make packed_object_info() resilient to pack corruptions
[git/git.git] / builtin-pack-objects.c
CommitLineData
5d4a6003 1#include "builtin.h"
c323ac7d 2#include "cache.h"
a74db82e 3#include "attr.h"
c323ac7d 4#include "object.h"
8e440259
PE
5#include "blob.h"
6#include "commit.h"
7#include "tag.h"
8#include "tree.h"
c323ac7d 9#include "delta.h"
a733cb60 10#include "pack.h"
3449f8c4 11#include "pack-revindex.h"
c38138cd 12#include "csum-file.h"
1b0c7174 13#include "tree-walk.h"
b5d97e6b
JH
14#include "diff.h"
15#include "revision.h"
16#include "list-objects.h"
96a02f8f 17#include "progress.h"
f0a24aa5 18#include "refs.h"
c323ac7d 19
8ecce684 20#ifdef THREADED_DELTA_SEARCH
833e3df1 21#include "thread-utils.h"
8ecce684
NP
22#include <pthread.h>
23#endif
24
231f240b 25static const char pack_usage[] = "\
f18d244a 26git pack-objects [{ -q | --progress | --all-progress }] \n\
a97773ce
BD
27 [--max-pack-size=N] [--local] [--incremental] \n\
28 [--window=N] [--window-memory=N] [--depth=N] \n\
fa736f72 29 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
367f4a43 30 [--threads=N] [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
ca11b212
NP
31 [--stdout | base-name] [--include-tag] \n\
32 [--keep-unreachable | --unpack-unreachable] \n\
f0a24aa5 33 [<ref-list | <object-list]";
c323ac7d 34
c323ac7d 35struct object_entry {
aa7e44bf 36 struct pack_idx_entry idx;
3f9ac8d2 37 unsigned long size; /* uncompressed size */
3f9ac8d2 38 struct packed_git *in_pack; /* already in pack */
6777a59f 39 off_t in_pack_offset;
a3fbf4df 40 struct object_entry *delta; /* delta base object */
82e5a82f 41 struct object_entry *delta_child; /* deltified objects who bases me */
15b4d577
JH
42 struct object_entry *delta_sibling; /* other deltified objects who
43 * uses the same base as me
44 */
074b2eea 45 void *delta_data; /* cached delta (uncompressed) */
a3fbf4df 46 unsigned long delta_size; /* delta data size (uncompressed) */
ed4a9031 47 unsigned long z_delta_size; /* delta data size (compressed) */
5a235b5e 48 unsigned int hash; /* name hint hash */
a3fbf4df
NP
49 enum object_type type;
50 enum object_type in_pack_type; /* could be delta */
51 unsigned char in_pack_header_size;
52 unsigned char preferred_base; /* we do not pack this, but is available
a74db82e 53 * to be used as the base object to delta
a3fbf4df
NP
54 * objects against.
55 */
a74db82e 56 unsigned char no_try_delta;
c323ac7d
LT
57};
58
3f9ac8d2 59/*
82e5a82f 60 * Objects we are going to pack are collected in objects array (dynamically
3f9ac8d2
JH
61 * expanded). nr_objects & nr_alloc controls this array. They are stored
62 * in the order we see -- typically rev-list --objects order that gives us
63 * nice "minimum seek" order.
3f9ac8d2 64 */
e4d58311 65static struct object_entry *objects;
79814f42 66static struct pack_idx_entry **written_list;
d01fb92f 67static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
3f9ac8d2 68
96f1e58f 69static int non_empty;
a7de7130 70static int reuse_delta = 1, reuse_object = 1;
e5e9714a 71static int keep_unreachable, unpack_unreachable, include_tag;
96f1e58f
DR
72static int local;
73static int incremental;
be6b1914 74static int allow_ofs_delta;
d01fb92f 75static const char *base_name;
024701f1 76static int progress = 1;
4812a93a 77static int window = 10;
2b84b5a8 78static uint32_t pack_size_limit, pack_size_limit_cfg;
618e613a 79static int depth = 50;
367f4a43 80static int delta_search_threads = 1;
df6d6101 81static int pack_to_stdout;
8d1d8f83 82static int num_preferred_base;
dc6a0757 83static struct progress *progress_state;
960ccca6
DH
84static int pack_compression_level = Z_DEFAULT_COMPRESSION;
85static int pack_compression_seen;
c323ac7d 86
074b2eea
MK
87static unsigned long delta_cache_size = 0;
88static unsigned long max_delta_cache_size = 0;
e3dfddb3 89static unsigned long cache_max_small_delta_size = 1000;
074b2eea 90
a97773ce
BD
91static unsigned long window_memory_limit = 0;
92
3f9ac8d2
JH
93/*
94 * The object names in objects array are hashed with this hashtable,
e4d58311 95 * to help looking up the entry by object name.
3f9ac8d2
JH
96 * This hashtable is built after all the objects are seen.
97 */
96f1e58f
DR
98static int *object_ix;
99static int object_ix_hashsz;
3f9ac8d2 100
3f9ac8d2
JH
101/*
102 * stats
103 */
7cadf491
SP
104static uint32_t written, written_delta;
105static uint32_t reused, reused_delta;
3f9ac8d2 106
780e6e73 107
3613f9b4 108static void *get_delta(struct object_entry *entry)
c323ac7d 109{
3613f9b4
NP
110 unsigned long size, base_size, delta_size;
111 void *buf, *base_buf, *delta_buf;
21666f1a 112 enum object_type type;
c323ac7d 113
3613f9b4
NP
114 buf = read_sha1_file(entry->idx.sha1, &type, &size);
115 if (!buf)
116 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
117 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
118 if (!base_buf)
aa7e44bf 119 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
3613f9b4 120 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 121 buf, size, &delta_size, 0);
3613f9b4 122 if (!delta_buf || delta_size != entry->delta_size)
a6080a0a 123 die("delta size changed");
3613f9b4
NP
124 free(buf);
125 free(base_buf);
c323ac7d
LT
126 return delta_buf;
127}
128
30ebb40a
NP
129static unsigned long do_compress(void **pptr, unsigned long size)
130{
131 z_stream stream;
132 void *in, *out;
133 unsigned long maxsize;
134
135 memset(&stream, 0, sizeof(stream));
136 deflateInit(&stream, pack_compression_level);
137 maxsize = deflateBound(&stream, size);
138
139 in = *pptr;
140 out = xmalloc(maxsize);
141 *pptr = out;
142
143 stream.next_in = in;
144 stream.avail_in = size;
145 stream.next_out = out;
146 stream.avail_out = maxsize;
147 while (deflate(&stream, Z_FINISH) == Z_OK)
148 ; /* nothing */
149 deflateEnd(&stream);
150
151 free(in);
152 return stream.total_out;
153}
154
a733cb60
LT
155/*
156 * The per-object header is a pretty dense thing, which is
157 * - first byte: low four bits are "size", then three bits of "type",
158 * and the high bit is "size continues".
159 * - each byte afterwards: low seven bits are size continuation,
160 * with the high bit being "size continues"
161 */
162static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
163{
01247d87 164 int n = 1;
a733cb60
LT
165 unsigned char c;
166
eb32d236 167 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
a733cb60
LT
168 die("bad type %d", type);
169
01247d87
LT
170 c = (type << 4) | (size & 15);
171 size >>= 4;
172 while (size) {
a733cb60 173 *hdr++ = c | 0x80;
01247d87
LT
174 c = size & 0x7f;
175 size >>= 7;
176 n++;
a733cb60
LT
177 }
178 *hdr = c;
179 return n;
180}
181
780e6e73
NP
182/*
183 * we are going to reuse the existing object data as is. make
184 * sure it is not corrupt.
185 */
079afb18
SP
186static int check_pack_inflate(struct packed_git *p,
187 struct pack_window **w_curs,
6777a59f
SP
188 off_t offset,
189 off_t len,
079afb18
SP
190 unsigned long expect)
191{
192 z_stream stream;
193 unsigned char fakebuf[4096], *in;
194 int st;
195
196 memset(&stream, 0, sizeof(stream));
197 inflateInit(&stream);
198 do {
199 in = use_pack(p, w_curs, offset, &stream.avail_in);
200 stream.next_in = in;
201 stream.next_out = fakebuf;
202 stream.avail_out = sizeof(fakebuf);
203 st = inflate(&stream, Z_FINISH);
204 offset += stream.next_in - in;
205 } while (st == Z_OK || st == Z_BUF_ERROR);
206 inflateEnd(&stream);
207 return (st == Z_STREAM_END &&
208 stream.total_out == expect &&
209 stream.total_in == len) ? 0 : -1;
210}
211
212static void copy_pack_data(struct sha1file *f,
213 struct packed_git *p,
214 struct pack_window **w_curs,
6777a59f
SP
215 off_t offset,
216 off_t len)
079afb18
SP
217{
218 unsigned char *in;
219 unsigned int avail;
220
221 while (len) {
222 in = use_pack(p, w_curs, offset, &avail);
223 if (avail > len)
6777a59f 224 avail = (unsigned int)len;
079afb18
SP
225 sha1write(f, in, avail);
226 offset += avail;
227 len -= avail;
228 }
229}
230
d7dd0223 231static unsigned long write_object(struct sha1file *f,
17b08f2c
DH
232 struct object_entry *entry,
233 off_t write_offset)
c323ac7d 234{
30ebb40a 235 unsigned long size, limit, datalen;
3f9ac8d2 236 void *buf;
2c5ef824 237 unsigned char header[10], dheader[10];
6777a59f 238 unsigned hdrlen;
2c5ef824
NP
239 enum object_type type;
240 int usable_delta, to_reuse;
c323ac7d 241
78d1e84f
NP
242 if (!pack_to_stdout)
243 crc32_begin(f);
244
2c5ef824
NP
245 type = entry->type;
246
247 /* write limit if limited packsize and not first object */
248 limit = pack_size_limit && nr_written ?
249 pack_size_limit - write_offset : 0;
250
251 if (!entry->delta)
252 usable_delta = 0; /* no delta */
253 else if (!pack_size_limit)
254 usable_delta = 1; /* unlimited packfile */
255 else if (entry->delta->idx.offset == (off_t)-1)
256 usable_delta = 0; /* base was written to another pack */
257 else if (entry->delta->idx.offset)
258 usable_delta = 1; /* base already exists in this pack */
259 else
260 usable_delta = 0; /* base could end up in another pack */
261
a7de7130 262 if (!reuse_object)
fa736f72
NP
263 to_reuse = 0; /* explicit */
264 else if (!entry->in_pack)
ab7cd7bb 265 to_reuse = 0; /* can't reuse what we don't have */
2c5ef824 266 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
17b08f2c
DH
267 /* check_object() decided it for us ... */
268 to_reuse = usable_delta;
269 /* ... but pack split may override that */
2c5ef824 270 else if (type != entry->in_pack_type)
ab7cd7bb
JH
271 to_reuse = 0; /* pack has delta which is unusable */
272 else if (entry->delta)
273 to_reuse = 0; /* we want to pack afresh */
274 else
275 to_reuse = 1; /* we have it in-pack undeltified,
276 * and we do not need to deltify it.
277 */
278
df6d6101 279 if (!to_reuse) {
5476a8ad 280 if (!usable_delta) {
2c5ef824 281 buf = read_sha1_file(entry->idx.sha1, &type, &size);
5476a8ad 282 if (!buf)
aa7e44bf 283 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
ed4a9031
NP
284 /*
285 * make sure no cached delta data remains from a
286 * previous attempt before a pack split occured.
287 */
288 free(entry->delta_data);
289 entry->delta_data = NULL;
290 entry->z_delta_size = 0;
5476a8ad 291 } else if (entry->delta_data) {
3f9ac8d2 292 size = entry->delta_size;
5476a8ad
NP
293 buf = entry->delta_data;
294 entry->delta_data = NULL;
2c5ef824 295 type = (allow_ofs_delta && entry->delta->idx.offset) ?
be6b1914 296 OBJ_OFS_DELTA : OBJ_REF_DELTA;
17b08f2c 297 } else {
3613f9b4 298 buf = get_delta(entry);
5476a8ad 299 size = entry->delta_size;
2c5ef824 300 type = (allow_ofs_delta && entry->delta->idx.offset) ?
5476a8ad 301 OBJ_OFS_DELTA : OBJ_REF_DELTA;
3f9ac8d2 302 }
ed4a9031
NP
303
304 if (entry->z_delta_size)
305 datalen = entry->z_delta_size;
306 else
307 datalen = do_compress(&buf, size);
8c3c7b2a 308
3f9ac8d2
JH
309 /*
310 * The object header is a byte of 'type' followed by zero or
be6b1914 311 * more bytes of length.
3f9ac8d2 312 */
2c5ef824 313 hdrlen = encode_header(type, size, header);
3f9ac8d2 314
2c5ef824 315 if (type == OBJ_OFS_DELTA) {
be6b1914
NP
316 /*
317 * Deltas with relative base contain an additional
318 * encoding of the relative offset for the delta
319 * base from this object's position in the pack.
320 */
aa7e44bf 321 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
322 unsigned pos = sizeof(dheader) - 1;
323 dheader[pos] = ofs & 127;
be6b1914 324 while (ofs >>= 7)
17b08f2c
DH
325 dheader[--pos] = 128 | (--ofs & 127);
326 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
17b08f2c
DH
327 free(buf);
328 return 0;
329 }
330 sha1write(f, header, hdrlen);
331 sha1write(f, dheader + pos, sizeof(dheader) - pos);
332 hdrlen += sizeof(dheader) - pos;
2c5ef824 333 } else if (type == OBJ_REF_DELTA) {
be6b1914
NP
334 /*
335 * Deltas with a base reference contain
336 * an additional 20 bytes for the base sha1.
337 */
17b08f2c 338 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
17b08f2c
DH
339 free(buf);
340 return 0;
341 }
342 sha1write(f, header, hdrlen);
aa7e44bf 343 sha1write(f, entry->delta->idx.sha1, 20);
3f9ac8d2 344 hdrlen += 20;
17b08f2c
DH
345 } else {
346 if (limit && hdrlen + datalen + 20 >= limit) {
17b08f2c
DH
347 free(buf);
348 return 0;
349 }
350 sha1write(f, header, hdrlen);
3f9ac8d2 351 }
30ebb40a 352 sha1write(f, buf, datalen);
3f9ac8d2 353 free(buf);
c323ac7d 354 }
3f9ac8d2
JH
355 else {
356 struct packed_git *p = entry->in_pack;
03e79c88 357 struct pack_window *w_curs = NULL;
91ecbeca 358 struct revindex_entry *revidx;
6777a59f 359 off_t offset;
3f9ac8d2 360
780e6e73 361 if (entry->delta) {
2c5ef824 362 type = (allow_ofs_delta && entry->delta->idx.offset) ?
780e6e73
NP
363 OBJ_OFS_DELTA : OBJ_REF_DELTA;
364 reused_delta++;
365 }
2c5ef824 366 hdrlen = encode_header(type, entry->size, header);
91ecbeca 367 offset = entry->in_pack_offset;
3449f8c4 368 revidx = find_pack_revindex(p, offset);
91ecbeca
NP
369 datalen = revidx[1].offset - offset;
370 if (!pack_to_stdout && p->index_version > 1 &&
371 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
aa7e44bf 372 die("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
91ecbeca
NP
373 offset += entry->in_pack_header_size;
374 datalen -= entry->in_pack_header_size;
2c5ef824 375 if (type == OBJ_OFS_DELTA) {
aa7e44bf 376 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
377 unsigned pos = sizeof(dheader) - 1;
378 dheader[pos] = ofs & 127;
379 while (ofs >>= 7)
380 dheader[--pos] = 128 | (--ofs & 127);
381 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
382 return 0;
383 sha1write(f, header, hdrlen);
384 sha1write(f, dheader + pos, sizeof(dheader) - pos);
385 hdrlen += sizeof(dheader) - pos;
2c5ef824 386 } else if (type == OBJ_REF_DELTA) {
17b08f2c
DH
387 if (limit && hdrlen + 20 + datalen + 20 >= limit)
388 return 0;
389 sha1write(f, header, hdrlen);
aa7e44bf 390 sha1write(f, entry->delta->idx.sha1, 20);
17b08f2c
DH
391 hdrlen += 20;
392 } else {
393 if (limit && hdrlen + datalen + 20 >= limit)
394 return 0;
395 sha1write(f, header, hdrlen);
396 }
397
91ecbeca
NP
398 if (!pack_to_stdout && p->index_version == 1 &&
399 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
aa7e44bf 400 die("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
079afb18 401 copy_pack_data(f, p, &w_curs, offset, datalen);
03e79c88 402 unuse_pack(&w_curs);
3f9ac8d2 403 reused++;
a733cb60 404 }
17b08f2c 405 if (usable_delta)
ab7cd7bb 406 written_delta++;
3f9ac8d2 407 written++;
78d1e84f 408 if (!pack_to_stdout)
aa7e44bf 409 entry->idx.crc32 = crc32_end(f);
c323ac7d
LT
410 return hdrlen + datalen;
411}
412
6ed7f25e 413static int write_one(struct sha1file *f,
9d5ab962 414 struct object_entry *e,
6ed7f25e 415 off_t *offset)
9d5ab962 416{
d7dd0223
NP
417 unsigned long size;
418
419 /* offset is non zero if object is written already. */
aa7e44bf 420 if (e->idx.offset || e->preferred_base)
6ed7f25e 421 return 1;
d7dd0223
NP
422
423 /* if we are deltified, write out base object first. */
6ed7f25e
NP
424 if (e->delta && !write_one(f, e->delta, offset))
425 return 0;
d7dd0223 426
6ed7f25e
NP
427 e->idx.offset = *offset;
428 size = write_object(f, e, *offset);
17b08f2c 429 if (!size) {
aa7e44bf 430 e->idx.offset = 0;
17b08f2c
DH
431 return 0;
432 }
79814f42 433 written_list[nr_written++] = &e->idx;
d7dd0223
NP
434
435 /* make sure off_t is sufficiently large not to wrap */
6ed7f25e 436 if (*offset > *offset + size)
d7dd0223 437 die("pack too large for current definition of off_t");
6ed7f25e
NP
438 *offset += size;
439 return 1;
9d5ab962
JH
440}
441
aa7e44bf 442/* forward declaration for write_pack_file */
d01fb92f
DH
443static int adjust_perm(const char *path, mode_t mode);
444
445static void write_pack_file(void)
c323ac7d 446{
ebe27b13 447 uint32_t i = 0, j;
d22b9290 448 struct sha1file *f;
6ed7f25e 449 off_t offset;
a733cb60 450 struct pack_header hdr;
ebe27b13 451 uint32_t nr_remaining = nr_result;
f746bae8 452 time_t last_mtime = 0;
81a216a5 453
bcd7954e 454 if (progress > pack_to_stdout)
dc6a0757 455 progress_state = start_progress("Writing objects", nr_result);
79814f42 456 written_list = xmalloc(nr_objects * sizeof(*written_list));
183bdb2c 457
ebe27b13 458 do {
aa7e44bf 459 unsigned char sha1[20];
7ba502c4 460 char *pack_tmp_name = NULL;
aa7e44bf 461
ebe27b13 462 if (pack_to_stdout) {
2a128d63 463 f = sha1fd_throughput(1, "<stdout>", progress_state);
ebe27b13 464 } else {
7ba502c4
NP
465 char tmpname[PATH_MAX];
466 int fd;
467 snprintf(tmpname, sizeof(tmpname),
8b4eb6b6 468 "%s/pack/tmp_pack_XXXXXX", get_object_directory());
7ba502c4 469 fd = xmkstemp(tmpname);
ebe27b13
DH
470 pack_tmp_name = xstrdup(tmpname);
471 f = sha1fd(fd, pack_tmp_name);
472 }
473
474 hdr.hdr_signature = htonl(PACK_SIGNATURE);
475 hdr.hdr_version = htonl(PACK_VERSION);
476 hdr.hdr_entries = htonl(nr_remaining);
477 sha1write(f, &hdr, sizeof(hdr));
478 offset = sizeof(hdr);
479 nr_written = 0;
480 for (; i < nr_objects; i++) {
6ed7f25e 481 if (!write_one(f, objects + i, &offset))
ebe27b13 482 break;
4d4fcc54 483 display_progress(progress_state, written);
ebe27b13 484 }
c553ca25 485
ebe27b13
DH
486 /*
487 * Did we write the wrong # entries in the header?
488 * If so, rewrite it like in fast-import
489 */
54352bb2
LT
490 if (pack_to_stdout) {
491 sha1close(f, sha1, CSUM_CLOSE);
492 } else if (nr_written == nr_remaining) {
493 sha1close(f, sha1, CSUM_FSYNC);
ebe27b13 494 } else {
ac0463ed 495 int fd = sha1close(f, sha1, 0);
abeb40e5 496 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
ac0463ed 497 nr_written, sha1, offset);
7ba502c4 498 close(fd);
ebe27b13
DH
499 }
500
501 if (!pack_to_stdout) {
d01fb92f 502 mode_t mode = umask(0);
f746bae8 503 struct stat st;
7ba502c4 504 char *idx_tmp_name, tmpname[PATH_MAX];
d01fb92f
DH
505
506 umask(mode);
507 mode = 0444 & ~mode;
508
79814f42
NP
509 idx_tmp_name = write_idx_file(NULL, written_list,
510 nr_written, sha1);
f746bae8 511
d01fb92f 512 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
aa7e44bf 513 base_name, sha1_to_hex(sha1));
d01fb92f
DH
514 if (adjust_perm(pack_tmp_name, mode))
515 die("unable to make temporary pack file readable: %s",
516 strerror(errno));
517 if (rename(pack_tmp_name, tmpname))
518 die("unable to rename temporary pack file: %s",
519 strerror(errno));
f746bae8
NP
520
521 /*
522 * Packs are runtime accessed in their mtime
523 * order since newer packs are more likely to contain
524 * younger objects. So if we are creating multiple
525 * packs then we should modify the mtime of later ones
526 * to preserve this property.
527 */
528 if (stat(tmpname, &st) < 0) {
529 warning("failed to stat %s: %s",
530 tmpname, strerror(errno));
531 } else if (!last_mtime) {
532 last_mtime = st.st_mtime;
533 } else {
534 struct utimbuf utb;
535 utb.actime = st.st_atime;
536 utb.modtime = --last_mtime;
537 if (utime(tmpname, &utb) < 0)
538 warning("failed utime() on %s: %s",
539 tmpname, strerror(errno));
540 }
541
d01fb92f 542 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
aa7e44bf 543 base_name, sha1_to_hex(sha1));
d01fb92f
DH
544 if (adjust_perm(idx_tmp_name, mode))
545 die("unable to make temporary index file readable: %s",
546 strerror(errno));
547 if (rename(idx_tmp_name, tmpname))
548 die("unable to rename temporary index file: %s",
549 strerror(errno));
f746bae8 550
7ba502c4
NP
551 free(idx_tmp_name);
552 free(pack_tmp_name);
aa7e44bf 553 puts(sha1_to_hex(sha1));
ebe27b13
DH
554 }
555
556 /* mark written objects as written to previous pack */
557 for (j = 0; j < nr_written; j++) {
79814f42 558 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
559 }
560 nr_remaining -= nr_written;
561 } while (nr_remaining && i < nr_objects);
562
563 free(written_list);
4d4fcc54 564 stop_progress(&progress_state);
67c08ce1 565 if (written != nr_result)
6e1c2344
RJ
566 die("wrote %"PRIu32" objects while expecting %"PRIu32,
567 written, nr_result);
01c12a23
DH
568 /*
569 * We have scanned through [0 ... i). Since we have written
570 * the correct number of objects, the remaining [i ... nr_objects)
571 * items must be either already written (due to out-of-order delta base)
572 * or a preferred base. Count those which are neither and complain if any.
573 */
ebe27b13
DH
574 for (j = 0; i < nr_objects; i++) {
575 struct object_entry *e = objects + i;
aa7e44bf 576 j += !e->idx.offset && !e->preferred_base;
d01fb92f 577 }
ebe27b13 578 if (j)
6e1c2344
RJ
579 die("wrote %"PRIu32" objects as expected but %"PRIu32
580 " unwritten", written, j);
c323ac7d
LT
581}
582
7a979d99
JH
583static int locate_object_entry_hash(const unsigned char *sha1)
584{
585 int i;
586 unsigned int ui;
587 memcpy(&ui, sha1, sizeof(unsigned int));
588 i = ui % object_ix_hashsz;
589 while (0 < object_ix[i]) {
aa7e44bf 590 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
7a979d99
JH
591 return i;
592 if (++i == object_ix_hashsz)
593 i = 0;
594 }
595 return -1 - i;
596}
597
598static struct object_entry *locate_object_entry(const unsigned char *sha1)
599{
600 int i;
601
602 if (!object_ix_hashsz)
603 return NULL;
604
605 i = locate_object_entry_hash(sha1);
606 if (0 <= i)
607 return &objects[object_ix[i]-1];
608 return NULL;
609}
610
611static void rehash_objects(void)
c323ac7d 612{
7cadf491 613 uint32_t i;
7a979d99
JH
614 struct object_entry *oe;
615
616 object_ix_hashsz = nr_objects * 3;
617 if (object_ix_hashsz < 1024)
618 object_ix_hashsz = 1024;
619 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 620 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99 621 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
aa7e44bf 622 int ix = locate_object_entry_hash(oe->idx.sha1);
7a979d99
JH
623 if (0 <= ix)
624 continue;
625 ix = -1 - ix;
626 object_ix[ix] = i + 1;
627 }
628}
629
ce0bd642 630static unsigned name_hash(const char *name)
1d6b38cc 631{
ce0bd642
LT
632 unsigned char c;
633 unsigned hash = 0;
634
bc32fed5
JH
635 if (!name)
636 return 0;
637
eeef7135 638 /*
ce0bd642
LT
639 * This effectively just creates a sortable number from the
640 * last sixteen non-whitespace characters. Last characters
641 * count "most", so things that end in ".c" sort together.
eeef7135 642 */
ce0bd642
LT
643 while ((c = *name++) != 0) {
644 if (isspace(c))
645 continue;
646 hash = (hash >> 2) + (c << 24);
647 }
1d6b38cc
JH
648 return hash;
649}
650
a74db82e
JH
651static void setup_delta_attr_check(struct git_attr_check *check)
652{
653 static struct git_attr *attr_delta;
654
655 if (!attr_delta)
656 attr_delta = git_attr("delta", 5);
657
658 check[0].attr = attr_delta;
659}
660
661static int no_try_delta(const char *path)
662{
663 struct git_attr_check check[1];
664
665 setup_delta_attr_check(check);
666 if (git_checkattr(path, ARRAY_SIZE(check), check))
667 return 0;
668 if (ATTR_FALSE(check->value))
669 return 1;
670 return 0;
671}
672
5c49c116 673static int add_object_entry(const unsigned char *sha1, enum object_type type,
bc32fed5 674 const char *name, int exclude)
c323ac7d 675{
c323ac7d 676 struct object_entry *entry;
29b734e4 677 struct packed_git *p, *found_pack = NULL;
6777a59f 678 off_t found_offset = 0;
29b734e4 679 int ix;
bc32fed5 680 unsigned hash = name_hash(name);
29b734e4
NP
681
682 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
683 if (ix >= 0) {
684 if (exclude) {
685 entry = objects + object_ix[ix] - 1;
81a216a5
NP
686 if (!entry->preferred_base)
687 nr_result--;
29b734e4
NP
688 entry->preferred_base = 1;
689 }
690 return 0;
691 }
c323ac7d 692
5c49c116
NP
693 for (p = packed_git; p; p = p->next) {
694 off_t offset = find_pack_entry_one(sha1, p);
695 if (offset) {
696 if (!found_pack) {
697 found_offset = offset;
698 found_pack = p;
64560374 699 }
5c49c116
NP
700 if (exclude)
701 break;
702 if (incremental)
703 return 0;
704 if (local && !p->pack_local)
705 return 0;
64560374
LT
706 }
707 }
eb019375 708
29b734e4
NP
709 if (nr_objects >= nr_alloc) {
710 nr_alloc = (nr_alloc + 1024) * 3 / 2;
7cadf491 711 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d 712 }
29b734e4
NP
713
714 entry = objects + nr_objects++;
c323ac7d 715 memset(entry, 0, sizeof(*entry));
aa7e44bf 716 hashcpy(entry->idx.sha1, sha1);
27225f2e 717 entry->hash = hash;
5c49c116
NP
718 if (type)
719 entry->type = type;
29b734e4
NP
720 if (exclude)
721 entry->preferred_base = 1;
81a216a5
NP
722 else
723 nr_result++;
29b734e4
NP
724 if (found_pack) {
725 entry->in_pack = found_pack;
726 entry->in_pack_offset = found_offset;
727 }
7a979d99
JH
728
729 if (object_ix_hashsz * 3 <= nr_objects * 4)
730 rehash_objects();
29b734e4
NP
731 else
732 object_ix[-1 - ix] = nr_objects;
7a979d99 733
4d4fcc54 734 display_progress(progress_state, nr_objects);
29b734e4 735
a74db82e
JH
736 if (name && no_try_delta(name))
737 entry->no_try_delta = 1;
738
29b734e4 739 return 1;
c323ac7d
LT
740}
741
5379a5c5
JH
742struct pbase_tree_cache {
743 unsigned char sha1[20];
744 int ref;
745 int temporary;
746 void *tree_data;
747 unsigned long tree_size;
748};
749
750static struct pbase_tree_cache *(pbase_tree_cache[256]);
751static int pbase_tree_cache_ix(const unsigned char *sha1)
752{
753 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
754}
755static int pbase_tree_cache_ix_incr(int ix)
756{
757 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
758}
759
760static struct pbase_tree {
761 struct pbase_tree *next;
762 /* This is a phony "cache" entry; we are not
763 * going to evict it nor find it through _get()
764 * mechanism -- this is for the toplevel node that
765 * would almost always change with any commit.
766 */
767 struct pbase_tree_cache pcache;
768} *pbase_tree;
769
770static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
771{
772 struct pbase_tree_cache *ent, *nent;
773 void *data;
774 unsigned long size;
21666f1a 775 enum object_type type;
5379a5c5
JH
776 int neigh;
777 int my_ix = pbase_tree_cache_ix(sha1);
778 int available_ix = -1;
779
780 /* pbase-tree-cache acts as a limited hashtable.
781 * your object will be found at your index or within a few
782 * slots after that slot if it is cached.
783 */
784 for (neigh = 0; neigh < 8; neigh++) {
785 ent = pbase_tree_cache[my_ix];
a89fccd2 786 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
787 ent->ref++;
788 return ent;
789 }
790 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
791 ((0 <= available_ix) &&
792 (!ent && pbase_tree_cache[available_ix])))
793 available_ix = my_ix;
794 if (!ent)
795 break;
796 my_ix = pbase_tree_cache_ix_incr(my_ix);
797 }
798
799 /* Did not find one. Either we got a bogus request or
800 * we need to read and perhaps cache.
801 */
21666f1a 802 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
803 if (!data)
804 return NULL;
21666f1a 805 if (type != OBJ_TREE) {
5379a5c5
JH
806 free(data);
807 return NULL;
808 }
809
810 /* We need to either cache or return a throwaway copy */
811
812 if (available_ix < 0)
813 ent = NULL;
814 else {
815 ent = pbase_tree_cache[available_ix];
816 my_ix = available_ix;
817 }
818
819 if (!ent) {
820 nent = xmalloc(sizeof(*nent));
821 nent->temporary = (available_ix < 0);
822 }
823 else {
824 /* evict and reuse */
825 free(ent->tree_data);
826 nent = ent;
827 }
e702496e 828 hashcpy(nent->sha1, sha1);
5379a5c5
JH
829 nent->tree_data = data;
830 nent->tree_size = size;
831 nent->ref = 1;
832 if (!nent->temporary)
833 pbase_tree_cache[my_ix] = nent;
834 return nent;
835}
836
837static void pbase_tree_put(struct pbase_tree_cache *cache)
838{
839 if (!cache->temporary) {
840 cache->ref--;
841 return;
842 }
843 free(cache->tree_data);
844 free(cache);
845}
846
847static int name_cmp_len(const char *name)
848{
849 int i;
850 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
851 ;
852 return i;
853}
854
855static void add_pbase_object(struct tree_desc *tree,
5379a5c5 856 const char *name,
ce0bd642
LT
857 int cmplen,
858 const char *fullname)
3f9ac8d2 859{
4c068a98 860 struct name_entry entry;
8a5a8d6c 861 int cmp;
4c068a98
LT
862
863 while (tree_entry(tree,&entry)) {
1211be6b
LT
864 if (S_ISGITLINK(entry.mode))
865 continue;
8a5a8d6c
NP
866 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
867 memcmp(name, entry.path, cmplen);
868 if (cmp > 0)
7a979d99 869 continue;
8a5a8d6c
NP
870 if (cmp < 0)
871 return;
5379a5c5 872 if (name[cmplen] != '/') {
5c49c116 873 add_object_entry(entry.sha1,
4d1012c3 874 object_type(entry.mode),
bc32fed5 875 fullname, 1);
5379a5c5
JH
876 return;
877 }
8a5a8d6c 878 if (S_ISDIR(entry.mode)) {
7a979d99 879 struct tree_desc sub;
5379a5c5
JH
880 struct pbase_tree_cache *tree;
881 const char *down = name+cmplen+1;
882 int downlen = name_cmp_len(down);
883
4c068a98 884 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
885 if (!tree)
886 return;
6fda5e51 887 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 888
ce0bd642 889 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
890 pbase_tree_put(tree);
891 }
892 }
893}
1d6b38cc 894
5379a5c5
JH
895static unsigned *done_pbase_paths;
896static int done_pbase_paths_num;
897static int done_pbase_paths_alloc;
898static int done_pbase_path_pos(unsigned hash)
899{
900 int lo = 0;
901 int hi = done_pbase_paths_num;
902 while (lo < hi) {
903 int mi = (hi + lo) / 2;
904 if (done_pbase_paths[mi] == hash)
905 return mi;
906 if (done_pbase_paths[mi] < hash)
907 hi = mi;
908 else
909 lo = mi + 1;
910 }
911 return -lo-1;
912}
913
914static int check_pbase_path(unsigned hash)
915{
916 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
917 if (0 <= pos)
918 return 1;
919 pos = -pos - 1;
920 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
921 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
922 done_pbase_paths = xrealloc(done_pbase_paths,
923 done_pbase_paths_alloc *
924 sizeof(unsigned));
925 }
926 done_pbase_paths_num++;
927 if (pos < done_pbase_paths_num)
928 memmove(done_pbase_paths + pos + 1,
929 done_pbase_paths + pos,
930 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
931 done_pbase_paths[pos] = hash;
932 return 0;
933}
934
bc32fed5 935static void add_preferred_base_object(const char *name)
5379a5c5
JH
936{
937 struct pbase_tree *it;
8a5a8d6c 938 int cmplen;
bc32fed5 939 unsigned hash = name_hash(name);
5379a5c5 940
8a5a8d6c 941 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
942 return;
943
8a5a8d6c 944 cmplen = name_cmp_len(name);
5379a5c5
JH
945 for (it = pbase_tree; it; it = it->next) {
946 if (cmplen == 0) {
bc32fed5 947 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
948 }
949 else {
950 struct tree_desc tree;
6fda5e51 951 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 952 add_pbase_object(&tree, name, cmplen, name);
7a979d99 953 }
3f9ac8d2 954 }
3f9ac8d2
JH
955}
956
7a979d99 957static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 958{
5379a5c5
JH
959 struct pbase_tree *it;
960 void *data;
961 unsigned long size;
962 unsigned char tree_sha1[20];
1d6b38cc 963
8d1d8f83
JH
964 if (window <= num_preferred_base++)
965 return;
966
5379a5c5
JH
967 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
968 if (!data)
7a979d99 969 return;
5379a5c5
JH
970
971 for (it = pbase_tree; it; it = it->next) {
a89fccd2 972 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
973 free(data);
974 return;
975 }
976 }
977
978 it = xcalloc(1, sizeof(*it));
979 it->next = pbase_tree;
980 pbase_tree = it;
981
e702496e 982 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
983 it->pcache.tree_data = data;
984 it->pcache.tree_size = size;
3f9ac8d2
JH
985}
986
c323ac7d
LT
987static void check_object(struct object_entry *entry)
988{
5c49c116 989 if (entry->in_pack) {
780e6e73 990 struct packed_git *p = entry->in_pack;
03e79c88 991 struct pack_window *w_curs = NULL;
5c49c116
NP
992 const unsigned char *base_ref = NULL;
993 struct object_entry *base_entry;
994 unsigned long used, used_0;
6777a59f 995 unsigned int avail;
5c49c116
NP
996 off_t ofs;
997 unsigned char *buf, c;
780e6e73 998
6777a59f 999 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1000
5c49c116 1001 /*
fa736f72
NP
1002 * We want in_pack_type even if we do not reuse delta
1003 * since non-delta representations could still be reused.
ab7cd7bb 1004 */
09ded04b 1005 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1006 &entry->in_pack_type,
1007 &entry->size);
ab7cd7bb 1008
5c49c116
NP
1009 /*
1010 * Determine if this is a delta and if so whether we can
1011 * reuse it or not. Otherwise let's find out as cheaply as
1012 * possible what the actual type and size for this object is.
3f9ac8d2 1013 */
5c49c116
NP
1014 switch (entry->in_pack_type) {
1015 default:
1016 /* Not a delta hence we've already got all we need. */
1017 entry->type = entry->in_pack_type;
1018 entry->in_pack_header_size = used;
1019 unuse_pack(&w_curs);
1020 return;
1021 case OBJ_REF_DELTA:
a7de7130 1022 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1023 base_ref = use_pack(p, &w_curs,
1024 entry->in_pack_offset + used, NULL);
1025 entry->in_pack_header_size = used + 20;
1026 break;
1027 case OBJ_OFS_DELTA:
1028 buf = use_pack(p, &w_curs,
1029 entry->in_pack_offset + used, NULL);
1030 used_0 = 0;
1031 c = buf[used_0++];
1032 ofs = c & 127;
1033 while (c & 128) {
1034 ofs += 1;
1035 if (!ofs || MSB(ofs, 7))
1036 die("delta base offset overflow in pack for %s",
aa7e44bf 1037 sha1_to_hex(entry->idx.sha1));
5c49c116
NP
1038 c = buf[used_0++];
1039 ofs = (ofs << 7) + (c & 127);
780e6e73 1040 }
d8f32556
NP
1041 ofs = entry->in_pack_offset - ofs;
1042 if (ofs <= 0 || ofs >= entry->in_pack_offset)
5c49c116 1043 die("delta base offset out of bound for %s",
aa7e44bf 1044 sha1_to_hex(entry->idx.sha1));
a7de7130 1045 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1046 struct revindex_entry *revidx;
1047 revidx = find_pack_revindex(p, ofs);
1048 base_ref = nth_packed_object_sha1(p, revidx->nr);
1049 }
5c49c116
NP
1050 entry->in_pack_header_size = used + used_0;
1051 break;
780e6e73 1052 }
780e6e73 1053
5c49c116
NP
1054 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1055 /*
1056 * If base_ref was set above that means we wish to
1057 * reuse delta data, and we even found that base
1058 * in the list of objects we want to pack. Goodie!
1059 *
1060 * Depth value does not matter - find_deltas() will
1061 * never consider reused delta as the base object to
1062 * deltify other objects against, in order to avoid
1063 * circular deltas.
3f9ac8d2 1064 */
780e6e73 1065 entry->type = entry->in_pack_type;
5c49c116 1066 entry->delta = base_entry;
15b4d577
JH
1067 entry->delta_sibling = base_entry->delta_child;
1068 base_entry->delta_child = entry;
5c49c116
NP
1069 unuse_pack(&w_curs);
1070 return;
1071 }
ab7cd7bb 1072
5c49c116
NP
1073 if (entry->type) {
1074 /*
1075 * This must be a delta and we already know what the
1076 * final object type is. Let's extract the actual
1077 * object size from the delta header.
1078 */
1079 entry->size = get_size_from_delta(p, &w_curs,
1080 entry->in_pack_offset + entry->in_pack_header_size);
1081 unuse_pack(&w_curs);
3f9ac8d2
JH
1082 return;
1083 }
5c49c116
NP
1084
1085 /*
1086 * No choice but to fall back to the recursive delta walk
1087 * with sha1_object_info() to find about the object type
1088 * at this point...
1089 */
1090 unuse_pack(&w_curs);
36e4d74a 1091 }
3f9ac8d2 1092
aa7e44bf 1093 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
6d6f9cdd
SP
1094 /*
1095 * The error condition is checked in prepare_pack(). This is
1096 * to permit a missing preferred base object to be ignored
1097 * as a preferred base. Doing so can result in a larger
1098 * pack file, but the transfer will still take place.
1099 */
3f9ac8d2
JH
1100}
1101
5c49c116
NP
1102static int pack_offset_sort(const void *_a, const void *_b)
1103{
1104 const struct object_entry *a = *(struct object_entry **)_a;
1105 const struct object_entry *b = *(struct object_entry **)_b;
1106
1107 /* avoid filesystem trashing with loose objects */
1108 if (!a->in_pack && !b->in_pack)
aa7e44bf 1109 return hashcmp(a->idx.sha1, b->idx.sha1);
5c49c116
NP
1110
1111 if (a->in_pack < b->in_pack)
1112 return -1;
1113 if (a->in_pack > b->in_pack)
1114 return 1;
1115 return a->in_pack_offset < b->in_pack_offset ? -1 :
1116 (a->in_pack_offset > b->in_pack_offset);
1117}
1118
c323ac7d
LT
1119static void get_object_details(void)
1120{
7cadf491 1121 uint32_t i;
5c49c116
NP
1122 struct object_entry **sorted_by_offset;
1123
1124 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1125 for (i = 0; i < nr_objects; i++)
1126 sorted_by_offset[i] = objects + i;
1127 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1128
5c49c116
NP
1129 for (i = 0; i < nr_objects; i++)
1130 check_object(sorted_by_offset[i]);
3449f8c4 1131
5c49c116 1132 free(sorted_by_offset);
c323ac7d
LT
1133}
1134
b904166c
NP
1135/*
1136 * We search for deltas in a list sorted by type, by filename hash, and then
1137 * by size, so that we see progressively smaller and smaller files.
1138 * That's because we prefer deltas to be from the bigger file
1139 * to the smaller -- deletes are potentially cheaper, but perhaps
1140 * more importantly, the bigger file is likely the more recent
1141 * one. The deepest deltas are therefore the oldest objects which are
1142 * less susceptible to be accessed often.
1143 */
9668cf59 1144static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1145{
9668cf59
NP
1146 const struct object_entry *a = *(struct object_entry **)_a;
1147 const struct object_entry *b = *(struct object_entry **)_b;
1148
c323ac7d 1149 if (a->type > b->type)
27225f2e 1150 return -1;
b904166c 1151 if (a->type < b->type)
27225f2e 1152 return 1;
b904166c 1153 if (a->hash > b->hash)
7a979d99 1154 return -1;
b904166c 1155 if (a->hash < b->hash)
7a979d99 1156 return 1;
b904166c 1157 if (a->preferred_base > b->preferred_base)
c323ac7d 1158 return -1;
b904166c
NP
1159 if (a->preferred_base < b->preferred_base)
1160 return 1;
c323ac7d 1161 if (a->size > b->size)
b904166c
NP
1162 return -1;
1163 if (a->size < b->size)
c323ac7d 1164 return 1;
b904166c 1165 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1166}
1167
1168struct unpacked {
1169 struct object_entry *entry;
1170 void *data;
f6c7081a 1171 struct delta_index *index;
5a235b5e 1172 unsigned depth;
c323ac7d
LT
1173};
1174
d250626c
NP
1175static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1176 unsigned long delta_size)
074b2eea
MK
1177{
1178 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1179 return 0;
1180
e3dfddb3
MK
1181 if (delta_size < cache_max_small_delta_size)
1182 return 1;
1183
074b2eea
MK
1184 /* cache delta, if objects are large enough compared to delta size */
1185 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1186 return 1;
1187
1188 return 0;
1189}
1190
8ecce684
NP
1191#ifdef THREADED_DELTA_SEARCH
1192
1193static pthread_mutex_t read_mutex = PTHREAD_MUTEX_INITIALIZER;
1194#define read_lock() pthread_mutex_lock(&read_mutex)
1195#define read_unlock() pthread_mutex_unlock(&read_mutex)
1196
3c701839
NP
1197static pthread_mutex_t cache_mutex = PTHREAD_MUTEX_INITIALIZER;
1198#define cache_lock() pthread_mutex_lock(&cache_mutex)
1199#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1200
8ecce684
NP
1201static pthread_mutex_t progress_mutex = PTHREAD_MUTEX_INITIALIZER;
1202#define progress_lock() pthread_mutex_lock(&progress_mutex)
1203#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1204
1205#else
1206
e1ef8673
JH
1207#define read_lock() (void)0
1208#define read_unlock() (void)0
1209#define cache_lock() (void)0
1210#define cache_unlock() (void)0
1211#define progress_lock() (void)0
1212#define progress_unlock() (void)0
8ecce684
NP
1213
1214#endif
1215
f6c7081a 1216static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1217 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1218{
f6c7081a
NP
1219 struct object_entry *trg_entry = trg->entry;
1220 struct object_entry *src_entry = src->entry;
560b25a8 1221 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1222 unsigned ref_depth;
21666f1a 1223 enum object_type type;
c323ac7d
LT
1224 void *delta_buf;
1225
1226 /* Don't bother doing diffs between different types */
f6c7081a 1227 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1228 return -1;
1229
51d1e83f
LT
1230 /*
1231 * We do not bother to try a delta that we discarded
8dbbd14e 1232 * on an earlier try, but only when reusing delta data.
51d1e83f 1233 */
a7de7130 1234 if (reuse_delta && trg_entry->in_pack &&
e9195b58
JH
1235 trg_entry->in_pack == src_entry->in_pack &&
1236 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1237 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1238 return 0;
1239
898b14ce 1240 /* Let's not bust the allowed depth. */
5a235b5e 1241 if (src->depth >= max_depth)
d116a45a 1242 return 0;
c323ac7d 1243
c3b06a69 1244 /* Now some size filtering heuristics. */
560b25a8 1245 trg_size = trg_entry->size;
c83f032e
NP
1246 if (!trg_entry->delta) {
1247 max_size = trg_size/2 - 20;
1248 ref_depth = 1;
1249 } else {
1250 max_size = trg_entry->delta_size;
5a235b5e 1251 ref_depth = trg->depth;
c83f032e 1252 }
5a235b5e 1253 max_size = max_size * (max_depth - src->depth) /
c83f032e 1254 (max_depth - ref_depth + 1);
c3b06a69
NP
1255 if (max_size == 0)
1256 return 0;
f6c7081a 1257 src_size = src_entry->size;
560b25a8 1258 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1259 if (sizediff >= max_size)
f527cb8c 1260 return 0;
a1dab41a
BD
1261 if (trg_size < src_size / 32)
1262 return 0;
f6c7081a 1263
560b25a8
NP
1264 /* Load data if not already done */
1265 if (!trg->data) {
8ecce684 1266 read_lock();
aa7e44bf 1267 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
8ecce684 1268 read_unlock();
2e3404c3
JH
1269 if (!trg->data)
1270 die("object %s cannot be read",
1271 sha1_to_hex(trg_entry->idx.sha1));
560b25a8
NP
1272 if (sz != trg_size)
1273 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1274 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
ef0316fc 1275 *mem_usage += sz;
560b25a8
NP
1276 }
1277 if (!src->data) {
8ecce684 1278 read_lock();
aa7e44bf 1279 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
8ecce684 1280 read_unlock();
2e3404c3
JH
1281 if (!src->data)
1282 die("object %s cannot be read",
1283 sha1_to_hex(src_entry->idx.sha1));
560b25a8
NP
1284 if (sz != src_size)
1285 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1286 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
ef0316fc 1287 *mem_usage += sz;
560b25a8
NP
1288 }
1289 if (!src->index) {
1290 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1291 if (!src->index) {
1292 static int warned = 0;
1293 if (!warned++)
1294 warning("suboptimal pack - out of memory");
1295 return 0;
1296 }
ef0316fc 1297 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1298 }
1299
1300 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1301 if (!delta_buf)
75c42d8c 1302 return 0;
f6c7081a 1303
9e2d57a0 1304 if (trg_entry->delta) {
848d732c
BD
1305 /* Prefer only shallower same-sized deltas. */
1306 if (delta_size == trg_entry->delta_size &&
5a235b5e 1307 src->depth + 1 >= trg->depth) {
848d732c
BD
1308 free(delta_buf);
1309 return 0;
1310 }
074b2eea 1311 }
9e2d57a0 1312
3c701839
NP
1313 /*
1314 * Handle memory allocation outside of the cache
1315 * accounting lock. Compiler will optimize the strangeness
1316 * away when THREADED_DELTA_SEARCH is not defined.
1317 */
8e0f7003 1318 free(trg_entry->delta_data);
3c701839 1319 cache_lock();
9e2d57a0
NP
1320 if (trg_entry->delta_data) {
1321 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1322 trg_entry->delta_data = NULL;
1323 }
d250626c 1324 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1325 delta_cache_size += delta_size;
3c701839
NP
1326 cache_unlock();
1327 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1328 } else {
1329 cache_unlock();
074b2eea 1330 free(delta_buf);
3c701839
NP
1331 }
1332
b7a28f78
NP
1333 trg_entry->delta = src_entry;
1334 trg_entry->delta_size = delta_size;
1335 trg->depth = src->depth + 1;
1336
f6c7081a 1337 return 1;
c323ac7d
LT
1338}
1339
898b14ce 1340static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1341{
898b14ce
NP
1342 struct object_entry *child = me->delta_child;
1343 unsigned int m = n;
1344 while (child) {
1345 unsigned int c = check_delta_limit(child, n + 1);
1346 if (m < c)
1347 m = c;
1348 child = child->delta_sibling;
1349 }
1350 return m;
b2504a0d
NP
1351}
1352
75ad235c 1353static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1354{
ef0316fc 1355 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1356 free_delta_index(n->index);
1357 n->index = NULL;
1358 if (n->data) {
ef0316fc 1359 freed_mem += n->entry->size;
a97773ce
BD
1360 free(n->data);
1361 n->data = NULL;
a97773ce
BD
1362 }
1363 n->entry = NULL;
7d7baa5e 1364 n->depth = 0;
ef0316fc 1365 return freed_mem;
a97773ce
BD
1366}
1367
384b32c0 1368static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1369 int window, int depth, unsigned *processed)
c323ac7d 1370{
384b32c0 1371 uint32_t i, idx = 0, count = 0;
7cadf491 1372 struct unpacked *array;
ef0316fc 1373 unsigned long mem_usage = 0;
c323ac7d 1374
19d4b416 1375 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 1376
384b32c0 1377 for (;;) {
421b488a 1378 struct object_entry *entry;
c323ac7d 1379 struct unpacked *n = array + idx;
ef0316fc 1380 int j, max_depth, best_base = -1;
c323ac7d 1381
384b32c0
NP
1382 progress_lock();
1383 if (!*list_size) {
1384 progress_unlock();
1385 break;
1386 }
421b488a 1387 entry = *list++;
384b32c0
NP
1388 (*list_size)--;
1389 if (!entry->preferred_base) {
1390 (*processed)++;
1391 display_progress(progress_state, *processed);
1392 }
1393 progress_unlock();
1394
ef0316fc 1395 mem_usage -= free_unpacked(n);
c323ac7d 1396 n->entry = entry;
ab7cd7bb 1397
a97773ce 1398 while (window_memory_limit &&
ef0316fc 1399 mem_usage > window_memory_limit &&
a97773ce
BD
1400 count > 1) {
1401 uint32_t tail = (idx + window - count) % window;
75ad235c 1402 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
1403 count--;
1404 }
1405
75d39853
NP
1406 /* We do not compute delta to *create* objects we are not
1407 * going to pack.
1408 */
1409 if (entry->preferred_base)
1410 goto next;
1411
898b14ce
NP
1412 /*
1413 * If the current object is at pack edge, take the depth the
1414 * objects that depend on the current object into account
1415 * otherwise they would become too deep.
1416 */
1417 max_depth = depth;
1418 if (entry->delta_child) {
1419 max_depth -= check_delta_limit(entry, 0);
1420 if (max_depth <= 0)
1421 goto next;
1422 }
1423
78817c15
LT
1424 j = window;
1425 while (--j > 0) {
77639870 1426 int ret;
7cadf491 1427 uint32_t other_idx = idx + j;
c323ac7d 1428 struct unpacked *m;
78817c15
LT
1429 if (other_idx >= window)
1430 other_idx -= window;
c323ac7d
LT
1431 m = array + other_idx;
1432 if (!m->entry)
1433 break;
ef0316fc 1434 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 1435 if (ret < 0)
c323ac7d 1436 break;
77639870
JH
1437 else if (ret > 0)
1438 best_base = other_idx;
c323ac7d 1439 }
898b14ce 1440
ed4a9031
NP
1441 /*
1442 * If we decided to cache the delta data, then it is best
1443 * to compress it right away. First because we have to do
1444 * it anyway, and doing it here while we're threaded will
1445 * save a lot of time in the non threaded write phase,
1446 * as well as allow for caching more deltas within
1447 * the same cache size limit.
1448 * ...
1449 * But only if not writing to stdout, since in that case
1450 * the network is most likely throttling writes anyway,
1451 * and therefore it is best to go to the write phase ASAP
1452 * instead, as we can afford spending more time compressing
1453 * between writes at that moment.
1454 */
1455 if (entry->delta_data && !pack_to_stdout) {
1456 entry->z_delta_size = do_compress(&entry->delta_data,
1457 entry->delta_size);
1458 cache_lock();
1459 delta_cache_size -= entry->delta_size;
1460 delta_cache_size += entry->z_delta_size;
1461 cache_unlock();
1462 }
1463
70ca1a3f
JH
1464 /* if we made n a delta, and if n is already at max
1465 * depth, leaving it in the window is pointless. we
1466 * should evict it first.
70ca1a3f 1467 */
70baf5d4 1468 if (entry->delta && max_depth <= n->depth)
70ca1a3f 1469 continue;
ff45715c 1470
77639870
JH
1471 /*
1472 * Move the best delta base up in the window, after the
1473 * currently deltified object, to keep it longer. It will
1474 * be the first base object to be attempted next.
1475 */
1476 if (entry->delta) {
1477 struct unpacked swap = array[best_base];
1478 int dist = (window + idx - best_base) % window;
1479 int dst = best_base;
1480 while (dist--) {
1481 int src = (dst + 1) % window;
1482 array[dst] = array[src];
1483 dst = src;
1484 }
1485 array[dst] = swap;
1486 }
1487
898b14ce 1488 next:
521a4f4c 1489 idx++;
a97773ce
BD
1490 if (count + 1 < window)
1491 count++;
521a4f4c
LT
1492 if (idx >= window)
1493 idx = 0;
384b32c0 1494 }
adee7bdf 1495
f6c7081a 1496 for (i = 0; i < window; ++i) {
ff45715c 1497 free_delta_index(array[i].index);
adee7bdf 1498 free(array[i].data);
f6c7081a 1499 }
adee7bdf 1500 free(array);
c323ac7d
LT
1501}
1502
8ecce684
NP
1503#ifdef THREADED_DELTA_SEARCH
1504
50f22ada
JS
1505/*
1506 * The main thread waits on the condition that (at least) one of the workers
1507 * has stopped working (which is indicated in the .working member of
1508 * struct thread_params).
1509 * When a work thread has completed its work, it sets .working to 0 and
1510 * signals the main thread and waits on the condition that .data_ready
1511 * becomes 1.
1512 */
1513
8ecce684
NP
1514struct thread_params {
1515 pthread_t thread;
1516 struct object_entry **list;
1517 unsigned list_size;
384b32c0 1518 unsigned remaining;
8ecce684
NP
1519 int window;
1520 int depth;
50f22ada
JS
1521 int working;
1522 int data_ready;
1523 pthread_mutex_t mutex;
1524 pthread_cond_t cond;
8ecce684
NP
1525 unsigned *processed;
1526};
1527
50f22ada 1528static pthread_cond_t progress_cond = PTHREAD_COND_INITIALIZER;
c2a33679 1529
8ecce684
NP
1530static void *threaded_find_deltas(void *arg)
1531{
c2a33679
NP
1532 struct thread_params *me = arg;
1533
50f22ada 1534 while (me->remaining) {
384b32c0 1535 find_deltas(me->list, &me->remaining,
c2a33679 1536 me->window, me->depth, me->processed);
50f22ada
JS
1537
1538 progress_lock();
1539 me->working = 0;
1540 pthread_cond_signal(&progress_cond);
1541 progress_unlock();
1542
1543 /*
1544 * We must not set ->data_ready before we wait on the
1545 * condition because the main thread may have set it to 1
1546 * before we get here. In order to be sure that new
1547 * work is available if we see 1 in ->data_ready, it
1548 * was initialized to 0 before this thread was spawned
1549 * and we reset it to 0 right away.
1550 */
1551 pthread_mutex_lock(&me->mutex);
1552 while (!me->data_ready)
1553 pthread_cond_wait(&me->cond, &me->mutex);
1554 me->data_ready = 0;
1555 pthread_mutex_unlock(&me->mutex);
c2a33679 1556 }
50f22ada
JS
1557 /* leave ->working 1 so that this doesn't get more work assigned */
1558 return NULL;
8ecce684
NP
1559}
1560
8ecce684
NP
1561static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1562 int window, int depth, unsigned *processed)
1563{
50f22ada 1564 struct thread_params p[delta_search_threads];
384b32c0 1565 int i, ret, active_threads = 0;
c2a33679 1566
367f4a43 1567 if (delta_search_threads <= 1) {
384b32c0 1568 find_deltas(list, &list_size, window, depth, processed);
367f4a43
NP
1569 return;
1570 }
1571
50f22ada 1572 /* Partition the work amongst work threads. */
367f4a43 1573 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
1574 unsigned sub_size = list_size / (delta_search_threads - i);
1575
8ecce684
NP
1576 p[i].window = window;
1577 p[i].depth = depth;
1578 p[i].processed = processed;
50f22ada
JS
1579 p[i].working = 1;
1580 p[i].data_ready = 0;
c2a33679 1581
59921b4b 1582 /* try to split chunks on "path" boundaries */
6fc74703
NP
1583 while (sub_size && sub_size < list_size &&
1584 list[sub_size]->hash &&
384b32c0
NP
1585 list[sub_size]->hash == list[sub_size-1]->hash)
1586 sub_size++;
1587
50f22ada
JS
1588 p[i].list = list;
1589 p[i].list_size = sub_size;
1590 p[i].remaining = sub_size;
59921b4b 1591
384b32c0
NP
1592 list += sub_size;
1593 list_size -= sub_size;
1594 }
1595
50f22ada
JS
1596 /* Start work threads. */
1597 for (i = 0; i < delta_search_threads; i++) {
1598 if (!p[i].list_size)
1599 continue;
68e6a4f8
JS
1600 pthread_mutex_init(&p[i].mutex, NULL);
1601 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
1602 ret = pthread_create(&p[i].thread, NULL,
1603 threaded_find_deltas, &p[i]);
1604 if (ret)
1605 die("unable to create thread: %s", strerror(ret));
1606 active_threads++;
1607 }
1608
384b32c0
NP
1609 /*
1610 * Now let's wait for work completion. Each time a thread is done
1611 * with its work, we steal half of the remaining work from the
1612 * thread with the largest number of unprocessed objects and give
1613 * it to that newly idle thread. This ensure good load balancing
1614 * until the remaining object list segments are simply too short
1615 * to be worth splitting anymore.
1616 */
50f22ada
JS
1617 while (active_threads) {
1618 struct thread_params *target = NULL;
384b32c0
NP
1619 struct thread_params *victim = NULL;
1620 unsigned sub_size = 0;
384b32c0
NP
1621
1622 progress_lock();
50f22ada
JS
1623 for (;;) {
1624 for (i = 0; !target && i < delta_search_threads; i++)
1625 if (!p[i].working)
1626 target = &p[i];
1627 if (target)
1628 break;
1629 pthread_cond_wait(&progress_cond, &progress_mutex);
1630 }
1631
384b32c0
NP
1632 for (i = 0; i < delta_search_threads; i++)
1633 if (p[i].remaining > 2*window &&
1634 (!victim || victim->remaining < p[i].remaining))
1635 victim = &p[i];
1636 if (victim) {
1637 sub_size = victim->remaining / 2;
1638 list = victim->list + victim->list_size - sub_size;
1639 while (sub_size && list[0]->hash &&
1640 list[0]->hash == list[-1]->hash) {
1641 list++;
1642 sub_size--;
1643 }
eb9688ff
NP
1644 if (!sub_size) {
1645 /*
1646 * It is possible for some "paths" to have
1647 * so many objects that no hash boundary
1648 * might be found. Let's just steal the
1649 * exact half in that case.
1650 */
1651 sub_size = victim->remaining / 2;
1652 list -= sub_size;
1653 }
384b32c0
NP
1654 target->list = list;
1655 victim->list_size -= sub_size;
1656 victim->remaining -= sub_size;
1657 }
384b32c0
NP
1658 target->list_size = sub_size;
1659 target->remaining = sub_size;
50f22ada
JS
1660 target->working = 1;
1661 progress_unlock();
1662
1663 pthread_mutex_lock(&target->mutex);
1664 target->data_ready = 1;
1665 pthread_cond_signal(&target->cond);
1666 pthread_mutex_unlock(&target->mutex);
c2a33679 1667
384b32c0 1668 if (!sub_size) {
b81d9af7 1669 pthread_join(target->thread, NULL);
50f22ada
JS
1670 pthread_cond_destroy(&target->cond);
1671 pthread_mutex_destroy(&target->mutex);
384b32c0 1672 active_threads--;
c2a33679 1673 }
50f22ada 1674 }
8ecce684
NP
1675}
1676
1677#else
384b32c0 1678#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
1679#endif
1680
f0a24aa5
SP
1681static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1682{
1683 unsigned char peeled[20];
1684
1685 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1686 !peel_ref(path, peeled) && /* peelable? */
1687 !is_null_sha1(peeled) && /* annotated tag? */
1688 locate_object_entry(peeled)) /* object packed? */
1689 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1690 return 0;
1691}
1692
f3123c4a
JH
1693static void prepare_pack(int window, int depth)
1694{
9668cf59 1695 struct object_entry **delta_list;
6e1c2344
RJ
1696 uint32_t i, nr_deltas;
1697 unsigned n;
9668cf59 1698
3f9ac8d2 1699 get_object_details();
9668cf59 1700
0e8189e2
NP
1701 /*
1702 * If we're locally repacking then we need to be doubly careful
1703 * from now on in order to make sure no stealth corruption gets
1704 * propagated to the new pack. Clients receiving streamed packs
1705 * should validate everything they get anyway so no need to incur
1706 * the additional cost here in that case.
1707 */
1708 if (!pack_to_stdout)
1709 do_check_packed_object_crc = 1;
1710
75d39853 1711 if (!nr_objects || !window || !depth)
9668cf59
NP
1712 return;
1713
1714 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
75d39853
NP
1715 nr_deltas = n = 0;
1716
1717 for (i = 0; i < nr_objects; i++) {
1718 struct object_entry *entry = objects + i;
1719
1720 if (entry->delta)
1721 /* This happens if we decided to reuse existing
a7de7130 1722 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
1723 */
1724 continue;
1725
1726 if (entry->size < 50)
1727 continue;
1728
1729 if (entry->no_try_delta)
1730 continue;
1731
6d6f9cdd 1732 if (!entry->preferred_base) {
75d39853 1733 nr_deltas++;
6d6f9cdd
SP
1734 if (entry->type < 0)
1735 die("unable to get type of object %s",
1736 sha1_to_hex(entry->idx.sha1));
eede9f42
NP
1737 } else {
1738 if (entry->type < 0) {
1739 /*
1740 * This object is not found, but we
1741 * don't have to include it anyway.
1742 */
1743 continue;
1744 }
6d6f9cdd 1745 }
75d39853
NP
1746
1747 delta_list[n++] = entry;
1748 }
1749
2f8b8947 1750 if (nr_deltas && n > 1) {
e334977d
NP
1751 unsigned nr_done = 0;
1752 if (progress)
dc6a0757
NP
1753 progress_state = start_progress("Compressing objects",
1754 nr_deltas);
75d39853 1755 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
8ecce684 1756 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 1757 stop_progress(&progress_state);
e334977d
NP
1758 if (nr_done != nr_deltas)
1759 die("inconsistency with delta count");
75d39853 1760 }
9668cf59 1761 free(delta_list);
f3123c4a
JH
1762}
1763
ef90d6d4 1764static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a
JK
1765{
1766 if(!strcmp(k, "pack.window")) {
1767 window = git_config_int(k, v);
1768 return 0;
1769 }
a97773ce
BD
1770 if (!strcmp(k, "pack.windowmemory")) {
1771 window_memory_limit = git_config_ulong(k, v);
1772 return 0;
1773 }
1774 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
1775 depth = git_config_int(k, v);
1776 return 0;
1777 }
960ccca6
DH
1778 if (!strcmp(k, "pack.compression")) {
1779 int level = git_config_int(k, v);
1780 if (level == -1)
1781 level = Z_DEFAULT_COMPRESSION;
1782 else if (level < 0 || level > Z_BEST_COMPRESSION)
1783 die("bad pack compression level %d", level);
1784 pack_compression_level = level;
1785 pack_compression_seen = 1;
1786 return 0;
1787 }
074b2eea
MK
1788 if (!strcmp(k, "pack.deltacachesize")) {
1789 max_delta_cache_size = git_config_int(k, v);
1790 return 0;
1791 }
e3dfddb3
MK
1792 if (!strcmp(k, "pack.deltacachelimit")) {
1793 cache_max_small_delta_size = git_config_int(k, v);
1794 return 0;
1795 }
693b86ff
NP
1796 if (!strcmp(k, "pack.threads")) {
1797 delta_search_threads = git_config_int(k, v);
833e3df1 1798 if (delta_search_threads < 0)
693b86ff
NP
1799 die("invalid number of threads specified (%d)",
1800 delta_search_threads);
1801#ifndef THREADED_DELTA_SEARCH
833e3df1 1802 if (delta_search_threads != 1)
693b86ff
NP
1803 warning("no threads support, ignoring %s", k);
1804#endif
1805 return 0;
1806 }
4d00bda2
NP
1807 if (!strcmp(k, "pack.indexversion")) {
1808 pack_idx_default_version = git_config_int(k, v);
1809 if (pack_idx_default_version > 2)
6e1c2344
RJ
1810 die("bad pack.indexversion=%"PRIu32,
1811 pack_idx_default_version);
4d00bda2
NP
1812 return 0;
1813 }
2b84b5a8
JS
1814 if (!strcmp(k, "pack.packsizelimit")) {
1815 pack_size_limit_cfg = git_config_ulong(k, v);
1816 return 0;
1817 }
ef90d6d4 1818 return git_default_config(k, v, cb);
4812a93a
JK
1819}
1820
b5d97e6b 1821static void read_object_list_from_stdin(void)
c323ac7d 1822{
b5d97e6b
JH
1823 char line[40 + 1 + PATH_MAX + 2];
1824 unsigned char sha1[20];
b2504a0d 1825
da93d12b 1826 for (;;) {
da93d12b
LT
1827 if (!fgets(line, sizeof(line), stdin)) {
1828 if (feof(stdin))
1829 break;
1830 if (!ferror(stdin))
1831 die("fgets returned NULL, not EOF, not error!");
687dd75c
JH
1832 if (errno != EINTR)
1833 die("fgets: %s", strerror(errno));
1834 clearerr(stdin);
1835 continue;
da93d12b 1836 }
7a979d99
JH
1837 if (line[0] == '-') {
1838 if (get_sha1_hex(line+1, sha1))
1839 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 1840 line);
8d1d8f83 1841 add_preferred_base(sha1);
7a979d99 1842 continue;
21fcd1bd 1843 }
c323ac7d 1844 if (get_sha1_hex(line, sha1))
ef07618f 1845 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 1846
bc32fed5
JH
1847 add_preferred_base_object(line+41);
1848 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 1849 }
b5d97e6b
JH
1850}
1851
08cdfb13
JH
1852#define OBJECT_ADDED (1u<<20)
1853
b5d97e6b
JH
1854static void show_commit(struct commit *commit)
1855{
bc32fed5 1856 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
08cdfb13 1857 commit->object.flags |= OBJECT_ADDED;
b5d97e6b
JH
1858}
1859
1860static void show_object(struct object_array_entry *p)
1861{
bc32fed5
JH
1862 add_preferred_base_object(p->name);
1863 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
08cdfb13 1864 p->item->flags |= OBJECT_ADDED;
b5d97e6b
JH
1865}
1866
8d1d8f83
JH
1867static void show_edge(struct commit *commit)
1868{
1869 add_preferred_base(commit->object.sha1);
1870}
1871
08cdfb13
JH
1872struct in_pack_object {
1873 off_t offset;
1874 struct object *object;
1875};
1876
1877struct in_pack {
1878 int alloc;
1879 int nr;
1880 struct in_pack_object *array;
1881};
1882
1883static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1884{
1885 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1886 in_pack->array[in_pack->nr].object = object;
1887 in_pack->nr++;
1888}
1889
1890/*
1891 * Compare the objects in the offset order, in order to emulate the
f18d244a 1892 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
1893 */
1894static int ofscmp(const void *a_, const void *b_)
1895{
1896 struct in_pack_object *a = (struct in_pack_object *)a_;
1897 struct in_pack_object *b = (struct in_pack_object *)b_;
1898
1899 if (a->offset < b->offset)
1900 return -1;
1901 else if (a->offset > b->offset)
1902 return 1;
1903 else
1904 return hashcmp(a->object->sha1, b->object->sha1);
1905}
1906
1907static void add_objects_in_unpacked_packs(struct rev_info *revs)
1908{
1909 struct packed_git *p;
1910 struct in_pack in_pack;
1911 uint32_t i;
1912
1913 memset(&in_pack, 0, sizeof(in_pack));
1914
1915 for (p = packed_git; p; p = p->next) {
1916 const unsigned char *sha1;
1917 struct object *o;
1918
1919 for (i = 0; i < revs->num_ignore_packed; i++) {
1920 if (matches_pack_name(p, revs->ignore_packed[i]))
1921 break;
1922 }
1923 if (revs->num_ignore_packed <= i)
1924 continue;
1925 if (open_pack_index(p))
1926 die("cannot open pack index");
1927
1928 ALLOC_GROW(in_pack.array,
1929 in_pack.nr + p->num_objects,
1930 in_pack.alloc);
1931
1932 for (i = 0; i < p->num_objects; i++) {
1933 sha1 = nth_packed_object_sha1(p, i);
1934 o = lookup_unknown_object(sha1);
1935 if (!(o->flags & OBJECT_ADDED))
1936 mark_in_pack_object(o, p, &in_pack);
1937 o->flags |= OBJECT_ADDED;
1938 }
1939 }
1940
1941 if (in_pack.nr) {
1942 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
1943 ofscmp);
1944 for (i = 0; i < in_pack.nr; i++) {
1945 struct object *o = in_pack.array[i].object;
1946 add_object_entry(o->sha1, o->type, "", 0);
1947 }
1948 }
1949 free(in_pack.array);
1950}
1951
ca11b212
NP
1952static void loosen_unused_packed_objects(struct rev_info *revs)
1953{
1954 struct packed_git *p;
1955 uint32_t i;
1956 const unsigned char *sha1;
1957
1958 for (p = packed_git; p; p = p->next) {
1959 for (i = 0; i < revs->num_ignore_packed; i++) {
1960 if (matches_pack_name(p, revs->ignore_packed[i]))
1961 break;
1962 }
1963 if (revs->num_ignore_packed <= i)
1964 continue;
1965
1966 if (open_pack_index(p))
1967 die("cannot open pack index");
1968
1969 for (i = 0; i < p->num_objects; i++) {
1970 sha1 = nth_packed_object_sha1(p, i);
1971 if (!locate_object_entry(sha1))
1972 if (force_object_loose(sha1, p->mtime))
1973 die("unable to force loose object");
1974 }
1975 }
1976}
1977
8d1d8f83 1978static void get_object_list(int ac, const char **av)
b5d97e6b
JH
1979{
1980 struct rev_info revs;
1981 char line[1000];
b5d97e6b
JH
1982 int flags = 0;
1983
b5d97e6b
JH
1984 init_revisions(&revs, NULL);
1985 save_commit_buffer = 0;
b5d97e6b
JH
1986 setup_revisions(ac, av, &revs, NULL);
1987
b5d97e6b
JH
1988 while (fgets(line, sizeof(line), stdin) != NULL) {
1989 int len = strlen(line);
872c930d 1990 if (len && line[len - 1] == '\n')
b5d97e6b
JH
1991 line[--len] = 0;
1992 if (!len)
1993 break;
1994 if (*line == '-') {
1995 if (!strcmp(line, "--not")) {
1996 flags ^= UNINTERESTING;
1997 continue;
1998 }
1999 die("not a rev '%s'", line);
2000 }
2001 if (handle_revision_arg(line, &revs, flags, 1))
2002 die("bad revision '%s'", line);
2003 }
2004
3d51e1b5
MK
2005 if (prepare_revision_walk(&revs))
2006 die("revision walk setup failed");
8d1d8f83 2007 mark_edges_uninteresting(revs.commits, &revs, show_edge);
b5d97e6b 2008 traverse_commit_list(&revs, show_commit, show_object);
08cdfb13
JH
2009
2010 if (keep_unreachable)
2011 add_objects_in_unpacked_packs(&revs);
ca11b212
NP
2012 if (unpack_unreachable)
2013 loosen_unused_packed_objects(&revs);
b5d97e6b
JH
2014}
2015
b6b32ccb
JH
2016static int adjust_perm(const char *path, mode_t mode)
2017{
2018 if (chmod(path, mode))
2019 return -1;
2020 return adjust_shared_perm(path);
2021}
2022
b5d97e6b
JH
2023int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2024{
b5d97e6b 2025 int use_internal_rev_list = 0;
8d1d8f83 2026 int thin = 0;
7cadf491 2027 uint32_t i;
ffa84ffb
RD
2028 const char **rp_av;
2029 int rp_ac_alloc = 64;
8d1d8f83
JH
2030 int rp_ac;
2031
ffa84ffb
RD
2032 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2033
8d1d8f83
JH
2034 rp_av[0] = "pack-objects";
2035 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2036 rp_ac = 2;
b5d97e6b 2037
ef90d6d4 2038 git_config(git_pack_config, NULL);
960ccca6
DH
2039 if (!pack_compression_seen && core_compression_seen)
2040 pack_compression_level = core_compression_level;
b5d97e6b
JH
2041
2042 progress = isatty(2);
2043 for (i = 1; i < argc; i++) {
2044 const char *arg = argv[i];
2045
2046 if (*arg != '-')
2047 break;
2048
2049 if (!strcmp("--non-empty", arg)) {
2050 non_empty = 1;
2051 continue;
2052 }
2053 if (!strcmp("--local", arg)) {
2054 local = 1;
2055 continue;
2056 }
b5d97e6b
JH
2057 if (!strcmp("--incremental", arg)) {
2058 incremental = 1;
2059 continue;
2060 }
960ccca6
DH
2061 if (!prefixcmp(arg, "--compression=")) {
2062 char *end;
2063 int level = strtoul(arg+14, &end, 0);
2064 if (!arg[14] || *end)
2065 usage(pack_usage);
2066 if (level == -1)
2067 level = Z_DEFAULT_COMPRESSION;
2068 else if (level < 0 || level > Z_BEST_COMPRESSION)
2069 die("bad pack compression level %d", level);
2070 pack_compression_level = level;
2071 continue;
2072 }
6b94b1a0
DH
2073 if (!prefixcmp(arg, "--max-pack-size=")) {
2074 char *end;
2b84b5a8 2075 pack_size_limit_cfg = 0;
6b94b1a0
DH
2076 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2077 if (!arg[16] || *end)
2078 usage(pack_usage);
2079 continue;
2080 }
599065a3 2081 if (!prefixcmp(arg, "--window=")) {
b5d97e6b
JH
2082 char *end;
2083 window = strtoul(arg+9, &end, 0);
2084 if (!arg[9] || *end)
2085 usage(pack_usage);
2086 continue;
2087 }
a97773ce
BD
2088 if (!prefixcmp(arg, "--window-memory=")) {
2089 if (!git_parse_ulong(arg+16, &window_memory_limit))
2090 usage(pack_usage);
2091 continue;
2092 }
367f4a43
NP
2093 if (!prefixcmp(arg, "--threads=")) {
2094 char *end;
2095 delta_search_threads = strtoul(arg+10, &end, 0);
833e3df1 2096 if (!arg[10] || *end || delta_search_threads < 0)
367f4a43
NP
2097 usage(pack_usage);
2098#ifndef THREADED_DELTA_SEARCH
833e3df1 2099 if (delta_search_threads != 1)
367f4a43
NP
2100 warning("no threads support, "
2101 "ignoring %s", arg);
2102#endif
2103 continue;
2104 }
599065a3 2105 if (!prefixcmp(arg, "--depth=")) {
b5d97e6b
JH
2106 char *end;
2107 depth = strtoul(arg+8, &end, 0);
2108 if (!arg[8] || *end)
2109 usage(pack_usage);
2110 continue;
2111 }
2112 if (!strcmp("--progress", arg)) {
2113 progress = 1;
2114 continue;
2115 }
231f240b
NP
2116 if (!strcmp("--all-progress", arg)) {
2117 progress = 2;
2118 continue;
2119 }
b5d97e6b
JH
2120 if (!strcmp("-q", arg)) {
2121 progress = 0;
2122 continue;
2123 }
2124 if (!strcmp("--no-reuse-delta", arg)) {
a7de7130 2125 reuse_delta = 0;
b5d97e6b
JH
2126 continue;
2127 }
fa736f72 2128 if (!strcmp("--no-reuse-object", arg)) {
a7de7130 2129 reuse_object = reuse_delta = 0;
fa736f72
NP
2130 continue;
2131 }
be6b1914 2132 if (!strcmp("--delta-base-offset", arg)) {
780e6e73 2133 allow_ofs_delta = 1;
be6b1914
NP
2134 continue;
2135 }
b5d97e6b
JH
2136 if (!strcmp("--stdout", arg)) {
2137 pack_to_stdout = 1;
2138 continue;
2139 }
2140 if (!strcmp("--revs", arg)) {
2141 use_internal_rev_list = 1;
2142 continue;
2143 }
08cdfb13
JH
2144 if (!strcmp("--keep-unreachable", arg)) {
2145 keep_unreachable = 1;
2146 continue;
2147 }
ca11b212
NP
2148 if (!strcmp("--unpack-unreachable", arg)) {
2149 unpack_unreachable = 1;
2150 continue;
2151 }
f0a24aa5
SP
2152 if (!strcmp("--include-tag", arg)) {
2153 include_tag = 1;
2154 continue;
2155 }
8d1d8f83 2156 if (!strcmp("--unpacked", arg) ||
599065a3 2157 !prefixcmp(arg, "--unpacked=") ||
63049292 2158 !strcmp("--reflog", arg) ||
8d1d8f83
JH
2159 !strcmp("--all", arg)) {
2160 use_internal_rev_list = 1;
ffa84ffb
RD
2161 if (rp_ac >= rp_ac_alloc - 1) {
2162 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2163 rp_av = xrealloc(rp_av,
2164 rp_ac_alloc * sizeof(*rp_av));
2165 }
8d1d8f83 2166 rp_av[rp_ac++] = arg;
b5d97e6b
JH
2167 continue;
2168 }
8d1d8f83
JH
2169 if (!strcmp("--thin", arg)) {
2170 use_internal_rev_list = 1;
2171 thin = 1;
2172 rp_av[1] = "--objects-edge";
b5d97e6b 2173 continue;
4ba7d711
NP
2174 }
2175 if (!prefixcmp(arg, "--index-version=")) {
2176 char *c;
aa7e44bf
GB
2177 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2178 if (pack_idx_default_version > 2)
4ba7d711
NP
2179 die("bad %s", arg);
2180 if (*c == ',')
aa7e44bf
GB
2181 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2182 if (*c || pack_idx_off32_limit & 0x80000000)
4ba7d711
NP
2183 die("bad %s", arg);
2184 continue;
b5d97e6b
JH
2185 }
2186 usage(pack_usage);
2187 }
2188
2189 /* Traditionally "pack-objects [options] base extra" failed;
2190 * we would however want to take refs parameter that would
2191 * have been given to upstream rev-list ourselves, which means
2192 * we somehow want to say what the base name is. So the
2193 * syntax would be:
2194 *
2195 * pack-objects [options] base <refs...>
2196 *
2197 * in other words, we would treat the first non-option as the
2198 * base_name and send everything else to the internal revision
2199 * walker.
2200 */
2201
2202 if (!pack_to_stdout)
2203 base_name = argv[i++];
2204
2205 if (pack_to_stdout != !base_name)
2206 usage(pack_usage);
2207
2b84b5a8
JS
2208 if (!pack_to_stdout && !pack_size_limit)
2209 pack_size_limit = pack_size_limit_cfg;
2210
01c12a23
DH
2211 if (pack_to_stdout && pack_size_limit)
2212 die("--max-pack-size cannot be used to build a pack for transfer.");
2213
8d1d8f83
JH
2214 if (!pack_to_stdout && thin)
2215 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 2216
ca11b212
NP
2217 if (keep_unreachable && unpack_unreachable)
2218 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2219
833e3df1
AE
2220#ifdef THREADED_DELTA_SEARCH
2221 if (!delta_search_threads) /* --threads=0 means autodetect */
2222 delta_search_threads = online_cpus();
2223#endif
2224
b5d97e6b
JH
2225 prepare_packed_git();
2226
13aaf148 2227 if (progress)
dc6a0757 2228 progress_state = start_progress("Counting objects", 0);
b5d97e6b
JH
2229 if (!use_internal_rev_list)
2230 read_object_list_from_stdin();
8d1d8f83
JH
2231 else {
2232 rp_av[rp_ac] = NULL;
2233 get_object_list(rp_ac, rp_av);
2234 }
f0a24aa5
SP
2235 if (include_tag && nr_result)
2236 for_each_ref(add_ref_tag, NULL);
4d4fcc54 2237 stop_progress(&progress_state);
96a02f8f 2238
f0b0af1b 2239 if (non_empty && !nr_result)
1c4a2912 2240 return 0;
f7ae6a93
NP
2241 if (nr_result)
2242 prepare_pack(window, depth);
d01fb92f 2243 write_pack_file();
ab7cd7bb 2244 if (progress)
6e1c2344
RJ
2245 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2246 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 2247 written, written_delta, reused, reused_delta);
c323ac7d
LT
2248 return 0;
2249}