make shallow repository deepening more network efficient
[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, use_include_tag;
31 static int no_progress;
32 static struct object_array have_obj;
33 static struct object_array want_obj;
34 static struct object_array extra_edge_obj;
35 static unsigned int timeout;
36 /* 0 for no sideband,
37 * otherwise maximum packet size (up to 65520 bytes).
38 */
39 static int use_sideband;
40 static int debug_fd;
41
42 static void reset_timeout(void)
43 {
44 alarm(timeout);
45 }
46
47 static int strip(char *line, int len)
48 {
49 if (len && line[len-1] == '\n')
50 line[--len] = 0;
51 return len;
52 }
53
54 static ssize_t send_client_data(int fd, const char *data, ssize_t sz)
55 {
56 if (use_sideband)
57 return send_sideband(1, fd, data, sz, use_sideband);
58 if (fd == 3)
59 /* emergency quit */
60 fd = 2;
61 if (fd == 2) {
62 /* XXX: are we happy to lose stuff here? */
63 xwrite(fd, data, sz);
64 return sz;
65 }
66 return safe_write(fd, data, sz);
67 }
68
69 static FILE *pack_pipe = NULL;
70 static void show_commit(struct commit *commit, void *data)
71 {
72 if (commit->object.flags & BOUNDARY)
73 fputc('-', pack_pipe);
74 if (fputs(sha1_to_hex(commit->object.sha1), pack_pipe) < 0)
75 die("broken output pipe");
76 fputc('\n', pack_pipe);
77 fflush(pack_pipe);
78 free(commit->buffer);
79 commit->buffer = NULL;
80 }
81
82 static void show_object(struct object *obj, const struct name_path *path, const char *component)
83 {
84 /* An object with name "foo\n0000000..." can be used to
85 * confuse downstream git-pack-objects very badly.
86 */
87 const char *name = path_name(path, component);
88 const char *ep = strchr(name, '\n');
89 if (ep) {
90 fprintf(pack_pipe, "%s %.*s\n", sha1_to_hex(obj->sha1),
91 (int) (ep - name),
92 name);
93 }
94 else
95 fprintf(pack_pipe, "%s %s\n",
96 sha1_to_hex(obj->sha1), name);
97 free((char *)name);
98 }
99
100 static void show_edge(struct commit *commit)
101 {
102 fprintf(pack_pipe, "-%s\n", sha1_to_hex(commit->object.sha1));
103 }
104
105 static int do_rev_list(int fd, void *create_full_pack)
106 {
107 int i;
108 struct rev_info revs;
109
110 pack_pipe = fdopen(fd, "w");
111 if (create_full_pack)
112 use_thin_pack = 0; /* no point doing it */
113 init_revisions(&revs, NULL);
114 revs.tag_objects = 1;
115 revs.tree_objects = 1;
116 revs.blob_objects = 1;
117 if (use_thin_pack)
118 revs.edge_hint = 1;
119
120 if (create_full_pack) {
121 const char *args[] = {"rev-list", "--all", NULL};
122 setup_revisions(2, args, &revs, NULL);
123 } else {
124 for (i = 0; i < want_obj.nr; i++) {
125 struct object *o = want_obj.objects[i].item;
126 /* why??? */
127 o->flags &= ~UNINTERESTING;
128 add_pending_object(&revs, o, NULL);
129 }
130 for (i = 0; i < have_obj.nr; i++) {
131 struct object *o = have_obj.objects[i].item;
132 o->flags |= UNINTERESTING;
133 add_pending_object(&revs, o, NULL);
134 }
135 setup_revisions(0, NULL, &revs, NULL);
136 }
137 if (prepare_revision_walk(&revs))
138 die("revision walk setup failed");
139 mark_edges_uninteresting(revs.commits, &revs, show_edge);
140 if (use_thin_pack)
141 for (i = 0; i < extra_edge_obj.nr; i++)
142 fprintf(pack_pipe, "-%s\n", sha1_to_hex(
143 extra_edge_obj.objects[i].item->sha1));
144 traverse_commit_list(&revs, show_commit, show_object, NULL);
145 fflush(pack_pipe);
146 fclose(pack_pipe);
147 return 0;
148 }
149
150 static void create_pack_file(void)
151 {
152 struct async rev_list;
153 struct child_process pack_objects;
154 int create_full_pack = (nr_our_refs == want_obj.nr && !have_obj.nr);
155 char data[8193], progress[128];
156 char abort_msg[] = "aborting due to possible repository "
157 "corruption on the remote side.";
158 int buffered = -1;
159 ssize_t sz;
160 const char *argv[10];
161 int arg = 0;
162
163 rev_list.proc = do_rev_list;
164 /* .data is just a boolean: any non-NULL value will do */
165 rev_list.data = create_full_pack ? &rev_list : NULL;
166 if (start_async(&rev_list))
167 die("git upload-pack: unable to fork git-rev-list");
168
169 argv[arg++] = "pack-objects";
170 argv[arg++] = "--stdout";
171 if (!no_progress)
172 argv[arg++] = "--progress";
173 if (use_ofs_delta)
174 argv[arg++] = "--delta-base-offset";
175 if (use_include_tag)
176 argv[arg++] = "--include-tag";
177 argv[arg++] = NULL;
178
179 memset(&pack_objects, 0, sizeof(pack_objects));
180 pack_objects.in = rev_list.out; /* start_command closes it */
181 pack_objects.out = -1;
182 pack_objects.err = -1;
183 pack_objects.git_cmd = 1;
184 pack_objects.argv = argv;
185
186 if (start_command(&pack_objects))
187 die("git upload-pack: unable to fork git-pack-objects");
188
189 /* We read from pack_objects.err to capture stderr output for
190 * progress bar, and pack_objects.out to capture the pack data.
191 */
192
193 while (1) {
194 struct pollfd pfd[2];
195 int pe, pu, pollsize;
196
197 reset_timeout();
198
199 pollsize = 0;
200 pe = pu = -1;
201
202 if (0 <= pack_objects.out) {
203 pfd[pollsize].fd = pack_objects.out;
204 pfd[pollsize].events = POLLIN;
205 pu = pollsize;
206 pollsize++;
207 }
208 if (0 <= pack_objects.err) {
209 pfd[pollsize].fd = pack_objects.err;
210 pfd[pollsize].events = POLLIN;
211 pe = pollsize;
212 pollsize++;
213 }
214
215 if (!pollsize)
216 break;
217
218 if (poll(pfd, pollsize, -1) < 0) {
219 if (errno != EINTR) {
220 error("poll failed, resuming: %s",
221 strerror(errno));
222 sleep(1);
223 }
224 continue;
225 }
226 if (0 <= pu && (pfd[pu].revents & (POLLIN|POLLHUP))) {
227 /* Data ready; we keep the last byte to ourselves
228 * in case we detect broken rev-list, so that we
229 * can leave the stream corrupted. This is
230 * unfortunate -- unpack-objects would happily
231 * accept a valid packdata with trailing garbage,
232 * so appending garbage after we pass all the
233 * pack data is not good enough to signal
234 * breakage to downstream.
235 */
236 char *cp = data;
237 ssize_t outsz = 0;
238 if (0 <= buffered) {
239 *cp++ = buffered;
240 outsz++;
241 }
242 sz = xread(pack_objects.out, cp,
243 sizeof(data) - outsz);
244 if (0 < sz)
245 ;
246 else if (sz == 0) {
247 close(pack_objects.out);
248 pack_objects.out = -1;
249 }
250 else
251 goto fail;
252 sz += outsz;
253 if (1 < sz) {
254 buffered = data[sz-1] & 0xFF;
255 sz--;
256 }
257 else
258 buffered = -1;
259 sz = send_client_data(1, data, sz);
260 if (sz < 0)
261 goto fail;
262 }
263 if (0 <= pe && (pfd[pe].revents & (POLLIN|POLLHUP))) {
264 /* Status ready; we ship that in the side-band
265 * or dump to the standard error.
266 */
267 sz = xread(pack_objects.err, progress,
268 sizeof(progress));
269 if (0 < sz)
270 send_client_data(2, progress, sz);
271 else if (sz == 0) {
272 close(pack_objects.err);
273 pack_objects.err = -1;
274 }
275 else
276 goto fail;
277 }
278 }
279
280 if (finish_command(&pack_objects)) {
281 error("git upload-pack: git-pack-objects died with error.");
282 goto fail;
283 }
284 if (finish_async(&rev_list))
285 goto fail; /* error was already reported */
286
287 /* flush the data */
288 if (0 <= buffered) {
289 data[0] = buffered;
290 sz = send_client_data(1, data, 1);
291 if (sz < 0)
292 goto fail;
293 fprintf(stderr, "flushed.\n");
294 }
295 if (use_sideband)
296 packet_flush(1);
297 return;
298
299 fail:
300 send_client_data(3, abort_msg, sizeof(abort_msg));
301 die("git upload-pack: %s", abort_msg);
302 }
303
304 static int got_sha1(char *hex, unsigned char *sha1)
305 {
306 struct object *o;
307 int we_knew_they_have = 0;
308
309 if (get_sha1_hex(hex, sha1))
310 die("git upload-pack: expected SHA1 object, got '%s'", hex);
311 if (!has_sha1_file(sha1))
312 return -1;
313
314 o = lookup_object(sha1);
315 if (!(o && o->parsed))
316 o = parse_object(sha1);
317 if (!o)
318 die("oops (%s)", sha1_to_hex(sha1));
319 if (o->type == OBJ_COMMIT) {
320 struct commit_list *parents;
321 struct commit *commit = (struct commit *)o;
322 if (o->flags & THEY_HAVE)
323 we_knew_they_have = 1;
324 else
325 o->flags |= THEY_HAVE;
326 if (!oldest_have || (commit->date < oldest_have))
327 oldest_have = commit->date;
328 for (parents = commit->parents;
329 parents;
330 parents = parents->next)
331 parents->item->object.flags |= THEY_HAVE;
332 }
333 if (!we_knew_they_have) {
334 add_object_array(o, NULL, &have_obj);
335 return 1;
336 }
337 return 0;
338 }
339
340 static int reachable(struct commit *want)
341 {
342 struct commit_list *work = NULL;
343
344 insert_by_date(want, &work);
345 while (work) {
346 struct commit_list *list = work->next;
347 struct commit *commit = work->item;
348 free(work);
349 work = list;
350
351 if (commit->object.flags & THEY_HAVE) {
352 want->object.flags |= COMMON_KNOWN;
353 break;
354 }
355 if (!commit->object.parsed)
356 parse_object(commit->object.sha1);
357 if (commit->object.flags & REACHABLE)
358 continue;
359 commit->object.flags |= REACHABLE;
360 if (commit->date < oldest_have)
361 continue;
362 for (list = commit->parents; list; list = list->next) {
363 struct commit *parent = list->item;
364 if (!(parent->object.flags & REACHABLE))
365 insert_by_date(parent, &work);
366 }
367 }
368 want->object.flags |= REACHABLE;
369 clear_commit_marks(want, REACHABLE);
370 free_commit_list(work);
371 return (want->object.flags & COMMON_KNOWN);
372 }
373
374 static int ok_to_give_up(void)
375 {
376 int i;
377
378 if (!have_obj.nr)
379 return 0;
380
381 for (i = 0; i < want_obj.nr; i++) {
382 struct object *want = want_obj.objects[i].item;
383
384 if (want->flags & COMMON_KNOWN)
385 continue;
386 want = deref_tag(want, "a want line", 0);
387 if (!want || want->type != OBJ_COMMIT) {
388 /* no way to tell if this is reachable by
389 * looking at the ancestry chain alone, so
390 * leave a note to ourselves not to worry about
391 * this object anymore.
392 */
393 want_obj.objects[i].item->flags |= COMMON_KNOWN;
394 continue;
395 }
396 if (!reachable((struct commit *)want))
397 return 0;
398 }
399 return 1;
400 }
401
402 static int get_common_commits(void)
403 {
404 static char line[1000];
405 unsigned char sha1[20];
406 char hex[41], last_hex[41];
407
408 save_commit_buffer = 0;
409
410 for(;;) {
411 int len = packet_read_line(0, line, sizeof(line));
412 reset_timeout();
413
414 if (!len) {
415 if (have_obj.nr == 0 || multi_ack)
416 packet_write(1, "NAK\n");
417 continue;
418 }
419 strip(line, len);
420 if (!prefixcmp(line, "have ")) {
421 switch (got_sha1(line+5, sha1)) {
422 case -1: /* they have what we do not */
423 if (multi_ack && ok_to_give_up())
424 packet_write(1, "ACK %s continue\n",
425 sha1_to_hex(sha1));
426 break;
427 default:
428 memcpy(hex, sha1_to_hex(sha1), 41);
429 if (multi_ack) {
430 const char *msg = "ACK %s continue\n";
431 packet_write(1, msg, hex);
432 memcpy(last_hex, hex, 41);
433 }
434 else if (have_obj.nr == 1)
435 packet_write(1, "ACK %s\n", hex);
436 break;
437 }
438 continue;
439 }
440 if (!strcmp(line, "done")) {
441 if (have_obj.nr > 0) {
442 if (multi_ack)
443 packet_write(1, "ACK %s\n", last_hex);
444 return 0;
445 }
446 packet_write(1, "NAK\n");
447 return -1;
448 }
449 die("git upload-pack: expected SHA1 list, got '%s'", line);
450 }
451 }
452
453 static void receive_needs(void)
454 {
455 struct object_array shallows = {0, 0, NULL};
456 static char line[1000];
457 int len, depth = 0;
458
459 if (debug_fd)
460 write_in_full(debug_fd, "#S\n", 3);
461 for (;;) {
462 struct object *o;
463 unsigned char sha1_buf[20];
464 len = packet_read_line(0, line, sizeof(line));
465 reset_timeout();
466 if (!len)
467 break;
468 if (debug_fd)
469 write_in_full(debug_fd, line, len);
470
471 if (!prefixcmp(line, "shallow ")) {
472 unsigned char sha1[20];
473 struct object *object;
474 if (get_sha1(line + 8, sha1))
475 die("invalid shallow line: %s", line);
476 object = parse_object(sha1);
477 if (!object)
478 die("did not find object for %s", line);
479 object->flags |= CLIENT_SHALLOW;
480 add_object_array(object, NULL, &shallows);
481 continue;
482 }
483 if (!prefixcmp(line, "deepen ")) {
484 char *end;
485 depth = strtol(line + 7, &end, 0);
486 if (end == line + 7 || depth <= 0)
487 die("Invalid deepen: %s", line);
488 continue;
489 }
490 if (prefixcmp(line, "want ") ||
491 get_sha1_hex(line+5, sha1_buf))
492 die("git upload-pack: protocol error, "
493 "expected to get sha, not '%s'", line);
494 if (strstr(line+45, "multi_ack"))
495 multi_ack = 1;
496 if (strstr(line+45, "thin-pack"))
497 use_thin_pack = 1;
498 if (strstr(line+45, "ofs-delta"))
499 use_ofs_delta = 1;
500 if (strstr(line+45, "side-band-64k"))
501 use_sideband = LARGE_PACKET_MAX;
502 else if (strstr(line+45, "side-band"))
503 use_sideband = DEFAULT_PACKET_MAX;
504 if (strstr(line+45, "no-progress"))
505 no_progress = 1;
506 if (strstr(line+45, "include-tag"))
507 use_include_tag = 1;
508
509 /* We have sent all our refs already, and the other end
510 * should have chosen out of them; otherwise they are
511 * asking for nonsense.
512 *
513 * Hmph. We may later want to allow "want" line that
514 * asks for something like "master~10" (symbolic)...
515 * would it make sense? I don't know.
516 */
517 o = lookup_object(sha1_buf);
518 if (!o || !(o->flags & OUR_REF))
519 die("git upload-pack: not our ref %s", line+5);
520 if (!(o->flags & WANTED)) {
521 o->flags |= WANTED;
522 add_object_array(o, NULL, &want_obj);
523 }
524 }
525 if (debug_fd)
526 write_in_full(debug_fd, "#E\n", 3);
527 if (depth == 0 && shallows.nr == 0)
528 return;
529 if (depth > 0) {
530 struct commit_list *result, *backup;
531 int i;
532 backup = result = get_shallow_commits(&want_obj, depth,
533 SHALLOW, NOT_SHALLOW);
534 while (result) {
535 struct object *object = &result->item->object;
536 if (!(object->flags & (CLIENT_SHALLOW|NOT_SHALLOW))) {
537 packet_write(1, "shallow %s",
538 sha1_to_hex(object->sha1));
539 register_shallow(object->sha1);
540 }
541 result = result->next;
542 }
543 free_commit_list(backup);
544 for (i = 0; i < shallows.nr; i++) {
545 struct object *object = shallows.objects[i].item;
546 if (object->flags & NOT_SHALLOW) {
547 struct commit_list *parents;
548 packet_write(1, "unshallow %s",
549 sha1_to_hex(object->sha1));
550 object->flags &= ~CLIENT_SHALLOW;
551 /* make sure the real parents are parsed */
552 unregister_shallow(object->sha1);
553 object->parsed = 0;
554 if (parse_commit((struct commit *)object))
555 die("invalid commit");
556 parents = ((struct commit *)object)->parents;
557 while (parents) {
558 add_object_array(&parents->item->object,
559 NULL, &want_obj);
560 parents = parents->next;
561 }
562 add_object_array(object, NULL, &extra_edge_obj);
563 }
564 /* make sure commit traversal conforms to client */
565 register_shallow(object->sha1);
566 }
567 packet_flush(1);
568 } else
569 if (shallows.nr > 0) {
570 int i;
571 for (i = 0; i < shallows.nr; i++)
572 register_shallow(shallows.objects[i].item->sha1);
573 }
574 free(shallows.objects);
575 }
576
577 static int send_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
578 {
579 static const char *capabilities = "multi_ack thin-pack side-band"
580 " side-band-64k ofs-delta shallow no-progress"
581 " include-tag";
582 struct object *o = parse_object(sha1);
583
584 if (!o)
585 die("git upload-pack: cannot find object %s:", sha1_to_hex(sha1));
586
587 if (capabilities)
588 packet_write(1, "%s %s%c%s\n", sha1_to_hex(sha1), refname,
589 0, capabilities);
590 else
591 packet_write(1, "%s %s\n", sha1_to_hex(sha1), refname);
592 capabilities = NULL;
593 if (!(o->flags & OUR_REF)) {
594 o->flags |= OUR_REF;
595 nr_our_refs++;
596 }
597 if (o->type == OBJ_TAG) {
598 o = deref_tag(o, refname, 0);
599 if (o)
600 packet_write(1, "%s %s^{}\n", sha1_to_hex(o->sha1), refname);
601 }
602 return 0;
603 }
604
605 static void upload_pack(void)
606 {
607 reset_timeout();
608 head_ref(send_ref, NULL);
609 for_each_ref(send_ref, NULL);
610 packet_flush(1);
611 receive_needs();
612 if (want_obj.nr) {
613 get_common_commits();
614 create_pack_file();
615 }
616 }
617
618 int main(int argc, char **argv)
619 {
620 char *dir;
621 int i;
622 int strict = 0;
623
624 git_extract_argv0_path(argv[0]);
625
626 for (i = 1; i < argc; i++) {
627 char *arg = argv[i];
628
629 if (arg[0] != '-')
630 break;
631 if (!strcmp(arg, "--strict")) {
632 strict = 1;
633 continue;
634 }
635 if (!prefixcmp(arg, "--timeout=")) {
636 timeout = atoi(arg+10);
637 continue;
638 }
639 if (!strcmp(arg, "--")) {
640 i++;
641 break;
642 }
643 }
644
645 if (i != argc-1)
646 usage(upload_pack_usage);
647
648 setup_path();
649
650 dir = argv[i];
651
652 if (!enter_repo(dir, strict))
653 die("'%s' does not appear to be a git repository", dir);
654 if (is_repository_shallow())
655 die("attempt to fetch/clone from a shallow repository");
656 if (getenv("GIT_DEBUG_SEND_PACK"))
657 debug_fd = atoi(getenv("GIT_DEBUG_SEND_PACK"));
658 upload_pack();
659 return 0;
660 }