From fe7e48db03de03c232cc5af28e37ce08c5ccd3d8 Mon Sep 17 00:00:00 2001 From: George Joseph Date: Fri, 1 Apr 2016 12:30:56 -0600 Subject: res_pjsip contact: Lock expiration/addition of contacts Contact expiration can occur in several places: res_pjsip_registrar, res_pjsip_registrar_expire, and automatically when anyone calls ast_sip_location_retrieve_aor_contact. At the same time, res_pjsip_registrar may also be attempting to renew or add a contact. Since none of this was locked it was possible for one thread to be renewing a contact and another thread to expire it immediately because it was working off of stale data. This was the casue of intermittent registration/inbound/nominal/multiple_contacts test failures. Now, the new named lock functionality is used to lock the aor during contact expire and add operations and res_pjsip_registrar_expire now checks the expiration with the lock held before deleting the contact. ASTERISK-25885 #close Reported-by: Josh Colp Change-Id: I83d413c46a47796f3ab052ca3b349f21cca47059 --- res/res_pjsip/location.c | 44 ++++++++++++++++++++++++++++++-- res/res_pjsip_registrar.c | 55 ++++++++++++++++++++++++++++++---------- res/res_pjsip_registrar_expire.c | 18 ++++++++++++- 3 files changed, 100 insertions(+), 17 deletions(-) (limited to 'res') diff --git a/res/res_pjsip/location.c b/res/res_pjsip/location.c index 735b18f46..3413708e9 100644 --- a/res/res_pjsip/location.c +++ b/res/res_pjsip/location.c @@ -27,6 +27,7 @@ #include "include/res_pjsip_private.h" #include "asterisk/res_pjsip_cli.h" #include "asterisk/statsd.h" +#include "asterisk/named_locks.h" /*! \brief Destructor for AOR */ static void aor_destroy(void *obj) @@ -173,7 +174,7 @@ struct ast_sip_contact *ast_sip_location_retrieve_first_aor_contact(const struct return contact; } -struct ao2_container *ast_sip_location_retrieve_aor_contacts(const struct ast_sip_aor *aor) +struct ao2_container *ast_sip_location_retrieve_aor_contacts_nolock(const struct ast_sip_aor *aor) { /* Give enough space for ^ at the beginning and ;@ at the end, since that is our object naming scheme */ char regex[strlen(ast_sorcery_object_get_id(aor)) + 4]; @@ -196,6 +197,24 @@ struct ao2_container *ast_sip_location_retrieve_aor_contacts(const struct ast_si return contacts; } +struct ao2_container *ast_sip_location_retrieve_aor_contacts(const struct ast_sip_aor *aor) +{ + struct ao2_container *contacts; + struct ast_named_lock *lock; + + lock = ast_named_lock_get(AST_NAMED_LOCK_TYPE_RWLOCK, "aor", ast_sorcery_object_get_id(aor)); + if (!lock) { + return NULL; + } + + ao2_wrlock(lock); + contacts = ast_sip_location_retrieve_aor_contacts_nolock(aor); + ao2_unlock(lock); + ast_named_lock_put(lock); + + return contacts; +} + void ast_sip_location_retrieve_contact_and_aor_from_list(const char *aor_list, struct ast_sip_aor **aor, struct ast_sip_contact **contact) { @@ -279,7 +298,7 @@ struct ast_sip_contact *ast_sip_location_retrieve_contact(const char *contact_na return ast_sorcery_retrieve_by_id(ast_sip_get_sorcery(), "contact", contact_name); } -int ast_sip_location_add_contact(struct ast_sip_aor *aor, const char *uri, +int ast_sip_location_add_contact_nolock(struct ast_sip_aor *aor, const char *uri, struct timeval expiration_time, const char *path_info, const char *user_agent, struct ast_sip_endpoint *endpoint) { @@ -316,6 +335,27 @@ int ast_sip_location_add_contact(struct ast_sip_aor *aor, const char *uri, return ast_sorcery_create(ast_sip_get_sorcery(), contact); } +int ast_sip_location_add_contact(struct ast_sip_aor *aor, const char *uri, + struct timeval expiration_time, const char *path_info, const char *user_agent, + struct ast_sip_endpoint *endpoint) +{ + int res; + struct ast_named_lock *lock; + + lock = ast_named_lock_get(AST_NAMED_LOCK_TYPE_RWLOCK, "aor", ast_sorcery_object_get_id(aor)); + if (!lock) { + return -1; + } + + ao2_wrlock(lock); + res = ast_sip_location_add_contact_nolock(aor, uri, expiration_time, path_info, user_agent, + endpoint); + ao2_unlock(lock); + ast_named_lock_put(lock); + + return res; +} + int ast_sip_location_update_contact(struct ast_sip_contact *contact) { return ast_sorcery_update(ast_sip_get_sorcery(), contact); diff --git a/res/res_pjsip_registrar.c b/res/res_pjsip_registrar.c index fb2b9daba..4a42102dc 100644 --- a/res/res_pjsip_registrar.c +++ b/res/res_pjsip_registrar.c @@ -32,6 +32,7 @@ #include "asterisk/test.h" #include "asterisk/taskprocessor.h" #include "asterisk/manager.h" +#include "asterisk/named_locks.h" #include "res_pjsip/include/res_pjsip_private.h" /*** DOCUMENTATION @@ -412,27 +413,21 @@ static int registrar_validate_path(struct rx_task_data *task_data, struct ast_st return -1; } -static int rx_task(void *data) +static int rx_task_core(struct rx_task_data *task_data, struct ao2_container *contacts, + const char *aor_name) { static const pj_str_t USER_AGENT = { "User-Agent", 10 }; - RAII_VAR(struct rx_task_data *, task_data, data, ao2_cleanup); - RAII_VAR(struct ao2_container *, contacts, NULL, ao2_cleanup); - int added = 0, updated = 0, deleted = 0; pjsip_contact_hdr *contact_hdr = NULL; struct registrar_contact_details details = { 0, }; pjsip_tx_data *tdata; - const char *aor_name = ast_sorcery_object_get_id(task_data->aor); RAII_VAR(struct ast_str *, path_str, NULL, ast_free); struct ast_sip_contact *response_contact; char *user_agent = NULL; pjsip_user_agent_hdr *user_agent_hdr; pjsip_expires_hdr *expires_hdr; - /* Retrieve the current contacts, we'll need to know whether to update or not */ - contacts = ast_sip_location_retrieve_aor_contacts(task_data->aor); - /* So we don't count static contacts against max_contacts we prune them out from the container */ ao2_callback(contacts, OBJ_NODATA | OBJ_UNLINK | OBJ_MULTIPLE, registrar_prune_static, NULL); @@ -503,7 +498,7 @@ static int rx_task(void *data) continue; } - if (ast_sip_location_add_contact(task_data->aor, contact_uri, ast_tvadd(ast_tvnow(), + if (ast_sip_location_add_contact_nolock(task_data->aor, contact_uri, ast_tvadd(ast_tvnow(), ast_samp2tv(expiration, 1)), path_str ? ast_str_buffer(path_str) : NULL, user_agent, task_data->endpoint)) { ast_log(LOG_ERROR, "Unable to bind contact '%s' to AOR '%s'\n", @@ -545,7 +540,7 @@ static int rx_task(void *data) if (ast_sip_location_update_contact(contact_update)) { ast_log(LOG_ERROR, "Failed to update contact '%s' expiration time to %d seconds.\n", contact->uri, expiration); - ast_sorcery_delete(ast_sip_get_sorcery(), contact); + ast_sip_location_delete_contact(contact); continue; } ast_debug(3, "Refreshed contact '%s' on AOR '%s' with new expiration of %d seconds\n", @@ -584,15 +579,14 @@ static int rx_task(void *data) ao2_callback(contacts, OBJ_NODATA | OBJ_MULTIPLE, registrar_delete_contact, NULL); } - /* Update the contacts as things will probably have changed */ - ao2_cleanup(contacts); - - contacts = ast_sip_location_retrieve_aor_contacts(task_data->aor); + /* Re-retrieve contacts. Caller will clean up the original container. */ + contacts = ast_sip_location_retrieve_aor_contacts_nolock(task_data->aor); response_contact = ao2_callback(contacts, 0, NULL, NULL); /* Send a response containing all of the contacts (including static) that are present on this AOR */ if (ast_sip_create_response(task_data->rdata, 200, response_contact, &tdata) != PJ_SUCCESS) { ao2_cleanup(response_contact); + ao2_cleanup(contacts); return PJ_TRUE; } ao2_cleanup(response_contact); @@ -601,6 +595,7 @@ static int rx_task(void *data) registrar_add_date_header(tdata); ao2_callback(contacts, 0, registrar_add_contact, tdata); + ao2_cleanup(contacts); if ((expires_hdr = pjsip_msg_find_hdr(task_data->rdata->msg_info.msg, PJSIP_H_EXPIRES, NULL))) { expires_hdr = pjsip_expires_hdr_create(tdata->pool, registrar_get_expiration(task_data->aor, NULL, task_data->rdata)); @@ -612,6 +607,38 @@ static int rx_task(void *data) return PJ_TRUE; } +static int rx_task(void *data) +{ + int res; + struct rx_task_data *task_data = data; + struct ao2_container *contacts = NULL; + struct ast_named_lock *lock; + const char *aor_name = ast_sorcery_object_get_id(task_data->aor); + + lock = ast_named_lock_get(AST_NAMED_LOCK_TYPE_RWLOCK, "aor", aor_name); + if (!lock) { + ao2_cleanup(task_data); + return PJ_TRUE; + } + + ao2_wrlock(lock); + contacts = ast_sip_location_retrieve_aor_contacts_nolock(task_data->aor); + if (!contacts) { + ao2_unlock(lock); + ast_named_lock_put(lock); + ao2_cleanup(task_data); + return PJ_TRUE; + } + + res = rx_task_core(task_data, contacts, aor_name); + ao2_cleanup(contacts); + ao2_unlock(lock); + ast_named_lock_put(lock); + ao2_cleanup(task_data); + + return res; +} + static pj_bool_t registrar_on_rx_request(struct pjsip_rx_data *rdata) { RAII_VAR(struct serializer *, ser, NULL, ao2_cleanup); diff --git a/res/res_pjsip_registrar_expire.c b/res/res_pjsip_registrar_expire.c index 97f2ca1c0..0fddbda91 100644 --- a/res/res_pjsip_registrar_expire.c +++ b/res/res_pjsip_registrar_expire.c @@ -30,6 +30,7 @@ #include "asterisk/res_pjsip.h" #include "asterisk/module.h" +#include "asterisk/named_locks.h" /*! \brief Thread keeping things alive */ static pthread_t check_thread = AST_PTHREADT_NULL; @@ -41,8 +42,23 @@ static unsigned int check_interval; static int expire_contact(void *obj, void *arg, int flags) { struct ast_sip_contact *contact = obj; + struct ast_named_lock *lock; - ast_sorcery_delete(ast_sip_get_sorcery(), contact); + lock = ast_named_lock_get(AST_NAMED_LOCK_TYPE_RWLOCK, "aor", contact->aor); + if (!lock) { + return 0; + } + + /* + * We need to check the expiration again with the aor lock held + * in case another thread is attempting to renew the contact. + */ + ao2_wrlock(lock); + if (ast_tvdiff_ms(ast_tvnow(), contact->expiration_time) > 0) { + ast_sip_location_delete_contact(contact); + } + ao2_unlock(lock); + ast_named_lock_put(lock); return 0; } -- cgit v1.2.3