1#ifndef PACK_OBJECTS_H 2#define PACK_OBJECTS_H 3 4#include"object-store.h" 5#include"thread-utils.h" 6#include"pack.h" 7 8#define DEFAULT_DELTA_CACHE_SIZE (256 * 1024 * 1024) 9 10#define OE_DFS_STATE_BITS 2 11#define OE_DEPTH_BITS 12 12#define OE_IN_PACK_BITS 10 13#define OE_Z_DELTA_BITS 20 14/* 15 * Note that oe_set_size() becomes expensive when the given size is 16 * above this limit. Don't lower it too much. 17 */ 18#define OE_SIZE_BITS 31 19#define OE_DELTA_SIZE_BITS 23 20 21/* 22 * State flags for depth-first search used for analyzing delta cycles. 23 * 24 * The depth is measured in delta-links to the base (so if A is a delta 25 * against B, then A has a depth of 1, and B a depth of 0). 26 */ 27enum dfs_state { 28 DFS_NONE =0, 29 DFS_ACTIVE, 30 DFS_DONE, 31 DFS_NUM_STATES 32}; 33 34/* 35 * The size of struct nearly determines pack-objects's memory 36 * consumption. This struct is packed tight for that reason. When you 37 * add or reorder something in this struct, think a bit about this. 38 * 39 * basic object info 40 * ----------------- 41 * idx.oid is filled up before delta searching starts. idx.crc32 is 42 * only valid after the object is written out and will be used for 43 * generating the index. idx.offset will be both gradually set and 44 * used in writing phase (base objects get offset first, then deltas 45 * refer to them) 46 * 47 * "size" is the uncompressed object size. Compressed size of the raw 48 * data for an object in a pack is not stored anywhere but is computed 49 * and made available when reverse .idx is made. Note that when a 50 * delta is reused, "size" is the uncompressed _delta_ size, not the 51 * canonical one after the delta has been applied. 52 * 53 * "hash" contains a path name hash which is used for sorting the 54 * delta list and also during delta searching. Once prepare_pack() 55 * returns it's no longer needed. 56 * 57 * source pack info 58 * ---------------- 59 * The (in_pack, in_pack_offset) tuple contains the location of the 60 * object in the source pack. in_pack_header_size allows quickly 61 * skipping the header and going straight to the zlib stream. 62 * 63 * "type" and "in_pack_type" both describe object type. in_pack_type 64 * may contain a delta type, while type is always the canonical type. 65 * 66 * deltas 67 * ------ 68 * Delta links (delta, delta_child and delta_sibling) are created to 69 * reflect that delta graph from the source pack then updated or added 70 * during delta searching phase when we find better deltas. 71 * 72 * delta_child and delta_sibling are last needed in 73 * compute_write_order(). "delta" and "delta_size" must remain valid 74 * at object writing phase in case the delta is not cached. 75 * 76 * If a delta is cached in memory and is compressed, delta_data points 77 * to the data and z_delta_size contains the compressed size. If it's 78 * uncompressed [1], z_delta_size must be zero. delta_size is always 79 * the uncompressed size and must be valid even if the delta is not 80 * cached. 81 * 82 * [1] during try_delta phase we don't bother with compressing because 83 * the delta could be quickly replaced with a better one. 84 */ 85struct object_entry { 86struct pack_idx_entry idx; 87void*delta_data;/* cached delta (uncompressed) */ 88 off_t in_pack_offset; 89uint32_t hash;/* name hint hash */ 90unsigned size_:OE_SIZE_BITS; 91unsigned size_valid:1; 92uint32_t delta_idx;/* delta base object */ 93uint32_t delta_child_idx;/* deltified objects who bases me */ 94uint32_t delta_sibling_idx;/* other deltified objects who 95 * uses the same base as me 96 */ 97unsigned delta_size_:OE_DELTA_SIZE_BITS;/* delta data size (uncompressed) */ 98unsigned delta_size_valid:1; 99unsigned char in_pack_header_size; 100unsigned in_pack_idx:OE_IN_PACK_BITS;/* already in pack */ 101unsigned z_delta_size:OE_Z_DELTA_BITS; 102unsigned type_valid:1; 103unsigned no_try_delta:1; 104unsigned type_:TYPE_BITS; 105unsigned in_pack_type:TYPE_BITS;/* could be delta */ 106 107unsigned preferred_base:1;/* 108 * we do not pack this, but is available 109 * to be used as the base object to delta 110 * objects against. 111 */ 112unsigned tagged:1;/* near the very tip of refs */ 113unsigned filled:1;/* assigned write-order */ 114unsigned dfs_state:OE_DFS_STATE_BITS; 115unsigned depth:OE_DEPTH_BITS; 116unsigned ext_base:1;/* delta_idx points outside packlist */ 117 118/* 119 * pahole results on 64-bit linux (gcc and clang) 120 * 121 * size: 80, bit_padding: 9 bits 122 * 123 * and on 32-bit (gcc) 124 * 125 * size: 76, bit_padding: 9 bits 126 */ 127}; 128 129struct packing_data { 130struct object_entry *objects; 131uint32_t nr_objects, nr_alloc; 132 133int32_t*index; 134uint32_t index_size; 135 136unsigned int*in_pack_pos; 137unsigned long*delta_size; 138 139/* 140 * Only one of these can be non-NULL and they have different 141 * sizes. if in_pack_by_idx is allocated, oe_in_pack() returns 142 * the pack of an object using in_pack_idx field. If not, 143 * in_pack[] array is used the same way as in_pack_pos[] 144 */ 145struct packed_git **in_pack_by_idx; 146struct packed_git **in_pack; 147 148 pthread_mutex_t lock; 149 150/* 151 * This list contains entries for bases which we know the other side 152 * has (e.g., via reachability bitmaps), but which aren't in our 153 * "objects" list. 154 */ 155struct object_entry *ext_bases; 156uint32_t nr_ext, alloc_ext; 157 158uintmax_t oe_size_limit; 159uintmax_t oe_delta_size_limit; 160 161/* delta islands */ 162unsigned int*tree_depth; 163unsigned char*layer; 164}; 165 166voidprepare_packing_data(struct packing_data *pdata); 167 168staticinlinevoidpacking_data_lock(struct packing_data *pdata) 169{ 170pthread_mutex_lock(&pdata->lock); 171} 172staticinlinevoidpacking_data_unlock(struct packing_data *pdata) 173{ 174pthread_mutex_unlock(&pdata->lock); 175} 176 177struct object_entry *packlist_alloc(struct packing_data *pdata, 178const unsigned char*sha1, 179uint32_t index_pos); 180 181struct object_entry *packlist_find(struct packing_data *pdata, 182const unsigned char*sha1, 183uint32_t*index_pos); 184 185staticinlineuint32_tpack_name_hash(const char*name) 186{ 187uint32_t c, hash =0; 188 189if(!name) 190return0; 191 192/* 193 * This effectively just creates a sortable number from the 194 * last sixteen non-whitespace characters. Last characters 195 * count "most", so things that end in ".c" sort together. 196 */ 197while((c = *name++) !=0) { 198if(isspace(c)) 199continue; 200 hash = (hash >>2) + (c <<24); 201} 202return hash; 203} 204 205staticinlineenum object_type oe_type(const struct object_entry *e) 206{ 207return e->type_valid ? e->type_ : OBJ_BAD; 208} 209 210staticinlinevoidoe_set_type(struct object_entry *e, 211enum object_type type) 212{ 213if(type >= OBJ_ANY) 214BUG("OBJ_ANY cannot be set in pack-objects code"); 215 216 e->type_valid = type >= OBJ_NONE; 217 e->type_ = (unsigned)type; 218} 219 220staticinlineunsigned intoe_in_pack_pos(const struct packing_data *pack, 221const struct object_entry *e) 222{ 223return pack->in_pack_pos[e - pack->objects]; 224} 225 226staticinlinevoidoe_set_in_pack_pos(const struct packing_data *pack, 227const struct object_entry *e, 228unsigned int pos) 229{ 230 pack->in_pack_pos[e - pack->objects] = pos; 231} 232 233staticinlinestruct packed_git *oe_in_pack(const struct packing_data *pack, 234const struct object_entry *e) 235{ 236if(pack->in_pack_by_idx) 237return pack->in_pack_by_idx[e->in_pack_idx]; 238else 239return pack->in_pack[e - pack->objects]; 240} 241 242voidoe_map_new_pack(struct packing_data *pack, 243struct packed_git *p); 244staticinlinevoidoe_set_in_pack(struct packing_data *pack, 245struct object_entry *e, 246struct packed_git *p) 247{ 248if(!p->index) 249oe_map_new_pack(pack, p); 250if(pack->in_pack_by_idx) 251 e->in_pack_idx = p->index; 252else 253 pack->in_pack[e - pack->objects] = p; 254} 255 256staticinlinestruct object_entry *oe_delta( 257const struct packing_data *pack, 258const struct object_entry *e) 259{ 260if(!e->delta_idx) 261return NULL; 262if(e->ext_base) 263return&pack->ext_bases[e->delta_idx -1]; 264else 265return&pack->objects[e->delta_idx -1]; 266} 267 268staticinlinevoidoe_set_delta(struct packing_data *pack, 269struct object_entry *e, 270struct object_entry *delta) 271{ 272if(delta) 273 e->delta_idx = (delta - pack->objects) +1; 274else 275 e->delta_idx =0; 276} 277 278voidoe_set_delta_ext(struct packing_data *pack, 279struct object_entry *e, 280const unsigned char*sha1); 281 282staticinlinestruct object_entry *oe_delta_child( 283const struct packing_data *pack, 284const struct object_entry *e) 285{ 286if(e->delta_child_idx) 287return&pack->objects[e->delta_child_idx -1]; 288return NULL; 289} 290 291staticinlinevoidoe_set_delta_child(struct packing_data *pack, 292struct object_entry *e, 293struct object_entry *delta) 294{ 295if(delta) 296 e->delta_child_idx = (delta - pack->objects) +1; 297else 298 e->delta_child_idx =0; 299} 300 301staticinlinestruct object_entry *oe_delta_sibling( 302const struct packing_data *pack, 303const struct object_entry *e) 304{ 305if(e->delta_sibling_idx) 306return&pack->objects[e->delta_sibling_idx -1]; 307return NULL; 308} 309 310staticinlinevoidoe_set_delta_sibling(struct packing_data *pack, 311struct object_entry *e, 312struct object_entry *delta) 313{ 314if(delta) 315 e->delta_sibling_idx = (delta - pack->objects) +1; 316else 317 e->delta_sibling_idx =0; 318} 319 320unsigned longoe_get_size_slow(struct packing_data *pack, 321const struct object_entry *e); 322staticinlineunsigned longoe_size(struct packing_data *pack, 323const struct object_entry *e) 324{ 325if(e->size_valid) 326return e->size_; 327 328returnoe_get_size_slow(pack, e); 329} 330 331staticinlineintoe_size_less_than(struct packing_data *pack, 332const struct object_entry *lhs, 333unsigned long rhs) 334{ 335if(lhs->size_valid) 336return lhs->size_ < rhs; 337if(rhs < pack->oe_size_limit)/* rhs < 2^x <= lhs ? */ 338return0; 339returnoe_get_size_slow(pack, lhs) < rhs; 340} 341 342staticinlineintoe_size_greater_than(struct packing_data *pack, 343const struct object_entry *lhs, 344unsigned long rhs) 345{ 346if(lhs->size_valid) 347return lhs->size_ > rhs; 348if(rhs < pack->oe_size_limit)/* rhs < 2^x <= lhs ? */ 349return1; 350returnoe_get_size_slow(pack, lhs) > rhs; 351} 352 353staticinlinevoidoe_set_size(struct packing_data *pack, 354struct object_entry *e, 355unsigned long size) 356{ 357if(size < pack->oe_size_limit) { 358 e->size_ = size; 359 e->size_valid =1; 360}else{ 361 e->size_valid =0; 362if(oe_get_size_slow(pack, e) != size) 363BUG("'size' is supposed to be the object size!"); 364} 365} 366 367staticinlineunsigned longoe_delta_size(struct packing_data *pack, 368const struct object_entry *e) 369{ 370if(e->delta_size_valid) 371return e->delta_size_; 372 373/* 374 * pack->delta_size[] can't be NULL because oe_set_delta_size() 375 * must have been called when a new delta is saved with 376 * oe_set_delta(). 377 * If oe_delta() returns NULL (i.e. default state, which means 378 * delta_size_valid is also false), then the caller must never 379 * call oe_delta_size(). 380 */ 381return pack->delta_size[e - pack->objects]; 382} 383 384staticinlinevoidoe_set_delta_size(struct packing_data *pack, 385struct object_entry *e, 386unsigned long size) 387{ 388if(size < pack->oe_delta_size_limit) { 389 e->delta_size_ = size; 390 e->delta_size_valid =1; 391}else{ 392packing_data_lock(pack); 393if(!pack->delta_size) 394ALLOC_ARRAY(pack->delta_size, pack->nr_alloc); 395packing_data_unlock(pack); 396 397 pack->delta_size[e - pack->objects] = size; 398 e->delta_size_valid =0; 399} 400} 401 402staticinlineunsigned intoe_tree_depth(struct packing_data *pack, 403struct object_entry *e) 404{ 405if(!pack->tree_depth) 406return0; 407return pack->tree_depth[e - pack->objects]; 408} 409 410staticinlinevoidoe_set_tree_depth(struct packing_data *pack, 411struct object_entry *e, 412unsigned int tree_depth) 413{ 414if(!pack->tree_depth) 415CALLOC_ARRAY(pack->tree_depth, pack->nr_alloc); 416 pack->tree_depth[e - pack->objects] = tree_depth; 417} 418 419staticinlineunsigned charoe_layer(struct packing_data *pack, 420struct object_entry *e) 421{ 422if(!pack->layer) 423return0; 424return pack->layer[e - pack->objects]; 425} 426 427staticinlinevoidoe_set_layer(struct packing_data *pack, 428struct object_entry *e, 429unsigned char layer) 430{ 431if(!pack->layer) 432CALLOC_ARRAY(pack->layer, pack->nr_alloc); 433 pack->layer[e - pack->objects] = layer; 434} 435 436#endif