Merge branch 'bw/submodule-with-bs-path'
[git/git.git] / pack-write.c
1 #include "cache.h"
2 #include "pack.h"
3 #include "csum-file.h"
4
5 void reset_pack_idx_option(struct pack_idx_option *opts)
6 {
7 memset(opts, 0, sizeof(*opts));
8 opts->version = 2;
9 opts->off32_limit = 0x7fffffff;
10 }
11
12 static int sha1_compare(const void *_a, const void *_b)
13 {
14 struct pack_idx_entry *a = *(struct pack_idx_entry **)_a;
15 struct pack_idx_entry *b = *(struct pack_idx_entry **)_b;
16 return hashcmp(a->sha1, b->sha1);
17 }
18
19 static int cmp_uint32(const void *a_, const void *b_)
20 {
21 uint32_t a = *((uint32_t *)a_);
22 uint32_t b = *((uint32_t *)b_);
23
24 return (a < b) ? -1 : (a != b);
25 }
26
27 static int need_large_offset(off_t offset, const struct pack_idx_option *opts)
28 {
29 uint32_t ofsval;
30
31 if ((offset >> 31) || (opts->off32_limit < offset))
32 return 1;
33 if (!opts->anomaly_nr)
34 return 0;
35 ofsval = offset;
36 return !!bsearch(&ofsval, opts->anomaly, opts->anomaly_nr,
37 sizeof(ofsval), cmp_uint32);
38 }
39
40 /*
41 * On entry *sha1 contains the pack content SHA1 hash, on exit it is
42 * the SHA1 hash of sorted object names. The objects array passed in
43 * will be sorted by SHA1 on exit.
44 */
45 const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects,
46 int nr_objects, const struct pack_idx_option *opts,
47 const unsigned char *sha1)
48 {
49 struct sha1file *f;
50 struct pack_idx_entry **sorted_by_sha, **list, **last;
51 off_t last_obj_offset = 0;
52 uint32_t array[256];
53 int i, fd;
54 uint32_t index_version;
55
56 if (nr_objects) {
57 sorted_by_sha = objects;
58 list = sorted_by_sha;
59 last = sorted_by_sha + nr_objects;
60 for (i = 0; i < nr_objects; ++i) {
61 if (objects[i]->offset > last_obj_offset)
62 last_obj_offset = objects[i]->offset;
63 }
64 QSORT(sorted_by_sha, nr_objects, sha1_compare);
65 }
66 else
67 sorted_by_sha = list = last = NULL;
68
69 if (opts->flags & WRITE_IDX_VERIFY) {
70 assert(index_name);
71 f = sha1fd_check(index_name);
72 } else {
73 if (!index_name) {
74 struct strbuf tmp_file = STRBUF_INIT;
75 fd = odb_mkstemp(&tmp_file, "pack/tmp_idx_XXXXXX");
76 index_name = strbuf_detach(&tmp_file, NULL);
77 } else {
78 unlink(index_name);
79 fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
80 if (fd < 0)
81 die_errno("unable to create '%s'", index_name);
82 }
83 f = sha1fd(fd, index_name);
84 }
85
86 /* if last object's offset is >= 2^31 we should use index V2 */
87 index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
88
89 /* index versions 2 and above need a header */
90 if (index_version >= 2) {
91 struct pack_idx_header hdr;
92 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
93 hdr.idx_version = htonl(index_version);
94 sha1write(f, &hdr, sizeof(hdr));
95 }
96
97 /*
98 * Write the first-level table (the list is sorted,
99 * but we use a 256-entry lookup to be able to avoid
100 * having to do eight extra binary search iterations).
101 */
102 for (i = 0; i < 256; i++) {
103 struct pack_idx_entry **next = list;
104 while (next < last) {
105 struct pack_idx_entry *obj = *next;
106 if (obj->sha1[0] != i)
107 break;
108 next++;
109 }
110 array[i] = htonl(next - sorted_by_sha);
111 list = next;
112 }
113 sha1write(f, array, 256 * 4);
114
115 /*
116 * Write the actual SHA1 entries..
117 */
118 list = sorted_by_sha;
119 for (i = 0; i < nr_objects; i++) {
120 struct pack_idx_entry *obj = *list++;
121 if (index_version < 2) {
122 uint32_t offset = htonl(obj->offset);
123 sha1write(f, &offset, 4);
124 }
125 sha1write(f, obj->sha1, 20);
126 if ((opts->flags & WRITE_IDX_STRICT) &&
127 (i && !hashcmp(list[-2]->sha1, obj->sha1)))
128 die("The same object %s appears twice in the pack",
129 sha1_to_hex(obj->sha1));
130 }
131
132 if (index_version >= 2) {
133 unsigned int nr_large_offset = 0;
134
135 /* write the crc32 table */
136 list = sorted_by_sha;
137 for (i = 0; i < nr_objects; i++) {
138 struct pack_idx_entry *obj = *list++;
139 uint32_t crc32_val = htonl(obj->crc32);
140 sha1write(f, &crc32_val, 4);
141 }
142
143 /* write the 32-bit offset table */
144 list = sorted_by_sha;
145 for (i = 0; i < nr_objects; i++) {
146 struct pack_idx_entry *obj = *list++;
147 uint32_t offset;
148
149 offset = (need_large_offset(obj->offset, opts)
150 ? (0x80000000 | nr_large_offset++)
151 : obj->offset);
152 offset = htonl(offset);
153 sha1write(f, &offset, 4);
154 }
155
156 /* write the large offset table */
157 list = sorted_by_sha;
158 while (nr_large_offset) {
159 struct pack_idx_entry *obj = *list++;
160 uint64_t offset = obj->offset;
161 uint32_t split[2];
162
163 if (!need_large_offset(offset, opts))
164 continue;
165 split[0] = htonl(offset >> 32);
166 split[1] = htonl(offset & 0xffffffff);
167 sha1write(f, split, 8);
168 nr_large_offset--;
169 }
170 }
171
172 sha1write(f, sha1, 20);
173 sha1close(f, NULL, ((opts->flags & WRITE_IDX_VERIFY)
174 ? CSUM_CLOSE : CSUM_FSYNC));
175 return index_name;
176 }
177
178 off_t write_pack_header(struct sha1file *f, uint32_t nr_entries)
179 {
180 struct pack_header hdr;
181
182 hdr.hdr_signature = htonl(PACK_SIGNATURE);
183 hdr.hdr_version = htonl(PACK_VERSION);
184 hdr.hdr_entries = htonl(nr_entries);
185 sha1write(f, &hdr, sizeof(hdr));
186 return sizeof(hdr);
187 }
188
189 /*
190 * Update pack header with object_count and compute new SHA1 for pack data
191 * associated to pack_fd, and write that SHA1 at the end. That new SHA1
192 * is also returned in new_pack_sha1.
193 *
194 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
195 * (without the header update) is computed and validated against the
196 * one provided in partial_pack_sha1. The validation is performed at
197 * partial_pack_offset bytes in the pack file. The SHA1 of the remaining
198 * data (i.e. from partial_pack_offset to the end) is then computed and
199 * returned in partial_pack_sha1.
200 *
201 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
202 * partial_pack_sha1 can refer to the same buffer if the caller is not
203 * interested in the resulting SHA1 of pack data above partial_pack_offset.
204 */
205 void fixup_pack_header_footer(int pack_fd,
206 unsigned char *new_pack_sha1,
207 const char *pack_name,
208 uint32_t object_count,
209 unsigned char *partial_pack_sha1,
210 off_t partial_pack_offset)
211 {
212 int aligned_sz, buf_sz = 8 * 1024;
213 git_SHA_CTX old_sha1_ctx, new_sha1_ctx;
214 struct pack_header hdr;
215 char *buf;
216
217 git_SHA1_Init(&old_sha1_ctx);
218 git_SHA1_Init(&new_sha1_ctx);
219
220 if (lseek(pack_fd, 0, SEEK_SET) != 0)
221 die_errno("Failed seeking to start of '%s'", pack_name);
222 if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
223 die_errno("Unable to reread header of '%s'", pack_name);
224 if (lseek(pack_fd, 0, SEEK_SET) != 0)
225 die_errno("Failed seeking to start of '%s'", pack_name);
226 git_SHA1_Update(&old_sha1_ctx, &hdr, sizeof(hdr));
227 hdr.hdr_entries = htonl(object_count);
228 git_SHA1_Update(&new_sha1_ctx, &hdr, sizeof(hdr));
229 write_or_die(pack_fd, &hdr, sizeof(hdr));
230 partial_pack_offset -= sizeof(hdr);
231
232 buf = xmalloc(buf_sz);
233 aligned_sz = buf_sz - sizeof(hdr);
234 for (;;) {
235 ssize_t m, n;
236 m = (partial_pack_sha1 && partial_pack_offset < aligned_sz) ?
237 partial_pack_offset : aligned_sz;
238 n = xread(pack_fd, buf, m);
239 if (!n)
240 break;
241 if (n < 0)
242 die_errno("Failed to checksum '%s'", pack_name);
243 git_SHA1_Update(&new_sha1_ctx, buf, n);
244
245 aligned_sz -= n;
246 if (!aligned_sz)
247 aligned_sz = buf_sz;
248
249 if (!partial_pack_sha1)
250 continue;
251
252 git_SHA1_Update(&old_sha1_ctx, buf, n);
253 partial_pack_offset -= n;
254 if (partial_pack_offset == 0) {
255 unsigned char sha1[20];
256 git_SHA1_Final(sha1, &old_sha1_ctx);
257 if (hashcmp(sha1, partial_pack_sha1) != 0)
258 die("Unexpected checksum for %s "
259 "(disk corruption?)", pack_name);
260 /*
261 * Now let's compute the SHA1 of the remainder of the
262 * pack, which also means making partial_pack_offset
263 * big enough not to matter anymore.
264 */
265 git_SHA1_Init(&old_sha1_ctx);
266 partial_pack_offset = ~partial_pack_offset;
267 partial_pack_offset -= MSB(partial_pack_offset, 1);
268 }
269 }
270 free(buf);
271
272 if (partial_pack_sha1)
273 git_SHA1_Final(partial_pack_sha1, &old_sha1_ctx);
274 git_SHA1_Final(new_pack_sha1, &new_sha1_ctx);
275 write_or_die(pack_fd, new_pack_sha1, 20);
276 fsync_or_die(pack_fd, pack_name);
277 }
278
279 char *index_pack_lockfile(int ip_out)
280 {
281 char packname[46];
282
283 /*
284 * The first thing we expect from index-pack's output
285 * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
286 * %40s is the newly created pack SHA1 name. In the "keep"
287 * case, we need it to remove the corresponding .keep file
288 * later on. If we don't get that then tough luck with it.
289 */
290 if (read_in_full(ip_out, packname, 46) == 46 && packname[45] == '\n') {
291 const char *name;
292 packname[45] = 0;
293 if (skip_prefix(packname, "keep\t", &name))
294 return xstrfmt("%s/pack/pack-%s.keep",
295 get_object_directory(), name);
296 }
297 return NULL;
298 }
299
300 /*
301 * The per-object header is a pretty dense thing, which is
302 * - first byte: low four bits are "size", then three bits of "type",
303 * and the high bit is "size continues".
304 * - each byte afterwards: low seven bits are size continuation,
305 * with the high bit being "size continues"
306 */
307 int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
308 enum object_type type, uintmax_t size)
309 {
310 int n = 1;
311 unsigned char c;
312
313 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
314 die("bad type %d", type);
315
316 c = (type << 4) | (size & 15);
317 size >>= 4;
318 while (size) {
319 if (n == hdr_len)
320 die("object size is too enormous to format");
321 *hdr++ = c | 0x80;
322 c = size & 0x7f;
323 size >>= 7;
324 n++;
325 }
326 *hdr = c;
327 return n;
328 }
329
330 struct sha1file *create_tmp_packfile(char **pack_tmp_name)
331 {
332 struct strbuf tmpname = STRBUF_INIT;
333 int fd;
334
335 fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
336 *pack_tmp_name = strbuf_detach(&tmpname, NULL);
337 return sha1fd(fd, *pack_tmp_name);
338 }
339
340 void finish_tmp_packfile(struct strbuf *name_buffer,
341 const char *pack_tmp_name,
342 struct pack_idx_entry **written_list,
343 uint32_t nr_written,
344 struct pack_idx_option *pack_idx_opts,
345 unsigned char sha1[])
346 {
347 const char *idx_tmp_name;
348 int basename_len = name_buffer->len;
349
350 if (adjust_shared_perm(pack_tmp_name))
351 die_errno("unable to make temporary pack file readable");
352
353 idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
354 pack_idx_opts, sha1);
355 if (adjust_shared_perm(idx_tmp_name))
356 die_errno("unable to make temporary index file readable");
357
358 strbuf_addf(name_buffer, "%s.pack", sha1_to_hex(sha1));
359
360 if (rename(pack_tmp_name, name_buffer->buf))
361 die_errno("unable to rename temporary pack file");
362
363 strbuf_setlen(name_buffer, basename_len);
364
365 strbuf_addf(name_buffer, "%s.idx", sha1_to_hex(sha1));
366 if (rename(idx_tmp_name, name_buffer->buf))
367 die_errno("unable to rename temporary index file");
368
369 strbuf_setlen(name_buffer, basename_len);
370
371 free((void *)idx_tmp_name);
372 }