2 * Copyright (c) 2020 Stefan Sperling <stsp@openbsd.org>
4 * Permission to use, copy, modify, and distribute this software for any
5 * purpose with or without fee is hereby granted, provided that the above
6 * copyright notice and this permission notice appear in all copies.
8 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
9 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
10 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
11 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
12 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
13 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
14 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17 #include "got_compat.h"
19 #include <sys/types.h>
20 #include <sys/queue.h>
23 #include <sys/socket.h>
36 #include "got_error.h"
37 #include "got_cancel.h"
38 #include "got_object.h"
39 #include "got_reference.h"
40 #include "got_repository.h"
41 #include "got_repository_admin.h"
42 #include "got_opentemp.h"
45 #include "got_lib_delta.h"
46 #include "got_lib_object.h"
47 #include "got_lib_object_idset.h"
48 #include "got_lib_object_cache.h"
49 #include "got_lib_pack.h"
50 #include "got_lib_privsep.h"
51 #include "got_lib_repository.h"
52 #include "got_lib_ratelimit.h"
53 #include "got_lib_pack_create.h"
54 #include "got_lib_hash.h"
55 #include "got_lib_lockfile.h"
58 #define nitems(_a) (sizeof((_a)) / sizeof((_a)[0]))
61 static const struct got_error *
62 get_reflist_object_ids(struct got_object_id ***ids, int *nobjects,
63 unsigned int wanted_obj_type_mask, struct got_reflist_head *refs,
64 struct got_repository *repo,
65 got_cancel_cb cancel_cb, void *cancel_arg)
67 const struct got_error *err = NULL;
68 const size_t alloc_chunksz = 256;
70 struct got_reflist_entry *re;
76 err = got_reflist_sort(refs,
77 got_ref_cmp_by_commit_timestamp_descending, repo);
81 *ids = reallocarray(NULL, alloc_chunksz, sizeof(struct got_object_id *));
83 return got_error_from_errno("reallocarray");
84 nalloc = alloc_chunksz;
86 TAILQ_FOREACH(re, refs, entry) {
87 struct got_object_id *id;
90 err = cancel_cb(cancel_arg);
95 err = got_ref_resolve(&id, repo, re->ref);
99 if (wanted_obj_type_mask != GOT_OBJ_TYPE_ANY) {
101 err = got_object_get_type(&obj_type, repo, id);
104 if ((wanted_obj_type_mask & (1 << obj_type)) == 0) {
111 if (nalloc <= *nobjects) {
112 struct got_object_id **new;
113 new = recallocarray(*ids, nalloc,
114 nalloc + alloc_chunksz,
115 sizeof(struct got_object_id *));
117 err = got_error_from_errno(
122 nalloc += alloc_chunksz;
124 (*ids)[*nobjects] = id;
125 if ((*ids)[*nobjects] == NULL) {
126 err = got_error_from_errno("got_object_id_dup");
133 for (i = 0; i < *nobjects; i++)
142 const struct got_error *
143 got_repo_pack_objects(FILE **packfile, struct got_object_id **pack_hash,
144 struct got_reflist_head *include_refs,
145 struct got_reflist_head *exclude_refs, struct got_repository *repo,
146 int loose_obj_only, int force_refdelta,
147 got_pack_progress_cb progress_cb, void *progress_arg,
148 got_cancel_cb cancel_cb, void *cancel_arg)
150 const struct got_error *err = NULL;
151 struct got_object_id **ours = NULL, **theirs = NULL;
152 int nours = 0, ntheirs = 0, packfd = -1, i;
153 char *tmpfile_path = NULL, *path = NULL, *packfile_path = NULL;
154 char *sha1_str = NULL;
155 FILE *delta_cache = NULL;
156 struct got_ratelimit rl;
161 got_ratelimit_init(&rl, 0, 500);
163 if (asprintf(&path, "%s/%s/packing.pack",
164 got_repo_get_path_git_dir(repo), GOT_OBJECTS_PACK_DIR) == -1) {
165 err = got_error_from_errno("asprintf");
168 err = got_opentemp_named_fd(&tmpfile_path, &packfd, path, "");
172 if (fchmod(packfd, GOT_DEFAULT_PACK_MODE) == -1) {
173 err = got_error_from_errno2("fchmod", tmpfile_path);
177 delta_cache = got_opentemp();
178 if (delta_cache == NULL) {
179 err = got_error_from_errno("got_opentemp");
183 err = get_reflist_object_ids(&ours, &nours,
184 (1 << GOT_OBJ_TYPE_COMMIT) | (1 << GOT_OBJ_TYPE_TAG),
185 include_refs, repo, cancel_cb, cancel_arg);
190 err = got_error(GOT_ERR_CANNOT_PACK);
194 if (!TAILQ_EMPTY(exclude_refs)) {
195 err = get_reflist_object_ids(&theirs, &ntheirs,
196 (1 << GOT_OBJ_TYPE_COMMIT) | (1 << GOT_OBJ_TYPE_TAG),
198 cancel_cb, cancel_arg);
203 *pack_hash = calloc(1, sizeof(**pack_hash));
204 if (*pack_hash == NULL) {
205 err = got_error_from_errno("calloc");
209 err = got_pack_create((*pack_hash)->sha1, packfd, delta_cache,
210 theirs, ntheirs, ours, nours, repo, loose_obj_only,
211 0, force_refdelta, progress_cb, progress_arg, &rl,
212 cancel_cb, cancel_arg);
216 err = got_object_id_str(&sha1_str, *pack_hash);
219 if (asprintf(&packfile_path, "%s/%s/pack-%s.pack",
220 got_repo_get_path_git_dir(repo), GOT_OBJECTS_PACK_DIR,
222 err = got_error_from_errno("asprintf");
226 if (lseek(packfd, 0L, SEEK_SET) == -1) {
227 err = got_error_from_errno("lseek");
230 if (rename(tmpfile_path, packfile_path) == -1) {
231 err = got_error_from_errno3("rename", tmpfile_path,
238 *packfile = fdopen(packfd, "w");
239 if (*packfile == NULL) {
240 err = got_error_from_errno2("fdopen", tmpfile_path);
245 for (i = 0; i < nours; i++)
248 for (i = 0; i < ntheirs; i++)
251 if (packfd != -1 && close(packfd) == -1 && err == NULL)
252 err = got_error_from_errno2("close", packfile_path);
253 if (delta_cache && fclose(delta_cache) == EOF && err == NULL)
254 err = got_error_from_errno("fclose");
255 if (tmpfile_path && unlink(tmpfile_path) == -1 && err == NULL)
256 err = got_error_from_errno2("unlink", tmpfile_path);
271 const struct got_error *
272 got_repo_index_pack(FILE *packfile, struct got_object_id *pack_hash,
273 struct got_repository *repo,
274 got_pack_index_progress_cb progress_cb, void *progress_arg,
275 got_cancel_cb cancel_cb, void *cancel_arg)
280 int npackfd = -1, idxfd = -1, nidxfd = -1;
282 int idxstatus, done = 0;
283 const struct got_error *err;
284 struct imsgbuf idxibuf;
286 char *tmpidxpath = NULL;
287 char *packfile_path = NULL, *idxpath = NULL, *id_str = NULL;
288 const char *repo_path = got_repo_get_path_git_dir(repo);
291 for (i = 0; i < nitems(tmpfds); i++)
294 if (asprintf(&path, "%s/%s/indexing.idx",
295 repo_path, GOT_OBJECTS_PACK_DIR) == -1) {
296 err = got_error_from_errno("asprintf");
299 err = got_opentemp_named_fd(&tmpidxpath, &idxfd, path, "");
303 if (fchmod(idxfd, GOT_DEFAULT_PACK_MODE) == -1) {
304 err = got_error_from_errno2("fchmod", tmpidxpath);
310 err = got_error_from_errno("dup");
314 for (i = 0; i < nitems(tmpfds); i++) {
315 tmpfds[i] = got_opentempfd();
316 if (tmpfds[i] == -1) {
317 err = got_error_from_errno("got_opentempfd");
322 err = got_object_id_str(&id_str, pack_hash);
326 if (asprintf(&packfile_path, "%s/%s/pack-%s.pack",
327 repo_path, GOT_OBJECTS_PACK_DIR, id_str) == -1) {
328 err = got_error_from_errno("asprintf");
332 if (fstat(fileno(packfile), &sb) == -1) {
333 err = got_error_from_errno2("fstat", packfile_path);
337 if (asprintf(&idxpath, "%s/%s/pack-%s.idx",
338 repo_path, GOT_OBJECTS_PACK_DIR, id_str) == -1) {
339 err = got_error_from_errno("asprintf");
343 if (socketpair(AF_UNIX, SOCK_STREAM, PF_UNSPEC, imsg_idxfds) == -1) {
344 err = got_error_from_errno("socketpair");
349 err= got_error_from_errno("fork");
351 } else if (idxpid == 0)
352 got_privsep_exec_child(imsg_idxfds,
353 GOT_PATH_PROG_INDEX_PACK, packfile_path);
354 if (close(imsg_idxfds[1]) == -1) {
355 err = got_error_from_errno("close");
358 imsg_init(&idxibuf, imsg_idxfds[0]);
360 npackfd = dup(fileno(packfile));
362 err = got_error_from_errno("dup");
365 err = got_privsep_send_index_pack_req(&idxibuf, pack_hash->sha1,
370 err = got_privsep_send_index_pack_outfd(&idxibuf, nidxfd);
374 for (i = 0; i < nitems(tmpfds); i++) {
375 err = got_privsep_send_tmpfd(&idxibuf, tmpfds[i]);
382 int nobj_total, nobj_indexed, nobj_loose, nobj_resolved;
385 err = cancel_cb(cancel_arg);
390 err = got_privsep_recv_index_progress(&done, &nobj_total,
391 &nobj_indexed, &nobj_loose, &nobj_resolved,
395 if (nobj_indexed != 0) {
396 err = progress_cb(progress_arg, sb.st_size,
397 nobj_total, nobj_indexed, nobj_loose,
403 if (close(imsg_idxfds[0]) == -1) {
404 err = got_error_from_errno("close");
407 if (waitpid(idxpid, &idxstatus, 0) == -1) {
408 err = got_error_from_errno("waitpid");
412 if (rename(tmpidxpath, idxpath) == -1) {
413 err = got_error_from_errno3("rename", tmpidxpath, idxpath);
420 if (tmpidxpath && unlink(tmpidxpath) == -1 && err == NULL)
421 err = got_error_from_errno2("unlink", tmpidxpath);
422 if (npackfd != -1 && close(npackfd) == -1 && err == NULL)
423 err = got_error_from_errno("close");
424 if (idxfd != -1 && close(idxfd) == -1 && err == NULL)
425 err = got_error_from_errno("close");
426 for (i = 0; i < nitems(tmpfds); i++) {
427 if (tmpfds[i] != -1 && close(tmpfds[i]) == -1 && err == NULL)
428 err = got_error_from_errno("close");
436 const struct got_error *
437 got_repo_find_pack(FILE **packfile, struct got_object_id **pack_hash,
438 struct got_repository *repo, const char *packfile_path)
440 const struct got_error *err = NULL;
441 const char *packdir_path = NULL;
442 char *packfile_name = NULL, *p, *dot;
443 struct got_object_id id;
449 packdir_path = got_repo_get_path_objects_pack(repo);
450 if (packdir_path == NULL)
451 return got_error_from_errno("got_repo_get_path_objects_pack");
453 if (!got_path_is_child(packfile_path, packdir_path,
454 strlen(packdir_path))) {
455 err = got_error_path(packfile_path, GOT_ERR_BAD_PATH);
460 err = got_path_basename(&packfile_name, packfile_path);
465 if (strncmp(p, "pack-", 5) != 0) {
466 err = got_error_fmt(GOT_ERR_BAD_PATH,
467 "'%s' is not a valid pack file name",
472 dot = strchr(p, '.');
474 err = got_error_fmt(GOT_ERR_BAD_PATH,
475 "'%s' is not a valid pack file name",
479 if (strcmp(dot + 1, "pack") != 0) {
480 err = got_error_fmt(GOT_ERR_BAD_PATH,
481 "'%s' is not a valid pack file name",
486 if (!got_parse_object_id(&id, p, repo->algo)) {
487 err = got_error_fmt(GOT_ERR_BAD_PATH,
488 "'%s' is not a valid pack file name",
493 *pack_hash = got_object_id_dup(&id);
494 if (*pack_hash == NULL) {
495 err = got_error_from_errno("got_object_id_dup");
499 packfd = open(packfile_path, O_RDONLY | O_NOFOLLOW | O_CLOEXEC);
501 err = got_error_from_errno2("open", packfile_path);
505 *packfile = fdopen(packfd, "r");
506 if (*packfile == NULL) {
507 err = got_error_from_errno2("fdopen", packfile_path);
512 if (packfd != -1 && close(packfd) == -1 && err == NULL)
513 err = got_error_from_errno2("close", packfile_path);
522 const struct got_error *
523 got_repo_list_pack(FILE *packfile, struct got_object_id *pack_hash,
524 struct got_repository *repo, got_pack_list_cb list_cb, void *list_arg,
525 got_cancel_cb cancel_cb, void *cancel_arg)
527 const struct got_error *err = NULL;
528 char *id_str = NULL, *idxpath = NULL, *packpath = NULL;
529 struct got_packidx *packidx = NULL;
530 struct got_pack *pack = NULL;
533 err = got_object_id_str(&id_str, pack_hash);
537 if (asprintf(&packpath, "%s/pack-%s.pack",
538 GOT_OBJECTS_PACK_DIR, id_str) == -1) {
539 err = got_error_from_errno("asprintf");
542 if (asprintf(&idxpath, "%s/pack-%s.idx",
543 GOT_OBJECTS_PACK_DIR, id_str) == -1) {
544 err = got_error_from_errno("asprintf");
548 err = got_packidx_open(&packidx, got_repo_get_fd(repo), idxpath, 1);
552 err = got_repo_cache_pack(&pack, repo, packpath, packidx);
556 nobj = be32toh(packidx->hdr.fanout_table[0xff]);
557 for (i = 0; i < nobj; i++) {
558 struct got_packidx_object_id *oid;
559 struct got_object_id id, base_id;
560 off_t offset, base_offset = 0;
566 err = cancel_cb(cancel_arg);
570 oid = &packidx->hdr.sorted_ids[i];
571 memcpy(id.sha1, oid->sha1, SHA1_DIGEST_LENGTH);
573 offset = got_packidx_get_object_offset(packidx, i);
575 err = got_error(GOT_ERR_BAD_PACKIDX);
579 err = got_pack_parse_object_type_and_size(&type, &size, &tslen,
585 case GOT_OBJ_TYPE_OFFSET_DELTA:
586 err = got_pack_parse_offset_delta(&base_offset, &len,
587 pack, offset, tslen);
591 case GOT_OBJ_TYPE_REF_DELTA:
592 err = got_pack_parse_ref_delta(&base_id,
593 pack, offset, tslen);
598 err = (*list_cb)(list_arg, &id, type, offset, size,
599 base_offset, &base_id);
609 got_packidx_close(packidx);
613 const struct got_error *
614 got_repo_cleanup_prepare(struct got_repository *repo,
615 struct got_lockfile **lk)
617 const struct got_error *err;
618 char myname[HOST_NAME_MAX + 1];
620 if (gethostname(myname, sizeof(myname)) == -1)
621 return got_error_from_errno("gethostname");
623 err = got_lockfile_lock(lk, "gc.pid", got_repo_get_fd(repo));
628 * Git uses these info to provide some verbiage when finds a
629 * lock during `git gc --force' so don't try too hard to avoid
630 * short writes and don't care if a race happens between the
631 * lockfile creation and the write itself.
633 if (dprintf((*lk)->fd, "%d %s", getpid(), myname) < 0)
634 return got_error_from_errno("dprintf");
639 const struct got_error *
640 got_repo_cleanup_complete(struct got_repository *repo,
641 struct got_lockfile *lk)
646 return got_lockfile_unlock(lk, got_repo_get_fd(repo));
649 static const struct got_error *
650 report_cleanup_progress(got_cleanup_progress_cb progress_cb,
651 void *progress_arg, struct got_ratelimit *rl,
652 int nloose, int ncommits, int npurged)
654 const struct got_error *err;
657 if (progress_cb == NULL)
660 err = got_ratelimit_check(&elapsed, rl);
664 return progress_cb(progress_arg, nloose, ncommits, npurged, -1);
667 static const struct got_error *
668 get_loose_object_ids(struct got_object_idset **loose_ids, off_t *ondisk_size,
669 got_cleanup_progress_cb progress_cb, void *progress_arg,
670 struct got_ratelimit *rl, struct got_repository *repo)
672 const struct got_error *err = NULL;
673 char *path_objects = NULL, *path = NULL;
675 struct got_object *obj = NULL;
676 struct got_object_id id;
681 *loose_ids = got_object_idset_alloc();
682 if (*loose_ids == NULL)
683 return got_error_from_errno("got_object_idset_alloc");
685 path_objects = got_repo_get_path_objects(repo);
686 if (path_objects == NULL) {
687 err = got_error_from_errno("got_repo_get_path_objects");
691 for (i = 0; i <= 0xff; i++) {
694 if (asprintf(&path, "%s/%.2x", path_objects, i) == -1) {
695 err = got_error_from_errno("asprintf");
701 if (errno == ENOENT) {
705 err = got_error_from_errno2("opendir", path);
709 while ((dent = readdir(dir)) != NULL) {
712 if (strcmp(dent->d_name, ".") == 0 ||
713 strcmp(dent->d_name, "..") == 0)
716 if (asprintf(&id_str, "%.2x%s", i, dent->d_name) == -1) {
717 err = got_error_from_errno("asprintf");
721 if (!got_parse_object_id(&id, id_str, repo->algo)) {
727 err = got_object_open_loose_fd(&fd, &id, repo);
730 if (fstat(fd, &sb) == -1) {
731 err = got_error_from_errno("fstat");
734 err = got_object_read_header_privsep(&obj, &id, repo,
738 fd = -1; /* already closed */
741 case GOT_OBJ_TYPE_COMMIT:
742 case GOT_OBJ_TYPE_TREE:
743 case GOT_OBJ_TYPE_BLOB:
744 case GOT_OBJ_TYPE_TAG:
747 err = got_error_fmt(GOT_ERR_OBJ_TYPE,
751 got_object_close(obj);
753 (*ondisk_size) += sb.st_size;
754 err = got_object_idset_add(*loose_ids, &id, NULL);
757 err = report_cleanup_progress(progress_cb,
759 got_object_idset_num_elements(*loose_ids), -1, -1);
764 if (closedir(dir) != 0) {
765 err = got_error_from_errno("closedir");
774 if (dir && closedir(dir) != 0 && err == NULL)
775 err = got_error_from_errno("closedir");
776 if (fd != -1 && close(fd) == -1 && err == NULL)
777 err = got_error_from_errno("close");
779 got_object_idset_free(*loose_ids);
783 got_object_close(obj);
789 static const struct got_error *
790 preserve_loose_object(struct got_object_idset *loose_ids,
791 struct got_object_id *id, struct got_repository *repo, int *npacked)
793 const struct got_error *err = NULL;
794 struct got_object *obj;
796 if (!got_object_idset_contains(loose_ids, id))
800 * Try to open this object from a pack file. This ensures that
801 * we do in fact have a valid packed copy of the object. Otherwise
802 * we should not delete the loose representation of this object.
804 err = got_object_open_packed(&obj, id, repo);
806 got_object_close(obj);
808 * The object is referenced and packed.
809 * We can purge the redundantly stored loose object.
813 } else if (err->code != GOT_ERR_NO_OBJ)
817 * This object is referenced and not packed.
818 * Remove it from our purge set.
820 return got_object_idset_remove(NULL, loose_ids, id);
823 static const struct got_error *
824 load_tree_entries(struct got_object_id_queue *ids,
825 struct got_object_idset *loose_ids,
826 struct got_object_idset *traversed_ids, struct got_object_id *tree_id,
827 const char *dpath, struct got_repository *repo, int *npacked,
828 got_cancel_cb cancel_cb, void *cancel_arg)
830 const struct got_error *err;
831 struct got_tree_object *tree;
835 err = got_object_open_as_tree(&tree, repo, tree_id);
839 for (i = 0; i < got_object_tree_get_nentries(tree); i++) {
840 struct got_tree_entry *e = got_object_tree_get_entry(tree, i);
841 struct got_object_id *id = got_tree_entry_get_id(e);
842 mode_t mode = got_tree_entry_get_mode(e);
845 err = (*cancel_cb)(cancel_arg);
850 if (got_object_tree_entry_is_symlink(e) ||
851 got_object_tree_entry_is_submodule(e) ||
852 got_object_idset_contains(traversed_ids, id))
855 if (asprintf(&p, "%s%s%s", dpath, dpath[0] != '\0' ? "/" : "",
856 got_tree_entry_get_name(e)) == -1) {
857 err = got_error_from_errno("asprintf");
862 struct got_object_qid *qid;
863 err = got_object_qid_alloc(&qid, id);
866 STAILQ_INSERT_TAIL(ids, qid, entry);
867 } else if (S_ISREG(mode)) {
868 /* This blob is referenced. */
869 err = preserve_loose_object(loose_ids, id, repo,
873 err = got_object_idset_add(traversed_ids, id, NULL);
881 got_object_tree_close(tree);
886 static const struct got_error *
887 load_tree(struct got_object_idset *loose_ids,
888 struct got_object_idset *traversed_ids, struct got_object_id *tree_id,
889 const char *dpath, struct got_repository *repo, int *npacked,
890 got_cancel_cb cancel_cb, void *cancel_arg)
892 const struct got_error *err = NULL;
893 struct got_object_id_queue tree_ids;
894 struct got_object_qid *qid;
896 err = got_object_qid_alloc(&qid, tree_id);
900 STAILQ_INIT(&tree_ids);
901 STAILQ_INSERT_TAIL(&tree_ids, qid, entry);
903 while (!STAILQ_EMPTY(&tree_ids)) {
905 err = (*cancel_cb)(cancel_arg);
910 qid = STAILQ_FIRST(&tree_ids);
911 STAILQ_REMOVE_HEAD(&tree_ids, entry);
913 if (got_object_idset_contains(traversed_ids, &qid->id)) {
914 got_object_qid_free(qid);
918 err = got_object_idset_add(traversed_ids, &qid->id, NULL);
920 got_object_qid_free(qid);
924 /* This tree is referenced. */
925 err = preserve_loose_object(loose_ids, &qid->id, repo, npacked);
929 err = load_tree_entries(&tree_ids, loose_ids, traversed_ids,
930 &qid->id, dpath, repo, npacked, cancel_cb, cancel_arg);
931 got_object_qid_free(qid);
936 got_object_id_queue_free(&tree_ids);
940 static const struct got_error *
941 load_commit_or_tag(struct got_object_idset *loose_ids, int *ncommits,
942 int *npacked, struct got_object_idset *traversed_ids,
943 struct got_object_id *id, struct got_repository *repo,
944 got_cleanup_progress_cb progress_cb, void *progress_arg,
945 struct got_ratelimit *rl, int nloose,
946 got_cancel_cb cancel_cb, void *cancel_arg)
948 const struct got_error *err;
949 struct got_commit_object *commit = NULL;
950 struct got_tag_object *tag = NULL;
951 struct got_object_id *tree_id = NULL;
952 struct got_object_id_queue ids;
953 struct got_object_qid *qid;
956 err = got_object_qid_alloc(&qid, id);
961 STAILQ_INSERT_TAIL(&ids, qid, entry);
963 while (!STAILQ_EMPTY(&ids)) {
965 err = (*cancel_cb)(cancel_arg);
970 qid = STAILQ_FIRST(&ids);
971 STAILQ_REMOVE_HEAD(&ids, entry);
973 if (got_object_idset_contains(traversed_ids, &qid->id)) {
974 got_object_qid_free(qid);
979 err = got_object_idset_add(traversed_ids, &qid->id, NULL);
983 /* This commit or tag is referenced. */
984 err = preserve_loose_object(loose_ids, &qid->id, repo, npacked);
988 err = got_object_get_type(&obj_type, repo, &qid->id);
992 case GOT_OBJ_TYPE_COMMIT:
993 err = got_object_open_as_commit(&commit, repo,
998 case GOT_OBJ_TYPE_TAG:
999 err = got_object_open_as_tag(&tag, repo, &qid->id);
1004 /* should not happen */
1005 err = got_error(GOT_ERR_OBJ_TYPE);
1009 /* Find a tree object to scan. */
1011 tree_id = got_object_commit_get_tree_id(commit);
1013 obj_type = got_object_tag_get_object_type(tag);
1015 case GOT_OBJ_TYPE_COMMIT:
1016 err = got_object_open_as_commit(&commit, repo,
1017 got_object_tag_get_object_id(tag));
1020 tree_id = got_object_commit_get_tree_id(commit);
1022 case GOT_OBJ_TYPE_TREE:
1023 tree_id = got_object_tag_get_object_id(tag);
1027 * Tag points at something other than a
1028 * commit or tree. Leave this weird tag object
1029 * and the object it points to on disk.
1031 err = got_object_idset_remove(NULL, loose_ids,
1033 if (err && err->code != GOT_ERR_NO_OBJ)
1035 err = got_object_idset_remove(NULL, loose_ids,
1036 got_object_tag_get_object_id(tag));
1037 if (err && err->code != GOT_ERR_NO_OBJ)
1045 err = load_tree(loose_ids, traversed_ids, tree_id, "",
1046 repo, npacked, cancel_cb, cancel_arg);
1052 (*ncommits)++; /* scanned tags are counted as commits */
1054 err = report_cleanup_progress(progress_cb, progress_arg, rl,
1055 nloose, *ncommits, -1);
1060 /* Find parent commits to scan. */
1061 const struct got_object_id_queue *parent_ids;
1062 parent_ids = got_object_commit_get_parent_ids(commit);
1063 err = got_object_id_queue_copy(parent_ids, &ids);
1066 got_object_commit_close(commit);
1070 got_object_tag_close(tag);
1073 got_object_qid_free(qid);
1078 got_object_qid_free(qid);
1080 got_object_commit_close(commit);
1082 got_object_tag_close(tag);
1083 got_object_id_queue_free(&ids);
1087 struct purge_loose_object_arg {
1088 struct got_repository *repo;
1089 got_cleanup_progress_cb progress_cb;
1091 struct got_ratelimit *rl;
1101 static const struct got_error *
1102 purge_loose_object(struct got_object_id *id, void *data, void *arg)
1104 struct purge_loose_object_arg *a = arg;
1105 const struct got_error *err, *unlock_err = NULL;
1109 struct got_lockfile *lf = NULL;
1111 err = got_object_get_path(&path, id, a->repo);
1115 err = got_object_open_loose_fd(&fd, id, a->repo);
1119 if (fstat(fd, &sb) == -1) {
1120 err = got_error_from_errno("fstat");
1125 * Do not delete objects which are younger than our maximum
1126 * modification time threshold. This prevents a race where
1127 * new objects which are being added to the repository
1128 * concurrently would be deleted.
1130 if (a->ignore_mtime || sb.st_mtime <= a->max_mtime) {
1132 err = got_lockfile_lock(&lf, path, -1);
1135 if (unlink(path) == -1) {
1136 err = got_error_from_errno2("unlink", path);
1142 a->size_purged += sb.st_size;
1143 err = report_cleanup_progress(a->progress_cb, a->progress_arg,
1144 a->rl, a->nloose, a->ncommits, a->npurged);
1149 if (fd != -1 && close(fd) == -1 && err == NULL)
1150 err = got_error_from_errno("close");
1153 unlock_err = got_lockfile_unlock(lf, -1);
1154 return err ? err : unlock_err;
1157 const struct got_error *
1158 got_repo_purge_unreferenced_loose_objects(struct got_repository *repo,
1159 off_t *size_before, off_t *size_after, int *npacked, int dry_run,
1160 int ignore_mtime, got_cleanup_progress_cb progress_cb, void *progress_arg,
1161 got_cancel_cb cancel_cb, void *cancel_arg)
1163 const struct got_error *err;
1164 struct got_object_idset *loose_ids;
1165 struct got_object_idset *traversed_ids;
1166 struct got_object_id **referenced_ids;
1167 int i, nreferenced, nloose, ncommits = 0;
1168 struct got_reflist_head refs;
1169 struct got_reflist_entry *re;
1170 struct purge_loose_object_arg arg;
1171 time_t max_mtime = 0;
1172 struct got_ratelimit rl;
1175 got_ratelimit_init(&rl, 0, 500);
1181 err = get_loose_object_ids(&loose_ids, size_before,
1182 progress_cb, progress_arg, &rl, repo);
1185 nloose = got_object_idset_num_elements(loose_ids);
1187 got_object_idset_free(loose_ids);
1189 err = progress_cb(progress_arg, 0, 0, 0, -1);
1196 traversed_ids = got_object_idset_alloc();
1197 if (traversed_ids == NULL) {
1198 err = got_error_from_errno("got_object_idset_alloc");
1202 err = got_ref_list(&refs, repo, "", got_ref_cmp_by_name, NULL);
1205 if (!ignore_mtime) {
1206 TAILQ_FOREACH(re, &refs, entry) {
1207 time_t mtime = got_ref_get_mtime(re->ref);
1208 if (mtime > max_mtime)
1212 * For safety, keep objects created within 10 minutes
1213 * before the youngest reference was created.
1215 if (max_mtime >= 600)
1219 err = get_reflist_object_ids(&referenced_ids, &nreferenced,
1220 (1 << GOT_OBJ_TYPE_COMMIT) | (1 << GOT_OBJ_TYPE_TAG),
1221 &refs, repo, cancel_cb, cancel_arg);
1225 for (i = 0; i < nreferenced; i++) {
1226 struct got_object_id *id = referenced_ids[i];
1227 err = load_commit_or_tag(loose_ids, &ncommits, npacked,
1228 traversed_ids, id, repo, progress_cb, progress_arg, &rl,
1229 nloose, cancel_cb, cancel_arg);
1234 /* Any remaining loose objects are unreferenced and can be purged. */
1236 arg.progress_arg = progress_arg;
1237 arg.progress_cb = progress_cb;
1239 arg.nloose = nloose;
1241 arg.size_purged = 0;
1242 arg.ncommits = ncommits;
1243 arg.dry_run = dry_run;
1244 arg.max_mtime = max_mtime;
1245 arg.ignore_mtime = ignore_mtime;
1246 err = got_object_idset_for_each(loose_ids, purge_loose_object, &arg);
1249 *size_after = *size_before - arg.size_purged;
1251 /* Produce a final progress report. */
1253 err = progress_cb(progress_arg, nloose, ncommits, arg.npurged,
1259 got_object_idset_free(loose_ids);
1260 got_object_idset_free(traversed_ids);
1264 static const struct got_error *
1265 purge_redundant_pack(struct got_repository *repo, const char *packidx_path,
1266 int dry_run, int *remove, off_t *size_before, off_t *size_after)
1268 static const char *ext[] = {".idx", ".pack", ".rev", ".bitmap",
1269 ".promisor", ".mtimes"};
1271 char *dot, path[PATH_MAX];
1274 if (strlcpy(path, packidx_path, sizeof(path)) >=
1276 return got_error(GOT_ERR_NO_SPACE);
1279 * For compatibility with Git, if a matching .keep file exist
1280 * don't delete the packfile.
1282 dot = strrchr(path, '.');
1284 if (strlcat(path, ".keep", sizeof(path)) >= sizeof(path))
1285 return got_error(GOT_ERR_NO_SPACE);
1286 if (faccessat(got_repo_get_fd(repo), path, F_OK, 0) == 0)
1289 for (i = 0; i < nitems(ext); ++i) {
1292 if (strlcat(path, ext[i], sizeof(path)) >=
1294 return got_error(GOT_ERR_NO_SPACE);
1296 if (fstatat(got_repo_get_fd(repo), path, &sb, 0) ==
1298 if (errno == ENOENT &&
1299 strcmp(ext[i], ".pack") != 0 &&
1300 strcmp(ext[i], ".idx") != 0)
1302 return got_error_from_errno2("fstatat", path);
1305 *size_before += sb.st_size;
1307 *size_after += sb.st_size;
1314 if (unlinkat(got_repo_get_fd(repo), path, 0) == -1) {
1315 if (errno == ENOENT)
1317 return got_error_from_errno2("unlinkat",
1325 static const struct got_error *
1326 pack_is_redundant(int *redundant, struct got_repository *repo,
1327 const char *packidx_path, struct got_object_idset *idset)
1329 const struct got_error *err;
1330 struct got_packidx *packidx;
1331 struct got_packidx_object_id *pid;
1332 struct got_object_id id;
1337 err = got_repo_get_packidx(&packidx, packidx_path, repo);
1341 nobjects = be32toh(packidx->hdr.fanout_table[0xff]);
1342 for (i = 0; i < nobjects; ++i) {
1343 pid = &packidx->hdr.sorted_ids[i];
1345 memset(&id, 0, sizeof(id));
1346 memcpy(&id.sha1, pid->sha1, sizeof(id.sha1));
1348 if (got_object_idset_contains(idset, &id))
1352 err = got_object_idset_add(idset, &id, NULL);
1366 pack_info_cmp(const void *a, const void *b)
1368 const struct pack_info *pa, *pb;
1372 if (pa->nobjects == pb->nobjects)
1373 return strcmp(pa->path, pb->path);
1374 if (pa->nobjects > pb->nobjects)
1379 const struct got_error *
1380 got_repo_purge_redundant_packfiles(struct got_repository *repo,
1381 off_t *size_before, off_t *size_after, int dry_run,
1382 got_cleanup_progress_cb progress_cb, void *progress_arg,
1383 got_cancel_cb cancel_cb, void *cancel_arg)
1385 const struct got_error *err;
1386 struct pack_info *pinfo, *sorted = NULL;
1387 struct got_packidx *packidx;
1388 struct got_object_idset *idset = NULL;
1389 struct got_pathlist_entry *pe;
1391 int remove, redundant_packs = 0;
1397 TAILQ_FOREACH(pe, &repo->packidx_paths, entry)
1403 sorted = calloc(npacks, sizeof(*sorted));
1405 return got_error_from_errno("calloc");
1408 TAILQ_FOREACH(pe, &repo->packidx_paths, entry) {
1409 err = got_repo_get_packidx(&packidx, pe->path, repo);
1413 pinfo = &sorted[i++];
1414 pinfo->path = pe->path;
1415 pinfo->nobjects = be32toh(packidx->hdr.fanout_table[0xff]);
1417 qsort(sorted, npacks, sizeof(*sorted), pack_info_cmp);
1419 idset = got_object_idset_alloc();
1420 if (idset == NULL) {
1421 err = got_error_from_errno("got_object_idset_alloc");
1425 for (i = 0; i < npacks; ++i) {
1427 err = (*cancel_cb)(cancel_arg);
1432 err = pack_is_redundant(&remove, repo, sorted[i].path, idset);
1435 err = purge_redundant_pack(repo, sorted[i].path, dry_run,
1436 &remove, size_before, size_after);
1441 err = progress_cb(progress_arg, -1, -1, -1,
1447 err = progress_cb(progress_arg, -1, -1, -1, redundant_packs);
1451 got_object_idset_free(idset);
1455 const struct got_error *
1456 got_repo_remove_lonely_packidx(struct got_repository *repo, int dry_run,
1457 got_lonely_packidx_progress_cb progress_cb, void *progress_arg,
1458 got_cancel_cb cancel_cb, void *cancel_arg)
1460 const struct got_error *err = NULL;
1461 DIR *packdir = NULL;
1462 struct dirent *dent;
1463 char *pack_relpath = NULL;
1467 packdir_fd = openat(got_repo_get_fd(repo),
1468 GOT_OBJECTS_PACK_DIR, O_DIRECTORY | O_CLOEXEC);
1469 if (packdir_fd == -1) {
1470 if (errno == ENOENT)
1472 return got_error_from_errno_fmt("openat: %s/%s",
1473 got_repo_get_path_git_dir(repo),
1474 GOT_OBJECTS_PACK_DIR);
1477 packdir = fdopendir(packdir_fd);
1478 if (packdir == NULL) {
1479 err = got_error_from_errno("fdopendir");
1483 while ((dent = readdir(packdir)) != NULL) {
1485 err = cancel_cb(cancel_arg);
1490 if (!got_repo_is_packidx_filename(dent->d_name,
1491 strlen(dent->d_name)))
1494 err = got_packidx_get_packfile_path(&pack_relpath,
1499 if (fstatat(packdir_fd, pack_relpath, &sb, 0) != -1) {
1501 pack_relpath = NULL;
1504 if (errno != ENOENT) {
1505 err = got_error_from_errno_fmt("fstatat: %s/%s/%s",
1506 got_repo_get_path_git_dir(repo),
1507 GOT_OBJECTS_PACK_DIR,
1513 if (unlinkat(packdir_fd, dent->d_name, 0) == -1) {
1514 err = got_error_from_errno("unlinkat");
1520 if (asprintf(&path, "%s/%s/%s",
1521 got_repo_get_path_git_dir(repo),
1522 GOT_OBJECTS_PACK_DIR,
1523 dent->d_name) == -1) {
1524 err = got_error_from_errno("asprintf");
1527 err = progress_cb(progress_arg, path);
1533 pack_relpath = NULL;
1536 if (packdir && closedir(packdir) != 0 && err == NULL)
1537 err = got_error_from_errno("closedir");