upload-pack: start pack-objects before async rev-list
[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=<n>] <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, use_include_tag;
31 static int no_progress, daemon_mode;
32 static int shallow_nr;
33 static struct object_array have_obj;
34 static struct object_array want_obj;
35 static struct object_array extra_edge_obj;
36 static unsigned int timeout;
37 /* 0 for no sideband,
38 * otherwise maximum packet size (up to 65520 bytes).
39 */
40 static int use_sideband;
41 static int debug_fd;
42 static int advertise_refs;
43 static int stateless_rpc;
44
45 static void reset_timeout(void)
46 {
47 alarm(timeout);
48 }
49
50 static int strip(char *line, int len)
51 {
52 if (len && line[len-1] == '\n')
53 line[--len] = 0;
54 return len;
55 }
56
57 static ssize_t send_client_data(int fd, const char *data, ssize_t sz)
58 {
59 if (use_sideband)
60 return send_sideband(1, fd, data, sz, use_sideband);
61 if (fd == 3)
62 /* emergency quit */
63 fd = 2;
64 if (fd == 2) {
65 /* XXX: are we happy to lose stuff here? */
66 xwrite(fd, data, sz);
67 return sz;
68 }
69 return safe_write(fd, data, sz);
70 }
71
72 static FILE *pack_pipe = NULL;
73 static void show_commit(struct commit *commit, void *data)
74 {
75 if (commit->object.flags & BOUNDARY)
76 fputc('-', pack_pipe);
77 if (fputs(sha1_to_hex(commit->object.sha1), pack_pipe) < 0)
78 die("broken output pipe");
79 fputc('\n', pack_pipe);
80 fflush(pack_pipe);
81 free(commit->buffer);
82 commit->buffer = NULL;
83 }
84
85 static void show_object(struct object *obj, const struct name_path *path, const char *component)
86 {
87 /* An object with name "foo\n0000000..." can be used to
88 * confuse downstream git-pack-objects very badly.
89 */
90 const char *name = path_name(path, component);
91 const char *ep = strchr(name, '\n');
92 if (ep) {
93 fprintf(pack_pipe, "%s %.*s\n", sha1_to_hex(obj->sha1),
94 (int) (ep - name),
95 name);
96 }
97 else
98 fprintf(pack_pipe, "%s %s\n",
99 sha1_to_hex(obj->sha1), name);
100 free((char *)name);
101 }
102
103 static void show_edge(struct commit *commit)
104 {
105 fprintf(pack_pipe, "-%s\n", sha1_to_hex(commit->object.sha1));
106 }
107
108 static int do_rev_list(int in, int out, void *user_data)
109 {
110 int i;
111 struct rev_info revs;
112
113 pack_pipe = xfdopen(out, "w");
114 init_revisions(&revs, NULL);
115 revs.tag_objects = 1;
116 revs.tree_objects = 1;
117 revs.blob_objects = 1;
118 if (use_thin_pack)
119 revs.edge_hint = 1;
120
121 for (i = 0; i < want_obj.nr; i++) {
122 struct object *o = want_obj.objects[i].item;
123 /* why??? */
124 o->flags &= ~UNINTERESTING;
125 add_pending_object(&revs, o, NULL);
126 }
127 for (i = 0; i < have_obj.nr; i++) {
128 struct object *o = have_obj.objects[i].item;
129 o->flags |= UNINTERESTING;
130 add_pending_object(&revs, o, NULL);
131 }
132 setup_revisions(0, NULL, &revs, NULL);
133 if (prepare_revision_walk(&revs))
134 die("revision walk setup failed");
135 mark_edges_uninteresting(revs.commits, &revs, show_edge);
136 if (use_thin_pack)
137 for (i = 0; i < extra_edge_obj.nr; i++)
138 fprintf(pack_pipe, "-%s\n", sha1_to_hex(
139 extra_edge_obj.objects[i].item->sha1));
140 traverse_commit_list(&revs, show_commit, show_object, NULL);
141 fflush(pack_pipe);
142 fclose(pack_pipe);
143 return 0;
144 }
145
146 static void create_pack_file(void)
147 {
148 struct async rev_list;
149 struct child_process pack_objects;
150 int create_full_pack = (nr_our_refs == want_obj.nr && !have_obj.nr);
151 char data[8193], progress[128];
152 char abort_msg[] = "aborting due to possible repository "
153 "corruption on the remote side.";
154 int buffered = -1;
155 ssize_t sz;
156 const char *argv[10];
157 int arg = 0;
158
159 argv[arg++] = "pack-objects";
160 if (!shallow_nr) {
161 argv[arg++] = "--revs";
162 if (create_full_pack)
163 argv[arg++] = "--all";
164 else if (use_thin_pack)
165 argv[arg++] = "--thin";
166 }
167
168 argv[arg++] = "--stdout";
169 if (!no_progress)
170 argv[arg++] = "--progress";
171 if (use_ofs_delta)
172 argv[arg++] = "--delta-base-offset";
173 if (use_include_tag)
174 argv[arg++] = "--include-tag";
175 argv[arg++] = NULL;
176
177 memset(&pack_objects, 0, sizeof(pack_objects));
178 pack_objects.in = -1;
179 pack_objects.out = -1;
180 pack_objects.err = -1;
181 pack_objects.git_cmd = 1;
182 pack_objects.argv = argv;
183
184 if (start_command(&pack_objects))
185 die("git upload-pack: unable to fork git-pack-objects");
186
187 if (shallow_nr) {
188 memset(&rev_list, 0, sizeof(rev_list));
189 rev_list.proc = do_rev_list;
190 rev_list.out = pack_objects.in;
191 if (start_async(&rev_list))
192 die("git upload-pack: unable to fork git-rev-list");
193 }
194 else {
195 FILE *pipe_fd = xfdopen(pack_objects.in, "w");
196 if (!create_full_pack) {
197 int i;
198 for (i = 0; i < want_obj.nr; i++)
199 fprintf(pipe_fd, "%s\n", sha1_to_hex(want_obj.objects[i].item->sha1));
200 fprintf(pipe_fd, "--not\n");
201 for (i = 0; i < have_obj.nr; i++)
202 fprintf(pipe_fd, "%s\n", sha1_to_hex(have_obj.objects[i].item->sha1));
203 }
204
205 fprintf(pipe_fd, "\n");
206 fflush(pipe_fd);
207 fclose(pipe_fd);
208 }
209
210
211 /* We read from pack_objects.err to capture stderr output for
212 * progress bar, and pack_objects.out to capture the pack data.
213 */
214
215 while (1) {
216 struct pollfd pfd[2];
217 int pe, pu, pollsize;
218
219 reset_timeout();
220
221 pollsize = 0;
222 pe = pu = -1;
223
224 if (0 <= pack_objects.out) {
225 pfd[pollsize].fd = pack_objects.out;
226 pfd[pollsize].events = POLLIN;
227 pu = pollsize;
228 pollsize++;
229 }
230 if (0 <= pack_objects.err) {
231 pfd[pollsize].fd = pack_objects.err;
232 pfd[pollsize].events = POLLIN;
233 pe = pollsize;
234 pollsize++;
235 }
236
237 if (!pollsize)
238 break;
239
240 if (poll(pfd, pollsize, -1) < 0) {
241 if (errno != EINTR) {
242 error("poll failed, resuming: %s",
243 strerror(errno));
244 sleep(1);
245 }
246 continue;
247 }
248 if (0 <= pe && (pfd[pe].revents & (POLLIN|POLLHUP))) {
249 /* Status ready; we ship that in the side-band
250 * or dump to the standard error.
251 */
252 sz = xread(pack_objects.err, progress,
253 sizeof(progress));
254 if (0 < sz)
255 send_client_data(2, progress, sz);
256 else if (sz == 0) {
257 close(pack_objects.err);
258 pack_objects.err = -1;
259 }
260 else
261 goto fail;
262 /* give priority to status messages */
263 continue;
264 }
265 if (0 <= pu && (pfd[pu].revents & (POLLIN|POLLHUP))) {
266 /* Data ready; we keep the last byte to ourselves
267 * in case we detect broken rev-list, so that we
268 * can leave the stream corrupted. This is
269 * unfortunate -- unpack-objects would happily
270 * accept a valid packdata with trailing garbage,
271 * so appending garbage after we pass all the
272 * pack data is not good enough to signal
273 * breakage to downstream.
274 */
275 char *cp = data;
276 ssize_t outsz = 0;
277 if (0 <= buffered) {
278 *cp++ = buffered;
279 outsz++;
280 }
281 sz = xread(pack_objects.out, cp,
282 sizeof(data) - outsz);
283 if (0 < sz)
284 ;
285 else if (sz == 0) {
286 close(pack_objects.out);
287 pack_objects.out = -1;
288 }
289 else
290 goto fail;
291 sz += outsz;
292 if (1 < sz) {
293 buffered = data[sz-1] & 0xFF;
294 sz--;
295 }
296 else
297 buffered = -1;
298 sz = send_client_data(1, data, sz);
299 if (sz < 0)
300 goto fail;
301 }
302 }
303
304 if (finish_command(&pack_objects)) {
305 error("git upload-pack: git-pack-objects died with error.");
306 goto fail;
307 }
308 if (shallow_nr && finish_async(&rev_list))
309 goto fail; /* error was already reported */
310
311 /* flush the data */
312 if (0 <= buffered) {
313 data[0] = buffered;
314 sz = send_client_data(1, data, 1);
315 if (sz < 0)
316 goto fail;
317 fprintf(stderr, "flushed.\n");
318 }
319 if (use_sideband)
320 packet_flush(1);
321 return;
322
323 fail:
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 commit_list_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 commit_list_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 last_hex[41];
431
432 save_commit_buffer = 0;
433
434 for (;;) {
435 int len = packet_read_line(0, line, sizeof(line));
436 reset_timeout();
437
438 if (!len) {
439 if (have_obj.nr == 0 || multi_ack)
440 packet_write(1, "NAK\n");
441 if (stateless_rpc)
442 exit(0);
443 continue;
444 }
445 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 const char *hex = sha1_to_hex(sha1);
451 if (multi_ack == 2)
452 packet_write(1, "ACK %s ready\n", hex);
453 else
454 packet_write(1, "ACK %s continue\n", hex);
455 }
456 break;
457 default:
458 memcpy(last_hex, sha1_to_hex(sha1), 41);
459 if (multi_ack == 2)
460 packet_write(1, "ACK %s common\n", last_hex);
461 else if (multi_ack)
462 packet_write(1, "ACK %s continue\n", last_hex);
463 else if (have_obj.nr == 1)
464 packet_write(1, "ACK %s\n", last_hex);
465 break;
466 }
467 continue;
468 }
469 if (!strcmp(line, "done")) {
470 if (have_obj.nr > 0) {
471 if (multi_ack)
472 packet_write(1, "ACK %s\n", last_hex);
473 return 0;
474 }
475 packet_write(1, "NAK\n");
476 return -1;
477 }
478 die("git upload-pack: expected SHA1 list, got '%s'", line);
479 }
480 }
481
482 static void receive_needs(void)
483 {
484 struct object_array shallows = OBJECT_ARRAY_INIT;
485 static char line[1000];
486 int len, depth = 0;
487
488 shallow_nr = 0;
489 if (debug_fd)
490 write_str_in_full(debug_fd, "#S\n");
491 for (;;) {
492 struct object *o;
493 unsigned char sha1_buf[20];
494 len = packet_read_line(0, line, sizeof(line));
495 reset_timeout();
496 if (!len)
497 break;
498 if (debug_fd)
499 write_in_full(debug_fd, line, len);
500
501 if (!prefixcmp(line, "shallow ")) {
502 unsigned char sha1[20];
503 struct object *object;
504 if (get_sha1(line + 8, sha1))
505 die("invalid shallow line: %s", line);
506 object = parse_object(sha1);
507 if (!object)
508 die("did not find object for %s", line);
509 object->flags |= CLIENT_SHALLOW;
510 add_object_array(object, NULL, &shallows);
511 continue;
512 }
513 if (!prefixcmp(line, "deepen ")) {
514 char *end;
515 depth = strtol(line + 7, &end, 0);
516 if (end == line + 7 || depth <= 0)
517 die("Invalid deepen: %s", line);
518 continue;
519 }
520 if (prefixcmp(line, "want ") ||
521 get_sha1_hex(line+5, sha1_buf))
522 die("git upload-pack: protocol error, "
523 "expected to get sha, not '%s'", line);
524 if (strstr(line+45, "multi_ack_detailed"))
525 multi_ack = 2;
526 else if (strstr(line+45, "multi_ack"))
527 multi_ack = 1;
528 if (strstr(line+45, "thin-pack"))
529 use_thin_pack = 1;
530 if (strstr(line+45, "ofs-delta"))
531 use_ofs_delta = 1;
532 if (strstr(line+45, "side-band-64k"))
533 use_sideband = LARGE_PACKET_MAX;
534 else if (strstr(line+45, "side-band"))
535 use_sideband = DEFAULT_PACKET_MAX;
536 if (strstr(line+45, "no-progress"))
537 no_progress = 1;
538 if (strstr(line+45, "include-tag"))
539 use_include_tag = 1;
540
541 /* We have sent all our refs already, and the other end
542 * should have chosen out of them; otherwise they are
543 * asking for nonsense.
544 *
545 * Hmph. We may later want to allow "want" line that
546 * asks for something like "master~10" (symbolic)...
547 * would it make sense? I don't know.
548 */
549 o = lookup_object(sha1_buf);
550 if (!o || !(o->flags & OUR_REF))
551 die("git upload-pack: not our ref %s",
552 sha1_to_hex(sha1_buf));
553 if (!(o->flags & WANTED)) {
554 o->flags |= WANTED;
555 add_object_array(o, NULL, &want_obj);
556 }
557 }
558 if (debug_fd)
559 write_str_in_full(debug_fd, "#E\n");
560
561 if (!use_sideband && daemon_mode)
562 no_progress = 1;
563
564 if (depth == 0 && shallows.nr == 0)
565 return;
566 if (depth > 0) {
567 struct commit_list *result, *backup;
568 int i;
569 backup = result = get_shallow_commits(&want_obj, depth,
570 SHALLOW, NOT_SHALLOW);
571 while (result) {
572 struct object *object = &result->item->object;
573 if (!(object->flags & (CLIENT_SHALLOW|NOT_SHALLOW))) {
574 packet_write(1, "shallow %s",
575 sha1_to_hex(object->sha1));
576 register_shallow(object->sha1);
577 shallow_nr++;
578 }
579 result = result->next;
580 }
581 free_commit_list(backup);
582 for (i = 0; i < shallows.nr; i++) {
583 struct object *object = shallows.objects[i].item;
584 if (object->flags & NOT_SHALLOW) {
585 struct commit_list *parents;
586 packet_write(1, "unshallow %s",
587 sha1_to_hex(object->sha1));
588 object->flags &= ~CLIENT_SHALLOW;
589 /* make sure the real parents are parsed */
590 unregister_shallow(object->sha1);
591 object->parsed = 0;
592 if (parse_commit((struct commit *)object))
593 die("invalid commit");
594 parents = ((struct commit *)object)->parents;
595 while (parents) {
596 add_object_array(&parents->item->object,
597 NULL, &want_obj);
598 parents = parents->next;
599 }
600 add_object_array(object, NULL, &extra_edge_obj);
601 }
602 /* make sure commit traversal conforms to client */
603 register_shallow(object->sha1);
604 }
605 packet_flush(1);
606 } else
607 if (shallows.nr > 0) {
608 int i;
609 for (i = 0; i < shallows.nr; i++)
610 register_shallow(shallows.objects[i].item->sha1);
611 }
612
613 shallow_nr += shallows.nr;
614 free(shallows.objects);
615 }
616
617 static int send_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
618 {
619 static const char *capabilities = "multi_ack thin-pack side-band"
620 " side-band-64k ofs-delta shallow no-progress"
621 " include-tag multi_ack_detailed";
622 struct object *o = parse_object(sha1);
623
624 if (!o)
625 die("git upload-pack: cannot find object %s:", sha1_to_hex(sha1));
626
627 if (capabilities)
628 packet_write(1, "%s %s%c%s\n", sha1_to_hex(sha1), refname,
629 0, capabilities);
630 else
631 packet_write(1, "%s %s\n", sha1_to_hex(sha1), refname);
632 capabilities = NULL;
633 if (!(o->flags & OUR_REF)) {
634 o->flags |= OUR_REF;
635 nr_our_refs++;
636 }
637 if (o->type == OBJ_TAG) {
638 o = deref_tag(o, refname, 0);
639 if (o)
640 packet_write(1, "%s %s^{}\n", sha1_to_hex(o->sha1), refname);
641 }
642 return 0;
643 }
644
645 static int mark_our_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
646 {
647 struct object *o = parse_object(sha1);
648 if (!o)
649 die("git upload-pack: cannot find object %s:", sha1_to_hex(sha1));
650 if (!(o->flags & OUR_REF)) {
651 o->flags |= OUR_REF;
652 nr_our_refs++;
653 }
654 return 0;
655 }
656
657 static void upload_pack(void)
658 {
659 if (advertise_refs || !stateless_rpc) {
660 reset_timeout();
661 head_ref(send_ref, NULL);
662 for_each_ref(send_ref, NULL);
663 packet_flush(1);
664 } else {
665 head_ref(mark_our_ref, NULL);
666 for_each_ref(mark_our_ref, NULL);
667 }
668 if (advertise_refs)
669 return;
670
671 receive_needs();
672 if (want_obj.nr) {
673 get_common_commits();
674 create_pack_file();
675 }
676 }
677
678 int main(int argc, char **argv)
679 {
680 char *dir;
681 int i;
682 int strict = 0;
683
684 git_extract_argv0_path(argv[0]);
685 read_replace_refs = 0;
686
687 for (i = 1; i < argc; i++) {
688 char *arg = argv[i];
689
690 if (arg[0] != '-')
691 break;
692 if (!strcmp(arg, "--advertise-refs")) {
693 advertise_refs = 1;
694 continue;
695 }
696 if (!strcmp(arg, "--stateless-rpc")) {
697 stateless_rpc = 1;
698 continue;
699 }
700 if (!strcmp(arg, "--strict")) {
701 strict = 1;
702 continue;
703 }
704 if (!prefixcmp(arg, "--timeout=")) {
705 timeout = atoi(arg+10);
706 daemon_mode = 1;
707 continue;
708 }
709 if (!strcmp(arg, "--")) {
710 i++;
711 break;
712 }
713 }
714
715 if (i != argc-1)
716 usage(upload_pack_usage);
717
718 setup_path();
719
720 dir = argv[i];
721
722 if (!enter_repo(dir, strict))
723 die("'%s' does not appear to be a git repository", dir);
724 if (is_repository_shallow())
725 die("attempt to fetch/clone from a shallow repository");
726 if (getenv("GIT_DEBUG_SEND_PACK"))
727 debug_fd = atoi(getenv("GIT_DEBUG_SEND_PACK"));
728 upload_pack();
729 return 0;
730 }