1357 lines
47 KiB
C
1357 lines
47 KiB
C
|
/*
|
||
|
* Asterisk -- An open source telephony toolkit.
|
||
|
*
|
||
|
* Copyright (C) 1999 - 2007, Digium, Inc.
|
||
|
*
|
||
|
* Joshua Colp <jcolp@digium.com>
|
||
|
*
|
||
|
* See http://www.asterisk.org for more information about
|
||
|
* the Asterisk project. Please do not directly contact
|
||
|
* any of the maintainers of this project for assistance;
|
||
|
* the project provides a web site, mailing lists and IRC
|
||
|
* channels for your use.
|
||
|
*
|
||
|
* This program is free software, distributed under the terms of
|
||
|
* the GNU General Public License Version 2. See the LICENSE file
|
||
|
* at the top of the source tree.
|
||
|
*/
|
||
|
|
||
|
/*! \file
|
||
|
*
|
||
|
* \brief Audiohooks Architecture
|
||
|
*
|
||
|
* \author Joshua Colp <jcolp@digium.com>
|
||
|
*/
|
||
|
|
||
|
/*** MODULEINFO
|
||
|
<support_level>core</support_level>
|
||
|
***/
|
||
|
|
||
|
#include "asterisk.h"
|
||
|
|
||
|
#include <signal.h>
|
||
|
|
||
|
#include "asterisk/channel.h"
|
||
|
#include "asterisk/utils.h"
|
||
|
#include "asterisk/lock.h"
|
||
|
#include "asterisk/linkedlists.h"
|
||
|
#include "asterisk/audiohook.h"
|
||
|
#include "asterisk/slinfactory.h"
|
||
|
#include "asterisk/frame.h"
|
||
|
#include "asterisk/translate.h"
|
||
|
#include "asterisk/format_cache.h"
|
||
|
|
||
|
#define AST_AUDIOHOOK_SYNC_TOLERANCE 100 /*!< Tolerance in milliseconds for audiohooks synchronization */
|
||
|
#define AST_AUDIOHOOK_SMALL_QUEUE_TOLERANCE 100 /*!< When small queue is enabled, this is the maximum amount of audio that can remain queued at a time. */
|
||
|
#define AST_AUDIOHOOK_LONG_QUEUE_TOLERANCE 500 /*!< Otheriwise we still don't want the queue to grow indefinitely */
|
||
|
|
||
|
#define DEFAULT_INTERNAL_SAMPLE_RATE 8000
|
||
|
|
||
|
struct ast_audiohook_translate {
|
||
|
struct ast_trans_pvt *trans_pvt;
|
||
|
struct ast_format *format;
|
||
|
};
|
||
|
|
||
|
struct ast_audiohook_list {
|
||
|
/* If all the audiohooks in this list are capable
|
||
|
* of processing slinear at any sample rate, this
|
||
|
* variable will be set and the sample rate will
|
||
|
* be preserved during ast_audiohook_write_list()*/
|
||
|
int native_slin_compatible;
|
||
|
int list_internal_samp_rate;/*!< Internal sample rate used when writing to the audiohook list */
|
||
|
|
||
|
struct ast_audiohook_translate in_translate[2];
|
||
|
struct ast_audiohook_translate out_translate[2];
|
||
|
AST_LIST_HEAD_NOLOCK(, ast_audiohook) spy_list;
|
||
|
AST_LIST_HEAD_NOLOCK(, ast_audiohook) whisper_list;
|
||
|
AST_LIST_HEAD_NOLOCK(, ast_audiohook) manipulate_list;
|
||
|
};
|
||
|
|
||
|
static int audiohook_set_internal_rate(struct ast_audiohook *audiohook, int rate, int reset)
|
||
|
{
|
||
|
struct ast_format *slin;
|
||
|
|
||
|
if (audiohook->hook_internal_samp_rate == rate) {
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
audiohook->hook_internal_samp_rate = rate;
|
||
|
|
||
|
slin = ast_format_cache_get_slin_by_rate(rate);
|
||
|
|
||
|
/* Setup the factories that are needed for this audiohook type */
|
||
|
switch (audiohook->type) {
|
||
|
case AST_AUDIOHOOK_TYPE_SPY:
|
||
|
case AST_AUDIOHOOK_TYPE_WHISPER:
|
||
|
if (reset) {
|
||
|
ast_slinfactory_destroy(&audiohook->read_factory);
|
||
|
ast_slinfactory_destroy(&audiohook->write_factory);
|
||
|
}
|
||
|
ast_slinfactory_init_with_format(&audiohook->read_factory, slin);
|
||
|
ast_slinfactory_init_with_format(&audiohook->write_factory, slin);
|
||
|
break;
|
||
|
default:
|
||
|
break;
|
||
|
}
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_init(struct ast_audiohook *audiohook, enum ast_audiohook_type type, const char *source, enum ast_audiohook_init_flags init_flags)
|
||
|
{
|
||
|
/* Need to keep the type and source */
|
||
|
audiohook->type = type;
|
||
|
audiohook->source = source;
|
||
|
|
||
|
/* Initialize lock that protects our audiohook */
|
||
|
ast_mutex_init(&audiohook->lock);
|
||
|
ast_cond_init(&audiohook->trigger, NULL);
|
||
|
|
||
|
audiohook->init_flags = init_flags;
|
||
|
|
||
|
/* initialize internal rate at 8khz, this will adjust if necessary */
|
||
|
audiohook_set_internal_rate(audiohook, DEFAULT_INTERNAL_SAMPLE_RATE, 0);
|
||
|
|
||
|
/* Since we are just starting out... this audiohook is new */
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_NEW);
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_destroy(struct ast_audiohook *audiohook)
|
||
|
{
|
||
|
/* Drop the factories used by this audiohook type */
|
||
|
switch (audiohook->type) {
|
||
|
case AST_AUDIOHOOK_TYPE_SPY:
|
||
|
case AST_AUDIOHOOK_TYPE_WHISPER:
|
||
|
ast_slinfactory_destroy(&audiohook->read_factory);
|
||
|
ast_slinfactory_destroy(&audiohook->write_factory);
|
||
|
break;
|
||
|
default:
|
||
|
break;
|
||
|
}
|
||
|
|
||
|
/* Destroy translation path if present */
|
||
|
if (audiohook->trans_pvt)
|
||
|
ast_translator_free_path(audiohook->trans_pvt);
|
||
|
|
||
|
ao2_cleanup(audiohook->format);
|
||
|
|
||
|
/* Lock and trigger be gone! */
|
||
|
ast_cond_destroy(&audiohook->trigger);
|
||
|
ast_mutex_destroy(&audiohook->lock);
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
#define SHOULD_MUTE(hook, dir) \
|
||
|
((ast_test_flag(hook, AST_AUDIOHOOK_MUTE_READ) && (dir == AST_AUDIOHOOK_DIRECTION_READ)) || \
|
||
|
(ast_test_flag(hook, AST_AUDIOHOOK_MUTE_WRITE) && (dir == AST_AUDIOHOOK_DIRECTION_WRITE)) || \
|
||
|
(ast_test_flag(hook, AST_AUDIOHOOK_MUTE_READ | AST_AUDIOHOOK_MUTE_WRITE) == (AST_AUDIOHOOK_MUTE_READ | AST_AUDIOHOOK_MUTE_WRITE)))
|
||
|
|
||
|
int ast_audiohook_write_frame(struct ast_audiohook *audiohook, enum ast_audiohook_direction direction, struct ast_frame *frame)
|
||
|
{
|
||
|
struct ast_slinfactory *factory = (direction == AST_AUDIOHOOK_DIRECTION_READ ? &audiohook->read_factory : &audiohook->write_factory);
|
||
|
struct ast_slinfactory *other_factory = (direction == AST_AUDIOHOOK_DIRECTION_READ ? &audiohook->write_factory : &audiohook->read_factory);
|
||
|
struct timeval *rwtime = (direction == AST_AUDIOHOOK_DIRECTION_READ ? &audiohook->read_time : &audiohook->write_time), previous_time = *rwtime;
|
||
|
int our_factory_samples;
|
||
|
int our_factory_ms;
|
||
|
int other_factory_samples;
|
||
|
int other_factory_ms;
|
||
|
|
||
|
/* Update last feeding time to be current */
|
||
|
*rwtime = ast_tvnow();
|
||
|
|
||
|
our_factory_samples = ast_slinfactory_available(factory);
|
||
|
our_factory_ms = ast_tvdiff_ms(*rwtime, previous_time) + (our_factory_samples / (audiohook->hook_internal_samp_rate / 1000));
|
||
|
other_factory_samples = ast_slinfactory_available(other_factory);
|
||
|
other_factory_ms = other_factory_samples / (audiohook->hook_internal_samp_rate / 1000);
|
||
|
|
||
|
if (ast_test_flag(audiohook, AST_AUDIOHOOK_TRIGGER_SYNC) && (our_factory_ms - other_factory_ms > AST_AUDIOHOOK_SYNC_TOLERANCE)) {
|
||
|
ast_debug(4, "Flushing audiohook %p so it remains in sync\n", audiohook);
|
||
|
ast_slinfactory_flush(factory);
|
||
|
ast_slinfactory_flush(other_factory);
|
||
|
}
|
||
|
|
||
|
if (ast_test_flag(audiohook, AST_AUDIOHOOK_SMALL_QUEUE) && ((our_factory_ms > AST_AUDIOHOOK_SMALL_QUEUE_TOLERANCE) || (other_factory_ms > AST_AUDIOHOOK_SMALL_QUEUE_TOLERANCE))) {
|
||
|
ast_debug(4, "Audiohook %p has stale audio in its factories. Flushing them both\n", audiohook);
|
||
|
ast_slinfactory_flush(factory);
|
||
|
ast_slinfactory_flush(other_factory);
|
||
|
} else if ((our_factory_ms > AST_AUDIOHOOK_LONG_QUEUE_TOLERANCE) || (other_factory_ms > AST_AUDIOHOOK_LONG_QUEUE_TOLERANCE)) {
|
||
|
ast_debug(4, "Audiohook %p has stale audio in its factories. Flushing them both\n", audiohook);
|
||
|
ast_slinfactory_flush(factory);
|
||
|
ast_slinfactory_flush(other_factory);
|
||
|
}
|
||
|
|
||
|
/* Write frame out to respective factory */
|
||
|
ast_slinfactory_feed(factory, frame);
|
||
|
|
||
|
/* If we need to notify the respective handler of this audiohook, do so */
|
||
|
if ((ast_test_flag(audiohook, AST_AUDIOHOOK_TRIGGER_MODE) == AST_AUDIOHOOK_TRIGGER_READ) && (direction == AST_AUDIOHOOK_DIRECTION_READ)) {
|
||
|
ast_cond_signal(&audiohook->trigger);
|
||
|
} else if ((ast_test_flag(audiohook, AST_AUDIOHOOK_TRIGGER_MODE) == AST_AUDIOHOOK_TRIGGER_WRITE) && (direction == AST_AUDIOHOOK_DIRECTION_WRITE)) {
|
||
|
ast_cond_signal(&audiohook->trigger);
|
||
|
} else if (ast_test_flag(audiohook, AST_AUDIOHOOK_TRIGGER_SYNC)) {
|
||
|
ast_cond_signal(&audiohook->trigger);
|
||
|
}
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
static struct ast_frame *audiohook_read_frame_single(struct ast_audiohook *audiohook, size_t samples, enum ast_audiohook_direction direction)
|
||
|
{
|
||
|
struct ast_slinfactory *factory = (direction == AST_AUDIOHOOK_DIRECTION_READ ? &audiohook->read_factory : &audiohook->write_factory);
|
||
|
int vol = (direction == AST_AUDIOHOOK_DIRECTION_READ ? audiohook->options.read_volume : audiohook->options.write_volume);
|
||
|
short buf[samples];
|
||
|
struct ast_frame frame = {
|
||
|
.frametype = AST_FRAME_VOICE,
|
||
|
.subclass.format = ast_format_cache_get_slin_by_rate(audiohook->hook_internal_samp_rate),
|
||
|
.data.ptr = buf,
|
||
|
.datalen = sizeof(buf),
|
||
|
.samples = samples,
|
||
|
};
|
||
|
|
||
|
/* Ensure the factory is able to give us the samples we want */
|
||
|
if (samples > ast_slinfactory_available(factory)) {
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
/* Read data in from factory */
|
||
|
if (!ast_slinfactory_read(factory, buf, samples)) {
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
if (SHOULD_MUTE(audiohook, direction)) {
|
||
|
/* Swap frame data for zeros if mute is required */
|
||
|
ast_frame_clear(&frame);
|
||
|
} else if (vol) {
|
||
|
/* If a volume adjustment needs to be applied apply it */
|
||
|
ast_frame_adjust_volume(&frame, vol);
|
||
|
}
|
||
|
|
||
|
return ast_frdup(&frame);
|
||
|
}
|
||
|
|
||
|
static struct ast_frame *audiohook_read_frame_both(struct ast_audiohook *audiohook, size_t samples, struct ast_frame **read_reference, struct ast_frame **write_reference)
|
||
|
{
|
||
|
int count;
|
||
|
int usable_read;
|
||
|
int usable_write;
|
||
|
short adjust_value;
|
||
|
short buf1[samples];
|
||
|
short buf2[samples];
|
||
|
short *read_buf = NULL;
|
||
|
short *write_buf = NULL;
|
||
|
struct ast_frame frame = {
|
||
|
.frametype = AST_FRAME_VOICE,
|
||
|
.datalen = sizeof(buf1),
|
||
|
.samples = samples,
|
||
|
};
|
||
|
|
||
|
/* Make sure both factories have the required samples */
|
||
|
usable_read = (ast_slinfactory_available(&audiohook->read_factory) >= samples ? 1 : 0);
|
||
|
usable_write = (ast_slinfactory_available(&audiohook->write_factory) >= samples ? 1 : 0);
|
||
|
|
||
|
if (!usable_read && !usable_write) {
|
||
|
/* If both factories are unusable bail out */
|
||
|
ast_debug(3, "Read factory %p and write factory %p both fail to provide %zu samples\n", &audiohook->read_factory, &audiohook->write_factory, samples);
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
/* If we want to provide only a read factory make sure we aren't waiting for other audio */
|
||
|
if (usable_read && !usable_write && (ast_tvdiff_ms(ast_tvnow(), audiohook->write_time) < (samples/8)*2)) {
|
||
|
ast_debug(3, "Write factory %p was pretty quick last time, waiting for them.\n", &audiohook->write_factory);
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
/* If we want to provide only a write factory make sure we aren't waiting for other audio */
|
||
|
if (usable_write && !usable_read && (ast_tvdiff_ms(ast_tvnow(), audiohook->read_time) < (samples/8)*2)) {
|
||
|
ast_debug(3, "Read factory %p was pretty quick last time, waiting for them.\n", &audiohook->read_factory);
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
/* Start with the read factory... if there are enough samples, read them in */
|
||
|
if (usable_read) {
|
||
|
if (ast_slinfactory_read(&audiohook->read_factory, buf1, samples)) {
|
||
|
read_buf = buf1;
|
||
|
|
||
|
if ((ast_test_flag(audiohook, AST_AUDIOHOOK_MUTE_READ))) {
|
||
|
/* Clear the frame data if we are muting */
|
||
|
memset(buf1, 0, sizeof(buf1));
|
||
|
} else if (audiohook->options.read_volume) {
|
||
|
/* Adjust read volume if need be */
|
||
|
adjust_value = abs(audiohook->options.read_volume);
|
||
|
for (count = 0; count < samples; count++) {
|
||
|
if (audiohook->options.read_volume > 0) {
|
||
|
ast_slinear_saturated_multiply(&buf1[count], &adjust_value);
|
||
|
} else if (audiohook->options.read_volume < 0) {
|
||
|
ast_slinear_saturated_divide(&buf1[count], &adjust_value);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
} else {
|
||
|
ast_debug(1, "Failed to get %d samples from read factory %p\n", (int)samples, &audiohook->read_factory);
|
||
|
}
|
||
|
|
||
|
/* Move on to the write factory... if there are enough samples, read them in */
|
||
|
if (usable_write) {
|
||
|
if (ast_slinfactory_read(&audiohook->write_factory, buf2, samples)) {
|
||
|
write_buf = buf2;
|
||
|
|
||
|
if ((ast_test_flag(audiohook, AST_AUDIOHOOK_MUTE_WRITE))) {
|
||
|
/* Clear the frame data if we are muting */
|
||
|
memset(buf2, 0, sizeof(buf2));
|
||
|
} else if (audiohook->options.write_volume) {
|
||
|
/* Adjust write volume if need be */
|
||
|
adjust_value = abs(audiohook->options.write_volume);
|
||
|
for (count = 0; count < samples; count++) {
|
||
|
if (audiohook->options.write_volume > 0) {
|
||
|
ast_slinear_saturated_multiply(&buf2[count], &adjust_value);
|
||
|
} else if (audiohook->options.write_volume < 0) {
|
||
|
ast_slinear_saturated_divide(&buf2[count], &adjust_value);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
} else {
|
||
|
ast_debug(3, "Failed to get %d samples from write factory %p\n", (int)samples, &audiohook->write_factory);
|
||
|
}
|
||
|
|
||
|
frame.subclass.format = ast_format_cache_get_slin_by_rate(audiohook->hook_internal_samp_rate);
|
||
|
|
||
|
/* Should we substitute silence if one side lacks audio? */
|
||
|
if ((ast_test_flag(audiohook, AST_AUDIOHOOK_SUBSTITUTE_SILENCE))) {
|
||
|
if (read_reference && !read_buf && write_buf) {
|
||
|
read_buf = buf1;
|
||
|
memset(buf1, 0, sizeof(buf1));
|
||
|
} else if (write_reference && read_buf && !write_buf) {
|
||
|
write_buf = buf2;
|
||
|
memset(buf2, 0, sizeof(buf2));
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/* Basically we figure out which buffer to use... and if mixing can be done here */
|
||
|
if (read_buf && read_reference) {
|
||
|
frame.data.ptr = read_buf;
|
||
|
*read_reference = ast_frdup(&frame);
|
||
|
}
|
||
|
if (write_buf && write_reference) {
|
||
|
frame.data.ptr = write_buf;
|
||
|
*write_reference = ast_frdup(&frame);
|
||
|
}
|
||
|
|
||
|
/* Make the correct buffer part of the built frame, so it gets duplicated. */
|
||
|
if (read_buf) {
|
||
|
frame.data.ptr = read_buf;
|
||
|
if (write_buf) {
|
||
|
for (count = 0; count < samples; count++) {
|
||
|
ast_slinear_saturated_add(read_buf++, write_buf++);
|
||
|
}
|
||
|
}
|
||
|
} else if (write_buf) {
|
||
|
frame.data.ptr = write_buf;
|
||
|
} else {
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
/* Yahoo, a combined copy of the audio! */
|
||
|
return ast_frdup(&frame);
|
||
|
}
|
||
|
|
||
|
static struct ast_frame *audiohook_read_frame_helper(struct ast_audiohook *audiohook, size_t samples, enum ast_audiohook_direction direction, struct ast_format *format, struct ast_frame **read_reference, struct ast_frame **write_reference)
|
||
|
{
|
||
|
struct ast_frame *read_frame = NULL, *final_frame = NULL;
|
||
|
struct ast_format *slin;
|
||
|
|
||
|
/*
|
||
|
* Update the rate if compatibility mode is turned off or if it is
|
||
|
* turned on and the format rate is higher than the current rate.
|
||
|
*
|
||
|
* This makes it so any unnecessary rate switching/resetting does
|
||
|
* not take place and also any associated audiohook_list's internal
|
||
|
* sample rate maintains the highest sample rate between hooks.
|
||
|
*/
|
||
|
if (!ast_test_flag(audiohook, AST_AUDIOHOOK_COMPATIBLE) ||
|
||
|
(ast_test_flag(audiohook, AST_AUDIOHOOK_COMPATIBLE) &&
|
||
|
ast_format_get_sample_rate(format) > audiohook->hook_internal_samp_rate)) {
|
||
|
audiohook_set_internal_rate(audiohook, ast_format_get_sample_rate(format), 1);
|
||
|
}
|
||
|
|
||
|
/* If the sample rate of the requested format differs from that of the underlying audiohook
|
||
|
* sample rate determine how many samples we actually need to get from the audiohook. This
|
||
|
* needs to occur as the signed linear factory stores them at the rate of the audiohook.
|
||
|
* We do this by determining the duration of audio they've requested and then determining
|
||
|
* how many samples that would be in the audiohook format.
|
||
|
*/
|
||
|
if (ast_format_get_sample_rate(format) != audiohook->hook_internal_samp_rate) {
|
||
|
samples = (audiohook->hook_internal_samp_rate / 1000) * (samples / (ast_format_get_sample_rate(format) / 1000));
|
||
|
}
|
||
|
|
||
|
if (!(read_frame = (direction == AST_AUDIOHOOK_DIRECTION_BOTH ?
|
||
|
audiohook_read_frame_both(audiohook, samples, read_reference, write_reference) :
|
||
|
audiohook_read_frame_single(audiohook, samples, direction)))) {
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
slin = ast_format_cache_get_slin_by_rate(audiohook->hook_internal_samp_rate);
|
||
|
|
||
|
/* If they don't want signed linear back out, we'll have to send it through the translation path */
|
||
|
if (ast_format_cmp(format, slin) != AST_FORMAT_CMP_EQUAL) {
|
||
|
/* Rebuild translation path if different format then previously */
|
||
|
if (ast_format_cmp(format, audiohook->format) == AST_FORMAT_CMP_NOT_EQUAL) {
|
||
|
if (audiohook->trans_pvt) {
|
||
|
ast_translator_free_path(audiohook->trans_pvt);
|
||
|
audiohook->trans_pvt = NULL;
|
||
|
}
|
||
|
|
||
|
/* Setup new translation path for this format... if we fail we can't very well return signed linear so free the frame and return nothing */
|
||
|
if (!(audiohook->trans_pvt = ast_translator_build_path(format, slin))) {
|
||
|
ast_frfree(read_frame);
|
||
|
return NULL;
|
||
|
}
|
||
|
ao2_replace(audiohook->format, format);
|
||
|
}
|
||
|
/* Convert to requested format, and allow the read in frame to be freed */
|
||
|
final_frame = ast_translate(audiohook->trans_pvt, read_frame, 1);
|
||
|
} else {
|
||
|
final_frame = read_frame;
|
||
|
}
|
||
|
|
||
|
return final_frame;
|
||
|
}
|
||
|
|
||
|
struct ast_frame *ast_audiohook_read_frame(struct ast_audiohook *audiohook, size_t samples, enum ast_audiohook_direction direction, struct ast_format *format)
|
||
|
{
|
||
|
return audiohook_read_frame_helper(audiohook, samples, direction, format, NULL, NULL);
|
||
|
}
|
||
|
|
||
|
struct ast_frame *ast_audiohook_read_frame_all(struct ast_audiohook *audiohook, size_t samples, struct ast_format *format, struct ast_frame **read_frame, struct ast_frame **write_frame)
|
||
|
{
|
||
|
return audiohook_read_frame_helper(audiohook, samples, AST_AUDIOHOOK_DIRECTION_BOTH, format, read_frame, write_frame);
|
||
|
}
|
||
|
|
||
|
static void audiohook_list_set_samplerate_compatibility(struct ast_audiohook_list *audiohook_list)
|
||
|
{
|
||
|
struct ast_audiohook *ah = NULL;
|
||
|
|
||
|
/*
|
||
|
* Anytime the samplerate compatibility is set (attach/remove an audiohook) the
|
||
|
* list's internal sample rate needs to be reset so that the next time processing
|
||
|
* through write_list, if needed, it will get updated to the correct rate.
|
||
|
*
|
||
|
* A list's internal rate always chooses the higher between its own rate and a
|
||
|
* given rate. If the current rate is being driven by an audiohook that wanted a
|
||
|
* higher rate then when this audiohook is removed the list's rate would remain
|
||
|
* at that level when it should be lower, and with no way to lower it since any
|
||
|
* rate compared against it would be lower.
|
||
|
*
|
||
|
* By setting it back to the lowest rate it can recalulate the new highest rate.
|
||
|
*/
|
||
|
audiohook_list->list_internal_samp_rate = DEFAULT_INTERNAL_SAMPLE_RATE;
|
||
|
|
||
|
audiohook_list->native_slin_compatible = 1;
|
||
|
AST_LIST_TRAVERSE(&audiohook_list->manipulate_list, ah, list) {
|
||
|
if (!(ah->init_flags & AST_AUDIOHOOK_MANIPULATE_ALL_RATES)) {
|
||
|
audiohook_list->native_slin_compatible = 0;
|
||
|
return;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_attach(struct ast_channel *chan, struct ast_audiohook *audiohook)
|
||
|
{
|
||
|
ast_channel_lock(chan);
|
||
|
|
||
|
/* Don't allow an audiohook to be attached to a channel that is already hung up.
|
||
|
* The hang up process is what actually notifies the audiohook that it should
|
||
|
* stop.
|
||
|
*/
|
||
|
if (ast_test_flag(ast_channel_flags(chan), AST_FLAG_ZOMBIE)) {
|
||
|
ast_channel_unlock(chan);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
if (!ast_channel_audiohooks(chan)) {
|
||
|
struct ast_audiohook_list *ahlist;
|
||
|
/* Whoops... allocate a new structure */
|
||
|
if (!(ahlist = ast_calloc(1, sizeof(*ahlist)))) {
|
||
|
ast_channel_unlock(chan);
|
||
|
return -1;
|
||
|
}
|
||
|
ast_channel_audiohooks_set(chan, ahlist);
|
||
|
AST_LIST_HEAD_INIT_NOLOCK(&ast_channel_audiohooks(chan)->spy_list);
|
||
|
AST_LIST_HEAD_INIT_NOLOCK(&ast_channel_audiohooks(chan)->whisper_list);
|
||
|
AST_LIST_HEAD_INIT_NOLOCK(&ast_channel_audiohooks(chan)->manipulate_list);
|
||
|
/* This sample rate will adjust as necessary when writing to the list. */
|
||
|
ast_channel_audiohooks(chan)->list_internal_samp_rate = DEFAULT_INTERNAL_SAMPLE_RATE;
|
||
|
}
|
||
|
|
||
|
/* Drop into respective list */
|
||
|
if (audiohook->type == AST_AUDIOHOOK_TYPE_SPY) {
|
||
|
AST_LIST_INSERT_TAIL(&ast_channel_audiohooks(chan)->spy_list, audiohook, list);
|
||
|
} else if (audiohook->type == AST_AUDIOHOOK_TYPE_WHISPER) {
|
||
|
AST_LIST_INSERT_TAIL(&ast_channel_audiohooks(chan)->whisper_list, audiohook, list);
|
||
|
} else if (audiohook->type == AST_AUDIOHOOK_TYPE_MANIPULATE) {
|
||
|
AST_LIST_INSERT_TAIL(&ast_channel_audiohooks(chan)->manipulate_list, audiohook, list);
|
||
|
}
|
||
|
|
||
|
/*
|
||
|
* Initialize the audiohook's rate to the default. If it needs to be,
|
||
|
* it will get updated later.
|
||
|
*/
|
||
|
audiohook_set_internal_rate(audiohook, DEFAULT_INTERNAL_SAMPLE_RATE, 1);
|
||
|
audiohook_list_set_samplerate_compatibility(ast_channel_audiohooks(chan));
|
||
|
|
||
|
/* Change status over to running since it is now attached */
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_RUNNING);
|
||
|
|
||
|
if (ast_channel_is_bridged(chan)) {
|
||
|
ast_channel_set_unbridged_nolock(chan, 1);
|
||
|
}
|
||
|
|
||
|
ast_channel_unlock(chan);
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
void ast_audiohook_update_status(struct ast_audiohook *audiohook, enum ast_audiohook_status status)
|
||
|
{
|
||
|
ast_audiohook_lock(audiohook);
|
||
|
if (audiohook->status != AST_AUDIOHOOK_STATUS_DONE) {
|
||
|
audiohook->status = status;
|
||
|
ast_cond_signal(&audiohook->trigger);
|
||
|
}
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_detach(struct ast_audiohook *audiohook)
|
||
|
{
|
||
|
if (audiohook->status == AST_AUDIOHOOK_STATUS_NEW || audiohook->status == AST_AUDIOHOOK_STATUS_DONE) {
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_SHUTDOWN);
|
||
|
|
||
|
while (audiohook->status != AST_AUDIOHOOK_STATUS_DONE) {
|
||
|
ast_audiohook_trigger_wait(audiohook);
|
||
|
}
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
void ast_audiohook_detach_list(struct ast_audiohook_list *audiohook_list)
|
||
|
{
|
||
|
int i;
|
||
|
struct ast_audiohook *audiohook;
|
||
|
|
||
|
if (!audiohook_list) {
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
/* Drop any spies */
|
||
|
while ((audiohook = AST_LIST_REMOVE_HEAD(&audiohook_list->spy_list, list))) {
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_DONE);
|
||
|
}
|
||
|
|
||
|
/* Drop any whispering sources */
|
||
|
while ((audiohook = AST_LIST_REMOVE_HEAD(&audiohook_list->whisper_list, list))) {
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_DONE);
|
||
|
}
|
||
|
|
||
|
/* Drop any manipulators */
|
||
|
while ((audiohook = AST_LIST_REMOVE_HEAD(&audiohook_list->manipulate_list, list))) {
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_DONE);
|
||
|
audiohook->manipulate_callback(audiohook, NULL, NULL, 0);
|
||
|
}
|
||
|
|
||
|
/* Drop translation paths if present */
|
||
|
for (i = 0; i < 2; i++) {
|
||
|
if (audiohook_list->in_translate[i].trans_pvt) {
|
||
|
ast_translator_free_path(audiohook_list->in_translate[i].trans_pvt);
|
||
|
ao2_cleanup(audiohook_list->in_translate[i].format);
|
||
|
}
|
||
|
if (audiohook_list->out_translate[i].trans_pvt) {
|
||
|
ast_translator_free_path(audiohook_list->out_translate[i].trans_pvt);
|
||
|
ao2_cleanup(audiohook_list->in_translate[i].format);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/* Free ourselves */
|
||
|
ast_free(audiohook_list);
|
||
|
}
|
||
|
|
||
|
/*! \brief find an audiohook based on its source
|
||
|
* \param audiohook_list The list of audiohooks to search in
|
||
|
* \param source The source of the audiohook we wish to find
|
||
|
* \return corresponding audiohook
|
||
|
* \retval NULL if it cannot be found
|
||
|
*/
|
||
|
static struct ast_audiohook *find_audiohook_by_source(struct ast_audiohook_list *audiohook_list, const char *source)
|
||
|
{
|
||
|
struct ast_audiohook *audiohook = NULL;
|
||
|
|
||
|
AST_LIST_TRAVERSE(&audiohook_list->spy_list, audiohook, list) {
|
||
|
if (!strcasecmp(audiohook->source, source)) {
|
||
|
return audiohook;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
AST_LIST_TRAVERSE(&audiohook_list->whisper_list, audiohook, list) {
|
||
|
if (!strcasecmp(audiohook->source, source)) {
|
||
|
return audiohook;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
AST_LIST_TRAVERSE(&audiohook_list->manipulate_list, audiohook, list) {
|
||
|
if (!strcasecmp(audiohook->source, source)) {
|
||
|
return audiohook;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
static void audiohook_move(struct ast_channel *old_chan, struct ast_channel *new_chan, struct ast_audiohook *audiohook)
|
||
|
{
|
||
|
enum ast_audiohook_status oldstatus;
|
||
|
|
||
|
/* By locking both channels and the audiohook, we can assure that
|
||
|
* another thread will not have a chance to read the audiohook's status
|
||
|
* as done, even though ast_audiohook_remove signals the trigger
|
||
|
* condition.
|
||
|
*/
|
||
|
ast_audiohook_lock(audiohook);
|
||
|
oldstatus = audiohook->status;
|
||
|
|
||
|
ast_audiohook_remove(old_chan, audiohook);
|
||
|
ast_audiohook_attach(new_chan, audiohook);
|
||
|
|
||
|
audiohook->status = oldstatus;
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
}
|
||
|
|
||
|
void ast_audiohook_move_by_source(struct ast_channel *old_chan, struct ast_channel *new_chan, const char *source)
|
||
|
{
|
||
|
struct ast_audiohook *audiohook;
|
||
|
|
||
|
if (!ast_channel_audiohooks(old_chan)) {
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
audiohook = find_audiohook_by_source(ast_channel_audiohooks(old_chan), source);
|
||
|
if (!audiohook) {
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
audiohook_move(old_chan, new_chan, audiohook);
|
||
|
}
|
||
|
|
||
|
void ast_audiohook_move_all(struct ast_channel *old_chan, struct ast_channel *new_chan)
|
||
|
{
|
||
|
struct ast_audiohook *audiohook;
|
||
|
struct ast_audiohook_list *audiohook_list;
|
||
|
|
||
|
audiohook_list = ast_channel_audiohooks(old_chan);
|
||
|
if (!audiohook_list) {
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
AST_LIST_TRAVERSE_SAFE_BEGIN(&audiohook_list->spy_list, audiohook, list) {
|
||
|
audiohook_move(old_chan, new_chan, audiohook);
|
||
|
}
|
||
|
AST_LIST_TRAVERSE_SAFE_END;
|
||
|
|
||
|
AST_LIST_TRAVERSE_SAFE_BEGIN(&audiohook_list->whisper_list, audiohook, list) {
|
||
|
audiohook_move(old_chan, new_chan, audiohook);
|
||
|
}
|
||
|
AST_LIST_TRAVERSE_SAFE_END;
|
||
|
|
||
|
AST_LIST_TRAVERSE_SAFE_BEGIN(&audiohook_list->manipulate_list, audiohook, list) {
|
||
|
audiohook_move(old_chan, new_chan, audiohook);
|
||
|
}
|
||
|
AST_LIST_TRAVERSE_SAFE_END;
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_detach_source(struct ast_channel *chan, const char *source)
|
||
|
{
|
||
|
struct ast_audiohook *audiohook = NULL;
|
||
|
|
||
|
ast_channel_lock(chan);
|
||
|
|
||
|
/* Ensure the channel has audiohooks on it */
|
||
|
if (!ast_channel_audiohooks(chan)) {
|
||
|
ast_channel_unlock(chan);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
audiohook = find_audiohook_by_source(ast_channel_audiohooks(chan), source);
|
||
|
|
||
|
ast_channel_unlock(chan);
|
||
|
|
||
|
if (audiohook && audiohook->status != AST_AUDIOHOOK_STATUS_DONE) {
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_SHUTDOWN);
|
||
|
}
|
||
|
|
||
|
return (audiohook ? 0 : -1);
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_remove(struct ast_channel *chan, struct ast_audiohook *audiohook)
|
||
|
{
|
||
|
ast_channel_lock(chan);
|
||
|
|
||
|
if (!ast_channel_audiohooks(chan)) {
|
||
|
ast_channel_unlock(chan);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
if (audiohook->type == AST_AUDIOHOOK_TYPE_SPY) {
|
||
|
AST_LIST_REMOVE(&ast_channel_audiohooks(chan)->spy_list, audiohook, list);
|
||
|
} else if (audiohook->type == AST_AUDIOHOOK_TYPE_WHISPER) {
|
||
|
AST_LIST_REMOVE(&ast_channel_audiohooks(chan)->whisper_list, audiohook, list);
|
||
|
} else if (audiohook->type == AST_AUDIOHOOK_TYPE_MANIPULATE) {
|
||
|
AST_LIST_REMOVE(&ast_channel_audiohooks(chan)->manipulate_list, audiohook, list);
|
||
|
}
|
||
|
|
||
|
audiohook_list_set_samplerate_compatibility(ast_channel_audiohooks(chan));
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_DONE);
|
||
|
|
||
|
if (ast_channel_is_bridged(chan)) {
|
||
|
ast_channel_set_unbridged_nolock(chan, 1);
|
||
|
}
|
||
|
|
||
|
ast_channel_unlock(chan);
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
/*! \brief Pass a DTMF frame off to be handled by the audiohook core
|
||
|
* \param chan Channel that the list is coming off of
|
||
|
* \param audiohook_list List of audiohooks
|
||
|
* \param direction Direction frame is coming in from
|
||
|
* \param frame The frame itself
|
||
|
* \return frame on success
|
||
|
* \retval NULL on failure
|
||
|
*/
|
||
|
static struct ast_frame *dtmf_audiohook_write_list(struct ast_channel *chan, struct ast_audiohook_list *audiohook_list, enum ast_audiohook_direction direction, struct ast_frame *frame)
|
||
|
{
|
||
|
struct ast_audiohook *audiohook = NULL;
|
||
|
int removed = 0;
|
||
|
|
||
|
AST_LIST_TRAVERSE_SAFE_BEGIN(&audiohook_list->manipulate_list, audiohook, list) {
|
||
|
ast_audiohook_lock(audiohook);
|
||
|
if (audiohook->status != AST_AUDIOHOOK_STATUS_RUNNING) {
|
||
|
AST_LIST_REMOVE_CURRENT(list);
|
||
|
removed = 1;
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_DONE);
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
audiohook->manipulate_callback(audiohook, NULL, NULL, 0);
|
||
|
if (ast_channel_is_bridged(chan)) {
|
||
|
ast_channel_set_unbridged_nolock(chan, 1);
|
||
|
}
|
||
|
continue;
|
||
|
}
|
||
|
if (ast_test_flag(audiohook, AST_AUDIOHOOK_WANTS_DTMF)) {
|
||
|
audiohook->manipulate_callback(audiohook, chan, frame, direction);
|
||
|
}
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
}
|
||
|
AST_LIST_TRAVERSE_SAFE_END;
|
||
|
|
||
|
/* if an audiohook got removed, reset samplerate compatibility */
|
||
|
if (removed) {
|
||
|
audiohook_list_set_samplerate_compatibility(audiohook_list);
|
||
|
}
|
||
|
return frame;
|
||
|
}
|
||
|
|
||
|
static struct ast_frame *audiohook_list_translate_to_slin(struct ast_audiohook_list *audiohook_list,
|
||
|
enum ast_audiohook_direction direction, struct ast_frame *frame)
|
||
|
{
|
||
|
struct ast_audiohook_translate *in_translate = (direction == AST_AUDIOHOOK_DIRECTION_READ ?
|
||
|
&audiohook_list->in_translate[0] : &audiohook_list->in_translate[1]);
|
||
|
struct ast_frame *new_frame = frame;
|
||
|
struct ast_format *slin;
|
||
|
|
||
|
/*
|
||
|
* If we are capable of sample rates other that 8khz, update the internal
|
||
|
* audiohook_list's rate and higher sample rate audio arrives. If native
|
||
|
* slin compatibility is turned on all audiohooks in the list will be
|
||
|
* updated as well during read/write processing.
|
||
|
*/
|
||
|
audiohook_list->list_internal_samp_rate =
|
||
|
MAX(ast_format_get_sample_rate(frame->subclass.format), audiohook_list->list_internal_samp_rate);
|
||
|
|
||
|
slin = ast_format_cache_get_slin_by_rate(audiohook_list->list_internal_samp_rate);
|
||
|
if (ast_format_cmp(frame->subclass.format, slin) == AST_FORMAT_CMP_EQUAL) {
|
||
|
return new_frame;
|
||
|
}
|
||
|
|
||
|
if (!in_translate->format ||
|
||
|
ast_format_cmp(frame->subclass.format, in_translate->format) != AST_FORMAT_CMP_EQUAL) {
|
||
|
struct ast_trans_pvt *new_trans;
|
||
|
|
||
|
new_trans = ast_translator_build_path(slin, frame->subclass.format);
|
||
|
if (!new_trans) {
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
if (in_translate->trans_pvt) {
|
||
|
ast_translator_free_path(in_translate->trans_pvt);
|
||
|
}
|
||
|
in_translate->trans_pvt = new_trans;
|
||
|
|
||
|
ao2_replace(in_translate->format, frame->subclass.format);
|
||
|
}
|
||
|
|
||
|
if (!(new_frame = ast_translate(in_translate->trans_pvt, frame, 0))) {
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
return new_frame;
|
||
|
}
|
||
|
|
||
|
static struct ast_frame *audiohook_list_translate_to_native(struct ast_audiohook_list *audiohook_list,
|
||
|
enum ast_audiohook_direction direction, struct ast_frame *slin_frame, struct ast_format *outformat)
|
||
|
{
|
||
|
struct ast_audiohook_translate *out_translate = (direction == AST_AUDIOHOOK_DIRECTION_READ ? &audiohook_list->out_translate[0] : &audiohook_list->out_translate[1]);
|
||
|
struct ast_frame *outframe = NULL;
|
||
|
if (ast_format_cmp(slin_frame->subclass.format, outformat) == AST_FORMAT_CMP_NOT_EQUAL) {
|
||
|
/* rebuild translators if necessary */
|
||
|
if (ast_format_cmp(out_translate->format, outformat) == AST_FORMAT_CMP_NOT_EQUAL) {
|
||
|
if (out_translate->trans_pvt) {
|
||
|
ast_translator_free_path(out_translate->trans_pvt);
|
||
|
}
|
||
|
if (!(out_translate->trans_pvt = ast_translator_build_path(outformat, slin_frame->subclass.format))) {
|
||
|
return NULL;
|
||
|
}
|
||
|
ao2_replace(out_translate->format, outformat);
|
||
|
}
|
||
|
/* translate back to the format the frame came in as. */
|
||
|
if (!(outframe = ast_translate(out_translate->trans_pvt, slin_frame, 0))) {
|
||
|
return NULL;
|
||
|
}
|
||
|
}
|
||
|
return outframe;
|
||
|
}
|
||
|
|
||
|
/*!
|
||
|
*\brief Set the audiohook's internal sample rate to the audiohook_list's rate,
|
||
|
* but only when native slin compatibility is turned on.
|
||
|
*
|
||
|
* \param audiohook_list audiohook_list data object
|
||
|
* \param audiohook the audiohook to update
|
||
|
* \param rate the current max internal sample rate
|
||
|
*/
|
||
|
static void audiohook_list_set_hook_rate(struct ast_audiohook_list *audiohook_list,
|
||
|
struct ast_audiohook *audiohook, int *rate)
|
||
|
{
|
||
|
/* The rate should always be the max between itself and the hook */
|
||
|
if (audiohook->hook_internal_samp_rate > *rate) {
|
||
|
*rate = audiohook->hook_internal_samp_rate;
|
||
|
}
|
||
|
|
||
|
/*
|
||
|
* If native slin compatibility is turned on then update the audiohook
|
||
|
* with the audiohook_list's current rate. Note, the audiohook's rate is
|
||
|
* set to the audiohook_list's rate and not the given rate. If there is
|
||
|
* a change in rate the hook's rate is changed on its next check.
|
||
|
*/
|
||
|
if (audiohook_list->native_slin_compatible) {
|
||
|
ast_set_flag(audiohook, AST_AUDIOHOOK_COMPATIBLE);
|
||
|
audiohook_set_internal_rate(audiohook, audiohook_list->list_internal_samp_rate, 1);
|
||
|
} else {
|
||
|
ast_clear_flag(audiohook, AST_AUDIOHOOK_COMPATIBLE);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/*!
|
||
|
* \brief Pass an AUDIO frame off to be handled by the audiohook core
|
||
|
*
|
||
|
* \details
|
||
|
* This function has 3 ast_frames and 3 parts to handle each. At the beginning of this
|
||
|
* function all 3 frames, start_frame, middle_frame, and end_frame point to the initial
|
||
|
* input frame.
|
||
|
*
|
||
|
* Part_1: Translate the start_frame into SLINEAR audio if it is not already in that
|
||
|
* format. The result of this part is middle_frame is guaranteed to be in
|
||
|
* SLINEAR format for Part_2.
|
||
|
* Part_2: Send middle_frame off to spies and manipulators. At this point middle_frame is
|
||
|
* either a new frame as result of the translation, or points directly to the start_frame
|
||
|
* because no translation to SLINEAR audio was required.
|
||
|
* Part_3: Translate end_frame's audio back into the format of start frame if necessary. This
|
||
|
* is only necessary if manipulation of middle_frame occurred.
|
||
|
*
|
||
|
* \param chan Channel that the list is coming off of
|
||
|
* \param audiohook_list List of audiohooks
|
||
|
* \param direction Direction frame is coming in from
|
||
|
* \param frame The frame itself
|
||
|
* \return frame on success
|
||
|
* \retval NULL on failure
|
||
|
*/
|
||
|
static struct ast_frame *audio_audiohook_write_list(struct ast_channel *chan, struct ast_audiohook_list *audiohook_list, enum ast_audiohook_direction direction, struct ast_frame *frame)
|
||
|
{
|
||
|
struct ast_frame *start_frame = frame, *middle_frame = frame, *end_frame = frame;
|
||
|
struct ast_audiohook *audiohook = NULL;
|
||
|
int samples;
|
||
|
int middle_frame_manipulated = 0;
|
||
|
int removed = 0;
|
||
|
int internal_sample_rate;
|
||
|
|
||
|
/* ---Part_1. translate start_frame to SLINEAR if necessary. */
|
||
|
if (!(middle_frame = audiohook_list_translate_to_slin(audiohook_list, direction, start_frame))) {
|
||
|
return frame;
|
||
|
}
|
||
|
|
||
|
/* If the translation resulted in an interpolated frame then immediately return as audiohooks
|
||
|
* rely on actual media being present to do things.
|
||
|
*/
|
||
|
if (!middle_frame->data.ptr) {
|
||
|
if (middle_frame != start_frame) {
|
||
|
ast_frfree(middle_frame);
|
||
|
}
|
||
|
return start_frame;
|
||
|
}
|
||
|
|
||
|
samples = middle_frame->samples;
|
||
|
|
||
|
/*
|
||
|
* While processing each audiohook check to see if the internal sample rate needs
|
||
|
* to be adjusted (it should be the highest rate specified between formats and
|
||
|
* hooks). The given audiohook_list's internal sample rate is then set to the
|
||
|
* updated value before returning.
|
||
|
*
|
||
|
* If slin compatibility mode is turned on then an audiohook's internal sample
|
||
|
* rate is set to its audiohook_list's rate. If an audiohook_list's rate is
|
||
|
* adjusted during this pass then the change is picked up by the audiohooks
|
||
|
* on the next pass.
|
||
|
*/
|
||
|
internal_sample_rate = audiohook_list->list_internal_samp_rate;
|
||
|
|
||
|
/* ---Part_2: Send middle_frame to spy and manipulator lists. middle_frame is guaranteed to be SLINEAR here.*/
|
||
|
/* Queue up signed linear frame to each spy */
|
||
|
AST_LIST_TRAVERSE_SAFE_BEGIN(&audiohook_list->spy_list, audiohook, list) {
|
||
|
ast_audiohook_lock(audiohook);
|
||
|
if (audiohook->status != AST_AUDIOHOOK_STATUS_RUNNING) {
|
||
|
AST_LIST_REMOVE_CURRENT(list);
|
||
|
removed = 1;
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_DONE);
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
if (ast_channel_is_bridged(chan)) {
|
||
|
ast_channel_set_unbridged_nolock(chan, 1);
|
||
|
}
|
||
|
continue;
|
||
|
}
|
||
|
audiohook_list_set_hook_rate(audiohook_list, audiohook, &internal_sample_rate);
|
||
|
ast_audiohook_write_frame(audiohook, direction, middle_frame);
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
}
|
||
|
AST_LIST_TRAVERSE_SAFE_END;
|
||
|
|
||
|
/* If this frame is being written out to the channel then we need to use whisper sources */
|
||
|
if (!AST_LIST_EMPTY(&audiohook_list->whisper_list)) {
|
||
|
int i = 0;
|
||
|
short read_buf[samples], combine_buf[samples], *data1 = NULL, *data2 = NULL;
|
||
|
memset(&combine_buf, 0, sizeof(combine_buf));
|
||
|
AST_LIST_TRAVERSE_SAFE_BEGIN(&audiohook_list->whisper_list, audiohook, list) {
|
||
|
struct ast_slinfactory *factory = (direction == AST_AUDIOHOOK_DIRECTION_READ ? &audiohook->read_factory : &audiohook->write_factory);
|
||
|
ast_audiohook_lock(audiohook);
|
||
|
if (audiohook->status != AST_AUDIOHOOK_STATUS_RUNNING) {
|
||
|
AST_LIST_REMOVE_CURRENT(list);
|
||
|
removed = 1;
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_DONE);
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
if (ast_channel_is_bridged(chan)) {
|
||
|
ast_channel_set_unbridged_nolock(chan, 1);
|
||
|
}
|
||
|
continue;
|
||
|
}
|
||
|
audiohook_list_set_hook_rate(audiohook_list, audiohook, &internal_sample_rate);
|
||
|
if (ast_slinfactory_available(factory) >= samples && ast_slinfactory_read(factory, read_buf, samples)) {
|
||
|
/* Take audio from this whisper source and combine it into our main buffer */
|
||
|
for (i = 0, data1 = combine_buf, data2 = read_buf; i < samples; i++, data1++, data2++) {
|
||
|
ast_slinear_saturated_add(data1, data2);
|
||
|
}
|
||
|
}
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
}
|
||
|
AST_LIST_TRAVERSE_SAFE_END;
|
||
|
/* We take all of the combined whisper sources and combine them into the audio being written out */
|
||
|
for (i = 0, data1 = middle_frame->data.ptr, data2 = combine_buf; i < samples; i++, data1++, data2++) {
|
||
|
ast_slinear_saturated_add(data1, data2);
|
||
|
}
|
||
|
middle_frame_manipulated = 1;
|
||
|
}
|
||
|
|
||
|
/* Pass off frame to manipulate audiohooks */
|
||
|
if (!AST_LIST_EMPTY(&audiohook_list->manipulate_list)) {
|
||
|
AST_LIST_TRAVERSE_SAFE_BEGIN(&audiohook_list->manipulate_list, audiohook, list) {
|
||
|
ast_audiohook_lock(audiohook);
|
||
|
if (audiohook->status != AST_AUDIOHOOK_STATUS_RUNNING) {
|
||
|
AST_LIST_REMOVE_CURRENT(list);
|
||
|
removed = 1;
|
||
|
ast_audiohook_update_status(audiohook, AST_AUDIOHOOK_STATUS_DONE);
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
/* We basically drop all of our links to the manipulate audiohook and prod it to do it's own destructive things */
|
||
|
audiohook->manipulate_callback(audiohook, chan, NULL, direction);
|
||
|
if (ast_channel_is_bridged(chan)) {
|
||
|
ast_channel_set_unbridged_nolock(chan, 1);
|
||
|
}
|
||
|
continue;
|
||
|
}
|
||
|
audiohook_list_set_hook_rate(audiohook_list, audiohook, &internal_sample_rate);
|
||
|
/*
|
||
|
* Feed in frame to manipulation.
|
||
|
*/
|
||
|
if (!audiohook->manipulate_callback(audiohook, chan, middle_frame, direction)) {
|
||
|
/*
|
||
|
* XXX FAILURES ARE IGNORED XXX
|
||
|
* If the manipulation fails then the frame will be returned in its original state.
|
||
|
* Since there are potentially more manipulator callbacks in the list, no action should
|
||
|
* be taken here to exit early.
|
||
|
*/
|
||
|
middle_frame_manipulated = 1;
|
||
|
}
|
||
|
ast_audiohook_unlock(audiohook);
|
||
|
}
|
||
|
AST_LIST_TRAVERSE_SAFE_END;
|
||
|
}
|
||
|
|
||
|
/* ---Part_3: Decide what to do with the end_frame (whether to transcode or not) */
|
||
|
if (middle_frame_manipulated) {
|
||
|
if (!(end_frame = audiohook_list_translate_to_native(audiohook_list, direction, middle_frame, start_frame->subclass.format))) {
|
||
|
/* translation failed, so just pass back the input frame */
|
||
|
end_frame = start_frame;
|
||
|
}
|
||
|
} else {
|
||
|
end_frame = start_frame;
|
||
|
}
|
||
|
/* clean up our middle_frame if required */
|
||
|
if (middle_frame != end_frame) {
|
||
|
ast_frfree(middle_frame);
|
||
|
middle_frame = NULL;
|
||
|
}
|
||
|
|
||
|
/* Before returning, if an audiohook got removed, reset samplerate compatibility */
|
||
|
if (removed) {
|
||
|
audiohook_list_set_samplerate_compatibility(audiohook_list);
|
||
|
} else {
|
||
|
/*
|
||
|
* Set the audiohook_list's rate to the updated rate. Note that if a hook
|
||
|
* was removed then the list's internal rate is reset to the default.
|
||
|
*/
|
||
|
audiohook_list->list_internal_samp_rate = internal_sample_rate;
|
||
|
}
|
||
|
|
||
|
return end_frame;
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_write_list_empty(struct ast_audiohook_list *audiohook_list)
|
||
|
{
|
||
|
return !audiohook_list
|
||
|
|| (AST_LIST_EMPTY(&audiohook_list->spy_list)
|
||
|
&& AST_LIST_EMPTY(&audiohook_list->whisper_list)
|
||
|
&& AST_LIST_EMPTY(&audiohook_list->manipulate_list));
|
||
|
}
|
||
|
|
||
|
struct ast_frame *ast_audiohook_write_list(struct ast_channel *chan, struct ast_audiohook_list *audiohook_list, enum ast_audiohook_direction direction, struct ast_frame *frame)
|
||
|
{
|
||
|
/* Pass off frame to it's respective list write function */
|
||
|
if (frame->frametype == AST_FRAME_VOICE) {
|
||
|
return audio_audiohook_write_list(chan, audiohook_list, direction, frame);
|
||
|
} else if (frame->frametype == AST_FRAME_DTMF) {
|
||
|
return dtmf_audiohook_write_list(chan, audiohook_list, direction, frame);
|
||
|
} else {
|
||
|
return frame;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/*! \brief Wait for audiohook trigger to be triggered
|
||
|
* \param audiohook Audiohook to wait on
|
||
|
*/
|
||
|
void ast_audiohook_trigger_wait(struct ast_audiohook *audiohook)
|
||
|
{
|
||
|
struct timeval wait;
|
||
|
struct timespec ts;
|
||
|
|
||
|
wait = ast_tvadd(ast_tvnow(), ast_samp2tv(50000, 1000));
|
||
|
ts.tv_sec = wait.tv_sec;
|
||
|
ts.tv_nsec = wait.tv_usec * 1000;
|
||
|
|
||
|
ast_cond_timedwait(&audiohook->trigger, &audiohook->lock, &ts);
|
||
|
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
/* Count number of channel audiohooks by type, regardless of type */
|
||
|
int ast_channel_audiohook_count_by_source(struct ast_channel *chan, const char *source, enum ast_audiohook_type type)
|
||
|
{
|
||
|
int count = 0;
|
||
|
struct ast_audiohook *ah = NULL;
|
||
|
|
||
|
if (!ast_channel_audiohooks(chan)) {
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
switch (type) {
|
||
|
case AST_AUDIOHOOK_TYPE_SPY:
|
||
|
AST_LIST_TRAVERSE(&ast_channel_audiohooks(chan)->spy_list, ah, list) {
|
||
|
if (!strcmp(ah->source, source)) {
|
||
|
count++;
|
||
|
}
|
||
|
}
|
||
|
break;
|
||
|
case AST_AUDIOHOOK_TYPE_WHISPER:
|
||
|
AST_LIST_TRAVERSE(&ast_channel_audiohooks(chan)->whisper_list, ah, list) {
|
||
|
if (!strcmp(ah->source, source)) {
|
||
|
count++;
|
||
|
}
|
||
|
}
|
||
|
break;
|
||
|
case AST_AUDIOHOOK_TYPE_MANIPULATE:
|
||
|
AST_LIST_TRAVERSE(&ast_channel_audiohooks(chan)->manipulate_list, ah, list) {
|
||
|
if (!strcmp(ah->source, source)) {
|
||
|
count++;
|
||
|
}
|
||
|
}
|
||
|
break;
|
||
|
default:
|
||
|
ast_debug(1, "Invalid audiohook type supplied, (%u)\n", type);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
return count;
|
||
|
}
|
||
|
|
||
|
/* Count number of channel audiohooks by type that are running */
|
||
|
int ast_channel_audiohook_count_by_source_running(struct ast_channel *chan, const char *source, enum ast_audiohook_type type)
|
||
|
{
|
||
|
int count = 0;
|
||
|
struct ast_audiohook *ah = NULL;
|
||
|
if (!ast_channel_audiohooks(chan))
|
||
|
return -1;
|
||
|
|
||
|
switch (type) {
|
||
|
case AST_AUDIOHOOK_TYPE_SPY:
|
||
|
AST_LIST_TRAVERSE(&ast_channel_audiohooks(chan)->spy_list, ah, list) {
|
||
|
if ((!strcmp(ah->source, source)) && (ah->status == AST_AUDIOHOOK_STATUS_RUNNING))
|
||
|
count++;
|
||
|
}
|
||
|
break;
|
||
|
case AST_AUDIOHOOK_TYPE_WHISPER:
|
||
|
AST_LIST_TRAVERSE(&ast_channel_audiohooks(chan)->whisper_list, ah, list) {
|
||
|
if ((!strcmp(ah->source, source)) && (ah->status == AST_AUDIOHOOK_STATUS_RUNNING))
|
||
|
count++;
|
||
|
}
|
||
|
break;
|
||
|
case AST_AUDIOHOOK_TYPE_MANIPULATE:
|
||
|
AST_LIST_TRAVERSE(&ast_channel_audiohooks(chan)->manipulate_list, ah, list) {
|
||
|
if ((!strcmp(ah->source, source)) && (ah->status == AST_AUDIOHOOK_STATUS_RUNNING))
|
||
|
count++;
|
||
|
}
|
||
|
break;
|
||
|
default:
|
||
|
ast_debug(1, "Invalid audiohook type supplied, (%u)\n", type);
|
||
|
return -1;
|
||
|
}
|
||
|
return count;
|
||
|
}
|
||
|
|
||
|
/*! \brief Audiohook volume adjustment structure */
|
||
|
struct audiohook_volume {
|
||
|
struct ast_audiohook audiohook; /*!< Audiohook attached to the channel */
|
||
|
int read_adjustment; /*!< Value to adjust frames read from the channel by */
|
||
|
int write_adjustment; /*!< Value to adjust frames written to the channel by */
|
||
|
};
|
||
|
|
||
|
/*! \brief Callback used to destroy the audiohook volume datastore
|
||
|
* \param data Volume information structure
|
||
|
*/
|
||
|
static void audiohook_volume_destroy(void *data)
|
||
|
{
|
||
|
struct audiohook_volume *audiohook_volume = data;
|
||
|
|
||
|
/* Destroy the audiohook as it is no longer in use */
|
||
|
ast_audiohook_destroy(&audiohook_volume->audiohook);
|
||
|
|
||
|
/* Finally free ourselves, we are of no more use */
|
||
|
ast_free(audiohook_volume);
|
||
|
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
/*! \brief Datastore used to store audiohook volume information */
|
||
|
static const struct ast_datastore_info audiohook_volume_datastore = {
|
||
|
.type = "Volume",
|
||
|
.destroy = audiohook_volume_destroy,
|
||
|
};
|
||
|
|
||
|
/*! \brief Helper function which actually gets called by audiohooks to perform the adjustment
|
||
|
* \param audiohook Audiohook attached to the channel
|
||
|
* \param chan Channel we are attached to
|
||
|
* \param frame Frame of audio we want to manipulate
|
||
|
* \param direction Direction the audio came in from
|
||
|
* \retval 0 on success
|
||
|
* \retval -1 on failure
|
||
|
*/
|
||
|
static int audiohook_volume_callback(struct ast_audiohook *audiohook, struct ast_channel *chan, struct ast_frame *frame, enum ast_audiohook_direction direction)
|
||
|
{
|
||
|
struct ast_datastore *datastore = NULL;
|
||
|
struct audiohook_volume *audiohook_volume = NULL;
|
||
|
int *gain = NULL;
|
||
|
|
||
|
/* If the audiohook is shutting down don't even bother */
|
||
|
if (audiohook->status == AST_AUDIOHOOK_STATUS_DONE) {
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
/* Try to find the datastore containg adjustment information, if we can't just bail out */
|
||
|
if (!(datastore = ast_channel_datastore_find(chan, &audiohook_volume_datastore, NULL))) {
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
audiohook_volume = datastore->data;
|
||
|
|
||
|
/* Based on direction grab the appropriate adjustment value */
|
||
|
if (direction == AST_AUDIOHOOK_DIRECTION_READ) {
|
||
|
gain = &audiohook_volume->read_adjustment;
|
||
|
} else if (direction == AST_AUDIOHOOK_DIRECTION_WRITE) {
|
||
|
gain = &audiohook_volume->write_adjustment;
|
||
|
}
|
||
|
|
||
|
/* If an adjustment value is present modify the frame */
|
||
|
if (gain && *gain) {
|
||
|
ast_frame_adjust_volume(frame, *gain);
|
||
|
}
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
/*! \brief Helper function which finds and optionally creates an audiohook_volume_datastore datastore on a channel
|
||
|
* \param chan Channel to look on
|
||
|
* \param create Whether to create the datastore if not found
|
||
|
* \return audiohook_volume structure on success
|
||
|
* \retval NULL on failure
|
||
|
*/
|
||
|
static struct audiohook_volume *audiohook_volume_get(struct ast_channel *chan, int create)
|
||
|
{
|
||
|
struct ast_datastore *datastore = NULL;
|
||
|
struct audiohook_volume *audiohook_volume = NULL;
|
||
|
|
||
|
/* If we are able to find the datastore return the contents (which is actually an audiohook_volume structure) */
|
||
|
if ((datastore = ast_channel_datastore_find(chan, &audiohook_volume_datastore, NULL))) {
|
||
|
return datastore->data;
|
||
|
}
|
||
|
|
||
|
/* If we are not allowed to create a datastore or if we fail to create a datastore, bail out now as we have nothing for them */
|
||
|
if (!create || !(datastore = ast_datastore_alloc(&audiohook_volume_datastore, NULL))) {
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
/* Create a new audiohook_volume structure to contain our adjustments and audiohook */
|
||
|
if (!(audiohook_volume = ast_calloc(1, sizeof(*audiohook_volume)))) {
|
||
|
ast_datastore_free(datastore);
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
/* Setup our audiohook structure so we can manipulate the audio */
|
||
|
ast_audiohook_init(&audiohook_volume->audiohook, AST_AUDIOHOOK_TYPE_MANIPULATE, "Volume", AST_AUDIOHOOK_MANIPULATE_ALL_RATES);
|
||
|
audiohook_volume->audiohook.manipulate_callback = audiohook_volume_callback;
|
||
|
|
||
|
/* Attach the audiohook_volume blob to the datastore and attach to the channel */
|
||
|
datastore->data = audiohook_volume;
|
||
|
ast_channel_datastore_add(chan, datastore);
|
||
|
|
||
|
/* All is well... put the audiohook into motion */
|
||
|
ast_audiohook_attach(chan, &audiohook_volume->audiohook);
|
||
|
|
||
|
return audiohook_volume;
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_volume_set(struct ast_channel *chan, enum ast_audiohook_direction direction, int volume)
|
||
|
{
|
||
|
struct audiohook_volume *audiohook_volume = NULL;
|
||
|
|
||
|
/* Attempt to find the audiohook volume information, but only create it if we are not setting the adjustment value to zero */
|
||
|
if (!(audiohook_volume = audiohook_volume_get(chan, (volume ? 1 : 0)))) {
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
/* Now based on the direction set the proper value */
|
||
|
if (direction == AST_AUDIOHOOK_DIRECTION_READ || direction == AST_AUDIOHOOK_DIRECTION_BOTH) {
|
||
|
audiohook_volume->read_adjustment = volume;
|
||
|
}
|
||
|
if (direction == AST_AUDIOHOOK_DIRECTION_WRITE || direction == AST_AUDIOHOOK_DIRECTION_BOTH) {
|
||
|
audiohook_volume->write_adjustment = volume;
|
||
|
}
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_volume_get(struct ast_channel *chan, enum ast_audiohook_direction direction)
|
||
|
{
|
||
|
struct audiohook_volume *audiohook_volume = NULL;
|
||
|
int adjustment = 0;
|
||
|
|
||
|
/* Attempt to find the audiohook volume information, but do not create it as we only want to look at the values */
|
||
|
if (!(audiohook_volume = audiohook_volume_get(chan, 0))) {
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
/* Grab the adjustment value based on direction given */
|
||
|
if (direction == AST_AUDIOHOOK_DIRECTION_READ) {
|
||
|
adjustment = audiohook_volume->read_adjustment;
|
||
|
} else if (direction == AST_AUDIOHOOK_DIRECTION_WRITE) {
|
||
|
adjustment = audiohook_volume->write_adjustment;
|
||
|
}
|
||
|
|
||
|
return adjustment;
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_volume_adjust(struct ast_channel *chan, enum ast_audiohook_direction direction, int volume)
|
||
|
{
|
||
|
struct audiohook_volume *audiohook_volume = NULL;
|
||
|
|
||
|
/* Attempt to find the audiohook volume information, and create an audiohook if none exists */
|
||
|
if (!(audiohook_volume = audiohook_volume_get(chan, 1))) {
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
/* Based on the direction change the specific adjustment value */
|
||
|
if (direction == AST_AUDIOHOOK_DIRECTION_READ || direction == AST_AUDIOHOOK_DIRECTION_BOTH) {
|
||
|
audiohook_volume->read_adjustment += volume;
|
||
|
}
|
||
|
if (direction == AST_AUDIOHOOK_DIRECTION_WRITE || direction == AST_AUDIOHOOK_DIRECTION_BOTH) {
|
||
|
audiohook_volume->write_adjustment += volume;
|
||
|
}
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
int ast_audiohook_set_mute(struct ast_channel *chan, const char *source, enum ast_audiohook_flags flag, int clear)
|
||
|
{
|
||
|
struct ast_audiohook *audiohook = NULL;
|
||
|
|
||
|
ast_channel_lock(chan);
|
||
|
|
||
|
/* Ensure the channel has audiohooks on it */
|
||
|
if (!ast_channel_audiohooks(chan)) {
|
||
|
ast_channel_unlock(chan);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
audiohook = find_audiohook_by_source(ast_channel_audiohooks(chan), source);
|
||
|
|
||
|
if (audiohook) {
|
||
|
if (clear) {
|
||
|
ast_clear_flag(audiohook, flag);
|
||
|
} else {
|
||
|
ast_set_flag(audiohook, flag);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
ast_channel_unlock(chan);
|
||
|
|
||
|
return (audiohook ? 0 : -1);
|
||
|
}
|