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 "list.h"
10#include "streaming.h"
11#include "sha1-lookup.h"
12#include "commit.h"
13#include "object.h"
14#include "tag.h"
15#include "tree-walk.h"
16#include "tree.h"
17#include "object-store.h"
18#include "midx.h"
19char *odb_pack_name(struct strbuf *buf,
21 const unsigned char *sha1,
22 const char *ext)
23{
24 strbuf_reset(buf);
25 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
26 sha1_to_hex(sha1), ext);
27 return buf->buf;
28}
29char *sha1_pack_name(const unsigned char *sha1)
31{
32 static struct strbuf buf = STRBUF_INIT;
33 return odb_pack_name(&buf, sha1, "pack");
34}
35char *sha1_pack_index_name(const unsigned char *sha1)
37{
38 static struct strbuf buf = STRBUF_INIT;
39 return odb_pack_name(&buf, sha1, "idx");
40}
41static unsigned int pack_used_ctr;
43static unsigned int pack_mmap_calls;
44static unsigned int peak_pack_open_windows;
45static unsigned int pack_open_windows;
46static unsigned int pack_open_fds;
47static unsigned int pack_max_fds;
48static size_t peak_pack_mapped;
49static size_t pack_mapped;
50#define SZ_FMT PRIuMAX
52static inline uintmax_t sz_fmt(size_t s) { return s; }
53void pack_report(void)
55{
56 fprintf(stderr,
57 "pack_report: getpagesize() = %10" SZ_FMT "\n"
58 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
60 sz_fmt(getpagesize()),
61 sz_fmt(packed_git_window_size),
62 sz_fmt(packed_git_limit));
63 fprintf(stderr,
64 "pack_report: pack_used_ctr = %10u\n"
65 "pack_report: pack_mmap_calls = %10u\n"
66 "pack_report: pack_open_windows = %10u / %10u\n"
67 "pack_report: pack_mapped = "
68 "%10" SZ_FMT " / %10" SZ_FMT "\n",
69 pack_used_ctr,
70 pack_mmap_calls,
71 pack_open_windows, peak_pack_open_windows,
72 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
73}
74/*
76 * Open and mmap the index file at path, perform a couple of
77 * consistency checks, then record its information to p. Return 0 on
78 * success.
79 */
80static int check_packed_git_idx(const char *path, struct packed_git *p)
81{
82 void *idx_map;
83 size_t idx_size;
84 int fd = git_open(path), ret;
85 struct stat st;
86 const unsigned int hashsz = the_hash_algo->rawsz;
87 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 + hashsz + hashsz) {
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 ret = load_idx(path, hashsz, idx_map, idx_size, p);
103 if (ret)
105 munmap(idx_map, idx_size);
106 return ret;
108}
109int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
111 size_t idx_size, struct packed_git *p)
112{
113 struct pack_idx_header *hdr = idx_map;
114 uint32_t version, nr, i, *index;
115 if (idx_size < 4 * 256 + hashsz + hashsz)
117 return error("index file %s is too small", path);
118 if (idx_map == NULL)
119 return error("empty data");
120 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
122 version = ntohl(hdr->idx_version);
123 if (version < 2 || version > 2)
124 return error("index file %s is version %"PRIu32
125 " and is not supported by this binary"
126 " (try upgrading GIT to a newer version)",
127 path, version);
128 } else
129 version = 1;
130 nr = 0;
132 index = idx_map;
133 if (version > 1)
134 index += 2; /* skip index header */
135 for (i = 0; i < 256; i++) {
136 uint32_t n = ntohl(index[i]);
137 if (n < nr)
138 return error("non-monotonic index %s", path);
139 nr = n;
140 }
141 if (version == 1) {
143 /*
144 * Total size:
145 * - 256 index entries 4 bytes each
146 * - 24-byte entries * nr (object ID + 4-byte offset)
147 * - hash of the packfile
148 * - file checksum
149 */
150 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
151 return error("wrong index v1 file size in %s", path);
152 } else if (version == 2) {
153 /*
154 * Minimum size:
155 * - 8 bytes of header
156 * - 256 index entries 4 bytes each
157 * - object ID entry * nr
158 * - 4-byte crc entry * nr
159 * - 4-byte offset entry * nr
160 * - hash of the packfile
161 * - file checksum
162 * And after the 4-byte offset table might be a
163 * variable sized table containing 8-byte entries
164 * for offsets larger than 2^31.
165 */
166 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
167 unsigned long max_size = min_size;
168 if (nr)
169 max_size += (nr - 1)*8;
170 if (idx_size < min_size || idx_size > max_size)
171 return error("wrong index v2 file size in %s", path);
172 if (idx_size != min_size &&
173 /*
174 * make sure we can deal with large pack offsets.
175 * 31-bit signed offset won't be enough, neither
176 * 32-bit unsigned one will be.
177 */
178 (sizeof(off_t) <= 4))
179 return error("pack too large for current definition of off_t in %s", path);
180 }
181 p->index_version = version;
183 p->index_data = idx_map;
184 p->index_size = idx_size;
185 p->num_objects = nr;
186 return 0;
187}
188int open_pack_index(struct packed_git *p)
190{
191 char *idx_name;
192 size_t len;
193 int ret;
194 if (p->index_data)
196 return 0;
197 if (!strip_suffix(p->pack_name, ".pack", &len))
199 BUG("pack_name does not end in .pack");
200 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
201 ret = check_packed_git_idx(idx_name, p);
202 free(idx_name);
203 return ret;
204}
205uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
207{
208 const uint32_t *level1_ofs = p->index_data;
209 if (!level1_ofs) {
211 if (open_pack_index(p))
212 return 0;
213 level1_ofs = p->index_data;
214 }
215 if (p->index_version > 1) {
217 level1_ofs += 2;
218 }
219 return ntohl(level1_ofs[value]);
221}
222static struct packed_git *alloc_packed_git(int extra)
224{
225 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
226 memset(p, 0, sizeof(*p));
227 p->pack_fd = -1;
228 return p;
229}
230struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
232{
233 const char *path = sha1_pack_name(sha1);
234 size_t alloc = st_add(strlen(path), 1);
235 struct packed_git *p = alloc_packed_git(alloc);
236 memcpy(p->pack_name, path, alloc); /* includes NUL */
238 hashcpy(p->hash, sha1);
239 if (check_packed_git_idx(idx_path, p)) {
240 free(p);
241 return NULL;
242 }
243 return p;
245}
246static void scan_windows(struct packed_git *p,
248 struct packed_git **lru_p,
249 struct pack_window **lru_w,
250 struct pack_window **lru_l)
251{
252 struct pack_window *w, *w_l;
253 for (w_l = NULL, w = p->windows; w; w = w->next) {
255 if (!w->inuse_cnt) {
256 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
257 *lru_p = p;
258 *lru_w = w;
259 *lru_l = w_l;
260 }
261 }
262 w_l = w;
263 }
264}
265static int unuse_one_window(struct packed_git *current)
267{
268 struct packed_git *p, *lru_p = NULL;
269 struct pack_window *lru_w = NULL, *lru_l = NULL;
270 if (current)
272 scan_windows(current, &lru_p, &lru_w, &lru_l);
273 for (p = the_repository->objects->packed_git; p; p = p->next)
274 scan_windows(p, &lru_p, &lru_w, &lru_l);
275 if (lru_p) {
276 munmap(lru_w->base, lru_w->len);
277 pack_mapped -= lru_w->len;
278 if (lru_l)
279 lru_l->next = lru_w->next;
280 else
281 lru_p->windows = lru_w->next;
282 free(lru_w);
283 pack_open_windows--;
284 return 1;
285 }
286 return 0;
287}
288void release_pack_memory(size_t need)
290{
291 size_t cur = pack_mapped;
292 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
293 ; /* nothing */
294}
295void close_pack_windows(struct packed_git *p)
297{
298 while (p->windows) {
299 struct pack_window *w = p->windows;
300 if (w->inuse_cnt)
302 die("pack '%s' still has open windows to it",
303 p->pack_name);
304 munmap(w->base, w->len);
305 pack_mapped -= w->len;
306 pack_open_windows--;
307 p->windows = w->next;
308 free(w);
309 }
310}
311int close_pack_fd(struct packed_git *p)
313{
314 if (p->pack_fd < 0)
315 return 0;
316 close(p->pack_fd);
318 pack_open_fds--;
319 p->pack_fd = -1;
320 return 1;
322}
323void close_pack_index(struct packed_git *p)
325{
326 if (p->index_data) {
327 munmap((void *)p->index_data, p->index_size);
328 p->index_data = NULL;
329 }
330}
331void close_pack(struct packed_git *p)
333{
334 close_pack_windows(p);
335 close_pack_fd(p);
336 close_pack_index(p);
337}
338void close_all_packs(struct raw_object_store *o)
340{
341 struct packed_git *p;
342 for (p = o->packed_git; p; p = p->next)
344 if (p->do_not_close)
345 BUG("want to close pack marked 'do-not-close'");
346 else
347 close_pack(p);
348 if (o->multi_pack_index) {
350 close_midx(o->multi_pack_index);
351 o->multi_pack_index = NULL;
352 }
353}
354/*
356 * The LRU pack is the one with the oldest MRU window, preferring packs
357 * with no used windows, or the oldest mtime if it has no windows allocated.
358 */
359static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
360{
361 struct pack_window *w, *this_mru_w;
362 int has_windows_inuse = 0;
363 /*
365 * Reject this pack if it has windows and the previously selected
366 * one does not. If this pack does not have windows, reject
367 * it if the pack file is newer than the previously selected one.
368 */
369 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
370 return;
371 for (w = this_mru_w = p->windows; w; w = w->next) {
373 /*
374 * Reject this pack if any of its windows are in use,
375 * but the previously selected pack did not have any
376 * inuse windows. Otherwise, record that this pack
377 * has windows in use.
378 */
379 if (w->inuse_cnt) {
380 if (*accept_windows_inuse)
381 has_windows_inuse = 1;
382 else
383 return;
384 }
385 if (w->last_used > this_mru_w->last_used)
387 this_mru_w = w;
388 /*
390 * Reject this pack if it has windows that have been
391 * used more recently than the previously selected pack.
392 * If the previously selected pack had windows inuse and
393 * we have not encountered a window in this pack that is
394 * inuse, skip this check since we prefer a pack with no
395 * inuse windows to one that has inuse windows.
396 */
397 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
398 this_mru_w->last_used > (*mru_w)->last_used)
399 return;
400 }
401 /*
403 * Select this pack.
404 */
405 *mru_w = this_mru_w;
406 *lru_p = p;
407 *accept_windows_inuse = has_windows_inuse;
408}
409static int close_one_pack(void)
411{
412 struct packed_git *p, *lru_p = NULL;
413 struct pack_window *mru_w = NULL;
414 int accept_windows_inuse = 1;
415 for (p = the_repository->objects->packed_git; p; p = p->next) {
417 if (p->pack_fd == -1)
418 continue;
419 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
420 }
421 if (lru_p)
423 return close_pack_fd(lru_p);
424 return 0;
426}
427static unsigned int get_max_fd_limit(void)
429{
430#ifdef RLIMIT_NOFILE
431 {
432 struct rlimit lim;
433 if (!getrlimit(RLIMIT_NOFILE, &lim))
435 return lim.rlim_cur;
436 }
437#endif
438#ifdef _SC_OPEN_MAX
440 {
441 long open_max = sysconf(_SC_OPEN_MAX);
442 if (0 < open_max)
443 return open_max;
444 /*
445 * Otherwise, we got -1 for one of the two
446 * reasons:
447 *
448 * (1) sysconf() did not understand _SC_OPEN_MAX
449 * and signaled an error with -1; or
450 * (2) sysconf() said there is no limit.
451 *
452 * We _could_ clear errno before calling sysconf() to
453 * tell these two cases apart and return a huge number
454 * in the latter case to let the caller cap it to a
455 * value that is not so selfish, but letting the
456 * fallback OPEN_MAX codepath take care of these cases
457 * is a lot simpler.
458 */
459 }
460#endif
461#ifdef OPEN_MAX
463 return OPEN_MAX;
464#else
465 return 1; /* see the caller ;-) */
466#endif
467}
468const char *pack_basename(struct packed_git *p)
470{
471 const char *ret = strrchr(p->pack_name, '/');
472 if (ret)
473 ret = ret + 1; /* skip past slash */
474 else
475 ret = p->pack_name; /* we only have a base */
476 return ret;
477}
478/*
480 * Do not call this directly as this leaks p->pack_fd on error return;
481 * call open_packed_git() instead.
482 */
483static int open_packed_git_1(struct packed_git *p)
484{
485 struct stat st;
486 struct pack_header hdr;
487 unsigned char hash[GIT_MAX_RAWSZ];
488 unsigned char *idx_hash;
489 long fd_flag;
490 ssize_t read_result;
491 const unsigned hashsz = the_hash_algo->rawsz;
492 if (!p->index_data) {
494 struct multi_pack_index *m;
495 const char *pack_name = pack_basename(p);
496 for (m = the_repository->objects->multi_pack_index;
498 m; m = m->next) {
499 if (midx_contains_pack(m, pack_name))
500 break;
501 }
502 if (!m && open_pack_index(p))
504 return error("packfile %s index unavailable", p->pack_name);
505 }
506 if (!pack_max_fds) {
508 unsigned int max_fds = get_max_fd_limit();
509 /* Save 3 for stdin/stdout/stderr, 22 for work */
511 if (25 < max_fds)
512 pack_max_fds = max_fds - 25;
513 else
514 pack_max_fds = 1;
515 }
516 while (pack_max_fds <= pack_open_fds && close_one_pack())
518 ; /* nothing */
519 p->pack_fd = git_open(p->pack_name);
521 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
522 return -1;
523 pack_open_fds++;
524 /* If we created the struct before we had the pack we lack size. */
526 if (!p->pack_size) {
527 if (!S_ISREG(st.st_mode))
528 return error("packfile %s not a regular file", p->pack_name);
529 p->pack_size = st.st_size;
530 } else if (p->pack_size != st.st_size)
531 return error("packfile %s size changed", p->pack_name);
532 /* We leave these file descriptors open with sliding mmap;
534 * there is no point keeping them open across exec(), though.
535 */
536 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
537 if (fd_flag < 0)
538 return error("cannot determine file descriptor flags");
539 fd_flag |= FD_CLOEXEC;
540 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
541 return error("cannot set FD_CLOEXEC");
542 /* Verify we recognize this pack file format. */
544 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
545 if (read_result < 0)
546 return error_errno("error reading from %s", p->pack_name);
547 if (read_result != sizeof(hdr))
548 return error("file %s is far too short to be a packfile", p->pack_name);
549 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
550 return error("file %s is not a GIT packfile", p->pack_name);
551 if (!pack_version_ok(hdr.hdr_version))
552 return error("packfile %s is version %"PRIu32" and not"
553 " supported (try upgrading GIT to a newer version)",
554 p->pack_name, ntohl(hdr.hdr_version));
555 /* Skip index checking if in multi-pack-index */
557 if (!p->index_data)
558 return 0;
559 /* Verify the pack matches its index. */
561 if (p->num_objects != ntohl(hdr.hdr_entries))
562 return error("packfile %s claims to have %"PRIu32" objects"
563 " while index indicates %"PRIu32" objects",
564 p->pack_name, ntohl(hdr.hdr_entries),
565 p->num_objects);
566 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
567 return error("end of packfile %s is unavailable", p->pack_name);
568 read_result = read_in_full(p->pack_fd, hash, hashsz);
569 if (read_result < 0)
570 return error_errno("error reading from %s", p->pack_name);
571 if (read_result != hashsz)
572 return error("packfile %s signature is unavailable", p->pack_name);
573 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
574 if (!hasheq(hash, idx_hash))
575 return error("packfile %s does not match index", p->pack_name);
576 return 0;
577}
578static int open_packed_git(struct packed_git *p)
580{
581 if (!open_packed_git_1(p))
582 return 0;
583 close_pack_fd(p);
584 return -1;
585}
586static int in_window(struct pack_window *win, off_t offset)
588{
589 /* We must promise at least one full hash after the
590 * offset is available from this window, otherwise the offset
591 * is not actually in this window and a different window (which
592 * has that one hash excess) must be used. This is to support
593 * the object header and delta base parsing routines below.
594 */
595 off_t win_off = win->offset;
596 return win_off <= offset
597 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
598}
599unsigned char *use_pack(struct packed_git *p,
601 struct pack_window **w_cursor,
602 off_t offset,
603 unsigned long *left)
604{
605 struct pack_window *win = *w_cursor;
606 /* Since packfiles end in a hash of their content and it's
608 * pointless to ask for an offset into the middle of that
609 * hash, and the in_window function above wouldn't match
610 * don't allow an offset too close to the end of the file.
611 */
612 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
613 die("packfile %s cannot be accessed", p->pack_name);
614 if (offset > (p->pack_size - the_hash_algo->rawsz))
615 die("offset beyond end of packfile (truncated pack?)");
616 if (offset < 0)
617 die(_("offset before end of packfile (broken .idx?)"));
618 if (!win || !in_window(win, offset)) {
620 if (win)
621 win->inuse_cnt--;
622 for (win = p->windows; win; win = win->next) {
623 if (in_window(win, offset))
624 break;
625 }
626 if (!win) {
627 size_t window_align = packed_git_window_size / 2;
628 off_t len;
629 if (p->pack_fd == -1 && open_packed_git(p))
631 die("packfile %s cannot be accessed", p->pack_name);
632 win = xcalloc(1, sizeof(*win));
634 win->offset = (offset / window_align) * window_align;
635 len = p->pack_size - win->offset;
636 if (len > packed_git_window_size)
637 len = packed_git_window_size;
638 win->len = (size_t)len;
639 pack_mapped += win->len;
640 while (packed_git_limit < pack_mapped
641 && unuse_one_window(p))
642 ; /* nothing */
643 win->base = xmmap(NULL, win->len,
644 PROT_READ, MAP_PRIVATE,
645 p->pack_fd, win->offset);
646 if (win->base == MAP_FAILED)
647 die_errno("packfile %s cannot be mapped",
648 p->pack_name);
649 if (!win->offset && win->len == p->pack_size
650 && !p->do_not_close)
651 close_pack_fd(p);
652 pack_mmap_calls++;
653 pack_open_windows++;
654 if (pack_mapped > peak_pack_mapped)
655 peak_pack_mapped = pack_mapped;
656 if (pack_open_windows > peak_pack_open_windows)
657 peak_pack_open_windows = pack_open_windows;
658 win->next = p->windows;
659 p->windows = win;
660 }
661 }
662 if (win != *w_cursor) {
663 win->last_used = pack_used_ctr++;
664 win->inuse_cnt++;
665 *w_cursor = win;
666 }
667 offset -= win->offset;
668 if (left)
669 *left = win->len - xsize_t(offset);
670 return win->base + offset;
671}
672void unuse_pack(struct pack_window **w_cursor)
674{
675 struct pack_window *w = *w_cursor;
676 if (w) {
677 w->inuse_cnt--;
678 *w_cursor = NULL;
679 }
680}
681static void try_to_free_pack_memory(size_t size)
683{
684 release_pack_memory(size);
685}
686struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
688{
689 static int have_set_try_to_free_routine;
690 struct stat st;
691 size_t alloc;
692 struct packed_git *p;
693 if (!have_set_try_to_free_routine) {
695 have_set_try_to_free_routine = 1;
696 set_try_to_free_routine(try_to_free_pack_memory);
697 }
698 /*
700 * Make sure a corresponding .pack file exists and that
701 * the index looks sane.
702 */
703 if (!strip_suffix_mem(path, &path_len, ".idx"))
704 return NULL;
705 /*
707 * ".promisor" is long enough to hold any suffix we're adding (and
708 * the use xsnprintf double-checks that)
709 */
710 alloc = st_add3(path_len, strlen(".promisor"), 1);
711 p = alloc_packed_git(alloc);
712 memcpy(p->pack_name, path, path_len);
713 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
715 if (!access(p->pack_name, F_OK))
716 p->pack_keep = 1;
717 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
719 if (!access(p->pack_name, F_OK))
720 p->pack_promisor = 1;
721 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
723 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
724 free(p);
725 return NULL;
726 }
727 /* ok, it looks sane as far as we can check without
729 * actually mapping the pack file.
730 */
731 p->pack_size = st.st_size;
732 p->pack_local = local;
733 p->mtime = st.st_mtime;
734 if (path_len < the_hash_algo->hexsz ||
735 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
736 hashclr(p->hash);
737 return p;
738}
739void install_packed_git(struct repository *r, struct packed_git *pack)
741{
742 if (pack->pack_fd != -1)
743 pack_open_fds++;
744 pack->next = r->objects->packed_git;
746 r->objects->packed_git = pack;
747}
748void (*report_garbage)(unsigned seen_bits, const char *path);
750static void report_helper(const struct string_list *list,
752 int seen_bits, int first, int last)
753{
754 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
755 return;
756 for (; first < last; first++)
758 report_garbage(seen_bits, list->items[first].string);
759}
760static void report_pack_garbage(struct string_list *list)
762{
763 int i, baselen = -1, first = 0, seen_bits = 0;
764 if (!report_garbage)
766 return;
767 string_list_sort(list);
769 for (i = 0; i < list->nr; i++) {
771 const char *path = list->items[i].string;
772 if (baselen != -1 &&
773 strncmp(path, list->items[first].string, baselen)) {
774 report_helper(list, seen_bits, first, i);
775 baselen = -1;
776 seen_bits = 0;
777 }
778 if (baselen == -1) {
779 const char *dot = strrchr(path, '.');
780 if (!dot) {
781 report_garbage(PACKDIR_FILE_GARBAGE, path);
782 continue;
783 }
784 baselen = dot - path + 1;
785 first = i;
786 }
787 if (!strcmp(path + baselen, "pack"))
788 seen_bits |= 1;
789 else if (!strcmp(path + baselen, "idx"))
790 seen_bits |= 2;
791 }
792 report_helper(list, seen_bits, first, list->nr);
793}
794void for_each_file_in_pack_dir(const char *objdir,
796 each_file_in_pack_dir_fn fn,
797 void *data)
798{
799 struct strbuf path = STRBUF_INIT;
800 size_t dirnamelen;
801 DIR *dir;
802 struct dirent *de;
803 strbuf_addstr(&path, objdir);
805 strbuf_addstr(&path, "/pack");
806 dir = opendir(path.buf);
807 if (!dir) {
808 if (errno != ENOENT)
809 error_errno("unable to open object pack directory: %s",
810 path.buf);
811 strbuf_release(&path);
812 return;
813 }
814 strbuf_addch(&path, '/');
815 dirnamelen = path.len;
816 while ((de = readdir(dir)) != NULL) {
817 if (is_dot_or_dotdot(de->d_name))
818 continue;
819 strbuf_setlen(&path, dirnamelen);
821 strbuf_addstr(&path, de->d_name);
822 fn(path.buf, path.len, de->d_name, data);
824 }
825 closedir(dir);
827 strbuf_release(&path);
828}
829struct prepare_pack_data {
831 struct repository *r;
832 struct string_list *garbage;
833 int local;
834 struct multi_pack_index *m;
835};
836static void prepare_pack(const char *full_name, size_t full_name_len,
838 const char *file_name, void *_data)
839{
840 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
841 struct packed_git *p;
842 size_t base_len = full_name_len;
843 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
845 !(data->m && midx_contains_pack(data->m, file_name))) {
846 /* Don't reopen a pack we already have. */
847 for (p = data->r->objects->packed_git; p; p = p->next) {
848 size_t len;
849 if (strip_suffix(p->pack_name, ".pack", &len) &&
850 len == base_len &&
851 !memcmp(p->pack_name, full_name, len))
852 break;
853 }
854 if (!p) {
856 p = add_packed_git(full_name, full_name_len, data->local);
857 if (p)
858 install_packed_git(data->r, p);
859 }
860 }
861 if (!report_garbage)
863 return;
864 if (!strcmp(file_name, "multi-pack-index"))
866 return;
867 if (ends_with(file_name, ".idx") ||
868 ends_with(file_name, ".pack") ||
869 ends_with(file_name, ".bitmap") ||
870 ends_with(file_name, ".keep") ||
871 ends_with(file_name, ".promisor"))
872 string_list_append(data->garbage, full_name);
873 else
874 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
875}
876static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
878{
879 struct prepare_pack_data data;
880 struct string_list garbage = STRING_LIST_INIT_DUP;
881 data.m = r->objects->multi_pack_index;
883 /* look for the multi-pack-index for this object directory */
885 while (data.m && strcmp(data.m->object_dir, objdir))
886 data.m = data.m->next;
887 data.r = r;
889 data.garbage = &garbage;
890 data.local = local;
891 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
893 report_pack_garbage(data.garbage);
895 string_list_clear(data.garbage, 0);
896}
897static void prepare_packed_git(struct repository *r);
899/*
900 * Give a fast, rough count of the number of objects in the repository. This
901 * ignores loose objects completely. If you have a lot of them, then either
902 * you should repack because your performance will be awful, or they are
903 * all unreachable objects about to be pruned, in which case they're not really
904 * interesting as a measure of repo size in the first place.
905 */
906unsigned long repo_approximate_object_count(struct repository *r)
907{
908 if (!r->objects->approximate_object_count_valid) {
909 unsigned long count;
910 struct multi_pack_index *m;
911 struct packed_git *p;
912 prepare_packed_git(r);
914 count = 0;
915 for (m = get_multi_pack_index(r); m; m = m->next)
916 count += m->num_objects;
917 for (p = r->objects->packed_git; p; p = p->next) {
918 if (open_pack_index(p))
919 continue;
920 count += p->num_objects;
921 }
922 r->objects->approximate_object_count = count;
923 }
924 return r->objects->approximate_object_count;
925}
926static void *get_next_packed_git(const void *p)
928{
929 return ((const struct packed_git *)p)->next;
930}
931static void set_next_packed_git(void *p, void *next)
933{
934 ((struct packed_git *)p)->next = next;
935}
936static int sort_pack(const void *a_, const void *b_)
938{
939 const struct packed_git *a = a_;
940 const struct packed_git *b = b_;
941 int st;
942 /*
944 * Local packs tend to contain objects specific to our
945 * variant of the project than remote ones. In addition,
946 * remote ones could be on a network mounted filesystem.
947 * Favor local ones for these reasons.
948 */
949 st = a->pack_local - b->pack_local;
950 if (st)
951 return -st;
952 /*
954 * Younger packs tend to contain more recent objects,
955 * and more recent objects tend to get accessed more
956 * often.
957 */
958 if (a->mtime < b->mtime)
959 return 1;
960 else if (a->mtime == b->mtime)
961 return 0;
962 return -1;
963}
964static void rearrange_packed_git(struct repository *r)
966{
967 r->objects->packed_git = llist_mergesort(
968 r->objects->packed_git, get_next_packed_git,
969 set_next_packed_git, sort_pack);
970}
971static void prepare_packed_git_mru(struct repository *r)
973{
974 struct packed_git *p;
975 INIT_LIST_HEAD(&r->objects->packed_git_mru);
977 for (p = r->objects->packed_git; p; p = p->next)
979 list_add_tail(&p->mru, &r->objects->packed_git_mru);
980}
981static void prepare_packed_git(struct repository *r)
983{
984 struct object_directory *odb;
985 if (r->objects->packed_git_initialized)
987 return;
988 prepare_alt_odb(r);
990 for (odb = r->objects->odb; odb; odb = odb->next) {
991 int local = (odb == r->objects->odb);
992 prepare_multi_pack_index_one(r, odb->path, local);
993 prepare_packed_git_one(r, odb->path, local);
994 }
995 rearrange_packed_git(r);
996 r->objects->all_packs = NULL;
998 prepare_packed_git_mru(r);
1000 r->objects->packed_git_initialized = 1;
1001}
1002void reprepare_packed_git(struct repository *r)
1004{
1005 struct object_directory *odb;
1006 for (odb = r->objects->odb; odb; odb = odb->next)
1008 odb_clear_loose_cache(odb);
1009 r->objects->approximate_object_count_valid = 0;
1011 r->objects->packed_git_initialized = 0;
1012 prepare_packed_git(r);
1013}
1014struct packed_git *get_packed_git(struct repository *r)
1016{
1017 prepare_packed_git(r);
1018 return r->objects->packed_git;
1019}
1020struct multi_pack_index *get_multi_pack_index(struct repository *r)
1022{
1023 prepare_packed_git(r);
1024 return r->objects->multi_pack_index;
1025}
1026struct packed_git *get_all_packs(struct repository *r)
1028{
1029 prepare_packed_git(r);
1030 if (!r->objects->all_packs) {
1032 struct packed_git *p = r->objects->packed_git;
1033 struct multi_pack_index *m;
1034 for (m = r->objects->multi_pack_index; m; m = m->next) {
1036 uint32_t i;
1037 for (i = 0; i < m->num_packs; i++) {
1038 if (!prepare_midx_pack(m, i)) {
1039 m->packs[i]->next = p;
1040 p = m->packs[i];
1041 }
1042 }
1043 }
1044 r->objects->all_packs = p;
1046 }
1047 return r->objects->all_packs;
1049}
1050struct list_head *get_packed_git_mru(struct repository *r)
1052{
1053 prepare_packed_git(r);
1054 return &r->objects->packed_git_mru;
1055}
1056unsigned long unpack_object_header_buffer(const unsigned char *buf,
1058 unsigned long len, enum object_type *type, unsigned long *sizep)
1059{
1060 unsigned shift;
1061 unsigned long size, c;
1062 unsigned long used = 0;
1063 c = buf[used++];
1065 *type = (c >> 4) & 7;
1066 size = c & 15;
1067 shift = 4;
1068 while (c & 0x80) {
1069 if (len <= used || bitsizeof(long) <= shift) {
1070 error("bad object header");
1071 size = used = 0;
1072 break;
1073 }
1074 c = buf[used++];
1075 size += (c & 0x7f) << shift;
1076 shift += 7;
1077 }
1078 *sizep = size;
1079 return used;
1080}
1081unsigned long get_size_from_delta(struct packed_git *p,
1083 struct pack_window **w_curs,
1084 off_t curpos)
1085{
1086 const unsigned char *data;
1087 unsigned char delta_head[20], *in;
1088 git_zstream stream;
1089 int st;
1090 memset(&stream, 0, sizeof(stream));
1092 stream.next_out = delta_head;
1093 stream.avail_out = sizeof(delta_head);
1094 git_inflate_init(&stream);
1096 do {
1097 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1098 stream.next_in = in;
1099 st = git_inflate(&stream, Z_FINISH);
1100 curpos += stream.next_in - in;
1101 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1102 stream.total_out < sizeof(delta_head));
1103 git_inflate_end(&stream);
1104 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1105 error("delta data unpack-initial failed");
1106 return 0;
1107 }
1108 /* Examine the initial part of the delta to figure out
1110 * the result size.
1111 */
1112 data = delta_head;
1113 /* ignore base size */
1115 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1116 /* Read the result size */
1118 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1119}
1120int unpack_object_header(struct packed_git *p,
1122 struct pack_window **w_curs,
1123 off_t *curpos,
1124 unsigned long *sizep)
1125{
1126 unsigned char *base;
1127 unsigned long left;
1128 unsigned long used;
1129 enum object_type type;
1130 /* use_pack() assures us we have [base, base + 20) available
1132 * as a range that we can look at. (Its actually the hash
1133 * size that is assured.) With our object header encoding
1134 * the maximum deflated object size is 2^137, which is just
1135 * insane, so we know won't exceed what we have been given.
1136 */
1137 base = use_pack(p, w_curs, *curpos, &left);
1138 used = unpack_object_header_buffer(base, left, &type, sizep);
1139 if (!used) {
1140 type = OBJ_BAD;
1141 } else
1142 *curpos += used;
1143 return type;
1145}
1146void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1148{
1149 unsigned i;
1150 const unsigned hashsz = the_hash_algo->rawsz;
1151 for (i = 0; i < p->num_bad_objects; i++)
1152 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1153 return;
1154 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1155 st_mult(GIT_MAX_RAWSZ,
1156 st_add(p->num_bad_objects, 1)));
1157 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1158 p->num_bad_objects++;
1159}
1160const struct packed_git *has_packed_and_bad(struct repository *r,
1162 const unsigned char *sha1)
1163{
1164 struct packed_git *p;
1165 unsigned i;
1166 for (p = r->objects->packed_git; p; p = p->next)
1168 for (i = 0; i < p->num_bad_objects; i++)
1169 if (hasheq(sha1,
1170 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1171 return p;
1172 return NULL;
1173}
1174static off_t get_delta_base(struct packed_git *p,
1176 struct pack_window **w_curs,
1177 off_t *curpos,
1178 enum object_type type,
1179 off_t delta_obj_offset)
1180{
1181 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1182 off_t base_offset;
1183 /* use_pack() assured us we have [base_info, base_info + 20)
1185 * as a range that we can look at without walking off the
1186 * end of the mapped window. Its actually the hash size
1187 * that is assured. An OFS_DELTA longer than the hash size
1188 * is stupid, as then a REF_DELTA would be smaller to store.
1189 */
1190 if (type == OBJ_OFS_DELTA) {
1191 unsigned used = 0;
1192 unsigned char c = base_info[used++];
1193 base_offset = c & 127;
1194 while (c & 128) {
1195 base_offset += 1;
1196 if (!base_offset || MSB(base_offset, 7))
1197 return 0; /* overflow */
1198 c = base_info[used++];
1199 base_offset = (base_offset << 7) + (c & 127);
1200 }
1201 base_offset = delta_obj_offset - base_offset;
1202 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1203 return 0; /* out of bound */
1204 *curpos += used;
1205 } else if (type == OBJ_REF_DELTA) {
1206 /* The base entry _must_ be in the same pack */
1207 base_offset = find_pack_entry_one(base_info, p);
1208 *curpos += the_hash_algo->rawsz;
1209 } else
1210 die("I am totally screwed");
1211 return base_offset;
1212}
1213/*
1215 * Like get_delta_base above, but we return the sha1 instead of the pack
1216 * offset. This means it is cheaper for REF deltas (we do not have to do
1217 * the final object lookup), but more expensive for OFS deltas (we
1218 * have to load the revidx to convert the offset back into a sha1).
1219 */
1220static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1221 struct pack_window **w_curs,
1222 off_t curpos,
1223 enum object_type type,
1224 off_t delta_obj_offset)
1225{
1226 if (type == OBJ_REF_DELTA) {
1227 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1228 return base;
1229 } else if (type == OBJ_OFS_DELTA) {
1230 struct revindex_entry *revidx;
1231 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1232 type, delta_obj_offset);
1233 if (!base_offset)
1235 return NULL;
1236 revidx = find_pack_revindex(p, base_offset);
1238 if (!revidx)
1239 return NULL;
1240 return nth_packed_object_sha1(p, revidx->nr);
1242 } else
1243 return NULL;
1244}
1245static int retry_bad_packed_offset(struct repository *r,
1247 struct packed_git *p,
1248 off_t obj_offset)
1249{
1250 int type;
1251 struct revindex_entry *revidx;
1252 struct object_id oid;
1253 revidx = find_pack_revindex(p, obj_offset);
1254 if (!revidx)
1255 return OBJ_BAD;
1256 nth_packed_object_oid(&oid, p, revidx->nr);
1257 mark_bad_packed_object(p, oid.hash);
1258 type = oid_object_info(r, &oid, NULL);
1259 if (type <= OBJ_NONE)
1260 return OBJ_BAD;
1261 return type;
1262}
1263#define POI_STACK_PREALLOC 64
1265static enum object_type packed_to_object_type(struct repository *r,
1267 struct packed_git *p,
1268 off_t obj_offset,
1269 enum object_type type,
1270 struct pack_window **w_curs,
1271 off_t curpos)
1272{
1273 off_t small_poi_stack[POI_STACK_PREALLOC];
1274 off_t *poi_stack = small_poi_stack;
1275 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1276 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1278 off_t base_offset;
1279 unsigned long size;
1280 /* Push the object we're going to leave behind */
1281 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1282 poi_stack_alloc = alloc_nr(poi_stack_nr);
1283 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1284 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1285 } else {
1286 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1287 }
1288 poi_stack[poi_stack_nr++] = obj_offset;
1289 /* If parsing the base offset fails, just unwind */
1290 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1291 if (!base_offset)
1292 goto unwind;
1293 curpos = obj_offset = base_offset;
1294 type = unpack_object_header(p, w_curs, &curpos, &size);
1295 if (type <= OBJ_NONE) {
1296 /* If getting the base itself fails, we first
1297 * retry the base, otherwise unwind */
1298 type = retry_bad_packed_offset(r, p, base_offset);
1299 if (type > OBJ_NONE)
1300 goto out;
1301 goto unwind;
1302 }
1303 }
1304 switch (type) {
1306 case OBJ_BAD:
1307 case OBJ_COMMIT:
1308 case OBJ_TREE:
1309 case OBJ_BLOB:
1310 case OBJ_TAG:
1311 break;
1312 default:
1313 error("unknown object type %i at offset %"PRIuMAX" in %s",
1314 type, (uintmax_t)obj_offset, p->pack_name);
1315 type = OBJ_BAD;
1316 }
1317out:
1319 if (poi_stack != small_poi_stack)
1320 free(poi_stack);
1321 return type;
1322unwind:
1324 while (poi_stack_nr) {
1325 obj_offset = poi_stack[--poi_stack_nr];
1326 type = retry_bad_packed_offset(r, p, obj_offset);
1327 if (type > OBJ_NONE)
1328 goto out;
1329 }
1330 type = OBJ_BAD;
1331 goto out;
1332}
1333static struct hashmap delta_base_cache;
1335static size_t delta_base_cached;
1336static LIST_HEAD(delta_base_cache_lru);
1338struct delta_base_cache_key {
1340 struct packed_git *p;
1341 off_t base_offset;
1342};
1343struct delta_base_cache_entry {
1345 struct hashmap hash;
1346 struct delta_base_cache_key key;
1347 struct list_head lru;
1348 void *data;
1349 unsigned long size;
1350 enum object_type type;
1351};
1352static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1354{
1355 unsigned int hash;
1356 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1358 hash += (hash >> 8) + (hash >> 16);
1359 return hash;
1360}
1361static struct delta_base_cache_entry *
1363get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1364{
1365 struct hashmap_entry entry;
1366 struct delta_base_cache_key key;
1367 if (!delta_base_cache.cmpfn)
1369 return NULL;
1370 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1372 key.p = p;
1373 key.base_offset = base_offset;
1374 return hashmap_get(&delta_base_cache, &entry, &key);
1375}
1376static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1378 const struct delta_base_cache_key *b)
1379{
1380 return a->p == b->p && a->base_offset == b->base_offset;
1381}
1382static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1384 const void *va, const void *vb,
1385 const void *vkey)
1386{
1387 const struct delta_base_cache_entry *a = va, *b = vb;
1388 const struct delta_base_cache_key *key = vkey;
1389 if (key)
1390 return !delta_base_cache_key_eq(&a->key, key);
1391 else
1392 return !delta_base_cache_key_eq(&a->key, &b->key);
1393}
1394static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1396{
1397 return !!get_delta_base_cache_entry(p, base_offset);
1398}
1399/*
1401 * Remove the entry from the cache, but do _not_ free the associated
1402 * entry data. The caller takes ownership of the "data" buffer, and
1403 * should copy out any fields it wants before detaching.
1404 */
1405static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1406{
1407 hashmap_remove(&delta_base_cache, ent, &ent->key);
1408 list_del(&ent->lru);
1409 delta_base_cached -= ent->size;
1410 free(ent);
1411}
1412static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1414 off_t base_offset, unsigned long *base_size,
1415 enum object_type *type)
1416{
1417 struct delta_base_cache_entry *ent;
1418 ent = get_delta_base_cache_entry(p, base_offset);
1420 if (!ent)
1421 return unpack_entry(r, p, base_offset, type, base_size);
1422 if (type)
1424 *type = ent->type;
1425 if (base_size)
1426 *base_size = ent->size;
1427 return xmemdupz(ent->data, ent->size);
1428}
1429static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1431{
1432 free(ent->data);
1433 detach_delta_base_cache_entry(ent);
1434}
1435void clear_delta_base_cache(void)
1437{
1438 struct list_head *lru, *tmp;
1439 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1440 struct delta_base_cache_entry *entry =
1441 list_entry(lru, struct delta_base_cache_entry, lru);
1442 release_delta_base_cache(entry);
1443 }
1444}
1445static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1447 void *base, unsigned long base_size, enum object_type type)
1448{
1449 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1450 struct list_head *lru, *tmp;
1451 delta_base_cached += base_size;
1453 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1455 struct delta_base_cache_entry *f =
1456 list_entry(lru, struct delta_base_cache_entry, lru);
1457 if (delta_base_cached <= delta_base_cache_limit)
1458 break;
1459 release_delta_base_cache(f);
1460 }
1461 ent->key.p = p;
1463 ent->key.base_offset = base_offset;
1464 ent->type = type;
1465 ent->data = base;
1466 ent->size = base_size;
1467 list_add_tail(&ent->lru, &delta_base_cache_lru);
1468 if (!delta_base_cache.cmpfn)
1470 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1471 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1472 hashmap_add(&delta_base_cache, ent);
1473}
1474int packed_object_info(struct repository *r, struct packed_git *p,
1476 off_t obj_offset, struct object_info *oi)
1477{
1478 struct pack_window *w_curs = NULL;
1479 unsigned long size;
1480 off_t curpos = obj_offset;
1481 enum object_type type;
1482 /*
1484 * We always get the representation type, but only convert it to
1485 * a "real" type later if the caller is interested.
1486 */
1487 if (oi->contentp) {
1488 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1489 &type);
1490 if (!*oi->contentp)
1491 type = OBJ_BAD;
1492 } else {
1493 type = unpack_object_header(p, &w_curs, &curpos, &size);
1494 }
1495 if (!oi->contentp && oi->sizep) {
1497 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1498 off_t tmp_pos = curpos;
1499 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1500 type, obj_offset);
1501 if (!base_offset) {
1502 type = OBJ_BAD;
1503 goto out;
1504 }
1505 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1506 if (*oi->sizep == 0) {
1507 type = OBJ_BAD;
1508 goto out;
1509 }
1510 } else {
1511 *oi->sizep = size;
1512 }
1513 }
1514 if (oi->disk_sizep) {
1516 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1517 *oi->disk_sizep = revidx[1].offset - obj_offset;
1518 }
1519 if (oi->typep || oi->type_name) {
1521 enum object_type ptot;
1522 ptot = packed_to_object_type(r, p, obj_offset,
1523 type, &w_curs, curpos);
1524 if (oi->typep)
1525 *oi->typep = ptot;
1526 if (oi->type_name) {
1527 const char *tn = type_name(ptot);
1528 if (tn)
1529 strbuf_addstr(oi->type_name, tn);
1530 }
1531 if (ptot < 0) {
1532 type = OBJ_BAD;
1533 goto out;
1534 }
1535 }
1536 if (oi->delta_base_sha1) {
1538 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1539 const unsigned char *base;
1540 base = get_delta_base_sha1(p, &w_curs, curpos,
1542 type, obj_offset);
1543 if (!base) {
1544 type = OBJ_BAD;
1545 goto out;
1546 }
1547 hashcpy(oi->delta_base_sha1, base);
1549 } else
1550 hashclr(oi->delta_base_sha1);
1551 }
1552 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1554 OI_PACKED;
1555out:
1557 unuse_pack(&w_curs);
1558 return type;
1559}
1560static void *unpack_compressed_entry(struct packed_git *p,
1562 struct pack_window **w_curs,
1563 off_t curpos,
1564 unsigned long size)
1565{
1566 int st;
1567 git_zstream stream;
1568 unsigned char *buffer, *in;
1569 buffer = xmallocz_gently(size);
1571 if (!buffer)
1572 return NULL;
1573 memset(&stream, 0, sizeof(stream));
1574 stream.next_out = buffer;
1575 stream.avail_out = size + 1;
1576 git_inflate_init(&stream);
1578 do {
1579 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1580 stream.next_in = in;
1581 st = git_inflate(&stream, Z_FINISH);
1582 if (!stream.avail_out)
1583 break; /* the payload is larger than it should be */
1584 curpos += stream.next_in - in;
1585 } while (st == Z_OK || st == Z_BUF_ERROR);
1586 git_inflate_end(&stream);
1587 if ((st != Z_STREAM_END) || stream.total_out != size) {
1588 free(buffer);
1589 return NULL;
1590 }
1591 /* versions of zlib can clobber unconsumed portion of outbuf */
1593 buffer[size] = '\0';
1594 return buffer;
1596}
1597static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1599{
1600 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1601 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1602 p->pack_name, (uintmax_t)obj_offset);
1603}
1604int do_check_packed_object_crc;
1606#define UNPACK_ENTRY_STACK_PREALLOC 64
1608struct unpack_entry_stack_ent {
1609 off_t obj_offset;
1610 off_t curpos;
1611 unsigned long size;
1612};
1613static void *read_object(struct repository *r,
1615 const struct object_id *oid,
1616 enum object_type *type,
1617 unsigned long *size)
1618{
1619 struct object_info oi = OBJECT_INFO_INIT;
1620 void *content;
1621 oi.typep = type;
1622 oi.sizep = size;
1623 oi.contentp = &content;
1624 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1626 return NULL;
1627 return content;
1628}
1629void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1631 enum object_type *final_type, unsigned long *final_size)
1632{
1633 struct pack_window *w_curs = NULL;
1634 off_t curpos = obj_offset;
1635 void *data = NULL;
1636 unsigned long size;
1637 enum object_type type;
1638 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1639 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1640 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1641 int base_from_cache = 0;
1642 write_pack_access_log(p, obj_offset);
1644 /* PHASE 1: drill down to the innermost base object */
1646 for (;;) {
1647 off_t base_offset;
1648 int i;
1649 struct delta_base_cache_entry *ent;
1650 ent = get_delta_base_cache_entry(p, curpos);
1652 if (ent) {
1653 type = ent->type;
1654 data = ent->data;
1655 size = ent->size;
1656 detach_delta_base_cache_entry(ent);
1657 base_from_cache = 1;
1658 break;
1659 }
1660 if (do_check_packed_object_crc && p->index_version > 1) {
1662 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1663 off_t len = revidx[1].offset - obj_offset;
1664 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1665 struct object_id oid;
1666 nth_packed_object_oid(&oid, p, revidx->nr);
1667 error("bad packed object CRC for %s",
1668 oid_to_hex(&oid));
1669 mark_bad_packed_object(p, oid.hash);
1670 data = NULL;
1671 goto out;
1672 }
1673 }
1674 type = unpack_object_header(p, &w_curs, &curpos, &size);
1676 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1677 break;
1678 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1680 if (!base_offset) {
1681 error("failed to validate delta base reference "
1682 "at offset %"PRIuMAX" from %s",
1683 (uintmax_t)curpos, p->pack_name);
1684 /* bail to phase 2, in hopes of recovery */
1685 data = NULL;
1686 break;
1687 }
1688 /* push object, proceed to base */
1690 if (delta_stack_nr >= delta_stack_alloc
1691 && delta_stack == small_delta_stack) {
1692 delta_stack_alloc = alloc_nr(delta_stack_nr);
1693 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1694 memcpy(delta_stack, small_delta_stack,
1695 sizeof(*delta_stack)*delta_stack_nr);
1696 } else {
1697 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1698 }
1699 i = delta_stack_nr++;
1700 delta_stack[i].obj_offset = obj_offset;
1701 delta_stack[i].curpos = curpos;
1702 delta_stack[i].size = size;
1703 curpos = obj_offset = base_offset;
1705 }
1706 /* PHASE 2: handle the base */
1708 switch (type) {
1709 case OBJ_OFS_DELTA:
1710 case OBJ_REF_DELTA:
1711 if (data)
1712 BUG("unpack_entry: left loop at a valid delta");
1713 break;
1714 case OBJ_COMMIT:
1715 case OBJ_TREE:
1716 case OBJ_BLOB:
1717 case OBJ_TAG:
1718 if (!base_from_cache)
1719 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1720 break;
1721 default:
1722 data = NULL;
1723 error("unknown object type %i at offset %"PRIuMAX" in %s",
1724 type, (uintmax_t)obj_offset, p->pack_name);
1725 }
1726 /* PHASE 3: apply deltas in order */
1728 /* invariants:
1730 * 'data' holds the base data, or NULL if there was corruption
1731 */
1732 while (delta_stack_nr) {
1733 void *delta_data;
1734 void *base = data;
1735 void *external_base = NULL;
1736 unsigned long delta_size, base_size = size;
1737 int i;
1738 data = NULL;
1740 if (base)
1742 add_delta_base_cache(p, obj_offset, base, base_size, type);
1743 if (!base) {
1745 /*
1746 * We're probably in deep shit, but let's try to fetch
1747 * the required base anyway from another pack or loose.
1748 * This is costly but should happen only in the presence
1749 * of a corrupted pack, and is better than failing outright.
1750 */
1751 struct revindex_entry *revidx;
1752 struct object_id base_oid;
1753 revidx = find_pack_revindex(p, obj_offset);
1754 if (revidx) {
1755 nth_packed_object_oid(&base_oid, p, revidx->nr);
1756 error("failed to read delta base object %s"
1757 " at offset %"PRIuMAX" from %s",
1758 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1759 p->pack_name);
1760 mark_bad_packed_object(p, base_oid.hash);
1761 base = read_object(r, &base_oid, &type, &base_size);
1762 external_base = base;
1763 }
1764 }
1765 i = --delta_stack_nr;
1767 obj_offset = delta_stack[i].obj_offset;
1768 curpos = delta_stack[i].curpos;
1769 delta_size = delta_stack[i].size;
1770 if (!base)
1772 continue;
1773 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1775 if (!delta_data) {
1777 error("failed to unpack compressed delta "
1778 "at offset %"PRIuMAX" from %s",
1779 (uintmax_t)curpos, p->pack_name);
1780 data = NULL;
1781 free(external_base);
1782 continue;
1783 }
1784 data = patch_delta(base, base_size,
1786 delta_data, delta_size,
1787 &size);
1788 /*
1790 * We could not apply the delta; warn the user, but keep going.
1791 * Our failure will be noticed either in the next iteration of
1792 * the loop, or if this is the final delta, in the caller when
1793 * we return NULL. Those code paths will take care of making
1794 * a more explicit warning and retrying with another copy of
1795 * the object.
1796 */
1797 if (!data)
1798 error("failed to apply delta");
1799 free(delta_data);
1801 free(external_base);
1802 }
1803 if (final_type)
1805 *final_type = type;
1806 if (final_size)
1807 *final_size = size;
1808out:
1810 unuse_pack(&w_curs);
1811 if (delta_stack != small_delta_stack)
1813 free(delta_stack);
1814 return data;
1816}
1817int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1819{
1820 const unsigned char *index_fanout = p->index_data;
1821 const unsigned char *index_lookup;
1822 const unsigned int hashsz = the_hash_algo->rawsz;
1823 int index_lookup_width;
1824 if (!index_fanout)
1826 BUG("bsearch_pack called without a valid pack-index");
1827 index_lookup = index_fanout + 4 * 256;
1829 if (p->index_version == 1) {
1830 index_lookup_width = hashsz + 4;
1831 index_lookup += 4;
1832 } else {
1833 index_lookup_width = hashsz;
1834 index_fanout += 8;
1835 index_lookup += 8;
1836 }
1837 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1839 index_lookup, index_lookup_width, result);
1840}
1841const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1843 uint32_t n)
1844{
1845 const unsigned char *index = p->index_data;
1846 const unsigned int hashsz = the_hash_algo->rawsz;
1847 if (!index) {
1848 if (open_pack_index(p))
1849 return NULL;
1850 index = p->index_data;
1851 }
1852 if (n >= p->num_objects)
1853 return NULL;
1854 index += 4 * 256;
1855 if (p->index_version == 1) {
1856 return index + (hashsz + 4) * n + 4;
1857 } else {
1858 index += 8;
1859 return index + hashsz * n;
1860 }
1861}
1862const struct object_id *nth_packed_object_oid(struct object_id *oid,
1864 struct packed_git *p,
1865 uint32_t n)
1866{
1867 const unsigned char *hash = nth_packed_object_sha1(p, n);
1868 if (!hash)
1869 return NULL;
1870 hashcpy(oid->hash, hash);
1871 return oid;
1872}
1873void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1875{
1876 const unsigned char *ptr = vptr;
1877 const unsigned char *start = p->index_data;
1878 const unsigned char *end = start + p->index_size;
1879 if (ptr < start)
1880 die(_("offset before start of pack index for %s (corrupt index?)"),
1881 p->pack_name);
1882 /* No need to check for underflow; .idx files must be at least 8 bytes */
1883 if (ptr >= end - 8)
1884 die(_("offset beyond end of pack index for %s (truncated index?)"),
1885 p->pack_name);
1886}
1887off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1889{
1890 const unsigned char *index = p->index_data;
1891 const unsigned int hashsz = the_hash_algo->rawsz;
1892 index += 4 * 256;
1893 if (p->index_version == 1) {
1894 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1895 } else {
1896 uint32_t off;
1897 index += 8 + p->num_objects * (hashsz + 4);
1898 off = ntohl(*((uint32_t *)(index + 4 * n)));
1899 if (!(off & 0x80000000))
1900 return off;
1901 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1902 check_pack_index_ptr(p, index);
1903 return get_be64(index);
1904 }
1905}
1906off_t find_pack_entry_one(const unsigned char *sha1,
1908 struct packed_git *p)
1909{
1910 const unsigned char *index = p->index_data;
1911 struct object_id oid;
1912 uint32_t result;
1913 if (!index) {
1915 if (open_pack_index(p))
1916 return 0;
1917 }
1918 hashcpy(oid.hash, sha1);
1920 if (bsearch_pack(&oid, p, &result))
1921 return nth_packed_object_offset(p, result);
1922 return 0;
1923}
1924int is_pack_valid(struct packed_git *p)
1926{
1927 /* An already open pack is known to be valid. */
1928 if (p->pack_fd != -1)
1929 return 1;
1930 /* If the pack has one window completely covering the
1932 * file size, the pack is known to be valid even if
1933 * the descriptor is not currently open.
1934 */
1935 if (p->windows) {
1936 struct pack_window *w = p->windows;
1937 if (!w->offset && w->len == p->pack_size)
1939 return 1;
1940 }
1941 /* Force the pack to open to prove its valid. */
1943 return !open_packed_git(p);
1944}
1945struct packed_git *find_sha1_pack(const unsigned char *sha1,
1947 struct packed_git *packs)
1948{
1949 struct packed_git *p;
1950 for (p = packs; p; p = p->next) {
1952 if (find_pack_entry_one(sha1, p))
1953 return p;
1954 }
1955 return NULL;
1956}
1958static int fill_pack_entry(const struct object_id *oid,
1960 struct pack_entry *e,
1961 struct packed_git *p)
1962{
1963 off_t offset;
1964 if (p->num_bad_objects) {
1966 unsigned i;
1967 for (i = 0; i < p->num_bad_objects; i++)
1968 if (hasheq(oid->hash,
1969 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1970 return 0;
1971 }
1972 offset = find_pack_entry_one(oid->hash, p);
1974 if (!offset)
1975 return 0;
1976 /*
1978 * We are about to tell the caller where they can locate the
1979 * requested object. We better make sure the packfile is
1980 * still here and can be accessed before supplying that
1981 * answer, as it may have been deleted since the index was
1982 * loaded!
1983 */
1984 if (!is_pack_valid(p))
1985 return 0;
1986 e->offset = offset;
1987 e->p = p;
1988 return 1;
1989}
1990int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1992{
1993 struct list_head *pos;
1994 struct multi_pack_index *m;
1995 prepare_packed_git(r);
1997 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1998 return 0;
1999 for (m = r->objects->multi_pack_index; m; m = m->next) {
2001 if (fill_midx_entry(oid, e, m))
2002 return 1;
2003 }
2004 list_for_each(pos, &r->objects->packed_git_mru) {
2006 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2007 if (fill_pack_entry(oid, e, p)) {
2008 list_move(&p->mru, &r->objects->packed_git_mru);
2009 return 1;
2010 }
2011 }
2012 return 0;
2013}
2014int has_object_pack(const struct object_id *oid)
2016{
2017 struct pack_entry e;
2018 return find_pack_entry(the_repository, oid, &e);
2019}
2020int has_pack_index(const unsigned char *sha1)
2022{
2023 struct stat st;
2024 if (stat(sha1_pack_index_name(sha1), &st))
2025 return 0;
2026 return 1;
2027}
2028int for_each_object_in_pack(struct packed_git *p,
2030 each_packed_object_fn cb, void *data,
2031 enum for_each_object_flags flags)
2032{
2033 uint32_t i;
2034 int r = 0;
2035 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2037 if (load_pack_revindex(p))
2038 return -1;
2039 }
2040 for (i = 0; i < p->num_objects; i++) {
2042 uint32_t pos;
2043 struct object_id oid;
2044 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2046 pos = p->revindex[i].nr;
2047 else
2048 pos = i;
2049 if (!nth_packed_object_oid(&oid, p, pos))
2051 return error("unable to get sha1 of object %u in %s",
2052 pos, p->pack_name);
2053 r = cb(&oid, p, pos, data);
2055 if (r)
2056 break;
2057 }
2058 return r;
2059}
2060int for_each_packed_object(each_packed_object_fn cb, void *data,
2062 enum for_each_object_flags flags)
2063{
2064 struct packed_git *p;
2065 int r = 0;
2066 int pack_errors = 0;
2067 prepare_packed_git(the_repository);
2069 for (p = get_all_packs(the_repository); p; p = p->next) {
2070 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2071 continue;
2072 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2073 !p->pack_promisor)
2074 continue;
2075 if (open_pack_index(p)) {
2076 pack_errors = 1;
2077 continue;
2078 }
2079 r = for_each_object_in_pack(p, cb, data, flags);
2080 if (r)
2081 break;
2082 }
2083 return r ? r : pack_errors;
2084}
2085static int add_promisor_object(const struct object_id *oid,
2087 struct packed_git *pack,
2088 uint32_t pos,
2089 void *set_)
2090{
2091 struct oidset *set = set_;
2092 struct object *obj = parse_object(the_repository, oid);
2093 if (!obj)
2094 return 1;
2095 oidset_insert(set, oid);
2097 /*
2099 * If this is a tree, commit, or tag, the objects it refers
2100 * to are also promisor objects. (Blobs refer to no objects->)
2101 */
2102 if (obj->type == OBJ_TREE) {
2103 struct tree *tree = (struct tree *)obj;
2104 struct tree_desc desc;
2105 struct name_entry entry;
2106 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2107 /*
2108 * Error messages are given when packs are
2109 * verified, so do not print any here.
2110 */
2111 return 0;
2112 while (tree_entry_gently(&desc, &entry))
2113 oidset_insert(set, &entry.oid);
2114 } else if (obj->type == OBJ_COMMIT) {
2115 struct commit *commit = (struct commit *) obj;
2116 struct commit_list *parents = commit->parents;
2117 oidset_insert(set, get_commit_tree_oid(commit));
2119 for (; parents; parents = parents->next)
2120 oidset_insert(set, &parents->item->object.oid);
2121 } else if (obj->type == OBJ_TAG) {
2122 struct tag *tag = (struct tag *) obj;
2123 oidset_insert(set, &tag->tagged->oid);
2124 }
2125 return 0;
2126}
2127int is_promisor_object(const struct object_id *oid)
2129{
2130 static struct oidset promisor_objects;
2131 static int promisor_objects_prepared;
2132 if (!promisor_objects_prepared) {
2134 if (repository_format_partial_clone) {
2135 for_each_packed_object(add_promisor_object,
2136 &promisor_objects,
2137 FOR_EACH_OBJECT_PROMISOR_ONLY);
2138 }
2139 promisor_objects_prepared = 1;
2140 }
2141 return oidset_contains(&promisor_objects, oid);
2142}