr = GetPtr(file_end); + mappable->ensure(ptr); + memset(ptr, 0, next_page - file_end); + } + if (mem_end > next_page) { + if (mprotect(GetPtr(next_page), mem_end - next_page, prot) < 0) { + ERROR("%s: Failed to mprotect", GetPath()); + return false; + } + } + } + return true; +} + +namespace { + +void debug_dyn(const char *type, const Dyn *dyn) +{ + DEBUG_LOG("%s 0x%08" PRIxAddr, type, dyn->d_un.d_val); +} + +} /* anonymous namespace */ + +bool +CustomElf::InitDyn(const Phdr *pt_dyn) +{ + /* Scan PT_DYNAMIC segment and gather some information */ + const Dyn *first_dyn = GetPtr<Dyn>(pt_dyn->p_vaddr); + const Dyn *end_dyn = GetPtr<Dyn>(pt_dyn->p_vaddr + pt_dyn->p_filesz); + std::vector<Word> dt_needed; + size_t symnum = 0; + for (const Dyn *dyn = first_dyn; dyn < end_dyn && dyn->d_tag; dyn++) { + switch (dyn->d_tag) { + case DT_NEEDED: + debug_dyn("DT_NEEDED", dyn); + dt_needed.push_back(dyn->d_un.d_val); + break; + case DT_HASH: + { + debug_dyn("DT_HASH", dyn); + const Word *hash_table_header = GetPtr<Word>(dyn->d_un.d_ptr); + symnum = hash_table_header[1]; + buckets.Init(&hash_table_header[2], hash_table_header[0]); + chains.Init(&*buckets.end()); + } + break; + case DT_STRTAB: + debug_dyn("DT_STRTAB", dyn); + strtab.Init(GetPtr(dyn->d_un.d_ptr)); + break; + case DT_SYMTAB: + debug_dyn("DT_SYMTAB", dyn); + symtab.Init(GetPtr(dyn->d_un.d_ptr)); + break; + case DT_SYMENT: + debug_dyn("DT_SYMENT", dyn); + if (dyn->d_un.d_val != sizeof(Sym)) { + ERROR("%s: Unsupported DT_SYMENT", GetPath()); + return false; + } + break; + case DT_TEXTREL: + if (strcmp("libflashplayer.so", GetName()) == 0) { + has_text_relocs = true; + } else { + ERROR("%s: Text relocations are not supported", GetPath()); + return false; + } + break; + case DT_STRSZ: /* Ignored */ + debug_dyn("DT_STRSZ", dyn); + break; + case UNSUPPORTED_RELOC(): + case UNSUPPORTED_RELOC(SZ): + case UNSUPPORTED_RELOC(ENT): + ERROR("%s: Unsupported relocations", GetPath()); + return false; + case RELOC(): + debug_dyn(STR_RELOC(), dyn); + relocations.Init(GetPtr(dyn->d_un.d_ptr)); + break; + case RELOC(SZ): + debug_dyn(STR_RELOC(SZ), dyn); + relocations.InitSize(dyn->d_un.d_val); + break; + case RELOC(ENT): + debug_dyn(STR_RELOC(ENT), dyn); + if (dyn->d_un.d_val != sizeof(Reloc)) { + ERROR("%s: Unsupported DT_RELENT", GetPath()); + return false; + } + break; + case DT_JMPREL: + debug_dyn("DT_JMPREL", dyn); + jumprels.Init(GetPtr(dyn->d_un.d_ptr)); + break; + case DT_PLTRELSZ: + debug_dyn("DT_PLTRELSZ", dyn); + jumprels.InitSize(dyn->d_un.d_val); + break; + case DT_PLTGOT: + debug_dyn("DT_PLTGOT", dyn); + break; + case DT_INIT: + debug_dyn("DT_INIT", dyn); + init = dyn->d_un.d_ptr; + break; + case DT_INIT_ARRAY: + debug_dyn("DT_INIT_ARRAY", dyn); + init_array.Init(GetPtr(dyn->d_un.d_ptr)); + break; + case DT_INIT_ARRAYSZ: + debug_dyn("DT_INIT_ARRAYSZ", dyn); + init_array.InitSize(dyn->d_un.d_val); + break; + case DT_FINI: + debug_dyn("DT_FINI", dyn); + fini = dyn->d_un.d_ptr; + break; + case DT_FINI_ARRAY: + debug_dyn("DT_FINI_ARRAY", dyn); + fini_array.Init(GetPtr(dyn->d_un.d_ptr)); + break; + case DT_FINI_ARRAYSZ: + debug_dyn("DT_FINI_ARRAYSZ", dyn); + fini_array.InitSize(dyn->d_un.d_val); + break; + case DT_PLTREL: + if (dyn->d_un.d_val != RELOC()) { + ERROR("%s: Error: DT_PLTREL is not " STR_RELOC(), GetPath()); + return false; + } + break; + case DT_FLAGS: + { + Addr flags = dyn->d_un.d_val; + /* Treat as a DT_TEXTREL tag */ + if (flags & DF_TEXTREL) { + if (strcmp("libflashplayer.so", GetName()) == 0) { + has_text_relocs = true; + } else { + ERROR("%s: Text relocations are not supported", GetPath()); + return false; + } + } + /* we can treat this like having a DT_SYMBOLIC tag */ + flags &= ~DF_SYMBOLIC; + if (flags) + WARN("%s: unhandled flags #%" PRIxAddr" not handled", + GetPath(), flags); + } + break; + case DT_SONAME: /* Should match GetName(), but doesn't matter */ + case DT_SYMBOLIC: /* Indicates internal symbols should be looked up in + * the library itself first instead of the executable, + * which is actually what this linker does by default */ + case RELOC(COUNT): /* Indicates how many relocations are relative, which + * is usually used to skip relocations on prelinked + * libraries. They are not supported anyways. */ + case UNSUPPORTED_RELOC(COUNT): /* This should error out, but it doesn't + * really matter. */ + case DT_FLAGS_1: /* Additional linker-internal flags that we don't care about. See + * DF_1_* values in src/include/elf/common.h in binutils. */ + case DT_VERSYM: /* DT_VER* entries are used for symbol versioning, which */ + case DT_VERDEF: /* this linker doesn't support yet. */ + case DT_VERDEFNUM: + case DT_VERNEED: + case DT_VERNEEDNUM: + /* Ignored */ + break; + default: + WARN("%s: dynamic header type #%" PRIxAddr" not handled", + GetPath(), dyn->d_tag); + } + } + + if (!buckets || !symnum) { + ERROR("%s: Missing or broken DT_HASH", GetPath()); + return false; + } + if (!strtab) { + ERROR("%s: Missing DT_STRTAB", GetPath()); + return false; + } + if (!symtab) { + ERROR("%s: Missing DT_SYMTAB", GetPath()); + return false; + } + + /* Load dependent libraries */ + for (size_t i = 0; i < dt_needed.size(); i++) { + const char *name = strtab.GetStringAt(dt_needed[i]); + RefPtr<LibHandle> handle = + ElfLoader::Singleton.Load(name, RTLD_GLOBAL | RTLD_LAZY, this); + if (!handle) + return false; + dependencies.push_back(handle); + } + + return true; +} + +bool +CustomElf::Relocate() +{ + DEBUG_LOG("Relocate %s @%p", GetPath(), static_cast<void *>(base)); + uint32_t symtab_index = (uint32_t) -1; + void *symptr = nullptr; + for (Array<Reloc>::iterator rel = relocations.begin(); + rel < relocations.end(); ++rel) { + /* Location of the relocation */ + void *ptr = GetPtr(rel->r_offset); + + /* R_*_RELATIVE relocations apply directly at the given location */ + if (ELF_R_TYPE(rel->r_info) == R_RELATIVE) { + *(void **) ptr = GetPtr(rel->GetAddend(base)); + continue; + } + /* Other relocation types need a symbol resolution */ + /* Avoid symbol resolution when it's the same symbol as last iteration */ + if (symtab_index != ELF_R_SYM(rel->r_info)) { + symtab_index = ELF_R_SYM(rel->r_info); + const Sym sym = symtab[symtab_index]; + if (sym.st_shndx != SHN_UNDEF) { + symptr = GetPtr(sym.st_value); + } else { + /* TODO: handle symbol resolving to nullptr vs. being undefined. */ + symptr = GetSymbolPtrInDeps(strtab.GetStringAt(sym.st_name)); + } + } + + if (symptr == nullptr) + WARN("%s: Relocation to NULL @0x%08" PRIxAddr, + GetPath(), rel->r_offset); + + /* Apply relocation */ + switch (ELF_R_TYPE(rel->r_info)) { + case R_GLOB_DAT: + /* R_*_GLOB_DAT relocations simply use the symbol value */ + *(void **) ptr = symptr; + break; + case R_ABS: + /* R_*_ABS* relocations add the relocation added to the symbol value */ + *(const char **) ptr = (const char *)symptr + rel->GetAddend(base); + break; + default: + ERROR("%s: Unsupported relocation type: 0x%" PRIxAddr, + GetPath(), ELF_R_TYPE(rel->r_info)); + return false; + } + } + return true; +} + +bool +CustomElf::RelocateJumps() +{ + /* TODO: Dynamic symbol resolution */ + for (Array<Reloc>::iterator rel = jumprels.begin(); + rel < jumprels.end(); ++rel) { + /* Location of the relocation */ + void *ptr = GetPtr(rel->r_offset); + + /* Only R_*_JMP_SLOT relocations are expected */ + if (ELF_R_TYPE(rel->r_info) != R_JMP_SLOT) { + ERROR("%s: Jump relocation type mismatch", GetPath()); + return false; + } + + /* TODO: Avoid code duplication with the relocations above */ + const Sym sym = symtab[ELF_R_SYM(rel->r_info)]; + void *symptr; + if (sym.st_shndx != SHN_UNDEF) + symptr = GetPtr(sym.st_value); + else + symptr = GetSymbolPtrInDeps(strtab.GetStringAt(sym.st_name)); + + if (symptr == nullptr) { + if (ELF_ST_BIND(sym.st_info) == STB_WEAK) { + WARN("%s: Relocation to NULL @0x%08" PRIxAddr " for symbol \"%s\"", + GetPath(), + rel->r_offset, strtab.GetStringAt(sym.st_name)); + } else { + ERROR("%s: Relocation to NULL @0x%08" PRIxAddr " for symbol \"%s\"", + GetPath(), + rel->r_offset, strtab.GetStringAt(sym.st_name)); + return false; + } + } + /* Apply relocation */ + *(void **) ptr = symptr; + } + return true; +} + +bool +CustomElf::CallInit() +{ + if (init) + CallFunction(init); + + for (Array<void *>::iterator it = init_array.begin(); + it < init_array.end(); ++it) { + /* Android x86 NDK wrongly puts 0xffffffff in INIT_ARRAY */ + if (*it && *it != reinterpret_cast<void *>(-1)) + CallFunction(*it); + } + initialized = true; + return true; +} + +void +CustomElf::CallFini() +{ + if (!initialized) + return; + for (Array<void *>::reverse_iterator it = fini_array.rbegin(); + it < fini_array.rend(); ++it) { + /* Android x86 NDK wrongly puts 0xffffffff in FINI_ARRAY */ + if (*it && *it != reinterpret_cast<void *>(-1)) + CallFunction(*it); + } + if (fini) + CallFunction(fini); +} + +Mappable * +CustomElf::GetMappable() const +{ + if (!mappable) + return nullptr; + if (mappable->GetKind() == Mappable::MAPPABLE_EXTRACT_FILE) + return mappable; + return ElfLoader::GetMappableFromPath(GetPath()); +} diff --git a/mozglue/linker/CustomElf.h b/mozglue/linker/CustomElf.h new file mode 100644 index 0000000000..5ecc7da085 --- /dev/null +++ b/mozglue/linker/CustomElf.h @@ -0,0 +1,159 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef CustomElf_h +#define CustomElf_h + +#include "ElfLoader.h" +#include "BaseElf.h" +#include "Logging.h" +#include "Elfxx.h" + +/** + * Library Handle class for ELF libraries we don't let the system linker + * handle. + */ +class CustomElf: public BaseElf, private ElfLoader::link_map +{ + friend class ElfLoader; + friend class SEGVHandler; +public: + /** + * Returns a new CustomElf using the given file descriptor to map ELF + * content. The file descriptor ownership is stolen, and it will be closed + * in CustomElf's destructor if an instance is created, or by the Load + * method otherwise. The path corresponds to the file descriptor, and flags + * are the same kind of flags that would be given to dlopen(), though + * currently, none are supported and the behaviour is more or less that of + * RTLD_GLOBAL | RTLD_BIND_NOW. + */ + static already_AddRefed<LibHandle> Load(Mappable *mappable, + const char *path, int flags); + + /** + * Inherited from LibHandle/BaseElf + */ + virtual ~CustomElf(); + +protected: + virtual Mappable *GetMappable() const; + +public: + /** + * Shows some stats about the Mappable instance. The when argument is to be + * used by the caller to give an identifier of the when the stats call is + * made. + */ + virtual void stats(const char *when) const; + + /** + * Returns the instance, casted as BaseElf. (short of a better way to do + * this without RTTI) + */ + virtual BaseElf *AsBaseElf() { return this; } + +private: + /** + * Scan dependent libraries to find the address corresponding to the + * given symbol name. This is used to find symbols that are undefined + * in the Elf object. + */ + void *GetSymbolPtrInDeps(const char *symbol) const; + + /** + * Private constructor + */ + CustomElf(Mappable *mappable, const char *path) + : BaseElf(path, mappable) + , link_map() + , init(0) + , fini(0) + , initialized(false) + , has_text_relocs(false) + { } + + /** + * Loads an Elf segment defined by the given PT_LOAD header. + * Returns whether this succeeded or failed. + */ + bool LoadSegment(const Elf::Phdr *pt_load) const; + + /** + * Initializes the library according to information found in the given + * PT_DYNAMIC header. + * Returns whether this succeeded or failed. + */ + bool InitDyn(const Elf::Phdr *pt_dyn); + + /** + * Apply .rel.dyn/.rela.dyn relocations. + * Returns whether this succeeded or failed. + */ + bool Relocate(); + + /** + * Apply .rel.plt/.rela.plt relocations. + * Returns whether this succeeded or failed. + */ + bool RelocateJumps(); + + /** + * Call initialization functions (.init/.init_array) + * Returns true; + */ + bool CallInit(); + + /** + * Call destructor functions (.fini_array/.fini) + * Returns whether this succeeded or failed. + */ + void CallFini(); + + /** + * Call a function given a pointer to its location. + */ + void CallFunction(void *ptr) const + { + /* C++ doesn't allow direct conversion between pointer-to-object + * and pointer-to-function. */ + union { + void *ptr; + void (*func)(void); + } f; + f.ptr = ptr; + DEBUG_LOG("%s: Calling function @%p", GetPath(), ptr); + f.func(); + } + + /** + * Call a function given a an address relative to the library base + */ + void CallFunction(Elf::Addr addr) const + { + return CallFunction(GetPtr(addr)); + } + + /* List of dependent libraries */ + std::vector<RefPtr<LibHandle> > dependencies; + + /* List of .rel.dyn/.rela.dyn relocations */ + Array<Elf::Reloc> relocations; + + /* List of .rel.plt/.rela.plt relocation */ + Array<Elf::Reloc> jumprels; + + /* Relative address of the initialization and destruction functions + * (.init/.fini) */ + Elf::Addr init, fini; + + /* List of initialization and destruction functions + * (.init_array/.fini_array) */ + Array<void *> init_array, fini_array; + + bool initialized; + + bool has_text_relocs; +}; + +#endif /* CustomElf_h */ diff --git a/mozglue/linker/ElfLoader.cpp b/mozglue/linker/ElfLoader.cpp new file mode 100644 index 0000000000..76225d1e77 --- /dev/null +++ b/mozglue/linker/ElfLoader.cpp @@ -0,0 +1,1310 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include <string> +#include <cstring> +#include <cstdlib> +#include <cstdio> +#include <dlfcn.h> +#include <unistd.h> +#include <errno.h> +#include <algorithm> +#include <fcntl.h> +#include "ElfLoader.h" +#include "BaseElf.h" +#include "CustomElf.h" +#include "Mappable.h" +#include "Logging.h" +#include <inttypes.h> + +#if defined(ANDROID) +#include <sys/syscall.h> + +#include <android/api-level.h> +#if __ANDROID_API__ < 8 +/* Android API < 8 doesn't provide sigaltstack */ + +extern "C" { + +inline int sigaltstack(const stack_t *ss, stack_t *oss) { + return syscall(__NR_sigaltstack, ss, oss); +} + +} /* extern "C" */ +#endif /* __ANDROID_API__ */ +#endif /* ANDROID */ + +#ifdef __ARM_EABI__ +extern "C" MOZ_EXPORT const void * +__gnu_Unwind_Find_exidx(void *pc, int *pcount) __attribute__((weak)); +#endif + +/* Pointer to the PT_DYNAMIC section of the executable or library + * containing this code. */ +extern "C" Elf::Dyn _DYNAMIC[]; + +using namespace mozilla; + +/** + * dlfcn.h replacements functions + */ + +void * +__wrap_dlopen(const char *path, int flags) +{ + RefPtr<LibHandle> handle = ElfLoader::Singleton.Load(path, flags); + if (handle) + handle->AddDirectRef(); + return handle; +} + +const char * +__wrap_dlerror(void) +{ + const char *error = ElfLoader::Singleton.lastError; + ElfLoader::Singleton.lastError = nullptr; + return error; +} + +void * +__wrap_dlsym(void *handle, const char *symbol) +{ + if (!handle) { + ElfLoader::Singleton.lastError = "dlsym(NULL, sym) unsupported"; + return nullptr; + } + if (handle != RTLD_DEFAULT && handle != RTLD_NEXT) { + LibHandle *h = reinterpret_cast<LibHandle *>(handle); + return h->GetSymbolPtr(symbol); + } + return dlsym(handle, symbol); +} + +int +__wrap_dlclose(void *handle) +{ + if (!handle) { + ElfLoader::Singleton.lastError = "No handle given to dlclose()"; + return -1; + } + reinterpret_cast<LibHandle *>(handle)->ReleaseDirectRef(); + return 0; +} + +int +__wrap_dladdr(void *addr, Dl_info *info) +{ + RefPtr<LibHandle> handle = ElfLoader::Singleton.GetHandleByPtr(addr); + if (!handle) { + return dladdr(addr, info); + } + info->dli_fname = handle->GetPath(); + info->dli_fbase = handle->GetBase(); + return 1; +} + +int +__wrap_dl_iterate_phdr(dl_phdr_cb callback, void *data) +{ + if (!ElfLoader::Singleton.dbg) + return -1; + + int pipefd[2]; + bool valid_pipe = (pipe(pipefd) == 0); + AutoCloseFD read_fd(pipefd[0]); + AutoCloseFD write_fd(pipefd[1]); + + for (ElfLoader::DebuggerHelper::iterator it = ElfLoader::Singleton.dbg.begin(); + it < ElfLoader::Singleton.dbg.end(); ++it) { + dl_phdr_info info; + info.dlpi_addr = reinterpret_cast<Elf::Addr>(it->l_addr); + info.dlpi_name = it->l_name; + info.dlpi_phdr = nullptr; + info.dlpi_phnum = 0; + + // Assuming l_addr points to Elf headers (in most cases, this is true), + // get the Phdr location from there. + // Unfortunately, when l_addr doesn't point to Elf headers, it may point + // to unmapped memory, or worse, unreadable memory. The only way to detect + // the latter without causing a SIGSEGV is to use the pointer in a system + // call that will try to read from there, and return an EFAULT error if + // it can't. One such system call is write(). It used to be possible to + // use a file descriptor on /dev/null for these kind of things, but recent + // Linux kernels never return an EFAULT error when using /dev/null. + // So instead, we use a self pipe. We do however need to read() from the + // read end of the pipe as well so as to not fill up the pipe buffer and + // block on subsequent writes. + // In the unlikely event reads from or write to the pipe fail for some + // other reason than EFAULT, we don't try any further and just skip setting + // the Phdr location for all subsequent libraries, rather than trying to + // start over with a new pipe. + int can_read = true; + if (valid_pipe) { + int ret; + char raw_ehdr[sizeof(Elf::Ehdr)]; + static_assert(sizeof(raw_ehdr) < PIPE_BUF, "PIPE_BUF is too small"); + do { + // writes are atomic when smaller than PIPE_BUF, per POSIX.1-2008. + ret = write(write_fd, it->l_addr, sizeof(raw_ehdr)); + } while (ret == -1 && errno == EINTR); + if (ret != sizeof(raw_ehdr)) { + if (ret == -1 && errno == EFAULT) { + can_read = false; + } else { + valid_pipe = false; + } + } else { + size_t nbytes = 0; + do { + // Per POSIX.1-2008, interrupted reads can return a length smaller + // than the given one instead of failing with errno EINTR. + ret = read(read_fd, raw_ehdr + nbytes, sizeof(raw_ehdr) - nbytes); + if (ret > 0) + nbytes += ret; + } while ((nbytes != sizeof(raw_ehdr) && ret > 0) || + (ret == -1 && errno == EINTR)); + if (nbytes != sizeof(raw_ehdr)) { + valid_pipe = false; + } + } + } + + if (valid_pipe && can_read) { + const Elf::Ehdr *ehdr = Elf::Ehdr::validate(it->l_addr); + if (ehdr) { + info.dlpi_phdr = reinterpret_cast<const Elf::Phdr *>( + reinterpret_cast<const char *>(ehdr) + ehdr->e_phoff); + info.dlpi_phnum = ehdr->e_phnum; + } + } + + int ret = callback(&info, sizeof(dl_phdr_info), data); + if (ret) + return ret; + } + return 0; +} + +#ifdef __ARM_EABI__ +const void * +__wrap___gnu_Unwind_Find_exidx(void *pc, int *pcount) +{ + RefPtr<LibHandle> handle = ElfLoader::Singleton.GetHandleByPtr(pc); + if (handle) + return handle->FindExidx(pcount); + if (__gnu_Unwind_Find_exidx) + return __gnu_Unwind_Find_exidx(pc, pcount); + *pcount = 0; + return nullptr; +} +#endif + +/** + * faulty.lib public API + */ + +MFBT_API size_t +__dl_get_mappable_length(void *handle) { + if (!handle) + return 0; + return reinterpret_cast<LibHandle *>(handle)->GetMappableLength(); +} + +MFBT_API void * +__dl_mmap(void *handle, void *addr, size_t length, off_t offset) +{ + if (!handle) + return nullptr; + return reinterpret_cast<LibHandle *>(handle)->MappableMMap(addr, length, + offset); +} + +MFBT_API void +__dl_munmap(void *handle, void *addr, size_t length) +{ + if (!handle) + return; + return reinterpret_cast<LibHandle *>(handle)->MappableMUnmap(addr, length); +} + +MFBT_API bool +IsSignalHandlingBroken() +{ + return ElfLoader::Singleton.isSignalHandlingBroken(); +} + +namespace { + +/** + * Returns the part after the last '/' for the given path + */ +const char * +LeafName(const char *path) +{ + const char *lastSlash = strrchr(path, '/'); + if (lastSlash) + return lastSlash + 1; + return path; +} + +} /* Anonymous namespace */ + +/** + * LibHandle + */ +LibHandle::~LibHandle() +{ + free(path); +} + +const char * +LibHandle::GetName() const +{ + return path ? LeafName(path) : nullptr; +} + +size_t +LibHandle::GetMappableLength() const +{ + if (!mappable) + mappable = GetMappable(); + if (!mappable) + return 0; + return mappable->GetLength(); +} + +void * +LibHandle::MappableMMap(void *addr, size_t length, off_t offset) const +{ + if (!mappable) + mappable = GetMappable(); + if (!mappable) + return MAP_FAILED; + void* mapped = mappable->mmap(addr, length, PROT_READ, MAP_PRIVATE, offset); + if (mapped != MAP_FAILED) { + /* Ensure the availability of all pages within the mapping */ + for (size_t off = 0; off < length; off += PageSize()) { + mappable->ensure(reinterpret_cast<char *>(mapped) + off); + } + } + return mapped; +} + +void +LibHandle::MappableMUnmap(void *addr, size_t length) const +{ + if (mappable) + mappable->munmap(addr, length); +} + +/** + * SystemElf + */ +already_AddRefed<LibHandle> +SystemElf::Load(const char *path, int flags) +{ + /* The Android linker returns a handle when the file name matches an + * already loaded library, even when the full path doesn't exist */ + if (path && path[0] == '/' && (access(path, F_OK) == -1)){ + DEBUG_LOG("dlopen(\"%s\", 0x%x) = %p", path, flags, (void *)nullptr); + return nullptr; + } + + void *handle = dlopen(path, flags); + DEBUG_LOG("dlopen(\"%s\", 0x%x) = %p", path, flags, handle); + ElfLoader::Singleton.lastError = dlerror(); + if (handle) { + SystemElf *elf = new SystemElf(path, handle); + ElfLoader::Singleton.Register(elf); + RefPtr<LibHandle> lib(elf); + return lib.forget(); + } + return nullptr; +} + +SystemElf::~SystemElf() +{ + if (!dlhandle) + return; + DEBUG_LOG("dlclose(%p [\"%s\"])", dlhandle, GetPath()); + dlclose(dlhandle); + ElfLoader::Singleton.lastError = dlerror(); + ElfLoader::Singleton.Forget(this); +} + +void * +SystemElf::GetSymbolPtr(const char *symbol) const +{ + void *sym = dlsym(dlhandle, symbol); + DEBUG_LOG("dlsym(%p [\"%s\"], \"%s\") = %p", dlhandle, GetPath(), symbol, sym); + ElfLoader::Singleton.lastError = dlerror(); + return sym; +} + +Mappable * +SystemElf::GetMappable() const +{ + const char *path = GetPath(); + if (!path) + return nullptr; +#ifdef ANDROID + /* On Android, if we don't have the full path, try in /system/lib */ + const char *name = LeafName(path); + std::string systemPath; + if (name == path) { + systemPath = "/system/lib/"; + systemPath += path; + path = systemPath.c_str(); + } +#endif + + return MappableFile::Create(path); +} + +#ifdef __ARM_EABI__ +const void * +SystemElf::FindExidx(int *pcount) const +{ + /* TODO: properly implement when ElfLoader::GetHandleByPtr + does return SystemElf handles */ + *pcount = 0; + return nullptr; +} +#endif + +/** + * ElfLoader + */ + +/* Unique ElfLoader instance */ +ElfLoader ElfLoader::Singleton; + +already_AddRefed<LibHandle> +ElfLoader::Load(const char *path, int flags, LibHandle *parent) +{ + /* Ensure logging is initialized or refresh if environment changed. */ + Logging::Init(); + + /* Ensure self_elf initialization. */ + if (!self_elf) + Init(); + + RefPtr<LibHandle> handle; + + /* Handle dlopen(nullptr) directly. */ + if (!path) { + handle = SystemElf::Load(nullptr, flags); + return handle.forget(); + } + + /* TODO: Handle relative paths correctly */ + const char *name = LeafName(path); + + /* Search the list of handles we already have for a match. When the given + * path is not absolute, compare file names, otherwise compare full paths. */ + if (name == path) { + for (LibHandleList::iterator it = handles.begin(); it < handles.end(); ++it) + if ((*it)->GetName() && (strcmp((*it)->GetName(), name) == 0)) { + handle = *it; + return handle.forget(); + } + } else { + for (LibHandleList::iterator it = handles.begin(); it < handles.end(); ++it) + if ((*it)->GetPath() && (strcmp((*it)->GetPath(), path) == 0)) { + handle = *it; + return handle.forget(); + } + } + + char *abs_path = nullptr; + const char *requested_path = path; + + /* When the path is not absolute and the library is being loaded for + * another, first try to load the library from the directory containing + * that parent library. */ + if ((name == path) && parent) { + const char *parentPath = parent->GetPath(); + abs_path = new char[strlen(parentPath) + strlen(path)]; + strcpy(abs_path, parentPath); + char *slash = strrchr(abs_path, '/'); + strcpy(slash + 1, path); + path = abs_path; + } + + Mappable *mappable = GetMappableFromPath(path); + + /* Try loading with the custom linker if we have a Mappable */ + if (mappable) + handle = CustomElf::Load(mappable, path, flags); + + /* Try loading with the system linker if everything above failed */ + if (!handle) + handle = SystemElf::Load(path, flags); + + /* If we didn't have an absolute path and haven't been able to load + * a library yet, try in the system search path */ + if (!handle && abs_path) + handle = SystemElf::Load(name, flags); + + delete [] abs_path; + DEBUG_LOG("ElfLoader::Load(\"%s\", 0x%x, %p [\"%s\"]) = %p", requested_path, flags, + reinterpret_cast<void *>(parent), parent ? parent->GetPath() : "", + static_cast<void *>(handle)); + + return handle.forget(); +} + +already_AddRefed<LibHandle> +ElfLoader::GetHandleByPtr(void *addr) +{ + /* Scan the list of handles we already have for a match */ + for (LibHandleList::iterator it = handles.begin(); it < handles.end(); ++it) { + if ((*it)->Contains(addr)) { + RefPtr<LibHandle> lib = *it; + return lib.forget(); + } + } + return nullptr; +} + +Mappable * +ElfLoader::GetMappableFromPath(const char *path) +{ + const char *name = LeafName(path); + Mappable *mappable = nullptr; + RefPtr<Zip> zip; + const char *subpath; + if ((subpath = strchr(path, '!'))) { + char *zip_path = strndup(path, subpath - path); + while (*(++subpath) == '/') { } + zip = ZipCollection::GetZip(zip_path); + free(zip_path); + Zip::Stream s; + if (zip && zip->GetStream(subpath, &s)) { + /* When the MOZ_LINKER_EXTRACT environment variable is set to "1", + * compressed libraries are going to be (temporarily) extracted as + * files, in the directory pointed by the MOZ_LINKER_CACHE + * environment variable. */ + const char *extract = getenv("MOZ_LINKER_EXTRACT"); + if (extract && !strncmp(extract, "1", 2 /* Including '\0' */)) + mappable = MappableExtractFile::Create(name, zip, &s); + if (!mappable) { + if (s.GetType() == Zip::Stream::DEFLATE) { + mappable = MappableDeflate::Create(name, zip, &s); + } else if (s.GetType() == Zip::Stream::STORE) { + mappable = MappableSeekableZStream::Create(name, zip, &s); + } + } + } + } + /* If we couldn't load above, try with a MappableFile */ + if (!mappable && !zip) + mappable = MappableFile::Create(path); + + return mappable; +} + +void +ElfLoader::Register(LibHandle *handle) +{ + handles.push_back(handle); +} + +void +ElfLoader::Register(CustomElf *handle) +{ + Register(static_cast<LibHandle *>(handle)); + if (dbg) { + dbg.Add(handle); + } +} + +void +ElfLoader::Forget(LibHandle *handle) +{ + /* Ensure logging is initialized or refresh if environment changed. */ + Logging::Init(); + + LibHandleList::iterator it = std::find(handles.begin(), handles.end(), handle); + if (it != handles.end()) { + DEBUG_LOG("ElfLoader::Forget(%p [\"%s\"])", reinterpret_cast<void *>(handle), + handle->GetPath()); + handles.erase(it); + } else { + DEBUG_LOG("ElfLoader::Forget(%p [\"%s\"]): Handle not found", + reinterpret_cast<void *>(handle), handle->GetPath()); + } +} + +void +ElfLoader::Forget(CustomElf *handle) +{ + Forget(static_cast<LibHandle *>(handle)); + if (dbg) { + dbg.Remove(handle); + } +} + +void +ElfLoader::Init() +{ + Dl_info info; + /* On Android < 4.1 can't reenter dl* functions. So when the library + * containing this code is dlopen()ed, it can't call dladdr from a + * static initializer. */ + if (dladdr(_DYNAMIC, &info) != 0) { + self_elf = LoadedElf::Create(info.dli_fname, info.dli_fbase); + } +#if defined(ANDROID) + if (dladdr(FunctionPtr(syscall), &info) != 0) { + libc = LoadedElf::Create(info.dli_fname, info.dli_fbase); + } +#endif +} + +ElfLoader::~ElfLoader() +{ + LibHandleList list; + + if (!Singleton.IsShutdownExpected()) { + MOZ_CRASH("Unexpected shutdown"); + } + + /* Release self_elf and libc */ + self_elf = nullptr; +#if defined(ANDROID) + libc = nullptr; +#endif + + /* Build up a list of all library handles with direct (external) references. + * We actually skip system library handles because we want to keep at least + * some of these open. Most notably, Mozilla codebase keeps a few libgnome + * libraries deliberately open because of the mess that libORBit destruction + * is. dlclose()ing these libraries actually leads to problems. */ + for (LibHandleList::reverse_iterator it = handles.rbegin(); + it < handles.rend(); ++it) { + if ((*it)->DirectRefCount()) { + if (SystemElf *se = (*it)->AsSystemElf()) { + se->Forget(); + } else { + list.push_back(*it); + } + } + } + /* Force release all external references to the handles collected above */ + for (LibHandleList::iterator it = list.begin(); it < list.end(); ++it) { + while ((*it)->ReleaseDirectRef()) { } + } + /* Remove the remaining system handles. */ + if (handles.size()) { + list = handles; + for (LibHandleList::reverse_iterator it = list.rbegin(); + it < list.rend(); ++it) { + if ((*it)->AsSystemElf()) { + DEBUG_LOG("ElfLoader::~ElfLoader(): Remaining handle for \"%s\" " + "[%d direct refs, %d refs total]", (*it)->GetPath(), + (*it)->DirectRefCount(), (*it)->refCount()); + } else { + DEBUG_LOG("ElfLoader::~ElfLoader(): Unexpected remaining handle for \"%s\" " + "[%d direct refs, %d refs total]", (*it)->GetPath(), + (*it)->DirectRefCount(), (*it)->refCount()); + /* Not removing, since it could have references to other libraries, + * destroying them as a side effect, and possibly leaving dangling + * pointers in the handle list we're scanning */ + } + } + } +} + +void +ElfLoader::stats(const char *when) +{ + if (MOZ_LIKELY(!Logging::isVerbose())) + return; + + for (LibHandleList::iterator it = Singleton.handles.begin(); + it < Singleton.handles.end(); ++it) + (*it)->stats(when); +} + +#ifdef __ARM_EABI__ +int +ElfLoader::__wrap_aeabi_atexit(void *that, ElfLoader::Destructor destructor, + void *dso_handle) +{ + Singleton.destructors.push_back( + DestructorCaller(destructor, that, dso_handle)); + return 0; +} +#else +int +ElfLoader::__wrap_cxa_atexit(ElfLoader::Destructor destructor, void *that, + void *dso_handle) +{ + Singleton.destructors.push_back( + DestructorCaller(destructor, that, dso_handle)); + return 0; +} +#endif + +void +ElfLoader::__wrap_cxa_finalize(void *dso_handle) +{ + /* Call all destructors for the given DSO handle in reverse order they were + * registered. */ + std::vector<DestructorCaller>::reverse_iterator it; + for (it = Singleton.destructors.rbegin(); + it < Singleton.destructors.rend(); ++it) { + if (it->IsForHandle(dso_handle)) { + it->Call(); + } + } +} + +void +ElfLoader::DestructorCaller::Call() +{ + if (destructor) { + DEBUG_LOG("ElfLoader::DestructorCaller::Call(%p, %p, %p)", + FunctionPtr(destructor), object, dso_handle); + destructor(object); + destructor = nullptr; + } +} + +ElfLoader::DebuggerHelper::DebuggerHelper(): dbg(nullptr), firstAdded(nullptr) +{ + /* Find ELF auxiliary vectors. + * + * The kernel stores the following data on the stack when starting a + * program: + * argc + * argv[0] (pointer into argv strings defined below) + * argv[1] (likewise) + * ... + * argv[argc - 1] (likewise) + * nullptr + * envp[0] (pointer into environment strings defined below) + * envp[1] (likewise) + * ... + * envp[n] (likewise) + * nullptr + * ... (more NULLs on some platforms such as Android 4.3) + * auxv[0] (first ELF auxiliary vector) + * auxv[1] (second ELF auxiliary vector) + * ... + * auxv[p] (last ELF auxiliary vector) + * (AT_NULL, nullptr) + * padding + * argv strings, separated with '\0' + * environment strings, separated with '\0' + * nullptr + * + * What we are after are the auxv values defined by the following struct. + */ + struct AuxVector { + Elf::Addr type; + Elf::Addr value; + }; + + /* Pointer to the environment variables list */ + extern char **environ; + + /* The environment may have changed since the program started, in which + * case the environ variables list isn't the list the kernel put on stack + * anymore. But in this new list, variables that didn't change still point + * to the strings the kernel put on stack. It is quite unlikely that two + * modified environment variables point to two consecutive strings in memory, + * so we assume that if two consecutive environment variables point to two + * consecutive strings, we found strings the kernel put on stack. */ + char **env; + for (env = environ; *env; env++) + if (*env + strlen(*env) + 1 == env[1]) + break; + if (!*env) + return; + + /* Next, we scan the stack backwards to find a pointer to one of those + * strings we found above, which will give us the location of the original + * envp list. As we are looking for pointers, we need to look at 32-bits or + * 64-bits aligned values, depening on the architecture. */ + char **scan = reinterpret_cast<char **>( + reinterpret_cast<uintptr_t>(*env) & ~(sizeof(void *) - 1)); + while (*env != *scan) + scan--; + + /* Finally, scan forward to find the last environment variable pointer and + * thus the first auxiliary vector. */ + while (*scan++); + + /* Some platforms have more NULLs here, so skip them if we encounter them */ + while (!*scan) + scan++; + + AuxVector *auxv = reinterpret_cast<AuxVector *>(scan); + + /* The two values of interest in the auxiliary vectors are AT_PHDR and + * AT_PHNUM, which gives us the the location and size of the ELF program + * headers. */ + Array<Elf::Phdr> phdrs; + char *base = nullptr; + while (auxv->type) { + if (auxv->type == AT_PHDR) { + phdrs.Init(reinterpret_cast<Elf::Phdr*>(auxv->value)); + /* Assume the base address is the first byte of the same page */ + base = reinterpret_cast<char *>(PageAlignedPtr(auxv->value)); + } + if (auxv->type == AT_PHNUM) + phdrs.Init(auxv->value); + auxv++; + } + + if (!phdrs) { + DEBUG_LOG("Couldn't find program headers"); + return; + } + + /* In some cases, the address for the program headers we get from the + * auxiliary vectors is not mapped, because of the PT_LOAD segments + * definitions in the program executable. Trying to map anonymous memory + * with a hint giving the base address will return a different address + * if something is mapped there, and the base address otherwise. */ + MappedPtr mem(MemoryRange::mmap(base, PageSize(), PROT_NONE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0)); + if (mem == base) { + /* If program headers aren't mapped, try to map them */ + int fd = open("/proc/self/exe", O_RDONLY); + if (fd == -1) { + DEBUG_LOG("Failed to open /proc/self/exe"); + return; + } + mem.Assign(MemoryRange::mmap(base, PageSize(), PROT_READ, MAP_PRIVATE, + fd, 0)); + /* If we don't manage to map at the right address, just give up. */ + if (mem != base) { + DEBUG_LOG("Couldn't read program headers"); + return; + } + } + /* Sanity check: the first bytes at the base address should be an ELF + * header. */ + if (!Elf::Ehdr::validate(base)) { + DEBUG_LOG("Couldn't find program base"); + return; + } + + /* Search for the program PT_DYNAMIC segment */ + Array<Elf::Dyn> dyns; + for (Array<Elf::Phdr>::iterator phdr = phdrs.begin(); phdr < phdrs.end(); + ++phdr) { + /* While the program headers are expected within the first mapped page of + * the program executable, the executable PT_LOADs may actually make them + * loaded at an address that is not the wanted base address of the + * library. We thus need to adjust the base address, compensating for the + * virtual address of the PT_LOAD segment corresponding to offset 0. */ + if (phdr->p_type == PT_LOAD && phdr->p_offset == 0) + base -= phdr->p_vaddr; + if (phdr->p_type == PT_DYNAMIC) + dyns.Init(base + phdr->p_vaddr, phdr->p_filesz); + } + if (!dyns) { + DEBUG_LOG("Failed to find PT_DYNAMIC section in program"); + return; + } + + /* Search for the DT_DEBUG information */ + for (Array<Elf::Dyn>::iterator dyn = dyns.begin(); dyn < dyns.end(); ++dyn) { + if (dyn->d_tag == DT_DEBUG) { + dbg = reinterpret_cast<r_debug *>(dyn->d_un.d_ptr); + break; + } + } + DEBUG_LOG("DT_DEBUG points at %p", static_cast<void *>(dbg)); +} + +/** + * Helper class to ensure the given pointer is writable within the scope of + * an instance. Permissions to the memory page where the pointer lies are + * restored to their original value when the instance is destroyed. + */ +class EnsureWritable +{ +public: + template <typename T> + EnsureWritable(T *ptr, size_t length_ = sizeof(T)) + { + MOZ_ASSERT(length_ < PageSize()); + prot = -1; + page = MAP_FAILED; + + char *firstPage = PageAlignedPtr(reinterpret_cast<char *>(ptr)); + char *lastPageEnd = PageAlignedEndPtr(reinterpret_cast<char *>(ptr) + length_); + length = lastPageEnd - firstPage; + uintptr_t start = reinterpret_cast<uintptr_t>(firstPage); + uintptr_t end; + + prot = getProt(start, &end); + if (prot == -1 || (start + length) > end) + MOZ_CRASH(); + + if (prot & PROT_WRITE) + return; + + page = firstPage; + mprotect(page, length, prot | PROT_WRITE); + } + + ~EnsureWritable() + { + if (page != MAP_FAILED) { + mprotect(page, length, prot); +} + } + +private: + int getProt(uintptr_t addr, uintptr_t *end) + { + /* The interesting part of the /proc/self/maps format looks like: + * startAddr-endAddr rwxp */ + int result = 0; + AutoCloseFILE f(fopen("/proc/self/maps", "r")); + while (f) { + unsigned long long startAddr, endAddr; + char perms[5]; + if (fscanf(f, "%llx-%llx %4s %*1024[^\n] ", &startAddr, &endAddr, perms) != 3) + return -1; + if (addr < startAddr || addr >= endAddr) + continue; + if (perms[0] == 'r') + result |= PROT_READ; + else if (perms[0] != '-') + return -1; + if (perms[1] == 'w') + result |= PROT_WRITE; + else if (perms[1] != '-') + return -1; + if (perms[2] == 'x') + result |= PROT_EXEC; + else if (perms[2] != '-') + return -1; + *end = endAddr; + return result; + } + return -1; + } + + int prot; + void *page; + size_t length; +}; + +/** + * The system linker maintains a doubly linked list of library it loads + * for use by the debugger. Unfortunately, it also uses the list pointers + * in a lot of operations and adding our data in the list is likely to + * trigger crashes when the linker tries to use data we don't provide or + * that fall off the amount data we allocated. Fortunately, the linker only + * traverses the list forward and accesses the head of the list from a + * private pointer instead of using the value in the r_debug structure. + * This means we can safely add members at the beginning of the list. + * Unfortunately, gdb checks the coherency of l_prev values, so we have + * to adjust the l_prev value for the first element the system linker + * knows about. Fortunately, it doesn't use l_prev, and the first element + * is not ever going to be released before our elements, since it is the + * program executable, so the system linker should not be changing + * r_debug::r_map. + */ +void +ElfLoader::DebuggerHelper::Add(ElfLoader::link_map *map) +{ + if (!dbg->r_brk) + return; + dbg->r_state = r_debug::RT_ADD; + dbg->r_brk(); + map->l_prev = nullptr; + map->l_next = dbg->r_map; + if (!firstAdded) { + firstAdded = map; + /* When adding a library for the first time, r_map points to data + * handled by the system linker, and that data may be read-only */ + EnsureWritable w(&dbg->r_map->l_prev); + dbg->r_map->l_prev = map; + } else + dbg->r_map->l_prev = map; + dbg->r_map = map; + dbg->r_state = r_debug::RT_CONSISTENT; + dbg->r_brk(); +} + +void +ElfLoader::DebuggerHelper::Remove(ElfLoader::link_map *map) +{ + if (!dbg->r_brk) + return; + dbg->r_state = r_debug::RT_DELETE; + dbg->r_brk(); + if (dbg->r_map == map) + dbg->r_map = map->l_next; + else if (map->l_prev) { + map->l_prev->l_next = map->l_next; + } + if (map == firstAdded) { + firstAdded = map->l_prev; + /* When removing the first added library, its l_next is going to be + * data handled by the system linker, and that data may be read-only */ + EnsureWritable w(&map->l_next->l_prev); + map->l_next->l_prev = map->l_prev; + } else if (map->l_next) { + map->l_next->l_prev = map->l_prev; + } + dbg->r_state = r_debug::RT_CONSISTENT; + dbg->r_brk(); +} + +#if defined(ANDROID) +/* As some system libraries may be calling signal() or sigaction() to + * set a SIGSEGV handler, effectively breaking MappableSeekableZStream, + * or worse, restore our SIGSEGV handler with wrong flags (which using + * signal() will do), we want to hook into the system's sigaction() to + * replace it with our own wrapper instead, so that our handler is never + * replaced. We used to only do that with libraries this linker loads, + * but it turns out at least one system library does call signal() and + * breaks us (libsc-a3xx.so on the Samsung Galaxy S4). + * As libc's signal (bsd_signal/sysv_signal, really) calls sigaction + * under the hood, instead of calling the signal system call directly, + * we only need to hook sigaction. This is true for both bionic and + * glibc. + */ + +/* libc's sigaction */ +extern "C" int +sigaction(int signum, const struct sigaction *act, + struct sigaction *oldact); + +/* Simple reimplementation of sigaction. This is roughly equivalent + * to the assembly that comes in bionic, but not quite equivalent to + * glibc's implementation, so we only use this on Android. */ +int +sys_sigaction(int signum, const struct sigaction *act, + struct sigaction *oldact) +{ + return syscall(__NR_sigaction, signum, act, oldact); +} + +/* Replace the first instructions of the given function with a jump + * to the given new function. */ +template <typename T> +static bool +Divert(T func, T new_func) +{ + void *ptr = FunctionPtr(func); + uintptr_t addr = reinterpret_cast<uintptr_t>(ptr); + +#if defined(__i386__) + // A 32-bit jump is a 5 bytes instruction. + EnsureWritable w(ptr, 5); + *reinterpret_cast<unsigned char *>(addr) = 0xe9; // jmp + *reinterpret_cast<intptr_t *>(addr + 1) = + reinterpret_cast<uintptr_t>(new_func) - addr - 5; // target displacement + return true; +#elif defined(__arm__) + const unsigned char trampoline[] = { + // .thumb + 0x46, 0x04, // nop + 0x78, 0x47, // bx pc + 0x46, 0x04, // nop + // .arm + 0x04, 0xf0, 0x1f, 0xe5, // ldr pc, [pc, #-4] + // .word <new_func> + }; + const unsigned char *start; + if (addr & 0x01) { + /* Function is thumb, the actual address of the code is without the + * least significant bit. */ + addr--; + /* The arm part of the trampoline needs to be 32-bit aligned */ + if (addr & 0x02) + start = trampoline; + else + start = trampoline + 2; + } else { + /* Function is arm, we only need the arm part of the trampoline */ + start = trampoline + 6; + } + + size_t len = sizeof(trampoline) - (start - trampoline); + EnsureWritable w(reinterpret_cast<void *>(addr), len + sizeof(void *)); + memcpy(reinterpret_cast<void *>(addr), start, len); + *reinterpret_cast<void **>(addr + len) = FunctionPtr(new_func); + cacheflush(addr, addr + len + sizeof(void *), 0); + return true; +#else + return false; +#endif +} +#else +#define sys_sigaction sigaction +template <typename T> +static bool +Divert(T func, T new_func) +{ + return false; +} +#endif + +namespace { + +/* Clock that only accounts for time spent in the current process. */ +static uint64_t ProcessTimeStamp_Now() +{ + struct timespec ts; + int rv = clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts); + + if (rv != 0) { + return 0; + } + + uint64_t baseNs = (uint64_t)ts.tv_sec * 1000000000; + return baseNs + (uint64_t)ts.tv_nsec; +} + +} + +/* Data structure used to pass data to the temporary signal handler, + * as well as triggering a test crash. */ +struct TmpData { + volatile int crash_int; + volatile uint64_t crash_timestamp; +}; + +SEGVHandler::SEGVHandler() +: initialized(false), registeredHandler(false), signalHandlingBroken(true) +, signalHandlingSlow(true) +{ + /* Ensure logging is initialized before the DEBUG_LOG in the test_handler. + * As this constructor runs before the ElfLoader constructor (by effect + * of ElfLoader inheriting from this class), this also initializes on behalf + * of ElfLoader and DebuggerHelper. */ + Logging::Init(); + + /* Initialize oldStack.ss_flags to an invalid value when used to set + * an alternative stack, meaning we haven't got information about the + * original alternative stack and thus don't mean to restore it in + * the destructor. */ + oldStack.ss_flags = SS_ONSTACK; + + /* Get the current segfault signal handler. */ + struct sigaction old_action; + sys_sigaction(SIGSEGV, nullptr, &old_action); + + /* Some devices don't provide useful information to their SIGSEGV handlers, + * making it impossible for on-demand decompression to work. To check if + * we're on such a device, setup a temporary handler and deliberately + * trigger a segfault. The handler will set signalHandlingBroken if the + * provided information is bogus. + * Some other devices have a kernel option enabled that makes SIGSEGV handler + * have an overhead so high that it affects how on-demand decompression + * performs. The handler will also set signalHandlingSlow if the triggered + * SIGSEGV took too much time. */ + struct sigaction action; + action.sa_sigaction = &SEGVHandler::test_handler; + sigemptyset(&action.sa_mask); + action.sa_flags = SA_SIGINFO | SA_NODEFER; + action.sa_restorer = nullptr; + stackPtr.Assign(MemoryRange::mmap(nullptr, PageSize(), + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0)); + if (stackPtr.get() == MAP_FAILED) + return; + if (sys_sigaction(SIGSEGV, &action, nullptr)) + return; + + TmpData *data = reinterpret_cast<TmpData*>(stackPtr.get()); + data->crash_timestamp = ProcessTimeStamp_Now(); + mprotect(stackPtr, stackPtr.GetLength(), PROT_NONE); + data->crash_int = 123; + /* Restore the original segfault signal handler. */ + sys_sigaction(SIGSEGV, &old_action, nullptr); + stackPtr.Assign(MAP_FAILED, 0); +} + +void +SEGVHandler::FinishInitialization() +{ + /* Ideally, we'd need some locking here, but in practice, we're not + * going to race with another thread. */ + initialized = true; + + if (signalHandlingBroken || signalHandlingSlow) + return; + + typedef int (*sigaction_func)(int, const struct sigaction *, + struct sigaction *); + + sigaction_func libc_sigaction; + +#if defined(ANDROID) + /* Android > 4.4 comes with a sigaction wrapper in a LD_PRELOADed library + * (libsigchain) for ART. That wrapper kind of does the same trick as we + * do, so we need extra care in handling it. + * - Divert the libc's sigaction, assuming the LD_PRELOADed library uses + * it under the hood (which is more or less true according to the source + * of that library, since it's doing a lookup in RTLD_NEXT) + * - With the LD_PRELOADed library in place, all calls to sigaction from + * from system libraries will go to the LD_PRELOADed library. + * - The LD_PRELOADed library calls to sigaction go to our __wrap_sigaction. + * - The calls to sigaction from libraries faulty.lib loads are sent to + * the LD_PRELOADed library. + * In practice, for signal handling, this means: + * - The signal handler registered to the kernel is ours. + * - Our handler redispatches to the LD_PRELOADed library's if there's a + * segfault we don't handle. + * - The LD_PRELOADed library redispatches according to whatever system + * library or faulty.lib-loaded library set with sigaction. + * + * When there is no sigaction wrapper in place: + * - Divert the libc's sigaction. + * - Calls to sigaction from system library and faulty.lib-loaded libraries + * all go to the libc's sigaction, which end up in our __wrap_sigaction. + * - The signal handler registered to the kernel is ours. + * - Our handler redispatches according to whatever system library or + * faulty.lib-loaded library set with sigaction. + */ + void *libc = dlopen("libc.so", RTLD_GLOBAL | RTLD_LAZY); + if (libc) { + /* + * Lollipop bionic only has a small trampoline in sigaction, with the real + * work happening in __sigaction. Divert there instead of sigaction if it exists. + * Bug 1154803 + */ + libc_sigaction = reinterpret_cast<sigaction_func>(dlsym(libc, "__sigaction")); + + if (!libc_sigaction) { + libc_sigaction = + reinterpret_cast<sigaction_func>(dlsym(libc, "sigaction")); + } + } else +#endif + { + libc_sigaction = sigaction; + } + + if (!Divert(libc_sigaction, __wrap_sigaction)) + return; + + /* Setup an alternative stack if the already existing one is not big + * enough, or if there is none. */ + if (sigaltstack(nullptr, &oldStack) == 0) { + if (oldStack.ss_flags == SS_ONSTACK) + oldStack.ss_flags = 0; + if (!oldStack.ss_sp || oldStack.ss_size < stackSize) { + stackPtr.Assign(MemoryRange::mmap(nullptr, stackSize, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0)); + if (stackPtr.get() == MAP_FAILED) + return; + stack_t stack; + stack.ss_sp = stackPtr; + stack.ss_size = stackSize; + stack.ss_flags = 0; + if (sigaltstack(&stack, nullptr) != 0) + return; + } + } + /* Register our own handler, and store the already registered one in + * SEGVHandler's struct sigaction member */ + action.sa_sigaction = &SEGVHandler::handler; + action.sa_flags = SA_SIGINFO | SA_NODEFER | SA_ONSTACK; + registeredHandler = !sys_sigaction(SIGSEGV, &action, &this->action); +} + +SEGVHandler::~SEGVHandler() +{ + /* Restore alternative stack for signals */ + if (oldStack.ss_flags != SS_ONSTACK) + sigaltstack(&oldStack, nullptr); + /* Restore original signal handler */ + if (registeredHandler) + sys_sigaction(SIGSEGV, &this->action, nullptr); +} + +/* Test handler for a deliberately triggered SIGSEGV that determines whether + * useful information is provided to signal handlers, particularly whether + * si_addr is filled in properly, and whether the segfault handler is called + * quickly enough. */ +void SEGVHandler::test_handler(int signum, siginfo_t *info, void *context) +{ + SEGVHandler &that = ElfLoader::Singleton; + if (signum == SIGSEGV && info && + info->si_addr == that.stackPtr.get()) + that.signalHandlingBroken = false; + mprotect(that.stackPtr, that.stackPtr.GetLength(), PROT_READ | PROT_WRITE); + TmpData *data = reinterpret_cast<TmpData*>(that.stackPtr.get()); + uint64_t latency = ProcessTimeStamp_Now() - data->crash_timestamp; + DEBUG_LOG("SEGVHandler latency: %" PRIu64, latency); + /* See bug 886736 for timings on different devices, 150 µs is reasonably above + * the latency on "working" devices and seems to be short enough to not incur + * a huge overhead to on-demand decompression. */ + if (latency <= 150000) + that.signalHandlingSlow = false; +} + +/* TODO: "properly" handle signal masks and flags */ +void SEGVHandler::handler(int signum, siginfo_t *info, void *context) +{ + //ASSERT(signum == SIGSEGV); + DEBUG_LOG("Caught segmentation fault @%p", info->si_addr); + + /* Check whether we segfaulted in the address space of a CustomElf. We're + * only expecting that to happen as an access error. */ + if (info->si_code == SEGV_ACCERR) { + RefPtr<LibHandle> handle = + ElfLoader::Singleton.GetHandleByPtr(info->si_addr); + BaseElf *elf; + if (handle && (elf = handle->AsBaseElf())) { + DEBUG_LOG("Within the address space of %s", handle->GetPath()); + if (elf->mappable && elf->mappable->ensure(info->si_addr)) { + return; + } + } + } + + /* Redispatch to the registered handler */ + SEGVHandler &that = ElfLoader::Singleton; + if (that.action.sa_flags & SA_SIGINFO) { + DEBUG_LOG("Redispatching to registered handler @%p", + FunctionPtr(that.action.sa_sigaction)); + that.action.sa_sigaction(signum, info, context); + } else if (that.action.sa_handler == SIG_DFL) { + DEBUG_LOG("Redispatching to default handler"); + /* Reset the handler to the default one, and trigger it. */ + sys_sigaction(signum, &that.action, nullptr); + raise(signum); + } else if (that.action.sa_handler != SIG_IGN) { + DEBUG_LOG("Redispatching to registered handler @%p", + FunctionPtr(that.action.sa_handler)); + that.action.sa_handler(signum); + } else { + DEBUG_LOG("Ignoring"); + } +} + +int +SEGVHandler::__wrap_sigaction(int signum, const struct sigaction *act, + struct sigaction *oldact) +{ + SEGVHandler &that = ElfLoader::Singleton; + + /* Use system sigaction() function for all but SIGSEGV signals. */ + if (!that.registeredHandler || (signum != SIGSEGV)) + return sys_sigaction(signum, act, oldact); + + if (oldact) + *oldact = that.action; + if (act) + that.action = *act; + return 0; +} + +Logging Logging::Singleton; diff --git a/mozglue/linker/ElfLoader.h b/mozglue/linker/ElfLoader.h new file mode 100644 index 0000000000..0d26a011ee --- /dev/null +++ b/mozglue/linker/ElfLoader.h @@ -0,0 +1,674 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef ElfLoader_h +#define ElfLoader_h + +#include <vector> +#include <dlfcn.h> +#include <signal.h> +#include "mozilla/RefCounted.h" +#include "mozilla/RefPtr.h" +#include "mozilla/UniquePtr.h" +#include "Zip.h" +#include "Elfxx.h" +#include "Mappable.h" + +/** + * dlfcn.h replacement functions + */ +extern "C" { + void *__wrap_dlopen(const char *path, int flags); + const char *__wrap_dlerror(void); + void *__wrap_dlsym(void *handle, const char *symbol); + int __wrap_dlclose(void *handle); + +#ifndef HAVE_DLADDR + typedef struct { + const char *dli_fname; + void *dli_fbase; + const char *dli_sname; + void *dli_saddr; + } Dl_info; +#endif + int __wrap_dladdr(void *addr, Dl_info *info); + + struct dl_phdr_info { + Elf::Addr dlpi_addr; + const char *dlpi_name; + const Elf::Phdr *dlpi_phdr; + Elf::Half dlpi_phnum; + }; + + typedef int (*dl_phdr_cb)(struct dl_phdr_info *, size_t, void *); + int __wrap_dl_iterate_phdr(dl_phdr_cb callback, void *data); + +#ifdef __ARM_EABI__ + const void *__wrap___gnu_Unwind_Find_exidx(void *pc, int *pcount); +#endif + +/** + * faulty.lib public API + */ +MFBT_API size_t +__dl_get_mappable_length(void *handle); + +MFBT_API void * +__dl_mmap(void *handle, void *addr, size_t length, off_t offset); + +MFBT_API void +__dl_munmap(void *handle, void *addr, size_t length); + +MFBT_API bool +IsSignalHandlingBroken(); + +} + +/* Forward declarations for use in LibHandle */ +class BaseElf; +class CustomElf; +class SystemElf; + +/** + * Specialize RefCounted template for LibHandle. We may get references to + * LibHandles during the execution of their destructor, so we need + * RefCounted<LibHandle>::Release to support some reentrancy. See further + * below. + */ +class LibHandle; + +namespace mozilla { +namespace detail { + +template <> inline void RefCounted<LibHandle, AtomicRefCount>::Release() const; + +template <> inline RefCounted<LibHandle, AtomicRefCount>::~RefCounted() +{ + MOZ_ASSERT(mRefCnt == 0x7fffdead); +} + +} /* namespace detail */ +} /* namespace mozilla */ + +/** + * Abstract class for loaded libraries. Libraries may be loaded through the + * system linker or this linker, both cases will be derived from this class. + */ +class LibHandle: public mozilla::external::AtomicRefCounted<LibHandle> +{ +public: + MOZ_DECLARE_REFCOUNTED_TYPENAME(LibHandle) + /** + * Constructor. Takes the path of the loaded library and will store a copy + * of the leaf name. + */ + LibHandle(const char *path) + : directRefCnt(0), path(path ? strdup(path) : nullptr), mappable(nullptr) { } + + /** + * Destructor. + */ + virtual ~LibHandle(); + + /** + * Returns the pointer to the address to which the given symbol resolves + * inside the library. It is not supposed to resolve the symbol in other + * libraries, although in practice, it will for system libraries. + */ + virtual void *GetSymbolPtr(const char *symbol) const = 0; + + /** + * Returns whether the given address is part of the virtual address space + * covered by the loaded library. + */ + virtual bool Contains(void *addr) const = 0; + + /** + * Returns the base address of the loaded library. + */ + virtual void *GetBase() const = 0; + + /** + * Returns the file name of the library without the containing directory. + */ + const char *GetName() const; + + /** + * Returns the full path of the library, when available. Otherwise, returns + * the file name. + */ + const char *GetPath() const + { + return path; + } + + /** + * Library handles can be referenced from other library handles or + * externally (when dlopen()ing using this linker). We need to be + * able to distinguish between the two kind of referencing for better + * bookkeeping. + */ + void AddDirectRef() + { + ++directRefCnt; + mozilla::external::AtomicRefCounted<LibHandle>::AddRef(); + } + + /** + * Releases a direct reference, and returns whether there are any direct + * references left. + */ + bool ReleaseDirectRef() + { + bool ret = false; + if (directRefCnt) { + MOZ_ASSERT(directRefCnt <= + mozilla::external::AtomicRefCounted<LibHandle>::refCount()); + if (--directRefCnt) + ret = true; + mozilla::external::AtomicRefCounted<LibHandle>::Release(); + } + return ret; + } + + /** + * Returns the number of direct references + */ + MozRefCountType DirectRefCount() + { + return directRefCnt; + } + + /** + * Returns the complete size of the file or stream behind the library + * handle. + */ + size_t GetMappableLength() const; + + /** + * Returns a memory mapping of the file or stream behind the library + * handle. + */ + void *MappableMMap(void *addr, size_t length, off_t offset) const; + + /** + * Unmaps a memory mapping of the file or stream behind the library + * handle. + */ + void MappableMUnmap(void *addr, size_t length) const; + +#ifdef __ARM_EABI__ + /** + * Find the address and entry count of the ARM.exidx section + * associated with the library + */ + virtual const void *FindExidx(int *pcount) const = 0; +#endif + + /** + * Shows some stats about the Mappable instance. The when argument is to be + * used by the caller to give an identifier of the when the stats call is + * made. + */ + virtual void stats(const char *when) const { }; + +protected: + /** + * Returns a mappable object for use by MappableMMap and related functions. + */ + virtual Mappable *GetMappable() const = 0; + + /** + * Returns the instance, casted as the wanted type. Returns nullptr if + * that's not the actual type. (short of a better way to do this without + * RTTI) + */ + friend class ElfLoader; + friend class CustomElf; + friend class SEGVHandler; + virtual BaseElf *AsBaseElf() { return nullptr; } + virtual SystemElf *AsSystemElf() { return nullptr; } + +private: + MozRefCountType directRefCnt; + char *path; + + /* Mappable object keeping the result of GetMappable() */ + mutable RefPtr<Mappable> mappable; +}; + +/** + * Specialized RefCounted<LibHandle>::Release. Under normal operation, when + * mRefCnt reaches 0, the LibHandle is deleted. Its mRefCnt is however + * increased to 1 on normal builds, and 0x7fffdead on debug builds so that the + * LibHandle can still be referenced while the destructor is executing. The + * mRefCnt is allowed to grow > 0x7fffdead, but not to decrease under that + * value, which would mean too many Releases from within the destructor. + */ +namespace mozilla { +namespace detail { + +template <> inline void RefCounted<LibHandle, AtomicRefCount>::Release() const { +#ifdef DEBUG + if (mRefCnt > 0x7fff0000) + MOZ_ASSERT(mRefCnt > 0x7fffdead); +#endif + MOZ_ASSERT(mRefCnt > 0); + if (mRefCnt > 0) { + if (0 == --mRefCnt) { +#ifdef DEBUG + mRefCnt = 0x7fffdead; +#else + mRefCnt = 1; +#endif + delete static_cast<const LibHandle*>(this); + } + } +} + +} /* namespace detail */ +} /* namespace mozilla */ + +/** + * Class handling libraries loaded by the system linker + */ +class SystemElf: public LibHandle +{ +public: + /** + * Returns a new SystemElf for the given path. The given flags are passed + * to dlopen(). + */ + static already_AddRefed<LibHandle> Load(const char *path, int flags); + + /** + * Inherited from LibHandle + */ + virtual ~SystemElf(); + virtual void *GetSymbolPtr(const char *symbol) const; + virtual bool Contains(void *addr) const { return false; /* UNIMPLEMENTED */ } + virtual void *GetBase() const { return nullptr; /* UNIMPLEMENTED */ } + +#ifdef __ARM_EABI__ + virtual const void *FindExidx(int *pcount) const; +#endif + +protected: + virtual Mappable *GetMappable() const; + + /** + * Returns the instance, casted as SystemElf. (short of a better way to do + * this without RTTI) + */ + friend class ElfLoader; + virtual SystemElf *AsSystemElf() { return this; } + + /** + * Remove the reference to the system linker handle. This avoids dlclose() + * being called when the instance is destroyed. + */ + void Forget() + { + dlhandle = nullptr; + } + +private: + /** + * Private constructor + */ + SystemElf(const char *path, void *handle) + : LibHandle(path), dlhandle(handle) { } + + /* Handle as returned by system dlopen() */ + void *dlhandle; +}; + +/** + * The ElfLoader registers its own SIGSEGV handler to handle segmentation + * faults within the address space of the loaded libraries. It however + * allows a handler to be set for faults in other places, and redispatches + * to the handler set through signal() or sigaction(). + */ +class SEGVHandler +{ +public: + bool hasRegisteredHandler() { + if (! initialized) + FinishInitialization(); + return registeredHandler; + } + + bool isSignalHandlingBroken() { + return signalHandlingBroken; + } + + static int __wrap_sigaction(int signum, const struct sigaction *act, + struct sigaction *oldact); + +protected: + SEGVHandler(); + ~SEGVHandler(); + +private: + + /** + * The constructor doesn't do all initialization, and the tail is done + * at a later time. + */ + void FinishInitialization(); + + /** + * SIGSEGV handler registered with __wrap_signal or __wrap_sigaction. + */ + struct sigaction action; + + /** + * ElfLoader SIGSEGV handler. + */ + static void handler(int signum, siginfo_t *info, void *context); + + /** + * Temporary test handler. + */ + static void test_handler(int signum, siginfo_t *info, void *context); + + /** + * Size of the alternative stack. The printf family requires more than 8KB + * of stack, and our signal handler may print a few things. + */ + static const size_t stackSize = 12 * 1024; + + /** + * Alternative stack information used before initialization. + */ + stack_t oldStack; + + /** + * Pointer to an alternative stack for signals. Only set if oldStack is + * not set or not big enough. + */ + MappedPtr stackPtr; + + bool initialized; + bool registeredHandler; + bool signalHandlingBroken; + bool signalHandlingSlow; +}; + +/** + * Elf Loader class in charge of loading and bookkeeping libraries. + */ +class ElfLoader: public SEGVHandler +{ +public: + /** + * The Elf Loader instance + */ + static ElfLoader Singleton; + + /** + * Loads the given library with the given flags. Equivalent to dlopen() + * The extra "parent" argument optionally gives the handle of the library + * requesting the given library to be loaded. The loader may look in the + * directory containing that parent library for the library to load. + */ + already_AddRefed<LibHandle> Load(const char *path, int flags, + LibHandle *parent = nullptr); + + /** + * Returns the handle of the library containing the given address in + * its virtual address space, i.e. the library handle for which + * LibHandle::Contains returns true. Its purpose is to allow to + * implement dladdr(). + */ + already_AddRefed<LibHandle> GetHandleByPtr(void *addr); + + /** + * Returns a Mappable object for the path. Paths in the form + * /foo/bar/baz/archive!/directory/lib.so + * try to load the directory/lib.so in /foo/bar/baz/archive, provided + * that file is a Zip archive. + */ + static Mappable *GetMappableFromPath(const char *path); + + void ExpectShutdown(bool val) { expect_shutdown = val; } + bool IsShutdownExpected() { return expect_shutdown; } + +private: + bool expect_shutdown; + +protected: + /** + * Registers the given handle. This method is meant to be called by + * LibHandle subclass creators. + */ + void Register(LibHandle *handle); + void Register(CustomElf *handle); + + /** + * Forget about the given handle. This method is meant to be called by + * LibHandle subclass destructors. + */ + void Forget(LibHandle *handle); + void Forget(CustomElf *handle); + + /* Last error. Used for dlerror() */ + friend class SystemElf; + friend const char *__wrap_dlerror(void); + friend void *__wrap_dlsym(void *handle, const char *symbol); + friend int __wrap_dlclose(void *handle); + const char *lastError; + +private: + ElfLoader() : expect_shutdown(true) {} + ~ElfLoader(); + + /* Initialization code that can't run during static initialization. */ + void Init(); + + /* System loader handle for the library/program containing our code. This + * is used to resolve wrapped functions. */ + RefPtr<LibHandle> self_elf; + +#if defined(ANDROID) + /* System loader handle for the libc. This is used to resolve weak symbols + * that some libcs contain that the Android linker won't dlsym(). Normally, + * we wouldn't treat non-Android differently, but glibc uses versioned + * symbols which this linker doesn't support. */ + RefPtr<LibHandle> libc; +#endif + + /* Bookkeeping */ + typedef std::vector<LibHandle *> LibHandleList; + LibHandleList handles; + +protected: + friend class CustomElf; + friend class LoadedElf; + /** + * Show some stats about Mappables in CustomElfs. The when argument is to + * be used by the caller to give an identifier of the when the stats call + * is made. + */ + static void stats(const char *when); + + /* Definition of static destructors as to be used for C++ ABI compatibility */ + typedef void (*Destructor)(void *object); + + /** + * C++ ABI makes static initializers register destructors through a specific + * atexit interface. On glibc/linux systems, the dso_handle is a pointer + * within a given library. On bionic/android systems, it is an undefined + * symbol. Making sense of the value is not really important, and all that + * is really important is that it is different for each loaded library, so + * that they can be discriminated when shutting down. For convenience, on + * systems where the dso handle is a symbol, that symbol is resolved to + * point at corresponding CustomElf. + * + * Destructors are registered with __*_atexit with an associated object to + * be passed as argument when it is called. + * + * When __cxa_finalize is called, destructors registered for the given + * DSO handle are called in the reverse order they were registered. + */ +#ifdef __ARM_EABI__ + static int __wrap_aeabi_atexit(void *that, Destructor destructor, + void *dso_handle); +#else + static int __wrap_cxa_atexit(Destructor destructor, void *that, + void *dso_handle); +#endif + + static void __wrap_cxa_finalize(void *dso_handle); + + /** + * Registered destructor. Keeps track of the destructor function pointer, + * associated object to call it with, and DSO handle. + */ + class DestructorCaller { + public: + DestructorCaller(Destructor destructor, void *object, void *dso_handle) + : destructor(destructor), object(object), dso_handle(dso_handle) { } + + /** + * Call the destructor function with the associated object. + * Call only once, see CustomElf::~CustomElf. + */ + void Call(); + + /** + * Returns whether the destructor is associated to the given DSO handle + */ + bool IsForHandle(void *handle) const + { + return handle == dso_handle; + } + + private: + Destructor destructor; + void *object; + void *dso_handle; + }; + +private: + /* Keep track of all registered destructors */ + std::vector<DestructorCaller> destructors; + + /* Forward declaration, see further below */ + class DebuggerHelper; +public: + /* Loaded object descriptor for the debugger interface below*/ + struct link_map { + /* Base address of the loaded object. */ + const void *l_addr; + /* File name */ + const char *l_name; + /* Address of the PT_DYNAMIC segment. */ + const void *l_ld; + + private: + friend class ElfLoader::DebuggerHelper; + /* Double linked list of loaded objects. */ + link_map *l_next, *l_prev; + }; + +private: + /* Data structure used by the linker to give details about shared objects it + * loaded to debuggers. This is normally defined in link.h, but Android + * headers lack this file. */ + struct r_debug { + /* Version number of the protocol. */ + int r_version; + + /* Head of the linked list of loaded objects. */ + link_map *r_map; + + /* Function to be called when updates to the linked list of loaded objects + * are going to occur. The function is to be called before and after + * changes. */ + void (*r_brk)(void); + + /* Indicates to the debugger what state the linked list of loaded objects + * is in when the function above is called. */ + enum { + RT_CONSISTENT, /* Changes are complete */ + RT_ADD, /* Beginning to add a new object */ + RT_DELETE /* Beginning to remove an object */ + } r_state; + }; + + /* Memory representation of ELF Auxiliary Vectors */ + struct AuxVector { + Elf::Addr type; + Elf::Addr value; + }; + + /* Helper class used to integrate libraries loaded by this linker in + * r_debug */ + class DebuggerHelper + { + public: + DebuggerHelper(); + + void Init(AuxVector *auvx); + + operator bool() + { + return dbg; + } + + /* Make the debugger aware of a new loaded object */ + void Add(link_map *map); + + /* Make the debugger aware of the unloading of an object */ + void Remove(link_map *map); + + /* Iterates over all link_maps */ + class iterator + { + public: + const link_map *operator ->() const + { + return item; + } + + const link_map &operator ++() + { + item = item->l_next; + return *item; + } + + bool operator<(const iterator &other) const + { + if (other.item == nullptr) + return item ? true : false; + MOZ_CRASH("DebuggerHelper::iterator::operator< called with something else than DebuggerHelper::end()"); + } + protected: + friend class DebuggerHelper; + iterator(const link_map *item): item(item) { } + + private: + const link_map *item; + }; + + iterator begin() const + { + return iterator(dbg ? dbg->r_map : nullptr); + } + + iterator end() const + { + return iterator(nullptr); + } + + private: + r_debug *dbg; + link_map *firstAdded; + }; + friend int __wrap_dl_iterate_phdr(dl_phdr_cb callback, void *data); + DebuggerHelper dbg; +}; + +#endif /* ElfLoader_h */ diff --git a/mozglue/linker/Elfxx.h b/mozglue/linker/Elfxx.h new file mode 100644 index 0000000000..b21a893365 --- /dev/null +++ b/mozglue/linker/Elfxx.h @@ -0,0 +1,241 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef Elfxx_h +#define Elfxx_h + +/** + * Android system headers have two different elf.h file. The one under linux/ + * is the most complete on older android API versions. + */ +#if defined(ANDROID) && __ANDROID_API__ < 21 +#include <linux/elf.h> +#else +#include <elf.h> +#endif +#include <endian.h> + +#if defined(__ARM_EABI__) && !defined(PT_ARM_EXIDX) +#define PT_ARM_EXIDX 0x70000001 +#endif + +/** + * Generic ELF macros for the target system + */ +#ifdef __LP64__ +#define Elf_(type) Elf64_ ## type +#define ELFCLASS ELFCLASS64 +#define ELF_R_TYPE ELF64_R_TYPE +#define ELF_R_SYM ELF64_R_SYM +#ifndef ELF_ST_BIND +#define ELF_ST_BIND ELF64_ST_BIND +#endif +#else +#define Elf_(type) Elf32_ ## type +#define ELFCLASS ELFCLASS32 +#define ELF_R_TYPE ELF32_R_TYPE +#define ELF_R_SYM ELF32_R_SYM +#ifndef ELF_ST_BIND +#define ELF_ST_BIND ELF32_ST_BIND +#endif +#endif + +#ifndef __BYTE_ORDER +#error Cannot find endianness +#endif + +#if __BYTE_ORDER == __LITTLE_ENDIAN +#define ELFDATA ELFDATA2LSB +#elif __BYTE_ORDER == __BIG_ENDIAN +#define ELFDATA ELFDATA2MSB +#endif + +#ifdef __linux__ +#define ELFOSABI ELFOSABI_LINUX +#ifdef EI_ABIVERSION +#define ELFABIVERSION 0 +#endif +#else +#error Unknown ELF OSABI +#endif + +#if defined(__i386__) +#define ELFMACHINE EM_386 + +// Doing this way probably doesn't scale to other architectures +#define R_ABS R_386_32 +#define R_GLOB_DAT R_386_GLOB_DAT +#define R_JMP_SLOT R_386_JMP_SLOT +#define R_RELATIVE R_386_RELATIVE +#define RELOC(n) DT_REL ## n +#define UNSUPPORTED_RELOC(n) DT_RELA ## n +#define STR_RELOC(n) "DT_REL" # n +#define Reloc Rel + +#elif defined(__x86_64__) +#define ELFMACHINE EM_X86_64 + +#define R_ABS R_X86_64_64 +#define R_GLOB_DAT R_X86_64_GLOB_DAT +#define R_JMP_SLOT R_X86_64_JUMP_SLOT +#define R_RELATIVE R_X86_64_RELATIVE +#define RELOC(n) DT_RELA ## n +#define UNSUPPORTED_RELOC(n) DT_REL ## n +#define STR_RELOC(n) "DT_RELA" # n +#define Reloc Rela + +#elif defined(__arm__) +#define ELFMACHINE EM_ARM + +#ifndef R_ARM_ABS32 +#define R_ARM_ABS32 2 +#endif +#ifndef R_ARM_GLOB_DAT +#define R_ARM_GLOB_DAT 21 +#endif +#ifndef R_ARM_JUMP_SLOT +#define R_ARM_JUMP_SLOT 22 +#endif +#ifndef R_ARM_RELATIVE +#define R_ARM_RELATIVE 23 +#endif + +#define R_ABS R_ARM_ABS32 +#define R_GLOB_DAT R_ARM_GLOB_DAT +#define R_JMP_SLOT R_ARM_JUMP_SLOT +#define R_RELATIVE R_ARM_RELATIVE +#define RELOC(n) DT_REL ## n +#define UNSUPPORTED_RELOC(n) DT_RELA ## n +#define STR_RELOC(n) "DT_REL" # n +#define Reloc Rel + +#else +#error Unknown ELF machine type +#endif + +/** + * Android system headers don't have all definitions + */ +#ifndef STN_UNDEF +#define STN_UNDEF 0 +#endif +#ifndef DT_INIT_ARRAY +#define DT_INIT_ARRAY 25 +#endif +#ifndef DT_FINI_ARRAY +#define DT_FINI_ARRAY 26 +#endif +#ifndef DT_INIT_ARRAYSZ +#define DT_INIT_ARRAYSZ 27 +#endif +#ifndef DT_FINI_ARRAYSZ +#define DT_FINI_ARRAYSZ 28 +#endif +#ifndef DT_RELACOUNT +#define DT_RELACOUNT 0x6ffffff9 +#endif +#ifndef DT_RELCOUNT +#define DT_RELCOUNT 0x6ffffffa +#endif +#ifndef DT_VERSYM +#define DT_VERSYM 0x6ffffff0 +#endif +#ifndef DT_VERDEF +#define DT_VERDEF 0x6ffffffc +#endif +#ifndef DT_VERDEFNUM +#define DT_VERDEFNUM 0x6ffffffd +#endif +#ifndef DT_VERNEED +#define DT_VERNEED 0x6ffffffe +#endif +#ifndef DT_VERNEEDNUM +#define DT_VERNEEDNUM 0x6fffffff +#endif +#ifndef DT_FLAGS_1 +#define DT_FLAGS_1 0x6ffffffb +#endif +#ifndef DT_FLAGS +#define DT_FLAGS 30 +#endif +#ifndef DF_SYMBOLIC +#define DF_SYMBOLIC 0x00000002 +#endif +#ifndef DF_TEXTREL +#define DF_TEXTREL 0x00000004 +#endif + +namespace Elf { + +/** + * Define a few basic Elf Types + */ +typedef Elf_(Phdr) Phdr; +typedef Elf_(Dyn) Dyn; +typedef Elf_(Sym) Sym; +typedef Elf_(Addr) Addr; +typedef Elf_(Word) Word; +typedef Elf_(Half) Half; + +/** + * Helper class around the standard Elf header struct + */ +struct Ehdr: public Elf_(Ehdr) +{ + /** + * Equivalent to reinterpret_cast<const Ehdr *>(buf), but additionally + * checking that this is indeed an Elf header and that the Elf type + * corresponds to that of the system + */ + static const Ehdr *validate(const void *buf); +}; + +/** + * Elf String table + */ +class Strtab: public UnsizedArray<const char> +{ +public: + /** + * Returns the string at the given index in the table + */ + const char *GetStringAt(off_t index) const + { + return &UnsizedArray<const char>::operator[](index); + } +}; + +/** + * Helper class around Elf relocation. + */ +struct Rel: public Elf_(Rel) +{ + /** + * Returns the addend for the relocation, which is the value stored + * at r_offset. + */ + Addr GetAddend(void *base) const + { + return *(reinterpret_cast<const Addr *>( + reinterpret_cast<const char *>(base) + r_offset)); + } +}; + +/** + * Helper class around Elf relocation with addend. + */ +struct Rela: public Elf_(Rela) +{ + /** + * Returns the addend for the relocation. + */ + Addr GetAddend(void *base) const + { + return r_addend; + } +}; + +} /* namespace Elf */ + +#endif /* Elfxx_h */ diff --git a/mozglue/linker/Logging.h b/mozglue/linker/Logging.h new file mode 100644 index 0000000000..046d918f4a --- /dev/null +++ b/mozglue/linker/Logging.h @@ -0,0 +1,87 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef Logging_h +#define Logging_h + +#include "mozilla/Likely.h" + +#ifdef ANDROID +#include <android/log.h> +#define LOG(...) __android_log_print(ANDROID_LOG_INFO, "GeckoLinker", __VA_ARGS__) +#define WARN(...) __android_log_print(ANDROID_LOG_WARN, "GeckoLinker", __VA_ARGS__) +#define ERROR(...) __android_log_print(ANDROID_LOG_ERROR, "GeckoLinker", __VA_ARGS__) +#else +#include <cstdio> + +/* Expand to 1 or m depending on whether there is one argument or more + * given. */ +#define MOZ_ONE_OR_MORE_ARGS_IMPL2(_1, _2, _3, _4, _5, _6, _7, _8, _9, N, ...) \ + N +#define MOZ_ONE_OR_MORE_ARGS_IMPL(args) MOZ_ONE_OR_MORE_ARGS_IMPL2 args +#define MOZ_ONE_OR_MORE_ARGS(...) \ + MOZ_ONE_OR_MORE_ARGS_IMPL((__VA_ARGS__, m, m, m, m, m, m, m, m, 1, 0)) + +#define MOZ_MACRO_GLUE(a, b) a b +#define MOZ_CONCAT2(a, b) a ## b +#define MOZ_CONCAT1(a, b) MOZ_CONCAT2(a, b) +#define MOZ_CONCAT(a, b) MOZ_CONCAT1(a, b) + +/* Some magic to choose between LOG1 and LOGm depending on the number of + * arguments */ +#define MOZ_CHOOSE_LOG(...) \ + MOZ_MACRO_GLUE(MOZ_CONCAT(LOG, MOZ_ONE_OR_MORE_ARGS(__VA_ARGS__)), \ + (__VA_ARGS__)) + +#define LOG1(format) fprintf(stderr, format "\n") +#define LOGm(format, ...) fprintf(stderr, format "\n", __VA_ARGS__) +#define LOG(...) MOZ_CHOOSE_LOG(__VA_ARGS__) +#define WARN(...) MOZ_CHOOSE_LOG("Warning: " __VA_ARGS__) +#define ERROR(...) MOZ_CHOOSE_LOG("Error: " __VA_ARGS__) + +#endif + +class Logging +{ +public: + static bool isVerbose() + { + return Singleton.verbose; + } + +private: + bool verbose; + +public: + static void Init() + { + const char *env = getenv("MOZ_DEBUG_LINKER"); + if (env && *env == '1') + Singleton.verbose = true; + } + +private: + static Logging Singleton; +}; + +#define DEBUG_LOG(...) \ + do { \ + if (MOZ_UNLIKELY(Logging::isVerbose())) { \ + LOG(__VA_ARGS__); \ + } \ + } while(0) + +#if defined(__LP64__) +# define PRIxAddr "lx" +# define PRIxSize "lx" +# define PRIdSize "ld" +# define PRIuSize "lu" +#else +# define PRIxAddr "x" +# define PRIxSize "x" +# define PRIdSize "d" +# define PRIuSize "u" +#endif + +#endif /* Logging_h */ diff --git a/mozglue/linker/Mappable.cpp b/mozglue/linker/Mappable.cpp new file mode 100644 index 0000000000..47b883d2d5 --- /dev/null +++ b/mozglue/linker/Mappable.cpp @@ -0,0 +1,681 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include <fcntl.h> +#include <unistd.h> +#include <sys/mman.h> +#include <sys/stat.h> +#include <cstring> +#include <cstdlib> +#include <cstdio> +#include <string> + +#include "Mappable.h" + +#include "mozilla/UniquePtr.h" + +#ifdef ANDROID +#include <linux/ashmem.h> +#endif +#include <sys/stat.h> +#include <errno.h> +#include "ElfLoader.h" +#include "SeekableZStream.h" +#include "XZStream.h" +#include "Logging.h" + +using mozilla::MakeUnique; +using mozilla::UniquePtr; + +class CacheValidator +{ +public: + CacheValidator(const char* aCachedLibPath, Zip* aZip, Zip::Stream* aStream) + : mCachedLibPath(aCachedLibPath) + { + static const char kChecksumSuffix[] = ".crc"; + + mCachedChecksumPath = + MakeUnique<char[]>(strlen(aCachedLibPath) + sizeof(kChecksumSuffix)); + sprintf(mCachedChecksumPath.get(), "%s%s", aCachedLibPath, kChecksumSuffix); + DEBUG_LOG("mCachedChecksumPath: %s", mCachedChecksumPath.get()); + + mChecksum = aStream->GetCRC32(); + DEBUG_LOG("mChecksum: %x", mChecksum); + } + + // Returns whether the cache is valid and up-to-date. + bool IsValid() const + { + // Validate based on checksum. + RefPtr<Mappable> checksumMap = MappableFile::Create(mCachedChecksumPath.get()); + if (!checksumMap) { + // Force caching if checksum is missing in cache. + return false; + } + + DEBUG_LOG("Comparing %x with %s", mChecksum, mCachedChecksumPath.get()); + MappedPtr checksumBuf = checksumMap->mmap(nullptr, checksumMap->GetLength(), + PROT_READ, MAP_PRIVATE, 0); + if (checksumBuf == MAP_FAILED) { + WARN("Couldn't map %s to validate checksum", mCachedChecksumPath.get()); + return false; + } + if (memcmp(checksumBuf, &mChecksum, sizeof(mChecksum))) { + return false; + } + return !access(mCachedLibPath.c_str(), R_OK); + } + + // Caches the APK-provided checksum used in future cache validations. + void CacheChecksum() const + { + AutoCloseFD fd(open(mCachedChecksumPath.get(), + O_TRUNC | O_RDWR | O_CREAT | O_NOATIME, + S_IRUSR | S_IWUSR)); + if (fd == -1) { + WARN("Couldn't open %s to update checksum", mCachedChecksumPath.get()); + return; + } + + DEBUG_LOG("Updating checksum %s", mCachedChecksumPath.get()); + + const size_t size = sizeof(mChecksum); + size_t written = 0; + while (written < size) { + ssize_t ret = write(fd, + reinterpret_cast<const uint8_t*>(&mChecksum) + written, + size - written); + if (ret >= 0) { + written += ret; + } else if (errno != EINTR) { + WARN("Writing checksum %s failed with errno %d", + mCachedChecksumPath.get(), errno); + break; + } + } + } + +private: + const std::string mCachedLibPath; + UniquePtr<char[]> mCachedChecksumPath; + uint32_t mChecksum; +}; + +Mappable * +MappableFile::Create(const char *path) +{ + int fd = open(path, O_RDONLY); + if (fd != -1) + return new MappableFile(fd); + return nullptr; +} + +MemoryRange +MappableFile::mmap(const void *addr, size_t length, int prot, int flags, + off_t offset) +{ + MOZ_ASSERT(fd != -1); + MOZ_ASSERT(!(flags & MAP_SHARED)); + flags |= MAP_PRIVATE; + + return MemoryRange::mmap(const_cast<void *>(addr), length, prot, flags, + fd, offset); +} + +void +MappableFile::finalize() +{ + /* Close file ; equivalent to close(fd.forget()) */ + fd = -1; +} + +size_t +MappableFile::GetLength() const +{ + struct stat st; + return fstat(fd, &st) ? 0 : st.st_size; +} + +Mappable * +MappableExtractFile::Create(const char *name, Zip *zip, Zip::Stream *stream) +{ + MOZ_ASSERT(zip && stream); + + const char *cachePath = getenv("MOZ_LINKER_CACHE"); + if (!cachePath || !*cachePath) { + WARN("MOZ_LINKER_EXTRACT is set, but not MOZ_LINKER_CACHE; " + "not extracting"); + return nullptr; + } + + // Ensure that the cache dir is private. + chmod(cachePath, 0770); + + UniquePtr<char[]> path = + MakeUnique<char[]>(strlen(cachePath) + strlen(name) + 2); + sprintf(path.get(), "%s/%s", cachePath, name); + + CacheValidator validator(path.get(), zip, stream); + if (validator.IsValid()) { + DEBUG_LOG("Reusing %s", static_cast<char *>(path.get())); + return MappableFile::Create(path.get()); + } + DEBUG_LOG("Extracting to %s", static_cast<char *>(path.get())); + AutoCloseFD fd; + fd = open(path.get(), O_TRUNC | O_RDWR | O_CREAT | O_NOATIME, + S_IRUSR | S_IWUSR); + if (fd == -1) { + ERROR("Couldn't open %s to decompress library", path.get()); + return nullptr; + } + AutoUnlinkFile file(path.release()); + if (stream->GetType() == Zip::Stream::DEFLATE) { + if (ftruncate(fd, stream->GetUncompressedSize()) == -1) { + ERROR("Couldn't ftruncate %s to decompress library", file.get()); + return nullptr; + } + /* Map the temporary file for use as inflate buffer */ + MappedPtr buffer(MemoryRange::mmap(nullptr, stream->GetUncompressedSize(), + PROT_WRITE, MAP_SHARED, fd, 0)); + if (buffer == MAP_FAILED) { + ERROR("Couldn't map %s to decompress library", file.get()); + return nullptr; + } + + zxx_stream zStream = stream->GetZStream(buffer); + + /* Decompress */ + if (inflateInit2(&zStream, -MAX_WBITS) != Z_OK) { + ERROR("inflateInit failed: %s", zStream.msg); + return nullptr; + } + if (inflate(&zStream, Z_FINISH) != Z_STREAM_END) { + ERROR("inflate failed: %s", zStream.msg); + return nullptr; + } + if (inflateEnd(&zStream) != Z_OK) { + ERROR("inflateEnd failed: %s", zStream.msg); + return nullptr; + } + if (zStream.total_out != stream->GetUncompressedSize()) { + ERROR("File not fully uncompressed! %ld / %d", zStream.total_out, + static_cast<unsigned int>(stream->GetUncompressedSize())); + return nullptr; + } + } else if (XZStream::IsXZ(stream->GetBuffer(), stream->GetSize())) { + XZStream xzStream(stream->GetBuffer(), stream->GetSize()); + + if (!xzStream.Init()) { + ERROR("Couldn't initialize XZ decoder"); + return nullptr; + } + DEBUG_LOG("XZStream created, compressed=%u, uncompressed=%u", + xzStream.Size(), xzStream.UncompressedSize()); + + if (ftruncate(fd, xzStream.UncompressedSize()) == -1) { + ERROR("Couldn't ftruncate %s to decompress library", file.get()); + return nullptr; + } + MappedPtr buffer(MemoryRange::mmap(nullptr, xzStream.UncompressedSize(), + PROT_WRITE, MAP_SHARED, fd, 0)); + if (buffer == MAP_FAILED) { + ERROR("Couldn't map %s to decompress library", file.get()); + return nullptr; + } + const size_t written = xzStream.Decode(buffer, buffer.GetLength()); + DEBUG_LOG("XZStream decoded %u", written); + if (written != buffer.GetLength()) { + ERROR("Error decoding XZ file %s", file.get()); + return nullptr; + } + } else if (stream->GetType() == Zip::Stream::STORE) { + SeekableZStream zStream; + if (!zStream.Init(stream->GetBuffer(), stream->GetSize())) { + ERROR("Couldn't initialize SeekableZStream for %s", name); + return nullptr; + } + if (ftruncate(fd, zStream.GetUncompressedSize()) == -1) { + ERROR("Couldn't ftruncate %s to decompress library", file.get()); + return nullptr; + } + MappedPtr buffer(MemoryRange::mmap(nullptr, zStream.GetUncompressedSize(), + PROT_WRITE, MAP_SHARED, fd, 0)); + if (buffer == MAP_FAILED) { + ERROR("Couldn't map %s to decompress library", file.get()); + return nullptr; + } + + if (!zStream.Decompress(buffer, 0, zStream.GetUncompressedSize())) { + ERROR("%s: failed to decompress", name); + return nullptr; + } + } else { + return nullptr; + } + + validator.CacheChecksum(); + return new MappableExtractFile(fd.forget(), file.release()); +} + +/** + * _MappableBuffer is a buffer which content can be mapped at different + * locations in the virtual address space. + * On Linux, uses a (deleted) temporary file on a tmpfs for sharable content. + * On Android, uses ashmem. + */ +class _MappableBuffer: public MappedPtr +{ +public: + /** + * Returns a _MappableBuffer instance with the given name and the given + * length. + */ + static _MappableBuffer *Create(const char *name, size_t length) + { + AutoCloseFD fd; +#ifdef ANDROID + /* On Android, initialize an ashmem region with the given length */ + fd = open("/" ASHMEM_NAME_DEF, O_RDWR, 0600); + if (fd == -1) + return nullptr; + char str[ASHMEM_NAME_LEN]; + strlcpy(str, name, sizeof(str)); + ioctl(fd, ASHMEM_SET_NAME, str); + if (ioctl(fd, ASHMEM_SET_SIZE, length)) + return nullptr; + + /* The Gecko crash reporter is confused by adjacent memory mappings of + * the same file and chances are we're going to map from the same file + * descriptor right away. To avoid problems with the crash reporter, + * create an empty anonymous page before or after the ashmem mapping, + * depending on how mappings grow in the address space. + */ +#if defined(__arm__) + void *buf = ::mmap(nullptr, length + PAGE_SIZE, PROT_READ | PROT_WRITE, + MAP_SHARED, fd, 0); + if (buf != MAP_FAILED) { + ::mmap(AlignedEndPtr(reinterpret_cast<char *>(buf) + length, PAGE_SIZE), + PAGE_SIZE, PROT_NONE, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + DEBUG_LOG("Decompression buffer of size 0x%x in ashmem \"%s\", mapped @%p", + length, str, buf); + return new _MappableBuffer(fd.forget(), buf, length); + } +#elif defined(__i386__) + size_t anon_mapping_length = length + PAGE_SIZE; + void *buf = ::mmap(nullptr, anon_mapping_length, PROT_NONE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + if (buf != MAP_FAILED) { + char *first_page = reinterpret_cast<char *>(buf); + char *map_page = first_page + PAGE_SIZE; + + void *actual_buf = ::mmap(map_page, length, PROT_READ | PROT_WRITE, + MAP_FIXED | MAP_SHARED, fd, 0); + if (actual_buf == MAP_FAILED) { + ::munmap(buf, anon_mapping_length); + DEBUG_LOG("Fixed allocation of decompression buffer at %p failed", map_page); + return nullptr; + } + + DEBUG_LOG("Decompression buffer of size 0x%x in ashmem \"%s\", mapped @%p", + length, str, actual_buf); + return new _MappableBuffer(fd.forget(), actual_buf, length); + } +#else +#error need to add a case for your CPU +#endif +#else + /* On Linux, use /dev/shm as base directory for temporary files, assuming + * it's on tmpfs */ + /* TODO: check that /dev/shm is tmpfs */ + char path[256]; + sprintf(path, "/dev/shm/%s.XXXXXX", name); + fd = mkstemp(path); + if (fd == -1) + return nullptr; + unlink(path); + ftruncate(fd, length); + + void *buf = ::mmap(nullptr, length, PROT_READ | PROT_WRITE, + MAP_SHARED, fd, 0); + if (buf != MAP_FAILED) { + DEBUG_LOG("Decompression buffer of size %ld in \"%s\", mapped @%p", + length, path, buf); + return new _MappableBuffer(fd.forget(), buf, length); + } +#endif + return nullptr; + } + + void *mmap(const void *addr, size_t length, int prot, int flags, off_t offset) + { + MOZ_ASSERT(fd != -1); +#ifdef ANDROID + /* Mapping ashmem MAP_PRIVATE is like mapping anonymous memory, even when + * there is content in the ashmem */ + if (flags & MAP_PRIVATE) { + flags &= ~MAP_PRIVATE; + flags |= MAP_SHARED; + } +#endif + return ::mmap(const_cast<void *>(addr), length, prot, flags, fd, offset); + } + +#ifdef ANDROID + ~_MappableBuffer() { + /* Free the additional page we allocated. See _MappableBuffer::Create */ +#if defined(__arm__) + ::munmap(AlignedEndPtr(*this + GetLength(), PAGE_SIZE), PAGE_SIZE); +#elif defined(__i386__) + ::munmap(*this - PAGE_SIZE, GetLength() + PAGE_SIZE); +#else +#error need to add a case for your CPU +#endif + } +#endif + +private: + _MappableBuffer(int fd, void *buf, size_t length) + : MappedPtr(buf, length), fd(fd) { } + + /* File descriptor for the temporary file or ashmem */ + AutoCloseFD fd; +}; + + +Mappable * +MappableDeflate::Create(const char *name, Zip *zip, Zip::Stream *stream) +{ + MOZ_ASSERT(stream->GetType() == Zip::Stream::DEFLATE); + _MappableBuffer *buf = _MappableBuffer::Create(name, stream->GetUncompressedSize()); + if (buf) + return new MappableDeflate(buf, zip, stream); + return nullptr; +} + +MappableDeflate::MappableDeflate(_MappableBuffer *buf, Zip *zip, + Zip::Stream *stream) +: zip(zip), buffer(buf), zStream(stream->GetZStream(*buf)) { } + +MappableDeflate::~MappableDeflate() { } + +MemoryRange +MappableDeflate::mmap(const void *addr, size_t length, int prot, int flags, off_t offset) +{ + MOZ_ASSERT(buffer); + MOZ_ASSERT(!(flags & MAP_SHARED)); + flags |= MAP_PRIVATE; + + /* The deflate stream is uncompressed up to the required offset + length, if + * it hasn't previously been uncompressed */ + ssize_t missing = offset + length + zStream.avail_out - buffer->GetLength(); + if (missing > 0) { + uInt avail_out = zStream.avail_out; + zStream.avail_out = missing; + if ((*buffer == zStream.next_out) && + (inflateInit2(&zStream, -MAX_WBITS) != Z_OK)) { + ERROR("inflateInit failed: %s", zStream.msg); + return MemoryRange(MAP_FAILED, 0); + } + int ret = inflate(&zStream, Z_SYNC_FLUSH); + if (ret < 0) { + ERROR("inflate failed: %s", zStream.msg); + return MemoryRange(MAP_FAILED, 0); + } + if (ret == Z_NEED_DICT) { + ERROR("zstream requires a dictionary. %s", zStream.msg); + return MemoryRange(MAP_FAILED, 0); + } + zStream.avail_out = avail_out - missing + zStream.avail_out; + if (ret == Z_STREAM_END) { + if (inflateEnd(&zStream) != Z_OK) { + ERROR("inflateEnd failed: %s", zStream.msg); + return MemoryRange(MAP_FAILED, 0); + } + if (zStream.total_out != buffer->GetLength()) { + ERROR("File not fully uncompressed! %ld / %d", zStream.total_out, + static_cast<unsigned int>(buffer->GetLength())); + return MemoryRange(MAP_FAILED, 0); + } + } + } +#if defined(ANDROID) && defined(__arm__) + if (prot & PROT_EXEC) { + /* We just extracted data that may be executed in the future. + * We thus need to ensure Instruction and Data cache coherency. */ + DEBUG_LOG("cacheflush(%p, %p)", *buffer + offset, *buffer + (offset + length)); + cacheflush(reinterpret_cast<uintptr_t>(*buffer + offset), + reinterpret_cast<uintptr_t>(*buffer + (offset + length)), 0); + } +#endif + + return MemoryRange(buffer->mmap(addr, length, prot, flags, offset), length); +} + +void +MappableDeflate::finalize() +{ + /* Free zlib internal buffers */ + inflateEnd(&zStream); + /* Free decompression buffer */ + buffer = nullptr; + /* Remove reference to Zip archive */ + zip = nullptr; +} + +size_t +MappableDeflate::GetLength() const +{ + return buffer->GetLength(); +} + +Mappable * +MappableSeekableZStream::Create(const char *name, Zip *zip, + Zip::Stream *stream) +{ + MOZ_ASSERT(stream->GetType() == Zip::Stream::STORE); + UniquePtr<MappableSeekableZStream> mappable(new MappableSeekableZStream(zip)); + + pthread_mutexattr_t recursiveAttr; + pthread_mutexattr_init(&recursiveAttr); + pthread_mutexattr_settype(&recursiveAttr, PTHREAD_MUTEX_RECURSIVE); + + if (pthread_mutex_init(&mappable->mutex, &recursiveAttr)) + return nullptr; + + if (!mappable->zStream.Init(stream->GetBuffer(), stream->GetSize())) + return nullptr; + + mappable->buffer.reset(_MappableBuffer::Create(name, + mappable->zStream.GetUncompressedSize())); + if (!mappable->buffer) + return nullptr; + + mappable->chunkAvail = MakeUnique<unsigned char[]>(mappable->zStream.GetChunksNum()); + + return mappable.release(); +} + +MappableSeekableZStream::MappableSeekableZStream(Zip *zip) +: zip(zip), chunkAvailNum(0) { } + +MappableSeekableZStream::~MappableSeekableZStream() +{ + pthread_mutex_destroy(&mutex); +} + +MemoryRange +MappableSeekableZStream::mmap(const void *addr, size_t length, int prot, + int flags, off_t offset) +{ + /* Map with PROT_NONE so that accessing the mapping would segfault, and + * bring us to ensure() */ + void *res = buffer->mmap(addr, length, PROT_NONE, flags, offset); + if (res == MAP_FAILED) + return MemoryRange(MAP_FAILED, 0); + + /* Store the mapping, ordered by offset and length */ + std::vector<LazyMap>::reverse_iterator it; + for (it = lazyMaps.rbegin(); it < lazyMaps.rend(); ++it) { + if ((it->offset < offset) || + ((it->offset == offset) && (it->length < length))) + break; + } + LazyMap map = { res, length, prot, offset }; + lazyMaps.insert(it.base(), map); + return MemoryRange(res, length); +} + +void +MappableSeekableZStream::munmap(void *addr, size_t length) +{ + std::vector<LazyMap>::iterator it; + for (it = lazyMaps.begin(); it < lazyMaps.end(); ++it) + if ((it->addr = addr) && (it->length == length)) { + lazyMaps.erase(it); + ::munmap(addr, length); + return; + } + MOZ_CRASH("munmap called with unknown mapping"); +} + +void +MappableSeekableZStream::finalize() { } + +bool +MappableSeekableZStream::ensure(const void *addr) +{ + DEBUG_LOG("ensure @%p", addr); + const void *addrPage = PageAlignedPtr(addr); + /* Find the mapping corresponding to the given page */ + std::vector<LazyMap>::iterator map; + for (map = lazyMaps.begin(); map < lazyMaps.end(); ++map) { + if (map->Contains(addrPage)) + break; + } + if (map == lazyMaps.end()) + return false; + + /* Find corresponding chunk */ + off_t mapOffset = map->offsetOf(addrPage); + off_t chunk = mapOffset / zStream.GetChunkSize(); + + /* In the typical case, we just need to decompress the chunk entirely. But + * when the current mapping ends in the middle of the chunk, we want to + * stop at the end of the corresponding page. + * However, if another mapping needs the last part of the chunk, we still + * need to continue. As mappings are ordered by offset and length, we don't + * need to scan the entire list of mappings. + * It is safe to run through lazyMaps here because the linker is never + * going to call mmap (which adds lazyMaps) while this function is + * called. */ + size_t length = zStream.GetChunkSize(chunk); + off_t chunkStart = chunk * zStream.GetChunkSize(); + off_t chunkEnd = chunkStart + length; + std::vector<LazyMap>::iterator it; + for (it = map; it < lazyMaps.end(); ++it) { + if (chunkEnd <= it->endOffset()) + break; + } + if ((it == lazyMaps.end()) || (chunkEnd > it->endOffset())) { + /* The mapping "it" points at now is past the interesting one */ + --it; + length = it->endOffset() - chunkStart; + } + + length = PageAlignedSize(length); + + /* The following lock can be re-acquired by the thread holding it. + * If this happens, it means the following code is interrupted somehow by + * some signal, and ends up retriggering a chunk decompression for the + * same MappableSeekableZStream. + * If the chunk to decompress is different the second time, then everything + * is safe as the only common data touched below is chunkAvailNum, and it is + * atomically updated (leaving out any chance of an interruption while it is + * updated affecting the result). If the chunk to decompress is the same, the + * worst thing that can happen is chunkAvailNum being incremented one too + * many times, which doesn't affect functionality. The chances of it + * happening being pretty slim, and the effect being harmless, we can just + * ignore the issue. Other than that, we'd just be wasting time decompressing + * the same chunk twice. */ + AutoLock lock(&mutex); + + /* The very first page is mapped and accessed separately of the rest, and + * as such, only the first page of the first chunk is decompressed this way. + * When we fault in the remaining pages of that chunk, we want to decompress + * the complete chunk again. Short of doing that, we would end up with + * no data between PageSize() and chunkSize, which would effectively corrupt + * symbol resolution in the underlying library. */ + if (chunkAvail[chunk] < PageNumber(length)) { + if (!zStream.DecompressChunk(*buffer + chunkStart, chunk, length)) + return false; + +#if defined(ANDROID) && defined(__arm__) + if (map->prot & PROT_EXEC) { + /* We just extracted data that may be executed in the future. + * We thus need to ensure Instruction and Data cache coherency. */ + DEBUG_LOG("cacheflush(%p, %p)", *buffer + chunkStart, *buffer + (chunkStart + length)); + cacheflush(reinterpret_cast<uintptr_t>(*buffer + chunkStart), + reinterpret_cast<uintptr_t>(*buffer + (chunkStart + length)), 0); + } +#endif + /* Only count if we haven't already decompressed parts of the chunk */ + if (chunkAvail[chunk] == 0) + chunkAvailNum++; + + chunkAvail[chunk] = PageNumber(length); + } + + /* Flip the chunk mapping protection to the recorded flags. We could + * also flip the protection for other mappings of the same chunk, + * but it's easier to skip that and let further segfaults call + * ensure again. */ + const void *chunkAddr = reinterpret_cast<const void *> + (reinterpret_cast<uintptr_t>(addrPage) + - mapOffset % zStream.GetChunkSize()); + const void *chunkEndAddr = reinterpret_cast<const void *> + (reinterpret_cast<uintptr_t>(chunkAddr) + length); + + const void *start = std::max(map->addr, chunkAddr); + const void *end = std::min(map->end(), chunkEndAddr); + length = reinterpret_cast<uintptr_t>(end) + - reinterpret_cast<uintptr_t>(start); + + if (mprotect(const_cast<void *>(start), length, map->prot) == 0) { + DEBUG_LOG("mprotect @%p, 0x%" PRIxSize ", 0x%x", start, length, map->prot); + return true; + } + + ERROR("mprotect @%p, 0x%" PRIxSize ", 0x%x failed with errno %d", + start, length, map->prot, errno); + return false; +} + +void +MappableSeekableZStream::stats(const char *when, const char *name) const +{ + size_t nEntries = zStream.GetChunksNum(); + DEBUG_LOG("%s: %s; %" PRIdSize "/%" PRIdSize " chunks decompressed", + name, when, static_cast<size_t>(chunkAvailNum), nEntries); + + size_t len = 64; + UniquePtr<char[]> map = MakeUnique<char[]>(len + 3); + map[0] = '['; + + for (size_t i = 0, j = 1; i < nEntries; i++, j++) { + map[j] = chunkAvail[i] ? '*' : '_'; + if ((j == len) || (i == nEntries - 1)) { + map[j + 1] = ']'; + map[j + 2] = '\0'; + DEBUG_LOG("%s", static_cast<char *>(map.get())); + j = 0; + } + } +} + +size_t +MappableSeekableZStream::GetLength() const +{ + return buffer->GetLength(); +} diff --git a/mozglue/linker/Mappable.h b/mozglue/linker/Mappable.h new file mode 100644 index 0000000000..0224ae3b51 --- /dev/null +++ b/mozglue/linker/Mappable.h @@ -0,0 +1,267 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef Mappable_h +#define Mappable_h + +#include "Zip.h" +#include "SeekableZStream.h" +#include "mozilla/RefPtr.h" +#include "mozilla/UniquePtr.h" +#include "zlib.h" + +/** + * Abstract class to handle mmap()ing from various kind of entities, such as + * plain files or Zip entries. The virtual members are meant to act as the + * equivalent system functions, except mapped memory is always MAP_PRIVATE, + * even though a given implementation may use something different internally. + */ +class Mappable: public mozilla::RefCounted<Mappable> +{ +public: + MOZ_DECLARE_REFCOUNTED_TYPENAME(Mappable) + virtual ~Mappable() { } + + virtual MemoryRange mmap(const void *addr, size_t length, int prot, int flags, + off_t offset) = 0; + + enum Kind { + MAPPABLE_FILE, + MAPPABLE_EXTRACT_FILE, + MAPPABLE_DEFLATE, + MAPPABLE_SEEKABLE_ZSTREAM + }; + + virtual Kind GetKind() const = 0; + +private: + virtual void munmap(void *addr, size_t length) { + ::munmap(addr, length); + } + /* Limit use of Mappable::munmap to classes that keep track of the address + * and size of the mapping. This allows to ignore ::munmap return value. */ + friend class Mappable1stPagePtr; + friend class LibHandle; + +public: + /** + * Ensures the availability of the memory pages for the page(s) containing + * the given address. Returns whether the pages were successfully made + * available. + */ + virtual bool ensure(const void *addr) { return true; } + + /** + * Indicate to a Mappable instance that no further mmap is going to happen. + */ + virtual void finalize() = 0; + + /** + * Shows some stats about the Mappable instance. + * Meant for MappableSeekableZStream only. + * As Mappables don't keep track of what they are instanciated for, the name + * argument is used to make the stats logging useful to the reader. The when + * argument is to be used by the caller to give an identifier of the when + * the stats call is made. + */ + virtual void stats(const char *when, const char *name) const { } + + /** + * Returns the maximum length that can be mapped from this Mappable for + * offset = 0. + */ + virtual size_t GetLength() const = 0; +}; + +/** + * Mappable implementation for plain files + */ +class MappableFile: public Mappable +{ +public: + ~MappableFile() { } + + /** + * Create a MappableFile instance for the given file path. + */ + static Mappable *Create(const char *path); + + /* Inherited from Mappable */ + virtual MemoryRange mmap(const void *addr, size_t length, int prot, int flags, off_t offset); + virtual void finalize(); + virtual size_t GetLength() const; + + virtual Kind GetKind() const { return MAPPABLE_FILE; }; +protected: + MappableFile(int fd): fd(fd) { } + +private: + /* File descriptor */ + AutoCloseFD fd; +}; + +/** + * Mappable implementation for deflated stream in a Zip archive + * Inflates the complete stream into a cache file. + */ +class MappableExtractFile: public MappableFile +{ +public: + ~MappableExtractFile() = default; + + /** + * Create a MappableExtractFile instance for the given Zip stream. The name + * argument is used to create the cache file in the cache directory. + */ + static Mappable *Create(const char *name, Zip *zip, Zip::Stream *stream); + + /* Override finalize from MappableFile */ + virtual void finalize() {} + + virtual Kind GetKind() const { return MAPPABLE_EXTRACT_FILE; }; +private: + /** + * AutoUnlinkFile keeps track of a file name and removes (unlinks) the file + * when the instance is destroyed. + */ + struct UnlinkFile + { + void operator()(char *value) { + unlink(value); + delete [] value; + } + }; + typedef mozilla::UniquePtr<char[], UnlinkFile> AutoUnlinkFile; + + MappableExtractFile(int fd, const char* path) + : MappableFile(fd), path(path) { } + + /* Extracted file path */ + mozilla::UniquePtr<const char[]> path; +}; + +class _MappableBuffer; + +/** + * Mappable implementation for deflated stream in a Zip archive. + * Inflates the mapped bits in a temporary buffer. + */ +class MappableDeflate: public Mappable +{ +public: + ~MappableDeflate(); + + /** + * Create a MappableDeflate instance for the given Zip stream. The name + * argument is used for an appropriately named temporary file, and the Zip + * instance is given for the MappableDeflate to keep a reference of it. + */ + static Mappable *Create(const char *name, Zip *zip, Zip::Stream *stream); + + /* Inherited from Mappable */ + virtual MemoryRange mmap(const void *addr, size_t length, int prot, int flags, off_t offset); + virtual void finalize(); + virtual size_t GetLength() const; + + virtual Kind GetKind() const { return MAPPABLE_DEFLATE; }; +private: + MappableDeflate(_MappableBuffer *buf, Zip *zip, Zip::Stream *stream); + + /* Zip reference */ + RefPtr<Zip> zip; + + /* Decompression buffer */ + mozilla::UniquePtr<_MappableBuffer> buffer; + + /* Zlib data */ + zxx_stream zStream; +}; + +/** + * Mappable implementation for seekable zStreams. + * Inflates the mapped bits in a temporary buffer, on demand. + */ +class MappableSeekableZStream: public Mappable +{ +public: + ~MappableSeekableZStream(); + + /** + * Create a MappableSeekableZStream instance for the given Zip stream. The + * name argument is used for an appropriately named temporary file, and the + * Zip instance is given for the MappableSeekableZStream to keep a reference + * of it. + */ + static Mappable *Create(const char *name, Zip *zip, + Zip::Stream *stream); + + /* Inherited from Mappable */ + virtual MemoryRange mmap(const void *addr, size_t length, int prot, int flags, off_t offset); + virtual void munmap(void *addr, size_t length); + virtual void finalize(); + virtual bool ensure(const void *addr); + virtual void stats(const char *when, const char *name) const; + virtual size_t GetLength() const; + + virtual Kind GetKind() const { return MAPPABLE_SEEKABLE_ZSTREAM; }; +private: + MappableSeekableZStream(Zip *zip); + + /* Zip reference */ + RefPtr<Zip> zip; + + /* Decompression buffer */ + mozilla::UniquePtr<_MappableBuffer> buffer; + + /* Seekable ZStream */ + SeekableZStream zStream; + + /* Keep track of mappings performed with MappableSeekableZStream::mmap so + * that they can be realized by MappableSeekableZStream::ensure. + * Values stored in the struct are those passed to mmap */ + struct LazyMap + { + const void *addr; + size_t length; + int prot; + off_t offset; + + /* Returns addr + length, as a pointer */ + const void *end() const { + return reinterpret_cast<const void *> + (reinterpret_cast<const unsigned char *>(addr) + length); + } + + /* Returns offset + length */ + off_t endOffset() const { + return offset + length; + } + + /* Returns the offset corresponding to the given address */ + off_t offsetOf(const void *ptr) const { + return reinterpret_cast<uintptr_t>(ptr) + - reinterpret_cast<uintptr_t>(addr) + offset; + } + + /* Returns whether the given address is in the LazyMap range */ + bool Contains(const void *ptr) const { + return (ptr >= addr) && (ptr < end()); + } + }; + + /* List of all mappings */ + std::vector<LazyMap> lazyMaps; + + /* Array keeping track of which chunks have already been decompressed. + * Each value is the number of pages decompressed for the given chunk. */ + mozilla::UniquePtr<unsigned char[]> chunkAvail; + + /* Number of chunks that have already been decompressed. */ + mozilla::Atomic<size_t> chunkAvailNum; + + /* Mutex protecting decompression */ + pthread_mutex_t mutex; +}; + +#endif /* Mappable_h */ diff --git a/mozglue/linker/SeekableZStream.cpp b/mozglue/linker/SeekableZStream.cpp new file mode 100644 index 0000000000..6dd0ef6d58 --- /dev/null +++ b/mozglue/linker/SeekableZStream.cpp @@ -0,0 +1,261 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include <algorithm> +#include "SeekableZStream.h" +#include "Logging.h" + +bool +SeekableZStream::Init(const void *buf, size_t length) +{ + const SeekableZStreamHeader *header = SeekableZStreamHeader::validate(buf); + if (!header) { + ERROR("Not a seekable zstream"); + return false; + } + + buffer = reinterpret_cast<const unsigned char *>(buf); + totalSize = header->totalSize; + chunkSize = header->chunkSize; + lastChunkSize = header->lastChunkSize; + windowBits = header->windowBits; + dictionary.Init(buffer + sizeof(SeekableZStreamHeader), header->dictSize); + offsetTable.Init(buffer + sizeof(SeekableZStreamHeader) + header->dictSize, + header->nChunks); + filter = GetFilter(header->filter); + + /* Sanity check */ + if ((chunkSize == 0) || + (!IsPageAlignedSize(chunkSize)) || + (chunkSize > 8 * PageSize()) || + (offsetTable.numElements() < 1) || + (lastChunkSize == 0) || + (lastChunkSize > chunkSize) || + (length < totalSize)) { + ERROR("Malformed or broken seekable zstream"); + return false; + } + + return true; +} + +bool +SeekableZStream::Decompress(void *where, size_t chunk, size_t length) +{ + while (length) { + size_t len = std::min(length, static_cast<size_t>(chunkSize)); + if (!DecompressChunk(where, chunk, len)) + return false; + where = reinterpret_cast<unsigned char *>(where) + len; + length -= len; + chunk++; + } + return true; +} + +bool +SeekableZStream::DecompressChunk(void *where, size_t chunk, size_t length) +{ + if (chunk >= offsetTable.numElements()) { + ERROR("DecompressChunk: chunk #%" PRIdSize " out of range [0-%" PRIdSize ")", + chunk, offsetTable.numElements()); + return false; + } + + bool isLastChunk = (chunk == offsetTable.numElements() - 1); + + size_t chunkLen = isLastChunk ? lastChunkSize : chunkSize; + + if (length == 0 || length > chunkLen) + length = chunkLen; + + DEBUG_LOG("DecompressChunk #%" PRIdSize " @%p (%" PRIdSize "/% " PRIdSize ")", + chunk, where, length, chunkLen); + zxx_stream zStream(&allocator); + zStream.avail_in = (isLastChunk ? totalSize : uint32_t(offsetTable[chunk + 1])) + - uint32_t(offsetTable[chunk]); + zStream.next_in = const_cast<Bytef *>(buffer + uint32_t(offsetTable[chunk])); + zStream.avail_out = length; + zStream.next_out = reinterpret_cast<Bytef *>(where); + + /* Decompress chunk */ + if (inflateInit2(&zStream, windowBits) != Z_OK) { + ERROR("inflateInit failed: %s", zStream.msg); + return false; + } + if (dictionary && inflateSetDictionary(&zStream, dictionary, + dictionary.numElements()) != Z_OK) { + ERROR("inflateSetDictionary failed: %s", zStream.msg); + return false; + } + if (inflate(&zStream, (length == chunkLen) ? Z_FINISH : Z_SYNC_FLUSH) + != (length == chunkLen) ? Z_STREAM_END : Z_OK) { + ERROR("inflate failed: %s", zStream.msg); + return false; + } + if (inflateEnd(&zStream) != Z_OK) { + ERROR("inflateEnd failed: %s", zStream.msg); + return false; + } + if (filter) + filter(chunk * chunkSize, UNFILTER, (unsigned char *)where, chunkLen); + + return true; +} + +/* Branch/Call/Jump conversion filter for Thumb, derived from xz-utils + * by Igor Pavlov and Lasse Collin, published in the public domain */ +static void +BCJ_Thumb_filter(off_t offset, SeekableZStream::FilterDirection dir, + unsigned char *buf, size_t size) +{ + size_t i; + for (i = 0; i + 4 <= size; i += 2) { + if ((buf[i + 1] & 0xf8) == 0xf0 && (buf[i + 3] & 0xf8) == 0xf8) { + uint32_t src = (buf[i] << 11) + | ((buf[i + 1] & 0x07) << 19) + | buf[i + 2] + | ((buf[i + 3] & 0x07) << 8); + src <<= 1; + uint32_t dest; + if (dir == SeekableZStream::FILTER) + dest = offset + (uint32_t)(i) + 4 + src; + else + dest = src - (offset + (uint32_t)(i) + 4); + + dest >>= 1; + buf[i] = dest >> 11; + buf[i + 1] = 0xf0 | ((dest >> 19) & 0x07); + buf[i + 2] = dest; + buf[i + 3] = 0xf8 | ((dest >> 8) & 0x07); + i += 2; + } + } +} + +/* Branch/Call/Jump conversion filter for ARM, derived from xz-utils + * by Igor Pavlov and Lasse Collin, published in the public domain */ +static void +BCJ_ARM_filter(off_t offset, SeekableZStream::FilterDirection dir, + unsigned char *buf, size_t size) +{ + size_t i; + for (i = 0; i + 4 <= size; i += 4) { + if (buf[i + 3] == 0xeb) { + uint32_t src = buf[i] + | (buf[i + 1] << 8) + | (buf[i + 2] << 16); + src <<= 2; + uint32_t dest; + if (dir == SeekableZStream::FILTER) + dest = offset + (uint32_t)(i) + 8 + src; + else + dest = src - (offset + (uint32_t)(i) + 8); + + dest >>= 2; + buf[i] = dest; + buf[i + 1] = dest >> 8; + buf[i + 2] = dest >> 16; + } + } +} + +/* Branch/Call/Jump conversion filter for x86, derived from xz-utils + * by Igor Pavlov and Lasse Collin, published in the public domain */ + +#define Test86MSByte(b) ((b) == 0 || (b) == 0xff) + +static void +BCJ_X86_filter(off_t offset, SeekableZStream::FilterDirection dir, + unsigned char *buf, size_t size) +{ + static const bool MASK_TO_ALLOWED_STATUS[8] = + { true, true, true, false, true, false, false, false }; + + static const uint32_t MASK_TO_BIT_NUMBER[8] = + { 0, 1, 2, 2, 3, 3, 3, 3 }; + + uint32_t prev_mask = 0; + uint32_t prev_pos = 0; + + for (size_t i = 0; i + 5 <= size;) { + uint8_t b = buf[i]; + if (b != 0xe8 && b != 0xe9) { + ++i; + continue; + } + + const uint32_t off = offset + (uint32_t)(i) - prev_pos; + prev_pos = offset + (uint32_t)(i); + + if (off > 5) { + prev_mask = 0; + } else { + for (uint32_t i = 0; i < off; ++i) { + prev_mask &= 0x77; + prev_mask <<= 1; + } + } + + b = buf[i + 4]; + + if (Test86MSByte(b) && MASK_TO_ALLOWED_STATUS[(prev_mask >> 1) & 0x7] + && (prev_mask >> 1) < 0x10) { + + uint32_t src = ((uint32_t)(b) << 24) + | ((uint32_t)(buf[i + 3]) << 16) + | ((uint32_t)(buf[i + 2]) << 8) + | (buf[i + 1]); + + uint32_t dest; + while (true) { + if (dir == SeekableZStream::FILTER) + dest = src + (offset + (uint32_t)(i) + 5); + else + dest = src - (offset + (uint32_t)(i) + 5); + + if (prev_mask == 0) + break; + + const uint32_t i = MASK_TO_BIT_NUMBER[prev_mask >> 1]; + + b = (uint8_t)(dest >> (24 - i * 8)); + + if (!Test86MSByte(b)) + break; + + src = dest ^ ((1 << (32 - i * 8)) - 1); + } + + buf[i + 4] = (uint8_t)(~(((dest >> 24) & 1) - 1)); + buf[i + 3] = (uint8_t)(dest >> 16); + buf[i + 2] = (uint8_t)(dest >> 8); + buf[i + 1] = (uint8_t)(dest); + i += 5; + prev_mask = 0; + + } else { + ++i; + prev_mask |= 1; + if (Test86MSByte(b)) + prev_mask |= 0x10; + } + } +} + +SeekableZStream::ZStreamFilter +SeekableZStream::GetFilter(SeekableZStream::FilterId id) +{ + switch (id) { + case BCJ_THUMB: + return BCJ_Thumb_filter; + case BCJ_ARM: + return BCJ_ARM_filter; + case BCJ_X86: + return BCJ_X86_filter; + default: + return nullptr; + } + return nullptr; +} diff --git a/mozglue/linker/SeekableZStream.h b/mozglue/linker/SeekableZStream.h new file mode 100644 index 0000000000..3505c681e8 --- /dev/null +++ b/mozglue/linker/SeekableZStream.h @@ -0,0 +1,154 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef SeekableZStream_h +#define SeekableZStream_h + +#include "Zip.h" + +/** + * Seekable compressed stream are created by splitting the original + * decompressed data in small chunks and compress these chunks + * individually. + * + * The seekable compressed file format consists in a header defined below, + * followed by a table of 32-bits words containing the offsets for each + * individual compressed chunk, then followed by the compressed chunks. + */ + +#pragma pack(1) +struct SeekableZStreamHeader: public Zip::SignedEntity<SeekableZStreamHeader> +{ + SeekableZStreamHeader() + : Zip::SignedEntity<SeekableZStreamHeader>(magic) + , totalSize(0), chunkSize(0), dictSize(0), nChunks(0), lastChunkSize(0) + , windowBits(0), filter(0) { } + + /* Reuse Zip::SignedEntity to handle the magic number used in the Seekable + * ZStream file format. The magic number is "SeZz". */ + static const uint32_t magic = 0x7a5a6553; + + /* Total size of the stream, including the 4 magic bytes. */ + le_uint32 totalSize; + + /* Chunk size */ + le_uint16 chunkSize; + + /* Size of the dictionary */ + le_uint16 dictSize; + + /* Number of chunks */ + le_uint32 nChunks; + + /* Size of last chunk (> 0, <= Chunk size) */ + le_uint16 lastChunkSize; + + /* windowBits value used when deflating */ + signed char windowBits; + + /* Filter Id */ + unsigned char filter; +}; +#pragma pack() + +static_assert(sizeof(SeekableZStreamHeader) == 5 * 4, + "SeekableZStreamHeader should be 5 32-bits words"); + +/** + * Helper class used to decompress Seekable ZStreams. + */ +class SeekableZStream { +public: + /* Initialize from the given buffer. Returns whether initialization + * succeeded (true) or failed (false). */ + bool Init(const void *buf, size_t length); + + /* Decompresses starting from the given chunk. The decompressed data is + * stored at the given location. The given length, in bytes, indicates + * how much data to decompress. If length is 0, then exactly one chunk + * is decompressed. + * Returns whether decompression succeeded (true) or failed (false). */ + bool Decompress(void *where, size_t chunk, size_t length = 0); + + /* Decompresses the given chunk at the given address. If a length is given, + * only decompresses that amount of data instead of the entire chunk. + * Returns whether decompression succeeded (true) or failed (false). */ + bool DecompressChunk(void *where, size_t chunk, size_t length = 0); + + /* Returns the uncompressed size of the complete zstream */ + size_t GetUncompressedSize() const + { + return (offsetTable.numElements() - 1) * chunkSize + lastChunkSize; + } + + /* Returns the chunk size of the given chunk */ + size_t GetChunkSize(size_t chunk = 0) const { + return (chunk == offsetTable.numElements() - 1) ? lastChunkSize : chunkSize; + } + + /* Returns the number of chunks */ + size_t GetChunksNum() const { + return offsetTable.numElements(); + } + + /** + * Filters used to improve compression rate. + */ + enum FilterDirection { + FILTER, + UNFILTER + }; + typedef void (*ZStreamFilter)(off_t, FilterDirection, + unsigned char *, size_t); + + enum FilterId { + NONE, + BCJ_THUMB, + BCJ_ARM, + BCJ_X86, + FILTER_MAX + }; + static ZStreamFilter GetFilter(FilterId id); + + static ZStreamFilter GetFilter(uint16_t id) { + return GetFilter(static_cast<FilterId>(id)); + } + +private: + /* RAW Seekable SZtream buffer */ + const unsigned char *buffer; + + /* Total size of the stream, including the 4 magic bytes. */ + uint32_t totalSize; + + /* Chunk size */ + uint32_t chunkSize; + + /* Size of last chunk (> 0, <= Chunk size) */ + uint32_t lastChunkSize; + + /* windowBits value used when deflating */ + int windowBits; + + /* Offsets table */ + Array<le_uint32> offsetTable; + + /* Filter */ + ZStreamFilter filter; + + /* Deflate dictionary */ + Array<unsigned char> dictionary; + + /* Special allocator for inflate to use the same buffers for every chunk */ + zxx_stream::StaticAllocator allocator; +}; + +inline void +operator++(SeekableZStream::FilterId &other) +{ + const int orig = static_cast<int>(other); + other = static_cast<SeekableZStream::FilterId>(orig + 1); +} + +#endif /* SeekableZStream_h */ diff --git a/mozglue/linker/Utils.h b/mozglue/linker/Utils.h new file mode 100644 index 0000000000..c5314ef603 --- /dev/null +++ b/mozglue/linker/Utils.h @@ -0,0 +1,618 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef Utils_h +#define Utils_h + +#include <pthread.h> +#include <stdint.h> +#include <stddef.h> +#include <sys/mman.h> +#include <unistd.h> +#include "mozilla/Assertions.h" +#include "mozilla/Scoped.h" + +/** + * On architectures that are little endian and that support unaligned reads, + * we can use direct type, but on others, we want to have a special class + * to handle conversion and alignment issues. + */ +#if !defined(DEBUG) && (defined(__i386__) || defined(__x86_64__)) +typedef uint16_t le_uint16; +typedef uint32_t le_uint32; +#else + +/** + * Template that allows to find an unsigned int type from a (computed) bit size + */ +template <int s> struct UInt { }; +template <> struct UInt<16> { typedef uint16_t Type; }; +template <> struct UInt<32> { typedef uint32_t Type; }; + +/** + * Template to access 2 n-bit sized words as a 2*n-bit sized word, doing + * conversion from little endian and avoiding alignment issues. + */ +template <typename T> +class le_to_cpu +{ +public: + typedef typename UInt<16 * sizeof(T)>::Type Type; + + operator Type() const + { + return (b << (sizeof(T) * 8)) | a; + } + + const le_to_cpu& operator =(const Type &v) + { + a = v & ((1 << (sizeof(T) * 8)) - 1); + b = v >> (sizeof(T) * 8); + return *this; + } + + le_to_cpu() { } + le_to_cpu(const Type &v) + { + operator =(v); + } + + const le_to_cpu& operator +=(const Type &v) + { + return operator =(operator Type() + v); + } + + const le_to_cpu& operator ++(int) + { + return operator =(operator Type() + 1); + } + +private: + T a, b; +}; + +/** + * Type definitions + */ +typedef le_to_cpu<unsigned char> le_uint16; +typedef le_to_cpu<le_uint16> le_uint32; +#endif + + +/** + * AutoCloseFD is a RAII wrapper for POSIX file descriptors + */ +struct AutoCloseFDTraits +{ + typedef int type; + static int empty() { return -1; } + static void release(int fd) { if (fd != -1) close(fd); } +}; +typedef mozilla::Scoped<AutoCloseFDTraits> AutoCloseFD; + +/** + * AutoCloseFILE is a RAII wrapper for POSIX streams + */ +struct AutoCloseFILETraits +{ + typedef FILE *type; + static FILE *empty() { return nullptr; } + static void release(FILE *f) { if (f) fclose(f); } +}; +typedef mozilla::Scoped<AutoCloseFILETraits> AutoCloseFILE; + +/** + * Page alignment helpers + */ +static inline size_t PageSize() +{ + return 4096; +} + +static inline uintptr_t AlignedPtr(uintptr_t ptr, size_t alignment) +{ + return ptr & ~(alignment - 1); +} + +template <typename T> +static inline T *AlignedPtr(T *ptr, size_t alignment) +{ + return reinterpret_cast<T *>( + AlignedPtr(reinterpret_cast<uintptr_t>(ptr), alignment)); +} + +template <typename T> +static inline T PageAlignedPtr(T ptr) +{ + return AlignedPtr(ptr, PageSize()); +} + +static inline uintptr_t AlignedEndPtr(uintptr_t ptr, size_t alignment) +{ + return AlignedPtr(ptr + alignment - 1, alignment); +} + +template <typename T> +static inline T *AlignedEndPtr(T *ptr, size_t alignment) +{ + return reinterpret_cast<T *>( + AlignedEndPtr(reinterpret_cast<uintptr_t>(ptr), alignment)); +} + +template <typename T> +static inline T PageAlignedEndPtr(T ptr) +{ + return AlignedEndPtr(ptr, PageSize()); +} + +static inline size_t AlignedSize(size_t size, size_t alignment) +{ + return (size + alignment - 1) & ~(alignment - 1); +} + +static inline size_t PageAlignedSize(size_t size) +{ + return AlignedSize(size, PageSize()); +} + +static inline bool IsAlignedPtr(uintptr_t ptr, size_t alignment) +{ + return ptr % alignment == 0; +} + +template <typename T> +static inline bool IsAlignedPtr(T *ptr, size_t alignment) +{ + return IsAlignedPtr(reinterpret_cast<uintptr_t>(ptr), alignment); +} + +template <typename T> +static inline bool IsPageAlignedPtr(T ptr) +{ + return IsAlignedPtr(ptr, PageSize()); +} + +static inline bool IsAlignedSize(size_t size, size_t alignment) +{ + return size % alignment == 0; +} + +static inline bool IsPageAlignedSize(size_t size) +{ + return IsAlignedSize(size, PageSize()); +} + +static inline size_t PageNumber(size_t size) +{ + return (size + PageSize() - 1) / PageSize(); +} + +/** + * MemoryRange stores a pointer, size pair. + */ +class MemoryRange +{ +public: + MemoryRange(void *buf, size_t length): buf(buf), length(length) { } + + void Assign(void *b, size_t len) { + buf = b; + length = len; + } + + void Assign(const MemoryRange& other) { + buf = other.buf; + length = other.length; + } + + void *get() const + { + return buf; + } + + operator void *() const + { + return buf; + } + + operator unsigned char *() const + { + return reinterpret_cast<unsigned char *>(buf); + } + + bool operator ==(void *ptr) const { + return buf == ptr; + } + + bool operator ==(unsigned char *ptr) const { + return buf == ptr; + } + + void *operator +(off_t offset) const + { + return reinterpret_cast<char *>(buf) + offset; + } + + /** + * Returns whether the given address is within the mapped range + */ + bool Contains(void *ptr) const + { + return (ptr >= buf) && (ptr < reinterpret_cast<char *>(buf) + length); + } + + /** + * Returns the length of the mapped range + */ + size_t GetLength() const + { + return length; + } + + static MemoryRange mmap(void *addr, size_t length, int prot, int flags, + int fd, off_t offset) { + return MemoryRange(::mmap(addr, length, prot, flags, fd, offset), length); + } + +private: + void *buf; + size_t length; +}; + +/** + * MappedPtr is a RAII wrapper for mmap()ed memory. It can be used as + * a simple void * or unsigned char *. + * + * It is defined as a derivative of a template that allows to use a + * different unmapping strategy. + */ +template <typename T> +class GenericMappedPtr: public MemoryRange +{ +public: + GenericMappedPtr(void *buf, size_t length): MemoryRange(buf, length) { } + GenericMappedPtr(const MemoryRange& other): MemoryRange(other) { } + GenericMappedPtr(): MemoryRange(MAP_FAILED, 0) { } + + void Assign(void *b, size_t len) { + if (get() != MAP_FAILED) + static_cast<T *>(this)->munmap(get(), GetLength()); + MemoryRange::Assign(b, len); + } + + void Assign(const MemoryRange& other) { + Assign(other.get(), other.GetLength()); + } + + ~GenericMappedPtr() + { + if (get() != MAP_FAILED) + static_cast<T *>(this)->munmap(get(), GetLength()); + } + + void release() + { + MemoryRange::Assign(MAP_FAILED, 0); + } +}; + +struct MappedPtr: public GenericMappedPtr<MappedPtr> +{ + MappedPtr(void *buf, size_t length) + : GenericMappedPtr<MappedPtr>(buf, length) { } + MappedPtr(const MemoryRange& other) + : GenericMappedPtr<MappedPtr>(other) { } + MappedPtr(): GenericMappedPtr<MappedPtr>() { } + +private: + friend class GenericMappedPtr<MappedPtr>; + void munmap(void *buf, size_t length) + { + ::munmap(buf, length); + } +}; + +/** + * UnsizedArray is a way to access raw arrays of data in memory. + * + * struct S { ... }; + * UnsizedArray<S> a(buf); + * UnsizedArray<S> b; b.Init(buf); + * + * This is roughly equivalent to + * const S *a = reinterpret_cast<const S *>(buf); + * const S *b = nullptr; b = reinterpret_cast<const S *>(buf); + * + * An UnsizedArray has no known length, and it's up to the caller to make + * sure the accessed memory is mapped and makes sense. + */ +template <typename T> +class UnsizedArray +{ +public: + typedef size_t idx_t; + + /** + * Constructors and Initializers + */ + UnsizedArray(): contents(nullptr) { } + UnsizedArray(const void *buf): contents(reinterpret_cast<const T *>(buf)) { } + + void Init(const void *buf) + { + MOZ_ASSERT(contents == nullptr); + contents = reinterpret_cast<const T *>(buf); + } + + /** + * Returns the nth element of the array + */ + const T &operator[](const idx_t index) const + { + MOZ_ASSERT(contents); + return contents[index]; + } + + operator const T *() const + { + return contents; + } + /** + * Returns whether the array points somewhere + */ + operator bool() const + { + return contents != nullptr; + } +private: + const T *contents; +}; + +/** + * Array, like UnsizedArray, is a way to access raw arrays of data in memory. + * Unlike UnsizedArray, it has a known length, and is enumerable with an + * iterator. + * + * struct S { ... }; + * Array<S> a(buf, len); + * UnsizedArray<S> b; b.Init(buf, len); + * + * In the above examples, len is the number of elements in the array. It is + * also possible to initialize an Array with the buffer size: + * + * Array<S> c; c.InitSize(buf, size); + * + * It is also possible to initialize an Array in two steps, only providing + * one data at a time: + * + * Array<S> d; + * d.Init(buf); + * d.Init(len); // or d.InitSize(size); + * + */ +template <typename T> +class Array: public UnsizedArray<T> +{ +public: + typedef typename UnsizedArray<T>::idx_t idx_t; + + /** + * Constructors and Initializers + */ + Array(): UnsizedArray<T>(), length(0) { } + Array(const void *buf, const idx_t length) + : UnsizedArray<T>(buf), length(length) { } + + void Init(const void *buf) + { + UnsizedArray<T>::Init(buf); + } + + void Init(const idx_t len) + { + MOZ_ASSERT(length == 0); + length = len; + } + + void InitSize(const idx_t size) + { + Init(size / sizeof(T)); + } + + void Init(const void *buf, const idx_t len) + { + UnsizedArray<T>::Init(buf); + Init(len); + } + + void InitSize(const void *buf, const idx_t size) + { + UnsizedArray<T>::Init(buf); + InitSize(size); + } + + /** + * Returns the nth element of the array + */ + const T &operator[](const idx_t index) const + { + MOZ_ASSERT(index < length); + MOZ_ASSERT(operator bool()); + return UnsizedArray<T>::operator[](index); + } + + /** + * Returns the number of elements in the array + */ + idx_t numElements() const + { + return length; + } + + /** + * Returns whether the array points somewhere and has at least one element. + */ + operator bool() const + { + return (length > 0) && UnsizedArray<T>::operator bool(); + } + + /** + * Iterator for an Array. Use is similar to that of STL const_iterators: + * + * struct S { ... }; + * Array<S> a(buf, len); + * for (Array<S>::iterator it = a.begin(); it < a.end(); ++it) { + * // Do something with *it. + * } + */ + class iterator + { + public: + iterator(): item(nullptr) { } + + const T &operator *() const + { + return *item; + } + + const T *operator ->() const + { + return item; + } + + iterator &operator ++() + { + ++item; + return *this; + } + + bool operator<(const iterator &other) const + { + return item < other.item; + } + protected: + friend class Array<T>; + iterator(const T &item): item(&item) { } + + private: + const T *item; + }; + + /** + * Returns an iterator pointing at the beginning of the Array + */ + iterator begin() const { + if (length) + return iterator(UnsizedArray<T>::operator[](0)); + return iterator(); + } + + /** + * Returns an iterator pointing past the end of the Array + */ + iterator end() const { + if (length) + return iterator(UnsizedArray<T>::operator[](length)); + return iterator(); + } + + /** + * Reverse iterator for an Array. Use is similar to that of STL + * const_reverse_iterators: + * + * struct S { ... }; + * Array<S> a(buf, len); + * for (Array<S>::reverse_iterator it = a.rbegin(); it < a.rend(); ++it) { + * // Do something with *it. + * } + */ + class reverse_iterator + { + public: + reverse_iterator(): item(nullptr) { } + + const T &operator *() const + { + const T *tmp = item; + return *--tmp; + } + + const T *operator ->() const + { + return &operator*(); + } + + reverse_iterator &operator ++() + { + --item; + return *this; + } + + bool operator<(const reverse_iterator &other) const + { + return item > other.item; + } + protected: + friend class Array<T>; + reverse_iterator(const T &item): item(&item) { } + + private: + const T *item; + }; + + /** + * Returns a reverse iterator pointing at the end of the Array + */ + reverse_iterator rbegin() const { + if (length) + return reverse_iterator(UnsizedArray<T>::operator[](length)); + return reverse_iterator(); + } + + /** + * Returns a reverse iterator pointing past the beginning of the Array + */ + reverse_iterator rend() const { + if (length) + return reverse_iterator(UnsizedArray<T>::operator[](0)); + return reverse_iterator(); + } +private: + idx_t length; +}; + +/** + * Transforms a pointer-to-function to a pointer-to-object pointing at the + * same address. + */ +template <typename T> +void *FunctionPtr(T func) +{ + union { + void *ptr; + T func; + } f; + f.func = func; + return f.ptr; +} + +class AutoLock { +public: + AutoLock(pthread_mutex_t *mutex): mutex(mutex) + { + if (pthread_mutex_lock(mutex)) + MOZ_CRASH("pthread_mutex_lock failed"); + } + ~AutoLock() + { + if (pthread_mutex_unlock(mutex)) + MOZ_CRASH("pthread_mutex_unlock failed"); + } +private: + pthread_mutex_t *mutex; +}; + +#endif /* Utils_h */ + diff --git a/mozglue/linker/XZStream.cpp b/mozglue/linker/XZStream.cpp new file mode 100644 index 0000000000..bd71655f52 --- /dev/null +++ b/mozglue/linker/XZStream.cpp @@ -0,0 +1,213 @@ +#include "XZStream.h" + +#include <algorithm> +#include "mozilla/Assertions.h" +#include "Logging.h" + +// LZMA dictionary size, should have a minimum size for the given compression +// rate, see XZ Utils docs for details. +static const uint32_t kDictSize = 1 << 24; + +static const size_t kFooterSize = 12; + +// Parses a variable-length integer (VLI), +// see http://tukaani.org/xz/xz-file-format.txt for details. +static size_t +ParseVarLenInt(const uint8_t* aBuf, size_t aBufSize, uint64_t* aValue) +{ + if (!aBufSize) { + return 0; + } + aBufSize = std::min(9u, aBufSize); + + *aValue = aBuf[0] & 0x7F; + size_t i = 0; + + while (aBuf[i++] & 0x80) { + if (i >= aBufSize || aBuf[i] == 0x0) { + return 0; + } + *aValue |= static_cast<uint64_t>(aBuf[i] & 0x7F) << (i * 7); + } + return i; +} + +/* static */ bool +XZStream::IsXZ(const void* aBuf, size_t aBufSize) +{ + static const uint8_t kXzMagic[] = {0xfd, '7', 'z', 'X', 'Z', 0x0}; + MOZ_ASSERT(aBuf); + return aBufSize > sizeof(kXzMagic) && + !memcmp(reinterpret_cast<const void*>(kXzMagic), aBuf, sizeof(kXzMagic)); +} + +XZStream::XZStream(const void* aInBuf, size_t aInSize) + : mInBuf(static_cast<const uint8_t*>(aInBuf)) + , mUncompSize(0) + , mDec(nullptr) +{ + mBuffers.in = mInBuf; + mBuffers.in_pos = 0; + mBuffers.in_size = aInSize; +} + +XZStream::~XZStream() +{ + xz_dec_end(mDec); +} + +bool +XZStream::Init() +{ +#ifdef XZ_USE_CRC64 + xz_crc64_init(); +#endif + xz_crc32_init(); + + mDec = xz_dec_init(XZ_DYNALLOC, kDictSize); + + if (!mDec) { + return false; + } + + mUncompSize = ParseUncompressedSize(); + + return true; +} + +size_t +XZStream::Decode(void* aOutBuf, size_t aOutSize) +{ + if (!mDec) { + return 0; + } + + mBuffers.out = static_cast<uint8_t*>(aOutBuf); + mBuffers.out_pos = 0; + mBuffers.out_size = aOutSize; + + while (mBuffers.in_pos < mBuffers.in_size && + mBuffers.out_pos < mBuffers.out_size) { + const xz_ret ret = xz_dec_run(mDec, &mBuffers); + + switch (ret) { + case XZ_STREAM_END: + // Stream ended, the next loop iteration should terminate. + MOZ_ASSERT(mBuffers.in_pos == mBuffers.in_size); + MOZ_FALLTHROUGH; +#ifdef XZ_DEC_ANY_CHECK + case XZ_UNSUPPORTED_CHECK: + // Ignore unsupported check. + MOZ_FALLTHROUGH; +#endif + case XZ_OK: + // Chunk decoded, proceed. + break; + + case XZ_MEM_ERROR: + ERROR("XZ decoding: memory allocation failed"); + return 0; + + case XZ_MEMLIMIT_ERROR: + ERROR("XZ decoding: memory usage limit reached"); + return 0; + + case XZ_FORMAT_ERROR: + ERROR("XZ decoding: invalid stream format"); + return 0; + + case XZ_OPTIONS_ERROR: + ERROR("XZ decoding: unsupported header options"); + return 0; + + case XZ_DATA_ERROR: + MOZ_FALLTHROUGH; + case XZ_BUF_ERROR: + ERROR("XZ decoding: corrupt input stream"); + return 0; + + default: + MOZ_ASSERT_UNREACHABLE("XZ decoding: unknown error condition"); + return 0; + } + } + return mBuffers.out_pos; +} + +size_t +XZStream::RemainingInput() const +{ + return mBuffers.in_size - mBuffers.in_pos; +} + +size_t +XZStream::Size() const +{ + return mBuffers.in_size; +} + +size_t +XZStream::UncompressedSize() const +{ + return mUncompSize; +} + +size_t +XZStream::ParseIndexSize() const +{ + static const uint8_t kFooterMagic[] = {'Y', 'Z'}; + + const uint8_t* footer = mInBuf + mBuffers.in_size - kFooterSize; + // The magic bytes are at the end of the footer. + if (memcmp(reinterpret_cast<const void*>(kFooterMagic), + footer + kFooterSize - sizeof(kFooterMagic), + sizeof(kFooterMagic))) { + // Not a valid footer at stream end. + return 0; + } + // Backward size is a 32 bit LE integer field positioned after the 32 bit CRC32 + // code. It encodes the index size as a multiple of 4 bytes with a minimum + // size of 4 bytes. + const uint32_t backwardSize = *(footer + 4); + return (backwardSize + 1) * 4; +} + +size_t +XZStream::ParseUncompressedSize() const +{ + static const uint8_t kIndexIndicator[] = {0x0}; + + const size_t indexSize = ParseIndexSize(); + if (!indexSize) { + return 0; + } + // The footer follows directly the index, so we can use it as a reference. + const uint8_t* end = mInBuf + mBuffers.in_size; + const uint8_t* index = end - kFooterSize - indexSize; + + // The index consists of a one byte indicator followed by a VLI field for the + // number of records (1 expected) followed by a list of records. One record + // contains a VLI field for unpadded size followed by a VLI field for + // uncompressed size. + if (memcmp(reinterpret_cast<const void*>(kIndexIndicator), + index, sizeof(kIndexIndicator))) { + // Not a valid index. + return 0; + } + + index += sizeof(kIndexIndicator); + uint64_t numRecords = 0; + index += ParseVarLenInt(index, end - index, &numRecords); + if (!numRecords) { + return 0; + } + uint64_t unpaddedSize = 0; + index += ParseVarLenInt(index, end - index, &unpaddedSize); + if (!unpaddedSize) { + return 0; + } + uint64_t uncompressedSize = 0; + index += ParseVarLenInt(index, end - index, &uncompressedSize); + + return uncompressedSize; +} diff --git a/mozglue/linker/XZStream.h b/mozglue/linker/XZStream.h new file mode 100644 index 0000000000..76a979b135 --- /dev/null +++ b/mozglue/linker/XZStream.h @@ -0,0 +1,48 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef XZSTREAM_h +#define XZSTREAM_h + +#include <cstdlib> + +#define XZ_DEC_DYNALLOC +#include "xz.h" + +// Used to decode XZ stream buffers. +class XZStream +{ +public: + // Returns whether the provided buffer is likely a XZ stream. + static bool IsXZ(const void* aBuf, size_t aBufSize); + + // Creates a XZ stream object for the given input buffer. + XZStream(const void* aInBuf, size_t aInSize); + ~XZStream(); + + // Initializes the decoder and returns whether decoding may commence. + bool Init(); + // Decodes the next chunk of input into the given output buffer. + size_t Decode(void* aOutBuf, size_t aOutSize); + // Returns the number of yet undecoded bytes in the input buffer. + size_t RemainingInput() const; + // Returns the total number of bytes in the input buffer (compressed size). + size_t Size() const; + // Returns the expected final number of bytes in the output buffer. + // Note: will return 0 before successful Init(). + size_t UncompressedSize() const; + +private: + // Parses the stream footer and returns the size of the index in bytes. + size_t ParseIndexSize() const; + // Parses the stream index and returns the expected uncompressed size in bytes. + size_t ParseUncompressedSize() const; + + const uint8_t* mInBuf; + size_t mUncompSize; + xz_buf mBuffers; + xz_dec* mDec; +}; + +#endif // XZSTREAM_h diff --git a/mozglue/linker/Zip.cpp b/mozglue/linker/Zip.cpp new file mode 100644 index 0000000000..d141eaa458 --- /dev/null +++ b/mozglue/linker/Zip.cpp @@ -0,0 +1,229 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include <sys/mman.h> +#include <sys/stat.h> +#include <fcntl.h> +#include <errno.h> +#include <unistd.h> +#include <cstdlib> +#include <algorithm> +#include "Logging.h" +#include "Zip.h" + +already_AddRefed<Zip> +Zip::Create(const char *filename) +{ + /* Open and map the file in memory */ + AutoCloseFD fd(open(filename, O_RDONLY)); + if (fd == -1) { + ERROR("Error opening %s: %s", filename, strerror(errno)); + return nullptr; + } + struct stat st; + if (fstat(fd, &st) == -1) { + ERROR("Error stating %s: %s", filename, strerror(errno)); + return nullptr; + } + size_t size = st.st_size; + if (size <= sizeof(CentralDirectoryEnd)) { + ERROR("Error reading %s: too short", filename); + return nullptr; + } + void *mapped = mmap(nullptr, size, PROT_READ, MAP_SHARED, fd, 0); + if (mapped == MAP_FAILED) { + ERROR("Error mmapping %s: %s", filename, strerror(errno)); + return nullptr; + } + DEBUG_LOG("Mapped %s @%p", filename, mapped); + + return Create(filename, mapped, size); +} + +already_AddRefed<Zip> +Zip::Create(const char *filename, void *mapped, size_t size) +{ + RefPtr<Zip> zip = new Zip(filename, mapped, size); + + // If neither the first Local File entry nor central directory entries + // have been found, the zip was invalid. + if (!zip->nextFile && !zip->entries) { + ERROR("%s - Invalid zip", filename); + return nullptr; + } + + ZipCollection::Singleton.Register(zip); + return zip.forget(); +} + +Zip::Zip(const char *filename, void *mapped, size_t size) +: name(filename ? strdup(filename) : nullptr) +, mapped(mapped) +, size(size) +, nextFile(LocalFile::validate(mapped)) // first Local File entry +, nextDir(nullptr) +, entries(nullptr) +{ + pthread_mutex_init(&mutex, nullptr); + // If the first local file entry couldn't be found (which can happen + // with optimized jars), check the first central directory entry. + if (!nextFile) + GetFirstEntry(); +} + +Zip::~Zip() +{ + ZipCollection::Forget(this); + if (name) { + munmap(mapped, size); + DEBUG_LOG("Unmapped %s @%p", name, mapped); + free(name); + } + pthread_mutex_destroy(&mutex); +} + +bool +Zip::GetStream(const char *path, Zip::Stream *out) const +{ + AutoLock lock(&mutex); + + DEBUG_LOG("%s - GetFile %s", name, path); + /* Fast path: if the Local File header on store matches, we can return the + * corresponding stream right away. + * However, the Local File header may not contain enough information, in + * which case the 3rd bit on the generalFlag is set. Unfortunately, this + * bit is also set in some archives even when we do have the data (most + * notably the android packages as built by the Mozilla build system). + * So instead of testing the generalFlag bit, only use the fast path when + * we haven't read the central directory entries yet, and when the + * compressed size as defined in the header is not filled (which is a + * normal condition for the bit to be set). */ + if (nextFile && nextFile->GetName().Equals(path) && + !entries && (nextFile->compressedSize != 0)) { + DEBUG_LOG("%s - %s was next file: fast path", name, path); + /* Fill Stream info from Local File header content */ + const char *data = reinterpret_cast<const char *>(nextFile->GetData()); + out->compressedBuf = data; + out->compressedSize = nextFile->compressedSize; + out->uncompressedSize = nextFile->uncompressedSize; + out->CRC32 = nextFile->CRC32; + out->type = static_cast<Stream::Type>(uint16_t(nextFile->compression)); + + /* Find the next Local File header. It is usually simply following the + * compressed stream, but in cases where the 3rd bit of the generalFlag + * is set, there is a Data Descriptor header before. */ + data += nextFile->compressedSize; + if ((nextFile->generalFlag & 0x8) && DataDescriptor::validate(data)) { + data += sizeof(DataDescriptor); + } + nextFile = LocalFile::validate(data); + return true; + } + + /* If the directory entry we have in store doesn't match, scan the Central + * Directory for the entry corresponding to the given path */ + if (!nextDir || !nextDir->GetName().Equals(path)) { + const DirectoryEntry *entry = GetFirstEntry(); + DEBUG_LOG("%s - Scan directory entries in search for %s", name, path); + while (entry && !entry->GetName().Equals(path)) { + entry = entry->GetNext(); + } + nextDir = entry; + } + if (!nextDir) { + DEBUG_LOG("%s - Couldn't find %s", name, path); + return false; + } + + /* Find the Local File header corresponding to the Directory entry that + * was found. */ + nextFile = LocalFile::validate(static_cast<const char *>(mapped) + + nextDir->offset); + if (!nextFile) { + ERROR("%s - Couldn't find the Local File header for %s", name, path); + return false; + } + + /* Fill Stream info from Directory entry content */ + const char *data = reinterpret_cast<const char *>(nextFile->GetData()); + out->compressedBuf = data; + out->compressedSize = nextDir->compressedSize; + out->uncompressedSize = nextDir->uncompressedSize; + out->CRC32 = nextDir->CRC32; + out->type = static_cast<Stream::Type>(uint16_t(nextDir->compression)); + + /* Store the next directory entry */ + nextDir = nextDir->GetNext(); + nextFile = nullptr; + return true; +} + +const Zip::DirectoryEntry * +Zip::GetFirstEntry() const +{ + if (entries) + return entries; + + const CentralDirectoryEnd *end = nullptr; + const char *_end = static_cast<const char *>(mapped) + size + - sizeof(CentralDirectoryEnd); + + /* Scan for the Central Directory End */ + for (; _end > mapped && !end; _end--) + end = CentralDirectoryEnd::validate(_end); + if (!end) { + ERROR("%s - Couldn't find end of central directory record", name); + return nullptr; + } + + entries = DirectoryEntry::validate(static_cast<const char *>(mapped) + + end->offset); + if (!entries) { + ERROR("%s - Couldn't find central directory record", name); + } + return entries; +} + +ZipCollection ZipCollection::Singleton; + +static pthread_mutex_t sZipCollectionMutex = PTHREAD_MUTEX_INITIALIZER; + +already_AddRefed<Zip> +ZipCollection::GetZip(const char *path) +{ + { + AutoLock lock(&sZipCollectionMutex); + /* Search the list of Zips we already have for a match */ + for (std::vector<Zip *>::iterator it = Singleton.zips.begin(); + it < Singleton.zips.end(); ++it) { + if ((*it)->GetName() && (strcmp((*it)->GetName(), path) == 0)) { + RefPtr<Zip> zip = *it; + return zip.forget(); + } + } + } + return Zip::Create(path); +} + +void +ZipCollection::Register(Zip *zip) +{ + AutoLock lock(&sZipCollectionMutex); + DEBUG_LOG("ZipCollection::Register(\"%s\")", zip->GetName()); + Singleton.zips.push_back(zip); +} + +void +ZipCollection::Forget(Zip *zip) +{ + AutoLock lock(&sZipCollectionMutex); + DEBUG_LOG("ZipCollection::Forget(\"%s\")", zip->GetName()); + std::vector<Zip *>::iterator it = std::find(Singleton.zips.begin(), + Singleton.zips.end(), zip); + if (*it == zip) { + Singleton.zips.erase(it); + } else { + DEBUG_LOG("ZipCollection::Forget: didn't find \"%s\" in bookkeeping", zip->GetName()); + } +} diff --git a/mozglue/linker/Zip.h b/mozglue/linker/Zip.h new file mode 100644 index 0000000000..29e42592ad --- /dev/null +++ b/mozglue/linker/Zip.h @@ -0,0 +1,500 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef Zip_h +#define Zip_h + +#include <cstring> +#include <stdint.h> +#include <vector> +#include <zlib.h> +#include <pthread.h> +#include "Utils.h" +#include "mozilla/Assertions.h" +#include "mozilla/RefCounted.h" +#include "mozilla/RefPtr.h" + +/** + * Helper class wrapping z_stream to avoid malloc() calls during + * inflate. Do not use for deflate. + * inflateInit allocates two buffers: + * - one for its internal state, which is "approximately 10K bytes" according + * to inflate.h from zlib. + * - one for the compression window, which depends on the window size passed + * to inflateInit2, but is never greater than 32K (1 << MAX_WBITS). + * Those buffers are created at instantiation time instead of when calling + * inflateInit2. When inflateInit2 is called, it will call zxx_stream::Alloc + * to get both these buffers. zxx_stream::Alloc will choose one of the + * pre-allocated buffers depending on the requested size. + */ +class zxx_stream: public z_stream +{ +public: + /* Forward declaration */ + class StaticAllocator; + + explicit zxx_stream(StaticAllocator *allocator_=nullptr) + : allocator(allocator_) + { + memset(this, 0, sizeof(z_stream)); + zalloc = Alloc; + zfree = Free; + opaque = this; + } + +private: + static void *Alloc(void *data, uInt items, uInt size) + { + zxx_stream *zStream = reinterpret_cast<zxx_stream *>(data); + if (zStream->allocator) { + return zStream->allocator->Alloc(items, size); + } + size_t buf_size = items * size; + return ::operator new(buf_size); + } + + static void Free(void *data, void *ptr) + { + zxx_stream *zStream = reinterpret_cast<zxx_stream *>(data); + if (zStream->allocator) { + zStream->allocator->Free(ptr); + } else { + ::operator delete(ptr); + } + } + + /** + * Helper class for each buffer in StaticAllocator. + */ + template <size_t Size> + class ZStreamBuf + { + public: + ZStreamBuf() : inUse(false) { } + + bool get(char*& out) + { + if (!inUse) { + inUse = true; + out = buf; + return true; + } else { + return false; + } + } + + void Release() + { + memset(buf, 0, Size); + inUse = false; + } + + bool Equals(const void *other) { return other == buf; } + + static const size_t size = Size; + + private: + char buf[Size]; + bool inUse; + }; + +public: + /** + * Special allocator that uses static buffers to allocate from. + */ + class StaticAllocator + { + public: + void *Alloc(uInt items, uInt size) + { + if (items == 1 && size <= stateBuf1.size) { + char* res = nullptr; + if (stateBuf1.get(res) || stateBuf2.get(res)) { + return res; + } + MOZ_CRASH("ZStreamBuf already in use"); + } else if (items * size == windowBuf1.size) { + char* res = nullptr; + if (windowBuf1.get(res) || windowBuf2.get(res)) { + return res; + } + MOZ_CRASH("ZStreamBuf already in use"); + } else { + MOZ_CRASH("No ZStreamBuf for allocation"); + } + } + + void Free(void *ptr) + { + if (stateBuf1.Equals(ptr)) { + stateBuf1.Release(); + } else if (stateBuf2.Equals(ptr)) { + stateBuf2.Release(); + }else if (windowBuf1.Equals(ptr)) { + windowBuf1.Release(); + } else if (windowBuf2.Equals(ptr)) { + windowBuf2.Release(); + } else { + MOZ_CRASH("Pointer doesn't match a ZStreamBuf"); + } + } + + // 0x3000 is an arbitrary size above 10K. + ZStreamBuf<0x3000> stateBuf1, stateBuf2; + ZStreamBuf<1 << MAX_WBITS> windowBuf1, windowBuf2; + }; + +private: + StaticAllocator *allocator; +}; + +/** + * Forward declaration + */ +class ZipCollection; + +/** + * Class to handle access to Zip archive streams. The Zip archive is mapped + * in memory, and streams are direct references to that mapped memory. + * Zip files are assumed to be correctly formed. No boundary checks are + * performed, which means hand-crafted malicious Zip archives can make the + * code fail in bad ways. However, since the only intended use is to load + * libraries from Zip archives, there is no interest in making this code + * safe, since the libraries could contain malicious code anyways. + */ +class Zip: public mozilla::external::AtomicRefCounted<Zip> +{ +public: + MOZ_DECLARE_REFCOUNTED_TYPENAME(Zip) + /** + * Create a Zip instance for the given file name. Returns nullptr in case + * of failure. + */ + static already_AddRefed<Zip> Create(const char *filename); + + /** + * Create a Zip instance using the given buffer. + */ + static already_AddRefed<Zip> Create(void *buffer, size_t size) { + return Create(nullptr, buffer, size); + } + +private: + static already_AddRefed<Zip> Create(const char *filename, + void *buffer, size_t size); + + /** + * Private constructor + */ + Zip(const char *filename, void *buffer, size_t size); + +public: + /** + * Destructor + */ + ~Zip(); + + /** + * Class used to access Zip archive item streams + */ + class Stream + { + public: + /** + * Stream types + */ + enum Type { + STORE = 0, + DEFLATE = 8 + }; + + /** + * Constructor + */ + Stream(): compressedBuf(nullptr), compressedSize(0), uncompressedSize(0) + , CRC32(0) + , type(STORE) { } + + /** + * Getters + */ + const void *GetBuffer() { return compressedBuf; } + size_t GetSize() { return compressedSize; } + size_t GetUncompressedSize() { return uncompressedSize; } + size_t GetCRC32() { return CRC32; } + Type GetType() { return type; } + + /** + * Returns a zxx_stream for use with inflate functions using the given + * buffer as inflate output. The caller is expected to allocate enough + * memory for the Stream uncompressed size. + */ + zxx_stream GetZStream(void *buf) + { + zxx_stream zStream; + zStream.avail_in = compressedSize; + zStream.next_in = reinterpret_cast<Bytef *>( + const_cast<void *>(compressedBuf)); + zStream.avail_out = uncompressedSize; + zStream.next_out = static_cast<Bytef *>(buf); + return zStream; + } + + protected: + friend class Zip; + const void *compressedBuf; + size_t compressedSize; + size_t uncompressedSize; + size_t CRC32; + Type type; + }; + + /** + * Returns a stream from the Zip archive. + */ + bool GetStream(const char *path, Stream *out) const; + + /** + * Returns the file name of the archive + */ + const char *GetName() const + { + return name; + } + +private: + /* File name of the archive */ + char *name; + /* Address where the Zip archive is mapped */ + void *mapped; + /* Size of the archive */ + size_t size; + + /** + * Strings (file names, comments, etc.) in the Zip headers are NOT zero + * terminated. This class is a helper around them. + */ + class StringBuf + { + public: + /** + * Constructor + */ + StringBuf(const char *buf, size_t length): buf(buf), length(length) { } + + /** + * Returns whether the string has the same content as the given zero + * terminated string. + */ + bool Equals(const char *str) const + { + return (strncmp(str, buf, length) == 0 && str[length] == '\0'); + } + + private: + const char *buf; + size_t length; + }; + +/* All the following types need to be packed */ +#pragma pack(1) +public: + /** + * A Zip archive is an aggregate of entities which all start with a + * signature giving their type. This template is to be used as a base + * class for these entities. + */ + template <typename T> + class SignedEntity + { + public: + /** + * Equivalent to reinterpret_cast<const T *>(buf), with an additional + * check of the signature. + */ + static const T *validate(const void *buf) + { + const T *ret = static_cast<const T *>(buf); + if (ret->signature == T::magic) + return ret; + return nullptr; + } + + SignedEntity(uint32_t magic): signature(magic) { } + private: + le_uint32 signature; + }; + +private: + /** + * Header used to describe a Local File entry. The header is followed by + * the file name and an extra field, then by the data stream. + */ + struct LocalFile: public SignedEntity<LocalFile> + { + /* Signature for a Local File header */ + static const uint32_t magic = 0x04034b50; + + /** + * Returns the file name + */ + StringBuf GetName() const + { + return StringBuf(reinterpret_cast<const char *>(this) + sizeof(*this), + filenameSize); + } + + /** + * Returns a pointer to the data associated with this header + */ + const void *GetData() const + { + return reinterpret_cast<const char *>(this) + sizeof(*this) + + filenameSize + extraFieldSize; + } + + le_uint16 minVersion; + le_uint16 generalFlag; + le_uint16 compression; + le_uint16 lastModifiedTime; + le_uint16 lastModifiedDate; + le_uint32 CRC32; + le_uint32 compressedSize; + le_uint32 uncompressedSize; + le_uint16 filenameSize; + le_uint16 extraFieldSize; + }; + + /** + * In some cases, when a zip archive is created, compressed size and CRC + * are not known when writing the Local File header. In these cases, the + * 3rd bit of the general flag in the Local File header is set, and there + * is an additional header following the compressed data. + */ + struct DataDescriptor: public SignedEntity<DataDescriptor> + { + /* Signature for a Data Descriptor header */ + static const uint32_t magic = 0x08074b50; + + le_uint32 CRC32; + le_uint32 compressedSize; + le_uint32 uncompressedSize; + }; + + /** + * Header used to describe a Central Directory Entry. The header is + * followed by the file name, an extra field, and a comment. + */ + struct DirectoryEntry: public SignedEntity<DirectoryEntry> + { + /* Signature for a Central Directory Entry header */ + static const uint32_t magic = 0x02014b50; + + /** + * Returns the file name + */ + StringBuf GetName() const + { + return StringBuf(reinterpret_cast<const char *>(this) + sizeof(*this), + filenameSize); + } + + /** + * Returns the Central Directory Entry following this one. + */ + const DirectoryEntry *GetNext() const + { + return validate(reinterpret_cast<const char *>(this) + sizeof(*this) + + filenameSize + extraFieldSize + fileCommentSize); + } + + le_uint16 creatorVersion; + le_uint16 minVersion; + le_uint16 generalFlag; + le_uint16 compression; + le_uint16 lastModifiedTime; + le_uint16 lastModifiedDate; + le_uint32 CRC32; + le_uint32 compressedSize; + le_uint32 uncompressedSize; + le_uint16 filenameSize; + le_uint16 extraFieldSize; + le_uint16 fileCommentSize; + le_uint16 diskNum; + le_uint16 internalAttributes; + le_uint32 externalAttributes; + le_uint32 offset; + }; + + /** + * Header used to describe the End of Central Directory Record. + */ + struct CentralDirectoryEnd: public SignedEntity<CentralDirectoryEnd> + { + /* Signature for the End of Central Directory Record */ + static const uint32_t magic = 0x06054b50; + + le_uint16 diskNum; + le_uint16 startDisk; + le_uint16 recordsOnDisk; + le_uint16 records; + le_uint32 size; + le_uint32 offset; + le_uint16 commentSize; + }; +#pragma pack() + + /** + * Returns the first Directory entry + */ + const DirectoryEntry *GetFirstEntry() const; + + /* Pointer to the Local File Entry following the last one GetStream() used. + * This is used by GetStream to avoid scanning the Directory Entries when the + * requested entry is that one. */ + mutable const LocalFile *nextFile; + + /* Likewise for the next Directory entry */ + mutable const DirectoryEntry *nextDir; + + /* Pointer to the Directory entries */ + mutable const DirectoryEntry *entries; + + mutable pthread_mutex_t mutex; +}; + +/** + * Class for bookkeeping Zip instances + */ +class ZipCollection +{ +public: + static ZipCollection Singleton; + + /** + * Get a Zip instance for the given path. If there is an existing one + * already, return that one, otherwise create a new one. + */ + static already_AddRefed<Zip> GetZip(const char *path); + +protected: + friend class Zip; + /** + * Register the given Zip instance. This method is meant to be called + * by Zip::Create. + */ + static void Register(Zip *zip); + + /** + * Forget about the given Zip instance. This method is meant to be called + * by the Zip destructor. + */ + static void Forget(Zip *zip); + +private: + /* Zip instances bookkept in this collection */ + std::vector<Zip *> zips; +}; + +#endif /* Zip_h */ diff --git a/mozglue/linker/dladdr.h b/mozglue/linker/dladdr.h new file mode 100644 index 0000000000..9a209547a6 --- /dev/null +++ b/mozglue/linker/dladdr.h @@ -0,0 +1,16 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include <dlfcn.h> + +#ifndef HAVE_DLADDR +typedef struct { + const char *dli_fname; + void *dli_fbase; + const char *dli_sname; + void *dli_saddr; +} Dl_info; +extern int dladdr(void *addr, Dl_info *info) __attribute__((weak)); +#define HAVE_DLADDR 1 +#endif diff --git a/mozglue/linker/moz.build b/mozglue/linker/moz.build new file mode 100644 index 0000000000..49f4d62f0c --- /dev/null +++ b/mozglue/linker/moz.build @@ -0,0 +1,54 @@ +# -*- Mode: python; indent-tabs-mode: nil; tab-width: 40 -*- +# vim: set filetype=python: +# This Source Code Form is subject to the terms of the Mozilla Public +# License, v. 2.0. If a copy of the MPL was not distributed with this +# file, You can obtain one at http://mozilla.org/MPL/2.0/. + +SOURCES += [ + 'BaseElf.cpp', + 'CustomElf.cpp', + 'ElfLoader.cpp', + 'Mappable.cpp', + 'SeekableZStream.cpp', + 'XZStream.cpp', + 'Zip.cpp', +] + +Library('linker') + +HOST_SOURCES += [ + 'SeekableZStream.cpp', + 'szip.cpp', +] + +HostProgram('szip') + +FINAL_LIBRARY = 'mozglue' + +DEFINES['IMPL_MFBT'] = True + +DISABLE_STL_WRAPPING = True + +TEST_DIRS += ['tests'] + +HOST_OS_LIBS += [ + 'z', +] + +if CONFIG['TARGET_CPU'] == 'arm': + if CONFIG['MOZ_THUMB2']: + HOST_DEFINES['TARGET_THUMB'] = True + else: + HOST_DEFINES['TARGET_ARM'] = True + +if CONFIG['CPU_ARCH'] == 'x86': + HOST_DEFINES['TARGET_X86'] = True + +if CONFIG['GNU_CXX']: + CXXFLAGS += ['-Wno-error=shadow'] + +DEFINES['XZ_USE_CRC64'] = 1 + +USE_LIBS += [ + 'xz-embedded', +] diff --git a/mozglue/linker/szip.cpp b/mozglue/linker/szip.cpp new file mode 100644 index 0000000000..bfc882fbea --- /dev/null +++ b/mozglue/linker/szip.cpp @@ -0,0 +1,593 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include <algorithm> +#include <map> +#include <sys/stat.h> +#include <string> +#include <sstream> +#include <cstring> +#include <cstdlib> +#include <zlib.h> +#include <fcntl.h> +#include <errno.h> +#include "mozilla/Assertions.h" +#include "mozilla/Scoped.h" +#include "mozilla/UniquePtr.h" +#include "SeekableZStream.h" +#include "Utils.h" +#include "Logging.h" + +Logging Logging::Singleton; + +const char *filterName[] = { + "none", + "thumb", + "arm", + "x86", + "auto" +}; + +/* Maximum supported size for chunkSize */ +static const size_t maxChunkSize = + 1 << (8 * std::min(sizeof(((SeekableZStreamHeader *)nullptr)->chunkSize), + sizeof(((SeekableZStreamHeader *)nullptr)->lastChunkSize)) - 1); + +class Buffer: public MappedPtr +{ +public: + virtual ~Buffer() { } + + virtual bool Resize(size_t size) + { + MemoryRange buf = mmap(nullptr, size, PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANON, -1, 0); + if (buf == MAP_FAILED) + return false; + if (*this != MAP_FAILED) + memcpy(buf, *this, std::min(size, GetLength())); + Assign(buf); + return true; + } + + bool Fill(Buffer &other) + { + size_t size = other.GetLength(); + if (!size || !Resize(size)) + return false; + memcpy(static_cast<void *>(*this), static_cast<void *>(other), size); + return true; + } +}; + +class FileBuffer: public Buffer +{ +public: + bool Init(const char *name, bool writable_ = false) + { + fd = open(name, writable_ ? O_RDWR | O_CREAT | O_TRUNC : O_RDONLY, 0666); + if (fd == -1) + return false; + writable = writable_; + return true; + } + + virtual bool Resize(size_t size) + { + if (writable) { + if (ftruncate(fd, size) == -1) + return false; + } + Assign(MemoryRange::mmap(nullptr, size, + PROT_READ | (writable ? PROT_WRITE : 0), + writable ? MAP_SHARED : MAP_PRIVATE, fd, 0)); + return this != MAP_FAILED; + } + + int getFd() + { + return fd; + } + +private: + AutoCloseFD fd; + bool writable; +}; + +class FilteredBuffer: public Buffer +{ +public: + void Filter(Buffer &other, SeekableZStream::FilterId filter, size_t chunkSize) + { + SeekableZStream::ZStreamFilter filterCB = + SeekableZStream::GetFilter(filter); + MOZ_ASSERT(filterCB); + Fill(other); + size_t size = other.GetLength(); + Bytef *data = reinterpret_cast<Bytef *>(static_cast<void *>(*this)); + size_t avail = 0; + /* Filter needs to be applied in chunks. */ + while (size) { + avail = std::min(size, chunkSize); + filterCB(data - static_cast<unsigned char *>(static_cast<void *>(*this)), + SeekableZStream::FILTER, data, avail); + size -= avail; + data += avail; + } + } +}; + +template <typename T> +class Dictionary: public Buffer +{ + typedef T piece; + typedef std::pair<piece, int> stat_pair; + + static bool stat_cmp(stat_pair a, stat_pair b) + { + return a.second < b.second; + } + +public: + Dictionary(Buffer &inBuf, size_t size) + { + if (!size || !Resize(size)) + return; + DEBUG_LOG("Creating dictionary"); + piece *origBufPieces = reinterpret_cast<piece *>( + static_cast<void *>(inBuf)); + std::map<piece, int> stats; + for (unsigned int i = 0; i < inBuf.GetLength() / sizeof(piece); i++) { + stats[origBufPieces[i]]++; + } + std::vector<stat_pair> statsVec(stats.begin(), stats.end()); + std::sort(statsVec.begin(), statsVec.end(), stat_cmp); + + piece *dictPieces = reinterpret_cast<piece *>( + static_cast<void *>(*this)); + typename std::vector<stat_pair>::reverse_iterator it = statsVec.rbegin(); + for (int i = size / sizeof(piece); i > 0 && it < statsVec.rend(); + i--, ++it) { + dictPieces[i - 1] = it->first; + } + } +}; + +class SzipAction +{ +public: + virtual int run(const char *name, Buffer &origBuf, + const char *outName, Buffer &outBuf) = 0; + + virtual ~SzipAction() {} +}; + +class SzipDecompress: public SzipAction +{ +public: + int run(const char *name, Buffer &origBuf, + const char *outName, Buffer &outBuf); +}; + + +class SzipCompress: public SzipAction +{ +public: + int run(const char *name, Buffer &origBuf, + const char *outName, Buffer &outBuf); + + SzipCompress(size_t aChunkSize, SeekableZStream::FilterId aFilter, + size_t aDictSize) + : chunkSize(aChunkSize ? aChunkSize : 16384) + , filter(aFilter) + , dictSize(aDictSize) + {} + + const static signed char winSizeLog = 15; + const static size_t winSize = 1 << winSizeLog; + + const static SeekableZStream::FilterId DEFAULT_FILTER = +#if defined(TARGET_THUMB) + SeekableZStream::BCJ_THUMB; +#elif defined(TARGET_ARM) + SeekableZStream::BCJ_ARM; +#elif defined(TARGET_X86) + SeekableZStream::BCJ_X86; +#else + SeekableZStream::NONE; +#endif + +private: + + int do_compress(Buffer &origBuf, Buffer &outBuf, const unsigned char *aDict, + size_t aDictSize, SeekableZStream::FilterId aFilter); + + size_t chunkSize; + SeekableZStream::FilterId filter; + size_t dictSize; +}; + +/* Decompress a seekable compressed stream */ +int SzipDecompress::run(const char *name, Buffer &origBuf, + const char *outName, Buffer &outBuf) +{ + size_t origSize = origBuf.GetLength(); + if (origSize < sizeof(SeekableZStreamHeader)) { + ERROR("%s is not compressed", name); + return 0; + } + + SeekableZStream zstream; + if (!zstream.Init(origBuf, origSize)) + return 0; + + size_t size = zstream.GetUncompressedSize(); + + /* Give enough room for the uncompressed data */ + if (!outBuf.Resize(size)) { + ERROR("Error resizing %s: %s", outName, strerror(errno)); + return 1; + } + + if (!zstream.Decompress(outBuf, 0, size)) + return 1; + + return 0; +} + +/* Generate a seekable compressed stream. */ +int SzipCompress::run(const char *name, Buffer &origBuf, + const char *outName, Buffer &outBuf) +{ + size_t origSize = origBuf.GetLength(); + if (origSize == 0) { + ERROR("Won't compress %s: it's empty", name); + return 1; + } + if (SeekableZStreamHeader::validate(origBuf)) { + WARN("Skipping %s: it's already a szip", name); + return 0; + } + bool compressed = false; + LOG("Size = %" PRIuSize, origSize); + + /* Allocate a buffer the size of the uncompressed data: we don't want + * a compressed file larger than that anyways. */ + if (!outBuf.Resize(origSize)) { + ERROR("Couldn't allocate output buffer: %s", strerror(errno)); + return 1; + } + + /* Find the most appropriate filter */ + SeekableZStream::FilterId firstFilter, lastFilter; + bool scanFilters; + if (filter == SeekableZStream::FILTER_MAX) { + firstFilter = SeekableZStream::NONE; + lastFilter = SeekableZStream::FILTER_MAX; + scanFilters = true; + } else { + firstFilter = lastFilter = filter; + ++lastFilter; + scanFilters = false; + } + + mozilla::UniquePtr<Buffer> filteredBuf; + Buffer *origData; + for (SeekableZStream::FilterId f = firstFilter; f < lastFilter; ++f) { + mozilla::UniquePtr<FilteredBuffer> filteredTmp; + Buffer tmpBuf; + if (f != SeekableZStream::NONE) { + DEBUG_LOG("Applying filter \"%s\"", filterName[f]); + filteredTmp = mozilla::MakeUnique<FilteredBuffer>(); + filteredTmp->Filter(origBuf, f, chunkSize); + origData = filteredTmp.get(); + } else { + origData = &origBuf; + } + if (dictSize && !scanFilters) { + filteredBuf = mozilla::Move(filteredTmp); + break; + } + DEBUG_LOG("Compressing with no dictionary"); + if (do_compress(*origData, tmpBuf, nullptr, 0, f) == 0) { + if (tmpBuf.GetLength() < outBuf.GetLength()) { + outBuf.Fill(tmpBuf); + compressed = true; + filter = f; + filteredBuf = mozilla::Move(filteredTmp); + continue; + } + } + } + + origData = filteredBuf ? filteredBuf.get() : &origBuf; + + if (dictSize) { + Dictionary<uint64_t> dict(*origData, dictSize ? SzipCompress::winSize : 0); + + /* Find the most appropriate dictionary size */ + size_t firstDictSize, lastDictSize; + if (dictSize == (size_t) -1) { + /* If we scanned for filters, we effectively already tried dictSize=0 */ + firstDictSize = scanFilters ? 4096 : 0; + lastDictSize = SzipCompress::winSize; + } else { + firstDictSize = lastDictSize = dictSize; + } + + Buffer tmpBuf; + for (size_t d = firstDictSize; d <= lastDictSize; d += 4096) { + DEBUG_LOG("Compressing with dictionary of size %" PRIuSize, d); + if (do_compress(*origData, tmpBuf, static_cast<unsigned char *>(dict) + + SzipCompress::winSize - d, d, filter)) + continue; + if (!compressed || tmpBuf.GetLength() < outBuf.GetLength()) { + outBuf.Fill(tmpBuf); + compressed = true; + dictSize = d; + } + } + } + + if (!compressed) { + outBuf.Fill(origBuf); + LOG("Not compressed"); + return 0; + } + + if (dictSize == (size_t) -1) + dictSize = 0; + + DEBUG_LOG("Used filter \"%s\" and dictionary size of %" PRIuSize, + filterName[filter], dictSize); + LOG("Compressed size is %" PRIuSize, outBuf.GetLength()); + + /* Sanity check */ + Buffer tmpBuf; + SzipDecompress decompress; + if (decompress.run("buffer", outBuf, "buffer", tmpBuf)) + return 1; + + size_t size = tmpBuf.GetLength(); + if (size != origSize) { + ERROR("Compression error: %" PRIuSize " != %" PRIuSize, size, origSize); + return 1; + } + if (memcmp(static_cast<void *>(origBuf), static_cast<void *>(tmpBuf), size)) { + ERROR("Compression error: content mismatch"); + return 1; + } + return 0; +} + +int SzipCompress::do_compress(Buffer &origBuf, Buffer &outBuf, + const unsigned char *aDict, size_t aDictSize, + SeekableZStream::FilterId aFilter) +{ + size_t origSize = origBuf.GetLength(); + MOZ_ASSERT(origSize != 0); + + /* Expected total number of chunks */ + size_t nChunks = ((origSize + chunkSize - 1) / chunkSize); + + /* The first chunk is going to be stored after the header, the dictionary + * and the offset table */ + size_t offset = sizeof(SeekableZStreamHeader) + aDictSize + + nChunks * sizeof(uint32_t); + + if (offset >= origSize) + return 1; + + /* Allocate a buffer the size of the uncompressed data: we don't want + * a compressed file larger than that anyways. */ + if (!outBuf.Resize(origSize)) { + ERROR("Couldn't allocate output buffer: %s", strerror(errno)); + return 1; + } + + SeekableZStreamHeader *header = new (outBuf) SeekableZStreamHeader; + unsigned char *dictionary = static_cast<unsigned char *>( + outBuf + sizeof(SeekableZStreamHeader)); + le_uint32 *entry = + reinterpret_cast<le_uint32 *>(dictionary + aDictSize); + + /* Initialize header */ + header->chunkSize = chunkSize; + header->dictSize = aDictSize; + header->totalSize = offset; + header->windowBits = -SzipCompress::winSizeLog; // Raw stream, + // window size of 32k. + header->filter = aFilter; + if (aDictSize) + memcpy(dictionary, aDict, aDictSize); + + /* Initialize zlib structure */ + z_stream zStream; + memset(&zStream, 0, sizeof(zStream)); + zStream.avail_out = origSize - offset; + zStream.next_out = static_cast<Bytef*>(outBuf) + offset; + + size_t avail = 0; + size_t size = origSize; + unsigned char *data = reinterpret_cast<unsigned char *>( + static_cast<void *>(origBuf)); + while (size) { + avail = std::min(size, chunkSize); + + /* Compress chunk */ + int ret = deflateInit2(&zStream, 9, Z_DEFLATED, header->windowBits, + MAX_MEM_LEVEL, Z_DEFAULT_STRATEGY); + if (aDictSize) + deflateSetDictionary(&zStream, dictionary, aDictSize); + MOZ_ASSERT(ret == Z_OK); + zStream.avail_in = avail; + zStream.next_in = data; + ret = deflate(&zStream, Z_FINISH); + /* Under normal conditions, deflate returns Z_STREAM_END. If there is not + * enough room to compress, deflate returns Z_OK and avail_out is 0. We + * still want to deflateEnd in that case, so fall through. It will bail + * on the avail_out test that follows. */ + MOZ_ASSERT(ret == Z_STREAM_END || ret == Z_OK); + ret = deflateEnd(&zStream); + MOZ_ASSERT(ret == Z_OK); + if (zStream.avail_out <= 0) + return 1; + + size_t len = origSize - offset - zStream.avail_out; + + /* Adjust headers */ + header->totalSize += len; + *entry++ = offset; + header->nChunks++; + + /* Prepare for next iteration */ + size -= avail; + data += avail; + offset += len; + } + header->lastChunkSize = avail; + MOZ_ASSERT(header->totalSize == offset); + MOZ_ASSERT(header->nChunks == nChunks); + + if (!outBuf.Resize(offset)) { + ERROR("Error truncating output: %s", strerror(errno)); + return 1; + } + + return 0; + +} + +bool GetSize(const char *str, size_t *out) +{ + char *end; + MOZ_ASSERT(out); + errno = 0; + *out = strtol(str, &end, 10); + return (!errno && !*end); +} + +int main(int argc, char* argv[]) +{ + mozilla::UniquePtr<SzipAction> action; + char **firstArg; + bool compress = true; + size_t chunkSize = 0; + SeekableZStream::FilterId filter = SzipCompress::DEFAULT_FILTER; + size_t dictSize = (size_t) 0; + + Logging::Init(); + + for (firstArg = &argv[1]; argc > 2; argc--, firstArg++) { + if (!firstArg[0] || firstArg[0][0] != '-') + break; + if (strcmp(firstArg[0], "-d") == 0) { + compress = false; + } else if (strcmp(firstArg[0], "-c") == 0) { + firstArg++; + argc--; + if (!firstArg[0]) + break; + if (!GetSize(firstArg[0], &chunkSize) || !chunkSize || + (chunkSize % 4096) || (chunkSize > maxChunkSize)) { + ERROR("Invalid chunk size"); + return 1; + } + } else if (strcmp(firstArg[0], "-f") == 0) { + firstArg++; + argc--; + if (!firstArg[0]) + break; + bool matched = false; + for (unsigned int i = 0; i < sizeof(filterName) / sizeof(char *); ++i) { + if (strcmp(firstArg[0], filterName[i]) == 0) { + filter = static_cast<SeekableZStream::FilterId>(i); + matched = true; + break; + } + } + if (!matched) { + ERROR("Invalid filter"); + return 1; + } + } else if (strcmp(firstArg[0], "-D") == 0) { + firstArg++; + argc--; + if (!firstArg[0]) + break; + if (strcmp(firstArg[0], "auto") == 0) { + dictSize = -1; + } else if (!GetSize(firstArg[0], &dictSize) || (dictSize >= 1 << 16)) { + ERROR("Invalid dictionary size"); + return 1; + } + } + } + + if (argc != 2 || !firstArg[0]) { + LOG("usage: %s [-d] [-c CHUNKSIZE] [-f FILTER] [-D DICTSIZE] file", + argv[0]); + return 1; + } + + if (compress) { + action.reset(new SzipCompress(chunkSize, filter, dictSize)); + } else { + if (chunkSize) { + ERROR("-c is incompatible with -d"); + return 1; + } + if (dictSize) { + ERROR("-D is incompatible with -d"); + return 1; + } + action.reset(new SzipDecompress()); + } + + std::stringstream tmpOutStream; + tmpOutStream << firstArg[0] << ".sz." << getpid(); + std::string tmpOut(tmpOutStream.str()); + int ret; + struct stat st; + { + FileBuffer origBuf; + if (!origBuf.Init(firstArg[0])) { + ERROR("Couldn't open %s: %s", firstArg[0], strerror(errno)); + return 1; + } + + ret = fstat(origBuf.getFd(), &st); + if (ret == -1) { + ERROR("Couldn't stat %s: %s", firstArg[0], strerror(errno)); + return 1; + } + + size_t origSize = st.st_size; + + /* Mmap the original file */ + if (!origBuf.Resize(origSize)) { + ERROR("Couldn't mmap %s: %s", firstArg[0], strerror(errno)); + return 1; + } + + /* Create the compressed file */ + FileBuffer outBuf; + if (!outBuf.Init(tmpOut.c_str(), true)) { + ERROR("Couldn't open %s: %s", tmpOut.c_str(), strerror(errno)); + return 1; + } + + ret = action->run(firstArg[0], origBuf, tmpOut.c_str(), outBuf); + if ((ret == 0) && (fstat(outBuf.getFd(), &st) == -1)) { + st.st_size = 0; + } + } + + if ((ret == 0) && st.st_size) { + rename(tmpOut.c_str(), firstArg[0]); + } else { + unlink(tmpOut.c_str()); + } + return ret; +} diff --git a/mozglue/linker/tests/TestZip.cpp b/mozglue/linker/tests/TestZip.cpp new file mode 100644 index 0000000000..25873f5e5f --- /dev/null +++ b/mozglue/linker/tests/TestZip.cpp @@ -0,0 +1,69 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include <cstdio> +#include <unistd.h> +#include "Zip.h" +#include "mozilla/RefPtr.h" + +extern "C" void report_mapping() { } +extern "C" void delete_mapping() { } + +/** + * test.zip is a basic test zip file with a central directory. It contains + * four entries, in the following order: + * "foo", "bar", "baz", "qux". + * The entries are going to be read out of order. + */ +const char *test_entries[] = { + "baz", "foo", "bar", "qux" +}; + +/** + * no_central_dir.zip is a hand crafted test zip with no central directory + * entries. The Zip reader is expected to be able to traverse these entries + * if requested in order, without reading a central directory + * - First entry is a file "a", STOREd. + * - Second entry is a file "b", STOREd, using a data descriptor. CRC is + * unknown, but compressed and uncompressed sizes are known in the local + * file header. + * - Third entry is a file "c", DEFLATEd, using a data descriptor. CRC, + * compressed and uncompressed sizes are known in the local file header. + * This is the kind of entry that can be found in a zip that went through + * zipalign if it had a data descriptor originally. + * - Fourth entry is a file "d", STOREd. + */ +const char *no_central_dir_entries[] = { + "a", "b", "c", "d" +}; + +int main(int argc, char *argv[]) +{ + if (argc != 2) { + fprintf(stderr, "TEST-FAIL | TestZip | Expecting the directory containing test Zips\n"); + return 1; + } + chdir(argv[1]); + Zip::Stream s; + RefPtr<Zip> z = ZipCollection::GetZip("test.zip"); + for (size_t i = 0; i < sizeof(test_entries) / sizeof(*test_entries); i++) { + if (!z->GetStream(test_entries[i], &s)) { + fprintf(stderr, "TEST-UNEXPECTED-FAIL | TestZip | test.zip: Couldn't get entry \"%s\"\n", test_entries[i]); + return 1; + } + } + fprintf(stderr, "TEST-PASS | TestZip | test.zip could be accessed fully\n"); + + z = ZipCollection::GetZip("no_central_dir.zip"); + for (size_t i = 0; i < sizeof(no_central_dir_entries) + / sizeof(*no_central_dir_entries); i++) { + if (!z->GetStream(no_central_dir_entries[i], &s)) { + fprintf(stderr, "TEST-UNEXPECTED-FAIL | TestZip | no_central_dir.zip: Couldn't get entry \"%s\"\n", no_central_dir_entries[i]); + return 1; + } + } + fprintf(stderr, "TEST-PASS | TestZip | no_central_dir.zip could be accessed in order\n"); + + return 0; +} diff --git a/mozglue/linker/tests/moz.build b/mozglue/linker/tests/moz.build new file mode 100644 index 0000000000..3d4c1c6162 --- /dev/null +++ b/mozglue/linker/tests/moz.build @@ -0,0 +1,23 @@ +# -*- Mode: python; indent-tabs-mode: nil; tab-width: 40 -*- +# vim: set filetype=python: +# This Source Code Form is subject to the terms of the Mozilla Public +# License, v. 2.0. If a copy of the MPL was not distributed with this +# file, You can obtain one at http://mozilla.org/MPL/2.0/. + +DIST_INSTALL = False + +SimplePrograms([ + 'TestZip', +]) +LOCAL_INCLUDES += ['..'] +USE_LIBS += [ + 'linker', + 'mfbt', +] +OS_LIBS += CONFIG['MOZ_ZLIB_LIBS'] +DISABLE_STL_WRAPPING = True + +PYTHON_UNIT_TESTS += ['run_test_zip.py'] + +if CONFIG['GNU_CXX']: + CXXFLAGS += ['-Wno-error=shadow'] diff --git a/mozglue/linker/tests/no_central_dir.zip b/mozglue/linker/tests/no_central_dir.zip Binary files differnew file mode 100644 index 0000000000..df882220d1 --- /dev/null +++ b/mozglue/linker/tests/no_central_dir.zip diff --git a/mozglue/linker/tests/run_test_zip.py b/mozglue/linker/tests/run_test_zip.py new file mode 100644 index 0000000000..e606e16d49 --- /dev/null +++ b/mozglue/linker/tests/run_test_zip.py @@ -0,0 +1,21 @@ +#!/usr/bin/env python +# +# Any copyright is dedicated to the Public Domain. +# http://creativecommons.org/publicdomain/zero/1.0/ + +import buildconfig +import mozpack.path as mozpath +import mozunit +import subprocess +import unittest + +class TestZip(unittest.TestCase): + def test_zip(self): + srcdir = mozpath.dirname(__file__) + relsrcdir = mozpath.relpath(srcdir, buildconfig.topsrcdir) + test_bin = mozpath.join(buildconfig.topobjdir, relsrcdir, + 'TestZip' + buildconfig.substs['BIN_SUFFIX']) + self.assertEqual(0, subprocess.call([test_bin, srcdir])) + +if __name__ == '__main__': + mozunit.main() diff --git a/mozglue/linker/tests/test.zip b/mozglue/linker/tests/test.zip Binary files differnew file mode 100644 index 0000000000..657835b0ca --- /dev/null +++ b/mozglue/linker/tests/test.zip |