1 /* Copyright 2002-2004 Justin Erenkrantz and Greg Stein
2 *
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at
6 *
7 * http://www.apache.org/licenses/LICENSE-2.0
8 *
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
14 */
15
16 #include <apr_lib.h>
17 #include <apr_strings.h>
18 #include <apr_date.h>
19
20 #include "serf.h"
21 #include "serf_bucket_util.h"
22 #include "serf_private.h"
23
24 typedef struct {
25 serf_bucket_t *stream;
26 serf_bucket_t *body; /* Pointer to the stream wrapping the body. */
27 serf_bucket_t *headers; /* holds parsed headers */
28
29 enum {
30 STATE_STATUS_LINE, /* reading status line */
31 STATE_HEADERS, /* reading headers */
32 STATE_BODY, /* reading body */
33 STATE_TRAILERS, /* reading trailers */
34 STATE_DONE /* we've sent EOF */
35 } state;
36
37 /* Buffer for accumulating a line from the response. */
38 serf_linebuf_t linebuf;
39
40 serf_status_line sl;
41
42 int chunked; /* Do we need to read trailers? */
43 int head_req; /* Was this a HEAD request? */
44 } response_context_t;
45
46 /* Returns 1 if according to RFC2626 this response can have a body, 0 if it
47 must not have a body. */
expect_body(response_context_t * ctx)48 static int expect_body(response_context_t *ctx)
49 {
50 if (ctx->head_req)
51 return 0;
52
53 /* 100 Continue and 101 Switching Protocols */
54 if (ctx->sl.code >= 100 && ctx->sl.code < 200)
55 return 0;
56
57 /* 204 No Content */
58 if (ctx->sl.code == 204)
59 return 0;
60
61 /* 205? */
62
63 /* 304 Not Modified */
64 if (ctx->sl.code == 304)
65 return 0;
66
67 return 1;
68 }
69
serf_bucket_response_create(serf_bucket_t * stream,serf_bucket_alloc_t * allocator)70 serf_bucket_t *serf_bucket_response_create(
71 serf_bucket_t *stream,
72 serf_bucket_alloc_t *allocator)
73 {
74 response_context_t *ctx;
75
76 ctx = serf_bucket_mem_alloc(allocator, sizeof(*ctx));
77 ctx->stream = stream;
78 ctx->body = NULL;
79 ctx->headers = serf_bucket_headers_create(allocator);
80 ctx->state = STATE_STATUS_LINE;
81 ctx->chunked = 0;
82 ctx->head_req = 0;
83
84 serf_linebuf_init(&ctx->linebuf);
85
86 return serf_bucket_create(&serf_bucket_type_response, allocator, ctx);
87 }
88
serf_bucket_response_set_head(serf_bucket_t * bucket)89 void serf_bucket_response_set_head(
90 serf_bucket_t *bucket)
91 {
92 response_context_t *ctx = bucket->data;
93
94 ctx->head_req = 1;
95 }
96
serf_bucket_response_get_headers(serf_bucket_t * bucket)97 serf_bucket_t *serf_bucket_response_get_headers(
98 serf_bucket_t *bucket)
99 {
100 return ((response_context_t *)bucket->data)->headers;
101 }
102
103
serf_response_destroy_and_data(serf_bucket_t * bucket)104 static void serf_response_destroy_and_data(serf_bucket_t *bucket)
105 {
106 response_context_t *ctx = bucket->data;
107
108 if (ctx->state != STATE_STATUS_LINE) {
109 serf_bucket_mem_free(bucket->allocator, (void*)ctx->sl.reason);
110 }
111
112 serf_bucket_destroy(ctx->stream);
113 if (ctx->body != NULL)
114 serf_bucket_destroy(ctx->body);
115 serf_bucket_destroy(ctx->headers);
116
117 serf_default_destroy_and_data(bucket);
118 }
119
fetch_line(response_context_t * ctx,int acceptable)120 static apr_status_t fetch_line(response_context_t *ctx, int acceptable)
121 {
122 return serf_linebuf_fetch(&ctx->linebuf, ctx->stream, acceptable);
123 }
124
parse_status_line(response_context_t * ctx,serf_bucket_alloc_t * allocator)125 static apr_status_t parse_status_line(response_context_t *ctx,
126 serf_bucket_alloc_t *allocator)
127 {
128 int res;
129 char *reason; /* ### stupid APR interface makes this non-const */
130
131 /* ctx->linebuf.line should be of form: HTTP/1.1 200 OK */
132 res = apr_date_checkmask(ctx->linebuf.line, "HTTP/#.# ###*");
133 if (!res) {
134 /* Not an HTTP response? Well, at least we won't understand it. */
135 return SERF_ERROR_BAD_HTTP_RESPONSE;
136 }
137
138 ctx->sl.version = SERF_HTTP_VERSION(ctx->linebuf.line[5] - '0',
139 ctx->linebuf.line[7] - '0');
140 ctx->sl.code = apr_strtoi64(ctx->linebuf.line + 8, &reason, 10);
141
142 /* Skip leading spaces for the reason string. */
143 if (apr_isspace(*reason)) {
144 reason++;
145 }
146
147 /* Copy the reason value out of the line buffer. */
148 ctx->sl.reason = serf_bstrmemdup(allocator, reason,
149 ctx->linebuf.used
150 - (reason - ctx->linebuf.line));
151
152 return APR_SUCCESS;
153 }
154
155 /* This code should be replaced with header buckets. */
fetch_headers(serf_bucket_t * bkt,response_context_t * ctx)156 static apr_status_t fetch_headers(serf_bucket_t *bkt, response_context_t *ctx)
157 {
158 apr_status_t status;
159
160 /* RFC 2616 says that CRLF is the only line ending, but we can easily
161 * accept any kind of line ending.
162 */
163 status = fetch_line(ctx, SERF_NEWLINE_ANY);
164 if (SERF_BUCKET_READ_ERROR(status)) {
165 return status;
166 }
167 /* Something was read. Process it. */
168
169 if (ctx->linebuf.state == SERF_LINEBUF_READY && ctx->linebuf.used) {
170 const char *end_key;
171 const char *c;
172
173 end_key = c = memchr(ctx->linebuf.line, ':', ctx->linebuf.used);
174 if (!c) {
175 /* Bad headers? */
176 return SERF_ERROR_BAD_HTTP_RESPONSE;
177 }
178
179 /* Skip over initial ':' */
180 c++;
181
182 /* And skip all whitespaces. */
183 for(; c < ctx->linebuf.line + ctx->linebuf.used; c++)
184 {
185 if (!apr_isspace(*c))
186 {
187 break;
188 }
189 }
190
191 /* Always copy the headers (from the linebuf into new mem). */
192 /* ### we should be able to optimize some mem copies */
193 serf_bucket_headers_setx(
194 ctx->headers,
195 ctx->linebuf.line, end_key - ctx->linebuf.line, 1,
196 c, ctx->linebuf.line + ctx->linebuf.used - c, 1);
197 }
198
199 return status;
200 }
201
202 /* Perform one iteration of the state machine.
203 *
204 * Will return when one the following conditions occurred:
205 * 1) a state change
206 * 2) an error
207 * 3) the stream is not ready or at EOF
208 * 4) APR_SUCCESS, meaning the machine can be run again immediately
209 */
run_machine(serf_bucket_t * bkt,response_context_t * ctx)210 static apr_status_t run_machine(serf_bucket_t *bkt, response_context_t *ctx)
211 {
212 apr_status_t status = APR_SUCCESS; /* initialize to avoid gcc warnings */
213
214 switch (ctx->state) {
215 case STATE_STATUS_LINE:
216 /* RFC 2616 says that CRLF is the only line ending, but we can easily
217 * accept any kind of line ending.
218 */
219 status = fetch_line(ctx, SERF_NEWLINE_ANY);
220 if (SERF_BUCKET_READ_ERROR(status))
221 return status;
222
223 if (ctx->linebuf.state == SERF_LINEBUF_READY) {
224 /* The Status-Line is in the line buffer. Process it. */
225 status = parse_status_line(ctx, bkt->allocator);
226 if (status)
227 return status;
228
229 /* Good times ahead: we're switching protocols! */
230 if (ctx->sl.code == 101) {
231 ctx->body =
232 serf_bucket_barrier_create(ctx->stream, bkt->allocator);
233 ctx->state = STATE_DONE;
234 break;
235 }
236
237 /* Okay... move on to reading the headers. */
238 ctx->state = STATE_HEADERS;
239 }
240 else {
241 /* The connection closed before we could get the next
242 * response. Treat the request as lost so that our upper
243 * end knows the server never tried to give us a response.
244 */
245 if (APR_STATUS_IS_EOF(status)) {
246 return SERF_ERROR_REQUEST_LOST;
247 }
248 }
249 break;
250 case STATE_HEADERS:
251 status = fetch_headers(bkt, ctx);
252 if (SERF_BUCKET_READ_ERROR(status))
253 return status;
254
255 /* If an empty line was read, then we hit the end of the headers.
256 * Move on to the body.
257 */
258 if (ctx->linebuf.state == SERF_LINEBUF_READY && !ctx->linebuf.used) {
259 const void *v;
260
261 /* Advance the state. */
262 ctx->state = STATE_BODY;
263
264 /* If this is a response to a HEAD request, or code == 1xx,204 or304
265 then we don't receive a real body. */
266 if (!expect_body(ctx)) {
267 ctx->body = serf_bucket_simple_create(NULL, 0, NULL, NULL,
268 bkt->allocator);
269 ctx->state = STATE_BODY;
270 break;
271 }
272
273 ctx->body =
274 serf_bucket_barrier_create(ctx->stream, bkt->allocator);
275
276 /* Are we C-L, chunked, or conn close? */
277 v = serf_bucket_headers_get(ctx->headers, "Content-Length");
278 if (v) {
279 apr_uint64_t length;
280 length = apr_strtoi64(v, NULL, 10);
281 if (errno == ERANGE) {
282 return APR_FROM_OS_ERROR(ERANGE);
283 }
284 ctx->body = serf_bucket_response_body_create(
285 ctx->body, length, bkt->allocator);
286 }
287 else {
288 v = serf_bucket_headers_get(ctx->headers, "Transfer-Encoding");
289
290 /* Need to handle multiple transfer-encoding. */
291 if (v && strcasecmp("chunked", v) == 0) {
292 ctx->chunked = 1;
293 ctx->body = serf_bucket_dechunk_create(ctx->body,
294 bkt->allocator);
295 }
296 }
297 v = serf_bucket_headers_get(ctx->headers, "Content-Encoding");
298 if (v) {
299 /* Need to handle multiple content-encoding. */
300 if (v && strcasecmp("gzip", v) == 0) {
301 ctx->body =
302 serf_bucket_deflate_create(ctx->body, bkt->allocator,
303 SERF_DEFLATE_GZIP);
304 }
305 else if (v && strcasecmp("deflate", v) == 0) {
306 ctx->body =
307 serf_bucket_deflate_create(ctx->body, bkt->allocator,
308 SERF_DEFLATE_DEFLATE);
309 }
310 }
311 }
312 break;
313 case STATE_BODY:
314 /* Don't do anything. */
315 break;
316 case STATE_TRAILERS:
317 status = fetch_headers(bkt, ctx);
318 if (SERF_BUCKET_READ_ERROR(status))
319 return status;
320
321 /* If an empty line was read, then we're done. */
322 if (ctx->linebuf.state == SERF_LINEBUF_READY && !ctx->linebuf.used) {
323 ctx->state = STATE_DONE;
324 return APR_EOF;
325 }
326 break;
327 case STATE_DONE:
328 return APR_EOF;
329 default:
330 /* Not reachable */
331 return APR_EGENERAL;
332 }
333
334 return status;
335 }
336
wait_for_body(serf_bucket_t * bkt,response_context_t * ctx)337 static apr_status_t wait_for_body(serf_bucket_t *bkt, response_context_t *ctx)
338 {
339 apr_status_t status;
340
341 /* Keep reading and moving through states if we aren't at the BODY */
342 while (ctx->state != STATE_BODY) {
343 status = run_machine(bkt, ctx);
344
345 /* Anything other than APR_SUCCESS means that we cannot immediately
346 * read again (for now).
347 */
348 if (status)
349 return status;
350 }
351 /* in STATE_BODY */
352
353 return APR_SUCCESS;
354 }
355
serf_bucket_response_wait_for_headers(serf_bucket_t * bucket)356 apr_status_t serf_bucket_response_wait_for_headers(
357 serf_bucket_t *bucket)
358 {
359 response_context_t *ctx = bucket->data;
360
361 return wait_for_body(bucket, ctx);
362 }
363
serf_bucket_response_status(serf_bucket_t * bkt,serf_status_line * sline)364 apr_status_t serf_bucket_response_status(
365 serf_bucket_t *bkt,
366 serf_status_line *sline)
367 {
368 response_context_t *ctx = bkt->data;
369 apr_status_t status;
370
371 if (ctx->state != STATE_STATUS_LINE) {
372 /* We already read it and moved on. Just return it. */
373 *sline = ctx->sl;
374 return APR_SUCCESS;
375 }
376
377 /* Running the state machine once will advance the machine, or state
378 * that the stream isn't ready with enough data. There isn't ever a
379 * need to run the machine more than once to try and satisfy this. We
380 * have to look at the state to tell whether it advanced, though, as
381 * it is quite possible to advance *and* to return APR_EAGAIN.
382 */
383 status = run_machine(bkt, ctx);
384 if (ctx->state == STATE_HEADERS) {
385 *sline = ctx->sl;
386 }
387 else {
388 /* Indicate that we don't have the information yet. */
389 sline->version = 0;
390 }
391
392 return status;
393 }
394
serf_response_read(serf_bucket_t * bucket,apr_size_t requested,const char ** data,apr_size_t * len)395 static apr_status_t serf_response_read(serf_bucket_t *bucket,
396 apr_size_t requested,
397 const char **data, apr_size_t *len)
398 {
399 response_context_t *ctx = bucket->data;
400 apr_status_t rv;
401
402 rv = wait_for_body(bucket, ctx);
403 if (rv) {
404 /* It's not possible to have read anything yet! */
405 if (APR_STATUS_IS_EOF(rv) || APR_STATUS_IS_EAGAIN(rv)) {
406 *len = 0;
407 }
408 return rv;
409 }
410
411 rv = serf_bucket_read(ctx->body, requested, data, len);
412 if (SERF_BUCKET_READ_ERROR(rv))
413 return rv;
414
415 if (APR_STATUS_IS_EOF(rv)) {
416 if (ctx->chunked) {
417 ctx->state = STATE_TRAILERS;
418 /* Mask the result. */
419 rv = APR_SUCCESS;
420 } else {
421 ctx->state = STATE_DONE;
422 }
423 }
424 return rv;
425 }
426
serf_response_readline(serf_bucket_t * bucket,int acceptable,int * found,const char ** data,apr_size_t * len)427 static apr_status_t serf_response_readline(serf_bucket_t *bucket,
428 int acceptable, int *found,
429 const char **data, apr_size_t *len)
430 {
431 response_context_t *ctx = bucket->data;
432 apr_status_t rv;
433
434 rv = wait_for_body(bucket, ctx);
435 if (rv) {
436 return rv;
437 }
438
439 /* Delegate to the stream bucket to do the readline. */
440 return serf_bucket_readline(ctx->body, acceptable, found, data, len);
441 }
442
serf_response_full_become_aggregate(serf_bucket_t * bucket)443 apr_status_t serf_response_full_become_aggregate(serf_bucket_t *bucket)
444 {
445 response_context_t *ctx = bucket->data;
446 serf_bucket_t *bkt;
447 char buf[256];
448 int size;
449
450 serf_bucket_aggregate_become(bucket);
451
452 /* Add reconstructed status line. */
453 size = apr_snprintf(buf, 256, "HTTP/%d.%d %d ",
454 SERF_HTTP_VERSION_MAJOR(ctx->sl.version),
455 SERF_HTTP_VERSION_MINOR(ctx->sl.version),
456 ctx->sl.code);
457 bkt = serf_bucket_simple_copy_create(buf, size,
458 bucket->allocator);
459 serf_bucket_aggregate_append(bucket, bkt);
460 bkt = serf_bucket_simple_copy_create(ctx->sl.reason, strlen(ctx->sl.reason),
461 bucket->allocator);
462 serf_bucket_aggregate_append(bucket, bkt);
463 bkt = SERF_BUCKET_SIMPLE_STRING_LEN("\r\n", 2,
464 bucket->allocator);
465 serf_bucket_aggregate_append(bucket, bkt);
466
467 /* Add headers and stream buckets in order. */
468 serf_bucket_aggregate_append(bucket, ctx->headers);
469 serf_bucket_aggregate_append(bucket, ctx->stream);
470
471 serf_bucket_mem_free(bucket->allocator, ctx);
472
473 return APR_SUCCESS;
474 }
475
476 /* ### need to implement */
477 #define serf_response_peek NULL
478
479 const serf_bucket_type_t serf_bucket_type_response = {
480 "RESPONSE",
481 serf_response_read,
482 serf_response_readline,
483 serf_default_read_iovec,
484 serf_default_read_for_sendfile,
485 serf_default_read_bucket,
486 serf_response_peek,
487 serf_response_destroy_and_data,
488 };
489