Merge branch 'jc/maint-read-tree-multi' into maint-1.6.2
[git/kusma.git] / pack-write.c
blob7053538f4cf44e15a788ab46dfb680ee85ce4fc2
1 #include "cache.h"
2 #include "pack.h"
3 #include "csum-file.h"
5 uint32_t pack_idx_default_version = 2;
6 uint32_t pack_idx_off32_limit = 0x7fffffff;
8 static int sha1_compare(const void *_a, const void *_b)
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);
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.
20 char *write_idx_file(char *index_name, struct pack_idx_entry **objects,
21 int nr_objects, unsigned char *sha1)
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;
28 git_SHA_CTX ctx;
29 uint32_t index_version;
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;
39 qsort(sorted_by_sha, nr_objects, sizeof(sorted_by_sha[0]),
40 sha1_compare);
42 else
43 sorted_by_sha = list = last = NULL;
45 if (!index_name) {
46 static char tmpfile[PATH_MAX];
47 fd = odb_mkstemp(tmpfile, sizeof(tmpfile), "pack/tmp_idx_XXXXXX");
48 index_name = xstrdup(tmpfile);
49 } else {
50 unlink(index_name);
51 fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
53 if (fd < 0)
54 die("unable to create %s: %s", index_name, strerror(errno));
55 f = sha1fd(fd, index_name);
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;
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));
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).
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++;
81 array[i] = htonl(next - sorted_by_sha);
82 list = next;
84 sha1write(f, array, 256 * 4);
86 /* compute the SHA1 hash of sorted object names. */
87 git_SHA1_Init(&ctx);
90 * Write the actual SHA1 entries..
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);
99 sha1write(f, obj->sha1, 20);
100 git_SHA1_Update(&ctx, obj->sha1, 20);
103 if (index_version >= 2) {
104 unsigned int nr_large_offset = 0;
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);
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);
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--;
139 sha1write(f, sha1, 20);
140 sha1close(f, NULL, CSUM_FSYNC);
141 git_SHA1_Final(sha1, &ctx);
142 return index_name;
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.
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.
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.
161 void fixup_pack_header_footer(int pack_fd,
162 unsigned char *new_pack_sha1,
163 const char *pack_name,
164 uint32_t object_count,
165 unsigned char *partial_pack_sha1,
166 off_t partial_pack_offset)
168 int aligned_sz, buf_sz = 8 * 1024;
169 git_SHA_CTX old_sha1_ctx, new_sha1_ctx;
170 struct pack_header hdr;
171 char *buf;
173 git_SHA1_Init(&old_sha1_ctx);
174 git_SHA1_Init(&new_sha1_ctx);
176 if (lseek(pack_fd, 0, SEEK_SET) != 0)
177 die("Failed seeking to start of %s: %s", pack_name, strerror(errno));
178 if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
179 die("Unable to reread header of %s: %s", pack_name, strerror(errno));
180 if (lseek(pack_fd, 0, SEEK_SET) != 0)
181 die("Failed seeking to start of %s: %s", pack_name, strerror(errno));
182 git_SHA1_Update(&old_sha1_ctx, &hdr, sizeof(hdr));
183 hdr.hdr_entries = htonl(object_count);
184 git_SHA1_Update(&new_sha1_ctx, &hdr, sizeof(hdr));
185 write_or_die(pack_fd, &hdr, sizeof(hdr));
186 partial_pack_offset -= sizeof(hdr);
188 buf = xmalloc(buf_sz);
189 aligned_sz = buf_sz - sizeof(hdr);
190 for (;;) {
191 ssize_t m, n;
192 m = (partial_pack_sha1 && partial_pack_offset < aligned_sz) ?
193 partial_pack_offset : aligned_sz;
194 n = xread(pack_fd, buf, m);
195 if (!n)
196 break;
197 if (n < 0)
198 die("Failed to checksum %s: %s", pack_name, strerror(errno));
199 git_SHA1_Update(&new_sha1_ctx, buf, n);
201 aligned_sz -= n;
202 if (!aligned_sz)
203 aligned_sz = buf_sz;
205 if (!partial_pack_sha1)
206 continue;
208 git_SHA1_Update(&old_sha1_ctx, buf, n);
209 partial_pack_offset -= n;
210 if (partial_pack_offset == 0) {
211 unsigned char sha1[20];
212 git_SHA1_Final(sha1, &old_sha1_ctx);
213 if (hashcmp(sha1, partial_pack_sha1) != 0)
214 die("Unexpected checksum for %s "
215 "(disk corruption?)", pack_name);
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.
221 git_SHA1_Init(&old_sha1_ctx);
222 partial_pack_offset = ~partial_pack_offset;
223 partial_pack_offset -= MSB(partial_pack_offset, 1);
226 free(buf);
228 if (partial_pack_sha1)
229 git_SHA1_Final(partial_pack_sha1, &old_sha1_ctx);
230 git_SHA1_Final(new_pack_sha1, &new_sha1_ctx);
231 write_or_die(pack_fd, new_pack_sha1, 20);
232 fsync_or_die(pack_fd, pack_name);
235 char *index_pack_lockfile(int ip_out)
237 char packname[46];
240 * The first thing we expect from index-pack's output
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.
246 if (read_in_full(ip_out, packname, 46) == 46 && packname[45] == '\n' &&
247 memcmp(packname, "keep\t", 5) == 0) {
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);
254 return NULL;