Re: cvs commit: apr/network_io/unix sendrecv.c

[Greg Stein <gs...@lyra.org>]

On Fri, Feb 16, 2001 at 09:34:09PM -0000, trawick@apache.org wrote:
>...
>   +    /* HP-UX can only send one header iovec and one footer iovec; try to
>   +     * only allocate storage to combine input iovecs when we really have to
>   +     */
>...
>   +        /* XXX:  BUHHH? wow, what a memory leak! */
>   +        headerbuf = hdtrarray[0].iov_base = apr_palloc(sock->cntxt, headerlen);

Use a subpool. Create if/when you need to do this apr_palloc() (or the one
for the trailers). Toss on exit from the function (watch out for early exit).

>...
>        do {
>   +        if (nbytes) {       /* any bytes to send from the file? */
>   +            rc = sendfile(sock->socketdes,      /* socket  */
>   +                          file->filedes,        /* file descriptor to send */
>   +                          *offset,              /* where in the file to start */
>   +                          nbytes,               /* number of bytes to send from file */
>   +                          hdtrarray,            /* Headers/footers */
>   +                          flags);               /* undefined, set to 0 */
>   +        }
>   +        else {              /* we can't call sendfile() for trailers only */
>   +            rc = write(sock->socketdes, hdtrarray[1].iov_base, hdtrarray[1].iov_len);
>   +        }
>   +    } while (rc == -1 && errno == EINTR);

If nbytes==0, then where/how are the headers sent? I just see the trailers.

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/

Re: cvs commit: apr/network_io/unix sendrecv.c

[Jeff Trawick <tr...@bellsouth.net>]

Greg Stein <gs...@lyra.org> writes:

> On Fri, Feb 16, 2001 at 09:34:09PM -0000, trawick@apache.org wrote:
> >...
> >   +    /* HP-UX can only send one header iovec and one footer iovec; try to
> >   +     * only allocate storage to combine input iovecs when we really have to
> >   +     */
> >...
> >   +        /* XXX:  BUHHH? wow, what a memory leak! */
> >   +        headerbuf = hdtrarray[0].iov_base = apr_palloc(sock->cntxt, headerlen);
> 
> Use a subpool. Create if/when you need to do this apr_palloc() (or the one
> for the trailers). Toss on exit from the function (watch out for
> early exit).

good idea; I'll work it into the HP and other platform versions of
apr_sendfile()

> >        do {
> >   +        if (nbytes) {       /* any bytes to send from the file? */
> >   +            rc = sendfile(sock->socketdes,      /* socket  */
> >   +                          file->filedes,        /* file descriptor to send */
> >   +                          *offset,              /* where in the file to start */
> >   +                          nbytes,               /* number of bytes to send from file */
> >   +                          hdtrarray,            /* Headers/footers */
> >   +                          flags);               /* undefined, set to 0 */
> >   +        }
> >   +        else {              /* we can't call sendfile() for trailers only */
> >   +            rc = write(sock->socketdes, hdtrarray[1].iov_base, hdtrarray[1].iov_len);
> >   +        }
> >   +    } while (rc == -1 && errno == EINTR);
> 
> If nbytes==0, then where/how are the headers sent? I just see the
> trailers.

yeah, a built-in assumption that apr_sendfile() is used in an
"appropriate" manner; i.e., you never have headers followed by 0 bytes
from the file followed by trailers, though in a repeat call on a
non-blocking socket you might have no headers (already sent) followed
by 0 bytes from the file (already sent) followed by trailers

I'll add some simple quick failure path to the FreeBSD and HP-UX
flavors.

Thanks!

-- 
Jeff Trawick | trawickj@bellsouth.net | PGP public key at web site:
       http://www.geocities.com/SiliconValley/Park/9289/
             Born in Roswell... married an alien...