Skip to contentSkip to conversation

nfs41: change PNFS_LAYOUTRET_ON_SETATTR to only return on truncation to smaller size

PATCH In review Peng Tao 4 participants 4 messages Sep 12, 2014 – Sep 17, 2014

Subsystems (2)
lore
Share & tools
lore

Peng Tao

author
#0 lore raw

PATCH v2 1/1 nfs41: change PNFS_LAYOUTRET_ON_SETATTR to only return on truncation to smaller size

Both blocks layout and objects layout want to use it to avoid CB_LAYOUTRECALL but that should only happen if client is doing truncation to a smaller size. For other cases, we let server decide if it wants to recall client's layouts. Change PNFS_LAYOUTRET_ON_SETATTR to follow the logic and not to send layoutreturn unnecessarily.

Cc Christoph Hellwig <[email protected]>
Cc Boaz Harrosh <[email protected]>
---
 fs/nfs/nfs4proc.c | 4 +++-
 fs/nfs/pnfs.h     | 3 ++-
 2 files changed, 5 insertions(+), 2 deletions(-)
fs/nfs/nfs4proc.c +3 -1
index 7dd8aca..63a5b80 100644
@@ -3213,7 +3213,9 @@ nfs4_proc_setattr(struct dentry *dentry, struct nfs_fattr *fattr,
 	struct nfs4_label *label = NULL;
 	int status;
 
-	if (pnfs_ld_layoutret_on_setattr(inode))
+	if (pnfs_ld_layoutret_on_setattr(inode) &&
+	    sattr->ia_valid & ATTR_SIZE &&
+	    sattr->ia_size < i_size_read(inode))
 		pnfs_commit_and_return_layout(inode);
 
 	nfs_fattr_init(fattr);
fs/nfs/pnfs.h +2 -1
index aca3dff..04a3646 100644
@@ -68,7 +68,8 @@ enum {
 };
 
 enum layoutdriver_policy_flags {
-	/* Should the pNFS client commit and return the layout upon a setattr */
+	/* Should the pNFS client commit and return the layout upon truncate to
+	 * a smaller size */
 	PNFS_LAYOUTRET_ON_SETATTR	= 1 << 0,
 	PNFS_LAYOUTRET_ON_ERROR		= 1 << 1,
 };
Signature · 1 line
1.9.1
#1 lore raw
Quoted 8 lines · Peng Tao

Both blocks layout and objects layout want to use it to avoid CB_LAYOUTRECALL but that should only happen if client is doing truncation to a smaller size. For other cases, we let server decide if it wants to recall client's layouts. Change PNFS_LAYOUTRET_ON_SETATTR to follow the logic and not to send layoutreturn unnecessarily.

Cc Christoph Hellwig <[email protected]>
Cc Boaz Harrosh <[email protected]>

Peng, Christoph Hi

This is in violation of the pnf-objects draft. We are obliged on chown to return layouts because of how the CAPs work, they have an embedded CAPs version which might increment when chown is changed, and also the credential keys. Which means that using the old layout will return an E_ACCESS from the OSD.

But..

I will agree with this patch. The above is true in an OSD protocol higher then NO_SEC. But none of the current implementation support anything other then NO_SEC.

Second also with none-NO_SEC once a client does a chown the server needs to RECALL all other clients, so as Peng says above the Server should/can also recall from us.

But some good sole needs to make an errata at the RFC draft so to explain.

So:

Quoted 36 lines
---
 fs/nfs/nfs4proc.c | 4 +++-
 fs/nfs/pnfs.h     | 3 ++-
 2 files changed, 5 insertions(+), 2 deletions(-)
fs/nfs/nfs4proc.c +3 -1
index 7dd8aca..63a5b80 100644
@@ -3213,7 +3213,9 @@ nfs4_proc_setattr(struct dentry *dentry, struct nfs_fattr *fattr,
 	struct nfs4_label *label = NULL;
 	int status;
 
-	if (pnfs_ld_layoutret_on_setattr(inode))
+	if (pnfs_ld_layoutret_on_setattr(inode) &&
+	    sattr->ia_valid & ATTR_SIZE &&
+	    sattr->ia_size < i_size_read(inode))
 		pnfs_commit_and_return_layout(inode);
 
 	nfs_fattr_init(fattr);
fs/nfs/pnfs.h +2 -1
index aca3dff..04a3646 100644
@@ -68,7 +68,8 @@ enum {
 };
 
 enum layoutdriver_policy_flags {
-	/* Should the pNFS client commit and return the layout upon a setattr */
+	/* Should the pNFS client commit and return the layout upon truncate to
+	 * a smaller size */
 	PNFS_LAYOUTRET_ON_SETATTR	= 1 << 0,
 	PNFS_LAYOUTRET_ON_ERROR		= 1 << 1,
 };
#2 lore raw
Quoted 1 line · Boaz Harrosh

This is in violation of the pnf-objects draft.

Ithought that's an RFC by now? :)

Quoted 5 lines

We are obliged
on chown to return layouts because of how the CAPs work, they have an embedded CAPs version which might increment when chown is changed, and also the credential keys. Which means that using the old layout will return an E_ACCESS from the OSD.

This is a good enough argument to just split the flags into one for chown and one for truncate. Can you confirm that any other setattr but a truncate to a smaller size or chown is fine with your interpretation of RFC5664?

Quoted 6 lines

But..

I will agree with this patch. The above is true in an OSD protocol higher then NO_SEC. But none of the current implementation support anything other then NO_SEC.

Quoted 6 lines

Second also with none-NO_SEC once a client does a chown the server needs to RECALL all other clients, so as Peng says above the Server should/can also recall from us.

But some good sole needs to make an errata at the RFC draft so to explain.

I'd rather not violate the existing RFC unless we have a good reason for it and an errata out. Having a flag to return on chown for the object layout seems like the easier way to go forward.

#3 lore raw
Quoted 24 lines · Christoph Hellwig
Quoted 1 line · Boaz Harrosh

This is in violation of the pnf-objects draft.

Ithought that's an RFC by now? :)

Quoted 5 lines

We are obliged
on chown to return layouts because of how the CAPs work, they have an embedded CAPs version which might increment when chown is changed, and also the credential keys. Which means that using the old layout will return an E_ACCESS from the OSD.

This is a good enough argument to just split the flags into one for chown and one for truncate. Can you confirm that any other setattr but a truncate to a smaller size or chown is fine with your interpretation of RFC5664?

Quoted 10 lines

But..

I will agree with this patch. The above is true in an OSD protocol higher then NO_SEC. But none of the current implementation support anything other then NO_SEC.

Second also with none-NO_SEC once a client does a chown the server needs to RECALL all other clients, so as Peng says above the Server should/can also recall from us.

Quoted 5 lines
Quoted 2 lines

But some good sole needs to make an errata at the RFC draft so to explain.

I'd rather not violate the existing RFC unless we have a good reason for it and an errata out. Having a flag to return on chown for the object layout seems like the easier way to go forward.

Ack

Benny

Quoted 4 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html