mirror of
https://github.com/jart/cosmopolitan.git
synced 2025-07-03 01:38:30 +00:00
Rewrite Cosmopolitan Ar
The build/bootstrap/ar.com program is now tinier. This change reduces its size from 140kb to 53kb. Nothing was traded away. Cosmopolitan Ar performance is now 2x better than llvm-ar largely thanks to using the copy_file_range() system call. This change homebrews a new allocation API that addresses the shortcomings of the C standard library design. Using these new balloc() and reballoc() functions I managed to reduce memory consumption so much that Cosmpolitan Ar should now use roughly 100x fewer bytes of peak resident memory compared to llvm-ar. Correct behavior with better compatibility has been assured. Binary output is now pretty much bit-identical to llvm-ar, as of this change. This can and should be the living proof we need to show that a better world is possible for software.
This commit is contained in:
Justine Tunney
parent
197aa0d465
commit
0c630d95b5
27 changed files with 916 additions and 341 deletions
|
|
@ -1,27 +1,162 @@
|
|||
#ifndef COSMOPOLITAN_LIBC_ELF_STRUCT_EHDR_H_
|
||||
#define COSMOPOLITAN_LIBC_ELF_STRUCT_EHDR_H_
|
||||
#include "libc/elf/def.h"
|
||||
#include "libc/elf/scalar.h"
|
||||
|
||||
#define EI_NIDENT 16
|
||||
|
||||
#if !(__ASSEMBLER__ + __LINKER__ + 0)
|
||||
|
||||
/*
|
||||
* ELF header.
|
||||
*/
|
||||
typedef struct Elf64_Ehdr {
|
||||
unsigned char e_ident[EI_NIDENT];
|
||||
|
||||
/*
|
||||
* Leading bytes of ELF header.
|
||||
*
|
||||
* - `e_ident[0]` is always `127`
|
||||
* - `e_ident[1]` is always `'E'`
|
||||
* - `e_ident[2]` is always `'L'`
|
||||
* - `e_ident[3]` is always `'F'`
|
||||
*
|
||||
* - `e_ident[EI_CLASS]` is mandatory and should be:
|
||||
*
|
||||
* - `ELFCLASSNONE64` if it's an Elf64 image
|
||||
* - `ELFCLASSNONE32` if it's an Elf32 image
|
||||
* - Otherwise we assume it's an Elf64 image
|
||||
*
|
||||
* - `e_ident[EI_DATA]` is advisory and could be:
|
||||
*
|
||||
* - `ELFDATANONE` isn't strictly valid
|
||||
* - `ELFDATA2LSB` for little-endian
|
||||
* - `ELFDATA2MSB` for big-endian
|
||||
*
|
||||
* - `e_ident[EI_VERSION]` is advisory and should be:
|
||||
*
|
||||
* - `EV_NONE` if it's zero or unspecified
|
||||
* - `EV_CURRENT` for current ELF version (which is 1)
|
||||
*
|
||||
* - `e_ident[EI_OSABI]` is mandatory and could be:
|
||||
*
|
||||
* - `ELFOSABI_NONE` is zero
|
||||
* - `ELFOSABI_GNU` is for GNU
|
||||
* - `ELFOSABI_SYSV` used by GNU
|
||||
* - `ELFOSABI_LINUX` doesn't care
|
||||
* - `ELFOSABI_FREEBSD` does care (recommended)
|
||||
* - `ELFOSABI_NETBSD` doesn't care (see `PT_NOTE`)
|
||||
* - `ELFOSABI_OPENBSD` doesn't care (see `PT_NOTE`)
|
||||
*
|
||||
* - `e_ident[EI_ABIVERSION]` is advisory
|
||||
*
|
||||
*/
|
||||
unsigned char e_ident[16];
|
||||
|
||||
/*
|
||||
* ELF image type.
|
||||
*
|
||||
* This field is mandatory and should be one of:
|
||||
*
|
||||
* - `ET_REL` for `.o` object files
|
||||
* - `ET_DYN` for `.so` files and `-pie` executables
|
||||
* - `ET_EXEC` for statically-linked executables
|
||||
*
|
||||
*/
|
||||
Elf64_Half e_type;
|
||||
|
||||
/*
|
||||
* ELF machine type.
|
||||
*
|
||||
* This field is mandatory and could be one of:
|
||||
*
|
||||
* - `EM_M32` for Bellmac
|
||||
* - `EM_X86_64` for Amd64
|
||||
* - `EM_AARCH64` for Arm64
|
||||
* - `EM_PPC64` for Raptors
|
||||
* - `EM_RISCV` for Berkeley
|
||||
* - `EM_S390` for System/360
|
||||
*
|
||||
*/
|
||||
Elf64_Half e_machine;
|
||||
|
||||
/*
|
||||
* ELF version.
|
||||
*
|
||||
* This field is advisory and could be:
|
||||
*
|
||||
* - `EV_NONE` if it's zero or unspecified
|
||||
* - `EV_CURRENT` for current ELF version (which is 1)
|
||||
*
|
||||
* @see `e_ident[EI_VERSION]`
|
||||
*/
|
||||
Elf64_Word e_version;
|
||||
|
||||
/*
|
||||
* ELF executable entrypoint.
|
||||
*
|
||||
* Static executables should use this field to store the virtual
|
||||
* address of the _start() function. This field may be zero, for
|
||||
* unspecified.
|
||||
*/
|
||||
Elf64_Addr e_entry;
|
||||
|
||||
/*
|
||||
* `Elf64_Phdr` file offset.
|
||||
*
|
||||
* This field is mandatory. Object files should set it to zero.
|
||||
*/
|
||||
Elf64_Off e_phoff;
|
||||
|
||||
/*
|
||||
* `Elf64_Shdr` file offset.
|
||||
*
|
||||
* This field is advisory.
|
||||
*/
|
||||
Elf64_Off e_shoff;
|
||||
|
||||
/*
|
||||
* ELF flags.
|
||||
*
|
||||
* This field is advisory.
|
||||
*/
|
||||
Elf64_Word e_flags;
|
||||
|
||||
/*
|
||||
* `Elf64_Ehdr` size.
|
||||
*
|
||||
* This field is advisory and should be 64.
|
||||
*/
|
||||
Elf64_Half e_ehsize;
|
||||
|
||||
/*
|
||||
* `Elf64_Phdr` element size.
|
||||
*
|
||||
* This field *is* cared about and should be set to 56. Cosmopolitan
|
||||
* permits larger values for the pleasure of it.
|
||||
*/
|
||||
Elf64_Half e_phentsize;
|
||||
|
||||
/*
|
||||
* `Elf64_Phdr` array count.
|
||||
*/
|
||||
Elf64_Half e_phnum;
|
||||
|
||||
/*
|
||||
* `Elf64_Shdr` element size.
|
||||
*
|
||||
* This field is advisory and should be set to 64. Cosmopolitan
|
||||
* permits larger values for the pleasure of it.
|
||||
*/
|
||||
Elf64_Half e_shentsize;
|
||||
|
||||
/*
|
||||
* `Elf64_Shdr` count.
|
||||
*
|
||||
* This field is advisory.
|
||||
*/
|
||||
Elf64_Half e_shnum;
|
||||
|
||||
/*
|
||||
* Section header index of section name string table.
|
||||
*/
|
||||
Elf64_Half e_shstrndx;
|
||||
|
||||
} Elf64_Ehdr;
|
||||
|
||||
#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
|
||||
|
|
|
|||
|
|
@ -3,9 +3,51 @@
|
|||
#include "libc/elf/scalar.h"
|
||||
#if !(__ASSEMBLER__ + __LINKER__ + 0)
|
||||
|
||||
/*
|
||||
* ELF relocation.
|
||||
*
|
||||
* Relocations let us easily apply fixups to compiled object code. This
|
||||
* data structure represents the contents of an `sh_type` w/ `SHT_REL`.
|
||||
*
|
||||
* @see Elf64_Rela
|
||||
*/
|
||||
typedef struct Elf64_Rel {
|
||||
|
||||
/*
|
||||
* Location to be modified.
|
||||
*
|
||||
* If `e_type` is `ET_REL` then this is a section data byte offset.
|
||||
*
|
||||
* If `e_type` isn't `ET_REL` then this is a virtual address.
|
||||
*/
|
||||
Elf64_Addr r_offset;
|
||||
Elf64_Xword r_info; /** @see ELF64_R_{SYM,SIZE,INFO} */
|
||||
|
||||
/*
|
||||
* Relocation type and symbol.
|
||||
*
|
||||
* This value may be created using:
|
||||
*
|
||||
* r_info = ELF64_R_INFO(sym, type);
|
||||
*
|
||||
* This value may be read using:
|
||||
*
|
||||
* Elf64_Word sym = ELF64_R_SYM(r_info);
|
||||
* Elf64_Word type = ELF64_R_TYPE(r_info);
|
||||
*
|
||||
* Where `sym` is a symbol index, and `type` might be:
|
||||
*
|
||||
* - `R_X86_64_64`
|
||||
* - `R_X86_64_PC32`
|
||||
* - `R_X86_64_GOTPCRELX`
|
||||
* - `R_AARCH64_ABS64`
|
||||
*
|
||||
* Each relocation type specifies a mathematical formula that's used
|
||||
* to compute the appropriate value for the fixed-up object code. If
|
||||
* if needs an addend, then this struct doesn't have one, but it can
|
||||
* still be embedded by the compiler in the location to be modified.
|
||||
*/
|
||||
Elf64_Xword r_info;
|
||||
|
||||
} Elf64_Rel;
|
||||
|
||||
#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
|
||||
|
|
|
|||
|
|
@ -3,15 +3,55 @@
|
|||
#include "libc/elf/scalar.h"
|
||||
#if !(__ASSEMBLER__ + __LINKER__ + 0)
|
||||
|
||||
/*
|
||||
* ELF relocation w/ explicit addend.
|
||||
*
|
||||
* Relocations let us easily apply fixups to compiled object code. This
|
||||
* data structure represents the contents of an `sh_type` w/ `SHT_RELA`
|
||||
*
|
||||
* @see Elf64_Rel
|
||||
*/
|
||||
typedef struct Elf64_Rela {
|
||||
/*u64*/ Elf64_Addr r_offset;
|
||||
|
||||
/*
|
||||
* ELF64_R_SYM(r_info) → sym
|
||||
* ELF64_R_TYPE(r_info) → R_X86_64_{64,PC32,GOTPCRELX,...}
|
||||
* ELF64_R_INFO(sym, type) → r_info
|
||||
* Location to be modified.
|
||||
*
|
||||
* If `e_type` is `ET_REL` then this is a section data byte offset.
|
||||
*
|
||||
* If `e_type` isn't `ET_REL` then this is a virtual address.
|
||||
*/
|
||||
/*u64*/ Elf64_Xword r_info; /* ELF64_R_{SYM,SIZE,INFO} */
|
||||
/*i64*/ Elf64_Sxword r_addend;
|
||||
Elf64_Addr r_offset;
|
||||
|
||||
/*
|
||||
* Relocation type and symbol.
|
||||
*
|
||||
* This value may be created using:
|
||||
*
|
||||
* r_info = ELF64_R_INFO(sym, type);
|
||||
*
|
||||
* This value may be read using:
|
||||
*
|
||||
* Elf64_Word sym = ELF64_R_SYM(r_info);
|
||||
* Elf64_Word type = ELF64_R_TYPE(r_info);
|
||||
*
|
||||
* Where `sym` is a symbol index, and `type` might be:
|
||||
*
|
||||
* - `R_X86_64_64`
|
||||
* - `R_X86_64_PC32`
|
||||
* - `R_X86_64_GOTPCRELX`
|
||||
* - `R_AARCH64_ABS64`
|
||||
*
|
||||
*/
|
||||
Elf64_Xword r_info;
|
||||
|
||||
/*
|
||||
* Relocation parameter.
|
||||
*
|
||||
* Each relocation type has its own mathematical formula, which should
|
||||
* incorporate this value in its own unique way.
|
||||
*/
|
||||
Elf64_Sxword r_addend;
|
||||
|
||||
} Elf64_Rela;
|
||||
|
||||
#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
|
||||
|
|
|
|||
|
|
@ -8,24 +8,44 @@
|
|||
* @see https://docs.oracle.com/cd/E19683-01/816-1386/chapter6-94076/index.html
|
||||
*/
|
||||
typedef struct Elf64_Shdr {
|
||||
|
||||
Elf64_Word sh_name;
|
||||
Elf64_Word sh_type; /* SHT_{PROGBITS,NOBITS,STRTAB,SYMTAB,RELA,...} */
|
||||
|
||||
Elf64_Word sh_type; /* SHT_{PROGBITS,NOBITS,STRTAB,SYMTAB,RELA,...} */
|
||||
|
||||
Elf64_Xword sh_flags; /* SHF_{WRITE,ALLOC,EXECINSTR,MERGE,STRINGS,...} */
|
||||
|
||||
Elf64_Addr sh_addr;
|
||||
|
||||
Elf64_Off sh_offset;
|
||||
|
||||
Elf64_Xword sh_size;
|
||||
|
||||
/*
|
||||
* If SHT_RELA: Index of section of associated symbol table.
|
||||
* If SHT_SYMTAB: Index of section of associated string table.
|
||||
* Index of linked section header.
|
||||
*
|
||||
* If `sh_type` is `SHT_RELA` then `sh_link` holds the section header
|
||||
* index of the associated symbol table.
|
||||
*
|
||||
* If `sh_type` is `SHT_SYMTAB` then `sh_link` holds the section
|
||||
* header index of the associated string table.
|
||||
*/
|
||||
Elf64_Word sh_link;
|
||||
|
||||
/*
|
||||
* If SHT_RELA: Index of section to which relocations apply.
|
||||
* If SHT_SYMTAB: One greater than symbol table index of last local symbol.
|
||||
* If `sh_type` is `SHT_RELA` then `sh_info` contains the index of the
|
||||
* section to which relocations apply.
|
||||
*
|
||||
* If `sh_type` is `SHT_SYMTAB` or `SHT_DYNSYM` then `sh_info`
|
||||
* contains an index that's one greater than symbol table index of
|
||||
* last `STB_LOCAL` symbol.
|
||||
*/
|
||||
Elf64_Word sh_info;
|
||||
|
||||
Elf64_Xword sh_addralign;
|
||||
|
||||
Elf64_Xword sh_entsize;
|
||||
|
||||
} Elf64_Shdr;
|
||||
|
||||
#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
|
||||
|
|
|
|||
|
|
@ -4,18 +4,91 @@
|
|||
#if !(__ASSEMBLER__ + __LINKER__ + 0)
|
||||
|
||||
typedef struct Elf64_Sym {
|
||||
|
||||
/*
|
||||
* Symbol name.
|
||||
*
|
||||
* This value is a byte offset into the `.strtab` section. If this
|
||||
* value is zero, then the symbol has no name.
|
||||
*/
|
||||
Elf64_Word st_name;
|
||||
/* ELF64_ST_TYPE(st_info) → STT_{NOTYPE,OBJECT,FUNC,SECTION,FILE,COMMON,...}
|
||||
* ELF64_ST_BIND(st_info) → STB_{LOCAL,GLOBAL,WEAK,...} */
|
||||
|
||||
/*
|
||||
* Symbol type and binding.
|
||||
*
|
||||
* This value may be created using:
|
||||
*
|
||||
* sym.st_info = ELF64_ST_INFO(bind, type);
|
||||
*
|
||||
* This value may be read using:
|
||||
*
|
||||
* int bind = ELF64_ST_BIND(sym.st_info);
|
||||
* int type = ELF64_ST_TYPE(sym.st_info);
|
||||
*
|
||||
* Where `bind` is typically:
|
||||
*
|
||||
* - `STB_LOCAL`
|
||||
* - `STB_GLOBAL`
|
||||
* - `STB_WEAK`
|
||||
*
|
||||
* Where `type` is typically:
|
||||
*
|
||||
* - `STT_NOTYPE`
|
||||
* - `STT_OBJECT`
|
||||
* - `STT_FUNC`
|
||||
* - `STT_SECTION`
|
||||
* - `STT_FILE`
|
||||
* - `STT_COMMON`
|
||||
*/
|
||||
uint8_t st_info;
|
||||
/* STV_{DEFAULT,INTERNAL,HIDDEN,PROTECTED} */
|
||||
|
||||
/*
|
||||
* Symbol visibility.
|
||||
*
|
||||
* This value should be accessed using:
|
||||
*
|
||||
* int visibility = ELF64_ST_VISIBILITY(sym.st_other);
|
||||
*
|
||||
* Where `visibility` is typically:
|
||||
*
|
||||
* - `STV_DEFAULT`
|
||||
* - `STV_INTERNAL`
|
||||
* - `STV_HIDDEN`
|
||||
* - `STV_PROTECTED`
|
||||
*/
|
||||
uint8_t st_other;
|
||||
/* SHN_UNDEF, <section index>, SHN_ABS, SHN_COMMON, etc. */
|
||||
|
||||
/*
|
||||
* Symbol section.
|
||||
*
|
||||
* If `st_shndx` is within `(SHN_UNDEF,SHN_LORESERVE)` then it holds
|
||||
* an index into the section header table.
|
||||
*
|
||||
* Otherwise `st_shndx` is usually one of the following magic numbers:
|
||||
*
|
||||
* - `SHN_UNDEF` means symbol is undefined
|
||||
* - `SHN_ABS` means symbol is a linker integer
|
||||
* - `SHN_COMMON` means symbol is defined traditionally
|
||||
*/
|
||||
Elf64_Section st_shndx;
|
||||
/* byte offset into GetElfSectionAddress(st_shndx) */
|
||||
|
||||
/*
|
||||
* Symbol value.
|
||||
*
|
||||
* If `e_type` is `ET_REL` and `st_shndx` is `SHN_COMMON`, then
|
||||
* `st_value` holds the required symbol alignment, or ≤ 1 if no
|
||||
* alignment is required.
|
||||
*
|
||||
* If `e_type` is `ET_REL` and `st_shndx` is a section index, then
|
||||
* `st_value` holds a byte offset into the section memory.
|
||||
*
|
||||
* If `e_type` isn't `ET_REL` then `st_value` holds a virtual address.
|
||||
*/
|
||||
Elf64_Addr st_value;
|
||||
|
||||
/* byte length optionally set by .size directive */
|
||||
Elf64_Xword st_size;
|
||||
|
||||
} Elf64_Sym;
|
||||
|
||||
#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue