Merge branch 'ab/merge-file-prefix' into maint
[git] / archive-tar.c
1 /*
2  * Copyright (c) 2005, 2006 Rene Scharfe
3  */
4 #include "cache.h"
5 #include "tar.h"
6 #include "archive.h"
7 #include "streaming.h"
8 #include "run-command.h"
9
10 #define RECORDSIZE      (512)
11 #define BLOCKSIZE       (RECORDSIZE * 20)
12
13 static char block[BLOCKSIZE];
14 static unsigned long offset;
15
16 static int tar_umask = 002;
17
18 static int write_tar_filter_archive(const struct archiver *ar,
19                                     struct archiver_args *args);
20
21 /* writes out the whole block, but only if it is full */
22 static void write_if_needed(void)
23 {
24         if (offset == BLOCKSIZE) {
25                 write_or_die(1, block, BLOCKSIZE);
26                 offset = 0;
27         }
28 }
29
30 /*
31  * queues up writes, so that all our write(2) calls write exactly one
32  * full block; pads writes to RECORDSIZE
33  */
34 static void do_write_blocked(const void *data, unsigned long size)
35 {
36         const char *buf = data;
37
38         if (offset) {
39                 unsigned long chunk = BLOCKSIZE - offset;
40                 if (size < chunk)
41                         chunk = size;
42                 memcpy(block + offset, buf, chunk);
43                 size -= chunk;
44                 offset += chunk;
45                 buf += chunk;
46                 write_if_needed();
47         }
48         while (size >= BLOCKSIZE) {
49                 write_or_die(1, buf, BLOCKSIZE);
50                 size -= BLOCKSIZE;
51                 buf += BLOCKSIZE;
52         }
53         if (size) {
54                 memcpy(block + offset, buf, size);
55                 offset += size;
56         }
57 }
58
59 static void finish_record(void)
60 {
61         unsigned long tail;
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 static void write_blocked(const void *data, unsigned long size)
71 {
72         do_write_blocked(data, size);
73         finish_record();
74 }
75
76 /*
77  * The end of tar archives is marked by 2*512 nul bytes and after that
78  * follows the rest of the block (if any).
79  */
80 static void write_trailer(void)
81 {
82         int tail = BLOCKSIZE - offset;
83         memset(block + offset, 0, tail);
84         write_or_die(1, block, BLOCKSIZE);
85         if (tail < 2 * RECORDSIZE) {
86                 memset(block, 0, offset);
87                 write_or_die(1, block, BLOCKSIZE);
88         }
89 }
90
91 /*
92  * queues up writes, so that all our write(2) calls write exactly one
93  * full block; pads writes to RECORDSIZE
94  */
95 static int stream_blocked(const unsigned char *sha1)
96 {
97         struct git_istream *st;
98         enum object_type type;
99         unsigned long sz;
100         char buf[BLOCKSIZE];
101         ssize_t readlen;
102
103         st = open_istream(sha1, &type, &sz, NULL);
104         if (!st)
105                 return error("cannot stream blob %s", sha1_to_hex(sha1));
106         for (;;) {
107                 readlen = read_istream(st, buf, sizeof(buf));
108                 if (readlen <= 0)
109                         break;
110                 do_write_blocked(buf, readlen);
111         }
112         close_istream(st);
113         if (!readlen)
114                 finish_record();
115         return readlen;
116 }
117
118 /*
119  * pax extended header records have the format "%u %s=%s\n".  %u contains
120  * the size of the whole string (including the %u), the first %s is the
121  * keyword, the second one is the value.  This function constructs such a
122  * string and appends it to a struct strbuf.
123  */
124 static void strbuf_append_ext_header(struct strbuf *sb, const char *keyword,
125                                      const char *value, unsigned int valuelen)
126 {
127         int len, tmp;
128
129         /* "%u %s=%s\n" */
130         len = 1 + 1 + strlen(keyword) + 1 + valuelen + 1;
131         for (tmp = len; tmp > 9; tmp /= 10)
132                 len++;
133
134         strbuf_grow(sb, len);
135         strbuf_addf(sb, "%u %s=", len, keyword);
136         strbuf_add(sb, value, valuelen);
137         strbuf_addch(sb, '\n');
138 }
139
140 static unsigned int ustar_header_chksum(const struct ustar_header *header)
141 {
142         const unsigned char *p = (const unsigned char *)header;
143         unsigned int chksum = 0;
144         while (p < (const unsigned char *)header->chksum)
145                 chksum += *p++;
146         chksum += sizeof(header->chksum) * ' ';
147         p += sizeof(header->chksum);
148         while (p < (const unsigned char *)header + sizeof(struct ustar_header))
149                 chksum += *p++;
150         return chksum;
151 }
152
153 static size_t get_path_prefix(const char *path, size_t pathlen, size_t maxlen)
154 {
155         size_t i = pathlen;
156         if (i > 1 && path[i - 1] == '/')
157                 i--;
158         if (i > maxlen)
159                 i = maxlen;
160         do {
161                 i--;
162         } while (i > 0 && path[i] != '/');
163         return i;
164 }
165
166 static void prepare_header(struct archiver_args *args,
167                            struct ustar_header *header,
168                            unsigned int mode, unsigned long size)
169 {
170         sprintf(header->mode, "%07o", mode & 07777);
171         sprintf(header->size, "%011lo", S_ISREG(mode) ? size : 0);
172         sprintf(header->mtime, "%011lo", (unsigned long) args->time);
173
174         sprintf(header->uid, "%07o", 0);
175         sprintf(header->gid, "%07o", 0);
176         strlcpy(header->uname, "root", sizeof(header->uname));
177         strlcpy(header->gname, "root", sizeof(header->gname));
178         sprintf(header->devmajor, "%07o", 0);
179         sprintf(header->devminor, "%07o", 0);
180
181         memcpy(header->magic, "ustar", 6);
182         memcpy(header->version, "00", 2);
183
184         sprintf(header->chksum, "%07o", ustar_header_chksum(header));
185 }
186
187 static int write_extended_header(struct archiver_args *args,
188                                  const unsigned char *sha1,
189                                  const void *buffer, unsigned long size)
190 {
191         struct ustar_header header;
192         unsigned int mode;
193         memset(&header, 0, sizeof(header));
194         *header.typeflag = TYPEFLAG_EXT_HEADER;
195         mode = 0100666;
196         sprintf(header.name, "%s.paxheader", sha1_to_hex(sha1));
197         prepare_header(args, &header, mode, size);
198         write_blocked(&header, sizeof(header));
199         write_blocked(buffer, size);
200         return 0;
201 }
202
203 static int write_tar_entry(struct archiver_args *args,
204                            const unsigned char *sha1,
205                            const char *path, size_t pathlen,
206                            unsigned int mode)
207 {
208         struct ustar_header header;
209         struct strbuf ext_header = STRBUF_INIT;
210         unsigned int old_mode = mode;
211         unsigned long size;
212         void *buffer;
213         int err = 0;
214
215         memset(&header, 0, sizeof(header));
216
217         if (S_ISDIR(mode) || S_ISGITLINK(mode)) {
218                 *header.typeflag = TYPEFLAG_DIR;
219                 mode = (mode | 0777) & ~tar_umask;
220         } else if (S_ISLNK(mode)) {
221                 *header.typeflag = TYPEFLAG_LNK;
222                 mode |= 0777;
223         } else if (S_ISREG(mode)) {
224                 *header.typeflag = TYPEFLAG_REG;
225                 mode = (mode | ((mode & 0100) ? 0777 : 0666)) & ~tar_umask;
226         } else {
227                 return error("unsupported file mode: 0%o (SHA1: %s)",
228                              mode, sha1_to_hex(sha1));
229         }
230         if (pathlen > sizeof(header.name)) {
231                 size_t plen = get_path_prefix(path, pathlen,
232                                               sizeof(header.prefix));
233                 size_t rest = pathlen - plen - 1;
234                 if (plen > 0 && rest <= sizeof(header.name)) {
235                         memcpy(header.prefix, path, plen);
236                                 memcpy(header.name, path + plen + 1, rest);
237                 } else {
238                         sprintf(header.name, "%s.data",
239                                 sha1_to_hex(sha1));
240                         strbuf_append_ext_header(&ext_header, "path",
241                                                  path, pathlen);
242                 }
243         } else
244                 memcpy(header.name, path, pathlen);
245
246         if (S_ISREG(mode) && !args->convert &&
247             sha1_object_info(sha1, &size) == OBJ_BLOB &&
248             size > big_file_threshold)
249                 buffer = NULL;
250         else if (S_ISLNK(mode) || S_ISREG(mode)) {
251                 enum object_type type;
252                 buffer = sha1_file_to_archive(args, path, sha1, old_mode, &type, &size);
253                 if (!buffer)
254                         return error("cannot read %s", sha1_to_hex(sha1));
255         } else {
256                 buffer = NULL;
257                 size = 0;
258         }
259
260         if (S_ISLNK(mode)) {
261                 if (size > sizeof(header.linkname)) {
262                         sprintf(header.linkname, "see %s.paxheader",
263                                 sha1_to_hex(sha1));
264                         strbuf_append_ext_header(&ext_header, "linkpath",
265                                                  buffer, size);
266                 } else
267                         memcpy(header.linkname, buffer, size);
268         }
269
270         prepare_header(args, &header, mode, size);
271
272         if (ext_header.len > 0) {
273                 err = write_extended_header(args, sha1, ext_header.buf,
274                                             ext_header.len);
275                 if (err) {
276                         free(buffer);
277                         return err;
278                 }
279         }
280         strbuf_release(&ext_header);
281         write_blocked(&header, sizeof(header));
282         if (S_ISREG(mode) && size > 0) {
283                 if (buffer)
284                         write_blocked(buffer, size);
285                 else
286                         err = stream_blocked(sha1);
287         }
288         free(buffer);
289         return err;
290 }
291
292 static int write_global_extended_header(struct archiver_args *args)
293 {
294         const unsigned char *sha1 = args->commit_sha1;
295         struct strbuf ext_header = STRBUF_INIT;
296         struct ustar_header header;
297         unsigned int mode;
298         int err = 0;
299
300         strbuf_append_ext_header(&ext_header, "comment", sha1_to_hex(sha1), 40);
301         memset(&header, 0, sizeof(header));
302         *header.typeflag = TYPEFLAG_GLOBAL_HEADER;
303         mode = 0100666;
304         strcpy(header.name, "pax_global_header");
305         prepare_header(args, &header, mode, ext_header.len);
306         write_blocked(&header, sizeof(header));
307         write_blocked(ext_header.buf, ext_header.len);
308         strbuf_release(&ext_header);
309         return err;
310 }
311
312 static struct archiver **tar_filters;
313 static int nr_tar_filters;
314 static int alloc_tar_filters;
315
316 static struct archiver *find_tar_filter(const char *name, int len)
317 {
318         int i;
319         for (i = 0; i < nr_tar_filters; i++) {
320                 struct archiver *ar = tar_filters[i];
321                 if (!strncmp(ar->name, name, len) && !ar->name[len])
322                         return ar;
323         }
324         return NULL;
325 }
326
327 static int tar_filter_config(const char *var, const char *value, void *data)
328 {
329         struct archiver *ar;
330         const char *name;
331         const char *type;
332         int namelen;
333
334         if (parse_config_key(var, "tar", &name, &namelen, &type) < 0 || !name)
335                 return 0;
336
337         ar = find_tar_filter(name, namelen);
338         if (!ar) {
339                 ar = xcalloc(1, sizeof(*ar));
340                 ar->name = xmemdupz(name, namelen);
341                 ar->write_archive = write_tar_filter_archive;
342                 ar->flags = ARCHIVER_WANT_COMPRESSION_LEVELS;
343                 ALLOC_GROW(tar_filters, nr_tar_filters + 1, alloc_tar_filters);
344                 tar_filters[nr_tar_filters++] = ar;
345         }
346
347         if (!strcmp(type, "command")) {
348                 if (!value)
349                         return config_error_nonbool(var);
350                 free(ar->data);
351                 ar->data = xstrdup(value);
352                 return 0;
353         }
354         if (!strcmp(type, "remote")) {
355                 if (git_config_bool(var, value))
356                         ar->flags |= ARCHIVER_REMOTE;
357                 else
358                         ar->flags &= ~ARCHIVER_REMOTE;
359                 return 0;
360         }
361
362         return 0;
363 }
364
365 static int git_tar_config(const char *var, const char *value, void *cb)
366 {
367         if (!strcmp(var, "tar.umask")) {
368                 if (value && !strcmp(value, "user")) {
369                         tar_umask = umask(0);
370                         umask(tar_umask);
371                 } else {
372                         tar_umask = git_config_int(var, value);
373                 }
374                 return 0;
375         }
376
377         return tar_filter_config(var, value, cb);
378 }
379
380 static int write_tar_archive(const struct archiver *ar,
381                              struct archiver_args *args)
382 {
383         int err = 0;
384
385         if (args->commit_sha1)
386                 err = write_global_extended_header(args);
387         if (!err)
388                 err = write_archive_entries(args, write_tar_entry);
389         if (!err)
390                 write_trailer();
391         return err;
392 }
393
394 static int write_tar_filter_archive(const struct archiver *ar,
395                                     struct archiver_args *args)
396 {
397         struct strbuf cmd = STRBUF_INIT;
398         struct child_process filter = CHILD_PROCESS_INIT;
399         const char *argv[2];
400         int r;
401
402         if (!ar->data)
403                 die("BUG: tar-filter archiver called with no filter defined");
404
405         strbuf_addstr(&cmd, ar->data);
406         if (args->compression_level >= 0)
407                 strbuf_addf(&cmd, " -%d", args->compression_level);
408
409         argv[0] = cmd.buf;
410         argv[1] = NULL;
411         filter.argv = argv;
412         filter.use_shell = 1;
413         filter.in = -1;
414
415         if (start_command(&filter) < 0)
416                 die_errno("unable to start '%s' filter", argv[0]);
417         close(1);
418         if (dup2(filter.in, 1) < 0)
419                 die_errno("unable to redirect descriptor");
420         close(filter.in);
421
422         r = write_tar_archive(ar, args);
423
424         close(1);
425         if (finish_command(&filter) != 0)
426                 die("'%s' filter reported error", argv[0]);
427
428         strbuf_release(&cmd);
429         return r;
430 }
431
432 static struct archiver tar_archiver = {
433         "tar",
434         write_tar_archive,
435         ARCHIVER_REMOTE
436 };
437
438 void init_tar_archiver(void)
439 {
440         int i;
441         register_archiver(&tar_archiver);
442
443         tar_filter_config("tar.tgz.command", "gzip -cn", NULL);
444         tar_filter_config("tar.tgz.remote", "true", NULL);
445         tar_filter_config("tar.tar.gz.command", "gzip -cn", NULL);
446         tar_filter_config("tar.tar.gz.remote", "true", NULL);
447         git_config(git_tar_config, NULL);
448         for (i = 0; i < nr_tar_filters; i++) {
449                 /* omit any filters that never had a command configured */
450                 if (tar_filters[i]->data)
451                         register_archiver(tar_filters[i]);
452         }
453 }