mailinfo: be more liberal with header whitespace
[git.git] / packfile.c
blob7e7c04e4d802e20438a8a94181c6a92fa5c5184b
1 #include "cache.h"
2 #include "list.h"
3 #include "pack.h"
4 #include "repository.h"
5 #include "dir.h"
6 #include "mergesort.h"
7 #include "packfile.h"
8 #include "delta.h"
9 #include "streaming.h"
10 #include "sha1-lookup.h"
11 #include "commit.h"
12 #include "object.h"
13 #include "tag.h"
14 #include "tree-walk.h"
15 #include "tree.h"
16 #include "object-store.h"
17 #include "midx.h"
18 #include "commit-graph.h"
19 #include "promisor-remote.h"
21 char *odb_pack_name(struct strbuf *buf,
22 const unsigned char *hash,
23 const char *ext)
25 strbuf_reset(buf);
26 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
27 hash_to_hex(hash), ext);
28 return buf->buf;
31 char *sha1_pack_name(const unsigned char *sha1)
33 static struct strbuf buf = STRBUF_INIT;
34 return odb_pack_name(&buf, sha1, "pack");
37 char *sha1_pack_index_name(const unsigned char *sha1)
39 static struct strbuf buf = STRBUF_INIT;
40 return odb_pack_name(&buf, sha1, "idx");
43 static unsigned int pack_used_ctr;
44 static unsigned int pack_mmap_calls;
45 static unsigned int peak_pack_open_windows;
46 static unsigned int pack_open_windows;
47 static unsigned int pack_open_fds;
48 static unsigned int pack_max_fds;
49 static size_t peak_pack_mapped;
50 static size_t pack_mapped;
52 #define SZ_FMT PRIuMAX
53 static inline uintmax_t sz_fmt(size_t s) { return s; }
55 void pack_report(void)
57 fprintf(stderr,
58 "pack_report: getpagesize() = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
60 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
61 sz_fmt(getpagesize()),
62 sz_fmt(packed_git_window_size),
63 sz_fmt(packed_git_limit));
64 fprintf(stderr,
65 "pack_report: pack_used_ctr = %10u\n"
66 "pack_report: pack_mmap_calls = %10u\n"
67 "pack_report: pack_open_windows = %10u / %10u\n"
68 "pack_report: pack_mapped = "
69 "%10" SZ_FMT " / %10" SZ_FMT "\n",
70 pack_used_ctr,
71 pack_mmap_calls,
72 pack_open_windows, peak_pack_open_windows,
73 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
77 * Open and mmap the index file at path, perform a couple of
78 * consistency checks, then record its information to p. Return 0 on
79 * success.
81 static int check_packed_git_idx(const char *path, struct packed_git *p)
83 void *idx_map;
84 size_t idx_size;
85 int fd = git_open(path), ret;
86 struct stat st;
87 const unsigned int hashsz = the_hash_algo->rawsz;
89 if (fd < 0)
90 return -1;
91 if (fstat(fd, &st)) {
92 close(fd);
93 return -1;
95 idx_size = xsize_t(st.st_size);
96 if (idx_size < 4 * 256 + hashsz + hashsz) {
97 close(fd);
98 return error("index file %s is too small", path);
100 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
101 close(fd);
103 ret = load_idx(path, hashsz, idx_map, idx_size, p);
105 if (ret)
106 munmap(idx_map, idx_size);
108 return ret;
111 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
112 size_t idx_size, struct packed_git *p)
114 struct pack_idx_header *hdr = idx_map;
115 uint32_t version, nr, i, *index;
117 if (idx_size < 4 * 256 + hashsz + hashsz)
118 return error("index file %s is too small", path);
119 if (idx_map == NULL)
120 return error("empty data");
122 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
123 version = ntohl(hdr->idx_version);
124 if (version < 2 || version > 2)
125 return error("index file %s is version %"PRIu32
126 " and is not supported by this binary"
127 " (try upgrading GIT to a newer version)",
128 path, version);
129 } else
130 version = 1;
132 nr = 0;
133 index = idx_map;
134 if (version > 1)
135 index += 2; /* skip index header */
136 for (i = 0; i < 256; i++) {
137 uint32_t n = ntohl(index[i]);
138 if (n < nr)
139 return error("non-monotonic index %s", path);
140 nr = n;
143 if (version == 1) {
145 * Total size:
146 * - 256 index entries 4 bytes each
147 * - 24-byte entries * nr (object ID + 4-byte offset)
148 * - hash of the packfile
149 * - file checksum
151 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
152 return error("wrong index v1 file size in %s", path);
153 } else if (version == 2) {
155 * Minimum size:
156 * - 8 bytes of header
157 * - 256 index entries 4 bytes each
158 * - object ID entry * nr
159 * - 4-byte crc entry * nr
160 * - 4-byte offset entry * nr
161 * - hash of the packfile
162 * - file checksum
163 * And after the 4-byte offset table might be a
164 * variable sized table containing 8-byte entries
165 * for offsets larger than 2^31.
167 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
168 unsigned long max_size = min_size;
169 if (nr)
170 max_size += (nr - 1)*8;
171 if (idx_size < min_size || idx_size > max_size)
172 return error("wrong index v2 file size in %s", path);
173 if (idx_size != min_size &&
175 * make sure we can deal with large pack offsets.
176 * 31-bit signed offset won't be enough, neither
177 * 32-bit unsigned one will be.
179 (sizeof(off_t) <= 4))
180 return error("pack too large for current definition of off_t in %s", path);
183 p->index_version = version;
184 p->index_data = idx_map;
185 p->index_size = idx_size;
186 p->num_objects = nr;
187 return 0;
190 int open_pack_index(struct packed_git *p)
192 char *idx_name;
193 size_t len;
194 int ret;
196 if (p->index_data)
197 return 0;
199 if (!strip_suffix(p->pack_name, ".pack", &len))
200 BUG("pack_name does not end in .pack");
201 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
202 ret = check_packed_git_idx(idx_name, p);
203 free(idx_name);
204 return ret;
207 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
209 const uint32_t *level1_ofs = p->index_data;
211 if (!level1_ofs) {
212 if (open_pack_index(p))
213 return 0;
214 level1_ofs = p->index_data;
217 if (p->index_version > 1) {
218 level1_ofs += 2;
221 return ntohl(level1_ofs[value]);
224 static struct packed_git *alloc_packed_git(int extra)
226 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
227 memset(p, 0, sizeof(*p));
228 p->pack_fd = -1;
229 return p;
232 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
234 const char *path = sha1_pack_name(sha1);
235 size_t alloc = st_add(strlen(path), 1);
236 struct packed_git *p = alloc_packed_git(alloc);
238 memcpy(p->pack_name, path, alloc); /* includes NUL */
239 hashcpy(p->hash, sha1);
240 if (check_packed_git_idx(idx_path, p)) {
241 free(p);
242 return NULL;
245 return p;
248 static void scan_windows(struct packed_git *p,
249 struct packed_git **lru_p,
250 struct pack_window **lru_w,
251 struct pack_window **lru_l)
253 struct pack_window *w, *w_l;
255 for (w_l = NULL, w = p->windows; w; w = w->next) {
256 if (!w->inuse_cnt) {
257 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
258 *lru_p = p;
259 *lru_w = w;
260 *lru_l = w_l;
263 w_l = w;
267 static int unuse_one_window(struct packed_git *current)
269 struct packed_git *p, *lru_p = NULL;
270 struct pack_window *lru_w = NULL, *lru_l = NULL;
272 if (current)
273 scan_windows(current, &lru_p, &lru_w, &lru_l);
274 for (p = the_repository->objects->packed_git; p; p = p->next)
275 scan_windows(p, &lru_p, &lru_w, &lru_l);
276 if (lru_p) {
277 munmap(lru_w->base, lru_w->len);
278 pack_mapped -= lru_w->len;
279 if (lru_l)
280 lru_l->next = lru_w->next;
281 else
282 lru_p->windows = lru_w->next;
283 free(lru_w);
284 pack_open_windows--;
285 return 1;
287 return 0;
290 void close_pack_windows(struct packed_git *p)
292 while (p->windows) {
293 struct pack_window *w = p->windows;
295 if (w->inuse_cnt)
296 die("pack '%s' still has open windows to it",
297 p->pack_name);
298 munmap(w->base, w->len);
299 pack_mapped -= w->len;
300 pack_open_windows--;
301 p->windows = w->next;
302 free(w);
306 int close_pack_fd(struct packed_git *p)
308 if (p->pack_fd < 0)
309 return 0;
311 close(p->pack_fd);
312 pack_open_fds--;
313 p->pack_fd = -1;
315 return 1;
318 void close_pack_index(struct packed_git *p)
320 if (p->index_data) {
321 munmap((void *)p->index_data, p->index_size);
322 p->index_data = NULL;
326 void close_pack(struct packed_git *p)
328 close_pack_windows(p);
329 close_pack_fd(p);
330 close_pack_index(p);
333 void close_object_store(struct raw_object_store *o)
335 struct packed_git *p;
337 for (p = o->packed_git; p; p = p->next)
338 if (p->do_not_close)
339 BUG("want to close pack marked 'do-not-close'");
340 else
341 close_pack(p);
343 if (o->multi_pack_index) {
344 close_midx(o->multi_pack_index);
345 o->multi_pack_index = NULL;
348 close_commit_graph(o);
351 void unlink_pack_path(const char *pack_name, int force_delete)
353 static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
354 int i;
355 struct strbuf buf = STRBUF_INIT;
356 size_t plen;
358 strbuf_addstr(&buf, pack_name);
359 strip_suffix_mem(buf.buf, &buf.len, ".pack");
360 plen = buf.len;
362 if (!force_delete) {
363 strbuf_addstr(&buf, ".keep");
364 if (!access(buf.buf, F_OK)) {
365 strbuf_release(&buf);
366 return;
370 for (i = 0; i < ARRAY_SIZE(exts); i++) {
371 strbuf_setlen(&buf, plen);
372 strbuf_addstr(&buf, exts[i]);
373 unlink(buf.buf);
376 strbuf_release(&buf);
380 * The LRU pack is the one with the oldest MRU window, preferring packs
381 * with no used windows, or the oldest mtime if it has no windows allocated.
383 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
385 struct pack_window *w, *this_mru_w;
386 int has_windows_inuse = 0;
389 * Reject this pack if it has windows and the previously selected
390 * one does not. If this pack does not have windows, reject
391 * it if the pack file is newer than the previously selected one.
393 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
394 return;
396 for (w = this_mru_w = p->windows; w; w = w->next) {
398 * Reject this pack if any of its windows are in use,
399 * but the previously selected pack did not have any
400 * inuse windows. Otherwise, record that this pack
401 * has windows in use.
403 if (w->inuse_cnt) {
404 if (*accept_windows_inuse)
405 has_windows_inuse = 1;
406 else
407 return;
410 if (w->last_used > this_mru_w->last_used)
411 this_mru_w = w;
414 * Reject this pack if it has windows that have been
415 * used more recently than the previously selected pack.
416 * If the previously selected pack had windows inuse and
417 * we have not encountered a window in this pack that is
418 * inuse, skip this check since we prefer a pack with no
419 * inuse windows to one that has inuse windows.
421 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
422 this_mru_w->last_used > (*mru_w)->last_used)
423 return;
427 * Select this pack.
429 *mru_w = this_mru_w;
430 *lru_p = p;
431 *accept_windows_inuse = has_windows_inuse;
434 static int close_one_pack(void)
436 struct packed_git *p, *lru_p = NULL;
437 struct pack_window *mru_w = NULL;
438 int accept_windows_inuse = 1;
440 for (p = the_repository->objects->packed_git; p; p = p->next) {
441 if (p->pack_fd == -1)
442 continue;
443 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
446 if (lru_p)
447 return close_pack_fd(lru_p);
449 return 0;
452 static unsigned int get_max_fd_limit(void)
454 #ifdef RLIMIT_NOFILE
456 struct rlimit lim;
458 if (!getrlimit(RLIMIT_NOFILE, &lim))
459 return lim.rlim_cur;
461 #endif
463 #ifdef _SC_OPEN_MAX
465 long open_max = sysconf(_SC_OPEN_MAX);
466 if (0 < open_max)
467 return open_max;
469 * Otherwise, we got -1 for one of the two
470 * reasons:
472 * (1) sysconf() did not understand _SC_OPEN_MAX
473 * and signaled an error with -1; or
474 * (2) sysconf() said there is no limit.
476 * We _could_ clear errno before calling sysconf() to
477 * tell these two cases apart and return a huge number
478 * in the latter case to let the caller cap it to a
479 * value that is not so selfish, but letting the
480 * fallback OPEN_MAX codepath take care of these cases
481 * is a lot simpler.
484 #endif
486 #ifdef OPEN_MAX
487 return OPEN_MAX;
488 #else
489 return 1; /* see the caller ;-) */
490 #endif
493 const char *pack_basename(struct packed_git *p)
495 const char *ret = strrchr(p->pack_name, '/');
496 if (ret)
497 ret = ret + 1; /* skip past slash */
498 else
499 ret = p->pack_name; /* we only have a base */
500 return ret;
504 * Do not call this directly as this leaks p->pack_fd on error return;
505 * call open_packed_git() instead.
507 static int open_packed_git_1(struct packed_git *p)
509 struct stat st;
510 struct pack_header hdr;
511 unsigned char hash[GIT_MAX_RAWSZ];
512 unsigned char *idx_hash;
513 ssize_t read_result;
514 const unsigned hashsz = the_hash_algo->rawsz;
516 if (!p->index_data) {
517 struct multi_pack_index *m;
518 const char *pack_name = pack_basename(p);
520 for (m = the_repository->objects->multi_pack_index;
521 m; m = m->next) {
522 if (midx_contains_pack(m, pack_name))
523 break;
526 if (!m && open_pack_index(p))
527 return error("packfile %s index unavailable", p->pack_name);
530 if (!pack_max_fds) {
531 unsigned int max_fds = get_max_fd_limit();
533 /* Save 3 for stdin/stdout/stderr, 22 for work */
534 if (25 < max_fds)
535 pack_max_fds = max_fds - 25;
536 else
537 pack_max_fds = 1;
540 while (pack_max_fds <= pack_open_fds && close_one_pack())
541 ; /* nothing */
543 p->pack_fd = git_open(p->pack_name);
544 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
545 return -1;
546 pack_open_fds++;
548 /* If we created the struct before we had the pack we lack size. */
549 if (!p->pack_size) {
550 if (!S_ISREG(st.st_mode))
551 return error("packfile %s not a regular file", p->pack_name);
552 p->pack_size = st.st_size;
553 } else if (p->pack_size != st.st_size)
554 return error("packfile %s size changed", p->pack_name);
556 /* Verify we recognize this pack file format. */
557 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
558 if (read_result < 0)
559 return error_errno("error reading from %s", p->pack_name);
560 if (read_result != sizeof(hdr))
561 return error("file %s is far too short to be a packfile", p->pack_name);
562 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
563 return error("file %s is not a GIT packfile", p->pack_name);
564 if (!pack_version_ok(hdr.hdr_version))
565 return error("packfile %s is version %"PRIu32" and not"
566 " supported (try upgrading GIT to a newer version)",
567 p->pack_name, ntohl(hdr.hdr_version));
569 /* Skip index checking if in multi-pack-index */
570 if (!p->index_data)
571 return 0;
573 /* Verify the pack matches its index. */
574 if (p->num_objects != ntohl(hdr.hdr_entries))
575 return error("packfile %s claims to have %"PRIu32" objects"
576 " while index indicates %"PRIu32" objects",
577 p->pack_name, ntohl(hdr.hdr_entries),
578 p->num_objects);
579 read_result = pread_in_full(p->pack_fd, hash, hashsz,
580 p->pack_size - hashsz);
581 if (read_result < 0)
582 return error_errno("error reading from %s", p->pack_name);
583 if (read_result != hashsz)
584 return error("packfile %s signature is unavailable", p->pack_name);
585 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
586 if (!hasheq(hash, idx_hash))
587 return error("packfile %s does not match index", p->pack_name);
588 return 0;
591 static int open_packed_git(struct packed_git *p)
593 if (!open_packed_git_1(p))
594 return 0;
595 close_pack_fd(p);
596 return -1;
599 static int in_window(struct pack_window *win, off_t offset)
601 /* We must promise at least one full hash after the
602 * offset is available from this window, otherwise the offset
603 * is not actually in this window and a different window (which
604 * has that one hash excess) must be used. This is to support
605 * the object header and delta base parsing routines below.
607 off_t win_off = win->offset;
608 return win_off <= offset
609 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
612 unsigned char *use_pack(struct packed_git *p,
613 struct pack_window **w_cursor,
614 off_t offset,
615 unsigned long *left)
617 struct pack_window *win = *w_cursor;
619 /* Since packfiles end in a hash of their content and it's
620 * pointless to ask for an offset into the middle of that
621 * hash, and the in_window function above wouldn't match
622 * don't allow an offset too close to the end of the file.
624 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
625 die("packfile %s cannot be accessed", p->pack_name);
626 if (offset > (p->pack_size - the_hash_algo->rawsz))
627 die("offset beyond end of packfile (truncated pack?)");
628 if (offset < 0)
629 die(_("offset before end of packfile (broken .idx?)"));
631 if (!win || !in_window(win, offset)) {
632 if (win)
633 win->inuse_cnt--;
634 for (win = p->windows; win; win = win->next) {
635 if (in_window(win, offset))
636 break;
638 if (!win) {
639 size_t window_align = packed_git_window_size / 2;
640 off_t len;
642 if (p->pack_fd == -1 && open_packed_git(p))
643 die("packfile %s cannot be accessed", p->pack_name);
645 win = xcalloc(1, sizeof(*win));
646 win->offset = (offset / window_align) * window_align;
647 len = p->pack_size - win->offset;
648 if (len > packed_git_window_size)
649 len = packed_git_window_size;
650 win->len = (size_t)len;
651 pack_mapped += win->len;
652 while (packed_git_limit < pack_mapped
653 && unuse_one_window(p))
654 ; /* nothing */
655 win->base = xmmap_gently(NULL, win->len,
656 PROT_READ, MAP_PRIVATE,
657 p->pack_fd, win->offset);
658 if (win->base == MAP_FAILED)
659 die_errno("packfile %s cannot be mapped",
660 p->pack_name);
661 if (!win->offset && win->len == p->pack_size
662 && !p->do_not_close)
663 close_pack_fd(p);
664 pack_mmap_calls++;
665 pack_open_windows++;
666 if (pack_mapped > peak_pack_mapped)
667 peak_pack_mapped = pack_mapped;
668 if (pack_open_windows > peak_pack_open_windows)
669 peak_pack_open_windows = pack_open_windows;
670 win->next = p->windows;
671 p->windows = win;
674 if (win != *w_cursor) {
675 win->last_used = pack_used_ctr++;
676 win->inuse_cnt++;
677 *w_cursor = win;
679 offset -= win->offset;
680 if (left)
681 *left = win->len - xsize_t(offset);
682 return win->base + offset;
685 void unuse_pack(struct pack_window **w_cursor)
687 struct pack_window *w = *w_cursor;
688 if (w) {
689 w->inuse_cnt--;
690 *w_cursor = NULL;
694 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
696 struct stat st;
697 size_t alloc;
698 struct packed_git *p;
701 * Make sure a corresponding .pack file exists and that
702 * the index looks sane.
704 if (!strip_suffix_mem(path, &path_len, ".idx"))
705 return NULL;
708 * ".promisor" is long enough to hold any suffix we're adding (and
709 * the use xsnprintf double-checks that)
711 alloc = st_add3(path_len, strlen(".promisor"), 1);
712 p = alloc_packed_git(alloc);
713 memcpy(p->pack_name, path, path_len);
715 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
716 if (!access(p->pack_name, F_OK))
717 p->pack_keep = 1;
719 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
720 if (!access(p->pack_name, F_OK))
721 p->pack_promisor = 1;
723 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
724 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
725 free(p);
726 return NULL;
729 /* ok, it looks sane as far as we can check without
730 * actually mapping the pack file.
732 p->pack_size = st.st_size;
733 p->pack_local = local;
734 p->mtime = st.st_mtime;
735 if (path_len < the_hash_algo->hexsz ||
736 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
737 hashclr(p->hash);
738 return p;
741 void install_packed_git(struct repository *r, struct packed_git *pack)
743 if (pack->pack_fd != -1)
744 pack_open_fds++;
746 pack->next = r->objects->packed_git;
747 r->objects->packed_git = pack;
749 hashmap_entry_init(&pack->packmap_ent, strhash(pack->pack_name));
750 hashmap_add(&r->objects->pack_map, &pack->packmap_ent);
753 void (*report_garbage)(unsigned seen_bits, const char *path);
755 static void report_helper(const struct string_list *list,
756 int seen_bits, int first, int last)
758 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
759 return;
761 for (; first < last; first++)
762 report_garbage(seen_bits, list->items[first].string);
765 static void report_pack_garbage(struct string_list *list)
767 int i, baselen = -1, first = 0, seen_bits = 0;
769 if (!report_garbage)
770 return;
772 string_list_sort(list);
774 for (i = 0; i < list->nr; i++) {
775 const char *path = list->items[i].string;
776 if (baselen != -1 &&
777 strncmp(path, list->items[first].string, baselen)) {
778 report_helper(list, seen_bits, first, i);
779 baselen = -1;
780 seen_bits = 0;
782 if (baselen == -1) {
783 const char *dot = strrchr(path, '.');
784 if (!dot) {
785 report_garbage(PACKDIR_FILE_GARBAGE, path);
786 continue;
788 baselen = dot - path + 1;
789 first = i;
791 if (!strcmp(path + baselen, "pack"))
792 seen_bits |= 1;
793 else if (!strcmp(path + baselen, "idx"))
794 seen_bits |= 2;
796 report_helper(list, seen_bits, first, list->nr);
799 void for_each_file_in_pack_dir(const char *objdir,
800 each_file_in_pack_dir_fn fn,
801 void *data)
803 struct strbuf path = STRBUF_INIT;
804 size_t dirnamelen;
805 DIR *dir;
806 struct dirent *de;
808 strbuf_addstr(&path, objdir);
809 strbuf_addstr(&path, "/pack");
810 dir = opendir(path.buf);
811 if (!dir) {
812 if (errno != ENOENT)
813 error_errno("unable to open object pack directory: %s",
814 path.buf);
815 strbuf_release(&path);
816 return;
818 strbuf_addch(&path, '/');
819 dirnamelen = path.len;
820 while ((de = readdir(dir)) != NULL) {
821 if (is_dot_or_dotdot(de->d_name))
822 continue;
824 strbuf_setlen(&path, dirnamelen);
825 strbuf_addstr(&path, de->d_name);
827 fn(path.buf, path.len, de->d_name, data);
830 closedir(dir);
831 strbuf_release(&path);
834 struct prepare_pack_data {
835 struct repository *r;
836 struct string_list *garbage;
837 int local;
838 struct multi_pack_index *m;
841 static void prepare_pack(const char *full_name, size_t full_name_len,
842 const char *file_name, void *_data)
844 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
845 struct packed_git *p;
846 size_t base_len = full_name_len;
848 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
849 !(data->m && midx_contains_pack(data->m, file_name))) {
850 struct hashmap_entry hent;
851 char *pack_name = xstrfmt("%.*s.pack", (int)base_len, full_name);
852 unsigned int hash = strhash(pack_name);
853 hashmap_entry_init(&hent, hash);
855 /* Don't reopen a pack we already have. */
856 if (!hashmap_get(&data->r->objects->pack_map, &hent, pack_name)) {
857 p = add_packed_git(full_name, full_name_len, data->local);
858 if (p)
859 install_packed_git(data->r, p);
861 free(pack_name);
864 if (!report_garbage)
865 return;
867 if (!strcmp(file_name, "multi-pack-index"))
868 return;
869 if (ends_with(file_name, ".idx") ||
870 ends_with(file_name, ".pack") ||
871 ends_with(file_name, ".bitmap") ||
872 ends_with(file_name, ".keep") ||
873 ends_with(file_name, ".promisor"))
874 string_list_append(data->garbage, full_name);
875 else
876 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
879 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
881 struct prepare_pack_data data;
882 struct string_list garbage = STRING_LIST_INIT_DUP;
884 data.m = r->objects->multi_pack_index;
886 /* look for the multi-pack-index for this object directory */
887 while (data.m && strcmp(data.m->object_dir, objdir))
888 data.m = data.m->next;
890 data.r = r;
891 data.garbage = &garbage;
892 data.local = local;
894 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
896 report_pack_garbage(data.garbage);
897 string_list_clear(data.garbage, 0);
900 static void prepare_packed_git(struct repository *r);
902 * Give a fast, rough count of the number of objects in the repository. This
903 * ignores loose objects completely. If you have a lot of them, then either
904 * you should repack because your performance will be awful, or they are
905 * all unreachable objects about to be pruned, in which case they're not really
906 * interesting as a measure of repo size in the first place.
908 unsigned long repo_approximate_object_count(struct repository *r)
910 if (!r->objects->approximate_object_count_valid) {
911 unsigned long count;
912 struct multi_pack_index *m;
913 struct packed_git *p;
915 prepare_packed_git(r);
916 count = 0;
917 for (m = get_multi_pack_index(r); m; m = m->next)
918 count += m->num_objects;
919 for (p = r->objects->packed_git; p; p = p->next) {
920 if (open_pack_index(p))
921 continue;
922 count += p->num_objects;
924 r->objects->approximate_object_count = count;
926 return r->objects->approximate_object_count;
929 static void *get_next_packed_git(const void *p)
931 return ((const struct packed_git *)p)->next;
934 static void set_next_packed_git(void *p, void *next)
936 ((struct packed_git *)p)->next = next;
939 static int sort_pack(const void *a_, const void *b_)
941 const struct packed_git *a = a_;
942 const struct packed_git *b = b_;
943 int st;
946 * Local packs tend to contain objects specific to our
947 * variant of the project than remote ones. In addition,
948 * remote ones could be on a network mounted filesystem.
949 * Favor local ones for these reasons.
951 st = a->pack_local - b->pack_local;
952 if (st)
953 return -st;
956 * Younger packs tend to contain more recent objects,
957 * and more recent objects tend to get accessed more
958 * often.
960 if (a->mtime < b->mtime)
961 return 1;
962 else if (a->mtime == b->mtime)
963 return 0;
964 return -1;
967 static void rearrange_packed_git(struct repository *r)
969 r->objects->packed_git = llist_mergesort(
970 r->objects->packed_git, get_next_packed_git,
971 set_next_packed_git, sort_pack);
974 static void prepare_packed_git_mru(struct repository *r)
976 struct packed_git *p;
978 INIT_LIST_HEAD(&r->objects->packed_git_mru);
980 for (p = r->objects->packed_git; p; p = p->next)
981 list_add_tail(&p->mru, &r->objects->packed_git_mru);
984 static void prepare_packed_git(struct repository *r)
986 struct object_directory *odb;
988 if (r->objects->packed_git_initialized)
989 return;
991 prepare_alt_odb(r);
992 for (odb = r->objects->odb; odb; odb = odb->next) {
993 int local = (odb == r->objects->odb);
994 prepare_multi_pack_index_one(r, odb->path, local);
995 prepare_packed_git_one(r, odb->path, local);
997 rearrange_packed_git(r);
999 prepare_packed_git_mru(r);
1000 r->objects->packed_git_initialized = 1;
1003 void reprepare_packed_git(struct repository *r)
1005 struct object_directory *odb;
1007 for (odb = r->objects->odb; odb; odb = odb->next)
1008 odb_clear_loose_cache(odb);
1010 r->objects->approximate_object_count_valid = 0;
1011 r->objects->packed_git_initialized = 0;
1012 prepare_packed_git(r);
1015 struct packed_git *get_packed_git(struct repository *r)
1017 prepare_packed_git(r);
1018 return r->objects->packed_git;
1021 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1023 prepare_packed_git(r);
1024 return r->objects->multi_pack_index;
1027 struct packed_git *get_all_packs(struct repository *r)
1029 struct multi_pack_index *m;
1031 prepare_packed_git(r);
1032 for (m = r->objects->multi_pack_index; m; m = m->next) {
1033 uint32_t i;
1034 for (i = 0; i < m->num_packs; i++)
1035 prepare_midx_pack(r, m, i);
1038 return r->objects->packed_git;
1041 struct list_head *get_packed_git_mru(struct repository *r)
1043 prepare_packed_git(r);
1044 return &r->objects->packed_git_mru;
1047 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1048 unsigned long len, enum object_type *type, unsigned long *sizep)
1050 unsigned shift;
1051 unsigned long size, c;
1052 unsigned long used = 0;
1054 c = buf[used++];
1055 *type = (c >> 4) & 7;
1056 size = c & 15;
1057 shift = 4;
1058 while (c & 0x80) {
1059 if (len <= used || bitsizeof(long) <= shift) {
1060 error("bad object header");
1061 size = used = 0;
1062 break;
1064 c = buf[used++];
1065 size += (c & 0x7f) << shift;
1066 shift += 7;
1068 *sizep = size;
1069 return used;
1072 unsigned long get_size_from_delta(struct packed_git *p,
1073 struct pack_window **w_curs,
1074 off_t curpos)
1076 const unsigned char *data;
1077 unsigned char delta_head[20], *in;
1078 git_zstream stream;
1079 int st;
1081 memset(&stream, 0, sizeof(stream));
1082 stream.next_out = delta_head;
1083 stream.avail_out = sizeof(delta_head);
1085 git_inflate_init(&stream);
1086 do {
1087 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1088 stream.next_in = in;
1089 st = git_inflate(&stream, Z_FINISH);
1090 curpos += stream.next_in - in;
1091 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1092 stream.total_out < sizeof(delta_head));
1093 git_inflate_end(&stream);
1094 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1095 error("delta data unpack-initial failed");
1096 return 0;
1099 /* Examine the initial part of the delta to figure out
1100 * the result size.
1102 data = delta_head;
1104 /* ignore base size */
1105 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1107 /* Read the result size */
1108 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1111 int unpack_object_header(struct packed_git *p,
1112 struct pack_window **w_curs,
1113 off_t *curpos,
1114 unsigned long *sizep)
1116 unsigned char *base;
1117 unsigned long left;
1118 unsigned long used;
1119 enum object_type type;
1121 /* use_pack() assures us we have [base, base + 20) available
1122 * as a range that we can look at. (Its actually the hash
1123 * size that is assured.) With our object header encoding
1124 * the maximum deflated object size is 2^137, which is just
1125 * insane, so we know won't exceed what we have been given.
1127 base = use_pack(p, w_curs, *curpos, &left);
1128 used = unpack_object_header_buffer(base, left, &type, sizep);
1129 if (!used) {
1130 type = OBJ_BAD;
1131 } else
1132 *curpos += used;
1134 return type;
1137 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1139 unsigned i;
1140 const unsigned hashsz = the_hash_algo->rawsz;
1141 for (i = 0; i < p->num_bad_objects; i++)
1142 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1143 return;
1144 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1145 st_mult(GIT_MAX_RAWSZ,
1146 st_add(p->num_bad_objects, 1)));
1147 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1148 p->num_bad_objects++;
1151 const struct packed_git *has_packed_and_bad(struct repository *r,
1152 const unsigned char *sha1)
1154 struct packed_git *p;
1155 unsigned i;
1157 for (p = r->objects->packed_git; p; p = p->next)
1158 for (i = 0; i < p->num_bad_objects; i++)
1159 if (hasheq(sha1,
1160 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1161 return p;
1162 return NULL;
1165 static off_t get_delta_base(struct packed_git *p,
1166 struct pack_window **w_curs,
1167 off_t *curpos,
1168 enum object_type type,
1169 off_t delta_obj_offset)
1171 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1172 off_t base_offset;
1174 /* use_pack() assured us we have [base_info, base_info + 20)
1175 * as a range that we can look at without walking off the
1176 * end of the mapped window. Its actually the hash size
1177 * that is assured. An OFS_DELTA longer than the hash size
1178 * is stupid, as then a REF_DELTA would be smaller to store.
1180 if (type == OBJ_OFS_DELTA) {
1181 unsigned used = 0;
1182 unsigned char c = base_info[used++];
1183 base_offset = c & 127;
1184 while (c & 128) {
1185 base_offset += 1;
1186 if (!base_offset || MSB(base_offset, 7))
1187 return 0; /* overflow */
1188 c = base_info[used++];
1189 base_offset = (base_offset << 7) + (c & 127);
1191 base_offset = delta_obj_offset - base_offset;
1192 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1193 return 0; /* out of bound */
1194 *curpos += used;
1195 } else if (type == OBJ_REF_DELTA) {
1196 /* The base entry _must_ be in the same pack */
1197 base_offset = find_pack_entry_one(base_info, p);
1198 *curpos += the_hash_algo->rawsz;
1199 } else
1200 die("I am totally screwed");
1201 return base_offset;
1205 * Like get_delta_base above, but we return the sha1 instead of the pack
1206 * offset. This means it is cheaper for REF deltas (we do not have to do
1207 * the final object lookup), but more expensive for OFS deltas (we
1208 * have to load the revidx to convert the offset back into a sha1).
1210 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1211 struct pack_window **w_curs,
1212 off_t curpos,
1213 enum object_type type,
1214 off_t delta_obj_offset)
1216 if (type == OBJ_REF_DELTA) {
1217 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1218 return base;
1219 } else if (type == OBJ_OFS_DELTA) {
1220 struct revindex_entry *revidx;
1221 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1222 type, delta_obj_offset);
1224 if (!base_offset)
1225 return NULL;
1227 revidx = find_pack_revindex(p, base_offset);
1228 if (!revidx)
1229 return NULL;
1231 return nth_packed_object_sha1(p, revidx->nr);
1232 } else
1233 return NULL;
1236 static int retry_bad_packed_offset(struct repository *r,
1237 struct packed_git *p,
1238 off_t obj_offset)
1240 int type;
1241 struct revindex_entry *revidx;
1242 struct object_id oid;
1243 revidx = find_pack_revindex(p, obj_offset);
1244 if (!revidx)
1245 return OBJ_BAD;
1246 nth_packed_object_oid(&oid, p, revidx->nr);
1247 mark_bad_packed_object(p, oid.hash);
1248 type = oid_object_info(r, &oid, NULL);
1249 if (type <= OBJ_NONE)
1250 return OBJ_BAD;
1251 return type;
1254 #define POI_STACK_PREALLOC 64
1256 static enum object_type packed_to_object_type(struct repository *r,
1257 struct packed_git *p,
1258 off_t obj_offset,
1259 enum object_type type,
1260 struct pack_window **w_curs,
1261 off_t curpos)
1263 off_t small_poi_stack[POI_STACK_PREALLOC];
1264 off_t *poi_stack = small_poi_stack;
1265 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1267 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1268 off_t base_offset;
1269 unsigned long size;
1270 /* Push the object we're going to leave behind */
1271 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1272 poi_stack_alloc = alloc_nr(poi_stack_nr);
1273 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1274 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1275 } else {
1276 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1278 poi_stack[poi_stack_nr++] = obj_offset;
1279 /* If parsing the base offset fails, just unwind */
1280 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1281 if (!base_offset)
1282 goto unwind;
1283 curpos = obj_offset = base_offset;
1284 type = unpack_object_header(p, w_curs, &curpos, &size);
1285 if (type <= OBJ_NONE) {
1286 /* If getting the base itself fails, we first
1287 * retry the base, otherwise unwind */
1288 type = retry_bad_packed_offset(r, p, base_offset);
1289 if (type > OBJ_NONE)
1290 goto out;
1291 goto unwind;
1295 switch (type) {
1296 case OBJ_BAD:
1297 case OBJ_COMMIT:
1298 case OBJ_TREE:
1299 case OBJ_BLOB:
1300 case OBJ_TAG:
1301 break;
1302 default:
1303 error("unknown object type %i at offset %"PRIuMAX" in %s",
1304 type, (uintmax_t)obj_offset, p->pack_name);
1305 type = OBJ_BAD;
1308 out:
1309 if (poi_stack != small_poi_stack)
1310 free(poi_stack);
1311 return type;
1313 unwind:
1314 while (poi_stack_nr) {
1315 obj_offset = poi_stack[--poi_stack_nr];
1316 type = retry_bad_packed_offset(r, p, obj_offset);
1317 if (type > OBJ_NONE)
1318 goto out;
1320 type = OBJ_BAD;
1321 goto out;
1324 static struct hashmap delta_base_cache;
1325 static size_t delta_base_cached;
1327 static LIST_HEAD(delta_base_cache_lru);
1329 struct delta_base_cache_key {
1330 struct packed_git *p;
1331 off_t base_offset;
1334 struct delta_base_cache_entry {
1335 struct hashmap_entry ent;
1336 struct delta_base_cache_key key;
1337 struct list_head lru;
1338 void *data;
1339 unsigned long size;
1340 enum object_type type;
1343 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1345 unsigned int hash;
1347 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1348 hash += (hash >> 8) + (hash >> 16);
1349 return hash;
1352 static struct delta_base_cache_entry *
1353 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1355 struct hashmap_entry entry, *e;
1356 struct delta_base_cache_key key;
1358 if (!delta_base_cache.cmpfn)
1359 return NULL;
1361 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1362 key.p = p;
1363 key.base_offset = base_offset;
1364 e = hashmap_get(&delta_base_cache, &entry, &key);
1365 return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1368 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1369 const struct delta_base_cache_key *b)
1371 return a->p == b->p && a->base_offset == b->base_offset;
1374 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1375 const struct hashmap_entry *va,
1376 const struct hashmap_entry *vb,
1377 const void *vkey)
1379 const struct delta_base_cache_entry *a, *b;
1380 const struct delta_base_cache_key *key = vkey;
1382 a = container_of(va, const struct delta_base_cache_entry, ent);
1383 b = container_of(vb, const struct delta_base_cache_entry, ent);
1385 if (key)
1386 return !delta_base_cache_key_eq(&a->key, key);
1387 else
1388 return !delta_base_cache_key_eq(&a->key, &b->key);
1391 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1393 return !!get_delta_base_cache_entry(p, base_offset);
1397 * Remove the entry from the cache, but do _not_ free the associated
1398 * entry data. The caller takes ownership of the "data" buffer, and
1399 * should copy out any fields it wants before detaching.
1401 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1403 hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1404 list_del(&ent->lru);
1405 delta_base_cached -= ent->size;
1406 free(ent);
1409 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1410 off_t base_offset, unsigned long *base_size,
1411 enum object_type *type)
1413 struct delta_base_cache_entry *ent;
1415 ent = get_delta_base_cache_entry(p, base_offset);
1416 if (!ent)
1417 return unpack_entry(r, p, base_offset, type, base_size);
1419 if (type)
1420 *type = ent->type;
1421 if (base_size)
1422 *base_size = ent->size;
1423 return xmemdupz(ent->data, ent->size);
1426 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1428 free(ent->data);
1429 detach_delta_base_cache_entry(ent);
1432 void clear_delta_base_cache(void)
1434 struct list_head *lru, *tmp;
1435 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1436 struct delta_base_cache_entry *entry =
1437 list_entry(lru, struct delta_base_cache_entry, lru);
1438 release_delta_base_cache(entry);
1442 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1443 void *base, unsigned long base_size, enum object_type type)
1445 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1446 struct list_head *lru, *tmp;
1448 delta_base_cached += base_size;
1450 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1451 struct delta_base_cache_entry *f =
1452 list_entry(lru, struct delta_base_cache_entry, lru);
1453 if (delta_base_cached <= delta_base_cache_limit)
1454 break;
1455 release_delta_base_cache(f);
1458 ent->key.p = p;
1459 ent->key.base_offset = base_offset;
1460 ent->type = type;
1461 ent->data = base;
1462 ent->size = base_size;
1463 list_add_tail(&ent->lru, &delta_base_cache_lru);
1465 if (!delta_base_cache.cmpfn)
1466 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1467 hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1468 hashmap_add(&delta_base_cache, &ent->ent);
1471 int packed_object_info(struct repository *r, struct packed_git *p,
1472 off_t obj_offset, struct object_info *oi)
1474 struct pack_window *w_curs = NULL;
1475 unsigned long size;
1476 off_t curpos = obj_offset;
1477 enum object_type type;
1480 * We always get the representation type, but only convert it to
1481 * a "real" type later if the caller is interested.
1483 if (oi->contentp) {
1484 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1485 &type);
1486 if (!*oi->contentp)
1487 type = OBJ_BAD;
1488 } else {
1489 type = unpack_object_header(p, &w_curs, &curpos, &size);
1492 if (!oi->contentp && oi->sizep) {
1493 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1494 off_t tmp_pos = curpos;
1495 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1496 type, obj_offset);
1497 if (!base_offset) {
1498 type = OBJ_BAD;
1499 goto out;
1501 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1502 if (*oi->sizep == 0) {
1503 type = OBJ_BAD;
1504 goto out;
1506 } else {
1507 *oi->sizep = size;
1511 if (oi->disk_sizep) {
1512 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1513 *oi->disk_sizep = revidx[1].offset - obj_offset;
1516 if (oi->typep || oi->type_name) {
1517 enum object_type ptot;
1518 ptot = packed_to_object_type(r, p, obj_offset,
1519 type, &w_curs, curpos);
1520 if (oi->typep)
1521 *oi->typep = ptot;
1522 if (oi->type_name) {
1523 const char *tn = type_name(ptot);
1524 if (tn)
1525 strbuf_addstr(oi->type_name, tn);
1527 if (ptot < 0) {
1528 type = OBJ_BAD;
1529 goto out;
1533 if (oi->delta_base_sha1) {
1534 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1535 const unsigned char *base;
1537 base = get_delta_base_sha1(p, &w_curs, curpos,
1538 type, obj_offset);
1539 if (!base) {
1540 type = OBJ_BAD;
1541 goto out;
1544 hashcpy(oi->delta_base_sha1, base);
1545 } else
1546 hashclr(oi->delta_base_sha1);
1549 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1550 OI_PACKED;
1552 out:
1553 unuse_pack(&w_curs);
1554 return type;
1557 static void *unpack_compressed_entry(struct packed_git *p,
1558 struct pack_window **w_curs,
1559 off_t curpos,
1560 unsigned long size)
1562 int st;
1563 git_zstream stream;
1564 unsigned char *buffer, *in;
1566 buffer = xmallocz_gently(size);
1567 if (!buffer)
1568 return NULL;
1569 memset(&stream, 0, sizeof(stream));
1570 stream.next_out = buffer;
1571 stream.avail_out = size + 1;
1573 git_inflate_init(&stream);
1574 do {
1575 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1576 stream.next_in = in;
1577 st = git_inflate(&stream, Z_FINISH);
1578 if (!stream.avail_out)
1579 break; /* the payload is larger than it should be */
1580 curpos += stream.next_in - in;
1581 } while (st == Z_OK || st == Z_BUF_ERROR);
1582 git_inflate_end(&stream);
1583 if ((st != Z_STREAM_END) || stream.total_out != size) {
1584 free(buffer);
1585 return NULL;
1588 /* versions of zlib can clobber unconsumed portion of outbuf */
1589 buffer[size] = '\0';
1591 return buffer;
1594 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1596 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1597 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1598 p->pack_name, (uintmax_t)obj_offset);
1601 int do_check_packed_object_crc;
1603 #define UNPACK_ENTRY_STACK_PREALLOC 64
1604 struct unpack_entry_stack_ent {
1605 off_t obj_offset;
1606 off_t curpos;
1607 unsigned long size;
1610 static void *read_object(struct repository *r,
1611 const struct object_id *oid,
1612 enum object_type *type,
1613 unsigned long *size)
1615 struct object_info oi = OBJECT_INFO_INIT;
1616 void *content;
1617 oi.typep = type;
1618 oi.sizep = size;
1619 oi.contentp = &content;
1621 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1622 return NULL;
1623 return content;
1626 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1627 enum object_type *final_type, unsigned long *final_size)
1629 struct pack_window *w_curs = NULL;
1630 off_t curpos = obj_offset;
1631 void *data = NULL;
1632 unsigned long size;
1633 enum object_type type;
1634 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1635 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1636 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1637 int base_from_cache = 0;
1639 write_pack_access_log(p, obj_offset);
1641 /* PHASE 1: drill down to the innermost base object */
1642 for (;;) {
1643 off_t base_offset;
1644 int i;
1645 struct delta_base_cache_entry *ent;
1647 ent = get_delta_base_cache_entry(p, curpos);
1648 if (ent) {
1649 type = ent->type;
1650 data = ent->data;
1651 size = ent->size;
1652 detach_delta_base_cache_entry(ent);
1653 base_from_cache = 1;
1654 break;
1657 if (do_check_packed_object_crc && p->index_version > 1) {
1658 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1659 off_t len = revidx[1].offset - obj_offset;
1660 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1661 struct object_id oid;
1662 nth_packed_object_oid(&oid, p, revidx->nr);
1663 error("bad packed object CRC for %s",
1664 oid_to_hex(&oid));
1665 mark_bad_packed_object(p, oid.hash);
1666 data = NULL;
1667 goto out;
1671 type = unpack_object_header(p, &w_curs, &curpos, &size);
1672 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1673 break;
1675 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1676 if (!base_offset) {
1677 error("failed to validate delta base reference "
1678 "at offset %"PRIuMAX" from %s",
1679 (uintmax_t)curpos, p->pack_name);
1680 /* bail to phase 2, in hopes of recovery */
1681 data = NULL;
1682 break;
1685 /* push object, proceed to base */
1686 if (delta_stack_nr >= delta_stack_alloc
1687 && delta_stack == small_delta_stack) {
1688 delta_stack_alloc = alloc_nr(delta_stack_nr);
1689 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1690 COPY_ARRAY(delta_stack, small_delta_stack,
1691 delta_stack_nr);
1692 } else {
1693 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1695 i = delta_stack_nr++;
1696 delta_stack[i].obj_offset = obj_offset;
1697 delta_stack[i].curpos = curpos;
1698 delta_stack[i].size = size;
1700 curpos = obj_offset = base_offset;
1703 /* PHASE 2: handle the base */
1704 switch (type) {
1705 case OBJ_OFS_DELTA:
1706 case OBJ_REF_DELTA:
1707 if (data)
1708 BUG("unpack_entry: left loop at a valid delta");
1709 break;
1710 case OBJ_COMMIT:
1711 case OBJ_TREE:
1712 case OBJ_BLOB:
1713 case OBJ_TAG:
1714 if (!base_from_cache)
1715 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1716 break;
1717 default:
1718 data = NULL;
1719 error("unknown object type %i at offset %"PRIuMAX" in %s",
1720 type, (uintmax_t)obj_offset, p->pack_name);
1723 /* PHASE 3: apply deltas in order */
1725 /* invariants:
1726 * 'data' holds the base data, or NULL if there was corruption
1728 while (delta_stack_nr) {
1729 void *delta_data;
1730 void *base = data;
1731 void *external_base = NULL;
1732 unsigned long delta_size, base_size = size;
1733 int i;
1735 data = NULL;
1737 if (base)
1738 add_delta_base_cache(p, obj_offset, base, base_size, type);
1740 if (!base) {
1742 * We're probably in deep shit, but let's try to fetch
1743 * the required base anyway from another pack or loose.
1744 * This is costly but should happen only in the presence
1745 * of a corrupted pack, and is better than failing outright.
1747 struct revindex_entry *revidx;
1748 struct object_id base_oid;
1749 revidx = find_pack_revindex(p, obj_offset);
1750 if (revidx) {
1751 nth_packed_object_oid(&base_oid, p, revidx->nr);
1752 error("failed to read delta base object %s"
1753 " at offset %"PRIuMAX" from %s",
1754 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1755 p->pack_name);
1756 mark_bad_packed_object(p, base_oid.hash);
1757 base = read_object(r, &base_oid, &type, &base_size);
1758 external_base = base;
1762 i = --delta_stack_nr;
1763 obj_offset = delta_stack[i].obj_offset;
1764 curpos = delta_stack[i].curpos;
1765 delta_size = delta_stack[i].size;
1767 if (!base)
1768 continue;
1770 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1772 if (!delta_data) {
1773 error("failed to unpack compressed delta "
1774 "at offset %"PRIuMAX" from %s",
1775 (uintmax_t)curpos, p->pack_name);
1776 data = NULL;
1777 free(external_base);
1778 continue;
1781 data = patch_delta(base, base_size,
1782 delta_data, delta_size,
1783 &size);
1786 * We could not apply the delta; warn the user, but keep going.
1787 * Our failure will be noticed either in the next iteration of
1788 * the loop, or if this is the final delta, in the caller when
1789 * we return NULL. Those code paths will take care of making
1790 * a more explicit warning and retrying with another copy of
1791 * the object.
1793 if (!data)
1794 error("failed to apply delta");
1796 free(delta_data);
1797 free(external_base);
1800 if (final_type)
1801 *final_type = type;
1802 if (final_size)
1803 *final_size = size;
1805 out:
1806 unuse_pack(&w_curs);
1808 if (delta_stack != small_delta_stack)
1809 free(delta_stack);
1811 return data;
1814 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1816 const unsigned char *index_fanout = p->index_data;
1817 const unsigned char *index_lookup;
1818 const unsigned int hashsz = the_hash_algo->rawsz;
1819 int index_lookup_width;
1821 if (!index_fanout)
1822 BUG("bsearch_pack called without a valid pack-index");
1824 index_lookup = index_fanout + 4 * 256;
1825 if (p->index_version == 1) {
1826 index_lookup_width = hashsz + 4;
1827 index_lookup += 4;
1828 } else {
1829 index_lookup_width = hashsz;
1830 index_fanout += 8;
1831 index_lookup += 8;
1834 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1835 index_lookup, index_lookup_width, result);
1838 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1839 uint32_t n)
1841 const unsigned char *index = p->index_data;
1842 const unsigned int hashsz = the_hash_algo->rawsz;
1843 if (!index) {
1844 if (open_pack_index(p))
1845 return NULL;
1846 index = p->index_data;
1848 if (n >= p->num_objects)
1849 return NULL;
1850 index += 4 * 256;
1851 if (p->index_version == 1) {
1852 return index + (hashsz + 4) * n + 4;
1853 } else {
1854 index += 8;
1855 return index + hashsz * n;
1859 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1860 struct packed_git *p,
1861 uint32_t n)
1863 const unsigned char *hash = nth_packed_object_sha1(p, n);
1864 if (!hash)
1865 return NULL;
1866 hashcpy(oid->hash, hash);
1867 return oid;
1870 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1872 const unsigned char *ptr = vptr;
1873 const unsigned char *start = p->index_data;
1874 const unsigned char *end = start + p->index_size;
1875 if (ptr < start)
1876 die(_("offset before start of pack index for %s (corrupt index?)"),
1877 p->pack_name);
1878 /* No need to check for underflow; .idx files must be at least 8 bytes */
1879 if (ptr >= end - 8)
1880 die(_("offset beyond end of pack index for %s (truncated index?)"),
1881 p->pack_name);
1884 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1886 const unsigned char *index = p->index_data;
1887 const unsigned int hashsz = the_hash_algo->rawsz;
1888 index += 4 * 256;
1889 if (p->index_version == 1) {
1890 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1891 } else {
1892 uint32_t off;
1893 index += 8 + p->num_objects * (hashsz + 4);
1894 off = ntohl(*((uint32_t *)(index + 4 * n)));
1895 if (!(off & 0x80000000))
1896 return off;
1897 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1898 check_pack_index_ptr(p, index);
1899 return get_be64(index);
1903 off_t find_pack_entry_one(const unsigned char *sha1,
1904 struct packed_git *p)
1906 const unsigned char *index = p->index_data;
1907 struct object_id oid;
1908 uint32_t result;
1910 if (!index) {
1911 if (open_pack_index(p))
1912 return 0;
1915 hashcpy(oid.hash, sha1);
1916 if (bsearch_pack(&oid, p, &result))
1917 return nth_packed_object_offset(p, result);
1918 return 0;
1921 int is_pack_valid(struct packed_git *p)
1923 /* An already open pack is known to be valid. */
1924 if (p->pack_fd != -1)
1925 return 1;
1927 /* If the pack has one window completely covering the
1928 * file size, the pack is known to be valid even if
1929 * the descriptor is not currently open.
1931 if (p->windows) {
1932 struct pack_window *w = p->windows;
1934 if (!w->offset && w->len == p->pack_size)
1935 return 1;
1938 /* Force the pack to open to prove its valid. */
1939 return !open_packed_git(p);
1942 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1943 struct packed_git *packs)
1945 struct packed_git *p;
1947 for (p = packs; p; p = p->next) {
1948 if (find_pack_entry_one(sha1, p))
1949 return p;
1951 return NULL;
1955 static int fill_pack_entry(const struct object_id *oid,
1956 struct pack_entry *e,
1957 struct packed_git *p)
1959 off_t offset;
1961 if (p->num_bad_objects) {
1962 unsigned i;
1963 for (i = 0; i < p->num_bad_objects; i++)
1964 if (hasheq(oid->hash,
1965 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1966 return 0;
1969 offset = find_pack_entry_one(oid->hash, p);
1970 if (!offset)
1971 return 0;
1974 * We are about to tell the caller where they can locate the
1975 * requested object. We better make sure the packfile is
1976 * still here and can be accessed before supplying that
1977 * answer, as it may have been deleted since the index was
1978 * loaded!
1980 if (!is_pack_valid(p))
1981 return 0;
1982 e->offset = offset;
1983 e->p = p;
1984 return 1;
1987 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1989 struct list_head *pos;
1990 struct multi_pack_index *m;
1992 prepare_packed_git(r);
1993 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1994 return 0;
1996 for (m = r->objects->multi_pack_index; m; m = m->next) {
1997 if (fill_midx_entry(r, oid, e, m))
1998 return 1;
2001 list_for_each(pos, &r->objects->packed_git_mru) {
2002 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2003 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2004 list_move(&p->mru, &r->objects->packed_git_mru);
2005 return 1;
2008 return 0;
2011 int has_object_pack(const struct object_id *oid)
2013 struct pack_entry e;
2014 return find_pack_entry(the_repository, oid, &e);
2017 int has_pack_index(const unsigned char *sha1)
2019 struct stat st;
2020 if (stat(sha1_pack_index_name(sha1), &st))
2021 return 0;
2022 return 1;
2025 int for_each_object_in_pack(struct packed_git *p,
2026 each_packed_object_fn cb, void *data,
2027 enum for_each_object_flags flags)
2029 uint32_t i;
2030 int r = 0;
2032 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2033 if (load_pack_revindex(p))
2034 return -1;
2037 for (i = 0; i < p->num_objects; i++) {
2038 uint32_t pos;
2039 struct object_id oid;
2041 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2042 pos = p->revindex[i].nr;
2043 else
2044 pos = i;
2046 if (!nth_packed_object_oid(&oid, p, pos))
2047 return error("unable to get sha1 of object %u in %s",
2048 pos, p->pack_name);
2050 r = cb(&oid, p, pos, data);
2051 if (r)
2052 break;
2054 return r;
2057 int for_each_packed_object(each_packed_object_fn cb, void *data,
2058 enum for_each_object_flags flags)
2060 struct packed_git *p;
2061 int r = 0;
2062 int pack_errors = 0;
2064 prepare_packed_git(the_repository);
2065 for (p = get_all_packs(the_repository); p; p = p->next) {
2066 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2067 continue;
2068 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2069 !p->pack_promisor)
2070 continue;
2071 if (open_pack_index(p)) {
2072 pack_errors = 1;
2073 continue;
2075 r = for_each_object_in_pack(p, cb, data, flags);
2076 if (r)
2077 break;
2079 return r ? r : pack_errors;
2082 static int add_promisor_object(const struct object_id *oid,
2083 struct packed_git *pack,
2084 uint32_t pos,
2085 void *set_)
2087 struct oidset *set = set_;
2088 struct object *obj = parse_object(the_repository, oid);
2089 if (!obj)
2090 return 1;
2092 oidset_insert(set, oid);
2095 * If this is a tree, commit, or tag, the objects it refers
2096 * to are also promisor objects. (Blobs refer to no objects->)
2098 if (obj->type == OBJ_TREE) {
2099 struct tree *tree = (struct tree *)obj;
2100 struct tree_desc desc;
2101 struct name_entry entry;
2102 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2104 * Error messages are given when packs are
2105 * verified, so do not print any here.
2107 return 0;
2108 while (tree_entry_gently(&desc, &entry))
2109 oidset_insert(set, &entry.oid);
2110 } else if (obj->type == OBJ_COMMIT) {
2111 struct commit *commit = (struct commit *) obj;
2112 struct commit_list *parents = commit->parents;
2114 oidset_insert(set, get_commit_tree_oid(commit));
2115 for (; parents; parents = parents->next)
2116 oidset_insert(set, &parents->item->object.oid);
2117 } else if (obj->type == OBJ_TAG) {
2118 struct tag *tag = (struct tag *) obj;
2119 oidset_insert(set, get_tagged_oid(tag));
2121 return 0;
2124 int is_promisor_object(const struct object_id *oid)
2126 static struct oidset promisor_objects;
2127 static int promisor_objects_prepared;
2129 if (!promisor_objects_prepared) {
2130 if (has_promisor_remote()) {
2131 for_each_packed_object(add_promisor_object,
2132 &promisor_objects,
2133 FOR_EACH_OBJECT_PROMISOR_ONLY);
2135 promisor_objects_prepared = 1;
2137 return oidset_contains(&promisor_objects, oid);