| 1 | #ifndef PACK_OBJECTS_H |
| 2 | #define PACK_OBJECTS_H |
| 3 | |
| 4 | #include "odb.h" |
| 5 | #include "thread-utils.h" |
| 6 | #include "pack.h" |
| 7 | #include "packfile.h" |
| 8 | |
| 9 | struct repository; |
| 10 | |
| 11 | #define DEFAULT_DELTA_CACHE_SIZE (256 * 1024 * 1024) |
| 12 | #define DEFAULT_DELTA_BASE_CACHE_LIMIT (96 * 1024 * 1024) |
| 13 | |
| 14 | #define OE_DFS_STATE_BITS 2 |
| 15 | #define OE_DEPTH_BITS 12 |
| 16 | #define OE_IN_PACK_BITS 10 |
| 17 | #define OE_Z_DELTA_BITS 20 |
| 18 | /* |
| 19 | * Note that oe_set_size() becomes expensive when the given size is |
| 20 | * above this limit. Don't lower it too much. |
| 21 | */ |
| 22 | #define OE_SIZE_BITS 31 |
| 23 | #define OE_DELTA_SIZE_BITS 23 |
| 24 | |
| 25 | /* |
| 26 | * State flags for depth-first search used for analyzing delta cycles. |
| 27 | * |
| 28 | * The depth is measured in delta-links to the base (so if A is a delta |
| 29 | * against B, then A has a depth of 1, and B a depth of 0). |
| 30 | */ |
| 31 | enum dfs_state { |
| 32 | DFS_NONE = 0, |
| 33 | DFS_ACTIVE, |
| 34 | DFS_DONE, |
| 35 | DFS_NUM_STATES |
| 36 | }; |
| 37 | |
| 38 | /* |
| 39 | * The size of struct nearly determines pack-objects's memory |
| 40 | * consumption. This struct is packed tight for that reason. When you |
| 41 | * add or reorder something in this struct, think a bit about this. |
| 42 | * |
| 43 | * basic object info |
| 44 | * ----------------- |
| 45 | * idx.oid is filled up before delta searching starts. idx.crc32 is |
| 46 | * only valid after the object is written out and will be used for |
| 47 | * generating the index. idx.offset will be both gradually set and |
| 48 | * used in writing phase (base objects get offset first, then deltas |
| 49 | * refer to them) |
| 50 | * |
| 51 | * "size" is the uncompressed object size. Compressed size of the raw |
| 52 | * data for an object in a pack is not stored anywhere but is computed |
| 53 | * and made available when reverse .idx is made. Note that when a |
| 54 | * delta is reused, "size" is the uncompressed _delta_ size, not the |
| 55 | * canonical one after the delta has been applied. |
| 56 | * |
| 57 | * "hash" contains a path name hash which is used for sorting the |
| 58 | * delta list and also during delta searching. Once prepare_pack() |
| 59 | * returns it's no longer needed. |
| 60 | * |
| 61 | * source pack info |
| 62 | * ---------------- |
| 63 | * The (in_pack, in_pack_offset) tuple contains the location of the |
| 64 | * object in the source pack. in_pack_header_size allows quickly |
| 65 | * skipping the header and going straight to the zlib stream. |
| 66 | * |
| 67 | * "type" and "in_pack_type" both describe object type. in_pack_type |
| 68 | * may contain a delta type, while type is always the canonical type. |
| 69 | * |
| 70 | * deltas |
| 71 | * ------ |
| 72 | * Delta links (delta, delta_child and delta_sibling) are created to |
| 73 | * reflect that delta graph from the source pack then updated or added |
| 74 | * during delta searching phase when we find better deltas. |
| 75 | * |
| 76 | * delta_child and delta_sibling are last needed in |
| 77 | * compute_write_order(). "delta" and "delta_size" must remain valid |
| 78 | * at object writing phase in case the delta is not cached. |
| 79 | * |
| 80 | * If a delta is cached in memory and is compressed, delta_data points |
| 81 | * to the data and z_delta_size contains the compressed size. If it's |
| 82 | * uncompressed [1], z_delta_size must be zero. delta_size is always |
| 83 | * the uncompressed size and must be valid even if the delta is not |
| 84 | * cached. |
| 85 | * |
| 86 | * [1] during try_delta phase we don't bother with compressing because |
| 87 | * the delta could be quickly replaced with a better one. |
| 88 | */ |
| 89 | struct object_entry { |
| 90 | struct pack_idx_entry idx; |
| 91 | void *delta_data; /* cached delta (uncompressed) */ |
| 92 | off_t in_pack_offset; |
| 93 | uint32_t hash; /* name hint hash */ |
| 94 | unsigned size_:OE_SIZE_BITS; |
| 95 | unsigned size_valid:1; |
| 96 | uint32_t delta_idx; /* delta base object */ |
| 97 | uint32_t delta_child_idx; /* deltified objects who bases me */ |
| 98 | uint32_t delta_sibling_idx; /* other deltified objects who |
| 99 | * uses the same base as me |
| 100 | */ |
| 101 | unsigned delta_size_:OE_DELTA_SIZE_BITS; /* delta data size (uncompressed) */ |
| 102 | unsigned delta_size_valid:1; |
| 103 | unsigned char in_pack_header_size; |
| 104 | unsigned in_pack_idx:OE_IN_PACK_BITS; /* already in pack */ |
| 105 | unsigned z_delta_size:OE_Z_DELTA_BITS; |
| 106 | unsigned type_valid:1; |
| 107 | unsigned no_try_delta:1; |
| 108 | unsigned type_:TYPE_BITS; |
| 109 | unsigned in_pack_type:TYPE_BITS; /* could be delta */ |
| 110 | |
| 111 | unsigned preferred_base:1; /* |
| 112 | * we do not pack this, but is available |
| 113 | * to be used as the base object to delta |
| 114 | * objects against. |
| 115 | */ |
| 116 | unsigned tagged:1; /* near the very tip of refs */ |
| 117 | unsigned filled:1; /* assigned write-order */ |
| 118 | unsigned dfs_state:OE_DFS_STATE_BITS; |
| 119 | unsigned depth:OE_DEPTH_BITS; |
| 120 | unsigned ext_base:1; /* delta_idx points outside packlist */ |
| 121 | }; |
| 122 | |
| 123 | /** |
| 124 | * A packing region is a section of the packing_data.objects array |
| 125 | * as given by a starting index and a number of elements. |
| 126 | */ |
| 127 | struct packing_region { |
| 128 | size_t start; |
| 129 | size_t nr; |
| 130 | }; |
| 131 | |
| 132 | struct packing_data { |
| 133 | struct repository *repo; |
| 134 | struct object_entry *objects; |
| 135 | uint32_t nr_objects, nr_alloc; |
| 136 | |
| 137 | struct packing_region *regions; |
| 138 | size_t nr_regions, nr_regions_alloc; |
| 139 | |
| 140 | int32_t *index; |
| 141 | uint32_t index_size; |
| 142 | |
| 143 | unsigned int *in_pack_pos; |
| 144 | size_t *delta_size; |
| 145 | |
| 146 | /* |
| 147 | * Only one of these can be non-NULL and they have different |
| 148 | * sizes. if in_pack_by_idx is allocated, oe_in_pack() returns |
| 149 | * the pack of an object using in_pack_idx field. If not, |
| 150 | * in_pack[] array is used the same way as in_pack_pos[] |
| 151 | */ |
| 152 | struct packed_git **in_pack_by_idx; |
| 153 | struct packed_git **in_pack; |
| 154 | |
| 155 | /* |
| 156 | * During packing with multiple threads, protect the in-core |
| 157 | * object database from concurrent accesses. |
| 158 | */ |
| 159 | pthread_mutex_t odb_lock; |
| 160 | |
| 161 | /* |
| 162 | * This list contains entries for bases which we know the other side |
| 163 | * has (e.g., via reachability bitmaps), but which aren't in our |
| 164 | * "objects" list. |
| 165 | */ |
| 166 | struct object_entry *ext_bases; |
| 167 | uint32_t nr_ext, alloc_ext; |
| 168 | |
| 169 | uintmax_t oe_size_limit; |
| 170 | uintmax_t oe_delta_size_limit; |
| 171 | |
| 172 | /* delta islands */ |
| 173 | unsigned int *tree_depth; |
| 174 | unsigned char *layer; |
| 175 | |
| 176 | /* |
| 177 | * Used when writing cruft packs. |
| 178 | * |
| 179 | * Object mtimes are stored in pack order when writing, but |
| 180 | * written out in lexicographic (index) order. |
| 181 | */ |
| 182 | uint32_t *cruft_mtime; |
| 183 | }; |
| 184 | |
| 185 | void prepare_packing_data(struct repository *r, struct packing_data *pdata); |
| 186 | void clear_packing_data(struct packing_data *pdata); |
| 187 | |
| 188 | /* Protect access to object database */ |
| 189 | static inline void packing_data_lock(struct packing_data *pdata) |
| 190 | { |
| 191 | pthread_mutex_lock(&pdata->odb_lock); |
| 192 | } |
| 193 | static inline void packing_data_unlock(struct packing_data *pdata) |
| 194 | { |
| 195 | pthread_mutex_unlock(&pdata->odb_lock); |
| 196 | } |
| 197 | |
| 198 | struct object_entry *packlist_alloc(struct packing_data *pdata, |
| 199 | const struct object_id *oid); |
| 200 | |
| 201 | struct object_entry *packlist_find(struct packing_data *pdata, |
| 202 | const struct object_id *oid); |
| 203 | |
| 204 | static inline uint32_t pack_name_hash(const char *name) |
| 205 | { |
| 206 | uint32_t c, hash = 0; |
| 207 | |
| 208 | if (!name) |
| 209 | return 0; |
| 210 | |
| 211 | /* |
| 212 | * This effectively just creates a sortable number from the |
| 213 | * last sixteen non-whitespace characters. Last characters |
| 214 | * count "most", so things that end in ".c" sort together. |
| 215 | */ |
| 216 | while ((c = *name++) != 0) { |
| 217 | if (isspace(c)) |
| 218 | continue; |
| 219 | hash = (hash >> 2) + (c << 24); |
| 220 | } |
| 221 | return hash; |
| 222 | } |
| 223 | |
| 224 | static inline uint32_t pack_name_hash_v2(const unsigned char *name) |
| 225 | { |
| 226 | uint32_t hash = 0, base = 0, c; |
| 227 | |
| 228 | if (!name) |
| 229 | return 0; |
| 230 | |
| 231 | while ((c = *name++)) { |
| 232 | if (isspace(c)) |
| 233 | continue; |
| 234 | if (c == '/') { |
| 235 | base = (base >> 6) ^ hash; |
| 236 | hash = 0; |
| 237 | } else { |
| 238 | /* |
| 239 | * 'c' is only a single byte. Reverse it and move |
| 240 | * it to the top of the hash, moving the rest to |
| 241 | * less-significant bits. |
| 242 | */ |
| 243 | c = (c & 0xF0) >> 4 | (c & 0x0F) << 4; |
| 244 | c = (c & 0xCC) >> 2 | (c & 0x33) << 2; |
| 245 | c = (c & 0xAA) >> 1 | (c & 0x55) << 1; |
| 246 | hash = (hash >> 2) + (c << 24); |
| 247 | } |
| 248 | } |
| 249 | return (base >> 6) ^ hash; |
| 250 | } |
| 251 | |
| 252 | static inline enum object_type oe_type(const struct object_entry *e) |
| 253 | { |
| 254 | return e->type_valid ? e->type_ : OBJ_BAD; |
| 255 | } |
| 256 | |
| 257 | static inline void oe_set_type(struct object_entry *e, |
| 258 | enum object_type type) |
| 259 | { |
| 260 | if (type >= OBJ_ANY) |
| 261 | BUG("OBJ_ANY cannot be set in pack-objects code"); |
| 262 | |
| 263 | e->type_valid = type >= OBJ_NONE; |
| 264 | e->type_ = (unsigned)type; |
| 265 | } |
| 266 | |
| 267 | static inline unsigned int oe_in_pack_pos(const struct packing_data *pack, |
| 268 | const struct object_entry *e) |
| 269 | { |
| 270 | return pack->in_pack_pos[e - pack->objects]; |
| 271 | } |
| 272 | |
| 273 | static inline void oe_set_in_pack_pos(const struct packing_data *pack, |
| 274 | const struct object_entry *e, |
| 275 | unsigned int pos) |
| 276 | { |
| 277 | pack->in_pack_pos[e - pack->objects] = pos; |
| 278 | } |
| 279 | |
| 280 | static inline struct packed_git *oe_in_pack(const struct packing_data *pack, |
| 281 | const struct object_entry *e) |
| 282 | { |
| 283 | if (pack->in_pack_by_idx) |
| 284 | return pack->in_pack_by_idx[e->in_pack_idx]; |
| 285 | else |
| 286 | return pack->in_pack[e - pack->objects]; |
| 287 | } |
| 288 | |
| 289 | void oe_map_new_pack(struct packing_data *pack); |
| 290 | |
| 291 | static inline void oe_set_in_pack(struct packing_data *pack, |
| 292 | struct object_entry *e, |
| 293 | struct packed_git *p) |
| 294 | { |
| 295 | if (pack->in_pack_by_idx) { |
| 296 | if (p->index) { |
| 297 | e->in_pack_idx = p->index; |
| 298 | return; |
| 299 | } |
| 300 | /* |
| 301 | * We're accessing packs by index, but this pack doesn't have |
| 302 | * an index (e.g., because it was added since we created the |
| 303 | * in_pack_by_idx array). Bail to oe_map_new_pack(), which |
| 304 | * will convert us to using the full in_pack array, and then |
| 305 | * fall through to our in_pack handling. |
| 306 | */ |
| 307 | oe_map_new_pack(pack); |
| 308 | } |
| 309 | pack->in_pack[e - pack->objects] = p; |
| 310 | } |
| 311 | |
| 312 | void oe_set_delta_ext(struct packing_data *pack, |
| 313 | struct object_entry *e, |
| 314 | const struct object_id *oid); |
| 315 | |
| 316 | static inline unsigned int oe_tree_depth(struct packing_data *pack, |
| 317 | struct object_entry *e) |
| 318 | { |
| 319 | if (!pack->tree_depth) |
| 320 | return 0; |
| 321 | return pack->tree_depth[e - pack->objects]; |
| 322 | } |
| 323 | |
| 324 | static inline void oe_set_layer(struct packing_data *pack, |
| 325 | struct object_entry *e, |
| 326 | unsigned char layer) |
| 327 | { |
| 328 | if (!pack->layer) |
| 329 | CALLOC_ARRAY(pack->layer, pack->nr_alloc); |
| 330 | pack->layer[e - pack->objects] = layer; |
| 331 | } |
| 332 | |
| 333 | static inline uint32_t oe_cruft_mtime(struct packing_data *pack, |
| 334 | struct object_entry *e) |
| 335 | { |
| 336 | if (!pack->cruft_mtime) |
| 337 | return 0; |
| 338 | return pack->cruft_mtime[e - pack->objects]; |
| 339 | } |
| 340 | |
| 341 | static inline void oe_set_cruft_mtime(struct packing_data *pack, |
| 342 | struct object_entry *e, |
| 343 | uint32_t mtime) |
| 344 | { |
| 345 | if (!pack->cruft_mtime) |
| 346 | CALLOC_ARRAY(pack->cruft_mtime, pack->nr_alloc); |
| 347 | pack->cruft_mtime[e - pack->objects] = mtime; |
| 348 | } |
| 349 | |
| 350 | #endif |