Fix crasher on encountering SHA1-like non-note in notes tree
[git] / builtin-reflog.c
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"
10
11 /*
12  * reflog expire
13  */
14
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>...";
19
20 static unsigned long default_reflog_expire;
21 static unsigned long default_reflog_expire_unreachable;
22
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;
33 };
34
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];
41 };
42
43 struct collected_reflog {
44         unsigned char sha1[20];
45         char reflog[FLEX_ARRAY];
46 };
47 struct collect_reflog_cb {
48         struct collected_reflog **e;
49         int alloc;
50         int nr;
51 };
52
53 #define INCOMPLETE      (1u<<10)
54 #define STUDYING        (1u<<11)
55 #define REACHABLE       (1u<<12)
56
57 static int tree_is_complete(const unsigned char *sha1)
58 {
59         struct tree_desc desc;
60         struct name_entry entry;
61         int complete;
62         struct tree *tree;
63
64         tree = lookup_tree(sha1);
65         if (!tree)
66                 return 0;
67         if (tree->object.flags & SEEN)
68                 return 1;
69         if (tree->object.flags & INCOMPLETE)
70                 return 0;
71
72         if (!tree->buffer) {
73                 enum object_type type;
74                 unsigned long size;
75                 void *data = read_sha1_file(sha1, &type, &size);
76                 if (!data) {
77                         tree->object.flags |= INCOMPLETE;
78                         return 0;
79                 }
80                 tree->buffer = data;
81                 tree->size = size;
82         }
83         init_tree_desc(&desc, tree->buffer, tree->size);
84         complete = 1;
85         while (tree_entry(&desc, &entry)) {
86                 if (!has_sha1_file(entry.sha1) ||
87                     (S_ISDIR(entry.mode) && !tree_is_complete(entry.sha1))) {
88                         tree->object.flags |= INCOMPLETE;
89                         complete = 0;
90                 }
91         }
92         free(tree->buffer);
93         tree->buffer = NULL;
94
95         if (complete)
96                 tree->object.flags |= SEEN;
97         return complete;
98 }
99
100 static int commit_is_complete(struct commit *commit)
101 {
102         struct object_array study;
103         struct object_array found;
104         int is_incomplete = 0;
105         int i;
106
107         /* early return */
108         if (commit->object.flags & SEEN)
109                 return 1;
110         if (commit->object.flags & INCOMPLETE)
111                 return 0;
112         /*
113          * Find all commits that are reachable and are not marked as
114          * SEEN.  Then make sure the trees and blobs contained are
115          * complete.  After that, mark these commits also as SEEN.
116          * If some of the objects that are needed to complete this
117          * commit are missing, mark this commit as INCOMPLETE.
118          */
119         memset(&study, 0, sizeof(study));
120         memset(&found, 0, sizeof(found));
121         add_object_array(&commit->object, NULL, &study);
122         add_object_array(&commit->object, NULL, &found);
123         commit->object.flags |= STUDYING;
124         while (study.nr) {
125                 struct commit *c;
126                 struct commit_list *parent;
127
128                 c = (struct commit *)study.objects[--study.nr].item;
129                 if (!c->object.parsed && !parse_object(c->object.sha1))
130                         c->object.flags |= INCOMPLETE;
131
132                 if (c->object.flags & INCOMPLETE) {
133                         is_incomplete = 1;
134                         break;
135                 }
136                 else if (c->object.flags & SEEN)
137                         continue;
138                 for (parent = c->parents; parent; parent = parent->next) {
139                         struct commit *p = parent->item;
140                         if (p->object.flags & STUDYING)
141                                 continue;
142                         p->object.flags |= STUDYING;
143                         add_object_array(&p->object, NULL, &study);
144                         add_object_array(&p->object, NULL, &found);
145                 }
146         }
147         if (!is_incomplete) {
148                 /*
149                  * make sure all commits in "found" array have all the
150                  * necessary objects.
151                  */
152                 for (i = 0; i < found.nr; i++) {
153                         struct commit *c =
154                                 (struct commit *)found.objects[i].item;
155                         if (!tree_is_complete(c->tree->object.sha1)) {
156                                 is_incomplete = 1;
157                                 c->object.flags |= INCOMPLETE;
158                         }
159                 }
160                 if (!is_incomplete) {
161                         /* mark all found commits as complete, iow SEEN */
162                         for (i = 0; i < found.nr; i++)
163                                 found.objects[i].item->flags |= SEEN;
164                 }
165         }
166         /* clear flags from the objects we traversed */
167         for (i = 0; i < found.nr; i++)
168                 found.objects[i].item->flags &= ~STUDYING;
169         if (is_incomplete)
170                 commit->object.flags |= INCOMPLETE;
171         else {
172                 /*
173                  * If we come here, we have (1) traversed the ancestry chain
174                  * from the "commit" until we reach SEEN commits (which are
175                  * known to be complete), and (2) made sure that the commits
176                  * encountered during the above traversal refer to trees that
177                  * are complete.  Which means that we know *all* the commits
178                  * we have seen during this process are complete.
179                  */
180                 for (i = 0; i < found.nr; i++)
181                         found.objects[i].item->flags |= SEEN;
182         }
183         /* free object arrays */
184         free(study.objects);
185         free(found.objects);
186         return !is_incomplete;
187 }
188
189 static int keep_entry(struct commit **it, unsigned char *sha1)
190 {
191         struct commit *commit;
192
193         if (is_null_sha1(sha1))
194                 return 1;
195         commit = lookup_commit_reference_gently(sha1, 1);
196         if (!commit)
197                 return 0;
198
199         /*
200          * Make sure everything in this commit exists.
201          *
202          * We have walked all the objects reachable from the refs
203          * and cache earlier.  The commits reachable by this commit
204          * must meet SEEN commits -- and then we should mark them as
205          * SEEN as well.
206          */
207         if (!commit_is_complete(commit))
208                 return 0;
209         *it = commit;
210         return 1;
211 }
212
213 static int unreachable(struct expire_reflog_cb *cb, struct commit *commit, unsigned char *sha1)
214 {
215         /*
216          * We may or may not have the commit yet - if not, look it
217          * up using the supplied sha1.
218          */
219         if (!commit) {
220                 if (is_null_sha1(sha1))
221                         return 0;
222
223                 commit = lookup_commit_reference_gently(sha1, 1);
224
225                 /* Not a commit -- keep it */
226                 if (!commit)
227                         return 0;
228         }
229
230         /* Reachable from the current ref?  Don't prune. */
231         if (commit->object.flags & REACHABLE)
232                 return 0;
233         if (in_merge_bases(commit, &cb->ref_commit, 1))
234                 return 0;
235
236         /* We can't reach it - prune it. */
237         return 1;
238 }
239
240 static void mark_reachable(struct commit *commit, unsigned long expire_limit)
241 {
242         /*
243          * We need to compute whether the commit on either side of a reflog
244          * entry is reachable from the tip of the ref for all entries.
245          * Mark commits that are reachable from the tip down to the
246          * time threshold first; we know a commit marked thusly is
247          * reachable from the tip without running in_merge_bases()
248          * at all.
249          */
250         struct commit_list *pending = NULL;
251
252         commit_list_insert(commit, &pending);
253         while (pending) {
254                 struct commit_list *entry = pending;
255                 struct commit_list *parent;
256                 pending = entry->next;
257                 commit = entry->item;
258                 free(entry);
259                 if (commit->object.flags & REACHABLE)
260                         continue;
261                 if (parse_commit(commit))
262                         continue;
263                 commit->object.flags |= REACHABLE;
264                 if (commit->date < expire_limit)
265                         continue;
266                 parent = commit->parents;
267                 while (parent) {
268                         commit = parent->item;
269                         parent = parent->next;
270                         if (commit->object.flags & REACHABLE)
271                                 continue;
272                         commit_list_insert(commit, &pending);
273                 }
274         }
275 }
276
277 static int expire_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
278                 const char *email, unsigned long timestamp, int tz,
279                 const char *message, void *cb_data)
280 {
281         struct expire_reflog_cb *cb = cb_data;
282         struct commit *old, *new;
283
284         if (timestamp < cb->cmd->expire_total)
285                 goto prune;
286
287         if (cb->cmd->rewrite)
288                 osha1 = cb->last_kept_sha1;
289
290         old = new = NULL;
291         if (cb->cmd->stalefix &&
292             (!keep_entry(&old, osha1) || !keep_entry(&new, nsha1)))
293                 goto prune;
294
295         if (timestamp < cb->cmd->expire_unreachable) {
296                 if (!cb->ref_commit)
297                         goto prune;
298                 if (unreachable(cb, old, osha1) || unreachable(cb, new, nsha1))
299                         goto prune;
300         }
301
302         if (cb->cmd->recno && --(cb->cmd->recno) == 0)
303                 goto prune;
304
305         if (cb->newlog) {
306                 char sign = (tz < 0) ? '-' : '+';
307                 int zone = (tz < 0) ? (-tz) : tz;
308                 fprintf(cb->newlog, "%s %s %s %lu %c%04d\t%s",
309                         sha1_to_hex(osha1), sha1_to_hex(nsha1),
310                         email, timestamp, sign, zone,
311                         message);
312                 hashcpy(cb->last_kept_sha1, nsha1);
313         }
314         if (cb->cmd->verbose)
315                 printf("keep %s", message);
316         return 0;
317  prune:
318         if (!cb->newlog || cb->cmd->verbose)
319                 printf("%sprune %s", cb->newlog ? "" : "would ", message);
320         return 0;
321 }
322
323 static int expire_reflog(const char *ref, const unsigned char *sha1, int unused, void *cb_data)
324 {
325         struct cmd_reflog_expire_cb *cmd = cb_data;
326         struct expire_reflog_cb cb;
327         struct ref_lock *lock;
328         char *log_file, *newlog_path = NULL;
329         int status = 0;
330
331         memset(&cb, 0, sizeof(cb));
332
333         /*
334          * we take the lock for the ref itself to prevent it from
335          * getting updated.
336          */
337         lock = lock_any_ref_for_update(ref, sha1, 0);
338         if (!lock)
339                 return error("cannot lock ref '%s'", ref);
340         log_file = git_pathdup("logs/%s", ref);
341         if (!file_exists(log_file))
342                 goto finish;
343         if (!cmd->dry_run) {
344                 newlog_path = git_pathdup("logs/%s.lock", ref);
345                 cb.newlog = fopen(newlog_path, "w");
346         }
347
348         cb.ref_commit = lookup_commit_reference_gently(sha1, 1);
349         cb.ref = ref;
350         cb.cmd = cmd;
351         if (cb.ref_commit)
352                 mark_reachable(cb.ref_commit, cmd->expire_total);
353         for_each_reflog_ent(ref, expire_reflog_ent, &cb);
354         if (cb.ref_commit)
355                 clear_commit_marks(cb.ref_commit, REACHABLE);
356  finish:
357         if (cb.newlog) {
358                 if (fclose(cb.newlog)) {
359                         status |= error("%s: %s", strerror(errno),
360                                         newlog_path);
361                         unlink(newlog_path);
362                 } else if (cmd->updateref &&
363                         (write_in_full(lock->lock_fd,
364                                 sha1_to_hex(cb.last_kept_sha1), 40) != 40 ||
365                          write_str_in_full(lock->lock_fd, "\n") != 1 ||
366                          close_ref(lock) < 0)) {
367                         status |= error("Couldn't write %s",
368                                 lock->lk->filename);
369                         unlink(newlog_path);
370                 } else if (rename(newlog_path, log_file)) {
371                         status |= error("cannot rename %s to %s",
372                                         newlog_path, log_file);
373                         unlink(newlog_path);
374                 } else if (cmd->updateref && commit_ref(lock)) {
375                         status |= error("Couldn't set %s", lock->ref_name);
376                 } else {
377                         adjust_shared_perm(log_file);
378                 }
379         }
380         free(newlog_path);
381         free(log_file);
382         unlock_ref(lock);
383         return status;
384 }
385
386 static int collect_reflog(const char *ref, const unsigned char *sha1, int unused, void *cb_data)
387 {
388         struct collected_reflog *e;
389         struct collect_reflog_cb *cb = cb_data;
390         size_t namelen = strlen(ref);
391
392         e = xmalloc(sizeof(*e) + namelen + 1);
393         hashcpy(e->sha1, sha1);
394         memcpy(e->reflog, ref, namelen + 1);
395         ALLOC_GROW(cb->e, cb->nr + 1, cb->alloc);
396         cb->e[cb->nr++] = e;
397         return 0;
398 }
399
400 static struct reflog_expire_cfg {
401         struct reflog_expire_cfg *next;
402         unsigned long expire_total;
403         unsigned long expire_unreachable;
404         size_t len;
405         char pattern[FLEX_ARRAY];
406 } *reflog_expire_cfg, **reflog_expire_cfg_tail;
407
408 static struct reflog_expire_cfg *find_cfg_ent(const char *pattern, size_t len)
409 {
410         struct reflog_expire_cfg *ent;
411
412         if (!reflog_expire_cfg_tail)
413                 reflog_expire_cfg_tail = &reflog_expire_cfg;
414
415         for (ent = reflog_expire_cfg; ent; ent = ent->next)
416                 if (ent->len == len &&
417                     !memcmp(ent->pattern, pattern, len))
418                         return ent;
419
420         ent = xcalloc(1, (sizeof(*ent) + len));
421         memcpy(ent->pattern, pattern, len);
422         ent->len = len;
423         *reflog_expire_cfg_tail = ent;
424         reflog_expire_cfg_tail = &(ent->next);
425         return ent;
426 }
427
428 static int parse_expire_cfg_value(const char *var, const char *value, unsigned long *expire)
429 {
430         if (!value)
431                 return config_error_nonbool(var);
432         if (!strcmp(value, "never") || !strcmp(value, "false")) {
433                 *expire = 0;
434                 return 0;
435         }
436         *expire = approxidate(value);
437         return 0;
438 }
439
440 /* expiry timer slot */
441 #define EXPIRE_TOTAL   01
442 #define EXPIRE_UNREACH 02
443
444 static int reflog_expire_config(const char *var, const char *value, void *cb)
445 {
446         const char *lastdot = strrchr(var, '.');
447         unsigned long expire;
448         int slot;
449         struct reflog_expire_cfg *ent;
450
451         if (!lastdot || prefixcmp(var, "gc."))
452                 return git_default_config(var, value, cb);
453
454         if (!strcmp(lastdot, ".reflogexpire")) {
455                 slot = EXPIRE_TOTAL;
456                 if (parse_expire_cfg_value(var, value, &expire))
457                         return -1;
458         } else if (!strcmp(lastdot, ".reflogexpireunreachable")) {
459                 slot = EXPIRE_UNREACH;
460                 if (parse_expire_cfg_value(var, value, &expire))
461                         return -1;
462         } else
463                 return git_default_config(var, value, cb);
464
465         if (lastdot == var + 2) {
466                 switch (slot) {
467                 case EXPIRE_TOTAL:
468                         default_reflog_expire = expire;
469                         break;
470                 case EXPIRE_UNREACH:
471                         default_reflog_expire_unreachable = expire;
472                         break;
473                 }
474                 return 0;
475         }
476
477         ent = find_cfg_ent(var + 3, lastdot - (var+3));
478         if (!ent)
479                 return -1;
480         switch (slot) {
481         case EXPIRE_TOTAL:
482                 ent->expire_total = expire;
483                 break;
484         case EXPIRE_UNREACH:
485                 ent->expire_unreachable = expire;
486                 break;
487         }
488         return 0;
489 }
490
491 static void set_reflog_expiry_param(struct cmd_reflog_expire_cb *cb, int slot, const char *ref)
492 {
493         struct reflog_expire_cfg *ent;
494
495         if (slot == (EXPIRE_TOTAL|EXPIRE_UNREACH))
496                 return; /* both given explicitly -- nothing to tweak */
497
498         for (ent = reflog_expire_cfg; ent; ent = ent->next) {
499                 if (!fnmatch(ent->pattern, ref, 0)) {
500                         if (!(slot & EXPIRE_TOTAL))
501                                 cb->expire_total = ent->expire_total;
502                         if (!(slot & EXPIRE_UNREACH))
503                                 cb->expire_unreachable = ent->expire_unreachable;
504                         return;
505                 }
506         }
507
508         /*
509          * If unconfigured, make stash never expire
510          */
511         if (!strcmp(ref, "refs/stash")) {
512                 if (!(slot & EXPIRE_TOTAL))
513                         cb->expire_total = 0;
514                 if (!(slot & EXPIRE_UNREACH))
515                         cb->expire_unreachable = 0;
516                 return;
517         }
518
519         /* Nothing matched -- use the default value */
520         if (!(slot & EXPIRE_TOTAL))
521                 cb->expire_total = default_reflog_expire;
522         if (!(slot & EXPIRE_UNREACH))
523                 cb->expire_unreachable = default_reflog_expire_unreachable;
524 }
525
526 static int cmd_reflog_expire(int argc, const char **argv, const char *prefix)
527 {
528         struct cmd_reflog_expire_cb cb;
529         unsigned long now = time(NULL);
530         int i, status, do_all;
531         int explicit_expiry = 0;
532
533         git_config(reflog_expire_config, NULL);
534
535         save_commit_buffer = 0;
536         do_all = status = 0;
537         memset(&cb, 0, sizeof(cb));
538
539         if (!default_reflog_expire_unreachable)
540                 default_reflog_expire_unreachable = now - 30 * 24 * 3600;
541         if (!default_reflog_expire)
542                 default_reflog_expire = now - 90 * 24 * 3600;
543         cb.expire_total = default_reflog_expire;
544         cb.expire_unreachable = default_reflog_expire_unreachable;
545
546         for (i = 1; i < argc; i++) {
547                 const char *arg = argv[i];
548                 if (!strcmp(arg, "--dry-run") || !strcmp(arg, "-n"))
549                         cb.dry_run = 1;
550                 else if (!prefixcmp(arg, "--expire=")) {
551                         cb.expire_total = approxidate(arg + 9);
552                         explicit_expiry |= EXPIRE_TOTAL;
553                 }
554                 else if (!prefixcmp(arg, "--expire-unreachable=")) {
555                         cb.expire_unreachable = approxidate(arg + 21);
556                         explicit_expiry |= EXPIRE_UNREACH;
557                 }
558                 else if (!strcmp(arg, "--stale-fix"))
559                         cb.stalefix = 1;
560                 else if (!strcmp(arg, "--rewrite"))
561                         cb.rewrite = 1;
562                 else if (!strcmp(arg, "--updateref"))
563                         cb.updateref = 1;
564                 else if (!strcmp(arg, "--all"))
565                         do_all = 1;
566                 else if (!strcmp(arg, "--verbose"))
567                         cb.verbose = 1;
568                 else if (!strcmp(arg, "--")) {
569                         i++;
570                         break;
571                 }
572                 else if (arg[0] == '-')
573                         usage(reflog_expire_usage);
574                 else
575                         break;
576         }
577
578         /*
579          * We can trust the commits and objects reachable from refs
580          * even in older repository.  We cannot trust what's reachable
581          * from reflog if the repository was pruned with older git.
582          */
583         if (cb.stalefix) {
584                 init_revisions(&cb.revs, prefix);
585                 if (cb.verbose)
586                         printf("Marking reachable objects...");
587                 mark_reachable_objects(&cb.revs, 0);
588                 if (cb.verbose)
589                         putchar('\n');
590         }
591
592         if (do_all) {
593                 struct collect_reflog_cb collected;
594                 int i;
595
596                 memset(&collected, 0, sizeof(collected));
597                 for_each_reflog(collect_reflog, &collected);
598                 for (i = 0; i < collected.nr; i++) {
599                         struct collected_reflog *e = collected.e[i];
600                         set_reflog_expiry_param(&cb, explicit_expiry, e->reflog);
601                         status |= expire_reflog(e->reflog, e->sha1, 0, &cb);
602                         free(e);
603                 }
604                 free(collected.e);
605         }
606
607         for (; i < argc; i++) {
608                 char *ref;
609                 unsigned char sha1[20];
610                 if (!dwim_log(argv[i], strlen(argv[i]), sha1, &ref)) {
611                         status |= error("%s points nowhere!", argv[i]);
612                         continue;
613                 }
614                 set_reflog_expiry_param(&cb, explicit_expiry, ref);
615                 status |= expire_reflog(ref, sha1, 0, &cb);
616         }
617         return status;
618 }
619
620 static int count_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
621                 const char *email, unsigned long timestamp, int tz,
622                 const char *message, void *cb_data)
623 {
624         struct cmd_reflog_expire_cb *cb = cb_data;
625         if (!cb->expire_total || timestamp < cb->expire_total)
626                 cb->recno++;
627         return 0;
628 }
629
630 static int cmd_reflog_delete(int argc, const char **argv, const char *prefix)
631 {
632         struct cmd_reflog_expire_cb cb;
633         int i, status = 0;
634
635         memset(&cb, 0, sizeof(cb));
636
637         for (i = 1; i < argc; i++) {
638                 const char *arg = argv[i];
639                 if (!strcmp(arg, "--dry-run") || !strcmp(arg, "-n"))
640                         cb.dry_run = 1;
641                 else if (!strcmp(arg, "--rewrite"))
642                         cb.rewrite = 1;
643                 else if (!strcmp(arg, "--updateref"))
644                         cb.updateref = 1;
645                 else if (!strcmp(arg, "--verbose"))
646                         cb.verbose = 1;
647                 else if (!strcmp(arg, "--")) {
648                         i++;
649                         break;
650                 }
651                 else if (arg[0] == '-')
652                         usage(reflog_delete_usage);
653                 else
654                         break;
655         }
656
657         if (argc - i < 1)
658                 return error("Nothing to delete?");
659
660         for ( ; i < argc; i++) {
661                 const char *spec = strstr(argv[i], "@{");
662                 unsigned char sha1[20];
663                 char *ep, *ref;
664                 int recno;
665
666                 if (!spec) {
667                         status |= error("Not a reflog: %s", argv[i]);
668                         continue;
669                 }
670
671                 if (!dwim_log(argv[i], spec - argv[i], sha1, &ref)) {
672                         status |= error("no reflog for '%s'", argv[i]);
673                         continue;
674                 }
675
676                 recno = strtoul(spec + 2, &ep, 10);
677                 if (*ep == '}') {
678                         cb.recno = -recno;
679                         for_each_reflog_ent(ref, count_reflog_ent, &cb);
680                 } else {
681                         cb.expire_total = approxidate(spec + 2);
682                         for_each_reflog_ent(ref, count_reflog_ent, &cb);
683                         cb.expire_total = 0;
684                 }
685
686                 status |= expire_reflog(ref, sha1, 0, &cb);
687                 free(ref);
688         }
689         return status;
690 }
691
692 /*
693  * main "reflog"
694  */
695
696 static const char reflog_usage[] =
697 "git reflog [ show | expire | delete ]";
698
699 int cmd_reflog(int argc, const char **argv, const char *prefix)
700 {
701         if (argc > 1 && !strcmp(argv[1], "-h"))
702                 usage(reflog_usage);
703
704         /* With no command, we default to showing it. */
705         if (argc < 2 || *argv[1] == '-')
706                 return cmd_log_reflog(argc, argv, prefix);
707
708         if (!strcmp(argv[1], "show"))
709                 return cmd_log_reflog(argc - 1, argv + 1, prefix);
710
711         if (!strcmp(argv[1], "expire"))
712                 return cmd_reflog_expire(argc - 1, argv + 1, prefix);
713
714         if (!strcmp(argv[1], "delete"))
715                 return cmd_reflog_delete(argc - 1, argv + 1, prefix);
716
717         /* Not a recognized reflog command..*/
718         usage(reflog_usage);
719 }