Blob


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>
5 *
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.
9 *
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.
17 */
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_inflate.h"
57 #include "got_lib_object.h"
58 #include "got_lib_object_parse.h"
59 #include "got_lib_object_create.h"
60 #include "got_lib_pack.h"
61 #include "got_lib_hash.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)
86 {
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;
118 };
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;
134 /*
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.
143 */
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 int obj_type;
173 err = got_ref_open(&ref, repo, refname, 0);
174 if (err)
175 return err;
177 if (got_ref_is_symbolic(ref)) {
178 err = got_error_fmt(GOT_ERR_BAD_REF_TYPE,
179 "cannot send symbolic reference %s", refname);
180 goto done;
183 err = got_ref_resolve(&id, repo, ref);
184 if (err)
185 goto done;
186 err = got_object_get_type(&obj_type, repo, id);
187 if (err)
188 goto done;
189 switch (obj_type) {
190 case GOT_OBJ_TYPE_COMMIT:
191 case GOT_OBJ_TYPE_TAG:
192 break;
193 default:
194 err = got_error_fmt(GOT_ERR_OBJ_TYPE," cannot send %s",
195 refname);
196 goto done;
199 err = got_pathlist_insert(NULL, refs, target_refname, id);
200 done:
201 if (ref)
202 got_ref_close(ref);
203 if (err)
204 free(id);
205 return err;
208 static const struct got_error *
209 check_common_ancestry(const char *refname, struct got_object_id *my_id,
210 struct got_object_id *their_id, struct got_repository *repo,
211 got_cancel_cb cancel_cb, void *cancel_arg)
213 const struct got_error *err = NULL;
214 struct got_object_id *yca_id;
215 int obj_type;
217 err = got_object_get_type(&obj_type, repo, their_id);
218 if (err)
219 return err;
220 if (obj_type != GOT_OBJ_TYPE_COMMIT)
221 return got_error_fmt(GOT_ERR_OBJ_TYPE,
222 "bad object type on server for %s", refname);
224 err = got_commit_graph_find_youngest_common_ancestor(&yca_id,
225 my_id, their_id, 0, 0, repo, cancel_cb, cancel_arg);
226 if (err)
227 return err;
228 if (yca_id == NULL)
229 return got_error_fmt(GOT_ERR_SEND_ANCESTRY, "%s", refname);
231 if (got_object_id_cmp(their_id, yca_id) != 0)
232 err = got_error_fmt(GOT_ERR_SEND_ANCESTRY, "%s", refname);
234 free(yca_id);
235 return err;
238 static const struct got_error *
239 realloc_ids(struct got_object_id ***ids, size_t *nalloc, size_t n)
241 struct got_object_id **new;
242 const size_t alloc_chunksz = 256;
244 if (*nalloc >= n)
245 return NULL;
247 new = recallocarray(*ids, *nalloc, *nalloc + alloc_chunksz,
248 sizeof(struct got_object_id));
249 if (new == NULL)
250 return got_error_from_errno("recallocarray");
252 *ids = new;
253 *nalloc += alloc_chunksz;
254 return NULL;
257 static struct got_pathlist_entry *
258 find_ref(struct got_pathlist_head *refs, const char *refname)
260 struct got_pathlist_entry *pe;
262 TAILQ_FOREACH(pe, refs, entry) {
263 if (got_path_cmp(pe->path, refname, strlen(pe->path),
264 strlen(refname)) == 0) {
265 return pe;
269 return NULL;
272 static const struct got_error *
273 get_remote_refname(char **remote_refname, const char *remote_name,
274 const char *refname)
276 if (strncmp(refname, "refs/", 5) == 0)
277 refname += 5;
278 if (strncmp(refname, "heads/", 6) == 0)
279 refname += 6;
281 if (asprintf(remote_refname, "refs/remotes/%s/%s",
282 remote_name, refname) == -1)
283 return got_error_from_errno("asprintf");
285 return NULL;
288 static const struct got_error *
289 update_remote_ref(struct got_pathlist_entry *my_ref, const char *remote_name,
290 struct got_repository *repo)
292 const struct got_error *err, *unlock_err;
293 const char *refname = my_ref->path;
294 struct got_object_id *my_id = my_ref->data;
295 struct got_reference *ref = NULL;
296 char *remote_refname = NULL;
297 int ref_locked = 0;
299 err = get_remote_refname(&remote_refname, remote_name, refname);
300 if (err)
301 goto done;
303 err = got_ref_open(&ref, repo, remote_refname, 1 /* lock */);
304 if (err) {
305 if (err->code != GOT_ERR_NOT_REF)
306 goto done;
307 err = got_ref_alloc(&ref, remote_refname, my_id);
308 if (err)
309 goto done;
310 } else {
311 ref_locked = 1;
312 err = got_ref_change_ref(ref, my_id);
313 if (err)
314 goto done;
317 err = got_ref_write(ref, repo);
318 done:
319 if (ref) {
320 if (ref_locked) {
321 unlock_err = got_ref_unlock(ref);
322 if (unlock_err && err == NULL)
323 err = unlock_err;
325 got_ref_close(ref);
327 free(remote_refname);
328 return err;
331 const struct got_error*
332 got_send_pack(const char *remote_name, struct got_pathlist_head *branch_names,
333 struct got_pathlist_head *tag_names,
334 struct got_pathlist_head *delete_branches,
335 int verbosity, int overwrite_refs, int sendfd,
336 struct got_repository *repo, got_send_progress_cb progress_cb,
337 void *progress_arg, got_cancel_cb cancel_cb, void *cancel_arg)
339 int imsg_sendfds[2];
340 int npackfd = -1, nsendfd = -1;
341 int sendstatus, done = 0;
342 const struct got_error *err;
343 struct imsgbuf sendibuf;
344 pid_t sendpid = -1;
345 struct got_pathlist_head have_refs;
346 struct got_pathlist_head their_refs;
347 struct got_pathlist_entry *pe;
348 struct got_object_id **our_ids = NULL;
349 struct got_object_id **their_ids = NULL;
350 int nours = 0, ntheirs = 0;
351 size_t nalloc_ours = 0, nalloc_theirs = 0;
352 int refs_to_send = 0, refs_to_delete = 0;
353 off_t bytes_sent = 0, bytes_sent_cur = 0;
354 struct pack_progress_arg ppa;
355 uint8_t packsha1[SHA1_DIGEST_LENGTH];
356 int packfd = -1;
357 FILE *delta_cache = NULL;
358 char *s = NULL;
360 TAILQ_INIT(&have_refs);
361 TAILQ_INIT(&their_refs);
363 TAILQ_FOREACH(pe, branch_names, entry) {
364 const char *branchname = pe->path;
365 const char *targetname = pe->data;
367 if (targetname == NULL)
368 targetname = branchname;
370 if (strncmp(targetname, "refs/heads/", 11) != 0) {
371 if (asprintf(&s, "refs/heads/%s", targetname) == -1) {
372 err = got_error_from_errno("asprintf");
373 goto done;
375 } else {
376 if ((s = strdup(targetname)) == NULL) {
377 err = got_error_from_errno("strdup");
378 goto done;
381 err = insert_sendable_ref(&have_refs, branchname, s, repo);
382 if (err)
383 goto done;
384 s = NULL;
387 TAILQ_FOREACH(pe, delete_branches, entry) {
388 const char *branchname = pe->path;
389 struct got_pathlist_entry *ref;
390 if (strncmp(branchname, "refs/heads/", 11) != 0) {
391 err = got_error_fmt(GOT_ERR_SEND_DELETE_REF, "%s",
392 branchname);
393 goto done;
395 ref = find_ref(&have_refs, branchname);
396 if (ref) {
397 err = got_error_fmt(GOT_ERR_SEND_DELETE_REF,
398 "changes on %s will be sent to server",
399 branchname);
400 goto done;
404 TAILQ_FOREACH(pe, tag_names, entry) {
405 const char *tagname = pe->path;
406 if (strncmp(tagname, "refs/tags/", 10) != 0) {
407 if (asprintf(&s, "refs/tags/%s", tagname) == -1) {
408 err = got_error_from_errno("asprintf");
409 goto done;
411 } else {
412 if ((s = strdup(pe->path)) == NULL) {
413 err = got_error_from_errno("strdup");
414 goto done;
417 err = insert_sendable_ref(&have_refs, s, s, repo);
418 if (err)
419 goto done;
420 s = NULL;
423 if (TAILQ_EMPTY(&have_refs) && TAILQ_EMPTY(delete_branches)) {
424 err = got_error(GOT_ERR_SEND_EMPTY);
425 goto done;
428 packfd = got_opentempfd();
429 if (packfd == -1) {
430 err = got_error_from_errno("got_opentempfd");
431 goto done;
434 delta_cache = got_opentemp();
435 if (delta_cache == NULL) {
436 err = got_error_from_errno("got_opentemp");
437 goto done;
440 if (socketpair(AF_UNIX, SOCK_STREAM, PF_UNSPEC, imsg_sendfds) == -1) {
441 err = got_error_from_errno("socketpair");
442 goto done;
445 sendpid = fork();
446 if (sendpid == -1) {
447 err = got_error_from_errno("fork");
448 goto done;
449 } else if (sendpid == 0){
450 got_privsep_exec_child(imsg_sendfds,
451 GOT_PATH_PROG_SEND_PACK, got_repo_get_path(repo));
454 if (close(imsg_sendfds[1]) == -1) {
455 err = got_error_from_errno("close");
456 goto done;
458 imsg_init(&sendibuf, imsg_sendfds[0]);
459 nsendfd = dup(sendfd);
460 if (nsendfd == -1) {
461 err = got_error_from_errno("dup");
462 goto done;
465 /*
466 * Prepare the array of our object IDs which
467 * will be needed for generating a pack file.
468 */
469 TAILQ_FOREACH(pe, &have_refs, entry) {
470 struct got_object_id *id = pe->data;
472 err = realloc_ids(&our_ids, &nalloc_ours, nours + 1);
473 if (err)
474 goto done;
475 our_ids[nours] = id;
476 nours++;
479 err = got_privsep_send_send_req(&sendibuf, nsendfd, &have_refs,
480 delete_branches, verbosity);
481 if (err)
482 goto done;
483 nsendfd = -1;
485 err = got_privsep_recv_send_remote_refs(&their_refs, &sendibuf);
486 if (err)
487 goto done;
488 /*
489 * Process references reported by the server.
490 * Push appropriate object IDs onto the "their IDs" array.
491 * This array will be used to exclude objects which already
492 * exist on the server from our pack file.
493 */
494 TAILQ_FOREACH(pe, &their_refs, entry) {
495 const char *refname = pe->path;
496 struct got_object_id *their_id = pe->data;
497 int have_their_id;
498 struct got_object *obj;
499 struct got_pathlist_entry *my_ref = NULL;
500 int is_tag = 0;
502 /* Don't blindly trust the server to send us valid names. */
503 if (!got_ref_name_is_valid(refname))
504 continue;
506 if (strncmp(refname, "refs/tags/", 10) == 0)
507 is_tag = 1;
508 /*
509 * Find out whether this is a reference we want to upload.
510 * Otherwise we can still use this reference as a hint to
511 * avoid uploading any objects the server already has.
512 */
513 my_ref = find_ref(&have_refs, refname);
514 if (my_ref) {
515 struct got_object_id *my_id = my_ref->data;
516 if (got_object_id_cmp(my_id, their_id) != 0) {
517 if (!overwrite_refs && is_tag) {
518 err = got_error_fmt(
519 GOT_ERR_SEND_TAG_EXISTS,
520 "%s", refname);
521 goto done;
523 refs_to_send++;
527 /* Check if their object exists locally. */
528 err = got_object_open(&obj, repo, their_id);
529 if (err) {
530 if (err->code != GOT_ERR_NO_OBJ)
531 goto done;
532 if (!overwrite_refs && my_ref != NULL) {
533 err = got_error_fmt(GOT_ERR_SEND_ANCESTRY,
534 "%s", refname);
535 goto done;
537 have_their_id = 0;
538 } else {
539 got_object_close(obj);
540 have_their_id = 1;
543 err = realloc_ids(&their_ids, &nalloc_theirs, ntheirs + 1);
544 if (err)
545 goto done;
547 if (have_their_id) {
548 /* Enforce linear ancestry if required. */
549 if (!overwrite_refs && my_ref && !is_tag) {
550 struct got_object_id *my_id = my_ref->data;
551 err = check_common_ancestry(refname, my_id,
552 their_id, repo, cancel_cb, cancel_arg);
553 if (err)
554 goto done;
556 /* Exclude any objects reachable via their ID. */
557 their_ids[ntheirs] = their_id;
558 ntheirs++;
559 } else if (!is_tag) {
560 char *remote_refname;
561 struct got_reference *ref;
562 /*
563 * Exclude any objects which exist on the server
564 * according to a locally cached remote reference.
565 */
566 err = get_remote_refname(&remote_refname,
567 remote_name, refname);
568 if (err)
569 goto done;
570 err = got_ref_open(&ref, repo, remote_refname, 0);
571 free(remote_refname);
572 if (err) {
573 if (err->code != GOT_ERR_NOT_REF)
574 goto done;
575 } else {
576 err = got_ref_resolve(&their_ids[ntheirs],
577 repo, ref);
578 got_ref_close(ref);
579 if (err)
580 goto done;
581 ntheirs++;
586 /* Account for any new references we are going to upload. */
587 TAILQ_FOREACH(pe, &have_refs, entry) {
588 const char *refname = pe->path;
589 if (find_ref(&their_refs, refname) == NULL)
590 refs_to_send++;
593 /* Account for any existing references we are going to delete. */
594 TAILQ_FOREACH(pe, delete_branches, entry) {
595 const char *branchname = pe->path;
596 if (find_ref(&their_refs, branchname))
597 refs_to_delete++;
600 if (refs_to_send == 0 && refs_to_delete == 0) {
601 got_privsep_send_stop(imsg_sendfds[0]);
602 goto done;
605 if (refs_to_send > 0) {
606 struct got_ratelimit rl;
607 got_ratelimit_init(&rl, 0, 500);
608 memset(&ppa, 0, sizeof(ppa));
609 ppa.progress_cb = progress_cb;
610 ppa.progress_arg = progress_arg;
611 ppa.sendfd = sendfd;
612 err = got_pack_create(packsha1, packfd, delta_cache,
613 their_ids, ntheirs, our_ids, nours, repo, 0, 1, 0,
614 pack_progress, &ppa, &rl, cancel_cb, cancel_arg);
615 if (err)
616 goto done;
618 npackfd = dup(packfd);
619 if (npackfd == -1) {
620 err = got_error_from_errno("dup");
621 goto done;
623 err = got_privsep_send_packfd(&sendibuf, npackfd);
624 if (err != NULL)
625 goto done;
626 npackfd = -1;
627 } else {
628 err = got_privsep_send_packfd(&sendibuf, -1);
629 if (err != NULL)
630 goto done;
633 while (!done) {
634 int success = 0;
635 char *refname = NULL;
636 char *errmsg = NULL;
638 if (cancel_cb) {
639 err = (*cancel_cb)(cancel_arg);
640 if (err)
641 goto done;
643 err = got_privsep_recv_send_progress(&done, &bytes_sent,
644 &success, &refname, &errmsg, &sendibuf);
645 if (err)
646 goto done;
647 if (refname && got_ref_name_is_valid(refname) && success &&
648 strncmp(refname, "refs/tags/", 10) != 0) {
649 struct got_pathlist_entry *my_ref;
650 /*
651 * The server has accepted our changes.
652 * Update our reference in refs/remotes/ accordingly.
653 */
654 my_ref = find_ref(&have_refs, refname);
655 if (my_ref) {
656 err = update_remote_ref(my_ref, remote_name,
657 repo);
658 if (err)
659 goto done;
662 if (refname != NULL ||
663 bytes_sent_cur != bytes_sent) {
664 err = progress_cb(progress_arg, ppa.ncolored,
665 ppa.nfound, ppa.ntrees, ppa.packfile_size,
666 ppa.ncommits, ppa.nobj_total, ppa.nobj_deltify,
667 ppa.nobj_written, bytes_sent,
668 refname, errmsg, success);
669 if (err) {
670 free(refname);
671 free(errmsg);
672 goto done;
674 bytes_sent_cur = bytes_sent;
676 free(refname);
677 free(errmsg);
679 done:
680 if (sendpid != -1) {
681 if (err)
682 got_privsep_send_stop(imsg_sendfds[0]);
683 if (waitpid(sendpid, &sendstatus, 0) == -1 && err == NULL)
684 err = got_error_from_errno("waitpid");
686 if (packfd != -1 && close(packfd) == -1 && err == NULL)
687 err = got_error_from_errno("close");
688 if (delta_cache && fclose(delta_cache) == EOF && err == NULL)
689 err = got_error_from_errno("fclose");
690 if (nsendfd != -1 && close(nsendfd) == -1 && err == NULL)
691 err = got_error_from_errno("close");
692 if (npackfd != -1 && close(npackfd) == -1 && err == NULL)
693 err = got_error_from_errno("close");
695 got_pathlist_free(&have_refs, GOT_PATHLIST_FREE_ALL);
696 got_pathlist_free(&their_refs, GOT_PATHLIST_FREE_ALL);
697 /*
698 * Object ids are owned by have_refs/their_refs and are already freed;
699 * Only the arrays must be freed.
700 */
701 free(our_ids);
702 free(their_ids);
703 free(s);
704 return err;