]>
Commit | Line | Data |
---|---|---|
8b0eca7c DH |
1 | #include "cache.h" |
2 | #include "pack.h" | |
aa7e44bf GB |
3 | #include "csum-file.h" |
4 | ||
c0a5e2d4 | 5 | uint32_t pack_idx_default_version = 2; |
aa7e44bf GB |
6 | uint32_t pack_idx_off32_limit = 0x7fffffff; |
7 | ||
8 | static int sha1_compare(const void *_a, const void *_b) | |
9 | { | |
10 | struct pack_idx_entry *a = *(struct pack_idx_entry **)_a; | |
11 | struct pack_idx_entry *b = *(struct pack_idx_entry **)_b; | |
12 | return hashcmp(a->sha1, b->sha1); | |
13 | } | |
14 | ||
15 | /* | |
16 | * On entry *sha1 contains the pack content SHA1 hash, on exit it is | |
17 | * the SHA1 hash of sorted object names. The objects array passed in | |
18 | * will be sorted by SHA1 on exit. | |
19 | */ | |
3bb72562 LT |
20 | const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects, |
21 | int nr_objects, unsigned char *sha1) | |
aa7e44bf GB |
22 | { |
23 | struct sha1file *f; | |
24 | struct pack_idx_entry **sorted_by_sha, **list, **last; | |
25 | off_t last_obj_offset = 0; | |
26 | uint32_t array[256]; | |
27 | int i, fd; | |
9126f009 | 28 | git_SHA_CTX ctx; |
aa7e44bf GB |
29 | uint32_t index_version; |
30 | ||
31 | if (nr_objects) { | |
32 | sorted_by_sha = objects; | |
33 | list = sorted_by_sha; | |
34 | last = sorted_by_sha + nr_objects; | |
35 | for (i = 0; i < nr_objects; ++i) { | |
36 | if (objects[i]->offset > last_obj_offset) | |
37 | last_obj_offset = objects[i]->offset; | |
38 | } | |
39 | qsort(sorted_by_sha, nr_objects, sizeof(sorted_by_sha[0]), | |
40 | sha1_compare); | |
41 | } | |
42 | else | |
43 | sorted_by_sha = list = last = NULL; | |
44 | ||
45 | if (!index_name) { | |
46 | static char tmpfile[PATH_MAX]; | |
6e180cdc | 47 | fd = odb_mkstemp(tmpfile, sizeof(tmpfile), "pack/tmp_idx_XXXXXX"); |
aa7e44bf GB |
48 | index_name = xstrdup(tmpfile); |
49 | } else { | |
50 | unlink(index_name); | |
51 | fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600); | |
52 | } | |
53 | if (fd < 0) | |
d824cbba | 54 | die_errno("unable to create '%s'", index_name); |
aa7e44bf GB |
55 | f = sha1fd(fd, index_name); |
56 | ||
57 | /* if last object's offset is >= 2^31 we should use index V2 */ | |
58 | index_version = (last_obj_offset >> 31) ? 2 : pack_idx_default_version; | |
59 | ||
60 | /* index versions 2 and above need a header */ | |
61 | if (index_version >= 2) { | |
62 | struct pack_idx_header hdr; | |
63 | hdr.idx_signature = htonl(PACK_IDX_SIGNATURE); | |
64 | hdr.idx_version = htonl(index_version); | |
65 | sha1write(f, &hdr, sizeof(hdr)); | |
66 | } | |
67 | ||
68 | /* | |
69 | * Write the first-level table (the list is sorted, | |
70 | * but we use a 256-entry lookup to be able to avoid | |
71 | * having to do eight extra binary search iterations). | |
72 | */ | |
73 | for (i = 0; i < 256; i++) { | |
74 | struct pack_idx_entry **next = list; | |
75 | while (next < last) { | |
76 | struct pack_idx_entry *obj = *next; | |
77 | if (obj->sha1[0] != i) | |
78 | break; | |
79 | next++; | |
80 | } | |
81 | array[i] = htonl(next - sorted_by_sha); | |
82 | list = next; | |
83 | } | |
84 | sha1write(f, array, 256 * 4); | |
85 | ||
86 | /* compute the SHA1 hash of sorted object names. */ | |
9126f009 | 87 | git_SHA1_Init(&ctx); |
aa7e44bf GB |
88 | |
89 | /* | |
90 | * Write the actual SHA1 entries.. | |
91 | */ | |
92 | list = sorted_by_sha; | |
93 | for (i = 0; i < nr_objects; i++) { | |
94 | struct pack_idx_entry *obj = *list++; | |
95 | if (index_version < 2) { | |
96 | uint32_t offset = htonl(obj->offset); | |
97 | sha1write(f, &offset, 4); | |
98 | } | |
99 | sha1write(f, obj->sha1, 20); | |
9126f009 | 100 | git_SHA1_Update(&ctx, obj->sha1, 20); |
aa7e44bf GB |
101 | } |
102 | ||
103 | if (index_version >= 2) { | |
104 | unsigned int nr_large_offset = 0; | |
105 | ||
106 | /* write the crc32 table */ | |
107 | list = sorted_by_sha; | |
108 | for (i = 0; i < nr_objects; i++) { | |
109 | struct pack_idx_entry *obj = *list++; | |
110 | uint32_t crc32_val = htonl(obj->crc32); | |
111 | sha1write(f, &crc32_val, 4); | |
112 | } | |
113 | ||
114 | /* write the 32-bit offset table */ | |
115 | list = sorted_by_sha; | |
116 | for (i = 0; i < nr_objects; i++) { | |
117 | struct pack_idx_entry *obj = *list++; | |
118 | uint32_t offset = (obj->offset <= pack_idx_off32_limit) ? | |
119 | obj->offset : (0x80000000 | nr_large_offset++); | |
120 | offset = htonl(offset); | |
121 | sha1write(f, &offset, 4); | |
122 | } | |
123 | ||
124 | /* write the large offset table */ | |
125 | list = sorted_by_sha; | |
126 | while (nr_large_offset) { | |
127 | struct pack_idx_entry *obj = *list++; | |
128 | uint64_t offset = obj->offset; | |
129 | if (offset > pack_idx_off32_limit) { | |
130 | uint32_t split[2]; | |
131 | split[0] = htonl(offset >> 32); | |
132 | split[1] = htonl(offset & 0xffffffff); | |
133 | sha1write(f, split, 8); | |
134 | nr_large_offset--; | |
135 | } | |
136 | } | |
137 | } | |
138 | ||
139 | sha1write(f, sha1, 20); | |
4c81b03e | 140 | sha1close(f, NULL, CSUM_FSYNC); |
9126f009 | 141 | git_SHA1_Final(sha1, &ctx); |
aa7e44bf GB |
142 | return index_name; |
143 | } | |
8b0eca7c | 144 | |
abeb40e5 NP |
145 | /* |
146 | * Update pack header with object_count and compute new SHA1 for pack data | |
147 | * associated to pack_fd, and write that SHA1 at the end. That new SHA1 | |
148 | * is also returned in new_pack_sha1. | |
149 | * | |
150 | * If partial_pack_sha1 is non null, then the SHA1 of the existing pack | |
151 | * (without the header update) is computed and validated against the | |
152 | * one provided in partial_pack_sha1. The validation is performed at | |
153 | * partial_pack_offset bytes in the pack file. The SHA1 of the remaining | |
154 | * data (i.e. from partial_pack_offset to the end) is then computed and | |
155 | * returned in partial_pack_sha1. | |
156 | * | |
157 | * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and | |
158 | * partial_pack_sha1 can refer to the same buffer if the caller is not | |
159 | * interested in the resulting SHA1 of pack data above partial_pack_offset. | |
160 | */ | |
8b0eca7c | 161 | void fixup_pack_header_footer(int pack_fd, |
abeb40e5 | 162 | unsigned char *new_pack_sha1, |
8b0eca7c | 163 | const char *pack_name, |
abeb40e5 NP |
164 | uint32_t object_count, |
165 | unsigned char *partial_pack_sha1, | |
166 | off_t partial_pack_offset) | |
8b0eca7c | 167 | { |
d35825da | 168 | int aligned_sz, buf_sz = 8 * 1024; |
9126f009 | 169 | git_SHA_CTX old_sha1_ctx, new_sha1_ctx; |
8b0eca7c DH |
170 | struct pack_header hdr; |
171 | char *buf; | |
172 | ||
9126f009 NP |
173 | git_SHA1_Init(&old_sha1_ctx); |
174 | git_SHA1_Init(&new_sha1_ctx); | |
abeb40e5 | 175 | |
8b0eca7c | 176 | if (lseek(pack_fd, 0, SEEK_SET) != 0) |
d824cbba | 177 | die_errno("Failed seeking to start of '%s'", pack_name); |
8b0eca7c | 178 | if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
d824cbba | 179 | die_errno("Unable to reread header of '%s'", pack_name); |
8b0eca7c | 180 | if (lseek(pack_fd, 0, SEEK_SET) != 0) |
d824cbba | 181 | die_errno("Failed seeking to start of '%s'", pack_name); |
9126f009 | 182 | git_SHA1_Update(&old_sha1_ctx, &hdr, sizeof(hdr)); |
8b0eca7c | 183 | hdr.hdr_entries = htonl(object_count); |
9126f009 | 184 | git_SHA1_Update(&new_sha1_ctx, &hdr, sizeof(hdr)); |
8b0eca7c | 185 | write_or_die(pack_fd, &hdr, sizeof(hdr)); |
abeb40e5 | 186 | partial_pack_offset -= sizeof(hdr); |
8b0eca7c DH |
187 | |
188 | buf = xmalloc(buf_sz); | |
d35825da | 189 | aligned_sz = buf_sz - sizeof(hdr); |
8b0eca7c | 190 | for (;;) { |
abeb40e5 | 191 | ssize_t m, n; |
d35825da NP |
192 | m = (partial_pack_sha1 && partial_pack_offset < aligned_sz) ? |
193 | partial_pack_offset : aligned_sz; | |
abeb40e5 | 194 | n = xread(pack_fd, buf, m); |
8b0eca7c DH |
195 | if (!n) |
196 | break; | |
197 | if (n < 0) | |
d824cbba | 198 | die_errno("Failed to checksum '%s'", pack_name); |
9126f009 | 199 | git_SHA1_Update(&new_sha1_ctx, buf, n); |
abeb40e5 | 200 | |
d35825da NP |
201 | aligned_sz -= n; |
202 | if (!aligned_sz) | |
203 | aligned_sz = buf_sz; | |
204 | ||
abeb40e5 NP |
205 | if (!partial_pack_sha1) |
206 | continue; | |
207 | ||
9126f009 | 208 | git_SHA1_Update(&old_sha1_ctx, buf, n); |
abeb40e5 NP |
209 | partial_pack_offset -= n; |
210 | if (partial_pack_offset == 0) { | |
211 | unsigned char sha1[20]; | |
9126f009 | 212 | git_SHA1_Final(sha1, &old_sha1_ctx); |
abeb40e5 NP |
213 | if (hashcmp(sha1, partial_pack_sha1) != 0) |
214 | die("Unexpected checksum for %s " | |
215 | "(disk corruption?)", pack_name); | |
216 | /* | |
217 | * Now let's compute the SHA1 of the remainder of the | |
218 | * pack, which also means making partial_pack_offset | |
219 | * big enough not to matter anymore. | |
220 | */ | |
9126f009 | 221 | git_SHA1_Init(&old_sha1_ctx); |
abeb40e5 NP |
222 | partial_pack_offset = ~partial_pack_offset; |
223 | partial_pack_offset -= MSB(partial_pack_offset, 1); | |
224 | } | |
8b0eca7c DH |
225 | } |
226 | free(buf); | |
227 | ||
abeb40e5 | 228 | if (partial_pack_sha1) |
9126f009 NP |
229 | git_SHA1_Final(partial_pack_sha1, &old_sha1_ctx); |
230 | git_SHA1_Final(new_pack_sha1, &new_sha1_ctx); | |
abeb40e5 | 231 | write_or_die(pack_fd, new_pack_sha1, 20); |
0c68d386 | 232 | fsync_or_die(pack_fd, pack_name); |
8b0eca7c | 233 | } |
106764e6 SP |
234 | |
235 | char *index_pack_lockfile(int ip_out) | |
236 | { | |
106764e6 SP |
237 | char packname[46]; |
238 | ||
239 | /* | |
6e180cdc | 240 | * The first thing we expect from index-pack's output |
106764e6 SP |
241 | * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where |
242 | * %40s is the newly created pack SHA1 name. In the "keep" | |
243 | * case, we need it to remove the corresponding .keep file | |
244 | * later on. If we don't get that then tough luck with it. | |
245 | */ | |
c697ad14 HO |
246 | if (read_in_full(ip_out, packname, 46) == 46 && packname[45] == '\n' && |
247 | memcmp(packname, "keep\t", 5) == 0) { | |
106764e6 SP |
248 | char path[PATH_MAX]; |
249 | packname[45] = 0; | |
250 | snprintf(path, sizeof(path), "%s/pack/pack-%s.keep", | |
251 | get_object_directory(), packname + 5); | |
252 | return xstrdup(path); | |
253 | } | |
254 | return NULL; | |
255 | } | |
f965c525 NP |
256 | |
257 | /* | |
258 | * The per-object header is a pretty dense thing, which is | |
259 | * - first byte: low four bits are "size", then three bits of "type", | |
260 | * and the high bit is "size continues". | |
261 | * - each byte afterwards: low seven bits are size continuation, | |
262 | * with the high bit being "size continues" | |
263 | */ | |
264 | int encode_in_pack_object_header(enum object_type type, uintmax_t size, unsigned char *hdr) | |
265 | { | |
266 | int n = 1; | |
267 | unsigned char c; | |
268 | ||
269 | if (type < OBJ_COMMIT || type > OBJ_REF_DELTA) | |
270 | die("bad type %d", type); | |
271 | ||
272 | c = (type << 4) | (size & 15); | |
273 | size >>= 4; | |
274 | while (size) { | |
275 | *hdr++ = c | 0x80; | |
276 | c = size & 0x7f; | |
277 | size >>= 7; | |
278 | n++; | |
279 | } | |
280 | *hdr = c; | |
281 | return n; | |
282 | } |