svn commit: r201999 - head/lib/libc/stdio
Jilles Tjoelker
jilles at stack.nl
Sun Jan 10 23:06:20 UTC 2010
On Sun, Jan 10, 2010 at 02:18:58PM -0800, Colin Percival wrote:
> Andrey Chernov wrote:
> > On Sun, Jan 10, 2010 at 02:30:30PM +0000, Colin Percival wrote:
> >> + if (((count | size) > 0xFFFF) &&
> >> + (count > SIZE_MAX / size)) {
> >> + errno = EINVAL;
> >> + fp->_flags |= __SERR;
> >> + return (0);
> >> + }
> > 2) fp->_flags |= __SERR;
> > This flag is for errors in the file stream, not for errors in
> > the arguments. Please back that line out.
> Quoting fread(3):
> The function fread() does not distinguish between end-of-file and error,
> and callers must use feof(3) and ferror(3) to determine which occurred.
> This would seem to imply that for any failed request, either feof or ferror
> should return a non-zero value. Removing this line would break the common
> while (fread(ptr, reclen, 1, f) == 1) {
> /* Do stuff here. */
> }
> if (ferror(f)) {
> /* Die due to read error */
> }
> /* No error? Ok, we must have hit EOF */
> idiom.
Agreed, this error should set the error indicator.
> > 3) errno should be EOVERFLOW, see other owerflow checks in the stdio.
> I picked EINVAL because this is the code used by read(2) and write(2) if they
> are passed nbytes > INT_MAX. It would seem odd to use one error code for a
> number of bytes between INT_MAX and SIZE_MAX and then switch to a different
> error code for > SIZE_MAX bytes.
I think the fact that you can only read(2)/write(2) INT_MAX at a time is
a bug, so basing additional code on it seems inappropriate.
Even then, fread()/fwrite() will not generate such large
read(2)/write(2) requests as they always go through the buffer which is
filled using smaller requests.
Also, EOVERFLOW will give a more understandable error message for users.
--
Jilles Tjoelker
More information about the svn-src-all
mailing list