packfile: add repository argument to unpack_entry
[git/git.git] / packfile.c
CommitLineData
4f39cd82 1#include "cache.h"
ec2dd32c 2#include "list.h"
0317f455 3#include "pack.h"
031dc927 4#include "repository.h"
0abe14f6
JT
5#include "dir.h"
6#include "mergesort.h"
7#include "packfile.h"
7b3aa75d 8#include "delta.h"
f1d8130b
JT
9#include "list.h"
10#include "streaming.h"
a2551953 11#include "sha1-lookup.h"
498f1f61
JT
12#include "commit.h"
13#include "object.h"
14#include "tag.h"
15#include "tree-walk.h"
16#include "tree.h"
0d4a1321 17#include "object-store.h"
4f39cd82
JT
18
19char *odb_pack_name(struct strbuf *buf,
20 const unsigned char *sha1,
21 const char *ext)
22{
23 strbuf_reset(buf);
24 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
25 sha1_to_hex(sha1), ext);
26 return buf->buf;
27}
28
29char *sha1_pack_name(const unsigned char *sha1)
30{
31 static struct strbuf buf = STRBUF_INIT;
32 return odb_pack_name(&buf, sha1, "pack");
33}
34
35char *sha1_pack_index_name(const unsigned char *sha1)
36{
37 static struct strbuf buf = STRBUF_INIT;
38 return odb_pack_name(&buf, sha1, "idx");
39}
6d6a80e0 40
84f80ad5
JT
41static unsigned int pack_used_ctr;
42static unsigned int pack_mmap_calls;
43static unsigned int peak_pack_open_windows;
44static unsigned int pack_open_windows;
e65f1862 45static unsigned int pack_open_fds;
84f80ad5
JT
46static unsigned int pack_max_fds;
47static size_t peak_pack_mapped;
48static size_t pack_mapped;
8e21176c
JT
49
50#define SZ_FMT PRIuMAX
51static inline uintmax_t sz_fmt(size_t s) { return s; }
52
53void pack_report(void)
54{
55 fprintf(stderr,
56 "pack_report: getpagesize() = %10" SZ_FMT "\n"
57 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
58 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
59 sz_fmt(getpagesize()),
60 sz_fmt(packed_git_window_size),
61 sz_fmt(packed_git_limit));
62 fprintf(stderr,
63 "pack_report: pack_used_ctr = %10u\n"
64 "pack_report: pack_mmap_calls = %10u\n"
65 "pack_report: pack_open_windows = %10u / %10u\n"
66 "pack_report: pack_mapped = "
67 "%10" SZ_FMT " / %10" SZ_FMT "\n",
68 pack_used_ctr,
69 pack_mmap_calls,
70 pack_open_windows, peak_pack_open_windows,
71 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
72}
0317f455
JT
73
74/*
75 * Open and mmap the index file at path, perform a couple of
76 * consistency checks, then record its information to p. Return 0 on
77 * success.
78 */
79static int check_packed_git_idx(const char *path, struct packed_git *p)
80{
81 void *idx_map;
82 struct pack_idx_header *hdr;
83 size_t idx_size;
84 uint32_t version, nr, i, *index;
85 int fd = git_open(path);
86 struct stat st;
87
88 if (fd < 0)
89 return -1;
90 if (fstat(fd, &st)) {
91 close(fd);
92 return -1;
93 }
94 idx_size = xsize_t(st.st_size);
95 if (idx_size < 4 * 256 + 20 + 20) {
96 close(fd);
97 return error("index file %s is too small", path);
98 }
99 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
100 close(fd);
101
102 hdr = idx_map;
103 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
104 version = ntohl(hdr->idx_version);
105 if (version < 2 || version > 2) {
106 munmap(idx_map, idx_size);
107 return error("index file %s is version %"PRIu32
108 " and is not supported by this binary"
109 " (try upgrading GIT to a newer version)",
110 path, version);
111 }
112 } else
113 version = 1;
114
115 nr = 0;
116 index = idx_map;
117 if (version > 1)
118 index += 2; /* skip index header */
119 for (i = 0; i < 256; i++) {
120 uint32_t n = ntohl(index[i]);
121 if (n < nr) {
122 munmap(idx_map, idx_size);
123 return error("non-monotonic index %s", path);
124 }
125 nr = n;
126 }
127
128 if (version == 1) {
129 /*
130 * Total size:
131 * - 256 index entries 4 bytes each
132 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
133 * - 20-byte SHA1 of the packfile
134 * - 20-byte SHA1 file checksum
135 */
136 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
137 munmap(idx_map, idx_size);
138 return error("wrong index v1 file size in %s", path);
139 }
140 } else if (version == 2) {
141 /*
142 * Minimum size:
143 * - 8 bytes of header
144 * - 256 index entries 4 bytes each
145 * - 20-byte sha1 entry * nr
146 * - 4-byte crc entry * nr
147 * - 4-byte offset entry * nr
148 * - 20-byte SHA1 of the packfile
149 * - 20-byte SHA1 file checksum
150 * And after the 4-byte offset table might be a
151 * variable sized table containing 8-byte entries
152 * for offsets larger than 2^31.
153 */
154 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
155 unsigned long max_size = min_size;
156 if (nr)
157 max_size += (nr - 1)*8;
158 if (idx_size < min_size || idx_size > max_size) {
159 munmap(idx_map, idx_size);
160 return error("wrong index v2 file size in %s", path);
161 }
162 if (idx_size != min_size &&
163 /*
164 * make sure we can deal with large pack offsets.
165 * 31-bit signed offset won't be enough, neither
166 * 32-bit unsigned one will be.
167 */
168 (sizeof(off_t) <= 4)) {
169 munmap(idx_map, idx_size);
170 return error("pack too large for current definition of off_t in %s", path);
171 }
172 }
173
174 p->index_version = version;
175 p->index_data = idx_map;
176 p->index_size = idx_size;
177 p->num_objects = nr;
178 return 0;
179}
180
181int open_pack_index(struct packed_git *p)
182{
183 char *idx_name;
184 size_t len;
185 int ret;
186
187 if (p->index_data)
188 return 0;
189
190 if (!strip_suffix(p->pack_name, ".pack", &len))
191 die("BUG: pack_name does not end in .pack");
192 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
193 ret = check_packed_git_idx(idx_name, p);
194 free(idx_name);
195 return ret;
196}
197
198static struct packed_git *alloc_packed_git(int extra)
199{
200 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
201 memset(p, 0, sizeof(*p));
202 p->pack_fd = -1;
203 return p;
204}
205
206struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
207{
208 const char *path = sha1_pack_name(sha1);
209 size_t alloc = st_add(strlen(path), 1);
210 struct packed_git *p = alloc_packed_git(alloc);
211
212 memcpy(p->pack_name, path, alloc); /* includes NUL */
213 hashcpy(p->sha1, sha1);
214 if (check_packed_git_idx(idx_path, p)) {
215 free(p);
216 return NULL;
217 }
218
219 return p;
220}
f0e17e86
JT
221
222static void scan_windows(struct packed_git *p,
223 struct packed_git **lru_p,
224 struct pack_window **lru_w,
225 struct pack_window **lru_l)
226{
227 struct pack_window *w, *w_l;
228
229 for (w_l = NULL, w = p->windows; w; w = w->next) {
230 if (!w->inuse_cnt) {
231 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
232 *lru_p = p;
233 *lru_w = w;
234 *lru_l = w_l;
235 }
236 }
237 w_l = w;
238 }
239}
240
84f80ad5 241static int unuse_one_window(struct packed_git *current)
f0e17e86
JT
242{
243 struct packed_git *p, *lru_p = NULL;
244 struct pack_window *lru_w = NULL, *lru_l = NULL;
245
246 if (current)
247 scan_windows(current, &lru_p, &lru_w, &lru_l);
a80d72db 248 for (p = the_repository->objects->packed_git; p; p = p->next)
f0e17e86
JT
249 scan_windows(p, &lru_p, &lru_w, &lru_l);
250 if (lru_p) {
251 munmap(lru_w->base, lru_w->len);
252 pack_mapped -= lru_w->len;
253 if (lru_l)
254 lru_l->next = lru_w->next;
255 else
256 lru_p->windows = lru_w->next;
257 free(lru_w);
258 pack_open_windows--;
259 return 1;
260 }
261 return 0;
262}
263
264void release_pack_memory(size_t need)
265{
266 size_t cur = pack_mapped;
267 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
268 ; /* nothing */
269}
3836d88a
JT
270
271void close_pack_windows(struct packed_git *p)
272{
273 while (p->windows) {
274 struct pack_window *w = p->windows;
275
276 if (w->inuse_cnt)
277 die("pack '%s' still has open windows to it",
278 p->pack_name);
279 munmap(w->base, w->len);
280 pack_mapped -= w->len;
281 pack_open_windows--;
282 p->windows = w->next;
283 free(w);
284 }
285}
286
84f80ad5 287static int close_pack_fd(struct packed_git *p)
3836d88a
JT
288{
289 if (p->pack_fd < 0)
290 return 0;
291
292 close(p->pack_fd);
293 pack_open_fds--;
294 p->pack_fd = -1;
295
296 return 1;
297}
298
299void close_pack_index(struct packed_git *p)
300{
301 if (p->index_data) {
302 munmap((void *)p->index_data, p->index_size);
303 p->index_data = NULL;
304 }
305}
306
307static void close_pack(struct packed_git *p)
308{
309 close_pack_windows(p);
310 close_pack_fd(p);
311 close_pack_index(p);
312}
313
d0b59866 314void close_all_packs(struct raw_object_store *o)
3836d88a
JT
315{
316 struct packed_git *p;
317
d0b59866 318 for (p = o->packed_git; p; p = p->next)
3836d88a
JT
319 if (p->do_not_close)
320 die("BUG: want to close pack marked 'do-not-close'");
321 else
322 close_pack(p);
323}
84f80ad5
JT
324
325/*
326 * The LRU pack is the one with the oldest MRU window, preferring packs
327 * with no used windows, or the oldest mtime if it has no windows allocated.
328 */
329static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
330{
331 struct pack_window *w, *this_mru_w;
332 int has_windows_inuse = 0;
333
334 /*
335 * Reject this pack if it has windows and the previously selected
336 * one does not. If this pack does not have windows, reject
337 * it if the pack file is newer than the previously selected one.
338 */
339 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
340 return;
341
342 for (w = this_mru_w = p->windows; w; w = w->next) {
343 /*
344 * Reject this pack if any of its windows are in use,
345 * but the previously selected pack did not have any
346 * inuse windows. Otherwise, record that this pack
347 * has windows in use.
348 */
349 if (w->inuse_cnt) {
350 if (*accept_windows_inuse)
351 has_windows_inuse = 1;
352 else
353 return;
354 }
355
356 if (w->last_used > this_mru_w->last_used)
357 this_mru_w = w;
358
359 /*
360 * Reject this pack if it has windows that have been
361 * used more recently than the previously selected pack.
362 * If the previously selected pack had windows inuse and
363 * we have not encountered a window in this pack that is
364 * inuse, skip this check since we prefer a pack with no
365 * inuse windows to one that has inuse windows.
366 */
367 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
368 this_mru_w->last_used > (*mru_w)->last_used)
369 return;
370 }
371
372 /*
373 * Select this pack.
374 */
375 *mru_w = this_mru_w;
376 *lru_p = p;
377 *accept_windows_inuse = has_windows_inuse;
378}
379
380static int close_one_pack(void)
381{
382 struct packed_git *p, *lru_p = NULL;
383 struct pack_window *mru_w = NULL;
384 int accept_windows_inuse = 1;
385
a80d72db 386 for (p = the_repository->objects->packed_git; p; p = p->next) {
84f80ad5
JT
387 if (p->pack_fd == -1)
388 continue;
389 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
390 }
391
392 if (lru_p)
393 return close_pack_fd(lru_p);
394
395 return 0;
396}
397
398static unsigned int get_max_fd_limit(void)
399{
400#ifdef RLIMIT_NOFILE
401 {
402 struct rlimit lim;
403
404 if (!getrlimit(RLIMIT_NOFILE, &lim))
405 return lim.rlim_cur;
406 }
407#endif
408
409#ifdef _SC_OPEN_MAX
410 {
411 long open_max = sysconf(_SC_OPEN_MAX);
412 if (0 < open_max)
413 return open_max;
414 /*
415 * Otherwise, we got -1 for one of the two
416 * reasons:
417 *
418 * (1) sysconf() did not understand _SC_OPEN_MAX
419 * and signaled an error with -1; or
420 * (2) sysconf() said there is no limit.
421 *
422 * We _could_ clear errno before calling sysconf() to
423 * tell these two cases apart and return a huge number
424 * in the latter case to let the caller cap it to a
425 * value that is not so selfish, but letting the
426 * fallback OPEN_MAX codepath take care of these cases
427 * is a lot simpler.
428 */
429 }
430#endif
431
432#ifdef OPEN_MAX
433 return OPEN_MAX;
434#else
435 return 1; /* see the caller ;-) */
436#endif
437}
438
439/*
440 * Do not call this directly as this leaks p->pack_fd on error return;
441 * call open_packed_git() instead.
442 */
443static int open_packed_git_1(struct packed_git *p)
444{
445 struct stat st;
446 struct pack_header hdr;
447 unsigned char sha1[20];
448 unsigned char *idx_sha1;
449 long fd_flag;
41dcc4dc 450 ssize_t read_result;
84f80ad5
JT
451
452 if (!p->index_data && open_pack_index(p))
453 return error("packfile %s index unavailable", p->pack_name);
454
455 if (!pack_max_fds) {
456 unsigned int max_fds = get_max_fd_limit();
457
458 /* Save 3 for stdin/stdout/stderr, 22 for work */
459 if (25 < max_fds)
460 pack_max_fds = max_fds - 25;
461 else
462 pack_max_fds = 1;
463 }
464
465 while (pack_max_fds <= pack_open_fds && close_one_pack())
466 ; /* nothing */
467
468 p->pack_fd = git_open(p->pack_name);
469 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
470 return -1;
471 pack_open_fds++;
472
473 /* If we created the struct before we had the pack we lack size. */
474 if (!p->pack_size) {
475 if (!S_ISREG(st.st_mode))
476 return error("packfile %s not a regular file", p->pack_name);
477 p->pack_size = st.st_size;
478 } else if (p->pack_size != st.st_size)
479 return error("packfile %s size changed", p->pack_name);
480
481 /* We leave these file descriptors open with sliding mmap;
482 * there is no point keeping them open across exec(), though.
483 */
484 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
485 if (fd_flag < 0)
486 return error("cannot determine file descriptor flags");
487 fd_flag |= FD_CLOEXEC;
488 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
489 return error("cannot set FD_CLOEXEC");
490
491 /* Verify we recognize this pack file format. */
41dcc4dc
JK
492 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
493 if (read_result < 0)
494 return error_errno("error reading from %s", p->pack_name);
495 if (read_result != sizeof(hdr))
84f80ad5
JT
496 return error("file %s is far too short to be a packfile", p->pack_name);
497 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
498 return error("file %s is not a GIT packfile", p->pack_name);
499 if (!pack_version_ok(hdr.hdr_version))
500 return error("packfile %s is version %"PRIu32" and not"
501 " supported (try upgrading GIT to a newer version)",
502 p->pack_name, ntohl(hdr.hdr_version));
503
504 /* Verify the pack matches its index. */
505 if (p->num_objects != ntohl(hdr.hdr_entries))
506 return error("packfile %s claims to have %"PRIu32" objects"
507 " while index indicates %"PRIu32" objects",
508 p->pack_name, ntohl(hdr.hdr_entries),
509 p->num_objects);
510 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
511 return error("end of packfile %s is unavailable", p->pack_name);
41dcc4dc
JK
512 read_result = read_in_full(p->pack_fd, sha1, sizeof(sha1));
513 if (read_result < 0)
514 return error_errno("error reading from %s", p->pack_name);
515 if (read_result != sizeof(sha1))
84f80ad5
JT
516 return error("packfile %s signature is unavailable", p->pack_name);
517 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
518 if (hashcmp(sha1, idx_sha1))
519 return error("packfile %s does not match index", p->pack_name);
520 return 0;
521}
522
a2551953 523static int open_packed_git(struct packed_git *p)
84f80ad5
JT
524{
525 if (!open_packed_git_1(p))
526 return 0;
527 close_pack_fd(p);
528 return -1;
529}
530
531static int in_window(struct pack_window *win, off_t offset)
532{
533 /* We must promise at least 20 bytes (one hash) after the
534 * offset is available from this window, otherwise the offset
535 * is not actually in this window and a different window (which
536 * has that one hash excess) must be used. This is to support
537 * the object header and delta base parsing routines below.
538 */
539 off_t win_off = win->offset;
540 return win_off <= offset
541 && (offset + 20) <= (win_off + win->len);
542}
543
544unsigned char *use_pack(struct packed_git *p,
545 struct pack_window **w_cursor,
546 off_t offset,
547 unsigned long *left)
548{
549 struct pack_window *win = *w_cursor;
550
551 /* Since packfiles end in a hash of their content and it's
552 * pointless to ask for an offset into the middle of that
553 * hash, and the in_window function above wouldn't match
554 * don't allow an offset too close to the end of the file.
555 */
556 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
557 die("packfile %s cannot be accessed", p->pack_name);
558 if (offset > (p->pack_size - 20))
559 die("offset beyond end of packfile (truncated pack?)");
560 if (offset < 0)
561 die(_("offset before end of packfile (broken .idx?)"));
562
563 if (!win || !in_window(win, offset)) {
564 if (win)
565 win->inuse_cnt--;
566 for (win = p->windows; win; win = win->next) {
567 if (in_window(win, offset))
568 break;
569 }
570 if (!win) {
571 size_t window_align = packed_git_window_size / 2;
572 off_t len;
573
574 if (p->pack_fd == -1 && open_packed_git(p))
575 die("packfile %s cannot be accessed", p->pack_name);
576
577 win = xcalloc(1, sizeof(*win));
578 win->offset = (offset / window_align) * window_align;
579 len = p->pack_size - win->offset;
580 if (len > packed_git_window_size)
581 len = packed_git_window_size;
582 win->len = (size_t)len;
583 pack_mapped += win->len;
584 while (packed_git_limit < pack_mapped
585 && unuse_one_window(p))
586 ; /* nothing */
587 win->base = xmmap(NULL, win->len,
588 PROT_READ, MAP_PRIVATE,
589 p->pack_fd, win->offset);
590 if (win->base == MAP_FAILED)
591 die_errno("packfile %s cannot be mapped",
592 p->pack_name);
593 if (!win->offset && win->len == p->pack_size
594 && !p->do_not_close)
595 close_pack_fd(p);
596 pack_mmap_calls++;
597 pack_open_windows++;
598 if (pack_mapped > peak_pack_mapped)
599 peak_pack_mapped = pack_mapped;
600 if (pack_open_windows > peak_pack_open_windows)
601 peak_pack_open_windows = pack_open_windows;
602 win->next = p->windows;
603 p->windows = win;
604 }
605 }
606 if (win != *w_cursor) {
607 win->last_used = pack_used_ctr++;
608 win->inuse_cnt++;
609 *w_cursor = win;
610 }
611 offset -= win->offset;
612 if (left)
613 *left = win->len - xsize_t(offset);
614 return win->base + offset;
615}
97de1803
JT
616
617void unuse_pack(struct pack_window **w_cursor)
618{
619 struct pack_window *w = *w_cursor;
620 if (w) {
621 w->inuse_cnt--;
622 *w_cursor = NULL;
623 }
624}
9a428653
JT
625
626static void try_to_free_pack_memory(size_t size)
627{
628 release_pack_memory(size);
629}
630
631struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
632{
633 static int have_set_try_to_free_routine;
634 struct stat st;
635 size_t alloc;
636 struct packed_git *p;
637
638 if (!have_set_try_to_free_routine) {
639 have_set_try_to_free_routine = 1;
640 set_try_to_free_routine(try_to_free_pack_memory);
641 }
642
643 /*
644 * Make sure a corresponding .pack file exists and that
645 * the index looks sane.
646 */
647 if (!strip_suffix_mem(path, &path_len, ".idx"))
648 return NULL;
649
650 /*
498f1f61 651 * ".promisor" is long enough to hold any suffix we're adding (and
9a428653
JT
652 * the use xsnprintf double-checks that)
653 */
498f1f61 654 alloc = st_add3(path_len, strlen(".promisor"), 1);
9a428653
JT
655 p = alloc_packed_git(alloc);
656 memcpy(p->pack_name, path, path_len);
657
658 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
659 if (!access(p->pack_name, F_OK))
660 p->pack_keep = 1;
661
498f1f61
JT
662 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
663 if (!access(p->pack_name, F_OK))
664 p->pack_promisor = 1;
665
9a428653
JT
666 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
667 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
668 free(p);
669 return NULL;
670 }
671
672 /* ok, it looks sane as far as we can check without
673 * actually mapping the pack file.
674 */
675 p->pack_size = st.st_size;
676 p->pack_local = local;
677 p->mtime = st.st_mtime;
678 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
679 hashclr(p->sha1);
680 return p;
681}
e65f1862 682
5babff16 683void install_packed_git(struct repository *r, struct packed_git *pack)
e65f1862
JT
684{
685 if (pack->pack_fd != -1)
686 pack_open_fds++;
687
5babff16
SB
688 pack->next = r->objects->packed_git;
689 r->objects->packed_git = pack;
e65f1862 690}
0abe14f6
JT
691
692void (*report_garbage)(unsigned seen_bits, const char *path);
693
694static void report_helper(const struct string_list *list,
695 int seen_bits, int first, int last)
696{
697 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
698 return;
699
700 for (; first < last; first++)
701 report_garbage(seen_bits, list->items[first].string);
702}
703
704static void report_pack_garbage(struct string_list *list)
705{
706 int i, baselen = -1, first = 0, seen_bits = 0;
707
708 if (!report_garbage)
709 return;
710
711 string_list_sort(list);
712
713 for (i = 0; i < list->nr; i++) {
714 const char *path = list->items[i].string;
715 if (baselen != -1 &&
716 strncmp(path, list->items[first].string, baselen)) {
717 report_helper(list, seen_bits, first, i);
718 baselen = -1;
719 seen_bits = 0;
720 }
721 if (baselen == -1) {
722 const char *dot = strrchr(path, '.');
723 if (!dot) {
724 report_garbage(PACKDIR_FILE_GARBAGE, path);
725 continue;
726 }
727 baselen = dot - path + 1;
728 first = i;
729 }
730 if (!strcmp(path + baselen, "pack"))
731 seen_bits |= 1;
732 else if (!strcmp(path + baselen, "idx"))
733 seen_bits |= 2;
734 }
735 report_helper(list, seen_bits, first, list->nr);
736}
737
935cdd69 738static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
0abe14f6
JT
739{
740 struct strbuf path = STRBUF_INIT;
741 size_t dirnamelen;
742 DIR *dir;
743 struct dirent *de;
744 struct string_list garbage = STRING_LIST_INIT_DUP;
745
746 strbuf_addstr(&path, objdir);
747 strbuf_addstr(&path, "/pack");
748 dir = opendir(path.buf);
749 if (!dir) {
750 if (errno != ENOENT)
751 error_errno("unable to open object pack directory: %s",
752 path.buf);
753 strbuf_release(&path);
754 return;
755 }
756 strbuf_addch(&path, '/');
757 dirnamelen = path.len;
758 while ((de = readdir(dir)) != NULL) {
759 struct packed_git *p;
760 size_t base_len;
761
762 if (is_dot_or_dotdot(de->d_name))
763 continue;
764
765 strbuf_setlen(&path, dirnamelen);
766 strbuf_addstr(&path, de->d_name);
767
768 base_len = path.len;
769 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
770 /* Don't reopen a pack we already have. */
935cdd69 771 for (p = r->objects->packed_git; p;
a80d72db 772 p = p->next) {
0abe14f6
JT
773 size_t len;
774 if (strip_suffix(p->pack_name, ".pack", &len) &&
775 len == base_len &&
776 !memcmp(p->pack_name, path.buf, len))
777 break;
778 }
779 if (p == NULL &&
780 /*
781 * See if it really is a valid .idx file with
782 * corresponding .pack file that we can map.
783 */
784 (p = add_packed_git(path.buf, path.len, local)) != NULL)
935cdd69 785 install_packed_git(r, p);
0abe14f6
JT
786 }
787
788 if (!report_garbage)
789 continue;
790
791 if (ends_with(de->d_name, ".idx") ||
792 ends_with(de->d_name, ".pack") ||
793 ends_with(de->d_name, ".bitmap") ||
498f1f61
JT
794 ends_with(de->d_name, ".keep") ||
795 ends_with(de->d_name, ".promisor"))
0abe14f6
JT
796 string_list_append(&garbage, path.buf);
797 else
798 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
799 }
800 closedir(dir);
801 report_pack_garbage(&garbage);
802 string_list_clear(&garbage, 0);
803 strbuf_release(&path);
804}
805
464416a2 806static void prepare_packed_git(struct repository *r);
0abe14f6
JT
807/*
808 * Give a fast, rough count of the number of objects in the repository. This
809 * ignores loose objects completely. If you have a lot of them, then either
810 * you should repack because your performance will be awful, or they are
811 * all unreachable objects about to be pruned, in which case they're not really
812 * interesting as a measure of repo size in the first place.
813 */
814unsigned long approximate_object_count(void)
815{
9a00580d
SB
816 if (!the_repository->objects->approximate_object_count_valid) {
817 unsigned long count;
0abe14f6
JT
818 struct packed_git *p;
819
6fdb4e9f 820 prepare_packed_git(the_repository);
0abe14f6 821 count = 0;
a80d72db 822 for (p = the_repository->objects->packed_git; p; p = p->next) {
0abe14f6
JT
823 if (open_pack_index(p))
824 continue;
825 count += p->num_objects;
826 }
9a00580d 827 the_repository->objects->approximate_object_count = count;
0abe14f6 828 }
9a00580d 829 return the_repository->objects->approximate_object_count;
0abe14f6
JT
830}
831
832static void *get_next_packed_git(const void *p)
833{
834 return ((const struct packed_git *)p)->next;
835}
836
837static void set_next_packed_git(void *p, void *next)
838{
839 ((struct packed_git *)p)->next = next;
840}
841
842static int sort_pack(const void *a_, const void *b_)
843{
844 const struct packed_git *a = a_;
845 const struct packed_git *b = b_;
846 int st;
847
848 /*
849 * Local packs tend to contain objects specific to our
850 * variant of the project than remote ones. In addition,
851 * remote ones could be on a network mounted filesystem.
852 * Favor local ones for these reasons.
853 */
854 st = a->pack_local - b->pack_local;
855 if (st)
856 return -st;
857
858 /*
859 * Younger packs tend to contain more recent objects,
860 * and more recent objects tend to get accessed more
861 * often.
862 */
863 if (a->mtime < b->mtime)
864 return 1;
865 else if (a->mtime == b->mtime)
866 return 0;
867 return -1;
868}
869
c235beac 870static void rearrange_packed_git(struct repository *r)
0abe14f6 871{
c235beac
SB
872 r->objects->packed_git = llist_mergesort(
873 r->objects->packed_git, get_next_packed_git,
a80d72db 874 set_next_packed_git, sort_pack);
0abe14f6
JT
875}
876
804be796 877static void prepare_packed_git_mru(struct repository *r)
0abe14f6
JT
878{
879 struct packed_git *p;
880
804be796 881 INIT_LIST_HEAD(&r->objects->packed_git_mru);
ec2dd32c 882
804be796
SB
883 for (p = r->objects->packed_git; p; p = p->next)
884 list_add_tail(&p->mru, &r->objects->packed_git_mru);
0abe14f6
JT
885}
886
464416a2 887static void prepare_packed_git(struct repository *r)
0abe14f6
JT
888{
889 struct alternate_object_database *alt;
890
0f90a9f2 891 if (r->objects->packed_git_initialized)
0abe14f6 892 return;
0f90a9f2
SB
893 prepare_packed_git_one(r, r->objects->objectdir, 1);
894 prepare_alt_odb(r);
895 for (alt = r->objects->alt_odb_list; alt; alt = alt->next)
896 prepare_packed_git_one(r, alt->path, 0);
897 rearrange_packed_git(r);
898 prepare_packed_git_mru(r);
899 r->objects->packed_git_initialized = 1;
0abe14f6
JT
900}
901
4c2a13b4 902void reprepare_packed_git(struct repository *r)
0abe14f6 903{
4c2a13b4
SB
904 r->objects->approximate_object_count_valid = 0;
905 r->objects->packed_git_initialized = 0;
906 prepare_packed_git(r);
0abe14f6 907}
32b42e15 908
a80d72db
SB
909struct packed_git *get_packed_git(struct repository *r)
910{
464416a2 911 prepare_packed_git(r);
a80d72db
SB
912 return r->objects->packed_git;
913}
914
915struct list_head *get_packed_git_mru(struct repository *r)
916{
464416a2 917 prepare_packed_git(r);
a80d72db
SB
918 return &r->objects->packed_git_mru;
919}
920
32b42e15
JT
921unsigned long unpack_object_header_buffer(const unsigned char *buf,
922 unsigned long len, enum object_type *type, unsigned long *sizep)
923{
924 unsigned shift;
925 unsigned long size, c;
926 unsigned long used = 0;
927
928 c = buf[used++];
929 *type = (c >> 4) & 7;
930 size = c & 15;
931 shift = 4;
932 while (c & 0x80) {
933 if (len <= used || bitsizeof(long) <= shift) {
934 error("bad object header");
935 size = used = 0;
936 break;
937 }
938 c = buf[used++];
939 size += (c & 0x7f) << shift;
940 shift += 7;
941 }
942 *sizep = size;
943 return used;
944}
7b3aa75d
JT
945
946unsigned long get_size_from_delta(struct packed_git *p,
947 struct pack_window **w_curs,
948 off_t curpos)
949{
950 const unsigned char *data;
951 unsigned char delta_head[20], *in;
952 git_zstream stream;
953 int st;
954
955 memset(&stream, 0, sizeof(stream));
956 stream.next_out = delta_head;
957 stream.avail_out = sizeof(delta_head);
958
959 git_inflate_init(&stream);
960 do {
961 in = use_pack(p, w_curs, curpos, &stream.avail_in);
962 stream.next_in = in;
963 st = git_inflate(&stream, Z_FINISH);
964 curpos += stream.next_in - in;
965 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
966 stream.total_out < sizeof(delta_head));
967 git_inflate_end(&stream);
968 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
969 error("delta data unpack-initial failed");
970 return 0;
971 }
972
973 /* Examine the initial part of the delta to figure out
974 * the result size.
975 */
976 data = delta_head;
977
978 /* ignore base size */
979 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
980
981 /* Read the result size */
982 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
983}
3588dd6e
JT
984
985int unpack_object_header(struct packed_git *p,
986 struct pack_window **w_curs,
987 off_t *curpos,
988 unsigned long *sizep)
989{
990 unsigned char *base;
991 unsigned long left;
992 unsigned long used;
993 enum object_type type;
994
995 /* use_pack() assures us we have [base, base + 20) available
996 * as a range that we can look at. (Its actually the hash
997 * size that is assured.) With our object header encoding
998 * the maximum deflated object size is 2^137, which is just
999 * insane, so we know won't exceed what we have been given.
1000 */
1001 base = use_pack(p, w_curs, *curpos, &left);
1002 used = unpack_object_header_buffer(base, left, &type, sizep);
1003 if (!used) {
1004 type = OBJ_BAD;
1005 } else
1006 *curpos += used;
1007
1008 return type;
1009}
f1d8130b
JT
1010
1011void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1012{
1013 unsigned i;
1014 for (i = 0; i < p->num_bad_objects; i++)
1015 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1016 return;
1017 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1018 st_mult(GIT_MAX_RAWSZ,
1019 st_add(p->num_bad_objects, 1)));
1020 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1021 p->num_bad_objects++;
1022}
1023
1024const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1025{
1026 struct packed_git *p;
1027 unsigned i;
1028
a80d72db 1029 for (p = the_repository->objects->packed_git; p; p = p->next)
f1d8130b
JT
1030 for (i = 0; i < p->num_bad_objects; i++)
1031 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1032 return p;
1033 return NULL;
1034}
1035
1036static off_t get_delta_base(struct packed_git *p,
1037 struct pack_window **w_curs,
1038 off_t *curpos,
1039 enum object_type type,
1040 off_t delta_obj_offset)
1041{
1042 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1043 off_t base_offset;
1044
1045 /* use_pack() assured us we have [base_info, base_info + 20)
1046 * as a range that we can look at without walking off the
1047 * end of the mapped window. Its actually the hash size
1048 * that is assured. An OFS_DELTA longer than the hash size
1049 * is stupid, as then a REF_DELTA would be smaller to store.
1050 */
1051 if (type == OBJ_OFS_DELTA) {
1052 unsigned used = 0;
1053 unsigned char c = base_info[used++];
1054 base_offset = c & 127;
1055 while (c & 128) {
1056 base_offset += 1;
1057 if (!base_offset || MSB(base_offset, 7))
1058 return 0; /* overflow */
1059 c = base_info[used++];
1060 base_offset = (base_offset << 7) + (c & 127);
1061 }
1062 base_offset = delta_obj_offset - base_offset;
1063 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1064 return 0; /* out of bound */
1065 *curpos += used;
1066 } else if (type == OBJ_REF_DELTA) {
1067 /* The base entry _must_ be in the same pack */
1068 base_offset = find_pack_entry_one(base_info, p);
1069 *curpos += 20;
1070 } else
1071 die("I am totally screwed");
1072 return base_offset;
1073}
1074
1075/*
1076 * Like get_delta_base above, but we return the sha1 instead of the pack
1077 * offset. This means it is cheaper for REF deltas (we do not have to do
1078 * the final object lookup), but more expensive for OFS deltas (we
1079 * have to load the revidx to convert the offset back into a sha1).
1080 */
1081static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1082 struct pack_window **w_curs,
1083 off_t curpos,
1084 enum object_type type,
1085 off_t delta_obj_offset)
1086{
1087 if (type == OBJ_REF_DELTA) {
1088 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1089 return base;
1090 } else if (type == OBJ_OFS_DELTA) {
1091 struct revindex_entry *revidx;
1092 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1093 type, delta_obj_offset);
1094
1095 if (!base_offset)
1096 return NULL;
1097
1098 revidx = find_pack_revindex(p, base_offset);
1099 if (!revidx)
1100 return NULL;
1101
1102 return nth_packed_object_sha1(p, revidx->nr);
1103 } else
1104 return NULL;
1105}
1106
0df23781
SB
1107#define retry_bad_packed_offset(r, p, o) \
1108 retry_bad_packed_offset_##r(p, o)
1109static int retry_bad_packed_offset_the_repository(struct packed_git *p, off_t obj_offset)
f1d8130b
JT
1110{
1111 int type;
1112 struct revindex_entry *revidx;
d169d664 1113 struct object_id oid;
f1d8130b
JT
1114 revidx = find_pack_revindex(p, obj_offset);
1115 if (!revidx)
1116 return OBJ_BAD;
d169d664 1117 nth_packed_object_oid(&oid, p, revidx->nr);
1118 mark_bad_packed_object(p, oid.hash);
0df8e965 1119 type = oid_object_info(the_repository, &oid, NULL);
f1d8130b
JT
1120 if (type <= OBJ_NONE)
1121 return OBJ_BAD;
1122 return type;
1123}
1124
1125#define POI_STACK_PREALLOC 64
1126
144f4948
SB
1127#define packed_to_object_type(r, p, o, t, w, c) \
1128 packed_to_object_type_##r(p, o, t, w, c)
1129static enum object_type packed_to_object_type_the_repository(struct packed_git *p,
1130 off_t obj_offset,
1131 enum object_type type,
1132 struct pack_window **w_curs,
1133 off_t curpos)
f1d8130b
JT
1134{
1135 off_t small_poi_stack[POI_STACK_PREALLOC];
1136 off_t *poi_stack = small_poi_stack;
1137 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1138
1139 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1140 off_t base_offset;
1141 unsigned long size;
1142 /* Push the object we're going to leave behind */
1143 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1144 poi_stack_alloc = alloc_nr(poi_stack_nr);
1145 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1146 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1147 } else {
1148 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1149 }
1150 poi_stack[poi_stack_nr++] = obj_offset;
1151 /* If parsing the base offset fails, just unwind */
1152 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1153 if (!base_offset)
1154 goto unwind;
1155 curpos = obj_offset = base_offset;
1156 type = unpack_object_header(p, w_curs, &curpos, &size);
1157 if (type <= OBJ_NONE) {
1158 /* If getting the base itself fails, we first
1159 * retry the base, otherwise unwind */
0df23781 1160 type = retry_bad_packed_offset(the_repository, p, base_offset);
f1d8130b
JT
1161 if (type > OBJ_NONE)
1162 goto out;
1163 goto unwind;
1164 }
1165 }
1166
1167 switch (type) {
1168 case OBJ_BAD:
1169 case OBJ_COMMIT:
1170 case OBJ_TREE:
1171 case OBJ_BLOB:
1172 case OBJ_TAG:
1173 break;
1174 default:
1175 error("unknown object type %i at offset %"PRIuMAX" in %s",
1176 type, (uintmax_t)obj_offset, p->pack_name);
1177 type = OBJ_BAD;
1178 }
1179
1180out:
1181 if (poi_stack != small_poi_stack)
1182 free(poi_stack);
1183 return type;
1184
1185unwind:
1186 while (poi_stack_nr) {
1187 obj_offset = poi_stack[--poi_stack_nr];
0df23781 1188 type = retry_bad_packed_offset(the_repository, p, obj_offset);
f1d8130b
JT
1189 if (type > OBJ_NONE)
1190 goto out;
1191 }
1192 type = OBJ_BAD;
1193 goto out;
1194}
1195
1196static struct hashmap delta_base_cache;
1197static size_t delta_base_cached;
1198
1199static LIST_HEAD(delta_base_cache_lru);
1200
1201struct delta_base_cache_key {
1202 struct packed_git *p;
1203 off_t base_offset;
1204};
1205
1206struct delta_base_cache_entry {
1207 struct hashmap hash;
1208 struct delta_base_cache_key key;
1209 struct list_head lru;
1210 void *data;
1211 unsigned long size;
1212 enum object_type type;
1213};
1214
1215static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1216{
1217 unsigned int hash;
1218
1219 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1220 hash += (hash >> 8) + (hash >> 16);
1221 return hash;
1222}
1223
1224static struct delta_base_cache_entry *
1225get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1226{
1227 struct hashmap_entry entry;
1228 struct delta_base_cache_key key;
1229
1230 if (!delta_base_cache.cmpfn)
1231 return NULL;
1232
1233 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1234 key.p = p;
1235 key.base_offset = base_offset;
1236 return hashmap_get(&delta_base_cache, &entry, &key);
1237}
1238
1239static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1240 const struct delta_base_cache_key *b)
1241{
1242 return a->p == b->p && a->base_offset == b->base_offset;
1243}
1244
1245static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1246 const void *va, const void *vb,
1247 const void *vkey)
1248{
1249 const struct delta_base_cache_entry *a = va, *b = vb;
1250 const struct delta_base_cache_key *key = vkey;
1251 if (key)
1252 return !delta_base_cache_key_eq(&a->key, key);
1253 else
1254 return !delta_base_cache_key_eq(&a->key, &b->key);
1255}
1256
1257static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1258{
1259 return !!get_delta_base_cache_entry(p, base_offset);
1260}
1261
1262/*
1263 * Remove the entry from the cache, but do _not_ free the associated
1264 * entry data. The caller takes ownership of the "data" buffer, and
1265 * should copy out any fields it wants before detaching.
1266 */
1267static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1268{
1269 hashmap_remove(&delta_base_cache, ent, &ent->key);
1270 list_del(&ent->lru);
1271 delta_base_cached -= ent->size;
1272 free(ent);
1273}
1274
1275static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1276 unsigned long *base_size, enum object_type *type)
1277{
1278 struct delta_base_cache_entry *ent;
1279
1280 ent = get_delta_base_cache_entry(p, base_offset);
1281 if (!ent)
57a6a500 1282 return unpack_entry(the_repository, p, base_offset, type, base_size);
f1d8130b
JT
1283
1284 if (type)
1285 *type = ent->type;
1286 if (base_size)
1287 *base_size = ent->size;
1288 return xmemdupz(ent->data, ent->size);
1289}
1290
1291static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1292{
1293 free(ent->data);
1294 detach_delta_base_cache_entry(ent);
1295}
1296
1297void clear_delta_base_cache(void)
1298{
1299 struct list_head *lru, *tmp;
1300 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1301 struct delta_base_cache_entry *entry =
1302 list_entry(lru, struct delta_base_cache_entry, lru);
1303 release_delta_base_cache(entry);
1304 }
1305}
1306
1307static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1308 void *base, unsigned long base_size, enum object_type type)
1309{
1310 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1311 struct list_head *lru, *tmp;
1312
1313 delta_base_cached += base_size;
1314
1315 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1316 struct delta_base_cache_entry *f =
1317 list_entry(lru, struct delta_base_cache_entry, lru);
1318 if (delta_base_cached <= delta_base_cache_limit)
1319 break;
1320 release_delta_base_cache(f);
1321 }
1322
1323 ent->key.p = p;
1324 ent->key.base_offset = base_offset;
1325 ent->type = type;
1326 ent->data = base;
1327 ent->size = base_size;
1328 list_add_tail(&ent->lru, &delta_base_cache_lru);
1329
1330 if (!delta_base_cache.cmpfn)
1331 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1332 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1333 hashmap_add(&delta_base_cache, ent);
1334}
1335
720aaa1a
JN
1336int packed_object_info_the_repository(struct packed_git *p, off_t obj_offset,
1337 struct object_info *oi)
f1d8130b
JT
1338{
1339 struct pack_window *w_curs = NULL;
1340 unsigned long size;
1341 off_t curpos = obj_offset;
1342 enum object_type type;
1343
1344 /*
1345 * We always get the representation type, but only convert it to
1346 * a "real" type later if the caller is interested.
1347 */
1348 if (oi->contentp) {
1349 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
1350 &type);
1351 if (!*oi->contentp)
1352 type = OBJ_BAD;
1353 } else {
1354 type = unpack_object_header(p, &w_curs, &curpos, &size);
1355 }
1356
1357 if (!oi->contentp && oi->sizep) {
1358 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1359 off_t tmp_pos = curpos;
1360 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1361 type, obj_offset);
1362 if (!base_offset) {
1363 type = OBJ_BAD;
1364 goto out;
1365 }
1366 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1367 if (*oi->sizep == 0) {
1368 type = OBJ_BAD;
1369 goto out;
1370 }
1371 } else {
1372 *oi->sizep = size;
1373 }
1374 }
1375
1376 if (oi->disk_sizep) {
1377 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1378 *oi->disk_sizep = revidx[1].offset - obj_offset;
1379 }
1380
6ca32f47 1381 if (oi->typep || oi->type_name) {
f1d8130b 1382 enum object_type ptot;
144f4948
SB
1383 ptot = packed_to_object_type(the_repository, p, obj_offset,
1384 type, &w_curs, curpos);
f1d8130b
JT
1385 if (oi->typep)
1386 *oi->typep = ptot;
6ca32f47 1387 if (oi->type_name) {
debca9d2 1388 const char *tn = type_name(ptot);
f1d8130b 1389 if (tn)
6ca32f47 1390 strbuf_addstr(oi->type_name, tn);
f1d8130b
JT
1391 }
1392 if (ptot < 0) {
1393 type = OBJ_BAD;
1394 goto out;
1395 }
1396 }
1397
1398 if (oi->delta_base_sha1) {
1399 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1400 const unsigned char *base;
1401
1402 base = get_delta_base_sha1(p, &w_curs, curpos,
1403 type, obj_offset);
1404 if (!base) {
1405 type = OBJ_BAD;
1406 goto out;
1407 }
1408
1409 hashcpy(oi->delta_base_sha1, base);
1410 } else
1411 hashclr(oi->delta_base_sha1);
1412 }
1413
1414 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1415 OI_PACKED;
1416
1417out:
1418 unuse_pack(&w_curs);
1419 return type;
1420}
1421
1422static void *unpack_compressed_entry(struct packed_git *p,
1423 struct pack_window **w_curs,
1424 off_t curpos,
1425 unsigned long size)
1426{
1427 int st;
1428 git_zstream stream;
1429 unsigned char *buffer, *in;
1430
1431 buffer = xmallocz_gently(size);
1432 if (!buffer)
1433 return NULL;
1434 memset(&stream, 0, sizeof(stream));
1435 stream.next_out = buffer;
1436 stream.avail_out = size + 1;
1437
1438 git_inflate_init(&stream);
1439 do {
1440 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1441 stream.next_in = in;
1442 st = git_inflate(&stream, Z_FINISH);
1443 if (!stream.avail_out)
1444 break; /* the payload is larger than it should be */
1445 curpos += stream.next_in - in;
1446 } while (st == Z_OK || st == Z_BUF_ERROR);
1447 git_inflate_end(&stream);
1448 if ((st != Z_STREAM_END) || stream.total_out != size) {
1449 free(buffer);
1450 return NULL;
1451 }
1452
1453 return buffer;
1454}
1455
1456static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1457{
1458 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1459 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1460 p->pack_name, (uintmax_t)obj_offset);
1461}
1462
1463int do_check_packed_object_crc;
1464
1465#define UNPACK_ENTRY_STACK_PREALLOC 64
1466struct unpack_entry_stack_ent {
1467 off_t obj_offset;
1468 off_t curpos;
1469 unsigned long size;
1470};
1471
5da6534d
SB
1472#define read_object(r, o, t, s) read_object_##r(o, t, s)
1473static void *read_object_the_repository(const struct object_id *oid,
1474 enum object_type *type,
1475 unsigned long *size)
f1d8130b
JT
1476{
1477 struct object_info oi = OBJECT_INFO_INIT;
1478 void *content;
1479 oi.typep = type;
1480 oi.sizep = size;
1481 oi.contentp = &content;
1482
7ecd8690 1483 if (oid_object_info_extended(the_repository, oid, &oi, 0) < 0)
f1d8130b
JT
1484 return NULL;
1485 return content;
1486}
1487
57a6a500
SB
1488void *unpack_entry_the_repository(struct packed_git *p, off_t obj_offset,
1489 enum object_type *final_type,
1490 unsigned long *final_size)
f1d8130b
JT
1491{
1492 struct pack_window *w_curs = NULL;
1493 off_t curpos = obj_offset;
1494 void *data = NULL;
1495 unsigned long size;
1496 enum object_type type;
1497 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1498 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1499 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1500 int base_from_cache = 0;
1501
1502 write_pack_access_log(p, obj_offset);
1503
1504 /* PHASE 1: drill down to the innermost base object */
1505 for (;;) {
1506 off_t base_offset;
1507 int i;
1508 struct delta_base_cache_entry *ent;
1509
1510 ent = get_delta_base_cache_entry(p, curpos);
1511 if (ent) {
1512 type = ent->type;
1513 data = ent->data;
1514 size = ent->size;
1515 detach_delta_base_cache_entry(ent);
1516 base_from_cache = 1;
1517 break;
1518 }
1519
1520 if (do_check_packed_object_crc && p->index_version > 1) {
1521 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1522 off_t len = revidx[1].offset - obj_offset;
1523 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
4310b0c4 1524 struct object_id oid;
1525 nth_packed_object_oid(&oid, p, revidx->nr);
f1d8130b 1526 error("bad packed object CRC for %s",
4310b0c4 1527 oid_to_hex(&oid));
1528 mark_bad_packed_object(p, oid.hash);
f1d8130b
JT
1529 data = NULL;
1530 goto out;
1531 }
1532 }
1533
1534 type = unpack_object_header(p, &w_curs, &curpos, &size);
1535 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1536 break;
1537
1538 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1539 if (!base_offset) {
1540 error("failed to validate delta base reference "
1541 "at offset %"PRIuMAX" from %s",
1542 (uintmax_t)curpos, p->pack_name);
1543 /* bail to phase 2, in hopes of recovery */
1544 data = NULL;
1545 break;
1546 }
1547
1548 /* push object, proceed to base */
1549 if (delta_stack_nr >= delta_stack_alloc
1550 && delta_stack == small_delta_stack) {
1551 delta_stack_alloc = alloc_nr(delta_stack_nr);
1552 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1553 memcpy(delta_stack, small_delta_stack,
1554 sizeof(*delta_stack)*delta_stack_nr);
1555 } else {
1556 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1557 }
1558 i = delta_stack_nr++;
1559 delta_stack[i].obj_offset = obj_offset;
1560 delta_stack[i].curpos = curpos;
1561 delta_stack[i].size = size;
1562
1563 curpos = obj_offset = base_offset;
1564 }
1565
1566 /* PHASE 2: handle the base */
1567 switch (type) {
1568 case OBJ_OFS_DELTA:
1569 case OBJ_REF_DELTA:
1570 if (data)
1571 die("BUG: unpack_entry: left loop at a valid delta");
1572 break;
1573 case OBJ_COMMIT:
1574 case OBJ_TREE:
1575 case OBJ_BLOB:
1576 case OBJ_TAG:
1577 if (!base_from_cache)
1578 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1579 break;
1580 default:
1581 data = NULL;
1582 error("unknown object type %i at offset %"PRIuMAX" in %s",
1583 type, (uintmax_t)obj_offset, p->pack_name);
1584 }
1585
1586 /* PHASE 3: apply deltas in order */
1587
1588 /* invariants:
1589 * 'data' holds the base data, or NULL if there was corruption
1590 */
1591 while (delta_stack_nr) {
1592 void *delta_data;
1593 void *base = data;
1594 void *external_base = NULL;
1595 unsigned long delta_size, base_size = size;
1596 int i;
1597
1598 data = NULL;
1599
1600 if (base)
1601 add_delta_base_cache(p, obj_offset, base, base_size, type);
1602
1603 if (!base) {
1604 /*
1605 * We're probably in deep shit, but let's try to fetch
1606 * the required base anyway from another pack or loose.
1607 * This is costly but should happen only in the presence
1608 * of a corrupted pack, and is better than failing outright.
1609 */
1610 struct revindex_entry *revidx;
4310b0c4 1611 struct object_id base_oid;
f1d8130b
JT
1612 revidx = find_pack_revindex(p, obj_offset);
1613 if (revidx) {
4310b0c4 1614 nth_packed_object_oid(&base_oid, p, revidx->nr);
f1d8130b
JT
1615 error("failed to read delta base object %s"
1616 " at offset %"PRIuMAX" from %s",
4310b0c4 1617 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
f1d8130b 1618 p->pack_name);
4310b0c4 1619 mark_bad_packed_object(p, base_oid.hash);
5da6534d 1620 base = read_object(the_repository, &base_oid, &type, &base_size);
f1d8130b
JT
1621 external_base = base;
1622 }
1623 }
1624
1625 i = --delta_stack_nr;
1626 obj_offset = delta_stack[i].obj_offset;
1627 curpos = delta_stack[i].curpos;
1628 delta_size = delta_stack[i].size;
1629
1630 if (!base)
1631 continue;
1632
1633 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1634
1635 if (!delta_data) {
1636 error("failed to unpack compressed delta "
1637 "at offset %"PRIuMAX" from %s",
1638 (uintmax_t)curpos, p->pack_name);
1639 data = NULL;
1640 free(external_base);
1641 continue;
1642 }
1643
1644 data = patch_delta(base, base_size,
1645 delta_data, delta_size,
1646 &size);
1647
1648 /*
1649 * We could not apply the delta; warn the user, but keep going.
1650 * Our failure will be noticed either in the next iteration of
1651 * the loop, or if this is the final delta, in the caller when
1652 * we return NULL. Those code paths will take care of making
1653 * a more explicit warning and retrying with another copy of
1654 * the object.
1655 */
1656 if (!data)
1657 error("failed to apply delta");
1658
1659 free(delta_data);
1660 free(external_base);
1661 }
1662
1663 if (final_type)
1664 *final_type = type;
1665 if (final_size)
1666 *final_size = size;
1667
1668out:
1669 unuse_pack(&w_curs);
1670
1671 if (delta_stack != small_delta_stack)
1672 free(delta_stack);
1673
1674 return data;
1675}
d5a16761 1676
3d475f46
DS
1677int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1678{
1679 const unsigned char *index_fanout = p->index_data;
1680 const unsigned char *index_lookup;
1681 int index_lookup_width;
1682
1683 if (!index_fanout)
1684 BUG("bsearch_pack called without a valid pack-index");
1685
1686 index_lookup = index_fanout + 4 * 256;
1687 if (p->index_version == 1) {
1688 index_lookup_width = 24;
1689 index_lookup += 4;
1690 } else {
1691 index_lookup_width = 20;
1692 index_fanout += 8;
1693 index_lookup += 8;
1694 }
1695
1696 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1697 index_lookup, index_lookup_width, result);
1698}
1699
d5a16761
JT
1700const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1701 uint32_t n)
1702{
1703 const unsigned char *index = p->index_data;
1704 if (!index) {
1705 if (open_pack_index(p))
1706 return NULL;
1707 index = p->index_data;
1708 }
1709 if (n >= p->num_objects)
1710 return NULL;
1711 index += 4 * 256;
1712 if (p->index_version == 1) {
1713 return index + 24 * n + 4;
1714 } else {
1715 index += 8;
1716 return index + 20 * n;
1717 }
1718}
1719
1720const struct object_id *nth_packed_object_oid(struct object_id *oid,
1721 struct packed_git *p,
1722 uint32_t n)
1723{
1724 const unsigned char *hash = nth_packed_object_sha1(p, n);
1725 if (!hash)
1726 return NULL;
1727 hashcpy(oid->hash, hash);
1728 return oid;
1729}
9e0f45f5
JT
1730
1731void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1732{
1733 const unsigned char *ptr = vptr;
1734 const unsigned char *start = p->index_data;
1735 const unsigned char *end = start + p->index_size;
1736 if (ptr < start)
1737 die(_("offset before start of pack index for %s (corrupt index?)"),
1738 p->pack_name);
1739 /* No need to check for underflow; .idx files must be at least 8 bytes */
1740 if (ptr >= end - 8)
1741 die(_("offset beyond end of pack index for %s (truncated index?)"),
1742 p->pack_name);
1743}
1744
1745off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1746{
1747 const unsigned char *index = p->index_data;
1748 index += 4 * 256;
1749 if (p->index_version == 1) {
1750 return ntohl(*((uint32_t *)(index + 24 * n)));
1751 } else {
1752 uint32_t off;
1753 index += 8 + p->num_objects * (20 + 4);
1754 off = ntohl(*((uint32_t *)(index + 4 * n)));
1755 if (!(off & 0x80000000))
1756 return off;
1757 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1758 check_pack_index_ptr(p, index);
ad622a25 1759 return get_be64(index);
9e0f45f5
JT
1760 }
1761}
a2551953
JT
1762
1763off_t find_pack_entry_one(const unsigned char *sha1,
1764 struct packed_git *p)
1765{
a2551953 1766 const unsigned char *index = p->index_data;
3d475f46 1767 struct object_id oid;
b4e00f73 1768 uint32_t result;
a2551953
JT
1769
1770 if (!index) {
1771 if (open_pack_index(p))
1772 return 0;
a2551953
JT
1773 }
1774
3d475f46
DS
1775 hashcpy(oid.hash, sha1);
1776 if (bsearch_pack(&oid, p, &result))
b4e00f73 1777 return nth_packed_object_offset(p, result);
a2551953
JT
1778 return 0;
1779}
1780
1781int is_pack_valid(struct packed_git *p)
1782{
1783 /* An already open pack is known to be valid. */
1784 if (p->pack_fd != -1)
1785 return 1;
1786
1787 /* If the pack has one window completely covering the
1788 * file size, the pack is known to be valid even if
1789 * the descriptor is not currently open.
1790 */
1791 if (p->windows) {
1792 struct pack_window *w = p->windows;
1793
1794 if (!w->offset && w->len == p->pack_size)
1795 return 1;
1796 }
1797
1798 /* Force the pack to open to prove its valid. */
1799 return !open_packed_git(p);
1800}
d6fe0036
JT
1801
1802struct packed_git *find_sha1_pack(const unsigned char *sha1,
1803 struct packed_git *packs)
1804{
1805 struct packed_git *p;
1806
1807 for (p = packs; p; p = p->next) {
1808 if (find_pack_entry_one(sha1, p))
1809 return p;
1810 }
1811 return NULL;
1812
1813}
1a1e5d4f
JT
1814
1815static int fill_pack_entry(const unsigned char *sha1,
1816 struct pack_entry *e,
1817 struct packed_git *p)
1818{
1819 off_t offset;
1820
1821 if (p->num_bad_objects) {
1822 unsigned i;
1823 for (i = 0; i < p->num_bad_objects; i++)
1824 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1825 return 0;
1826 }
1827
1828 offset = find_pack_entry_one(sha1, p);
1829 if (!offset)
1830 return 0;
1831
1832 /*
1833 * We are about to tell the caller where they can locate the
1834 * requested object. We better make sure the packfile is
1835 * still here and can be accessed before supplying that
1836 * answer, as it may have been deleted since the index was
1837 * loaded!
1838 */
1839 if (!is_pack_valid(p))
1840 return 0;
1841 e->offset = offset;
1842 e->p = p;
1843 hashcpy(e->sha1, sha1);
1844 return 1;
1845}
1846
0a0dd632 1847int find_pack_entry(struct repository *r, const unsigned char *sha1, struct pack_entry *e)
1a1e5d4f 1848{
8865859d 1849 struct list_head *pos;
1a1e5d4f 1850
0a0dd632
SB
1851 prepare_packed_git(r);
1852 if (!r->objects->packed_git)
1a1e5d4f
JT
1853 return 0;
1854
0a0dd632 1855 list_for_each(pos, &r->objects->packed_git_mru) {
ec2dd32c
GS
1856 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1857 if (fill_pack_entry(sha1, e, p)) {
0a0dd632 1858 list_move(&p->mru, &r->objects->packed_git_mru);
1a1e5d4f
JT
1859 return 1;
1860 }
1861 }
1862 return 0;
1863}
150e3001
JT
1864
1865int has_sha1_pack(const unsigned char *sha1)
1866{
1867 struct pack_entry e;
613b42f2 1868 return find_pack_entry(the_repository, sha1, &e);
150e3001 1869}
f9a8672a
JT
1870
1871int has_pack_index(const unsigned char *sha1)
1872{
1873 struct stat st;
1874 if (stat(sha1_pack_index_name(sha1), &st))
1875 return 0;
1876 return 1;
1877}
7709f468
JT
1878
1879static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
1880{
1881 uint32_t i;
1882 int r = 0;
1883
1884 for (i = 0; i < p->num_objects; i++) {
1885 struct object_id oid;
1886
1887 if (!nth_packed_object_oid(&oid, p, i))
1888 return error("unable to get sha1 of object %u in %s",
1889 i, p->pack_name);
1890
1891 r = cb(&oid, p, i, data);
1892 if (r)
1893 break;
1894 }
1895 return r;
1896}
1897
1898int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
1899{
1900 struct packed_git *p;
1901 int r = 0;
1902 int pack_errors = 0;
1903
6fdb4e9f 1904 prepare_packed_git(the_repository);
a80d72db 1905 for (p = the_repository->objects->packed_git; p; p = p->next) {
7709f468
JT
1906 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
1907 continue;
498f1f61
JT
1908 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
1909 !p->pack_promisor)
1910 continue;
7709f468
JT
1911 if (open_pack_index(p)) {
1912 pack_errors = 1;
1913 continue;
1914 }
1915 r = for_each_object_in_pack(p, cb, data);
1916 if (r)
1917 break;
1918 }
1919 return r ? r : pack_errors;
1920}
498f1f61
JT
1921
1922static int add_promisor_object(const struct object_id *oid,
1923 struct packed_git *pack,
1924 uint32_t pos,
1925 void *set_)
1926{
1927 struct oidset *set = set_;
1928 struct object *obj = parse_object(oid);
1929 if (!obj)
1930 return 1;
1931
1932 oidset_insert(set, oid);
1933
1934 /*
1935 * If this is a tree, commit, or tag, the objects it refers
a80d72db 1936 * to are also promisor objects. (Blobs refer to no objects->)
498f1f61
JT
1937 */
1938 if (obj->type == OBJ_TREE) {
1939 struct tree *tree = (struct tree *)obj;
1940 struct tree_desc desc;
1941 struct name_entry entry;
1942 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
1943 /*
1944 * Error messages are given when packs are
1945 * verified, so do not print any here.
1946 */
1947 return 0;
1948 while (tree_entry_gently(&desc, &entry))
1949 oidset_insert(set, entry.oid);
1950 } else if (obj->type == OBJ_COMMIT) {
1951 struct commit *commit = (struct commit *) obj;
1952 struct commit_list *parents = commit->parents;
1953
1954 oidset_insert(set, &commit->tree->object.oid);
1955 for (; parents; parents = parents->next)
1956 oidset_insert(set, &parents->item->object.oid);
1957 } else if (obj->type == OBJ_TAG) {
1958 struct tag *tag = (struct tag *) obj;
1959 oidset_insert(set, &tag->tagged->oid);
1960 }
1961 return 0;
1962}
1963
1964int is_promisor_object(const struct object_id *oid)
1965{
1966 static struct oidset promisor_objects;
1967 static int promisor_objects_prepared;
1968
1969 if (!promisor_objects_prepared) {
1970 if (repository_format_partial_clone) {
1971 for_each_packed_object(add_promisor_object,
1972 &promisor_objects,
1973 FOR_EACH_OBJECT_PROMISOR_ONLY);
1974 }
1975 promisor_objects_prepared = 1;
1976 }
1977 return oidset_contains(&promisor_objects, oid);
1978}