6 #define PREV_BUF_SIZE 4096
12 struct packed_git
*packs
;
13 struct alt_base
*next
;
16 enum object_request_state
{
25 struct walker
*walker
;
26 unsigned char sha1
[20];
27 struct alt_base
*repo
;
29 char filename
[PATH_MAX
];
30 char tmpfile
[PATH_MAX
];
32 enum object_request_state state
;
34 char errorstr
[CURL_ERROR_SIZE
];
36 unsigned char real_sha1
[20];
41 struct active_request_slot
*slot
;
42 struct object_request
*next
;
45 struct alternates_request
{
46 struct walker
*walker
;
49 struct strbuf
*buffer
;
50 struct active_request_slot
*slot
;
60 static struct object_request
*object_queue_head
;
62 static size_t fwrite_sha1_file(void *ptr
, size_t eltsize
, size_t nmemb
,
65 unsigned char expn
[4096];
66 size_t size
= eltsize
* nmemb
;
68 struct object_request
*obj_req
= (struct object_request
*)data
;
70 ssize_t retval
= xwrite(obj_req
->local
,
71 (char *) ptr
+ posn
, size
- posn
);
75 } while (posn
< size
);
77 obj_req
->stream
.avail_in
= size
;
78 obj_req
->stream
.next_in
= ptr
;
80 obj_req
->stream
.next_out
= expn
;
81 obj_req
->stream
.avail_out
= sizeof(expn
);
82 obj_req
->zret
= git_inflate(&obj_req
->stream
, Z_SYNC_FLUSH
);
83 git_SHA1_Update(&obj_req
->c
, expn
,
84 sizeof(expn
) - obj_req
->stream
.avail_out
);
85 } while (obj_req
->stream
.avail_in
&& obj_req
->zret
== Z_OK
);
90 static void fetch_alternates(struct walker
*walker
, const char *base
);
92 static void process_object_response(void *callback_data
);
94 static void start_object_request(struct walker
*walker
,
95 struct object_request
*obj_req
)
97 char *hex
= sha1_to_hex(obj_req
->sha1
);
98 char prevfile
[PATH_MAX
];
102 unsigned char prev_buf
[PREV_BUF_SIZE
];
103 ssize_t prev_read
= 0;
105 char range
[RANGE_HEADER_SIZE
];
106 struct curl_slist
*range_header
= NULL
;
107 struct active_request_slot
*slot
;
109 snprintf(prevfile
, sizeof(prevfile
), "%s.prev", obj_req
->filename
);
110 unlink_or_warn(prevfile
);
111 rename(obj_req
->tmpfile
, prevfile
);
112 unlink_or_warn(obj_req
->tmpfile
);
114 if (obj_req
->local
!= -1)
115 error("fd leakage in start: %d", obj_req
->local
);
116 obj_req
->local
= open(obj_req
->tmpfile
,
117 O_WRONLY
| O_CREAT
| O_EXCL
, 0666);
119 * This could have failed due to the "lazy directory creation";
120 * try to mkdir the last path component.
122 if (obj_req
->local
< 0 && errno
== ENOENT
) {
123 char *dir
= strrchr(obj_req
->tmpfile
, '/');
126 mkdir(obj_req
->tmpfile
, 0777);
129 obj_req
->local
= open(obj_req
->tmpfile
,
130 O_WRONLY
| O_CREAT
| O_EXCL
, 0666);
133 if (obj_req
->local
< 0) {
134 obj_req
->state
= ABORTED
;
135 error("Couldn't create temporary file %s for %s: %s",
136 obj_req
->tmpfile
, obj_req
->filename
, strerror(errno
));
140 memset(&obj_req
->stream
, 0, sizeof(obj_req
->stream
));
142 git_inflate_init(&obj_req
->stream
);
144 git_SHA1_Init(&obj_req
->c
);
146 url
= xmalloc(strlen(obj_req
->repo
->base
) + 51);
147 obj_req
->url
= xmalloc(strlen(obj_req
->repo
->base
) + 51);
148 strcpy(url
, obj_req
->repo
->base
);
149 posn
= url
+ strlen(obj_req
->repo
->base
);
150 strcpy(posn
, "/objects/");
152 memcpy(posn
, hex
, 2);
155 strcpy(posn
, hex
+ 2);
156 strcpy(obj_req
->url
, url
);
159 * If a previous temp file is present, process what was already
162 prevlocal
= open(prevfile
, O_RDONLY
);
163 if (prevlocal
!= -1) {
165 prev_read
= xread(prevlocal
, prev_buf
, PREV_BUF_SIZE
);
167 if (fwrite_sha1_file(prev_buf
,
170 obj_req
) == prev_read
)
171 prev_posn
+= prev_read
;
175 } while (prev_read
> 0);
178 unlink_or_warn(prevfile
);
181 * Reset inflate/SHA1 if there was an error reading the previous temp
182 * file; also rewind to the beginning of the local file.
184 if (prev_read
== -1) {
185 memset(&obj_req
->stream
, 0, sizeof(obj_req
->stream
));
186 git_inflate_init(&obj_req
->stream
);
187 git_SHA1_Init(&obj_req
->c
);
190 lseek(obj_req
->local
, 0, SEEK_SET
);
191 ftruncate(obj_req
->local
, 0);
195 slot
= get_active_slot();
196 slot
->callback_func
= process_object_response
;
197 slot
->callback_data
= obj_req
;
198 obj_req
->slot
= slot
;
200 curl_easy_setopt(slot
->curl
, CURLOPT_FILE
, obj_req
);
201 curl_easy_setopt(slot
->curl
, CURLOPT_WRITEFUNCTION
, fwrite_sha1_file
);
202 curl_easy_setopt(slot
->curl
, CURLOPT_ERRORBUFFER
, obj_req
->errorstr
);
203 curl_easy_setopt(slot
->curl
, CURLOPT_URL
, url
);
204 curl_easy_setopt(slot
->curl
, CURLOPT_HTTPHEADER
, no_pragma_header
);
207 * If we have successfully processed data from a previous fetch
208 * attempt, only fetch the data we don't already have.
211 if (walker
->get_verbosely
)
213 "Resuming fetch of object %s at byte %ld\n",
215 sprintf(range
, "Range: bytes=%ld-", prev_posn
);
216 range_header
= curl_slist_append(range_header
, range
);
217 curl_easy_setopt(slot
->curl
,
218 CURLOPT_HTTPHEADER
, range_header
);
221 /* Try to get the request started, abort the request on error */
222 obj_req
->state
= ACTIVE
;
223 if (!start_active_slot(slot
)) {
224 obj_req
->state
= ABORTED
;
225 obj_req
->slot
= NULL
;
226 close(obj_req
->local
);
233 static void finish_object_request(struct object_request
*obj_req
)
237 close(obj_req
->local
);
240 if (obj_req
->http_code
== 416) {
241 fprintf(stderr
, "Warning: requested range invalid; we may already have all the data.\n");
242 } else if (obj_req
->curl_result
!= CURLE_OK
) {
243 if (stat(obj_req
->tmpfile
, &st
) == 0)
245 unlink_or_warn(obj_req
->tmpfile
);
249 git_inflate_end(&obj_req
->stream
);
250 git_SHA1_Final(obj_req
->real_sha1
, &obj_req
->c
);
251 if (obj_req
->zret
!= Z_STREAM_END
) {
252 unlink_or_warn(obj_req
->tmpfile
);
255 if (hashcmp(obj_req
->sha1
, obj_req
->real_sha1
)) {
256 unlink_or_warn(obj_req
->tmpfile
);
260 move_temp_to_file(obj_req
->tmpfile
, obj_req
->filename
);
262 if (obj_req
->rename
== 0)
263 walker_say(obj_req
->walker
, "got %s\n", sha1_to_hex(obj_req
->sha1
));
266 static void process_object_response(void *callback_data
)
268 struct object_request
*obj_req
=
269 (struct object_request
*)callback_data
;
270 struct walker
*walker
= obj_req
->walker
;
271 struct walker_data
*data
= walker
->data
;
272 struct alt_base
*alt
= data
->alt
;
274 obj_req
->curl_result
= obj_req
->slot
->curl_result
;
275 obj_req
->http_code
= obj_req
->slot
->http_code
;
276 obj_req
->slot
= NULL
;
277 obj_req
->state
= COMPLETE
;
279 /* Use alternates if necessary */
280 if (missing_target(obj_req
)) {
281 fetch_alternates(walker
, alt
->base
);
282 if (obj_req
->repo
->next
!= NULL
) {
285 close(obj_req
->local
);
287 start_object_request(walker
, obj_req
);
292 finish_object_request(obj_req
);
295 static void release_object_request(struct object_request
*obj_req
)
297 struct object_request
*entry
= object_queue_head
;
299 if (obj_req
->local
!= -1)
300 error("fd leakage in release: %d", obj_req
->local
);
301 if (obj_req
== object_queue_head
) {
302 object_queue_head
= obj_req
->next
;
304 while (entry
->next
!= NULL
&& entry
->next
!= obj_req
)
306 if (entry
->next
== obj_req
)
307 entry
->next
= entry
->next
->next
;
314 #ifdef USE_CURL_MULTI
315 static int fill_active_slot(struct walker
*walker
)
317 struct object_request
*obj_req
;
319 for (obj_req
= object_queue_head
; obj_req
; obj_req
= obj_req
->next
) {
320 if (obj_req
->state
== WAITING
) {
321 if (has_sha1_file(obj_req
->sha1
))
322 obj_req
->state
= COMPLETE
;
324 start_object_request(walker
, obj_req
);
333 static void prefetch(struct walker
*walker
, unsigned char *sha1
)
335 struct object_request
*newreq
;
336 struct object_request
*tail
;
337 struct walker_data
*data
= walker
->data
;
338 char *filename
= sha1_file_name(sha1
);
340 newreq
= xmalloc(sizeof(*newreq
));
341 newreq
->walker
= walker
;
342 hashcpy(newreq
->sha1
, sha1
);
343 newreq
->repo
= data
->alt
;
346 newreq
->state
= WAITING
;
347 snprintf(newreq
->filename
, sizeof(newreq
->filename
), "%s", filename
);
348 snprintf(newreq
->tmpfile
, sizeof(newreq
->tmpfile
),
349 "%s.temp", filename
);
353 http_is_verbose
= walker
->get_verbosely
;
355 if (object_queue_head
== NULL
) {
356 object_queue_head
= newreq
;
358 tail
= object_queue_head
;
359 while (tail
->next
!= NULL
)
364 #ifdef USE_CURL_MULTI
370 static void process_alternates_response(void *callback_data
)
372 struct alternates_request
*alt_req
=
373 (struct alternates_request
*)callback_data
;
374 struct walker
*walker
= alt_req
->walker
;
375 struct walker_data
*cdata
= walker
->data
;
376 struct active_request_slot
*slot
= alt_req
->slot
;
377 struct alt_base
*tail
= cdata
->alt
;
378 const char *base
= alt_req
->base
;
379 static const char null_byte
= '\0';
383 if (alt_req
->http_specific
) {
384 if (slot
->curl_result
!= CURLE_OK
||
385 !alt_req
->buffer
->len
) {
387 /* Try reusing the slot to get non-http alternates */
388 alt_req
->http_specific
= 0;
389 sprintf(alt_req
->url
, "%s/objects/info/alternates",
391 curl_easy_setopt(slot
->curl
, CURLOPT_URL
,
395 if (slot
->finished
!= NULL
)
396 (*slot
->finished
) = 0;
397 if (!start_active_slot(slot
)) {
398 cdata
->got_alternates
= -1;
400 if (slot
->finished
!= NULL
)
401 (*slot
->finished
) = 1;
405 } else if (slot
->curl_result
!= CURLE_OK
) {
406 if (!missing_target(slot
)) {
407 cdata
->got_alternates
= -1;
412 fwrite_buffer(&null_byte
, 1, 1, alt_req
->buffer
);
413 alt_req
->buffer
->len
--;
414 data
= alt_req
->buffer
->buf
;
416 while (i
< alt_req
->buffer
->len
) {
418 while (posn
< alt_req
->buffer
->len
&& data
[posn
] != '\n')
420 if (data
[posn
] == '\n') {
423 struct alt_base
*newalt
;
425 if (data
[i
] == '/') {
428 * http://git.host/pub/scm/linux.git/
430 * so memcpy(dst, base, serverlen) will
431 * copy up to "...git.host".
433 const char *colon_ss
= strstr(base
,"://");
435 serverlen
= (strchr(colon_ss
+ 3, '/')
439 } else if (!memcmp(data
+ i
, "../", 3)) {
441 * Relative URL; chop the corresponding
442 * number of subpath from base (and ../
443 * from data), and concatenate the result.
445 * The code first drops ../ from data, and
446 * then drops one ../ from data and one path
447 * from base. IOW, one extra ../ is dropped
448 * from data than path is dropped from base.
450 * This is not wrong. The alternate in
451 * http://git.host/pub/scm/linux.git/
453 * http://git.host/pub/scm/linus.git/
454 * is ../../linus.git/objects/. You need
455 * two ../../ to borrow from your direct
459 serverlen
= strlen(base
);
460 while (i
+ 2 < posn
&&
461 !memcmp(data
+ i
, "../", 3)) {
464 } while (serverlen
&&
465 base
[serverlen
- 1] != '/');
468 /* If the server got removed, give up. */
469 okay
= strchr(base
, ':') - base
+ 3 <
471 } else if (alt_req
->http_specific
) {
472 char *colon
= strchr(data
+ i
, ':');
473 char *slash
= strchr(data
+ i
, '/');
474 if (colon
&& slash
&& colon
< data
+ posn
&&
475 slash
< data
+ posn
&& colon
< slash
) {
479 /* skip "objects\n" at end */
481 target
= xmalloc(serverlen
+ posn
- i
- 6);
482 memcpy(target
, base
, serverlen
);
483 memcpy(target
+ serverlen
, data
+ i
,
485 target
[serverlen
+ posn
- i
- 7] = 0;
486 if (walker
->get_verbosely
)
488 "Also look at %s\n", target
);
489 newalt
= xmalloc(sizeof(*newalt
));
491 newalt
->base
= target
;
492 newalt
->got_indices
= 0;
493 newalt
->packs
= NULL
;
495 while (tail
->next
!= NULL
)
503 cdata
->got_alternates
= 1;
506 static void fetch_alternates(struct walker
*walker
, const char *base
)
508 struct strbuf buffer
= STRBUF_INIT
;
510 struct active_request_slot
*slot
;
511 struct alternates_request alt_req
;
512 struct walker_data
*cdata
= walker
->data
;
515 * If another request has already started fetching alternates,
516 * wait for them to arrive and return to processing this request's
519 #ifdef USE_CURL_MULTI
520 while (cdata
->got_alternates
== 0) {
525 /* Nothing to do if they've already been fetched */
526 if (cdata
->got_alternates
== 1)
529 /* Start the fetch */
530 cdata
->got_alternates
= 0;
532 if (walker
->get_verbosely
)
533 fprintf(stderr
, "Getting alternates list for %s\n", base
);
535 url
= xmalloc(strlen(base
) + 31);
536 sprintf(url
, "%s/objects/info/http-alternates", base
);
539 * Use a callback to process the result, since another request
540 * may fail and need to have alternates loaded before continuing
542 slot
= get_active_slot();
543 slot
->callback_func
= process_alternates_response
;
544 alt_req
.walker
= walker
;
545 slot
->callback_data
= &alt_req
;
547 curl_easy_setopt(slot
->curl
, CURLOPT_FILE
, &buffer
);
548 curl_easy_setopt(slot
->curl
, CURLOPT_WRITEFUNCTION
, fwrite_buffer
);
549 curl_easy_setopt(slot
->curl
, CURLOPT_URL
, url
);
553 alt_req
.buffer
= &buffer
;
554 alt_req
.http_specific
= 1;
557 if (start_active_slot(slot
))
558 run_active_slot(slot
);
560 cdata
->got_alternates
= -1;
562 strbuf_release(&buffer
);
566 static int fetch_indices(struct walker
*walker
, struct alt_base
*repo
)
570 if (repo
->got_indices
)
573 if (walker
->get_verbosely
)
574 fprintf(stderr
, "Getting pack list for %s\n", repo
->base
);
576 switch (http_get_info_packs(repo
->base
, &repo
->packs
)) {
578 case HTTP_MISSING_TARGET
:
579 repo
->got_indices
= 1;
583 repo
->got_indices
= 0;
590 static int fetch_pack(struct walker
*walker
, struct alt_base
*repo
, unsigned char *sha1
)
592 struct packed_git
*target
;
594 struct slot_results results
;
595 struct http_pack_request
*preq
;
597 if (fetch_indices(walker
, repo
))
599 target
= find_sha1_pack(sha1
, repo
->packs
);
603 if (walker
->get_verbosely
) {
604 fprintf(stderr
, "Getting pack %s\n",
605 sha1_to_hex(target
->sha1
));
606 fprintf(stderr
, " which contains %s\n",
610 preq
= new_http_pack_request(target
, repo
->base
);
613 preq
->lst
= &repo
->packs
;
614 preq
->slot
->results
= &results
;
616 if (start_active_slot(preq
->slot
)) {
617 run_active_slot(preq
->slot
);
618 if (results
.curl_result
!= CURLE_OK
) {
619 error("Unable to get pack file %s\n%s", preq
->url
,
624 error("Unable to start request");
628 ret
= finish_http_pack_request(preq
);
629 release_http_pack_request(preq
);
639 static void abort_object_request(struct object_request
*obj_req
)
641 if (obj_req
->local
>= 0) {
642 close(obj_req
->local
);
645 unlink_or_warn(obj_req
->tmpfile
);
647 release_active_slot(obj_req
->slot
);
648 obj_req
->slot
= NULL
;
650 release_object_request(obj_req
);
653 static int fetch_object(struct walker
*walker
, struct alt_base
*repo
, unsigned char *sha1
)
655 char *hex
= sha1_to_hex(sha1
);
657 struct object_request
*obj_req
= object_queue_head
;
659 while (obj_req
!= NULL
&& hashcmp(obj_req
->sha1
, sha1
))
660 obj_req
= obj_req
->next
;
662 return error("Couldn't find request for %s in the queue", hex
);
664 if (has_sha1_file(obj_req
->sha1
)) {
665 abort_object_request(obj_req
);
669 #ifdef USE_CURL_MULTI
670 while (obj_req
->state
== WAITING
)
673 start_object_request(walker
, obj_req
);
676 while (obj_req
->state
== ACTIVE
)
677 run_active_slot(obj_req
->slot
);
679 if (obj_req
->local
!= -1) {
680 close(obj_req
->local
);
684 if (obj_req
->state
== ABORTED
) {
685 ret
= error("Request for %s aborted", hex
);
686 } else if (obj_req
->curl_result
!= CURLE_OK
&&
687 obj_req
->http_code
!= 416) {
688 if (missing_target(obj_req
))
689 ret
= -1; /* Be silent, it is probably in a pack. */
691 ret
= error("%s (curl_result = %d, http_code = %ld, sha1 = %s)",
692 obj_req
->errorstr
, obj_req
->curl_result
,
693 obj_req
->http_code
, hex
);
694 } else if (obj_req
->zret
!= Z_STREAM_END
) {
695 walker
->corrupt_object_found
++;
696 ret
= error("File %s (%s) corrupt", hex
, obj_req
->url
);
697 } else if (hashcmp(obj_req
->sha1
, obj_req
->real_sha1
)) {
698 ret
= error("File %s has bad hash", hex
);
699 } else if (obj_req
->rename
< 0) {
700 ret
= error("unable to write sha1 filename %s",
704 release_object_request(obj_req
);
708 static int fetch(struct walker
*walker
, unsigned char *sha1
)
710 struct walker_data
*data
= walker
->data
;
711 struct alt_base
*altbase
= data
->alt
;
713 if (!fetch_object(walker
, altbase
, sha1
))
716 if (!fetch_pack(walker
, altbase
, sha1
))
718 fetch_alternates(walker
, data
->alt
->base
);
719 altbase
= altbase
->next
;
721 return error("Unable to find %s under %s", sha1_to_hex(sha1
),
725 static int fetch_ref(struct walker
*walker
, struct ref
*ref
)
727 struct walker_data
*data
= walker
->data
;
728 return http_fetch_ref(data
->alt
->base
, ref
);
731 static void cleanup(struct walker
*walker
)
736 struct walker
*get_http_walker(const char *url
, struct remote
*remote
)
739 struct walker_data
*data
= xmalloc(sizeof(struct walker_data
));
740 struct walker
*walker
= xmalloc(sizeof(struct walker
));
744 data
->alt
= xmalloc(sizeof(*data
->alt
));
745 data
->alt
->base
= xmalloc(strlen(url
) + 1);
746 strcpy(data
->alt
->base
, url
);
747 for (s
= data
->alt
->base
+ strlen(data
->alt
->base
) - 1; *s
== '/'; --s
)
750 data
->alt
->got_indices
= 0;
751 data
->alt
->packs
= NULL
;
752 data
->alt
->next
= NULL
;
753 data
->got_alternates
= -1;
755 walker
->corrupt_object_found
= 0;
756 walker
->fetch
= fetch
;
757 walker
->fetch_ref
= fetch_ref
;
758 walker
->prefetch
= prefetch
;
759 walker
->cleanup
= cleanup
;
762 #ifdef USE_CURL_MULTI
763 add_fill_function(walker
, (int (*)(void *)) fill_active_slot
);