implement some resilience against pack corruptions
[git/jnareb-git/bp-gitweb.git] / builtin-reflog.c
blobb151e24ff94e06840db7e57014fe4a73ea811575
1 #include "cache.h"
2 #include "builtin.h"
3 #include "commit.h"
4 #include "refs.h"
5 #include "dir.h"
6 #include "tree-walk.h"
7 #include "diff.h"
8 #include "revision.h"
9 #include "reachable.h"
12 * reflog expire
15 static const char reflog_expire_usage[] =
16 "git-reflog (show|expire) [--verbose] [--dry-run] [--stale-fix] [--expire=<time>] [--expire-unreachable=<time>] [--all] <refs>...";
17 static const char reflog_delete_usage[] =
18 "git-reflog delete [--verbose] [--dry-run] [--rewrite] [--updateref] <refs>...";
20 static unsigned long default_reflog_expire;
21 static unsigned long default_reflog_expire_unreachable;
23 struct cmd_reflog_expire_cb {
24 struct rev_info revs;
25 int dry_run;
26 int stalefix;
27 int rewrite;
28 int updateref;
29 int verbose;
30 unsigned long expire_total;
31 unsigned long expire_unreachable;
32 int recno;
35 struct expire_reflog_cb {
36 FILE *newlog;
37 const char *ref;
38 struct commit *ref_commit;
39 struct cmd_reflog_expire_cb *cmd;
40 unsigned char last_kept_sha1[20];
43 struct collected_reflog {
44 unsigned char sha1[20];
45 char reflog[FLEX_ARRAY];
47 struct collect_reflog_cb {
48 struct collected_reflog **e;
49 int alloc;
50 int nr;
53 #define INCOMPLETE (1u<<10)
54 #define STUDYING (1u<<11)
56 static int tree_is_complete(const unsigned char *sha1)
58 struct tree_desc desc;
59 struct name_entry entry;
60 int complete;
61 struct tree *tree;
63 tree = lookup_tree(sha1);
64 if (!tree)
65 return 0;
66 if (tree->object.flags & SEEN)
67 return 1;
68 if (tree->object.flags & INCOMPLETE)
69 return 0;
71 if (!tree->buffer) {
72 enum object_type type;
73 unsigned long size;
74 void *data = read_sha1_file(sha1, &type, &size);
75 if (!data) {
76 tree->object.flags |= INCOMPLETE;
77 return 0;
79 tree->buffer = data;
80 tree->size = size;
82 init_tree_desc(&desc, tree->buffer, tree->size);
83 complete = 1;
84 while (tree_entry(&desc, &entry)) {
85 if (!has_sha1_file(entry.sha1) ||
86 (S_ISDIR(entry.mode) && !tree_is_complete(entry.sha1))) {
87 tree->object.flags |= INCOMPLETE;
88 complete = 0;
91 free(tree->buffer);
92 tree->buffer = NULL;
94 if (complete)
95 tree->object.flags |= SEEN;
96 return complete;
99 static int commit_is_complete(struct commit *commit)
101 struct object_array study;
102 struct object_array found;
103 int is_incomplete = 0;
104 int i;
106 /* early return */
107 if (commit->object.flags & SEEN)
108 return 1;
109 if (commit->object.flags & INCOMPLETE)
110 return 0;
112 * Find all commits that are reachable and are not marked as
113 * SEEN. Then make sure the trees and blobs contained are
114 * complete. After that, mark these commits also as SEEN.
115 * If some of the objects that are needed to complete this
116 * commit are missing, mark this commit as INCOMPLETE.
118 memset(&study, 0, sizeof(study));
119 memset(&found, 0, sizeof(found));
120 add_object_array(&commit->object, NULL, &study);
121 add_object_array(&commit->object, NULL, &found);
122 commit->object.flags |= STUDYING;
123 while (study.nr) {
124 struct commit *c;
125 struct commit_list *parent;
127 c = (struct commit *)study.objects[--study.nr].item;
128 if (!c->object.parsed && !parse_object(c->object.sha1))
129 c->object.flags |= INCOMPLETE;
131 if (c->object.flags & INCOMPLETE) {
132 is_incomplete = 1;
133 break;
135 else if (c->object.flags & SEEN)
136 continue;
137 for (parent = c->parents; parent; parent = parent->next) {
138 struct commit *p = parent->item;
139 if (p->object.flags & STUDYING)
140 continue;
141 p->object.flags |= STUDYING;
142 add_object_array(&p->object, NULL, &study);
143 add_object_array(&p->object, NULL, &found);
146 if (!is_incomplete) {
148 * make sure all commits in "found" array have all the
149 * necessary objects.
151 for (i = 0; i < found.nr; i++) {
152 struct commit *c =
153 (struct commit *)found.objects[i].item;
154 if (!tree_is_complete(c->tree->object.sha1)) {
155 is_incomplete = 1;
156 c->object.flags |= INCOMPLETE;
159 if (!is_incomplete) {
160 /* mark all found commits as complete, iow SEEN */
161 for (i = 0; i < found.nr; i++)
162 found.objects[i].item->flags |= SEEN;
165 /* clear flags from the objects we traversed */
166 for (i = 0; i < found.nr; i++)
167 found.objects[i].item->flags &= ~STUDYING;
168 if (is_incomplete)
169 commit->object.flags |= INCOMPLETE;
170 else {
172 * If we come here, we have (1) traversed the ancestry chain
173 * from the "commit" until we reach SEEN commits (which are
174 * known to be complete), and (2) made sure that the commits
175 * encountered during the above traversal refer to trees that
176 * are complete. Which means that we know *all* the commits
177 * we have seen during this process are complete.
179 for (i = 0; i < found.nr; i++)
180 found.objects[i].item->flags |= SEEN;
182 /* free object arrays */
183 free(study.objects);
184 free(found.objects);
185 return !is_incomplete;
188 static int keep_entry(struct commit **it, unsigned char *sha1)
190 struct commit *commit;
192 if (is_null_sha1(sha1))
193 return 1;
194 commit = lookup_commit_reference_gently(sha1, 1);
195 if (!commit)
196 return 0;
199 * Make sure everything in this commit exists.
201 * We have walked all the objects reachable from the refs
202 * and cache earlier. The commits reachable by this commit
203 * must meet SEEN commits -- and then we should mark them as
204 * SEEN as well.
206 if (!commit_is_complete(commit))
207 return 0;
208 *it = commit;
209 return 1;
212 static int expire_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
213 const char *email, unsigned long timestamp, int tz,
214 const char *message, void *cb_data)
216 struct expire_reflog_cb *cb = cb_data;
217 struct commit *old, *new;
219 if (timestamp < cb->cmd->expire_total)
220 goto prune;
222 if (cb->cmd->rewrite)
223 osha1 = cb->last_kept_sha1;
225 old = new = NULL;
226 if (cb->cmd->stalefix &&
227 (!keep_entry(&old, osha1) || !keep_entry(&new, nsha1)))
228 goto prune;
230 if (timestamp < cb->cmd->expire_unreachable) {
231 if (!cb->ref_commit)
232 goto prune;
233 if (!old && !is_null_sha1(osha1))
234 old = lookup_commit_reference_gently(osha1, 1);
235 if (!new && !is_null_sha1(nsha1))
236 new = lookup_commit_reference_gently(nsha1, 1);
237 if ((old && !in_merge_bases(old, &cb->ref_commit, 1)) ||
238 (new && !in_merge_bases(new, &cb->ref_commit, 1)))
239 goto prune;
242 if (cb->cmd->recno && --(cb->cmd->recno) == 0)
243 goto prune;
245 if (cb->newlog) {
246 char sign = (tz < 0) ? '-' : '+';
247 int zone = (tz < 0) ? (-tz) : tz;
248 fprintf(cb->newlog, "%s %s %s %lu %c%04d\t%s",
249 sha1_to_hex(osha1), sha1_to_hex(nsha1),
250 email, timestamp, sign, zone,
251 message);
252 hashcpy(cb->last_kept_sha1, nsha1);
254 if (cb->cmd->verbose)
255 printf("keep %s", message);
256 return 0;
257 prune:
258 if (!cb->newlog || cb->cmd->verbose)
259 printf("%sprune %s", cb->newlog ? "" : "would ", message);
260 return 0;
263 static int expire_reflog(const char *ref, const unsigned char *sha1, int unused, void *cb_data)
265 struct cmd_reflog_expire_cb *cmd = cb_data;
266 struct expire_reflog_cb cb;
267 struct ref_lock *lock;
268 char *log_file, *newlog_path = NULL;
269 int status = 0;
271 memset(&cb, 0, sizeof(cb));
272 /* we take the lock for the ref itself to prevent it from
273 * getting updated.
275 lock = lock_any_ref_for_update(ref, sha1, 0);
276 if (!lock)
277 return error("cannot lock ref '%s'", ref);
278 log_file = xstrdup(git_path("logs/%s", ref));
279 if (!file_exists(log_file))
280 goto finish;
281 if (!cmd->dry_run) {
282 newlog_path = xstrdup(git_path("logs/%s.lock", ref));
283 cb.newlog = fopen(newlog_path, "w");
286 cb.ref_commit = lookup_commit_reference_gently(sha1, 1);
287 cb.ref = ref;
288 cb.cmd = cmd;
289 for_each_reflog_ent(ref, expire_reflog_ent, &cb);
290 finish:
291 if (cb.newlog) {
292 if (fclose(cb.newlog)) {
293 status |= error("%s: %s", strerror(errno),
294 newlog_path);
295 unlink(newlog_path);
296 } else if (cmd->updateref &&
297 (write_in_full(lock->lock_fd,
298 sha1_to_hex(cb.last_kept_sha1), 40) != 40 ||
299 write_in_full(lock->lock_fd, "\n", 1) != 1 ||
300 close_ref(lock) < 0)) {
301 status |= error("Couldn't write %s",
302 lock->lk->filename);
303 unlink(newlog_path);
304 } else if (rename(newlog_path, log_file)) {
305 status |= error("cannot rename %s to %s",
306 newlog_path, log_file);
307 unlink(newlog_path);
308 } else if (cmd->updateref && commit_ref(lock)) {
309 status |= error("Couldn't set %s", lock->ref_name);
310 } else {
311 adjust_shared_perm(log_file);
314 free(newlog_path);
315 free(log_file);
316 unlock_ref(lock);
317 return status;
320 static int collect_reflog(const char *ref, const unsigned char *sha1, int unused, void *cb_data)
322 struct collected_reflog *e;
323 struct collect_reflog_cb *cb = cb_data;
324 size_t namelen = strlen(ref);
326 e = xmalloc(sizeof(*e) + namelen + 1);
327 hashcpy(e->sha1, sha1);
328 memcpy(e->reflog, ref, namelen + 1);
329 ALLOC_GROW(cb->e, cb->nr + 1, cb->alloc);
330 cb->e[cb->nr++] = e;
331 return 0;
334 static int reflog_expire_config(const char *var, const char *value, void *cb)
336 if (!strcmp(var, "gc.reflogexpire")) {
337 if (!value)
338 config_error_nonbool(var);
339 default_reflog_expire = approxidate(value);
340 return 0;
342 if (!strcmp(var, "gc.reflogexpireunreachable")) {
343 if (!value)
344 config_error_nonbool(var);
345 default_reflog_expire_unreachable = approxidate(value);
346 return 0;
348 return git_default_config(var, value, cb);
351 static int cmd_reflog_expire(int argc, const char **argv, const char *prefix)
353 struct cmd_reflog_expire_cb cb;
354 unsigned long now = time(NULL);
355 int i, status, do_all;
357 git_config(reflog_expire_config, NULL);
359 save_commit_buffer = 0;
360 do_all = status = 0;
361 memset(&cb, 0, sizeof(cb));
363 if (!default_reflog_expire_unreachable)
364 default_reflog_expire_unreachable = now - 30 * 24 * 3600;
365 if (!default_reflog_expire)
366 default_reflog_expire = now - 90 * 24 * 3600;
367 cb.expire_total = default_reflog_expire;
368 cb.expire_unreachable = default_reflog_expire_unreachable;
371 * We can trust the commits and objects reachable from refs
372 * even in older repository. We cannot trust what's reachable
373 * from reflog if the repository was pruned with older git.
376 for (i = 1; i < argc; i++) {
377 const char *arg = argv[i];
378 if (!strcmp(arg, "--dry-run") || !strcmp(arg, "-n"))
379 cb.dry_run = 1;
380 else if (!prefixcmp(arg, "--expire="))
381 cb.expire_total = approxidate(arg + 9);
382 else if (!prefixcmp(arg, "--expire-unreachable="))
383 cb.expire_unreachable = approxidate(arg + 21);
384 else if (!strcmp(arg, "--stale-fix"))
385 cb.stalefix = 1;
386 else if (!strcmp(arg, "--rewrite"))
387 cb.rewrite = 1;
388 else if (!strcmp(arg, "--updateref"))
389 cb.updateref = 1;
390 else if (!strcmp(arg, "--all"))
391 do_all = 1;
392 else if (!strcmp(arg, "--verbose"))
393 cb.verbose = 1;
394 else if (!strcmp(arg, "--")) {
395 i++;
396 break;
398 else if (arg[0] == '-')
399 usage(reflog_expire_usage);
400 else
401 break;
403 if (cb.stalefix) {
404 init_revisions(&cb.revs, prefix);
405 if (cb.verbose)
406 printf("Marking reachable objects...");
407 mark_reachable_objects(&cb.revs, 0);
408 if (cb.verbose)
409 putchar('\n');
412 if (do_all) {
413 struct collect_reflog_cb collected;
414 int i;
416 memset(&collected, 0, sizeof(collected));
417 for_each_reflog(collect_reflog, &collected);
418 for (i = 0; i < collected.nr; i++) {
419 struct collected_reflog *e = collected.e[i];
420 status |= expire_reflog(e->reflog, e->sha1, 0, &cb);
421 free(e);
423 free(collected.e);
426 while (i < argc) {
427 const char *ref = argv[i++];
428 unsigned char sha1[20];
429 if (!resolve_ref(ref, sha1, 1, NULL)) {
430 status |= error("%s points nowhere!", ref);
431 continue;
433 status |= expire_reflog(ref, sha1, 0, &cb);
435 return status;
438 static int count_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
439 const char *email, unsigned long timestamp, int tz,
440 const char *message, void *cb_data)
442 struct cmd_reflog_expire_cb *cb = cb_data;
443 if (!cb->expire_total || timestamp < cb->expire_total)
444 cb->recno++;
445 return 0;
448 static int cmd_reflog_delete(int argc, const char **argv, const char *prefix)
450 struct cmd_reflog_expire_cb cb;
451 int i, status = 0;
453 memset(&cb, 0, sizeof(cb));
455 for (i = 1; i < argc; i++) {
456 const char *arg = argv[i];
457 if (!strcmp(arg, "--dry-run") || !strcmp(arg, "-n"))
458 cb.dry_run = 1;
459 else if (!strcmp(arg, "--rewrite"))
460 cb.rewrite = 1;
461 else if (!strcmp(arg, "--updateref"))
462 cb.updateref = 1;
463 else if (!strcmp(arg, "--verbose"))
464 cb.verbose = 1;
465 else if (!strcmp(arg, "--")) {
466 i++;
467 break;
469 else if (arg[0] == '-')
470 usage(reflog_delete_usage);
471 else
472 break;
475 if (argc - i < 1)
476 return error("Nothing to delete?");
478 for ( ; i < argc; i++) {
479 const char *spec = strstr(argv[i], "@{");
480 unsigned char sha1[20];
481 char *ep, *ref;
482 int recno;
484 if (!spec) {
485 status |= error("Not a reflog: %s", argv[i]);
486 continue;
489 if (!dwim_ref(argv[i], spec - argv[i], sha1, &ref)) {
490 status |= error("%s points nowhere!", argv[i]);
491 continue;
494 recno = strtoul(spec + 2, &ep, 10);
495 if (*ep == '}') {
496 cb.recno = -recno;
497 for_each_reflog_ent(ref, count_reflog_ent, &cb);
498 } else {
499 cb.expire_total = approxidate(spec + 2);
500 for_each_reflog_ent(ref, count_reflog_ent, &cb);
501 cb.expire_total = 0;
504 status |= expire_reflog(ref, sha1, 0, &cb);
505 free(ref);
507 return status;
511 * main "reflog"
514 static const char reflog_usage[] =
515 "git-reflog (expire | ...)";
517 int cmd_reflog(int argc, const char **argv, const char *prefix)
519 /* With no command, we default to showing it. */
520 if (argc < 2 || *argv[1] == '-')
521 return cmd_log_reflog(argc, argv, prefix);
523 if (!strcmp(argv[1], "show"))
524 return cmd_log_reflog(argc - 1, argv + 1, prefix);
526 if (!strcmp(argv[1], "expire"))
527 return cmd_reflog_expire(argc - 1, argv + 1, prefix);
529 if (!strcmp(argv[1], "delete"))
530 return cmd_reflog_delete(argc - 1, argv + 1, prefix);
532 /* Not a recognized reflog command..*/
533 usage(reflog_usage);