diff mbox

[RFC,v0,10/49] pnfsd: use sbid hash table to map super_blocks to devid major identifiers

Message ID 1380220831-13136-1-git-send-email-bhalevy@primarydata.com (mailing list archive)
State New, archived
Headers show

Commit Message

Benny Halevy Sept. 26, 2013, 6:40 p.m. UTC 
From: Benny Halevy <bhalevy@panasas.com>

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[pnfsd: alloc_sid should kmalloc a object not a pointer]
Signed-off-by: Bian Naimeng <biannm@cn.fujitsu.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@primarydata.com>
---
 fs/nfsd/nfs4pnfsd.c | 120 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/nfsd/pnfsd.h     |   2 +
 2 files changed, 122 insertions(+)

Comments

Bruce Fields Oct. 1, 2013, 10:14 p.m. UTC | #1 
See previous comments.  What guarantees these superblock pointers stay
good as long as they're in the cache?

--b.

On Thu, Sep 26, 2013 at 02:40:31PM -0400, Benny Halevy wrote:
> From: Benny Halevy <bhalevy@panasas.com>
> 
> Signed-off-by: Benny Halevy <bhalevy@panasas.com>
> [pnfsd: alloc_sid should kmalloc a object not a pointer]
> Signed-off-by: Bian Naimeng <biannm@cn.fujitsu.com>
> Signed-off-by: Benny Halevy <bhalevy@panasas.com>
> Signed-off-by: Benny Halevy <bhalevy@primarydata.com>
> ---
>  fs/nfsd/nfs4pnfsd.c | 120 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>  fs/nfsd/pnfsd.h     |   2 +
>  2 files changed, 122 insertions(+)
> 
> diff --git a/fs/nfsd/nfs4pnfsd.c b/fs/nfsd/nfs4pnfsd.c
> index cb28207..9a7cbc9 100644
> --- a/fs/nfsd/nfs4pnfsd.c
> +++ b/fs/nfsd/nfs4pnfsd.c
> @@ -25,3 +25,123 @@
>  
>  #define NFSDDBG_FACILITY                NFSDDBG_PNFS
>  
> +static DEFINE_SPINLOCK(layout_lock);
> +
> +/* hash table for nfsd4_pnfs_deviceid.sbid */
> +#define SBID_HASH_BITS	8
> +#define SBID_HASH_SIZE	(1 << SBID_HASH_BITS)
> +#define SBID_HASH_MASK	(SBID_HASH_SIZE - 1)
> +
> +struct sbid_tracker {
> +	u64 id;
> +	struct super_block *sb;
> +	struct list_head hash;
> +};
> +
> +static u64 current_sbid;
> +static struct list_head sbid_hashtbl[SBID_HASH_SIZE];
> +
> +static unsigned long
> +sbid_hashval(struct super_block *sb)
> +{
> +	return hash_ptr(sb, SBID_HASH_BITS);
> +}
> +
> +static struct sbid_tracker *
> +alloc_sbid(void)
> +{
> +	return kmalloc(sizeof(struct sbid_tracker), GFP_KERNEL);
> +}
> +
> +static void
> +destroy_sbid(struct sbid_tracker *sbid)
> +{
> +	spin_lock(&layout_lock);
> +	list_del(&sbid->hash);
> +	spin_unlock(&layout_lock);
> +	kfree(sbid);
> +}
> +
> +void
> +nfsd4_free_pnfs_slabs(void)
> +{
> +	int i;
> +	struct sbid_tracker *sbid;
> +
> +	for (i = 0; i < SBID_HASH_SIZE; i++) {
> +		while (!list_empty(&sbid_hashtbl[i])) {
> +			sbid = list_first_entry(&sbid_hashtbl[i],
> +						struct sbid_tracker,
> +						hash);
> +			destroy_sbid(sbid);
> +		}
> +	}
> +}
> +
> +int
> +nfsd4_init_pnfs_slabs(void)
> +{
> +	int i;
> +
> +	for (i = 0; i < SBID_HASH_SIZE; i++)
> +		INIT_LIST_HEAD(&sbid_hashtbl[i]);
> +
> +	return 0;
> +}
> +
> +static u64
> +alloc_init_sbid(struct super_block *sb)
> +{
> +	struct sbid_tracker *sbid;
> +	struct sbid_tracker *new = alloc_sbid();
> +	unsigned long hash_idx = sbid_hashval(sb);
> +	u64 id = 0;
> +
> +	if (likely(new)) {
> +		spin_lock(&layout_lock);
> +		id = ++current_sbid;
> +		new->id = (id << SBID_HASH_BITS) | (hash_idx & SBID_HASH_MASK);
> +		id = new->id;
> +		BUG_ON(id == 0);
> +		new->sb = sb;
> +
> +		list_for_each_entry (sbid, &sbid_hashtbl[hash_idx], hash)
> +			if (sbid->sb == sb) {
> +				kfree(new);
> +				id = sbid->id;
> +				spin_unlock(&layout_lock);
> +				return id;
> +			}
> +		list_add(&new->hash, &sbid_hashtbl[hash_idx]);
> +		spin_unlock(&layout_lock);
> +	}
> +	return id;
> +}
> +
> +static u64
> +find_create_sbid(struct super_block *sb)
> +{
> +	struct sbid_tracker *sbid;
> +	unsigned long hash_idx = sbid_hashval(sb);
> +	int pos = 0;
> +	u64 id = 0;
> +
> +	spin_lock(&layout_lock);
> +	list_for_each_entry (sbid, &sbid_hashtbl[hash_idx], hash) {
> +		pos++;
> +		if (sbid->sb != sb)
> +			continue;
> +		if (pos > 1) {
> +			list_del(&sbid->hash);
> +			list_add(&sbid->hash, &sbid_hashtbl[hash_idx]);
> +		}
> +		id = sbid->id;
> +		break;
> +	}
> +	spin_unlock(&layout_lock);
> +
> +	if (!id)
> +		id = alloc_init_sbid(sb);
> +
> +	return id;
> +}
> diff --git a/fs/nfsd/pnfsd.h b/fs/nfsd/pnfsd.h
> index 7c46791..29ea2e7 100644
> --- a/fs/nfsd/pnfsd.h
> +++ b/fs/nfsd/pnfsd.h
> @@ -36,4 +36,6 @@
>  
>  #include <linux/nfsd/nfsd4_pnfs.h>
>  
> +#include "xdr4.h"
> +
>  #endif /* LINUX_NFSD_PNFSD_H */
> -- 
> 1.8.3.1
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Benny Halevy Oct. 2, 2013, 2:32 p.m. UTC | #2 
On 2013-10-02 01:14, J. Bruce Fields wrote:
> See previous comments.  What guarantees these superblock pointers stay
> good as long as they're in the cache?

Currently, the dependency on nfsd.ko should hold them but that should go away.
Trying to think about referencing svc_export instead, we use find_sbid_id
to get to the superblock in nfsd4_getdevinfo since we have no current fh.
And we need the superblock to call into the fs sb->s_pnfs_op->get_device_info
later in nfsd4_encode_getdevinfo.

Just to make sure, we can safely get to the sb via exp->ex_path.dentry->d_inode->i_sb
right?

Benny

> 
> --b.
> 
> On Thu, Sep 26, 2013 at 02:40:31PM -0400, Benny Halevy wrote:
>> From: Benny Halevy <bhalevy@panasas.com>
>>
>> Signed-off-by: Benny Halevy <bhalevy@panasas.com>
>> [pnfsd: alloc_sid should kmalloc a object not a pointer]
>> Signed-off-by: Bian Naimeng <biannm@cn.fujitsu.com>
>> Signed-off-by: Benny Halevy <bhalevy@panasas.com>
>> Signed-off-by: Benny Halevy <bhalevy@primarydata.com>
>> ---
>>  fs/nfsd/nfs4pnfsd.c | 120 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>>  fs/nfsd/pnfsd.h     |   2 +
>>  2 files changed, 122 insertions(+)
>>
>> diff --git a/fs/nfsd/nfs4pnfsd.c b/fs/nfsd/nfs4pnfsd.c
>> index cb28207..9a7cbc9 100644
>> --- a/fs/nfsd/nfs4pnfsd.c
>> +++ b/fs/nfsd/nfs4pnfsd.c
>> @@ -25,3 +25,123 @@
>>  
>>  #define NFSDDBG_FACILITY                NFSDDBG_PNFS
>>  
>> +static DEFINE_SPINLOCK(layout_lock);
>> +
>> +/* hash table for nfsd4_pnfs_deviceid.sbid */
>> +#define SBID_HASH_BITS	8
>> +#define SBID_HASH_SIZE	(1 << SBID_HASH_BITS)
>> +#define SBID_HASH_MASK	(SBID_HASH_SIZE - 1)
>> +
>> +struct sbid_tracker {
>> +	u64 id;
>> +	struct super_block *sb;
>> +	struct list_head hash;
>> +};
>> +
>> +static u64 current_sbid;
>> +static struct list_head sbid_hashtbl[SBID_HASH_SIZE];
>> +
>> +static unsigned long
>> +sbid_hashval(struct super_block *sb)
>> +{
>> +	return hash_ptr(sb, SBID_HASH_BITS);
>> +}
>> +
>> +static struct sbid_tracker *
>> +alloc_sbid(void)
>> +{
>> +	return kmalloc(sizeof(struct sbid_tracker), GFP_KERNEL);
>> +}
>> +
>> +static void
>> +destroy_sbid(struct sbid_tracker *sbid)
>> +{
>> +	spin_lock(&layout_lock);
>> +	list_del(&sbid->hash);
>> +	spin_unlock(&layout_lock);
>> +	kfree(sbid);
>> +}
>> +
>> +void
>> +nfsd4_free_pnfs_slabs(void)
>> +{
>> +	int i;
>> +	struct sbid_tracker *sbid;
>> +
>> +	for (i = 0; i < SBID_HASH_SIZE; i++) {
>> +		while (!list_empty(&sbid_hashtbl[i])) {
>> +			sbid = list_first_entry(&sbid_hashtbl[i],
>> +						struct sbid_tracker,
>> +						hash);
>> +			destroy_sbid(sbid);
>> +		}
>> +	}
>> +}
>> +
>> +int
>> +nfsd4_init_pnfs_slabs(void)
>> +{
>> +	int i;
>> +
>> +	for (i = 0; i < SBID_HASH_SIZE; i++)
>> +		INIT_LIST_HEAD(&sbid_hashtbl[i]);
>> +
>> +	return 0;
>> +}
>> +
>> +static u64
>> +alloc_init_sbid(struct super_block *sb)
>> +{
>> +	struct sbid_tracker *sbid;
>> +	struct sbid_tracker *new = alloc_sbid();
>> +	unsigned long hash_idx = sbid_hashval(sb);
>> +	u64 id = 0;
>> +
>> +	if (likely(new)) {
>> +		spin_lock(&layout_lock);
>> +		id = ++current_sbid;
>> +		new->id = (id << SBID_HASH_BITS) | (hash_idx & SBID_HASH_MASK);
>> +		id = new->id;
>> +		BUG_ON(id == 0);
>> +		new->sb = sb;
>> +
>> +		list_for_each_entry (sbid, &sbid_hashtbl[hash_idx], hash)
>> +			if (sbid->sb == sb) {
>> +				kfree(new);
>> +				id = sbid->id;
>> +				spin_unlock(&layout_lock);
>> +				return id;
>> +			}
>> +		list_add(&new->hash, &sbid_hashtbl[hash_idx]);
>> +		spin_unlock(&layout_lock);
>> +	}
>> +	return id;
>> +}
>> +
>> +static u64
>> +find_create_sbid(struct super_block *sb)
>> +{
>> +	struct sbid_tracker *sbid;
>> +	unsigned long hash_idx = sbid_hashval(sb);
>> +	int pos = 0;
>> +	u64 id = 0;
>> +
>> +	spin_lock(&layout_lock);
>> +	list_for_each_entry (sbid, &sbid_hashtbl[hash_idx], hash) {
>> +		pos++;
>> +		if (sbid->sb != sb)
>> +			continue;
>> +		if (pos > 1) {
>> +			list_del(&sbid->hash);
>> +			list_add(&sbid->hash, &sbid_hashtbl[hash_idx]);
>> +		}
>> +		id = sbid->id;
>> +		break;
>> +	}
>> +	spin_unlock(&layout_lock);
>> +
>> +	if (!id)
>> +		id = alloc_init_sbid(sb);
>> +
>> +	return id;
>> +}
>> diff --git a/fs/nfsd/pnfsd.h b/fs/nfsd/pnfsd.h
>> index 7c46791..29ea2e7 100644
>> --- a/fs/nfsd/pnfsd.h
>> +++ b/fs/nfsd/pnfsd.h
>> @@ -36,4 +36,6 @@
>>  
>>  #include <linux/nfsd/nfsd4_pnfs.h>
>>  
>> +#include "xdr4.h"
>> +
>>  #endif /* LINUX_NFSD_PNFSD_H */
>> -- 
>> 1.8.3.1
>>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Bruce Fields Oct. 2, 2013, 3:24 p.m. UTC | #3 
On Wed, Oct 02, 2013 at 05:32:32PM +0300, Benny Halevy wrote:
> On 2013-10-02 01:14, J. Bruce Fields wrote:
> > See previous comments.  What guarantees these superblock pointers stay
> > good as long as they're in the cache?
> 
> Currently, the dependency on nfsd.ko should hold them but that should go away.

I don't see how that prevents anyone from unmounting a filesystem.

> Trying to think about referencing svc_export instead, we use find_sbid_id
> to get to the superblock in nfsd4_getdevinfo since we have no current fh.
> And we need the superblock to call into the fs sb->s_pnfs_op->get_device_info
> later in nfsd4_encode_getdevinfo.
> 
> Just to make sure, we can safely get to the sb via exp->ex_path.dentry->d_inode->i_sb
> right?

Right.

--b.

> 
> Benny
> 
> > 
> > --b.
> > 
> > On Thu, Sep 26, 2013 at 02:40:31PM -0400, Benny Halevy wrote:
> >> From: Benny Halevy <bhalevy@panasas.com>
> >>
> >> Signed-off-by: Benny Halevy <bhalevy@panasas.com>
> >> [pnfsd: alloc_sid should kmalloc a object not a pointer]
> >> Signed-off-by: Bian Naimeng <biannm@cn.fujitsu.com>
> >> Signed-off-by: Benny Halevy <bhalevy@panasas.com>
> >> Signed-off-by: Benny Halevy <bhalevy@primarydata.com>
> >> ---
> >>  fs/nfsd/nfs4pnfsd.c | 120 ++++++++++++++++++++++++++++++++++++++++++++++++++++
> >>  fs/nfsd/pnfsd.h     |   2 +
> >>  2 files changed, 122 insertions(+)
> >>
> >> diff --git a/fs/nfsd/nfs4pnfsd.c b/fs/nfsd/nfs4pnfsd.c
> >> index cb28207..9a7cbc9 100644
> >> --- a/fs/nfsd/nfs4pnfsd.c
> >> +++ b/fs/nfsd/nfs4pnfsd.c
> >> @@ -25,3 +25,123 @@
> >>  
> >>  #define NFSDDBG_FACILITY                NFSDDBG_PNFS
> >>  
> >> +static DEFINE_SPINLOCK(layout_lock);
> >> +
> >> +/* hash table for nfsd4_pnfs_deviceid.sbid */
> >> +#define SBID_HASH_BITS	8
> >> +#define SBID_HASH_SIZE	(1 << SBID_HASH_BITS)
> >> +#define SBID_HASH_MASK	(SBID_HASH_SIZE - 1)
> >> +
> >> +struct sbid_tracker {
> >> +	u64 id;
> >> +	struct super_block *sb;
> >> +	struct list_head hash;
> >> +};
> >> +
> >> +static u64 current_sbid;
> >> +static struct list_head sbid_hashtbl[SBID_HASH_SIZE];
> >> +
> >> +static unsigned long
> >> +sbid_hashval(struct super_block *sb)
> >> +{
> >> +	return hash_ptr(sb, SBID_HASH_BITS);
> >> +}
> >> +
> >> +static struct sbid_tracker *
> >> +alloc_sbid(void)
> >> +{
> >> +	return kmalloc(sizeof(struct sbid_tracker), GFP_KERNEL);
> >> +}
> >> +
> >> +static void
> >> +destroy_sbid(struct sbid_tracker *sbid)
> >> +{
> >> +	spin_lock(&layout_lock);
> >> +	list_del(&sbid->hash);
> >> +	spin_unlock(&layout_lock);
> >> +	kfree(sbid);
> >> +}
> >> +
> >> +void
> >> +nfsd4_free_pnfs_slabs(void)
> >> +{
> >> +	int i;
> >> +	struct sbid_tracker *sbid;
> >> +
> >> +	for (i = 0; i < SBID_HASH_SIZE; i++) {
> >> +		while (!list_empty(&sbid_hashtbl[i])) {
> >> +			sbid = list_first_entry(&sbid_hashtbl[i],
> >> +						struct sbid_tracker,
> >> +						hash);
> >> +			destroy_sbid(sbid);
> >> +		}
> >> +	}
> >> +}
> >> +
> >> +int
> >> +nfsd4_init_pnfs_slabs(void)
> >> +{
> >> +	int i;
> >> +
> >> +	for (i = 0; i < SBID_HASH_SIZE; i++)
> >> +		INIT_LIST_HEAD(&sbid_hashtbl[i]);
> >> +
> >> +	return 0;
> >> +}
> >> +
> >> +static u64
> >> +alloc_init_sbid(struct super_block *sb)
> >> +{
> >> +	struct sbid_tracker *sbid;
> >> +	struct sbid_tracker *new = alloc_sbid();
> >> +	unsigned long hash_idx = sbid_hashval(sb);
> >> +	u64 id = 0;
> >> +
> >> +	if (likely(new)) {
> >> +		spin_lock(&layout_lock);
> >> +		id = ++current_sbid;
> >> +		new->id = (id << SBID_HASH_BITS) | (hash_idx & SBID_HASH_MASK);
> >> +		id = new->id;
> >> +		BUG_ON(id == 0);
> >> +		new->sb = sb;
> >> +
> >> +		list_for_each_entry (sbid, &sbid_hashtbl[hash_idx], hash)
> >> +			if (sbid->sb == sb) {
> >> +				kfree(new);
> >> +				id = sbid->id;
> >> +				spin_unlock(&layout_lock);
> >> +				return id;
> >> +			}
> >> +		list_add(&new->hash, &sbid_hashtbl[hash_idx]);
> >> +		spin_unlock(&layout_lock);
> >> +	}
> >> +	return id;
> >> +}
> >> +
> >> +static u64
> >> +find_create_sbid(struct super_block *sb)
> >> +{
> >> +	struct sbid_tracker *sbid;
> >> +	unsigned long hash_idx = sbid_hashval(sb);
> >> +	int pos = 0;
> >> +	u64 id = 0;
> >> +
> >> +	spin_lock(&layout_lock);
> >> +	list_for_each_entry (sbid, &sbid_hashtbl[hash_idx], hash) {
> >> +		pos++;
> >> +		if (sbid->sb != sb)
> >> +			continue;
> >> +		if (pos > 1) {
> >> +			list_del(&sbid->hash);
> >> +			list_add(&sbid->hash, &sbid_hashtbl[hash_idx]);
> >> +		}
> >> +		id = sbid->id;
> >> +		break;
> >> +	}
> >> +	spin_unlock(&layout_lock);
> >> +
> >> +	if (!id)
> >> +		id = alloc_init_sbid(sb);
> >> +
> >> +	return id;
> >> +}
> >> diff --git a/fs/nfsd/pnfsd.h b/fs/nfsd/pnfsd.h
> >> index 7c46791..29ea2e7 100644
> >> --- a/fs/nfsd/pnfsd.h
> >> +++ b/fs/nfsd/pnfsd.h
> >> @@ -36,4 +36,6 @@
> >>  
> >>  #include <linux/nfsd/nfsd4_pnfs.h>
> >>  
> >> +#include "xdr4.h"
> >> +
> >>  #endif /* LINUX_NFSD_PNFSD_H */
> >> -- 
> >> 1.8.3.1
> >>
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > 
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Christoph Hellwig Oct. 11, 2013, 7:56 p.m. UTC | #4 
On Thu, Sep 26, 2013 at 02:40:31PM -0400, Benny Halevy wrote:
> From: Benny Halevy <bhalevy@panasas.com>

This is entirely unessecary.  Just make sure all layouts set the sbid
field in getdevicelist to the s_dev value of the filesystem and you
can just use user_get_super.

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Benny Halevy Oct. 13, 2013, 6:11 a.m. UTC | #5 
On 2013-10-11 22:56, Christoph Hellwig wrote:
> On Thu, Sep 26, 2013 at 02:40:31PM -0400, Benny Halevy wrote:
>> From: Benny Halevy <bhalevy@panasas.com>
> 
> This is entirely unessecary.  Just make sure all layouts set the sbid
> field in getdevicelist to the s_dev value of the filesystem and you
> can just use user_get_super.

That's true for filesystmes that have a meaningful s_dev, unlike exofs,
but this functionality can be added later respectively.

Benny
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Christoph Hellwig Oct. 13, 2013, 11:08 a.m. UTC | #6 
On Sun, Oct 13, 2013 at 09:11:40AM +0300, Benny Halevy wrote:
> On 2013-10-11 22:56, Christoph Hellwig wrote:
> > On Thu, Sep 26, 2013 at 02:40:31PM -0400, Benny Halevy wrote:
> >> From: Benny Halevy <bhalevy@panasas.com>
> > 
> > This is entirely unessecary.  Just make sure all layouts set the sbid
> > field in getdevicelist to the s_dev value of the filesystem and you
> > can just use user_get_super.
> 
> That's true for filesystmes that have a meaningful s_dev, unlike exofs,
> but this functionality can be added later respectively.

Even exofs has a s_dev from the anon dev_T allocator.  Given that
layouts aren't supposed to surived over reboots of the server I don't
see a problem with it.

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Benny Halevy Oct. 13, 2013, 12:44 p.m. UTC | #7 
On 2013-10-13 14:08, Christoph Hellwig wrote:
> On Sun, Oct 13, 2013 at 09:11:40AM +0300, Benny Halevy wrote:
>> On 2013-10-11 22:56, Christoph Hellwig wrote:
>>> On Thu, Sep 26, 2013 at 02:40:31PM -0400, Benny Halevy wrote:
>>>> From: Benny Halevy <bhalevy@panasas.com>
>>>
>>> This is entirely unessecary.  Just make sure all layouts set the sbid
>>> field in getdevicelist to the s_dev value of the filesystem and you
>>> can just use user_get_super.
>>
>> That's true for filesystmes that have a meaningful s_dev, unlike exofs,
>> but this functionality can be added later respectively.
> 
> Even exofs has a s_dev from the anon dev_T allocator.  Given that
> layouts aren't supposed to surived over reboots of the server I don't
> see a problem with it.

As far as I can see, exofs doesn't actually doing that.
Currently exofs_fill_super sets s_dev = 0;
Boaz, did I miss anything?

Benny

> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Christoph Hellwig Oct. 14, 2013, 2:15 p.m. UTC | #8 
On Sun, Oct 13, 2013 at 03:44:30PM +0300, Benny Halevy wrote:
> As far as I can see, exofs doesn't actually doing that.
> Currently exofs_fill_super sets s_dev = 0;
> Boaz, did I miss anything?

Just removing that line should fix it, mount_nodev already takes care of
getting a proper s_dev through set_anon_super.  It'll also need to
switch .kill_sb to kill_anon_super just like the other filesystems using
mount_nodev.

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

Patch

diff --git a/fs/nfsd/nfs4pnfsd.c b/fs/nfsd/nfs4pnfsd.c
index cb28207..9a7cbc9 100644
--- a/fs/nfsd/nfs4pnfsd.c
+++ b/fs/nfsd/nfs4pnfsd.c
@@ -25,3 +25,123 @@ 
 
 #define NFSDDBG_FACILITY                NFSDDBG_PNFS
 
+static DEFINE_SPINLOCK(layout_lock);
+
+/* hash table for nfsd4_pnfs_deviceid.sbid */
+#define SBID_HASH_BITS	8
+#define SBID_HASH_SIZE	(1 << SBID_HASH_BITS)
+#define SBID_HASH_MASK	(SBID_HASH_SIZE - 1)
+
+struct sbid_tracker {
+	u64 id;
+	struct super_block *sb;
+	struct list_head hash;
+};
+
+static u64 current_sbid;
+static struct list_head sbid_hashtbl[SBID_HASH_SIZE];
+
+static unsigned long
+sbid_hashval(struct super_block *sb)
+{
+	return hash_ptr(sb, SBID_HASH_BITS);
+}
+
+static struct sbid_tracker *
+alloc_sbid(void)
+{
+	return kmalloc(sizeof(struct sbid_tracker), GFP_KERNEL);
+}
+
+static void
+destroy_sbid(struct sbid_tracker *sbid)
+{
+	spin_lock(&layout_lock);
+	list_del(&sbid->hash);
+	spin_unlock(&layout_lock);
+	kfree(sbid);
+}
+
+void
+nfsd4_free_pnfs_slabs(void)
+{
+	int i;
+	struct sbid_tracker *sbid;
+
+	for (i = 0; i < SBID_HASH_SIZE; i++) {
+		while (!list_empty(&sbid_hashtbl[i])) {
+			sbid = list_first_entry(&sbid_hashtbl[i],
+						struct sbid_tracker,
+						hash);
+			destroy_sbid(sbid);
+		}
+	}
+}
+
+int
+nfsd4_init_pnfs_slabs(void)
+{
+	int i;
+
+	for (i = 0; i < SBID_HASH_SIZE; i++)
+		INIT_LIST_HEAD(&sbid_hashtbl[i]);
+
+	return 0;
+}
+
+static u64
+alloc_init_sbid(struct super_block *sb)
+{
+	struct sbid_tracker *sbid;
+	struct sbid_tracker *new = alloc_sbid();
+	unsigned long hash_idx = sbid_hashval(sb);
+	u64 id = 0;
+
+	if (likely(new)) {
+		spin_lock(&layout_lock);
+		id = ++current_sbid;
+		new->id = (id << SBID_HASH_BITS) | (hash_idx & SBID_HASH_MASK);
+		id = new->id;
+		BUG_ON(id == 0);
+		new->sb = sb;
+
+		list_for_each_entry (sbid, &sbid_hashtbl[hash_idx], hash)
+			if (sbid->sb == sb) {
+				kfree(new);
+				id = sbid->id;
+				spin_unlock(&layout_lock);
+				return id;
+			}
+		list_add(&new->hash, &sbid_hashtbl[hash_idx]);
+		spin_unlock(&layout_lock);
+	}
+	return id;
+}
+
+static u64
+find_create_sbid(struct super_block *sb)
+{
+	struct sbid_tracker *sbid;
+	unsigned long hash_idx = sbid_hashval(sb);
+	int pos = 0;
+	u64 id = 0;
+
+	spin_lock(&layout_lock);
+	list_for_each_entry (sbid, &sbid_hashtbl[hash_idx], hash) {
+		pos++;
+		if (sbid->sb != sb)
+			continue;
+		if (pos > 1) {
+			list_del(&sbid->hash);
+			list_add(&sbid->hash, &sbid_hashtbl[hash_idx]);
+		}
+		id = sbid->id;
+		break;
+	}
+	spin_unlock(&layout_lock);
+
+	if (!id)
+		id = alloc_init_sbid(sb);
+
+	return id;
+}
diff --git a/fs/nfsd/pnfsd.h b/fs/nfsd/pnfsd.h
index 7c46791..29ea2e7 100644
--- a/fs/nfsd/pnfsd.h
+++ b/fs/nfsd/pnfsd.h
@@ -36,4 +36,6 @@ 
 
 #include <linux/nfsd/nfsd4_pnfs.h>
 
+#include "xdr4.h"
+
 #endif /* LINUX_NFSD_PNFSD_H */