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       trace(2, "parse") << argv[i] << end();
 38       ifstream fin(argv[i]);
 39       if (!fin) {
 40         cerr << "could not open " << argv[i] << '\n';
 41         return 1;
 42       }
 43       parse(fin, p);
 44       if (trace_contains_errors()) return 1;
 45     }
 46   }
 47   if (p.segments.empty()) {
 48     print_translate_usage();
 49     cerr << "nothing to do; must provide at least one file to read\n";
 50     exit(1);
 51   }
 52   if (output_filename.empty()) {
 53     print_translate_usage();
 54     cerr << "must provide a filename to write to using '-o'\n";
 55     exit(1);
 56   }
 57   trace(2, "transform") << "begin" << end();
 58   transform(p);
 59   if (trace_contains_errors()) return 1;
 60   trace(2, "translate") << "begin" << end();
 61   save_elf(p, output_filename);
 62   if (trace_contains_errors()) {
 63     unlink(output_filename.c_str());
 64     return 1;
 65   }
 66   // End subx translate
 67   return 0;
 68 }
 69 
 70 :(code)
 71 void print_translate_usage() {
 72   cerr << "Usage: subx translate file1 file2 ... -o output\n";
 73 }
 74 
 75 // write out a program to a bare-bones ELF file
 76 void save_elf(const program& p, const string& filename) {
 77   ofstream out(filename.c_str(), ios::binary);
 78   write_elf_header(out, p);
 79   for (size_t i = 0;  i < p.segments.size();  ++i)
 80     write_segment(p.segments.at(i), out);
 81   out.close();
 82 }
 83 
 84 void write_elf_header(ostream& out, const program& p) {
 85   char c = '\0';
 86 #define O(X)  c = (X); out.write(&c, sizeof(c))
 87 // host is required to be little-endian
 88 #define emit(X)  out.write(reinterpret_cast<const char*>(&X), sizeof(X))
 89   //// ehdr
 90   // e_ident
 91   O(0x7f); O(/*E*/0x45); O(/*L*/0x4c); O(/*F*/0x46);
 92     O(0x1);  // 32-bit format
 93     O(0x1);  // little-endian
 94     O(0x1); O(0x0);
 95   for (size_t i = 0;  i < 8;  ++i) { O(0x0); }
 96   // e_type
 97   O(0x02); O(0x00);
 98   // e_machine
 99   O(0x03); O(0x00);
100   // e_version
101   O(0x01); O(0x00); O(0x00); O(0x00);
102   // e_entry
103   uint32_t e_entry = find(p, "code")->start;
104   // Override e_entry
105   emit(e_entry);
106   // e_phoff -- immediately after ELF header
107   uint32_t e_phoff = 0x34;
108   emit(e_phoff);
109   // e_shoff; unused
110   uint32_t dummy32 = 0;
111   emit(dummy32);
112   // e_flags; unused
113   emit(dummy32);
114   // e_ehsize
115   uint16_t e_ehsize = 0x34;
116   emit(e_ehsize);
117   // e_phentsize
118   uint16_t e_phentsize = 0x20;
119   emit(e_phentsize);
120   // e_phnum
121   uint16_t e_phnum = SIZE(p.segments);
122   emit(e_phnum);
123   // e_shentsize
124   uint16_t dummy16 = 0x0;
125   emit(dummy16);
126   // e_shnum
127   emit(dummy16);
128   // e_shstrndx
129   emit(dummy16);
130 
131   uint32_t p_offset = /*size of ehdr*/0x34 + SIZE(p.segments)*0x20/*size of each phdr*/;
132   for (int i = 0;  i < SIZE(p.segments);  ++i) {
133     const segment& curr = p.segments.at(i);
134     //// phdr
135     // p_type
136     uint32_t p_type = 0x1;
137     emit(p_type);
138     // p_offset
139     emit(p_offset);
140     // p_vaddr
141     uint32_t p_start = curr.start;
142     emit(p_start);
143     // p_paddr
144     emit(p_start);
145     // p_filesz
146     uint32_t size = num_words(curr);
147     assert(p_offset + size < SEGMENT_ALIGNMENT);
148     emit(size);
149     // p_memsz
150     emit(size);
151     // p_flags
152     uint32_t p_flags = (curr.name == "code") ? /*r-x*/0x5 : /*rw-*/0x6;
153     emit(p_flags);
154 
155     // p_align
156     // "As the system creates or augments a process image, it logically copies
157     // a file's segment to a virtual memory segment.  When—and if— the system
158     // physically reads the file depends on the program's execution behavior,
159     // system load, and so on.  A process does not require a physical page
160     // unless it references the logical page during execution, and processes
161     // commonly leave many pages unreferenced. Therefore delaying physical
162     // reads frequently obviates them, improving system performance. To obtain
163     // this efficiency in practice, executable and shared object files must
164     // have segment images whose file offsets and virtual addresses are
165     // congruent, modulo the page size." -- http://refspecs.linuxbase.org/elf/elf.pdf (page 95)
166     uint32_t p_align = 0x1000;  // default page size on linux
167     emit(p_align);
168     if (p_offset % p_align != p_start % p_align) {
169       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();
170       return;
171     }
172 
173     // prepare for next segment
174     p_offset += size;
175   }
176 #undef O
177 #undef emit
178 }
179 
180 void write_segment(const segment& s, ostream& out) {
181   for (int i = 0;  i < SIZE(s.lines);  ++i) {
182     const vector<word>& w = s.lines.at(i).words;
183     for (int j = 0;  j < SIZE(w);  ++j) {
184       uint8_t x = hex_byte(w.at(j).data);  // we're done with metadata by this point
185       out.write(reinterpret_cast<const char*>(&x), /*sizeof(byte)*/1);
186     }
187   }
188 }
189 
190 uint32_t num_words(const segment& s) {
191   uint32_t sum = 0;
192   for (int i = 0;  i < SIZE(s.lines);  ++i)
193     sum += SIZE(s.lines.at(i).words);
194   return sum;
195 }
196 
197 :(before "End Includes")
198 using std::ios;