sha1_name: convert struct min_abbrev_data to object_id
[git/git.git] / packfile.c
CommitLineData
4f39cd82 1#include "cache.h"
ec2dd32c 2#include "list.h"
0317f455 3#include "pack.h"
0abe14f6
JT
4#include "dir.h"
5#include "mergesort.h"
6#include "packfile.h"
7b3aa75d 7#include "delta.h"
f1d8130b
JT
8#include "list.h"
9#include "streaming.h"
a2551953 10#include "sha1-lookup.h"
498f1f61
JT
11#include "commit.h"
12#include "object.h"
13#include "tag.h"
14#include "tree-walk.h"
15#include "tree.h"
4f39cd82
JT
16
17char *odb_pack_name(struct strbuf *buf,
18 const unsigned char *sha1,
19 const char *ext)
20{
21 strbuf_reset(buf);
22 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
23 sha1_to_hex(sha1), ext);
24 return buf->buf;
25}
26
27char *sha1_pack_name(const unsigned char *sha1)
28{
29 static struct strbuf buf = STRBUF_INIT;
30 return odb_pack_name(&buf, sha1, "pack");
31}
32
33char *sha1_pack_index_name(const unsigned char *sha1)
34{
35 static struct strbuf buf = STRBUF_INIT;
36 return odb_pack_name(&buf, sha1, "idx");
37}
6d6a80e0 38
84f80ad5
JT
39static unsigned int pack_used_ctr;
40static unsigned int pack_mmap_calls;
41static unsigned int peak_pack_open_windows;
42static unsigned int pack_open_windows;
e65f1862 43static unsigned int pack_open_fds;
84f80ad5
JT
44static unsigned int pack_max_fds;
45static size_t peak_pack_mapped;
46static size_t pack_mapped;
6d6a80e0 47struct packed_git *packed_git;
ec2dd32c 48LIST_HEAD(packed_git_mru);
8e21176c
JT
49
50#define SZ_FMT PRIuMAX
51static inline uintmax_t sz_fmt(size_t s) { return s; }
52
53void pack_report(void)
54{
55 fprintf(stderr,
56 "pack_report: getpagesize() = %10" SZ_FMT "\n"
57 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
58 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
59 sz_fmt(getpagesize()),
60 sz_fmt(packed_git_window_size),
61 sz_fmt(packed_git_limit));
62 fprintf(stderr,
63 "pack_report: pack_used_ctr = %10u\n"
64 "pack_report: pack_mmap_calls = %10u\n"
65 "pack_report: pack_open_windows = %10u / %10u\n"
66 "pack_report: pack_mapped = "
67 "%10" SZ_FMT " / %10" SZ_FMT "\n",
68 pack_used_ctr,
69 pack_mmap_calls,
70 pack_open_windows, peak_pack_open_windows,
71 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
72}
0317f455
JT
73
74/*
75 * Open and mmap the index file at path, perform a couple of
76 * consistency checks, then record its information to p. Return 0 on
77 * success.
78 */
79static int check_packed_git_idx(const char *path, struct packed_git *p)
80{
81 void *idx_map;
82 struct pack_idx_header *hdr;
83 size_t idx_size;
84 uint32_t version, nr, i, *index;
85 int fd = git_open(path);
86 struct stat st;
87
88 if (fd < 0)
89 return -1;
90 if (fstat(fd, &st)) {
91 close(fd);
92 return -1;
93 }
94 idx_size = xsize_t(st.st_size);
95 if (idx_size < 4 * 256 + 20 + 20) {
96 close(fd);
97 return error("index file %s is too small", path);
98 }
99 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
100 close(fd);
101
102 hdr = idx_map;
103 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
104 version = ntohl(hdr->idx_version);
105 if (version < 2 || version > 2) {
106 munmap(idx_map, idx_size);
107 return error("index file %s is version %"PRIu32
108 " and is not supported by this binary"
109 " (try upgrading GIT to a newer version)",
110 path, version);
111 }
112 } else
113 version = 1;
114
115 nr = 0;
116 index = idx_map;
117 if (version > 1)
118 index += 2; /* skip index header */
119 for (i = 0; i < 256; i++) {
120 uint32_t n = ntohl(index[i]);
121 if (n < nr) {
122 munmap(idx_map, idx_size);
123 return error("non-monotonic index %s", path);
124 }
125 nr = n;
126 }
127
128 if (version == 1) {
129 /*
130 * Total size:
131 * - 256 index entries 4 bytes each
132 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
133 * - 20-byte SHA1 of the packfile
134 * - 20-byte SHA1 file checksum
135 */
136 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
137 munmap(idx_map, idx_size);
138 return error("wrong index v1 file size in %s", path);
139 }
140 } else if (version == 2) {
141 /*
142 * Minimum size:
143 * - 8 bytes of header
144 * - 256 index entries 4 bytes each
145 * - 20-byte sha1 entry * nr
146 * - 4-byte crc entry * nr
147 * - 4-byte offset entry * nr
148 * - 20-byte SHA1 of the packfile
149 * - 20-byte SHA1 file checksum
150 * And after the 4-byte offset table might be a
151 * variable sized table containing 8-byte entries
152 * for offsets larger than 2^31.
153 */
154 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
155 unsigned long max_size = min_size;
156 if (nr)
157 max_size += (nr - 1)*8;
158 if (idx_size < min_size || idx_size > max_size) {
159 munmap(idx_map, idx_size);
160 return error("wrong index v2 file size in %s", path);
161 }
162 if (idx_size != min_size &&
163 /*
164 * make sure we can deal with large pack offsets.
165 * 31-bit signed offset won't be enough, neither
166 * 32-bit unsigned one will be.
167 */
168 (sizeof(off_t) <= 4)) {
169 munmap(idx_map, idx_size);
170 return error("pack too large for current definition of off_t in %s", path);
171 }
172 }
173
174 p->index_version = version;
175 p->index_data = idx_map;
176 p->index_size = idx_size;
177 p->num_objects = nr;
178 return 0;
179}
180
181int open_pack_index(struct packed_git *p)
182{
183 char *idx_name;
184 size_t len;
185 int ret;
186
187 if (p->index_data)
188 return 0;
189
190 if (!strip_suffix(p->pack_name, ".pack", &len))
191 die("BUG: pack_name does not end in .pack");
192 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
193 ret = check_packed_git_idx(idx_name, p);
194 free(idx_name);
195 return ret;
196}
197
198static struct packed_git *alloc_packed_git(int extra)
199{
200 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
201 memset(p, 0, sizeof(*p));
202 p->pack_fd = -1;
203 return p;
204}
205
206struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
207{
208 const char *path = sha1_pack_name(sha1);
209 size_t alloc = st_add(strlen(path), 1);
210 struct packed_git *p = alloc_packed_git(alloc);
211
212 memcpy(p->pack_name, path, alloc); /* includes NUL */
213 hashcpy(p->sha1, sha1);
214 if (check_packed_git_idx(idx_path, p)) {
215 free(p);
216 return NULL;
217 }
218
219 return p;
220}
f0e17e86
JT
221
222static void scan_windows(struct packed_git *p,
223 struct packed_git **lru_p,
224 struct pack_window **lru_w,
225 struct pack_window **lru_l)
226{
227 struct pack_window *w, *w_l;
228
229 for (w_l = NULL, w = p->windows; w; w = w->next) {
230 if (!w->inuse_cnt) {
231 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
232 *lru_p = p;
233 *lru_w = w;
234 *lru_l = w_l;
235 }
236 }
237 w_l = w;
238 }
239}
240
84f80ad5 241static int unuse_one_window(struct packed_git *current)
f0e17e86
JT
242{
243 struct packed_git *p, *lru_p = NULL;
244 struct pack_window *lru_w = NULL, *lru_l = NULL;
245
246 if (current)
247 scan_windows(current, &lru_p, &lru_w, &lru_l);
248 for (p = packed_git; p; p = p->next)
249 scan_windows(p, &lru_p, &lru_w, &lru_l);
250 if (lru_p) {
251 munmap(lru_w->base, lru_w->len);
252 pack_mapped -= lru_w->len;
253 if (lru_l)
254 lru_l->next = lru_w->next;
255 else
256 lru_p->windows = lru_w->next;
257 free(lru_w);
258 pack_open_windows--;
259 return 1;
260 }
261 return 0;
262}
263
264void release_pack_memory(size_t need)
265{
266 size_t cur = pack_mapped;
267 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
268 ; /* nothing */
269}
3836d88a
JT
270
271void close_pack_windows(struct packed_git *p)
272{
273 while (p->windows) {
274 struct pack_window *w = p->windows;
275
276 if (w->inuse_cnt)
277 die("pack '%s' still has open windows to it",
278 p->pack_name);
279 munmap(w->base, w->len);
280 pack_mapped -= w->len;
281 pack_open_windows--;
282 p->windows = w->next;
283 free(w);
284 }
285}
286
84f80ad5 287static int close_pack_fd(struct packed_git *p)
3836d88a
JT
288{
289 if (p->pack_fd < 0)
290 return 0;
291
292 close(p->pack_fd);
293 pack_open_fds--;
294 p->pack_fd = -1;
295
296 return 1;
297}
298
299void close_pack_index(struct packed_git *p)
300{
301 if (p->index_data) {
302 munmap((void *)p->index_data, p->index_size);
303 p->index_data = NULL;
304 }
305}
306
307static void close_pack(struct packed_git *p)
308{
309 close_pack_windows(p);
310 close_pack_fd(p);
311 close_pack_index(p);
312}
313
314void close_all_packs(void)
315{
316 struct packed_git *p;
317
318 for (p = packed_git; p; p = p->next)
319 if (p->do_not_close)
320 die("BUG: want to close pack marked 'do-not-close'");
321 else
322 close_pack(p);
323}
84f80ad5
JT
324
325/*
326 * The LRU pack is the one with the oldest MRU window, preferring packs
327 * with no used windows, or the oldest mtime if it has no windows allocated.
328 */
329static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
330{
331 struct pack_window *w, *this_mru_w;
332 int has_windows_inuse = 0;
333
334 /*
335 * Reject this pack if it has windows and the previously selected
336 * one does not. If this pack does not have windows, reject
337 * it if the pack file is newer than the previously selected one.
338 */
339 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
340 return;
341
342 for (w = this_mru_w = p->windows; w; w = w->next) {
343 /*
344 * Reject this pack if any of its windows are in use,
345 * but the previously selected pack did not have any
346 * inuse windows. Otherwise, record that this pack
347 * has windows in use.
348 */
349 if (w->inuse_cnt) {
350 if (*accept_windows_inuse)
351 has_windows_inuse = 1;
352 else
353 return;
354 }
355
356 if (w->last_used > this_mru_w->last_used)
357 this_mru_w = w;
358
359 /*
360 * Reject this pack if it has windows that have been
361 * used more recently than the previously selected pack.
362 * If the previously selected pack had windows inuse and
363 * we have not encountered a window in this pack that is
364 * inuse, skip this check since we prefer a pack with no
365 * inuse windows to one that has inuse windows.
366 */
367 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
368 this_mru_w->last_used > (*mru_w)->last_used)
369 return;
370 }
371
372 /*
373 * Select this pack.
374 */
375 *mru_w = this_mru_w;
376 *lru_p = p;
377 *accept_windows_inuse = has_windows_inuse;
378}
379
380static int close_one_pack(void)
381{
382 struct packed_git *p, *lru_p = NULL;
383 struct pack_window *mru_w = NULL;
384 int accept_windows_inuse = 1;
385
386 for (p = packed_git; p; p = p->next) {
387 if (p->pack_fd == -1)
388 continue;
389 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
390 }
391
392 if (lru_p)
393 return close_pack_fd(lru_p);
394
395 return 0;
396}
397
398static unsigned int get_max_fd_limit(void)
399{
400#ifdef RLIMIT_NOFILE
401 {
402 struct rlimit lim;
403
404 if (!getrlimit(RLIMIT_NOFILE, &lim))
405 return lim.rlim_cur;
406 }
407#endif
408
409#ifdef _SC_OPEN_MAX
410 {
411 long open_max = sysconf(_SC_OPEN_MAX);
412 if (0 < open_max)
413 return open_max;
414 /*
415 * Otherwise, we got -1 for one of the two
416 * reasons:
417 *
418 * (1) sysconf() did not understand _SC_OPEN_MAX
419 * and signaled an error with -1; or
420 * (2) sysconf() said there is no limit.
421 *
422 * We _could_ clear errno before calling sysconf() to
423 * tell these two cases apart and return a huge number
424 * in the latter case to let the caller cap it to a
425 * value that is not so selfish, but letting the
426 * fallback OPEN_MAX codepath take care of these cases
427 * is a lot simpler.
428 */
429 }
430#endif
431
432#ifdef OPEN_MAX
433 return OPEN_MAX;
434#else
435 return 1; /* see the caller ;-) */
436#endif
437}
438
439/*
440 * Do not call this directly as this leaks p->pack_fd on error return;
441 * call open_packed_git() instead.
442 */
443static int open_packed_git_1(struct packed_git *p)
444{
445 struct stat st;
446 struct pack_header hdr;
447 unsigned char sha1[20];
448 unsigned char *idx_sha1;
449 long fd_flag;
41dcc4dc 450 ssize_t read_result;
84f80ad5
JT
451
452 if (!p->index_data && open_pack_index(p))
453 return error("packfile %s index unavailable", p->pack_name);
454
455 if (!pack_max_fds) {
456 unsigned int max_fds = get_max_fd_limit();
457
458 /* Save 3 for stdin/stdout/stderr, 22 for work */
459 if (25 < max_fds)
460 pack_max_fds = max_fds - 25;
461 else
462 pack_max_fds = 1;
463 }
464
465 while (pack_max_fds <= pack_open_fds && close_one_pack())
466 ; /* nothing */
467
468 p->pack_fd = git_open(p->pack_name);
469 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
470 return -1;
471 pack_open_fds++;
472
473 /* If we created the struct before we had the pack we lack size. */
474 if (!p->pack_size) {
475 if (!S_ISREG(st.st_mode))
476 return error("packfile %s not a regular file", p->pack_name);
477 p->pack_size = st.st_size;
478 } else if (p->pack_size != st.st_size)
479 return error("packfile %s size changed", p->pack_name);
480
481 /* We leave these file descriptors open with sliding mmap;
482 * there is no point keeping them open across exec(), though.
483 */
484 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
485 if (fd_flag < 0)
486 return error("cannot determine file descriptor flags");
487 fd_flag |= FD_CLOEXEC;
488 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
489 return error("cannot set FD_CLOEXEC");
490
491 /* Verify we recognize this pack file format. */
41dcc4dc
JK
492 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
493 if (read_result < 0)
494 return error_errno("error reading from %s", p->pack_name);
495 if (read_result != sizeof(hdr))
84f80ad5
JT
496 return error("file %s is far too short to be a packfile", p->pack_name);
497 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
498 return error("file %s is not a GIT packfile", p->pack_name);
499 if (!pack_version_ok(hdr.hdr_version))
500 return error("packfile %s is version %"PRIu32" and not"
501 " supported (try upgrading GIT to a newer version)",
502 p->pack_name, ntohl(hdr.hdr_version));
503
504 /* Verify the pack matches its index. */
505 if (p->num_objects != ntohl(hdr.hdr_entries))
506 return error("packfile %s claims to have %"PRIu32" objects"
507 " while index indicates %"PRIu32" objects",
508 p->pack_name, ntohl(hdr.hdr_entries),
509 p->num_objects);
510 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
511 return error("end of packfile %s is unavailable", p->pack_name);
41dcc4dc
JK
512 read_result = read_in_full(p->pack_fd, sha1, sizeof(sha1));
513 if (read_result < 0)
514 return error_errno("error reading from %s", p->pack_name);
515 if (read_result != sizeof(sha1))
84f80ad5
JT
516 return error("packfile %s signature is unavailable", p->pack_name);
517 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
518 if (hashcmp(sha1, idx_sha1))
519 return error("packfile %s does not match index", p->pack_name);
520 return 0;
521}
522
a2551953 523static int open_packed_git(struct packed_git *p)
84f80ad5
JT
524{
525 if (!open_packed_git_1(p))
526 return 0;
527 close_pack_fd(p);
528 return -1;
529}
530
531static int in_window(struct pack_window *win, off_t offset)
532{
533 /* We must promise at least 20 bytes (one hash) after the
534 * offset is available from this window, otherwise the offset
535 * is not actually in this window and a different window (which
536 * has that one hash excess) must be used. This is to support
537 * the object header and delta base parsing routines below.
538 */
539 off_t win_off = win->offset;
540 return win_off <= offset
541 && (offset + 20) <= (win_off + win->len);
542}
543
544unsigned char *use_pack(struct packed_git *p,
545 struct pack_window **w_cursor,
546 off_t offset,
547 unsigned long *left)
548{
549 struct pack_window *win = *w_cursor;
550
551 /* Since packfiles end in a hash of their content and it's
552 * pointless to ask for an offset into the middle of that
553 * hash, and the in_window function above wouldn't match
554 * don't allow an offset too close to the end of the file.
555 */
556 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
557 die("packfile %s cannot be accessed", p->pack_name);
558 if (offset > (p->pack_size - 20))
559 die("offset beyond end of packfile (truncated pack?)");
560 if (offset < 0)
561 die(_("offset before end of packfile (broken .idx?)"));
562
563 if (!win || !in_window(win, offset)) {
564 if (win)
565 win->inuse_cnt--;
566 for (win = p->windows; win; win = win->next) {
567 if (in_window(win, offset))
568 break;
569 }
570 if (!win) {
571 size_t window_align = packed_git_window_size / 2;
572 off_t len;
573
574 if (p->pack_fd == -1 && open_packed_git(p))
575 die("packfile %s cannot be accessed", p->pack_name);
576
577 win = xcalloc(1, sizeof(*win));
578 win->offset = (offset / window_align) * window_align;
579 len = p->pack_size - win->offset;
580 if (len > packed_git_window_size)
581 len = packed_git_window_size;
582 win->len = (size_t)len;
583 pack_mapped += win->len;
584 while (packed_git_limit < pack_mapped
585 && unuse_one_window(p))
586 ; /* nothing */
587 win->base = xmmap(NULL, win->len,
588 PROT_READ, MAP_PRIVATE,
589 p->pack_fd, win->offset);
590 if (win->base == MAP_FAILED)
591 die_errno("packfile %s cannot be mapped",
592 p->pack_name);
593 if (!win->offset && win->len == p->pack_size
594 && !p->do_not_close)
595 close_pack_fd(p);
596 pack_mmap_calls++;
597 pack_open_windows++;
598 if (pack_mapped > peak_pack_mapped)
599 peak_pack_mapped = pack_mapped;
600 if (pack_open_windows > peak_pack_open_windows)
601 peak_pack_open_windows = pack_open_windows;
602 win->next = p->windows;
603 p->windows = win;
604 }
605 }
606 if (win != *w_cursor) {
607 win->last_used = pack_used_ctr++;
608 win->inuse_cnt++;
609 *w_cursor = win;
610 }
611 offset -= win->offset;
612 if (left)
613 *left = win->len - xsize_t(offset);
614 return win->base + offset;
615}
97de1803
JT
616
617void unuse_pack(struct pack_window **w_cursor)
618{
619 struct pack_window *w = *w_cursor;
620 if (w) {
621 w->inuse_cnt--;
622 *w_cursor = NULL;
623 }
624}
9a428653
JT
625
626static void try_to_free_pack_memory(size_t size)
627{
628 release_pack_memory(size);
629}
630
631struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
632{
633 static int have_set_try_to_free_routine;
634 struct stat st;
635 size_t alloc;
636 struct packed_git *p;
637
638 if (!have_set_try_to_free_routine) {
639 have_set_try_to_free_routine = 1;
640 set_try_to_free_routine(try_to_free_pack_memory);
641 }
642
643 /*
644 * Make sure a corresponding .pack file exists and that
645 * the index looks sane.
646 */
647 if (!strip_suffix_mem(path, &path_len, ".idx"))
648 return NULL;
649
650 /*
498f1f61 651 * ".promisor" is long enough to hold any suffix we're adding (and
9a428653
JT
652 * the use xsnprintf double-checks that)
653 */
498f1f61 654 alloc = st_add3(path_len, strlen(".promisor"), 1);
9a428653
JT
655 p = alloc_packed_git(alloc);
656 memcpy(p->pack_name, path, path_len);
657
658 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
659 if (!access(p->pack_name, F_OK))
660 p->pack_keep = 1;
661
498f1f61
JT
662 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
663 if (!access(p->pack_name, F_OK))
664 p->pack_promisor = 1;
665
9a428653
JT
666 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
667 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
668 free(p);
669 return NULL;
670 }
671
672 /* ok, it looks sane as far as we can check without
673 * actually mapping the pack file.
674 */
675 p->pack_size = st.st_size;
676 p->pack_local = local;
677 p->mtime = st.st_mtime;
678 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
679 hashclr(p->sha1);
680 return p;
681}
e65f1862
JT
682
683void install_packed_git(struct packed_git *pack)
684{
685 if (pack->pack_fd != -1)
686 pack_open_fds++;
687
688 pack->next = packed_git;
689 packed_git = pack;
690}
0abe14f6
JT
691
692void (*report_garbage)(unsigned seen_bits, const char *path);
693
694static void report_helper(const struct string_list *list,
695 int seen_bits, int first, int last)
696{
697 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
698 return;
699
700 for (; first < last; first++)
701 report_garbage(seen_bits, list->items[first].string);
702}
703
704static void report_pack_garbage(struct string_list *list)
705{
706 int i, baselen = -1, first = 0, seen_bits = 0;
707
708 if (!report_garbage)
709 return;
710
711 string_list_sort(list);
712
713 for (i = 0; i < list->nr; i++) {
714 const char *path = list->items[i].string;
715 if (baselen != -1 &&
716 strncmp(path, list->items[first].string, baselen)) {
717 report_helper(list, seen_bits, first, i);
718 baselen = -1;
719 seen_bits = 0;
720 }
721 if (baselen == -1) {
722 const char *dot = strrchr(path, '.');
723 if (!dot) {
724 report_garbage(PACKDIR_FILE_GARBAGE, path);
725 continue;
726 }
727 baselen = dot - path + 1;
728 first = i;
729 }
730 if (!strcmp(path + baselen, "pack"))
731 seen_bits |= 1;
732 else if (!strcmp(path + baselen, "idx"))
733 seen_bits |= 2;
734 }
735 report_helper(list, seen_bits, first, list->nr);
736}
737
738static void prepare_packed_git_one(char *objdir, int local)
739{
740 struct strbuf path = STRBUF_INIT;
741 size_t dirnamelen;
742 DIR *dir;
743 struct dirent *de;
744 struct string_list garbage = STRING_LIST_INIT_DUP;
745
746 strbuf_addstr(&path, objdir);
747 strbuf_addstr(&path, "/pack");
748 dir = opendir(path.buf);
749 if (!dir) {
750 if (errno != ENOENT)
751 error_errno("unable to open object pack directory: %s",
752 path.buf);
753 strbuf_release(&path);
754 return;
755 }
756 strbuf_addch(&path, '/');
757 dirnamelen = path.len;
758 while ((de = readdir(dir)) != NULL) {
759 struct packed_git *p;
760 size_t base_len;
761
762 if (is_dot_or_dotdot(de->d_name))
763 continue;
764
765 strbuf_setlen(&path, dirnamelen);
766 strbuf_addstr(&path, de->d_name);
767
768 base_len = path.len;
769 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
770 /* Don't reopen a pack we already have. */
771 for (p = packed_git; p; p = p->next) {
772 size_t len;
773 if (strip_suffix(p->pack_name, ".pack", &len) &&
774 len == base_len &&
775 !memcmp(p->pack_name, path.buf, len))
776 break;
777 }
778 if (p == NULL &&
779 /*
780 * See if it really is a valid .idx file with
781 * corresponding .pack file that we can map.
782 */
783 (p = add_packed_git(path.buf, path.len, local)) != NULL)
784 install_packed_git(p);
785 }
786
787 if (!report_garbage)
788 continue;
789
790 if (ends_with(de->d_name, ".idx") ||
791 ends_with(de->d_name, ".pack") ||
792 ends_with(de->d_name, ".bitmap") ||
498f1f61
JT
793 ends_with(de->d_name, ".keep") ||
794 ends_with(de->d_name, ".promisor"))
0abe14f6
JT
795 string_list_append(&garbage, path.buf);
796 else
797 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
798 }
799 closedir(dir);
800 report_pack_garbage(&garbage);
801 string_list_clear(&garbage, 0);
802 strbuf_release(&path);
803}
804
805static int approximate_object_count_valid;
806
807/*
808 * Give a fast, rough count of the number of objects in the repository. This
809 * ignores loose objects completely. If you have a lot of them, then either
810 * you should repack because your performance will be awful, or they are
811 * all unreachable objects about to be pruned, in which case they're not really
812 * interesting as a measure of repo size in the first place.
813 */
814unsigned long approximate_object_count(void)
815{
816 static unsigned long count;
817 if (!approximate_object_count_valid) {
818 struct packed_git *p;
819
820 prepare_packed_git();
821 count = 0;
822 for (p = packed_git; p; p = p->next) {
823 if (open_pack_index(p))
824 continue;
825 count += p->num_objects;
826 }
827 }
828 return count;
829}
830
831static void *get_next_packed_git(const void *p)
832{
833 return ((const struct packed_git *)p)->next;
834}
835
836static void set_next_packed_git(void *p, void *next)
837{
838 ((struct packed_git *)p)->next = next;
839}
840
841static int sort_pack(const void *a_, const void *b_)
842{
843 const struct packed_git *a = a_;
844 const struct packed_git *b = b_;
845 int st;
846
847 /*
848 * Local packs tend to contain objects specific to our
849 * variant of the project than remote ones. In addition,
850 * remote ones could be on a network mounted filesystem.
851 * Favor local ones for these reasons.
852 */
853 st = a->pack_local - b->pack_local;
854 if (st)
855 return -st;
856
857 /*
858 * Younger packs tend to contain more recent objects,
859 * and more recent objects tend to get accessed more
860 * often.
861 */
862 if (a->mtime < b->mtime)
863 return 1;
864 else if (a->mtime == b->mtime)
865 return 0;
866 return -1;
867}
868
869static void rearrange_packed_git(void)
870{
871 packed_git = llist_mergesort(packed_git, get_next_packed_git,
872 set_next_packed_git, sort_pack);
873}
874
875static void prepare_packed_git_mru(void)
876{
877 struct packed_git *p;
878
ec2dd32c
GS
879 INIT_LIST_HEAD(&packed_git_mru);
880
0abe14f6 881 for (p = packed_git; p; p = p->next)
ec2dd32c 882 list_add_tail(&p->mru, &packed_git_mru);
0abe14f6
JT
883}
884
885static int prepare_packed_git_run_once = 0;
886void prepare_packed_git(void)
887{
888 struct alternate_object_database *alt;
889
890 if (prepare_packed_git_run_once)
891 return;
892 prepare_packed_git_one(get_object_directory(), 1);
893 prepare_alt_odb();
894 for (alt = alt_odb_list; alt; alt = alt->next)
895 prepare_packed_git_one(alt->path, 0);
896 rearrange_packed_git();
897 prepare_packed_git_mru();
898 prepare_packed_git_run_once = 1;
899}
900
901void reprepare_packed_git(void)
902{
903 approximate_object_count_valid = 0;
904 prepare_packed_git_run_once = 0;
905 prepare_packed_git();
906}
32b42e15
JT
907
908unsigned long unpack_object_header_buffer(const unsigned char *buf,
909 unsigned long len, enum object_type *type, unsigned long *sizep)
910{
911 unsigned shift;
912 unsigned long size, c;
913 unsigned long used = 0;
914
915 c = buf[used++];
916 *type = (c >> 4) & 7;
917 size = c & 15;
918 shift = 4;
919 while (c & 0x80) {
920 if (len <= used || bitsizeof(long) <= shift) {
921 error("bad object header");
922 size = used = 0;
923 break;
924 }
925 c = buf[used++];
926 size += (c & 0x7f) << shift;
927 shift += 7;
928 }
929 *sizep = size;
930 return used;
931}
7b3aa75d
JT
932
933unsigned long get_size_from_delta(struct packed_git *p,
934 struct pack_window **w_curs,
935 off_t curpos)
936{
937 const unsigned char *data;
938 unsigned char delta_head[20], *in;
939 git_zstream stream;
940 int st;
941
942 memset(&stream, 0, sizeof(stream));
943 stream.next_out = delta_head;
944 stream.avail_out = sizeof(delta_head);
945
946 git_inflate_init(&stream);
947 do {
948 in = use_pack(p, w_curs, curpos, &stream.avail_in);
949 stream.next_in = in;
950 st = git_inflate(&stream, Z_FINISH);
951 curpos += stream.next_in - in;
952 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
953 stream.total_out < sizeof(delta_head));
954 git_inflate_end(&stream);
955 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
956 error("delta data unpack-initial failed");
957 return 0;
958 }
959
960 /* Examine the initial part of the delta to figure out
961 * the result size.
962 */
963 data = delta_head;
964
965 /* ignore base size */
966 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
967
968 /* Read the result size */
969 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
970}
3588dd6e
JT
971
972int unpack_object_header(struct packed_git *p,
973 struct pack_window **w_curs,
974 off_t *curpos,
975 unsigned long *sizep)
976{
977 unsigned char *base;
978 unsigned long left;
979 unsigned long used;
980 enum object_type type;
981
982 /* use_pack() assures us we have [base, base + 20) available
983 * as a range that we can look at. (Its actually the hash
984 * size that is assured.) With our object header encoding
985 * the maximum deflated object size is 2^137, which is just
986 * insane, so we know won't exceed what we have been given.
987 */
988 base = use_pack(p, w_curs, *curpos, &left);
989 used = unpack_object_header_buffer(base, left, &type, sizep);
990 if (!used) {
991 type = OBJ_BAD;
992 } else
993 *curpos += used;
994
995 return type;
996}
f1d8130b
JT
997
998void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
999{
1000 unsigned i;
1001 for (i = 0; i < p->num_bad_objects; i++)
1002 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1003 return;
1004 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1005 st_mult(GIT_MAX_RAWSZ,
1006 st_add(p->num_bad_objects, 1)));
1007 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1008 p->num_bad_objects++;
1009}
1010
1011const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1012{
1013 struct packed_git *p;
1014 unsigned i;
1015
1016 for (p = packed_git; p; p = p->next)
1017 for (i = 0; i < p->num_bad_objects; i++)
1018 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1019 return p;
1020 return NULL;
1021}
1022
1023static off_t get_delta_base(struct packed_git *p,
1024 struct pack_window **w_curs,
1025 off_t *curpos,
1026 enum object_type type,
1027 off_t delta_obj_offset)
1028{
1029 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1030 off_t base_offset;
1031
1032 /* use_pack() assured us we have [base_info, base_info + 20)
1033 * as a range that we can look at without walking off the
1034 * end of the mapped window. Its actually the hash size
1035 * that is assured. An OFS_DELTA longer than the hash size
1036 * is stupid, as then a REF_DELTA would be smaller to store.
1037 */
1038 if (type == OBJ_OFS_DELTA) {
1039 unsigned used = 0;
1040 unsigned char c = base_info[used++];
1041 base_offset = c & 127;
1042 while (c & 128) {
1043 base_offset += 1;
1044 if (!base_offset || MSB(base_offset, 7))
1045 return 0; /* overflow */
1046 c = base_info[used++];
1047 base_offset = (base_offset << 7) + (c & 127);
1048 }
1049 base_offset = delta_obj_offset - base_offset;
1050 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1051 return 0; /* out of bound */
1052 *curpos += used;
1053 } else if (type == OBJ_REF_DELTA) {
1054 /* The base entry _must_ be in the same pack */
1055 base_offset = find_pack_entry_one(base_info, p);
1056 *curpos += 20;
1057 } else
1058 die("I am totally screwed");
1059 return base_offset;
1060}
1061
1062/*
1063 * Like get_delta_base above, but we return the sha1 instead of the pack
1064 * offset. This means it is cheaper for REF deltas (we do not have to do
1065 * the final object lookup), but more expensive for OFS deltas (we
1066 * have to load the revidx to convert the offset back into a sha1).
1067 */
1068static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1069 struct pack_window **w_curs,
1070 off_t curpos,
1071 enum object_type type,
1072 off_t delta_obj_offset)
1073{
1074 if (type == OBJ_REF_DELTA) {
1075 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1076 return base;
1077 } else if (type == OBJ_OFS_DELTA) {
1078 struct revindex_entry *revidx;
1079 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1080 type, delta_obj_offset);
1081
1082 if (!base_offset)
1083 return NULL;
1084
1085 revidx = find_pack_revindex(p, base_offset);
1086 if (!revidx)
1087 return NULL;
1088
1089 return nth_packed_object_sha1(p, revidx->nr);
1090 } else
1091 return NULL;
1092}
1093
1094static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1095{
1096 int type;
1097 struct revindex_entry *revidx;
d169d664 1098 struct object_id oid;
f1d8130b
JT
1099 revidx = find_pack_revindex(p, obj_offset);
1100 if (!revidx)
1101 return OBJ_BAD;
d169d664 1102 nth_packed_object_oid(&oid, p, revidx->nr);
1103 mark_bad_packed_object(p, oid.hash);
abef9020 1104 type = oid_object_info(&oid, NULL);
f1d8130b
JT
1105 if (type <= OBJ_NONE)
1106 return OBJ_BAD;
1107 return type;
1108}
1109
1110#define POI_STACK_PREALLOC 64
1111
1112static enum object_type packed_to_object_type(struct packed_git *p,
1113 off_t obj_offset,
1114 enum object_type type,
1115 struct pack_window **w_curs,
1116 off_t curpos)
1117{
1118 off_t small_poi_stack[POI_STACK_PREALLOC];
1119 off_t *poi_stack = small_poi_stack;
1120 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1121
1122 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1123 off_t base_offset;
1124 unsigned long size;
1125 /* Push the object we're going to leave behind */
1126 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1127 poi_stack_alloc = alloc_nr(poi_stack_nr);
1128 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1129 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1130 } else {
1131 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1132 }
1133 poi_stack[poi_stack_nr++] = obj_offset;
1134 /* If parsing the base offset fails, just unwind */
1135 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1136 if (!base_offset)
1137 goto unwind;
1138 curpos = obj_offset = base_offset;
1139 type = unpack_object_header(p, w_curs, &curpos, &size);
1140 if (type <= OBJ_NONE) {
1141 /* If getting the base itself fails, we first
1142 * retry the base, otherwise unwind */
1143 type = retry_bad_packed_offset(p, base_offset);
1144 if (type > OBJ_NONE)
1145 goto out;
1146 goto unwind;
1147 }
1148 }
1149
1150 switch (type) {
1151 case OBJ_BAD:
1152 case OBJ_COMMIT:
1153 case OBJ_TREE:
1154 case OBJ_BLOB:
1155 case OBJ_TAG:
1156 break;
1157 default:
1158 error("unknown object type %i at offset %"PRIuMAX" in %s",
1159 type, (uintmax_t)obj_offset, p->pack_name);
1160 type = OBJ_BAD;
1161 }
1162
1163out:
1164 if (poi_stack != small_poi_stack)
1165 free(poi_stack);
1166 return type;
1167
1168unwind:
1169 while (poi_stack_nr) {
1170 obj_offset = poi_stack[--poi_stack_nr];
1171 type = retry_bad_packed_offset(p, obj_offset);
1172 if (type > OBJ_NONE)
1173 goto out;
1174 }
1175 type = OBJ_BAD;
1176 goto out;
1177}
1178
1179static struct hashmap delta_base_cache;
1180static size_t delta_base_cached;
1181
1182static LIST_HEAD(delta_base_cache_lru);
1183
1184struct delta_base_cache_key {
1185 struct packed_git *p;
1186 off_t base_offset;
1187};
1188
1189struct delta_base_cache_entry {
1190 struct hashmap hash;
1191 struct delta_base_cache_key key;
1192 struct list_head lru;
1193 void *data;
1194 unsigned long size;
1195 enum object_type type;
1196};
1197
1198static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1199{
1200 unsigned int hash;
1201
1202 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1203 hash += (hash >> 8) + (hash >> 16);
1204 return hash;
1205}
1206
1207static struct delta_base_cache_entry *
1208get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1209{
1210 struct hashmap_entry entry;
1211 struct delta_base_cache_key key;
1212
1213 if (!delta_base_cache.cmpfn)
1214 return NULL;
1215
1216 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1217 key.p = p;
1218 key.base_offset = base_offset;
1219 return hashmap_get(&delta_base_cache, &entry, &key);
1220}
1221
1222static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1223 const struct delta_base_cache_key *b)
1224{
1225 return a->p == b->p && a->base_offset == b->base_offset;
1226}
1227
1228static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1229 const void *va, const void *vb,
1230 const void *vkey)
1231{
1232 const struct delta_base_cache_entry *a = va, *b = vb;
1233 const struct delta_base_cache_key *key = vkey;
1234 if (key)
1235 return !delta_base_cache_key_eq(&a->key, key);
1236 else
1237 return !delta_base_cache_key_eq(&a->key, &b->key);
1238}
1239
1240static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1241{
1242 return !!get_delta_base_cache_entry(p, base_offset);
1243}
1244
1245/*
1246 * Remove the entry from the cache, but do _not_ free the associated
1247 * entry data. The caller takes ownership of the "data" buffer, and
1248 * should copy out any fields it wants before detaching.
1249 */
1250static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1251{
1252 hashmap_remove(&delta_base_cache, ent, &ent->key);
1253 list_del(&ent->lru);
1254 delta_base_cached -= ent->size;
1255 free(ent);
1256}
1257
1258static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1259 unsigned long *base_size, enum object_type *type)
1260{
1261 struct delta_base_cache_entry *ent;
1262
1263 ent = get_delta_base_cache_entry(p, base_offset);
1264 if (!ent)
1265 return unpack_entry(p, base_offset, type, base_size);
1266
1267 if (type)
1268 *type = ent->type;
1269 if (base_size)
1270 *base_size = ent->size;
1271 return xmemdupz(ent->data, ent->size);
1272}
1273
1274static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1275{
1276 free(ent->data);
1277 detach_delta_base_cache_entry(ent);
1278}
1279
1280void clear_delta_base_cache(void)
1281{
1282 struct list_head *lru, *tmp;
1283 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1284 struct delta_base_cache_entry *entry =
1285 list_entry(lru, struct delta_base_cache_entry, lru);
1286 release_delta_base_cache(entry);
1287 }
1288}
1289
1290static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1291 void *base, unsigned long base_size, enum object_type type)
1292{
1293 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1294 struct list_head *lru, *tmp;
1295
1296 delta_base_cached += base_size;
1297
1298 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1299 struct delta_base_cache_entry *f =
1300 list_entry(lru, struct delta_base_cache_entry, lru);
1301 if (delta_base_cached <= delta_base_cache_limit)
1302 break;
1303 release_delta_base_cache(f);
1304 }
1305
1306 ent->key.p = p;
1307 ent->key.base_offset = base_offset;
1308 ent->type = type;
1309 ent->data = base;
1310 ent->size = base_size;
1311 list_add_tail(&ent->lru, &delta_base_cache_lru);
1312
1313 if (!delta_base_cache.cmpfn)
1314 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1315 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1316 hashmap_add(&delta_base_cache, ent);
1317}
1318
1319int packed_object_info(struct packed_git *p, off_t obj_offset,
1320 struct object_info *oi)
1321{
1322 struct pack_window *w_curs = NULL;
1323 unsigned long size;
1324 off_t curpos = obj_offset;
1325 enum object_type type;
1326
1327 /*
1328 * We always get the representation type, but only convert it to
1329 * a "real" type later if the caller is interested.
1330 */
1331 if (oi->contentp) {
1332 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
1333 &type);
1334 if (!*oi->contentp)
1335 type = OBJ_BAD;
1336 } else {
1337 type = unpack_object_header(p, &w_curs, &curpos, &size);
1338 }
1339
1340 if (!oi->contentp && oi->sizep) {
1341 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1342 off_t tmp_pos = curpos;
1343 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1344 type, obj_offset);
1345 if (!base_offset) {
1346 type = OBJ_BAD;
1347 goto out;
1348 }
1349 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1350 if (*oi->sizep == 0) {
1351 type = OBJ_BAD;
1352 goto out;
1353 }
1354 } else {
1355 *oi->sizep = size;
1356 }
1357 }
1358
1359 if (oi->disk_sizep) {
1360 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1361 *oi->disk_sizep = revidx[1].offset - obj_offset;
1362 }
1363
6ca32f47 1364 if (oi->typep || oi->type_name) {
f1d8130b
JT
1365 enum object_type ptot;
1366 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
1367 curpos);
1368 if (oi->typep)
1369 *oi->typep = ptot;
6ca32f47 1370 if (oi->type_name) {
debca9d2 1371 const char *tn = type_name(ptot);
f1d8130b 1372 if (tn)
6ca32f47 1373 strbuf_addstr(oi->type_name, tn);
f1d8130b
JT
1374 }
1375 if (ptot < 0) {
1376 type = OBJ_BAD;
1377 goto out;
1378 }
1379 }
1380
1381 if (oi->delta_base_sha1) {
1382 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1383 const unsigned char *base;
1384
1385 base = get_delta_base_sha1(p, &w_curs, curpos,
1386 type, obj_offset);
1387 if (!base) {
1388 type = OBJ_BAD;
1389 goto out;
1390 }
1391
1392 hashcpy(oi->delta_base_sha1, base);
1393 } else
1394 hashclr(oi->delta_base_sha1);
1395 }
1396
1397 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1398 OI_PACKED;
1399
1400out:
1401 unuse_pack(&w_curs);
1402 return type;
1403}
1404
1405static void *unpack_compressed_entry(struct packed_git *p,
1406 struct pack_window **w_curs,
1407 off_t curpos,
1408 unsigned long size)
1409{
1410 int st;
1411 git_zstream stream;
1412 unsigned char *buffer, *in;
1413
1414 buffer = xmallocz_gently(size);
1415 if (!buffer)
1416 return NULL;
1417 memset(&stream, 0, sizeof(stream));
1418 stream.next_out = buffer;
1419 stream.avail_out = size + 1;
1420
1421 git_inflate_init(&stream);
1422 do {
1423 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1424 stream.next_in = in;
1425 st = git_inflate(&stream, Z_FINISH);
1426 if (!stream.avail_out)
1427 break; /* the payload is larger than it should be */
1428 curpos += stream.next_in - in;
1429 } while (st == Z_OK || st == Z_BUF_ERROR);
1430 git_inflate_end(&stream);
1431 if ((st != Z_STREAM_END) || stream.total_out != size) {
1432 free(buffer);
1433 return NULL;
1434 }
1435
1436 return buffer;
1437}
1438
1439static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1440{
1441 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1442 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1443 p->pack_name, (uintmax_t)obj_offset);
1444}
1445
1446int do_check_packed_object_crc;
1447
1448#define UNPACK_ENTRY_STACK_PREALLOC 64
1449struct unpack_entry_stack_ent {
1450 off_t obj_offset;
1451 off_t curpos;
1452 unsigned long size;
1453};
1454
4310b0c4 1455static void *read_object(const struct object_id *oid, enum object_type *type,
f1d8130b
JT
1456 unsigned long *size)
1457{
1458 struct object_info oi = OBJECT_INFO_INIT;
1459 void *content;
1460 oi.typep = type;
1461 oi.sizep = size;
1462 oi.contentp = &content;
1463
abef9020 1464 if (oid_object_info_extended(oid, &oi, 0) < 0)
f1d8130b
JT
1465 return NULL;
1466 return content;
1467}
1468
1469void *unpack_entry(struct packed_git *p, off_t obj_offset,
1470 enum object_type *final_type, unsigned long *final_size)
1471{
1472 struct pack_window *w_curs = NULL;
1473 off_t curpos = obj_offset;
1474 void *data = NULL;
1475 unsigned long size;
1476 enum object_type type;
1477 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1478 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1479 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1480 int base_from_cache = 0;
1481
1482 write_pack_access_log(p, obj_offset);
1483
1484 /* PHASE 1: drill down to the innermost base object */
1485 for (;;) {
1486 off_t base_offset;
1487 int i;
1488 struct delta_base_cache_entry *ent;
1489
1490 ent = get_delta_base_cache_entry(p, curpos);
1491 if (ent) {
1492 type = ent->type;
1493 data = ent->data;
1494 size = ent->size;
1495 detach_delta_base_cache_entry(ent);
1496 base_from_cache = 1;
1497 break;
1498 }
1499
1500 if (do_check_packed_object_crc && p->index_version > 1) {
1501 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1502 off_t len = revidx[1].offset - obj_offset;
1503 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
4310b0c4 1504 struct object_id oid;
1505 nth_packed_object_oid(&oid, p, revidx->nr);
f1d8130b 1506 error("bad packed object CRC for %s",
4310b0c4 1507 oid_to_hex(&oid));
1508 mark_bad_packed_object(p, oid.hash);
f1d8130b
JT
1509 data = NULL;
1510 goto out;
1511 }
1512 }
1513
1514 type = unpack_object_header(p, &w_curs, &curpos, &size);
1515 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1516 break;
1517
1518 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1519 if (!base_offset) {
1520 error("failed to validate delta base reference "
1521 "at offset %"PRIuMAX" from %s",
1522 (uintmax_t)curpos, p->pack_name);
1523 /* bail to phase 2, in hopes of recovery */
1524 data = NULL;
1525 break;
1526 }
1527
1528 /* push object, proceed to base */
1529 if (delta_stack_nr >= delta_stack_alloc
1530 && delta_stack == small_delta_stack) {
1531 delta_stack_alloc = alloc_nr(delta_stack_nr);
1532 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1533 memcpy(delta_stack, small_delta_stack,
1534 sizeof(*delta_stack)*delta_stack_nr);
1535 } else {
1536 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1537 }
1538 i = delta_stack_nr++;
1539 delta_stack[i].obj_offset = obj_offset;
1540 delta_stack[i].curpos = curpos;
1541 delta_stack[i].size = size;
1542
1543 curpos = obj_offset = base_offset;
1544 }
1545
1546 /* PHASE 2: handle the base */
1547 switch (type) {
1548 case OBJ_OFS_DELTA:
1549 case OBJ_REF_DELTA:
1550 if (data)
1551 die("BUG: unpack_entry: left loop at a valid delta");
1552 break;
1553 case OBJ_COMMIT:
1554 case OBJ_TREE:
1555 case OBJ_BLOB:
1556 case OBJ_TAG:
1557 if (!base_from_cache)
1558 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1559 break;
1560 default:
1561 data = NULL;
1562 error("unknown object type %i at offset %"PRIuMAX" in %s",
1563 type, (uintmax_t)obj_offset, p->pack_name);
1564 }
1565
1566 /* PHASE 3: apply deltas in order */
1567
1568 /* invariants:
1569 * 'data' holds the base data, or NULL if there was corruption
1570 */
1571 while (delta_stack_nr) {
1572 void *delta_data;
1573 void *base = data;
1574 void *external_base = NULL;
1575 unsigned long delta_size, base_size = size;
1576 int i;
1577
1578 data = NULL;
1579
1580 if (base)
1581 add_delta_base_cache(p, obj_offset, base, base_size, type);
1582
1583 if (!base) {
1584 /*
1585 * We're probably in deep shit, but let's try to fetch
1586 * the required base anyway from another pack or loose.
1587 * This is costly but should happen only in the presence
1588 * of a corrupted pack, and is better than failing outright.
1589 */
1590 struct revindex_entry *revidx;
4310b0c4 1591 struct object_id base_oid;
f1d8130b
JT
1592 revidx = find_pack_revindex(p, obj_offset);
1593 if (revidx) {
4310b0c4 1594 nth_packed_object_oid(&base_oid, p, revidx->nr);
f1d8130b
JT
1595 error("failed to read delta base object %s"
1596 " at offset %"PRIuMAX" from %s",
4310b0c4 1597 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
f1d8130b 1598 p->pack_name);
4310b0c4 1599 mark_bad_packed_object(p, base_oid.hash);
1600 base = read_object(&base_oid, &type, &base_size);
f1d8130b
JT
1601 external_base = base;
1602 }
1603 }
1604
1605 i = --delta_stack_nr;
1606 obj_offset = delta_stack[i].obj_offset;
1607 curpos = delta_stack[i].curpos;
1608 delta_size = delta_stack[i].size;
1609
1610 if (!base)
1611 continue;
1612
1613 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1614
1615 if (!delta_data) {
1616 error("failed to unpack compressed delta "
1617 "at offset %"PRIuMAX" from %s",
1618 (uintmax_t)curpos, p->pack_name);
1619 data = NULL;
1620 free(external_base);
1621 continue;
1622 }
1623
1624 data = patch_delta(base, base_size,
1625 delta_data, delta_size,
1626 &size);
1627
1628 /*
1629 * We could not apply the delta; warn the user, but keep going.
1630 * Our failure will be noticed either in the next iteration of
1631 * the loop, or if this is the final delta, in the caller when
1632 * we return NULL. Those code paths will take care of making
1633 * a more explicit warning and retrying with another copy of
1634 * the object.
1635 */
1636 if (!data)
1637 error("failed to apply delta");
1638
1639 free(delta_data);
1640 free(external_base);
1641 }
1642
1643 if (final_type)
1644 *final_type = type;
1645 if (final_size)
1646 *final_size = size;
1647
1648out:
1649 unuse_pack(&w_curs);
1650
1651 if (delta_stack != small_delta_stack)
1652 free(delta_stack);
1653
1654 return data;
1655}
d5a16761
JT
1656
1657const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1658 uint32_t n)
1659{
1660 const unsigned char *index = p->index_data;
1661 if (!index) {
1662 if (open_pack_index(p))
1663 return NULL;
1664 index = p->index_data;
1665 }
1666 if (n >= p->num_objects)
1667 return NULL;
1668 index += 4 * 256;
1669 if (p->index_version == 1) {
1670 return index + 24 * n + 4;
1671 } else {
1672 index += 8;
1673 return index + 20 * n;
1674 }
1675}
1676
1677const struct object_id *nth_packed_object_oid(struct object_id *oid,
1678 struct packed_git *p,
1679 uint32_t n)
1680{
1681 const unsigned char *hash = nth_packed_object_sha1(p, n);
1682 if (!hash)
1683 return NULL;
1684 hashcpy(oid->hash, hash);
1685 return oid;
1686}
9e0f45f5
JT
1687
1688void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1689{
1690 const unsigned char *ptr = vptr;
1691 const unsigned char *start = p->index_data;
1692 const unsigned char *end = start + p->index_size;
1693 if (ptr < start)
1694 die(_("offset before start of pack index for %s (corrupt index?)"),
1695 p->pack_name);
1696 /* No need to check for underflow; .idx files must be at least 8 bytes */
1697 if (ptr >= end - 8)
1698 die(_("offset beyond end of pack index for %s (truncated index?)"),
1699 p->pack_name);
1700}
1701
1702off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1703{
1704 const unsigned char *index = p->index_data;
1705 index += 4 * 256;
1706 if (p->index_version == 1) {
1707 return ntohl(*((uint32_t *)(index + 24 * n)));
1708 } else {
1709 uint32_t off;
1710 index += 8 + p->num_objects * (20 + 4);
1711 off = ntohl(*((uint32_t *)(index + 4 * n)));
1712 if (!(off & 0x80000000))
1713 return off;
1714 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1715 check_pack_index_ptr(p, index);
ad622a25 1716 return get_be64(index);
9e0f45f5
JT
1717 }
1718}
a2551953
JT
1719
1720off_t find_pack_entry_one(const unsigned char *sha1,
1721 struct packed_git *p)
1722{
1723 const uint32_t *level1_ofs = p->index_data;
1724 const unsigned char *index = p->index_data;
b4e00f73
JT
1725 unsigned stride;
1726 uint32_t result;
a2551953
JT
1727
1728 if (!index) {
1729 if (open_pack_index(p))
1730 return 0;
1731 level1_ofs = p->index_data;
1732 index = p->index_data;
1733 }
1734 if (p->index_version > 1) {
1735 level1_ofs += 2;
1736 index += 8;
1737 }
1738 index += 4 * 256;
a2551953
JT
1739 if (p->index_version > 1) {
1740 stride = 20;
1741 } else {
1742 stride = 24;
1743 index += 4;
1744 }
1745
b4e00f73
JT
1746 if (bsearch_hash(sha1, level1_ofs, index, stride, &result))
1747 return nth_packed_object_offset(p, result);
a2551953
JT
1748 return 0;
1749}
1750
1751int is_pack_valid(struct packed_git *p)
1752{
1753 /* An already open pack is known to be valid. */
1754 if (p->pack_fd != -1)
1755 return 1;
1756
1757 /* If the pack has one window completely covering the
1758 * file size, the pack is known to be valid even if
1759 * the descriptor is not currently open.
1760 */
1761 if (p->windows) {
1762 struct pack_window *w = p->windows;
1763
1764 if (!w->offset && w->len == p->pack_size)
1765 return 1;
1766 }
1767
1768 /* Force the pack to open to prove its valid. */
1769 return !open_packed_git(p);
1770}
d6fe0036
JT
1771
1772struct packed_git *find_sha1_pack(const unsigned char *sha1,
1773 struct packed_git *packs)
1774{
1775 struct packed_git *p;
1776
1777 for (p = packs; p; p = p->next) {
1778 if (find_pack_entry_one(sha1, p))
1779 return p;
1780 }
1781 return NULL;
1782
1783}
1a1e5d4f
JT
1784
1785static int fill_pack_entry(const unsigned char *sha1,
1786 struct pack_entry *e,
1787 struct packed_git *p)
1788{
1789 off_t offset;
1790
1791 if (p->num_bad_objects) {
1792 unsigned i;
1793 for (i = 0; i < p->num_bad_objects; i++)
1794 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1795 return 0;
1796 }
1797
1798 offset = find_pack_entry_one(sha1, p);
1799 if (!offset)
1800 return 0;
1801
1802 /*
1803 * We are about to tell the caller where they can locate the
1804 * requested object. We better make sure the packfile is
1805 * still here and can be accessed before supplying that
1806 * answer, as it may have been deleted since the index was
1807 * loaded!
1808 */
1809 if (!is_pack_valid(p))
1810 return 0;
1811 e->offset = offset;
1812 e->p = p;
1813 hashcpy(e->sha1, sha1);
1814 return 1;
1815}
1816
1817/*
1818 * Iff a pack file contains the object named by sha1, return true and
1819 * store its location to e.
1820 */
1821int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1822{
8865859d 1823 struct list_head *pos;
1a1e5d4f
JT
1824
1825 prepare_packed_git();
1826 if (!packed_git)
1827 return 0;
1828
ec2dd32c
GS
1829 list_for_each(pos, &packed_git_mru) {
1830 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1831 if (fill_pack_entry(sha1, e, p)) {
1832 list_move(&p->mru, &packed_git_mru);
1a1e5d4f
JT
1833 return 1;
1834 }
1835 }
1836 return 0;
1837}
150e3001
JT
1838
1839int has_sha1_pack(const unsigned char *sha1)
1840{
1841 struct pack_entry e;
1842 return find_pack_entry(sha1, &e);
1843}
f9a8672a
JT
1844
1845int has_pack_index(const unsigned char *sha1)
1846{
1847 struct stat st;
1848 if (stat(sha1_pack_index_name(sha1), &st))
1849 return 0;
1850 return 1;
1851}
7709f468
JT
1852
1853static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
1854{
1855 uint32_t i;
1856 int r = 0;
1857
1858 for (i = 0; i < p->num_objects; i++) {
1859 struct object_id oid;
1860
1861 if (!nth_packed_object_oid(&oid, p, i))
1862 return error("unable to get sha1 of object %u in %s",
1863 i, p->pack_name);
1864
1865 r = cb(&oid, p, i, data);
1866 if (r)
1867 break;
1868 }
1869 return r;
1870}
1871
1872int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
1873{
1874 struct packed_git *p;
1875 int r = 0;
1876 int pack_errors = 0;
1877
1878 prepare_packed_git();
1879 for (p = packed_git; p; p = p->next) {
1880 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
1881 continue;
498f1f61
JT
1882 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
1883 !p->pack_promisor)
1884 continue;
7709f468
JT
1885 if (open_pack_index(p)) {
1886 pack_errors = 1;
1887 continue;
1888 }
1889 r = for_each_object_in_pack(p, cb, data);
1890 if (r)
1891 break;
1892 }
1893 return r ? r : pack_errors;
1894}
498f1f61
JT
1895
1896static int add_promisor_object(const struct object_id *oid,
1897 struct packed_git *pack,
1898 uint32_t pos,
1899 void *set_)
1900{
1901 struct oidset *set = set_;
1902 struct object *obj = parse_object(oid);
1903 if (!obj)
1904 return 1;
1905
1906 oidset_insert(set, oid);
1907
1908 /*
1909 * If this is a tree, commit, or tag, the objects it refers
1910 * to are also promisor objects. (Blobs refer to no objects.)
1911 */
1912 if (obj->type == OBJ_TREE) {
1913 struct tree *tree = (struct tree *)obj;
1914 struct tree_desc desc;
1915 struct name_entry entry;
1916 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
1917 /*
1918 * Error messages are given when packs are
1919 * verified, so do not print any here.
1920 */
1921 return 0;
1922 while (tree_entry_gently(&desc, &entry))
1923 oidset_insert(set, entry.oid);
1924 } else if (obj->type == OBJ_COMMIT) {
1925 struct commit *commit = (struct commit *) obj;
1926 struct commit_list *parents = commit->parents;
1927
1928 oidset_insert(set, &commit->tree->object.oid);
1929 for (; parents; parents = parents->next)
1930 oidset_insert(set, &parents->item->object.oid);
1931 } else if (obj->type == OBJ_TAG) {
1932 struct tag *tag = (struct tag *) obj;
1933 oidset_insert(set, &tag->tagged->oid);
1934 }
1935 return 0;
1936}
1937
1938int is_promisor_object(const struct object_id *oid)
1939{
1940 static struct oidset promisor_objects;
1941 static int promisor_objects_prepared;
1942
1943 if (!promisor_objects_prepared) {
1944 if (repository_format_partial_clone) {
1945 for_each_packed_object(add_promisor_object,
1946 &promisor_objects,
1947 FOR_EACH_OBJECT_PROMISOR_ONLY);
1948 }
1949 promisor_objects_prepared = 1;
1950 }
1951 return oidset_contains(&promisor_objects, oid);
1952}