Re: [Qemu-devel] [PATCH v4] iothread: fix epollfd leak in the process of delIOThread

[Fam Zheng]

On Wed, 05/16 19:43, Peter Xu wrote:
> On Wed, May 16, 2018 at 07:14:53PM +0800, WangJie (Pluto) wrote:
> > Hi, Peter Xu:
> > If call aio_epoll_disable() here, aio_epoll_disable() will return 
> > before close ctx->epollfd,
> > Because the ctx->epoll_enabled is false in the moment.
> > In the process of addIOThread, aio_context_setup created epoll without 
> > call aio_epoll_try_enable,
> > so ctx->epoll_enabled have no chance to set true.
> 
> I see that epoll_available will only be set if epollfd != -1, so it
> seems to me to make more sense if we swap the two variables in
> aio_epoll_disable(), from current version:
> 
> static void aio_epoll_disable(AioContext *ctx)
> {
> ctx->epoll_available = false;
> if (!ctx->epoll_enabled) {
> return;
> }
> ctx->epoll_enabled = false;
> close(ctx->epollfd);
> }
> 
> To:
> 
> static void aio_epoll_disable(AioContext *ctx)
> {
> ctx->epoll_enabled = false;
> if (!ctx->epoll_available) {
> return;
> }
> ctx->epoll_available = false;
> close(ctx->epollfd);
> }
> 
> What do you think?  And Fam?

Looks good.

Fam

> 
> > 
> > On 2018/5/16 16:36, Jie Wang wrote:
> > > +void aio_context_destroy(AioContext *ctx)
> > > +{
> > > +#ifdef CONFIG_EPOLL_CREATE1
> > > +if (ctx->epollfd >= 0) {
> > > +close(ctx->epollfd);
> > > +}
> > > +#endif
> > > +}
> > > +
> > >  void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
> > >   int64_t grow, int64_t shrink, Error 
> > > **errp)
> > 
> 
> -- 
> Peter Xu

Re: [Qemu-devel] [PATCH v4] iothread: fix epollfd leak in the process of delIOThread

[WangJie (Pluto)]

I agree, wait for a reply from Fam

On 2018/5/16 19:43, Peter Xu wrote:
> On Wed, May 16, 2018 at 07:14:53PM +0800, WangJie (Pluto) wrote:
>> Hi, Peter Xu:
>>  If call aio_epoll_disable() here, aio_epoll_disable() will return 
>> before close ctx->epollfd,
>> Because the ctx->epoll_enabled is false in the moment.
>>  In the process of addIOThread, aio_context_setup created epoll without 
>> call aio_epoll_try_enable,
>> so ctx->epoll_enabled have no chance to set true.
> 
> I see that epoll_available will only be set if epollfd != -1, so it
> seems to me to make more sense if we swap the two variables in
> aio_epoll_disable(), from current version:
> 
> static void aio_epoll_disable(AioContext *ctx)
> {
> ctx->epoll_available = false;
> if (!ctx->epoll_enabled) {
> return;
> }
> ctx->epoll_enabled = false;
> close(ctx->epollfd);
> }
> 
> To:
> 
> static void aio_epoll_disable(AioContext *ctx)
> {
> ctx->epoll_enabled = false;
> if (!ctx->epoll_available) {
> return;
> }
> ctx->epoll_available = false;
> close(ctx->epollfd);
> }
> 
> What do you think?  And Fam?
> 
>>
>> On 2018/5/16 16:36, Jie Wang wrote:
>>> +void aio_context_destroy(AioContext *ctx)
>>> +{
>>> +#ifdef CONFIG_EPOLL_CREATE1
>>> +if (ctx->epollfd >= 0) {
>>> +close(ctx->epollfd);
>>> +}
>>> +#endif
>>> +}
>>> +
>>>  void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
>>>   int64_t grow, int64_t shrink, Error 
>>> **errp)
>>
> 

Re: [Qemu-devel] [PATCH v4] iothread: fix epollfd leak in the process of delIOThread

[Peter Xu]

On Wed, May 16, 2018 at 07:14:53PM +0800, WangJie (Pluto) wrote:
> Hi, Peter Xu:
>   If call aio_epoll_disable() here, aio_epoll_disable() will return 
> before close ctx->epollfd,
> Because the ctx->epoll_enabled is false in the moment.
>   In the process of addIOThread, aio_context_setup created epoll without 
> call aio_epoll_try_enable,
> so ctx->epoll_enabled have no chance to set true.

I see that epoll_available will only be set if epollfd != -1, so it
seems to me to make more sense if we swap the two variables in
aio_epoll_disable(), from current version:

static void aio_epoll_disable(AioContext *ctx)
{
ctx->epoll_available = false;
if (!ctx->epoll_enabled) {
return;
}
ctx->epoll_enabled = false;
close(ctx->epollfd);
}

To:

static void aio_epoll_disable(AioContext *ctx)
{
ctx->epoll_enabled = false;
if (!ctx->epoll_available) {
return;
}
ctx->epoll_available = false;
close(ctx->epollfd);
}

What do you think?  And Fam?

> 
> On 2018/5/16 16:36, Jie Wang wrote:
> > +void aio_context_destroy(AioContext *ctx)
> > +{
> > +#ifdef CONFIG_EPOLL_CREATE1
> > +if (ctx->epollfd >= 0) {
> > +close(ctx->epollfd);
> > +}
> > +#endif
> > +}
> > +
> >  void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
> >   int64_t grow, int64_t shrink, Error 
> > **errp)
> 

-- 
Peter Xu

Re: [Qemu-devel] [PATCH v4] iothread: fix epollfd leak in the process of delIOThread

[WangJie (Pluto)]

Hi, Peter Xu:
If call aio_epoll_disable() in aio_context_destroy, aio_epoll_disable() 
will return before close(ctx->epollfd),
Because the ctx->epoll_enabled is false in the moment.
In the process of addIOThread, aio_context_setup created epoll without 
call aio_epoll_try_enable,
so ctx->epoll_enabled have no chance to set true.


On 2018/5/16 16:36, Jie Wang wrote:
> When we call addIOThread, the epollfd created in aio_context_setup,
> but not close it in the process of delIOThread, so the epollfd will leak.
> 
> Signed-off-by: Jie Wang 
> ---
>  include/block/aio.h | 8 
>  util/aio-posix.c| 9 +
>  util/aio-win32.c| 4 
>  util/async.c| 1 +
>  4 files changed, 22 insertions(+)
> 
> diff --git a/include/block/aio.h b/include/block/aio.h
> index a1d6b9e..ae6f354 100644
> --- a/include/block/aio.h
> +++ b/include/block/aio.h
> @@ -555,6 +555,14 @@ static inline bool in_aio_context_home_thread(AioContext 
> *ctx)
>  void aio_context_setup(AioContext *ctx);
>  
>  /**
> + * aio_context_destroy:
> + * @ctx: the aio context
> + *
> + * Destroy the aio context.
> + */
> +void aio_context_destroy(AioContext *ctx);
> +
> +/**
>   * aio_context_set_poll_params:
>   * @ctx: the aio context
>   * @max_ns: how long to busy poll for, in nanoseconds
> diff --git a/util/aio-posix.c b/util/aio-posix.c
> index d8f0cb4..0ade2c7 100644
> --- a/util/aio-posix.c
> +++ b/util/aio-posix.c
> @@ -713,6 +713,15 @@ void aio_context_setup(AioContext *ctx)
>  #endif
>  }
>  
> +void aio_context_destroy(AioContext *ctx)
> +{
> +#ifdef CONFIG_EPOLL_CREATE1
> +if (ctx->epollfd >= 0) {
> +close(ctx->epollfd);
> +}
> +#endif
> +}
> +
>  void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
>   int64_t grow, int64_t shrink, Error **errp)
>  {
> diff --git a/util/aio-win32.c b/util/aio-win32.c
> index a67b00c..e676a8d 100644
> --- a/util/aio-win32.c
> +++ b/util/aio-win32.c
> @@ -407,6 +407,10 @@ void aio_context_setup(AioContext *ctx)
>  {
>  }
>  
> +void aio_context_destroy(AioContext *ctx)
> +{
> +}
> +
>  void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
>   int64_t grow, int64_t shrink, Error **errp)
>  {
> diff --git a/util/async.c b/util/async.c
> index 4dd9d95..03f6278 100644
> --- a/util/async.c
> +++ b/util/async.c
> @@ -298,6 +298,7 @@ aio_ctx_finalize(GSource *source)
>  qemu_rec_mutex_destroy(>lock);
>  qemu_lockcnt_destroy(>list_lock);
>  timerlistgroup_deinit(>tlg);
> +aio_context_destroy(ctx);
>  }
>  
>  static GSourceFuncs aio_source_funcs = {
> 

Re: [Qemu-devel] [PATCH v4] iothread: fix epollfd leak in the process of delIOThread

[WangJie (Pluto)]

Hi, Peter Xu:
If call aio_epoll_disable() here, aio_epoll_disable() will return 
before close ctx->epollfd,
Because the ctx->epoll_enabled is false in the moment.
In the process of addIOThread, aio_context_setup created epoll without 
call aio_epoll_try_enable,
so ctx->epoll_enabled have no chance to set true.

On 2018/5/16 16:36, Jie Wang wrote:
> +void aio_context_destroy(AioContext *ctx)
> +{
> +#ifdef CONFIG_EPOLL_CREATE1
> +if (ctx->epollfd >= 0) {
> +close(ctx->epollfd);
> +}
> +#endif
> +}
> +
>  void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
>   int64_t grow, int64_t shrink, Error **errp)