> On Nov 21, 2024, at 8:51 AM, Cedric Blancher <cedric.blancher@xxxxxxxxx> wrote: > > On Thu, 21 Nov 2024 at 00:17, Dan Shelton <dan.f.shelton@xxxxxxxxx> wrote: >> >> On Wed, 20 Nov 2024 at 21:56, Steve Dickson <steved@xxxxxxxxxx> wrote: >>> >>> >>> >>> On 11/19/24 11:59 AM, Steve Dickson wrote: >>>> From: Seiichi Ikarashi <s.ikarashi@xxxxxxxxxxx> >>>> >>>> The rsize/wsize values are not multiples of 1024 but multiples of the >>>> system's page size or powers of 2 if < system's page size as defined >>>> in fs/nfs/internal.h:nfs_io_size(). >>>> >>>> Signed-off-by: Steve Dickson <steved@xxxxxxxxxx> >>> Committed... (tag: nfs-utils-2-8-2-rc2) >>> >>> I know we are still discussing this but I think >>> this version is better than what we have. >> >> Nope. The code is IMO wrong, and the docs are buggy too. Let's be absolutely clear. A code bug (or "the code is wrong/broken") occurs when code does not behave as it was designed to behave. That is not what is happening here: the Linux NFS client is working as intended. What you and Dan are asking for is a change in design or policy. The software industry uses the term "Request For Enhancement" for that. The patch Steve committed makes the man page match the current kernel code. That is not a bad thing to do, and it is certainly not meant to be a permanent and final answer to this issue. Now let's put aside the accusations and hostility, and please let's stick with technical concerns. >>> So update patches are welcome! >> >> Solaris, HPUX, FreeBSD and Windows NFSv3/v4 implementations all count in bytes. I don't understand what this statement implies. Linux's nfs(5) states these values are in units of bytes. AFAICT the proposed man page patch does not change that. On Linux we say "rsize=4096" and that means the maximum size of a READ on the wire is 4096 bytes. Our rsize and wsize options are specified just like on other operating systems. > I hereby concur, this is better to be consistent across operating > systems. In any case this is better than some machine- or > hardware-specific config option called "default page size", which no > one knows at boot time. The only difference with the Linux implementation is that the kernel might round off the specified values a little differently than other implementations. The description of the kernel's rounding behavior is what is being modified. -- Chuck Lever
