]> git.ipfire.org Git - thirdparty/git.git/blob - builtin/repack.c
object-store-ll.h: split this header out of object-store.h
[thirdparty/git.git] / builtin / repack.c
1 #include "builtin.h"
2 #include "alloc.h"
3 #include "config.h"
4 #include "dir.h"
5 #include "environment.h"
6 #include "gettext.h"
7 #include "hex.h"
8 #include "parse-options.h"
9 #include "path.h"
10 #include "run-command.h"
11 #include "server-info.h"
12 #include "sigchain.h"
13 #include "strbuf.h"
14 #include "string-list.h"
15 #include "strvec.h"
16 #include "midx.h"
17 #include "packfile.h"
18 #include "prune-packed.h"
19 #include "object-store-ll.h"
20 #include "promisor-remote.h"
21 #include "shallow.h"
22 #include "pack.h"
23 #include "pack-bitmap.h"
24 #include "refs.h"
25
26 #define ALL_INTO_ONE 1
27 #define LOOSEN_UNREACHABLE 2
28 #define PACK_CRUFT 4
29
30 #define DELETE_PACK 1
31 #define CRUFT_PACK 2
32
33 static int pack_everything;
34 static int delta_base_offset = 1;
35 static int pack_kept_objects = -1;
36 static int write_bitmaps = -1;
37 static int use_delta_islands;
38 static int run_update_server_info = 1;
39 static char *packdir, *packtmp_name, *packtmp;
40
41 static const char *const git_repack_usage[] = {
42 N_("git repack [<options>]"),
43 NULL
44 };
45
46 static const char incremental_bitmap_conflict_error[] = N_(
47 "Incremental repacks are incompatible with bitmap indexes. Use\n"
48 "--no-write-bitmap-index or disable the pack.writeBitmaps configuration."
49 );
50
51 struct pack_objects_args {
52 const char *window;
53 const char *window_memory;
54 const char *depth;
55 const char *threads;
56 const char *max_pack_size;
57 int no_reuse_delta;
58 int no_reuse_object;
59 int quiet;
60 int local;
61 };
62
63 static int repack_config(const char *var, const char *value, void *cb)
64 {
65 struct pack_objects_args *cruft_po_args = cb;
66 if (!strcmp(var, "repack.usedeltabaseoffset")) {
67 delta_base_offset = git_config_bool(var, value);
68 return 0;
69 }
70 if (!strcmp(var, "repack.packkeptobjects")) {
71 pack_kept_objects = git_config_bool(var, value);
72 return 0;
73 }
74 if (!strcmp(var, "repack.writebitmaps") ||
75 !strcmp(var, "pack.writebitmaps")) {
76 write_bitmaps = git_config_bool(var, value);
77 return 0;
78 }
79 if (!strcmp(var, "repack.usedeltaislands")) {
80 use_delta_islands = git_config_bool(var, value);
81 return 0;
82 }
83 if (strcmp(var, "repack.updateserverinfo") == 0) {
84 run_update_server_info = git_config_bool(var, value);
85 return 0;
86 }
87 if (!strcmp(var, "repack.cruftwindow"))
88 return git_config_string(&cruft_po_args->window, var, value);
89 if (!strcmp(var, "repack.cruftwindowmemory"))
90 return git_config_string(&cruft_po_args->window_memory, var, value);
91 if (!strcmp(var, "repack.cruftdepth"))
92 return git_config_string(&cruft_po_args->depth, var, value);
93 if (!strcmp(var, "repack.cruftthreads"))
94 return git_config_string(&cruft_po_args->threads, var, value);
95 return git_default_config(var, value, cb);
96 }
97
98 /*
99 * Adds all packs hex strings to either fname_nonkept_list or
100 * fname_kept_list based on whether each pack has a corresponding
101 * .keep file or not. Packs without a .keep file are not to be kept
102 * if we are going to pack everything into one file.
103 */
104 static void collect_pack_filenames(struct string_list *fname_nonkept_list,
105 struct string_list *fname_kept_list,
106 const struct string_list *extra_keep)
107 {
108 DIR *dir;
109 struct dirent *e;
110 char *fname;
111
112 if (!(dir = opendir(packdir)))
113 return;
114
115 while ((e = readdir(dir)) != NULL) {
116 size_t len;
117 int i;
118
119 if (!strip_suffix(e->d_name, ".pack", &len))
120 continue;
121
122 for (i = 0; i < extra_keep->nr; i++)
123 if (!fspathcmp(e->d_name, extra_keep->items[i].string))
124 break;
125
126 fname = xmemdupz(e->d_name, len);
127
128 if ((extra_keep->nr > 0 && i < extra_keep->nr) ||
129 (file_exists(mkpath("%s/%s.keep", packdir, fname)))) {
130 string_list_append_nodup(fname_kept_list, fname);
131 } else {
132 struct string_list_item *item;
133 item = string_list_append_nodup(fname_nonkept_list,
134 fname);
135 if (file_exists(mkpath("%s/%s.mtimes", packdir, fname)))
136 item->util = (void*)(uintptr_t)CRUFT_PACK;
137 }
138 }
139 closedir(dir);
140
141 string_list_sort(fname_kept_list);
142 }
143
144 static void remove_redundant_pack(const char *dir_name, const char *base_name)
145 {
146 struct strbuf buf = STRBUF_INIT;
147 struct multi_pack_index *m = get_local_multi_pack_index(the_repository);
148 strbuf_addf(&buf, "%s.pack", base_name);
149 if (m && midx_contains_pack(m, buf.buf))
150 clear_midx_file(the_repository);
151 strbuf_insertf(&buf, 0, "%s/", dir_name);
152 unlink_pack_path(buf.buf, 1);
153 strbuf_release(&buf);
154 }
155
156 static void prepare_pack_objects(struct child_process *cmd,
157 const struct pack_objects_args *args,
158 const char *out)
159 {
160 strvec_push(&cmd->args, "pack-objects");
161 if (args->window)
162 strvec_pushf(&cmd->args, "--window=%s", args->window);
163 if (args->window_memory)
164 strvec_pushf(&cmd->args, "--window-memory=%s", args->window_memory);
165 if (args->depth)
166 strvec_pushf(&cmd->args, "--depth=%s", args->depth);
167 if (args->threads)
168 strvec_pushf(&cmd->args, "--threads=%s", args->threads);
169 if (args->max_pack_size)
170 strvec_pushf(&cmd->args, "--max-pack-size=%s", args->max_pack_size);
171 if (args->no_reuse_delta)
172 strvec_pushf(&cmd->args, "--no-reuse-delta");
173 if (args->no_reuse_object)
174 strvec_pushf(&cmd->args, "--no-reuse-object");
175 if (args->local)
176 strvec_push(&cmd->args, "--local");
177 if (args->quiet)
178 strvec_push(&cmd->args, "--quiet");
179 if (delta_base_offset)
180 strvec_push(&cmd->args, "--delta-base-offset");
181 strvec_push(&cmd->args, out);
182 cmd->git_cmd = 1;
183 cmd->out = -1;
184 }
185
186 /*
187 * Write oid to the given struct child_process's stdin, starting it first if
188 * necessary.
189 */
190 static int write_oid(const struct object_id *oid,
191 struct packed_git *pack UNUSED,
192 uint32_t pos UNUSED, void *data)
193 {
194 struct child_process *cmd = data;
195
196 if (cmd->in == -1) {
197 if (start_command(cmd))
198 die(_("could not start pack-objects to repack promisor objects"));
199 }
200
201 xwrite(cmd->in, oid_to_hex(oid), the_hash_algo->hexsz);
202 xwrite(cmd->in, "\n", 1);
203 return 0;
204 }
205
206 static struct {
207 const char *name;
208 unsigned optional:1;
209 } exts[] = {
210 {".pack"},
211 {".rev", 1},
212 {".mtimes", 1},
213 {".bitmap", 1},
214 {".promisor", 1},
215 {".idx"},
216 };
217
218 struct generated_pack_data {
219 struct tempfile *tempfiles[ARRAY_SIZE(exts)];
220 };
221
222 static struct generated_pack_data *populate_pack_exts(const char *name)
223 {
224 struct stat statbuf;
225 struct strbuf path = STRBUF_INIT;
226 struct generated_pack_data *data = xcalloc(1, sizeof(*data));
227 int i;
228
229 for (i = 0; i < ARRAY_SIZE(exts); i++) {
230 strbuf_reset(&path);
231 strbuf_addf(&path, "%s-%s%s", packtmp, name, exts[i].name);
232
233 if (stat(path.buf, &statbuf))
234 continue;
235
236 data->tempfiles[i] = register_tempfile(path.buf);
237 }
238
239 strbuf_release(&path);
240 return data;
241 }
242
243 static void repack_promisor_objects(const struct pack_objects_args *args,
244 struct string_list *names)
245 {
246 struct child_process cmd = CHILD_PROCESS_INIT;
247 FILE *out;
248 struct strbuf line = STRBUF_INIT;
249
250 prepare_pack_objects(&cmd, args, packtmp);
251 cmd.in = -1;
252
253 /*
254 * NEEDSWORK: Giving pack-objects only the OIDs without any ordering
255 * hints may result in suboptimal deltas in the resulting pack. See if
256 * the OIDs can be sent with fake paths such that pack-objects can use a
257 * {type -> existing pack order} ordering when computing deltas instead
258 * of a {type -> size} ordering, which may produce better deltas.
259 */
260 for_each_packed_object(write_oid, &cmd,
261 FOR_EACH_OBJECT_PROMISOR_ONLY);
262
263 if (cmd.in == -1) {
264 /* No packed objects; cmd was never started */
265 child_process_clear(&cmd);
266 return;
267 }
268
269 close(cmd.in);
270
271 out = xfdopen(cmd.out, "r");
272 while (strbuf_getline_lf(&line, out) != EOF) {
273 struct string_list_item *item;
274 char *promisor_name;
275
276 if (line.len != the_hash_algo->hexsz)
277 die(_("repack: Expecting full hex object ID lines only from pack-objects."));
278 item = string_list_append(names, line.buf);
279
280 /*
281 * pack-objects creates the .pack and .idx files, but not the
282 * .promisor file. Create the .promisor file, which is empty.
283 *
284 * NEEDSWORK: fetch-pack sometimes generates non-empty
285 * .promisor files containing the ref names and associated
286 * hashes at the point of generation of the corresponding
287 * packfile, but this would not preserve their contents. Maybe
288 * concatenate the contents of all .promisor files instead of
289 * just creating a new empty file.
290 */
291 promisor_name = mkpathdup("%s-%s.promisor", packtmp,
292 line.buf);
293 write_promisor_file(promisor_name, NULL, 0);
294
295 item->util = populate_pack_exts(item->string);
296
297 free(promisor_name);
298 }
299 fclose(out);
300 if (finish_command(&cmd))
301 die(_("could not finish pack-objects to repack promisor objects"));
302 }
303
304 struct pack_geometry {
305 struct packed_git **pack;
306 uint32_t pack_nr, pack_alloc;
307 uint32_t split;
308 };
309
310 static uint32_t geometry_pack_weight(struct packed_git *p)
311 {
312 if (open_pack_index(p))
313 die(_("cannot open index for %s"), p->pack_name);
314 return p->num_objects;
315 }
316
317 static int geometry_cmp(const void *va, const void *vb)
318 {
319 uint32_t aw = geometry_pack_weight(*(struct packed_git **)va),
320 bw = geometry_pack_weight(*(struct packed_git **)vb);
321
322 if (aw < bw)
323 return -1;
324 if (aw > bw)
325 return 1;
326 return 0;
327 }
328
329 static void init_pack_geometry(struct pack_geometry **geometry_p,
330 struct string_list *existing_kept_packs,
331 const struct pack_objects_args *args)
332 {
333 struct packed_git *p;
334 struct pack_geometry *geometry;
335 struct strbuf buf = STRBUF_INIT;
336
337 *geometry_p = xcalloc(1, sizeof(struct pack_geometry));
338 geometry = *geometry_p;
339
340 for (p = get_all_packs(the_repository); p; p = p->next) {
341 if (args->local && !p->pack_local)
342 /*
343 * When asked to only repack local packfiles we skip
344 * over any packfiles that are borrowed from alternate
345 * object directories.
346 */
347 continue;
348
349 if (!pack_kept_objects) {
350 /*
351 * Any pack that has its pack_keep bit set will appear
352 * in existing_kept_packs below, but this saves us from
353 * doing a more expensive check.
354 */
355 if (p->pack_keep)
356 continue;
357
358 /*
359 * The pack may be kept via the --keep-pack option;
360 * check 'existing_kept_packs' to determine whether to
361 * ignore it.
362 */
363 strbuf_reset(&buf);
364 strbuf_addstr(&buf, pack_basename(p));
365 strbuf_strip_suffix(&buf, ".pack");
366
367 if (string_list_has_string(existing_kept_packs, buf.buf))
368 continue;
369 }
370 if (p->is_cruft)
371 continue;
372
373 ALLOC_GROW(geometry->pack,
374 geometry->pack_nr + 1,
375 geometry->pack_alloc);
376
377 geometry->pack[geometry->pack_nr] = p;
378 geometry->pack_nr++;
379 }
380
381 QSORT(geometry->pack, geometry->pack_nr, geometry_cmp);
382 strbuf_release(&buf);
383 }
384
385 static void split_pack_geometry(struct pack_geometry *geometry, int factor)
386 {
387 uint32_t i;
388 uint32_t split;
389 off_t total_size = 0;
390
391 if (!geometry->pack_nr) {
392 geometry->split = geometry->pack_nr;
393 return;
394 }
395
396 /*
397 * First, count the number of packs (in descending order of size) which
398 * already form a geometric progression.
399 */
400 for (i = geometry->pack_nr - 1; i > 0; i--) {
401 struct packed_git *ours = geometry->pack[i];
402 struct packed_git *prev = geometry->pack[i - 1];
403
404 if (unsigned_mult_overflows(factor, geometry_pack_weight(prev)))
405 die(_("pack %s too large to consider in geometric "
406 "progression"),
407 prev->pack_name);
408
409 if (geometry_pack_weight(ours) < factor * geometry_pack_weight(prev))
410 break;
411 }
412
413 split = i;
414
415 if (split) {
416 /*
417 * Move the split one to the right, since the top element in the
418 * last-compared pair can't be in the progression. Only do this
419 * when we split in the middle of the array (otherwise if we got
420 * to the end, then the split is in the right place).
421 */
422 split++;
423 }
424
425 /*
426 * Then, anything to the left of 'split' must be in a new pack. But,
427 * creating that new pack may cause packs in the heavy half to no longer
428 * form a geometric progression.
429 *
430 * Compute an expected size of the new pack, and then determine how many
431 * packs in the heavy half need to be joined into it (if any) to restore
432 * the geometric progression.
433 */
434 for (i = 0; i < split; i++) {
435 struct packed_git *p = geometry->pack[i];
436
437 if (unsigned_add_overflows(total_size, geometry_pack_weight(p)))
438 die(_("pack %s too large to roll up"), p->pack_name);
439 total_size += geometry_pack_weight(p);
440 }
441 for (i = split; i < geometry->pack_nr; i++) {
442 struct packed_git *ours = geometry->pack[i];
443
444 if (unsigned_mult_overflows(factor, total_size))
445 die(_("pack %s too large to roll up"), ours->pack_name);
446
447 if (geometry_pack_weight(ours) < factor * total_size) {
448 if (unsigned_add_overflows(total_size,
449 geometry_pack_weight(ours)))
450 die(_("pack %s too large to roll up"),
451 ours->pack_name);
452
453 split++;
454 total_size += geometry_pack_weight(ours);
455 } else
456 break;
457 }
458
459 geometry->split = split;
460 }
461
462 static struct packed_git *get_preferred_pack(struct pack_geometry *geometry)
463 {
464 uint32_t i;
465
466 if (!geometry) {
467 /*
468 * No geometry means either an all-into-one repack (in which
469 * case there is only one pack left and it is the largest) or an
470 * incremental one.
471 *
472 * If repacking incrementally, then we could check the size of
473 * all packs to determine which should be preferred, but leave
474 * this for later.
475 */
476 return NULL;
477 }
478 if (geometry->split == geometry->pack_nr)
479 return NULL;
480
481 /*
482 * The preferred pack is the largest pack above the split line. In
483 * other words, it is the largest pack that does not get rolled up in
484 * the geometric repack.
485 */
486 for (i = geometry->pack_nr; i > geometry->split; i--)
487 /*
488 * A pack that is not local would never be included in a
489 * multi-pack index. We thus skip over any non-local packs.
490 */
491 if (geometry->pack[i - 1]->pack_local)
492 return geometry->pack[i - 1];
493
494 return NULL;
495 }
496
497 static void clear_pack_geometry(struct pack_geometry *geometry)
498 {
499 if (!geometry)
500 return;
501
502 free(geometry->pack);
503 geometry->pack_nr = 0;
504 geometry->pack_alloc = 0;
505 geometry->split = 0;
506 }
507
508 struct midx_snapshot_ref_data {
509 struct tempfile *f;
510 struct oidset seen;
511 int preferred;
512 };
513
514 static int midx_snapshot_ref_one(const char *refname UNUSED,
515 const struct object_id *oid,
516 int flag UNUSED, void *_data)
517 {
518 struct midx_snapshot_ref_data *data = _data;
519 struct object_id peeled;
520
521 if (!peel_iterated_oid(oid, &peeled))
522 oid = &peeled;
523
524 if (oidset_insert(&data->seen, oid))
525 return 0; /* already seen */
526
527 if (oid_object_info(the_repository, oid, NULL) != OBJ_COMMIT)
528 return 0;
529
530 fprintf(data->f->fp, "%s%s\n", data->preferred ? "+" : "",
531 oid_to_hex(oid));
532
533 return 0;
534 }
535
536 static void midx_snapshot_refs(struct tempfile *f)
537 {
538 struct midx_snapshot_ref_data data;
539 const struct string_list *preferred = bitmap_preferred_tips(the_repository);
540
541 data.f = f;
542 data.preferred = 0;
543 oidset_init(&data.seen, 0);
544
545 if (!fdopen_tempfile(f, "w"))
546 die(_("could not open tempfile %s for writing"),
547 get_tempfile_path(f));
548
549 if (preferred) {
550 struct string_list_item *item;
551
552 data.preferred = 1;
553 for_each_string_list_item(item, preferred)
554 for_each_ref_in(item->string, midx_snapshot_ref_one, &data);
555 data.preferred = 0;
556 }
557
558 for_each_ref(midx_snapshot_ref_one, &data);
559
560 if (close_tempfile_gently(f)) {
561 int save_errno = errno;
562 delete_tempfile(&f);
563 errno = save_errno;
564 die_errno(_("could not close refs snapshot tempfile"));
565 }
566
567 oidset_clear(&data.seen);
568 }
569
570 static void midx_included_packs(struct string_list *include,
571 struct string_list *existing_nonkept_packs,
572 struct string_list *existing_kept_packs,
573 struct string_list *names,
574 struct pack_geometry *geometry)
575 {
576 struct string_list_item *item;
577
578 for_each_string_list_item(item, existing_kept_packs)
579 string_list_insert(include, xstrfmt("%s.idx", item->string));
580 for_each_string_list_item(item, names)
581 string_list_insert(include, xstrfmt("pack-%s.idx", item->string));
582 if (geometry) {
583 struct strbuf buf = STRBUF_INIT;
584 uint32_t i;
585 for (i = geometry->split; i < geometry->pack_nr; i++) {
586 struct packed_git *p = geometry->pack[i];
587
588 /*
589 * The multi-pack index never refers to packfiles part
590 * of an alternate object database, so we skip these.
591 * While git-multi-pack-index(1) would silently ignore
592 * them anyway, this allows us to skip executing the
593 * command completely when we have only non-local
594 * packfiles.
595 */
596 if (!p->pack_local)
597 continue;
598
599 strbuf_addstr(&buf, pack_basename(p));
600 strbuf_strip_suffix(&buf, ".pack");
601 strbuf_addstr(&buf, ".idx");
602
603 string_list_insert(include, strbuf_detach(&buf, NULL));
604 }
605
606 for_each_string_list_item(item, existing_nonkept_packs) {
607 if (!((uintptr_t)item->util & CRUFT_PACK)) {
608 /*
609 * no need to check DELETE_PACK, since we're not
610 * doing an ALL_INTO_ONE repack
611 */
612 continue;
613 }
614 string_list_insert(include, xstrfmt("%s.idx", item->string));
615 }
616 } else {
617 for_each_string_list_item(item, existing_nonkept_packs) {
618 if ((uintptr_t)item->util & DELETE_PACK)
619 continue;
620 string_list_insert(include, xstrfmt("%s.idx", item->string));
621 }
622 }
623 }
624
625 static int write_midx_included_packs(struct string_list *include,
626 struct pack_geometry *geometry,
627 const char *refs_snapshot,
628 int show_progress, int write_bitmaps)
629 {
630 struct child_process cmd = CHILD_PROCESS_INIT;
631 struct string_list_item *item;
632 struct packed_git *preferred = get_preferred_pack(geometry);
633 FILE *in;
634 int ret;
635
636 if (!include->nr)
637 return 0;
638
639 cmd.in = -1;
640 cmd.git_cmd = 1;
641
642 strvec_push(&cmd.args, "multi-pack-index");
643 strvec_pushl(&cmd.args, "write", "--stdin-packs", NULL);
644
645 if (show_progress)
646 strvec_push(&cmd.args, "--progress");
647 else
648 strvec_push(&cmd.args, "--no-progress");
649
650 if (write_bitmaps)
651 strvec_push(&cmd.args, "--bitmap");
652
653 if (preferred)
654 strvec_pushf(&cmd.args, "--preferred-pack=%s",
655 pack_basename(preferred));
656
657 if (refs_snapshot)
658 strvec_pushf(&cmd.args, "--refs-snapshot=%s", refs_snapshot);
659
660 ret = start_command(&cmd);
661 if (ret)
662 return ret;
663
664 in = xfdopen(cmd.in, "w");
665 for_each_string_list_item(item, include)
666 fprintf(in, "%s\n", item->string);
667 fclose(in);
668
669 return finish_command(&cmd);
670 }
671
672 static void remove_redundant_bitmaps(struct string_list *include,
673 const char *packdir)
674 {
675 struct strbuf path = STRBUF_INIT;
676 struct string_list_item *item;
677 size_t packdir_len;
678
679 strbuf_addstr(&path, packdir);
680 strbuf_addch(&path, '/');
681 packdir_len = path.len;
682
683 /*
684 * Remove any pack bitmaps corresponding to packs which are now
685 * included in the MIDX.
686 */
687 for_each_string_list_item(item, include) {
688 strbuf_addstr(&path, item->string);
689 strbuf_strip_suffix(&path, ".idx");
690 strbuf_addstr(&path, ".bitmap");
691
692 if (unlink(path.buf) && errno != ENOENT)
693 warning_errno(_("could not remove stale bitmap: %s"),
694 path.buf);
695
696 strbuf_setlen(&path, packdir_len);
697 }
698 strbuf_release(&path);
699 }
700
701 static int write_cruft_pack(const struct pack_objects_args *args,
702 const char *destination,
703 const char *pack_prefix,
704 const char *cruft_expiration,
705 struct string_list *names,
706 struct string_list *existing_packs,
707 struct string_list *existing_kept_packs)
708 {
709 struct child_process cmd = CHILD_PROCESS_INIT;
710 struct strbuf line = STRBUF_INIT;
711 struct string_list_item *item;
712 FILE *in, *out;
713 int ret;
714 const char *scratch;
715 int local = skip_prefix(destination, packdir, &scratch);
716
717 prepare_pack_objects(&cmd, args, destination);
718
719 strvec_push(&cmd.args, "--cruft");
720 if (cruft_expiration)
721 strvec_pushf(&cmd.args, "--cruft-expiration=%s",
722 cruft_expiration);
723
724 strvec_push(&cmd.args, "--honor-pack-keep");
725 strvec_push(&cmd.args, "--non-empty");
726 strvec_push(&cmd.args, "--max-pack-size=0");
727
728 cmd.in = -1;
729
730 ret = start_command(&cmd);
731 if (ret)
732 return ret;
733
734 /*
735 * names has a confusing double use: it both provides the list
736 * of just-written new packs, and accepts the name of the cruft
737 * pack we are writing.
738 *
739 * By the time it is read here, it contains only the pack(s)
740 * that were just written, which is exactly the set of packs we
741 * want to consider kept.
742 *
743 * If `--expire-to` is given, the double-use served by `names`
744 * ensures that the pack written to `--expire-to` excludes any
745 * objects contained in the cruft pack.
746 */
747 in = xfdopen(cmd.in, "w");
748 for_each_string_list_item(item, names)
749 fprintf(in, "%s-%s.pack\n", pack_prefix, item->string);
750 for_each_string_list_item(item, existing_packs)
751 fprintf(in, "-%s.pack\n", item->string);
752 for_each_string_list_item(item, existing_kept_packs)
753 fprintf(in, "%s.pack\n", item->string);
754 fclose(in);
755
756 out = xfdopen(cmd.out, "r");
757 while (strbuf_getline_lf(&line, out) != EOF) {
758 struct string_list_item *item;
759
760 if (line.len != the_hash_algo->hexsz)
761 die(_("repack: Expecting full hex object ID lines only "
762 "from pack-objects."));
763 /*
764 * avoid putting packs written outside of the repository in the
765 * list of names
766 */
767 if (local) {
768 item = string_list_append(names, line.buf);
769 item->util = populate_pack_exts(line.buf);
770 }
771 }
772 fclose(out);
773
774 strbuf_release(&line);
775
776 return finish_command(&cmd);
777 }
778
779 int cmd_repack(int argc, const char **argv, const char *prefix)
780 {
781 struct child_process cmd = CHILD_PROCESS_INIT;
782 struct string_list_item *item;
783 struct string_list names = STRING_LIST_INIT_DUP;
784 struct string_list existing_nonkept_packs = STRING_LIST_INIT_DUP;
785 struct string_list existing_kept_packs = STRING_LIST_INIT_DUP;
786 struct pack_geometry *geometry = NULL;
787 struct strbuf line = STRBUF_INIT;
788 struct tempfile *refs_snapshot = NULL;
789 int i, ext, ret;
790 FILE *out;
791 int show_progress;
792
793 /* variables to be filled by option parsing */
794 int delete_redundant = 0;
795 const char *unpack_unreachable = NULL;
796 int keep_unreachable = 0;
797 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
798 struct pack_objects_args po_args = {NULL};
799 struct pack_objects_args cruft_po_args = {NULL};
800 int geometric_factor = 0;
801 int write_midx = 0;
802 const char *cruft_expiration = NULL;
803 const char *expire_to = NULL;
804
805 struct option builtin_repack_options[] = {
806 OPT_BIT('a', NULL, &pack_everything,
807 N_("pack everything in a single pack"), ALL_INTO_ONE),
808 OPT_BIT('A', NULL, &pack_everything,
809 N_("same as -a, and turn unreachable objects loose"),
810 LOOSEN_UNREACHABLE | ALL_INTO_ONE),
811 OPT_BIT(0, "cruft", &pack_everything,
812 N_("same as -a, pack unreachable cruft objects separately"),
813 PACK_CRUFT),
814 OPT_STRING(0, "cruft-expiration", &cruft_expiration, N_("approxidate"),
815 N_("with --cruft, expire objects older than this")),
816 OPT_BOOL('d', NULL, &delete_redundant,
817 N_("remove redundant packs, and run git-prune-packed")),
818 OPT_BOOL('f', NULL, &po_args.no_reuse_delta,
819 N_("pass --no-reuse-delta to git-pack-objects")),
820 OPT_BOOL('F', NULL, &po_args.no_reuse_object,
821 N_("pass --no-reuse-object to git-pack-objects")),
822 OPT_NEGBIT('n', NULL, &run_update_server_info,
823 N_("do not run git-update-server-info"), 1),
824 OPT__QUIET(&po_args.quiet, N_("be quiet")),
825 OPT_BOOL('l', "local", &po_args.local,
826 N_("pass --local to git-pack-objects")),
827 OPT_BOOL('b', "write-bitmap-index", &write_bitmaps,
828 N_("write bitmap index")),
829 OPT_BOOL('i', "delta-islands", &use_delta_islands,
830 N_("pass --delta-islands to git-pack-objects")),
831 OPT_STRING(0, "unpack-unreachable", &unpack_unreachable, N_("approxidate"),
832 N_("with -A, do not loosen objects older than this")),
833 OPT_BOOL('k', "keep-unreachable", &keep_unreachable,
834 N_("with -a, repack unreachable objects")),
835 OPT_STRING(0, "window", &po_args.window, N_("n"),
836 N_("size of the window used for delta compression")),
837 OPT_STRING(0, "window-memory", &po_args.window_memory, N_("bytes"),
838 N_("same as the above, but limit memory size instead of entries count")),
839 OPT_STRING(0, "depth", &po_args.depth, N_("n"),
840 N_("limits the maximum delta depth")),
841 OPT_STRING(0, "threads", &po_args.threads, N_("n"),
842 N_("limits the maximum number of threads")),
843 OPT_STRING(0, "max-pack-size", &po_args.max_pack_size, N_("bytes"),
844 N_("maximum size of each packfile")),
845 OPT_BOOL(0, "pack-kept-objects", &pack_kept_objects,
846 N_("repack objects in packs marked with .keep")),
847 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
848 N_("do not repack this pack")),
849 OPT_INTEGER('g', "geometric", &geometric_factor,
850 N_("find a geometric progression with factor <N>")),
851 OPT_BOOL('m', "write-midx", &write_midx,
852 N_("write a multi-pack index of the resulting packs")),
853 OPT_STRING(0, "expire-to", &expire_to, N_("dir"),
854 N_("pack prefix to store a pack containing pruned objects")),
855 OPT_END()
856 };
857
858 git_config(repack_config, &cruft_po_args);
859
860 argc = parse_options(argc, argv, prefix, builtin_repack_options,
861 git_repack_usage, 0);
862
863 if (delete_redundant && repository_format_precious_objects)
864 die(_("cannot delete packs in a precious-objects repo"));
865
866 if (keep_unreachable &&
867 (unpack_unreachable || (pack_everything & LOOSEN_UNREACHABLE)))
868 die(_("options '%s' and '%s' cannot be used together"), "--keep-unreachable", "-A");
869
870 if (pack_everything & PACK_CRUFT) {
871 pack_everything |= ALL_INTO_ONE;
872
873 if (unpack_unreachable || (pack_everything & LOOSEN_UNREACHABLE))
874 die(_("options '%s' and '%s' cannot be used together"), "--cruft", "-A");
875 if (keep_unreachable)
876 die(_("options '%s' and '%s' cannot be used together"), "--cruft", "-k");
877 }
878
879 if (write_bitmaps < 0) {
880 if (!write_midx &&
881 (!(pack_everything & ALL_INTO_ONE) || !is_bare_repository()))
882 write_bitmaps = 0;
883 } else if (write_bitmaps &&
884 git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0) &&
885 git_env_bool(GIT_TEST_MULTI_PACK_INDEX_WRITE_BITMAP, 0)) {
886 write_bitmaps = 0;
887 }
888 if (pack_kept_objects < 0)
889 pack_kept_objects = write_bitmaps > 0 && !write_midx;
890
891 if (write_bitmaps && !(pack_everything & ALL_INTO_ONE) && !write_midx)
892 die(_(incremental_bitmap_conflict_error));
893
894 if (write_bitmaps && po_args.local && has_alt_odb(the_repository)) {
895 /*
896 * When asked to do a local repack, but we have
897 * packfiles that are inherited from an alternate, then
898 * we cannot guarantee that the multi-pack-index would
899 * have full coverage of all objects. We thus disable
900 * writing bitmaps in that case.
901 */
902 warning(_("disabling bitmap writing, as some objects are not being packed"));
903 write_bitmaps = 0;
904 }
905
906 if (write_midx && write_bitmaps) {
907 struct strbuf path = STRBUF_INIT;
908
909 strbuf_addf(&path, "%s/%s_XXXXXX", get_object_directory(),
910 "bitmap-ref-tips");
911
912 refs_snapshot = xmks_tempfile(path.buf);
913 midx_snapshot_refs(refs_snapshot);
914
915 strbuf_release(&path);
916 }
917
918 packdir = mkpathdup("%s/pack", get_object_directory());
919 packtmp_name = xstrfmt(".tmp-%d-pack", (int)getpid());
920 packtmp = mkpathdup("%s/%s", packdir, packtmp_name);
921
922 collect_pack_filenames(&existing_nonkept_packs, &existing_kept_packs,
923 &keep_pack_list);
924
925 if (geometric_factor) {
926 if (pack_everything)
927 die(_("options '%s' and '%s' cannot be used together"), "--geometric", "-A/-a");
928 init_pack_geometry(&geometry, &existing_kept_packs, &po_args);
929 split_pack_geometry(geometry, geometric_factor);
930 }
931
932 prepare_pack_objects(&cmd, &po_args, packtmp);
933
934 show_progress = !po_args.quiet && isatty(2);
935
936 strvec_push(&cmd.args, "--keep-true-parents");
937 if (!pack_kept_objects)
938 strvec_push(&cmd.args, "--honor-pack-keep");
939 for (i = 0; i < keep_pack_list.nr; i++)
940 strvec_pushf(&cmd.args, "--keep-pack=%s",
941 keep_pack_list.items[i].string);
942 strvec_push(&cmd.args, "--non-empty");
943 if (!geometry) {
944 /*
945 * We need to grab all reachable objects, including those that
946 * are reachable from reflogs and the index.
947 *
948 * When repacking into a geometric progression of packs,
949 * however, we ask 'git pack-objects --stdin-packs', and it is
950 * not about packing objects based on reachability but about
951 * repacking all the objects in specified packs and loose ones
952 * (indeed, --stdin-packs is incompatible with these options).
953 */
954 strvec_push(&cmd.args, "--all");
955 strvec_push(&cmd.args, "--reflog");
956 strvec_push(&cmd.args, "--indexed-objects");
957 }
958 if (repo_has_promisor_remote(the_repository))
959 strvec_push(&cmd.args, "--exclude-promisor-objects");
960 if (!write_midx) {
961 if (write_bitmaps > 0)
962 strvec_push(&cmd.args, "--write-bitmap-index");
963 else if (write_bitmaps < 0)
964 strvec_push(&cmd.args, "--write-bitmap-index-quiet");
965 }
966 if (use_delta_islands)
967 strvec_push(&cmd.args, "--delta-islands");
968
969 if (pack_everything & ALL_INTO_ONE) {
970 repack_promisor_objects(&po_args, &names);
971
972 if (existing_nonkept_packs.nr && delete_redundant &&
973 !(pack_everything & PACK_CRUFT)) {
974 for_each_string_list_item(item, &names) {
975 strvec_pushf(&cmd.args, "--keep-pack=%s-%s.pack",
976 packtmp_name, item->string);
977 }
978 if (unpack_unreachable) {
979 strvec_pushf(&cmd.args,
980 "--unpack-unreachable=%s",
981 unpack_unreachable);
982 } else if (pack_everything & LOOSEN_UNREACHABLE) {
983 strvec_push(&cmd.args,
984 "--unpack-unreachable");
985 } else if (keep_unreachable) {
986 strvec_push(&cmd.args, "--keep-unreachable");
987 strvec_push(&cmd.args, "--pack-loose-unreachable");
988 }
989 }
990 } else if (geometry) {
991 strvec_push(&cmd.args, "--stdin-packs");
992 strvec_push(&cmd.args, "--unpacked");
993 } else {
994 strvec_push(&cmd.args, "--unpacked");
995 strvec_push(&cmd.args, "--incremental");
996 }
997
998 if (geometry)
999 cmd.in = -1;
1000 else
1001 cmd.no_stdin = 1;
1002
1003 ret = start_command(&cmd);
1004 if (ret)
1005 goto cleanup;
1006
1007 if (geometry) {
1008 FILE *in = xfdopen(cmd.in, "w");
1009 /*
1010 * The resulting pack should contain all objects in packs that
1011 * are going to be rolled up, but exclude objects in packs which
1012 * are being left alone.
1013 */
1014 for (i = 0; i < geometry->split; i++)
1015 fprintf(in, "%s\n", pack_basename(geometry->pack[i]));
1016 for (i = geometry->split; i < geometry->pack_nr; i++)
1017 fprintf(in, "^%s\n", pack_basename(geometry->pack[i]));
1018 fclose(in);
1019 }
1020
1021 out = xfdopen(cmd.out, "r");
1022 while (strbuf_getline_lf(&line, out) != EOF) {
1023 struct string_list_item *item;
1024
1025 if (line.len != the_hash_algo->hexsz)
1026 die(_("repack: Expecting full hex object ID lines only from pack-objects."));
1027 item = string_list_append(&names, line.buf);
1028 item->util = populate_pack_exts(item->string);
1029 }
1030 strbuf_release(&line);
1031 fclose(out);
1032 ret = finish_command(&cmd);
1033 if (ret)
1034 goto cleanup;
1035
1036 if (!names.nr && !po_args.quiet)
1037 printf_ln(_("Nothing new to pack."));
1038
1039 if (pack_everything & PACK_CRUFT) {
1040 const char *pack_prefix;
1041 if (!skip_prefix(packtmp, packdir, &pack_prefix))
1042 die(_("pack prefix %s does not begin with objdir %s"),
1043 packtmp, packdir);
1044 if (*pack_prefix == '/')
1045 pack_prefix++;
1046
1047 if (!cruft_po_args.window)
1048 cruft_po_args.window = po_args.window;
1049 if (!cruft_po_args.window_memory)
1050 cruft_po_args.window_memory = po_args.window_memory;
1051 if (!cruft_po_args.depth)
1052 cruft_po_args.depth = po_args.depth;
1053 if (!cruft_po_args.threads)
1054 cruft_po_args.threads = po_args.threads;
1055
1056 cruft_po_args.local = po_args.local;
1057 cruft_po_args.quiet = po_args.quiet;
1058
1059 ret = write_cruft_pack(&cruft_po_args, packtmp, pack_prefix,
1060 cruft_expiration, &names,
1061 &existing_nonkept_packs,
1062 &existing_kept_packs);
1063 if (ret)
1064 goto cleanup;
1065
1066 if (delete_redundant && expire_to) {
1067 /*
1068 * If `--expire-to` is given with `-d`, it's possible
1069 * that we're about to prune some objects. With cruft
1070 * packs, pruning is implicit: any objects from existing
1071 * packs that weren't picked up by new packs are removed
1072 * when their packs are deleted.
1073 *
1074 * Generate an additional cruft pack, with one twist:
1075 * `names` now includes the name of the cruft pack
1076 * written in the previous step. So the contents of
1077 * _this_ cruft pack exclude everything contained in the
1078 * existing cruft pack (that is, all of the unreachable
1079 * objects which are no older than
1080 * `--cruft-expiration`).
1081 *
1082 * To make this work, cruft_expiration must become NULL
1083 * so that this cruft pack doesn't actually prune any
1084 * objects. If it were non-NULL, this call would always
1085 * generate an empty pack (since every object not in the
1086 * cruft pack generated above will have an mtime older
1087 * than the expiration).
1088 */
1089 ret = write_cruft_pack(&cruft_po_args, expire_to,
1090 pack_prefix,
1091 NULL,
1092 &names,
1093 &existing_nonkept_packs,
1094 &existing_kept_packs);
1095 if (ret)
1096 goto cleanup;
1097 }
1098 }
1099
1100 string_list_sort(&names);
1101
1102 close_object_store(the_repository->objects);
1103
1104 /*
1105 * Ok we have prepared all new packfiles.
1106 */
1107 for_each_string_list_item(item, &names) {
1108 struct generated_pack_data *data = item->util;
1109
1110 for (ext = 0; ext < ARRAY_SIZE(exts); ext++) {
1111 char *fname;
1112
1113 fname = mkpathdup("%s/pack-%s%s",
1114 packdir, item->string, exts[ext].name);
1115
1116 if (data->tempfiles[ext]) {
1117 const char *fname_old = get_tempfile_path(data->tempfiles[ext]);
1118 struct stat statbuffer;
1119
1120 if (!stat(fname_old, &statbuffer)) {
1121 statbuffer.st_mode &= ~(S_IWUSR | S_IWGRP | S_IWOTH);
1122 chmod(fname_old, statbuffer.st_mode);
1123 }
1124
1125 if (rename_tempfile(&data->tempfiles[ext], fname))
1126 die_errno(_("renaming pack to '%s' failed"), fname);
1127 } else if (!exts[ext].optional)
1128 die(_("pack-objects did not write a '%s' file for pack %s-%s"),
1129 exts[ext].name, packtmp, item->string);
1130 else if (unlink(fname) < 0 && errno != ENOENT)
1131 die_errno(_("could not unlink: %s"), fname);
1132
1133 free(fname);
1134 }
1135 }
1136 /* End of pack replacement. */
1137
1138 if (delete_redundant && pack_everything & ALL_INTO_ONE) {
1139 const int hexsz = the_hash_algo->hexsz;
1140 for_each_string_list_item(item, &existing_nonkept_packs) {
1141 char *sha1;
1142 size_t len = strlen(item->string);
1143 if (len < hexsz)
1144 continue;
1145 sha1 = item->string + len - hexsz;
1146 /*
1147 * Mark this pack for deletion, which ensures that this
1148 * pack won't be included in a MIDX (if `--write-midx`
1149 * was given) and that we will actually delete this pack
1150 * (if `-d` was given).
1151 */
1152 if (!string_list_has_string(&names, sha1))
1153 item->util = (void*)(uintptr_t)((size_t)item->util | DELETE_PACK);
1154 }
1155 }
1156
1157 if (write_midx) {
1158 struct string_list include = STRING_LIST_INIT_NODUP;
1159 midx_included_packs(&include, &existing_nonkept_packs,
1160 &existing_kept_packs, &names, geometry);
1161
1162 ret = write_midx_included_packs(&include, geometry,
1163 refs_snapshot ? get_tempfile_path(refs_snapshot) : NULL,
1164 show_progress, write_bitmaps > 0);
1165
1166 if (!ret && write_bitmaps)
1167 remove_redundant_bitmaps(&include, packdir);
1168
1169 string_list_clear(&include, 0);
1170
1171 if (ret)
1172 goto cleanup;
1173 }
1174
1175 reprepare_packed_git(the_repository);
1176
1177 if (delete_redundant) {
1178 int opts = 0;
1179 for_each_string_list_item(item, &existing_nonkept_packs) {
1180 if (!((uintptr_t)item->util & DELETE_PACK))
1181 continue;
1182 remove_redundant_pack(packdir, item->string);
1183 }
1184
1185 if (geometry) {
1186 struct strbuf buf = STRBUF_INIT;
1187
1188 uint32_t i;
1189 for (i = 0; i < geometry->split; i++) {
1190 struct packed_git *p = geometry->pack[i];
1191 if (string_list_has_string(&names,
1192 hash_to_hex(p->hash)))
1193 continue;
1194
1195 strbuf_reset(&buf);
1196 strbuf_addstr(&buf, pack_basename(p));
1197 strbuf_strip_suffix(&buf, ".pack");
1198
1199 if ((p->pack_keep) ||
1200 (string_list_has_string(&existing_kept_packs,
1201 buf.buf)))
1202 continue;
1203
1204 remove_redundant_pack(packdir, buf.buf);
1205 }
1206 strbuf_release(&buf);
1207 }
1208 if (show_progress)
1209 opts |= PRUNE_PACKED_VERBOSE;
1210 prune_packed_objects(opts);
1211
1212 if (!keep_unreachable &&
1213 (!(pack_everything & LOOSEN_UNREACHABLE) ||
1214 unpack_unreachable) &&
1215 is_repository_shallow(the_repository))
1216 prune_shallow(PRUNE_QUICK);
1217 }
1218
1219 if (run_update_server_info)
1220 update_server_info(0);
1221
1222 if (git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0)) {
1223 unsigned flags = 0;
1224 if (git_env_bool(GIT_TEST_MULTI_PACK_INDEX_WRITE_BITMAP, 0))
1225 flags |= MIDX_WRITE_BITMAP | MIDX_WRITE_REV_INDEX;
1226 write_midx_file(get_object_directory(), NULL, NULL, flags);
1227 }
1228
1229 cleanup:
1230 string_list_clear(&names, 1);
1231 string_list_clear(&existing_nonkept_packs, 0);
1232 string_list_clear(&existing_kept_packs, 0);
1233 clear_pack_geometry(geometry);
1234
1235 return ret;
1236 }