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 <sys/types.h>
18 #include <sys/queue.h>
22 #include <sys/socket.h>
39 #include "got_error.h"
40 #include "got_cancel.h"
41 #include "got_object.h"
42 #include "got_reference.h"
43 #include "got_repository.h"
44 #include "got_repository_admin.h"
45 #include "got_opentemp.h"
48 #include "got_lib_delta.h"
49 #include "got_lib_object.h"
50 #include "got_lib_object_idset.h"
51 #include "got_lib_object_cache.h"
52 #include "got_lib_pack.h"
53 #include "got_lib_privsep.h"
54 #include "got_lib_repository.h"
55 #include "got_lib_ratelimit.h"
56 #include "got_lib_pack_create.h"
57 #include "got_lib_hash.h"
58 #include "got_lib_lockfile.h"
61 #define nitems(_a) (sizeof((_a)) / sizeof((_a)[0]))
64 static const struct got_error *
65 get_reflist_object_ids(struct got_object_id ***ids, int *nobjects,
66 unsigned int wanted_obj_type_mask, struct got_reflist_head *refs,
67 struct got_repository *repo,
68 got_cancel_cb cancel_cb, void *cancel_arg)
70 const struct got_error *err = NULL;
71 const size_t alloc_chunksz = 256;
73 struct got_reflist_entry *re;
79 err = got_reflist_sort(refs,
80 got_ref_cmp_by_commit_timestamp_descending, repo);
84 *ids = reallocarray(NULL, alloc_chunksz, sizeof(struct got_object_id *));
86 return got_error_from_errno("reallocarray");
87 nalloc = alloc_chunksz;
89 TAILQ_FOREACH(re, refs, entry) {
90 struct got_object_id *id;
93 err = cancel_cb(cancel_arg);
98 err = got_ref_resolve(&id, repo, re->ref);
102 if (wanted_obj_type_mask != GOT_OBJ_TYPE_ANY) {
104 err = got_object_get_type(&obj_type, repo, id);
107 if ((wanted_obj_type_mask & (1 << obj_type)) == 0) {
114 if (nalloc <= *nobjects) {
115 struct got_object_id **new;
116 new = recallocarray(*ids, nalloc,
117 nalloc + alloc_chunksz,
118 sizeof(struct got_object_id *));
120 err = got_error_from_errno(
125 nalloc += alloc_chunksz;
127 (*ids)[*nobjects] = id;
128 if ((*ids)[*nobjects] == NULL) {
129 err = got_error_from_errno("got_object_id_dup");
136 for (i = 0; i < *nobjects; i++)
145 const struct got_error *
146 got_repo_pack_objects(FILE **packfile, struct got_object_id **pack_hash,
147 struct got_reflist_head *include_refs,
148 struct got_reflist_head *exclude_refs, struct got_repository *repo,
149 int loose_obj_only, int force_refdelta,
150 got_pack_progress_cb progress_cb, void *progress_arg,
151 got_cancel_cb cancel_cb, void *cancel_arg)
153 const struct got_error *err = NULL;
154 struct got_object_id **ours = NULL, **theirs = NULL;
155 int nours = 0, ntheirs = 0, packfd = -1, i;
156 char *tmpfile_path = NULL, *path = NULL, *packfile_path = NULL;
157 char *sha1_str = NULL;
158 FILE *delta_cache = NULL;
159 struct got_ratelimit rl;
164 got_ratelimit_init(&rl, 0, 500);
166 if (asprintf(&path, "%s/%s/packing.pack",
167 got_repo_get_path_git_dir(repo), GOT_OBJECTS_PACK_DIR) == -1) {
168 err = got_error_from_errno("asprintf");
171 err = got_opentemp_named_fd(&tmpfile_path, &packfd, path, "");
175 if (fchmod(packfd, GOT_DEFAULT_PACK_MODE) == -1) {
176 err = got_error_from_errno2("fchmod", tmpfile_path);
180 delta_cache = got_opentemp();
181 if (delta_cache == NULL) {
182 err = got_error_from_errno("got_opentemp");
186 err = get_reflist_object_ids(&ours, &nours,
187 (1 << GOT_OBJ_TYPE_COMMIT) | (1 << GOT_OBJ_TYPE_TAG),
188 include_refs, repo, cancel_cb, cancel_arg);
193 err = got_error(GOT_ERR_CANNOT_PACK);
197 if (!TAILQ_EMPTY(exclude_refs)) {
198 err = get_reflist_object_ids(&theirs, &ntheirs,
199 (1 << GOT_OBJ_TYPE_COMMIT) | (1 << GOT_OBJ_TYPE_TAG),
201 cancel_cb, cancel_arg);
206 *pack_hash = calloc(1, sizeof(**pack_hash));
207 if (*pack_hash == NULL) {
208 err = got_error_from_errno("calloc");
212 err = got_pack_create((*pack_hash)->sha1, packfd, delta_cache,
213 theirs, ntheirs, ours, nours, repo, loose_obj_only,
214 0, force_refdelta, progress_cb, progress_arg, &rl,
215 cancel_cb, cancel_arg);
219 err = got_object_id_str(&sha1_str, *pack_hash);
222 if (asprintf(&packfile_path, "%s/%s/pack-%s.pack",
223 got_repo_get_path_git_dir(repo), GOT_OBJECTS_PACK_DIR,
225 err = got_error_from_errno("asprintf");
229 if (lseek(packfd, 0L, SEEK_SET) == -1) {
230 err = got_error_from_errno("lseek");
233 if (rename(tmpfile_path, packfile_path) == -1) {
234 err = got_error_from_errno3("rename", tmpfile_path,
241 *packfile = fdopen(packfd, "w");
242 if (*packfile == NULL) {
243 err = got_error_from_errno2("fdopen", tmpfile_path);
248 for (i = 0; i < nours; i++)
251 for (i = 0; i < ntheirs; i++)
254 if (packfd != -1 && close(packfd) == -1 && err == NULL)
255 err = got_error_from_errno2("close", packfile_path);
256 if (delta_cache && fclose(delta_cache) == EOF && err == NULL)
257 err = got_error_from_errno("fclose");
258 if (tmpfile_path && unlink(tmpfile_path) == -1 && err == NULL)
259 err = got_error_from_errno2("unlink", tmpfile_path);
274 const struct got_error *
275 got_repo_index_pack(FILE *packfile, struct got_object_id *pack_hash,
276 struct got_repository *repo,
277 got_pack_index_progress_cb progress_cb, void *progress_arg,
278 got_cancel_cb cancel_cb, void *cancel_arg)
283 int npackfd = -1, idxfd = -1, nidxfd = -1;
285 int idxstatus, done = 0;
286 const struct got_error *err;
287 struct imsgbuf idxibuf;
289 char *tmpidxpath = NULL;
290 char *packfile_path = NULL, *idxpath = NULL, *id_str = NULL;
291 const char *repo_path = got_repo_get_path_git_dir(repo);
294 for (i = 0; i < nitems(tmpfds); i++)
297 if (asprintf(&path, "%s/%s/indexing.idx",
298 repo_path, GOT_OBJECTS_PACK_DIR) == -1) {
299 err = got_error_from_errno("asprintf");
302 err = got_opentemp_named_fd(&tmpidxpath, &idxfd, path, "");
306 if (fchmod(idxfd, GOT_DEFAULT_PACK_MODE) == -1) {
307 err = got_error_from_errno2("fchmod", tmpidxpath);
313 err = got_error_from_errno("dup");
317 for (i = 0; i < nitems(tmpfds); i++) {
318 tmpfds[i] = got_opentempfd();
319 if (tmpfds[i] == -1) {
320 err = got_error_from_errno("got_opentempfd");
325 err = got_object_id_str(&id_str, pack_hash);
329 if (asprintf(&packfile_path, "%s/%s/pack-%s.pack",
330 repo_path, GOT_OBJECTS_PACK_DIR, id_str) == -1) {
331 err = got_error_from_errno("asprintf");
335 if (fstat(fileno(packfile), &sb) == -1) {
336 err = got_error_from_errno2("fstat", packfile_path);
340 if (asprintf(&idxpath, "%s/%s/pack-%s.idx",
341 repo_path, GOT_OBJECTS_PACK_DIR, id_str) == -1) {
342 err = got_error_from_errno("asprintf");
346 if (socketpair(AF_UNIX, SOCK_STREAM, PF_UNSPEC, imsg_idxfds) == -1) {
347 err = got_error_from_errno("socketpair");
352 err= got_error_from_errno("fork");
354 } else if (idxpid == 0)
355 got_privsep_exec_child(imsg_idxfds,
356 GOT_PATH_PROG_INDEX_PACK, packfile_path);
357 if (close(imsg_idxfds[1]) == -1) {
358 err = got_error_from_errno("close");
361 imsg_init(&idxibuf, imsg_idxfds[0]);
363 npackfd = dup(fileno(packfile));
365 err = got_error_from_errno("dup");
368 err = got_privsep_send_index_pack_req(&idxibuf, pack_hash->sha1,
373 err = got_privsep_send_index_pack_outfd(&idxibuf, nidxfd);
377 for (i = 0; i < nitems(tmpfds); i++) {
378 err = got_privsep_send_tmpfd(&idxibuf, tmpfds[i]);
385 int nobj_total, nobj_indexed, nobj_loose, nobj_resolved;
388 err = cancel_cb(cancel_arg);
393 err = got_privsep_recv_index_progress(&done, &nobj_total,
394 &nobj_indexed, &nobj_loose, &nobj_resolved,
398 if (nobj_indexed != 0) {
399 err = progress_cb(progress_arg, sb.st_size,
400 nobj_total, nobj_indexed, nobj_loose,
406 if (close(imsg_idxfds[0]) == -1) {
407 err = got_error_from_errno("close");
410 if (waitpid(idxpid, &idxstatus, 0) == -1) {
411 err = got_error_from_errno("waitpid");
415 if (rename(tmpidxpath, idxpath) == -1) {
416 err = got_error_from_errno3("rename", tmpidxpath, idxpath);
423 if (tmpidxpath && unlink(tmpidxpath) == -1 && err == NULL)
424 err = got_error_from_errno2("unlink", tmpidxpath);
425 if (npackfd != -1 && close(npackfd) == -1 && err == NULL)
426 err = got_error_from_errno("close");
427 if (idxfd != -1 && close(idxfd) == -1 && err == NULL)
428 err = got_error_from_errno("close");
429 for (i = 0; i < nitems(tmpfds); i++) {
430 if (tmpfds[i] != -1 && close(tmpfds[i]) == -1 && err == NULL)
431 err = got_error_from_errno("close");
439 const struct got_error *
440 got_repo_find_pack(FILE **packfile, struct got_object_id **pack_hash,
441 struct got_repository *repo, const char *packfile_path)
443 const struct got_error *err = NULL;
444 const char *packdir_path = NULL;
445 char *packfile_name = NULL, *p, *dot;
446 struct got_object_id id;
452 packdir_path = got_repo_get_path_objects_pack(repo);
453 if (packdir_path == NULL)
454 return got_error_from_errno("got_repo_get_path_objects_pack");
456 if (!got_path_is_child(packfile_path, packdir_path,
457 strlen(packdir_path))) {
458 err = got_error_path(packfile_path, GOT_ERR_BAD_PATH);
463 err = got_path_basename(&packfile_name, packfile_path);
468 if (strncmp(p, "pack-", 5) != 0) {
469 err = got_error_fmt(GOT_ERR_BAD_PATH,
470 "'%s' is not a valid pack file name",
475 dot = strchr(p, '.');
477 err = got_error_fmt(GOT_ERR_BAD_PATH,
478 "'%s' is not a valid pack file name",
482 if (strcmp(dot + 1, "pack") != 0) {
483 err = got_error_fmt(GOT_ERR_BAD_PATH,
484 "'%s' is not a valid pack file name",
489 if (!got_parse_object_id(&id, p, repo->algo)) {
490 err = got_error_fmt(GOT_ERR_BAD_PATH,
491 "'%s' is not a valid pack file name",
496 *pack_hash = got_object_id_dup(&id);
497 if (*pack_hash == NULL) {
498 err = got_error_from_errno("got_object_id_dup");
502 packfd = open(packfile_path, O_RDONLY | O_NOFOLLOW | O_CLOEXEC);
504 err = got_error_from_errno2("open", packfile_path);
508 *packfile = fdopen(packfd, "r");
509 if (*packfile == NULL) {
510 err = got_error_from_errno2("fdopen", packfile_path);
515 if (packfd != -1 && close(packfd) == -1 && err == NULL)
516 err = got_error_from_errno2("close", packfile_path);
525 const struct got_error *
526 got_repo_list_pack(FILE *packfile, struct got_object_id *pack_hash,
527 struct got_repository *repo, got_pack_list_cb list_cb, void *list_arg,
528 got_cancel_cb cancel_cb, void *cancel_arg)
530 const struct got_error *err = NULL;
531 char *id_str = NULL, *idxpath = NULL, *packpath = NULL;
532 struct got_packidx *packidx = NULL;
533 struct got_pack *pack = NULL;
536 err = got_object_id_str(&id_str, pack_hash);
540 if (asprintf(&packpath, "%s/pack-%s.pack",
541 GOT_OBJECTS_PACK_DIR, id_str) == -1) {
542 err = got_error_from_errno("asprintf");
545 if (asprintf(&idxpath, "%s/pack-%s.idx",
546 GOT_OBJECTS_PACK_DIR, id_str) == -1) {
547 err = got_error_from_errno("asprintf");
551 err = got_packidx_open(&packidx, got_repo_get_fd(repo), idxpath, 1);
555 err = got_repo_cache_pack(&pack, repo, packpath, packidx);
559 nobj = be32toh(packidx->hdr.fanout_table[0xff]);
560 for (i = 0; i < nobj; i++) {
561 struct got_packidx_object_id *oid;
562 struct got_object_id id, base_id;
563 off_t offset, base_offset = 0;
569 err = cancel_cb(cancel_arg);
573 oid = &packidx->hdr.sorted_ids[i];
574 memcpy(id.sha1, oid->sha1, SHA1_DIGEST_LENGTH);
576 offset = got_packidx_get_object_offset(packidx, i);
578 err = got_error(GOT_ERR_BAD_PACKIDX);
582 err = got_pack_parse_object_type_and_size(&type, &size, &tslen,
588 case GOT_OBJ_TYPE_OFFSET_DELTA:
589 err = got_pack_parse_offset_delta(&base_offset, &len,
590 pack, offset, tslen);
594 case GOT_OBJ_TYPE_REF_DELTA:
595 err = got_pack_parse_ref_delta(&base_id,
596 pack, offset, tslen);
601 err = (*list_cb)(list_arg, &id, type, offset, size,
602 base_offset, &base_id);
612 got_packidx_close(packidx);
616 const struct got_error *
617 got_repo_cleanup_prepare(struct got_repository *repo,
618 struct got_lockfile **lk)
620 const struct got_error *err;
621 char myname[_POSIX_HOST_NAME_MAX + 1];
623 if (gethostname(myname, sizeof(myname)) == -1)
624 return got_error_from_errno("gethostname");
626 err = got_lockfile_lock(lk, "gc.pid", got_repo_get_fd(repo));
631 * Git uses these info to provide some verbiage when finds a
632 * lock during `git gc --force' so don't try too hard to avoid
633 * short writes and don't care if a race happens between the
634 * lockfile creation and the write itself.
636 if (dprintf((*lk)->fd, "%d %s", getpid(), myname) < 0)
637 return got_error_from_errno("dprintf");
642 const struct got_error *
643 got_repo_cleanup_complete(struct got_repository *repo,
644 struct got_lockfile *lk)
649 return got_lockfile_unlock(lk, got_repo_get_fd(repo));
652 static const struct got_error *
653 report_cleanup_progress(got_cleanup_progress_cb progress_cb,
654 void *progress_arg, struct got_ratelimit *rl,
655 int nloose, int ncommits, int npurged, int nredundant)
657 const struct got_error *err;
660 if (progress_cb == NULL)
663 err = got_ratelimit_check(&elapsed, rl);
667 return progress_cb(progress_arg, nloose, ncommits, npurged, nredundant);
670 static const struct got_error *
671 get_loose_object_ids(struct got_object_idset **loose_ids, off_t *ondisk_size,
672 got_cleanup_progress_cb progress_cb, void *progress_arg,
673 struct got_ratelimit *rl, struct got_repository *repo)
675 const struct got_error *err = NULL;
676 char *path_objects = NULL, *path = NULL;
678 struct got_object *obj = NULL;
679 struct got_object_id id;
684 *loose_ids = got_object_idset_alloc();
685 if (*loose_ids == NULL)
686 return got_error_from_errno("got_object_idset_alloc");
688 path_objects = got_repo_get_path_objects(repo);
689 if (path_objects == NULL) {
690 err = got_error_from_errno("got_repo_get_path_objects");
694 for (i = 0; i <= 0xff; i++) {
697 if (asprintf(&path, "%s/%.2x", path_objects, i) == -1) {
698 err = got_error_from_errno("asprintf");
704 if (errno == ENOENT) {
708 err = got_error_from_errno2("opendir", path);
712 while ((dent = readdir(dir)) != NULL) {
715 if (strcmp(dent->d_name, ".") == 0 ||
716 strcmp(dent->d_name, "..") == 0)
719 if (asprintf(&id_str, "%.2x%s", i, dent->d_name) == -1) {
720 err = got_error_from_errno("asprintf");
724 if (!got_parse_object_id(&id, id_str, repo->algo)) {
730 err = got_object_open_loose_fd(&fd, &id, repo);
733 if (fstat(fd, &sb) == -1) {
734 err = got_error_from_errno("fstat");
737 err = got_object_read_header_privsep(&obj, &id, repo,
741 fd = -1; /* already closed */
744 case GOT_OBJ_TYPE_COMMIT:
745 case GOT_OBJ_TYPE_TREE:
746 case GOT_OBJ_TYPE_BLOB:
747 case GOT_OBJ_TYPE_TAG:
750 err = got_error_fmt(GOT_ERR_OBJ_TYPE,
754 got_object_close(obj);
756 (*ondisk_size) += sb.st_size;
757 err = got_object_idset_add(*loose_ids, &id, NULL);
760 err = report_cleanup_progress(progress_cb,
762 got_object_idset_num_elements(*loose_ids),
768 if (closedir(dir) != 0) {
769 err = got_error_from_errno("closedir");
778 if (dir && closedir(dir) != 0 && err == NULL)
779 err = got_error_from_errno("closedir");
780 if (fd != -1 && close(fd) == -1 && err == NULL)
781 err = got_error_from_errno("close");
783 got_object_idset_free(*loose_ids);
787 got_object_close(obj);
793 static const struct got_error *
794 preserve_loose_object(struct got_object_idset *loose_ids,
795 struct got_object_id *id, struct got_repository *repo, int *npacked)
797 const struct got_error *err = NULL;
798 struct got_object *obj;
800 if (!got_object_idset_contains(loose_ids, id))
804 * Try to open this object from a pack file. This ensures that
805 * we do in fact have a valid packed copy of the object. Otherwise
806 * we should not delete the loose representation of this object.
808 err = got_object_open_packed(&obj, id, repo);
810 got_object_close(obj);
812 * The object is referenced and packed.
813 * We can purge the redundantly stored loose object.
817 } else if (err->code != GOT_ERR_NO_OBJ)
821 * This object is referenced and not packed.
822 * Remove it from our purge set.
824 return got_object_idset_remove(NULL, loose_ids, id);
827 static const struct got_error *
828 load_tree_entries(struct got_object_id_queue *ids,
829 struct got_object_idset *loose_ids,
830 struct got_object_idset *traversed_ids, struct got_object_id *tree_id,
831 const char *dpath, struct got_repository *repo, int *npacked,
832 got_cancel_cb cancel_cb, void *cancel_arg)
834 const struct got_error *err;
835 struct got_tree_object *tree;
839 err = got_object_open_as_tree(&tree, repo, tree_id);
843 for (i = 0; i < got_object_tree_get_nentries(tree); i++) {
844 struct got_tree_entry *e = got_object_tree_get_entry(tree, i);
845 struct got_object_id *id = got_tree_entry_get_id(e);
846 mode_t mode = got_tree_entry_get_mode(e);
849 err = (*cancel_cb)(cancel_arg);
854 if (got_object_tree_entry_is_symlink(e) ||
855 got_object_tree_entry_is_submodule(e) ||
856 got_object_idset_contains(traversed_ids, id))
859 if (asprintf(&p, "%s%s%s", dpath, dpath[0] != '\0' ? "/" : "",
860 got_tree_entry_get_name(e)) == -1) {
861 err = got_error_from_errno("asprintf");
866 struct got_object_qid *qid;
867 err = got_object_qid_alloc(&qid, id);
870 STAILQ_INSERT_TAIL(ids, qid, entry);
871 } else if (S_ISREG(mode)) {
872 /* This blob is referenced. */
873 err = preserve_loose_object(loose_ids, id, repo,
877 err = got_object_idset_add(traversed_ids, id, NULL);
885 got_object_tree_close(tree);
890 static const struct got_error *
891 load_tree(struct got_object_idset *loose_ids,
892 struct got_object_idset *traversed_ids, struct got_object_id *tree_id,
893 const char *dpath, struct got_repository *repo, int *npacked,
894 got_cancel_cb cancel_cb, void *cancel_arg)
896 const struct got_error *err = NULL;
897 struct got_object_id_queue tree_ids;
898 struct got_object_qid *qid;
900 err = got_object_qid_alloc(&qid, tree_id);
904 STAILQ_INIT(&tree_ids);
905 STAILQ_INSERT_TAIL(&tree_ids, qid, entry);
907 while (!STAILQ_EMPTY(&tree_ids)) {
909 err = (*cancel_cb)(cancel_arg);
914 qid = STAILQ_FIRST(&tree_ids);
915 STAILQ_REMOVE_HEAD(&tree_ids, entry);
917 if (got_object_idset_contains(traversed_ids, &qid->id)) {
918 got_object_qid_free(qid);
922 err = got_object_idset_add(traversed_ids, &qid->id, NULL);
924 got_object_qid_free(qid);
928 /* This tree is referenced. */
929 err = preserve_loose_object(loose_ids, &qid->id, repo, npacked);
933 err = load_tree_entries(&tree_ids, loose_ids, traversed_ids,
934 &qid->id, dpath, repo, npacked, cancel_cb, cancel_arg);
935 got_object_qid_free(qid);
940 got_object_id_queue_free(&tree_ids);
944 static const struct got_error *
945 load_commit_or_tag(struct got_object_idset *loose_ids, int *ncommits,
946 int *npacked, struct got_object_idset *traversed_ids,
947 struct got_object_id *id, struct got_repository *repo,
948 got_cleanup_progress_cb progress_cb, void *progress_arg,
949 struct got_ratelimit *rl, int nloose,
950 got_cancel_cb cancel_cb, void *cancel_arg)
952 const struct got_error *err;
953 struct got_commit_object *commit = NULL;
954 struct got_tag_object *tag = NULL;
955 struct got_object_id *tree_id = NULL;
956 struct got_object_id_queue ids;
957 struct got_object_qid *qid;
960 err = got_object_qid_alloc(&qid, id);
965 STAILQ_INSERT_TAIL(&ids, qid, entry);
967 while (!STAILQ_EMPTY(&ids)) {
969 err = (*cancel_cb)(cancel_arg);
974 qid = STAILQ_FIRST(&ids);
975 STAILQ_REMOVE_HEAD(&ids, entry);
977 if (got_object_idset_contains(traversed_ids, &qid->id)) {
978 got_object_qid_free(qid);
983 err = got_object_idset_add(traversed_ids, &qid->id, NULL);
987 /* This commit or tag is referenced. */
988 err = preserve_loose_object(loose_ids, &qid->id, repo, npacked);
992 err = got_object_get_type(&obj_type, repo, &qid->id);
996 case GOT_OBJ_TYPE_COMMIT:
997 err = got_object_open_as_commit(&commit, repo,
1002 case GOT_OBJ_TYPE_TAG:
1003 err = got_object_open_as_tag(&tag, repo, &qid->id);
1008 /* should not happen */
1009 err = got_error(GOT_ERR_OBJ_TYPE);
1013 /* Find a tree object to scan. */
1015 tree_id = got_object_commit_get_tree_id(commit);
1017 obj_type = got_object_tag_get_object_type(tag);
1019 case GOT_OBJ_TYPE_COMMIT:
1020 err = got_object_open_as_commit(&commit, repo,
1021 got_object_tag_get_object_id(tag));
1024 tree_id = got_object_commit_get_tree_id(commit);
1026 case GOT_OBJ_TYPE_TREE:
1027 tree_id = got_object_tag_get_object_id(tag);
1031 * Tag points at something other than a
1032 * commit or tree. Leave this weird tag object
1033 * and the object it points to on disk.
1035 err = got_object_idset_remove(NULL, loose_ids,
1037 if (err && err->code != GOT_ERR_NO_OBJ)
1039 err = got_object_idset_remove(NULL, loose_ids,
1040 got_object_tag_get_object_id(tag));
1041 if (err && err->code != GOT_ERR_NO_OBJ)
1049 err = load_tree(loose_ids, traversed_ids, tree_id, "",
1050 repo, npacked, cancel_cb, cancel_arg);
1056 (*ncommits)++; /* scanned tags are counted as commits */
1058 err = report_cleanup_progress(progress_cb, progress_arg, rl,
1059 nloose, *ncommits, -1, -1);
1064 /* Find parent commits to scan. */
1065 const struct got_object_id_queue *parent_ids;
1066 parent_ids = got_object_commit_get_parent_ids(commit);
1067 err = got_object_id_queue_copy(parent_ids, &ids);
1070 got_object_commit_close(commit);
1074 got_object_tag_close(tag);
1077 got_object_qid_free(qid);
1082 got_object_qid_free(qid);
1084 got_object_commit_close(commit);
1086 got_object_tag_close(tag);
1087 got_object_id_queue_free(&ids);
1091 struct purge_loose_object_arg {
1092 struct got_repository *repo;
1093 got_cleanup_progress_cb progress_cb;
1095 struct got_ratelimit *rl;
1105 static const struct got_error *
1106 purge_loose_object(struct got_object_id *id, void *data, void *arg)
1108 struct purge_loose_object_arg *a = arg;
1109 const struct got_error *err, *unlock_err = NULL;
1113 struct got_lockfile *lf = NULL;
1115 err = got_object_get_path(&path, id, a->repo);
1119 err = got_object_open_loose_fd(&fd, id, a->repo);
1123 if (fstat(fd, &sb) == -1) {
1124 err = got_error_from_errno("fstat");
1129 * Do not delete objects which are younger than our maximum
1130 * modification time threshold. This prevents a race where
1131 * new objects which are being added to the repository
1132 * concurrently would be deleted.
1134 if (a->ignore_mtime || sb.st_mtime <= a->max_mtime) {
1136 err = got_lockfile_lock(&lf, path, -1);
1139 if (unlink(path) == -1) {
1140 err = got_error_from_errno2("unlink", path);
1146 a->size_purged += sb.st_size;
1147 err = report_cleanup_progress(a->progress_cb, a->progress_arg,
1148 a->rl, a->nloose, a->ncommits, a->npurged, -1);
1153 if (fd != -1 && close(fd) == -1 && err == NULL)
1154 err = got_error_from_errno("close");
1157 unlock_err = got_lockfile_unlock(lf, -1);
1158 return err ? err : unlock_err;
1161 const struct got_error *
1162 got_repo_purge_unreferenced_loose_objects(struct got_repository *repo,
1163 off_t *size_before, off_t *size_after, int *ncommits, int *nloose,
1164 int *npacked, int dry_run, int ignore_mtime,
1165 got_cleanup_progress_cb progress_cb, void *progress_arg,
1166 got_cancel_cb cancel_cb, void *cancel_arg)
1168 const struct got_error *err;
1169 struct got_object_idset *loose_ids;
1170 struct got_object_idset *traversed_ids;
1171 struct got_object_id **referenced_ids;
1173 struct got_reflist_head refs;
1174 struct got_reflist_entry *re;
1175 struct purge_loose_object_arg arg;
1176 time_t max_mtime = 0;
1177 struct got_ratelimit rl;
1180 got_ratelimit_init(&rl, 0, 500);
1186 err = get_loose_object_ids(&loose_ids, size_before,
1187 progress_cb, progress_arg, &rl, repo);
1190 *nloose = got_object_idset_num_elements(loose_ids);
1192 got_object_idset_free(loose_ids);
1194 err = progress_cb(progress_arg, 0, 0, 0, -1);
1201 traversed_ids = got_object_idset_alloc();
1202 if (traversed_ids == NULL) {
1203 err = got_error_from_errno("got_object_idset_alloc");
1207 err = got_ref_list(&refs, repo, "", got_ref_cmp_by_name, NULL);
1210 if (!ignore_mtime) {
1211 TAILQ_FOREACH(re, &refs, entry) {
1212 time_t mtime = got_ref_get_mtime(re->ref);
1213 if (mtime > max_mtime)
1217 * For safety, keep objects created within 10 minutes
1218 * before the youngest reference was created.
1220 if (max_mtime >= 600)
1224 err = get_reflist_object_ids(&referenced_ids, &nreferenced,
1225 (1 << GOT_OBJ_TYPE_COMMIT) | (1 << GOT_OBJ_TYPE_TAG),
1226 &refs, repo, cancel_cb, cancel_arg);
1230 for (i = 0; i < nreferenced; i++) {
1231 struct got_object_id *id = referenced_ids[i];
1232 err = load_commit_or_tag(loose_ids, ncommits, npacked,
1233 traversed_ids, id, repo, progress_cb, progress_arg, &rl,
1234 *nloose, cancel_cb, cancel_arg);
1239 /* Any remaining loose objects are unreferenced and can be purged. */
1241 arg.progress_arg = progress_arg;
1242 arg.progress_cb = progress_cb;
1244 arg.nloose = *nloose;
1246 arg.size_purged = 0;
1247 arg.ncommits = *ncommits;
1248 arg.dry_run = dry_run;
1249 arg.max_mtime = max_mtime;
1250 arg.ignore_mtime = ignore_mtime;
1251 err = got_object_idset_for_each(loose_ids, purge_loose_object, &arg);
1254 *size_after = *size_before - arg.size_purged;
1256 /* Produce a final progress report. */
1258 err = progress_cb(progress_arg, *nloose, *ncommits, arg.npurged,
1264 got_object_idset_free(loose_ids);
1265 got_object_idset_free(traversed_ids);
1269 static const struct got_error *
1270 purge_redundant_pack(struct got_repository *repo, const char *packidx_path,
1271 int dry_run, int *remove, off_t *size_before, off_t *size_after)
1273 static const char *ext[] = {".idx", ".pack", ".rev", ".bitmap",
1274 ".promisor", ".mtimes"};
1276 char *dot, path[PATH_MAX];
1279 if (strlcpy(path, packidx_path, sizeof(path)) >= sizeof(path))
1280 return got_error(GOT_ERR_NO_SPACE);
1283 * For compatibility with Git, if a matching .keep file exist
1284 * don't delete the packfile.
1286 dot = strrchr(path, '.');
1288 if (strlcat(path, ".keep", sizeof(path)) >= sizeof(path))
1289 return got_error(GOT_ERR_NO_SPACE);
1290 if (faccessat(got_repo_get_fd(repo), path, F_OK, 0) == 0)
1293 for (i = 0; i < nitems(ext); ++i) {
1296 if (strlcat(path, ext[i], sizeof(path)) >=
1298 return got_error(GOT_ERR_NO_SPACE);
1300 if (fstatat(got_repo_get_fd(repo), path, &sb, 0) ==
1302 if (errno == ENOENT &&
1303 strcmp(ext[i], ".pack") != 0 &&
1304 strcmp(ext[i], ".idx") != 0)
1306 return got_error_from_errno2("fstatat", path);
1309 *size_before += sb.st_size;
1311 *size_after += sb.st_size;
1318 if (unlinkat(got_repo_get_fd(repo), path, 0) == -1) {
1319 if (errno == ENOENT)
1321 return got_error_from_errno2("unlinkat",
1329 static const struct got_error *
1330 pack_is_redundant(int *redundant, struct got_repository *repo,
1331 const char *packidx_path, struct got_object_idset *idset)
1333 const struct got_error *err;
1334 struct got_packidx *packidx;
1335 struct got_packidx_object_id *pid;
1336 struct got_object_id id;
1341 err = got_repo_get_packidx(&packidx, packidx_path, repo);
1345 nobjects = be32toh(packidx->hdr.fanout_table[0xff]);
1346 for (i = 0; i < nobjects; ++i) {
1347 pid = &packidx->hdr.sorted_ids[i];
1349 memset(&id, 0, sizeof(id));
1350 memcpy(&id.sha1, pid->sha1, sizeof(id.sha1));
1352 if (got_object_idset_contains(idset, &id))
1356 err = got_object_idset_add(idset, &id, NULL);
1370 pack_info_cmp(const void *a, const void *b)
1372 const struct pack_info *pa, *pb;
1376 if (pa->nobjects == pb->nobjects)
1377 return strcmp(pa->path, pb->path);
1378 if (pa->nobjects > pb->nobjects)
1383 const struct got_error *
1384 got_repo_purge_redundant_packfiles(struct got_repository *repo,
1385 off_t *size_before, off_t *size_after, int dry_run,
1386 int nloose, int ncommits, int npurged,
1387 got_cleanup_progress_cb progress_cb, void *progress_arg,
1388 got_cancel_cb cancel_cb, void *cancel_arg)
1390 const struct got_error *err;
1391 struct pack_info *pinfo, *sorted = NULL;
1392 struct got_packidx *packidx;
1393 struct got_object_idset *idset = NULL;
1394 struct got_pathlist_entry *pe;
1396 int remove, redundant_packs = 0;
1397 struct got_ratelimit rl;
1399 got_ratelimit_init(&rl, 0, 500);
1405 TAILQ_FOREACH(pe, &repo->packidx_paths, entry)
1411 sorted = calloc(npacks, sizeof(*sorted));
1413 return got_error_from_errno("calloc");
1416 TAILQ_FOREACH(pe, &repo->packidx_paths, entry) {
1417 err = got_repo_get_packidx(&packidx, pe->path, repo);
1421 pinfo = &sorted[i++];
1422 pinfo->path = pe->path;
1423 pinfo->nobjects = be32toh(packidx->hdr.fanout_table[0xff]);
1425 qsort(sorted, npacks, sizeof(*sorted), pack_info_cmp);
1427 idset = got_object_idset_alloc();
1428 if (idset == NULL) {
1429 err = got_error_from_errno("got_object_idset_alloc");
1433 for (i = 0; i < npacks; ++i) {
1435 err = (*cancel_cb)(cancel_arg);
1440 err = pack_is_redundant(&remove, repo, sorted[i].path, idset);
1443 err = purge_redundant_pack(repo, sorted[i].path, dry_run,
1444 &remove, size_before, size_after);
1449 err = report_cleanup_progress(progress_cb, progress_arg,
1450 &rl, nloose, ncommits, npurged, ++redundant_packs);
1455 /* Produce a final progress report. */
1457 err = progress_cb(progress_arg, nloose, ncommits, npurged,
1465 got_object_idset_free(idset);
1469 const struct got_error *
1470 got_repo_remove_lonely_packidx(struct got_repository *repo, int dry_run,
1471 got_lonely_packidx_progress_cb progress_cb, void *progress_arg,
1472 got_cancel_cb cancel_cb, void *cancel_arg)
1474 const struct got_error *err = NULL;
1475 DIR *packdir = NULL;
1476 struct dirent *dent;
1477 char *pack_relpath = NULL;
1481 packdir_fd = openat(got_repo_get_fd(repo),
1482 GOT_OBJECTS_PACK_DIR, O_DIRECTORY | O_CLOEXEC);
1483 if (packdir_fd == -1) {
1484 if (errno == ENOENT)
1486 return got_error_from_errno_fmt("openat: %s/%s",
1487 got_repo_get_path_git_dir(repo),
1488 GOT_OBJECTS_PACK_DIR);
1491 packdir = fdopendir(packdir_fd);
1492 if (packdir == NULL) {
1493 err = got_error_from_errno("fdopendir");
1497 while ((dent = readdir(packdir)) != NULL) {
1499 err = cancel_cb(cancel_arg);
1504 if (!got_repo_is_packidx_filename(dent->d_name, dent->d_namlen))
1507 err = got_packidx_get_packfile_path(&pack_relpath,
1512 if (fstatat(packdir_fd, pack_relpath, &sb, 0) != -1) {
1514 pack_relpath = NULL;
1517 if (errno != ENOENT) {
1518 err = got_error_from_errno_fmt("fstatat: %s/%s/%s",
1519 got_repo_get_path_git_dir(repo),
1520 GOT_OBJECTS_PACK_DIR,
1526 if (unlinkat(packdir_fd, dent->d_name, 0) == -1) {
1527 err = got_error_from_errno("unlinkat");
1533 if (asprintf(&path, "%s/%s/%s",
1534 got_repo_get_path_git_dir(repo),
1535 GOT_OBJECTS_PACK_DIR,
1536 dent->d_name) == -1) {
1537 err = got_error_from_errno("asprintf");
1540 err = progress_cb(progress_arg, path);
1546 pack_relpath = NULL;
1549 if (packdir && closedir(packdir) != 0 && err == NULL)
1550 err = got_error_from_errno("closedir");