Speed-up recursive by flushing index only once for all entries
[git] / archive-tar.c
1 /*
2  * Copyright (c) 2005, 2006 Rene Scharfe
3  */
4 #include "cache.h"
5 #include "commit.h"
6 #include "strbuf.h"
7 #include "tar.h"
8 #include "builtin.h"
9 #include "archive.h"
10
11 #define RECORDSIZE      (512)
12 #define BLOCKSIZE       (RECORDSIZE * 20)
13
14 static char block[BLOCKSIZE];
15 static unsigned long offset;
16
17 static time_t archive_time;
18 static int tar_umask = 002;
19 static int verbose;
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 write_blocked(const void *data, unsigned long size)
35 {
36         const char *buf = data;
37         unsigned long tail;
38
39         if (offset) {
40                 unsigned long chunk = BLOCKSIZE - offset;
41                 if (size < chunk)
42                         chunk = size;
43                 memcpy(block + offset, buf, chunk);
44                 size -= chunk;
45                 offset += chunk;
46                 buf += chunk;
47                 write_if_needed();
48         }
49         while (size >= BLOCKSIZE) {
50                 write_or_die(1, buf, BLOCKSIZE);
51                 size -= BLOCKSIZE;
52                 buf += BLOCKSIZE;
53         }
54         if (size) {
55                 memcpy(block + offset, buf, size);
56                 offset += size;
57         }
58         tail = offset % RECORDSIZE;
59         if (tail)  {
60                 memset(block + offset, 0, RECORDSIZE - tail);
61                 offset += RECORDSIZE - tail;
62         }
63         write_if_needed();
64 }
65
66 /*
67  * The end of tar archives is marked by 2*512 nul bytes and after that
68  * follows the rest of the block (if any).
69  */
70 static void write_trailer(void)
71 {
72         int tail = BLOCKSIZE - offset;
73         memset(block + offset, 0, tail);
74         write_or_die(1, block, BLOCKSIZE);
75         if (tail < 2 * RECORDSIZE) {
76                 memset(block, 0, offset);
77                 write_or_die(1, block, BLOCKSIZE);
78         }
79 }
80
81 static void strbuf_append_string(struct strbuf *sb, const char *s)
82 {
83         int slen = strlen(s);
84         int total = sb->len + slen;
85         if (total > sb->alloc) {
86                 sb->buf = xrealloc(sb->buf, total);
87                 sb->alloc = total;
88         }
89         memcpy(sb->buf + sb->len, s, slen);
90         sb->len = total;
91 }
92
93 /*
94  * pax extended header records have the format "%u %s=%s\n".  %u contains
95  * the size of the whole string (including the %u), the first %s is the
96  * keyword, the second one is the value.  This function constructs such a
97  * string and appends it to a struct strbuf.
98  */
99 static void strbuf_append_ext_header(struct strbuf *sb, const char *keyword,
100                                      const char *value, unsigned int valuelen)
101 {
102         char *p;
103         int len, total, tmp;
104
105         /* "%u %s=%s\n" */
106         len = 1 + 1 + strlen(keyword) + 1 + valuelen + 1;
107         for (tmp = len; tmp > 9; tmp /= 10)
108                 len++;
109
110         total = sb->len + len;
111         if (total > sb->alloc) {
112                 sb->buf = xrealloc(sb->buf, total);
113                 sb->alloc = total;
114         }
115
116         p = sb->buf;
117         p += sprintf(p, "%u %s=", len, keyword);
118         memcpy(p, value, valuelen);
119         p += valuelen;
120         *p = '\n';
121         sb->len = total;
122 }
123
124 static unsigned int ustar_header_chksum(const struct ustar_header *header)
125 {
126         char *p = (char *)header;
127         unsigned int chksum = 0;
128         while (p < header->chksum)
129                 chksum += *p++;
130         chksum += sizeof(header->chksum) * ' ';
131         p += sizeof(header->chksum);
132         while (p < (char *)header + sizeof(struct ustar_header))
133                 chksum += *p++;
134         return chksum;
135 }
136
137 static int get_path_prefix(const struct strbuf *path, int maxlen)
138 {
139         int i = path->len;
140         if (i > maxlen)
141                 i = maxlen;
142         do {
143                 i--;
144         } while (i > 0 && path->buf[i] != '/');
145         return i;
146 }
147
148 static void write_entry(const unsigned char *sha1, struct strbuf *path,
149                         unsigned int mode, void *buffer, unsigned long size)
150 {
151         struct ustar_header header;
152         struct strbuf ext_header;
153
154         memset(&header, 0, sizeof(header));
155         ext_header.buf = NULL;
156         ext_header.len = ext_header.alloc = 0;
157
158         if (!sha1) {
159                 *header.typeflag = TYPEFLAG_GLOBAL_HEADER;
160                 mode = 0100666;
161                 strcpy(header.name, "pax_global_header");
162         } else if (!path) {
163                 *header.typeflag = TYPEFLAG_EXT_HEADER;
164                 mode = 0100666;
165                 sprintf(header.name, "%s.paxheader", sha1_to_hex(sha1));
166         } else {
167                 if (verbose)
168                         fprintf(stderr, "%.*s\n", path->len, path->buf);
169                 if (S_ISDIR(mode)) {
170                         *header.typeflag = TYPEFLAG_DIR;
171                         mode = (mode | 0777) & ~tar_umask;
172                 } else if (S_ISLNK(mode)) {
173                         *header.typeflag = TYPEFLAG_LNK;
174                         mode |= 0777;
175                 } else if (S_ISREG(mode)) {
176                         *header.typeflag = TYPEFLAG_REG;
177                         mode = (mode | ((mode & 0100) ? 0777 : 0666)) & ~tar_umask;
178                 } else {
179                         error("unsupported file mode: 0%o (SHA1: %s)",
180                               mode, sha1_to_hex(sha1));
181                         return;
182                 }
183                 if (path->len > sizeof(header.name)) {
184                         int plen = get_path_prefix(path, sizeof(header.prefix));
185                         int rest = path->len - plen - 1;
186                         if (plen > 0 && rest <= sizeof(header.name)) {
187                                 memcpy(header.prefix, path->buf, plen);
188                                 memcpy(header.name, path->buf + plen + 1, rest);
189                         } else {
190                                 sprintf(header.name, "%s.data",
191                                         sha1_to_hex(sha1));
192                                 strbuf_append_ext_header(&ext_header, "path",
193                                                          path->buf, path->len);
194                         }
195                 } else
196                         memcpy(header.name, path->buf, path->len);
197         }
198
199         if (S_ISLNK(mode) && buffer) {
200                 if (size > sizeof(header.linkname)) {
201                         sprintf(header.linkname, "see %s.paxheader",
202                                 sha1_to_hex(sha1));
203                         strbuf_append_ext_header(&ext_header, "linkpath",
204                                                  buffer, size);
205                 } else
206                         memcpy(header.linkname, buffer, size);
207         }
208
209         sprintf(header.mode, "%07o", mode & 07777);
210         sprintf(header.size, "%011lo", S_ISREG(mode) ? size : 0);
211         sprintf(header.mtime, "%011lo", archive_time);
212
213         sprintf(header.uid, "%07o", 0);
214         sprintf(header.gid, "%07o", 0);
215         strlcpy(header.uname, "root", sizeof(header.uname));
216         strlcpy(header.gname, "root", sizeof(header.gname));
217         sprintf(header.devmajor, "%07o", 0);
218         sprintf(header.devminor, "%07o", 0);
219
220         memcpy(header.magic, "ustar", 6);
221         memcpy(header.version, "00", 2);
222
223         sprintf(header.chksum, "%07o", ustar_header_chksum(&header));
224
225         if (ext_header.len > 0) {
226                 write_entry(sha1, NULL, 0, ext_header.buf, ext_header.len);
227                 free(ext_header.buf);
228         }
229         write_blocked(&header, sizeof(header));
230         if (S_ISREG(mode) && buffer && size > 0)
231                 write_blocked(buffer, size);
232 }
233
234 static void write_global_extended_header(const unsigned char *sha1)
235 {
236         struct strbuf ext_header;
237         ext_header.buf = NULL;
238         ext_header.len = ext_header.alloc = 0;
239         strbuf_append_ext_header(&ext_header, "comment", sha1_to_hex(sha1), 40);
240         write_entry(NULL, NULL, 0, ext_header.buf, ext_header.len);
241         free(ext_header.buf);
242 }
243
244 static int git_tar_config(const char *var, const char *value)
245 {
246         if (!strcmp(var, "tar.umask")) {
247                 if (!strcmp(value, "user")) {
248                         tar_umask = umask(0);
249                         umask(tar_umask);
250                 } else {
251                         tar_umask = git_config_int(var, value);
252                 }
253                 return 0;
254         }
255         return git_default_config(var, value);
256 }
257
258 static int write_tar_entry(const unsigned char *sha1,
259                            const char *base, int baselen,
260                            const char *filename, unsigned mode, int stage)
261 {
262         static struct strbuf path;
263         int filenamelen = strlen(filename);
264         void *buffer;
265         char type[20];
266         unsigned long size;
267
268         if (!path.alloc) {
269                 path.buf = xmalloc(PATH_MAX);
270                 path.alloc = PATH_MAX;
271                 path.len = path.eof = 0;
272         }
273         if (path.alloc < baselen + filenamelen) {
274                 free(path.buf);
275                 path.buf = xmalloc(baselen + filenamelen);
276                 path.alloc = baselen + filenamelen;
277         }
278         memcpy(path.buf, base, baselen);
279         memcpy(path.buf + baselen, filename, filenamelen);
280         path.len = baselen + filenamelen;
281         if (S_ISDIR(mode)) {
282                 strbuf_append_string(&path, "/");
283                 buffer = NULL;
284                 size = 0;
285         } else {
286                 buffer = read_sha1_file(sha1, type, &size);
287                 if (!buffer)
288                         die("cannot read %s", sha1_to_hex(sha1));
289         }
290
291         write_entry(sha1, &path, mode, buffer, size);
292         free(buffer);
293
294         return READ_TREE_RECURSIVE;
295 }
296
297 int write_tar_archive(struct archiver_args *args)
298 {
299         int plen = args->base ? strlen(args->base) : 0;
300
301         git_config(git_tar_config);
302
303         archive_time = args->time;
304         verbose = args->verbose;
305
306         if (args->commit_sha1)
307                 write_global_extended_header(args->commit_sha1);
308
309         if (args->base && plen > 0 && args->base[plen - 1] == '/') {
310                 char *base = xstrdup(args->base);
311                 int baselen = strlen(base);
312
313                 while (baselen > 0 && base[baselen - 1] == '/')
314                         base[--baselen] = '\0';
315                 write_tar_entry(args->tree->object.sha1, "", 0, base, 040777, 0);
316                 free(base);
317         }
318         read_tree_recursive(args->tree, args->base, plen, 0,
319                             args->pathspec, write_tar_entry);
320         write_trailer();
321
322         return 0;
323 }