Re: [PATCH 51/87] fs/nfsd: convert to new inode {a,m}time accessors

[Chuck Lever]

On Thu, Sep 28, 2023 at 10:09:19AM -0400, Jeff Layton wrote:
> On Thu, 2023-09-28 at 09:56 -0400, Chuck Lever wrote:
> > On Thu, Sep 28, 2023 at 07:03:00AM -0400, Jeff Layton wrote:
> > > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
> > > ---
> > >  fs/nfsd/blocklayout.c | 3 ++-
> > >  fs/nfsd/nfs3proc.c    | 4 ++--
> > >  fs/nfsd/nfs4proc.c    | 8 ++++----
> > >  fs/nfsd/nfsctl.c      | 2 +-
> > >  4 files changed, 9 insertions(+), 8 deletions(-)
> > > 
> > > diff --git a/fs/nfsd/blocklayout.c b/fs/nfsd/blocklayout.c
> > > index 01d7fd108cf3..bdc582777738 100644
> > > --- a/fs/nfsd/blocklayout.c
> > > +++ b/fs/nfsd/blocklayout.c
> > > @@ -119,10 +119,11 @@ nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp,
> > >  {
> > >  	loff_t new_size = lcp->lc_last_wr + 1;
> > >  	struct iattr iattr = { .ia_valid = 0 };
> > > +	struct timespec64 mtime = inode_get_mtime(inode);
> > 
> > Nit: Please use reverse Christmas tree for new variable declarations.
> > 
> 
> Ok
> 
> > 
> > >  	int error;
> > >  
> > >  	if (lcp->lc_mtime.tv_nsec == UTIME_NOW ||
> > > -	    timespec64_compare(&lcp->lc_mtime, &inode->i_mtime) < 0)
> > > +	    timespec64_compare(&lcp->lc_mtime, &mtime) < 0)
> > >  		lcp->lc_mtime = current_time(inode);
> > >  	iattr.ia_valid |= ATTR_ATIME | ATTR_CTIME | ATTR_MTIME;
> > >  	iattr.ia_atime = iattr.ia_ctime = iattr.ia_mtime = lcp->lc_mtime;
> > > diff --git a/fs/nfsd/nfs3proc.c b/fs/nfsd/nfs3proc.c
> > > index 268ef57751c4..b1c90a901d3e 100644
> > > --- a/fs/nfsd/nfs3proc.c
> > > +++ b/fs/nfsd/nfs3proc.c
> > > @@ -294,8 +294,8 @@ nfsd3_create_file(struct svc_rqst *rqstp, struct svc_fh *fhp,
> > >  			status = nfserr_exist;
> > >  			break;
> > >  		case NFS3_CREATE_EXCLUSIVE:
> > > -			if (d_inode(child)->i_mtime.tv_sec == v_mtime &&
> > > -			    d_inode(child)->i_atime.tv_sec == v_atime &&
> > > +			if (inode_get_mtime(d_inode(child)).tv_sec == v_mtime &&
> > > +			    inode_get_atime(d_inode(child)).tv_sec == v_atime &&
> > 
> > "inode_get_atime(yada).tv_sec" seems to be a frequently-repeated
> > idiom, at least in this patch. Would it be helpful to have an
> > additional helper that extracted just the seconds field, and one
> > that extracts just the nsec field?
> > 
> 
> I don't know that extra helpers will make that any clearer.

To clarify my review comment:

I understand that eventually the timestamps stored in the inode will
be a single scalar value that is to be converted to a timespec64
structure.

So for accessors who want only one of the tv_sec or tv_nsec fields:

   scalar -> timespec64 -> tv_sec

It might be more efficient to skip extracting the tv_nsec field when
that value isn't going to be used. Perhaps the compiler might
observe that the tv_nsec result isn't used and remove that dead
code. But IMO it would be easier for humans to understand and more
dependably optimized to write the helpers so the tv_nsec part of the
computation wasn't even in there in these cases.


> > >  			    d_inode(child)->i_size == 0) {
> > >  				break;
> > >  			}
> > > diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
> > > index 4199ede0583c..b17309aac0d5 100644
> > > --- a/fs/nfsd/nfs4proc.c
> > > +++ b/fs/nfsd/nfs4proc.c
> > > @@ -322,8 +322,8 @@ nfsd4_create_file(struct svc_rqst *rqstp, struct svc_fh *fhp,
> > >  			status = nfserr_exist;
> > >  			break;
> > >  		case NFS4_CREATE_EXCLUSIVE:
> > > -			if (d_inode(child)->i_mtime.tv_sec == v_mtime &&
> > > -			    d_inode(child)->i_atime.tv_sec == v_atime &&
> > > +			if (inode_get_mtime(d_inode(child)).tv_sec == v_mtime &&
> > > +			    inode_get_atime(d_inode(child)).tv_sec == v_atime &&
> > >  			    d_inode(child)->i_size == 0) {
> > >  				open->op_created = true;
> > >  				break;		/* subtle */
> > > @@ -331,8 +331,8 @@ nfsd4_create_file(struct svc_rqst *rqstp, struct svc_fh *fhp,
> > >  			status = nfserr_exist;
> > >  			break;
> > >  		case NFS4_CREATE_EXCLUSIVE4_1:
> > > -			if (d_inode(child)->i_mtime.tv_sec == v_mtime &&
> > > -			    d_inode(child)->i_atime.tv_sec == v_atime &&
> > > +			if (inode_get_mtime(d_inode(child)).tv_sec == v_mtime &&
> > > +			    inode_get_atime(d_inode(child)).tv_sec == v_atime &&
> > >  			    d_inode(child)->i_size == 0) {
> > >  				open->op_created = true;
> > >  				goto set_attr;	/* subtle */
> > > diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> > > index 7ed02fb88a36..846559e4769b 100644
> > > --- a/fs/nfsd/nfsctl.c
> > > +++ b/fs/nfsd/nfsctl.c
> > > @@ -1132,7 +1132,7 @@ static struct inode *nfsd_get_inode(struct super_block *sb, umode_t mode)
> > >  	/* Following advice from simple_fill_super documentation: */
> > >  	inode->i_ino = iunique(sb, NFSD_MaxReserved);
> > >  	inode->i_mode = mode;
> > > -	inode->i_atime = inode->i_mtime = inode_set_ctime_current(inode);
> > > +	simple_inode_init_ts(inode);
> > 
> > An observation about the whole series: Should these helpers use the
> > usual naming convention of:
> > 
> >   <subsystem>-<subject>-<verb>
> > 
> > So we get:
> > 
> >   simple_inode_ts_init(inode);
> > 
> >   inode_atime_get(inode)
> > 
> 
> This was already bikeshedded during the ctime series, and the near
> universal preference at the time was to go with inode_set_ctime and
> inode_get_ctime. I'm just following suit with the new accessors.

When this was reviewed before, there were only ctime accessors. Now
we have two more sets of accessor utilities. Just an observation. I
can drop it here.


> > >  	switch (mode & S_IFMT) {
> > >  	case S_IFDIR:
> > >  		inode->i_fop = &simple_dir_operations;
> > > -- 
> > > 2.41.0
> > > 
> > 
> > Otherwise, for the patch(es) touching nfsd:
> > 
> > Acked-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
> > 
> 
> Thanks!
> -- 
> Jeff Layton <jlayton@xxxxxxxxxx>

-- 
Chuck Lever