libevent / libevent/libevent

When to free bufferevent+http+mbedtls context

Open
#1,808 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C
Stars
12k
Forks
3.5k
Avg merge
12h 55m
Merged PRs (30d)
6

Description

I'm creating a HTTP request with evhttp_request_new, which is made using a HTTP connection associated to a bufferevent. As I understand, this will ultimately trigger an event chain in bufferevent, which can invoke events even after we call evhttp_connection_free (which in turn calls bufferevent_free). If I'm not mistaken, this is also true even if we are doing evhttp_connection_free_on_completion, to avoid manually freeing stuff.

You can see my code here, which is based on this sample.

As you can see in the code, to keep things simple, I am storing all mbedtls context in a struct that is passed as an argument to evhttp_request_new. This allows for a context-free rest header.

Then, my first idea was to free that context later in its callback, which will cause a crash because of the mentioned chain of events that might be invoked after the callback and even after evhttp_connection_free. This is the context:

typedef struct {
    mbedtls_x509_crt cacert;
    mbedtls_ctr_drbg_context ctr_drbg;
    mbedtls_entropy_context entropy;
    mbedtls_ssl_config config;

    struct bufferevent *bev;
    struct evhttp_connection *evcon;
    mbedtls_dyncontext *ssl;
} mbedtls_ctx_t;

Specifically, the crash is happening because of doing this...:

    mbedtls_x509_crt_free(&ctx->cacert);
    mbedtls_ssl_config_free(&ctx->config);
    mbedtls_ctr_drbg_free(&ctx->ctr_drbg);
    mbedtls_entropy_free(&ctx->entropy);

    free(ctx);

...before, internally, the library executing bufferevent_finalize_cb_, which is calling some destruct pointer function that ends up calling bufferevent_mbedtls_dyncontext_free.

To solve this, I ended up adding a new callback that is executed once bufferevent has finished all of its chain of events (I think) and once bufferevent_mbedtls_dyncontext_free has been executed by the library (that's for sure). See the patch here.

Is this a good approach to solve this problem or am I overkilling it?

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Read event/event_rest.c alongside sample/https-client.c, then trace bufferevent_finalize_cb_ and bufferevent_mbedtls_dyncontext_free to establish the callback and cleanup order. Review the proposed callback in the linked patch and determine whether it provides a safe ownership boundary. Done means documenting or implementing an agreed lifetime strategy without the reported use-after-free crash.

Written by the indexing model from the issue text.

Assessment

Tech stack
c
Domain
networking
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.