1 //: Loading SubX programs from ELF binaries.
  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   trace(2, "run") << "=== Starting to run" << end();
 10   assert(argc > 2);
 11   reset();
 12   cerr << std::hex;
 13   load_elf(argv[2], argc, argv);
 14   while (EIP < End_of_program)  // weak final-gasp termination check
 15     run_one_instruction();
 16   raise << "executed past end of the world: " << EIP << " vs " << End_of_program << '\n' << end();
 17   return 1;
 18 }
 19 
 20 :(code)
 21 void load_elf(const string& filename, int argc, char* argv[]) {
 22   int fd = open(filename.c_str(), O_RDONLY);
 23   if (fd < 0) raise << filename.c_str() << ": open" << perr() << '\n' << die();
 24   off_t size = lseek(fd, 0, SEEK_END);
 25   lseek(fd, 0, SEEK_SET);
 26   uint8_t* elf_contents = static_cast<uint8_t*>(malloc(size));
 27   if (elf_contents == NULL) raise << "malloc(" << size << ')' << perr() << '\n' << die();
 28   ssize_t read_size = read(fd, elf_contents, size);
 29   if (size != read_size) raise << "read → " << size << " (!= " << read_size << ')' << perr() << '\n' << die();
 30   load_elf_contents(elf_contents, size, argc, argv);
 31   free(elf_contents);
 32 }
 33 
 34 void load_elf_contents(uint8_t* elf_contents, size_t size, int argc, char* argv[]) {
 35   uint8_t magic[5] = {0};
 36   memcpy(magic, elf_contents, 4);
 37   if (memcmp(magic, "\177ELF", 4) != 0)
 38     raise << "Invalid ELF file; starts with \"" << magic << '"' << die();
 39   if (elf_contents[4] != 1)
 40     raise << "Only 32-bit ELF files (4-byte words; virtual addresses up to 4GB) supported.\n" << die();
 41   if (elf_contents[5] != 1)
 42     raise << "Only little-endian ELF files supported.\n" << die();
 43   // unused: remaining 10 bytes of e_ident
 44   uint32_t e_machine_type = u32_in(&elf_contents[16]);
 45   if (e_machine_type != 0x00030002)
 46     raise << "ELF type/machine 0x" << HEXWORD << e_machine_type << " isn't i386 executable\n" << die();
 47   // unused: e_version. We only support version 1, and later versions will be backwards compatible.
 48   uint32_t e_entry = u32_in(&elf_contents[24]);
 49   uint32_t e_phoff = u32_in(&elf_contents[28]);
 50   // unused: e_shoff
 51   // unused: e_flags
 52   uint32_t e_ehsize = u16_in(&elf_contents[40]);
 53   if (e_ehsize < 52) raise << "Invalid binary; ELF header too small\n" << die();
 54   uint32_t e_phentsize = u16_in(&elf_contents[42]);
 55   uint32_t e_phnum = u16_in(&elf_contents[44]);
 56   trace(90, "load") << e_phnum << " entries in the program header, each " << e_phentsize << " bytes long" << end();
 57   // unused: e_shentsize
 58   // unused: e_shnum
 59   // unused: e_shstrndx
 60 
 61   set<uint32_t> overlap;  // to detect overlapping segments
 62   for (size_t i = 0;  i < e_phnum;  ++i)
 63     load_segment_from_program_header(elf_contents, i, size, e_phoff + i*e_phentsize, e_ehsize, overlap);
 64 
 65   // initialize code and stack
 66   assert(overlap.find(STACK_SEGMENT) == overlap.end());
 67   Mem.push_back(vma(STACK_SEGMENT));
 68   assert(overlap.find(AFTER_STACK) == overlap.end());
 69   Reg[ESP].u = AFTER_STACK;
 70   Reg[EBP].u = 0;
 71   EIP = e_entry;
 72 
 73   // initialize args on stack
 74   // no envp for now
 75   // we wastefully use a separate page of memory for argv
 76   Mem.push_back(vma(ARGV_DATA_SEGMENT));
 77   uint32_t argv_data = ARGV_DATA_SEGMENT;
 78   for (int i = argc-1;  i >= /*skip 'subx_bin' and 'run'*/2;  --i) {
 79     push(argv_data);
 80     for (size_t j = 0;  j <= strlen(argv[i]);  ++j) {
 81       assert(overlap.find(argv_data) == overlap.end());  // don't bother comparing ARGV and STACK
 82       write_mem_u8(argv_data, argv[i][j]);
 83       argv_data += sizeof(char);
 84       assert(argv_data < ARGV_DATA_SEGMENT + SEGMENT_ALIGNMENT);
 85     }
 86   }
 87   push(argc-/*skip 'subx_bin' and 'run'*/2);
 88 }
 89 
 90 void push(uint32_t val) {
 91   Reg[ESP].u -= 4;
 92   trace(Callstack_depth+1, "run") << "decrementing ESP to 0x" << HEXWORD << Reg[ESP].u << end();
 93   trace(Callstack_depth+1, "run") << "pushing value 0x" << HEXWORD << val << end();
 94   write_mem_u32(Reg[ESP].u, val);
 95 }
 96 
 97 void load_segment_from_program_header(uint8_t* elf_contents, int segment_index, size_t size, uint32_t offset, uint32_t e_ehsize, set<uint32_t>& overlap) {
 98   uint32_t p_type = u32_in(&elf_contents[offset]);
 99   trace(90, "load") << "program header at offset " << offset << ": type " << p_type << end();
100   if (p_type != 1) {
101     trace(90, "load") << "ignoring segment at offset " << offset << " of non PT_LOAD type " << p_type << " (see http://refspecs.linuxbase.org/elf/elf.pdf)" << end();
102     return;
103   }
104   uint32_t p_offset = u32_in(&elf_contents[offset + 4]);
105   uint32_t p_vaddr = u32_in(&elf_contents[offset + 8]);
106   if (e_ehsize > p_vaddr) raise << "Invalid binary; program header overlaps ELF header\n" << die();
107   // unused: p_paddr
108   uint32_t p_filesz = u32_in(&elf_contents[offset + 16]);
109   uint32_t p_memsz = u32_in(&elf_contents[offset + 20]);
110   if (p_filesz != p_memsz)
111     raise << "Can't yet handle segments where p_filesz != p_memsz (see http://refspecs.linuxbase.org/elf/elf.pdf)\n" << die();
112 
113   if (p_offset + p_filesz > size)
114     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();
115   if (p_memsz >= SEGMENT_ALIGNMENT) {
116     raise << "Code segment too small for SubX; for now please manually increase SEGMENT_ALIGNMENT.\n" << end();
117     return;
118   }
119   trace(90, "load") << "blitting file offsets (" << p_offset << ", " << (p_offset+p_filesz) << ") to addresses (" << p_vaddr << ", " << (p_vaddr+p_memsz) << ')' << end();
120   if (size > p_memsz) size = p_memsz;
121   Mem.push_back(vma(p_vaddr));
122   for (size_t i = 0;  i < p_filesz;  ++i) {
123     assert(overlap.find(p_vaddr+i) == overlap.end());
124     write_mem_u8(p_vaddr+i, elf_contents[p_offset+i]);
125     overlap.insert(p_vaddr+i);
126   }
127   if (segment_index == 0 && End_of_program < p_vaddr+p_memsz)
128     End_of_program = p_vaddr+p_memsz;
129 }
130 
131 :(before "End Includes")
132 // Very primitive/fixed/insecure ELF segments for now: just consecutive VMAs.
133 //   code: 0x09000000 -> 0x09ffffff
134 //   data/heap: 0x0a000000 -> 0x0affffff
135 //   stack: 0x0b000ffc -> 0x0b000000 (downward)
136 const int CODE_SEGMENT = 0x09000000;
137 const int DATA_SEGMENT = 0x0a000000;  // keep sync'd with `Heap.limit` in allocate.subx
138 const int STACK_SEGMENT = 0x0b000000;
139 const int AFTER_STACK = 0x0c000000;
140 const int ARGV_DATA_SEGMENT = 0x0c000000;
141 :(before "End Dump Info for Instruction")
142 //? dump_stack();  // slow
143 :(code)
144 void dump_stack() {
145   ostringstream out;
146   trace(Callstack_depth+1, "run") << "stack:" << end();
147   for (uint32_t a = AFTER_STACK-4;  a > Reg[ESP].u;  a -= 4)
148     trace(Callstack_depth+2, "run") << "  0x" << HEXWORD << a << " => 0x" << HEXWORD << read_mem_u32(a) << end();
149   trace(Callstack_depth+2, "run") << "  0x" << HEXWORD << Reg[ESP].u << " => 0x" << HEXWORD << read_mem_u32(Reg[ESP].u) << "  <=== ESP" << end();
150   for (uint32_t a = Reg[ESP].u-4;  a > Reg[ESP].u-40;  a -= 4)
151     trace(Callstack_depth+2, "run") << "  0x" << HEXWORD << a << " => 0x" << HEXWORD << read_mem_u32(a) << end();
152 }
153 
154 inline uint32_t u32_in(uint8_t* p) {
155   return p[0] | p[1] << 8 | p[2] << 16 | p[3] << 24;
156 }
157 
158 inline uint16_t u16_in(uint8_t* p) {
159   return p[0] | p[1] << 8;
160 }
161 
162 :(before "End Types")
163 struct perr {};
164 :(code)
165 ostream& operator<<(ostream& os, perr /*unused*/) {
166   if (errno)
167     os << ": " << strerror(errno);
168   return os;
169 }
170 
171 :(before "End Types")
172 struct die {};
173 :(code)
174 ostream& operator<<(ostream& /*unused*/, die /*unused*/) {
175   if (Trace_stream) Trace_stream->newline();
176   exit(1);
177 }
178 
179 :(before "End Includes")
180 #include <sys/types.h>
181 #include <sys/stat.h>
182 #include <fcntl.h>
183 #include <stdarg.h>
184 #include <errno.h>
185 #include <unistd.h>