index-pack: --verify
[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
19/*
20 * On entry *sha1 contains the pack content SHA1 hash, on exit it is
21 * the SHA1 hash of sorted object names. The objects array passed in
22 * will be sorted by SHA1 on exit.
23 */
3bb72562 24const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects,
ebcfb379
JH
25 int nr_objects, const struct pack_idx_option *opts,
26 unsigned char *sha1)
aa7e44bf
GB
27{
28 struct sha1file *f;
29 struct pack_idx_entry **sorted_by_sha, **list, **last;
30 off_t last_obj_offset = 0;
31 uint32_t array[256];
32 int i, fd;
9126f009 33 git_SHA_CTX ctx;
aa7e44bf
GB
34 uint32_t index_version;
35
36 if (nr_objects) {
37 sorted_by_sha = objects;
38 list = sorted_by_sha;
39 last = sorted_by_sha + nr_objects;
40 for (i = 0; i < nr_objects; ++i) {
41 if (objects[i]->offset > last_obj_offset)
42 last_obj_offset = objects[i]->offset;
43 }
44 qsort(sorted_by_sha, nr_objects, sizeof(sorted_by_sha[0]),
45 sha1_compare);
46 }
47 else
48 sorted_by_sha = list = last = NULL;
49
e337a04d
JH
50 if (opts->flags & WRITE_IDX_VERIFY) {
51 assert(index_name);
52 f = sha1fd_check(index_name);
aa7e44bf 53 } else {
e337a04d
JH
54 if (!index_name) {
55 static char tmpfile[PATH_MAX];
56 fd = odb_mkstemp(tmpfile, sizeof(tmpfile), "pack/tmp_idx_XXXXXX");
57 index_name = xstrdup(tmpfile);
58 } else {
59 unlink(index_name);
60 fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
61 }
62 if (fd < 0)
63 die_errno("unable to create '%s'", index_name);
64 f = sha1fd(fd, index_name);
aa7e44bf 65 }
aa7e44bf
GB
66
67 /* if last object's offset is >= 2^31 we should use index V2 */
ebcfb379 68 index_version = (last_obj_offset >> 31) ? 2 : opts->version;
aa7e44bf
GB
69
70 /* index versions 2 and above need a header */
71 if (index_version >= 2) {
72 struct pack_idx_header hdr;
73 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
74 hdr.idx_version = htonl(index_version);
75 sha1write(f, &hdr, sizeof(hdr));
76 }
77
78 /*
79 * Write the first-level table (the list is sorted,
80 * but we use a 256-entry lookup to be able to avoid
81 * having to do eight extra binary search iterations).
82 */
83 for (i = 0; i < 256; i++) {
84 struct pack_idx_entry **next = list;
85 while (next < last) {
86 struct pack_idx_entry *obj = *next;
87 if (obj->sha1[0] != i)
88 break;
89 next++;
90 }
91 array[i] = htonl(next - sorted_by_sha);
92 list = next;
93 }
94 sha1write(f, array, 256 * 4);
95
96 /* compute the SHA1 hash of sorted object names. */
9126f009 97 git_SHA1_Init(&ctx);
aa7e44bf
GB
98
99 /*
100 * Write the actual SHA1 entries..
101 */
102 list = sorted_by_sha;
103 for (i = 0; i < nr_objects; i++) {
104 struct pack_idx_entry *obj = *list++;
105 if (index_version < 2) {
106 uint32_t offset = htonl(obj->offset);
107 sha1write(f, &offset, 4);
108 }
109 sha1write(f, obj->sha1, 20);
9126f009 110 git_SHA1_Update(&ctx, obj->sha1, 20);
aa7e44bf
GB
111 }
112
113 if (index_version >= 2) {
114 unsigned int nr_large_offset = 0;
115
116 /* write the crc32 table */
117 list = sorted_by_sha;
118 for (i = 0; i < nr_objects; i++) {
119 struct pack_idx_entry *obj = *list++;
120 uint32_t crc32_val = htonl(obj->crc32);
121 sha1write(f, &crc32_val, 4);
122 }
123
124 /* write the 32-bit offset table */
125 list = sorted_by_sha;
126 for (i = 0; i < nr_objects; i++) {
127 struct pack_idx_entry *obj = *list++;
ebcfb379 128 uint32_t offset = (obj->offset <= opts->off32_limit) ?
aa7e44bf
GB
129 obj->offset : (0x80000000 | nr_large_offset++);
130 offset = htonl(offset);
131 sha1write(f, &offset, 4);
132 }
133
134 /* write the large offset table */
135 list = sorted_by_sha;
136 while (nr_large_offset) {
137 struct pack_idx_entry *obj = *list++;
138 uint64_t offset = obj->offset;
ebcfb379 139 if (offset > opts->off32_limit) {
aa7e44bf
GB
140 uint32_t split[2];
141 split[0] = htonl(offset >> 32);
142 split[1] = htonl(offset & 0xffffffff);
143 sha1write(f, split, 8);
144 nr_large_offset--;
145 }
146 }
147 }
148
149 sha1write(f, sha1, 20);
e337a04d
JH
150 sha1close(f, NULL, ((opts->flags & WRITE_IDX_VERIFY)
151 ? CSUM_CLOSE : CSUM_FSYNC));
9126f009 152 git_SHA1_Final(sha1, &ctx);
aa7e44bf
GB
153 return index_name;
154}
8b0eca7c 155
abeb40e5
NP
156/*
157 * Update pack header with object_count and compute new SHA1 for pack data
158 * associated to pack_fd, and write that SHA1 at the end. That new SHA1
159 * is also returned in new_pack_sha1.
160 *
161 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
162 * (without the header update) is computed and validated against the
163 * one provided in partial_pack_sha1. The validation is performed at
164 * partial_pack_offset bytes in the pack file. The SHA1 of the remaining
165 * data (i.e. from partial_pack_offset to the end) is then computed and
166 * returned in partial_pack_sha1.
167 *
168 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
169 * partial_pack_sha1 can refer to the same buffer if the caller is not
170 * interested in the resulting SHA1 of pack data above partial_pack_offset.
171 */
8b0eca7c 172void fixup_pack_header_footer(int pack_fd,
abeb40e5 173 unsigned char *new_pack_sha1,
8b0eca7c 174 const char *pack_name,
abeb40e5
NP
175 uint32_t object_count,
176 unsigned char *partial_pack_sha1,
177 off_t partial_pack_offset)
8b0eca7c 178{
d35825da 179 int aligned_sz, buf_sz = 8 * 1024;
9126f009 180 git_SHA_CTX old_sha1_ctx, new_sha1_ctx;
8b0eca7c
DH
181 struct pack_header hdr;
182 char *buf;
183
9126f009
NP
184 git_SHA1_Init(&old_sha1_ctx);
185 git_SHA1_Init(&new_sha1_ctx);
abeb40e5 186
8b0eca7c 187 if (lseek(pack_fd, 0, SEEK_SET) != 0)
d824cbba 188 die_errno("Failed seeking to start of '%s'", pack_name);
8b0eca7c 189 if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
d824cbba 190 die_errno("Unable to reread header of '%s'", pack_name);
8b0eca7c 191 if (lseek(pack_fd, 0, SEEK_SET) != 0)
d824cbba 192 die_errno("Failed seeking to start of '%s'", pack_name);
9126f009 193 git_SHA1_Update(&old_sha1_ctx, &hdr, sizeof(hdr));
8b0eca7c 194 hdr.hdr_entries = htonl(object_count);
9126f009 195 git_SHA1_Update(&new_sha1_ctx, &hdr, sizeof(hdr));
8b0eca7c 196 write_or_die(pack_fd, &hdr, sizeof(hdr));
abeb40e5 197 partial_pack_offset -= sizeof(hdr);
8b0eca7c
DH
198
199 buf = xmalloc(buf_sz);
d35825da 200 aligned_sz = buf_sz - sizeof(hdr);
8b0eca7c 201 for (;;) {
abeb40e5 202 ssize_t m, n;
d35825da
NP
203 m = (partial_pack_sha1 && partial_pack_offset < aligned_sz) ?
204 partial_pack_offset : aligned_sz;
abeb40e5 205 n = xread(pack_fd, buf, m);
8b0eca7c
DH
206 if (!n)
207 break;
208 if (n < 0)
d824cbba 209 die_errno("Failed to checksum '%s'", pack_name);
9126f009 210 git_SHA1_Update(&new_sha1_ctx, buf, n);
abeb40e5 211
d35825da
NP
212 aligned_sz -= n;
213 if (!aligned_sz)
214 aligned_sz = buf_sz;
215
abeb40e5
NP
216 if (!partial_pack_sha1)
217 continue;
218
9126f009 219 git_SHA1_Update(&old_sha1_ctx, buf, n);
abeb40e5
NP
220 partial_pack_offset -= n;
221 if (partial_pack_offset == 0) {
222 unsigned char sha1[20];
9126f009 223 git_SHA1_Final(sha1, &old_sha1_ctx);
abeb40e5
NP
224 if (hashcmp(sha1, partial_pack_sha1) != 0)
225 die("Unexpected checksum for %s "
226 "(disk corruption?)", pack_name);
227 /*
228 * Now let's compute the SHA1 of the remainder of the
229 * pack, which also means making partial_pack_offset
230 * big enough not to matter anymore.
231 */
9126f009 232 git_SHA1_Init(&old_sha1_ctx);
abeb40e5
NP
233 partial_pack_offset = ~partial_pack_offset;
234 partial_pack_offset -= MSB(partial_pack_offset, 1);
235 }
8b0eca7c
DH
236 }
237 free(buf);
238
abeb40e5 239 if (partial_pack_sha1)
9126f009
NP
240 git_SHA1_Final(partial_pack_sha1, &old_sha1_ctx);
241 git_SHA1_Final(new_pack_sha1, &new_sha1_ctx);
abeb40e5 242 write_or_die(pack_fd, new_pack_sha1, 20);
0c68d386 243 fsync_or_die(pack_fd, pack_name);
8b0eca7c 244}
106764e6
SP
245
246char *index_pack_lockfile(int ip_out)
247{
106764e6
SP
248 char packname[46];
249
250 /*
6e180cdc 251 * The first thing we expect from index-pack's output
106764e6
SP
252 * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
253 * %40s is the newly created pack SHA1 name. In the "keep"
254 * case, we need it to remove the corresponding .keep file
255 * later on. If we don't get that then tough luck with it.
256 */
c697ad14
HO
257 if (read_in_full(ip_out, packname, 46) == 46 && packname[45] == '\n' &&
258 memcmp(packname, "keep\t", 5) == 0) {
106764e6
SP
259 char path[PATH_MAX];
260 packname[45] = 0;
261 snprintf(path, sizeof(path), "%s/pack/pack-%s.keep",
262 get_object_directory(), packname + 5);
263 return xstrdup(path);
264 }
265 return NULL;
266}
f965c525
NP
267
268/*
269 * The per-object header is a pretty dense thing, which is
270 * - first byte: low four bits are "size", then three bits of "type",
271 * and the high bit is "size continues".
272 * - each byte afterwards: low seven bits are size continuation,
273 * with the high bit being "size continues"
274 */
275int encode_in_pack_object_header(enum object_type type, uintmax_t size, unsigned char *hdr)
276{
277 int n = 1;
278 unsigned char c;
279
280 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
281 die("bad type %d", type);
282
283 c = (type << 4) | (size & 15);
284 size >>= 4;
285 while (size) {
286 *hdr++ = c | 0x80;
287 c = size & 0x7f;
288 size >>= 7;
289 n++;
290 }
291 *hdr = c;
292 return n;
293}