Why is the expire_time_attr export config parameter not respected when attributes are being refreshed, I don't see where we check for expire_time_attr before calling the FSAL getattr. Did I miss it ? Thanks, Marc.
Hi Daniel, I am trying to understand the support for expire_time_attr. First there is this comment in the code /** Expiration time interval in seconds for attributes. Settable with Attr_Expiration_Time (should never be set for client export_perms. */ int32_t expire_time_attr; But I see this initialization code: #define GLOBAL_EXPORT_PERMS_INITIALIZER \ .def.anonymous_uid = ANON_UID, \ .def.anonymous_gid = ANON_GID, \ .def.expire_time_attr = 60, \ My question is in which configuration block you set Attr_Expiration_Time for export_perms and which block set it for attrlist Thanks, Marc.
On 3/1/19 11:45 PM, Marc Eshel wrote:
Hi Daniel,
I am trying to understand the support for expire_time_attr. First there is this comment in the code
/** Expiration time interval in seconds for attributes. Settable with Attr_Expiration_Time (should never be set for client export_perms. */ int32_t expire_time_attr;
But I see this initialization code:
#define GLOBAL_EXPORT_PERMS_INITIALIZER \ .def.anonymous_uid = ANON_UID, \ .def.anonymous_gid = ANON_GID, \ .def.expire_time_attr = 60, \
My question is in which configuration block you set Attr_Expiration_Time for export_perms and which block set it for attrlist
Thanks, Marc.
Attr_Expiration_Time is an export parameter, so it can be set in the Export block or the Export_Defaults block. When a new entry is created (in mdcache_new_entry()), that value is copied from the export into the attrlist for that entry. Thereafter, it's referenced directly from the attrlist. So, it's only configured on an export, but it's referenced from the attrlist of the entry itself. Daniel
Thanks for the information, I still don't know what is this comment referring to, what is setting client export_perms?
Attr_Expiration_Time (should never be set for client
export_perms.
From: Daniel Gryniewicz <dang@redhat.com> To: Marc Eshel <eshel@us.ibm.com> Cc: 'Ganesha-devel' <devel@lists.nfs-ganesha.org> Date: 03/04/2019 07:12 AM Subject: Re: expire_time_attr
On 3/1/19 11:45 PM, Marc Eshel wrote:
Hi Daniel,
I am trying to understand the support for expire_time_attr. First
Marc. Daniel Gryniewicz <dang@redhat.com> wrote on 03/04/2019 07:12:09 AM: there is
this comment in the code
/** Expiration time interval in seconds for attributes. Settable with Attr_Expiration_Time (should never be set for client export_perms. */ int32_t expire_time_attr;
But I see this initialization code:
#define GLOBAL_EXPORT_PERMS_INITIALIZER \ .def.anonymous_uid = ANON_UID, \ .def.anonymous_gid = ANON_GID, \ .def.expire_time_attr = 60, \
My question is in which configuration block you set Attr_Expiration_Time for export_perms and which block set it for attrlist
Thanks, Marc.
Attr_Expiration_Time is an export parameter, so it can be set in the Export block or the Export_Defaults block. When a new entry is created (in mdcache_new_entry()), that value is copied from the export into the attrlist for that entry. Thereafter, it's referenced directly from the attrlist.
So, it's only configured on an export, but it's referenced from the attrlist of the entry itself.
Daniel
A client entry has a copy of export_perms in it (named client_perms), but we don't want expire_time_attr to be configurable in the client block. To do this, we left expire_time_attr out of CONF_EXPORT_PERMS(), and it's explicitly added to export_params[] and export_update_params[]. This keeps it from being configured in client_perms. All this is in support/exports.c Daniel On 3/4/19 12:37 PM, Marc Eshel wrote:
Thanks for the information, I still don't know what is this comment referring to, what is setting client export_perms?
Attr_Expiration_Time (should never be set for client
export_perms.
Marc.
Daniel Gryniewicz <dang@redhat.com> wrote on 03/04/2019 07:12:09 AM:
From: Daniel Gryniewicz <dang@redhat.com> To: Marc Eshel <eshel@us.ibm.com> Cc: 'Ganesha-devel' <devel@lists.nfs-ganesha.org> Date: 03/04/2019 07:12 AM Subject: Re: expire_time_attr
Hi Daniel,
I am trying to understand the support for expire_time_attr. First
On 3/1/19 11:45 PM, Marc Eshel wrote: there is
this comment in the code
/** Expiration time interval in seconds for attributes. Settable with Attr_Expiration_Time (should never be set for client export_perms. */ int32_t expire_time_attr;
But I see this initialization code:
#define GLOBAL_EXPORT_PERMS_INITIALIZER \ .def.anonymous_uid = ANON_UID, \ .def.anonymous_gid = ANON_GID, \ .def.expire_time_attr = 60, \
My question is in which configuration block you set Attr_Expiration_Time for export_perms and which block set it for attrlist
Thanks, Marc.
Attr_Expiration_Time is an export parameter, so it can be set in the Export block or the Export_Defaults block. When a new entry is created (in mdcache_new_entry()), that value is copied from the export into the attrlist for that entry. Thereafter, it's referenced directly from the attrlist.
So, it's only configured on an export, but it's referenced from the attrlist of the entry itself.
Daniel
A client entry has a copy of export_perms in it (named client_perms), but we don't want expire_time_attr to be configurable in the client block. To do this, we left expire_time_attr out of CONF_EXPORT_PERMS(), and it's explicitly added to export_params[] and export_update_params[]. This keeps it from being configured in client_perms.
All this is in support/exports.c
Right, and the reason for that is that the expire_time_attr is a property of an MDCACHE inode. Most of the time those belong to a single export, so setting expire_time_attr per export will work fine (should an inode actually be shared by multiple exports, and they have different expire_time_attr, then the expire_time_attr will be set from whichever export a particular instance of the inode is first accessed via). Having the setting in the client block would mean the setting would be dependent on whichever client first accessed the inode... If we have a per-filesystem config block, that would be a better place for the parameter, but we don't have such a config block... Frank
On 3/4/19 12:37 PM, Marc Eshel wrote:
Thanks for the information, I still don't know what is this comment referring to, what is setting client export_perms?
Attr_Expiration_Time (should never be set for client
export_perms.
Marc.
Daniel Gryniewicz <dang@redhat.com> wrote on 03/04/2019 07:12:09 AM:
From: Daniel Gryniewicz <dang@redhat.com> To: Marc Eshel <eshel@us.ibm.com> Cc: 'Ganesha-devel' <devel@lists.nfs-ganesha.org> Date: 03/04/2019 07:12 AM Subject: Re: expire_time_attr
Hi Daniel,
I am trying to understand the support for expire_time_attr. First
On 3/1/19 11:45 PM, Marc Eshel wrote: there is
this comment in the code
/** Expiration time interval in seconds for attributes. Settable with Attr_Expiration_Time (should never be set for client export_perms. */ int32_t expire_time_attr;
But I see this initialization code:
#define GLOBAL_EXPORT_PERMS_INITIALIZER \ .def.anonymous_uid = ANON_UID, \ .def.anonymous_gid = ANON_GID, \ .def.expire_time_attr = 60, \
My question is in which configuration block you set Attr_Expiration_Time for export_perms and which block set it for attrlist
Thanks, Marc.
Attr_Expiration_Time is an export parameter, so it can be set in the Export block or the Export_Defaults block. When a new entry is created (in mdcache_new_entry()), that value is copied from the export into the attrlist for that entry. Thereafter, it's referenced directly from the attrlist.
So, it's only configured on an export, but it's referenced from the attrlist of the entry itself.
Daniel
_______________________________________________ Devel mailing list -- devel@lists.nfs-ganesha.org To unsubscribe send an email to devel-leave@lists.nfs-ganesha.org
On 3/1/19 3:43 PM, Marc Eshel wrote:
Why is the expire_time_attr export config parameter not respected when attributes are being refreshed, I don't see where we check for expire_time_attr before calling the FSAL getattr. Did I miss it ?
These are checked in mdcache_is_attrs_valid(), and set on the entry at creation from the export parameters in mdcache_new_entry(). Daniel
participants (3)
-
Daniel Gryniewicz -
Frank Filz -
Marc Eshel