add a howto document about corrupted blob recovery
[git/repo.git] / builtin-reflog.c
blobce093cad78ce8008cd8a60d3ab6be5663a712a9d
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>...";
18 static unsigned long default_reflog_expire;
19 static unsigned long default_reflog_expire_unreachable;
21 struct cmd_reflog_expire_cb {
22 struct rev_info revs;
23 int dry_run;
24 int stalefix;
25 int verbose;
26 unsigned long expire_total;
27 unsigned long expire_unreachable;
30 struct expire_reflog_cb {
31 FILE *newlog;
32 const char *ref;
33 struct commit *ref_commit;
34 struct cmd_reflog_expire_cb *cmd;
37 #define INCOMPLETE (1u<<10)
38 #define STUDYING (1u<<11)
40 static int tree_is_complete(const unsigned char *sha1)
42 struct tree_desc desc;
43 struct name_entry entry;
44 int complete;
45 struct tree *tree;
47 tree = lookup_tree(sha1);
48 if (!tree)
49 return 0;
50 if (tree->object.flags & SEEN)
51 return 1;
52 if (tree->object.flags & INCOMPLETE)
53 return 0;
55 if (!tree->buffer) {
56 enum object_type type;
57 unsigned long size;
58 void *data = read_sha1_file(sha1, &type, &size);
59 if (!data) {
60 tree->object.flags |= INCOMPLETE;
61 return 0;
63 tree->buffer = data;
64 tree->size = size;
66 init_tree_desc(&desc, tree->buffer, tree->size);
67 complete = 1;
68 while (tree_entry(&desc, &entry)) {
69 if (!has_sha1_file(entry.sha1) ||
70 (S_ISDIR(entry.mode) && !tree_is_complete(entry.sha1))) {
71 tree->object.flags |= INCOMPLETE;
72 complete = 0;
75 free(tree->buffer);
76 tree->buffer = NULL;
78 if (complete)
79 tree->object.flags |= SEEN;
80 return complete;
83 static int commit_is_complete(struct commit *commit)
85 struct object_array study;
86 struct object_array found;
87 int is_incomplete = 0;
88 int i;
90 /* early return */
91 if (commit->object.flags & SEEN)
92 return 1;
93 if (commit->object.flags & INCOMPLETE)
94 return 0;
96 * Find all commits that are reachable and are not marked as
97 * SEEN. Then make sure the trees and blobs contained are
98 * complete. After that, mark these commits also as SEEN.
99 * If some of the objects that are needed to complete this
100 * commit are missing, mark this commit as INCOMPLETE.
102 memset(&study, 0, sizeof(study));
103 memset(&found, 0, sizeof(found));
104 add_object_array(&commit->object, NULL, &study);
105 add_object_array(&commit->object, NULL, &found);
106 commit->object.flags |= STUDYING;
107 while (study.nr) {
108 struct commit *c;
109 struct commit_list *parent;
111 c = (struct commit *)study.objects[--study.nr].item;
112 if (!c->object.parsed && !parse_object(c->object.sha1))
113 c->object.flags |= INCOMPLETE;
115 if (c->object.flags & INCOMPLETE) {
116 is_incomplete = 1;
117 break;
119 else if (c->object.flags & SEEN)
120 continue;
121 for (parent = c->parents; parent; parent = parent->next) {
122 struct commit *p = parent->item;
123 if (p->object.flags & STUDYING)
124 continue;
125 p->object.flags |= STUDYING;
126 add_object_array(&p->object, NULL, &study);
127 add_object_array(&p->object, NULL, &found);
130 if (!is_incomplete) {
132 * make sure all commits in "found" array have all the
133 * necessary objects.
135 for (i = 0; i < found.nr; i++) {
136 struct commit *c =
137 (struct commit *)found.objects[i].item;
138 if (!tree_is_complete(c->tree->object.sha1)) {
139 is_incomplete = 1;
140 c->object.flags |= INCOMPLETE;
143 if (!is_incomplete) {
144 /* mark all found commits as complete, iow SEEN */
145 for (i = 0; i < found.nr; i++)
146 found.objects[i].item->flags |= SEEN;
149 /* clear flags from the objects we traversed */
150 for (i = 0; i < found.nr; i++)
151 found.objects[i].item->flags &= ~STUDYING;
152 if (is_incomplete)
153 commit->object.flags |= INCOMPLETE;
154 else {
156 * If we come here, we have (1) traversed the ancestry chain
157 * from the "commit" until we reach SEEN commits (which are
158 * known to be complete), and (2) made sure that the commits
159 * encountered during the above traversal refer to trees that
160 * are complete. Which means that we know *all* the commits
161 * we have seen during this process are complete.
163 for (i = 0; i < found.nr; i++)
164 found.objects[i].item->flags |= SEEN;
166 /* free object arrays */
167 free(study.objects);
168 free(found.objects);
169 return !is_incomplete;
172 static int keep_entry(struct commit **it, unsigned char *sha1)
174 struct commit *commit;
176 if (is_null_sha1(sha1))
177 return 1;
178 commit = lookup_commit_reference_gently(sha1, 1);
179 if (!commit)
180 return 0;
183 * Make sure everything in this commit exists.
185 * We have walked all the objects reachable from the refs
186 * and cache earlier. The commits reachable by this commit
187 * must meet SEEN commits -- and then we should mark them as
188 * SEEN as well.
190 if (!commit_is_complete(commit))
191 return 0;
192 *it = commit;
193 return 1;
196 static int expire_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
197 const char *email, unsigned long timestamp, int tz,
198 const char *message, void *cb_data)
200 struct expire_reflog_cb *cb = cb_data;
201 struct commit *old, *new;
203 if (timestamp < cb->cmd->expire_total)
204 goto prune;
206 old = new = NULL;
207 if (cb->cmd->stalefix &&
208 (!keep_entry(&old, osha1) || !keep_entry(&new, nsha1)))
209 goto prune;
211 if (timestamp < cb->cmd->expire_unreachable) {
212 if (!cb->ref_commit)
213 goto prune;
214 if (!old && !is_null_sha1(osha1))
215 old = lookup_commit_reference_gently(osha1, 1);
216 if (!new && !is_null_sha1(nsha1))
217 new = lookup_commit_reference_gently(nsha1, 1);
218 if ((old && !in_merge_bases(old, &cb->ref_commit, 1)) ||
219 (new && !in_merge_bases(new, &cb->ref_commit, 1)))
220 goto prune;
223 if (cb->newlog) {
224 char sign = (tz < 0) ? '-' : '+';
225 int zone = (tz < 0) ? (-tz) : tz;
226 fprintf(cb->newlog, "%s %s %s %lu %c%04d\t%s",
227 sha1_to_hex(osha1), sha1_to_hex(nsha1),
228 email, timestamp, sign, zone,
229 message);
231 if (cb->cmd->verbose)
232 printf("keep %s", message);
233 return 0;
234 prune:
235 if (!cb->newlog || cb->cmd->verbose)
236 printf("%sprune %s", cb->newlog ? "" : "would ", message);
237 return 0;
240 static int expire_reflog(const char *ref, const unsigned char *sha1, int unused, void *cb_data)
242 struct cmd_reflog_expire_cb *cmd = cb_data;
243 struct expire_reflog_cb cb;
244 struct ref_lock *lock;
245 char *log_file, *newlog_path = NULL;
246 int status = 0;
248 memset(&cb, 0, sizeof(cb));
249 /* we take the lock for the ref itself to prevent it from
250 * getting updated.
252 lock = lock_any_ref_for_update(ref, sha1, 0);
253 if (!lock)
254 return error("cannot lock ref '%s'", ref);
255 log_file = xstrdup(git_path("logs/%s", ref));
256 if (!file_exists(log_file))
257 goto finish;
258 if (!cmd->dry_run) {
259 newlog_path = xstrdup(git_path("logs/%s.lock", ref));
260 cb.newlog = fopen(newlog_path, "w");
263 cb.ref_commit = lookup_commit_reference_gently(sha1, 1);
264 cb.ref = ref;
265 cb.cmd = cmd;
266 for_each_reflog_ent(ref, expire_reflog_ent, &cb);
267 finish:
268 if (cb.newlog) {
269 if (fclose(cb.newlog))
270 status |= error("%s: %s", strerror(errno),
271 newlog_path);
272 if (rename(newlog_path, log_file)) {
273 status |= error("cannot rename %s to %s",
274 newlog_path, log_file);
275 unlink(newlog_path);
278 free(newlog_path);
279 free(log_file);
280 unlock_ref(lock);
281 return status;
284 static int reflog_expire_config(const char *var, const char *value)
286 if (!strcmp(var, "gc.reflogexpire"))
287 default_reflog_expire = approxidate(value);
288 else if (!strcmp(var, "gc.reflogexpireunreachable"))
289 default_reflog_expire_unreachable = approxidate(value);
290 else
291 return git_default_config(var, value);
292 return 0;
295 static int cmd_reflog_expire(int argc, const char **argv, const char *prefix)
297 struct cmd_reflog_expire_cb cb;
298 unsigned long now = time(NULL);
299 int i, status, do_all;
301 git_config(reflog_expire_config);
303 save_commit_buffer = 0;
304 do_all = status = 0;
305 memset(&cb, 0, sizeof(cb));
307 if (!default_reflog_expire_unreachable)
308 default_reflog_expire_unreachable = now - 30 * 24 * 3600;
309 if (!default_reflog_expire)
310 default_reflog_expire = now - 90 * 24 * 3600;
311 cb.expire_total = default_reflog_expire;
312 cb.expire_unreachable = default_reflog_expire_unreachable;
315 * We can trust the commits and objects reachable from refs
316 * even in older repository. We cannot trust what's reachable
317 * from reflog if the repository was pruned with older git.
320 for (i = 1; i < argc; i++) {
321 const char *arg = argv[i];
322 if (!strcmp(arg, "--dry-run") || !strcmp(arg, "-n"))
323 cb.dry_run = 1;
324 else if (!prefixcmp(arg, "--expire="))
325 cb.expire_total = approxidate(arg + 9);
326 else if (!prefixcmp(arg, "--expire-unreachable="))
327 cb.expire_unreachable = approxidate(arg + 21);
328 else if (!strcmp(arg, "--stale-fix"))
329 cb.stalefix = 1;
330 else if (!strcmp(arg, "--all"))
331 do_all = 1;
332 else if (!strcmp(arg, "--verbose"))
333 cb.verbose = 1;
334 else if (!strcmp(arg, "--")) {
335 i++;
336 break;
338 else if (arg[0] == '-')
339 usage(reflog_expire_usage);
340 else
341 break;
343 if (cb.stalefix) {
344 init_revisions(&cb.revs, prefix);
345 if (cb.verbose)
346 printf("Marking reachable objects...");
347 mark_reachable_objects(&cb.revs, 0);
348 if (cb.verbose)
349 putchar('\n');
352 if (do_all)
353 status |= for_each_reflog(expire_reflog, &cb);
354 while (i < argc) {
355 const char *ref = argv[i++];
356 unsigned char sha1[20];
357 if (!resolve_ref(ref, sha1, 1, NULL)) {
358 status |= error("%s points nowhere!", ref);
359 continue;
361 status |= expire_reflog(ref, sha1, 0, &cb);
363 return status;
367 * main "reflog"
370 static const char reflog_usage[] =
371 "git-reflog (expire | ...)";
373 int cmd_reflog(int argc, const char **argv, const char *prefix)
375 /* With no command, we default to showing it. */
376 if (argc < 2 || *argv[1] == '-')
377 return cmd_log_reflog(argc, argv, prefix);
379 if (!strcmp(argv[1], "show"))
380 return cmd_log_reflog(argc - 1, argv + 1, prefix);
382 if (!strcmp(argv[1], "expire"))
383 return cmd_reflog_expire(argc - 1, argv + 1, prefix);
385 /* Not a recognized reflog command..*/
386 usage(reflog_usage);