diff --binary generates full index on binary files.
[git] / builtin-tar-tree.c
1 /*
2  * Copyright (c) 2005, 2006 Rene Scharfe
3  */
4 #include <time.h>
5 #include "cache.h"
6 #include "tree-walk.h"
7 #include "commit.h"
8 #include "strbuf.h"
9 #include "tar.h"
10 #include "builtin.h"
11 #include "pkt-line.h"
12
13 #define RECORDSIZE      (512)
14 #define BLOCKSIZE       (RECORDSIZE * 20)
15
16 static const char tar_tree_usage[] =
17 "git-tar-tree [--remote=<repo>] <tree-ish> [basedir]";
18
19 static char block[BLOCKSIZE];
20 static unsigned long offset;
21
22 static time_t archive_time;
23 static int tar_umask;
24
25 /* writes out the whole block, but only if it is full */
26 static void write_if_needed(void)
27 {
28         if (offset == BLOCKSIZE) {
29                 write_or_die(1, block, BLOCKSIZE);
30                 offset = 0;
31         }
32 }
33
34 /*
35  * queues up writes, so that all our write(2) calls write exactly one
36  * full block; pads writes to RECORDSIZE
37  */
38 static void write_blocked(const void *data, unsigned long size)
39 {
40         const char *buf = data;
41         unsigned long tail;
42
43         if (offset) {
44                 unsigned long chunk = BLOCKSIZE - offset;
45                 if (size < chunk)
46                         chunk = size;
47                 memcpy(block + offset, buf, chunk);
48                 size -= chunk;
49                 offset += chunk;
50                 buf += chunk;
51                 write_if_needed();
52         }
53         while (size >= BLOCKSIZE) {
54                 write_or_die(1, buf, BLOCKSIZE);
55                 size -= BLOCKSIZE;
56                 buf += BLOCKSIZE;
57         }
58         if (size) {
59                 memcpy(block + offset, buf, size);
60                 offset += size;
61         }
62         tail = offset % RECORDSIZE;
63         if (tail)  {
64                 memset(block + offset, 0, RECORDSIZE - tail);
65                 offset += RECORDSIZE - tail;
66         }
67         write_if_needed();
68 }
69
70 /*
71  * The end of tar archives is marked by 2*512 nul bytes and after that
72  * follows the rest of the block (if any).
73  */
74 static void write_trailer(void)
75 {
76         int tail = BLOCKSIZE - offset;
77         memset(block + offset, 0, tail);
78         write_or_die(1, block, BLOCKSIZE);
79         if (tail < 2 * RECORDSIZE) {
80                 memset(block, 0, offset);
81                 write_or_die(1, block, BLOCKSIZE);
82         }
83 }
84
85 static void strbuf_append_string(struct strbuf *sb, const char *s)
86 {
87         int slen = strlen(s);
88         int total = sb->len + slen;
89         if (total > sb->alloc) {
90                 sb->buf = xrealloc(sb->buf, total);
91                 sb->alloc = total;
92         }
93         memcpy(sb->buf + sb->len, s, slen);
94         sb->len = total;
95 }
96
97 /*
98  * pax extended header records have the format "%u %s=%s\n".  %u contains
99  * the size of the whole string (including the %u), the first %s is the
100  * keyword, the second one is the value.  This function constructs such a
101  * string and appends it to a struct strbuf.
102  */
103 static void strbuf_append_ext_header(struct strbuf *sb, const char *keyword,
104                                      const char *value, unsigned int valuelen)
105 {
106         char *p;
107         int len, total, tmp;
108
109         /* "%u %s=%s\n" */
110         len = 1 + 1 + strlen(keyword) + 1 + valuelen + 1;
111         for (tmp = len; tmp > 9; tmp /= 10)
112                 len++;
113
114         total = sb->len + len;
115         if (total > sb->alloc) {
116                 sb->buf = xrealloc(sb->buf, total);
117                 sb->alloc = total;
118         }
119
120         p = sb->buf;
121         p += sprintf(p, "%u %s=", len, keyword);
122         memcpy(p, value, valuelen);
123         p += valuelen;
124         *p = '\n';
125         sb->len = total;
126 }
127
128 static unsigned int ustar_header_chksum(const struct ustar_header *header)
129 {
130         char *p = (char *)header;
131         unsigned int chksum = 0;
132         while (p < header->chksum)
133                 chksum += *p++;
134         chksum += sizeof(header->chksum) * ' ';
135         p += sizeof(header->chksum);
136         while (p < (char *)header + sizeof(struct ustar_header))
137                 chksum += *p++;
138         return chksum;
139 }
140
141 static int get_path_prefix(const struct strbuf *path, int maxlen)
142 {
143         int i = path->len;
144         if (i > maxlen)
145                 i = maxlen;
146         do {
147                 i--;
148         } while (i > 0 && path->buf[i] != '/');
149         return i;
150 }
151
152 static void write_entry(const unsigned char *sha1, struct strbuf *path,
153                         unsigned int mode, void *buffer, unsigned long size)
154 {
155         struct ustar_header header;
156         struct strbuf ext_header;
157
158         memset(&header, 0, sizeof(header));
159         ext_header.buf = NULL;
160         ext_header.len = ext_header.alloc = 0;
161
162         if (!sha1) {
163                 *header.typeflag = TYPEFLAG_GLOBAL_HEADER;
164                 mode = 0100666;
165                 strcpy(header.name, "pax_global_header");
166         } else if (!path) {
167                 *header.typeflag = TYPEFLAG_EXT_HEADER;
168                 mode = 0100666;
169                 sprintf(header.name, "%s.paxheader", sha1_to_hex(sha1));
170         } else {
171                 if (S_ISDIR(mode)) {
172                         *header.typeflag = TYPEFLAG_DIR;
173                         mode = (mode | 0777) & ~tar_umask;
174                 } else if (S_ISLNK(mode)) {
175                         *header.typeflag = TYPEFLAG_LNK;
176                         mode |= 0777;
177                 } else if (S_ISREG(mode)) {
178                         *header.typeflag = TYPEFLAG_REG;
179                         mode = (mode | ((mode & 0100) ? 0777 : 0666)) & ~tar_umask;
180                 } else {
181                         error("unsupported file mode: 0%o (SHA1: %s)",
182                               mode, sha1_to_hex(sha1));
183                         return;
184                 }
185                 if (path->len > sizeof(header.name)) {
186                         int plen = get_path_prefix(path, sizeof(header.prefix));
187                         int rest = path->len - plen - 1;
188                         if (plen > 0 && rest <= sizeof(header.name)) {
189                                 memcpy(header.prefix, path->buf, plen);
190                                 memcpy(header.name, path->buf + plen + 1, rest);
191                         } else {
192                                 sprintf(header.name, "%s.data",
193                                         sha1_to_hex(sha1));
194                                 strbuf_append_ext_header(&ext_header, "path",
195                                                          path->buf, path->len);
196                         }
197                 } else
198                         memcpy(header.name, path->buf, path->len);
199         }
200
201         if (S_ISLNK(mode) && buffer) {
202                 if (size > sizeof(header.linkname)) {
203                         sprintf(header.linkname, "see %s.paxheader",
204                                 sha1_to_hex(sha1));
205                         strbuf_append_ext_header(&ext_header, "linkpath",
206                                                  buffer, size);
207                 } else
208                         memcpy(header.linkname, buffer, size);
209         }
210
211         sprintf(header.mode, "%07o", mode & 07777);
212         sprintf(header.size, "%011lo", S_ISREG(mode) ? size : 0);
213         sprintf(header.mtime, "%011lo", archive_time);
214
215         /* XXX: should we provide more meaningful info here? */
216         sprintf(header.uid, "%07o", 0);
217         sprintf(header.gid, "%07o", 0);
218         strlcpy(header.uname, "git", sizeof(header.uname));
219         strlcpy(header.gname, "git", sizeof(header.gname));
220         sprintf(header.devmajor, "%07o", 0);
221         sprintf(header.devminor, "%07o", 0);
222
223         memcpy(header.magic, "ustar", 6);
224         memcpy(header.version, "00", 2);
225
226         sprintf(header.chksum, "%07o", ustar_header_chksum(&header));
227
228         if (ext_header.len > 0) {
229                 write_entry(sha1, NULL, 0, ext_header.buf, ext_header.len);
230                 free(ext_header.buf);
231         }
232         write_blocked(&header, sizeof(header));
233         if (S_ISREG(mode) && buffer && size > 0)
234                 write_blocked(buffer, size);
235 }
236
237 static void write_global_extended_header(const unsigned char *sha1)
238 {
239         struct strbuf ext_header;
240         ext_header.buf = NULL;
241         ext_header.len = ext_header.alloc = 0;
242         strbuf_append_ext_header(&ext_header, "comment", sha1_to_hex(sha1), 40);
243         write_entry(NULL, NULL, 0, ext_header.buf, ext_header.len);
244         free(ext_header.buf);
245 }
246
247 static void traverse_tree(struct tree_desc *tree, struct strbuf *path)
248 {
249         int pathlen = path->len;
250         struct name_entry entry;
251
252         while (tree_entry(tree, &entry)) {
253                 void *eltbuf;
254                 char elttype[20];
255                 unsigned long eltsize;
256
257                 eltbuf = read_sha1_file(entry.sha1, elttype, &eltsize);
258                 if (!eltbuf)
259                         die("cannot read %s", sha1_to_hex(entry.sha1));
260
261                 path->len = pathlen;
262                 strbuf_append_string(path, entry.path);
263                 if (S_ISDIR(entry.mode))
264                         strbuf_append_string(path, "/");
265
266                 write_entry(entry.sha1, path, entry.mode, eltbuf, eltsize);
267
268                 if (S_ISDIR(entry.mode)) {
269                         struct tree_desc subtree;
270                         subtree.buf = eltbuf;
271                         subtree.size = eltsize;
272                         traverse_tree(&subtree, path);
273                 }
274                 free(eltbuf);
275         }
276 }
277
278 static int git_tar_config(const char *var, const char *value)
279 {
280         if (!strcmp(var, "tar.umask")) {
281                 if (!strcmp(value, "user")) {
282                         tar_umask = umask(0);
283                         umask(tar_umask);
284                 } else {
285                         tar_umask = git_config_int(var, value);
286                 }
287                 return 0;
288         }
289         return git_default_config(var, value);
290 }
291
292 static int generate_tar(int argc, const char **argv, const char *prefix)
293 {
294         unsigned char sha1[20], tree_sha1[20];
295         struct commit *commit;
296         struct tree_desc tree;
297         struct strbuf current_path;
298         void *buffer;
299
300         current_path.buf = xmalloc(PATH_MAX);
301         current_path.alloc = PATH_MAX;
302         current_path.len = current_path.eof = 0;
303
304         git_config(git_tar_config);
305
306         switch (argc) {
307         case 3:
308                 strbuf_append_string(&current_path, argv[2]);
309                 strbuf_append_string(&current_path, "/");
310                 /* FALLTHROUGH */
311         case 2:
312                 if (get_sha1(argv[1], sha1))
313                         die("Not a valid object name %s", argv[1]);
314                 break;
315         default:
316                 usage(tar_tree_usage);
317         }
318
319         commit = lookup_commit_reference_gently(sha1, 1);
320         if (commit) {
321                 write_global_extended_header(commit->object.sha1);
322                 archive_time = commit->date;
323         } else
324                 archive_time = time(NULL);
325
326         tree.buf = buffer = read_object_with_reference(sha1, tree_type,
327                                                        &tree.size, tree_sha1);
328         if (!tree.buf)
329                 die("not a reference to a tag, commit or tree object: %s",
330                     sha1_to_hex(sha1));
331
332         if (current_path.len > 0)
333                 write_entry(tree_sha1, &current_path, 040777, NULL, 0);
334         traverse_tree(&tree, &current_path);
335         write_trailer();
336         free(buffer);
337         free(current_path.buf);
338         return 0;
339 }
340
341 static const char *exec = "git-upload-tar";
342
343 static int remote_tar(int argc, const char **argv)
344 {
345         int fd[2], ret, len;
346         pid_t pid;
347         char buf[1024];
348         char *url;
349
350         if (argc < 3 || 4 < argc)
351                 usage(tar_tree_usage);
352
353         /* --remote=<repo> */
354         url = xstrdup(argv[1]+9);
355         pid = git_connect(fd, url, exec);
356         if (pid < 0)
357                 return 1;
358
359         packet_write(fd[1], "want %s\n", argv[2]);
360         if (argv[3])
361                 packet_write(fd[1], "base %s\n", argv[3]);
362         packet_flush(fd[1]);
363
364         len = packet_read_line(fd[0], buf, sizeof(buf));
365         if (!len)
366                 die("git-tar-tree: expected ACK/NAK, got EOF");
367         if (buf[len-1] == '\n')
368                 buf[--len] = 0;
369         if (strcmp(buf, "ACK")) {
370                 if (5 < len && !strncmp(buf, "NACK ", 5))
371                         die("git-tar-tree: NACK %s", buf + 5);
372                 die("git-tar-tree: protocol error");
373         }
374         /* expect a flush */
375         len = packet_read_line(fd[0], buf, sizeof(buf));
376         if (len)
377                 die("git-tar-tree: expected a flush");
378
379         /* Now, start reading from fd[0] and spit it out to stdout */
380         ret = copy_fd(fd[0], 1);
381         close(fd[0]);
382
383         ret |= finish_connect(pid);
384         return !!ret;
385 }
386
387 int cmd_tar_tree(int argc, const char **argv, const char *prefix)
388 {
389         if (argc < 2)
390                 usage(tar_tree_usage);
391         if (!strncmp("--remote=", argv[1], 9))
392                 return remote_tar(argc, argv);
393         return generate_tar(argc, argv, prefix);
394 }
395
396 /* ustar header + extended global header content */
397 #define HEADERSIZE (2 * RECORDSIZE)
398
399 int cmd_get_tar_commit_id(int argc, const char **argv, const char *prefix)
400 {
401         char buffer[HEADERSIZE];
402         struct ustar_header *header = (struct ustar_header *)buffer;
403         char *content = buffer + RECORDSIZE;
404         ssize_t n;
405
406         n = xread(0, buffer, HEADERSIZE);
407         if (n < HEADERSIZE)
408                 die("git-get-tar-commit-id: read error");
409         if (header->typeflag[0] != 'g')
410                 return 1;
411         if (memcmp(content, "52 comment=", 11))
412                 return 1;
413
414         n = xwrite(1, content + 11, 41);
415         if (n < 41)
416                 die("git-get-tar-commit-id: write error");
417
418         return 0;
419 }