2 #include "pack-revindex.h"
3 #include "object-store.h"
7 struct revindex_entry {
13 * Pack index for existing packs give us easy access to the offsets into
14 * corresponding pack file where each object's data starts, but the entries
15 * do not store the size of the compressed representation (uncompressed
16 * size is easily available by examining the pack entry header). It is
17 * also rather expensive to find the sha1 for an object given its offset.
19 * The pack index file is sorted by object name mapping to offset;
20 * this revindex array is a list of offset/index_nr pairs
21 * ordered by offset, so if you know the offset of an object, next offset
22 * is where its packed representation ends and the index_nr can be used to
23 * get the object sha1 from the main index.
27 * This is a least-significant-digit radix sort.
29 * It sorts each of the "n" items in "entries" by its offset field. The "max"
30 * parameter must be at least as large as the largest offset in the array,
31 * and lets us quit the sort early.
33 static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
36 * We use a "digit" size of 16 bits. That keeps our memory
37 * usage reasonable, and we can generally (for a 4G or smaller
38 * packfile) quit after two rounds of radix-sorting.
40 #define DIGIT_SIZE (16)
41 #define BUCKETS (1 << DIGIT_SIZE)
43 * We want to know the bucket that a[i] will go into when we are using
44 * the digit that is N bits from the (least significant) end.
46 #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
49 * We need O(n) temporary storage. Rather than do an extra copy of the
50 * partial results into "entries", we sort back and forth between the
51 * real array and temporary storage. In each iteration of the loop, we
52 * keep track of them with alias pointers, always sorting from "from"
55 struct revindex_entry *tmp, *from, *to;
59 ALLOC_ARRAY(pos, BUCKETS);
65 * If (max >> bits) is zero, then we know that the radix digit we are
66 * on (and any higher) will be zero for all entries, and our loop will
67 * be a no-op, as everybody lands in the same zero-th bucket.
69 for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
72 memset(pos, 0, BUCKETS * sizeof(*pos));
75 * We want pos[i] to store the index of the last element that
76 * will go in bucket "i" (actually one past the last element).
77 * To do this, we first count the items that will go in each
78 * bucket, which gives us a relative offset from the last
79 * bucket. We can then cumulatively add the index from the
80 * previous bucket to get the true index.
82 for (i = 0; i < n; i++)
83 pos[BUCKET_FOR(from, i, bits)]++;
84 for (i = 1; i < BUCKETS; i++)
88 * Now we can drop the elements into their correct buckets (in
89 * our temporary array). We iterate the pos counter backwards
90 * to avoid using an extra index to count up. And since we are
91 * going backwards there, we must also go backwards through the
92 * array itself, to keep the sort stable.
94 * Note that we use an unsigned iterator to make sure we can
95 * handle 2^32-1 objects, even on a 32-bit system. But this
96 * means we cannot use the more obvious "i >= 0" loop condition
97 * for counting backwards, and must instead check for
98 * wrap-around with UINT_MAX.
100 for (i = n - 1; i != UINT_MAX; i--)
101 to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
104 * Now "to" contains the most sorted list, so we swap "from" and
105 * "to" for the next iteration.
111 * If we ended with our data in the original array, great. If not,
112 * we have to move it back from the temporary storage.
115 COPY_ARRAY(entries, tmp, n);
125 * Ordered list of offsets of objects in the pack.
127 static void create_pack_revindex(struct packed_git *p)
129 const unsigned num_ent = p->num_objects;
131 const char *index = p->index_data;
132 const unsigned hashsz = the_hash_algo->rawsz;
134 ALLOC_ARRAY(p->revindex, num_ent + 1);
137 if (p->index_version > 1) {
138 const uint32_t *off_32 =
139 (uint32_t *)(index + 8 + (size_t)p->num_objects * (hashsz + 4));
140 const uint32_t *off_64 = off_32 + p->num_objects;
141 for (i = 0; i < num_ent; i++) {
142 const uint32_t off = ntohl(*off_32++);
143 if (!(off & 0x80000000)) {
144 p->revindex[i].offset = off;
146 p->revindex[i].offset = get_be64(off_64);
149 p->revindex[i].nr = i;
152 for (i = 0; i < num_ent; i++) {
153 const uint32_t hl = *((uint32_t *)(index + (hashsz + 4) * i));
154 p->revindex[i].offset = ntohl(hl);
155 p->revindex[i].nr = i;
160 * This knows the pack format -- the hash trailer
161 * follows immediately after the last object data.
163 p->revindex[num_ent].offset = p->pack_size - hashsz;
164 p->revindex[num_ent].nr = -1;
165 sort_revindex(p->revindex, num_ent, p->pack_size);
168 static int create_pack_revindex_in_memory(struct packed_git *p)
170 if (git_env_bool(GIT_TEST_REV_INDEX_DIE_IN_MEMORY, 0))
171 die("dying as requested by '%s'",
172 GIT_TEST_REV_INDEX_DIE_IN_MEMORY);
173 if (open_pack_index(p))
175 create_pack_revindex(p);
179 static char *pack_revindex_filename(struct packed_git *p)
182 if (!strip_suffix(p->pack_name, ".pack", &len))
183 BUG("pack_name does not end in .pack");
184 return xstrfmt("%.*s.rev", (int)len, p->pack_name);
187 #define RIDX_HEADER_SIZE (12)
188 #define RIDX_MIN_SIZE (RIDX_HEADER_SIZE + (2 * the_hash_algo->rawsz))
190 struct revindex_header {
196 static int load_revindex_from_disk(char *revindex_name,
197 uint32_t num_objects,
198 const uint32_t **data_p, size_t *len_p)
203 size_t revindex_size;
204 struct revindex_header *hdr;
206 fd = git_open(revindex_name);
212 if (fstat(fd, &st)) {
213 ret = error_errno(_("failed to read %s"), revindex_name);
217 revindex_size = xsize_t(st.st_size);
219 if (revindex_size < RIDX_MIN_SIZE) {
220 ret = error(_("reverse-index file %s is too small"), revindex_name);
224 if (revindex_size - RIDX_MIN_SIZE != st_mult(sizeof(uint32_t), num_objects)) {
225 ret = error(_("reverse-index file %s is corrupt"), revindex_name);
229 data = xmmap(NULL, revindex_size, PROT_READ, MAP_PRIVATE, fd, 0);
232 if (ntohl(hdr->signature) != RIDX_SIGNATURE) {
233 ret = error(_("reverse-index file %s has unknown signature"), revindex_name);
236 if (ntohl(hdr->version) != 1) {
237 ret = error(_("reverse-index file %s has unsupported version %"PRIu32),
238 revindex_name, ntohl(hdr->version));
241 if (!(ntohl(hdr->hash_id) == 1 || ntohl(hdr->hash_id) == 2)) {
242 ret = error(_("reverse-index file %s has unsupported hash id %"PRIu32),
243 revindex_name, ntohl(hdr->hash_id));
250 munmap(data, revindex_size);
252 *len_p = revindex_size;
253 *data_p = (const uint32_t *)data;
261 static int load_pack_revindex_from_disk(struct packed_git *p)
265 if (open_pack_index(p))
268 revindex_name = pack_revindex_filename(p);
270 ret = load_revindex_from_disk(revindex_name,
277 p->revindex_data = (const uint32_t *)((const char *)p->revindex_map + RIDX_HEADER_SIZE);
284 int load_pack_revindex(struct packed_git *p)
286 if (p->revindex || p->revindex_data)
289 if (!load_pack_revindex_from_disk(p))
291 else if (!create_pack_revindex_in_memory(p))
296 int offset_to_pack_pos(struct packed_git *p, off_t ofs, uint32_t *pos)
300 if (load_pack_revindex(p) < 0)
304 hi = p->num_objects + 1;
307 const unsigned mi = lo + (hi - lo) / 2;
308 off_t got = pack_pos_to_offset(p, mi);
313 } else if (ofs < got)
319 error("bad offset for revindex");
323 uint32_t pack_pos_to_index(struct packed_git *p, uint32_t pos)
325 if (!(p->revindex || p->revindex_data))
326 BUG("pack_pos_to_index: reverse index not yet loaded");
327 if (p->num_objects <= pos)
328 BUG("pack_pos_to_index: out-of-bounds object at %"PRIu32, pos);
331 return p->revindex[pos].nr;
333 return get_be32(p->revindex_data + pos);
336 off_t pack_pos_to_offset(struct packed_git *p, uint32_t pos)
338 if (!(p->revindex || p->revindex_data))
339 BUG("pack_pos_to_index: reverse index not yet loaded");
340 if (p->num_objects < pos)
341 BUG("pack_pos_to_offset: out-of-bounds object at %"PRIu32, pos);
344 return p->revindex[pos].offset;
345 else if (pos == p->num_objects)
346 return p->pack_size - the_hash_algo->rawsz;
348 return nth_packed_object_offset(p, pack_pos_to_index(p, pos));