Re: [PATCH v2] fs/9p: Don't open remote file with APPEND mode when writeback cache is used

[Christian Schoenebeck <linux_oss@crudebyte.com>]

On Monday, November 3, 2025 8:34:19 AM CET Dominique Martinet wrote:
> Thanks for the v2
> 
> Tingmao Wang wrote on Sun, Nov 02, 2025 at 11:56:30PM +0000:
> > I haven't done a bisect to figure out if this regression was introduced by
> > a change or was this behaviour always present - 6.14 has the same problem
> > and 6.13 did not compile for me due to some problem with bool / true /
> > false being a keyword in c23, and it could not compile with c17 either.
> 
> Ok so you got me at the "it happens in cache=loose too", so I went ahead
> with bisect... I had the same problem with newer gcc being c23 and older
> kernels being silly, you'd think `KCFLAGS=-std=gnu11` is enough but some
> arch makefiles ignore that so I went in heavy patching
> arch/x86/boot/Makefile and arch/x86/boot/compressed/Makefile . . .
> ... Anyway, it certainly didn't break recently, but it's not ages ago
> either: the repro you made (thank you!!) starts failing in 6.4 cache
> rework: the first bad commit is actually 21e26d5e54ab ("fs/9p: Fix bit
> operation logic error"), but that's just a fixup of the previous commit
> so 4eb3117888a9 ("fs/9p: Rework cache modes and add new options to
> Documentation") is the cluprit... and... I'm not quite sure I get why
> without digging deeper...
> (But at least that made me realize the commit just before, 1543b4c5071c
> ("fs/9p: remove writeback fid and fix per-file modes"), removed the
> writeback fid I was talking about... But that commit was working (with a
> sane ~P9_OWRITE mask), so it's not directly that removal that broke
> things, but something else in the rework)

Haven't tested, but to me it looks like 1543b4c5071c ("fs/9p: remove writeback 
fid and fix per-file modes") was the causing change. Up to that commit the 
open flags were hard coded to exactly O_RDWR for all writeback fids:

struct p9_fid *v9fs_writeback_fid(struct dentry *dentry)
{
    ...
    /*
     * writeback fid will only be used to write back the
     * dirty pages. We always request for the open fid in read-write
     * mode so that a partial page write which result in page
     * read can work.
     */
    err = p9_client_open(fid, O_RDWR);
    ...
}

/Christian

> Anyway, I think this commit is a strict improvement over the current
> situation, I'll just slap a Fixes tag and push to -next for now, and if
> time allows I'll try to have a closer look...
> 
> Thanks,