asterisk/main/optional_api.c

271 lines
7.0 KiB
C
Raw Permalink Normal View History

optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
/*
* Asterisk -- An open source telephony toolkit.
*
* Copyright (C) 2013, Digium, Inc.
*
* David M. Lee, II <dlee@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.
*/
#include "asterisk.h"
#include "asterisk/optional_api.h"
#include "asterisk/utils.h"
#include "asterisk/vector.h"
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
#if defined(OPTIONAL_API)
/*!
* \file
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
*
* The calls to ast_optional_api_*() happen implicitly from \c __constructor__
* calls which are defined in header files. This means that some number of them
* happen before main() is called. This makes calling most Asterisk APIs
* dangerous, since we could be called before they are initialized. This
* includes things like AO2, malloc debug, and static mutexes.
*
* Another limitation is that most functions are called from the midst of
* dlopen() or dlclose(), and there is no opportunity to return a failure code.
* The best we can do is log an error, and call ast_do_crash().
*
* Fortunately, there are some constraints that help us out. The \c
* ast_optional_api_*() are called during module loads, which happens either
* before main(), or during dlopen() calls. These are already serialized, so we
* don't have to lock ourselves.
*/
/*! \brief A user of an optional API */
struct optional_api_user {
/*! Pointer to function pointer to link */
ast_optional_fn *optional_ref;
/*! Stub to use when impl is unavailable */
ast_optional_fn stub;
/*! Name of the module using the API */
char module[];
};
/*! \brief An optional API */
struct optional_api {
/*! Pointer to the implementation function; could be null */
ast_optional_fn impl;
/*! Users of the API */
AST_VECTOR(, struct optional_api_user *) users;
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
/*! Name of the optional API function */
char symname[];
};
/*! Vector of \ref optional_api functions */
AST_VECTOR(, struct optional_api *) apis;
#define USER_OPTIONAL_REF_CMP(ele, value) (ele->optional_ref == value)
#define OPTIONAL_API_SYMNAME_CMP(ele, value) (!strcmp(ele->symname, value))
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
/*!
* \brief Free an \ref optional_api_user.
*
* \param user User struct to free.
*/
static void optional_api_user_destroy(struct optional_api_user *user)
{
*user->optional_ref = user->stub;
ast_free(user);
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
/*!
* \brief Create an \ref optional_api_user.
*
* \param optional_ref Pointer-to-function-pointer to link to impl/stub.
* \param stub Stub function to link to when impl is not available.
* \param module Name of the module requesting the API.
*
* \return New \ref optional_api_user.
* \retval NULL on error.
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
*/
static struct optional_api_user *optional_api_user_create(
ast_optional_fn *optional_ref, ast_optional_fn stub, const char *module)
{
struct optional_api_user *user;
size_t size = sizeof(*user) + strlen(module) + 1;
user = ast_calloc(1, size);
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
if (!user) {
ast_do_crash();
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
return NULL;
}
user->optional_ref = optional_ref;
user->stub = stub;
strcpy(user->module, module); /* SAFE */
return user;
}
/*!
* \brief Free an \ref optional_api.
*
* \param api API struct to free.
*/
static void optional_api_destroy(struct optional_api *api)
{
AST_VECTOR_REMOVE_CMP_UNORDERED(&apis, api,
AST_VECTOR_ELEM_DEFAULT_CMP, AST_VECTOR_ELEM_CLEANUP_NOOP);
AST_VECTOR_CALLBACK_VOID(&api->users, optional_api_user_destroy);
AST_VECTOR_FREE(&api->users);
ast_free(api);
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
/*!
* \brief Create and link an \ref optional_api.
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
*
* \param symname Name of the optional function.
* \return New \ref optional_api.
* \retval NULL on error.
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
*/
static struct optional_api *optional_api_create(const char *symname)
{
struct optional_api *api;
api = ast_calloc(1, sizeof(*api) + strlen(symname) + 1);
if (!api || AST_VECTOR_APPEND(&apis, api)) {
ast_free(api);
ast_do_crash();
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
return NULL;
}
strcpy(api->symname, symname); /* SAFE */
return api;
}
/*!
* \brief Gets (or creates) the \ref optional_api for the given function.
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
*
* \param symname Name of the function to look up.
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
* \return Corresponding \ref optional_api.
* \retval NULL on error.
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
*/
static struct optional_api *get_api(const char *symname)
{
struct optional_api **api;
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
/* Find one, if we already have it */
api = AST_VECTOR_GET_CMP(&apis, symname, OPTIONAL_API_SYMNAME_CMP);
if (api) {
return *api;
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
/* API not found. Build one */
return optional_api_create(symname);
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
/*!
* \brief Re-links a given \a user against its associated \a api.
*
* If the \a api has an implementation, the \a user is linked to that
* implementation. Otherwise, the \a user is linked to its \a stub.
*
* \param user \ref optional_api_user to link.
* \param api \ref optional_api to link.
*/
static void optional_api_user_relink(struct optional_api_user *user,
struct optional_api *api)
{
if (api->impl && *user->optional_ref != api->impl) {
*user->optional_ref = api->impl;
} else if (!api->impl && *user->optional_ref != user->stub) {
*user->optional_ref = user->stub;
}
}
/*!
* \brief Sets the implementation function pointer for an \a api.
*
* \param api API to implement/stub out.
* \param impl Pointer to implementation function. Can be 0 to remove
* implementation.
*/
static void optional_api_set_impl(struct optional_api *api,
ast_optional_fn impl)
{
api->impl = impl;
/* re-link all users */
if (AST_VECTOR_SIZE(&api->users)) {
AST_VECTOR_CALLBACK_VOID(&api->users, optional_api_user_relink, api);
} else if (!impl) {
/* No users or impl means we should delete this api. */
optional_api_destroy(api);
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
}
void ast_optional_api_provide(const char *symname, ast_optional_fn impl)
{
struct optional_api *api;
api = get_api(symname);
if (api) {
optional_api_set_impl(api, impl);
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
}
void ast_optional_api_unprovide(const char *symname, ast_optional_fn impl)
{
struct optional_api *api;
api = get_api(symname);
if (api) {
optional_api_set_impl(api, 0);
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
}
void ast_optional_api_use(const char *symname, ast_optional_fn *optional_ref,
ast_optional_fn stub, const char *module)
{
struct optional_api_user *user;
struct optional_api *api;
api = get_api(symname);
if (!api) {
return;
}
user = optional_api_user_create(optional_ref, stub, module);
if (!user) {
return;
}
/* Add user to the API */
if (!AST_VECTOR_APPEND(&api->users, user)) {
optional_api_user_relink(user, api);
} else {
optional_api_user_destroy(user);
ast_do_crash();
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
}
void ast_optional_api_unuse(const char *symname, ast_optional_fn *optional_ref,
const char *module)
{
struct optional_api *api;
api = get_api(symname);
if (api) {
AST_VECTOR_REMOVE_CMP_UNORDERED(&api->users, optional_ref, USER_OPTIONAL_REF_CMP, optional_api_user_destroy);
if (!api->impl && !AST_VECTOR_SIZE(&api->users)) {
optional_api_destroy(api);
optional_api: Fix linking problems between modules that export global symbols With the new work in Asterisk 12, there are some uses of the optional_api that are prone to failure. The details are rather involved, and captured on [the wiki][1]. This patch addresses the issue by removing almost all of the magic from the optional API implementation. Instead of relying on weak symbol resolution, a new optional_api.c module was added to Asterisk core. For modules providing an optional API, the pointer to the implementation function is registered with the core. For modules that use an optional API, a pointer to a stub function, along with a optional_ref function pointer are registered with the core. The optional_ref function pointers is set to the implementation function when it's provided, or the stub function when it's now. Since the implementation no longer relies on magic, it is now supported on all platforms. In the spirit of choice, an OPTIONAL_API flag was added, so we can disable the optional_api if needed (maybe it's buggy on some bizarre platform I haven't tested on) The AST_OPTIONAL_API*() macros themselves remained unchanged, so existing code could remain unchanged. But to help with debugging the optional_api, the patch limits the #include of optional API's to just the modules using the API. This also reduces resource waste maintaining optional_ref pointers that aren't used. Other changes made as a part of this patch: * The stubs for http_websocket that wrap system calls set errno to ENOSYS. * res_http_websocket now properly increments module use count. * In loader.c, the while() wrappers around dlclose() were removed. The while(!dlclose()) is actually an anti-pattern, which can lead to infinite loops if the module you're attempting to unload exports a symbol that was directly linked to. * The special handling of nonoptreq on systems without weak symbol support was removed, since we no longer rely on weak symbols for optional_api. [1]: https://wiki.asterisk.org/wiki/x/wACUAQ (closes issue ASTERISK-22296) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2797/ ........ Merged revisions 397989 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397990 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-08-30 13:40:27 +00:00
}
}
}
#endif /* defined(OPTIONAL_API) */