Re: [lwip-users] Freeing memory after http transfer

[Bernhard 'Gustl' Bauer]

address@hidden schrieb:
> Bernhard,
> I'm currently (though very infrequently) in the process of working on 
> httpd, so I'm not really sure on which version you are based.

I'm based on lwip 1.3.0

> However, there are two correct ways when to free application buffers:
> - when data is copied with tcp_write(), application data can be freed 
> when tcp_write() returns ERR_OK since then data is bufferd internally 
> (this is left == 0)
> - when data is *not* copied with tcp_write(), these buffers must stay 
> unchanged until the data is actually ACKed (i.e. will not be sent again 
> by lwIP tcp). This is the case when the number of bytes acked (the len 
> argument of http_sent()) equals the size of data you wrote.
>
> Since the httpd in contrib tells tcp_write() to copy data to internal 
> buffers, the version of http_sent below is OK.

tcp_write is used in my copy too. And I see your point. But whenever I 
use the original http_sent the data in the last packet is cleared. I 
just verified it again.

Instead of clearing the memory I just filled it with its addresses. The 
addresses are not in the buffers of the LWIP, so I do not accidentally 
clear the wrong buffers :-)

This is my memcpy:
void pmemcpy(void *dst, void *src, unsigned int len) {
  unsigned int i;
        
  for (i=0;(i+8)<=len;i+=8) {
    memcpy((void *)((unsigned int)dst+i),(void *)((unsigned int)src+i),8);
  }
  if (i<len) {
    memcpy((void *)((unsigned int)dst+i),(void *)((unsigned 
int)src+i),len-i);
  }
}
It is split up to give IRQs a chance.

So it looks still as if the memory is used until the last packed is 
ackowledged.


> Simon
>
>
> Bernhard 'Gustl' Bauer wrote:
> > Kieran Mansley schrieb:
> >  
> > > Each time tcp_sent is called it tells you how much data has been
> > > released.  Keep a count of this.  When your count equals the size of
> > > data you sent, you know it's all done.
> > >     
> >
> > This is the original http_sent:
> >
> > static err_t http_sent(void *arg, struct tcp_pcb *pcb, u16_t len) {
> >    struct http_state *hs;
> >
> >    LWIP_UNUSED_ARG(len);
> >    hs = arg;
> >    hs->retries = 0;
> >
> >    if (hs->left > 0) {
> >      send_data(pcb, hs);
> >    } else {
> >      close_conn(pcb, hs);
> >    }
> >    return ERR_OK;
> > }
> >
> > If hs->left == 0 this does not mean that all data has already been 
> > sent out. The last packet is just somewere in the pipeline. So the 
> > connection    cannot be closed yet.
> >
> > I can remember something puzzled my when doing the first steps: Used 
> > buffers where freed by a function called from the timers. Probably 
> > this was the cause.
> >
> > I would suggest to add bytes_to_send and bytes_acked to struct 
> > http_state and to change the function like this:
> >
> > static err_t http_sent(void *arg, struct tcp_pcb *pcb, u16_t len) {
> >    struct http_state *hs;
> >
> >    LWIP_UNUSED_ARG(len);
> >    hs = arg;
> >    hs->retries = 0;
> >    hs->bytes_acked+=len;
> >     
> >    if (hs->left > 0) {
> >      send_data(pcb, hs);
> >    }
> >    if (hs->bytes_acked==hs->bytes_to_send) {
> >      close_conn(pcb, hs);
> >    }
> >    return ERR_OK;
> > }
> >
> >
> >
> > _______________________________________________
> > lwip-users mailing list
> > address@hidden
> > http://lists.nongnu.org/mailman/listinfo/lwip-users
> >
> >   
>
>
>
> _______________________________________________
> lwip-users mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/lwip-users