1 //: The bedrock level 1 of abstraction is now done, and we're going to start
  2 //: building levels above it that make programming in x86 machine code a
  3 //: little more ergonomic.
  4 //:
  5 //: All levels will be "pass through by default". Whatever they don't
  6 //: understand they will silently pass through to lower levels.
  7 //:
  8 //: Since raw hex bytes of machine code are always possible to inject, SubX is
  9 //: not a language, and we aren't building a compiler. This is something
 10 //: deliberately leakier. Levels are more for improving auditing, checks and
 11 //: error messages rather than for hiding low-level details.
 12 
 13 //: Translator workflow: read 'source' file. Run a series of transforms on it,
 14 //: each passing through what it doesn't understand. The final program should
 15 //: be just machine code, suitable to write to an ELF binary.
 16 //:
 17 //: Higher levels usually transform code on the basis of metadata.
 18 
 19 :(before "End Main")
 20 if (is_equal(argv[1], "translate")) {
 21   START_TRACING_UNTIL_END_OF_SCOPE;
 22   reset();
 23   // Begin subx translate
 24   program p;
 25   string output_filename;
 26   for (int i = /*skip 'subx translate'*/2;  i < argc;  ++i) {
 27     if (is_equal(argv[i], "-o")) {
 28       ++i;
 29       if (i >= argc) {
 30         print_translate_usage();
 31         cerr << "'-o' must be followed by a filename to write results to\n";
 32         exit(1);
 33       }
 34       output_filename = argv[i];
 35     }
 36     else {
 37       ifstream fin(argv[i]);
 38       if (!fin) {
 39         cerr << "could not open " << argv[i] << '\n';
 40         return 1;
 41       }
 42       parse(fin, p);
 43       if (trace_contains_errors()) return 1;
 44     }
 45   }
 46   if (p.segments.empty()) {
 47     print_translate_usage();
 48     cerr << "nothing to do; must provide at least one file to read\n";
 49     exit(1);
 50   }
 51   if (output_filename.empty()) {
 52     print_translate_usage();
 53     cerr << "must provide a filename to write to using '-o'\n";
 54     exit(1);
 55   }
 56   transform(p);
 57   if (trace_contains_errors()) return 1;
 58   save_elf(p, output_filename);
 59   if (trace_contains_errors()) {
 60     unlink(output_filename.c_str());
 61     return 1;
 62   }
 63   // End subx translate
 64   return 0;
 65 }
 66 
 67 :(code)
 68 void print_translate_usage() {
 69   cerr << "Usage: subx translate file1 file2 ... -o output\n";
 70 }
 71 
 72 // write out a program to a bare-bones ELF file
 73 void save_elf(const program& p, const string& filename) {
 74   ofstream out(filename.c_str(), ios::binary);
 75   write_elf_header(out, p);
 76   for (size_t i = 0;  i < p.segments.size();  ++i)
 77     write_segment(p.segments.at(i), out);
 78   out.close();
 79 }
 80 
 81 void write_elf_header(ostream& out, const program& p) {
 82   char c = '\0';
 83 #define O(X)  c = (X); out.write(&c, sizeof(c))
 84 // host is required to be little-endian
 85 #define emit(X)  out.write(reinterpret_cast<const char*>(&X), sizeof(X))
 86   //// ehdr
 87   // e_ident
 88   O(0x7f); O(/*E*/0x45); O(/*L*/0x4c); O(/*F*/0x46);
 89     O(0x1);  // 32-bit format
 90     O(0x1);  // little-endian
 91     O(0x1); O(0x0);
 92   for (size_t i = 0;  i < 8;  ++i) { O(0x0); }
 93   // e_type
 94   O(0x02); O(0x00);
 95   // e_machine
 96   O(0x03); O(0x00);
 97   // e_version
 98   O(0x01); O(0x00); O(0x00); O(0x00);
 99   // e_entry
100   int e_entry = p.segments.at(0).start;  // convention
101   emit(e_entry);
102   // e_phoff -- immediately after ELF header
103   int e_phoff = 0x34;
104   emit(e_phoff);
105   // e_shoff; unused
106   int dummy32 = 0;
107   emit(dummy32);
108   // e_flags; unused
109   emit(dummy32);
110   // e_ehsize
111   uint16_t e_ehsize = 0x34;
112   emit(e_ehsize);
113   // e_phentsize
114   uint16_t e_phentsize = 0x20;
115   emit(e_phentsize);
116   // e_phnum
117   uint16_t e_phnum = SIZE(p.segments);
118   emit(e_phnum);
119   // e_shentsize
120   uint16_t dummy16 = 0x0;
121   emit(dummy16);
122   // e_shnum
123   emit(dummy16);
124   // e_shstrndx
125   emit(dummy16);
126 
127   uint32_t p_offset = /*size of ehdr*/0x34 + SIZE(p.segments)*0x20/*size of each phdr*/;
128   for (int i = 0;  i < SIZE(p.segments);  ++i) {
129     //// phdr
130     // p_type
131     uint32_t p_type = 0x1;
132     emit(p_type);
133     // p_offset
134     emit(p_offset);
135     // p_vaddr
136     uint32_t p_start = p.segments.at(i).start;
137     emit(p_start);
138     // p_paddr
139     emit(p_start);
140     // p_filesz
141     uint32_t size = num_words(p.segments.at(i));
142     assert(p_offset + size < SEGMENT_ALIGNMENT);
143     emit(size);
144     // p_memsz
145     emit(size);
146     // p_flags
147     uint32_t p_flags = (i == 0) ? /*r-x*/0x5 : /*rw-*/0x6;  // convention: only first segment is code
148     emit(p_flags);
149 
150     // p_align
151     // "As the system creates or augments a process image, it logically copies
152     // a file's segment to a virtual memory segment.  When—and if— the system
153     // physically reads the file depends on the program's execution behavior,
154     // system load, and so on.  A process does not require a physical page
155     // unless it references the logical page during execution, and processes
156     // commonly leave many pages unreferenced. Therefore delaying physical
157     // reads frequently obviates them, improving system performance. To obtain
158     // this efficiency in practice, executable and shared object files must
159     // have segment images whose file offsets and virtual addresses are
160     // congruent, modulo the page size." -- http://refspecs.linuxbase.org/elf/elf.pdf (page 95)
161     uint32_t p_align = 0x1000;  // default page size on linux
162     emit(p_align);
163     if (p_offset % p_align != p_start % p_align) {
164       raise << "segment starting at 0x" << HEXWORD << p_start << " is improperly aligned; alignment for p_offset " << p_offset << " should be " << (p_offset % p_align) << " but is " << (p_start % p_align) << '\n' << end();
165       return;
166     }
167 
168     // prepare for next segment
169     p_offset += size;
170   }
171 #undef O
172 #undef emit
173 }
174 
175 void write_segment(const segment& s, ostream& out) {
176   for (int i = 0;  i < SIZE(s.lines);  ++i) {
177     const vector<word>& w = s.lines.at(i).words;
178     for (int j = 0;  j < SIZE(w);  ++j) {
179       uint8_t x = hex_byte(w.at(j).data);  // we're done with metadata by this point
180       out.write(reinterpret_cast<const char*>(&x), /*sizeof(byte)*/1);
181     }
182   }
183 }
184 
185 uint32_t num_words(const segment& s) {
186   uint32_t sum = 0;
187   for (int i = 0;  i < SIZE(s.lines);  ++i)
188     sum += SIZE(s.lines.at(i).words);
189   return sum;
190 }
191 
192 :(before "End Includes")
193 using std::ios;