gpfs 3.5 and Samba Compatibility

[Ralph Wuerthner]

On Sun, 3 Jun 2012 11:00:53 +0300
Nimrod Sapir <NIMRODS at il.ibm.com> wrote:

> You are right, of course. Also, I can't really think of a possible 
> solution from the GPFS side that will solve the problem while keeping 
> backward-compatibility. Is gpfs 3.5 compatibility included in the
> Samba roadmap? for which version? Is there anyone in the community
> (beside our team) which already started to examine this
> configuration? 
> 
> Thanks!
> 
> Christof Schmitt <christof.schmitt at us.ibm.com> wrote on 01/06/2012 
> 00:55:13:
> 
> 
> > From a brief look, making this change would tie Samba to GPFS
> > 3.5. A proper solution should allow a binary of the vfs_gpfs
> > module to work with GPFS 3.5 and older versions.
> > 
> > I am not sure if this can be done automatically, there is always
> > the fallback of adding a config option to specify the format.
> > When reading an ACL, simply allocating enough space and then
> > deciding according to the gpfs_acl header could be
> > enough. Writing might be more complicated as the Samba module
> > would not know beforehand which format to try.
> > 
> > Regards,
> > 
> > Christof Schmitt || IBM || SONAS System Development || Tucson, AZ
> > christof.schmitt at us.ibm.com  ||  +1-520-799-2469  (T/L: 321-2469)

I noticed the same build failure with GPFS 3.5 last week when I tried
to build Samba packages on a GPFS 3.5 system. I prepared the following
patch to address this build issue:

---
commit bcfe3632b5223c21e8bd0d22f06471cf89a49597
Author: Ralph Wuerthner <ralph.wuerthner at de.ibm.com>
Date:   Fri Jun 1 13:43:11 2012 +0200

    vsf_gpfs: fix ACL length calculation
    
    GPFS 3.5 introduces ACL enhancements which are breaking our ACL length
    calculations.
    
    Signed-off-by: Ralph Wuerthner <ralph.wuerthner at de.ibm.com>

diff --git a/source3/modules/vfs_gpfs.c b/source3/modules/vfs_gpfs.c
index ba756f6..d54405c 100644
--- a/source3/modules/vfs_gpfs.c
+++ b/source3/modules/vfs_gpfs.c
@@ -401,9 +401,10 @@ static bool gpfsacl_process_smbacl(files_struct *fsp, SMB4ACL_T *smbacl)
 	SMB4ACE_T	*smbace;
 	struct gpfs_acl *gacl;
 	TALLOC_CTX *mem_ctx  = talloc_tos();
+	int gacl_base_len = ((caddr_t)gacl->ace_v1) - ((caddr_t)gacl);
 
-	gacl_len = sizeof(struct gpfs_acl) +
-		(smb_get_naces(smbacl)-1)*sizeof(gpfs_ace_v4_t);
+	gacl_len = gacl_base_len + smb_get_naces(smbacl) *
+		sizeof(gpfs_ace_v4_t);
 
 	gacl = (struct gpfs_acl *)TALLOC_SIZE(mem_ctx, gacl_len);
 	if (gacl == NULL) {
@@ -662,16 +663,11 @@ static struct gpfs_acl *smb2gpfs_acl(const SMB_ACL_T pacl,
 	gpfs_aclLen_t len;
 	struct gpfs_acl *result;
 	int i;
-	union gpfs_ace_union
-	{
-		gpfs_ace_v1_t  ace_v1[1]; /* when GPFS_ACL_VERSION_POSIX */
-		gpfs_ace_v4_t  ace_v4[1]; /* when GPFS_ACL_VERSION_NFS4  */
-	};
+	int base_len = ((caddr_t)result->ace_v1) - ((caddr_t)result);
 
 	DEBUG(10, ("smb2gpfs_acl: Got ACL with %d entries\n", pacl->count));
 
-	len = sizeof(struct gpfs_acl) - sizeof(union gpfs_ace_union) +
-		(pacl->count)*sizeof(gpfs_ace_v1_t);
+	len = base_len + (pacl->count)*sizeof(gpfs_ace_v1_t);
 
 	result = (struct gpfs_acl *)SMB_MALLOC(len);
 	if (result == NULL) {

---

In my opinion the ultimate solution would be to ask GPFS to provide a
function/macro similar like this:

size_t gpfs_acl_len(gpfs_aclLevel_t acl_level,
                    gpfs_aclVersion_t acl_version,
                    gpfs_aclType_t    acl_type,
                    gpfs_aclCount_t   acl_nace);

This should allow maximum backward compatibility and allow GPFS to
enhance the ACL type if future features require this.


Regards

   Ralph