1 //: Generating ELF binaries for SubX programs.
  2 //: This will allow us to run them natively on a Linux kernel.
  3 //: Based on https://github.com/kragen/stoneknifeforth/blob/702d2ebe1b/386.c
  4 
  5 :(before "End Main")
  6 assert(argc > 1);
  7 if (is_equal(argv[1], "run")) {
  8   START_TRACING_UNTIL_END_OF_SCOPE;
  9   assert(argc > 2);
 10   reset();
 11   cerr << std::hex;
 12   initialize_mem();
 13   Mem_offset = CODE_START;
 14   load_elf(argv[2], argc, argv);
 15   while (EIP < End_of_program)  // weak final-gasp termination check
 16     run_one_instruction();
 17   trace(90, "load") << "executed past end of the world: " << EIP << " vs " << End_of_program << end();
 18   return 0;
 19 }
 20 
 21 :(code)
 22 void load_elf(const string& filename, int argc, char* argv[]) {
 23   int fd = open(filename.c_str(), O_RDONLY);
 24   if (fd < 0) raise << filename.c_str() << ": open" << perr() << '\n' << die();
 25   off_t size = lseek(fd, 0, SEEK_END);
 26   lseek(fd, 0, SEEK_SET);
 27   uint8_t* elf_contents = static_cast<uint8_t*>(malloc(size));
 28   if (elf_contents == NULL) raise << "malloc(" << size << ')' << perr() << '\n' << die();
 29   ssize_t read_size = read(fd, elf_contents, size);
 30   if (size != read_size) raise << "read → " << size << " (!= " << read_size << ')' << perr() << '\n' << die();
 31   load_elf_contents(elf_contents, size, argc, argv);
 32   free(elf_contents);
 33 }
 34 
 35 void load_elf_contents(uint8_t* elf_contents, size_t size, int argc, char* argv[]) {
 36   uint8_t magic[5] = {0};
 37   memcpy(magic, elf_contents, 4);
 38   if (memcmp(magic, "\177ELF", 4) != 0)
 39     raise << "Invalid ELF file; starts with \"" << magic << '"' << die();
 40   if (elf_contents[4] != 1)
 41     raise << "Only 32-bit ELF files (4-byte words; virtual addresses up to 4GB) supported.\n" << die();
 42   if (elf_contents[5] != 1)
 43     raise << "Only little-endian ELF files supported.\n" << die();
 44   // unused: remaining 10 bytes of e_ident
 45   uint32_t e_machine_type = u32_in(&elf_contents[16]);
 46   if (e_machine_type != 0x00030002)
 47     raise << "ELF type/machine 0x" << HEXWORD << e_machine_type << " isn't i386 executable\n" << die();
 48   // unused: e_version. We only support version 1, and later versions will be backwards compatible.
 49   uint32_t e_entry = u32_in(&elf_contents[24]);
 50   uint32_t e_phoff = u32_in(&elf_contents[28]);
 51   // unused: e_shoff
 52   // unused: e_flags
 53   uint32_t e_ehsize = u16_in(&elf_contents[40]);
 54   if (e_ehsize < 52) raise << "Invalid binary; ELF header too small\n" << die();
 55   uint32_t e_phentsize = u16_in(&elf_contents[42]);
 56   uint32_t e_phnum = u16_in(&elf_contents[44]);
 57   trace(90, "load") << e_phnum << " entries in the program header, each " << e_phentsize << " bytes long" << end();
 58   // unused: e_shentsize
 59   // unused: e_shnum
 60   // unused: e_shstrndx
 61 
 62   for (size_t i = 0;  i < e_phnum;  ++i)
 63     load_segment_from_program_header(elf_contents, size, e_phoff + i*e_phentsize, e_ehsize);
 64 
 65   // initialize code and stack
 66   Reg[ESP].u = AFTER_STACK;
 67   Reg[EBP].u = 0;
 68   EIP = e_entry;
 69 
 70   // initialize args on stack
 71   // no envp for now
 72   // we wastefully use a separate page of memory for argv
 73   uint32_t argv_data = ARGV_DATA_SEGMENT;
 74   for (int i = argc-1;  i >= /*skip 'subx_bin' and 'run'*/2;  --i) {
 75     push(argv_data);
 76     for (size_t j = 0;  j <= strlen(argv[i]);  ++j) {
 77       write_mem_u8(argv_data, argv[i][j]);
 78       argv_data += sizeof(char);
 79       assert(argv_data < ARGV_DATA_SEGMENT + SEGMENT_SIZE);
 80     }
 81   }
 82   push(argc-/*skip 'subx_bin' and 'run'*/2);
 83 }
 84 
 85 void push(uint32_t val) {
 86   Reg[ESP].u -= 4;
 87   trace(90, "run") << "decrementing ESP to 0x" << HEXWORD << Reg[ESP].u << end();
 88   trace(90, "run") << "pushing value 0x" << HEXWORD << val << end();
 89   write_mem_u32(Reg[ESP].u, val);
 90 }
 91 
 92 void load_segment_from_program_header(uint8_t* elf_contents, size_t size, uint32_t offset, uint32_t e_ehsize) {
 93   uint32_t p_type = u32_in(&elf_contents[offset]);
 94   trace(90, "load") << "program header at offset " << offset << ": type " << p_type << end();
 95   if (p_type != 1) {
 96     trace(90, "load") << "ignoring segment at offset " << offset << " of non PT_LOAD type " << p_type << " (see http://refspecs.linuxbase.org/elf/elf.pdf)" << end();
 97     return;
 98   }
 99   uint32_t p_offset = u32_in(&elf_contents[offset + 4]);
100   uint32_t p_vaddr = u32_in(&elf_contents[offset + 8]);
101   if (e_ehsize > p_vaddr) raise << "Invalid binary; program header overlaps ELF header\n" << die();
102   // unused: p_paddr
103   uint32_t p_filesz = u32_in(&elf_contents[offset + 16]);
104   uint32_t p_memsz = u32_in(&elf_contents[offset + 20]);
105   if (p_filesz != p_memsz)
106     raise << "Can't handle segments where p_filesz != p_memsz (see http://refspecs.linuxbase.org/elf/elf.pdf)\n" << die();
107 
108   if (p_offset + p_filesz > size)
109     raise << "Invalid binary; segment at offset " << offset << " is too large: wants to end at " << p_offset+p_filesz << " but the file ends at " << size << '\n' << die();
110   if (Mem.size() < p_vaddr + p_memsz)
111     Mem.resize(p_vaddr + p_memsz);
112   if (size > p_memsz) size = p_memsz;
113   trace(90, "load") << "blitting file offsets (" << p_offset << ", " << (p_offset+p_filesz) << ") to addresses (" << p_vaddr << ", " << (p_vaddr+p_memsz) << ')' << end();
114   for (size_t i = 0;  i < p_filesz;  ++i)
115     write_mem_u8(p_vaddr+i, elf_contents[p_offset+i]);
116   if (End_of_program < p_vaddr+p_memsz)
117     End_of_program = p_vaddr+p_memsz;
118 }
119 
120 :(before "End Includes")
121 // Very primitive/fixed/insecure ELF segments for now.
122 //   code: 0x08048000 -> 0x08048fff
123 //   data: 0x08049000 -> 0x08049fff
124 //   heap: 0x0804a000 -> 0x0804afff
125 //   stack: 0x0804bfff -> 0x0804b000 (downward)
126 const int CODE_START = 0x08048000;
127 const int SEGMENT_SIZE = 0x1000;
128 const int AFTER_STACK = 0x0804c000;
129 const int ARGV_DATA_SEGMENT = 0x0804e000;
130 :(code)
131 void initialize_mem() {
132   Mem.resize(AFTER_STACK - CODE_START);
133 }
134 
135 inline uint32_t u32_in(uint8_t* p) {
136   return p[0] | p[1] << 8 | p[2] << 16 | p[3] << 24;
137 }
138 
139 inline uint16_t u16_in(uint8_t* p) {
140   return p[0] | p[1] << 8;
141 }
142 
143 :(before "End Types")
144 struct perr {};
145 :(code)
146 ostream& operator<<(ostream& os, perr /*unused*/) {
147   if (errno)
148     os << ": " << strerror(errno);
149   return os;
150 }
151 
152 :(before "End Types")
153 struct die {};
154 :(code)
155 ostream& operator<<(ostream& /*unused*/, die /*unused*/) {
156   if (Trace_stream) Trace_stream->newline();
157   exit(1);
158 }
159 
160 :(before "End Includes")
161 #include <sys/types.h>
162 #include <sys/stat.h>
163 #include <fcntl.h>
164 #include <stdarg.h>
165 #include <errno.h>
166 #include <unistd.h>