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