On Sun, 11 Aug 2002, Maxim Konovalov wrote:
> This is sendfile(2) mis-behaviour arised after rev.1.109
> sys/kern/uipc_syscalls.c but I think the real problem in vn_rdwr(),
> sys/kern/vfs_vnops.c. Here is my patch but I really need somebody with
> vfs clue. I CC'ed Robert Watson as an author of sendfile(2)
> modification and our vfs expert.
>
> Index: sys/kern/vfs_vnops.c
> ===================================================================
> RCS file: /home/ncvs/src/sys/kern/vfs_vnops.c,v
> retrieving revision 1.159
> diff -u -r1.159 vfs_vnops.c
> --- sys/kern/vfs_vnops.c 8 Aug 2002 12:45:30 -0000 1.159
> +++ sys/kern/vfs_vnops.c 11 Aug 2002 10:19:47 -0000
> @@ -401,7 +401,7 @@
> if (aresid)
> *aresid = auio.uio_resid;
> else
> - if (auio.uio_resid && error == 0)
> + if (auio.uio_resid && error != 0)
> error = EIO;
> if ((ioflg & IO_NODELOCKED) == 0) {
> if (rw == UIO_WRITE)
>
> %%%
>
> With this patch sendfile(2) and ftpd(8) work as expected but I cannot
> believe vn_rdwr() has been broken since 1994.
I think the caller (do_sendfile() here?) is at fault for not passing
a non-NULL aresid. The EIO case is basically a kludge to handle short
i/o's. Since the caller hasn't passed a place to return the (residual)
i/o count, it can't even tell if a short i/o occurred, so we pretend
that short i/o's are i/o errors.
There are a lot of sloppy callers that pass a null aresid. But these are
less broken than link_aout_load_file() and link_elf_load_file which pass
a non-null one and then never check it.
Bruce
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
