[PATCH 0/2] NFSd: NFSv4 state - lease and grace times per net

[PATCH 0/2] NFSd: NFSv4 state - lease and grace times per net

[Stanislav Kinsbursky]

This two variables are parts of NFSv4 state and should be containerized too.

The following series implements...

---

Stanislav Kinsbursky (2):
      nfsd: make NFSv4 lease time per net
      nfsd: make NFSv4 grace time per net


 fs/nfsd/netns.h        |    3 +++
 fs/nfsd/nfs4callback.c |    8 +++++---
 fs/nfsd/nfs4state.c    |   18 +++++++-----------
 fs/nfsd/nfs4xdr.c      |    4 +++-
 fs/nfsd/nfsctl.c       |    9 +++++++--
 fs/nfsd/nfsd.h         |    3 ---
 6 files changed, 25 insertions(+), 20 deletions(-)

[PATCH 1/2] nfsd: make NFSv4 lease time per net

[Stanislav Kinsbursky]

Lease time is a part of NFSv4 state engine, which is constructed per network
namespace.

Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
---
 fs/nfsd/netns.h        |    2 ++
 fs/nfsd/nfs4callback.c |    8 +++++---
 fs/nfsd/nfs4state.c    |   11 +++++------
 fs/nfsd/nfs4xdr.c      |    4 +++-
 fs/nfsd/nfsctl.c       |    5 ++++-
 fs/nfsd/nfsd.h         |    1 -
 6 files changed, 19 insertions(+), 12 deletions(-)

diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
index 9047706..0c20be8 100644
--- a/fs/nfsd/netns.h
+++ b/fs/nfsd/netns.h
@@ -87,6 +87,8 @@ struct nfsd_net {
 
 	struct file *rec_file;
 	bool in_grace;
+
+	time_t nfsd4_lease;
 };
 
 extern int nfsd_net_id;
diff --git a/fs/nfsd/nfs4callback.c b/fs/nfsd/nfs4callback.c
index 826cc26..99bc85f 100644
--- a/fs/nfsd/nfs4callback.c
+++ b/fs/nfsd/nfs4callback.c
@@ -36,6 +36,7 @@
 #include <linux/slab.h>
 #include "nfsd.h"
 #include "state.h"
+#include "netns.h"
 
 #define NFSDDBG_FACILITY                NFSDDBG_PROC
 
@@ -625,9 +626,10 @@ static const struct rpc_program cb_program = {
 	.pipe_dir_name		= "nfsd4_cb",
 };
 
-static int max_cb_time(void)
+static int max_cb_time(struct net *net)
 {
-	return max(nfsd4_lease/10, (time_t)1) * HZ;
+	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
+	return max(nn->nfsd4_lease/10, (time_t)1) * HZ;
 }
 
 static struct rpc_cred *callback_cred;
@@ -659,7 +661,7 @@ static struct rpc_cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc
 static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
 {
 	struct rpc_timeout	timeparms = {
-		.to_initval	= max_cb_time(),
+		.to_initval	= max_cb_time(clp->net),
 		.to_retries	= 0,
 	};
 	struct rpc_create_args args = {
diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index 93c7882..56b0ebb 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -51,7 +51,6 @@
 #define NFSDDBG_FACILITY                NFSDDBG_PROC
 
 /* Globals */
-time_t nfsd4_lease = 90;     /* default lease time */
 time_t nfsd4_grace = 90;
 
 #define all_ones {{~0,~0},~0}
@@ -3184,7 +3183,7 @@ nfsd4_end_grace(struct nfsd_net *nn)
 	 * to see the (possibly new, possibly shorter) lease time, we
 	 * can safely set the next grace time to the current lease time:
 	 */
-	nfsd4_grace = nfsd4_lease;
+	nfsd4_grace = nn->nfsd4_lease;
 }
 
 static time_t
@@ -3194,9 +3193,9 @@ nfs4_laundromat(struct nfsd_net *nn)
 	struct nfs4_openowner *oo;
 	struct nfs4_delegation *dp;
 	struct list_head *pos, *next, reaplist;
-	time_t cutoff = get_seconds() - nfsd4_lease;
-	time_t t, clientid_val = nfsd4_lease;
-	time_t u, test_val = nfsd4_lease;
+	time_t cutoff = get_seconds() - nn->nfsd4_lease;
+	time_t t, clientid_val = nn->nfsd4_lease;
+	time_t u, test_val = nn->nfsd4_lease;
 
 	nfs4_lock_state();
 
@@ -3245,7 +3244,7 @@ nfs4_laundromat(struct nfsd_net *nn)
 		dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
 		unhash_delegation(dp);
 	}
-	test_val = nfsd4_lease;
+	test_val = nn->nfsd4_lease;
 	list_for_each_safe(pos, next, &nn->close_lru) {
 		oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
 		if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
index 406d0c4..6a5ab39 100644
--- a/fs/nfsd/nfs4xdr.c
+++ b/fs/nfsd/nfs4xdr.c
@@ -53,6 +53,7 @@
 #include "vfs.h"
 #include "state.h"
 #include "cache.h"
+#include "netns.h"
 
 #define NFSDDBG_FACILITY		NFSDDBG_XDR
 
@@ -2091,6 +2092,7 @@ nfsd4_encode_fattr(struct svc_fh *fhp, struct svc_export *exp,
 		.mnt	= exp->ex_path.mnt,
 		.dentry	= dentry,
 	};
+	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 
 	BUG_ON(bmval1 & NFSD_WRITEONLY_ATTRS_WORD1);
 	BUG_ON(bmval0 & ~nfsd_suppattrs0(minorversion));
@@ -2251,7 +2253,7 @@ nfsd4_encode_fattr(struct svc_fh *fhp, struct svc_export *exp,
 	if (bmval0 & FATTR4_WORD0_LEASE_TIME) {
 		if ((buflen -= 4) < 0)
 			goto out_resource;
-		WRITE32(nfsd4_lease);
+		WRITE32(nn->nfsd4_lease);
 	}
 	if (bmval0 & FATTR4_WORD0_RDATTR_ERROR) {
 		if ((buflen -= 4) < 0)
diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
index dab350d..4930981 100644
--- a/fs/nfsd/nfsctl.c
+++ b/fs/nfsd/nfsctl.c
@@ -912,7 +912,8 @@ static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size, time_
  */
 static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
 {
-	return nfsd4_write_time(file, buf, size, &nfsd4_lease);
+	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
+	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease);
 }
 
 /**
@@ -1063,6 +1064,7 @@ int nfsd_net_id;
 static __net_init int nfsd_init_net(struct net *net)
 {
 	int retval;
+	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 
 	retval = nfsd_export_init(net);
 	if (retval)
@@ -1070,6 +1072,7 @@ static __net_init int nfsd_init_net(struct net *net)
 	retval = nfsd_idmap_init(net);
 	if (retval)
 		goto out_idmap_error;
+	nn->nfsd4_lease = 90;	/* default lease time */
 	return 0;
 
 out_idmap_error:
diff --git a/fs/nfsd/nfsd.h b/fs/nfsd/nfsd.h
index d7b210b..a8f7325 100644
--- a/fs/nfsd/nfsd.h
+++ b/fs/nfsd/nfsd.h
@@ -276,7 +276,6 @@ extern struct timeval	nfssvc_boot;
 
 #ifdef CONFIG_NFSD_V4
 
-extern time_t nfsd4_lease;
 extern time_t nfsd4_grace;
 
 /* before processing a COMPOUND operation, we have to check that there

[PATCH 2/2] nfsd: make NFSv4 grace time per net

[Stanislav Kinsbursky]

Grace time is a part of NFSv4 state engine, which is constructed per network
namespace.

Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
---
 fs/nfsd/netns.h     |    1 +
 fs/nfsd/nfs4state.c |    9 +++------
 fs/nfsd/nfsctl.c    |    4 +++-
 fs/nfsd/nfsd.h      |    2 --
 4 files changed, 7 insertions(+), 9 deletions(-)

diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
index 0c20be8..2c4b2e2 100644
--- a/fs/nfsd/netns.h
+++ b/fs/nfsd/netns.h
@@ -89,6 +89,7 @@ struct nfsd_net {
 	bool in_grace;
 
 	time_t nfsd4_lease;
+	time_t nfsd4_grace;
 };
 
 extern int nfsd_net_id;
diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index 56b0ebb..36d3983 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -50,9 +50,6 @@
 
 #define NFSDDBG_FACILITY                NFSDDBG_PROC
 
-/* Globals */
-time_t nfsd4_grace = 90;
-
 #define all_ones {{~0,~0},~0}
 static const stateid_t one_stateid = {
 	.si_generation = ~0,
@@ -3183,7 +3180,7 @@ nfsd4_end_grace(struct nfsd_net *nn)
 	 * to see the (possibly new, possibly shorter) lease time, we
 	 * can safely set the next grace time to the current lease time:
 	 */
-	nfsd4_grace = nn->nfsd4_lease;
+	nn->nfsd4_grace = nn->nfsd4_lease;
 }
 
 static time_t
@@ -4880,8 +4877,8 @@ nfs4_state_start_net(struct net *net)
 	locks_start_grace(net, &nn->nfsd4_manager);
 	nn->grace_ended = false;
 	printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
-	       nfsd4_grace, net);
-	queue_delayed_work(laundry_wq, &nn->laundromat_work, nfsd4_grace * HZ);
+	       nn->nfsd4_grace, net);
+	queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
 	return 0;
 }
 
diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
index 4930981..7e41bd1 100644
--- a/fs/nfsd/nfsctl.c
+++ b/fs/nfsd/nfsctl.c
@@ -928,7 +928,8 @@ static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
  */
 static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
 {
-	return nfsd4_write_time(file, buf, size, &nfsd4_grace);
+	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
+	return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace);
 }
 
 static ssize_t __write_recoverydir(struct file *file, char *buf, size_t size)
@@ -1073,6 +1074,7 @@ static __net_init int nfsd_init_net(struct net *net)
 	if (retval)
 		goto out_idmap_error;
 	nn->nfsd4_lease = 90;	/* default lease time */
+	nn->nfsd4_grace = 90;
 	return 0;
 
 out_idmap_error:
diff --git a/fs/nfsd/nfsd.h b/fs/nfsd/nfsd.h
index a8f7325..5eea0f5 100644
--- a/fs/nfsd/nfsd.h
+++ b/fs/nfsd/nfsd.h
@@ -276,8 +276,6 @@ extern struct timeval	nfssvc_boot;
 
 #ifdef CONFIG_NFSD_V4
 
-extern time_t nfsd4_grace;
-
 /* before processing a COMPOUND operation, we have to check that there
  * is enough space in the buffer for XDR encode to succeed.  otherwise,
  * we might process an operation with side effects, and be unable to

Re: [PATCH 1/2] nfsd: make NFSv4 lease time per net

[J. Bruce Fields]

On Tue, Nov 27, 2012 at 02:11:44PM +0300, Stanislav Kinsbursky wrote:
> diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> index dab350d..4930981 100644
> --- a/fs/nfsd/nfsctl.c
> +++ b/fs/nfsd/nfsctl.c
> @@ -912,7 +912,8 @@ static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size, time_
>   */
>  static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
>  {
> -	return nfsd4_write_time(file, buf, size, &nfsd4_lease);
> +	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
> +	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease);

This is called in the context of whatever process writes to
nfsv4leasetime, so should be using its network namespace, right?

--b.

Re: [PATCH 2/2] nfsd: make NFSv4 grace time per net

[J. Bruce Fields]

On Tue, Nov 27, 2012 at 02:11:49PM +0300, Stanislav Kinsbursky wrote:
> diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> index 4930981..7e41bd1 100644
> --- a/fs/nfsd/nfsctl.c
> +++ b/fs/nfsd/nfsctl.c
> @@ -928,7 +928,8 @@ static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
>   */
>  static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
>  {
> -	return nfsd4_write_time(file, buf, size, &nfsd4_grace);
> +	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
> +	return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace);
>  }

Same comment as for lease time.--b.

Re: [PATCH 1/2] nfsd: make NFSv4 lease time per net

[Stanislav Kinsbursky]

28.11.2012 19:09, J. Bruce Fields пишет:
> On Tue, Nov 27, 2012 at 02:11:44PM +0300, Stanislav Kinsbursky wrote:
>> diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
>> index dab350d..4930981 100644
>> --- a/fs/nfsd/nfsctl.c
>> +++ b/fs/nfsd/nfsctl.c
>> @@ -912,7 +912,8 @@ static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size, time_
>>    */
>>   static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
>>   {
>> -	return nfsd4_write_time(file, buf, size, &nfsd4_lease);
>> +	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
>> +	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease);
>
> This is called in the context of whatever process writes to
> nfsv4leasetime, so should be using its network namespace, right?
>

This is, actually, a interim solution to preserve existent logic.
I.e. I'm going to convert "nfsd" filesystem into per-net one (like rpc_pipefs). I, actually, already done it in my tree.
Thus proper network namespace will be taken from nfsd superblock.

> --b.
>


-- 
Best regards,
Stanislav Kinsbursky

Re: [PATCH 1/2] nfsd: make NFSv4 lease time per net

[J. Bruce Fields]

On Wed, Nov 28, 2012 at 07:12:03PM +0400, Stanislav Kinsbursky wrote:
> 28.11.2012 19:09, J. Bruce Fields пишет:
> >On Tue, Nov 27, 2012 at 02:11:44PM +0300, Stanislav Kinsbursky wrote:
> >>diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> >>index dab350d..4930981 100644
> >>--- a/fs/nfsd/nfsctl.c
> >>+++ b/fs/nfsd/nfsctl.c
> >>@@ -912,7 +912,8 @@ static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size, time_
> >>   */
> >>  static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
> >>  {
> >>-	return nfsd4_write_time(file, buf, size, &nfsd4_lease);
> >>+	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
> >>+	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease);
> >
> >This is called in the context of whatever process writes to
> >nfsv4leasetime, so should be using its network namespace, right?
> >
> 
> This is, actually, a interim solution to preserve existent logic.
> I.e. I'm going to convert "nfsd" filesystem into per-net one (like rpc_pipefs). I, actually, already done it in my tree.
> Thus proper network namespace will be taken from nfsd superblock.

OK, remind me how that works?  It's mounted just once, but each network
namespace gets a different view of the filesystem?

--b.

Re: [PATCH 1/2] nfsd: make NFSv4 lease time per net

[Stanislav Kinsbursky]

28.11.2012 19:15, J. Bruce Fields пишет:
> On Wed, Nov 28, 2012 at 07:12:03PM +0400, Stanislav Kinsbursky wrote:
>> 28.11.2012 19:09, J. Bruce Fields пишет:
>>> On Tue, Nov 27, 2012 at 02:11:44PM +0300, Stanislav Kinsbursky wrote:
>>>> diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
>>>> index dab350d..4930981 100644
>>>> --- a/fs/nfsd/nfsctl.c
>>>> +++ b/fs/nfsd/nfsctl.c
>>>> @@ -912,7 +912,8 @@ static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size, time_
>>>>    */
>>>>   static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
>>>>   {
>>>> -	return nfsd4_write_time(file, buf, size, &nfsd4_lease);
>>>> +	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
>>>> +	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease);
>>>
>>> This is called in the context of whatever process writes to
>>> nfsv4leasetime, so should be using its network namespace, right?
>>>
>>
>> This is, actually, a interim solution to preserve existent logic.
>> I.e. I'm going to convert "nfsd" filesystem into per-net one (like rpc_pipefs). I, actually, already done it in my tree.
>> Thus proper network namespace will be taken from nfsd superblock.
>
> OK, remind me how that works?  It's mounted just once, but each network
> namespace gets a different view of the filesystem?
>

Nope. It's a single mount point, but per-namespace (network, in our case) - not globally.
Pointer to namespace will be placed on sb->s_fs_info.

> --b.
>


-- 
Best regards,
Stanislav Kinsbursky

Re: [PATCH 1/2] nfsd: make NFSv4 lease time per net

[J. Bruce Fields]

On Wed, Nov 28, 2012 at 07:35:20PM +0400, Stanislav Kinsbursky wrote:
> 28.11.2012 19:15, J. Bruce Fields пишет:
> >On Wed, Nov 28, 2012 at 07:12:03PM +0400, Stanislav Kinsbursky wrote:
> >>28.11.2012 19:09, J. Bruce Fields пишет:
> >>>On Tue, Nov 27, 2012 at 02:11:44PM +0300, Stanislav Kinsbursky wrote:
> >>>>diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> >>>>index dab350d..4930981 100644
> >>>>--- a/fs/nfsd/nfsctl.c
> >>>>+++ b/fs/nfsd/nfsctl.c
> >>>>@@ -912,7 +912,8 @@ static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size, time_
> >>>>   */
> >>>>  static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
> >>>>  {
> >>>>-	return nfsd4_write_time(file, buf, size, &nfsd4_lease);
> >>>>+	struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
> >>>>+	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease);
> >>>
> >>>This is called in the context of whatever process writes to
> >>>nfsv4leasetime, so should be using its network namespace, right?
> >>>
> >>
> >>This is, actually, a interim solution to preserve existent logic.
> >>I.e. I'm going to convert "nfsd" filesystem into per-net one (like rpc_pipefs). I, actually, already done it in my tree.
> >>Thus proper network namespace will be taken from nfsd superblock.
> >
> >OK, remind me how that works?  It's mounted just once, but each network
> >namespace gets a different view of the filesystem?
> >
> 
> Nope. It's a single mount point, but per-namespace (network, in our case) - not globally.
> Pointer to namespace will be placed on sb->s_fs_info.

OK, anyway, applying these.

--b.