Merge branch 'vd/doc-unpack-objects' into maint
[git.git] / upload-pack.c
blob8327dc0b7c80b61fc24342a1816e323009818729
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 #include "sigchain.h"
14 #include "version.h"
15 #include "string-list.h"
17 static const char upload_pack_usage[] = "git upload-pack [--strict] [--timeout=<n>] <dir>";
19 /* bits #0..7 in revision.h, #8..10 in commit.c */
20 #define THEY_HAVE (1u << 11)
21 #define OUR_REF (1u << 12)
22 #define WANTED (1u << 13)
23 #define COMMON_KNOWN (1u << 14)
24 #define REACHABLE (1u << 15)
26 #define SHALLOW (1u << 16)
27 #define NOT_SHALLOW (1u << 17)
28 #define CLIENT_SHALLOW (1u << 18)
29 #define HIDDEN_REF (1u << 19)
31 static unsigned long oldest_have;
33 static int multi_ack;
34 static int no_done;
35 static int use_thin_pack, use_ofs_delta, use_include_tag;
36 static int no_progress, daemon_mode;
37 static int allow_tip_sha1_in_want;
38 static int shallow_nr;
39 static struct object_array have_obj;
40 static struct object_array want_obj;
41 static struct object_array extra_edge_obj;
42 static unsigned int timeout;
43 static int keepalive = 5;
44 /* 0 for no sideband,
45 * otherwise maximum packet size (up to 65520 bytes).
47 static int use_sideband;
48 static int advertise_refs;
49 static int stateless_rpc;
51 static void reset_timeout(void)
53 alarm(timeout);
56 static ssize_t send_client_data(int fd, const char *data, ssize_t sz)
58 if (use_sideband)
59 return send_sideband(1, fd, data, sz, use_sideband);
60 if (fd == 3)
61 /* emergency quit */
62 fd = 2;
63 if (fd == 2) {
64 /* XXX: are we happy to lose stuff here? */
65 xwrite(fd, data, sz);
66 return sz;
68 write_or_die(fd, data, sz);
69 return sz;
72 static void create_pack_file(void)
74 struct child_process pack_objects;
75 char data[8193], progress[128];
76 char abort_msg[] = "aborting due to possible repository "
77 "corruption on the remote side.";
78 int buffered = -1;
79 ssize_t sz;
80 const char *argv[12];
81 int i, arg = 0;
82 FILE *pipe_fd;
83 char *shallow_file = NULL;
85 if (shallow_nr) {
86 shallow_file = setup_temporary_shallow();
87 argv[arg++] = "--shallow-file";
88 argv[arg++] = shallow_file;
90 argv[arg++] = "pack-objects";
91 argv[arg++] = "--revs";
92 if (use_thin_pack)
93 argv[arg++] = "--thin";
95 argv[arg++] = "--stdout";
96 if (!no_progress)
97 argv[arg++] = "--progress";
98 if (use_ofs_delta)
99 argv[arg++] = "--delta-base-offset";
100 if (use_include_tag)
101 argv[arg++] = "--include-tag";
102 argv[arg++] = NULL;
104 memset(&pack_objects, 0, sizeof(pack_objects));
105 pack_objects.in = -1;
106 pack_objects.out = -1;
107 pack_objects.err = -1;
108 pack_objects.git_cmd = 1;
109 pack_objects.argv = argv;
111 if (start_command(&pack_objects))
112 die("git upload-pack: unable to fork git-pack-objects");
114 pipe_fd = xfdopen(pack_objects.in, "w");
116 for (i = 0; i < want_obj.nr; i++)
117 fprintf(pipe_fd, "%s\n",
118 sha1_to_hex(want_obj.objects[i].item->sha1));
119 fprintf(pipe_fd, "--not\n");
120 for (i = 0; i < have_obj.nr; i++)
121 fprintf(pipe_fd, "%s\n",
122 sha1_to_hex(have_obj.objects[i].item->sha1));
123 for (i = 0; i < extra_edge_obj.nr; i++)
124 fprintf(pipe_fd, "%s\n",
125 sha1_to_hex(extra_edge_obj.objects[i].item->sha1));
126 fprintf(pipe_fd, "\n");
127 fflush(pipe_fd);
128 fclose(pipe_fd);
130 /* We read from pack_objects.err to capture stderr output for
131 * progress bar, and pack_objects.out to capture the pack data.
134 while (1) {
135 struct pollfd pfd[2];
136 int pe, pu, pollsize;
137 int ret;
139 reset_timeout();
141 pollsize = 0;
142 pe = pu = -1;
144 if (0 <= pack_objects.out) {
145 pfd[pollsize].fd = pack_objects.out;
146 pfd[pollsize].events = POLLIN;
147 pu = pollsize;
148 pollsize++;
150 if (0 <= pack_objects.err) {
151 pfd[pollsize].fd = pack_objects.err;
152 pfd[pollsize].events = POLLIN;
153 pe = pollsize;
154 pollsize++;
157 if (!pollsize)
158 break;
160 ret = poll(pfd, pollsize, 1000 * keepalive);
161 if (ret < 0) {
162 if (errno != EINTR) {
163 error("poll failed, resuming: %s",
164 strerror(errno));
165 sleep(1);
167 continue;
169 if (0 <= pe && (pfd[pe].revents & (POLLIN|POLLHUP))) {
170 /* Status ready; we ship that in the side-band
171 * or dump to the standard error.
173 sz = xread(pack_objects.err, progress,
174 sizeof(progress));
175 if (0 < sz)
176 send_client_data(2, progress, sz);
177 else if (sz == 0) {
178 close(pack_objects.err);
179 pack_objects.err = -1;
181 else
182 goto fail;
183 /* give priority to status messages */
184 continue;
186 if (0 <= pu && (pfd[pu].revents & (POLLIN|POLLHUP))) {
187 /* Data ready; we keep the last byte to ourselves
188 * in case we detect broken rev-list, so that we
189 * can leave the stream corrupted. This is
190 * unfortunate -- unpack-objects would happily
191 * accept a valid packdata with trailing garbage,
192 * so appending garbage after we pass all the
193 * pack data is not good enough to signal
194 * breakage to downstream.
196 char *cp = data;
197 ssize_t outsz = 0;
198 if (0 <= buffered) {
199 *cp++ = buffered;
200 outsz++;
202 sz = xread(pack_objects.out, cp,
203 sizeof(data) - outsz);
204 if (0 < sz)
206 else if (sz == 0) {
207 close(pack_objects.out);
208 pack_objects.out = -1;
210 else
211 goto fail;
212 sz += outsz;
213 if (1 < sz) {
214 buffered = data[sz-1] & 0xFF;
215 sz--;
217 else
218 buffered = -1;
219 sz = send_client_data(1, data, sz);
220 if (sz < 0)
221 goto fail;
225 * We hit the keepalive timeout without saying anything; send
226 * an empty message on the data sideband just to let the other
227 * side know we're still working on it, but don't have any data
228 * yet.
230 * If we don't have a sideband channel, there's no room in the
231 * protocol to say anything, so those clients are just out of
232 * luck.
234 if (!ret && use_sideband) {
235 static const char buf[] = "0005\1";
236 write_or_die(1, buf, 5);
240 if (finish_command(&pack_objects)) {
241 error("git upload-pack: git-pack-objects died with error.");
242 goto fail;
244 if (shallow_file) {
245 if (*shallow_file)
246 unlink(shallow_file);
247 free(shallow_file);
250 /* flush the data */
251 if (0 <= buffered) {
252 data[0] = buffered;
253 sz = send_client_data(1, data, 1);
254 if (sz < 0)
255 goto fail;
256 fprintf(stderr, "flushed.\n");
258 if (use_sideband)
259 packet_flush(1);
260 return;
262 fail:
263 send_client_data(3, abort_msg, sizeof(abort_msg));
264 die("git upload-pack: %s", abort_msg);
267 static int got_sha1(char *hex, unsigned char *sha1)
269 struct object *o;
270 int we_knew_they_have = 0;
272 if (get_sha1_hex(hex, sha1))
273 die("git upload-pack: expected SHA1 object, got '%s'", hex);
274 if (!has_sha1_file(sha1))
275 return -1;
277 o = parse_object(sha1);
278 if (!o)
279 die("oops (%s)", sha1_to_hex(sha1));
280 if (o->type == OBJ_COMMIT) {
281 struct commit_list *parents;
282 struct commit *commit = (struct commit *)o;
283 if (o->flags & THEY_HAVE)
284 we_knew_they_have = 1;
285 else
286 o->flags |= THEY_HAVE;
287 if (!oldest_have || (commit->date < oldest_have))
288 oldest_have = commit->date;
289 for (parents = commit->parents;
290 parents;
291 parents = parents->next)
292 parents->item->object.flags |= THEY_HAVE;
294 if (!we_knew_they_have) {
295 add_object_array(o, NULL, &have_obj);
296 return 1;
298 return 0;
301 static int reachable(struct commit *want)
303 struct commit_list *work = NULL;
305 commit_list_insert_by_date(want, &work);
306 while (work) {
307 struct commit_list *list = work->next;
308 struct commit *commit = work->item;
309 free(work);
310 work = list;
312 if (commit->object.flags & THEY_HAVE) {
313 want->object.flags |= COMMON_KNOWN;
314 break;
316 if (!commit->object.parsed)
317 parse_object(commit->object.sha1);
318 if (commit->object.flags & REACHABLE)
319 continue;
320 commit->object.flags |= REACHABLE;
321 if (commit->date < oldest_have)
322 continue;
323 for (list = commit->parents; list; list = list->next) {
324 struct commit *parent = list->item;
325 if (!(parent->object.flags & REACHABLE))
326 commit_list_insert_by_date(parent, &work);
329 want->object.flags |= REACHABLE;
330 clear_commit_marks(want, REACHABLE);
331 free_commit_list(work);
332 return (want->object.flags & COMMON_KNOWN);
335 static int ok_to_give_up(void)
337 int i;
339 if (!have_obj.nr)
340 return 0;
342 for (i = 0; i < want_obj.nr; i++) {
343 struct object *want = want_obj.objects[i].item;
345 if (want->flags & COMMON_KNOWN)
346 continue;
347 want = deref_tag(want, "a want line", 0);
348 if (!want || want->type != OBJ_COMMIT) {
349 /* no way to tell if this is reachable by
350 * looking at the ancestry chain alone, so
351 * leave a note to ourselves not to worry about
352 * this object anymore.
354 want_obj.objects[i].item->flags |= COMMON_KNOWN;
355 continue;
357 if (!reachable((struct commit *)want))
358 return 0;
360 return 1;
363 static int get_common_commits(void)
365 unsigned char sha1[20];
366 char last_hex[41];
367 int got_common = 0;
368 int got_other = 0;
369 int sent_ready = 0;
371 save_commit_buffer = 0;
373 for (;;) {
374 char *line = packet_read_line(0, NULL);
375 reset_timeout();
377 if (!line) {
378 if (multi_ack == 2 && got_common
379 && !got_other && ok_to_give_up()) {
380 sent_ready = 1;
381 packet_write(1, "ACK %s ready\n", last_hex);
383 if (have_obj.nr == 0 || multi_ack)
384 packet_write(1, "NAK\n");
386 if (no_done && sent_ready) {
387 packet_write(1, "ACK %s\n", last_hex);
388 return 0;
390 if (stateless_rpc)
391 exit(0);
392 got_common = 0;
393 got_other = 0;
394 continue;
396 if (!prefixcmp(line, "have ")) {
397 switch (got_sha1(line+5, sha1)) {
398 case -1: /* they have what we do not */
399 got_other = 1;
400 if (multi_ack && ok_to_give_up()) {
401 const char *hex = sha1_to_hex(sha1);
402 if (multi_ack == 2) {
403 sent_ready = 1;
404 packet_write(1, "ACK %s ready\n", hex);
405 } else
406 packet_write(1, "ACK %s continue\n", hex);
408 break;
409 default:
410 got_common = 1;
411 memcpy(last_hex, sha1_to_hex(sha1), 41);
412 if (multi_ack == 2)
413 packet_write(1, "ACK %s common\n", last_hex);
414 else if (multi_ack)
415 packet_write(1, "ACK %s continue\n", last_hex);
416 else if (have_obj.nr == 1)
417 packet_write(1, "ACK %s\n", last_hex);
418 break;
420 continue;
422 if (!strcmp(line, "done")) {
423 if (have_obj.nr > 0) {
424 if (multi_ack)
425 packet_write(1, "ACK %s\n", last_hex);
426 return 0;
428 packet_write(1, "NAK\n");
429 return -1;
431 die("git upload-pack: expected SHA1 list, got '%s'", line);
435 static int is_our_ref(struct object *o)
437 return o->flags &
438 ((allow_tip_sha1_in_want ? HIDDEN_REF : 0) | OUR_REF);
441 static void check_non_tip(void)
443 static const char *argv[] = {
444 "rev-list", "--stdin", NULL,
446 static struct child_process cmd;
447 struct object *o;
448 char namebuf[42]; /* ^ + SHA-1 + LF */
449 int i;
451 /* In the normal in-process case non-tip request can never happen */
452 if (!stateless_rpc)
453 goto error;
455 cmd.argv = argv;
456 cmd.git_cmd = 1;
457 cmd.no_stderr = 1;
458 cmd.in = -1;
459 cmd.out = -1;
461 if (start_command(&cmd))
462 goto error;
465 * If rev-list --stdin encounters an unknown commit, it
466 * terminates, which will cause SIGPIPE in the write loop
467 * below.
469 sigchain_push(SIGPIPE, SIG_IGN);
471 namebuf[0] = '^';
472 namebuf[41] = '\n';
473 for (i = get_max_object_index(); 0 < i; ) {
474 o = get_indexed_object(--i);
475 if (!o)
476 continue;
477 if (!is_our_ref(o))
478 continue;
479 memcpy(namebuf + 1, sha1_to_hex(o->sha1), 40);
480 if (write_in_full(cmd.in, namebuf, 42) < 0)
481 goto error;
483 namebuf[40] = '\n';
484 for (i = 0; i < want_obj.nr; i++) {
485 o = want_obj.objects[i].item;
486 if (is_our_ref(o))
487 continue;
488 memcpy(namebuf, sha1_to_hex(o->sha1), 40);
489 if (write_in_full(cmd.in, namebuf, 41) < 0)
490 goto error;
492 close(cmd.in);
494 sigchain_pop(SIGPIPE);
497 * The commits out of the rev-list are not ancestors of
498 * our ref.
500 i = read_in_full(cmd.out, namebuf, 1);
501 if (i)
502 goto error;
503 close(cmd.out);
506 * rev-list may have died by encountering a bad commit
507 * in the history, in which case we do want to bail out
508 * even when it showed no commit.
510 if (finish_command(&cmd))
511 goto error;
513 /* All the non-tip ones are ancestors of what we advertised */
514 return;
516 error:
517 /* Pick one of them (we know there at least is one) */
518 for (i = 0; i < want_obj.nr; i++) {
519 o = want_obj.objects[i].item;
520 if (!is_our_ref(o))
521 die("git upload-pack: not our ref %s",
522 sha1_to_hex(o->sha1));
526 static void receive_needs(void)
528 struct object_array shallows = OBJECT_ARRAY_INIT;
529 int depth = 0;
530 int has_non_tip = 0;
532 shallow_nr = 0;
533 for (;;) {
534 struct object *o;
535 const char *features;
536 unsigned char sha1_buf[20];
537 char *line = packet_read_line(0, NULL);
538 reset_timeout();
539 if (!line)
540 break;
542 if (!prefixcmp(line, "shallow ")) {
543 unsigned char sha1[20];
544 struct object *object;
545 if (get_sha1_hex(line + 8, sha1))
546 die("invalid shallow line: %s", line);
547 object = parse_object(sha1);
548 if (!object)
549 continue;
550 if (object->type != OBJ_COMMIT)
551 die("invalid shallow object %s", sha1_to_hex(sha1));
552 if (!(object->flags & CLIENT_SHALLOW)) {
553 object->flags |= CLIENT_SHALLOW;
554 add_object_array(object, NULL, &shallows);
556 continue;
558 if (!prefixcmp(line, "deepen ")) {
559 char *end;
560 depth = strtol(line + 7, &end, 0);
561 if (end == line + 7 || depth <= 0)
562 die("Invalid deepen: %s", line);
563 continue;
565 if (prefixcmp(line, "want ") ||
566 get_sha1_hex(line+5, sha1_buf))
567 die("git upload-pack: protocol error, "
568 "expected to get sha, not '%s'", line);
570 features = line + 45;
572 if (parse_feature_request(features, "multi_ack_detailed"))
573 multi_ack = 2;
574 else if (parse_feature_request(features, "multi_ack"))
575 multi_ack = 1;
576 if (parse_feature_request(features, "no-done"))
577 no_done = 1;
578 if (parse_feature_request(features, "thin-pack"))
579 use_thin_pack = 1;
580 if (parse_feature_request(features, "ofs-delta"))
581 use_ofs_delta = 1;
582 if (parse_feature_request(features, "side-band-64k"))
583 use_sideband = LARGE_PACKET_MAX;
584 else if (parse_feature_request(features, "side-band"))
585 use_sideband = DEFAULT_PACKET_MAX;
586 if (parse_feature_request(features, "no-progress"))
587 no_progress = 1;
588 if (parse_feature_request(features, "include-tag"))
589 use_include_tag = 1;
591 o = parse_object(sha1_buf);
592 if (!o)
593 die("git upload-pack: not our ref %s",
594 sha1_to_hex(sha1_buf));
595 if (!(o->flags & WANTED)) {
596 o->flags |= WANTED;
597 if (!is_our_ref(o))
598 has_non_tip = 1;
599 add_object_array(o, NULL, &want_obj);
604 * We have sent all our refs already, and the other end
605 * should have chosen out of them. When we are operating
606 * in the stateless RPC mode, however, their choice may
607 * have been based on the set of older refs advertised
608 * by another process that handled the initial request.
610 if (has_non_tip)
611 check_non_tip();
613 if (!use_sideband && daemon_mode)
614 no_progress = 1;
616 if (depth == 0 && shallows.nr == 0)
617 return;
618 if (depth > 0) {
619 struct commit_list *result = NULL, *backup = NULL;
620 int i;
621 if (depth == INFINITE_DEPTH)
622 for (i = 0; i < shallows.nr; i++) {
623 struct object *object = shallows.objects[i].item;
624 object->flags |= NOT_SHALLOW;
626 else
627 backup = result =
628 get_shallow_commits(&want_obj, depth,
629 SHALLOW, NOT_SHALLOW);
630 while (result) {
631 struct object *object = &result->item->object;
632 if (!(object->flags & (CLIENT_SHALLOW|NOT_SHALLOW))) {
633 packet_write(1, "shallow %s",
634 sha1_to_hex(object->sha1));
635 register_shallow(object->sha1);
636 shallow_nr++;
638 result = result->next;
640 free_commit_list(backup);
641 for (i = 0; i < shallows.nr; i++) {
642 struct object *object = shallows.objects[i].item;
643 if (object->flags & NOT_SHALLOW) {
644 struct commit_list *parents;
645 packet_write(1, "unshallow %s",
646 sha1_to_hex(object->sha1));
647 object->flags &= ~CLIENT_SHALLOW;
648 /* make sure the real parents are parsed */
649 unregister_shallow(object->sha1);
650 object->parsed = 0;
651 if (parse_commit((struct commit *)object))
652 die("invalid commit");
653 parents = ((struct commit *)object)->parents;
654 while (parents) {
655 add_object_array(&parents->item->object,
656 NULL, &want_obj);
657 parents = parents->next;
659 add_object_array(object, NULL, &extra_edge_obj);
661 /* make sure commit traversal conforms to client */
662 register_shallow(object->sha1);
664 packet_flush(1);
665 } else
666 if (shallows.nr > 0) {
667 int i;
668 for (i = 0; i < shallows.nr; i++)
669 register_shallow(shallows.objects[i].item->sha1);
672 shallow_nr += shallows.nr;
673 free(shallows.objects);
676 /* return non-zero if the ref is hidden, otherwise 0 */
677 static int mark_our_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
679 struct object *o = lookup_unknown_object(sha1);
681 if (ref_is_hidden(refname)) {
682 o->flags |= HIDDEN_REF;
683 return 1;
685 if (!o)
686 die("git upload-pack: cannot find object %s:", sha1_to_hex(sha1));
687 o->flags |= OUR_REF;
688 return 0;
691 static int send_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
693 static const char *capabilities = "multi_ack thin-pack side-band"
694 " side-band-64k ofs-delta shallow no-progress"
695 " include-tag multi_ack_detailed";
696 const char *refname_nons = strip_namespace(refname);
697 unsigned char peeled[20];
699 if (mark_our_ref(refname, sha1, flag, cb_data))
700 return 0;
702 if (capabilities)
703 packet_write(1, "%s %s%c%s%s%s agent=%s\n",
704 sha1_to_hex(sha1), refname_nons,
705 0, capabilities,
706 allow_tip_sha1_in_want ? " allow-tip-sha1-in-want" : "",
707 stateless_rpc ? " no-done" : "",
708 git_user_agent_sanitized());
709 else
710 packet_write(1, "%s %s\n", sha1_to_hex(sha1), refname_nons);
711 capabilities = NULL;
712 if (!peel_ref(refname, peeled))
713 packet_write(1, "%s %s^{}\n", sha1_to_hex(peeled), refname_nons);
714 return 0;
717 static void upload_pack(void)
719 if (advertise_refs || !stateless_rpc) {
720 reset_timeout();
721 head_ref_namespaced(send_ref, NULL);
722 for_each_namespaced_ref(send_ref, NULL);
723 packet_flush(1);
724 } else {
725 head_ref_namespaced(mark_our_ref, NULL);
726 for_each_namespaced_ref(mark_our_ref, NULL);
728 if (advertise_refs)
729 return;
731 receive_needs();
732 if (want_obj.nr) {
733 get_common_commits();
734 create_pack_file();
738 static int upload_pack_config(const char *var, const char *value, void *unused)
740 if (!strcmp("uploadpack.allowtipsha1inwant", var))
741 allow_tip_sha1_in_want = git_config_bool(var, value);
742 else if (!strcmp("uploadpack.keepalive", var)) {
743 keepalive = git_config_int(var, value);
744 if (!keepalive)
745 keepalive = -1;
747 return parse_hide_refs_config(var, value, "uploadpack");
750 int main(int argc, char **argv)
752 char *dir;
753 int i;
754 int strict = 0;
756 git_setup_gettext();
758 packet_trace_identity("upload-pack");
759 git_extract_argv0_path(argv[0]);
760 read_replace_refs = 0;
762 for (i = 1; i < argc; i++) {
763 char *arg = argv[i];
765 if (arg[0] != '-')
766 break;
767 if (!strcmp(arg, "--advertise-refs")) {
768 advertise_refs = 1;
769 continue;
771 if (!strcmp(arg, "--stateless-rpc")) {
772 stateless_rpc = 1;
773 continue;
775 if (!strcmp(arg, "--strict")) {
776 strict = 1;
777 continue;
779 if (!prefixcmp(arg, "--timeout=")) {
780 timeout = atoi(arg+10);
781 daemon_mode = 1;
782 continue;
784 if (!strcmp(arg, "--")) {
785 i++;
786 break;
790 if (i != argc-1)
791 usage(upload_pack_usage);
793 setup_path();
795 dir = argv[i];
797 if (!enter_repo(dir, strict))
798 die("'%s' does not appear to be a git repository", dir);
799 if (is_repository_shallow())
800 die("attempt to fetch/clone from a shallow repository");
801 git_config(upload_pack_config, NULL);
802 upload_pack();
803 return 0;