[v8,03/12] ceph: handle idmapped mounts in create_request_message()

Message ID 20230803135955.230449-4-aleksandr.mikhalitsyn@canonical.com
State New
Headers
Series ceph: support idmapped mounts |

Commit Message

Aleksandr Mikhalitsyn Aug. 3, 2023, 1:59 p.m. UTC
  From: Christian Brauner <brauner@kernel.org>

Inode operations that create a new filesystem object such as ->mknod,
->create, ->mkdir() and others don't take a {g,u}id argument explicitly.
Instead the caller's fs{g,u}id is used for the {g,u}id of the new
filesystem object.

In order to ensure that the correct {g,u}id is used map the caller's
fs{g,u}id for creation requests. This doesn't require complex changes.
It suffices to pass in the relevant idmapping recorded in the request
message. If this request message was triggered from an inode operation
that creates filesystem objects it will have passed down the relevant
idmaping. If this is a request message that was triggered from an inode
operation that doens't need to take idmappings into account the initial
idmapping is passed down which is an identity mapping.

This change uses a new cephfs protocol extension CEPHFS_FEATURE_HAS_OWNER_UIDGID
which adds two new fields (owner_{u,g}id) to the request head structure.
So, we need to ensure that MDS supports it otherwise we need to fail
any IO that comes through an idmapped mount because we can't process it
in a proper way. MDS server without such an extension will use caller_{u,g}id
fields to set a new inode owner UID/GID which is incorrect because caller_{u,g}id
values are unmapped. At the same time we can't map these fields with an
idmapping as it can break UID/GID-based permission checks logic on the
MDS side. This problem was described with a lot of details at [1], [2].

[1] https://lore.kernel.org/lkml/CAEivzxfw1fHO2TFA4dx3u23ZKK6Q+EThfzuibrhA3RKM=ZOYLg@mail.gmail.com/
[2] https://lore.kernel.org/all/20220104140414.155198-3-brauner@kernel.org/

https://github.com/ceph/ceph/pull/52575
https://tracker.ceph.com/issues/62217

Cc: Xiubo Li <xiubli@redhat.com>
Cc: Jeff Layton <jlayton@kernel.org>
Cc: Ilya Dryomov <idryomov@gmail.com>
Cc: ceph-devel@vger.kernel.org
Co-Developed-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
---
v7:
	- reworked to use two new fields for owner UID/GID (https://github.com/ceph/ceph/pull/52575)
v8:
	- properly handled case when old MDS used with new kernel client
---
 fs/ceph/mds_client.c         | 46 +++++++++++++++++++++++++++++++++---
 fs/ceph/mds_client.h         |  5 +++-
 include/linux/ceph/ceph_fs.h |  4 +++-
 3 files changed, 50 insertions(+), 5 deletions(-)
  

Comments

Xiubo Li Aug. 4, 2023, 2:26 a.m. UTC | #1
On 8/3/23 21:59, Alexander Mikhalitsyn wrote:
> From: Christian Brauner <brauner@kernel.org>
>
> Inode operations that create a new filesystem object such as ->mknod,
> ->create, ->mkdir() and others don't take a {g,u}id argument explicitly.
> Instead the caller's fs{g,u}id is used for the {g,u}id of the new
> filesystem object.
>
> In order to ensure that the correct {g,u}id is used map the caller's
> fs{g,u}id for creation requests. This doesn't require complex changes.
> It suffices to pass in the relevant idmapping recorded in the request
> message. If this request message was triggered from an inode operation
> that creates filesystem objects it will have passed down the relevant
> idmaping. If this is a request message that was triggered from an inode
> operation that doens't need to take idmappings into account the initial
> idmapping is passed down which is an identity mapping.
>
> This change uses a new cephfs protocol extension CEPHFS_FEATURE_HAS_OWNER_UIDGID
> which adds two new fields (owner_{u,g}id) to the request head structure.
> So, we need to ensure that MDS supports it otherwise we need to fail
> any IO that comes through an idmapped mount because we can't process it
> in a proper way. MDS server without such an extension will use caller_{u,g}id
> fields to set a new inode owner UID/GID which is incorrect because caller_{u,g}id
> values are unmapped. At the same time we can't map these fields with an
> idmapping as it can break UID/GID-based permission checks logic on the
> MDS side. This problem was described with a lot of details at [1], [2].
>
> [1] https://lore.kernel.org/lkml/CAEivzxfw1fHO2TFA4dx3u23ZKK6Q+EThfzuibrhA3RKM=ZOYLg@mail.gmail.com/
> [2] https://lore.kernel.org/all/20220104140414.155198-3-brauner@kernel.org/
>
> https://github.com/ceph/ceph/pull/52575
> https://tracker.ceph.com/issues/62217
>
> Cc: Xiubo Li <xiubli@redhat.com>
> Cc: Jeff Layton <jlayton@kernel.org>
> Cc: Ilya Dryomov <idryomov@gmail.com>
> Cc: ceph-devel@vger.kernel.org
> Co-Developed-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
> Signed-off-by: Christian Brauner <brauner@kernel.org>
> Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
> ---
> v7:
> 	- reworked to use two new fields for owner UID/GID (https://github.com/ceph/ceph/pull/52575)
> v8:
> 	- properly handled case when old MDS used with new kernel client
> ---
>   fs/ceph/mds_client.c         | 46 +++++++++++++++++++++++++++++++++---
>   fs/ceph/mds_client.h         |  5 +++-
>   include/linux/ceph/ceph_fs.h |  4 +++-
>   3 files changed, 50 insertions(+), 5 deletions(-)
>
> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> index 8829f55103da..7d3106d3b726 100644
> --- a/fs/ceph/mds_client.c
> +++ b/fs/ceph/mds_client.c
> @@ -2902,6 +2902,17 @@ static void encode_mclientrequest_tail(void **p, const struct ceph_mds_request *
>   	}
>   }
>   
> +static inline u16 mds_supported_head_version(struct ceph_mds_session *session)
> +{
> +	if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, &session->s_features))
> +		return 1;
> +
> +	if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features))
> +		return 2;
> +
> +	return CEPH_MDS_REQUEST_HEAD_VERSION;
> +}
> +
>   static struct ceph_mds_request_head_legacy *
>   find_legacy_request_head(void *p, u64 features)
>   {
> @@ -2923,6 +2934,7 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
>   {
>   	int mds = session->s_mds;
>   	struct ceph_mds_client *mdsc = session->s_mdsc;
> +	struct ceph_client *cl = mdsc->fsc->client;
>   	struct ceph_msg *msg;
>   	struct ceph_mds_request_head_legacy *lhead;
>   	const char *path1 = NULL;
> @@ -2936,7 +2948,7 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
>   	void *p, *end;
>   	int ret;
>   	bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
> -	bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, &session->s_features);
> +	u16 request_head_version = mds_supported_head_version(session);
>   
>   	ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
>   			      req->r_parent, req->r_path1, req->r_ino1.ino,
> @@ -2977,8 +2989,10 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
>   	 */
>   	if (legacy)
>   		len = sizeof(struct ceph_mds_request_head_legacy);
> -	else if (old_version)
> +	else if (request_head_version == 1)
>   		len = sizeof(struct ceph_mds_request_head_old);
> +	else if (request_head_version == 2)
> +		len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
>   	else
>   		len = sizeof(struct ceph_mds_request_head);
>   

This is not what we suppose to. If we do this again and again when 
adding new members it will make the code very complicated to maintain.

Once the CEPHFS_FEATURE_32BITS_RETRY_FWD has been supported the ceph 
should correctly decode it and if CEPHFS_FEATURE_HAS_OWNER_UIDGID is not 
supported the decoder should skip it directly.

Is the MDS side buggy ? Why you last version didn't work ?

Thanks

- Xiubo

> @@ -3028,6 +3042,16 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
>   	lhead = find_legacy_request_head(msg->front.iov_base,
>   					 session->s_con.peer_features);
>   
> +	if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
> +	    !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features)) {
> +		pr_err_ratelimited_client(cl,
> +			"idmapped mount is used and CEPHFS_FEATURE_HAS_OWNER_UIDGID"
> +			" is not supported by MDS. Fail request with -EIO.\n");
> +
> +		ret = -EIO;
> +		goto out_err;
> +	}
> +
>   	/*
>   	 * The ceph_mds_request_head_legacy didn't contain a version field, and
>   	 * one was added when we moved the message version from 3->4.
> @@ -3035,17 +3059,33 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
>   	if (legacy) {
>   		msg->hdr.version = cpu_to_le16(3);
>   		p = msg->front.iov_base + sizeof(*lhead);
> -	} else if (old_version) {
> +	} else if (request_head_version == 1) {
>   		struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
>   
>   		msg->hdr.version = cpu_to_le16(4);
>   		ohead->version = cpu_to_le16(1);
>   		p = msg->front.iov_base + sizeof(*ohead);
> +	} else if (request_head_version == 2) {
> +		struct ceph_mds_request_head *nhead = msg->front.iov_base;
> +
> +		msg->hdr.version = cpu_to_le16(6);
> +		nhead->version = cpu_to_le16(2);
> +
> +		p = msg->front.iov_base + offsetofend(struct ceph_mds_request_head, ext_num_fwd);
>   	} else {
>   		struct ceph_mds_request_head *nhead = msg->front.iov_base;
> +		kuid_t owner_fsuid;
> +		kgid_t owner_fsgid;
>   
>   		msg->hdr.version = cpu_to_le16(6);
>   		nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
> +
> +		owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
> +					  VFSUIDT_INIT(req->r_cred->fsuid));
> +		owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
> +					  VFSGIDT_INIT(req->r_cred->fsgid));
> +		nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns, owner_fsuid));
> +		nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns, owner_fsgid));
>   		p = msg->front.iov_base + sizeof(*nhead);
>   	}
>   
> diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
> index e3bbf3ba8ee8..8f683e8203bd 100644
> --- a/fs/ceph/mds_client.h
> +++ b/fs/ceph/mds_client.h
> @@ -33,8 +33,10 @@ enum ceph_feature_type {
>   	CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
>   	CEPHFS_FEATURE_OP_GETVXATTR,
>   	CEPHFS_FEATURE_32BITS_RETRY_FWD,
> +	CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
> +	CEPHFS_FEATURE_HAS_OWNER_UIDGID,
>   
> -	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
> +	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_HAS_OWNER_UIDGID,
>   };
>   
>   #define CEPHFS_FEATURES_CLIENT_SUPPORTED {	\
> @@ -49,6 +51,7 @@ enum ceph_feature_type {
>   	CEPHFS_FEATURE_NOTIFY_SESSION_STATE,	\
>   	CEPHFS_FEATURE_OP_GETVXATTR,		\
>   	CEPHFS_FEATURE_32BITS_RETRY_FWD,	\
> +	CEPHFS_FEATURE_HAS_OWNER_UIDGID,	\
>   }
>   
>   /*
> diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
> index 5f2301ee88bc..6eb83a51341c 100644
> --- a/include/linux/ceph/ceph_fs.h
> +++ b/include/linux/ceph/ceph_fs.h
> @@ -499,7 +499,7 @@ struct ceph_mds_request_head_legacy {
>   	union ceph_mds_request_args args;
>   } __attribute__ ((packed));
>   
> -#define CEPH_MDS_REQUEST_HEAD_VERSION  2
> +#define CEPH_MDS_REQUEST_HEAD_VERSION  3
>   
>   struct ceph_mds_request_head_old {
>   	__le16 version;                /* struct version */
> @@ -530,6 +530,8 @@ struct ceph_mds_request_head {
>   
>   	__le32 ext_num_retry;          /* new count retry attempts */
>   	__le32 ext_num_fwd;            /* new count fwd attempts */
> +
> +	__le32 owner_uid, owner_gid;   /* used for OPs which create inodes */
>   } __attribute__ ((packed));
>   
>   /* cap/lease release record */
  
Xiubo Li Aug. 4, 2023, 3:24 a.m. UTC | #2
On 8/4/23 10:26, Xiubo Li wrote:
>
> On 8/3/23 21:59, Alexander Mikhalitsyn wrote:
>> From: Christian Brauner <brauner@kernel.org>
>>
>> Inode operations that create a new filesystem object such as ->mknod,
>> ->create, ->mkdir() and others don't take a {g,u}id argument explicitly.
>> Instead the caller's fs{g,u}id is used for the {g,u}id of the new
>> filesystem object.
>>
>> In order to ensure that the correct {g,u}id is used map the caller's
>> fs{g,u}id for creation requests. This doesn't require complex changes.
>> It suffices to pass in the relevant idmapping recorded in the request
>> message. If this request message was triggered from an inode operation
>> that creates filesystem objects it will have passed down the relevant
>> idmaping. If this is a request message that was triggered from an inode
>> operation that doens't need to take idmappings into account the initial
>> idmapping is passed down which is an identity mapping.
>>
>> This change uses a new cephfs protocol extension 
>> CEPHFS_FEATURE_HAS_OWNER_UIDGID
>> which adds two new fields (owner_{u,g}id) to the request head structure.
>> So, we need to ensure that MDS supports it otherwise we need to fail
>> any IO that comes through an idmapped mount because we can't process it
>> in a proper way. MDS server without such an extension will use 
>> caller_{u,g}id
>> fields to set a new inode owner UID/GID which is incorrect because 
>> caller_{u,g}id
>> values are unmapped. At the same time we can't map these fields with an
>> idmapping as it can break UID/GID-based permission checks logic on the
>> MDS side. This problem was described with a lot of details at [1], [2].
>>
>> [1] 
>> https://lore.kernel.org/lkml/CAEivzxfw1fHO2TFA4dx3u23ZKK6Q+EThfzuibrhA3RKM=ZOYLg@mail.gmail.com/
>> [2] 
>> https://lore.kernel.org/all/20220104140414.155198-3-brauner@kernel.org/
>>
>> https://github.com/ceph/ceph/pull/52575
>> https://tracker.ceph.com/issues/62217
>>
>> Cc: Xiubo Li <xiubli@redhat.com>
>> Cc: Jeff Layton <jlayton@kernel.org>
>> Cc: Ilya Dryomov <idryomov@gmail.com>
>> Cc: ceph-devel@vger.kernel.org
>> Co-Developed-by: Alexander Mikhalitsyn 
>> <aleksandr.mikhalitsyn@canonical.com>
>> Signed-off-by: Christian Brauner <brauner@kernel.org>
>> Signed-off-by: Alexander Mikhalitsyn 
>> <aleksandr.mikhalitsyn@canonical.com>
>> ---
>> v7:
>>     - reworked to use two new fields for owner UID/GID 
>> (https://github.com/ceph/ceph/pull/52575)
>> v8:
>>     - properly handled case when old MDS used with new kernel client
>> ---
>>   fs/ceph/mds_client.c         | 46 +++++++++++++++++++++++++++++++++---
>>   fs/ceph/mds_client.h         |  5 +++-
>>   include/linux/ceph/ceph_fs.h |  4 +++-
>>   3 files changed, 50 insertions(+), 5 deletions(-)
>>
>> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
>> index 8829f55103da..7d3106d3b726 100644
>> --- a/fs/ceph/mds_client.c
>> +++ b/fs/ceph/mds_client.c
>> @@ -2902,6 +2902,17 @@ static void encode_mclientrequest_tail(void 
>> **p, const struct ceph_mds_request *
>>       }
>>   }
>>   +static inline u16 mds_supported_head_version(struct 
>> ceph_mds_session *session)
>> +{
>> +    if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, 
>> &session->s_features))
>> +        return 1;
>> +
>> +    if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, 
>> &session->s_features))
>> +        return 2;
>> +
>> +    return CEPH_MDS_REQUEST_HEAD_VERSION;
>> +}
>> +
>>   static struct ceph_mds_request_head_legacy *
>>   find_legacy_request_head(void *p, u64 features)
>>   {
>> @@ -2923,6 +2934,7 @@ static struct ceph_msg 
>> *create_request_message(struct ceph_mds_session *session,
>>   {
>>       int mds = session->s_mds;
>>       struct ceph_mds_client *mdsc = session->s_mdsc;
>> +    struct ceph_client *cl = mdsc->fsc->client;
>>       struct ceph_msg *msg;
>>       struct ceph_mds_request_head_legacy *lhead;
>>       const char *path1 = NULL;
>> @@ -2936,7 +2948,7 @@ static struct ceph_msg 
>> *create_request_message(struct ceph_mds_session *session,
>>       void *p, *end;
>>       int ret;
>>       bool legacy = !(session->s_con.peer_features & 
>> CEPH_FEATURE_FS_BTIME);
>> -    bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, 
>> &session->s_features);
>> +    u16 request_head_version = mds_supported_head_version(session);
>>         ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
>>                     req->r_parent, req->r_path1, req->r_ino1.ino,
>> @@ -2977,8 +2989,10 @@ static struct ceph_msg 
>> *create_request_message(struct ceph_mds_session *session,
>>        */
>>       if (legacy)
>>           len = sizeof(struct ceph_mds_request_head_legacy);
>> -    else if (old_version)
>> +    else if (request_head_version == 1)
>>           len = sizeof(struct ceph_mds_request_head_old);
>> +    else if (request_head_version == 2)
>> +        len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
>>       else
>>           len = sizeof(struct ceph_mds_request_head);
>
> This is not what we suppose to. If we do this again and again when 
> adding new members it will make the code very complicated to maintain.
>
> Once the CEPHFS_FEATURE_32BITS_RETRY_FWD has been supported the ceph 
> should correctly decode it and if CEPHFS_FEATURE_HAS_OWNER_UIDGID is 
> not supported the decoder should skip it directly.
>
> Is the MDS side buggy ? Why you last version didn't work ?
>

I think the ceph side is buggy. Possibly we should add one new `length` 
member in struct `struct ceph_mds_request_head` and just skip the extra 
bytes when decoding it.

Could you fix it together with your ceph PR ?

Thanks

- Xiubo


> Thanks
>
> - Xiubo
>
>> @@ -3028,6 +3042,16 @@ static struct ceph_msg 
>> *create_request_message(struct ceph_mds_session *session,
>>       lhead = find_legacy_request_head(msg->front.iov_base,
>>                        session->s_con.peer_features);
>>   +    if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
>> +        !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, 
>> &session->s_features)) {
>> +        pr_err_ratelimited_client(cl,
>> +            "idmapped mount is used and 
>> CEPHFS_FEATURE_HAS_OWNER_UIDGID"
>> +            " is not supported by MDS. Fail request with -EIO.\n");
>> +
>> +        ret = -EIO;
>> +        goto out_err;
>> +    }
>> +
>>       /*
>>        * The ceph_mds_request_head_legacy didn't contain a version 
>> field, and
>>        * one was added when we moved the message version from 3->4.
>> @@ -3035,17 +3059,33 @@ static struct ceph_msg 
>> *create_request_message(struct ceph_mds_session *session,
>>       if (legacy) {
>>           msg->hdr.version = cpu_to_le16(3);
>>           p = msg->front.iov_base + sizeof(*lhead);
>> -    } else if (old_version) {
>> +    } else if (request_head_version == 1) {
>>           struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
>>             msg->hdr.version = cpu_to_le16(4);
>>           ohead->version = cpu_to_le16(1);
>>           p = msg->front.iov_base + sizeof(*ohead);
>> +    } else if (request_head_version == 2) {
>> +        struct ceph_mds_request_head *nhead = msg->front.iov_base;
>> +
>> +        msg->hdr.version = cpu_to_le16(6);
>> +        nhead->version = cpu_to_le16(2);
>> +
>> +        p = msg->front.iov_base + offsetofend(struct 
>> ceph_mds_request_head, ext_num_fwd);
>>       } else {
>>           struct ceph_mds_request_head *nhead = msg->front.iov_base;
>> +        kuid_t owner_fsuid;
>> +        kgid_t owner_fsgid;
>>             msg->hdr.version = cpu_to_le16(6);
>>           nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
>> +
>> +        owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
>> +                      VFSUIDT_INIT(req->r_cred->fsuid));
>> +        owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
>> +                      VFSGIDT_INIT(req->r_cred->fsgid));
>> +        nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns, 
>> owner_fsuid));
>> +        nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns, 
>> owner_fsgid));
>>           p = msg->front.iov_base + sizeof(*nhead);
>>       }
>>   diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
>> index e3bbf3ba8ee8..8f683e8203bd 100644
>> --- a/fs/ceph/mds_client.h
>> +++ b/fs/ceph/mds_client.h
>> @@ -33,8 +33,10 @@ enum ceph_feature_type {
>>       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
>>       CEPHFS_FEATURE_OP_GETVXATTR,
>>       CEPHFS_FEATURE_32BITS_RETRY_FWD,
>> +    CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
>> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,
>>   -    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
>> +    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_HAS_OWNER_UIDGID,
>>   };
>>     #define CEPHFS_FEATURES_CLIENT_SUPPORTED {    \
>> @@ -49,6 +51,7 @@ enum ceph_feature_type {
>>       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,    \
>>       CEPHFS_FEATURE_OP_GETVXATTR,        \
>>       CEPHFS_FEATURE_32BITS_RETRY_FWD,    \
>> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,    \
>>   }
>>     /*
>> diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
>> index 5f2301ee88bc..6eb83a51341c 100644
>> --- a/include/linux/ceph/ceph_fs.h
>> +++ b/include/linux/ceph/ceph_fs.h
>> @@ -499,7 +499,7 @@ struct ceph_mds_request_head_legacy {
>>       union ceph_mds_request_args args;
>>   } __attribute__ ((packed));
>>   -#define CEPH_MDS_REQUEST_HEAD_VERSION  2
>> +#define CEPH_MDS_REQUEST_HEAD_VERSION  3
>>     struct ceph_mds_request_head_old {
>>       __le16 version;                /* struct version */
>> @@ -530,6 +530,8 @@ struct ceph_mds_request_head {
>>         __le32 ext_num_retry;          /* new count retry attempts */
>>       __le32 ext_num_fwd;            /* new count fwd attempts */
>> +
>> +    __le32 owner_uid, owner_gid;   /* used for OPs which create 
>> inodes */
>>   } __attribute__ ((packed));
>>     /* cap/lease release record */
  
Aleksandr Mikhalitsyn Aug. 4, 2023, 6:25 a.m. UTC | #3
On Fri, Aug 4, 2023 at 4:26 AM Xiubo Li <xiubli@redhat.com> wrote:
>
>
> On 8/3/23 21:59, Alexander Mikhalitsyn wrote:
> > From: Christian Brauner <brauner@kernel.org>
> >
> > Inode operations that create a new filesystem object such as ->mknod,
> > ->create, ->mkdir() and others don't take a {g,u}id argument explicitly.
> > Instead the caller's fs{g,u}id is used for the {g,u}id of the new
> > filesystem object.
> >
> > In order to ensure that the correct {g,u}id is used map the caller's
> > fs{g,u}id for creation requests. This doesn't require complex changes.
> > It suffices to pass in the relevant idmapping recorded in the request
> > message. If this request message was triggered from an inode operation
> > that creates filesystem objects it will have passed down the relevant
> > idmaping. If this is a request message that was triggered from an inode
> > operation that doens't need to take idmappings into account the initial
> > idmapping is passed down which is an identity mapping.
> >
> > This change uses a new cephfs protocol extension CEPHFS_FEATURE_HAS_OWNER_UIDGID
> > which adds two new fields (owner_{u,g}id) to the request head structure.
> > So, we need to ensure that MDS supports it otherwise we need to fail
> > any IO that comes through an idmapped mount because we can't process it
> > in a proper way. MDS server without such an extension will use caller_{u,g}id
> > fields to set a new inode owner UID/GID which is incorrect because caller_{u,g}id
> > values are unmapped. At the same time we can't map these fields with an
> > idmapping as it can break UID/GID-based permission checks logic on the
> > MDS side. This problem was described with a lot of details at [1], [2].
> >
> > [1] https://lore.kernel.org/lkml/CAEivzxfw1fHO2TFA4dx3u23ZKK6Q+EThfzuibrhA3RKM=ZOYLg@mail.gmail.com/
> > [2] https://lore.kernel.org/all/20220104140414.155198-3-brauner@kernel.org/
> >
> > https://github.com/ceph/ceph/pull/52575
> > https://tracker.ceph.com/issues/62217
> >
> > Cc: Xiubo Li <xiubli@redhat.com>
> > Cc: Jeff Layton <jlayton@kernel.org>
> > Cc: Ilya Dryomov <idryomov@gmail.com>
> > Cc: ceph-devel@vger.kernel.org
> > Co-Developed-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
> > Signed-off-by: Christian Brauner <brauner@kernel.org>
> > Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
> > ---
> > v7:
> >       - reworked to use two new fields for owner UID/GID (https://github.com/ceph/ceph/pull/52575)
> > v8:
> >       - properly handled case when old MDS used with new kernel client
> > ---
> >   fs/ceph/mds_client.c         | 46 +++++++++++++++++++++++++++++++++---
> >   fs/ceph/mds_client.h         |  5 +++-
> >   include/linux/ceph/ceph_fs.h |  4 +++-
> >   3 files changed, 50 insertions(+), 5 deletions(-)
> >
> > diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> > index 8829f55103da..7d3106d3b726 100644
> > --- a/fs/ceph/mds_client.c
> > +++ b/fs/ceph/mds_client.c
> > @@ -2902,6 +2902,17 @@ static void encode_mclientrequest_tail(void **p, const struct ceph_mds_request *
> >       }
> >   }
> >
> > +static inline u16 mds_supported_head_version(struct ceph_mds_session *session)
> > +{
> > +     if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, &session->s_features))
> > +             return 1;
> > +
> > +     if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features))
> > +             return 2;
> > +
> > +     return CEPH_MDS_REQUEST_HEAD_VERSION;
> > +}
> > +
> >   static struct ceph_mds_request_head_legacy *
> >   find_legacy_request_head(void *p, u64 features)
> >   {
> > @@ -2923,6 +2934,7 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
> >   {
> >       int mds = session->s_mds;
> >       struct ceph_mds_client *mdsc = session->s_mdsc;
> > +     struct ceph_client *cl = mdsc->fsc->client;
> >       struct ceph_msg *msg;
> >       struct ceph_mds_request_head_legacy *lhead;
> >       const char *path1 = NULL;
> > @@ -2936,7 +2948,7 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
> >       void *p, *end;
> >       int ret;
> >       bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
> > -     bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, &session->s_features);
> > +     u16 request_head_version = mds_supported_head_version(session);
> >
> >       ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
> >                             req->r_parent, req->r_path1, req->r_ino1.ino,
> > @@ -2977,8 +2989,10 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
> >        */
> >       if (legacy)
> >               len = sizeof(struct ceph_mds_request_head_legacy);
> > -     else if (old_version)
> > +     else if (request_head_version == 1)
> >               len = sizeof(struct ceph_mds_request_head_old);
> > +     else if (request_head_version == 2)
> > +             len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
> >       else
> >               len = sizeof(struct ceph_mds_request_head);
> >
>
> This is not what we suppose to. If we do this again and again when
> adding new members it will make the code very complicated to maintain.
>
> Once the CEPHFS_FEATURE_32BITS_RETRY_FWD has been supported the ceph
> should correctly decode it and if CEPHFS_FEATURE_HAS_OWNER_UIDGID is not
> supported the decoder should skip it directly.

I thought that too. But it doesn't work. Just try - take kernel client
testing branch, and then
add a new field to the struct ceph_mds_request_head. Compile and try to mount.
It will stop to work and on the MDS side you will see something like:

2023-08-03T13:15:40.871+0200 7fe64ef5e640 10 mds.c ms_handle_accept
v1:192.168.2.136:0/49354629 con 0x563962206880 session 0x563967054000
2023-08-03T13:15:40.871+0200 7fe650f62640 -1 failed to decode message
of type 24 v6: End of buffer [buffer:2]
2023-08-03T13:15:40.871+0200 7fe650f62640  1 dump:
00000000  03 00 01 00 00 00 00 00  00 00 10 00 00 00 00 00  |................|
00000010  00 00 00 00 00 00 01 01  00 00 00 00 00 00 00 00  |................|
00000020  00 00 00 00 00 00 00 00  00 00 01 00 00 00 00 00  |................|
00000030  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
00000070  00 00 01 01 00 00 00 00  00 00 00 00 00 00 00 01  |................|
00000080  00 00 00 00 00 00 00 00  00 00 00 00 5b 8c cb 64  |............[..d|
00000090  64 78 11 13 01 00 00 00  00 00 00 00 00 00 00 00  |dx..............|
000000a0  00 00 00 00 00 00 00 00  00 00 00 00              |............|
000000ac

As I understand, the MDS side is not ready to see struct
ceph_mds_request_head bigger in size
than supported.

>
> Is the MDS side buggy ? Why you last version didn't work ?
>
> Thanks
>
> - Xiubo
>
> > @@ -3028,6 +3042,16 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
> >       lhead = find_legacy_request_head(msg->front.iov_base,
> >                                        session->s_con.peer_features);
> >
> > +     if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
> > +         !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features)) {
> > +             pr_err_ratelimited_client(cl,
> > +                     "idmapped mount is used and CEPHFS_FEATURE_HAS_OWNER_UIDGID"
> > +                     " is not supported by MDS. Fail request with -EIO.\n");
> > +
> > +             ret = -EIO;
> > +             goto out_err;
> > +     }
> > +
> >       /*
> >        * The ceph_mds_request_head_legacy didn't contain a version field, and
> >        * one was added when we moved the message version from 3->4.
> > @@ -3035,17 +3059,33 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
> >       if (legacy) {
> >               msg->hdr.version = cpu_to_le16(3);
> >               p = msg->front.iov_base + sizeof(*lhead);
> > -     } else if (old_version) {
> > +     } else if (request_head_version == 1) {
> >               struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
> >
> >               msg->hdr.version = cpu_to_le16(4);
> >               ohead->version = cpu_to_le16(1);
> >               p = msg->front.iov_base + sizeof(*ohead);
> > +     } else if (request_head_version == 2) {
> > +             struct ceph_mds_request_head *nhead = msg->front.iov_base;
> > +
> > +             msg->hdr.version = cpu_to_le16(6);
> > +             nhead->version = cpu_to_le16(2);
> > +
> > +             p = msg->front.iov_base + offsetofend(struct ceph_mds_request_head, ext_num_fwd);
> >       } else {
> >               struct ceph_mds_request_head *nhead = msg->front.iov_base;
> > +             kuid_t owner_fsuid;
> > +             kgid_t owner_fsgid;
> >
> >               msg->hdr.version = cpu_to_le16(6);
> >               nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
> > +
> > +             owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
> > +                                       VFSUIDT_INIT(req->r_cred->fsuid));
> > +             owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
> > +                                       VFSGIDT_INIT(req->r_cred->fsgid));
> > +             nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns, owner_fsuid));
> > +             nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns, owner_fsgid));
> >               p = msg->front.iov_base + sizeof(*nhead);
> >       }
> >
> > diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
> > index e3bbf3ba8ee8..8f683e8203bd 100644
> > --- a/fs/ceph/mds_client.h
> > +++ b/fs/ceph/mds_client.h
> > @@ -33,8 +33,10 @@ enum ceph_feature_type {
> >       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
> >       CEPHFS_FEATURE_OP_GETVXATTR,
> >       CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > +     CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
> > +     CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> >
> > -     CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > +     CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> >   };
> >
> >   #define CEPHFS_FEATURES_CLIENT_SUPPORTED {  \
> > @@ -49,6 +51,7 @@ enum ceph_feature_type {
> >       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,    \
> >       CEPHFS_FEATURE_OP_GETVXATTR,            \
> >       CEPHFS_FEATURE_32BITS_RETRY_FWD,        \
> > +     CEPHFS_FEATURE_HAS_OWNER_UIDGID,        \
> >   }
> >
> >   /*
> > diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
> > index 5f2301ee88bc..6eb83a51341c 100644
> > --- a/include/linux/ceph/ceph_fs.h
> > +++ b/include/linux/ceph/ceph_fs.h
> > @@ -499,7 +499,7 @@ struct ceph_mds_request_head_legacy {
> >       union ceph_mds_request_args args;
> >   } __attribute__ ((packed));
> >
> > -#define CEPH_MDS_REQUEST_HEAD_VERSION  2
> > +#define CEPH_MDS_REQUEST_HEAD_VERSION  3
> >
> >   struct ceph_mds_request_head_old {
> >       __le16 version;                /* struct version */
> > @@ -530,6 +530,8 @@ struct ceph_mds_request_head {
> >
> >       __le32 ext_num_retry;          /* new count retry attempts */
> >       __le32 ext_num_fwd;            /* new count fwd attempts */
> > +
> > +     __le32 owner_uid, owner_gid;   /* used for OPs which create inodes */
> >   } __attribute__ ((packed));
> >
> >   /* cap/lease release record */
>
  
Aleksandr Mikhalitsyn Aug. 4, 2023, 6:35 a.m. UTC | #4
On Fri, Aug 4, 2023 at 5:24 AM Xiubo Li <xiubli@redhat.com> wrote:
>
>
> On 8/4/23 10:26, Xiubo Li wrote:
> >
> > On 8/3/23 21:59, Alexander Mikhalitsyn wrote:
> >> From: Christian Brauner <brauner@kernel.org>
> >>
> >> Inode operations that create a new filesystem object such as ->mknod,
> >> ->create, ->mkdir() and others don't take a {g,u}id argument explicitly.
> >> Instead the caller's fs{g,u}id is used for the {g,u}id of the new
> >> filesystem object.
> >>
> >> In order to ensure that the correct {g,u}id is used map the caller's
> >> fs{g,u}id for creation requests. This doesn't require complex changes.
> >> It suffices to pass in the relevant idmapping recorded in the request
> >> message. If this request message was triggered from an inode operation
> >> that creates filesystem objects it will have passed down the relevant
> >> idmaping. If this is a request message that was triggered from an inode
> >> operation that doens't need to take idmappings into account the initial
> >> idmapping is passed down which is an identity mapping.
> >>
> >> This change uses a new cephfs protocol extension
> >> CEPHFS_FEATURE_HAS_OWNER_UIDGID
> >> which adds two new fields (owner_{u,g}id) to the request head structure.
> >> So, we need to ensure that MDS supports it otherwise we need to fail
> >> any IO that comes through an idmapped mount because we can't process it
> >> in a proper way. MDS server without such an extension will use
> >> caller_{u,g}id
> >> fields to set a new inode owner UID/GID which is incorrect because
> >> caller_{u,g}id
> >> values are unmapped. At the same time we can't map these fields with an
> >> idmapping as it can break UID/GID-based permission checks logic on the
> >> MDS side. This problem was described with a lot of details at [1], [2].
> >>
> >> [1]
> >> https://lore.kernel.org/lkml/CAEivzxfw1fHO2TFA4dx3u23ZKK6Q+EThfzuibrhA3RKM=ZOYLg@mail.gmail.com/
> >> [2]
> >> https://lore.kernel.org/all/20220104140414.155198-3-brauner@kernel.org/
> >>
> >> https://github.com/ceph/ceph/pull/52575
> >> https://tracker.ceph.com/issues/62217
> >>
> >> Cc: Xiubo Li <xiubli@redhat.com>
> >> Cc: Jeff Layton <jlayton@kernel.org>
> >> Cc: Ilya Dryomov <idryomov@gmail.com>
> >> Cc: ceph-devel@vger.kernel.org
> >> Co-Developed-by: Alexander Mikhalitsyn
> >> <aleksandr.mikhalitsyn@canonical.com>
> >> Signed-off-by: Christian Brauner <brauner@kernel.org>
> >> Signed-off-by: Alexander Mikhalitsyn
> >> <aleksandr.mikhalitsyn@canonical.com>
> >> ---
> >> v7:
> >>     - reworked to use two new fields for owner UID/GID
> >> (https://github.com/ceph/ceph/pull/52575)
> >> v8:
> >>     - properly handled case when old MDS used with new kernel client
> >> ---
> >>   fs/ceph/mds_client.c         | 46 +++++++++++++++++++++++++++++++++---
> >>   fs/ceph/mds_client.h         |  5 +++-
> >>   include/linux/ceph/ceph_fs.h |  4 +++-
> >>   3 files changed, 50 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> >> index 8829f55103da..7d3106d3b726 100644
> >> --- a/fs/ceph/mds_client.c
> >> +++ b/fs/ceph/mds_client.c
> >> @@ -2902,6 +2902,17 @@ static void encode_mclientrequest_tail(void
> >> **p, const struct ceph_mds_request *
> >>       }
> >>   }
> >>   +static inline u16 mds_supported_head_version(struct
> >> ceph_mds_session *session)
> >> +{
> >> +    if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
> >> &session->s_features))
> >> +        return 1;
> >> +
> >> +    if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> >> &session->s_features))
> >> +        return 2;
> >> +
> >> +    return CEPH_MDS_REQUEST_HEAD_VERSION;
> >> +}
> >> +
> >>   static struct ceph_mds_request_head_legacy *
> >>   find_legacy_request_head(void *p, u64 features)
> >>   {
> >> @@ -2923,6 +2934,7 @@ static struct ceph_msg
> >> *create_request_message(struct ceph_mds_session *session,
> >>   {
> >>       int mds = session->s_mds;
> >>       struct ceph_mds_client *mdsc = session->s_mdsc;
> >> +    struct ceph_client *cl = mdsc->fsc->client;
> >>       struct ceph_msg *msg;
> >>       struct ceph_mds_request_head_legacy *lhead;
> >>       const char *path1 = NULL;
> >> @@ -2936,7 +2948,7 @@ static struct ceph_msg
> >> *create_request_message(struct ceph_mds_session *session,
> >>       void *p, *end;
> >>       int ret;
> >>       bool legacy = !(session->s_con.peer_features &
> >> CEPH_FEATURE_FS_BTIME);
> >> -    bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
> >> &session->s_features);
> >> +    u16 request_head_version = mds_supported_head_version(session);
> >>         ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
> >>                     req->r_parent, req->r_path1, req->r_ino1.ino,
> >> @@ -2977,8 +2989,10 @@ static struct ceph_msg
> >> *create_request_message(struct ceph_mds_session *session,
> >>        */
> >>       if (legacy)
> >>           len = sizeof(struct ceph_mds_request_head_legacy);
> >> -    else if (old_version)
> >> +    else if (request_head_version == 1)
> >>           len = sizeof(struct ceph_mds_request_head_old);
> >> +    else if (request_head_version == 2)
> >> +        len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
> >>       else
> >>           len = sizeof(struct ceph_mds_request_head);
> >
> > This is not what we suppose to. If we do this again and again when
> > adding new members it will make the code very complicated to maintain.
> >
> > Once the CEPHFS_FEATURE_32BITS_RETRY_FWD has been supported the ceph
> > should correctly decode it and if CEPHFS_FEATURE_HAS_OWNER_UIDGID is
> > not supported the decoder should skip it directly.
> >
> > Is the MDS side buggy ? Why you last version didn't work ?
> >
>
> I think the ceph side is buggy. Possibly we should add one new `length`
> member in struct `struct ceph_mds_request_head` and just skip the extra
> bytes when decoding it.

Hm, I think I found something suspicious. In cephfs code we have many
places that
call the DECODE_FINISH macro, but in our decoder we don't have it.

From documentation it follows that DECODE_FINISH purpose is precisely
about this problem.

What do you think?

>
> Could you fix it together with your ceph PR ?
>
> Thanks
>
> - Xiubo
>
>
> > Thanks
> >
> > - Xiubo
> >
> >> @@ -3028,6 +3042,16 @@ static struct ceph_msg
> >> *create_request_message(struct ceph_mds_session *session,
> >>       lhead = find_legacy_request_head(msg->front.iov_base,
> >>                        session->s_con.peer_features);
> >>   +    if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
> >> +        !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> >> &session->s_features)) {
> >> +        pr_err_ratelimited_client(cl,
> >> +            "idmapped mount is used and
> >> CEPHFS_FEATURE_HAS_OWNER_UIDGID"
> >> +            " is not supported by MDS. Fail request with -EIO.\n");
> >> +
> >> +        ret = -EIO;
> >> +        goto out_err;
> >> +    }
> >> +
> >>       /*
> >>        * The ceph_mds_request_head_legacy didn't contain a version
> >> field, and
> >>        * one was added when we moved the message version from 3->4.
> >> @@ -3035,17 +3059,33 @@ static struct ceph_msg
> >> *create_request_message(struct ceph_mds_session *session,
> >>       if (legacy) {
> >>           msg->hdr.version = cpu_to_le16(3);
> >>           p = msg->front.iov_base + sizeof(*lhead);
> >> -    } else if (old_version) {
> >> +    } else if (request_head_version == 1) {
> >>           struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
> >>             msg->hdr.version = cpu_to_le16(4);
> >>           ohead->version = cpu_to_le16(1);
> >>           p = msg->front.iov_base + sizeof(*ohead);
> >> +    } else if (request_head_version == 2) {
> >> +        struct ceph_mds_request_head *nhead = msg->front.iov_base;
> >> +
> >> +        msg->hdr.version = cpu_to_le16(6);
> >> +        nhead->version = cpu_to_le16(2);
> >> +
> >> +        p = msg->front.iov_base + offsetofend(struct
> >> ceph_mds_request_head, ext_num_fwd);
> >>       } else {
> >>           struct ceph_mds_request_head *nhead = msg->front.iov_base;
> >> +        kuid_t owner_fsuid;
> >> +        kgid_t owner_fsgid;
> >>             msg->hdr.version = cpu_to_le16(6);
> >>           nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
> >> +
> >> +        owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
> >> +                      VFSUIDT_INIT(req->r_cred->fsuid));
> >> +        owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
> >> +                      VFSGIDT_INIT(req->r_cred->fsgid));
> >> +        nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns,
> >> owner_fsuid));
> >> +        nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns,
> >> owner_fsgid));
> >>           p = msg->front.iov_base + sizeof(*nhead);
> >>       }
> >>   diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
> >> index e3bbf3ba8ee8..8f683e8203bd 100644
> >> --- a/fs/ceph/mds_client.h
> >> +++ b/fs/ceph/mds_client.h
> >> @@ -33,8 +33,10 @@ enum ceph_feature_type {
> >>       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
> >>       CEPHFS_FEATURE_OP_GETVXATTR,
> >>       CEPHFS_FEATURE_32BITS_RETRY_FWD,
> >> +    CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
> >> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> >>   -    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
> >> +    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> >>   };
> >>     #define CEPHFS_FEATURES_CLIENT_SUPPORTED {    \
> >> @@ -49,6 +51,7 @@ enum ceph_feature_type {
> >>       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,    \
> >>       CEPHFS_FEATURE_OP_GETVXATTR,        \
> >>       CEPHFS_FEATURE_32BITS_RETRY_FWD,    \
> >> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,    \
> >>   }
> >>     /*
> >> diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
> >> index 5f2301ee88bc..6eb83a51341c 100644
> >> --- a/include/linux/ceph/ceph_fs.h
> >> +++ b/include/linux/ceph/ceph_fs.h
> >> @@ -499,7 +499,7 @@ struct ceph_mds_request_head_legacy {
> >>       union ceph_mds_request_args args;
> >>   } __attribute__ ((packed));
> >>   -#define CEPH_MDS_REQUEST_HEAD_VERSION  2
> >> +#define CEPH_MDS_REQUEST_HEAD_VERSION  3
> >>     struct ceph_mds_request_head_old {
> >>       __le16 version;                /* struct version */
> >> @@ -530,6 +530,8 @@ struct ceph_mds_request_head {
> >>         __le32 ext_num_retry;          /* new count retry attempts */
> >>       __le32 ext_num_fwd;            /* new count fwd attempts */
> >> +
> >> +    __le32 owner_uid, owner_gid;   /* used for OPs which create
> >> inodes */
> >>   } __attribute__ ((packed));
> >>     /* cap/lease release record */
>
  
Aleksandr Mikhalitsyn Aug. 4, 2023, 6:43 a.m. UTC | #5
On Fri, Aug 4, 2023 at 8:35 AM Aleksandr Mikhalitsyn
<aleksandr.mikhalitsyn@canonical.com> wrote:
>
> On Fri, Aug 4, 2023 at 5:24 AM Xiubo Li <xiubli@redhat.com> wrote:
> >
> >
> > On 8/4/23 10:26, Xiubo Li wrote:
> > >
> > > On 8/3/23 21:59, Alexander Mikhalitsyn wrote:
> > >> From: Christian Brauner <brauner@kernel.org>
> > >>
> > >> Inode operations that create a new filesystem object such as ->mknod,
> > >> ->create, ->mkdir() and others don't take a {g,u}id argument explicitly.
> > >> Instead the caller's fs{g,u}id is used for the {g,u}id of the new
> > >> filesystem object.
> > >>
> > >> In order to ensure that the correct {g,u}id is used map the caller's
> > >> fs{g,u}id for creation requests. This doesn't require complex changes.
> > >> It suffices to pass in the relevant idmapping recorded in the request
> > >> message. If this request message was triggered from an inode operation
> > >> that creates filesystem objects it will have passed down the relevant
> > >> idmaping. If this is a request message that was triggered from an inode
> > >> operation that doens't need to take idmappings into account the initial
> > >> idmapping is passed down which is an identity mapping.
> > >>
> > >> This change uses a new cephfs protocol extension
> > >> CEPHFS_FEATURE_HAS_OWNER_UIDGID
> > >> which adds two new fields (owner_{u,g}id) to the request head structure.
> > >> So, we need to ensure that MDS supports it otherwise we need to fail
> > >> any IO that comes through an idmapped mount because we can't process it
> > >> in a proper way. MDS server without such an extension will use
> > >> caller_{u,g}id
> > >> fields to set a new inode owner UID/GID which is incorrect because
> > >> caller_{u,g}id
> > >> values are unmapped. At the same time we can't map these fields with an
> > >> idmapping as it can break UID/GID-based permission checks logic on the
> > >> MDS side. This problem was described with a lot of details at [1], [2].
> > >>
> > >> [1]
> > >> https://lore.kernel.org/lkml/CAEivzxfw1fHO2TFA4dx3u23ZKK6Q+EThfzuibrhA3RKM=ZOYLg@mail.gmail.com/
> > >> [2]
> > >> https://lore.kernel.org/all/20220104140414.155198-3-brauner@kernel.org/
> > >>
> > >> https://github.com/ceph/ceph/pull/52575
> > >> https://tracker.ceph.com/issues/62217
> > >>
> > >> Cc: Xiubo Li <xiubli@redhat.com>
> > >> Cc: Jeff Layton <jlayton@kernel.org>
> > >> Cc: Ilya Dryomov <idryomov@gmail.com>
> > >> Cc: ceph-devel@vger.kernel.org
> > >> Co-Developed-by: Alexander Mikhalitsyn
> > >> <aleksandr.mikhalitsyn@canonical.com>
> > >> Signed-off-by: Christian Brauner <brauner@kernel.org>
> > >> Signed-off-by: Alexander Mikhalitsyn
> > >> <aleksandr.mikhalitsyn@canonical.com>
> > >> ---
> > >> v7:
> > >>     - reworked to use two new fields for owner UID/GID
> > >> (https://github.com/ceph/ceph/pull/52575)
> > >> v8:
> > >>     - properly handled case when old MDS used with new kernel client
> > >> ---
> > >>   fs/ceph/mds_client.c         | 46 +++++++++++++++++++++++++++++++++---
> > >>   fs/ceph/mds_client.h         |  5 +++-
> > >>   include/linux/ceph/ceph_fs.h |  4 +++-
> > >>   3 files changed, 50 insertions(+), 5 deletions(-)
> > >>
> > >> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> > >> index 8829f55103da..7d3106d3b726 100644
> > >> --- a/fs/ceph/mds_client.c
> > >> +++ b/fs/ceph/mds_client.c
> > >> @@ -2902,6 +2902,17 @@ static void encode_mclientrequest_tail(void
> > >> **p, const struct ceph_mds_request *
> > >>       }
> > >>   }
> > >>   +static inline u16 mds_supported_head_version(struct
> > >> ceph_mds_session *session)
> > >> +{
> > >> +    if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > >> &session->s_features))
> > >> +        return 1;
> > >> +
> > >> +    if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> > >> &session->s_features))
> > >> +        return 2;
> > >> +
> > >> +    return CEPH_MDS_REQUEST_HEAD_VERSION;
> > >> +}
> > >> +
> > >>   static struct ceph_mds_request_head_legacy *
> > >>   find_legacy_request_head(void *p, u64 features)
> > >>   {
> > >> @@ -2923,6 +2934,7 @@ static struct ceph_msg
> > >> *create_request_message(struct ceph_mds_session *session,
> > >>   {
> > >>       int mds = session->s_mds;
> > >>       struct ceph_mds_client *mdsc = session->s_mdsc;
> > >> +    struct ceph_client *cl = mdsc->fsc->client;
> > >>       struct ceph_msg *msg;
> > >>       struct ceph_mds_request_head_legacy *lhead;
> > >>       const char *path1 = NULL;
> > >> @@ -2936,7 +2948,7 @@ static struct ceph_msg
> > >> *create_request_message(struct ceph_mds_session *session,
> > >>       void *p, *end;
> > >>       int ret;
> > >>       bool legacy = !(session->s_con.peer_features &
> > >> CEPH_FEATURE_FS_BTIME);
> > >> -    bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > >> &session->s_features);
> > >> +    u16 request_head_version = mds_supported_head_version(session);
> > >>         ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
> > >>                     req->r_parent, req->r_path1, req->r_ino1.ino,
> > >> @@ -2977,8 +2989,10 @@ static struct ceph_msg
> > >> *create_request_message(struct ceph_mds_session *session,
> > >>        */
> > >>       if (legacy)
> > >>           len = sizeof(struct ceph_mds_request_head_legacy);
> > >> -    else if (old_version)
> > >> +    else if (request_head_version == 1)
> > >>           len = sizeof(struct ceph_mds_request_head_old);
> > >> +    else if (request_head_version == 2)
> > >> +        len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
> > >>       else
> > >>           len = sizeof(struct ceph_mds_request_head);
> > >
> > > This is not what we suppose to. If we do this again and again when
> > > adding new members it will make the code very complicated to maintain.
> > >
> > > Once the CEPHFS_FEATURE_32BITS_RETRY_FWD has been supported the ceph
> > > should correctly decode it and if CEPHFS_FEATURE_HAS_OWNER_UIDGID is
> > > not supported the decoder should skip it directly.
> > >
> > > Is the MDS side buggy ? Why you last version didn't work ?
> > >
> >
> > I think the ceph side is buggy. Possibly we should add one new `length`
> > member in struct `struct ceph_mds_request_head` and just skip the extra
> > bytes when decoding it.
>
> Hm, I think I found something suspicious. In cephfs code we have many
> places that
> call the DECODE_FINISH macro, but in our decoder we don't have it.
>
> From documentation it follows that DECODE_FINISH purpose is precisely
> about this problem.
>
> What do you think?

Upd: this thing also changes on-wire format and adds field to store length.
But this will be a massive and incompatible protocol change. I don't think that
we want to do this in the scope of this task.

>
> >
> > Could you fix it together with your ceph PR ?
> >
> > Thanks
> >
> > - Xiubo
> >
> >
> > > Thanks
> > >
> > > - Xiubo
> > >
> > >> @@ -3028,6 +3042,16 @@ static struct ceph_msg
> > >> *create_request_message(struct ceph_mds_session *session,
> > >>       lhead = find_legacy_request_head(msg->front.iov_base,
> > >>                        session->s_con.peer_features);
> > >>   +    if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
> > >> +        !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> > >> &session->s_features)) {
> > >> +        pr_err_ratelimited_client(cl,
> > >> +            "idmapped mount is used and
> > >> CEPHFS_FEATURE_HAS_OWNER_UIDGID"
> > >> +            " is not supported by MDS. Fail request with -EIO.\n");
> > >> +
> > >> +        ret = -EIO;
> > >> +        goto out_err;
> > >> +    }
> > >> +
> > >>       /*
> > >>        * The ceph_mds_request_head_legacy didn't contain a version
> > >> field, and
> > >>        * one was added when we moved the message version from 3->4.
> > >> @@ -3035,17 +3059,33 @@ static struct ceph_msg
> > >> *create_request_message(struct ceph_mds_session *session,
> > >>       if (legacy) {
> > >>           msg->hdr.version = cpu_to_le16(3);
> > >>           p = msg->front.iov_base + sizeof(*lhead);
> > >> -    } else if (old_version) {
> > >> +    } else if (request_head_version == 1) {
> > >>           struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
> > >>             msg->hdr.version = cpu_to_le16(4);
> > >>           ohead->version = cpu_to_le16(1);
> > >>           p = msg->front.iov_base + sizeof(*ohead);
> > >> +    } else if (request_head_version == 2) {
> > >> +        struct ceph_mds_request_head *nhead = msg->front.iov_base;
> > >> +
> > >> +        msg->hdr.version = cpu_to_le16(6);
> > >> +        nhead->version = cpu_to_le16(2);
> > >> +
> > >> +        p = msg->front.iov_base + offsetofend(struct
> > >> ceph_mds_request_head, ext_num_fwd);
> > >>       } else {
> > >>           struct ceph_mds_request_head *nhead = msg->front.iov_base;
> > >> +        kuid_t owner_fsuid;
> > >> +        kgid_t owner_fsgid;
> > >>             msg->hdr.version = cpu_to_le16(6);
> > >>           nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
> > >> +
> > >> +        owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
> > >> +                      VFSUIDT_INIT(req->r_cred->fsuid));
> > >> +        owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
> > >> +                      VFSGIDT_INIT(req->r_cred->fsgid));
> > >> +        nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns,
> > >> owner_fsuid));
> > >> +        nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns,
> > >> owner_fsgid));
> > >>           p = msg->front.iov_base + sizeof(*nhead);
> > >>       }
> > >>   diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
> > >> index e3bbf3ba8ee8..8f683e8203bd 100644
> > >> --- a/fs/ceph/mds_client.h
> > >> +++ b/fs/ceph/mds_client.h
> > >> @@ -33,8 +33,10 @@ enum ceph_feature_type {
> > >>       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
> > >>       CEPHFS_FEATURE_OP_GETVXATTR,
> > >>       CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > >> +    CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
> > >> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> > >>   -    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > >> +    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> > >>   };
> > >>     #define CEPHFS_FEATURES_CLIENT_SUPPORTED {    \
> > >> @@ -49,6 +51,7 @@ enum ceph_feature_type {
> > >>       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,    \
> > >>       CEPHFS_FEATURE_OP_GETVXATTR,        \
> > >>       CEPHFS_FEATURE_32BITS_RETRY_FWD,    \
> > >> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,    \
> > >>   }
> > >>     /*
> > >> diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
> > >> index 5f2301ee88bc..6eb83a51341c 100644
> > >> --- a/include/linux/ceph/ceph_fs.h
> > >> +++ b/include/linux/ceph/ceph_fs.h
> > >> @@ -499,7 +499,7 @@ struct ceph_mds_request_head_legacy {
> > >>       union ceph_mds_request_args args;
> > >>   } __attribute__ ((packed));
> > >>   -#define CEPH_MDS_REQUEST_HEAD_VERSION  2
> > >> +#define CEPH_MDS_REQUEST_HEAD_VERSION  3
> > >>     struct ceph_mds_request_head_old {
> > >>       __le16 version;                /* struct version */
> > >> @@ -530,6 +530,8 @@ struct ceph_mds_request_head {
> > >>         __le32 ext_num_retry;          /* new count retry attempts */
> > >>       __le32 ext_num_fwd;            /* new count fwd attempts */
> > >> +
> > >> +    __le32 owner_uid, owner_gid;   /* used for OPs which create
> > >> inodes */
> > >>   } __attribute__ ((packed));
> > >>     /* cap/lease release record */
> >
  
Aleksandr Mikhalitsyn Aug. 4, 2023, 7:33 a.m. UTC | #6
On Fri, Aug 4, 2023 at 8:43 AM Aleksandr Mikhalitsyn
<aleksandr.mikhalitsyn@canonical.com> wrote:
>
> On Fri, Aug 4, 2023 at 8:35 AM Aleksandr Mikhalitsyn
> <aleksandr.mikhalitsyn@canonical.com> wrote:
> >
> > On Fri, Aug 4, 2023 at 5:24 AM Xiubo Li <xiubli@redhat.com> wrote:
> > >
> > >
> > > On 8/4/23 10:26, Xiubo Li wrote:
> > > >
> > > > On 8/3/23 21:59, Alexander Mikhalitsyn wrote:
> > > >> From: Christian Brauner <brauner@kernel.org>
> > > >>
> > > >> Inode operations that create a new filesystem object such as ->mknod,
> > > >> ->create, ->mkdir() and others don't take a {g,u}id argument explicitly.
> > > >> Instead the caller's fs{g,u}id is used for the {g,u}id of the new
> > > >> filesystem object.
> > > >>
> > > >> In order to ensure that the correct {g,u}id is used map the caller's
> > > >> fs{g,u}id for creation requests. This doesn't require complex changes.
> > > >> It suffices to pass in the relevant idmapping recorded in the request
> > > >> message. If this request message was triggered from an inode operation
> > > >> that creates filesystem objects it will have passed down the relevant
> > > >> idmaping. If this is a request message that was triggered from an inode
> > > >> operation that doens't need to take idmappings into account the initial
> > > >> idmapping is passed down which is an identity mapping.
> > > >>
> > > >> This change uses a new cephfs protocol extension
> > > >> CEPHFS_FEATURE_HAS_OWNER_UIDGID
> > > >> which adds two new fields (owner_{u,g}id) to the request head structure.
> > > >> So, we need to ensure that MDS supports it otherwise we need to fail
> > > >> any IO that comes through an idmapped mount because we can't process it
> > > >> in a proper way. MDS server without such an extension will use
> > > >> caller_{u,g}id
> > > >> fields to set a new inode owner UID/GID which is incorrect because
> > > >> caller_{u,g}id
> > > >> values are unmapped. At the same time we can't map these fields with an
> > > >> idmapping as it can break UID/GID-based permission checks logic on the
> > > >> MDS side. This problem was described with a lot of details at [1], [2].
> > > >>
> > > >> [1]
> > > >> https://lore.kernel.org/lkml/CAEivzxfw1fHO2TFA4dx3u23ZKK6Q+EThfzuibrhA3RKM=ZOYLg@mail.gmail.com/
> > > >> [2]
> > > >> https://lore.kernel.org/all/20220104140414.155198-3-brauner@kernel.org/
> > > >>
> > > >> https://github.com/ceph/ceph/pull/52575
> > > >> https://tracker.ceph.com/issues/62217
> > > >>
> > > >> Cc: Xiubo Li <xiubli@redhat.com>
> > > >> Cc: Jeff Layton <jlayton@kernel.org>
> > > >> Cc: Ilya Dryomov <idryomov@gmail.com>
> > > >> Cc: ceph-devel@vger.kernel.org
> > > >> Co-Developed-by: Alexander Mikhalitsyn
> > > >> <aleksandr.mikhalitsyn@canonical.com>
> > > >> Signed-off-by: Christian Brauner <brauner@kernel.org>
> > > >> Signed-off-by: Alexander Mikhalitsyn
> > > >> <aleksandr.mikhalitsyn@canonical.com>
> > > >> ---
> > > >> v7:
> > > >>     - reworked to use two new fields for owner UID/GID
> > > >> (https://github.com/ceph/ceph/pull/52575)
> > > >> v8:
> > > >>     - properly handled case when old MDS used with new kernel client
> > > >> ---
> > > >>   fs/ceph/mds_client.c         | 46 +++++++++++++++++++++++++++++++++---
> > > >>   fs/ceph/mds_client.h         |  5 +++-
> > > >>   include/linux/ceph/ceph_fs.h |  4 +++-
> > > >>   3 files changed, 50 insertions(+), 5 deletions(-)
> > > >>
> > > >> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> > > >> index 8829f55103da..7d3106d3b726 100644
> > > >> --- a/fs/ceph/mds_client.c
> > > >> +++ b/fs/ceph/mds_client.c
> > > >> @@ -2902,6 +2902,17 @@ static void encode_mclientrequest_tail(void
> > > >> **p, const struct ceph_mds_request *
> > > >>       }
> > > >>   }
> > > >>   +static inline u16 mds_supported_head_version(struct
> > > >> ceph_mds_session *session)
> > > >> +{
> > > >> +    if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > > >> &session->s_features))
> > > >> +        return 1;
> > > >> +
> > > >> +    if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> > > >> &session->s_features))
> > > >> +        return 2;
> > > >> +
> > > >> +    return CEPH_MDS_REQUEST_HEAD_VERSION;
> > > >> +}
> > > >> +
> > > >>   static struct ceph_mds_request_head_legacy *
> > > >>   find_legacy_request_head(void *p, u64 features)
> > > >>   {
> > > >> @@ -2923,6 +2934,7 @@ static struct ceph_msg
> > > >> *create_request_message(struct ceph_mds_session *session,
> > > >>   {
> > > >>       int mds = session->s_mds;
> > > >>       struct ceph_mds_client *mdsc = session->s_mdsc;
> > > >> +    struct ceph_client *cl = mdsc->fsc->client;
> > > >>       struct ceph_msg *msg;
> > > >>       struct ceph_mds_request_head_legacy *lhead;
> > > >>       const char *path1 = NULL;
> > > >> @@ -2936,7 +2948,7 @@ static struct ceph_msg
> > > >> *create_request_message(struct ceph_mds_session *session,
> > > >>       void *p, *end;
> > > >>       int ret;
> > > >>       bool legacy = !(session->s_con.peer_features &
> > > >> CEPH_FEATURE_FS_BTIME);
> > > >> -    bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > > >> &session->s_features);
> > > >> +    u16 request_head_version = mds_supported_head_version(session);
> > > >>         ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
> > > >>                     req->r_parent, req->r_path1, req->r_ino1.ino,
> > > >> @@ -2977,8 +2989,10 @@ static struct ceph_msg
> > > >> *create_request_message(struct ceph_mds_session *session,
> > > >>        */
> > > >>       if (legacy)
> > > >>           len = sizeof(struct ceph_mds_request_head_legacy);
> > > >> -    else if (old_version)
> > > >> +    else if (request_head_version == 1)
> > > >>           len = sizeof(struct ceph_mds_request_head_old);
> > > >> +    else if (request_head_version == 2)
> > > >> +        len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
> > > >>       else
> > > >>           len = sizeof(struct ceph_mds_request_head);
> > > >
> > > > This is not what we suppose to. If we do this again and again when
> > > > adding new members it will make the code very complicated to maintain.
> > > >
> > > > Once the CEPHFS_FEATURE_32BITS_RETRY_FWD has been supported the ceph
> > > > should correctly decode it and if CEPHFS_FEATURE_HAS_OWNER_UIDGID is
> > > > not supported the decoder should skip it directly.
> > > >
> > > > Is the MDS side buggy ? Why you last version didn't work ?
> > > >
> > >
> > > I think the ceph side is buggy. Possibly we should add one new `length`
> > > member in struct `struct ceph_mds_request_head` and just skip the extra
> > > bytes when decoding it.
> >
> > Hm, I think I found something suspicious. In cephfs code we have many
> > places that
> > call the DECODE_FINISH macro, but in our decoder we don't have it.
> >
> > From documentation it follows that DECODE_FINISH purpose is precisely
> > about this problem.
> >
> > What do you think?
>
> Upd: this thing also changes on-wire format and adds field to store length.
> But this will be a massive and incompatible protocol change. I don't think that
> we want to do this in the scope of this task.

https://github.com/ceph/ceph/pull/52575#issuecomment-1665141641

>
> >
> > >
> > > Could you fix it together with your ceph PR ?
> > >
> > > Thanks
> > >
> > > - Xiubo
> > >
> > >
> > > > Thanks
> > > >
> > > > - Xiubo
> > > >
> > > >> @@ -3028,6 +3042,16 @@ static struct ceph_msg
> > > >> *create_request_message(struct ceph_mds_session *session,
> > > >>       lhead = find_legacy_request_head(msg->front.iov_base,
> > > >>                        session->s_con.peer_features);
> > > >>   +    if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
> > > >> +        !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> > > >> &session->s_features)) {
> > > >> +        pr_err_ratelimited_client(cl,
> > > >> +            "idmapped mount is used and
> > > >> CEPHFS_FEATURE_HAS_OWNER_UIDGID"
> > > >> +            " is not supported by MDS. Fail request with -EIO.\n");
> > > >> +
> > > >> +        ret = -EIO;
> > > >> +        goto out_err;
> > > >> +    }
> > > >> +
> > > >>       /*
> > > >>        * The ceph_mds_request_head_legacy didn't contain a version
> > > >> field, and
> > > >>        * one was added when we moved the message version from 3->4.
> > > >> @@ -3035,17 +3059,33 @@ static struct ceph_msg
> > > >> *create_request_message(struct ceph_mds_session *session,
> > > >>       if (legacy) {
> > > >>           msg->hdr.version = cpu_to_le16(3);
> > > >>           p = msg->front.iov_base + sizeof(*lhead);
> > > >> -    } else if (old_version) {
> > > >> +    } else if (request_head_version == 1) {
> > > >>           struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
> > > >>             msg->hdr.version = cpu_to_le16(4);
> > > >>           ohead->version = cpu_to_le16(1);
> > > >>           p = msg->front.iov_base + sizeof(*ohead);
> > > >> +    } else if (request_head_version == 2) {
> > > >> +        struct ceph_mds_request_head *nhead = msg->front.iov_base;
> > > >> +
> > > >> +        msg->hdr.version = cpu_to_le16(6);
> > > >> +        nhead->version = cpu_to_le16(2);
> > > >> +
> > > >> +        p = msg->front.iov_base + offsetofend(struct
> > > >> ceph_mds_request_head, ext_num_fwd);
> > > >>       } else {
> > > >>           struct ceph_mds_request_head *nhead = msg->front.iov_base;
> > > >> +        kuid_t owner_fsuid;
> > > >> +        kgid_t owner_fsgid;
> > > >>             msg->hdr.version = cpu_to_le16(6);
> > > >>           nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
> > > >> +
> > > >> +        owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
> > > >> +                      VFSUIDT_INIT(req->r_cred->fsuid));
> > > >> +        owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
> > > >> +                      VFSGIDT_INIT(req->r_cred->fsgid));
> > > >> +        nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns,
> > > >> owner_fsuid));
> > > >> +        nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns,
> > > >> owner_fsgid));
> > > >>           p = msg->front.iov_base + sizeof(*nhead);
> > > >>       }
> > > >>   diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
> > > >> index e3bbf3ba8ee8..8f683e8203bd 100644
> > > >> --- a/fs/ceph/mds_client.h
> > > >> +++ b/fs/ceph/mds_client.h
> > > >> @@ -33,8 +33,10 @@ enum ceph_feature_type {
> > > >>       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
> > > >>       CEPHFS_FEATURE_OP_GETVXATTR,
> > > >>       CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > > >> +    CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
> > > >> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> > > >>   -    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
> > > >> +    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_HAS_OWNER_UIDGID,
> > > >>   };
> > > >>     #define CEPHFS_FEATURES_CLIENT_SUPPORTED {    \
> > > >> @@ -49,6 +51,7 @@ enum ceph_feature_type {
> > > >>       CEPHFS_FEATURE_NOTIFY_SESSION_STATE,    \
> > > >>       CEPHFS_FEATURE_OP_GETVXATTR,        \
> > > >>       CEPHFS_FEATURE_32BITS_RETRY_FWD,    \
> > > >> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,    \
> > > >>   }
> > > >>     /*
> > > >> diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
> > > >> index 5f2301ee88bc..6eb83a51341c 100644
> > > >> --- a/include/linux/ceph/ceph_fs.h
> > > >> +++ b/include/linux/ceph/ceph_fs.h
> > > >> @@ -499,7 +499,7 @@ struct ceph_mds_request_head_legacy {
> > > >>       union ceph_mds_request_args args;
> > > >>   } __attribute__ ((packed));
> > > >>   -#define CEPH_MDS_REQUEST_HEAD_VERSION  2
> > > >> +#define CEPH_MDS_REQUEST_HEAD_VERSION  3
> > > >>     struct ceph_mds_request_head_old {
> > > >>       __le16 version;                /* struct version */
> > > >> @@ -530,6 +530,8 @@ struct ceph_mds_request_head {
> > > >>         __le32 ext_num_retry;          /* new count retry attempts */
> > > >>       __le32 ext_num_fwd;            /* new count fwd attempts */
> > > >> +
> > > >> +    __le32 owner_uid, owner_gid;   /* used for OPs which create
> > > >> inodes */
> > > >>   } __attribute__ ((packed));
> > > >>     /* cap/lease release record */
> > >
  
Xiubo Li Aug. 7, 2023, 12:56 a.m. UTC | #7
On 8/4/23 14:35, Aleksandr Mikhalitsyn wrote:
> On Fri, Aug 4, 2023 at 5:24 AM Xiubo Li <xiubli@redhat.com> wrote:
>>
>> On 8/4/23 10:26, Xiubo Li wrote:
>>> On 8/3/23 21:59, Alexander Mikhalitsyn wrote:
>>>> From: Christian Brauner <brauner@kernel.org>
>>>>
>>>> Inode operations that create a new filesystem object such as ->mknod,
>>>> ->create, ->mkdir() and others don't take a {g,u}id argument explicitly.
>>>> Instead the caller's fs{g,u}id is used for the {g,u}id of the new
>>>> filesystem object.
>>>>
>>>> In order to ensure that the correct {g,u}id is used map the caller's
>>>> fs{g,u}id for creation requests. This doesn't require complex changes.
>>>> It suffices to pass in the relevant idmapping recorded in the request
>>>> message. If this request message was triggered from an inode operation
>>>> that creates filesystem objects it will have passed down the relevant
>>>> idmaping. If this is a request message that was triggered from an inode
>>>> operation that doens't need to take idmappings into account the initial
>>>> idmapping is passed down which is an identity mapping.
>>>>
>>>> This change uses a new cephfs protocol extension
>>>> CEPHFS_FEATURE_HAS_OWNER_UIDGID
>>>> which adds two new fields (owner_{u,g}id) to the request head structure.
>>>> So, we need to ensure that MDS supports it otherwise we need to fail
>>>> any IO that comes through an idmapped mount because we can't process it
>>>> in a proper way. MDS server without such an extension will use
>>>> caller_{u,g}id
>>>> fields to set a new inode owner UID/GID which is incorrect because
>>>> caller_{u,g}id
>>>> values are unmapped. At the same time we can't map these fields with an
>>>> idmapping as it can break UID/GID-based permission checks logic on the
>>>> MDS side. This problem was described with a lot of details at [1], [2].
>>>>
>>>> [1]
>>>> https://lore.kernel.org/lkml/CAEivzxfw1fHO2TFA4dx3u23ZKK6Q+EThfzuibrhA3RKM=ZOYLg@mail.gmail.com/
>>>> [2]
>>>> https://lore.kernel.org/all/20220104140414.155198-3-brauner@kernel.org/
>>>>
>>>> https://github.com/ceph/ceph/pull/52575
>>>> https://tracker.ceph.com/issues/62217
>>>>
>>>> Cc: Xiubo Li <xiubli@redhat.com>
>>>> Cc: Jeff Layton <jlayton@kernel.org>
>>>> Cc: Ilya Dryomov <idryomov@gmail.com>
>>>> Cc: ceph-devel@vger.kernel.org
>>>> Co-Developed-by: Alexander Mikhalitsyn
>>>> <aleksandr.mikhalitsyn@canonical.com>
>>>> Signed-off-by: Christian Brauner <brauner@kernel.org>
>>>> Signed-off-by: Alexander Mikhalitsyn
>>>> <aleksandr.mikhalitsyn@canonical.com>
>>>> ---
>>>> v7:
>>>>      - reworked to use two new fields for owner UID/GID
>>>> (https://github.com/ceph/ceph/pull/52575)
>>>> v8:
>>>>      - properly handled case when old MDS used with new kernel client
>>>> ---
>>>>    fs/ceph/mds_client.c         | 46 +++++++++++++++++++++++++++++++++---
>>>>    fs/ceph/mds_client.h         |  5 +++-
>>>>    include/linux/ceph/ceph_fs.h |  4 +++-
>>>>    3 files changed, 50 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
>>>> index 8829f55103da..7d3106d3b726 100644
>>>> --- a/fs/ceph/mds_client.c
>>>> +++ b/fs/ceph/mds_client.c
>>>> @@ -2902,6 +2902,17 @@ static void encode_mclientrequest_tail(void
>>>> **p, const struct ceph_mds_request *
>>>>        }
>>>>    }
>>>>    +static inline u16 mds_supported_head_version(struct
>>>> ceph_mds_session *session)
>>>> +{
>>>> +    if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
>>>> &session->s_features))
>>>> +        return 1;
>>>> +
>>>> +    if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID,
>>>> &session->s_features))
>>>> +        return 2;
>>>> +
>>>> +    return CEPH_MDS_REQUEST_HEAD_VERSION;
>>>> +}
>>>> +
>>>>    static struct ceph_mds_request_head_legacy *
>>>>    find_legacy_request_head(void *p, u64 features)
>>>>    {
>>>> @@ -2923,6 +2934,7 @@ static struct ceph_msg
>>>> *create_request_message(struct ceph_mds_session *session,
>>>>    {
>>>>        int mds = session->s_mds;
>>>>        struct ceph_mds_client *mdsc = session->s_mdsc;
>>>> +    struct ceph_client *cl = mdsc->fsc->client;
>>>>        struct ceph_msg *msg;
>>>>        struct ceph_mds_request_head_legacy *lhead;
>>>>        const char *path1 = NULL;
>>>> @@ -2936,7 +2948,7 @@ static struct ceph_msg
>>>> *create_request_message(struct ceph_mds_session *session,
>>>>        void *p, *end;
>>>>        int ret;
>>>>        bool legacy = !(session->s_con.peer_features &
>>>> CEPH_FEATURE_FS_BTIME);
>>>> -    bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
>>>> &session->s_features);
>>>> +    u16 request_head_version = mds_supported_head_version(session);
>>>>          ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
>>>>                      req->r_parent, req->r_path1, req->r_ino1.ino,
>>>> @@ -2977,8 +2989,10 @@ static struct ceph_msg
>>>> *create_request_message(struct ceph_mds_session *session,
>>>>         */
>>>>        if (legacy)
>>>>            len = sizeof(struct ceph_mds_request_head_legacy);
>>>> -    else if (old_version)
>>>> +    else if (request_head_version == 1)
>>>>            len = sizeof(struct ceph_mds_request_head_old);
>>>> +    else if (request_head_version == 2)
>>>> +        len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
>>>>        else
>>>>            len = sizeof(struct ceph_mds_request_head);
>>> This is not what we suppose to. If we do this again and again when
>>> adding new members it will make the code very complicated to maintain.
>>>
>>> Once the CEPHFS_FEATURE_32BITS_RETRY_FWD has been supported the ceph
>>> should correctly decode it and if CEPHFS_FEATURE_HAS_OWNER_UIDGID is
>>> not supported the decoder should skip it directly.
>>>
>>> Is the MDS side buggy ? Why you last version didn't work ?
>>>
>> I think the ceph side is buggy. Possibly we should add one new `length`
>> member in struct `struct ceph_mds_request_head` and just skip the extra
>> bytes when decoding it.
> Hm, I think I found something suspicious. In cephfs code we have many
> places that
> call the DECODE_FINISH macro, but in our decoder we don't have it.
>
>  From documentation it follows that DECODE_FINISH purpose is precisely
> about this problem.
>
> What do you think?

Yeah, correct.

We also need to do it like this.

Thanks

- Xiubo


>> Could you fix it together with your ceph PR ?
>>
>> Thanks
>>
>> - Xiubo
>>
>>
>>> Thanks
>>>
>>> - Xiubo
>>>
>>>> @@ -3028,6 +3042,16 @@ static struct ceph_msg
>>>> *create_request_message(struct ceph_mds_session *session,
>>>>        lhead = find_legacy_request_head(msg->front.iov_base,
>>>>                         session->s_con.peer_features);
>>>>    +    if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
>>>> +        !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID,
>>>> &session->s_features)) {
>>>> +        pr_err_ratelimited_client(cl,
>>>> +            "idmapped mount is used and
>>>> CEPHFS_FEATURE_HAS_OWNER_UIDGID"
>>>> +            " is not supported by MDS. Fail request with -EIO.\n");
>>>> +
>>>> +        ret = -EIO;
>>>> +        goto out_err;
>>>> +    }
>>>> +
>>>>        /*
>>>>         * The ceph_mds_request_head_legacy didn't contain a version
>>>> field, and
>>>>         * one was added when we moved the message version from 3->4.
>>>> @@ -3035,17 +3059,33 @@ static struct ceph_msg
>>>> *create_request_message(struct ceph_mds_session *session,
>>>>        if (legacy) {
>>>>            msg->hdr.version = cpu_to_le16(3);
>>>>            p = msg->front.iov_base + sizeof(*lhead);
>>>> -    } else if (old_version) {
>>>> +    } else if (request_head_version == 1) {
>>>>            struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
>>>>              msg->hdr.version = cpu_to_le16(4);
>>>>            ohead->version = cpu_to_le16(1);
>>>>            p = msg->front.iov_base + sizeof(*ohead);
>>>> +    } else if (request_head_version == 2) {
>>>> +        struct ceph_mds_request_head *nhead = msg->front.iov_base;
>>>> +
>>>> +        msg->hdr.version = cpu_to_le16(6);
>>>> +        nhead->version = cpu_to_le16(2);
>>>> +
>>>> +        p = msg->front.iov_base + offsetofend(struct
>>>> ceph_mds_request_head, ext_num_fwd);
>>>>        } else {
>>>>            struct ceph_mds_request_head *nhead = msg->front.iov_base;
>>>> +        kuid_t owner_fsuid;
>>>> +        kgid_t owner_fsgid;
>>>>              msg->hdr.version = cpu_to_le16(6);
>>>>            nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
>>>> +
>>>> +        owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
>>>> +                      VFSUIDT_INIT(req->r_cred->fsuid));
>>>> +        owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
>>>> +                      VFSGIDT_INIT(req->r_cred->fsgid));
>>>> +        nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns,
>>>> owner_fsuid));
>>>> +        nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns,
>>>> owner_fsgid));
>>>>            p = msg->front.iov_base + sizeof(*nhead);
>>>>        }
>>>>    diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
>>>> index e3bbf3ba8ee8..8f683e8203bd 100644
>>>> --- a/fs/ceph/mds_client.h
>>>> +++ b/fs/ceph/mds_client.h
>>>> @@ -33,8 +33,10 @@ enum ceph_feature_type {
>>>>        CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
>>>>        CEPHFS_FEATURE_OP_GETVXATTR,
>>>>        CEPHFS_FEATURE_32BITS_RETRY_FWD,
>>>> +    CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
>>>> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,
>>>>    -    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
>>>> +    CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_HAS_OWNER_UIDGID,
>>>>    };
>>>>      #define CEPHFS_FEATURES_CLIENT_SUPPORTED {    \
>>>> @@ -49,6 +51,7 @@ enum ceph_feature_type {
>>>>        CEPHFS_FEATURE_NOTIFY_SESSION_STATE,    \
>>>>        CEPHFS_FEATURE_OP_GETVXATTR,        \
>>>>        CEPHFS_FEATURE_32BITS_RETRY_FWD,    \
>>>> +    CEPHFS_FEATURE_HAS_OWNER_UIDGID,    \
>>>>    }
>>>>      /*
>>>> diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
>>>> index 5f2301ee88bc..6eb83a51341c 100644
>>>> --- a/include/linux/ceph/ceph_fs.h
>>>> +++ b/include/linux/ceph/ceph_fs.h
>>>> @@ -499,7 +499,7 @@ struct ceph_mds_request_head_legacy {
>>>>        union ceph_mds_request_args args;
>>>>    } __attribute__ ((packed));
>>>>    -#define CEPH_MDS_REQUEST_HEAD_VERSION  2
>>>> +#define CEPH_MDS_REQUEST_HEAD_VERSION  3
>>>>      struct ceph_mds_request_head_old {
>>>>        __le16 version;                /* struct version */
>>>> @@ -530,6 +530,8 @@ struct ceph_mds_request_head {
>>>>          __le32 ext_num_retry;          /* new count retry attempts */
>>>>        __le32 ext_num_fwd;            /* new count fwd attempts */
>>>> +
>>>> +    __le32 owner_uid, owner_gid;   /* used for OPs which create
>>>> inodes */
>>>>    } __attribute__ ((packed));
>>>>      /* cap/lease release record */
  

Patch

diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 8829f55103da..7d3106d3b726 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -2902,6 +2902,17 @@  static void encode_mclientrequest_tail(void **p, const struct ceph_mds_request *
 	}
 }
 
+static inline u16 mds_supported_head_version(struct ceph_mds_session *session)
+{
+	if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, &session->s_features))
+		return 1;
+
+	if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features))
+		return 2;
+
+	return CEPH_MDS_REQUEST_HEAD_VERSION;
+}
+
 static struct ceph_mds_request_head_legacy *
 find_legacy_request_head(void *p, u64 features)
 {
@@ -2923,6 +2934,7 @@  static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
 {
 	int mds = session->s_mds;
 	struct ceph_mds_client *mdsc = session->s_mdsc;
+	struct ceph_client *cl = mdsc->fsc->client;
 	struct ceph_msg *msg;
 	struct ceph_mds_request_head_legacy *lhead;
 	const char *path1 = NULL;
@@ -2936,7 +2948,7 @@  static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
 	void *p, *end;
 	int ret;
 	bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
-	bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, &session->s_features);
+	u16 request_head_version = mds_supported_head_version(session);
 
 	ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
 			      req->r_parent, req->r_path1, req->r_ino1.ino,
@@ -2977,8 +2989,10 @@  static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
 	 */
 	if (legacy)
 		len = sizeof(struct ceph_mds_request_head_legacy);
-	else if (old_version)
+	else if (request_head_version == 1)
 		len = sizeof(struct ceph_mds_request_head_old);
+	else if (request_head_version == 2)
+		len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
 	else
 		len = sizeof(struct ceph_mds_request_head);
 
@@ -3028,6 +3042,16 @@  static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
 	lhead = find_legacy_request_head(msg->front.iov_base,
 					 session->s_con.peer_features);
 
+	if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
+	    !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features)) {
+		pr_err_ratelimited_client(cl,
+			"idmapped mount is used and CEPHFS_FEATURE_HAS_OWNER_UIDGID"
+			" is not supported by MDS. Fail request with -EIO.\n");
+
+		ret = -EIO;
+		goto out_err;
+	}
+
 	/*
 	 * The ceph_mds_request_head_legacy didn't contain a version field, and
 	 * one was added when we moved the message version from 3->4.
@@ -3035,17 +3059,33 @@  static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
 	if (legacy) {
 		msg->hdr.version = cpu_to_le16(3);
 		p = msg->front.iov_base + sizeof(*lhead);
-	} else if (old_version) {
+	} else if (request_head_version == 1) {
 		struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
 
 		msg->hdr.version = cpu_to_le16(4);
 		ohead->version = cpu_to_le16(1);
 		p = msg->front.iov_base + sizeof(*ohead);
+	} else if (request_head_version == 2) {
+		struct ceph_mds_request_head *nhead = msg->front.iov_base;
+
+		msg->hdr.version = cpu_to_le16(6);
+		nhead->version = cpu_to_le16(2);
+
+		p = msg->front.iov_base + offsetofend(struct ceph_mds_request_head, ext_num_fwd);
 	} else {
 		struct ceph_mds_request_head *nhead = msg->front.iov_base;
+		kuid_t owner_fsuid;
+		kgid_t owner_fsgid;
 
 		msg->hdr.version = cpu_to_le16(6);
 		nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
+
+		owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
+					  VFSUIDT_INIT(req->r_cred->fsuid));
+		owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
+					  VFSGIDT_INIT(req->r_cred->fsgid));
+		nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns, owner_fsuid));
+		nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns, owner_fsgid));
 		p = msg->front.iov_base + sizeof(*nhead);
 	}
 
diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
index e3bbf3ba8ee8..8f683e8203bd 100644
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -33,8 +33,10 @@  enum ceph_feature_type {
 	CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
 	CEPHFS_FEATURE_OP_GETVXATTR,
 	CEPHFS_FEATURE_32BITS_RETRY_FWD,
+	CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
+	CEPHFS_FEATURE_HAS_OWNER_UIDGID,
 
-	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
+	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_HAS_OWNER_UIDGID,
 };
 
 #define CEPHFS_FEATURES_CLIENT_SUPPORTED {	\
@@ -49,6 +51,7 @@  enum ceph_feature_type {
 	CEPHFS_FEATURE_NOTIFY_SESSION_STATE,	\
 	CEPHFS_FEATURE_OP_GETVXATTR,		\
 	CEPHFS_FEATURE_32BITS_RETRY_FWD,	\
+	CEPHFS_FEATURE_HAS_OWNER_UIDGID,	\
 }
 
 /*
diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
index 5f2301ee88bc..6eb83a51341c 100644
--- a/include/linux/ceph/ceph_fs.h
+++ b/include/linux/ceph/ceph_fs.h
@@ -499,7 +499,7 @@  struct ceph_mds_request_head_legacy {
 	union ceph_mds_request_args args;
 } __attribute__ ((packed));
 
-#define CEPH_MDS_REQUEST_HEAD_VERSION  2
+#define CEPH_MDS_REQUEST_HEAD_VERSION  3
 
 struct ceph_mds_request_head_old {
 	__le16 version;                /* struct version */
@@ -530,6 +530,8 @@  struct ceph_mds_request_head {
 
 	__le32 ext_num_retry;          /* new count retry attempts */
 	__le32 ext_num_fwd;            /* new count fwd attempts */
+
+	__le32 owner_uid, owner_gid;   /* used for OPs which create inodes */
 } __attribute__ ((packed));
 
 /* cap/lease release record */