summaryrefslogtreecommitdiffstats
path: root/axiom/src/attachments/mime_parser.c
diff options
context:
space:
mode:
Diffstat (limited to 'axiom/src/attachments/mime_parser.c')
-rw-r--r--axiom/src/attachments/mime_parser.c2304
1 files changed, 2304 insertions, 0 deletions
diff --git a/axiom/src/attachments/mime_parser.c b/axiom/src/attachments/mime_parser.c
new file mode 100644
index 0000000..f36286c
--- /dev/null
+++ b/axiom/src/attachments/mime_parser.c
@@ -0,0 +1,2304 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <axiom_mime_parser.h>
+#include <axutil_string.h>
+#include <axiom_data_handler.h>
+#include <stdio.h>
+#include <ctype.h>
+#include <axutil_http_chunked_stream.h>
+#include <axiom_mtom_caching_callback.h>
+#include <axutil_class_loader.h>
+#include <axutil_url.h>
+
+struct axiom_mime_parser
+{
+ /* This will keep the attachment and its info*/
+ axutil_hash_t *mime_parts_map;
+
+ /* This is the actual SOAP part len */
+ size_t soap_body_len;
+
+ /* The SOAP part of the message */
+ axis2_char_t *soap_body_str;
+
+ /* The size of the buffer we give to the callback to
+ * read data */
+ size_t buffer_size;
+
+ /* The number of buffers */
+ int max_buffers;
+
+ /* The attachment dir name, in the case of caching */
+ axis2_char_t *attachment_dir;
+
+ /*A pointer to the caching callback */
+ axiom_mtom_caching_callback_t *mtom_caching_callback;
+
+ /* The caching callback name specified */
+ axis2_char_t *callback_name;
+
+ axis2_char_t **buf_array;
+
+ size_t *len_array;
+
+ int current_buf_num;
+
+ axis2_bool_t end_of_mime;
+
+ axis2_char_t *mime_boundary;
+
+};
+
+struct axiom_search_info
+{
+ /*String need to be searched*/
+ const axis2_char_t *search_str;
+
+ /*The buffers and the lengths need to be searched*/
+ axis2_char_t *buffer1;
+ size_t len1;
+ axis2_char_t *buffer2;
+ size_t len2;
+
+ /*Flag to keep what type of search is this buffer has done*/
+ axis2_bool_t primary_search;
+
+ /*The offset where we found the pattern entirely in one buffer*/
+ size_t match_len1;
+
+ /*when pattern contains in two buffers the length of partial pattern
+ in buffer2 */
+ size_t match_len2;
+
+ /*Whether we need caching or not*/
+ axis2_bool_t cached;
+
+ /*A pointer to a user provided storage to which we cache the attachment*/
+ void *handler;
+
+ /* Size of the binary when writing to the buffer*/
+ size_t binary_size;
+};
+
+typedef struct axiom_search_info axiom_search_info_t;
+
+#define AXIOM_MIME_PARSER_CONTENT_ID "content-id"
+#define AXIOM_MIME_PARSER_CONTENT_TYPE "content-type"
+
+#define AXIOM_MIME_PARSER_END_OF_MIME_MAX_COUNT 100
+
+static axis2_char_t *
+axiom_mime_parser_search_for_soap(
+ const axutil_env_t * env,
+ AXIS2_READ_INPUT_CALLBACK callback,
+ void *callback_ctx,
+ int *buf_num,
+ size_t *len_array,
+ axis2_char_t **buf_array,
+ axiom_search_info_t *search_info,
+ size_t size,
+ axis2_char_t *mime_boundary,
+ axiom_mime_parser_t *mime_parser);
+
+static axis2_char_t *
+axiom_mime_parser_search_for_crlf(
+ const axutil_env_t * env,
+ AXIS2_READ_INPUT_CALLBACK callback,
+ void *callback_ctx,
+ int *buf_num,
+ size_t *len_array,
+ axis2_char_t **buf_array,
+ axiom_search_info_t *search_info,
+ size_t size,
+ axiom_mime_parser_t *mime_parser);
+
+static size_t
+axiom_mime_parser_calculate_part_len(
+ const axutil_env_t *env,
+ int buf_num,
+ size_t *len_list,
+ int maker,
+ axis2_char_t *pos,
+ axis2_char_t *buf);
+
+static axis2_char_t *
+axiom_mime_parser_create_part(
+ const axutil_env_t *env,
+ size_t part_len,
+ int buf_num,
+ size_t *len_list,
+ int marker,
+ axis2_char_t *pos,
+ axis2_char_t **buf_list,
+ axiom_mime_parser_t *mime_parser);
+
+static axis2_char_t *
+axiom_mime_parser_search_string(
+ axiom_search_info_t *search_info,
+ const axutil_env_t *env);
+
+static axis2_char_t *
+axiom_mime_parser_search_for_attachment(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t * env,
+ AXIS2_READ_INPUT_CALLBACK callback,
+ void *callback_ctx,
+ int *buf_num,
+ size_t *len_array,
+ axis2_char_t **buf_array,
+ axiom_search_info_t *search_info,
+ size_t size,
+ axis2_char_t *mime_boundary,
+ axis2_char_t *mime_id,
+ void *user_param);
+
+static axis2_status_t
+axiom_mime_parser_store_attachment(
+ const axutil_env_t *env,
+ axiom_mime_parser_t *mime_parser,
+ axis2_char_t *mime_id,
+ axis2_char_t *mime_type,
+ axis2_char_t *mime_binary,
+ size_t mime_binary_len,
+ axis2_bool_t cached);
+
+static void
+axiom_mime_parser_clear_buffers(
+ const axutil_env_t *env,
+ axis2_char_t **buf_list,
+ int free_from,
+ int free_to);
+
+static axis2_status_t
+axiom_mime_parser_cache_to_buffer(
+ const axutil_env_t *env,
+ axis2_char_t *buf,
+ size_t buf_len,
+ axiom_search_info_t *search_info,
+ axiom_mime_parser_t *mime_parser);
+
+static axis2_bool_t
+axiom_mime_parser_is_more_data(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ axis2_callback_info_t *callback_info);
+
+static axis2_char_t *
+axiom_mime_parser_process_mime_headers(
+ const axutil_env_t *env,
+ axiom_mime_parser_t *mime_parser,
+ axis2_char_t **mime_id,
+ axis2_char_t *mime_headers);
+
+static axis2_status_t
+axiom_mime_parser_cache_to_file(
+ const axutil_env_t* env,
+ axis2_char_t *buf,
+ size_t buf_len,
+ void *handler);
+
+static void*
+axiom_mime_parser_initiate_callback(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ axis2_char_t *mime_id,
+ void *user_param);
+
+AXIS2_EXTERN axiom_mime_parser_t *AXIS2_CALL
+axiom_mime_parser_create(
+ const axutil_env_t * env)
+{
+ axiom_mime_parser_t *mime_parser = NULL;
+
+ AXIS2_ENV_CHECK(env, NULL);
+ mime_parser = (axiom_mime_parser_t *)AXIS2_MALLOC(env->allocator, sizeof(axiom_mime_parser_t));
+
+ if(!mime_parser)
+ {
+ AXIS2_ERROR_SET(env->error, AXIS2_ERROR_NO_MEMORY, AXIS2_FAILURE);
+ return NULL;
+ }
+
+ mime_parser->mime_parts_map = NULL;
+ mime_parser->soap_body_len = 0;
+ mime_parser->soap_body_str = NULL; /* shallow copy */
+ mime_parser->buffer_size = 1;
+ mime_parser->max_buffers = AXIOM_MIME_PARSER_MAX_BUFFERS;
+ mime_parser->attachment_dir = NULL;
+ mime_parser->mtom_caching_callback = NULL;
+ mime_parser->callback_name = NULL;
+ mime_parser->buf_array = NULL;
+ mime_parser->len_array = NULL;
+ mime_parser->current_buf_num = 0;
+ mime_parser->end_of_mime = AXIS2_FALSE;
+ mime_parser->mime_boundary = NULL;
+
+ mime_parser->mime_parts_map = axutil_hash_make(env);
+ if(!(mime_parser->mime_parts_map))
+ {
+ axiom_mime_parser_free(mime_parser, env);
+ return NULL;
+ }
+
+ return mime_parser;
+}
+
+AXIS2_EXTERN void AXIS2_CALL
+axiom_mime_parser_free(
+ axiom_mime_parser_t * mime_parser,
+ const axutil_env_t * env)
+{
+ AXIS2_ENV_CHECK(env, AXIS2_FAILURE);
+
+ /* The map is passed on to SOAP builder, and SOAP builder take over the
+ ownership of the map */
+
+ /* We will unload the callback at the end */
+
+ if(mime_parser->mtom_caching_callback)
+ {
+ axutil_param_t *param = NULL;
+ param = mime_parser->mtom_caching_callback->param;
+
+ AXIOM_MTOM_CACHING_CALLBACK_FREE(mime_parser->mtom_caching_callback, env);
+ mime_parser->mtom_caching_callback = NULL;
+
+ if(param)
+ {
+ axutil_param_free(param, env);
+ param = NULL;
+ }
+ }
+
+ if(mime_parser->buf_array)
+ {
+ AXIS2_FREE(env->allocator, mime_parser->buf_array);
+ mime_parser->buf_array = NULL;
+ }
+
+ if(mime_parser->len_array)
+ {
+ AXIS2_FREE(env->allocator, mime_parser->len_array);
+ mime_parser->len_array = NULL;
+ }
+
+ if(mime_parser)
+ {
+ AXIS2_FREE(env->allocator, mime_parser);
+ }
+
+ return;
+}
+
+AXIS2_EXTERN axis2_status_t AXIS2_CALL
+axiom_mime_parser_parse_for_soap(
+ axiom_mime_parser_t * mime_parser,
+ const axutil_env_t * env,
+ AXIS2_READ_INPUT_CALLBACK callback,
+ void *callback_ctx,
+ axis2_char_t * mime_boundary)
+{
+ size_t size = 0;
+ axis2_char_t *soap_str = NULL;
+ size_t soap_len = 0;
+ size_t temp_mime_boundary_size = 0;
+ axis2_char_t *temp_mime_boundary = NULL;
+ axis2_char_t **buf_array = NULL;
+ size_t *len_array = NULL;
+ int buf_num = 0;
+ axis2_char_t *pos = NULL;
+ axiom_search_info_t *search_info = NULL;
+ int part_start = 0;
+ axis2_bool_t end_of_mime = AXIS2_FALSE;
+ size_t len = 0;
+ axis2_char_t *buffer = NULL;
+ size_t malloc_len = 0;
+ axis2_callback_info_t *callback_info = NULL;
+
+ callback_info = (axis2_callback_info_t *)callback_ctx;
+
+ /* The user will specify the mime_parser->buffer_size */
+
+ size = AXIOM_MIME_PARSER_BUFFER_SIZE * (mime_parser->buffer_size);
+
+ /*An array to keep the set of buffers*/
+
+ buf_array = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t *) * (mime_parser->max_buffers));
+
+ if(!buf_array)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "No memory. Failed in creating buffer array");
+ return AXIS2_FAILURE;
+ }
+
+ /*Keeps the corresponding lengths of buffers in buf_array*/
+
+ len_array = AXIS2_MALLOC(env->allocator, sizeof(size_t) * (mime_parser->max_buffers));
+
+ if(!len_array)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "No memory. Failed in creating length array");
+ return AXIS2_FAILURE;
+ }
+
+ mime_parser->buf_array = buf_array;
+ mime_parser->len_array = len_array;
+
+ temp_mime_boundary = axutil_stracat(env, "--", mime_boundary);
+ temp_mime_boundary_size = strlen(mime_boundary) + 2;
+
+ /*This struct keeps the pre-post search informations*/
+ search_info = AXIS2_MALLOC(env->allocator, sizeof(axiom_search_info_t));
+
+ /* The first buffer is created */
+ buf_array[buf_num] = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ /* The buffer is filled from the callback */
+
+ if(buf_array[buf_num])
+ {
+ len = callback(buf_array[buf_num], (int)size, (void *)callback_ctx);
+ }
+ if(len > 0)
+ {
+ len_array[buf_num] = len;
+ }
+ else
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error reading from the stream");
+ return AXIS2_FAILURE;
+ }
+
+ /*starting buffer for the current search*/
+ part_start = buf_num;
+
+ /*We are passing the address of the buf_num , beacause that value
+ is changing inside the method.*/
+
+ /* Following call to the method will search first \r\n\r\n */
+
+ pos = axiom_mime_parser_search_for_crlf(env, callback, callback_ctx, &buf_num, len_array,
+ buf_array, search_info, size, mime_parser);
+
+ if(!pos)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in the message.");
+ return AXIS2_FAILURE;
+ }
+
+ /* The patteren contains in one buffer */
+
+ if((search_info->match_len2 == 0))
+ {
+ /*Readjusting the buffers for the next search and discarding the prevoius
+ buffers*/
+
+ /* We need the remaining part in the buffer after the \r\n\r\n*/
+
+ malloc_len = buf_array[buf_num] + len_array[buf_num] - pos - 4;
+ if(malloc_len < 0)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in parsing.");
+ return AXIS2_FAILURE;
+ }
+ else
+ {
+ /* Here we will create a new buffer of predefined size fill the
+ * first portion from the remaining part after previous search
+ * and then fill the remaining from the callback */
+
+ buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(malloc_len > 0)
+ {
+ memcpy(buffer, pos + 4, malloc_len);
+ }
+
+ /* Here we need to check for more data, because if the message is too small
+ * comapred to the reading size there may be no data in the stream , instead
+ * all the remaining data may be in the buffer.And if there are no more data
+ * we will set the len to be 0. Otherwise len_array will contain wrong lenghts.
+ */
+
+ if(axiom_mime_parser_is_more_data(mime_parser, env, callback_info))
+ {
+ /* There is more data so fill the remaining from the stream*/
+
+ len = callback(buffer + malloc_len, (int)(size - malloc_len), (void *)callback_ctx);
+ }
+ else
+ {
+ len = 0;
+ }
+
+ /* We do not need the data in the previous buffers once we found a particular
+ * string and after worked with those buffers */
+
+ axiom_mime_parser_clear_buffers(env, buf_array, part_start, buf_num);
+
+ /* Adding the new buffer to the buffer list */
+
+ if(len >= 0)
+ {
+ buf_array[buf_num] = buffer;
+ len_array[buf_num] = malloc_len + len;
+ }
+ }
+ }
+
+ /*The pattern divides among two buffers*/
+
+ else if(search_info->match_len2 > 0)
+ {
+ malloc_len = len_array[buf_num] - search_info->match_len2;
+ if(malloc_len < 0)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in parsing.");
+ return AXIS2_FAILURE;
+ }
+ else
+ {
+ buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ /* Here the buf_num is the second buffer. We will copy the remaining data
+ * after the partial string in the second buffer */
+
+ if(malloc_len > 0)
+ {
+ memcpy(buffer, buf_array[buf_num] + search_info->match_len2, malloc_len);
+ }
+ if(axiom_mime_parser_is_more_data(mime_parser, env, callback_info))
+ {
+ len = callback(buffer + malloc_len, (int)(size - malloc_len), (void *)callback_ctx);
+ }
+ else
+ {
+ len = 0;
+ }
+ axiom_mime_parser_clear_buffers(env, buf_array, part_start, buf_num);
+ if(len >= 0)
+ {
+ buf_array[buf_num] = buffer;
+ len_array[buf_num] = malloc_len + len;
+ }
+ }
+ }
+ else
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in parsing.");
+ return AXIS2_FAILURE;
+ }
+
+ /*Resetting the previous search data and getting ready
+ for the next search */
+
+ part_start = buf_num;
+ pos = NULL;
+ malloc_len = 0;
+
+ search_info->match_len1 = 0;
+ search_info->match_len2 = 0;
+
+ /*In order to extract the soap envelope we need to search for the first
+ --MIMEBOUNDARY */
+
+ pos = axiom_mime_parser_search_for_soap(env, callback, callback_ctx, &buf_num, len_array,
+ buf_array, search_info, size, temp_mime_boundary, mime_parser);
+
+ if(!pos)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error while searching for the SOAP part ");
+ return AXIS2_FAILURE;
+ }
+
+ if(search_info->match_len2 == 0)
+ {
+ /*Calculating the length of the SOAP str*/
+
+ soap_len = axiom_mime_parser_calculate_part_len(env, buf_num, len_array, part_start, pos,
+ buf_array[buf_num]);
+ if(soap_len > 0)
+ {
+ /* Get the SOAP string from the starting and end buffers containing
+ * the SOAP part */
+
+ soap_str = axiom_mime_parser_create_part(env, soap_len, buf_num, len_array, part_start,
+ pos, buf_array, mime_parser);
+ if(!soap_str)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI,
+ "Error while creating the SOAP part from the message ");
+ return AXIS2_FAILURE;
+ }
+
+ malloc_len = len_array[buf_num] - search_info->match_len1 - temp_mime_boundary_size;
+ if(malloc_len < 0)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in parsing for mime.");
+ return AXIS2_FAILURE;
+ }
+ else
+ {
+ /* This will fill the new buffer with remaining data after the
+ * SOAP */
+
+ buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ memset(buffer, 0, size + 1);
+ if(malloc_len > 0)
+ {
+ memcpy(buffer, pos + temp_mime_boundary_size, malloc_len);
+ }
+ if(axiom_mime_parser_is_more_data(mime_parser, env, callback_info))
+ {
+ len = callback(buffer + malloc_len,(int)(size - malloc_len),(void *)callback_ctx);
+ }
+ else
+ {
+ len = 0;
+ }
+ axiom_mime_parser_clear_buffers(env, buf_array, part_start, buf_num);
+ if(len >= 0)
+ {
+ buf_array[buf_num] = buffer;
+ len_array[buf_num] = malloc_len + len;
+ }
+ }
+ }
+ else
+ {
+ return AXIS2_FAILURE;
+ }
+ }
+
+ /* This is the condition where the --MIMEBOUNDARY is divided among two
+ * buffers */
+
+ else if(search_info->match_len2 > 0)
+ {
+ soap_len = axiom_mime_parser_calculate_part_len(env, buf_num - 1, len_array, part_start,
+ pos, buf_array[buf_num - 1]);
+
+ if(soap_len > 0)
+ {
+ /* Here we pass buf_num-1 because buf_num does not have any thing we want to
+ * for this particular part. It begins with the latter part of the search string */
+
+ soap_str = axiom_mime_parser_create_part(env, soap_len, buf_num - 1, len_array,
+ part_start, pos, buf_array, mime_parser);
+ if(!soap_str)
+ {
+ return AXIS2_FAILURE;
+ }
+
+ malloc_len = len_array[buf_num] - search_info->match_len2;
+ if(malloc_len < 0)
+ {
+ return AXIS2_FAILURE;
+ }
+ else
+ {
+ buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(malloc_len > 0)
+ {
+ memcpy(buffer, buf_array[buf_num] + search_info->match_len2, malloc_len);
+ }
+
+ if(axiom_mime_parser_is_more_data(mime_parser, env, callback_info))
+ {
+ len = callback(buffer + malloc_len,(int)(size - malloc_len),(void *)callback_ctx);
+ }
+ else
+ {
+ len = 0;
+ }
+ axiom_mime_parser_clear_buffers(env, buf_array, part_start, buf_num);
+ if(len >= 0)
+ {
+ buf_array[buf_num] = buffer;
+ len_array[buf_num] = malloc_len + len;
+ }
+ }
+ }
+ else
+ {
+ return AXIS2_FAILURE;
+ }
+ }
+
+ mime_parser->soap_body_str = soap_str;
+ mime_parser->soap_body_len = soap_len;
+ mime_parser->current_buf_num = buf_num;
+
+ /* There are multipart/related messages which does not contain attachments
+ * The only mime_part is the soap envelope. So for those messages the mime
+ * boundary after the soap will end up with --
+ * So we will check that here and if we found then the logic inside the
+ * while loop will not be executed */
+
+ end_of_mime = (AXIOM_MIME_BOUNDARY_BYTE == *(buf_array[buf_num])) && (AXIOM_MIME_BOUNDARY_BYTE
+ == *(buf_array[buf_num] + 1));
+ if(end_of_mime)
+ {
+ AXIS2_FREE(env->allocator, buf_array[buf_num]);
+ buf_array[buf_num] = NULL;
+ }
+
+ if(temp_mime_boundary)
+ {
+ AXIS2_FREE(env->allocator, temp_mime_boundary);
+ temp_mime_boundary = NULL;
+ }
+
+ if(search_info)
+ {
+ AXIS2_FREE(env->allocator, search_info);
+ search_info = NULL;
+ }
+
+ mime_parser->end_of_mime = end_of_mime;
+
+ return AXIS2_SUCCESS;
+}
+
+AXIS2_EXTERN axutil_hash_t *AXIS2_CALL
+axiom_mime_parser_parse_for_attachments(
+ axiom_mime_parser_t * mime_parser,
+ const axutil_env_t * env,
+ AXIS2_READ_INPUT_CALLBACK callback,
+ void *callback_ctx,
+ axis2_char_t * mime_boundary,
+ void *user_param)
+{
+ int count = 0;
+ axiom_search_info_t *search_info = NULL;
+ axis2_char_t *pos = NULL;
+ int part_start = 0;
+ axis2_char_t **buf_array = NULL;
+ size_t *len_array = NULL;
+ int buf_num = 0;
+ size_t size = 0;
+ size_t malloc_len = 0;
+ axis2_callback_info_t *callback_info = NULL;
+ axis2_char_t *temp_mime_boundary = NULL;
+ size_t temp_mime_boundary_size = 0;
+ axis2_bool_t end_of_mime = AXIS2_FALSE;
+
+ callback_info = (axis2_callback_info_t *)callback_ctx;
+
+ search_info = AXIS2_MALLOC(env->allocator, sizeof(axiom_search_info_t));
+
+ size = AXIOM_MIME_PARSER_BUFFER_SIZE * (mime_parser->buffer_size);
+
+ buf_array = mime_parser->buf_array;
+ len_array = mime_parser->len_array;
+ buf_num = mime_parser->current_buf_num;
+
+ /*<SOAP></SOAP>--MIMEBOUNDARY
+ mime_headr1:.......
+ mime_headr2:....
+
+ Binarstart.................
+ ...............--MIMEBOUNDARY
+ */
+
+ /* This loop will extract all the attachments in the message. The condition
+ * with the count is needed because if the sender not marked the end of the
+ * attachment wiht -- then this loop may run infinitely. To prevent that
+ * this additional condition has been put */
+
+ temp_mime_boundary = axutil_stracat(env, "--", mime_boundary);
+ temp_mime_boundary_size = strlen(mime_boundary) + 2;
+
+ while((!(mime_parser->end_of_mime)) && count < AXIOM_MIME_PARSER_END_OF_MIME_MAX_COUNT)
+ {
+ /*First we will search for \r\n\r\n*/
+ axis2_char_t *mime_id = NULL;
+ axis2_char_t *mime_type = NULL;
+ size_t mime_headers_len = 0;
+ size_t mime_binary_len = 0;
+ axis2_char_t *mime_binary = NULL;
+ axis2_char_t *mime_headers = NULL;
+ axis2_char_t *buffer = NULL;
+ int len = 0;
+ axis2_status_t status = AXIS2_FAILURE;
+
+ search_info->match_len1 = 0;
+ search_info->match_len2 = 0;
+ pos = NULL;
+ part_start = buf_num;
+
+ malloc_len = 0;
+
+ count++;
+
+ pos = axiom_mime_parser_search_for_crlf(env, callback, callback_ctx, &buf_num, len_array,
+ buf_array, search_info, size, mime_parser);
+
+ if(!pos)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in parsing for mime.");
+ return NULL;
+ }
+
+ /*The pattern contains in one buffer*/
+ if(search_info->match_len2 == 0)
+ {
+ /*We found it . so lets seperates the details of this binary into
+ mime headers.*/
+
+ mime_headers_len = axiom_mime_parser_calculate_part_len(env, buf_num, len_array,
+ part_start, pos, buf_array[buf_num]);
+ if(mime_headers_len > 0)
+ {
+ mime_headers = axiom_mime_parser_create_part(env, mime_headers_len, buf_num,
+ len_array, part_start, pos, buf_array, mime_parser);
+
+ if(!mime_headers)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in parsing for mime headers.");
+ return NULL;
+ }
+ malloc_len = buf_array[buf_num] + len_array[buf_num] - pos - 4;
+
+ /*This should be > 0 , > 0 means there is some part to copy = 0 means
+ there is nothing to copy*/
+ if(malloc_len < 0)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in parsing for mime headers");
+ return NULL;
+ }
+
+ else
+ {
+ buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(malloc_len > 0)
+ {
+ memcpy(buffer, pos + 4, malloc_len);
+ }
+
+ if(axiom_mime_parser_is_more_data(mime_parser, env, callback_info))
+ {
+ len = callback(buffer + malloc_len,(int)(size - malloc_len),(void *)callback_ctx);
+ }
+ else
+ {
+ len = 0;
+ }
+ axiom_mime_parser_clear_buffers(env, buf_array, part_start, buf_num);
+ if(len >= 0)
+ {
+ buf_array[buf_num] = buffer;
+ len_array[buf_num] = malloc_len + len;
+ }
+ }
+ }
+ else
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Error in parsing for mime headers.");
+ return NULL;
+ }
+ }
+
+ else if(search_info->match_len2 > 0)
+ {
+ /*Now we extract the mime headers */
+
+ mime_headers_len = axiom_mime_parser_calculate_part_len(env, buf_num - 1, len_array,
+ part_start, pos, buf_array[buf_num - 1]);
+
+ if(mime_headers_len > 0)
+ {
+ mime_headers = axiom_mime_parser_create_part(env, mime_headers_len, buf_num - 1,
+ len_array, part_start, pos, buf_array, mime_parser);
+ if(!mime_headers)
+ {
+ return NULL;
+ }
+
+ malloc_len = len_array[buf_num] - search_info->match_len2;
+ if(malloc_len < 0)
+ {
+ return NULL;
+ }
+ else
+ {
+ buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(malloc_len > 0)
+ {
+ memcpy(buffer, buf_array[buf_num] + search_info->match_len2, malloc_len);
+ }
+ if(axiom_mime_parser_is_more_data(mime_parser, env, callback_info))
+ {
+ len = callback(buffer + malloc_len,(int)(size - malloc_len),(void*)callback_ctx);
+ }
+ else
+ {
+ len = 0;
+ }
+ axiom_mime_parser_clear_buffers(env, buf_array, part_start, buf_num);
+ if(len >= 0)
+ {
+ buf_array[buf_num] = buffer;
+ len_array[buf_num] = malloc_len + len;
+ }
+ }
+ }
+ else
+ {
+ return NULL;
+ }
+ }
+ else
+ {
+ return NULL;
+ }
+
+ pos = NULL;
+
+ search_info->match_len1 = 0;
+ search_info->match_len2 = 0;
+
+ part_start = buf_num;
+ malloc_len = 0;
+
+ mime_type
+ = axiom_mime_parser_process_mime_headers(env, mime_parser, &mime_id, mime_headers);
+
+ if(!mime_id)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI,
+ "Error in parsing for mime headers.Mime id did not find");
+ return NULL;
+ }
+
+ if(!mime_type)
+ {
+ AXIS2_LOG_DEBUG(env->log, AXIS2_LOG_SI, "Mime type did not find");
+ }
+
+ /*We extract the mime headers. So lets search for the attachment.*/
+
+ pos = axiom_mime_parser_search_for_attachment(mime_parser, env, callback, callback_ctx,
+ &buf_num, len_array, buf_array, search_info, size, temp_mime_boundary, mime_id,
+ user_param);
+
+ if(pos)
+ {
+ /*If it is small we are not caching. Hence the attachment
+ is in memory. So store it in a buffer. */
+
+ if(!search_info->cached)
+ {
+ if(search_info->match_len2 == 0)
+ {
+ /* mime_binary contains the attachment when it does not
+ * cached */
+
+ mime_binary_len = axiom_mime_parser_calculate_part_len(env, buf_num, len_array,
+ part_start, pos, buf_array[buf_num]);
+ if(mime_binary_len > 0)
+ {
+ mime_binary = axiom_mime_parser_create_part(env, mime_binary_len, buf_num,
+ len_array, part_start, pos, buf_array, mime_parser);
+ if(!mime_binary)
+ {
+ return NULL;
+ }
+ }
+ else
+ {
+ return NULL;
+ }
+ }
+
+ else if(search_info->match_len2 > 0)
+ {
+ mime_binary_len = axiom_mime_parser_calculate_part_len(env, buf_num - 1,
+ len_array, part_start, pos, buf_array[buf_num - 1]);
+
+ if(mime_binary_len > 0)
+ {
+ mime_binary = axiom_mime_parser_create_part(env, mime_binary_len, buf_num
+ - 1, len_array, part_start, pos, buf_array, mime_parser);
+ if(!mime_binary)
+ {
+ return NULL;
+ }
+ }
+ else
+ {
+ return NULL;
+ }
+ }
+ }
+
+ /* The functionality below is common when it is cached or not. It deals with remaining
+ * after a particualr attachment, Those may be related to a end of mime_boundary or
+ * another attachment */
+
+ if(search_info->match_len2 == 0)
+ {
+ malloc_len = len_array[buf_num] - search_info->match_len1 - temp_mime_boundary_size;
+ if(malloc_len < 0)
+ {
+ return NULL;
+ }
+ else
+ {
+ buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(malloc_len > 0)
+ {
+ memcpy(buffer, pos + temp_mime_boundary_size, malloc_len);
+ }
+
+ /*When the last buffer only containing -- we know this is the end
+ of the attachments. Hence we don't need to read again*/
+
+ if(malloc_len != 2)
+ {
+ if(axiom_mime_parser_is_more_data(mime_parser, env, callback_info))
+ {
+ len = callback(buffer + malloc_len,(int)(size - malloc_len),(void*)callback_ctx);
+ }
+ else
+ {
+ len = 0;
+ }
+ if(len >= 0)
+ {
+ len_array[buf_num] = malloc_len + len;
+ }
+ }
+
+ /* This means there is another attachment */
+ else
+ {
+ len_array[buf_num] = malloc_len;
+ }
+ axiom_mime_parser_clear_buffers(env, buf_array, part_start, buf_num);
+ buf_array[buf_num] = buffer;
+ }
+ }
+ else if(search_info->match_len2 > 0)
+ {
+ malloc_len = len_array[buf_num] - search_info->match_len2;
+
+ if(malloc_len < 0)
+ {
+ return NULL;
+ }
+ else
+ {
+ buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(malloc_len > 0)
+ {
+ memcpy(buffer, buf_array[buf_num] + search_info->match_len2, malloc_len);
+ }
+ if(malloc_len != 2)
+ {
+ if(axiom_mime_parser_is_more_data(mime_parser, env, callback_info))
+ {
+ len = callback(buffer + malloc_len,(int)(size - malloc_len),(void *)callback_ctx);
+ }
+ else
+ {
+ len = 0;
+ }
+ if(len >= 0)
+ {
+ len_array[buf_num] = malloc_len + len;
+ }
+ }
+ else
+ {
+ len_array[buf_num] = malloc_len;
+ }
+ axiom_mime_parser_clear_buffers(env, buf_array, part_start, buf_num);
+ buf_array[buf_num] = buffer;
+ }
+ }
+ }
+ else
+ {
+ return NULL;
+ }
+
+ /*We have the attachment now either cached or not. So lets put it in the mime_parts
+ * hash map with the mime_id. Remember at this moment we have already processed the
+ * mime_headers and mime_id is already there */
+
+ /* In the case user has not specified the callback or the attachment dir . So we cached it to a memory
+ * buffer. Hence the data_handler type we need to create is different */
+
+ if((search_info->cached) && (!mime_parser->attachment_dir) && (!mime_parser->callback_name))
+ {
+ mime_binary = (axis2_char_t *)search_info->handler;
+ mime_binary_len = search_info->binary_size;
+ }
+
+ /* Storing the attachment in the hash map with the id*/
+
+ status = axiom_mime_parser_store_attachment(env, mime_parser, mime_id, mime_type,
+ mime_binary, mime_binary_len, search_info->cached);
+
+ /*Check wether we encounter --MIMEBOUNDARY-- to find the end of mime*/
+
+ if(buf_array[buf_num])
+ {
+ /* Here we check for the end of mime */
+
+ end_of_mime = (AXIOM_MIME_BOUNDARY_BYTE == *(buf_array[buf_num]))
+ && (AXIOM_MIME_BOUNDARY_BYTE == *(buf_array[buf_num] + 1));
+ if(end_of_mime)
+ {
+ AXIS2_FREE(env->allocator, buf_array[buf_num]);
+ buf_array[buf_num] = NULL;
+ }
+ mime_parser->end_of_mime = end_of_mime;
+ }
+
+ if(mime_headers)
+ {
+ AXIS2_FREE(env->allocator, mime_headers);
+ mime_headers = NULL;
+ }
+
+ if(status != AXIS2_SUCCESS)
+ {
+ return NULL;
+ }
+ }
+
+ /*Do the necessary cleaning */
+
+ /*if (buf_array)
+ {
+ AXIS2_FREE(env->allocator, buf_array);
+ buf_array = NULL;
+ }
+
+ if (len_array)
+ {
+ AXIS2_FREE(env->allocator, len_array);
+ len_array = NULL;
+ }*/
+
+ if(temp_mime_boundary)
+ {
+ AXIS2_FREE(env->allocator, temp_mime_boundary);
+ temp_mime_boundary = NULL;
+ }
+
+ if(search_info)
+ {
+ AXIS2_FREE(env->allocator, search_info);
+ search_info = NULL;
+ }
+
+ return mime_parser->mime_parts_map;
+
+}
+
+/*This method will search for \r\n\r\n */
+
+static axis2_char_t *
+axiom_mime_parser_search_for_crlf(
+ const axutil_env_t * env,
+ AXIS2_READ_INPUT_CALLBACK callback,
+ void *callback_ctx,
+ int *buf_num,
+ size_t *len_array,
+ axis2_char_t **buf_array,
+ axiom_search_info_t *search_info,
+ size_t size,
+ axiom_mime_parser_t *mime_parser)
+{
+ axis2_char_t *found = NULL;
+ int len = 0;
+
+ search_info->search_str = "\r\n\r\n";
+ search_info->buffer1 = NULL;
+ search_info->buffer2 = NULL;
+ search_info->len1 = 0;
+ search_info->len2 = 0;
+ search_info->match_len1 = 0;
+ search_info->match_len2 = 0;
+ search_info->primary_search = AXIS2_FALSE;
+ search_info->cached = AXIS2_FALSE;
+ search_info->handler = NULL;
+ search_info->binary_size = 0;
+
+ /*First do a search in the first buffer*/
+
+ if(buf_array[*buf_num])
+ {
+ search_info->buffer1 = buf_array[*buf_num];
+ search_info->len1 = len_array[*buf_num];
+ found = axiom_mime_parser_search_string(search_info, env);
+ }
+
+ while(!found)
+ {
+ /*Let's read another buffer and do a boundary search in both*/
+
+ *buf_num = *buf_num + 1;
+ buf_array[*buf_num] = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(buf_array[*buf_num])
+ {
+ len = callback(buf_array[*buf_num], (int)size, (void *)callback_ctx);
+ }
+ if(len > 0)
+ {
+ len_array[*buf_num] = len;
+ search_info->buffer2 = buf_array[*buf_num];
+ search_info->len2 = len;
+ found = axiom_mime_parser_search_string(search_info, env);
+ }
+ else
+ {
+ break;
+ }
+ if(!found)
+ {
+ /*Let's do a full search in the second buffer*/
+
+ search_info->buffer1 = buf_array[*buf_num];
+ search_info->len1 = len_array[*buf_num];
+ search_info->primary_search = AXIS2_FALSE;
+ search_info->buffer2 = NULL;
+ search_info->len2 = 0;
+ found = axiom_mime_parser_search_string(search_info, env);
+ }
+ }
+
+ return found;
+}
+
+/* This method will search for the mime_boundary after the SOAP part
+ * of the message */
+
+static axis2_char_t *
+axiom_mime_parser_search_for_soap(
+ const axutil_env_t * env,
+ AXIS2_READ_INPUT_CALLBACK callback,
+ void *callback_ctx,
+ int *buf_num,
+ size_t *len_array,
+ axis2_char_t **buf_array,
+ axiom_search_info_t *search_info,
+ size_t size,
+ axis2_char_t *mime_boundary,
+ axiom_mime_parser_t *mime_parser)
+{
+ axis2_char_t *found = NULL;
+ int len = 0;
+
+ /* What we need to search is the mime_boundary */
+
+ search_info->search_str = mime_boundary;
+ search_info->buffer1 = NULL;
+ search_info->buffer2 = NULL;
+ search_info->len1 = 0;
+ search_info->len2 = 0;
+ search_info->match_len1 = 0;
+ search_info->match_len2 = 0;
+ search_info->primary_search = AXIS2_FALSE;
+
+ if(buf_array[*buf_num])
+ {
+ search_info->buffer1 = buf_array[*buf_num];
+ search_info->len1 = len_array[*buf_num];
+ found = axiom_mime_parser_search_string(search_info, env);
+
+ /* Inside this search primary_search flag will be set to TRUE */
+ }
+
+ while(!found)
+ {
+ /* We need to create the second buffer and do the search for the
+ * mime_boundary in the both the buffers */
+
+ *buf_num = *buf_num + 1;
+ buf_array[*buf_num] = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(buf_array[*buf_num])
+ {
+ len = callback(buf_array[*buf_num], (int)size, (void *)callback_ctx);
+ }
+ if(len > 0)
+ {
+ /* In this search we are matching end part of the first
+ * buffer and starting part of the previous buffer */
+ len_array[*buf_num] = len;
+ search_info->buffer2 = buf_array[*buf_num];
+ search_info->len2 = len;
+ found = axiom_mime_parser_search_string(search_info, env);
+ }
+ else
+ {
+ break;
+ }
+ if(!found)
+ {
+ search_info->buffer1 = buf_array[*buf_num];
+ search_info->len1 = len_array[*buf_num];
+ search_info->primary_search = AXIS2_FALSE;
+ search_info->buffer2 = NULL;
+ search_info->len2 = 0;
+ found = axiom_mime_parser_search_string(search_info, env);
+ }
+ }
+
+ return found;
+}
+
+/*The caching is done in this function. Caching happens when we did not
+ find the mime_boundary in initial two buffers. So the maximum size
+ that we are keeping in memory is 2 * size. This size can be configurable from
+ the aixs.xml. The caching may starts when the search failed with the
+ second buffer.
+ In this logic first we will search for a callback to cache. If it is not
+ there then we will search for a directory to save the file. If it is also
+ not there then the attachment will be in memory.
+ */
+
+static axis2_char_t *
+axiom_mime_parser_search_for_attachment(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t * env,
+ AXIS2_READ_INPUT_CALLBACK callback,
+ void *callback_ctx,
+ int *buf_num,
+ size_t *len_array,
+ axis2_char_t **buf_array,
+ axiom_search_info_t *search_info,
+ size_t size,
+ axis2_char_t *mime_boundary,
+ axis2_char_t *mime_id,
+ void *user_param)
+{
+ axis2_char_t *found = NULL;
+ int len = 0;
+ axis2_status_t status = AXIS2_FAILURE;
+ axis2_char_t *temp = NULL;
+ size_t temp_length = 0;
+ axis2_char_t *file_name = NULL;
+
+ search_info->search_str = mime_boundary;
+ search_info->buffer1 = NULL;
+ search_info->buffer2 = NULL;
+ search_info->len1 = 0;
+ search_info->len2 = 0;
+ search_info->match_len1 = 0;
+ search_info->match_len2 = 0;
+ search_info->primary_search = AXIS2_FALSE;
+ search_info->cached = AXIS2_FALSE;
+ search_info->handler = NULL;
+
+ /*First search in the incoming buffer*/
+
+ if(buf_array[*buf_num])
+ {
+ search_info->buffer1 = buf_array[*buf_num];
+ search_info->len1 = len_array[*buf_num];
+ found = axiom_mime_parser_search_string(search_info, env);
+ }
+
+ while(!found)
+ {
+ if(search_info->cached)
+ {
+ if(mime_parser->callback_name)
+ {
+ if(!(search_info->handler))
+ {
+ /* If the callback is not loaded yet then we load it*/
+ if(!mime_parser->mtom_caching_callback)
+ {
+ search_info->handler = axiom_mime_parser_initiate_callback(mime_parser,
+ env, mime_id, user_param);
+ if(!(search_info->handler))
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI,
+ "Caching Callback is not loaded");
+ return NULL;
+ }
+ }
+ }
+ /*So lets cache the previous buffer which has undergone the
+ full search and the partial search. */
+
+ if(mime_parser->mtom_caching_callback)
+ {
+ /* Caching callback is loaded. So we can cache the previous buffer */
+ status
+ = AXIOM_MTOM_CACHING_CALLBACK_CACHE(mime_parser->mtom_caching_callback,
+ env, buf_array[*buf_num - 1], (int)len_array[*buf_num - 1],
+ search_info->handler);
+ }
+ }
+
+ else if(mime_parser->attachment_dir)
+ {
+ if(!(search_info->handler))
+ {
+ /* If the File is not opened yet we will open it*/
+
+ axis2_char_t *encoded_mime_id = NULL;
+
+ /* Some times content-ids urls, hence we need to encode them
+ * becasue we can't create files with / */
+
+ encoded_mime_id = AXIS2_MALLOC(env->allocator, (sizeof(axis2_char_t))
+ * (strlen(mime_id)));
+ memset(encoded_mime_id, 0, strlen(mime_id));
+ encoded_mime_id = axutil_url_encode(env, encoded_mime_id, mime_id, (int)strlen(
+ mime_id));
+ if(!encoded_mime_id)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Mime Id encoding failed");
+ return NULL;
+ }
+
+ file_name = axutil_stracat(env, mime_parser->attachment_dir, encoded_mime_id);
+ AXIS2_FREE(env->allocator, encoded_mime_id);
+ encoded_mime_id = NULL;
+
+ if(!file_name)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Caching file name creation error");
+ return NULL;
+ }
+ search_info->handler = (void *)fopen(file_name, "ab+");
+ if(!(search_info->handler))
+ {
+ return NULL;
+ }
+ }
+
+ /*So lets cache the previous buffer which has undergone the
+ full search and the partial search. */
+
+ status = axiom_mime_parser_cache_to_file(env, buf_array[*buf_num - 1],
+ len_array[*buf_num - 1], search_info->handler);
+ }
+
+ else
+ {
+ /* Here the user has not specified the caching File location. So we are
+ * not going to cache. Instead we store the attachment in the buffer */
+ status = axiom_mime_parser_cache_to_buffer(env, buf_array[*buf_num - 1],
+ len_array[*buf_num - 1], search_info, mime_parser);
+ }
+
+ if(status == AXIS2_FAILURE)
+ {
+ return NULL;
+ }
+ /*Here we interchange the buffers.*/
+
+ temp = buf_array[*buf_num - 1];
+ buf_array[*buf_num - 1] = buf_array[*buf_num];
+ buf_array[*buf_num] = temp;
+ temp_length = len_array[*buf_num - 1];
+ len_array[*buf_num - 1] = len_array[*buf_num];
+ len_array[*buf_num] = temp_length;
+ if(buf_array[*buf_num])
+ {
+ /*The cached buffer is the one which get filled.*/
+ len = callback(buf_array[*buf_num], (int)size, (void *)callback_ctx);
+ }
+ }
+
+ /*Size of the data in memory not yet risen to the caching threasold
+ *So we can create the second buffer */
+ else
+ {
+ *buf_num = *buf_num + 1;
+ buf_array[*buf_num] = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (size + 1));
+
+ if(buf_array[*buf_num])
+ {
+ len = callback(buf_array[*buf_num], (int)size, (void *)callback_ctx);
+ }
+ }
+
+ /*Doing a complete search in newly cretaed buffer*/
+
+ if(len > 0)
+ {
+ len_array[*buf_num] = len;
+ search_info->buffer2 = buf_array[*buf_num];
+ search_info->len2 = len;
+ found = axiom_mime_parser_search_string(search_info, env);
+ }
+ else
+ {
+ break;
+ }
+
+ /* Now there are two buffers. If the searching string is not
+ * here then we must cache the first buffer */
+
+ if(!found)
+ {
+ /*So now we must start caching*/
+ search_info->buffer1 = buf_array[*buf_num];
+ search_info->len1 = len_array[*buf_num];
+ search_info->primary_search = AXIS2_FALSE;
+ search_info->buffer2 = NULL;
+ search_info->len2 = 0;
+ found = axiom_mime_parser_search_string(search_info, env);
+ if(!found)
+ {
+ /* So at the begining of the next search we start
+ * that only after caching this data */
+ search_info->cached = AXIS2_TRUE;
+ }
+ }
+ }
+
+ /* Here we are out of the loop. If there is no error then this means
+ * the searching string is found */
+ if(search_info->cached && found)
+ {
+ /* If the attachment is cached then we need to cache the
+ * final buffer */
+
+ if(search_info->match_len2 == 0)
+ {
+ /* This is the case where we found the whole string in one buffer
+ * So we need to cache previous buffer and the data up to the starting
+ * point of the search string in the current buffer */
+
+ /* deduct last 2 CRLF character.
+ * For buffering case, it will be done when creating datahandler.*/
+
+ if(mime_parser->mtom_caching_callback)
+ {
+ status = AXIOM_MTOM_CACHING_CALLBACK_CACHE(mime_parser->mtom_caching_callback, env,
+ buf_array[*buf_num - 1], (int)len_array[*buf_num - 1], search_info->handler);
+ if(status == AXIS2_SUCCESS)
+ {
+ status
+ = AXIOM_MTOM_CACHING_CALLBACK_CACHE(mime_parser->mtom_caching_callback,
+ env, buf_array[*buf_num], (int)(found - buf_array[*buf_num] - 2),
+ search_info->handler);
+ }
+ }
+
+ else if(mime_parser->attachment_dir)
+ {
+ status = axiom_mime_parser_cache_to_file(env, buf_array[*buf_num - 1],
+ len_array[*buf_num - 1], search_info->handler);
+ if(status == AXIS2_SUCCESS)
+ {
+ status = axiom_mime_parser_cache_to_file(env, buf_array[*buf_num], found
+ - buf_array[*buf_num] - 2, search_info->handler);
+ }
+ }
+
+ /* If the callback or a file is not there then the data is appended to the buffer */
+
+ else
+ {
+ status = axiom_mime_parser_cache_to_buffer(env, buf_array[*buf_num - 1],
+ len_array[*buf_num - 1], search_info, mime_parser);
+ if(status == AXIS2_SUCCESS)
+ {
+ status = axiom_mime_parser_cache_to_buffer(env, buf_array[*buf_num], found
+ - buf_array[*buf_num], search_info, mime_parser);
+ }
+ }
+ }
+ else if(search_info->match_len2 > 0)
+ {
+ /*Here the curent buffer has partial mime boundary. So we need
+ to cache only the previous buffer. */
+
+ if(mime_parser->mtom_caching_callback)
+ {
+ status = AXIOM_MTOM_CACHING_CALLBACK_CACHE(mime_parser->mtom_caching_callback, env,
+ buf_array[*buf_num - 1], (int)(search_info->match_len1 - 2), search_info->handler);
+ }
+
+ else if(mime_parser->attachment_dir)
+ {
+ status = axiom_mime_parser_cache_to_file(env, buf_array[*buf_num - 1],
+ search_info->match_len1 - 2, search_info->handler);
+ }
+ else
+ {
+ status = axiom_mime_parser_cache_to_buffer(env, buf_array[*buf_num - 1],
+ search_info->match_len1, search_info, mime_parser);
+ }
+ }
+ else
+ {
+ return NULL;
+ }
+
+ if(status == AXIS2_FAILURE)
+ {
+ return NULL;
+ }
+ }
+
+ /* Parsing is done so lets close the relative handlers */
+
+ if(search_info->handler)
+ {
+ if(mime_parser->mtom_caching_callback)
+ {
+ status = AXIOM_MTOM_CACHING_CALLBACK_CLOSE_HANDLER(mime_parser->mtom_caching_callback,
+ env, search_info->handler);
+ if(status == AXIS2_FAILURE)
+ {
+ return NULL;
+ }
+ }
+
+ else if(mime_parser->attachment_dir)
+ {
+ if(fclose((FILE *)(search_info->handler)) == 0)
+ {
+ status = AXIS2_SUCCESS;
+ }
+ else
+ {
+ status = AXIS2_FAILURE;
+ }
+
+ AXIS2_FREE(env->allocator, file_name);
+ file_name = NULL;
+
+ if(status == AXIS2_FAILURE)
+ {
+ return NULL;
+ }
+ }
+ }
+ return found;
+}
+
+/*following two functions are used to extract important information
+ from the buffer list. eg: SOAP, MIME_HEADERS*/
+
+/*marker is the starting buffer of the required
+ part and pos is the end point of that part */
+
+static size_t
+axiom_mime_parser_calculate_part_len(
+ const axutil_env_t *env,
+ int buf_num,
+ size_t *len_list,
+ int marker,
+ axis2_char_t *pos,
+ axis2_char_t *buf)
+{
+ size_t part_len = 0;
+ int i = 0;
+
+ for(i = marker; i < buf_num; i++)
+ {
+ part_len += len_list[i];
+ }
+
+ part_len = part_len + (pos - buf);
+
+ return part_len;
+}
+
+static axis2_char_t *
+axiom_mime_parser_create_part(
+ const axutil_env_t *env,
+ size_t part_len,
+ int buf_num,
+ size_t *len_list,
+ int marker,
+ axis2_char_t *pos,
+ axis2_char_t **buf_list,
+ axiom_mime_parser_t *mime_parser)
+{
+ /*We will copy the set of buffers which contains the required part.
+ This part can be the SOAP message , mime headers or the mime
+ binary in the case of none cahced.*/
+
+ axis2_char_t *part_str = NULL;
+ int i = 0;
+ size_t temp = 0;
+
+ part_str = AXIS2_MALLOC(env->allocator, sizeof(char) * (part_len + 1));
+
+ if(!part_str)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "No memory. Failed in creating buffer");
+ return NULL;
+ }
+
+ /* Copy from the part starting buffer to the
+ * curent buffer */
+
+ for(i = marker; i < buf_num; i++)
+ {
+ if(buf_list[i])
+ {
+ memcpy(part_str + temp, buf_list[i], len_list[i]);
+ temp += len_list[i];
+ }
+ }
+ /* Finally we are copying from the final portion */
+
+ memcpy(part_str + temp, buf_list[i], pos - buf_list[i]);
+
+ part_str[part_len] = '\0';
+
+ return part_str;
+}
+
+AXIS2_EXTERN axutil_hash_t *AXIS2_CALL
+axiom_mime_parser_get_mime_parts_map(
+ axiom_mime_parser_t * mime_parser,
+ const axutil_env_t * env)
+{
+ return mime_parser->mime_parts_map;
+}
+
+AXIS2_EXTERN size_t AXIS2_CALL
+axiom_mime_parser_get_soap_body_len(
+ axiom_mime_parser_t * mime_parser,
+ const axutil_env_t * env)
+{
+ return mime_parser->soap_body_len;
+}
+
+AXIS2_EXTERN axis2_char_t *AXIS2_CALL
+axiom_mime_parser_get_soap_body_str(
+ axiom_mime_parser_t * mime_parser,
+ const axutil_env_t * env)
+{
+ return mime_parser->soap_body_str;
+}
+
+/*This is the new search function. This will first do a
+ search for the entire search string.Then will do a search
+ for the partial string which can be divided among two buffers.*/
+
+static axis2_char_t *
+axiom_mime_parser_search_string(
+ axiom_search_info_t *search_info,
+ const axutil_env_t *env)
+{
+ axis2_char_t *pos = NULL;
+ axis2_char_t *old_pos = NULL;
+ axis2_char_t *found = NULL;
+ size_t str_length = 0;
+ size_t search_length = 0;
+
+ str_length = strlen(search_info->search_str);
+
+ /*First lets search the entire buffer*/
+ if(!search_info->primary_search)
+ {
+ old_pos = search_info->buffer1;
+
+ do
+ {
+ /*find the first byte. We need to adhere to this
+ approach rather than straightaway using strstr
+ because the buffer1 can be containg binary data*/
+
+ pos = NULL;
+
+ search_length = search_info->buffer1 + search_info->len1 - old_pos - str_length + 1;
+
+ if(search_length < 0)
+ {
+ break;
+ }
+
+ if(old_pos)
+ {
+ pos = memchr(old_pos, *(search_info->search_str), search_length);
+ }
+
+ /* found it so lets check the remaining */
+
+ if(pos)
+ {
+ found = axutil_strstr(pos, search_info->search_str);
+ if(found)
+ {
+ search_info->match_len1 = found - search_info->buffer1;
+ break;
+ }
+ else
+ {
+ old_pos = pos + 1;
+ }
+ }
+ }
+ while(pos);
+ }
+
+ search_info->primary_search = AXIS2_TRUE;
+
+ if(found)
+ {
+ return found;
+ }
+
+ /*So we didn't find the string in the buffer
+ lets check whether it is divided in two buffers*/
+
+ else
+ {
+ size_t offset = 0;
+ pos = NULL;
+ old_pos = NULL;
+ found = NULL;
+ search_length = 0;
+
+ if(search_info->buffer2)
+ {
+ old_pos = search_info->buffer1 + search_info->len1 - str_length + 1;
+ do
+ {
+ /*First check the starting byte*/
+ pos = NULL;
+ found = NULL;
+
+ search_length = search_info->buffer1 + search_info->len1 - old_pos;
+
+ if(search_length < 0)
+ {
+ break;
+ }
+
+ pos = memchr(old_pos, *(search_info->search_str), search_length);
+
+ if(pos)
+ {
+ offset = search_info->buffer1 + search_info->len1 - pos;
+
+ /*First match the beginng to offset in buffer1*/
+
+ if(offset > 0)
+ {
+ if(memcmp(pos, search_info->search_str, offset) == 0)
+ {
+ found = pos;
+ }
+
+ /*We found something in buffer1 so lets match the
+ remaining in buffer2*/
+
+ if(found)
+ {
+ if(memcmp(search_info->buffer2, search_info->search_str + offset,
+ str_length - offset) == 0)
+ {
+ search_info->match_len2 = str_length - offset;
+ search_info->match_len1 = found - search_info->buffer1;
+ break;
+ }
+ else
+ {
+ old_pos = pos + 1;
+ }
+ }
+ else
+ {
+ old_pos = pos + 1;
+ }
+ }
+ }
+ }
+ while(pos);
+
+ /* We will set this to AXIS2_FALSE so when the next time this
+ * search method is called it will do a full search first for buffer1 */
+ search_info->primary_search = AXIS2_FALSE;
+
+ return found;
+ }
+ else
+ {
+ return NULL;
+ }
+ }
+}
+
+/* This method creates a data_handler out of the attachment
+ * and store the data_handler in the mime_parts map */
+
+static axis2_status_t
+axiom_mime_parser_store_attachment(
+ const axutil_env_t *env,
+ axiom_mime_parser_t *mime_parser,
+ axis2_char_t *mime_id,
+ axis2_char_t *mime_type,
+ axis2_char_t *mime_binary,
+ size_t mime_binary_len,
+ axis2_bool_t cached)
+{
+ if(mime_parser->mime_parts_map)
+ {
+ if(mime_id)
+ {
+ axiom_data_handler_t *data_handler = NULL;
+
+ /* Handling the case where attachment is cached using a callback */
+
+ if(mime_parser->callback_name && cached)
+ {
+ data_handler = axiom_data_handler_create(env, NULL, mime_type);
+ if(data_handler)
+ {
+ axiom_data_handler_set_cached(data_handler, env, AXIS2_TRUE);
+ axiom_data_handler_set_data_handler_type(data_handler, env,
+ AXIOM_DATA_HANDLER_TYPE_CALLBACK);
+ }
+ }
+
+ /* Handling the case where attachment is cached to a file*/
+
+ else if(mime_parser->attachment_dir && cached)
+ {
+ axis2_char_t *attachment_location = NULL;
+ axis2_char_t *encoded_mime_id = NULL;
+
+ /* Some times content-ids urls, hence we need to encode them
+ * becasue we can't create files with / */
+
+ encoded_mime_id = AXIS2_MALLOC(env->allocator, (sizeof(axis2_char_t)) * (strlen(
+ mime_id)));
+ memset(encoded_mime_id, 0, strlen(mime_id));
+ encoded_mime_id = axutil_url_encode(
+ env, encoded_mime_id, mime_id, (int)strlen(mime_id));
+ if(!encoded_mime_id)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Mime Id encoding failed");
+ return AXIS2_FAILURE;
+ }
+
+ attachment_location = axutil_stracat(env, mime_parser->attachment_dir,
+ encoded_mime_id);
+
+ AXIS2_FREE(env->allocator, encoded_mime_id);
+ encoded_mime_id = NULL;
+
+ if(attachment_location)
+ {
+
+ data_handler = axiom_data_handler_create(env, attachment_location, mime_type);
+ if(data_handler)
+ {
+ axiom_data_handler_set_cached(data_handler, env, AXIS2_TRUE);
+
+ }
+ AXIS2_FREE(env->allocator, attachment_location);
+ attachment_location = NULL;
+ }
+ }
+
+ /* Attachment is in memory, either it is small to be cached or
+ * user does not provided the attachment cached directory */
+
+ else if(mime_binary)
+ {
+ data_handler = axiom_data_handler_create(env, NULL, mime_type);
+ if(data_handler)
+ {
+ axiom_data_handler_set_binary_data(data_handler, env, mime_binary,
+ mime_binary_len - 2);
+ }
+ }
+ axiom_data_handler_set_mime_id(data_handler, env, mime_id);
+
+ axutil_hash_set(mime_parser->mime_parts_map, mime_id, AXIS2_HASH_KEY_STRING,
+ data_handler);
+ if(mime_type)
+ {
+ AXIS2_FREE(env->allocator, mime_type);
+ }
+ }
+ else
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Mime Id or Mime rype not found");
+ return AXIS2_FAILURE;
+ /*axis2_char_t temp_boundry[1024];
+ sprintf(temp_boundry, "--%s--", mime_boundary);
+ if (body_mime && axutil_strstr(body_mime, temp_boundry))
+ {
+ break;
+ }*/
+ }
+ return AXIS2_SUCCESS;
+ }
+ else
+ {
+ return AXIS2_FAILURE;
+ }
+}
+
+/* This method will process the mime_headers for a particualr
+ attacment and return the mime_id */
+
+static axis2_char_t *
+axiom_mime_parser_process_mime_headers(
+ const axutil_env_t *env,
+ axiom_mime_parser_t *mime_parser,
+ axis2_char_t **mime_id,
+ axis2_char_t *mime_headers)
+{
+ axis2_char_t *id = NULL;
+ axis2_char_t *type = NULL;
+ axis2_char_t *pos = NULL;
+
+ /* Get the MIME ID */
+ if(mime_headers)
+ {
+ id = axutil_strcasestr(mime_headers, AXIOM_MIME_HEADER_CONTENT_ID);
+ type = axutil_strcasestr(mime_headers, AXIOM_MIME_HEADER_CONTENT_TYPE);
+ if(type)
+ {
+ axis2_char_t *end = NULL;
+ axis2_char_t *temp_type = NULL;
+ type += axutil_strlen(AXIOM_MIME_HEADER_CONTENT_TYPE);
+ while(type && *type && *type != ':')
+ {
+ type++;
+ }
+ type++;
+ while(type && *type && *type == ' ')
+ {
+ type++;
+ }
+ end = type;
+ while(end && *end && !isspace((int)*end))
+ {
+ end++;
+ }
+ if((end - type) > 0)
+ {
+ temp_type = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * ((end - type) + 1));
+ if(!temp_type)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI,
+ "No memory. Failed in creating Content-Type");
+ return NULL;
+ }
+ memcpy(temp_type, type, (end - type));
+ temp_type[end - type] = '\0';
+ type = temp_type;
+ }
+ }
+ if(id)
+ {
+ id += axutil_strlen(AXIOM_MIME_HEADER_CONTENT_ID);
+ while(id && *id && *id != ':')
+ {
+ id++;
+ }
+ if(id)
+ {
+ while(id && *id && *id != '<')
+ {
+ id++;
+ }
+ id++;
+ pos = axutil_strstr(id, ">");
+ if(pos)
+ {
+ int mime_id_len = 0;
+ mime_id_len = (int)(pos - id);
+ *mime_id = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * mime_id_len + 1);
+ /* The MIME ID will be freed by the SOAP builder */
+ if(*mime_id)
+ {
+ memcpy(*mime_id, id, mime_id_len);
+ (*mime_id)[mime_id_len] = '\0';
+ }
+ else
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI,
+ "No memory. Failed in creating MIME ID");
+ return NULL;
+ }
+ }
+ }
+ }
+ else
+ {
+ /*axis2_char_t temp_boundry[1024];
+ sprintf(temp_boundry, "--%s--", mime_boundary);
+ if (body_mime && axutil_strstr(body_mime, temp_boundry))
+ {
+ break;
+ }*/
+ return NULL;
+ }
+ return type;
+ }
+ else
+ {
+ return NULL;
+ }
+}
+
+/*This is used to free some unwanted buffers. For example we did
+ not want the buffers which contains the data before the soap
+ envelope starts. */
+
+static void
+axiom_mime_parser_clear_buffers(
+ const axutil_env_t *env,
+ axis2_char_t **buf_list,
+ int free_from,
+ int free_to)
+{
+ int i = 0;
+
+ for(i = free_from; i <= free_to; i++)
+ {
+ if(buf_list[i])
+ {
+ AXIS2_FREE(env->allocator, buf_list[i]);
+ buf_list[i] = NULL;
+ }
+ }
+ return;
+}
+
+/* Instead of caching to a file this method will cache it
+ * to a buffer */
+
+static axis2_status_t
+axiom_mime_parser_cache_to_buffer(
+ const axutil_env_t *env,
+ axis2_char_t *buf,
+ size_t buf_len,
+ axiom_search_info_t *search_info,
+ axiom_mime_parser_t *mime_parser)
+{
+ axis2_char_t *data_buffer = NULL;
+ axis2_char_t *temp_buf = NULL;
+ size_t mime_binary_len = 0;
+
+ temp_buf = (axis2_char_t *)search_info->handler;
+ mime_binary_len = search_info->binary_size + buf_len;
+
+ if(mime_binary_len > 0)
+ {
+ data_buffer = AXIS2_MALLOC(env->allocator, sizeof(axis2_char_t) * (mime_binary_len));
+
+ if(data_buffer)
+ {
+ if(temp_buf && search_info->binary_size > 0)
+ {
+ memcpy(data_buffer, temp_buf, search_info->binary_size);
+ AXIS2_FREE(env->allocator, temp_buf);
+ temp_buf = NULL;
+ }
+ memcpy(data_buffer + (search_info->binary_size), buf, buf_len);
+ search_info->binary_size = mime_binary_len;
+ search_info->handler = (void *)data_buffer;
+
+ return AXIS2_SUCCESS;
+ }
+ else
+ {
+ return AXIS2_FAILURE;
+ }
+ }
+ else
+ {
+ return AXIS2_FAILURE;
+ }
+}
+
+AXIS2_EXTERN void AXIS2_CALL
+axiom_mime_parser_set_buffer_size(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ int size)
+{
+ mime_parser->buffer_size = size;
+}
+
+AXIS2_EXTERN void AXIS2_CALL
+axiom_mime_parser_set_max_buffers(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ int num)
+{
+ mime_parser->max_buffers = num;
+}
+
+AXIS2_EXTERN void AXIS2_CALL
+axiom_mime_parser_set_attachment_dir(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ axis2_char_t *attachment_dir)
+{
+ mime_parser->attachment_dir = attachment_dir;
+}
+
+/* Set the path of the caching callnack to be loaded */
+
+AXIS2_EXTERN void AXIS2_CALL
+axiom_mime_parser_set_caching_callback_name(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ axis2_char_t *callback_name)
+{
+ mime_parser->callback_name = callback_name;
+}
+
+AXIS2_EXTERN void AXIS2_CALL
+axiom_mime_parser_set_mime_boundary(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ axis2_char_t *mime_boundary)
+{
+ mime_parser->mime_boundary = mime_boundary;
+}
+
+AXIS2_EXTERN axis2_char_t *AXIS2_CALL
+axiom_mime_parser_get_mime_boundary(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env)
+{
+ return mime_parser->mime_boundary;
+}
+
+/* Load the caching callback dll */
+
+static void*
+axiom_mime_parser_initiate_callback(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ axis2_char_t *mime_id,
+ void *user_param)
+{
+ axutil_dll_desc_t *dll_desc = NULL;
+ axutil_param_t *impl_info_param = NULL;
+ void *ptr = NULL;
+
+ if(mime_parser->callback_name)
+ {
+ AXIS2_LOG_DEBUG(env->log, AXIS2_LOG_SI, "Trying to load module = %s",
+ mime_parser->callback_name);
+ dll_desc = axutil_dll_desc_create(env);
+ axutil_dll_desc_set_name(dll_desc, env, mime_parser->callback_name);
+ impl_info_param = axutil_param_create(env, NULL, dll_desc);
+ /*Set the free function*/
+ axutil_param_set_value_free(impl_info_param, env, axutil_dll_desc_free_void_arg);
+ axutil_class_loader_init(env);
+ ptr = axutil_class_loader_create_dll(env, impl_info_param);
+
+ if(!ptr)
+ {
+ AXIS2_LOG_ERROR(env->log, AXIS2_LOG_SI, "Unable to load the module %s. ERROR",
+ mime_parser->callback_name);
+ return NULL;
+ }
+
+ mime_parser->mtom_caching_callback = (axiom_mtom_caching_callback_t *)ptr;
+ mime_parser->mtom_caching_callback->param = impl_info_param;
+ mime_parser->mtom_caching_callback->user_param = user_param;
+
+ return AXIOM_MTOM_CACHING_CALLBACK_INIT_HANDLER(mime_parser->mtom_caching_callback, env,
+ mime_id);
+ }
+
+ else
+ {
+ return NULL;
+ }
+
+}
+
+/* This method will tell whether there are more data in the
+ * stream */
+
+static axis2_bool_t
+axiom_mime_parser_is_more_data(
+ axiom_mime_parser_t *mime_parser,
+ const axutil_env_t *env,
+ axis2_callback_info_t *callback_info)
+{
+ /* In the case of axutil_http_chunked stream it is the
+ * end of chunk */
+
+ if(callback_info->chunked_stream)
+ {
+ if(axutil_http_chunked_stream_get_end_of_chunks(callback_info->chunked_stream, env))
+ {
+ return AXIS2_FALSE;
+ }
+ else
+ {
+ return AXIS2_TRUE;
+ }
+ }
+
+ /* When we are using content length or any wrapped
+ * stream it will be the unread_length */
+
+ else if(callback_info->unread_len == 0)
+ {
+ return AXIS2_FALSE;
+ }
+ else
+ {
+ return AXIS2_TRUE;
+ }
+}
+
+static axis2_status_t
+axiom_mime_parser_cache_to_file(
+ const axutil_env_t* env,
+ axis2_char_t *buf,
+ size_t buf_len,
+ void *handler)
+{
+ size_t len = 0;
+ FILE *fp = NULL;
+
+ fp = (FILE *)handler;
+
+ len = fwrite(buf, 1, buf_len, fp);
+ if(len < 0)
+ {
+ return AXIS2_FAILURE;
+ }
+ else
+ {
+ return AXIS2_SUCCESS;
+ }
+}
+