archive: delegate blob reading to backend
[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 "run-command.h"
8
9 #define RECORDSIZE      (512)
10 #define BLOCKSIZE       (RECORDSIZE * 20)
11
12 static char block[BLOCKSIZE];
13 static unsigned long offset;
14
15 static int tar_umask = 002;
16
17 static int write_tar_filter_archive(const struct archiver *ar,
18                                     struct archiver_args *args);
19
20 /* writes out the whole block, but only if it is full */
21 static void write_if_needed(void)
22 {
23         if (offset == BLOCKSIZE) {
24                 write_or_die(1, block, BLOCKSIZE);
25                 offset = 0;
26         }
27 }
28
29 /*
30  * queues up writes, so that all our write(2) calls write exactly one
31  * full block; pads writes to RECORDSIZE
32  */
33 static void write_blocked(const void *data, unsigned long size)
34 {
35         const char *buf = data;
36         unsigned long tail;
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         tail = offset % RECORDSIZE;
58         if (tail)  {
59                 memset(block + offset, 0, RECORDSIZE - tail);
60                 offset += RECORDSIZE - tail;
61         }
62         write_if_needed();
63 }
64
65 /*
66  * The end of tar archives is marked by 2*512 nul bytes and after that
67  * follows the rest of the block (if any).
68  */
69 static void write_trailer(void)
70 {
71         int tail = BLOCKSIZE - offset;
72         memset(block + offset, 0, tail);
73         write_or_die(1, block, BLOCKSIZE);
74         if (tail < 2 * RECORDSIZE) {
75                 memset(block, 0, offset);
76                 write_or_die(1, block, BLOCKSIZE);
77         }
78 }
79
80 /*
81  * pax extended header records have the format "%u %s=%s\n".  %u contains
82  * the size of the whole string (including the %u), the first %s is the
83  * keyword, the second one is the value.  This function constructs such a
84  * string and appends it to a struct strbuf.
85  */
86 static void strbuf_append_ext_header(struct strbuf *sb, const char *keyword,
87                                      const char *value, unsigned int valuelen)
88 {
89         int len, tmp;
90
91         /* "%u %s=%s\n" */
92         len = 1 + 1 + strlen(keyword) + 1 + valuelen + 1;
93         for (tmp = len; tmp > 9; tmp /= 10)
94                 len++;
95
96         strbuf_grow(sb, len);
97         strbuf_addf(sb, "%u %s=", len, keyword);
98         strbuf_add(sb, value, valuelen);
99         strbuf_addch(sb, '\n');
100 }
101
102 static unsigned int ustar_header_chksum(const struct ustar_header *header)
103 {
104         char *p = (char *)header;
105         unsigned int chksum = 0;
106         while (p < header->chksum)
107                 chksum += *p++;
108         chksum += sizeof(header->chksum) * ' ';
109         p += sizeof(header->chksum);
110         while (p < (char *)header + sizeof(struct ustar_header))
111                 chksum += *p++;
112         return chksum;
113 }
114
115 static size_t get_path_prefix(const char *path, size_t pathlen, size_t maxlen)
116 {
117         size_t i = pathlen;
118         if (i > maxlen)
119                 i = maxlen;
120         do {
121                 i--;
122         } while (i > 0 && path[i] != '/');
123         return i;
124 }
125
126 static void prepare_header(struct archiver_args *args,
127                            struct ustar_header *header,
128                            unsigned int mode, unsigned long size)
129 {
130         sprintf(header->mode, "%07o", mode & 07777);
131         sprintf(header->size, "%011lo", S_ISREG(mode) ? size : 0);
132         sprintf(header->mtime, "%011lo", (unsigned long) args->time);
133
134         sprintf(header->uid, "%07o", 0);
135         sprintf(header->gid, "%07o", 0);
136         strlcpy(header->uname, "root", sizeof(header->uname));
137         strlcpy(header->gname, "root", sizeof(header->gname));
138         sprintf(header->devmajor, "%07o", 0);
139         sprintf(header->devminor, "%07o", 0);
140
141         memcpy(header->magic, "ustar", 6);
142         memcpy(header->version, "00", 2);
143
144         sprintf(header->chksum, "%07o", ustar_header_chksum(header));
145 }
146
147 static int write_extended_header(struct archiver_args *args,
148                                  const unsigned char *sha1,
149                                  const void *buffer, unsigned long size)
150 {
151         struct ustar_header header;
152         unsigned int mode;
153         memset(&header, 0, sizeof(header));
154         *header.typeflag = TYPEFLAG_EXT_HEADER;
155         mode = 0100666;
156         sprintf(header.name, "%s.paxheader", sha1_to_hex(sha1));
157         prepare_header(args, &header, mode, size);
158         write_blocked(&header, sizeof(header));
159         write_blocked(buffer, size);
160         return 0;
161 }
162
163 static int write_tar_entry(struct archiver_args *args,
164                            const unsigned char *sha1,
165                            const char *path, size_t pathlen,
166                            unsigned int mode)
167 {
168         struct ustar_header header;
169         struct strbuf ext_header = STRBUF_INIT;
170         unsigned int old_mode = mode;
171         unsigned long size;
172         void *buffer;
173         int err = 0;
174
175         memset(&header, 0, sizeof(header));
176
177         if (S_ISDIR(mode) || S_ISGITLINK(mode)) {
178                 *header.typeflag = TYPEFLAG_DIR;
179                 mode = (mode | 0777) & ~tar_umask;
180         } else if (S_ISLNK(mode)) {
181                 *header.typeflag = TYPEFLAG_LNK;
182                 mode |= 0777;
183         } else if (S_ISREG(mode)) {
184                 *header.typeflag = TYPEFLAG_REG;
185                 mode = (mode | ((mode & 0100) ? 0777 : 0666)) & ~tar_umask;
186         } else {
187                 return error("unsupported file mode: 0%o (SHA1: %s)",
188                              mode, sha1_to_hex(sha1));
189         }
190         if (pathlen > sizeof(header.name)) {
191                 size_t plen = get_path_prefix(path, pathlen,
192                                               sizeof(header.prefix));
193                 size_t rest = pathlen - plen - 1;
194                 if (plen > 0 && rest <= sizeof(header.name)) {
195                         memcpy(header.prefix, path, plen);
196                                 memcpy(header.name, path + plen + 1, rest);
197                 } else {
198                         sprintf(header.name, "%s.data",
199                                 sha1_to_hex(sha1));
200                         strbuf_append_ext_header(&ext_header, "path",
201                                                  path, pathlen);
202                 }
203         } else
204                 memcpy(header.name, path, pathlen);
205
206         if (S_ISLNK(mode) || S_ISREG(mode)) {
207                 enum object_type type;
208                 buffer = sha1_file_to_archive(args, path, sha1, old_mode, &type, &size);
209                 if (!buffer)
210                         return error("cannot read %s", sha1_to_hex(sha1));
211         } else {
212                 buffer = NULL;
213                 size = 0;
214         }
215
216         if (S_ISLNK(mode)) {
217                 if (size > sizeof(header.linkname)) {
218                         sprintf(header.linkname, "see %s.paxheader",
219                                 sha1_to_hex(sha1));
220                         strbuf_append_ext_header(&ext_header, "linkpath",
221                                                  buffer, size);
222                 } else
223                         memcpy(header.linkname, buffer, size);
224         }
225
226         prepare_header(args, &header, mode, size);
227
228         if (ext_header.len > 0) {
229                 err = write_extended_header(args, sha1, ext_header.buf,
230                                             ext_header.len);
231                 if (err) {
232                         free(buffer);
233                         return err;
234                 }
235         }
236         strbuf_release(&ext_header);
237         write_blocked(&header, sizeof(header));
238         if (S_ISREG(mode) && buffer && size > 0)
239                 write_blocked(buffer, size);
240         free(buffer);
241         return err;
242 }
243
244 static int write_global_extended_header(struct archiver_args *args)
245 {
246         const unsigned char *sha1 = args->commit_sha1;
247         struct strbuf ext_header = STRBUF_INIT;
248         struct ustar_header header;
249         unsigned int mode;
250         int err = 0;
251
252         strbuf_append_ext_header(&ext_header, "comment", sha1_to_hex(sha1), 40);
253         memset(&header, 0, sizeof(header));
254         *header.typeflag = TYPEFLAG_GLOBAL_HEADER;
255         mode = 0100666;
256         strcpy(header.name, "pax_global_header");
257         prepare_header(args, &header, mode, ext_header.len);
258         write_blocked(&header, sizeof(header));
259         write_blocked(ext_header.buf, ext_header.len);
260         strbuf_release(&ext_header);
261         return err;
262 }
263
264 static struct archiver **tar_filters;
265 static int nr_tar_filters;
266 static int alloc_tar_filters;
267
268 static struct archiver *find_tar_filter(const char *name, int len)
269 {
270         int i;
271         for (i = 0; i < nr_tar_filters; i++) {
272                 struct archiver *ar = tar_filters[i];
273                 if (!strncmp(ar->name, name, len) && !ar->name[len])
274                         return ar;
275         }
276         return NULL;
277 }
278
279 static int tar_filter_config(const char *var, const char *value, void *data)
280 {
281         struct archiver *ar;
282         const char *dot;
283         const char *name;
284         const char *type;
285         int namelen;
286
287         if (prefixcmp(var, "tar."))
288                 return 0;
289         dot = strrchr(var, '.');
290         if (dot == var + 9)
291                 return 0;
292
293         name = var + 4;
294         namelen = dot - name;
295         type = dot + 1;
296
297         ar = find_tar_filter(name, namelen);
298         if (!ar) {
299                 ar = xcalloc(1, sizeof(*ar));
300                 ar->name = xmemdupz(name, namelen);
301                 ar->write_archive = write_tar_filter_archive;
302                 ar->flags = ARCHIVER_WANT_COMPRESSION_LEVELS;
303                 ALLOC_GROW(tar_filters, nr_tar_filters + 1, alloc_tar_filters);
304                 tar_filters[nr_tar_filters++] = ar;
305         }
306
307         if (!strcmp(type, "command")) {
308                 if (!value)
309                         return config_error_nonbool(var);
310                 free(ar->data);
311                 ar->data = xstrdup(value);
312                 return 0;
313         }
314         if (!strcmp(type, "remote")) {
315                 if (git_config_bool(var, value))
316                         ar->flags |= ARCHIVER_REMOTE;
317                 else
318                         ar->flags &= ~ARCHIVER_REMOTE;
319                 return 0;
320         }
321
322         return 0;
323 }
324
325 static int git_tar_config(const char *var, const char *value, void *cb)
326 {
327         if (!strcmp(var, "tar.umask")) {
328                 if (value && !strcmp(value, "user")) {
329                         tar_umask = umask(0);
330                         umask(tar_umask);
331                 } else {
332                         tar_umask = git_config_int(var, value);
333                 }
334                 return 0;
335         }
336
337         return tar_filter_config(var, value, cb);
338 }
339
340 static int write_tar_archive(const struct archiver *ar,
341                              struct archiver_args *args)
342 {
343         int err = 0;
344
345         if (args->commit_sha1)
346                 err = write_global_extended_header(args);
347         if (!err)
348                 err = write_archive_entries(args, write_tar_entry);
349         if (!err)
350                 write_trailer();
351         return err;
352 }
353
354 static int write_tar_filter_archive(const struct archiver *ar,
355                                     struct archiver_args *args)
356 {
357         struct strbuf cmd = STRBUF_INIT;
358         struct child_process filter;
359         const char *argv[2];
360         int r;
361
362         if (!ar->data)
363                 die("BUG: tar-filter archiver called with no filter defined");
364
365         strbuf_addstr(&cmd, ar->data);
366         if (args->compression_level >= 0)
367                 strbuf_addf(&cmd, " -%d", args->compression_level);
368
369         memset(&filter, 0, sizeof(filter));
370         argv[0] = cmd.buf;
371         argv[1] = NULL;
372         filter.argv = argv;
373         filter.use_shell = 1;
374         filter.in = -1;
375
376         if (start_command(&filter) < 0)
377                 die_errno("unable to start '%s' filter", argv[0]);
378         close(1);
379         if (dup2(filter.in, 1) < 0)
380                 die_errno("unable to redirect descriptor");
381         close(filter.in);
382
383         r = write_tar_archive(ar, args);
384
385         close(1);
386         if (finish_command(&filter) != 0)
387                 die("'%s' filter reported error", argv[0]);
388
389         strbuf_release(&cmd);
390         return r;
391 }
392
393 static struct archiver tar_archiver = {
394         "tar",
395         write_tar_archive,
396         ARCHIVER_REMOTE
397 };
398
399 void init_tar_archiver(void)
400 {
401         int i;
402         register_archiver(&tar_archiver);
403
404         tar_filter_config("tar.tgz.command", "gzip -cn", NULL);
405         tar_filter_config("tar.tgz.remote", "true", NULL);
406         tar_filter_config("tar.tar.gz.command", "gzip -cn", NULL);
407         tar_filter_config("tar.tar.gz.remote", "true", NULL);
408         git_config(git_tar_config, NULL);
409         for (i = 0; i < nr_tar_filters; i++) {
410                 /* omit any filters that never had a command configured */
411                 if (tar_filters[i]->data)
412                         register_archiver(tar_filters[i]);
413         }
414 }