Add a timeout to side requests so they do not hang forever. Tested with SIGSTOPping sssd_be and monitor and then waiting for the request and later the idle timer to fire. Seem to work as expected here.
Simo. -- Simo Sorce * Red Hat, Inc * New York
>From 32f03cebf2570540c9d087c5d6d8fd8e46ec583e Mon Sep 17 00:00:00 2001 From: Simo Sorce <[email protected]> Date: Fri, 14 Dec 2012 14:59:21 -0500 Subject: [PATCH] responder_dp: Add timeout to side requets This is an additional proteciont in case the provider misbheaves to avoid having requests pending forever. Fixes: https://fedorahosted.org/sssd/ticket/1717 --- src/responder/common/responder_dp.c | 26 +++++++++++++++++++++++++- 1 file changed, 25 insertions(+), 1 deletion(-) diff --git a/src/responder/common/responder_dp.c b/src/responder/common/responder_dp.c index ecd5f39cd2aeb5e46f590a44fd76cf3ead65ac49..d15ca4dc9108d410cf750b2685e27fb56fb520a7 100644 --- a/src/responder/common/responder_dp.c +++ b/src/responder/common/responder_dp.c @@ -45,7 +45,6 @@ struct sss_dp_req { hash_key_t *key; - struct tevent_timer *tev; struct sss_dp_callback *cb_list; dbus_uint16_t dp_err; @@ -126,6 +125,15 @@ static int sss_dp_req_destructor(void *ptr) return 0; } +static void sss_dp_req_timeout(struct tevent_context *ev, + struct tevent_timer *te, + struct timeval t, void *ptr) +{ + /* ptr is a pointer to sidereq */ + /* Just free it to kill all waiting requests when the timeout fires */ + talloc_zfree(ptr); +} + void handle_requests_after_reconnect(struct resp_ctx *rctx) { int ret; @@ -252,6 +260,8 @@ sss_dp_issue_request(TALLOC_CTX *mem_ctx, struct resp_ctx *rctx, struct tevent_req *sidereq; struct sss_dp_req *sdp_req; struct sss_dp_callback *cb; + struct tevent_timer *te; + struct timeval tv; DBusMessage *msg; TALLOC_CTX *tmp_ctx = NULL; errno_t ret; @@ -306,6 +316,20 @@ sss_dp_issue_request(TALLOC_CTX *mem_ctx, struct resp_ctx *rctx, } tevent_req_set_callback(sidereq, sss_dp_req_done, NULL); + /* add timeout handling so we do not hang forever should something + * go worng in the provider. Use 2 sec less than the idle timeout to + * give it a chance to reply to the client before closing the + * connection. */ + tv = tevent_timeval_current_ofs(rctx->client_idle_timeout - 2, 0); + te = tevent_add_timer(rctx->ev, sidereq, tv, + sss_dp_req_timeout, sidereq); + if (!te) { + /* Nothing much we can do */ + DEBUG(SSSDBG_CRIT_FAILURE, ("Out of memory?!\n")); + ret = ENOMEM; + goto fail; + } + /* We should now be able to find the sdp_req in the hash table */ hret = hash_lookup(rctx->dp_request_table, key, &value); if (hret != HASH_SUCCESS) { -- 1.8.0.1
_______________________________________________ sssd-devel mailing list [email protected] https://lists.fedorahosted.org/mailman/listinfo/sssd-devel
