On Mon, 14 Sep 2015 12:10:15 -0400
"bfields@xxxxxxxxxxxx" <bfields@xxxxxxxxxxxx> wrote:
> On Sat, Sep 12, 2015 at 06:24:54AM -0400, Jeff Layton wrote:
> > On Sat, 12 Sep 2015 04:41:33 +0000
> > "Dilger, Andreas" <andreas.dilger@xxxxxxxxx> wrote:
> >
> > > On 2015/09/11, 4:20 AM, "HPDD-discuss on behalf of Jeff Layton"
> > > <hpdd-discuss-bounces@xxxxxxxxxxxx on behalf of jlayton@xxxxxxxxxxxxxxx>
> > > wrote:
> > >
> > > >With NFSv3 nfsd will always attempt to send along WCC data to the
> > > >client. This generally involves saving off the in-core inode information
> > > >prior to doing the operation on the given filehandle, and then issuing a
> > > >vfs_getattr to it after the op.
> > > >
> > > >Some filesystems (particularly clustered or networked ones) have an
> > > >expensive ->getattr inode operation. Atomicitiy is also often difficult
> > > >or impossible to guarantee on such filesystems. For those, we're best
> > > >off not trying to provide WCC information to the client at all, and to
> > > >simply allow it to poll for that information as needed with a GETATTR
> > > >RPC.
> > > >
> > > >This patch adds a new flags field to struct export_operations, and
> > > >defines a new EXPORT_OP_NOWCC flag that filesystems can use to indicate
> > > >that nfsd should not attempt to provide WCC info in NFSv3 replies. It
> > > >also adds a blurb about the new flags field and flag to the exporting
> > > >documentation.
> > > >
> > > >The server will also now skip collecting this information for NFSv2 as
> > > >well, since that info is never used there anyway.
> > > >
> > > >Note that this patch does not add this flag to any filesystem
> > > >export_operations structures. This was originally developed to allow
> > > >reexporting nfs via nfsd. That code is not (and may never be) suitable
> > > >for merging into mainline.
> > > >
> > > >Other filesystems may want to consider enabling this flag too. It's hard
> > > >to tell however which ones have export operations to enable export via
> > > >knfsd and which ones mostly rely on them for open-by-filehandle support,
> > > >so I'm leaving that up to the individual maintainers to decide. I am
> > > >cc'ing the relevant lists for those filesystems that I think may want to
> > > >consider adding this though.
> > > >
> > > >Cc: HPDD-discuss@xxxxxxxxxxxx
> > > >Cc: ceph-devel@xxxxxxxxxxxxxxx
> > > >Cc: cluster-devel@xxxxxxxxxx
> > > >Cc: fuse-devel@xxxxxxxxxxxxxxxxxxxxx
> > > >Cc: ocfs2-devel@xxxxxxxxxxxxxx
> > > >Signed-off-by: Jeff Layton <jeff.layton@xxxxxxxxxxxxxxx>
> > > >---
> > > > Documentation/filesystems/nfs/Exporting | 27 +++++++++++++++++++++++++++
> > > > fs/nfsd/nfs3xdr.c | 5 ++++-
> > > > fs/nfsd/nfsfh.c | 14 ++++++++++++++
> > > > fs/nfsd/nfsfh.h | 5 ++++-
> > > > include/linux/exportfs.h | 2 ++
> > > > 5 files changed, 51 insertions(+), 2 deletions(-)
> > > >
> > > >diff --git a/Documentation/filesystems/nfs/Exporting
> > > >b/Documentation/filesystems/nfs/Exporting
> > > >index 520a4becb75c..fa636cde3907 100644
> > > >--- a/Documentation/filesystems/nfs/Exporting
> > > >+++ b/Documentation/filesystems/nfs/Exporting
> > > >@@ -138,6 +138,11 @@ struct which has the following members:
> > > > to find potential names, and matches inode numbers to find the
> > > >correct
> > > > match.
> > > >
> > > >+ flags
> > > >+ Some filesystems may need to be handled differently than others. The
> > > >+ export_operations struct also includes a flags field that allows the
> > > >+ filesystem to communicate such information to nfsd. See the Export
> > > >+ Operations Flags section below for more explanation.
> > > >
> > > > A filehandle fragment consists of an array of 1 or more 4byte words,
> > > > together with a one byte "type".
> > > >@@ -147,3 +152,25 @@ generated by encode_fh, in which case it will have
> > > >been padded with
> > > > nuls. Rather, the encode_fh routine should choose a "type" which
> > > > indicates the decode_fh how much of the filehandle is valid, and how
> > > > it should be interpreted.
> > > >+
> > > >+Export Operations Flags
> > > >+-----------------------
> > > >+In addition to the operation vector pointers, struct export_operations
> > > >also
> > > >+contains a "flags" field that allows the filesystem to communicate to
> > > >nfsd
> > > >+that it may want to do things differently when dealing with it. The
> > > >+following flags are defined:
> > > >+
> > > >+ EXPORT_OP_NOWCC
> > > >+ RFC 1813 recommends that servers always send weak cache consistency
> > > >+ (WCC) data to the client after each operation. The server should
> > > >+ atomically collect attributes about the inode, do an operation on it,
> > > >+ and then collect the attributes afterward. This allows the client to
> > > >+ skip issuing GETATTRs in some situations but means that the server
> > > >+ is calling vfs_getattr for almost all RPCs. On some filesystems
> > > >+ (particularly those that are clustered or networked) this is
> > > >expensive
> > > >+ and atomicity is difficult to guarantee. This flag indicates to nfsd
> > > >+ that it should skip providing WCC attributes to the client in NFSv3
> > > >+ replies when doing operations on this filesystem. Consider enabling
> > > >+ this on filesystems that have an expensive ->getattr inode operation,
> > > >+ or when atomicity between pre and post operation attribute collection
> > > >+ is impossible to guarantee.
> > > >diff --git a/fs/nfsd/nfs3xdr.c b/fs/nfsd/nfs3xdr.c
> > > >index 01dcd494f781..c30c8c604e2a 100644
> > > >--- a/fs/nfsd/nfs3xdr.c
> > > >+++ b/fs/nfsd/nfs3xdr.c
> > > >@@ -203,7 +203,7 @@ static __be32 *
> > > > encode_post_op_attr(struct svc_rqst *rqstp, __be32 *p, struct svc_fh
> > > >*fhp)
> > > > {
> > > > struct dentry *dentry = fhp->fh_dentry;
> > > >- if (dentry && d_really_is_positive(dentry)) {
> > > >+ if (!fhp->fh_no_wcc && dentry && d_really_is_positive(dentry)) {
> > > > __be32 err;
> > > > struct kstat stat;
> > > >
> > > >@@ -256,6 +256,9 @@ void fill_post_wcc(struct svc_fh *fhp)
> > > > {
> > > > __be32 err;
> > > >
> > > >+ if (fhp->fh_no_wcc)
> > > >+ return;
> > > >+
> > > > if (fhp->fh_post_saved)
> > > > printk("nfsd: inode locked twice during operation.\n");
> > > >
> > > >diff --git a/fs/nfsd/nfsfh.c b/fs/nfsd/nfsfh.c
> > > >index 350041a40fe5..29ae37f62b9b 100644
> > > >--- a/fs/nfsd/nfsfh.c
> > > >+++ b/fs/nfsd/nfsfh.c
> > > >@@ -267,6 +267,16 @@ static __be32 nfsd_set_fh_dentry(struct svc_rqst
> > > >*rqstp, struct svc_fh *fhp)
> > > >
> > > > fhp->fh_dentry = dentry;
> > > > fhp->fh_export = exp;
> > > >+
> > > >+ switch (rqstp->rq_vers) {
> > > >+ case 3:
> > > >+ if (!(dentry->d_sb->s_export_op->flags & EXPORT_OP_NOWCC))
> > > >+ break;
> > > >+ /* Fallthrough */
> > > >+ case 2:
> > > >+ fhp->fh_no_wcc = true;
> > > >+ }
> > > >+
> > > > return 0;
> > > > out:
> > > > exp_put(exp);
> > > >@@ -535,6 +545,9 @@ fh_compose(struct svc_fh *fhp, struct svc_export
> > > >*exp, struct dentry *dentry,
> > > > */
> > > > set_version_and_fsid_type(fhp, exp, ref_fh);
> > > >
> > > >+ /* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
> > > >+ fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
> > > >+
> > > > if (ref_fh == fhp)
> > > > fh_put(ref_fh);
> > > >
> > > >@@ -641,6 +654,7 @@ fh_put(struct svc_fh *fhp)
> > > > exp_put(exp);
> > > > fhp->fh_export = NULL;
> > > > }
> > > >+ fhp->fh_no_wcc = false;
> > > > return;
> > > > }
> > > >
> > > >diff --git a/fs/nfsd/nfsfh.h b/fs/nfsd/nfsfh.h
> > > >index 1e90dad4926b..9ddead4d98f8 100644
> > > >--- a/fs/nfsd/nfsfh.h
> > > >+++ b/fs/nfsd/nfsfh.h
> > > >@@ -32,6 +32,7 @@ typedef struct svc_fh {
> > > >
> > > > unsigned char fh_locked; /* inode locked by us */
> > > > unsigned char fh_want_write; /* remount protection taken */
> > > >+ bool fh_no_wcc; /* no wcc data needed */
> > >
> > > This increases the size of svc_fh because it splits the four unsigned
> > > chars.
> > > You could change all of these (fh_locked, fh_want_write,
> > > fh_{pre,post}saved)
> > > to be bools to avoid that and make it more clear they are only used as
> > > booleans (I verified that they all are only assigned 0 or 1).
> > >
> >
> > I don't think it matters, at least not on x86_64. bools and chars both
> > require a byte. pahole does show this adding a new hole, but that's
> > just because this brings the code up to 5 flags and the next field
> > (fh_pre_size) needs to be aligned.
> >
> > I do agree that replacing those other unsigned chars with bools is more
> > clear however. Maybe we should even replace them all with a single
> > unsigned int and use bitops to set flags in there. That would be more
> > space efficient now that we're at 5 flags.
>
> Makes sense to me.--b.
I played around with this a little today, and it turns out not to make
a lot of difference. Here's what pahole says about the existing code
(once I moved fh_maxsize to snuggle up to fh_handle to plug a hole):
struct svc_fh {
struct knfsd_fh fh_handle; /* 0 132 */
/* --- cacheline 2 boundary (128 bytes) was 4 bytes ago --- */
int fh_maxsize; /* 132 4 */
struct dentry * fh_dentry; /* 136 8 */
struct svc_export * fh_export; /* 144 8 */
unsigned char fh_locked; /* 152 1 */
unsigned char fh_want_write; /* 153 1 */
unsigned char fh_post_saved; /* 154 1 */
unsigned char fh_pre_saved; /* 155 1 */
/* XXX 4 bytes hole, try to pack */
__u64 fh_pre_size; /* 160 8 */
struct timespec fh_pre_mtime; /* 168 16 */
struct timespec fh_pre_ctime; /* 184 16 */
/* --- cacheline 3 boundary (192 bytes) was 8 bytes ago --- */
u64 fh_pre_change; /* 200 8 */
struct kstat fh_post_attr; /* 208 104 */
/* --- cacheline 4 boundary (256 bytes) was 56 bytes ago --- */
u64 fh_post_change; /* 312 8 */
/* --- cacheline 5 boundary (320 bytes) --- */
/* size: 320, cachelines: 5, members: 14 */
/* sum members: 316, holes: 1, sum holes: 4 */
};
...and here's what it looks like with everything converted to a
fh_flags field:
struct svc_fh {
struct knfsd_fh fh_handle; /* 0 132 */
/* --- cacheline 2 boundary (128 bytes) was 4 bytes ago --- */
int fh_maxsize; /* 132 4 */
struct dentry * fh_dentry; /* 136 8 */
struct svc_export * fh_export; /* 144 8 */
long unsigned int fh_flags; /* 152 8 */
__u64 fh_pre_size; /* 160 8 */
struct timespec fh_pre_mtime; /* 168 16 */
struct timespec fh_pre_ctime; /* 184 16 */
/* --- cacheline 3 boundary (192 bytes) was 8 bytes ago --- */
u64 fh_pre_change; /* 200 8 */
struct kstat fh_post_attr; /* 208 104 */
/* --- cacheline 4 boundary (256 bytes) was 56 bytes ago --- */
u64 fh_post_change; /* 312 8 */
/* --- cacheline 5 boundary (320 bytes) --- */
/* size: 320, cachelines: 5, members: 11 */
};
...no change in size.
I used an unsigned long for fh_flags since we might as well. Making it
smaller just adds a hole in there since the compiler wants to align the
fh_pre_size. Moving it around doesn't help either as it just moves the
hole around. Note that this is x86_64. It might look different on a
32-bit arch, but I doubt it really matters much in the big scheme of
things.
Bruce, I'll send out the patches that change this if you like, but I'm
inclined to just leave this alone since it doesn't seem to have a
tangible benefit.
--
Jeff Layton <jlayton@xxxxxxxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
