packfile.c: speed up loading lots of packfiles
[git] / packfile.c
1 #include "cache.h"
2 #include "list.h"
3 #include "pack.h"
4 #include "repository.h"
5 #include "dir.h"
6 #include "mergesort.h"
7 #include "packfile.h"
8 #include "delta.h"
9 #include "streaming.h"
10 #include "sha1-lookup.h"
11 #include "commit.h"
12 #include "object.h"
13 #include "tag.h"
14 #include "tree-walk.h"
15 #include "tree.h"
16 #include "object-store.h"
17 #include "midx.h"
18 #include "commit-graph.h"
19 #include "promisor-remote.h"
20
21 char *odb_pack_name(struct strbuf *buf,
22                     const unsigned char *hash,
23                     const char *ext)
24 {
25         strbuf_reset(buf);
26         strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
27                     hash_to_hex(hash), ext);
28         return buf->buf;
29 }
30
31 char *sha1_pack_name(const unsigned char *sha1)
32 {
33         static struct strbuf buf = STRBUF_INIT;
34         return odb_pack_name(&buf, sha1, "pack");
35 }
36
37 char *sha1_pack_index_name(const unsigned char *sha1)
38 {
39         static struct strbuf buf = STRBUF_INIT;
40         return odb_pack_name(&buf, sha1, "idx");
41 }
42
43 static unsigned int pack_used_ctr;
44 static unsigned int pack_mmap_calls;
45 static unsigned int peak_pack_open_windows;
46 static unsigned int pack_open_windows;
47 static unsigned int pack_open_fds;
48 static unsigned int pack_max_fds;
49 static size_t peak_pack_mapped;
50 static size_t pack_mapped;
51
52 #define SZ_FMT PRIuMAX
53 static inline uintmax_t sz_fmt(size_t s) { return s; }
54
55 void pack_report(void)
56 {
57         fprintf(stderr,
58                 "pack_report: getpagesize()            = %10" SZ_FMT "\n"
59                 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
60                 "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
61                 sz_fmt(getpagesize()),
62                 sz_fmt(packed_git_window_size),
63                 sz_fmt(packed_git_limit));
64         fprintf(stderr,
65                 "pack_report: pack_used_ctr            = %10u\n"
66                 "pack_report: pack_mmap_calls          = %10u\n"
67                 "pack_report: pack_open_windows        = %10u / %10u\n"
68                 "pack_report: pack_mapped              = "
69                         "%10" SZ_FMT " / %10" SZ_FMT "\n",
70                 pack_used_ctr,
71                 pack_mmap_calls,
72                 pack_open_windows, peak_pack_open_windows,
73                 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
74 }
75
76 /*
77  * Open and mmap the index file at path, perform a couple of
78  * consistency checks, then record its information to p.  Return 0 on
79  * success.
80  */
81 static int check_packed_git_idx(const char *path, struct packed_git *p)
82 {
83         void *idx_map;
84         size_t idx_size;
85         int fd = git_open(path), ret;
86         struct stat st;
87         const unsigned int hashsz = the_hash_algo->rawsz;
88
89         if (fd < 0)
90                 return -1;
91         if (fstat(fd, &st)) {
92                 close(fd);
93                 return -1;
94         }
95         idx_size = xsize_t(st.st_size);
96         if (idx_size < 4 * 256 + hashsz + hashsz) {
97                 close(fd);
98                 return error("index file %s is too small", path);
99         }
100         idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
101         close(fd);
102
103         ret = load_idx(path, hashsz, idx_map, idx_size, p);
104
105         if (ret)
106                 munmap(idx_map, idx_size);
107
108         return ret;
109 }
110
111 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
112              size_t idx_size, struct packed_git *p)
113 {
114         struct pack_idx_header *hdr = idx_map;
115         uint32_t version, nr, i, *index;
116
117         if (idx_size < 4 * 256 + hashsz + hashsz)
118                 return error("index file %s is too small", path);
119         if (idx_map == NULL)
120                 return error("empty data");
121
122         if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
123                 version = ntohl(hdr->idx_version);
124                 if (version < 2 || version > 2)
125                         return error("index file %s is version %"PRIu32
126                                      " and is not supported by this binary"
127                                      " (try upgrading GIT to a newer version)",
128                                      path, version);
129         } else
130                 version = 1;
131
132         nr = 0;
133         index = idx_map;
134         if (version > 1)
135                 index += 2;  /* skip index header */
136         for (i = 0; i < 256; i++) {
137                 uint32_t n = ntohl(index[i]);
138                 if (n < nr)
139                         return error("non-monotonic index %s", path);
140                 nr = n;
141         }
142
143         if (version == 1) {
144                 /*
145                  * Total size:
146                  *  - 256 index entries 4 bytes each
147                  *  - 24-byte entries * nr (object ID + 4-byte offset)
148                  *  - hash of the packfile
149                  *  - file checksum
150                  */
151                 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
152                         return error("wrong index v1 file size in %s", path);
153         } else if (version == 2) {
154                 /*
155                  * Minimum size:
156                  *  - 8 bytes of header
157                  *  - 256 index entries 4 bytes each
158                  *  - object ID entry * nr
159                  *  - 4-byte crc entry * nr
160                  *  - 4-byte offset entry * nr
161                  *  - hash of the packfile
162                  *  - file checksum
163                  * And after the 4-byte offset table might be a
164                  * variable sized table containing 8-byte entries
165                  * for offsets larger than 2^31.
166                  */
167                 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
168                 unsigned long max_size = min_size;
169                 if (nr)
170                         max_size += (nr - 1)*8;
171                 if (idx_size < min_size || idx_size > max_size)
172                         return error("wrong index v2 file size in %s", path);
173                 if (idx_size != min_size &&
174                     /*
175                      * make sure we can deal with large pack offsets.
176                      * 31-bit signed offset won't be enough, neither
177                      * 32-bit unsigned one will be.
178                      */
179                     (sizeof(off_t) <= 4))
180                         return error("pack too large for current definition of off_t in %s", path);
181         }
182
183         p->index_version = version;
184         p->index_data = idx_map;
185         p->index_size = idx_size;
186         p->num_objects = nr;
187         return 0;
188 }
189
190 int open_pack_index(struct packed_git *p)
191 {
192         char *idx_name;
193         size_t len;
194         int ret;
195
196         if (p->index_data)
197                 return 0;
198
199         if (!strip_suffix(p->pack_name, ".pack", &len))
200                 BUG("pack_name does not end in .pack");
201         idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
202         ret = check_packed_git_idx(idx_name, p);
203         free(idx_name);
204         return ret;
205 }
206
207 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
208 {
209         const uint32_t *level1_ofs = p->index_data;
210
211         if (!level1_ofs) {
212                 if (open_pack_index(p))
213                         return 0;
214                 level1_ofs = p->index_data;
215         }
216
217         if (p->index_version > 1) {
218                 level1_ofs += 2;
219         }
220
221         return ntohl(level1_ofs[value]);
222 }
223
224 static struct packed_git *alloc_packed_git(int extra)
225 {
226         struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
227         memset(p, 0, sizeof(*p));
228         p->pack_fd = -1;
229         return p;
230 }
231
232 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
233 {
234         const char *path = sha1_pack_name(sha1);
235         size_t alloc = st_add(strlen(path), 1);
236         struct packed_git *p = alloc_packed_git(alloc);
237
238         memcpy(p->pack_name, path, alloc); /* includes NUL */
239         hashcpy(p->hash, sha1);
240         if (check_packed_git_idx(idx_path, p)) {
241                 free(p);
242                 return NULL;
243         }
244
245         return p;
246 }
247
248 static void scan_windows(struct packed_git *p,
249         struct packed_git **lru_p,
250         struct pack_window **lru_w,
251         struct pack_window **lru_l)
252 {
253         struct pack_window *w, *w_l;
254
255         for (w_l = NULL, w = p->windows; w; w = w->next) {
256                 if (!w->inuse_cnt) {
257                         if (!*lru_w || w->last_used < (*lru_w)->last_used) {
258                                 *lru_p = p;
259                                 *lru_w = w;
260                                 *lru_l = w_l;
261                         }
262                 }
263                 w_l = w;
264         }
265 }
266
267 static int unuse_one_window(struct packed_git *current)
268 {
269         struct packed_git *p, *lru_p = NULL;
270         struct pack_window *lru_w = NULL, *lru_l = NULL;
271
272         if (current)
273                 scan_windows(current, &lru_p, &lru_w, &lru_l);
274         for (p = the_repository->objects->packed_git; p; p = p->next)
275                 scan_windows(p, &lru_p, &lru_w, &lru_l);
276         if (lru_p) {
277                 munmap(lru_w->base, lru_w->len);
278                 pack_mapped -= lru_w->len;
279                 if (lru_l)
280                         lru_l->next = lru_w->next;
281                 else
282                         lru_p->windows = lru_w->next;
283                 free(lru_w);
284                 pack_open_windows--;
285                 return 1;
286         }
287         return 0;
288 }
289
290 void close_pack_windows(struct packed_git *p)
291 {
292         while (p->windows) {
293                 struct pack_window *w = p->windows;
294
295                 if (w->inuse_cnt)
296                         die("pack '%s' still has open windows to it",
297                             p->pack_name);
298                 munmap(w->base, w->len);
299                 pack_mapped -= w->len;
300                 pack_open_windows--;
301                 p->windows = w->next;
302                 free(w);
303         }
304 }
305
306 int close_pack_fd(struct packed_git *p)
307 {
308         if (p->pack_fd < 0)
309                 return 0;
310
311         close(p->pack_fd);
312         pack_open_fds--;
313         p->pack_fd = -1;
314
315         return 1;
316 }
317
318 void close_pack_index(struct packed_git *p)
319 {
320         if (p->index_data) {
321                 munmap((void *)p->index_data, p->index_size);
322                 p->index_data = NULL;
323         }
324 }
325
326 void close_pack(struct packed_git *p)
327 {
328         close_pack_windows(p);
329         close_pack_fd(p);
330         close_pack_index(p);
331 }
332
333 void close_object_store(struct raw_object_store *o)
334 {
335         struct packed_git *p;
336
337         for (p = o->packed_git; p; p = p->next)
338                 if (p->do_not_close)
339                         BUG("want to close pack marked 'do-not-close'");
340                 else
341                         close_pack(p);
342
343         if (o->multi_pack_index) {
344                 close_midx(o->multi_pack_index);
345                 o->multi_pack_index = NULL;
346         }
347
348         close_commit_graph(o);
349 }
350
351 void unlink_pack_path(const char *pack_name, int force_delete)
352 {
353         static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
354         int i;
355         struct strbuf buf = STRBUF_INIT;
356         size_t plen;
357
358         strbuf_addstr(&buf, pack_name);
359         strip_suffix_mem(buf.buf, &buf.len, ".pack");
360         plen = buf.len;
361
362         if (!force_delete) {
363                 strbuf_addstr(&buf, ".keep");
364                 if (!access(buf.buf, F_OK)) {
365                         strbuf_release(&buf);
366                         return;
367                 }
368         }
369
370         for (i = 0; i < ARRAY_SIZE(exts); i++) {
371                 strbuf_setlen(&buf, plen);
372                 strbuf_addstr(&buf, exts[i]);
373                 unlink(buf.buf);
374         }
375
376         strbuf_release(&buf);
377 }
378
379 /*
380  * The LRU pack is the one with the oldest MRU window, preferring packs
381  * with no used windows, or the oldest mtime if it has no windows allocated.
382  */
383 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
384 {
385         struct pack_window *w, *this_mru_w;
386         int has_windows_inuse = 0;
387
388         /*
389          * Reject this pack if it has windows and the previously selected
390          * one does not.  If this pack does not have windows, reject
391          * it if the pack file is newer than the previously selected one.
392          */
393         if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
394                 return;
395
396         for (w = this_mru_w = p->windows; w; w = w->next) {
397                 /*
398                  * Reject this pack if any of its windows are in use,
399                  * but the previously selected pack did not have any
400                  * inuse windows.  Otherwise, record that this pack
401                  * has windows in use.
402                  */
403                 if (w->inuse_cnt) {
404                         if (*accept_windows_inuse)
405                                 has_windows_inuse = 1;
406                         else
407                                 return;
408                 }
409
410                 if (w->last_used > this_mru_w->last_used)
411                         this_mru_w = w;
412
413                 /*
414                  * Reject this pack if it has windows that have been
415                  * used more recently than the previously selected pack.
416                  * If the previously selected pack had windows inuse and
417                  * we have not encountered a window in this pack that is
418                  * inuse, skip this check since we prefer a pack with no
419                  * inuse windows to one that has inuse windows.
420                  */
421                 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
422                     this_mru_w->last_used > (*mru_w)->last_used)
423                         return;
424         }
425
426         /*
427          * Select this pack.
428          */
429         *mru_w = this_mru_w;
430         *lru_p = p;
431         *accept_windows_inuse = has_windows_inuse;
432 }
433
434 static int close_one_pack(void)
435 {
436         struct packed_git *p, *lru_p = NULL;
437         struct pack_window *mru_w = NULL;
438         int accept_windows_inuse = 1;
439
440         for (p = the_repository->objects->packed_git; p; p = p->next) {
441                 if (p->pack_fd == -1)
442                         continue;
443                 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
444         }
445
446         if (lru_p)
447                 return close_pack_fd(lru_p);
448
449         return 0;
450 }
451
452 static unsigned int get_max_fd_limit(void)
453 {
454 #ifdef RLIMIT_NOFILE
455         {
456                 struct rlimit lim;
457
458                 if (!getrlimit(RLIMIT_NOFILE, &lim))
459                         return lim.rlim_cur;
460         }
461 #endif
462
463 #ifdef _SC_OPEN_MAX
464         {
465                 long open_max = sysconf(_SC_OPEN_MAX);
466                 if (0 < open_max)
467                         return open_max;
468                 /*
469                  * Otherwise, we got -1 for one of the two
470                  * reasons:
471                  *
472                  * (1) sysconf() did not understand _SC_OPEN_MAX
473                  *     and signaled an error with -1; or
474                  * (2) sysconf() said there is no limit.
475                  *
476                  * We _could_ clear errno before calling sysconf() to
477                  * tell these two cases apart and return a huge number
478                  * in the latter case to let the caller cap it to a
479                  * value that is not so selfish, but letting the
480                  * fallback OPEN_MAX codepath take care of these cases
481                  * is a lot simpler.
482                  */
483         }
484 #endif
485
486 #ifdef OPEN_MAX
487         return OPEN_MAX;
488 #else
489         return 1; /* see the caller ;-) */
490 #endif
491 }
492
493 const char *pack_basename(struct packed_git *p)
494 {
495         const char *ret = strrchr(p->pack_name, '/');
496         if (ret)
497                 ret = ret + 1; /* skip past slash */
498         else
499                 ret = p->pack_name; /* we only have a base */
500         return ret;
501 }
502
503 /*
504  * Do not call this directly as this leaks p->pack_fd on error return;
505  * call open_packed_git() instead.
506  */
507 static int open_packed_git_1(struct packed_git *p)
508 {
509         struct stat st;
510         struct pack_header hdr;
511         unsigned char hash[GIT_MAX_RAWSZ];
512         unsigned char *idx_hash;
513         long fd_flag;
514         ssize_t read_result;
515         const unsigned hashsz = the_hash_algo->rawsz;
516
517         if (!p->index_data) {
518                 struct multi_pack_index *m;
519                 const char *pack_name = pack_basename(p);
520
521                 for (m = the_repository->objects->multi_pack_index;
522                      m; m = m->next) {
523                         if (midx_contains_pack(m, pack_name))
524                                 break;
525                 }
526
527                 if (!m && open_pack_index(p))
528                         return error("packfile %s index unavailable", p->pack_name);
529         }
530
531         if (!pack_max_fds) {
532                 unsigned int max_fds = get_max_fd_limit();
533
534                 /* Save 3 for stdin/stdout/stderr, 22 for work */
535                 if (25 < max_fds)
536                         pack_max_fds = max_fds - 25;
537                 else
538                         pack_max_fds = 1;
539         }
540
541         while (pack_max_fds <= pack_open_fds && close_one_pack())
542                 ; /* nothing */
543
544         p->pack_fd = git_open(p->pack_name);
545         if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
546                 return -1;
547         pack_open_fds++;
548
549         /* If we created the struct before we had the pack we lack size. */
550         if (!p->pack_size) {
551                 if (!S_ISREG(st.st_mode))
552                         return error("packfile %s not a regular file", p->pack_name);
553                 p->pack_size = st.st_size;
554         } else if (p->pack_size != st.st_size)
555                 return error("packfile %s size changed", p->pack_name);
556
557         /* We leave these file descriptors open with sliding mmap;
558          * there is no point keeping them open across exec(), though.
559          */
560         fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
561         if (fd_flag < 0)
562                 return error("cannot determine file descriptor flags");
563         fd_flag |= FD_CLOEXEC;
564         if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
565                 return error("cannot set FD_CLOEXEC");
566
567         /* Verify we recognize this pack file format. */
568         read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
569         if (read_result < 0)
570                 return error_errno("error reading from %s", p->pack_name);
571         if (read_result != sizeof(hdr))
572                 return error("file %s is far too short to be a packfile", p->pack_name);
573         if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
574                 return error("file %s is not a GIT packfile", p->pack_name);
575         if (!pack_version_ok(hdr.hdr_version))
576                 return error("packfile %s is version %"PRIu32" and not"
577                         " supported (try upgrading GIT to a newer version)",
578                         p->pack_name, ntohl(hdr.hdr_version));
579
580         /* Skip index checking if in multi-pack-index */
581         if (!p->index_data)
582                 return 0;
583
584         /* Verify the pack matches its index. */
585         if (p->num_objects != ntohl(hdr.hdr_entries))
586                 return error("packfile %s claims to have %"PRIu32" objects"
587                              " while index indicates %"PRIu32" objects",
588                              p->pack_name, ntohl(hdr.hdr_entries),
589                              p->num_objects);
590         if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
591                 return error("end of packfile %s is unavailable", p->pack_name);
592         read_result = read_in_full(p->pack_fd, hash, hashsz);
593         if (read_result < 0)
594                 return error_errno("error reading from %s", p->pack_name);
595         if (read_result != hashsz)
596                 return error("packfile %s signature is unavailable", p->pack_name);
597         idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
598         if (!hasheq(hash, idx_hash))
599                 return error("packfile %s does not match index", p->pack_name);
600         return 0;
601 }
602
603 static int open_packed_git(struct packed_git *p)
604 {
605         if (!open_packed_git_1(p))
606                 return 0;
607         close_pack_fd(p);
608         return -1;
609 }
610
611 static int in_window(struct pack_window *win, off_t offset)
612 {
613         /* We must promise at least one full hash after the
614          * offset is available from this window, otherwise the offset
615          * is not actually in this window and a different window (which
616          * has that one hash excess) must be used.  This is to support
617          * the object header and delta base parsing routines below.
618          */
619         off_t win_off = win->offset;
620         return win_off <= offset
621                 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
622 }
623
624 unsigned char *use_pack(struct packed_git *p,
625                 struct pack_window **w_cursor,
626                 off_t offset,
627                 unsigned long *left)
628 {
629         struct pack_window *win = *w_cursor;
630
631         /* Since packfiles end in a hash of their content and it's
632          * pointless to ask for an offset into the middle of that
633          * hash, and the in_window function above wouldn't match
634          * don't allow an offset too close to the end of the file.
635          */
636         if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
637                 die("packfile %s cannot be accessed", p->pack_name);
638         if (offset > (p->pack_size - the_hash_algo->rawsz))
639                 die("offset beyond end of packfile (truncated pack?)");
640         if (offset < 0)
641                 die(_("offset before end of packfile (broken .idx?)"));
642
643         if (!win || !in_window(win, offset)) {
644                 if (win)
645                         win->inuse_cnt--;
646                 for (win = p->windows; win; win = win->next) {
647                         if (in_window(win, offset))
648                                 break;
649                 }
650                 if (!win) {
651                         size_t window_align = packed_git_window_size / 2;
652                         off_t len;
653
654                         if (p->pack_fd == -1 && open_packed_git(p))
655                                 die("packfile %s cannot be accessed", p->pack_name);
656
657                         win = xcalloc(1, sizeof(*win));
658                         win->offset = (offset / window_align) * window_align;
659                         len = p->pack_size - win->offset;
660                         if (len > packed_git_window_size)
661                                 len = packed_git_window_size;
662                         win->len = (size_t)len;
663                         pack_mapped += win->len;
664                         while (packed_git_limit < pack_mapped
665                                 && unuse_one_window(p))
666                                 ; /* nothing */
667                         win->base = xmmap_gently(NULL, win->len,
668                                 PROT_READ, MAP_PRIVATE,
669                                 p->pack_fd, win->offset);
670                         if (win->base == MAP_FAILED)
671                                 die_errno("packfile %s cannot be mapped",
672                                           p->pack_name);
673                         if (!win->offset && win->len == p->pack_size
674                                 && !p->do_not_close)
675                                 close_pack_fd(p);
676                         pack_mmap_calls++;
677                         pack_open_windows++;
678                         if (pack_mapped > peak_pack_mapped)
679                                 peak_pack_mapped = pack_mapped;
680                         if (pack_open_windows > peak_pack_open_windows)
681                                 peak_pack_open_windows = pack_open_windows;
682                         win->next = p->windows;
683                         p->windows = win;
684                 }
685         }
686         if (win != *w_cursor) {
687                 win->last_used = pack_used_ctr++;
688                 win->inuse_cnt++;
689                 *w_cursor = win;
690         }
691         offset -= win->offset;
692         if (left)
693                 *left = win->len - xsize_t(offset);
694         return win->base + offset;
695 }
696
697 void unuse_pack(struct pack_window **w_cursor)
698 {
699         struct pack_window *w = *w_cursor;
700         if (w) {
701                 w->inuse_cnt--;
702                 *w_cursor = NULL;
703         }
704 }
705
706 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
707 {
708         struct stat st;
709         size_t alloc;
710         struct packed_git *p;
711
712         /*
713          * Make sure a corresponding .pack file exists and that
714          * the index looks sane.
715          */
716         if (!strip_suffix_mem(path, &path_len, ".idx"))
717                 return NULL;
718
719         /*
720          * ".promisor" is long enough to hold any suffix we're adding (and
721          * the use xsnprintf double-checks that)
722          */
723         alloc = st_add3(path_len, strlen(".promisor"), 1);
724         p = alloc_packed_git(alloc);
725         memcpy(p->pack_name, path, path_len);
726
727         xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
728         if (!access(p->pack_name, F_OK))
729                 p->pack_keep = 1;
730
731         xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
732         if (!access(p->pack_name, F_OK))
733                 p->pack_promisor = 1;
734
735         xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
736         if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
737                 free(p);
738                 return NULL;
739         }
740
741         /* ok, it looks sane as far as we can check without
742          * actually mapping the pack file.
743          */
744         p->pack_size = st.st_size;
745         p->pack_local = local;
746         p->mtime = st.st_mtime;
747         if (path_len < the_hash_algo->hexsz ||
748             get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
749                 hashclr(p->hash);
750         return p;
751 }
752
753 void install_packed_git(struct repository *r, struct packed_git *pack)
754 {
755         if (pack->pack_fd != -1)
756                 pack_open_fds++;
757
758         pack->next = r->objects->packed_git;
759         r->objects->packed_git = pack;
760
761         hashmap_entry_init(&pack->packmap_ent, strhash(pack->pack_name));
762         hashmap_add(&r->objects->pack_map, &pack->packmap_ent);
763 }
764
765 void (*report_garbage)(unsigned seen_bits, const char *path);
766
767 static void report_helper(const struct string_list *list,
768                           int seen_bits, int first, int last)
769 {
770         if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
771                 return;
772
773         for (; first < last; first++)
774                 report_garbage(seen_bits, list->items[first].string);
775 }
776
777 static void report_pack_garbage(struct string_list *list)
778 {
779         int i, baselen = -1, first = 0, seen_bits = 0;
780
781         if (!report_garbage)
782                 return;
783
784         string_list_sort(list);
785
786         for (i = 0; i < list->nr; i++) {
787                 const char *path = list->items[i].string;
788                 if (baselen != -1 &&
789                     strncmp(path, list->items[first].string, baselen)) {
790                         report_helper(list, seen_bits, first, i);
791                         baselen = -1;
792                         seen_bits = 0;
793                 }
794                 if (baselen == -1) {
795                         const char *dot = strrchr(path, '.');
796                         if (!dot) {
797                                 report_garbage(PACKDIR_FILE_GARBAGE, path);
798                                 continue;
799                         }
800                         baselen = dot - path + 1;
801                         first = i;
802                 }
803                 if (!strcmp(path + baselen, "pack"))
804                         seen_bits |= 1;
805                 else if (!strcmp(path + baselen, "idx"))
806                         seen_bits |= 2;
807         }
808         report_helper(list, seen_bits, first, list->nr);
809 }
810
811 void for_each_file_in_pack_dir(const char *objdir,
812                                each_file_in_pack_dir_fn fn,
813                                void *data)
814 {
815         struct strbuf path = STRBUF_INIT;
816         size_t dirnamelen;
817         DIR *dir;
818         struct dirent *de;
819
820         strbuf_addstr(&path, objdir);
821         strbuf_addstr(&path, "/pack");
822         dir = opendir(path.buf);
823         if (!dir) {
824                 if (errno != ENOENT)
825                         error_errno("unable to open object pack directory: %s",
826                                     path.buf);
827                 strbuf_release(&path);
828                 return;
829         }
830         strbuf_addch(&path, '/');
831         dirnamelen = path.len;
832         while ((de = readdir(dir)) != NULL) {
833                 if (is_dot_or_dotdot(de->d_name))
834                         continue;
835
836                 strbuf_setlen(&path, dirnamelen);
837                 strbuf_addstr(&path, de->d_name);
838
839                 fn(path.buf, path.len, de->d_name, data);
840         }
841
842         closedir(dir);
843         strbuf_release(&path);
844 }
845
846 struct prepare_pack_data {
847         struct repository *r;
848         struct string_list *garbage;
849         int local;
850         struct multi_pack_index *m;
851 };
852
853 static void prepare_pack(const char *full_name, size_t full_name_len,
854                          const char *file_name, void *_data)
855 {
856         struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
857         struct packed_git *p;
858         size_t base_len = full_name_len;
859
860         if (strip_suffix_mem(full_name, &base_len, ".idx") &&
861             !(data->m && midx_contains_pack(data->m, file_name))) {
862                 struct hashmap_entry hent;
863                 char *pack_name = xstrfmt("%.*s.pack", (int)base_len, full_name);
864                 unsigned int hash = strhash(pack_name);
865                 hashmap_entry_init(&hent, hash);
866
867                 /* Don't reopen a pack we already have. */
868                 if (!hashmap_get(&data->r->objects->pack_map, &hent, pack_name)) {
869                         p = add_packed_git(full_name, full_name_len, data->local);
870                         if (p)
871                                 install_packed_git(data->r, p);
872                 }
873                 free(pack_name);
874         }
875
876         if (!report_garbage)
877                 return;
878
879         if (!strcmp(file_name, "multi-pack-index"))
880                 return;
881         if (ends_with(file_name, ".idx") ||
882             ends_with(file_name, ".pack") ||
883             ends_with(file_name, ".bitmap") ||
884             ends_with(file_name, ".keep") ||
885             ends_with(file_name, ".promisor"))
886                 string_list_append(data->garbage, full_name);
887         else
888                 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
889 }
890
891 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
892 {
893         struct prepare_pack_data data;
894         struct string_list garbage = STRING_LIST_INIT_DUP;
895
896         data.m = r->objects->multi_pack_index;
897
898         /* look for the multi-pack-index for this object directory */
899         while (data.m && strcmp(data.m->object_dir, objdir))
900                 data.m = data.m->next;
901
902         data.r = r;
903         data.garbage = &garbage;
904         data.local = local;
905
906         for_each_file_in_pack_dir(objdir, prepare_pack, &data);
907
908         report_pack_garbage(data.garbage);
909         string_list_clear(data.garbage, 0);
910 }
911
912 static void prepare_packed_git(struct repository *r);
913 /*
914  * Give a fast, rough count of the number of objects in the repository. This
915  * ignores loose objects completely. If you have a lot of them, then either
916  * you should repack because your performance will be awful, or they are
917  * all unreachable objects about to be pruned, in which case they're not really
918  * interesting as a measure of repo size in the first place.
919  */
920 unsigned long repo_approximate_object_count(struct repository *r)
921 {
922         if (!r->objects->approximate_object_count_valid) {
923                 unsigned long count;
924                 struct multi_pack_index *m;
925                 struct packed_git *p;
926
927                 prepare_packed_git(r);
928                 count = 0;
929                 for (m = get_multi_pack_index(r); m; m = m->next)
930                         count += m->num_objects;
931                 for (p = r->objects->packed_git; p; p = p->next) {
932                         if (open_pack_index(p))
933                                 continue;
934                         count += p->num_objects;
935                 }
936                 r->objects->approximate_object_count = count;
937         }
938         return r->objects->approximate_object_count;
939 }
940
941 static void *get_next_packed_git(const void *p)
942 {
943         return ((const struct packed_git *)p)->next;
944 }
945
946 static void set_next_packed_git(void *p, void *next)
947 {
948         ((struct packed_git *)p)->next = next;
949 }
950
951 static int sort_pack(const void *a_, const void *b_)
952 {
953         const struct packed_git *a = a_;
954         const struct packed_git *b = b_;
955         int st;
956
957         /*
958          * Local packs tend to contain objects specific to our
959          * variant of the project than remote ones.  In addition,
960          * remote ones could be on a network mounted filesystem.
961          * Favor local ones for these reasons.
962          */
963         st = a->pack_local - b->pack_local;
964         if (st)
965                 return -st;
966
967         /*
968          * Younger packs tend to contain more recent objects,
969          * and more recent objects tend to get accessed more
970          * often.
971          */
972         if (a->mtime < b->mtime)
973                 return 1;
974         else if (a->mtime == b->mtime)
975                 return 0;
976         return -1;
977 }
978
979 static void rearrange_packed_git(struct repository *r)
980 {
981         r->objects->packed_git = llist_mergesort(
982                 r->objects->packed_git, get_next_packed_git,
983                 set_next_packed_git, sort_pack);
984 }
985
986 static void prepare_packed_git_mru(struct repository *r)
987 {
988         struct packed_git *p;
989
990         INIT_LIST_HEAD(&r->objects->packed_git_mru);
991
992         for (p = r->objects->packed_git; p; p = p->next)
993                 list_add_tail(&p->mru, &r->objects->packed_git_mru);
994 }
995
996 static void prepare_packed_git(struct repository *r)
997 {
998         struct object_directory *odb;
999
1000         if (r->objects->packed_git_initialized)
1001                 return;
1002
1003         prepare_alt_odb(r);
1004         for (odb = r->objects->odb; odb; odb = odb->next) {
1005                 int local = (odb == r->objects->odb);
1006                 prepare_multi_pack_index_one(r, odb->path, local);
1007                 prepare_packed_git_one(r, odb->path, local);
1008         }
1009         rearrange_packed_git(r);
1010
1011         prepare_packed_git_mru(r);
1012         r->objects->packed_git_initialized = 1;
1013 }
1014
1015 void reprepare_packed_git(struct repository *r)
1016 {
1017         struct object_directory *odb;
1018
1019         for (odb = r->objects->odb; odb; odb = odb->next)
1020                 odb_clear_loose_cache(odb);
1021
1022         r->objects->approximate_object_count_valid = 0;
1023         r->objects->packed_git_initialized = 0;
1024         prepare_packed_git(r);
1025 }
1026
1027 struct packed_git *get_packed_git(struct repository *r)
1028 {
1029         prepare_packed_git(r);
1030         return r->objects->packed_git;
1031 }
1032
1033 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1034 {
1035         prepare_packed_git(r);
1036         return r->objects->multi_pack_index;
1037 }
1038
1039 struct packed_git *get_all_packs(struct repository *r)
1040 {
1041         struct multi_pack_index *m;
1042
1043         prepare_packed_git(r);
1044         for (m = r->objects->multi_pack_index; m; m = m->next) {
1045                 uint32_t i;
1046                 for (i = 0; i < m->num_packs; i++)
1047                         prepare_midx_pack(r, m, i);
1048         }
1049
1050         return r->objects->packed_git;
1051 }
1052
1053 struct list_head *get_packed_git_mru(struct repository *r)
1054 {
1055         prepare_packed_git(r);
1056         return &r->objects->packed_git_mru;
1057 }
1058
1059 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1060                 unsigned long len, enum object_type *type, unsigned long *sizep)
1061 {
1062         unsigned shift;
1063         unsigned long size, c;
1064         unsigned long used = 0;
1065
1066         c = buf[used++];
1067         *type = (c >> 4) & 7;
1068         size = c & 15;
1069         shift = 4;
1070         while (c & 0x80) {
1071                 if (len <= used || bitsizeof(long) <= shift) {
1072                         error("bad object header");
1073                         size = used = 0;
1074                         break;
1075                 }
1076                 c = buf[used++];
1077                 size += (c & 0x7f) << shift;
1078                 shift += 7;
1079         }
1080         *sizep = size;
1081         return used;
1082 }
1083
1084 unsigned long get_size_from_delta(struct packed_git *p,
1085                                   struct pack_window **w_curs,
1086                                   off_t curpos)
1087 {
1088         const unsigned char *data;
1089         unsigned char delta_head[20], *in;
1090         git_zstream stream;
1091         int st;
1092
1093         memset(&stream, 0, sizeof(stream));
1094         stream.next_out = delta_head;
1095         stream.avail_out = sizeof(delta_head);
1096
1097         git_inflate_init(&stream);
1098         do {
1099                 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1100                 stream.next_in = in;
1101                 st = git_inflate(&stream, Z_FINISH);
1102                 curpos += stream.next_in - in;
1103         } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1104                  stream.total_out < sizeof(delta_head));
1105         git_inflate_end(&stream);
1106         if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1107                 error("delta data unpack-initial failed");
1108                 return 0;
1109         }
1110
1111         /* Examine the initial part of the delta to figure out
1112          * the result size.
1113          */
1114         data = delta_head;
1115
1116         /* ignore base size */
1117         get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1118
1119         /* Read the result size */
1120         return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1121 }
1122
1123 int unpack_object_header(struct packed_git *p,
1124                          struct pack_window **w_curs,
1125                          off_t *curpos,
1126                          unsigned long *sizep)
1127 {
1128         unsigned char *base;
1129         unsigned long left;
1130         unsigned long used;
1131         enum object_type type;
1132
1133         /* use_pack() assures us we have [base, base + 20) available
1134          * as a range that we can look at.  (Its actually the hash
1135          * size that is assured.)  With our object header encoding
1136          * the maximum deflated object size is 2^137, which is just
1137          * insane, so we know won't exceed what we have been given.
1138          */
1139         base = use_pack(p, w_curs, *curpos, &left);
1140         used = unpack_object_header_buffer(base, left, &type, sizep);
1141         if (!used) {
1142                 type = OBJ_BAD;
1143         } else
1144                 *curpos += used;
1145
1146         return type;
1147 }
1148
1149 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1150 {
1151         unsigned i;
1152         const unsigned hashsz = the_hash_algo->rawsz;
1153         for (i = 0; i < p->num_bad_objects; i++)
1154                 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1155                         return;
1156         p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1157                                       st_mult(GIT_MAX_RAWSZ,
1158                                               st_add(p->num_bad_objects, 1)));
1159         hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1160         p->num_bad_objects++;
1161 }
1162
1163 const struct packed_git *has_packed_and_bad(struct repository *r,
1164                                             const unsigned char *sha1)
1165 {
1166         struct packed_git *p;
1167         unsigned i;
1168
1169         for (p = r->objects->packed_git; p; p = p->next)
1170                 for (i = 0; i < p->num_bad_objects; i++)
1171                         if (hasheq(sha1,
1172                                    p->bad_object_sha1 + the_hash_algo->rawsz * i))
1173                                 return p;
1174         return NULL;
1175 }
1176
1177 static off_t get_delta_base(struct packed_git *p,
1178                                     struct pack_window **w_curs,
1179                                     off_t *curpos,
1180                                     enum object_type type,
1181                                     off_t delta_obj_offset)
1182 {
1183         unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1184         off_t base_offset;
1185
1186         /* use_pack() assured us we have [base_info, base_info + 20)
1187          * as a range that we can look at without walking off the
1188          * end of the mapped window.  Its actually the hash size
1189          * that is assured.  An OFS_DELTA longer than the hash size
1190          * is stupid, as then a REF_DELTA would be smaller to store.
1191          */
1192         if (type == OBJ_OFS_DELTA) {
1193                 unsigned used = 0;
1194                 unsigned char c = base_info[used++];
1195                 base_offset = c & 127;
1196                 while (c & 128) {
1197                         base_offset += 1;
1198                         if (!base_offset || MSB(base_offset, 7))
1199                                 return 0;  /* overflow */
1200                         c = base_info[used++];
1201                         base_offset = (base_offset << 7) + (c & 127);
1202                 }
1203                 base_offset = delta_obj_offset - base_offset;
1204                 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1205                         return 0;  /* out of bound */
1206                 *curpos += used;
1207         } else if (type == OBJ_REF_DELTA) {
1208                 /* The base entry _must_ be in the same pack */
1209                 base_offset = find_pack_entry_one(base_info, p);
1210                 *curpos += the_hash_algo->rawsz;
1211         } else
1212                 die("I am totally screwed");
1213         return base_offset;
1214 }
1215
1216 /*
1217  * Like get_delta_base above, but we return the sha1 instead of the pack
1218  * offset. This means it is cheaper for REF deltas (we do not have to do
1219  * the final object lookup), but more expensive for OFS deltas (we
1220  * have to load the revidx to convert the offset back into a sha1).
1221  */
1222 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1223                                                 struct pack_window **w_curs,
1224                                                 off_t curpos,
1225                                                 enum object_type type,
1226                                                 off_t delta_obj_offset)
1227 {
1228         if (type == OBJ_REF_DELTA) {
1229                 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1230                 return base;
1231         } else if (type == OBJ_OFS_DELTA) {
1232                 struct revindex_entry *revidx;
1233                 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1234                                                    type, delta_obj_offset);
1235
1236                 if (!base_offset)
1237                         return NULL;
1238
1239                 revidx = find_pack_revindex(p, base_offset);
1240                 if (!revidx)
1241                         return NULL;
1242
1243                 return nth_packed_object_sha1(p, revidx->nr);
1244         } else
1245                 return NULL;
1246 }
1247
1248 static int retry_bad_packed_offset(struct repository *r,
1249                                    struct packed_git *p,
1250                                    off_t obj_offset)
1251 {
1252         int type;
1253         struct revindex_entry *revidx;
1254         struct object_id oid;
1255         revidx = find_pack_revindex(p, obj_offset);
1256         if (!revidx)
1257                 return OBJ_BAD;
1258         nth_packed_object_oid(&oid, p, revidx->nr);
1259         mark_bad_packed_object(p, oid.hash);
1260         type = oid_object_info(r, &oid, NULL);
1261         if (type <= OBJ_NONE)
1262                 return OBJ_BAD;
1263         return type;
1264 }
1265
1266 #define POI_STACK_PREALLOC 64
1267
1268 static enum object_type packed_to_object_type(struct repository *r,
1269                                               struct packed_git *p,
1270                                               off_t obj_offset,
1271                                               enum object_type type,
1272                                               struct pack_window **w_curs,
1273                                               off_t curpos)
1274 {
1275         off_t small_poi_stack[POI_STACK_PREALLOC];
1276         off_t *poi_stack = small_poi_stack;
1277         int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1278
1279         while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1280                 off_t base_offset;
1281                 unsigned long size;
1282                 /* Push the object we're going to leave behind */
1283                 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1284                         poi_stack_alloc = alloc_nr(poi_stack_nr);
1285                         ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1286                         COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1287                 } else {
1288                         ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1289                 }
1290                 poi_stack[poi_stack_nr++] = obj_offset;
1291                 /* If parsing the base offset fails, just unwind */
1292                 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1293                 if (!base_offset)
1294                         goto unwind;
1295                 curpos = obj_offset = base_offset;
1296                 type = unpack_object_header(p, w_curs, &curpos, &size);
1297                 if (type <= OBJ_NONE) {
1298                         /* If getting the base itself fails, we first
1299                          * retry the base, otherwise unwind */
1300                         type = retry_bad_packed_offset(r, p, base_offset);
1301                         if (type > OBJ_NONE)
1302                                 goto out;
1303                         goto unwind;
1304                 }
1305         }
1306
1307         switch (type) {
1308         case OBJ_BAD:
1309         case OBJ_COMMIT:
1310         case OBJ_TREE:
1311         case OBJ_BLOB:
1312         case OBJ_TAG:
1313                 break;
1314         default:
1315                 error("unknown object type %i at offset %"PRIuMAX" in %s",
1316                       type, (uintmax_t)obj_offset, p->pack_name);
1317                 type = OBJ_BAD;
1318         }
1319
1320 out:
1321         if (poi_stack != small_poi_stack)
1322                 free(poi_stack);
1323         return type;
1324
1325 unwind:
1326         while (poi_stack_nr) {
1327                 obj_offset = poi_stack[--poi_stack_nr];
1328                 type = retry_bad_packed_offset(r, p, obj_offset);
1329                 if (type > OBJ_NONE)
1330                         goto out;
1331         }
1332         type = OBJ_BAD;
1333         goto out;
1334 }
1335
1336 static struct hashmap delta_base_cache;
1337 static size_t delta_base_cached;
1338
1339 static LIST_HEAD(delta_base_cache_lru);
1340
1341 struct delta_base_cache_key {
1342         struct packed_git *p;
1343         off_t base_offset;
1344 };
1345
1346 struct delta_base_cache_entry {
1347         struct hashmap_entry ent;
1348         struct delta_base_cache_key key;
1349         struct list_head lru;
1350         void *data;
1351         unsigned long size;
1352         enum object_type type;
1353 };
1354
1355 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1356 {
1357         unsigned int hash;
1358
1359         hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1360         hash += (hash >> 8) + (hash >> 16);
1361         return hash;
1362 }
1363
1364 static struct delta_base_cache_entry *
1365 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1366 {
1367         struct hashmap_entry entry, *e;
1368         struct delta_base_cache_key key;
1369
1370         if (!delta_base_cache.cmpfn)
1371                 return NULL;
1372
1373         hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1374         key.p = p;
1375         key.base_offset = base_offset;
1376         e = hashmap_get(&delta_base_cache, &entry, &key);
1377         return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1378 }
1379
1380 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1381                                    const struct delta_base_cache_key *b)
1382 {
1383         return a->p == b->p && a->base_offset == b->base_offset;
1384 }
1385
1386 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1387                                      const struct hashmap_entry *va,
1388                                      const struct hashmap_entry *vb,
1389                                      const void *vkey)
1390 {
1391         const struct delta_base_cache_entry *a, *b;
1392         const struct delta_base_cache_key *key = vkey;
1393
1394         a = container_of(va, const struct delta_base_cache_entry, ent);
1395         b = container_of(vb, const struct delta_base_cache_entry, ent);
1396
1397         if (key)
1398                 return !delta_base_cache_key_eq(&a->key, key);
1399         else
1400                 return !delta_base_cache_key_eq(&a->key, &b->key);
1401 }
1402
1403 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1404 {
1405         return !!get_delta_base_cache_entry(p, base_offset);
1406 }
1407
1408 /*
1409  * Remove the entry from the cache, but do _not_ free the associated
1410  * entry data. The caller takes ownership of the "data" buffer, and
1411  * should copy out any fields it wants before detaching.
1412  */
1413 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1414 {
1415         hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1416         list_del(&ent->lru);
1417         delta_base_cached -= ent->size;
1418         free(ent);
1419 }
1420
1421 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1422                                    off_t base_offset, unsigned long *base_size,
1423                                    enum object_type *type)
1424 {
1425         struct delta_base_cache_entry *ent;
1426
1427         ent = get_delta_base_cache_entry(p, base_offset);
1428         if (!ent)
1429                 return unpack_entry(r, p, base_offset, type, base_size);
1430
1431         if (type)
1432                 *type = ent->type;
1433         if (base_size)
1434                 *base_size = ent->size;
1435         return xmemdupz(ent->data, ent->size);
1436 }
1437
1438 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1439 {
1440         free(ent->data);
1441         detach_delta_base_cache_entry(ent);
1442 }
1443
1444 void clear_delta_base_cache(void)
1445 {
1446         struct list_head *lru, *tmp;
1447         list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1448                 struct delta_base_cache_entry *entry =
1449                         list_entry(lru, struct delta_base_cache_entry, lru);
1450                 release_delta_base_cache(entry);
1451         }
1452 }
1453
1454 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1455         void *base, unsigned long base_size, enum object_type type)
1456 {
1457         struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1458         struct list_head *lru, *tmp;
1459
1460         delta_base_cached += base_size;
1461
1462         list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1463                 struct delta_base_cache_entry *f =
1464                         list_entry(lru, struct delta_base_cache_entry, lru);
1465                 if (delta_base_cached <= delta_base_cache_limit)
1466                         break;
1467                 release_delta_base_cache(f);
1468         }
1469
1470         ent->key.p = p;
1471         ent->key.base_offset = base_offset;
1472         ent->type = type;
1473         ent->data = base;
1474         ent->size = base_size;
1475         list_add_tail(&ent->lru, &delta_base_cache_lru);
1476
1477         if (!delta_base_cache.cmpfn)
1478                 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1479         hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1480         hashmap_add(&delta_base_cache, &ent->ent);
1481 }
1482
1483 int packed_object_info(struct repository *r, struct packed_git *p,
1484                        off_t obj_offset, struct object_info *oi)
1485 {
1486         struct pack_window *w_curs = NULL;
1487         unsigned long size;
1488         off_t curpos = obj_offset;
1489         enum object_type type;
1490
1491         /*
1492          * We always get the representation type, but only convert it to
1493          * a "real" type later if the caller is interested.
1494          */
1495         if (oi->contentp) {
1496                 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1497                                                       &type);
1498                 if (!*oi->contentp)
1499                         type = OBJ_BAD;
1500         } else {
1501                 type = unpack_object_header(p, &w_curs, &curpos, &size);
1502         }
1503
1504         if (!oi->contentp && oi->sizep) {
1505                 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1506                         off_t tmp_pos = curpos;
1507                         off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1508                                                            type, obj_offset);
1509                         if (!base_offset) {
1510                                 type = OBJ_BAD;
1511                                 goto out;
1512                         }
1513                         *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1514                         if (*oi->sizep == 0) {
1515                                 type = OBJ_BAD;
1516                                 goto out;
1517                         }
1518                 } else {
1519                         *oi->sizep = size;
1520                 }
1521         }
1522
1523         if (oi->disk_sizep) {
1524                 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1525                 *oi->disk_sizep = revidx[1].offset - obj_offset;
1526         }
1527
1528         if (oi->typep || oi->type_name) {
1529                 enum object_type ptot;
1530                 ptot = packed_to_object_type(r, p, obj_offset,
1531                                              type, &w_curs, curpos);
1532                 if (oi->typep)
1533                         *oi->typep = ptot;
1534                 if (oi->type_name) {
1535                         const char *tn = type_name(ptot);
1536                         if (tn)
1537                                 strbuf_addstr(oi->type_name, tn);
1538                 }
1539                 if (ptot < 0) {
1540                         type = OBJ_BAD;
1541                         goto out;
1542                 }
1543         }
1544
1545         if (oi->delta_base_sha1) {
1546                 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1547                         const unsigned char *base;
1548
1549                         base = get_delta_base_sha1(p, &w_curs, curpos,
1550                                                    type, obj_offset);
1551                         if (!base) {
1552                                 type = OBJ_BAD;
1553                                 goto out;
1554                         }
1555
1556                         hashcpy(oi->delta_base_sha1, base);
1557                 } else
1558                         hashclr(oi->delta_base_sha1);
1559         }
1560
1561         oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1562                                                           OI_PACKED;
1563
1564 out:
1565         unuse_pack(&w_curs);
1566         return type;
1567 }
1568
1569 static void *unpack_compressed_entry(struct packed_git *p,
1570                                     struct pack_window **w_curs,
1571                                     off_t curpos,
1572                                     unsigned long size)
1573 {
1574         int st;
1575         git_zstream stream;
1576         unsigned char *buffer, *in;
1577
1578         buffer = xmallocz_gently(size);
1579         if (!buffer)
1580                 return NULL;
1581         memset(&stream, 0, sizeof(stream));
1582         stream.next_out = buffer;
1583         stream.avail_out = size + 1;
1584
1585         git_inflate_init(&stream);
1586         do {
1587                 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1588                 stream.next_in = in;
1589                 st = git_inflate(&stream, Z_FINISH);
1590                 if (!stream.avail_out)
1591                         break; /* the payload is larger than it should be */
1592                 curpos += stream.next_in - in;
1593         } while (st == Z_OK || st == Z_BUF_ERROR);
1594         git_inflate_end(&stream);
1595         if ((st != Z_STREAM_END) || stream.total_out != size) {
1596                 free(buffer);
1597                 return NULL;
1598         }
1599
1600         /* versions of zlib can clobber unconsumed portion of outbuf */
1601         buffer[size] = '\0';
1602
1603         return buffer;
1604 }
1605
1606 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1607 {
1608         static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1609         trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1610                          p->pack_name, (uintmax_t)obj_offset);
1611 }
1612
1613 int do_check_packed_object_crc;
1614
1615 #define UNPACK_ENTRY_STACK_PREALLOC 64
1616 struct unpack_entry_stack_ent {
1617         off_t obj_offset;
1618         off_t curpos;
1619         unsigned long size;
1620 };
1621
1622 static void *read_object(struct repository *r,
1623                          const struct object_id *oid,
1624                          enum object_type *type,
1625                          unsigned long *size)
1626 {
1627         struct object_info oi = OBJECT_INFO_INIT;
1628         void *content;
1629         oi.typep = type;
1630         oi.sizep = size;
1631         oi.contentp = &content;
1632
1633         if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1634                 return NULL;
1635         return content;
1636 }
1637
1638 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1639                    enum object_type *final_type, unsigned long *final_size)
1640 {
1641         struct pack_window *w_curs = NULL;
1642         off_t curpos = obj_offset;
1643         void *data = NULL;
1644         unsigned long size;
1645         enum object_type type;
1646         struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1647         struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1648         int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1649         int base_from_cache = 0;
1650
1651         write_pack_access_log(p, obj_offset);
1652
1653         /* PHASE 1: drill down to the innermost base object */
1654         for (;;) {
1655                 off_t base_offset;
1656                 int i;
1657                 struct delta_base_cache_entry *ent;
1658
1659                 ent = get_delta_base_cache_entry(p, curpos);
1660                 if (ent) {
1661                         type = ent->type;
1662                         data = ent->data;
1663                         size = ent->size;
1664                         detach_delta_base_cache_entry(ent);
1665                         base_from_cache = 1;
1666                         break;
1667                 }
1668
1669                 if (do_check_packed_object_crc && p->index_version > 1) {
1670                         struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1671                         off_t len = revidx[1].offset - obj_offset;
1672                         if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1673                                 struct object_id oid;
1674                                 nth_packed_object_oid(&oid, p, revidx->nr);
1675                                 error("bad packed object CRC for %s",
1676                                       oid_to_hex(&oid));
1677                                 mark_bad_packed_object(p, oid.hash);
1678                                 data = NULL;
1679                                 goto out;
1680                         }
1681                 }
1682
1683                 type = unpack_object_header(p, &w_curs, &curpos, &size);
1684                 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1685                         break;
1686
1687                 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1688                 if (!base_offset) {
1689                         error("failed to validate delta base reference "
1690                               "at offset %"PRIuMAX" from %s",
1691                               (uintmax_t)curpos, p->pack_name);
1692                         /* bail to phase 2, in hopes of recovery */
1693                         data = NULL;
1694                         break;
1695                 }
1696
1697                 /* push object, proceed to base */
1698                 if (delta_stack_nr >= delta_stack_alloc
1699                     && delta_stack == small_delta_stack) {
1700                         delta_stack_alloc = alloc_nr(delta_stack_nr);
1701                         ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1702                         COPY_ARRAY(delta_stack, small_delta_stack,
1703                                    delta_stack_nr);
1704                 } else {
1705                         ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1706                 }
1707                 i = delta_stack_nr++;
1708                 delta_stack[i].obj_offset = obj_offset;
1709                 delta_stack[i].curpos = curpos;
1710                 delta_stack[i].size = size;
1711
1712                 curpos = obj_offset = base_offset;
1713         }
1714
1715         /* PHASE 2: handle the base */
1716         switch (type) {
1717         case OBJ_OFS_DELTA:
1718         case OBJ_REF_DELTA:
1719                 if (data)
1720                         BUG("unpack_entry: left loop at a valid delta");
1721                 break;
1722         case OBJ_COMMIT:
1723         case OBJ_TREE:
1724         case OBJ_BLOB:
1725         case OBJ_TAG:
1726                 if (!base_from_cache)
1727                         data = unpack_compressed_entry(p, &w_curs, curpos, size);
1728                 break;
1729         default:
1730                 data = NULL;
1731                 error("unknown object type %i at offset %"PRIuMAX" in %s",
1732                       type, (uintmax_t)obj_offset, p->pack_name);
1733         }
1734
1735         /* PHASE 3: apply deltas in order */
1736
1737         /* invariants:
1738          *   'data' holds the base data, or NULL if there was corruption
1739          */
1740         while (delta_stack_nr) {
1741                 void *delta_data;
1742                 void *base = data;
1743                 void *external_base = NULL;
1744                 unsigned long delta_size, base_size = size;
1745                 int i;
1746
1747                 data = NULL;
1748
1749                 if (base)
1750                         add_delta_base_cache(p, obj_offset, base, base_size, type);
1751
1752                 if (!base) {
1753                         /*
1754                          * We're probably in deep shit, but let's try to fetch
1755                          * the required base anyway from another pack or loose.
1756                          * This is costly but should happen only in the presence
1757                          * of a corrupted pack, and is better than failing outright.
1758                          */
1759                         struct revindex_entry *revidx;
1760                         struct object_id base_oid;
1761                         revidx = find_pack_revindex(p, obj_offset);
1762                         if (revidx) {
1763                                 nth_packed_object_oid(&base_oid, p, revidx->nr);
1764                                 error("failed to read delta base object %s"
1765                                       " at offset %"PRIuMAX" from %s",
1766                                       oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1767                                       p->pack_name);
1768                                 mark_bad_packed_object(p, base_oid.hash);
1769                                 base = read_object(r, &base_oid, &type, &base_size);
1770                                 external_base = base;
1771                         }
1772                 }
1773
1774                 i = --delta_stack_nr;
1775                 obj_offset = delta_stack[i].obj_offset;
1776                 curpos = delta_stack[i].curpos;
1777                 delta_size = delta_stack[i].size;
1778
1779                 if (!base)
1780                         continue;
1781
1782                 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1783
1784                 if (!delta_data) {
1785                         error("failed to unpack compressed delta "
1786                               "at offset %"PRIuMAX" from %s",
1787                               (uintmax_t)curpos, p->pack_name);
1788                         data = NULL;
1789                         free(external_base);
1790                         continue;
1791                 }
1792
1793                 data = patch_delta(base, base_size,
1794                                    delta_data, delta_size,
1795                                    &size);
1796
1797                 /*
1798                  * We could not apply the delta; warn the user, but keep going.
1799                  * Our failure will be noticed either in the next iteration of
1800                  * the loop, or if this is the final delta, in the caller when
1801                  * we return NULL. Those code paths will take care of making
1802                  * a more explicit warning and retrying with another copy of
1803                  * the object.
1804                  */
1805                 if (!data)
1806                         error("failed to apply delta");
1807
1808                 free(delta_data);
1809                 free(external_base);
1810         }
1811
1812         if (final_type)
1813                 *final_type = type;
1814         if (final_size)
1815                 *final_size = size;
1816
1817 out:
1818         unuse_pack(&w_curs);
1819
1820         if (delta_stack != small_delta_stack)
1821                 free(delta_stack);
1822
1823         return data;
1824 }
1825
1826 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1827 {
1828         const unsigned char *index_fanout = p->index_data;
1829         const unsigned char *index_lookup;
1830         const unsigned int hashsz = the_hash_algo->rawsz;
1831         int index_lookup_width;
1832
1833         if (!index_fanout)
1834                 BUG("bsearch_pack called without a valid pack-index");
1835
1836         index_lookup = index_fanout + 4 * 256;
1837         if (p->index_version == 1) {
1838                 index_lookup_width = hashsz + 4;
1839                 index_lookup += 4;
1840         } else {
1841                 index_lookup_width = hashsz;
1842                 index_fanout += 8;
1843                 index_lookup += 8;
1844         }
1845
1846         return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1847                             index_lookup, index_lookup_width, result);
1848 }
1849
1850 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1851                                             uint32_t n)
1852 {
1853         const unsigned char *index = p->index_data;
1854         const unsigned int hashsz = the_hash_algo->rawsz;
1855         if (!index) {
1856                 if (open_pack_index(p))
1857                         return NULL;
1858                 index = p->index_data;
1859         }
1860         if (n >= p->num_objects)
1861                 return NULL;
1862         index += 4 * 256;
1863         if (p->index_version == 1) {
1864                 return index + (hashsz + 4) * n + 4;
1865         } else {
1866                 index += 8;
1867                 return index + hashsz * n;
1868         }
1869 }
1870
1871 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1872                                               struct packed_git *p,
1873                                               uint32_t n)
1874 {
1875         const unsigned char *hash = nth_packed_object_sha1(p, n);
1876         if (!hash)
1877                 return NULL;
1878         hashcpy(oid->hash, hash);
1879         return oid;
1880 }
1881
1882 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1883 {
1884         const unsigned char *ptr = vptr;
1885         const unsigned char *start = p->index_data;
1886         const unsigned char *end = start + p->index_size;
1887         if (ptr < start)
1888                 die(_("offset before start of pack index for %s (corrupt index?)"),
1889                     p->pack_name);
1890         /* No need to check for underflow; .idx files must be at least 8 bytes */
1891         if (ptr >= end - 8)
1892                 die(_("offset beyond end of pack index for %s (truncated index?)"),
1893                     p->pack_name);
1894 }
1895
1896 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1897 {
1898         const unsigned char *index = p->index_data;
1899         const unsigned int hashsz = the_hash_algo->rawsz;
1900         index += 4 * 256;
1901         if (p->index_version == 1) {
1902                 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1903         } else {
1904                 uint32_t off;
1905                 index += 8 + p->num_objects * (hashsz + 4);
1906                 off = ntohl(*((uint32_t *)(index + 4 * n)));
1907                 if (!(off & 0x80000000))
1908                         return off;
1909                 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1910                 check_pack_index_ptr(p, index);
1911                 return get_be64(index);
1912         }
1913 }
1914
1915 off_t find_pack_entry_one(const unsigned char *sha1,
1916                                   struct packed_git *p)
1917 {
1918         const unsigned char *index = p->index_data;
1919         struct object_id oid;
1920         uint32_t result;
1921
1922         if (!index) {
1923                 if (open_pack_index(p))
1924                         return 0;
1925         }
1926
1927         hashcpy(oid.hash, sha1);
1928         if (bsearch_pack(&oid, p, &result))
1929                 return nth_packed_object_offset(p, result);
1930         return 0;
1931 }
1932
1933 int is_pack_valid(struct packed_git *p)
1934 {
1935         /* An already open pack is known to be valid. */
1936         if (p->pack_fd != -1)
1937                 return 1;
1938
1939         /* If the pack has one window completely covering the
1940          * file size, the pack is known to be valid even if
1941          * the descriptor is not currently open.
1942          */
1943         if (p->windows) {
1944                 struct pack_window *w = p->windows;
1945
1946                 if (!w->offset && w->len == p->pack_size)
1947                         return 1;
1948         }
1949
1950         /* Force the pack to open to prove its valid. */
1951         return !open_packed_git(p);
1952 }
1953
1954 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1955                                   struct packed_git *packs)
1956 {
1957         struct packed_git *p;
1958
1959         for (p = packs; p; p = p->next) {
1960                 if (find_pack_entry_one(sha1, p))
1961                         return p;
1962         }
1963         return NULL;
1964
1965 }
1966
1967 static int fill_pack_entry(const struct object_id *oid,
1968                            struct pack_entry *e,
1969                            struct packed_git *p)
1970 {
1971         off_t offset;
1972
1973         if (p->num_bad_objects) {
1974                 unsigned i;
1975                 for (i = 0; i < p->num_bad_objects; i++)
1976                         if (hasheq(oid->hash,
1977                                    p->bad_object_sha1 + the_hash_algo->rawsz * i))
1978                                 return 0;
1979         }
1980
1981         offset = find_pack_entry_one(oid->hash, p);
1982         if (!offset)
1983                 return 0;
1984
1985         /*
1986          * We are about to tell the caller where they can locate the
1987          * requested object.  We better make sure the packfile is
1988          * still here and can be accessed before supplying that
1989          * answer, as it may have been deleted since the index was
1990          * loaded!
1991          */
1992         if (!is_pack_valid(p))
1993                 return 0;
1994         e->offset = offset;
1995         e->p = p;
1996         return 1;
1997 }
1998
1999 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2000 {
2001         struct list_head *pos;
2002         struct multi_pack_index *m;
2003
2004         prepare_packed_git(r);
2005         if (!r->objects->packed_git && !r->objects->multi_pack_index)
2006                 return 0;
2007
2008         for (m = r->objects->multi_pack_index; m; m = m->next) {
2009                 if (fill_midx_entry(r, oid, e, m))
2010                         return 1;
2011         }
2012
2013         list_for_each(pos, &r->objects->packed_git_mru) {
2014                 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2015                 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2016                         list_move(&p->mru, &r->objects->packed_git_mru);
2017                         return 1;
2018                 }
2019         }
2020         return 0;
2021 }
2022
2023 int has_object_pack(const struct object_id *oid)
2024 {
2025         struct pack_entry e;
2026         return find_pack_entry(the_repository, oid, &e);
2027 }
2028
2029 int has_pack_index(const unsigned char *sha1)
2030 {
2031         struct stat st;
2032         if (stat(sha1_pack_index_name(sha1), &st))
2033                 return 0;
2034         return 1;
2035 }
2036
2037 int for_each_object_in_pack(struct packed_git *p,
2038                             each_packed_object_fn cb, void *data,
2039                             enum for_each_object_flags flags)
2040 {
2041         uint32_t i;
2042         int r = 0;
2043
2044         if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2045                 if (load_pack_revindex(p))
2046                         return -1;
2047         }
2048
2049         for (i = 0; i < p->num_objects; i++) {
2050                 uint32_t pos;
2051                 struct object_id oid;
2052
2053                 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2054                         pos = p->revindex[i].nr;
2055                 else
2056                         pos = i;
2057
2058                 if (!nth_packed_object_oid(&oid, p, pos))
2059                         return error("unable to get sha1 of object %u in %s",
2060                                      pos, p->pack_name);
2061
2062                 r = cb(&oid, p, pos, data);
2063                 if (r)
2064                         break;
2065         }
2066         return r;
2067 }
2068
2069 int for_each_packed_object(each_packed_object_fn cb, void *data,
2070                            enum for_each_object_flags flags)
2071 {
2072         struct packed_git *p;
2073         int r = 0;
2074         int pack_errors = 0;
2075
2076         prepare_packed_git(the_repository);
2077         for (p = get_all_packs(the_repository); p; p = p->next) {
2078                 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2079                         continue;
2080                 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2081                     !p->pack_promisor)
2082                         continue;
2083                 if (open_pack_index(p)) {
2084                         pack_errors = 1;
2085                         continue;
2086                 }
2087                 r = for_each_object_in_pack(p, cb, data, flags);
2088                 if (r)
2089                         break;
2090         }
2091         return r ? r : pack_errors;
2092 }
2093
2094 static int add_promisor_object(const struct object_id *oid,
2095                                struct packed_git *pack,
2096                                uint32_t pos,
2097                                void *set_)
2098 {
2099         struct oidset *set = set_;
2100         struct object *obj = parse_object(the_repository, oid);
2101         if (!obj)
2102                 return 1;
2103
2104         oidset_insert(set, oid);
2105
2106         /*
2107          * If this is a tree, commit, or tag, the objects it refers
2108          * to are also promisor objects. (Blobs refer to no objects->)
2109          */
2110         if (obj->type == OBJ_TREE) {
2111                 struct tree *tree = (struct tree *)obj;
2112                 struct tree_desc desc;
2113                 struct name_entry entry;
2114                 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2115                         /*
2116                          * Error messages are given when packs are
2117                          * verified, so do not print any here.
2118                          */
2119                         return 0;
2120                 while (tree_entry_gently(&desc, &entry))
2121                         oidset_insert(set, &entry.oid);
2122         } else if (obj->type == OBJ_COMMIT) {
2123                 struct commit *commit = (struct commit *) obj;
2124                 struct commit_list *parents = commit->parents;
2125
2126                 oidset_insert(set, get_commit_tree_oid(commit));
2127                 for (; parents; parents = parents->next)
2128                         oidset_insert(set, &parents->item->object.oid);
2129         } else if (obj->type == OBJ_TAG) {
2130                 struct tag *tag = (struct tag *) obj;
2131                 oidset_insert(set, get_tagged_oid(tag));
2132         }
2133         return 0;
2134 }
2135
2136 int is_promisor_object(const struct object_id *oid)
2137 {
2138         static struct oidset promisor_objects;
2139         static int promisor_objects_prepared;
2140
2141         if (!promisor_objects_prepared) {
2142                 if (has_promisor_remote()) {
2143                         for_each_packed_object(add_promisor_object,
2144                                                &promisor_objects,
2145                                                FOR_EACH_OBJECT_PROMISOR_ONLY);
2146                 }
2147                 promisor_objects_prepared = 1;
2148         }
2149         return oidset_contains(&promisor_objects, oid);
2150 }