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