Merge branch 'ah/filter-branch-setup'
[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 /*
22  * This is the max value that a ustar size header can specify, as it is fixed
23  * at 11 octal digits. POSIX specifies that we switch to extended headers at
24  * this size.
25  *
26  * Likewise for the mtime (which happens to use a buffer of the same size).
27  */
28 #if ULONG_MAX == 0xFFFFFFFF
29 #define USTAR_MAX_SIZE ULONG_MAX
30 #else
31 #define USTAR_MAX_SIZE 077777777777UL
32 #endif
33 #if TIME_MAX == 0xFFFFFFFF
34 #define USTAR_MAX_MTIME TIME_MAX
35 #else
36 #define USTAR_MAX_MTIME 077777777777ULL
37 #endif
38
39 /* writes out the whole block, but only if it is full */
40 static void write_if_needed(void)
41 {
42         if (offset == BLOCKSIZE) {
43                 write_or_die(1, block, BLOCKSIZE);
44                 offset = 0;
45         }
46 }
47
48 /*
49  * queues up writes, so that all our write(2) calls write exactly one
50  * full block; pads writes to RECORDSIZE
51  */
52 static void do_write_blocked(const void *data, unsigned long size)
53 {
54         const char *buf = data;
55
56         if (offset) {
57                 unsigned long chunk = BLOCKSIZE - offset;
58                 if (size < chunk)
59                         chunk = size;
60                 memcpy(block + offset, buf, chunk);
61                 size -= chunk;
62                 offset += chunk;
63                 buf += chunk;
64                 write_if_needed();
65         }
66         while (size >= BLOCKSIZE) {
67                 write_or_die(1, buf, BLOCKSIZE);
68                 size -= BLOCKSIZE;
69                 buf += BLOCKSIZE;
70         }
71         if (size) {
72                 memcpy(block + offset, buf, size);
73                 offset += size;
74         }
75 }
76
77 static void finish_record(void)
78 {
79         unsigned long tail;
80         tail = offset % RECORDSIZE;
81         if (tail)  {
82                 memset(block + offset, 0, RECORDSIZE - tail);
83                 offset += RECORDSIZE - tail;
84         }
85         write_if_needed();
86 }
87
88 static void write_blocked(const void *data, unsigned long size)
89 {
90         do_write_blocked(data, size);
91         finish_record();
92 }
93
94 /*
95  * The end of tar archives is marked by 2*512 nul bytes and after that
96  * follows the rest of the block (if any).
97  */
98 static void write_trailer(void)
99 {
100         int tail = BLOCKSIZE - offset;
101         memset(block + offset, 0, tail);
102         write_or_die(1, block, BLOCKSIZE);
103         if (tail < 2 * RECORDSIZE) {
104                 memset(block, 0, offset);
105                 write_or_die(1, block, BLOCKSIZE);
106         }
107 }
108
109 /*
110  * queues up writes, so that all our write(2) calls write exactly one
111  * full block; pads writes to RECORDSIZE
112  */
113 static int stream_blocked(const unsigned char *sha1)
114 {
115         struct git_istream *st;
116         enum object_type type;
117         unsigned long sz;
118         char buf[BLOCKSIZE];
119         ssize_t readlen;
120
121         st = open_istream(sha1, &type, &sz, NULL);
122         if (!st)
123                 return error("cannot stream blob %s", sha1_to_hex(sha1));
124         for (;;) {
125                 readlen = read_istream(st, buf, sizeof(buf));
126                 if (readlen <= 0)
127                         break;
128                 do_write_blocked(buf, readlen);
129         }
130         close_istream(st);
131         if (!readlen)
132                 finish_record();
133         return readlen;
134 }
135
136 /*
137  * pax extended header records have the format "%u %s=%s\n".  %u contains
138  * the size of the whole string (including the %u), the first %s is the
139  * keyword, the second one is the value.  This function constructs such a
140  * string and appends it to a struct strbuf.
141  */
142 static void strbuf_append_ext_header(struct strbuf *sb, const char *keyword,
143                                      const char *value, unsigned int valuelen)
144 {
145         int len, tmp;
146
147         /* "%u %s=%s\n" */
148         len = 1 + 1 + strlen(keyword) + 1 + valuelen + 1;
149         for (tmp = len; tmp > 9; tmp /= 10)
150                 len++;
151
152         strbuf_grow(sb, len);
153         strbuf_addf(sb, "%u %s=", len, keyword);
154         strbuf_add(sb, value, valuelen);
155         strbuf_addch(sb, '\n');
156 }
157
158 /*
159  * Like strbuf_append_ext_header, but for numeric values.
160  */
161 static void strbuf_append_ext_header_uint(struct strbuf *sb,
162                                           const char *keyword,
163                                           uintmax_t value)
164 {
165         char buf[40]; /* big enough for 2^128 in decimal, plus NUL */
166         int len;
167
168         len = xsnprintf(buf, sizeof(buf), "%"PRIuMAX, value);
169         strbuf_append_ext_header(sb, keyword, buf, len);
170 }
171
172 static unsigned int ustar_header_chksum(const struct ustar_header *header)
173 {
174         const unsigned char *p = (const unsigned char *)header;
175         unsigned int chksum = 0;
176         while (p < (const unsigned char *)header->chksum)
177                 chksum += *p++;
178         chksum += sizeof(header->chksum) * ' ';
179         p += sizeof(header->chksum);
180         while (p < (const unsigned char *)header + sizeof(struct ustar_header))
181                 chksum += *p++;
182         return chksum;
183 }
184
185 static size_t get_path_prefix(const char *path, size_t pathlen, size_t maxlen)
186 {
187         size_t i = pathlen;
188         if (i > 1 && path[i - 1] == '/')
189                 i--;
190         if (i > maxlen)
191                 i = maxlen;
192         do {
193                 i--;
194         } while (i > 0 && path[i] != '/');
195         return i;
196 }
197
198 static void prepare_header(struct archiver_args *args,
199                            struct ustar_header *header,
200                            unsigned int mode, unsigned long size)
201 {
202         xsnprintf(header->mode, sizeof(header->mode), "%07o", mode & 07777);
203         xsnprintf(header->size, sizeof(header->size), "%011lo", S_ISREG(mode) ? size : 0);
204         xsnprintf(header->mtime, sizeof(header->mtime), "%011lo", (unsigned long) args->time);
205
206         xsnprintf(header->uid, sizeof(header->uid), "%07o", 0);
207         xsnprintf(header->gid, sizeof(header->gid), "%07o", 0);
208         strlcpy(header->uname, "root", sizeof(header->uname));
209         strlcpy(header->gname, "root", sizeof(header->gname));
210         xsnprintf(header->devmajor, sizeof(header->devmajor), "%07o", 0);
211         xsnprintf(header->devminor, sizeof(header->devminor), "%07o", 0);
212
213         memcpy(header->magic, "ustar", 6);
214         memcpy(header->version, "00", 2);
215
216         xsnprintf(header->chksum, sizeof(header->chksum), "%07o", ustar_header_chksum(header));
217 }
218
219 static void write_extended_header(struct archiver_args *args,
220                                   const unsigned char *sha1,
221                                   const void *buffer, unsigned long size)
222 {
223         struct ustar_header header;
224         unsigned int mode;
225         memset(&header, 0, sizeof(header));
226         *header.typeflag = TYPEFLAG_EXT_HEADER;
227         mode = 0100666;
228         xsnprintf(header.name, sizeof(header.name), "%s.paxheader", sha1_to_hex(sha1));
229         prepare_header(args, &header, mode, size);
230         write_blocked(&header, sizeof(header));
231         write_blocked(buffer, size);
232 }
233
234 static int write_tar_entry(struct archiver_args *args,
235                            const unsigned char *sha1,
236                            const char *path, size_t pathlen,
237                            unsigned int mode)
238 {
239         struct ustar_header header;
240         struct strbuf ext_header = STRBUF_INIT;
241         unsigned int old_mode = mode;
242         unsigned long size, size_in_header;
243         void *buffer;
244         int err = 0;
245
246         memset(&header, 0, sizeof(header));
247
248         if (S_ISDIR(mode) || S_ISGITLINK(mode)) {
249                 *header.typeflag = TYPEFLAG_DIR;
250                 mode = (mode | 0777) & ~tar_umask;
251         } else if (S_ISLNK(mode)) {
252                 *header.typeflag = TYPEFLAG_LNK;
253                 mode |= 0777;
254         } else if (S_ISREG(mode)) {
255                 *header.typeflag = TYPEFLAG_REG;
256                 mode = (mode | ((mode & 0100) ? 0777 : 0666)) & ~tar_umask;
257         } else {
258                 return error("unsupported file mode: 0%o (SHA1: %s)",
259                              mode, sha1_to_hex(sha1));
260         }
261         if (pathlen > sizeof(header.name)) {
262                 size_t plen = get_path_prefix(path, pathlen,
263                                               sizeof(header.prefix));
264                 size_t rest = pathlen - plen - 1;
265                 if (plen > 0 && rest <= sizeof(header.name)) {
266                         memcpy(header.prefix, path, plen);
267                         memcpy(header.name, path + plen + 1, rest);
268                 } else {
269                         xsnprintf(header.name, sizeof(header.name), "%s.data",
270                                   sha1_to_hex(sha1));
271                         strbuf_append_ext_header(&ext_header, "path",
272                                                  path, pathlen);
273                 }
274         } else
275                 memcpy(header.name, path, pathlen);
276
277         if (S_ISREG(mode) && !args->convert &&
278             sha1_object_info(sha1, &size) == OBJ_BLOB &&
279             size > big_file_threshold)
280                 buffer = NULL;
281         else if (S_ISLNK(mode) || S_ISREG(mode)) {
282                 enum object_type type;
283                 buffer = sha1_file_to_archive(args, path, sha1, old_mode, &type, &size);
284                 if (!buffer)
285                         return error("cannot read %s", sha1_to_hex(sha1));
286         } else {
287                 buffer = NULL;
288                 size = 0;
289         }
290
291         if (S_ISLNK(mode)) {
292                 if (size > sizeof(header.linkname)) {
293                         xsnprintf(header.linkname, sizeof(header.linkname),
294                                   "see %s.paxheader", sha1_to_hex(sha1));
295                         strbuf_append_ext_header(&ext_header, "linkpath",
296                                                  buffer, size);
297                 } else
298                         memcpy(header.linkname, buffer, size);
299         }
300
301         size_in_header = size;
302         if (S_ISREG(mode) && size > USTAR_MAX_SIZE) {
303                 size_in_header = 0;
304                 strbuf_append_ext_header_uint(&ext_header, "size", size);
305         }
306
307         prepare_header(args, &header, mode, size_in_header);
308
309         if (ext_header.len > 0) {
310                 write_extended_header(args, sha1, ext_header.buf,
311                                       ext_header.len);
312         }
313         strbuf_release(&ext_header);
314         write_blocked(&header, sizeof(header));
315         if (S_ISREG(mode) && size > 0) {
316                 if (buffer)
317                         write_blocked(buffer, size);
318                 else
319                         err = stream_blocked(sha1);
320         }
321         free(buffer);
322         return err;
323 }
324
325 static void write_global_extended_header(struct archiver_args *args)
326 {
327         const unsigned char *sha1 = args->commit_sha1;
328         struct strbuf ext_header = STRBUF_INIT;
329         struct ustar_header header;
330         unsigned int mode;
331
332         if (sha1)
333                 strbuf_append_ext_header(&ext_header, "comment",
334                                          sha1_to_hex(sha1), 40);
335         if (args->time > USTAR_MAX_MTIME) {
336                 strbuf_append_ext_header_uint(&ext_header, "mtime",
337                                               args->time);
338                 args->time = USTAR_MAX_MTIME;
339         }
340
341         if (!ext_header.len)
342                 return;
343
344         memset(&header, 0, sizeof(header));
345         *header.typeflag = TYPEFLAG_GLOBAL_HEADER;
346         mode = 0100666;
347         xsnprintf(header.name, sizeof(header.name), "pax_global_header");
348         prepare_header(args, &header, mode, ext_header.len);
349         write_blocked(&header, sizeof(header));
350         write_blocked(ext_header.buf, ext_header.len);
351         strbuf_release(&ext_header);
352 }
353
354 static struct archiver **tar_filters;
355 static int nr_tar_filters;
356 static int alloc_tar_filters;
357
358 static struct archiver *find_tar_filter(const char *name, int len)
359 {
360         int i;
361         for (i = 0; i < nr_tar_filters; i++) {
362                 struct archiver *ar = tar_filters[i];
363                 if (!strncmp(ar->name, name, len) && !ar->name[len])
364                         return ar;
365         }
366         return NULL;
367 }
368
369 static int tar_filter_config(const char *var, const char *value, void *data)
370 {
371         struct archiver *ar;
372         const char *name;
373         const char *type;
374         int namelen;
375
376         if (parse_config_key(var, "tar", &name, &namelen, &type) < 0 || !name)
377                 return 0;
378
379         ar = find_tar_filter(name, namelen);
380         if (!ar) {
381                 ar = xcalloc(1, sizeof(*ar));
382                 ar->name = xmemdupz(name, namelen);
383                 ar->write_archive = write_tar_filter_archive;
384                 ar->flags = ARCHIVER_WANT_COMPRESSION_LEVELS;
385                 ALLOC_GROW(tar_filters, nr_tar_filters + 1, alloc_tar_filters);
386                 tar_filters[nr_tar_filters++] = ar;
387         }
388
389         if (!strcmp(type, "command")) {
390                 if (!value)
391                         return config_error_nonbool(var);
392                 free(ar->data);
393                 ar->data = xstrdup(value);
394                 return 0;
395         }
396         if (!strcmp(type, "remote")) {
397                 if (git_config_bool(var, value))
398                         ar->flags |= ARCHIVER_REMOTE;
399                 else
400                         ar->flags &= ~ARCHIVER_REMOTE;
401                 return 0;
402         }
403
404         return 0;
405 }
406
407 static int git_tar_config(const char *var, const char *value, void *cb)
408 {
409         if (!strcmp(var, "tar.umask")) {
410                 if (value && !strcmp(value, "user")) {
411                         tar_umask = umask(0);
412                         umask(tar_umask);
413                 } else {
414                         tar_umask = git_config_int(var, value);
415                 }
416                 return 0;
417         }
418
419         return tar_filter_config(var, value, cb);
420 }
421
422 static int write_tar_archive(const struct archiver *ar,
423                              struct archiver_args *args)
424 {
425         int err = 0;
426
427         write_global_extended_header(args);
428         err = write_archive_entries(args, write_tar_entry);
429         if (!err)
430                 write_trailer();
431         return err;
432 }
433
434 static int write_tar_filter_archive(const struct archiver *ar,
435                                     struct archiver_args *args)
436 {
437         struct strbuf cmd = STRBUF_INIT;
438         struct child_process filter = CHILD_PROCESS_INIT;
439         const char *argv[2];
440         int r;
441
442         if (!ar->data)
443                 die("BUG: tar-filter archiver called with no filter defined");
444
445         strbuf_addstr(&cmd, ar->data);
446         if (args->compression_level >= 0)
447                 strbuf_addf(&cmd, " -%d", args->compression_level);
448
449         argv[0] = cmd.buf;
450         argv[1] = NULL;
451         filter.argv = argv;
452         filter.use_shell = 1;
453         filter.in = -1;
454
455         if (start_command(&filter) < 0)
456                 die_errno("unable to start '%s' filter", argv[0]);
457         close(1);
458         if (dup2(filter.in, 1) < 0)
459                 die_errno("unable to redirect descriptor");
460         close(filter.in);
461
462         r = write_tar_archive(ar, args);
463
464         close(1);
465         if (finish_command(&filter) != 0)
466                 die("'%s' filter reported error", argv[0]);
467
468         strbuf_release(&cmd);
469         return r;
470 }
471
472 static struct archiver tar_archiver = {
473         "tar",
474         write_tar_archive,
475         ARCHIVER_REMOTE
476 };
477
478 void init_tar_archiver(void)
479 {
480         int i;
481         register_archiver(&tar_archiver);
482
483         tar_filter_config("tar.tgz.command", "gzip -cn", NULL);
484         tar_filter_config("tar.tgz.remote", "true", NULL);
485         tar_filter_config("tar.tar.gz.command", "gzip -cn", NULL);
486         tar_filter_config("tar.tar.gz.remote", "true", NULL);
487         git_config(git_tar_config, NULL);
488         for (i = 0; i < nr_tar_filters; i++) {
489                 /* omit any filters that never had a command configured */
490                 if (tar_filters[i]->data)
491                         register_archiver(tar_filters[i]);
492         }
493 }