tog: group state assignments and zap dup repo assignment
[got-portable.git] / lib / send.c
blobaaf32475a0d851e8fbe461a042b06ec0adaa7fba
1 /*
2 * Copyright (c) 2018, 2019 Ori Bernstein <ori@openbsd.org>
3 * Copyright (c) 2021 Stefan Sperling <stsp@openbsd.org>
4 * Copyright (c) 2023 Josh Rickmar <jrick@zettaport.com>
6 * Permission to use, copy, modify, and distribute this software for any
7 * purpose with or without fee is hereby granted, provided that the above
8 * copyright notice and this permission notice appear in all copies.
10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19 #include "got_compat.h"
21 #include <sys/types.h>
22 #include <sys/stat.h>
23 #include <sys/queue.h>
24 #include <sys/uio.h>
25 #include <sys/socket.h>
26 #include <sys/wait.h>
27 #include <sys/resource.h>
28 #include <sys/socket.h>
30 #include <errno.h>
31 #include <err.h>
32 #include <fcntl.h>
33 #include <stdio.h>
34 #include <stdlib.h>
35 #include <string.h>
36 #include <stdint.h>
37 #include <unistd.h>
38 #include <zlib.h>
39 #include <ctype.h>
40 #include <limits.h>
41 #include <time.h>
43 #include "got_error.h"
44 #include "got_reference.h"
45 #include "got_repository.h"
46 #include "got_path.h"
47 #include "got_cancel.h"
48 #include "got_worktree.h"
49 #include "got_object.h"
50 #include "got_opentemp.h"
51 #include "got_send.h"
52 #include "got_repository_admin.h"
53 #include "got_commit_graph.h"
55 #include "got_lib_delta.h"
56 #include "got_lib_hash.h"
57 #include "got_lib_inflate.h"
58 #include "got_lib_object.h"
59 #include "got_lib_object_parse.h"
60 #include "got_lib_object_create.h"
61 #include "got_lib_pack.h"
62 #include "got_lib_privsep.h"
63 #include "got_lib_object_cache.h"
64 #include "got_lib_repository.h"
65 #include "got_lib_ratelimit.h"
66 #include "got_lib_pack_create.h"
67 #include "got_lib_dial.h"
68 #include "got_lib_worktree_cvg.h"
69 #include "got_lib_poll.h"
71 #ifndef nitems
72 #define nitems(_a) (sizeof((_a)) / sizeof((_a)[0]))
73 #endif
75 #ifndef ssizeof
76 #define ssizeof(_x) ((ssize_t)(sizeof(_x)))
77 #endif
79 #ifndef MIN
80 #define MIN(_a,_b) ((_a) < (_b) ? (_a) : (_b))
81 #endif
83 const struct got_error *
84 got_send_connect(pid_t *sendpid, int *sendfd, const char *proto,
85 const char *host, const char *port, const char *server_path, int verbosity)
87 const struct got_error *err = NULL;
89 *sendpid = -1;
90 *sendfd = -1;
92 if (strcmp(proto, "ssh") == 0 || strcmp(proto, "git+ssh") == 0)
93 err = got_dial_ssh(sendpid, sendfd, host, port, server_path,
94 GOT_DIAL_CMD_SEND, verbosity);
95 else if (strcmp(proto, "git") == 0)
96 err = got_dial_git(sendfd, host, port, server_path,
97 GOT_DIAL_CMD_SEND);
98 else if (strcmp(proto, "http") == 0 || strcmp(proto, "git+http") == 0)
99 err = got_error_path(proto, GOT_ERR_NOT_IMPL);
100 else
101 err = got_error_path(proto, GOT_ERR_BAD_PROTO);
102 return err;
105 struct pack_progress_arg {
106 got_send_progress_cb progress_cb;
107 void *progress_arg;
108 int sendfd;
110 int ncolored;
111 int nfound;
112 int ntrees;
113 off_t packfile_size;
114 int ncommits;
115 int nobj_total;
116 int nobj_deltify;
117 int nobj_written;
120 static const struct got_error *
121 pack_progress(void *arg, int ncolored, int nfound, int ntrees,
122 off_t packfile_size, int ncommits, int nobj_total, int nobj_deltify,
123 int nobj_written)
125 const struct got_error *err;
126 struct pack_progress_arg *a = arg;
128 err = a->progress_cb(a->progress_arg, ncolored, nfound, ntrees,
129 packfile_size, ncommits, nobj_total, nobj_deltify,
130 nobj_written, 0, NULL, NULL, 0);
131 if (err)
132 return err;
135 * Detect the server closing our connection while we are
136 * busy creating a pack file.
138 * XXX This should be a temporary workaround. A better fix would
139 * be to avoid use of an on-disk tempfile for pack file data.
140 * Instead we could stream pack file data to got-send-pack while
141 * the pack file is being generated. Write errors in got-send-pack
142 * would then automatically abort the creation of pack file data.
144 err = got_poll_fd(a->sendfd, 0, 0);
145 if (err && err->code != GOT_ERR_TIMEOUT) {
146 if (err->code == GOT_ERR_EOF) {
147 err = got_error_msg(GOT_ERR_EOF,
148 "server unexpectedly closed the connection");
150 return err;
153 a->ncolored= ncolored;
154 a->nfound = nfound;
155 a->ntrees = ntrees;
156 a->packfile_size = packfile_size;
157 a->ncommits = ncommits;
158 a->nobj_total = nobj_total;
159 a->nobj_deltify = nobj_deltify;
160 a->nobj_written = nobj_written;
161 return NULL;
164 static const struct got_error *
165 insert_sendable_ref(struct got_pathlist_head *refs, const char *refname,
166 const char *target_refname, struct got_repository *repo)
168 const struct got_error *err;
169 struct got_reference *ref;
170 struct got_object_id *id = NULL;
171 struct got_pathlist_entry *new = NULL;
172 int obj_type;
174 err = got_ref_open(&ref, repo, refname, 0);
175 if (err)
176 return err;
178 if (got_ref_is_symbolic(ref)) {
179 err = got_error_fmt(GOT_ERR_BAD_REF_TYPE,
180 "cannot send symbolic reference %s", refname);
181 goto done;
184 err = got_ref_resolve(&id, repo, ref);
185 if (err)
186 goto done;
187 err = got_object_get_type(&obj_type, repo, id);
188 if (err)
189 goto done;
190 switch (obj_type) {
191 case GOT_OBJ_TYPE_COMMIT:
192 case GOT_OBJ_TYPE_TAG:
193 break;
194 default:
195 err = got_error_fmt(GOT_ERR_OBJ_TYPE, "cannot send %s",
196 refname);
197 goto done;
200 err = got_pathlist_insert(&new, refs, target_refname, id);
201 if (new == NULL && err == NULL)
202 err = got_error(GOT_ERR_REF_DUP_ENTRY);
204 done:
205 if (ref)
206 got_ref_close(ref);
207 if (err)
208 free(id);
209 return err;
212 static const struct got_error *
213 check_common_ancestry(const char *refname, struct got_object_id *my_id,
214 struct got_object_id *their_id, struct got_repository *repo,
215 got_cancel_cb cancel_cb, void *cancel_arg)
217 const struct got_error *err = NULL;
218 struct got_object_id *yca_id;
219 int obj_type;
221 err = got_object_get_type(&obj_type, repo, their_id);
222 if (err)
223 return err;
224 if (obj_type != GOT_OBJ_TYPE_COMMIT)
225 return got_error_fmt(GOT_ERR_OBJ_TYPE,
226 "bad object type on server for %s", refname);
228 err = got_commit_graph_find_youngest_common_ancestor(&yca_id,
229 my_id, their_id, 0, 1, repo, cancel_cb, cancel_arg);
230 if (err)
231 return err;
232 if (yca_id == NULL)
233 return got_error_fmt(GOT_ERR_SEND_ANCESTRY, "%s", refname);
235 if (got_object_id_cmp(their_id, yca_id) != 0)
236 err = got_error_fmt(GOT_ERR_SEND_ANCESTRY, "%s", refname);
238 free(yca_id);
239 return err;
242 static const struct got_error *
243 realloc_ids(struct got_object_id ***ids, size_t *nalloc, size_t n)
245 struct got_object_id **new;
246 const size_t alloc_chunksz = 256;
248 if (*nalloc >= n)
249 return NULL;
251 new = recallocarray(*ids, *nalloc, *nalloc + alloc_chunksz,
252 sizeof(struct got_object_id));
253 if (new == NULL)
254 return got_error_from_errno("recallocarray");
256 *ids = new;
257 *nalloc += alloc_chunksz;
258 return NULL;
261 static struct got_pathlist_entry *
262 find_ref(struct got_pathlist_head *refs, const char *refname)
264 struct got_pathlist_entry *pe;
266 TAILQ_FOREACH(pe, refs, entry) {
267 if (got_path_cmp(pe->path, refname, strlen(pe->path),
268 strlen(refname)) == 0) {
269 return pe;
273 return NULL;
276 static const struct got_error *
277 get_remote_refname(char **remote_refname, const char *remote_name,
278 const char *refname)
280 if (strncmp(refname, "refs/", 5) == 0)
281 refname += 5;
282 if (strncmp(refname, "heads/", 6) == 0)
283 refname += 6;
285 if (asprintf(remote_refname, "refs/remotes/%s/%s",
286 remote_name, refname) == -1)
287 return got_error_from_errno("asprintf");
289 return NULL;
292 static const struct got_error *
293 update_remote_ref(struct got_pathlist_entry *my_ref, const char *remote_name,
294 struct got_repository *repo)
296 const struct got_error *err, *unlock_err;
297 const char *refname = my_ref->path;
298 struct got_object_id *my_id = my_ref->data;
299 struct got_reference *ref = NULL;
300 char *remote_refname = NULL;
301 int ref_locked = 0;
303 err = get_remote_refname(&remote_refname, remote_name, refname);
304 if (err)
305 goto done;
307 err = got_ref_open(&ref, repo, remote_refname, 1 /* lock */);
308 if (err) {
309 if (err->code != GOT_ERR_NOT_REF)
310 goto done;
311 err = got_ref_alloc(&ref, remote_refname, my_id);
312 if (err)
313 goto done;
314 } else {
315 ref_locked = 1;
316 err = got_ref_change_ref(ref, my_id);
317 if (err)
318 goto done;
321 err = got_ref_write(ref, repo);
322 done:
323 if (ref) {
324 if (ref_locked) {
325 unlock_err = got_ref_unlock(ref);
326 if (unlock_err && err == NULL)
327 err = unlock_err;
329 got_ref_close(ref);
331 free(remote_refname);
332 return err;
335 const struct got_error*
336 got_send_pack(const char *remote_name, struct got_pathlist_head *branch_names,
337 struct got_pathlist_head *tag_names,
338 struct got_pathlist_head *delete_branches,
339 int verbosity, int overwrite_refs, int sendfd,
340 struct got_repository *repo, got_send_progress_cb progress_cb,
341 void *progress_arg, got_cancel_cb cancel_cb, void *cancel_arg)
343 int imsg_sendfds[2] = { -1, -1 };
344 int npackfd = -1, nsendfd = -1;
345 int sendstatus, done = 0;
346 const struct got_error *err;
347 struct imsgbuf sendibuf;
348 pid_t sendpid = -1;
349 struct got_pathlist_head have_refs;
350 struct got_pathlist_head their_refs;
351 struct got_pathlist_entry *pe;
352 struct got_object_id **our_ids = NULL;
353 struct got_object_id **their_ids = NULL;
354 int nours = 0, ntheirs = 0;
355 size_t nalloc_ours = 0, nalloc_theirs = 0;
356 int refs_to_send = 0, refs_to_delete = 0;
357 off_t bytes_sent = 0, bytes_sent_cur = 0;
358 struct pack_progress_arg ppa;
359 struct got_object_id packhash;
360 int packfd = -1;
361 FILE *delta_cache = NULL;
362 char *s = NULL;
364 TAILQ_INIT(&have_refs);
365 TAILQ_INIT(&their_refs);
367 if (got_repo_get_object_format(repo) != GOT_HASH_SHA1)
368 return got_error_fmt(GOT_ERR_NOT_IMPL,
369 "sha256 object IDs unsupported in network protocol");
371 TAILQ_FOREACH(pe, branch_names, entry) {
372 const char *branchname = pe->path;
373 const char *targetname = pe->data;
375 if (targetname == NULL)
376 targetname = branchname;
378 if (strncmp(targetname, "refs/heads/", 11) != 0) {
379 if (asprintf(&s, "refs/heads/%s", targetname) == -1) {
380 err = got_error_from_errno("asprintf");
381 goto done;
383 } else {
384 if ((s = strdup(targetname)) == NULL) {
385 err = got_error_from_errno("strdup");
386 goto done;
389 err = insert_sendable_ref(&have_refs, branchname, s, repo);
390 if (err) {
391 if (err->code != GOT_ERR_REF_DUP_ENTRY)
392 goto done;
393 err = NULL;
394 free(s);
396 s = NULL;
399 TAILQ_FOREACH(pe, delete_branches, entry) {
400 const char *branchname = pe->path;
401 struct got_pathlist_entry *ref;
402 if (strncmp(branchname, "refs/heads/", 11) != 0) {
403 err = got_error_fmt(GOT_ERR_SEND_DELETE_REF, "%s",
404 branchname);
405 goto done;
407 ref = find_ref(&have_refs, branchname);
408 if (ref) {
409 err = got_error_fmt(GOT_ERR_SEND_DELETE_REF,
410 "changes on %s will be sent to server",
411 branchname);
412 goto done;
416 TAILQ_FOREACH(pe, tag_names, entry) {
417 const char *tagname = pe->path;
418 if (strncmp(tagname, "refs/tags/", 10) != 0) {
419 if (asprintf(&s, "refs/tags/%s", tagname) == -1) {
420 err = got_error_from_errno("asprintf");
421 goto done;
423 } else {
424 if ((s = strdup(pe->path)) == NULL) {
425 err = got_error_from_errno("strdup");
426 goto done;
429 err = insert_sendable_ref(&have_refs, s, s, repo);
430 if (err) {
431 if (err->code != GOT_ERR_REF_DUP_ENTRY)
432 goto done;
433 err = NULL;
434 free(s);
436 s = NULL;
439 if (TAILQ_EMPTY(&have_refs) && TAILQ_EMPTY(delete_branches)) {
440 err = got_error(GOT_ERR_SEND_EMPTY);
441 goto done;
444 packfd = got_opentempfd();
445 if (packfd == -1) {
446 err = got_error_from_errno("got_opentempfd");
447 goto done;
450 delta_cache = got_opentemp();
451 if (delta_cache == NULL) {
452 err = got_error_from_errno("got_opentemp");
453 goto done;
456 if (socketpair(AF_UNIX, SOCK_STREAM, PF_UNSPEC, imsg_sendfds) == -1) {
457 err = got_error_from_errno("socketpair");
458 goto done;
461 sendpid = fork();
462 if (sendpid == -1) {
463 err = got_error_from_errno("fork");
464 goto done;
465 } else if (sendpid == 0) {
466 got_privsep_exec_child(imsg_sendfds,
467 GOT_PATH_PROG_SEND_PACK, got_repo_get_path(repo));
470 if (close(imsg_sendfds[1]) == -1) {
471 err = got_error_from_errno("close");
472 goto done;
474 imsg_sendfds[1] = -1;
475 if (imsgbuf_init(&sendibuf, imsg_sendfds[0]) == -1) {
476 err = got_error_from_errno("imsgbuf_init");
477 goto done;
479 imsgbuf_allow_fdpass(&sendibuf);
480 nsendfd = dup(sendfd);
481 if (nsendfd == -1) {
482 err = got_error_from_errno("dup");
483 goto done;
487 * Prepare the array of our object IDs which
488 * will be needed for generating a pack file.
490 TAILQ_FOREACH(pe, &have_refs, entry) {
491 struct got_object_id *id = pe->data;
493 err = realloc_ids(&our_ids, &nalloc_ours, nours + 1);
494 if (err)
495 goto done;
496 our_ids[nours] = id;
497 nours++;
500 err = got_privsep_send_send_req(&sendibuf, nsendfd, &have_refs,
501 delete_branches, verbosity);
502 if (err)
503 goto done;
504 nsendfd = -1;
506 err = got_privsep_recv_send_remote_refs(&their_refs, &sendibuf);
507 if (err)
508 goto done;
510 * Process references reported by the server.
511 * Push appropriate object IDs onto the "their IDs" array.
512 * This array will be used to exclude objects which already
513 * exist on the server from our pack file.
515 TAILQ_FOREACH(pe, &their_refs, entry) {
516 const char *refname = pe->path;
517 struct got_object_id *their_id = pe->data;
518 int have_their_id;
519 struct got_object *obj;
520 struct got_pathlist_entry *my_ref = NULL;
521 int is_tag = 0;
523 /* Don't blindly trust the server to send us valid names. */
524 if (!got_ref_name_is_valid(refname))
525 continue;
527 if (strncmp(refname, "refs/tags/", 10) == 0)
528 is_tag = 1;
530 * Find out whether this is a reference we want to upload.
531 * Otherwise we can still use this reference as a hint to
532 * avoid uploading any objects the server already has.
534 my_ref = find_ref(&have_refs, refname);
535 if (my_ref) {
536 struct got_object_id *my_id = my_ref->data;
537 if (got_object_id_cmp(my_id, their_id) != 0) {
538 if (!overwrite_refs && is_tag) {
539 err = got_error_fmt(
540 GOT_ERR_SEND_TAG_EXISTS,
541 "%s", refname);
542 goto done;
544 refs_to_send++;
548 /* Check if their object exists locally. */
549 err = got_object_open(&obj, repo, their_id);
550 if (err) {
551 if (err->code != GOT_ERR_NO_OBJ)
552 goto done;
553 if (!overwrite_refs && my_ref != NULL) {
554 err = got_error_fmt(GOT_ERR_SEND_ANCESTRY,
555 "%s", refname);
556 goto done;
558 have_their_id = 0;
559 } else {
560 got_object_close(obj);
561 have_their_id = 1;
564 err = realloc_ids(&their_ids, &nalloc_theirs, ntheirs + 1);
565 if (err)
566 goto done;
568 if (have_their_id) {
569 /* Enforce linear ancestry if required. */
570 if (!overwrite_refs && my_ref && !is_tag) {
571 struct got_object_id *my_id = my_ref->data;
572 err = check_common_ancestry(refname, my_id,
573 their_id, repo, cancel_cb, cancel_arg);
574 if (err)
575 goto done;
577 /* Exclude any objects reachable via their ID. */
578 their_ids[ntheirs] = their_id;
579 ntheirs++;
580 } else if (!is_tag) {
581 char *remote_refname;
582 struct got_reference *ref;
584 * Exclude any objects which exist on the server
585 * according to a locally cached remote reference.
587 err = get_remote_refname(&remote_refname,
588 remote_name, refname);
589 if (err)
590 goto done;
591 err = got_ref_open(&ref, repo, remote_refname, 0);
592 free(remote_refname);
593 if (err) {
594 if (err->code != GOT_ERR_NOT_REF)
595 goto done;
596 err = NULL;
597 } else {
598 err = got_ref_resolve(&their_ids[ntheirs],
599 repo, ref);
600 got_ref_close(ref);
601 if (err)
602 goto done;
603 ntheirs++;
608 /* Account for any new references we are going to upload. */
609 TAILQ_FOREACH(pe, &have_refs, entry) {
610 const char *refname = pe->path;
611 if (find_ref(&their_refs, refname) == NULL)
612 refs_to_send++;
615 /* Account for any existing references we are going to delete. */
616 TAILQ_FOREACH(pe, delete_branches, entry) {
617 const char *branchname = pe->path;
618 if (find_ref(&their_refs, branchname))
619 refs_to_delete++;
622 if (refs_to_send == 0 && refs_to_delete == 0) {
623 got_privsep_send_stop(imsg_sendfds[0]);
624 goto done;
627 if (refs_to_send > 0) {
628 struct got_ratelimit rl;
629 got_ratelimit_init(&rl, 0, 500);
630 memset(&ppa, 0, sizeof(ppa));
631 ppa.progress_cb = progress_cb;
632 ppa.progress_arg = progress_arg;
633 ppa.sendfd = sendfd;
634 err = got_pack_create(&packhash, packfd, delta_cache,
635 their_ids, ntheirs, our_ids, nours, repo, 0, 1, 0,
636 pack_progress, &ppa, &rl, cancel_cb, cancel_arg);
637 if (err)
638 goto done;
640 npackfd = dup(packfd);
641 if (npackfd == -1) {
642 err = got_error_from_errno("dup");
643 goto done;
645 err = got_privsep_send_packfd(&sendibuf, npackfd);
646 if (err != NULL)
647 goto done;
648 npackfd = -1;
649 } else {
650 err = got_privsep_send_packfd(&sendibuf, -1);
651 if (err != NULL)
652 goto done;
655 while (!done) {
656 int success = 0;
657 char *refname = NULL;
658 char *errmsg = NULL;
660 if (cancel_cb) {
661 err = (*cancel_cb)(cancel_arg);
662 if (err)
663 goto done;
665 err = got_privsep_recv_send_progress(&done, &bytes_sent,
666 &success, &refname, &errmsg, &sendibuf);
667 if (err)
668 goto done;
669 if (refname && got_ref_name_is_valid(refname) && success &&
670 strncmp(refname, "refs/tags/", 10) != 0) {
671 struct got_pathlist_entry *my_ref;
673 * The server has accepted our changes.
674 * Update our reference in refs/remotes/ accordingly.
676 my_ref = find_ref(&have_refs, refname);
677 if (my_ref) {
678 err = update_remote_ref(my_ref, remote_name,
679 repo);
680 if (err)
681 goto done;
684 if (refname != NULL ||
685 bytes_sent_cur != bytes_sent) {
686 err = progress_cb(progress_arg, ppa.ncolored,
687 ppa.nfound, ppa.ntrees, ppa.packfile_size,
688 ppa.ncommits, ppa.nobj_total, ppa.nobj_deltify,
689 ppa.nobj_written, bytes_sent,
690 refname, errmsg, success);
691 if (err) {
692 free(refname);
693 free(errmsg);
694 goto done;
696 bytes_sent_cur = bytes_sent;
698 free(refname);
699 free(errmsg);
701 done:
702 if (sendpid != -1) {
703 if (err)
704 got_privsep_send_stop(imsg_sendfds[0]);
705 if (waitpid(sendpid, &sendstatus, 0) == -1 && err == NULL)
706 err = got_error_from_errno("waitpid");
708 if (imsg_sendfds[0] != -1 && close(imsg_sendfds[0]) == -1 && err == NULL)
709 err = got_error_from_errno("close");
710 if (imsg_sendfds[1] != -1 && close(imsg_sendfds[1]) == -1 && err == NULL)
711 err = got_error_from_errno("close");
712 if (packfd != -1 && close(packfd) == -1 && err == NULL)
713 err = got_error_from_errno("close");
714 if (delta_cache && fclose(delta_cache) == EOF && err == NULL)
715 err = got_error_from_errno("fclose");
716 if (nsendfd != -1 && close(nsendfd) == -1 && err == NULL)
717 err = got_error_from_errno("close");
718 if (npackfd != -1 && close(npackfd) == -1 && err == NULL)
719 err = got_error_from_errno("close");
721 got_pathlist_free(&have_refs, GOT_PATHLIST_FREE_ALL);
722 got_pathlist_free(&their_refs, GOT_PATHLIST_FREE_ALL);
724 * Object ids are owned by have_refs/their_refs and are already freed;
725 * Only the arrays must be freed.
727 free(our_ids);
728 free(their_ids);
729 free(s);
730 return err;