archive-tar: split long paths more carefully
[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 > 1 && path[i - 1] == '/')
119                 i--;
120         if (i > maxlen)
121                 i = maxlen;
122         do {
123                 i--;
124         } while (i > 0 && path[i] != '/');
125         return i;
126 }
127
128 static int write_tar_entry(struct archiver_args *args,
129                 const unsigned char *sha1, const char *path, size_t pathlen,
130                 unsigned int mode, void *buffer, unsigned long size)
131 {
132         struct ustar_header header;
133         struct strbuf ext_header = STRBUF_INIT;
134         int err = 0;
135
136         memset(&header, 0, sizeof(header));
137
138         if (!sha1) {
139                 *header.typeflag = TYPEFLAG_GLOBAL_HEADER;
140                 mode = 0100666;
141                 strcpy(header.name, "pax_global_header");
142         } else if (!path) {
143                 *header.typeflag = TYPEFLAG_EXT_HEADER;
144                 mode = 0100666;
145                 sprintf(header.name, "%s.paxheader", sha1_to_hex(sha1));
146         } else {
147                 if (S_ISDIR(mode) || S_ISGITLINK(mode)) {
148                         *header.typeflag = TYPEFLAG_DIR;
149                         mode = (mode | 0777) & ~tar_umask;
150                 } else if (S_ISLNK(mode)) {
151                         *header.typeflag = TYPEFLAG_LNK;
152                         mode |= 0777;
153                 } else if (S_ISREG(mode)) {
154                         *header.typeflag = TYPEFLAG_REG;
155                         mode = (mode | ((mode & 0100) ? 0777 : 0666)) & ~tar_umask;
156                 } else {
157                         return error("unsupported file mode: 0%o (SHA1: %s)",
158                                         mode, sha1_to_hex(sha1));
159                 }
160                 if (pathlen > sizeof(header.name)) {
161                         size_t plen = get_path_prefix(path, pathlen,
162                                         sizeof(header.prefix));
163                         size_t rest = pathlen - plen - 1;
164                         if (plen > 0 && rest <= sizeof(header.name)) {
165                                 memcpy(header.prefix, path, plen);
166                                 memcpy(header.name, path + plen + 1, rest);
167                         } else {
168                                 sprintf(header.name, "%s.data",
169                                         sha1_to_hex(sha1));
170                                 strbuf_append_ext_header(&ext_header, "path",
171                                                 path, pathlen);
172                         }
173                 } else
174                         memcpy(header.name, path, pathlen);
175         }
176
177         if (S_ISLNK(mode) && buffer) {
178                 if (size > sizeof(header.linkname)) {
179                         sprintf(header.linkname, "see %s.paxheader",
180                                 sha1_to_hex(sha1));
181                         strbuf_append_ext_header(&ext_header, "linkpath",
182                                                  buffer, size);
183                 } else
184                         memcpy(header.linkname, buffer, size);
185         }
186
187         sprintf(header.mode, "%07o", mode & 07777);
188         sprintf(header.size, "%011lo", S_ISREG(mode) ? size : 0);
189         sprintf(header.mtime, "%011lo", (unsigned long) args->time);
190
191         sprintf(header.uid, "%07o", 0);
192         sprintf(header.gid, "%07o", 0);
193         strlcpy(header.uname, "root", sizeof(header.uname));
194         strlcpy(header.gname, "root", sizeof(header.gname));
195         sprintf(header.devmajor, "%07o", 0);
196         sprintf(header.devminor, "%07o", 0);
197
198         memcpy(header.magic, "ustar", 6);
199         memcpy(header.version, "00", 2);
200
201         sprintf(header.chksum, "%07o", ustar_header_chksum(&header));
202
203         if (ext_header.len > 0) {
204                 err = write_tar_entry(args, sha1, NULL, 0, 0, ext_header.buf,
205                                 ext_header.len);
206                 if (err)
207                         return err;
208         }
209         strbuf_release(&ext_header);
210         write_blocked(&header, sizeof(header));
211         if (S_ISREG(mode) && buffer && size > 0)
212                 write_blocked(buffer, size);
213         return err;
214 }
215
216 static int write_global_extended_header(struct archiver_args *args)
217 {
218         const unsigned char *sha1 = args->commit_sha1;
219         struct strbuf ext_header = STRBUF_INIT;
220         int err;
221
222         strbuf_append_ext_header(&ext_header, "comment", sha1_to_hex(sha1), 40);
223         err = write_tar_entry(args, NULL, NULL, 0, 0, ext_header.buf,
224                         ext_header.len);
225         strbuf_release(&ext_header);
226         return err;
227 }
228
229 static struct archiver **tar_filters;
230 static int nr_tar_filters;
231 static int alloc_tar_filters;
232
233 static struct archiver *find_tar_filter(const char *name, int len)
234 {
235         int i;
236         for (i = 0; i < nr_tar_filters; i++) {
237                 struct archiver *ar = tar_filters[i];
238                 if (!strncmp(ar->name, name, len) && !ar->name[len])
239                         return ar;
240         }
241         return NULL;
242 }
243
244 static int tar_filter_config(const char *var, const char *value, void *data)
245 {
246         struct archiver *ar;
247         const char *dot;
248         const char *name;
249         const char *type;
250         int namelen;
251
252         if (prefixcmp(var, "tar."))
253                 return 0;
254         dot = strrchr(var, '.');
255         if (dot == var + 9)
256                 return 0;
257
258         name = var + 4;
259         namelen = dot - name;
260         type = dot + 1;
261
262         ar = find_tar_filter(name, namelen);
263         if (!ar) {
264                 ar = xcalloc(1, sizeof(*ar));
265                 ar->name = xmemdupz(name, namelen);
266                 ar->write_archive = write_tar_filter_archive;
267                 ar->flags = ARCHIVER_WANT_COMPRESSION_LEVELS;
268                 ALLOC_GROW(tar_filters, nr_tar_filters + 1, alloc_tar_filters);
269                 tar_filters[nr_tar_filters++] = ar;
270         }
271
272         if (!strcmp(type, "command")) {
273                 if (!value)
274                         return config_error_nonbool(var);
275                 free(ar->data);
276                 ar->data = xstrdup(value);
277                 return 0;
278         }
279         if (!strcmp(type, "remote")) {
280                 if (git_config_bool(var, value))
281                         ar->flags |= ARCHIVER_REMOTE;
282                 else
283                         ar->flags &= ~ARCHIVER_REMOTE;
284                 return 0;
285         }
286
287         return 0;
288 }
289
290 static int git_tar_config(const char *var, const char *value, void *cb)
291 {
292         if (!strcmp(var, "tar.umask")) {
293                 if (value && !strcmp(value, "user")) {
294                         tar_umask = umask(0);
295                         umask(tar_umask);
296                 } else {
297                         tar_umask = git_config_int(var, value);
298                 }
299                 return 0;
300         }
301
302         return tar_filter_config(var, value, cb);
303 }
304
305 static int write_tar_archive(const struct archiver *ar,
306                              struct archiver_args *args)
307 {
308         int err = 0;
309
310         if (args->commit_sha1)
311                 err = write_global_extended_header(args);
312         if (!err)
313                 err = write_archive_entries(args, write_tar_entry);
314         if (!err)
315                 write_trailer();
316         return err;
317 }
318
319 static int write_tar_filter_archive(const struct archiver *ar,
320                                     struct archiver_args *args)
321 {
322         struct strbuf cmd = STRBUF_INIT;
323         struct child_process filter;
324         const char *argv[2];
325         int r;
326
327         if (!ar->data)
328                 die("BUG: tar-filter archiver called with no filter defined");
329
330         strbuf_addstr(&cmd, ar->data);
331         if (args->compression_level >= 0)
332                 strbuf_addf(&cmd, " -%d", args->compression_level);
333
334         memset(&filter, 0, sizeof(filter));
335         argv[0] = cmd.buf;
336         argv[1] = NULL;
337         filter.argv = argv;
338         filter.use_shell = 1;
339         filter.in = -1;
340
341         if (start_command(&filter) < 0)
342                 die_errno("unable to start '%s' filter", argv[0]);
343         close(1);
344         if (dup2(filter.in, 1) < 0)
345                 die_errno("unable to redirect descriptor");
346         close(filter.in);
347
348         r = write_tar_archive(ar, args);
349
350         close(1);
351         if (finish_command(&filter) != 0)
352                 die("'%s' filter reported error", argv[0]);
353
354         strbuf_release(&cmd);
355         return r;
356 }
357
358 static struct archiver tar_archiver = {
359         "tar",
360         write_tar_archive,
361         ARCHIVER_REMOTE
362 };
363
364 void init_tar_archiver(void)
365 {
366         int i;
367         register_archiver(&tar_archiver);
368
369         tar_filter_config("tar.tgz.command", "gzip -cn", NULL);
370         tar_filter_config("tar.tgz.remote", "true", NULL);
371         tar_filter_config("tar.tar.gz.command", "gzip -cn", NULL);
372         tar_filter_config("tar.tar.gz.remote", "true", NULL);
373         git_config(git_tar_config, NULL);
374         for (i = 0; i < nr_tar_filters; i++) {
375                 /* omit any filters that never had a command configured */
376                 if (tar_filters[i]->data)
377                         register_archiver(tar_filters[i]);
378         }
379 }