pack: convert struct pack_idx_entry to struct object_id
[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
GB
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;
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);
71 f = sha1fd_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 }
e337a04d 83 f = sha1fd(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);
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;
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 }
113 sha1write(f, array, 256 * 4);
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);
123 sha1write(f, &offset, 4);
124 }
e6a492b7 125 sha1write(f, obj->oid.hash, 20);
68be2fea 126 if ((opts->flags & WRITE_IDX_STRICT) &&
e6a492b7 127 (i && !oidcmp(&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);
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++;
fb956c1f
JH
147 uint32_t offset;
148
149 offset = (need_large_offset(obj->offset, opts)
150 ? (0x80000000 | nr_large_offset++)
151 : obj->offset);
aa7e44bf
GB
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;
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);
167 sha1write(f, split, 8);
168 nr_large_offset--;
aa7e44bf
GB
169 }
170 }
171
172 sha1write(f, sha1, 20);
e337a04d
JH
173 sha1close(f, NULL, ((opts->flags & WRITE_IDX_VERIFY)
174 ? CSUM_CLOSE : CSUM_FSYNC));
aa7e44bf
GB
175 return index_name;
176}
8b0eca7c 177
c0ad4657
JH
178off_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);
9af270e8 185 sha1write(f, &hdr, sizeof(hdr));
c0ad4657
JH
186 return sizeof(hdr);
187}
188
abeb40e5
NP
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 */
8b0eca7c 205void fixup_pack_header_footer(int pack_fd,
abeb40e5 206 unsigned char *new_pack_sha1,
8b0eca7c 207 const char *pack_name,
abeb40e5
NP
208 uint32_t object_count,
209 unsigned char *partial_pack_sha1,
210 off_t partial_pack_offset)
8b0eca7c 211{
d35825da 212 int aligned_sz, buf_sz = 8 * 1024;
9126f009 213 git_SHA_CTX old_sha1_ctx, new_sha1_ctx;
8b0eca7c
DH
214 struct pack_header hdr;
215 char *buf;
216
9126f009
NP
217 git_SHA1_Init(&old_sha1_ctx);
218 git_SHA1_Init(&new_sha1_ctx);
abeb40e5 219
8b0eca7c 220 if (lseek(pack_fd, 0, SEEK_SET) != 0)
d824cbba 221 die_errno("Failed seeking to start of '%s'", pack_name);
8b0eca7c 222 if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
d824cbba 223 die_errno("Unable to reread header of '%s'", pack_name);
8b0eca7c 224 if (lseek(pack_fd, 0, SEEK_SET) != 0)
d824cbba 225 die_errno("Failed seeking to start of '%s'", pack_name);
9126f009 226 git_SHA1_Update(&old_sha1_ctx, &hdr, sizeof(hdr));
8b0eca7c 227 hdr.hdr_entries = htonl(object_count);
9126f009 228 git_SHA1_Update(&new_sha1_ctx, &hdr, sizeof(hdr));
8b0eca7c 229 write_or_die(pack_fd, &hdr, sizeof(hdr));
abeb40e5 230 partial_pack_offset -= sizeof(hdr);
8b0eca7c
DH
231
232 buf = xmalloc(buf_sz);
d35825da 233 aligned_sz = buf_sz - sizeof(hdr);
8b0eca7c 234 for (;;) {
abeb40e5 235 ssize_t m, n;
d35825da
NP
236 m = (partial_pack_sha1 && partial_pack_offset < aligned_sz) ?
237 partial_pack_offset : aligned_sz;
abeb40e5 238 n = xread(pack_fd, buf, m);
8b0eca7c
DH
239 if (!n)
240 break;
241 if (n < 0)
d824cbba 242 die_errno("Failed to checksum '%s'", pack_name);
9126f009 243 git_SHA1_Update(&new_sha1_ctx, buf, n);
abeb40e5 244
d35825da
NP
245 aligned_sz -= n;
246 if (!aligned_sz)
247 aligned_sz = buf_sz;
248
abeb40e5
NP
249 if (!partial_pack_sha1)
250 continue;
251
9126f009 252 git_SHA1_Update(&old_sha1_ctx, buf, n);
abeb40e5
NP
253 partial_pack_offset -= n;
254 if (partial_pack_offset == 0) {
255 unsigned char sha1[20];
9126f009 256 git_SHA1_Final(sha1, &old_sha1_ctx);
abeb40e5
NP
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 */
9126f009 265 git_SHA1_Init(&old_sha1_ctx);
abeb40e5
NP
266 partial_pack_offset = ~partial_pack_offset;
267 partial_pack_offset -= MSB(partial_pack_offset, 1);
268 }
8b0eca7c
DH
269 }
270 free(buf);
271
abeb40e5 272 if (partial_pack_sha1)
9126f009
NP
273 git_SHA1_Final(partial_pack_sha1, &old_sha1_ctx);
274 git_SHA1_Final(new_pack_sha1, &new_sha1_ctx);
abeb40e5 275 write_or_die(pack_fd, new_pack_sha1, 20);
0c68d386 276 fsync_or_die(pack_fd, pack_name);
8b0eca7c 277}
106764e6
SP
278
279char *index_pack_lockfile(int ip_out)
280{
106764e6
SP
281 char packname[46];
282
283 /*
6e180cdc 284 * The first thing we expect from index-pack's output
106764e6
SP
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 */
d7731444
RS
290 if (read_in_full(ip_out, packname, 46) == 46 && packname[45] == '\n') {
291 const char *name;
106764e6 292 packname[45] = 0;
d7731444
RS
293 if (skip_prefix(packname, "keep\t", &name))
294 return xstrfmt("%s/pack/pack-%s.keep",
295 get_object_directory(), name);
106764e6
SP
296 }
297 return NULL;
298}
f965c525
NP
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 */
7202a6fa
JK
307int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
308 enum object_type type, uintmax_t size)
f965c525
NP
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) {
7202a6fa
JK
319 if (n == hdr_len)
320 die("object size is too enormous to format");
f965c525
NP
321 *hdr++ = c | 0x80;
322 c = size & 0x7f;
323 size >>= 7;
324 n++;
325 }
326 *hdr = c;
327 return n;
328}
cdf9db3c
JH
329
330struct sha1file *create_tmp_packfile(char **pack_tmp_name)
331{
594fa999 332 struct strbuf tmpname = STRBUF_INIT;
cdf9db3c
JH
333 int fd;
334
594fa999
JK
335 fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
336 *pack_tmp_name = strbuf_detach(&tmpname, NULL);
cdf9db3c
JH
337 return sha1fd(fd, *pack_tmp_name);
338}
0e990530 339
58892711 340void finish_tmp_packfile(struct strbuf *name_buffer,
0e990530
JH
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;
58892711 348 int basename_len = name_buffer->len;
0e990530
JH
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
58892711 358 strbuf_addf(name_buffer, "%s.pack", sha1_to_hex(sha1));
0e990530 359
58892711 360 if (rename(pack_tmp_name, name_buffer->buf))
0e990530
JH
361 die_errno("unable to rename temporary pack file");
362
58892711
SH
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))
0e990530
JH
367 die_errno("unable to rename temporary index file");
368
58892711 369 strbuf_setlen(name_buffer, basename_len);
7cc8f971 370
0e990530
JH
371 free((void *)idx_tmp_name);
372}