Documentation/technical: describe remembering renames optimization
[git/debian.git] / pack-write.c
blobe9bb3fd949b36d1ea1e5a61fb6d08592047223ff
1 #include "cache.h"
2 #include "pack.h"
3 #include "csum-file.h"
4 #include "remote.h"
6 void reset_pack_idx_option(struct pack_idx_option *opts)
8 memset(opts, 0, sizeof(*opts));
9 opts->version = 2;
10 opts->off32_limit = 0x7fffffff;
13 static int sha1_compare(const void *_a, const void *_b)
15 struct pack_idx_entry *a = *(struct pack_idx_entry **)_a;
16 struct pack_idx_entry *b = *(struct pack_idx_entry **)_b;
17 return oidcmp(&a->oid, &b->oid);
20 static int cmp_uint32(const void *a_, const void *b_)
22 uint32_t a = *((uint32_t *)a_);
23 uint32_t b = *((uint32_t *)b_);
25 return (a < b) ? -1 : (a != b);
28 static int need_large_offset(off_t offset, const struct pack_idx_option *opts)
30 uint32_t ofsval;
32 if ((offset >> 31) || (opts->off32_limit < offset))
33 return 1;
34 if (!opts->anomaly_nr)
35 return 0;
36 ofsval = offset;
37 return !!bsearch(&ofsval, opts->anomaly, opts->anomaly_nr,
38 sizeof(ofsval), cmp_uint32);
42 * The *sha1 contains the pack content SHA1 hash.
43 * The objects array passed in will be sorted by SHA1 on exit.
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)
49 struct hashfile *f;
50 struct pack_idx_entry **sorted_by_sha, **list, **last;
51 off_t last_obj_offset = 0;
52 int i, fd;
53 uint32_t index_version;
55 if (nr_objects) {
56 sorted_by_sha = objects;
57 list = sorted_by_sha;
58 last = sorted_by_sha + nr_objects;
59 for (i = 0; i < nr_objects; ++i) {
60 if (objects[i]->offset > last_obj_offset)
61 last_obj_offset = objects[i]->offset;
63 QSORT(sorted_by_sha, nr_objects, sha1_compare);
65 else
66 sorted_by_sha = list = last = NULL;
68 if (opts->flags & WRITE_IDX_VERIFY) {
69 assert(index_name);
70 f = hashfd_check(index_name);
71 } else {
72 if (!index_name) {
73 struct strbuf tmp_file = STRBUF_INIT;
74 fd = odb_mkstemp(&tmp_file, "pack/tmp_idx_XXXXXX");
75 index_name = strbuf_detach(&tmp_file, NULL);
76 } else {
77 unlink(index_name);
78 fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
79 if (fd < 0)
80 die_errno("unable to create '%s'", index_name);
82 f = hashfd(fd, index_name);
85 /* if last object's offset is >= 2^31 we should use index V2 */
86 index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
88 /* index versions 2 and above need a header */
89 if (index_version >= 2) {
90 struct pack_idx_header hdr;
91 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
92 hdr.idx_version = htonl(index_version);
93 hashwrite(f, &hdr, sizeof(hdr));
97 * Write the first-level table (the list is sorted,
98 * but we use a 256-entry lookup to be able to avoid
99 * having to do eight extra binary search iterations).
101 for (i = 0; i < 256; i++) {
102 struct pack_idx_entry **next = list;
103 while (next < last) {
104 struct pack_idx_entry *obj = *next;
105 if (obj->oid.hash[0] != i)
106 break;
107 next++;
109 hashwrite_be32(f, next - sorted_by_sha);
110 list = next;
114 * Write the actual SHA1 entries..
116 list = sorted_by_sha;
117 for (i = 0; i < nr_objects; i++) {
118 struct pack_idx_entry *obj = *list++;
119 if (index_version < 2)
120 hashwrite_be32(f, obj->offset);
121 hashwrite(f, obj->oid.hash, the_hash_algo->rawsz);
122 if ((opts->flags & WRITE_IDX_STRICT) &&
123 (i && oideq(&list[-2]->oid, &obj->oid)))
124 die("The same object %s appears twice in the pack",
125 oid_to_hex(&obj->oid));
128 if (index_version >= 2) {
129 unsigned int nr_large_offset = 0;
131 /* write the crc32 table */
132 list = sorted_by_sha;
133 for (i = 0; i < nr_objects; i++) {
134 struct pack_idx_entry *obj = *list++;
135 hashwrite_be32(f, obj->crc32);
138 /* write the 32-bit offset table */
139 list = sorted_by_sha;
140 for (i = 0; i < nr_objects; i++) {
141 struct pack_idx_entry *obj = *list++;
142 uint32_t offset;
144 offset = (need_large_offset(obj->offset, opts)
145 ? (0x80000000 | nr_large_offset++)
146 : obj->offset);
147 hashwrite_be32(f, offset);
150 /* write the large offset table */
151 list = sorted_by_sha;
152 while (nr_large_offset) {
153 struct pack_idx_entry *obj = *list++;
154 uint64_t offset = obj->offset;
156 if (!need_large_offset(offset, opts))
157 continue;
158 hashwrite_be64(f, offset);
159 nr_large_offset--;
163 hashwrite(f, sha1, the_hash_algo->rawsz);
164 finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_CLOSE |
165 ((opts->flags & WRITE_IDX_VERIFY)
166 ? 0 : CSUM_FSYNC));
167 return index_name;
170 off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)
172 struct pack_header hdr;
174 hdr.hdr_signature = htonl(PACK_SIGNATURE);
175 hdr.hdr_version = htonl(PACK_VERSION);
176 hdr.hdr_entries = htonl(nr_entries);
177 hashwrite(f, &hdr, sizeof(hdr));
178 return sizeof(hdr);
182 * Update pack header with object_count and compute new SHA1 for pack data
183 * associated to pack_fd, and write that SHA1 at the end. That new SHA1
184 * is also returned in new_pack_sha1.
186 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
187 * (without the header update) is computed and validated against the
188 * one provided in partial_pack_sha1. The validation is performed at
189 * partial_pack_offset bytes in the pack file. The SHA1 of the remaining
190 * data (i.e. from partial_pack_offset to the end) is then computed and
191 * returned in partial_pack_sha1.
193 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
194 * partial_pack_sha1 can refer to the same buffer if the caller is not
195 * interested in the resulting SHA1 of pack data above partial_pack_offset.
197 void fixup_pack_header_footer(int pack_fd,
198 unsigned char *new_pack_hash,
199 const char *pack_name,
200 uint32_t object_count,
201 unsigned char *partial_pack_hash,
202 off_t partial_pack_offset)
204 int aligned_sz, buf_sz = 8 * 1024;
205 git_hash_ctx old_hash_ctx, new_hash_ctx;
206 struct pack_header hdr;
207 char *buf;
208 ssize_t read_result;
210 the_hash_algo->init_fn(&old_hash_ctx);
211 the_hash_algo->init_fn(&new_hash_ctx);
213 if (lseek(pack_fd, 0, SEEK_SET) != 0)
214 die_errno("Failed seeking to start of '%s'", pack_name);
215 read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));
216 if (read_result < 0)
217 die_errno("Unable to reread header of '%s'", pack_name);
218 else if (read_result != sizeof(hdr))
219 die_errno("Unexpected short read for header of '%s'",
220 pack_name);
221 if (lseek(pack_fd, 0, SEEK_SET) != 0)
222 die_errno("Failed seeking to start of '%s'", pack_name);
223 the_hash_algo->update_fn(&old_hash_ctx, &hdr, sizeof(hdr));
224 hdr.hdr_entries = htonl(object_count);
225 the_hash_algo->update_fn(&new_hash_ctx, &hdr, sizeof(hdr));
226 write_or_die(pack_fd, &hdr, sizeof(hdr));
227 partial_pack_offset -= sizeof(hdr);
229 buf = xmalloc(buf_sz);
230 aligned_sz = buf_sz - sizeof(hdr);
231 for (;;) {
232 ssize_t m, n;
233 m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?
234 partial_pack_offset : aligned_sz;
235 n = xread(pack_fd, buf, m);
236 if (!n)
237 break;
238 if (n < 0)
239 die_errno("Failed to checksum '%s'", pack_name);
240 the_hash_algo->update_fn(&new_hash_ctx, buf, n);
242 aligned_sz -= n;
243 if (!aligned_sz)
244 aligned_sz = buf_sz;
246 if (!partial_pack_hash)
247 continue;
249 the_hash_algo->update_fn(&old_hash_ctx, buf, n);
250 partial_pack_offset -= n;
251 if (partial_pack_offset == 0) {
252 unsigned char hash[GIT_MAX_RAWSZ];
253 the_hash_algo->final_fn(hash, &old_hash_ctx);
254 if (!hasheq(hash, partial_pack_hash))
255 die("Unexpected checksum for %s "
256 "(disk corruption?)", pack_name);
258 * Now let's compute the SHA1 of the remainder of the
259 * pack, which also means making partial_pack_offset
260 * big enough not to matter anymore.
262 the_hash_algo->init_fn(&old_hash_ctx);
263 partial_pack_offset = ~partial_pack_offset;
264 partial_pack_offset -= MSB(partial_pack_offset, 1);
267 free(buf);
269 if (partial_pack_hash)
270 the_hash_algo->final_fn(partial_pack_hash, &old_hash_ctx);
271 the_hash_algo->final_fn(new_pack_hash, &new_hash_ctx);
272 write_or_die(pack_fd, new_pack_hash, the_hash_algo->rawsz);
273 fsync_or_die(pack_fd, pack_name);
276 char *index_pack_lockfile(int ip_out)
278 char packname[GIT_MAX_HEXSZ + 6];
279 const int len = the_hash_algo->hexsz + 6;
282 * The first thing we expect from index-pack's output
283 * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
284 * %40s is the newly created pack SHA1 name. In the "keep"
285 * case, we need it to remove the corresponding .keep file
286 * later on. If we don't get that then tough luck with it.
288 if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
289 const char *name;
290 packname[len-1] = 0;
291 if (skip_prefix(packname, "keep\t", &name))
292 return xstrfmt("%s/pack/pack-%s.keep",
293 get_object_directory(), name);
295 return NULL;
299 * The per-object header is a pretty dense thing, which is
300 * - first byte: low four bits are "size", then three bits of "type",
301 * and the high bit is "size continues".
302 * - each byte afterwards: low seven bits are size continuation,
303 * with the high bit being "size continues"
305 int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
306 enum object_type type, uintmax_t size)
308 int n = 1;
309 unsigned char c;
311 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
312 die("bad type %d", type);
314 c = (type << 4) | (size & 15);
315 size >>= 4;
316 while (size) {
317 if (n == hdr_len)
318 die("object size is too enormous to format");
319 *hdr++ = c | 0x80;
320 c = size & 0x7f;
321 size >>= 7;
322 n++;
324 *hdr = c;
325 return n;
328 struct hashfile *create_tmp_packfile(char **pack_tmp_name)
330 struct strbuf tmpname = STRBUF_INIT;
331 int fd;
333 fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
334 *pack_tmp_name = strbuf_detach(&tmpname, NULL);
335 return hashfd(fd, *pack_tmp_name);
338 void finish_tmp_packfile(struct strbuf *name_buffer,
339 const char *pack_tmp_name,
340 struct pack_idx_entry **written_list,
341 uint32_t nr_written,
342 struct pack_idx_option *pack_idx_opts,
343 unsigned char hash[])
345 const char *idx_tmp_name;
346 int basename_len = name_buffer->len;
348 if (adjust_shared_perm(pack_tmp_name))
349 die_errno("unable to make temporary pack file readable");
351 idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
352 pack_idx_opts, hash);
353 if (adjust_shared_perm(idx_tmp_name))
354 die_errno("unable to make temporary index file readable");
356 strbuf_addf(name_buffer, "%s.pack", hash_to_hex(hash));
358 if (rename(pack_tmp_name, name_buffer->buf))
359 die_errno("unable to rename temporary pack file");
361 strbuf_setlen(name_buffer, basename_len);
363 strbuf_addf(name_buffer, "%s.idx", hash_to_hex(hash));
364 if (rename(idx_tmp_name, name_buffer->buf))
365 die_errno("unable to rename temporary index file");
367 strbuf_setlen(name_buffer, basename_len);
369 free((void *)idx_tmp_name);
372 void write_promisor_file(const char *promisor_name, struct ref **sought, int nr_sought)
374 int i, err;
375 FILE *output = xfopen(promisor_name, "w");
377 for (i = 0; i < nr_sought; i++)
378 fprintf(output, "%s %s\n", oid_to_hex(&sought[i]->old_oid),
379 sought[i]->name);
381 err = ferror(output);
382 err |= fclose(output);
383 if (err)
384 die(_("could not write '%s' promisor file"), promisor_name);