git-push: add documentation for the newly added --mirror mode
[git/git.git] / upload-pack.c
1 #include "cache.h"
2 #include "refs.h"
3 #include "pkt-line.h"
4 #include "sideband.h"
5 #include "tag.h"
6 #include "object.h"
7 #include "commit.h"
8 #include "exec_cmd.h"
9 #include "diff.h"
10 #include "revision.h"
11 #include "list-objects.h"
12 #include "run-command.h"
13
14 static const char upload_pack_usage[] = "git-upload-pack [--strict] [--timeout=nn] <dir>";
15
16 /* bits #0..7 in revision.h, #8..10 in commit.c */
17 #define THEY_HAVE (1u << 11)
18 #define OUR_REF (1u << 12)
19 #define WANTED (1u << 13)
20 #define COMMON_KNOWN (1u << 14)
21 #define REACHABLE (1u << 15)
22
23 #define SHALLOW (1u << 16)
24 #define NOT_SHALLOW (1u << 17)
25 #define CLIENT_SHALLOW (1u << 18)
26
27 static unsigned long oldest_have;
28
29 static int multi_ack, nr_our_refs;
30 static int use_thin_pack, use_ofs_delta, no_progress;
31 static struct object_array have_obj;
32 static struct object_array want_obj;
33 static unsigned int timeout;
34 /* 0 for no sideband,
35 * otherwise maximum packet size (up to 65520 bytes).
36 */
37 static int use_sideband;
38
39 static void reset_timeout(void)
40 {
41 alarm(timeout);
42 }
43
44 static int strip(char *line, int len)
45 {
46 if (len && line[len-1] == '\n')
47 line[--len] = 0;
48 return len;
49 }
50
51 static ssize_t send_client_data(int fd, const char *data, ssize_t sz)
52 {
53 if (use_sideband)
54 return send_sideband(1, fd, data, sz, use_sideband);
55 if (fd == 3)
56 /* emergency quit */
57 fd = 2;
58 if (fd == 2) {
59 /* XXX: are we happy to lose stuff here? */
60 xwrite(fd, data, sz);
61 return sz;
62 }
63 return safe_write(fd, data, sz);
64 }
65
66 static FILE *pack_pipe = NULL;
67 static void show_commit(struct commit *commit)
68 {
69 if (commit->object.flags & BOUNDARY)
70 fputc('-', pack_pipe);
71 if (fputs(sha1_to_hex(commit->object.sha1), pack_pipe) < 0)
72 die("broken output pipe");
73 fputc('\n', pack_pipe);
74 fflush(pack_pipe);
75 free(commit->buffer);
76 commit->buffer = NULL;
77 }
78
79 static void show_object(struct object_array_entry *p)
80 {
81 /* An object with name "foo\n0000000..." can be used to
82 * confuse downstream git-pack-objects very badly.
83 */
84 const char *ep = strchr(p->name, '\n');
85 if (ep) {
86 fprintf(pack_pipe, "%s %.*s\n", sha1_to_hex(p->item->sha1),
87 (int) (ep - p->name),
88 p->name);
89 }
90 else
91 fprintf(pack_pipe, "%s %s\n",
92 sha1_to_hex(p->item->sha1), p->name);
93 }
94
95 static void show_edge(struct commit *commit)
96 {
97 fprintf(pack_pipe, "-%s\n", sha1_to_hex(commit->object.sha1));
98 }
99
100 static int do_rev_list(int fd, void *create_full_pack)
101 {
102 int i;
103 struct rev_info revs;
104
105 pack_pipe = fdopen(fd, "w");
106 if (create_full_pack)
107 use_thin_pack = 0; /* no point doing it */
108 init_revisions(&revs, NULL);
109 revs.tag_objects = 1;
110 revs.tree_objects = 1;
111 revs.blob_objects = 1;
112 if (use_thin_pack)
113 revs.edge_hint = 1;
114
115 if (create_full_pack) {
116 const char *args[] = {"rev-list", "--all", NULL};
117 setup_revisions(2, args, &revs, NULL);
118 } else {
119 for (i = 0; i < want_obj.nr; i++) {
120 struct object *o = want_obj.objects[i].item;
121 /* why??? */
122 o->flags &= ~UNINTERESTING;
123 add_pending_object(&revs, o, NULL);
124 }
125 for (i = 0; i < have_obj.nr; i++) {
126 struct object *o = have_obj.objects[i].item;
127 o->flags |= UNINTERESTING;
128 add_pending_object(&revs, o, NULL);
129 }
130 setup_revisions(0, NULL, &revs, NULL);
131 }
132 prepare_revision_walk(&revs);
133 mark_edges_uninteresting(revs.commits, &revs, show_edge);
134 traverse_commit_list(&revs, show_commit, show_object);
135 return 0;
136 }
137
138 static void create_pack_file(void)
139 {
140 struct async rev_list;
141 struct child_process pack_objects;
142 int create_full_pack = (nr_our_refs == want_obj.nr && !have_obj.nr);
143 char data[8193], progress[128];
144 char abort_msg[] = "aborting due to possible repository "
145 "corruption on the remote side.";
146 int buffered = -1;
147 const char *argv[10];
148 int arg = 0;
149
150 rev_list.proc = do_rev_list;
151 /* .data is just a boolean: any non-NULL value will do */
152 rev_list.data = create_full_pack ? &rev_list : NULL;
153 if (start_async(&rev_list))
154 die("git-upload-pack: unable to fork git-rev-list");
155
156 argv[arg++] = "pack-objects";
157 argv[arg++] = "--stdout";
158 if (!no_progress)
159 argv[arg++] = "--progress";
160 if (use_ofs_delta)
161 argv[arg++] = "--delta-base-offset";
162 argv[arg++] = NULL;
163
164 memset(&pack_objects, 0, sizeof(pack_objects));
165 pack_objects.in = rev_list.out; /* start_command closes it */
166 pack_objects.out = -1;
167 pack_objects.err = -1;
168 pack_objects.git_cmd = 1;
169 pack_objects.argv = argv;
170
171 if (start_command(&pack_objects)) {
172 /* daemon sets things up to ignore TERM */
173 kill(rev_list.pid, SIGKILL);
174 die("git-upload-pack: unable to fork git-pack-objects");
175 }
176
177 /* We read from pack_objects.err to capture stderr output for
178 * progress bar, and pack_objects.out to capture the pack data.
179 */
180
181 while (1) {
182 const char *who;
183 struct pollfd pfd[2];
184 pid_t pid;
185 int status;
186 ssize_t sz;
187 int pe, pu, pollsize;
188
189 reset_timeout();
190
191 pollsize = 0;
192 pe = pu = -1;
193
194 if (0 <= pack_objects.out) {
195 pfd[pollsize].fd = pack_objects.out;
196 pfd[pollsize].events = POLLIN;
197 pu = pollsize;
198 pollsize++;
199 }
200 if (0 <= pack_objects.err) {
201 pfd[pollsize].fd = pack_objects.err;
202 pfd[pollsize].events = POLLIN;
203 pe = pollsize;
204 pollsize++;
205 }
206
207 if (pollsize) {
208 if (poll(pfd, pollsize, -1) < 0) {
209 if (errno != EINTR) {
210 error("poll failed, resuming: %s",
211 strerror(errno));
212 sleep(1);
213 }
214 continue;
215 }
216 if (0 <= pu && (pfd[pu].revents & (POLLIN|POLLHUP))) {
217 /* Data ready; we keep the last byte
218 * to ourselves in case we detect
219 * broken rev-list, so that we can
220 * leave the stream corrupted. This
221 * is unfortunate -- unpack-objects
222 * would happily accept a valid pack
223 * data with trailing garbage, so
224 * appending garbage after we pass all
225 * the pack data is not good enough to
226 * signal breakage to downstream.
227 */
228 char *cp = data;
229 ssize_t outsz = 0;
230 if (0 <= buffered) {
231 *cp++ = buffered;
232 outsz++;
233 }
234 sz = xread(pack_objects.out, cp,
235 sizeof(data) - outsz);
236 if (0 < sz)
237 ;
238 else if (sz == 0) {
239 close(pack_objects.out);
240 pack_objects.out = -1;
241 }
242 else
243 goto fail;
244 sz += outsz;
245 if (1 < sz) {
246 buffered = data[sz-1] & 0xFF;
247 sz--;
248 }
249 else
250 buffered = -1;
251 sz = send_client_data(1, data, sz);
252 if (sz < 0)
253 goto fail;
254 }
255 if (0 <= pe && (pfd[pe].revents & (POLLIN|POLLHUP))) {
256 /* Status ready; we ship that in the side-band
257 * or dump to the standard error.
258 */
259 sz = xread(pack_objects.err, progress,
260 sizeof(progress));
261 if (0 < sz)
262 send_client_data(2, progress, sz);
263 else if (sz == 0) {
264 close(pack_objects.err);
265 pack_objects.err = -1;
266 }
267 else
268 goto fail;
269 }
270 }
271
272 /* See if the children are still there */
273 if (rev_list.pid || pack_objects.pid) {
274 pid = waitpid(-1, &status, WNOHANG);
275 if (!pid)
276 continue;
277 who = ((pid == rev_list.pid) ? "git-rev-list" :
278 (pid == pack_objects.pid) ? "git-pack-objects" :
279 NULL);
280 if (!who) {
281 if (pid < 0) {
282 error("git-upload-pack: %s",
283 strerror(errno));
284 goto fail;
285 }
286 error("git-upload-pack: we weren't "
287 "waiting for %d", pid);
288 continue;
289 }
290 if (!WIFEXITED(status) || WEXITSTATUS(status) > 0) {
291 error("git-upload-pack: %s died with error.",
292 who);
293 goto fail;
294 }
295 if (pid == rev_list.pid)
296 rev_list.pid = 0;
297 if (pid == pack_objects.pid)
298 pack_objects.pid = 0;
299 if (rev_list.pid || pack_objects.pid)
300 continue;
301 }
302
303 /* both died happily */
304 if (pollsize)
305 continue;
306
307 /* flush the data */
308 if (0 <= buffered) {
309 data[0] = buffered;
310 sz = send_client_data(1, data, 1);
311 if (sz < 0)
312 goto fail;
313 fprintf(stderr, "flushed.\n");
314 }
315 if (use_sideband)
316 packet_flush(1);
317 return;
318 }
319 fail:
320 if (pack_objects.pid)
321 kill(pack_objects.pid, SIGKILL);
322 if (rev_list.pid)
323 kill(rev_list.pid, SIGKILL);
324 send_client_data(3, abort_msg, sizeof(abort_msg));
325 die("git-upload-pack: %s", abort_msg);
326 }
327
328 static int got_sha1(char *hex, unsigned char *sha1)
329 {
330 struct object *o;
331 int we_knew_they_have = 0;
332
333 if (get_sha1_hex(hex, sha1))
334 die("git-upload-pack: expected SHA1 object, got '%s'", hex);
335 if (!has_sha1_file(sha1))
336 return -1;
337
338 o = lookup_object(sha1);
339 if (!(o && o->parsed))
340 o = parse_object(sha1);
341 if (!o)
342 die("oops (%s)", sha1_to_hex(sha1));
343 if (o->type == OBJ_COMMIT) {
344 struct commit_list *parents;
345 struct commit *commit = (struct commit *)o;
346 if (o->flags & THEY_HAVE)
347 we_knew_they_have = 1;
348 else
349 o->flags |= THEY_HAVE;
350 if (!oldest_have || (commit->date < oldest_have))
351 oldest_have = commit->date;
352 for (parents = commit->parents;
353 parents;
354 parents = parents->next)
355 parents->item->object.flags |= THEY_HAVE;
356 }
357 if (!we_knew_they_have) {
358 add_object_array(o, NULL, &have_obj);
359 return 1;
360 }
361 return 0;
362 }
363
364 static int reachable(struct commit *want)
365 {
366 struct commit_list *work = NULL;
367
368 insert_by_date(want, &work);
369 while (work) {
370 struct commit_list *list = work->next;
371 struct commit *commit = work->item;
372 free(work);
373 work = list;
374
375 if (commit->object.flags & THEY_HAVE) {
376 want->object.flags |= COMMON_KNOWN;
377 break;
378 }
379 if (!commit->object.parsed)
380 parse_object(commit->object.sha1);
381 if (commit->object.flags & REACHABLE)
382 continue;
383 commit->object.flags |= REACHABLE;
384 if (commit->date < oldest_have)
385 continue;
386 for (list = commit->parents; list; list = list->next) {
387 struct commit *parent = list->item;
388 if (!(parent->object.flags & REACHABLE))
389 insert_by_date(parent, &work);
390 }
391 }
392 want->object.flags |= REACHABLE;
393 clear_commit_marks(want, REACHABLE);
394 free_commit_list(work);
395 return (want->object.flags & COMMON_KNOWN);
396 }
397
398 static int ok_to_give_up(void)
399 {
400 int i;
401
402 if (!have_obj.nr)
403 return 0;
404
405 for (i = 0; i < want_obj.nr; i++) {
406 struct object *want = want_obj.objects[i].item;
407
408 if (want->flags & COMMON_KNOWN)
409 continue;
410 want = deref_tag(want, "a want line", 0);
411 if (!want || want->type != OBJ_COMMIT) {
412 /* no way to tell if this is reachable by
413 * looking at the ancestry chain alone, so
414 * leave a note to ourselves not to worry about
415 * this object anymore.
416 */
417 want_obj.objects[i].item->flags |= COMMON_KNOWN;
418 continue;
419 }
420 if (!reachable((struct commit *)want))
421 return 0;
422 }
423 return 1;
424 }
425
426 static int get_common_commits(void)
427 {
428 static char line[1000];
429 unsigned char sha1[20];
430 char hex[41], last_hex[41];
431 int len;
432
433 track_object_refs = 0;
434 save_commit_buffer = 0;
435
436 for(;;) {
437 len = packet_read_line(0, line, sizeof(line));
438 reset_timeout();
439
440 if (!len) {
441 if (have_obj.nr == 0 || multi_ack)
442 packet_write(1, "NAK\n");
443 continue;
444 }
445 len = strip(line, len);
446 if (!prefixcmp(line, "have ")) {
447 switch (got_sha1(line+5, sha1)) {
448 case -1: /* they have what we do not */
449 if (multi_ack && ok_to_give_up())
450 packet_write(1, "ACK %s continue\n",
451 sha1_to_hex(sha1));
452 break;
453 default:
454 memcpy(hex, sha1_to_hex(sha1), 41);
455 if (multi_ack) {
456 const char *msg = "ACK %s continue\n";
457 packet_write(1, msg, hex);
458 memcpy(last_hex, hex, 41);
459 }
460 else if (have_obj.nr == 1)
461 packet_write(1, "ACK %s\n", hex);
462 break;
463 }
464 continue;
465 }
466 if (!strcmp(line, "done")) {
467 if (have_obj.nr > 0) {
468 if (multi_ack)
469 packet_write(1, "ACK %s\n", last_hex);
470 return 0;
471 }
472 packet_write(1, "NAK\n");
473 return -1;
474 }
475 die("git-upload-pack: expected SHA1 list, got '%s'", line);
476 }
477 }
478
479 static void receive_needs(void)
480 {
481 struct object_array shallows = {0, 0, NULL};
482 static char line[1000];
483 int len, depth = 0;
484
485 for (;;) {
486 struct object *o;
487 unsigned char sha1_buf[20];
488 len = packet_read_line(0, line, sizeof(line));
489 reset_timeout();
490 if (!len)
491 break;
492
493 if (!prefixcmp(line, "shallow ")) {
494 unsigned char sha1[20];
495 struct object *object;
496 use_thin_pack = 0;
497 if (get_sha1(line + 8, sha1))
498 die("invalid shallow line: %s", line);
499 object = parse_object(sha1);
500 if (!object)
501 die("did not find object for %s", line);
502 object->flags |= CLIENT_SHALLOW;
503 add_object_array(object, NULL, &shallows);
504 continue;
505 }
506 if (!prefixcmp(line, "deepen ")) {
507 char *end;
508 use_thin_pack = 0;
509 depth = strtol(line + 7, &end, 0);
510 if (end == line + 7 || depth <= 0)
511 die("Invalid deepen: %s", line);
512 continue;
513 }
514 if (prefixcmp(line, "want ") ||
515 get_sha1_hex(line+5, sha1_buf))
516 die("git-upload-pack: protocol error, "
517 "expected to get sha, not '%s'", line);
518 if (strstr(line+45, "multi_ack"))
519 multi_ack = 1;
520 if (strstr(line+45, "thin-pack"))
521 use_thin_pack = 1;
522 if (strstr(line+45, "ofs-delta"))
523 use_ofs_delta = 1;
524 if (strstr(line+45, "side-band-64k"))
525 use_sideband = LARGE_PACKET_MAX;
526 else if (strstr(line+45, "side-band"))
527 use_sideband = DEFAULT_PACKET_MAX;
528 if (strstr(line+45, "no-progress"))
529 no_progress = 1;
530
531 /* We have sent all our refs already, and the other end
532 * should have chosen out of them; otherwise they are
533 * asking for nonsense.
534 *
535 * Hmph. We may later want to allow "want" line that
536 * asks for something like "master~10" (symbolic)...
537 * would it make sense? I don't know.
538 */
539 o = lookup_object(sha1_buf);
540 if (!o || !(o->flags & OUR_REF))
541 die("git-upload-pack: not our ref %s", line+5);
542 if (!(o->flags & WANTED)) {
543 o->flags |= WANTED;
544 add_object_array(o, NULL, &want_obj);
545 }
546 }
547 if (depth == 0 && shallows.nr == 0)
548 return;
549 if (depth > 0) {
550 struct commit_list *result, *backup;
551 int i;
552 backup = result = get_shallow_commits(&want_obj, depth,
553 SHALLOW, NOT_SHALLOW);
554 while (result) {
555 struct object *object = &result->item->object;
556 if (!(object->flags & (CLIENT_SHALLOW|NOT_SHALLOW))) {
557 packet_write(1, "shallow %s",
558 sha1_to_hex(object->sha1));
559 register_shallow(object->sha1);
560 }
561 result = result->next;
562 }
563 free_commit_list(backup);
564 for (i = 0; i < shallows.nr; i++) {
565 struct object *object = shallows.objects[i].item;
566 if (object->flags & NOT_SHALLOW) {
567 struct commit_list *parents;
568 packet_write(1, "unshallow %s",
569 sha1_to_hex(object->sha1));
570 object->flags &= ~CLIENT_SHALLOW;
571 /* make sure the real parents are parsed */
572 unregister_shallow(object->sha1);
573 object->parsed = 0;
574 parse_commit((struct commit *)object);
575 parents = ((struct commit *)object)->parents;
576 while (parents) {
577 add_object_array(&parents->item->object,
578 NULL, &want_obj);
579 parents = parents->next;
580 }
581 }
582 /* make sure commit traversal conforms to client */
583 register_shallow(object->sha1);
584 }
585 packet_flush(1);
586 } else
587 if (shallows.nr > 0) {
588 int i;
589 for (i = 0; i < shallows.nr; i++)
590 register_shallow(shallows.objects[i].item->sha1);
591 }
592 free(shallows.objects);
593 }
594
595 static int send_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
596 {
597 static const char *capabilities = "multi_ack thin-pack side-band"
598 " side-band-64k ofs-delta shallow no-progress";
599 struct object *o = parse_object(sha1);
600
601 if (!o)
602 die("git-upload-pack: cannot find object %s:", sha1_to_hex(sha1));
603
604 if (capabilities)
605 packet_write(1, "%s %s%c%s\n", sha1_to_hex(sha1), refname,
606 0, capabilities);
607 else
608 packet_write(1, "%s %s\n", sha1_to_hex(sha1), refname);
609 capabilities = NULL;
610 if (!(o->flags & OUR_REF)) {
611 o->flags |= OUR_REF;
612 nr_our_refs++;
613 }
614 if (o->type == OBJ_TAG) {
615 o = deref_tag(o, refname, 0);
616 packet_write(1, "%s %s^{}\n", sha1_to_hex(o->sha1), refname);
617 }
618 return 0;
619 }
620
621 static void upload_pack(void)
622 {
623 reset_timeout();
624 head_ref(send_ref, NULL);
625 for_each_ref(send_ref, NULL);
626 packet_flush(1);
627 receive_needs();
628 if (want_obj.nr) {
629 get_common_commits();
630 create_pack_file();
631 }
632 }
633
634 int main(int argc, char **argv)
635 {
636 char *dir;
637 int i;
638 int strict = 0;
639
640 for (i = 1; i < argc; i++) {
641 char *arg = argv[i];
642
643 if (arg[0] != '-')
644 break;
645 if (!strcmp(arg, "--strict")) {
646 strict = 1;
647 continue;
648 }
649 if (!prefixcmp(arg, "--timeout=")) {
650 timeout = atoi(arg+10);
651 continue;
652 }
653 if (!strcmp(arg, "--")) {
654 i++;
655 break;
656 }
657 }
658
659 if (i != argc-1)
660 usage(upload_pack_usage);
661 dir = argv[i];
662
663 if (!enter_repo(dir, strict))
664 die("'%s': unable to chdir or not a git archive", dir);
665 if (is_repository_shallow())
666 die("attempt to fetch/clone from a shallow repository");
667 upload_pack();
668 return 0;
669 }