decode_device

Discussion:

decode_device

Marc Eshel

2014-10-08 01:33:17 UTC

I see that NFS4_PNFS_MAX_STRIPE_CNT is 4096 but when give the client a cnt
that is bigger than 1024 it fails, is that the limitation of kcalloc() ?

366341.304362] <-- _nfs4_proc_getdeviceinfo status=0
[366341.304364] filelayout_get_device_info getdevice info returns 0
[366341.304368] decode_device stripe count 1512
[366341.304370] decode_device ERROR: returning NULL
[366341.304372] NFS: decode_and_add_device: Could not decode or add device
[366341.304428] <-- filelayout_get_device_info dsaddr (null)
[366341.304430] --> filelayout_check_layout returns -22
[366341.304433] pnfs_layout_process: Could not allocate layout: error -12

decode_device()

...
dprintk("%s stripe count %d\n", __func__, cnt);
if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
printk(KERN_WARNING "NFS: %s: stripe count %d greater than
"
"supported maximum %d\n", __func__,
cnt, NFS4_PNFS_MAX_STRIPE_CNT);
goto out_err_free_scratch;
}

/* read stripe indices */
stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
if (!stripe_indices)
goto out_err_free_scratch;

Thanks, Marc.

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

Tom Haynes

2014-10-08 12:46:24 UTC

Permalink

Post by Marc Eshel
I see that NFS4_PNFS_MAX_STRIPE_CNT is 4096 but when give the client a cnt
that is bigger than 1024 it fails, is that the limitation of kcalloc() ?
366341.304362] <-- _nfs4_proc_getdeviceinfo status=0
[366341.304364] filelayout_get_device_info getdevice info returns 0
[366341.304368] decode_device stripe count 1512
[366341.304370] decode_device ERROR: returning NULL
[366341.304372] NFS: decode_and_add_device: Could not decode or add device
[366341.304428] <-- filelayout_get_device_info dsaddr (null)
[366341.304430] --> filelayout_check_layout returns -22
[366341.304433] pnfs_layout_process: Could not allocate layout: error -12
decode_device()
...
dprintk("%s stripe count %d\n", __func__, cnt);
if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
printk(KERN_WARNING "NFS: %s: stripe count %d greater than
"
"supported maximum %d\n", __func__,
cnt, NFS4_PNFS_MAX_STRIPE_CNT);
goto out_err_free_scratch;
}
/* read stripe indices */
stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
if (!stripe_indices)
goto out_err_free_scratch;
Thanks, Marc.

Hi Marc,

You are asking for 1512 bytes of memory. That would be looking for a 2k slab.

What flags are being passed down in gfp_flags?

Are you experiencing any other memory issues?

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

Marc Eshel

2014-10-08 16:00:05 UTC

Permalink

Date: 10/08/2014 05:46 AM
Subject: Re: decode_device

Hi Marc,
You are asking for 1512 bytes of memory. That would be looking for a 2k slab.
What flags are being passed down in gfp_flags?
Are you experiencing any other memory issues?

Hi Tom,
It was working fine for a list of up to 1024 so I assumed it was a
kcalloc, I will have to add more debugging to find the problem.
Thanks, Marc.

Tom
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
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-***@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

Marc Eshel

2014-10-08 18:41:16 UTC

Permalink

It looks like the problem is in the following line in decode_device() p
= xdr_inline_decode(&stream, cnt << 2);
p is set to NULL
is there a limitation in xdr to decode a stream that is bigger than 4096?
it works fine for cnt of 1024 but not 1512

Marc.

From: Marc Eshel/Almaden/IBM
To: Tom Haynes <thomas.haynes-7I+n7zu2hftEKMMhf/***@public.gmane.org>,
Cc: linux-nfs-***@public.gmane.org, linux-nfs-owner-***@public.gmane.org, Marc
Eshel/Almaden/***@IBMUS, Trond Myklebust <trond.myklebust-7I+n7zu2hftEKMMhf/***@public.gmane.org>
Date: 10/08/2014 08:05 AM
Subject: Re: decode_device

Date: 10/08/2014 05:46 AM
Subject: Re: decode_device

Hi Marc,
You are asking for 1512 bytes of memory. That would be looking for a 2k slab.
What flags are being passed down in gfp_flags?
Are you experiencing any other memory issues?

Hi Tom,
It was working fine for a list of up to 1024 so I assumed it was a
kcalloc, I will have to add more debugging to find the problem.
Thanks, Marc.

Tom
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
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-***@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

Peng Tao

2014-10-09 15:33:35 UTC

Permalink

Post by Marc Eshel
It looks like the problem is in the following line in decode_device() p
= xdr_inline_decode(&stream, cnt << 2);
p is set to NULL
is there a limitation in xdr to decode a stream that is bigger than 4096?

Here's the limitation in decode_device():
78 /* set up xdr stream */
79 scratch = alloc_page(gfp_flags);
80 if (!scratch)
81 goto out_err;
82
83 xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
84 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);

Post by Marc Eshel
it works fine for cnt of 1024 but not 1512
Marc.
From: Marc Eshel/Almaden/IBM
Date: 10/08/2014 08:05 AM
Subject: Re: decode_device

Date: 10/08/2014 05:46 AM
Subject: Re: decode_device

Post by Marc Eshel
I see that NFS4_PNFS_MAX_STRIPE_CNT is 4096 but when give the client a

cnt

Post by Marc Eshel
that is bigger than 1024 it fails, is that the limitation of kcalloc()

Post by Marc Eshel
366341.304362] <-- _nfs4_proc_getdeviceinfo status=0
[366341.304364] filelayout_get_device_info getdevice info returns 0
[366341.304368] decode_device stripe count 1512
[366341.304370] decode_device ERROR: returning NULL
[366341.304372] NFS: decode_and_add_device: Could not decode or add

device

Post by Marc Eshel
[366341.304428] <-- filelayout_get_device_info dsaddr (null)
[366341.304430] --> filelayout_check_layout returns -22
[366341.304433] pnfs_layout_process: Could not allocate layout: error

-12

Post by Marc Eshel
decode_device()
...
dprintk("%s stripe count %d\n", __func__, cnt);
if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
printk(KERN_WARNING "NFS: %s: stripe count %d greater

than

Post by Marc Eshel
"
"supported maximum %d\n", __func__,
cnt, NFS4_PNFS_MAX_STRIPE_CNT);
goto out_err_free_scratch;
}
/* read stripe indices */
stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
if (!stripe_indices)
goto out_err_free_scratch;
Thanks, Marc.

Hi Marc,
You are asking for 1512 bytes of memory. That would be looking for a 2k

slab.

What flags are being passed down in gfp_flags?
Are you experiencing any other memory issues?

Hi Tom,
It was working fine for a list of up to 1024 so I assumed it was a
kcalloc, I will have to add more debugging to find the problem.
Thanks, Marc.

Tom
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
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-***@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

Marc Eshel

2014-10-09 18:02:50 UTC

Permalink

Date: 10/09/2014 08:40 AM
Subject: Re: decode_device

Post by Marc Eshel
It looks like the problem is in the following line in decode_device()

Post by Marc Eshel
= xdr_inline_decode(&stream, cnt << 2);
p is set to NULL
is there a limitation in xdr to decode a stream that is bigger than 4096?

78 /* set up xdr stream */
79 scratch = alloc_page(gfp_flags);
80 if (!scratch)
81 goto out_err;
82
83 xdr_init_decode_pages(&stream, &buf, pdev->pages,

pdev->pglen);

84 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);

Yes, so NFS4_PNFS_MAX_STRIPE_CNT that is now set to 4096 will not be
possible it can only be 1024 if we are limited to page size.
Is it possible to fix xdr or do we have to accept this limit?
Thanks, Marc.

Post by Marc Eshel
it works fine for cnt of 1024 but not 1512
Marc.
From: Marc Eshel/Almaden/IBM

Marc

Post by Marc Eshel
Date: 10/08/2014 08:05 AM
Subject: Re: decode_device

Date: 10/08/2014 05:46 AM
Subject: Re: decode_device

Post by Marc Eshel
I see that NFS4_PNFS_MAX_STRIPE_CNT is 4096 but when give the client a

cnt

Post by Marc Eshel
that is bigger than 1024 it fails, is that the limitation of kcalloc()

device

-12

Post by Marc Eshel
decode_device()
...
dprintk("%s stripe count %d\n", __func__, cnt);
if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
printk(KERN_WARNING "NFS: %s: stripe count %d greater

than

Hi Marc,
You are asking for 1512 bytes of memory. That would be looking for a 2k

slab.

What flags are being passed down in gfp_flags?
Are you experiencing any other memory issues?

Hi Tom,
It was working fine for a list of up to 1024 so I assumed it was a
kcalloc, I will have to add more debugging to find the problem.
Thanks, Marc.

Tom
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
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-***@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html