Merge branch 'js/gitdir-at-unc-root'
[git/git.git] / pack-write.c
CommitLineData
8b0eca7c
DH
1#include "cache.h"
2#include "pack.h"
aa7e44bf
GB
3#include "csum-file.h"
4
ebcfb379
JH
5void 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}
aa7e44bf
GB
11
12static 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;
e6a492b7 16 return oidcmp(&a->oid, &b->oid);
aa7e44bf
GB
17}
18
3c9fc074
JH
19static 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
fb956c1f
JH
27static int need_large_offset(off_t offset, const struct pack_idx_option *opts)
28{
3c9fc074
JH
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);
fb956c1f
JH
38}
39
aa7e44bf
GB
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 */
3bb72562 45const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects,
ebcfb379 46 int nr_objects, const struct pack_idx_option *opts,
1190a1ac 47 const unsigned char *sha1)
aa7e44bf 48{
98a3beab 49 struct hashfile *f;
aa7e44bf
GB
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;
aa7e44bf
GB
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 }
9ed0d8d6 64 QSORT(sorted_by_sha, nr_objects, sha1_compare);
aa7e44bf
GB
65 }
66 else
67 sorted_by_sha = list = last = NULL;
68
e337a04d
JH
69 if (opts->flags & WRITE_IDX_VERIFY) {
70 assert(index_name);
98a3beab 71 f = hashfd_check(index_name);
aa7e44bf 72 } else {
e337a04d 73 if (!index_name) {
594fa999
JK
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);
e337a04d
JH
77 } else {
78 unlink(index_name);
79 fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
892e723a
JK
80 if (fd < 0)
81 die_errno("unable to create '%s'", index_name);
e337a04d 82 }
98a3beab 83 f = hashfd(fd, index_name);
aa7e44bf 84 }
aa7e44bf
GB
85
86 /* if last object's offset is >= 2^31 we should use index V2 */
fb956c1f 87 index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
aa7e44bf
GB
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);
98a3beab 94 hashwrite(f, &hdr, sizeof(hdr));
aa7e44bf
GB
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;
e6a492b7 106 if (obj->oid.hash[0] != i)
aa7e44bf
GB
107 break;
108 next++;
109 }
110 array[i] = htonl(next - sorted_by_sha);
111 list = next;
112 }
98a3beab 113 hashwrite(f, array, 256 * 4);
aa7e44bf 114
aa7e44bf
GB
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);
98a3beab 123 hashwrite(f, &offset, 4);
aa7e44bf 124 }
98a3beab 125 hashwrite(f, obj->oid.hash, the_hash_algo->rawsz);
68be2fea 126 if ((opts->flags & WRITE_IDX_STRICT) &&
4a7e27e9 127 (i && oideq(&list[-2]->oid, &obj->oid)))
68be2fea 128 die("The same object %s appears twice in the pack",
e6a492b7 129 oid_to_hex(&obj->oid));
aa7e44bf
GB
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);
98a3beab 140 hashwrite(f, &crc32_val, 4);
aa7e44bf
GB
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++;
fb956c1f
JH
147 uint32_t offset;
148
149 offset = (need_large_offset(obj->offset, opts)
150 ? (0x80000000 | nr_large_offset++)
151 : obj->offset);
aa7e44bf 152 offset = htonl(offset);
98a3beab 153 hashwrite(f, &offset, 4);
aa7e44bf
GB
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;
fb956c1f
JH
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);
98a3beab 167 hashwrite(f, split, 8);
fb956c1f 168 nr_large_offset--;
aa7e44bf
GB
169 }
170 }
171
98a3beab 172 hashwrite(f, sha1, the_hash_algo->rawsz);
cfe83216
DS
173 finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_CLOSE |
174 ((opts->flags & WRITE_IDX_VERIFY)
175 ? 0 : CSUM_FSYNC));
aa7e44bf
GB
176 return index_name;
177}
8b0eca7c 178
98a3beab 179off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)
c0ad4657
JH
180{
181 struct pack_header hdr;
182
183 hdr.hdr_signature = htonl(PACK_SIGNATURE);
184 hdr.hdr_version = htonl(PACK_VERSION);
185 hdr.hdr_entries = htonl(nr_entries);
98a3beab 186 hashwrite(f, &hdr, sizeof(hdr));
c0ad4657
JH
187 return sizeof(hdr);
188}
189
abeb40e5
NP
190/*
191 * Update pack header with object_count and compute new SHA1 for pack data
192 * associated to pack_fd, and write that SHA1 at the end. That new SHA1
193 * is also returned in new_pack_sha1.
194 *
195 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
196 * (without the header update) is computed and validated against the
197 * one provided in partial_pack_sha1. The validation is performed at
198 * partial_pack_offset bytes in the pack file. The SHA1 of the remaining
199 * data (i.e. from partial_pack_offset to the end) is then computed and
200 * returned in partial_pack_sha1.
201 *
202 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
203 * partial_pack_sha1 can refer to the same buffer if the caller is not
204 * interested in the resulting SHA1 of pack data above partial_pack_offset.
205 */
8b0eca7c 206void fixup_pack_header_footer(int pack_fd,
81c58cd4 207 unsigned char *new_pack_hash,
8b0eca7c 208 const char *pack_name,
abeb40e5 209 uint32_t object_count,
81c58cd4 210 unsigned char *partial_pack_hash,
abeb40e5 211 off_t partial_pack_offset)
8b0eca7c 212{
d35825da 213 int aligned_sz, buf_sz = 8 * 1024;
81c58cd4 214 git_hash_ctx old_hash_ctx, new_hash_ctx;
8b0eca7c
DH
215 struct pack_header hdr;
216 char *buf;
90dca671 217 ssize_t read_result;
8b0eca7c 218
81c58cd4 219 the_hash_algo->init_fn(&old_hash_ctx);
220 the_hash_algo->init_fn(&new_hash_ctx);
abeb40e5 221
8b0eca7c 222 if (lseek(pack_fd, 0, SEEK_SET) != 0)
d824cbba 223 die_errno("Failed seeking to start of '%s'", pack_name);
90dca671
JK
224 read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));
225 if (read_result < 0)
d824cbba 226 die_errno("Unable to reread header of '%s'", pack_name);
90dca671
JK
227 else if (read_result != sizeof(hdr))
228 die_errno("Unexpected short read for header of '%s'",
229 pack_name);
8b0eca7c 230 if (lseek(pack_fd, 0, SEEK_SET) != 0)
d824cbba 231 die_errno("Failed seeking to start of '%s'", pack_name);
81c58cd4 232 the_hash_algo->update_fn(&old_hash_ctx, &hdr, sizeof(hdr));
8b0eca7c 233 hdr.hdr_entries = htonl(object_count);
81c58cd4 234 the_hash_algo->update_fn(&new_hash_ctx, &hdr, sizeof(hdr));
8b0eca7c 235 write_or_die(pack_fd, &hdr, sizeof(hdr));
abeb40e5 236 partial_pack_offset -= sizeof(hdr);
8b0eca7c
DH
237
238 buf = xmalloc(buf_sz);
d35825da 239 aligned_sz = buf_sz - sizeof(hdr);
8b0eca7c 240 for (;;) {
abeb40e5 241 ssize_t m, n;
81c58cd4 242 m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?
d35825da 243 partial_pack_offset : aligned_sz;
abeb40e5 244 n = xread(pack_fd, buf, m);
8b0eca7c
DH
245 if (!n)
246 break;
247 if (n < 0)
d824cbba 248 die_errno("Failed to checksum '%s'", pack_name);
81c58cd4 249 the_hash_algo->update_fn(&new_hash_ctx, buf, n);
abeb40e5 250
d35825da
NP
251 aligned_sz -= n;
252 if (!aligned_sz)
253 aligned_sz = buf_sz;
254
81c58cd4 255 if (!partial_pack_hash)
abeb40e5
NP
256 continue;
257
81c58cd4 258 the_hash_algo->update_fn(&old_hash_ctx, buf, n);
abeb40e5
NP
259 partial_pack_offset -= n;
260 if (partial_pack_offset == 0) {
81c58cd4 261 unsigned char hash[GIT_MAX_RAWSZ];
262 the_hash_algo->final_fn(hash, &old_hash_ctx);
67947c34 263 if (!hasheq(hash, partial_pack_hash))
abeb40e5
NP
264 die("Unexpected checksum for %s "
265 "(disk corruption?)", pack_name);
266 /*
267 * Now let's compute the SHA1 of the remainder of the
268 * pack, which also means making partial_pack_offset
269 * big enough not to matter anymore.
270 */
81c58cd4 271 the_hash_algo->init_fn(&old_hash_ctx);
abeb40e5
NP
272 partial_pack_offset = ~partial_pack_offset;
273 partial_pack_offset -= MSB(partial_pack_offset, 1);
274 }
8b0eca7c
DH
275 }
276 free(buf);
277
81c58cd4 278 if (partial_pack_hash)
279 the_hash_algo->final_fn(partial_pack_hash, &old_hash_ctx);
280 the_hash_algo->final_fn(new_pack_hash, &new_hash_ctx);
281 write_or_die(pack_fd, new_pack_hash, the_hash_algo->rawsz);
0c68d386 282 fsync_or_die(pack_fd, pack_name);
8b0eca7c 283}
106764e6
SP
284
285char *index_pack_lockfile(int ip_out)
286{
81c58cd4 287 char packname[GIT_MAX_HEXSZ + 6];
288 const int len = the_hash_algo->hexsz + 6;
106764e6
SP
289
290 /*
6e180cdc 291 * The first thing we expect from index-pack's output
106764e6
SP
292 * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
293 * %40s is the newly created pack SHA1 name. In the "keep"
294 * case, we need it to remove the corresponding .keep file
295 * later on. If we don't get that then tough luck with it.
296 */
81c58cd4 297 if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
d7731444 298 const char *name;
81c58cd4 299 packname[len-1] = 0;
d7731444
RS
300 if (skip_prefix(packname, "keep\t", &name))
301 return xstrfmt("%s/pack/pack-%s.keep",
302 get_object_directory(), name);
106764e6
SP
303 }
304 return NULL;
305}
f965c525
NP
306
307/*
308 * The per-object header is a pretty dense thing, which is
309 * - first byte: low four bits are "size", then three bits of "type",
310 * and the high bit is "size continues".
311 * - each byte afterwards: low seven bits are size continuation,
312 * with the high bit being "size continues"
313 */
7202a6fa
JK
314int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
315 enum object_type type, uintmax_t size)
f965c525
NP
316{
317 int n = 1;
318 unsigned char c;
319
320 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
321 die("bad type %d", type);
322
323 c = (type << 4) | (size & 15);
324 size >>= 4;
325 while (size) {
7202a6fa
JK
326 if (n == hdr_len)
327 die("object size is too enormous to format");
f965c525
NP
328 *hdr++ = c | 0x80;
329 c = size & 0x7f;
330 size >>= 7;
331 n++;
332 }
333 *hdr = c;
334 return n;
335}
cdf9db3c 336
98a3beab 337struct hashfile *create_tmp_packfile(char **pack_tmp_name)
cdf9db3c 338{
594fa999 339 struct strbuf tmpname = STRBUF_INIT;
cdf9db3c
JH
340 int fd;
341
594fa999
JK
342 fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
343 *pack_tmp_name = strbuf_detach(&tmpname, NULL);
98a3beab 344 return hashfd(fd, *pack_tmp_name);
cdf9db3c 345}
0e990530 346
58892711 347void finish_tmp_packfile(struct strbuf *name_buffer,
0e990530
JH
348 const char *pack_tmp_name,
349 struct pack_idx_entry **written_list,
350 uint32_t nr_written,
351 struct pack_idx_option *pack_idx_opts,
352 unsigned char sha1[])
353{
354 const char *idx_tmp_name;
58892711 355 int basename_len = name_buffer->len;
0e990530
JH
356
357 if (adjust_shared_perm(pack_tmp_name))
358 die_errno("unable to make temporary pack file readable");
359
360 idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
361 pack_idx_opts, sha1);
362 if (adjust_shared_perm(idx_tmp_name))
363 die_errno("unable to make temporary index file readable");
364
58892711 365 strbuf_addf(name_buffer, "%s.pack", sha1_to_hex(sha1));
0e990530 366
58892711 367 if (rename(pack_tmp_name, name_buffer->buf))
0e990530
JH
368 die_errno("unable to rename temporary pack file");
369
58892711
SH
370 strbuf_setlen(name_buffer, basename_len);
371
372 strbuf_addf(name_buffer, "%s.idx", sha1_to_hex(sha1));
373 if (rename(idx_tmp_name, name_buffer->buf))
0e990530
JH
374 die_errno("unable to rename temporary index file");
375
58892711 376 strbuf_setlen(name_buffer, basename_len);
7cc8f971 377
0e990530
JH
378 free((void *)idx_tmp_name);
379}