ceph: use correct index when encoding client supported features

[PATCH] ceph: use correct index when encoding client supported features

Posted by Luís Henriques 3 years, 11 months ago

Feature bits have to be encoded into the correct locations.  This hasn't
been an issue so far because the only hole in the feature bits was in bit
10 (CEPHFS_FEATURE_RECLAIM_CLIENT), which is located in the 2nd byte.  When
adding more bits that go beyond the this 2nd byte, the bug will show up.

Fixes: 9ba1e224538a ("ceph: allocate the correct amount of extra bytes for the session features")
Signed-off-by: Luís Henriques <lhenriques@suse.de>
---
 fs/ceph/mds_client.c | 7 +++++--
 fs/ceph/mds_client.h | 2 --
 2 files changed, 5 insertions(+), 4 deletions(-)

I hope I got this code right.  I found this issue when trying to add an
extra feature bit that would go beyond bit 15 and that wasn't showing up
on the MDS side.

diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 1bd3e1bb0fdf..77e742b6fd30 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -1220,14 +1220,17 @@ static int encode_supported_features(void **p, void *end)
 	if (count > 0) {
 		size_t i;
 		size_t size = FEATURE_BYTES(count);
+		unsigned long bit;
 
 		if (WARN_ON_ONCE(*p + 4 + size > end))
 			return -ERANGE;
 
 		ceph_encode_32(p, size);
 		memset(*p, 0, size);
-		for (i = 0; i < count; i++)
-			((unsigned char*)(*p))[i / 8] |= BIT(feature_bits[i] % 8);
+		for (i = 0; i < count; i++) {
+			bit = feature_bits[i];
+			((unsigned char *)(*p))[bit / 8] |= BIT(bit % 8);
+		}
 		*p += size;
 	} else {
 		if (WARN_ON_ONCE(*p + 4 > end))
diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
index 33497846e47e..12901e3a6823 100644
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -45,8 +45,6 @@ enum ceph_feature_type {
 	CEPHFS_FEATURE_MULTI_RECONNECT,		\
 	CEPHFS_FEATURE_DELEG_INO,		\
 	CEPHFS_FEATURE_METRIC_COLLECT,		\
-						\
-	CEPHFS_FEATURE_MAX,			\
 }
 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}

Re: [PATCH] ceph: use correct index when encoding client supported features

Posted by Xiubo Li 3 years, 11 months ago

On 5/25/22 12:06 AM, Luís Henriques wrote:
> Feature bits have to be encoded into the correct locations.  This hasn't
> been an issue so far because the only hole in the feature bits was in bit
> 10 (CEPHFS_FEATURE_RECLAIM_CLIENT), which is located in the 2nd byte.  When
> adding more bits that go beyond the this 2nd byte, the bug will show up.
>
> Fixes: 9ba1e224538a ("ceph: allocate the correct amount of extra bytes for the session features")
> Signed-off-by: Luís Henriques <lhenriques@suse.de>
> ---
>   fs/ceph/mds_client.c | 7 +++++--
>   fs/ceph/mds_client.h | 2 --
>   2 files changed, 5 insertions(+), 4 deletions(-)
>
> I hope I got this code right.  I found this issue when trying to add an
> extra feature bit that would go beyond bit 15 and that wasn't showing up
> on the MDS side.
>
> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> index 1bd3e1bb0fdf..77e742b6fd30 100644
> --- a/fs/ceph/mds_client.c
> +++ b/fs/ceph/mds_client.c
> @@ -1220,14 +1220,17 @@ static int encode_supported_features(void **p, void *end)
>   	if (count > 0) {
>   		size_t i;
>   		size_t size = FEATURE_BYTES(count);
> +		unsigned long bit;
>   
>   		if (WARN_ON_ONCE(*p + 4 + size > end))
>   			return -ERANGE;
>   
>   		ceph_encode_32(p, size);
>   		memset(*p, 0, size);
> -		for (i = 0; i < count; i++)
> -			((unsigned char*)(*p))[i / 8] |= BIT(feature_bits[i] % 8);
> +		for (i = 0; i < count; i++) {
> +			bit = feature_bits[i];
> +			((unsigned char *)(*p))[bit / 8] |= BIT(bit % 8);
> +		}
>   		*p += size;
>   	} else {
>   		if (WARN_ON_ONCE(*p + 4 > end))
> diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
> index 33497846e47e..12901e3a6823 100644
> --- a/fs/ceph/mds_client.h
> +++ b/fs/ceph/mds_client.h
> @@ -45,8 +45,6 @@ enum ceph_feature_type {
>   	CEPHFS_FEATURE_MULTI_RECONNECT,		\
>   	CEPHFS_FEATURE_DELEG_INO,		\
>   	CEPHFS_FEATURE_METRIC_COLLECT,		\
> -						\
> -	CEPHFS_FEATURE_MAX,			\
>   }
>   #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
>   

LGTM. Just one small fix in the comment, since you have removed the 
'CEPHFS_FEATURE_MAX', which makes no sense any more:


diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
index 33497846e47e..ac49344ea79e 100644
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -33,10 +33,6 @@ enum ceph_feature_type {
         CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT,
  };

-/*
- * This will always have the highest feature bit value
- * as the last element of the array.
- */
  #define CEPHFS_FEATURES_CLIENT_SUPPORTED {     \
         0, 1, 2, 3, 4, 5, 6, 7,                 \
         CEPHFS_FEATURE_MIMIC,                   \

Merged into the testing branch.

Thanks Luis.

-- Xiubo

Re: [PATCH] ceph: use correct index when encoding client supported features

Posted by Luís Henriques 3 years, 11 months ago

Xiubo Li <xiubli@redhat.com> writes:

> On 5/25/22 12:06 AM, Luís Henriques wrote:
>> Feature bits have to be encoded into the correct locations.  This hasn't
>> been an issue so far because the only hole in the feature bits was in bit
>> 10 (CEPHFS_FEATURE_RECLAIM_CLIENT), which is located in the 2nd byte.  When
>> adding more bits that go beyond the this 2nd byte, the bug will show up.
>>
>> Fixes: 9ba1e224538a ("ceph: allocate the correct amount of extra bytes for the session features")
>> Signed-off-by: Luís Henriques <lhenriques@suse.de>
>> ---
>>   fs/ceph/mds_client.c | 7 +++++--
>>   fs/ceph/mds_client.h | 2 --
>>   2 files changed, 5 insertions(+), 4 deletions(-)
>>
>> I hope I got this code right.  I found this issue when trying to add an
>> extra feature bit that would go beyond bit 15 and that wasn't showing up
>> on the MDS side.
>>
>> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
>> index 1bd3e1bb0fdf..77e742b6fd30 100644
>> --- a/fs/ceph/mds_client.c
>> +++ b/fs/ceph/mds_client.c
>> @@ -1220,14 +1220,17 @@ static int encode_supported_features(void **p, void *end)
>>   	if (count > 0) {
>>   		size_t i;
>>   		size_t size = FEATURE_BYTES(count);
>> +		unsigned long bit;
>>   
>>   		if (WARN_ON_ONCE(*p + 4 + size > end))
>>   			return -ERANGE;
>>   
>>   		ceph_encode_32(p, size);
>>   		memset(*p, 0, size);
>> -		for (i = 0; i < count; i++)
>> -			((unsigned char*)(*p))[i / 8] |= BIT(feature_bits[i] % 8);
>> +		for (i = 0; i < count; i++) {
>> +			bit = feature_bits[i];
>> +			((unsigned char *)(*p))[bit / 8] |= BIT(bit % 8);
>> +		}
>>   		*p += size;
>>   	} else {
>>   		if (WARN_ON_ONCE(*p + 4 > end))
>> diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
>> index 33497846e47e..12901e3a6823 100644
>> --- a/fs/ceph/mds_client.h
>> +++ b/fs/ceph/mds_client.h
>> @@ -45,8 +45,6 @@ enum ceph_feature_type {
>>   	CEPHFS_FEATURE_MULTI_RECONNECT,		\
>>   	CEPHFS_FEATURE_DELEG_INO,		\
>>   	CEPHFS_FEATURE_METRIC_COLLECT,		\
>> -						\
>> -	CEPHFS_FEATURE_MAX,			\
>>   }
>>   #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
>>   
>
> LGTM. Just one small fix in the comment, since you have removed the 
> 'CEPHFS_FEATURE_MAX', which makes no sense any more:

Ah, yeah.  I missed that.  Thanks!

Cheers,
-- 
Luís


> diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
> index 33497846e47e..ac49344ea79e 100644
> --- a/fs/ceph/mds_client.h
> +++ b/fs/ceph/mds_client.h
> @@ -33,10 +33,6 @@ enum ceph_feature_type {
>          CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT,
>   };
>
> -/*
> - * This will always have the highest feature bit value
> - * as the last element of the array.
> - */
>   #define CEPHFS_FEATURES_CLIENT_SUPPORTED {     \
>          0, 1, 2, 3, 4, 5, 6, 7,                 \
>          CEPHFS_FEATURE_MIMIC,                   \
>
> Merged into the testing branch.
>
> Thanks Luis.
>
> -- Xiubo
>

Re: [PATCH] ceph: use correct index when encoding client supported features

Posted by Jeff Layton 3 years, 11 months ago

On Tue, 2022-05-24 at 17:06 +0100, Luís Henriques wrote:
> Feature bits have to be encoded into the correct locations.  This hasn't
> been an issue so far because the only hole in the feature bits was in bit
> 10 (CEPHFS_FEATURE_RECLAIM_CLIENT), which is located in the 2nd byte.  When
> adding more bits that go beyond the this 2nd byte, the bug will show up.
> 
> Fixes: 9ba1e224538a ("ceph: allocate the correct amount of extra bytes for the session features")
> Signed-off-by: Luís Henriques <lhenriques@suse.de>
> ---
>  fs/ceph/mds_client.c | 7 +++++--
>  fs/ceph/mds_client.h | 2 --
>  2 files changed, 5 insertions(+), 4 deletions(-)
> 
> I hope I got this code right.  I found this issue when trying to add an
> extra feature bit that would go beyond bit 15 and that wasn't showing up
> on the MDS side.
> 
> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> index 1bd3e1bb0fdf..77e742b6fd30 100644
> --- a/fs/ceph/mds_client.c
> +++ b/fs/ceph/mds_client.c
> @@ -1220,14 +1220,17 @@ static int encode_supported_features(void **p, void *end)
>  	if (count > 0) {
>  		size_t i;
>  		size_t size = FEATURE_BYTES(count);
> +		unsigned long bit;
>  
>  		if (WARN_ON_ONCE(*p + 4 + size > end))
>  			return -ERANGE;
>  
>  		ceph_encode_32(p, size);
>  		memset(*p, 0, size);
> -		for (i = 0; i < count; i++)
> -			((unsigned char*)(*p))[i / 8] |= BIT(feature_bits[i] % 8);
> +		for (i = 0; i < count; i++) {
> +			bit = feature_bits[i];
> +			((unsigned char *)(*p))[bit / 8] |= BIT(bit % 8);
> +		}

I wish we could just use __set_bit/__clear_bit here. It would be a lot
easier to reason this out. Your logic looks correct to me though.


>  		*p += size;
>  	} else {
>  		if (WARN_ON_ONCE(*p + 4 > end))
> diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
> index 33497846e47e..12901e3a6823 100644
> --- a/fs/ceph/mds_client.h
> +++ b/fs/ceph/mds_client.h
> @@ -45,8 +45,6 @@ enum ceph_feature_type {
>  	CEPHFS_FEATURE_MULTI_RECONNECT,		\
>  	CEPHFS_FEATURE_DELEG_INO,		\
>  	CEPHFS_FEATURE_METRIC_COLLECT,		\
> -						\
> -	CEPHFS_FEATURE_MAX,			\
>  }
>  #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
>  

Reviewed-by: Jeff Layton <jlayton@kernel.org>