1/* file.c - describe file type
2 *
3 * Copyright 2016 The Android Open Source Project
4 *
5 * See http://pubs.opengroup.org/onlinepubs/9699919799/utilities/file.html
6
7USE_FILE(NEWTOY(file, "<1hL[!hL]", TOYFLAG_USR|TOYFLAG_BIN))
8
9config FILE
10  bool "file"
11  default y
12  help
13    usage: file [-hL] [file...]
14
15    Examine the given files and describe their content types.
16
17    -h	don't follow symlinks (default)
18    -L	follow symlinks
19*/
20
21#define FOR_file
22#include "toys.h"
23
24GLOBALS(
25  int max_name_len;
26)
27
28// We don't trust elf.h to be there, and two codepaths for 32/64 is awkward
29// anyway, so calculate struct offsets manually. (It's a fixed ABI.)
30static void do_elf_file(int fd, struct stat *sb)
31{
32  int endian = toybuf[5], bits = toybuf[4], i, j;
33  int64_t (*elf_int)(void *ptr, unsigned size) = peek_le;
34  // Values from include/linux/elf-em.h (plus arch/*/include/asm/elf.h)
35  // Names are linux/arch/ directory (sometimes before 32/64 bit merges)
36  struct {int val; char *name;} type[] = {{0x9026, "alpha"}, {93, "arc"},
37    {195, "arcv2"}, {40, "arm"}, {183, "arm64"}, {0x18ad, "avr32"},
38    {106, "blackfin"}, {140, "c6x"}, {23, "cell"}, {76, "cris"},
39    {0x5441, "frv"}, {46, "h8300"}, {164, "hexagon"}, {50, "ia64"},
40    {88, "m32r"}, {0x9041, "m32r"}, {4, "m68k"}, {174, "metag"},
41    {0xbaab, "microblaze"}, {8, "mips"}, {10, "mips-old"}, {89, "mn10300"},
42    {0xbeef, "mn10300-old"}, {113, "nios2"}, {92, "openrisc"},
43    {0x8472, "openrisc-old"}, {15, "parisc"}, {20, "ppc"}, {21, "ppc64"},
44    {22, "s390"}, {0xa390, "s390-old"}, {135, "score"}, {42, "sh"},
45    {2, "sparc"}, {18, "sparc8+"}, {43, "sparc9"}, {188, "tile"},
46    {191, "tilegx"}, {3, "386"}, {6, "486"}, {62, "x86-64"}, {94, "xtensa"},
47    {0xabc7, "xtensa-old"}
48  };
49  int dynamic = 0;
50  int stripped = 1;
51  char *map;
52  off_t phoff, shoff;
53  int phentsize, phnum, shsize, shnum;
54
55  printf("ELF ");
56
57  // executable (ELF says this is short but reality says byte, not MSB swapped)
58  i = toybuf[16];
59  if (i == 1) printf("relocatable");
60  else if (i == 2) printf("executable");
61  else if (i == 3) printf("shared object");
62  else if (i == 4) printf("core dump");
63  else printf("(bad type %d)", i);
64  printf(", ");
65
66  // "64-bit"
67  if (bits == 1) printf("32-bit ");
68  else if (bits == 2) printf("64-bit ");
69  else {
70    printf("(bad class %d) ", bits);
71    bits = 0;
72  }
73
74  // "LSB"
75  if (endian == 1) printf("LSB ");
76  else if (endian == 2) {
77    printf("MSB ");
78    elf_int = peek_be;
79  } else {
80    printf("(bad endian %d) \n", endian);
81    endian = 0;
82  }
83
84  // e_machine, ala "x86", from big table above
85  j = elf_int(toybuf+18, 2);
86  for (i = 0; i<ARRAY_LEN(type); i++) if (j==type[i].val) break;
87  if (i<ARRAY_LEN(type)) printf("%s", type[i].name);
88  else printf("(unknown arch %d)", j);
89
90  bits--;
91  // If what we've seen so far doesn't seem consistent, bail.
92  if (!((bits&1)==bits && endian)) {
93    printf(", corrupt?\n");
94    return;
95  }
96
97  // Stash what we need from the header; it's okay to reuse toybuf after this.
98  phentsize = elf_int(toybuf+42+12*bits, 2);
99  phnum = elf_int(toybuf+44+12*bits, 2);
100  phoff = elf_int(toybuf+28+4*bits, 4+4*bits);
101  shsize = elf_int(toybuf+46+12*bits, 2);
102  shnum = elf_int(toybuf+48+12*bits, 2);
103  shoff = elf_int(toybuf+32+8*bits, 4+4*bits);
104
105  // With binutils, phentsize seems to only be non-zero if phnum is non-zero.
106  // Such ELF files are rare, but do exist. (Android's crtbegin files, say.)
107  if (phnum && (phentsize != 32+24*bits)) {
108    printf(", corrupt phentsize %d?\n", phentsize);
109    return;
110  }
111
112  map = xmmap(0, sb->st_size, PROT_READ, MAP_SHARED, fd, 0);
113
114  // We need to read the phdrs for dynamic vs static.
115  // (Note: fields got reordered for 64 bit)
116  for (i = 0; i<phnum; i++) {
117    char *phdr = map+phoff+i*phentsize;
118    int p_type = elf_int(phdr, 4);
119    long long p_offset, p_filesz;
120
121    if (p_type==2 /*PT_DYNAMIC*/) dynamic = 1;
122    if (p_type!=3 /*PT_INTERP*/ && p_type!=4 /*PT_NOTE*/) continue;
123
124    j = bits+1;
125    p_offset = elf_int(phdr+4*j, 4*j);
126    p_filesz = elf_int(phdr+16*j, 4*j);
127
128    if (p_type==3 /*PT_INTERP*/)
129      printf(", dynamic (%.*s)", (int)p_filesz, map+p_offset);
130  }
131  if (!dynamic) printf(", static");
132
133  // We need to read the shdrs for stripped/unstripped and any notes.
134  // Notes are in program headers *and* section headers, but some files don't
135  // contain program headers, so we prefer to check here.
136  // (Note: fields got reordered for 64 bit)
137  for (i = 0; i<shnum; i++) {
138    char *shdr = map+shoff+i*shsize;
139    int sh_type = elf_int(shdr+4, 4);
140    long sh_offset = elf_int(shdr+8+8*(bits+1), 4*(bits+1));
141    int sh_size = elf_int(shdr+8+12*(bits+1), 4);
142
143    if (sh_type == 2 /*SHT_SYMTAB*/) {
144      stripped = 0;
145      break;
146    } else if (sh_type == 7 /*SHT_NOTE*/) {
147      char *note = map+sh_offset;
148
149      // An ELF note is a sequence of entries, each consisting of an
150      // ndhr followed by n_namesz+n_descsz bytes of data (each of those
151      // rounded up to the next 4 bytes, without this being reflected in
152      // the header byte counts themselves).
153      while (sh_size >= 3*4) { // Don't try to read a truncated entry.
154        int n_namesz = elf_int(note, 4);
155        int n_descsz = elf_int(note+4, 4);
156        int n_type = elf_int(note+8, 4);
157        int notesz = 3*4 + ((n_namesz+3)&~3) + ((n_descsz+3)&~3);
158
159        if (n_namesz==4 && !memcmp(note+12, "GNU", 4)) {
160          if (n_type==3 /*NT_GNU_BUILD_ID*/) {
161            printf(", BuildID=");
162            for (j = 0; j < n_descsz; ++j) printf("%02x", note[16 + j]);
163          }
164        } else if (n_namesz==8 && !memcmp(note+12, "Android", 8)) {
165          if (n_type==1 /*.android.note.ident*/) {
166            printf(", for Android %d", (int)elf_int(note+20, 4));
167            if (n_descsz > 24)
168              printf(", built by NDK %.64s (%.64s)", note+24, note+24+64);
169          }
170        }
171
172        note += notesz;
173        sh_size -= notesz;
174      }
175    }
176  }
177  printf(", %sstripped", stripped ? "" : "not ");
178  xputc('\n');
179
180  munmap(map, sb->st_size);
181}
182
183static void do_regular_file(int fd, char *name, struct stat *sb)
184{
185  char *s;
186  int len = read(fd, s = toybuf, sizeof(toybuf)-256);
187  int magic;
188
189  if (len<0) perror_msg("%s", name);
190
191  if (len>40 && strstart(&s, "\177ELF")) do_elf_file(fd, sb);
192  else if (len>=8 && strstart(&s, "!<arch>\n")) xprintf("ar archive\n");
193  else if (len>28 && strstart(&s, "\x89PNG\x0d\x0a\x1a\x0a")) {
194    // PNG is big-endian: https://www.w3.org/TR/PNG/#7Integers-and-byte-order
195    int chunk_length = peek_be(s, 4);
196
197    xprintf("PNG image data");
198
199    // The IHDR chunk comes first: https://www.w3.org/TR/PNG/#11IHDR
200    s += 4;
201    if (chunk_length == 13 && strstart(&s, "IHDR")) {
202      // https://www.w3.org/TR/PNG/#6Colour-values
203      char *c = 0, *colors[] = {"grayscale", 0, "color RGB", "indexed color",
204                                "grayscale with alpha", 0, "color RGBA"};
205
206      if (s[9]<ARRAY_LEN(colors)) c = colors[s[9]];
207      if (!c) c = "unknown";
208
209      xprintf(", %d x %d, %d-bit/%s, %sinterlaced", (int)peek_be(s, 4),
210        (int)peek_be(s+4, 4), s[8], c, s[12] ? "" : "non-");
211    }
212
213    xputc('\n');
214
215  // https://www.w3.org/Graphics/GIF/spec-gif89a.txt
216  } else if (len>16 && (strstart(&s, "GIF87a") || strstart(&s, "GIF89a")))
217    xprintf("GIF image data, %d x %d\n",
218      (int)peek_le(s, 2), (int)peek_le(s+8, 2));
219
220  // TODO: parsing JPEG for width/height is harder than GIF or PNG.
221  else if (len>32 && !memcmp(toybuf, "\xff\xd8", 2)) xputs("JPEG image data");
222
223  // https://docs.oracle.com/javase/specs/jvms/se7/html/jvms-4.html
224  else if (len>8 && strstart(&s, "\xca\xfe\xba\xbe"))
225    xprintf("Java class file, version %d.%d\n",
226      (int)peek_be(s+2, 2), (int)peek_be(s, 2));
227
228  // https://people.freebsd.org/~kientzle/libarchive/man/cpio.5.txt
229  // the lengths for cpio are size of header + 9 bytes, since any valid
230  // cpio archive ends with a record for "TARGET!!!"
231  else if (len>85 && strstart(&s, "07070")) {
232    char *cpioformat = "unknown type";
233
234    if (toybuf[5] == '7') cpioformat = "pre-SVR4 or odc";
235    else if (toybuf[5] == '1') cpioformat = "SVR4 with no CRC";
236    else if (toybuf[5] == '2') cpioformat = "SVR4 with CRC";
237    xprintf("ASCII cpio archive (%s)\n", cpioformat);
238  } else if (len>33 && (magic=peek(&s,2), magic==0143561 || magic==070707)) {
239    if (magic == 0143561) printf("byte-swapped ");
240    xprintf("cpio archive\n");
241  // tar archive (ustar/pax or gnu)
242  } else if (len>500 && !strncmp(s+257, "ustar", 5))
243    xprintf("POSIX tar archive%s\n", strncmp(s+262,"  ",2)?"":" (GNU)");
244  // zip/jar/apk archive, ODF/OOXML document, or such
245  else if (len>5 && strstart(&s, "PK\03\04")) {
246    int ver = toybuf[4];
247
248    xprintf("Zip archive data");
249    if (ver) xprintf(", requires at least v%d.%d to extract", ver/10, ver%10);
250    xputc('\n');
251  } else if (len>4 && strstart(&s, "BZh") && isdigit(*s))
252    xprintf("bzip2 compressed data, block size = %c00k\n", *s);
253  else if (len>10 && strstart(&s, "\x1f\x8b")) xputs("gzip compressed data");
254  else {
255    char *what = 0;
256    int i, bytes;
257
258    // If shell script, report which interpreter
259    if (len>3 && strstart(&s, "#!")) {
260      // Whitespace is allowed between the #! and the interpreter
261      while (isspace(*s)) s++;
262      if (strstart(&s, "/usr/bin/env")) while (isspace(*s)) s++;
263      for (what = s; (s-toybuf)<len && !isspace(*s); s++);
264      strcpy(s, " script");
265
266    // Distinguish ASCII text, UTF-8 text, or data
267    } else for (i = 0; i<len; ++i) {
268      if (!(isprint(toybuf[i]) || isspace(toybuf[i]))) {
269        wchar_t wc;
270        if ((bytes = mbrtowc(&wc, s+i, len-i, 0))>0 && wcwidth(wc)>=0) {
271          i += bytes-1;
272          if (!what) what = "UTF-8 text";
273        } else {
274          what = "data";
275          break;
276        }
277      }
278    }
279    xputs(what ? what : "ASCII text");
280  }
281}
282
283void file_main(void)
284{
285  char **arg;
286
287  for (arg = toys.optargs; *arg; ++arg) {
288    int name_len = strlen(*arg);
289
290    if (name_len > TT.max_name_len) TT.max_name_len = name_len;
291  }
292
293  // Can't use loopfiles here because it doesn't call function when can't open
294  for (arg = toys.optargs; *arg; arg++) {
295    char *name = *arg, *what = "cannot open";
296    struct stat sb;
297    int fd = !strcmp(name, "-");
298
299    xprintf("%s: %*s", name, (int)(TT.max_name_len - strlen(name)), "");
300
301    if (fd || !((toys.optflags & FLAG_L) ? stat : lstat)(name, &sb)) {
302      if (fd || S_ISREG(sb.st_mode)) {
303        if (!sb.st_size) what = "empty";
304        else if ((fd = openro(name, O_RDONLY)) != -1) {
305          do_regular_file(fd, name, &sb);
306          if (fd) close(fd);
307          continue;
308        }
309      } else if (S_ISFIFO(sb.st_mode)) what = "fifo";
310      else if (S_ISBLK(sb.st_mode)) what = "block special";
311      else if (S_ISCHR(sb.st_mode)) what = "character special";
312      else if (S_ISDIR(sb.st_mode)) what = "directory";
313      else if (S_ISSOCK(sb.st_mode)) what = "socket";
314      else if (S_ISLNK(sb.st_mode)) what = "symbolic link";
315      else what = "unknown";
316    }
317
318    xputs(what);
319  }
320}
321