static void * APR_THREAD_FUNC parser_thread(apr_thread_t *thread, void *data)
{
apr_status_t status;
apr_pool_t *pool, *subpool;
parser_baton_t *ctx;
ctx = (parser_baton_t*)data;
pool = apr_thread_pool_get(thread);
apr_pool_create(&subpool, pool);
while (1) {
doc_path_t *dup;
apr_pool_clear(subpool);
/* Grab it. */
apr_thread_mutex_lock(ctx->mutex);
/* Sleep. */
apr_thread_cond_wait(ctx->condvar, ctx->mutex);
/* Fetch the doc off the list. */
if (ctx->doc_queue->nelts) {
dup = *(doc_path_t**)(apr_array_pop(ctx->doc_queue));
/* dup = (ctx->doc_queue->conns->elts)[0]; */
}
else {
dup = NULL;
}
/* Don't need the mutex now. */
apr_thread_mutex_unlock(ctx->mutex);
/* Parse the doc/url pair. */
if (dup) {
status = find_href_doc(dup->doc, dup->path, ctx, subpool);
if (status) {
printf("Error finding hrefs: %d %s\n", status, dup->path);
}
/* Free the doc pair and its pool. */
apr_pool_destroy(dup->pool);
serf_bucket_mem_free(ctx->doc_queue_alloc, dup->path);
serf_bucket_mem_free(ctx->doc_queue_alloc, dup);
}
/* Hey are we done? */
if (!apr_atomic_read32(ctx->requests_outstanding)) {
break;
}
}
return NULL;
}
static void serf_iovec_destroy(serf_bucket_t *bucket)
{
iovec_context_t *ctx = bucket->data;
serf_bucket_mem_free(bucket->allocator, ctx->vecs);
serf_default_destroy_and_data(bucket);
}
static void serf_ssl_encrypt_destroy_and_data(serf_bucket_t *bucket)
{
ssl_context_t *ctx = bucket->data;
serf_ssl_context_t *ssl_ctx = ctx->ssl_ctx;
if (ssl_ctx->encrypt.stream == *ctx->our_stream) {
serf_bucket_destroy(*ctx->our_stream);
serf_bucket_destroy(ssl_ctx->encrypt.pending);
/* Reset our encrypted status and databuf. */
ssl_ctx->encrypt.status = APR_SUCCESS;
ssl_ctx->encrypt.databuf.status = APR_SUCCESS;
/* Advance to the next stream - if we have one. */
if (ssl_ctx->encrypt.stream_next == NULL) {
ssl_ctx->encrypt.stream = NULL;
ssl_ctx->encrypt.pending = NULL;
}
else {
bucket_list_t *cur;
cur = ssl_ctx->encrypt.stream_next;
ssl_ctx->encrypt.stream = cur->bucket;
ssl_ctx->encrypt.pending =
serf_bucket_aggregate_create(cur->bucket->allocator);
ssl_ctx->encrypt.stream_next = cur->next;
serf_bucket_mem_free(ssl_ctx->allocator, cur);
}
}
else {
/* Ah, darn. We haven't sent this one along yet. */
return;
}
serf_ssl_destroy_and_data(bucket);
}
void serf_default_destroy(serf_bucket_t *bucket)
{
#ifdef SERF_DEBUG_BUCKET_USE
serf_debug__bucket_destroy(bucket);
#endif
serf_bucket_mem_free(bucket->allocator, bucket);
}
static void serf_headers_destroy_and_data(serf_bucket_t *bucket)
{
headers_context_t *ctx = bucket->data;
header_list_t *scan = ctx->list;
while (scan) {
header_list_t *next_hdr = scan->next;
if (scan->alloc_flags & ALLOC_HEADER)
serf_bucket_mem_free(bucket->allocator, (void *)scan->header);
if (scan->alloc_flags & ALLOC_VALUE)
serf_bucket_mem_free(bucket->allocator, (void *)scan->value);
serf_bucket_mem_free(bucket->allocator, scan);
scan = next_hdr;
}
serf_default_destroy_and_data(bucket);
}
static void serf_response_destroy_and_data(serf_bucket_t *bucket)
{
response_context_t *ctx = bucket->data;
if (ctx->state != STATE_STATUS_LINE) {
serf_bucket_mem_free(bucket->allocator, (void*)ctx->sl.reason);
}
serf_bucket_destroy(ctx->stream);
if (ctx->body != NULL)
serf_bucket_destroy(ctx->body);
serf_bucket_destroy(ctx->headers);
serf_default_destroy_and_data(bucket);
}
static void serf_aggregate_destroy_and_data(serf_bucket_t *bucket)
{
aggregate_context_t *ctx = bucket->data;
bucket_list_t *next_ctx;
while (ctx->list) {
if (ctx->bucket_owner) {
serf_bucket_destroy(ctx->list->bucket);
}
next_ctx = ctx->list->next;
serf_bucket_mem_free(bucket->allocator, ctx->list);
ctx->list = next_ctx;
}
cleanup_aggregate(ctx, bucket->allocator);
serf_default_destroy_and_data(bucket);
}
const char *serf_bucket_headers_get(
serf_bucket_t *headers_bucket,
const char *header)
{
headers_context_t *ctx = headers_bucket->data;
header_list_t *found = ctx->list;
const char *val = NULL;
int value_size = 0;
int val_alloc = 0;
while (found) {
if (strcasecmp(found->header, header) == 0) {
if (val) {
/* The header is already present. RFC 2616, section 4.2
indicates that we should append the new value, separated by
a comma. Reasoning: for headers whose values are known to
be comma-separated, that is clearly the correct behavior;
for others, the correct behavior is undefined anyway. */
/* The "+1" is for the comma; serf_bstrmemdup() will also add
one slot for the terminating '\0'. */
apr_size_t new_size = found->value_size + value_size + 1;
char *new_val = serf_bucket_mem_alloc(headers_bucket->allocator,
new_size);
memcpy(new_val, val, value_size);
new_val[value_size] = ',';
memcpy(new_val + value_size + 1, found->value,
found->value_size);
new_val[new_size] = '\0';
/* Copy the new value over the already existing value. */
if (val_alloc)
serf_bucket_mem_free(headers_bucket->allocator, (void*)val);
val_alloc |= ALLOC_VALUE;
val = new_val;
value_size = new_size;
}
else {
val = found->value;
value_size = found->value_size;
}
}
found = found->next;
}
return val;
}
static void cleanup_aggregate(aggregate_context_t *ctx,
serf_bucket_alloc_t *allocator)
{
bucket_list_t *next_list;
/* If we finished reading a bucket during the previous read, then
* we can now toss that bucket.
*/
while (ctx->done != NULL) {
next_list = ctx->done->next;
if (ctx->bucket_owner) {
serf_bucket_destroy(ctx->done->bucket);
}
serf_bucket_mem_free(allocator, ctx->done);
ctx->done = next_list;
}
}
static apr_status_t ssl_free_context(
serf_ssl_context_t *ssl_ctx)
{
apr_pool_t *p;
/* If never had the pending buckets, don't try to free them. */
if (ssl_ctx->decrypt.pending != NULL) {
serf_bucket_destroy(ssl_ctx->decrypt.pending);
}
if (ssl_ctx->encrypt.pending != NULL) {
serf_bucket_destroy(ssl_ctx->encrypt.pending);
}
/* SSL_free implicitly frees the underlying BIO. */
SSL_free(ssl_ctx->ssl);
SSL_CTX_free(ssl_ctx->ctx);
p = ssl_ctx->pool;
serf_bucket_mem_free(ssl_ctx->allocator, ssl_ctx);
apr_pool_destroy(p);
return APR_SUCCESS;
}
int main(int argc, const char **argv)
{
apr_status_t status;
apr_pool_t *pool;
apr_sockaddr_t *address;
serf_context_t *context;
serf_connection_t *connection;
app_baton_t app_ctx;
handler_baton_t *handler_ctx;
apr_uri_t url;
const char *raw_url, *method;
int count;
apr_getopt_t *opt;
char opt_c;
char *authn = NULL;
const char *opt_arg;
/* For the parser threads */
apr_thread_t *thread[3];
apr_threadattr_t *tattr;
apr_status_t parser_status;
parser_baton_t *parser_ctx;
apr_initialize();
atexit(apr_terminate);
apr_pool_create(&pool, NULL);
apr_atomic_init(pool);
/* serf_initialize(); */
/* Default to one round of fetching. */
count = 1;
/* Default to GET. */
method = "GET";
apr_getopt_init(&opt, pool, argc, argv);
while ((status = apr_getopt(opt, "a:hv", &opt_c, &opt_arg)) ==
APR_SUCCESS) {
int srclen, enclen;
switch (opt_c) {
case 'a':
srclen = strlen(opt_arg);
enclen = apr_base64_encode_len(srclen);
authn = apr_palloc(pool, enclen + 6);
strcpy(authn, "Basic ");
(void) apr_base64_encode(&authn[6], opt_arg, srclen);
break;
case 'h':
print_usage(pool);
exit(0);
break;
case 'v':
puts("Serf version: " SERF_VERSION_STRING);
exit(0);
default:
break;
}
}
if (opt->ind != opt->argc - 1) {
print_usage(pool);
exit(-1);
}
raw_url = argv[opt->ind];
apr_uri_parse(pool, raw_url, &url);
if (!url.port) {
url.port = apr_uri_port_of_scheme(url.scheme);
}
if (!url.path) {
url.path = "/";
}
if (strcasecmp(url.scheme, "https") == 0) {
app_ctx.using_ssl = 1;
}
else {
app_ctx.using_ssl = 0;
}
status = apr_sockaddr_info_get(&address,
url.hostname, APR_UNSPEC, url.port, 0,
pool);
if (status) {
printf("Error creating address: %d\n", status);
exit(1);
}
context = serf_context_create(pool);
/* ### Connection or Context should have an allocator? */
app_ctx.bkt_alloc = serf_bucket_allocator_create(pool, NULL, NULL);
app_ctx.ssl_ctx = NULL;
app_ctx.authn = authn;
connection = serf_connection_create(context, address,
conn_setup, &app_ctx,
closed_connection, &app_ctx,
pool);
handler_ctx = (handler_baton_t*)serf_bucket_mem_alloc(app_ctx.bkt_alloc,
sizeof(handler_baton_t));
handler_ctx->allocator = app_ctx.bkt_alloc;
handler_ctx->doc_queue = apr_array_make(pool, 1, sizeof(doc_path_t*));
handler_ctx->doc_queue_alloc = app_ctx.bkt_alloc;
handler_ctx->requests_outstanding =
(apr_uint32_t*)serf_bucket_mem_alloc(app_ctx.bkt_alloc,
sizeof(apr_uint32_t));
apr_atomic_set32(handler_ctx->requests_outstanding, 0);
handler_ctx->hdr_read = 0;
parser_ctx = (void*)serf_bucket_mem_alloc(app_ctx.bkt_alloc,
sizeof(parser_baton_t));
parser_ctx->requests_outstanding = handler_ctx->requests_outstanding;
parser_ctx->connection = connection;
parser_ctx->app_ctx = &app_ctx;
parser_ctx->doc_queue = handler_ctx->doc_queue;
parser_ctx->doc_queue_alloc = handler_ctx->doc_queue_alloc;
/* Restrict ourselves to this host. */
parser_ctx->hostinfo = url.hostinfo;
status = apr_thread_mutex_create(&parser_ctx->mutex,
APR_THREAD_MUTEX_DEFAULT, pool);
if (status) {
printf("Couldn't create mutex %d\n", status);
return status;
}
status = apr_thread_cond_create(&parser_ctx->condvar, pool);
if (status) {
printf("Couldn't create condvar: %d\n", status);
return status;
}
/* Let the handler now which condvar to use. */
handler_ctx->doc_queue_condvar = parser_ctx->condvar;
apr_threadattr_create(&tattr, pool);
/* Start the parser thread. */
apr_thread_create(&thread[0], tattr, parser_thread, parser_ctx, pool);
/* Deliver the first request. */
create_request(url.hostinfo, url.path, NULL, NULL, parser_ctx, pool);
/* Go run our normal thread. */
while (1) {
int tries = 0;
status = serf_context_run(context, SERF_DURATION_FOREVER, pool);
if (APR_STATUS_IS_TIMEUP(status))
continue;
if (status) {
char buf[200];
printf("Error running context: (%d) %s\n", status,
apr_strerror(status, buf, sizeof(buf)));
exit(1);
}
/* We run this check to allow our parser threads to add more
* requests to our queue.
*/
for (tries = 0; tries < 3; tries++) {
if (!apr_atomic_read32(handler_ctx->requests_outstanding)) {
#ifdef SERF_VERBOSE
printf("Waiting...");
#endif
apr_sleep(100000);
#ifdef SERF_VERBOSE
printf("Done\n");
#endif
}
else {
break;
}
}
if (tries >= 3) {
break;
}
/* Debugging purposes only! */
serf_debug__closed_conn(app_ctx.bkt_alloc);
}
printf("Quitting...\n");
serf_connection_close(connection);
/* wake up the parser via condvar signal */
apr_thread_cond_signal(parser_ctx->condvar);
status = apr_thread_join(&parser_status, thread[0]);
if (status) {
printf("Error joining thread: %d\n", status);
return status;
}
serf_bucket_mem_free(app_ctx.bkt_alloc, handler_ctx->requests_outstanding);
serf_bucket_mem_free(app_ctx.bkt_alloc, parser_ctx);
apr_pool_destroy(pool);
return 0;
}
static apr_status_t handle_response(serf_request_t *request,
serf_bucket_t *response,
void *handler_baton,
apr_pool_t *pool)
{
const char *data;
apr_size_t len;
serf_status_line sl;
apr_status_t status;
handler_baton_t *ctx = handler_baton;
if (!response) {
/* Oh no! We've been cancelled! */
abort();
}
status = serf_bucket_response_status(response, &sl);
if (status) {
if (APR_STATUS_IS_EAGAIN(status)) {
return APR_SUCCESS;
}
abort();
}
while (1) {
status = serf_bucket_read(response, 2048, &data, &len);
if (SERF_BUCKET_READ_ERROR(status))
return status;
/*fwrite(data, 1, len, stdout);*/
if (!ctx->hdr_read) {
serf_bucket_t *hdrs;
const char *val;
printf("Processing %s\n", ctx->path);
hdrs = serf_bucket_response_get_headers(response);
val = serf_bucket_headers_get(hdrs, "Content-Type");
/* FIXME: This check isn't quite right because Content-Type could
* be decorated; ideally strcasestr would be correct.
*/
if (val && strcasecmp(val, "text/html") == 0) {
ctx->is_html = 1;
apr_pool_create(&ctx->parser_pool, NULL);
ctx->parser = apr_xml_parser_create(ctx->parser_pool);
}
else {
ctx->is_html = 0;
}
ctx->hdr_read = 1;
}
if (ctx->is_html) {
apr_status_t xs;
xs = apr_xml_parser_feed(ctx->parser, data, len);
/* Uh-oh. */
if (xs) {
#ifdef SERF_VERBOSE
printf("XML parser error (feed): %d\n", xs);
#endif
ctx->is_html = 0;
}
}
/* are we done yet? */
if (APR_STATUS_IS_EOF(status)) {
if (ctx->is_html) {
apr_xml_doc *xmld;
apr_status_t xs;
doc_path_t *dup;
xs = apr_xml_parser_done(ctx->parser, &xmld);
if (xs) {
#ifdef SERF_VERBOSE
printf("XML parser error (done): %d\n", xs);
#endif
return xs;
}
dup = (doc_path_t*)
serf_bucket_mem_alloc(ctx->doc_queue_alloc,
sizeof(doc_path_t));
dup->doc = xmld;
dup->path = (char*)serf_bucket_mem_alloc(ctx->doc_queue_alloc,
ctx->path_len);
memcpy(dup->path, ctx->path, ctx->path_len);
dup->pool = ctx->parser_pool;
*(doc_path_t **)apr_array_push(ctx->doc_queue) = dup;
apr_thread_cond_signal(ctx->doc_queue_condvar);
}
apr_atomic_dec32(ctx->requests_outstanding);
serf_bucket_mem_free(ctx->allocator, ctx->path);
if (ctx->query) {
serf_bucket_mem_free(ctx->allocator, ctx->query);
serf_bucket_mem_free(ctx->allocator, ctx->full_path);
}
if (ctx->fragment) {
serf_bucket_mem_free(ctx->allocator, ctx->fragment);
}
serf_bucket_mem_free(ctx->allocator, ctx);
return APR_EOF;
}
/* have we drained the response so far? */
if (APR_STATUS_IS_EAGAIN(status))
return APR_SUCCESS;
/* loop to read some more. */
}
/* NOTREACHED */
}
void serf_default_destroy_and_data(serf_bucket_t *bucket)
{
serf_bucket_mem_free(bucket->allocator, bucket->data);
serf_default_destroy(bucket);
}
/* This function reads a decrypted stream and returns an encrypted stream. */
static apr_status_t ssl_encrypt(void *baton, apr_size_t bufsize,
char *buf, apr_size_t *len)
{
const char *data;
apr_size_t interim_bufsize;
serf_ssl_context_t *ctx = baton;
apr_status_t status;
#ifdef SSL_VERBOSE
printf("ssl_encrypt: begin %d\n", bufsize);
#endif
/* Try to read already encrypted but unread data first. */
status = serf_bucket_read(ctx->encrypt.pending, bufsize, &data, len);
if (SERF_BUCKET_READ_ERROR(status)) {
return status;
}
/* Aha, we read something. Return that now. */
if (*len) {
memcpy(buf, data, *len);
if (APR_STATUS_IS_EOF(status)) {
status = APR_SUCCESS;
}
#ifdef SSL_VERBOSE
printf("ssl_encrypt: %d %d %d (quick read)\n", status, *len,
BIO_get_retry_flags(ctx->bio));
#endif
return status;
}
if (BIO_should_retry(ctx->bio) && BIO_should_write(ctx->bio)) {
#ifdef SSL_VERBOSE
printf("ssl_encrypt: %d %d %d (should write exit)\n", status, *len,
BIO_get_retry_flags(ctx->bio));
#endif
return APR_EAGAIN;
}
/* If we were previously blocked, unblock ourselves now. */
if (BIO_should_read(ctx->bio)) {
#ifdef SSL_VERBOSE
printf("ssl_encrypt: reset %d %d (%d %d %d)\n", status,
ctx->encrypt.status,
BIO_should_retry(ctx->bio), BIO_should_read(ctx->bio),
BIO_get_retry_flags(ctx->bio));
#endif
ctx->encrypt.status = APR_SUCCESS;
ctx->encrypt.exhausted_reset = 0;
}
/* Oh well, read from our stream now. */
interim_bufsize = bufsize;
do {
apr_size_t interim_len;
if (!ctx->encrypt.status) {
struct iovec vecs[64];
int vecs_read;
status = serf_bucket_read_iovec(ctx->encrypt.stream,
interim_bufsize, 64, vecs,
&vecs_read);
if (!SERF_BUCKET_READ_ERROR(status) && vecs_read) {
char *vecs_data;
int i, cur, vecs_data_len;
int ssl_len;
/* Combine the buffers of the iovec into one buffer, as
that is with SSL_write requires. */
vecs_data_len = 0;
for (i = 0; i < vecs_read; i++) {
vecs_data_len += vecs[i].iov_len;
}
vecs_data = serf_bucket_mem_alloc(ctx->allocator,
vecs_data_len);
cur = 0;
for (i = 0; i < vecs_read; i++) {
memcpy(vecs_data + cur, vecs[i].iov_base, vecs[i].iov_len);
cur += vecs[i].iov_len;
}
interim_bufsize -= vecs_data_len;
interim_len = vecs_data_len;
#ifdef SSL_VERBOSE
printf("ssl_encrypt: bucket read %d bytes; status %d\n",
interim_len, status);
printf("---\n%s\n-(%d)-\n", vecs_data, interim_len);
#endif
/* Stash our status away. */
ctx->encrypt.status = status;
ssl_len = SSL_write(ctx->ssl, vecs_data, interim_len);
#ifdef SSL_VERBOSE
printf("ssl_encrypt: SSL write: %d\n", ssl_len);
#endif
/* We're done. */
serf_bucket_mem_free(ctx->allocator, vecs_data);
/* If we failed to write... */
if (ssl_len < 0) {
int ssl_err;
/* Ah, bugger. We need to put that data back. */
serf_bucket_aggregate_prepend_iovec(ctx->encrypt.stream,
vecs, vecs_read);
ssl_err = SSL_get_error(ctx->ssl, ssl_len);
#ifdef SSL_VERBOSE
printf("ssl_encrypt: SSL write error: %d\n", ssl_err);
#endif
if (ssl_err == SSL_ERROR_SYSCALL) {
status = ctx->encrypt.status;
if (SERF_BUCKET_READ_ERROR(status)) {
return status;
}
}
else {
/* Oh, no. */
if (ssl_err == SSL_ERROR_WANT_READ) {
status = SERF_ERROR_WAIT_CONN;
}
else {
status = APR_EGENERAL;
}
}
#ifdef SSL_VERBOSE
printf("ssl_encrypt: SSL write error: %d %d\n", status, *len);
#endif
}
}
}
else {
interim_len = 0;
*len = 0;
status = ctx->encrypt.status;
}
} while (!status && interim_bufsize);
/* Okay, we exhausted our underlying stream. */
if (!SERF_BUCKET_READ_ERROR(status)) {
apr_status_t agg_status;
struct iovec vecs[64];
int vecs_read, i;
/* We read something! */
agg_status = serf_bucket_read_iovec(ctx->encrypt.pending, bufsize,
64, vecs, &vecs_read);
*len = 0;
for (i = 0; i < vecs_read; i++) {
memcpy(buf + *len, vecs[i].iov_base, vecs[i].iov_len);
*len += vecs[i].iov_len;
}
#ifdef SSL_VERBOSE
printf("ssl_encrypt read agg: %d %d %d %d\n", status, agg_status,
ctx->encrypt.status, *len);
#endif
if (!agg_status) {
status = agg_status;
}
}
if (status == SERF_ERROR_WAIT_CONN
&& BIO_should_retry(ctx->bio) && BIO_should_read(ctx->bio)) {
ctx->encrypt.exhausted = ctx->encrypt.status;
ctx->encrypt.status = SERF_ERROR_WAIT_CONN;
}
#ifdef SSL_VERBOSE
printf("ssl_encrypt finished: %d %d (%d %d %d)\n", status, *len,
BIO_should_retry(ctx->bio), BIO_should_read(ctx->bio),
BIO_get_retry_flags(ctx->bio));
#endif
return status;
}
static void free_copied_data(void *baton, const char *data)
{
serf_bucket_mem_free(baton, (char*)data);
}