create_tmp_packfile(): a helper function
[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;
16 return hashcmp(a->sha1, b->sha1);
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
JH
46 int nr_objects, const struct pack_idx_option *opts,
47 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;
9126f009 54 git_SHA_CTX ctx;
aa7e44bf
GB
55 uint32_t index_version;
56
57 if (nr_objects) {
58 sorted_by_sha = objects;
59 list = sorted_by_sha;
60 last = sorted_by_sha + nr_objects;
61 for (i = 0; i < nr_objects; ++i) {
62 if (objects[i]->offset > last_obj_offset)
63 last_obj_offset = objects[i]->offset;
64 }
65 qsort(sorted_by_sha, nr_objects, sizeof(sorted_by_sha[0]),
66 sha1_compare);
67 }
68 else
69 sorted_by_sha = list = last = NULL;
70
e337a04d
JH
71 if (opts->flags & WRITE_IDX_VERIFY) {
72 assert(index_name);
73 f = sha1fd_check(index_name);
aa7e44bf 74 } else {
e337a04d
JH
75 if (!index_name) {
76 static char tmpfile[PATH_MAX];
77 fd = odb_mkstemp(tmpfile, sizeof(tmpfile), "pack/tmp_idx_XXXXXX");
78 index_name = xstrdup(tmpfile);
79 } else {
80 unlink(index_name);
81 fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
82 }
83 if (fd < 0)
84 die_errno("unable to create '%s'", index_name);
85 f = sha1fd(fd, index_name);
aa7e44bf 86 }
aa7e44bf
GB
87
88 /* if last object's offset is >= 2^31 we should use index V2 */
fb956c1f 89 index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
aa7e44bf
GB
90
91 /* index versions 2 and above need a header */
92 if (index_version >= 2) {
93 struct pack_idx_header hdr;
94 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
95 hdr.idx_version = htonl(index_version);
96 sha1write(f, &hdr, sizeof(hdr));
97 }
98
99 /*
100 * Write the first-level table (the list is sorted,
101 * but we use a 256-entry lookup to be able to avoid
102 * having to do eight extra binary search iterations).
103 */
104 for (i = 0; i < 256; i++) {
105 struct pack_idx_entry **next = list;
106 while (next < last) {
107 struct pack_idx_entry *obj = *next;
108 if (obj->sha1[0] != i)
109 break;
110 next++;
111 }
112 array[i] = htonl(next - sorted_by_sha);
113 list = next;
114 }
115 sha1write(f, array, 256 * 4);
116
117 /* compute the SHA1 hash of sorted object names. */
9126f009 118 git_SHA1_Init(&ctx);
aa7e44bf
GB
119
120 /*
121 * Write the actual SHA1 entries..
122 */
123 list = sorted_by_sha;
124 for (i = 0; i < nr_objects; i++) {
125 struct pack_idx_entry *obj = *list++;
126 if (index_version < 2) {
127 uint32_t offset = htonl(obj->offset);
128 sha1write(f, &offset, 4);
129 }
130 sha1write(f, obj->sha1, 20);
9126f009 131 git_SHA1_Update(&ctx, obj->sha1, 20);
aa7e44bf
GB
132 }
133
134 if (index_version >= 2) {
135 unsigned int nr_large_offset = 0;
136
137 /* write the crc32 table */
138 list = sorted_by_sha;
139 for (i = 0; i < nr_objects; i++) {
140 struct pack_idx_entry *obj = *list++;
141 uint32_t crc32_val = htonl(obj->crc32);
142 sha1write(f, &crc32_val, 4);
143 }
144
145 /* write the 32-bit offset table */
146 list = sorted_by_sha;
147 for (i = 0; i < nr_objects; i++) {
148 struct pack_idx_entry *obj = *list++;
fb956c1f
JH
149 uint32_t offset;
150
151 offset = (need_large_offset(obj->offset, opts)
152 ? (0x80000000 | nr_large_offset++)
153 : obj->offset);
aa7e44bf
GB
154 offset = htonl(offset);
155 sha1write(f, &offset, 4);
156 }
157
158 /* write the large offset table */
159 list = sorted_by_sha;
160 while (nr_large_offset) {
161 struct pack_idx_entry *obj = *list++;
162 uint64_t offset = obj->offset;
fb956c1f
JH
163 uint32_t split[2];
164
165 if (!need_large_offset(offset, opts))
166 continue;
167 split[0] = htonl(offset >> 32);
168 split[1] = htonl(offset & 0xffffffff);
169 sha1write(f, split, 8);
170 nr_large_offset--;
aa7e44bf
GB
171 }
172 }
173
174 sha1write(f, sha1, 20);
e337a04d
JH
175 sha1close(f, NULL, ((opts->flags & WRITE_IDX_VERIFY)
176 ? CSUM_CLOSE : CSUM_FSYNC));
9126f009 177 git_SHA1_Final(sha1, &ctx);
aa7e44bf
GB
178 return index_name;
179}
8b0eca7c 180
c0ad4657
JH
181off_t write_pack_header(struct sha1file *f, uint32_t nr_entries)
182{
183 struct pack_header hdr;
184
185 hdr.hdr_signature = htonl(PACK_SIGNATURE);
186 hdr.hdr_version = htonl(PACK_VERSION);
187 hdr.hdr_entries = htonl(nr_entries);
188 if (sha1write(f, &hdr, sizeof(hdr)))
189 return 0;
190 return sizeof(hdr);
191}
192
abeb40e5
NP
193/*
194 * Update pack header with object_count and compute new SHA1 for pack data
195 * associated to pack_fd, and write that SHA1 at the end. That new SHA1
196 * is also returned in new_pack_sha1.
197 *
198 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
199 * (without the header update) is computed and validated against the
200 * one provided in partial_pack_sha1. The validation is performed at
201 * partial_pack_offset bytes in the pack file. The SHA1 of the remaining
202 * data (i.e. from partial_pack_offset to the end) is then computed and
203 * returned in partial_pack_sha1.
204 *
205 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
206 * partial_pack_sha1 can refer to the same buffer if the caller is not
207 * interested in the resulting SHA1 of pack data above partial_pack_offset.
208 */
8b0eca7c 209void fixup_pack_header_footer(int pack_fd,
abeb40e5 210 unsigned char *new_pack_sha1,
8b0eca7c 211 const char *pack_name,
abeb40e5
NP
212 uint32_t object_count,
213 unsigned char *partial_pack_sha1,
214 off_t partial_pack_offset)
8b0eca7c 215{
d35825da 216 int aligned_sz, buf_sz = 8 * 1024;
9126f009 217 git_SHA_CTX old_sha1_ctx, new_sha1_ctx;
8b0eca7c
DH
218 struct pack_header hdr;
219 char *buf;
220
9126f009
NP
221 git_SHA1_Init(&old_sha1_ctx);
222 git_SHA1_Init(&new_sha1_ctx);
abeb40e5 223
8b0eca7c 224 if (lseek(pack_fd, 0, SEEK_SET) != 0)
d824cbba 225 die_errno("Failed seeking to start of '%s'", pack_name);
8b0eca7c 226 if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
d824cbba 227 die_errno("Unable to reread header of '%s'", pack_name);
8b0eca7c 228 if (lseek(pack_fd, 0, SEEK_SET) != 0)
d824cbba 229 die_errno("Failed seeking to start of '%s'", pack_name);
9126f009 230 git_SHA1_Update(&old_sha1_ctx, &hdr, sizeof(hdr));
8b0eca7c 231 hdr.hdr_entries = htonl(object_count);
9126f009 232 git_SHA1_Update(&new_sha1_ctx, &hdr, sizeof(hdr));
8b0eca7c 233 write_or_die(pack_fd, &hdr, sizeof(hdr));
abeb40e5 234 partial_pack_offset -= sizeof(hdr);
8b0eca7c
DH
235
236 buf = xmalloc(buf_sz);
d35825da 237 aligned_sz = buf_sz - sizeof(hdr);
8b0eca7c 238 for (;;) {
abeb40e5 239 ssize_t m, n;
d35825da
NP
240 m = (partial_pack_sha1 && partial_pack_offset < aligned_sz) ?
241 partial_pack_offset : aligned_sz;
abeb40e5 242 n = xread(pack_fd, buf, m);
8b0eca7c
DH
243 if (!n)
244 break;
245 if (n < 0)
d824cbba 246 die_errno("Failed to checksum '%s'", pack_name);
9126f009 247 git_SHA1_Update(&new_sha1_ctx, buf, n);
abeb40e5 248
d35825da
NP
249 aligned_sz -= n;
250 if (!aligned_sz)
251 aligned_sz = buf_sz;
252
abeb40e5
NP
253 if (!partial_pack_sha1)
254 continue;
255
9126f009 256 git_SHA1_Update(&old_sha1_ctx, buf, n);
abeb40e5
NP
257 partial_pack_offset -= n;
258 if (partial_pack_offset == 0) {
259 unsigned char sha1[20];
9126f009 260 git_SHA1_Final(sha1, &old_sha1_ctx);
abeb40e5
NP
261 if (hashcmp(sha1, partial_pack_sha1) != 0)
262 die("Unexpected checksum for %s "
263 "(disk corruption?)", pack_name);
264 /*
265 * Now let's compute the SHA1 of the remainder of the
266 * pack, which also means making partial_pack_offset
267 * big enough not to matter anymore.
268 */
9126f009 269 git_SHA1_Init(&old_sha1_ctx);
abeb40e5
NP
270 partial_pack_offset = ~partial_pack_offset;
271 partial_pack_offset -= MSB(partial_pack_offset, 1);
272 }
8b0eca7c
DH
273 }
274 free(buf);
275
abeb40e5 276 if (partial_pack_sha1)
9126f009
NP
277 git_SHA1_Final(partial_pack_sha1, &old_sha1_ctx);
278 git_SHA1_Final(new_pack_sha1, &new_sha1_ctx);
abeb40e5 279 write_or_die(pack_fd, new_pack_sha1, 20);
0c68d386 280 fsync_or_die(pack_fd, pack_name);
8b0eca7c 281}
106764e6
SP
282
283char *index_pack_lockfile(int ip_out)
284{
106764e6
SP
285 char packname[46];
286
287 /*
6e180cdc 288 * The first thing we expect from index-pack's output
106764e6
SP
289 * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
290 * %40s is the newly created pack SHA1 name. In the "keep"
291 * case, we need it to remove the corresponding .keep file
292 * later on. If we don't get that then tough luck with it.
293 */
c697ad14
HO
294 if (read_in_full(ip_out, packname, 46) == 46 && packname[45] == '\n' &&
295 memcmp(packname, "keep\t", 5) == 0) {
106764e6
SP
296 char path[PATH_MAX];
297 packname[45] = 0;
298 snprintf(path, sizeof(path), "%s/pack/pack-%s.keep",
299 get_object_directory(), packname + 5);
300 return xstrdup(path);
301 }
302 return NULL;
303}
f965c525
NP
304
305/*
306 * The per-object header is a pretty dense thing, which is
307 * - first byte: low four bits are "size", then three bits of "type",
308 * and the high bit is "size continues".
309 * - each byte afterwards: low seven bits are size continuation,
310 * with the high bit being "size continues"
311 */
312int encode_in_pack_object_header(enum object_type type, uintmax_t size, unsigned char *hdr)
313{
314 int n = 1;
315 unsigned char c;
316
317 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
318 die("bad type %d", type);
319
320 c = (type << 4) | (size & 15);
321 size >>= 4;
322 while (size) {
323 *hdr++ = c | 0x80;
324 c = size & 0x7f;
325 size >>= 7;
326 n++;
327 }
328 *hdr = c;
329 return n;
330}
cdf9db3c
JH
331
332struct sha1file *create_tmp_packfile(char **pack_tmp_name)
333{
334 char tmpname[PATH_MAX];
335 int fd;
336
337 fd = odb_mkstemp(tmpname, sizeof(tmpname), "pack/tmp_pack_XXXXXX");
338 *pack_tmp_name = xstrdup(tmpname);
339 return sha1fd(fd, *pack_tmp_name);
340}