bundle: allowing to read from an unseekable fd
[git] / bundle.c
1 #include "cache.h"
2 #include "bundle.h"
3 #include "object.h"
4 #include "commit.h"
5 #include "diff.h"
6 #include "revision.h"
7 #include "list-objects.h"
8 #include "run-command.h"
9 #include "refs.h"
10
11 static const char bundle_signature[] = "# v2 git bundle\n";
12
13 static void add_to_ref_list(const unsigned char *sha1, const char *name,
14                 struct ref_list *list)
15 {
16         if (list->nr + 1 >= list->alloc) {
17                 list->alloc = alloc_nr(list->nr + 1);
18                 list->list = xrealloc(list->list,
19                                 list->alloc * sizeof(list->list[0]));
20         }
21         memcpy(list->list[list->nr].sha1, sha1, 20);
22         list->list[list->nr].name = xstrdup(name);
23         list->nr++;
24 }
25
26 /* Eventually this should go to strbuf.[ch] */
27 static int strbuf_readline_fd(struct strbuf *sb, int fd)
28 {
29         strbuf_reset(sb);
30
31         while (1) {
32                 char ch;
33                 ssize_t len = xread(fd, &ch, 1);
34                 if (len < 0)
35                         return -1;
36                 strbuf_addch(sb, ch);
37                 if (ch == '\n')
38                         break;
39         }
40         return 0;
41 }
42
43 int read_bundle_header(const char *path, struct bundle_header *header)
44 {
45         struct strbuf buf = STRBUF_INIT;
46         int fd = open(path, O_RDONLY);
47         int status = 0;
48
49         if (fd < 0)
50                 return error("could not open '%s'", path);
51
52         /* The bundle header begins with the signature */
53         if (strbuf_readline_fd(&buf, fd) ||
54             strcmp(buf.buf, bundle_signature)) {
55                 error("'%s' does not look like a v2 bundle file", path);
56                 status = -1;
57                 goto abort;
58         }
59
60         /* The bundle header ends with an empty line */
61         while (!strbuf_readline_fd(&buf, fd) &&
62                buf.len && buf.buf[0] != '\n') {
63                 unsigned char sha1[20];
64                 int is_prereq = 0;
65
66                 if (*buf.buf == '-') {
67                         is_prereq = 1;
68                         strbuf_remove(&buf, 0, 1);
69                 }
70                 strbuf_rtrim(&buf);
71
72                 /*
73                  * Tip lines have object name, SP, and refname.
74                  * Prerequisites have object name that is optionally
75                  * followed by SP and subject line.
76                  */
77                 if (get_sha1_hex(buf.buf, sha1) ||
78                     (40 <= buf.len && !isspace(buf.buf[40])) ||
79                     (!is_prereq && buf.len <= 40)) {
80                         error("unrecognized header: %s%s (%d)",
81                               (is_prereq ? "-" : ""), buf.buf, (int)buf.len);
82                         status = -1;
83                         break;
84                 } else {
85                         if (is_prereq)
86                                 add_to_ref_list(sha1, "", &header->prerequisites);
87                         else
88                                 add_to_ref_list(sha1, buf.buf + 41, &header->references);
89                 }
90         }
91
92  abort:
93         if (status) {
94                 close(fd);
95                 fd = -1;
96         }
97         strbuf_release(&buf);
98         return fd;
99 }
100
101 static int list_refs(struct ref_list *r, int argc, const char **argv)
102 {
103         int i;
104
105         for (i = 0; i < r->nr; i++) {
106                 if (argc > 1) {
107                         int j;
108                         for (j = 1; j < argc; j++)
109                                 if (!strcmp(r->list[i].name, argv[j]))
110                                         break;
111                         if (j == argc)
112                                 continue;
113                 }
114                 printf("%s %s\n", sha1_to_hex(r->list[i].sha1),
115                                 r->list[i].name);
116         }
117         return 0;
118 }
119
120 #define PREREQ_MARK (1u<<16)
121
122 int verify_bundle(struct bundle_header *header, int verbose)
123 {
124         /*
125          * Do fast check, then if any prereqs are missing then go line by line
126          * to be verbose about the errors
127          */
128         struct ref_list *p = &header->prerequisites;
129         struct rev_info revs;
130         const char *argv[] = {NULL, "--all", NULL};
131         struct object_array refs;
132         struct commit *commit;
133         int i, ret = 0, req_nr;
134         const char *message = "Repository lacks these prerequisite commits:";
135
136         init_revisions(&revs, NULL);
137         for (i = 0; i < p->nr; i++) {
138                 struct ref_list_entry *e = p->list + i;
139                 struct object *o = parse_object(e->sha1);
140                 if (o) {
141                         o->flags |= PREREQ_MARK;
142                         add_pending_object(&revs, o, e->name);
143                         continue;
144                 }
145                 if (++ret == 1)
146                         error("%s", message);
147                 error("%s %s", sha1_to_hex(e->sha1), e->name);
148         }
149         if (revs.pending.nr != p->nr)
150                 return ret;
151         req_nr = revs.pending.nr;
152         setup_revisions(2, argv, &revs, NULL);
153
154         memset(&refs, 0, sizeof(struct object_array));
155         for (i = 0; i < revs.pending.nr; i++) {
156                 struct object_array_entry *e = revs.pending.objects + i;
157                 add_object_array(e->item, e->name, &refs);
158         }
159
160         if (prepare_revision_walk(&revs))
161                 die("revision walk setup failed");
162
163         i = req_nr;
164         while (i && (commit = get_revision(&revs)))
165                 if (commit->object.flags & PREREQ_MARK)
166                         i--;
167
168         for (i = 0; i < req_nr; i++)
169                 if (!(refs.objects[i].item->flags & SHOWN)) {
170                         if (++ret == 1)
171                                 error("%s", message);
172                         error("%s %s", sha1_to_hex(refs.objects[i].item->sha1),
173                                 refs.objects[i].name);
174                 }
175
176         for (i = 0; i < refs.nr; i++)
177                 clear_commit_marks((struct commit *)refs.objects[i].item, -1);
178
179         if (verbose) {
180                 struct ref_list *r;
181
182                 r = &header->references;
183                 printf("The bundle contains %d ref%s\n",
184                        r->nr, (1 < r->nr) ? "s" : "");
185                 list_refs(r, 0, NULL);
186                 r = &header->prerequisites;
187                 printf("The bundle requires these %d ref%s\n",
188                        r->nr, (1 < r->nr) ? "s" : "");
189                 list_refs(r, 0, NULL);
190         }
191         return ret;
192 }
193
194 int list_bundle_refs(struct bundle_header *header, int argc, const char **argv)
195 {
196         return list_refs(&header->references, argc, argv);
197 }
198
199 static int is_tag_in_date_range(struct object *tag, struct rev_info *revs)
200 {
201         unsigned long size;
202         enum object_type type;
203         char *buf, *line, *lineend;
204         unsigned long date;
205
206         if (revs->max_age == -1 && revs->min_age == -1)
207                 return 1;
208
209         buf = read_sha1_file(tag->sha1, &type, &size);
210         if (!buf)
211                 return 1;
212         line = memmem(buf, size, "\ntagger ", 8);
213         if (!line++)
214                 return 1;
215         lineend = memchr(line, buf + size - line, '\n');
216         line = memchr(line, lineend ? lineend - line : buf + size - line, '>');
217         if (!line++)
218                 return 1;
219         date = strtoul(line, NULL, 10);
220         free(buf);
221         return (revs->max_age == -1 || revs->max_age < date) &&
222                 (revs->min_age == -1 || revs->min_age > date);
223 }
224
225 int create_bundle(struct bundle_header *header, const char *path,
226                 int argc, const char **argv)
227 {
228         static struct lock_file lock;
229         int bundle_fd = -1;
230         int bundle_to_stdout;
231         const char **argv_boundary = xmalloc((argc + 4) * sizeof(const char *));
232         const char **argv_pack = xmalloc(6 * sizeof(const char *));
233         int i, ref_count = 0;
234         char buffer[1024];
235         struct rev_info revs;
236         struct child_process rls;
237         FILE *rls_fout;
238
239         bundle_to_stdout = !strcmp(path, "-");
240         if (bundle_to_stdout)
241                 bundle_fd = 1;
242         else
243                 bundle_fd = hold_lock_file_for_update(&lock, path,
244                                                       LOCK_DIE_ON_ERROR);
245
246         /* write signature */
247         write_or_die(bundle_fd, bundle_signature, strlen(bundle_signature));
248
249         /* init revs to list objects for pack-objects later */
250         save_commit_buffer = 0;
251         init_revisions(&revs, NULL);
252
253         /* write prerequisites */
254         memcpy(argv_boundary + 3, argv + 1, argc * sizeof(const char *));
255         argv_boundary[0] = "rev-list";
256         argv_boundary[1] = "--boundary";
257         argv_boundary[2] = "--pretty=oneline";
258         argv_boundary[argc + 2] = NULL;
259         memset(&rls, 0, sizeof(rls));
260         rls.argv = argv_boundary;
261         rls.out = -1;
262         rls.git_cmd = 1;
263         if (start_command(&rls))
264                 return -1;
265         rls_fout = xfdopen(rls.out, "r");
266         while (fgets(buffer, sizeof(buffer), rls_fout)) {
267                 unsigned char sha1[20];
268                 if (buffer[0] == '-') {
269                         write_or_die(bundle_fd, buffer, strlen(buffer));
270                         if (!get_sha1_hex(buffer + 1, sha1)) {
271                                 struct object *object = parse_object(sha1);
272                                 object->flags |= UNINTERESTING;
273                                 add_pending_object(&revs, object, buffer);
274                         }
275                 } else if (!get_sha1_hex(buffer, sha1)) {
276                         struct object *object = parse_object(sha1);
277                         object->flags |= SHOWN;
278                 }
279         }
280         fclose(rls_fout);
281         if (finish_command(&rls))
282                 return error("rev-list died");
283
284         /* write references */
285         argc = setup_revisions(argc, argv, &revs, NULL);
286
287         if (argc > 1)
288                 return error("unrecognized argument: %s'", argv[1]);
289
290         object_array_remove_duplicates(&revs.pending);
291
292         for (i = 0; i < revs.pending.nr; i++) {
293                 struct object_array_entry *e = revs.pending.objects + i;
294                 unsigned char sha1[20];
295                 char *ref;
296                 const char *display_ref;
297                 int flag;
298
299                 if (e->item->flags & UNINTERESTING)
300                         continue;
301                 if (dwim_ref(e->name, strlen(e->name), sha1, &ref) != 1)
302                         continue;
303                 if (!resolve_ref(e->name, sha1, 1, &flag))
304                         flag = 0;
305                 display_ref = (flag & REF_ISSYMREF) ? e->name : ref;
306
307                 if (e->item->type == OBJ_TAG &&
308                                 !is_tag_in_date_range(e->item, &revs)) {
309                         e->item->flags |= UNINTERESTING;
310                         continue;
311                 }
312
313                 /*
314                  * Make sure the refs we wrote out is correct; --max-count and
315                  * other limiting options could have prevented all the tips
316                  * from getting output.
317                  *
318                  * Non commit objects such as tags and blobs do not have
319                  * this issue as they are not affected by those extra
320                  * constraints.
321                  */
322                 if (!(e->item->flags & SHOWN) && e->item->type == OBJ_COMMIT) {
323                         warning("ref '%s' is excluded by the rev-list options",
324                                 e->name);
325                         free(ref);
326                         continue;
327                 }
328                 /*
329                  * If you run "git bundle create bndl v1.0..v2.0", the
330                  * name of the positive ref is "v2.0" but that is the
331                  * commit that is referenced by the tag, and not the tag
332                  * itself.
333                  */
334                 if (hashcmp(sha1, e->item->sha1)) {
335                         /*
336                          * Is this the positive end of a range expressed
337                          * in terms of a tag (e.g. v2.0 from the range
338                          * "v1.0..v2.0")?
339                          */
340                         struct commit *one = lookup_commit_reference(sha1);
341                         struct object *obj;
342
343                         if (e->item == &(one->object)) {
344                                 /*
345                                  * Need to include e->name as an
346                                  * independent ref to the pack-objects
347                                  * input, so that the tag is included
348                                  * in the output; otherwise we would
349                                  * end up triggering "empty bundle"
350                                  * error.
351                                  */
352                                 obj = parse_object(sha1);
353                                 obj->flags |= SHOWN;
354                                 add_pending_object(&revs, obj, e->name);
355                         }
356                         free(ref);
357                         continue;
358                 }
359
360                 ref_count++;
361                 write_or_die(bundle_fd, sha1_to_hex(e->item->sha1), 40);
362                 write_or_die(bundle_fd, " ", 1);
363                 write_or_die(bundle_fd, display_ref, strlen(display_ref));
364                 write_or_die(bundle_fd, "\n", 1);
365                 free(ref);
366         }
367         if (!ref_count)
368                 die ("Refusing to create empty bundle.");
369
370         /* end header */
371         write_or_die(bundle_fd, "\n", 1);
372
373         /* write pack */
374         argv_pack[0] = "pack-objects";
375         argv_pack[1] = "--all-progress-implied";
376         argv_pack[2] = "--stdout";
377         argv_pack[3] = "--thin";
378         argv_pack[4] = "--delta-base-offset";
379         argv_pack[5] = NULL;
380         memset(&rls, 0, sizeof(rls));
381         rls.argv = argv_pack;
382         rls.in = -1;
383         rls.out = bundle_fd;
384         rls.git_cmd = 1;
385         if (start_command(&rls))
386                 return error("Could not spawn pack-objects");
387
388         /*
389          * start_command closed bundle_fd if it was > 1
390          * so set the lock fd to -1 so commit_lock_file()
391          * won't fail trying to close it.
392          */
393         lock.fd = -1;
394
395         for (i = 0; i < revs.pending.nr; i++) {
396                 struct object *object = revs.pending.objects[i].item;
397                 if (object->flags & UNINTERESTING)
398                         write_or_die(rls.in, "^", 1);
399                 write_or_die(rls.in, sha1_to_hex(object->sha1), 40);
400                 write_or_die(rls.in, "\n", 1);
401         }
402         close(rls.in);
403         if (finish_command(&rls))
404                 return error ("pack-objects died");
405         if (!bundle_to_stdout) {
406                 if (commit_lock_file(&lock))
407                         die_errno("cannot create '%s'", path);
408         }
409         return 0;
410 }
411
412 int unbundle(struct bundle_header *header, int bundle_fd)
413 {
414         const char *argv_index_pack[] = {"index-pack",
415                 "--fix-thin", "--stdin", NULL};
416         struct child_process ip;
417
418         if (verify_bundle(header, 0))
419                 return -1;
420         memset(&ip, 0, sizeof(ip));
421         ip.argv = argv_index_pack;
422         ip.in = bundle_fd;
423         ip.no_stdout = 1;
424         ip.git_cmd = 1;
425         if (run_command(&ip))
426                 return error("index-pack died");
427         return 0;
428 }