Smart HTTP fetch: gzip requests
[git.git] / remote-curl.c
blob0d7cf16e9c06d556cad8d92b6598a8da079426a4
1 #include "cache.h"
2 #include "remote.h"
3 #include "strbuf.h"
4 #include "walker.h"
5 #include "http.h"
6 #include "exec_cmd.h"
7 #include "run-command.h"
8 #include "pkt-line.h"
9 #include "sideband.h"
11 static struct remote *remote;
12 static const char *url;
13 static struct walker *walker;
15 struct options {
16 int verbosity;
17 unsigned long depth;
18 unsigned progress : 1,
19 followtags : 1,
20 dry_run : 1,
21 thin : 1;
23 static struct options options;
25 static void init_walker(void)
27 if (!walker)
28 walker = get_http_walker(url, remote);
31 static int set_option(const char *name, const char *value)
33 if (!strcmp(name, "verbosity")) {
34 char *end;
35 int v = strtol(value, &end, 10);
36 if (value == end || *end)
37 return -1;
38 options.verbosity = v;
39 return 0;
41 else if (!strcmp(name, "progress")) {
42 if (!strcmp(value, "true"))
43 options.progress = 1;
44 else if (!strcmp(value, "false"))
45 options.progress = 0;
46 else
47 return -1;
48 return 0;
50 else if (!strcmp(name, "depth")) {
51 char *end;
52 unsigned long v = strtoul(value, &end, 10);
53 if (value == end || *end)
54 return -1;
55 options.depth = v;
56 return 0;
58 else if (!strcmp(name, "followtags")) {
59 if (!strcmp(value, "true"))
60 options.followtags = 1;
61 else if (!strcmp(value, "false"))
62 options.followtags = 0;
63 else
64 return -1;
65 return 0;
67 else if (!strcmp(name, "dry-run")) {
68 if (!strcmp(value, "true"))
69 options.dry_run = 1;
70 else if (!strcmp(value, "false"))
71 options.dry_run = 0;
72 else
73 return -1;
74 return 0;
76 else {
77 return 1 /* unsupported */;
81 struct discovery {
82 const char *service;
83 char *buf_alloc;
84 char *buf;
85 size_t len;
86 unsigned proto_git : 1;
88 static struct discovery *last_discovery;
90 static void free_discovery(struct discovery *d)
92 if (d) {
93 if (d == last_discovery)
94 last_discovery = NULL;
95 free(d->buf_alloc);
96 free(d);
100 static struct discovery* discover_refs(const char *service)
102 struct strbuf buffer = STRBUF_INIT;
103 struct discovery *last = last_discovery;
104 char *refs_url;
105 int http_ret, is_http = 0;
107 if (last && !strcmp(service, last->service))
108 return last;
109 free_discovery(last);
111 strbuf_addf(&buffer, "%s/info/refs", url);
112 if (!prefixcmp(url, "http://") || !prefixcmp(url, "https://")) {
113 is_http = 1;
114 if (!strchr(url, '?'))
115 strbuf_addch(&buffer, '?');
116 else
117 strbuf_addch(&buffer, '&');
118 strbuf_addf(&buffer, "service=%s", service);
120 refs_url = strbuf_detach(&buffer, NULL);
122 init_walker();
123 http_ret = http_get_strbuf(refs_url, &buffer, HTTP_NO_CACHE);
124 switch (http_ret) {
125 case HTTP_OK:
126 break;
127 case HTTP_MISSING_TARGET:
128 die("%s not found: did you run git update-server-info on the"
129 " server?", refs_url);
130 default:
131 http_error(refs_url, http_ret);
132 die("HTTP request failed");
135 last= xcalloc(1, sizeof(*last_discovery));
136 last->service = service;
137 last->buf_alloc = strbuf_detach(&buffer, &last->len);
138 last->buf = last->buf_alloc;
140 if (is_http && 5 <= last->len && last->buf[4] == '#') {
141 /* smart HTTP response; validate that the service
142 * pkt-line matches our request.
144 struct strbuf exp = STRBUF_INIT;
146 if (packet_get_line(&buffer, &last->buf, &last->len) <= 0)
147 die("%s has invalid packet header", refs_url);
148 if (buffer.len && buffer.buf[buffer.len - 1] == '\n')
149 strbuf_setlen(&buffer, buffer.len - 1);
151 strbuf_addf(&exp, "# service=%s", service);
152 if (strbuf_cmp(&exp, &buffer))
153 die("invalid server response; got '%s'", buffer.buf);
154 strbuf_release(&exp);
156 /* The header can include additional metadata lines, up
157 * until a packet flush marker. Ignore these now, but
158 * in the future we might start to scan them.
160 strbuf_reset(&buffer);
161 while (packet_get_line(&buffer, &last->buf, &last->len) > 0)
162 strbuf_reset(&buffer);
164 last->proto_git = 1;
167 free(refs_url);
168 strbuf_release(&buffer);
169 last_discovery = last;
170 return last;
173 static int write_discovery(int fd, void *data)
175 struct discovery *heads = data;
176 int err = 0;
177 if (write_in_full(fd, heads->buf, heads->len) != heads->len)
178 err = 1;
179 close(fd);
180 return err;
183 static struct ref *parse_git_refs(struct discovery *heads)
185 struct ref *list = NULL;
186 struct async async;
188 memset(&async, 0, sizeof(async));
189 async.proc = write_discovery;
190 async.data = heads;
192 if (start_async(&async))
193 die("cannot start thread to parse advertised refs");
194 get_remote_heads(async.out, &list, 0, NULL, 0, NULL);
195 close(async.out);
196 if (finish_async(&async))
197 die("ref parsing thread failed");
198 return list;
201 static struct ref *parse_info_refs(struct discovery *heads)
203 char *data, *start, *mid;
204 char *ref_name;
205 int i = 0;
207 struct ref *refs = NULL;
208 struct ref *ref = NULL;
209 struct ref *last_ref = NULL;
211 data = heads->buf;
212 start = NULL;
213 mid = data;
214 while (i < heads->len) {
215 if (!start) {
216 start = &data[i];
218 if (data[i] == '\t')
219 mid = &data[i];
220 if (data[i] == '\n') {
221 data[i] = 0;
222 ref_name = mid + 1;
223 ref = xmalloc(sizeof(struct ref) +
224 strlen(ref_name) + 1);
225 memset(ref, 0, sizeof(struct ref));
226 strcpy(ref->name, ref_name);
227 get_sha1_hex(start, ref->old_sha1);
228 if (!refs)
229 refs = ref;
230 if (last_ref)
231 last_ref->next = ref;
232 last_ref = ref;
233 start = NULL;
235 i++;
238 init_walker();
239 ref = alloc_ref("HEAD");
240 if (!walker->fetch_ref(walker, ref) &&
241 !resolve_remote_symref(ref, refs)) {
242 ref->next = refs;
243 refs = ref;
244 } else {
245 free(ref);
248 return refs;
251 static struct ref *get_refs(int for_push)
253 struct discovery *heads;
255 if (for_push)
256 heads = discover_refs("git-receive-pack");
257 else
258 heads = discover_refs("git-upload-pack");
260 if (heads->proto_git)
261 return parse_git_refs(heads);
262 return parse_info_refs(heads);
265 static void output_refs(struct ref *refs)
267 struct ref *posn;
268 for (posn = refs; posn; posn = posn->next) {
269 if (posn->symref)
270 printf("@%s %s\n", posn->symref, posn->name);
271 else
272 printf("%s %s\n", sha1_to_hex(posn->old_sha1), posn->name);
274 printf("\n");
275 fflush(stdout);
276 free_refs(refs);
279 struct rpc_state {
280 const char *service_name;
281 const char **argv;
282 char *service_url;
283 char *hdr_content_type;
284 char *hdr_accept;
285 char *buf;
286 size_t alloc;
287 size_t len;
288 size_t pos;
289 int in;
290 int out;
291 struct strbuf result;
292 unsigned gzip_request : 1;
295 static size_t rpc_out(void *ptr, size_t eltsize,
296 size_t nmemb, void *buffer_)
298 size_t max = eltsize * nmemb;
299 struct rpc_state *rpc = buffer_;
300 size_t avail = rpc->len - rpc->pos;
302 if (!avail) {
303 avail = packet_read_line(rpc->out, rpc->buf, rpc->alloc);
304 if (!avail)
305 return 0;
306 rpc->pos = 0;
307 rpc->len = avail;
310 if (max < avail);
311 avail = max;
312 memcpy(ptr, rpc->buf + rpc->pos, avail);
313 rpc->pos += avail;
314 return avail;
317 static size_t rpc_in(const void *ptr, size_t eltsize,
318 size_t nmemb, void *buffer_)
320 size_t size = eltsize * nmemb;
321 struct rpc_state *rpc = buffer_;
322 write_or_die(rpc->in, ptr, size);
323 return size;
326 static int post_rpc(struct rpc_state *rpc)
328 struct active_request_slot *slot;
329 struct slot_results results;
330 struct curl_slist *headers = NULL;
331 int use_gzip = rpc->gzip_request;
332 char *gzip_body = NULL;
333 int err = 0, large_request = 0;
335 /* Try to load the entire request, if we can fit it into the
336 * allocated buffer space we can use HTTP/1.0 and avoid the
337 * chunked encoding mess.
339 while (1) {
340 size_t left = rpc->alloc - rpc->len;
341 char *buf = rpc->buf + rpc->len;
342 int n;
344 if (left < LARGE_PACKET_MAX) {
345 large_request = 1;
346 use_gzip = 0;
347 break;
350 n = packet_read_line(rpc->out, buf, left);
351 if (!n)
352 break;
353 rpc->len += n;
356 slot = get_active_slot();
357 slot->results = &results;
359 curl_easy_setopt(slot->curl, CURLOPT_POST, 1);
360 curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0);
361 curl_easy_setopt(slot->curl, CURLOPT_URL, rpc->service_url);
362 curl_easy_setopt(slot->curl, CURLOPT_ENCODING, "");
364 headers = curl_slist_append(headers, rpc->hdr_content_type);
365 headers = curl_slist_append(headers, rpc->hdr_accept);
367 if (large_request) {
368 /* The request body is large and the size cannot be predicted.
369 * We must use chunked encoding to send it.
371 headers = curl_slist_append(headers, "Expect: 100-continue");
372 headers = curl_slist_append(headers, "Transfer-Encoding: chunked");
373 curl_easy_setopt(slot->curl, CURLOPT_READFUNCTION, rpc_out);
374 curl_easy_setopt(slot->curl, CURLOPT_INFILE, rpc);
375 if (options.verbosity > 1) {
376 fprintf(stderr, "POST %s (chunked)\n", rpc->service_name);
377 fflush(stderr);
380 } else if (use_gzip && 1024 < rpc->len) {
381 /* The client backend isn't giving us compressed data so
382 * we can try to deflate it ourselves, this may save on.
383 * the transfer time.
385 size_t size;
386 z_stream stream;
387 int ret;
389 memset(&stream, 0, sizeof(stream));
390 ret = deflateInit2(&stream, Z_BEST_COMPRESSION,
391 Z_DEFLATED, (15 + 16),
392 8, Z_DEFAULT_STRATEGY);
393 if (ret != Z_OK)
394 die("cannot deflate request; zlib init error %d", ret);
395 size = deflateBound(&stream, rpc->len);
396 gzip_body = xmalloc(size);
398 stream.next_in = (unsigned char *)rpc->buf;
399 stream.avail_in = rpc->len;
400 stream.next_out = (unsigned char *)gzip_body;
401 stream.avail_out = size;
403 ret = deflate(&stream, Z_FINISH);
404 if (ret != Z_STREAM_END)
405 die("cannot deflate request; zlib deflate error %d", ret);
407 ret = deflateEnd(&stream);
408 if (ret != Z_OK)
409 die("cannot deflate request; zlib end error %d", ret);
411 size = stream.total_out;
413 headers = curl_slist_append(headers, "Content-Encoding: gzip");
414 curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDS, gzip_body);
415 curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDSIZE, size);
417 if (options.verbosity > 1) {
418 fprintf(stderr, "POST %s (gzip %lu to %lu bytes)\n",
419 rpc->service_name,
420 (unsigned long)rpc->len, (unsigned long)size);
421 fflush(stderr);
423 } else {
424 /* We know the complete request size in advance, use the
425 * more normal Content-Length approach.
427 curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDS, rpc->buf);
428 curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDSIZE, rpc->len);
429 if (options.verbosity > 1) {
430 fprintf(stderr, "POST %s (%lu bytes)\n",
431 rpc->service_name, (unsigned long)rpc->len);
432 fflush(stderr);
436 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, headers);
437 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, rpc_in);
438 curl_easy_setopt(slot->curl, CURLOPT_FILE, rpc);
440 slot->curl_result = curl_easy_perform(slot->curl);
441 finish_active_slot(slot);
443 if (results.curl_result != CURLE_OK) {
444 err |= error("RPC failed; result=%d, HTTP code = %ld",
445 results.curl_result, results.http_code);
448 curl_slist_free_all(headers);
449 free(gzip_body);
450 return err;
453 static int rpc_service(struct rpc_state *rpc, struct discovery *heads)
455 const char *svc = rpc->service_name;
456 struct strbuf buf = STRBUF_INIT;
457 struct child_process client;
458 int err = 0;
460 init_walker();
461 memset(&client, 0, sizeof(client));
462 client.in = -1;
463 client.out = -1;
464 client.git_cmd = 1;
465 client.argv = rpc->argv;
466 if (start_command(&client))
467 exit(1);
468 if (heads)
469 write_or_die(client.in, heads->buf, heads->len);
471 rpc->alloc = http_post_buffer;
472 rpc->buf = xmalloc(rpc->alloc);
473 rpc->in = client.in;
474 rpc->out = client.out;
475 strbuf_init(&rpc->result, 0);
477 strbuf_addf(&buf, "%s/%s", url, svc);
478 rpc->service_url = strbuf_detach(&buf, NULL);
480 strbuf_addf(&buf, "Content-Type: application/x-%s-request", svc);
481 rpc->hdr_content_type = strbuf_detach(&buf, NULL);
483 strbuf_addf(&buf, "Accept: application/x-%s-response", svc);
484 rpc->hdr_accept = strbuf_detach(&buf, NULL);
486 while (!err) {
487 int n = packet_read_line(rpc->out, rpc->buf, rpc->alloc);
488 if (!n)
489 break;
490 rpc->pos = 0;
491 rpc->len = n;
492 err |= post_rpc(rpc);
494 strbuf_read(&rpc->result, client.out, 0);
496 close(client.in);
497 close(client.out);
498 client.in = -1;
499 client.out = -1;
501 err |= finish_command(&client);
502 free(rpc->service_url);
503 free(rpc->hdr_content_type);
504 free(rpc->hdr_accept);
505 free(rpc->buf);
506 strbuf_release(&buf);
507 return err;
510 static int fetch_dumb(int nr_heads, struct ref **to_fetch)
512 char **targets = xmalloc(nr_heads * sizeof(char*));
513 int ret, i;
515 if (options.depth)
516 die("dumb http transport does not support --depth");
517 for (i = 0; i < nr_heads; i++)
518 targets[i] = xstrdup(sha1_to_hex(to_fetch[i]->old_sha1));
520 init_walker();
521 walker->get_all = 1;
522 walker->get_tree = 1;
523 walker->get_history = 1;
524 walker->get_verbosely = options.verbosity >= 3;
525 walker->get_recover = 0;
526 ret = walker_fetch(walker, nr_heads, targets, NULL, NULL);
528 for (i = 0; i < nr_heads; i++)
529 free(targets[i]);
530 free(targets);
532 return ret ? error("Fetch failed.") : 0;
535 static int fetch_git(struct discovery *heads,
536 int nr_heads, struct ref **to_fetch)
538 struct rpc_state rpc;
539 char *depth_arg = NULL;
540 const char **argv;
541 int argc = 0, i, err;
543 argv = xmalloc((15 + nr_heads) * sizeof(char*));
544 argv[argc++] = "fetch-pack";
545 argv[argc++] = "--stateless-rpc";
546 argv[argc++] = "--lock-pack";
547 if (options.followtags)
548 argv[argc++] = "--include-tag";
549 if (options.thin)
550 argv[argc++] = "--thin";
551 if (options.verbosity >= 3) {
552 argv[argc++] = "-v";
553 argv[argc++] = "-v";
555 if (!options.progress)
556 argv[argc++] = "--no-progress";
557 if (options.depth) {
558 struct strbuf buf = STRBUF_INIT;
559 strbuf_addf(&buf, "--depth=%lu", options.depth);
560 depth_arg = strbuf_detach(&buf, NULL);
561 argv[argc++] = depth_arg;
563 argv[argc++] = url;
564 for (i = 0; i < nr_heads; i++) {
565 struct ref *ref = to_fetch[i];
566 if (!ref->name || !*ref->name)
567 die("cannot fetch by sha1 over smart http");
568 argv[argc++] = ref->name;
570 argv[argc++] = NULL;
572 memset(&rpc, 0, sizeof(rpc));
573 rpc.service_name = "git-upload-pack",
574 rpc.argv = argv;
575 rpc.gzip_request = 1;
577 err = rpc_service(&rpc, heads);
578 if (rpc.result.len)
579 safe_write(1, rpc.result.buf, rpc.result.len);
580 strbuf_release(&rpc.result);
581 free(argv);
582 free(depth_arg);
583 return err;
586 static int fetch(int nr_heads, struct ref **to_fetch)
588 struct discovery *d = discover_refs("git-upload-pack");
589 if (d->proto_git)
590 return fetch_git(d, nr_heads, to_fetch);
591 else
592 return fetch_dumb(nr_heads, to_fetch);
595 static void parse_fetch(struct strbuf *buf)
597 struct ref **to_fetch = NULL;
598 struct ref *list_head = NULL;
599 struct ref **list = &list_head;
600 int alloc_heads = 0, nr_heads = 0;
602 do {
603 if (!prefixcmp(buf->buf, "fetch ")) {
604 char *p = buf->buf + strlen("fetch ");
605 char *name;
606 struct ref *ref;
607 unsigned char old_sha1[20];
609 if (strlen(p) < 40 || get_sha1_hex(p, old_sha1))
610 die("protocol error: expected sha/ref, got %s'", p);
611 if (p[40] == ' ')
612 name = p + 41;
613 else if (!p[40])
614 name = "";
615 else
616 die("protocol error: expected sha/ref, got %s'", p);
618 ref = alloc_ref(name);
619 hashcpy(ref->old_sha1, old_sha1);
621 *list = ref;
622 list = &ref->next;
624 ALLOC_GROW(to_fetch, nr_heads + 1, alloc_heads);
625 to_fetch[nr_heads++] = ref;
627 else
628 die("http transport does not support %s", buf->buf);
630 strbuf_reset(buf);
631 if (strbuf_getline(buf, stdin, '\n') == EOF)
632 return;
633 if (!*buf->buf)
634 break;
635 } while (1);
637 if (fetch(nr_heads, to_fetch))
638 exit(128); /* error already reported */
639 free_refs(list_head);
640 free(to_fetch);
642 printf("\n");
643 fflush(stdout);
644 strbuf_reset(buf);
647 static int push_dav(int nr_spec, char **specs)
649 const char **argv = xmalloc((10 + nr_spec) * sizeof(char*));
650 int argc = 0, i;
652 argv[argc++] = "http-push";
653 argv[argc++] = "--helper-status";
654 if (options.dry_run)
655 argv[argc++] = "--dry-run";
656 if (options.verbosity > 1)
657 argv[argc++] = "--verbose";
658 argv[argc++] = url;
659 for (i = 0; i < nr_spec; i++)
660 argv[argc++] = specs[i];
661 argv[argc++] = NULL;
663 if (run_command_v_opt(argv, RUN_GIT_CMD))
664 die("git-%s failed", argv[0]);
665 free(argv);
666 return 0;
669 static int push_git(struct discovery *heads, int nr_spec, char **specs)
671 struct rpc_state rpc;
672 const char **argv;
673 int argc = 0, i, err;
675 argv = xmalloc((10 + nr_spec) * sizeof(char*));
676 argv[argc++] = "send-pack";
677 argv[argc++] = "--stateless-rpc";
678 argv[argc++] = "--helper-status";
679 if (options.thin)
680 argv[argc++] = "--thin";
681 if (options.dry_run)
682 argv[argc++] = "--dry-run";
683 if (options.verbosity > 1)
684 argv[argc++] = "--verbose";
685 argv[argc++] = url;
686 for (i = 0; i < nr_spec; i++)
687 argv[argc++] = specs[i];
688 argv[argc++] = NULL;
690 memset(&rpc, 0, sizeof(rpc));
691 rpc.service_name = "git-receive-pack",
692 rpc.argv = argv;
694 err = rpc_service(&rpc, heads);
695 if (rpc.result.len)
696 safe_write(1, rpc.result.buf, rpc.result.len);
697 strbuf_release(&rpc.result);
698 free(argv);
699 return err;
702 static int push(int nr_spec, char **specs)
704 struct discovery *heads = discover_refs("git-receive-pack");
705 int ret;
707 if (heads->proto_git)
708 ret = push_git(heads, nr_spec, specs);
709 else
710 ret = push_dav(nr_spec, specs);
711 free_discovery(heads);
712 return ret;
715 static void parse_push(struct strbuf *buf)
717 char **specs = NULL;
718 int alloc_spec = 0, nr_spec = 0, i;
720 do {
721 if (!prefixcmp(buf->buf, "push ")) {
722 ALLOC_GROW(specs, nr_spec + 1, alloc_spec);
723 specs[nr_spec++] = xstrdup(buf->buf + 5);
725 else
726 die("http transport does not support %s", buf->buf);
728 strbuf_reset(buf);
729 if (strbuf_getline(buf, stdin, '\n') == EOF)
730 return;
731 if (!*buf->buf)
732 break;
733 } while (1);
735 if (push(nr_spec, specs))
736 exit(128); /* error already reported */
737 for (i = 0; i < nr_spec; i++)
738 free(specs[i]);
739 free(specs);
741 printf("\n");
742 fflush(stdout);
745 int main(int argc, const char **argv)
747 struct strbuf buf = STRBUF_INIT;
749 git_extract_argv0_path(argv[0]);
750 setup_git_directory();
751 if (argc < 2) {
752 fprintf(stderr, "Remote needed\n");
753 return 1;
756 options.verbosity = 1;
757 options.progress = !!isatty(2);
758 options.thin = 1;
760 remote = remote_get(argv[1]);
762 if (argc > 2) {
763 url = argv[2];
764 } else {
765 url = remote->url[0];
768 do {
769 if (strbuf_getline(&buf, stdin, '\n') == EOF)
770 break;
771 if (!prefixcmp(buf.buf, "fetch ")) {
772 parse_fetch(&buf);
774 } else if (!strcmp(buf.buf, "list") || !prefixcmp(buf.buf, "list ")) {
775 int for_push = !!strstr(buf.buf + 4, "for-push");
776 output_refs(get_refs(for_push));
778 } else if (!prefixcmp(buf.buf, "push ")) {
779 parse_push(&buf);
781 } else if (!prefixcmp(buf.buf, "option ")) {
782 char *name = buf.buf + strlen("option ");
783 char *value = strchr(name, ' ');
784 int result;
786 if (value)
787 *value++ = '\0';
788 else
789 value = "true";
791 result = set_option(name, value);
792 if (!result)
793 printf("ok\n");
794 else if (result < 0)
795 printf("error invalid value\n");
796 else
797 printf("unsupported\n");
798 fflush(stdout);
800 } else if (!strcmp(buf.buf, "capabilities")) {
801 printf("fetch\n");
802 printf("option\n");
803 printf("push\n");
804 printf("\n");
805 fflush(stdout);
806 } else {
807 return 1;
809 strbuf_reset(&buf);
810 } while (1);
811 return 0;