ceph: add getvxattr support
Subsystems (2)
cover ceph: add getvxattr support
Add getvxattr op support to handle fetching xattr values for:
* ceph.dir.pin*
* ceph.dir.layout*
* ceph.file.layout*
Milind Changire (1):
ceph: add getvxattr op
fs/ceph/inode.c | 51 ++++++++++++++++++++++++++++ fs/ceph/mds_client.c | 27 ++++++++++++++- fs/ceph/mds_client.h | 12 ++++++- fs/ceph/strings.c | 1 + fs/ceph/super.h | 1 + fs/ceph/xattr.c | 65 ++++++++++++++++++++++++++++++++++++ include/linux/ceph/ceph_fs.h | 1 + 7 files changed, 156 insertions(+), 2 deletions(-)
Signature · 1 line
2.34.1
PATCH 1/1 ceph: add getvxattr op
Problem:
Directory vxattrs like ceph.dir.pin* and ceph.dir.layout* may not be propagated to the client as frequently to keep them updated. This creates vxattr availability problems.
Solution:
Adds new getvxattr op to fetch ceph.dir.pin*, ceph.dir.layout* and ceph.file.layout* vxattrs.
If the entire layout for a dir or a file is being set, then it is expected that the layout be set in standard JSON format. Individual field value retrieval is not wrapped in JSON. The JSON format also applies while setting the vxattr if the entire layout is being set in one go.
As a temporary measure, setting a vxattr can also be done in the old format. The old format will be deprecated in the future.
URL: https://tracker.ceph.com/issues/51062
--- fs/ceph/inode.c | 51 ++++++++++++++++++++++++++++ fs/ceph/mds_client.c | 27 ++++++++++++++- fs/ceph/mds_client.h | 12 ++++++- fs/ceph/strings.c | 1 + fs/ceph/super.h | 1 + fs/ceph/xattr.c | 65 ++++++++++++++++++++++++++++++++++++ include/linux/ceph/ceph_fs.h | 1 + 7 files changed, 156 insertions(+), 2 deletions(-)
fs/ceph/inode.c +51 -0
@@ -2291,6 +2291,57 @@ int __ceph_do_getattr(struct inode *inode, struct page *locked_page,return err; } +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value, +size_t size) +{ +struct ceph_fs_client *fsc =ceph_sb_to_client (inode->i_sb); +struct ceph_mds_client *mdsc = fsc->mdsc; +struct ceph_mds_request *req; +int mode =USE_AUTH_MDS ; +int err; +char *xattr_value; +size_t xattr_value_len; + + req =ceph_mdsc_create_request (mdsc,CEPH_MDS_OP_GETVXATTR , mode); +if (IS_ERR (req)) { + err = -ENOMEM ; + goto out; + } + + req->r_path2 =kstrdup (name,GFP_NOFS ); +if (!req->r_path2) { + err = -ENOMEM ; + goto put; + } + +ihold (inode); + req->r_inode = inode; + err =ceph_mdsc_do_request (mdsc,NULL , req); +if (err <0 ) + goto put; + + xattr_value = req->r_reply_info.xattr_info.xattr_value; + xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len; + +dout ("do_getvxattr xattr_value_len:%lu, size:%lu\n" , xattr_value_len, size); + + err = xattr_value_len; +if (size ==0 ) + goto put; + +if (xattr_value_len > size) { + err = -ERANGE ; + goto put; + } + +memcpy (value, xattr_value, xattr_value_len); +put: +ceph_mdsc_put_request (req); +out: +dout ("do_getvxattr result=%d\n" , err); +return err; +} +/* * Check inode permissions. We verify we have a valid value for
fs/ceph/mds_client.c +26 -1
@@ -555,6 +555,29 @@ static int parse_reply_info_create(void **p, void *end,return -EIO ; } +static int parse_reply_info_getvxattr (void **p,void *end, +struct ceph_mds_reply_info_parsed *info, +u64 features) +{ +u8 struct_v, struct_compat; +u32 struct_len; +u32 value_len; + +ceph_decode_8_safe (p, end, struct_v, bad); +ceph_decode_8_safe (p, end, struct_compat, bad); +ceph_decode_32_safe (p, end, struct_len, bad); +ceph_decode_32_safe (p, end, value_len, bad); + +if (value_len == end - *p) { + info->xattr_info.xattr_value = *p; + info->xattr_info.xattr_value_len = end - *p; + *p = end; +return info->xattr_info.xattr_value_len; + } +bad: +return -EIO ; +} +/* * parse extra results */
@@ -570,6 +593,8 @@ static int parse_reply_info_extra(void **p, void *end,return parse_reply_info_readdir (p, end, info, features); else if (op == CEPH_MDS_OP_CREATE) return parse_reply_info_create(p, end, info, features, s); + else if (op == CEPH_MDS_OP_GETVXATTR) + return parse_reply_info_getvxattr(p, end, info, features); else return -EIO; }
@@ -615,7 +640,7 @@ static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,if (p != end) goto bad; -return 0 ; +return err; bad: err = -EIO ;
fs/ceph/mds_client.h +11 -1
@@ -29,8 +29,10 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT ,CEPHFS_FEATURE_DELEG_INO ,CEPHFS_FEATURE_METRIC_COLLECT , +CEPHFS_FEATURE_ALTERNATE_NAME , +CEPHFS_FEATURE_GETVXATTR , - CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT, + CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_GETVXATTR, };/*
@@ -45,6 +47,8 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT , \CEPHFS_FEATURE_DELEG_INO , \CEPHFS_FEATURE_METRIC_COLLECT , \ +CEPHFS_FEATURE_ALTERNATE_NAME , \ +CEPHFS_FEATURE_GETVXATTR , \ \ CEPHFS_FEATURE_MAX, \ }
@@ -100,6 +104,11 @@ struct ceph_mds_reply_dir_entry {loff_t offset; }; +struct ceph_mds_reply_xattr { +char *xattr_value; +size_t xattr_value_len; +}; +/* * parsed info about an mds reply, including information about * either: 1) the target inode and/or its parent directory and dentry,
@@ -115,6 +124,7 @@ struct ceph_mds_reply_info_parsed {char *dname;u32 dname_len;struct ceph_mds_reply_lease *dlease; +struct ceph_mds_reply_xattr xattr_info;/* extra */ union {
fs/ceph/strings.c +1 -0
@@ -60,6 +60,7 @@ const char *ceph_mds_op_name(int op)case CEPH_MDS_OP_LOOKUPINO :return "lookupino" ;case CEPH_MDS_OP_LOOKUPNAME :return "lookupname" ;case CEPH_MDS_OP_GETATTR :return "getattr" ; +case CEPH_MDS_OP_GETVXATTR :return "getvxattr" ;case CEPH_MDS_OP_SETXATTR :return "setxattr" ;case CEPH_MDS_OP_SETATTR :return "setattr" ;case CEPH_MDS_OP_RMXATTR :return "rmxattr" ;
fs/ceph/super.h +1 -0
@@ -1043,6 +1043,7 @@ static inline bool ceph_inode_is_shutdown(struct inode *inode)/* xattr.c */ int __ceph_setxattr (struct inode *,const char *,const void *,size_t ,int ); +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value,size_t size);ssize_t __ceph_getxattr (struct inode *,const char *,void *,size_t );extern ssize_t ceph_listxattr (struct dentry *,char *,size_t );extern struct ceph_buffer *__ceph_build_xattrs_blob (struct ceph_inode_info *ci);
fs/ceph/xattr.c +65 -0
77 lines not on this page · raw message
include/linux/ceph/ceph_fs.h +1 -0
@@ -328,6 +328,7 @@ enum {CEPH_MDS_OP_LOOKUPPARENT =0x00103 ,CEPH_MDS_OP_LOOKUPINO =0x00104 ,CEPH_MDS_OP_LOOKUPNAME =0x00105 , +CEPH_MDS_OP_GETVXATTR =0x00106 ,CEPH_MDS_OP_SETXATTR =0x01105 , CEPH_MDS_OP_RMXATTR = 0x01106,
Signature · 1 line
2.34.1
Re: [PATCH 1/1] ceph: add getvxattr op
Quoted 107 lines · Milind Changire
Problem:
Directory vxattrs like ceph.dir.pin* and ceph.dir.layout* may not be propagated to the client as frequently to keep them updated. This creates vxattr availability problems.Solution:
Adds new getvxattr op to fetch ceph.dir.pin*, ceph.dir.layout* and ceph.file.layout* vxattrs.
If the entire layout for a dir or a file is being set, then it is expected that the layout be set in standard JSON format. Individual field value retrieval is not wrapped in JSON. The JSON format also applies while setting the vxattr if the entire layout is being set in one go.
As a temporary measure, setting a vxattr can also be done in the old format. The old format will be deprecated in the future.URL: https://tracker.ceph.com/issues/51062
Signed-off-by Milind Changire <[email protected]>--- fs/ceph/inode.c | 51 ++++++++++++++++++++++++++++ fs/ceph/mds_client.c | 27 ++++++++++++++- fs/ceph/mds_client.h | 12 ++++++- fs/ceph/strings.c | 1 + fs/ceph/super.h | 1 + fs/ceph/xattr.c | 65 ++++++++++++++++++++++++++++++++++++ include/linux/ceph/ceph_fs.h | 1 + 7 files changed, 156 insertions(+), 2 deletions(-)fs/ceph/inode.c +51
-0@@ -2291,6 +2291,57 @@ int __ceph_do_getattr(struct inode *inode, struct page *locked_page,return err; } +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value, +size_t size) +{ +struct ceph_fs_client *fsc =ceph_sb_to_client (inode->i_sb); +struct ceph_mds_client *mdsc = fsc->mdsc; +struct ceph_mds_request *req; +int mode =USE_AUTH_MDS ; +int err; +char *xattr_value; +size_t xattr_value_len; + + req =ceph_mdsc_create_request (mdsc,CEPH_MDS_OP_GETVXATTR , mode); +if (IS_ERR (req)) { + err = -ENOMEM ; + goto out; + } + + req->r_path2 =kstrdup (name,GFP_NOFS ); +if (!req->r_path2) { + err = -ENOMEM ; + goto put; + } + +ihold (inode); + req->r_inode = inode; + err =ceph_mdsc_do_request (mdsc,NULL , req); +if (err <0 ) + goto put; + + xattr_value = req->r_reply_info.xattr_info.xattr_value; + xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len; + +dout ("do_getvxattr xattr_value_len:%lu, size:%lu\n" , xattr_value_len, size); + + err = xattr_value_len; +if (size ==0 ) + goto put; + +if (xattr_value_len > size) { + err = -ERANGE ; + goto put; + } + +memcpy (value, xattr_value, xattr_value_len); +put: +ceph_mdsc_put_request (req); +out: +dout ("do_getvxattr result=%d\n" , err); +return err; +} +/* * Check inode permissions. We verify we have a valid value for fs/ceph/mds_client.c +9
-0@@ -555,6 +555,29 @@ static int parse_reply_info_create(void **p, void *end,return -EIO ; } +static int parse_reply_info_getvxattr (void **p,void *end, +struct ceph_mds_reply_info_parsed *info, +u64 features) +{ +u8 struct_v, struct_compat; +u32 struct_len; +u32 value_len; + +ceph_decode_8_safe (p, end, struct_v, bad);
fs/ceph/mds_client.c static int parse_reply_info_create(void **p, void *end, · line 576
Quoted 10 lines
fs/ceph/mds_client.c +10
-0+ceph_decode_8_safe (p, end, struct_compat, bad); +ceph_decode_32_safe (p, end, struct_len, bad); +ceph_decode_32_safe (p, end, value_len, bad); + +if (value_len == end - *p) { + info->xattr_info.xattr_value = *p; + info->xattr_info.xattr_value_len = end - *p; + *p = end; +return info->xattr_info.xattr_value_len; + }
The kernel uses tab indent almost exclusively. Can you fix the indention above (and anywhere else I missed)?
Quoted 125 lines
fs/ceph/mds_client.c +7
-1+bad: +return -EIO ; +} +/* * parse extra results */ @@ -570,6 +593,8 @@ static int parse_reply_info_extra(void **p, void *end,return parse_reply_info_readdir (p, end, info, features); else if (op == CEPH_MDS_OP_CREATE) return parse_reply_info_create(p, end, info, features, s); + else if (op == CEPH_MDS_OP_GETVXATTR) + return parse_reply_info_getvxattr(p, end, info, features); else return -EIO; }@@ -615,7 +640,7 @@ static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,if (p != end) goto bad; -return 0 ; +return err; bad: err = -EIO ;fs/ceph/mds_client.h +11
-1@@ -29,8 +29,10 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT ,CEPHFS_FEATURE_DELEG_INO ,CEPHFS_FEATURE_METRIC_COLLECT , +CEPHFS_FEATURE_ALTERNATE_NAME , +CEPHFS_FEATURE_GETVXATTR , - CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT, + CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_GETVXATTR, };/* @@ -45,6 +47,8 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT , \CEPHFS_FEATURE_DELEG_INO , \CEPHFS_FEATURE_METRIC_COLLECT , \ +CEPHFS_FEATURE_ALTERNATE_NAME , \ +CEPHFS_FEATURE_GETVXATTR , \ \ CEPHFS_FEATURE_MAX, \ }@@ -100,6 +104,11 @@ struct ceph_mds_reply_dir_entry {loff_t offset; }; +struct ceph_mds_reply_xattr { +char *xattr_value; +size_t xattr_value_len; +}; +/* * parsed info about an mds reply, including information about * either: 1) the target inode and/or its parent directory and dentry, @@ -115,6 +124,7 @@ struct ceph_mds_reply_info_parsed {char *dname;u32 dname_len;struct ceph_mds_reply_lease *dlease; +struct ceph_mds_reply_xattr xattr_info;/* extra */ union {fs/ceph/strings.c +1
-0@@ -60,6 +60,7 @@ const char *ceph_mds_op_name(int op)case CEPH_MDS_OP_LOOKUPINO :return "lookupino" ;case CEPH_MDS_OP_LOOKUPNAME :return "lookupname" ;case CEPH_MDS_OP_GETATTR :return "getattr" ; +case CEPH_MDS_OP_GETVXATTR :return "getvxattr" ;case CEPH_MDS_OP_SETXATTR :return "setxattr" ;case CEPH_MDS_OP_SETATTR :return "setattr" ;case CEPH_MDS_OP_RMXATTR :return "rmxattr" ;fs/ceph/super.h +1
-0@@ -1043,6 +1043,7 @@ static inline bool ceph_inode_is_shutdown(struct inode *inode)/* xattr.c */ int __ceph_setxattr (struct inode *,const char *,const void *,size_t ,int ); +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value,size_t size);ssize_t __ceph_getxattr (struct inode *,const char *,void *,size_t );extern ssize_t ceph_listxattr (struct dentry *,char *,size_t );extern struct ceph_buffer *__ceph_build_xattrs_blob (struct ceph_inode_info *ci);fs/ceph/xattr.c +23
-0@@ -918,6 +918,64 @@ static inline int __get_request_mask(struct inode *in) {return mask; } +static inline bool ceph_is_passthrough_vxattr (const char *name) +{ +const char *vxattr_list[] = { +"ceph.dir.pin" , +"ceph.dir.pin.random" , +"ceph.dir.pin.distributed" , +"ceph.dir.layout" , +"ceph.dir.layout.object_size" , +"ceph.dir.layout.stripe_count" , +"ceph.dir.layout.stripe_unit" , +"ceph.dir.layout.pool" , +"ceph.dir.layout.pool_name" , +"ceph.dir.layout.pool_id" , +"ceph.dir.layout.pool_namespace" , +"ceph.file.layout" , +"ceph.file.layout.object_size" , +"ceph.file.layout.stripe_count" , +"ceph.file.layout.stripe_unit" , +"ceph.file.layout.pool" , +"ceph.file.layout.pool_name" , +"ceph.file.layout.pool_id" , +"ceph.file.layout.pool_namespace" , + };
fs/ceph/xattr.c static inline int __get_request_mask(struct inode *in) { · line 953
Quoted 10 lines
fs/ceph/xattr.c +10
-0+int i =0 ; +int n =sizeof (vxattr_list)/sizeof (vxattr_list[0 ]); + +while (i < n) { +if (!strcmp (name, vxattr_list[i])) +return true; + i++; + } +return false; +}
This part, I'm not so crazy about. This list will be hard to keep in sync with the userland code and you're almost guaranteed to have a mismatch between what the kernel and userland supports since they're on different release schedules.
I think what we probably ought to do is run the ceph_match_vxattr call first, and if it doesn't match the name and the name starts with "ceph.", then we'll send a GETVXATTR call to the MDS.
Sound ok?
Quoted 54 lines
fs/ceph/xattr.c +32
-0+ +/* check if the entire cluster supports the given feature */ +static inline bool ceph_cluster_has_feature (struct inode *inode,int feature_bit) +{ +int64_t i; +struct ceph_fs_client *fsc =ceph_inode_to_client (inode); +struct ceph_mds_session **sessions = fsc->mdsc->sessions; +int64_t num_sessions =atomic_read (&fsc->mdsc->num_sessions); + +if (fsc->mdsc->stopping) +return false; + +if (!sessions) +return false; + +for (i =0 ; i < num_sessions; i++) { +struct ceph_mds_session *session = sessions[i]; +if (!session) +return false; +if (!test_bit (feature_bit, &session->s_features)) +return false; + } +return true; +} +ssize_t __ceph_getxattr (struct inode *inode,const char *name,void *value,size_t size) {@@ -927,6 +985,13 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,int req_mask;ssize_t err; +if (ceph_is_passthrough_vxattr (name) && +ceph_cluster_has_feature (inode,CEPHFS_FEATURE_GETVXATTR )) { + err =ceph_do_getvxattr (inode, name, value, size); +return err; + } +dout ("vxattr is not passthrough or kclient doesn't support GETVXATTR\n" ); +/* let's see if a virtual xattr was requested */ vxattr =ceph_match_vxattr (inode, name);if (vxattr) {include/linux/ceph/ceph_fs.h +1
-0@@ -328,6 +328,7 @@ enum {CEPH_MDS_OP_LOOKUPPARENT =0x00103 ,CEPH_MDS_OP_LOOKUPINO =0x00104 ,CEPH_MDS_OP_LOOKUPNAME =0x00105 , +CEPH_MDS_OP_GETVXATTR =0x00106 ,CEPH_MDS_OP_SETXATTR =0x01105 , CEPH_MDS_OP_RMXATTR = 0x01106,
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH 1/1] ceph: add getvxattr op
Quoted 114 lines · Jeff Layton
Quoted 112 lines · Milind Changire
Problem:
Directory vxattrs like ceph.dir.pin* and ceph.dir.layout* may not be propagated to the client as frequently to keep them updated. This creates vxattr availability problems.Solution:
Adds new getvxattr op to fetch ceph.dir.pin*, ceph.dir.layout* and ceph.file.layout* vxattrs.
If the entire layout for a dir or a file is being set, then it is expected that the layout be set in standard JSON format. Individual field value retrieval is not wrapped in JSON. The JSON format also applies while setting the vxattr if the entire layout is being set in one go.
As a temporary measure, setting a vxattr can also be done in the old format. The old format will be deprecated in the future.URL: https://tracker.ceph.com/issues/51062
Signed-off-by Milind Changire <[email protected]>--- fs/ceph/inode.c | 51 ++++++++++++++++++++++++++++ fs/ceph/mds_client.c | 27 ++++++++++++++- fs/ceph/mds_client.h | 12 ++++++- fs/ceph/strings.c | 1 + fs/ceph/super.h | 1 + fs/ceph/xattr.c | 65 ++++++++++++++++++++++++++++++++++++ include/linux/ceph/ceph_fs.h | 1 + 7 files changed, 156 insertions(+), 2 deletions(-)fs/ceph/inode.c +51
-0@@ -2291,6 +2291,57 @@ int __ceph_do_getattr(struct inode *inode, struct page *locked_page,return err; } +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value, +size_t size) +{ +struct ceph_fs_client *fsc =ceph_sb_to_client (inode->i_sb); +struct ceph_mds_client *mdsc = fsc->mdsc; +struct ceph_mds_request *req; +int mode =USE_AUTH_MDS ; +int err; +char *xattr_value; +size_t xattr_value_len; + + req =ceph_mdsc_create_request (mdsc,CEPH_MDS_OP_GETVXATTR , mode); +if (IS_ERR (req)) { + err = -ENOMEM ; + goto out; + } + + req->r_path2 =kstrdup (name,GFP_NOFS ); +if (!req->r_path2) { + err = -ENOMEM ; + goto put; + } + +ihold (inode); + req->r_inode = inode; + err =ceph_mdsc_do_request (mdsc,NULL , req); +if (err <0 ) + goto put; + + xattr_value = req->r_reply_info.xattr_info.xattr_value; + xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len; + +dout ("do_getvxattr xattr_value_len:%lu, size:%lu\n" , xattr_value_len, size); + + err = xattr_value_len; +if (size ==0 ) + goto put; + +if (xattr_value_len > size) { + err = -ERANGE ; + goto put; + } + +memcpy (value, xattr_value, xattr_value_len); +put: +ceph_mdsc_put_request (req); +out: +dout ("do_getvxattr result=%d\n" , err); +return err; +} +/* * Check inode permissions. We verify we have a valid value for fs/ceph/mds_client.c +14
-0@@ -555,6 +555,29 @@ static int parse_reply_info_create(void **p, void *end,return -EIO ; } +static int parse_reply_info_getvxattr (void **p,void *end, +struct ceph_mds_reply_info_parsed *info, +u64 features) +{ +u8 struct_v, struct_compat; +u32 struct_len; +u32 value_len; + +ceph_decode_8_safe (p, end, struct_v, bad); +ceph_decode_8_safe (p, end, struct_compat, bad); +ceph_decode_32_safe (p, end, struct_len, bad); +ceph_decode_32_safe (p, end, value_len, bad); + +if (value_len == end - *p) {
Quoted 8 lines
fs/ceph/mds_client.c static int parse_reply_info_create(void **p, void *end, · line 576
Quoted 5 lines
fs/ceph/mds_client.c +5
-0+ info->xattr_info.xattr_value = *p; + info->xattr_info.xattr_value_len = end - *p; + *p = end; +return info->xattr_info.xattr_value_len; + }The kernel uses tab indent almost exclusively. Can you fix the indention above (and anywhere else I missed)?
My bad; newbie issues
I'll see how can I fix this.
Quoted 142 lines
Quoted 126 lines
fs/ceph/mds_client.c +7
-1+bad: +return -EIO ; +} +/* * parse extra results */ @@ -570,6 +593,8 @@ static int parse_reply_info_extra(void **p, void *end,return parse_reply_info_readdir (p, end, info, features); else if (op == CEPH_MDS_OP_CREATE) return parse_reply_info_create(p, end, info, features, s); + else if (op == CEPH_MDS_OP_GETVXATTR) + return parse_reply_info_getvxattr(p, end, info, features); else return -EIO; }@@ -615,7 +640,7 @@ static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,if (p != end) goto bad; -return 0 ; +return err; bad: err = -EIO ;fs/ceph/mds_client.h +11
-1@@ -29,8 +29,10 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT ,CEPHFS_FEATURE_DELEG_INO ,CEPHFS_FEATURE_METRIC_COLLECT , +CEPHFS_FEATURE_ALTERNATE_NAME , +CEPHFS_FEATURE_GETVXATTR , - CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT, + CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_GETVXATTR, };/* @@ -45,6 +47,8 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT , \CEPHFS_FEATURE_DELEG_INO , \CEPHFS_FEATURE_METRIC_COLLECT , \ +CEPHFS_FEATURE_ALTERNATE_NAME , \ +CEPHFS_FEATURE_GETVXATTR , \ \ CEPHFS_FEATURE_MAX, \ }@@ -100,6 +104,11 @@ struct ceph_mds_reply_dir_entry {loff_t offset; }; +struct ceph_mds_reply_xattr { +char *xattr_value; +size_t xattr_value_len; +}; +/* * parsed info about an mds reply, including information about * either: 1) the target inode and/or its parent directory and dentry, @@ -115,6 +124,7 @@ struct ceph_mds_reply_info_parsed {char *dname;u32 dname_len;struct ceph_mds_reply_lease *dlease; +struct ceph_mds_reply_xattr xattr_info;/* extra */ union {fs/ceph/strings.c +1
-0@@ -60,6 +60,7 @@ const char *ceph_mds_op_name(int op)case CEPH_MDS_OP_LOOKUPINO :return "lookupino" ;case CEPH_MDS_OP_LOOKUPNAME :return "lookupname" ;case CEPH_MDS_OP_GETATTR :return "getattr" ; +case CEPH_MDS_OP_GETVXATTR :return "getvxattr" ;case CEPH_MDS_OP_SETXATTR :return "setxattr" ;case CEPH_MDS_OP_SETATTR :return "setattr" ;case CEPH_MDS_OP_RMXATTR :return "rmxattr" ;fs/ceph/super.h +1
-0@@ -1043,6 +1043,7 @@ static inline bool ceph_inode_is_shutdown(struct inode *inode)/* xattr.c */ int __ceph_setxattr (struct inode *,const char *,const void *,size_t ,int ); +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value,size_t size);ssize_t __ceph_getxattr (struct inode *,const char *,void *,size_t );extern ssize_t ceph_listxattr (struct dentry *,char *,size_t );extern struct ceph_buffer *__ceph_build_xattrs_blob (struct ceph_inode_info *ci);fs/ceph/xattr.c +23
-0@@ -918,6 +918,64 @@ static inline int __get_request_mask(struct inode *in) {return mask; } +static inline bool ceph_is_passthrough_vxattr (const char *name) +{ +const char *vxattr_list[] = { +"ceph.dir.pin" , +"ceph.dir.pin.random" , +"ceph.dir.pin.distributed" , +"ceph.dir.layout" , +"ceph.dir.layout.object_size" , +"ceph.dir.layout.stripe_count" , +"ceph.dir.layout.stripe_unit" , +"ceph.dir.layout.pool" , +"ceph.dir.layout.pool_name" , +"ceph.dir.layout.pool_id" , +"ceph.dir.layout.pool_namespace" , +"ceph.file.layout" , +"ceph.file.layout.object_size" , +"ceph.file.layout.stripe_count" , +"ceph.file.layout.stripe_unit" , +"ceph.file.layout.pool" , +"ceph.file.layout.pool_name" , +"ceph.file.layout.pool_id" , +"ceph.file.layout.pool_namespace" , + };fs/ceph/xattr.c static inline int __get_request_mask(struct inode *in) { · line 953
Quoted 10 lines
fs/ceph/xattr.c +10
-0+int i =0 ; +int n =sizeof (vxattr_list)/sizeof (vxattr_list[0 ]); + +while (i < n) { +if (!strcmp (name, vxattr_list[i])) +return true; + i++; + } +return false; +}This part, I'm not so crazy about. This list will be hard to keep in sync with the userland code and you're almost guaranteed to have a mismatch between what the kernel and userland supports since they're on different release schedules.
Quoted 3 lines
I think what we probably ought to do is run the
ceph_match_vxattrcall first, and if it doesn't match the name and the name starts with "ceph.", then we'll send a GETVXATTR call to the MDS.
I was trying to preempt a getvxattr call if the full xattr name was invalid. I could just test for the prefixes: ceph.dir.pin, ceph.dir.layout and ceph.file.layout and send them over. Eventually, we'll get an ENODATA if xattr name validation fails on the MDS side.
The problem with doing a ceph_match_vxattr() first is that it will succeed for any of these vxattrs, but we'll land in the same place as the one you've described in the tracker.
The very idea of implementing getvxattr is to bypass the existing "cached" service mechanism for special vxattrs.
Quoted 60 lines
Sound ok?
Quoted 47 lines
fs/ceph/xattr.c +32
-0+ +/* check if the entire cluster supports the given feature */ +static inline bool ceph_cluster_has_feature (struct inode *inode,int feature_bit) +{ +int64_t i; +struct ceph_fs_client *fsc =ceph_inode_to_client (inode); +struct ceph_mds_session **sessions = fsc->mdsc->sessions; +int64_t num_sessions =atomic_read (&fsc->mdsc->num_sessions); + +if (fsc->mdsc->stopping) +return false; + +if (!sessions) +return false; + +for (i =0 ; i < num_sessions; i++) { +struct ceph_mds_session *session = sessions[i]; +if (!session) +return false; +if (!test_bit (feature_bit, &session->s_features)) +return false; + } +return true; +} +ssize_t __ceph_getxattr (struct inode *inode,const char *name,void *value,size_t size) {@@ -927,6 +985,13 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,int req_mask;ssize_t err; +if (ceph_is_passthrough_vxattr (name) && +ceph_cluster_has_feature (inode,CEPHFS_FEATURE_GETVXATTR )) { + err =ceph_do_getvxattr (inode, name, value, size); +return err; + } +dout ("vxattr is not passthrough or kclient doesn't support GETVXATTR\n" ); +/* let's see if a virtual xattr was requested */ vxattr =ceph_match_vxattr (inode, name);if (vxattr) {include/linux/ceph/ceph_fs.h +0
-0include/linux/ceph/ceph_fs.h enum { · line 334
Quoted 7 lines
include/linux/ceph/ceph_fs.h +1
-0@@ -328,6 +328,7 @@ enum {CEPH_MDS_OP_LOOKUPPARENT =0x00103 ,CEPH_MDS_OP_LOOKUPINO =0x00104 ,CEPH_MDS_OP_LOOKUPNAME =0x00105 , +CEPH_MDS_OP_GETVXATTR =0x00106 ,CEPH_MDS_OP_SETXATTR =0x01105 , CEPH_MDS_OP_RMXATTR = 0x01106,--
Jeff Layton <[email protected]>
A message that is not in the archive (sent off-list or not yet received): CANmksPRpzAEB70xGVmwKCgwv55bXYwuAruDcdu7ovc4suQqUZA@mail.gmail.com
Re: [PATCH 1/1] ceph: add getvxattr op
Quoted 303 lines · Milind Changire
responses below ...
Quoted 126 lines · Jeff Layton
Quoted 125 lines · Milind Changire
Problem:
Directory vxattrs like ceph.dir.pin* and ceph.dir.layout* may not be
propagated to the client as frequently to keep them updated. This creates vxattr availability problems.Solution:
Adds new getvxattr op to fetch ceph.dir.pin*, ceph.dir.layout* and ceph.file.layout* vxattrs.
If the entire layout for a dir or a file is being set, then it is expected that the layout be set in standard JSON format. Individual
field value retrieval is not wrapped in JSON. The JSON format also applies while setting the vxattr if the entire layout is being set in
one go.
As a temporary measure, setting a vxattr can also be done in the old
format. The old format will be deprecated in the future.URL: https://tracker.ceph.com/issues/51062
Signed-off-by Milind Changire <[email protected]>---
fs/ceph/inode.c | 51 ++++++++++++++++++++++++++++ fs/ceph/mds_client.c | 27 ++++++++++++++- fs/ceph/mds_client.h | 12 ++++++- fs/ceph/strings.c | 1 +
fs/ceph/super.h | 1 +
fs/ceph/xattr.c | 65
++++++++++++++++++++++++++++++++++++
include/linux/ceph/ceph_fs.h| 1 +7 files changed, 156 insertions(+), 2 deletions(-)fs/ceph/inode.c +0
-0@@ -2291,6 +2291,57 @@ int __ceph_do_getattr(struct inode *inode,struct page *locked_page, return err; }
+int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
+ size_t size)fs/ceph/inode.c +10
-0+{ + struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); + struct ceph_mds_client *mdsc = fsc->mdsc; + struct ceph_mds_request *req; + int mode = USE_AUTH_MDS; + int err; + char *xattr_value; + size_t xattr_value_len; + + req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR,mode);
+ if (IS_ERR(req)) {
+ err =-ENOMEM;
+ goto out;
+ }fs/ceph/inode.c +15
-0+ + req->r_path2 = kstrdup(name, GFP_NOFS); + if (!req->r_path2) { + err = -ENOMEM; + goto put; + } + + ihold(inode); + req->r_inode = inode; + err = ceph_mdsc_do_request(mdsc, NULL, req); + if (err < 0) + goto put; + + xattr_value = req->r_reply_info.xattr_info.xattr_value; + xattr_value_len = req-Quoted 1 line
r_reply_info.xattr_info.xattr_value_len;
+
+ dout("do_getvxattr xattr_value_len:%lu, size:%lu\n",xattr_value_len, size);fs/ceph/inode.c +18
-0+ + err = xattr_value_len; + if (size == 0) + goto put; + + if (xattr_value_len > size) { + err = -ERANGE; + goto put; + } + + memcpy(value, xattr_value, xattr_value_len); +put: + ceph_mdsc_put_request(req); +out: + dout("do_getvxattr result=%d\n", err); + return err; +} +
/*
* Check inode permissions. We verify we have a valid value forfs/ceph/mds_client.c +0
-0@@ -555,6 +555,29 @@ static int parse_reply_info_create(void **p,void *end, return -EIO; }
+static int parse_reply_info_getvxattr(void **p, void *end, + structceph_mds_reply_info_parsed*info,
+ u64 features)fs/ceph/mds_client.c +12
-0+{ + u8 struct_v, struct_compat; + u32 struct_len; + u32 value_len; + + ceph_decode_8_safe(p, end, struct_v, bad); + ceph_decode_8_safe(p, end, struct_compat, bad); + ceph_decode_32_safe(p, end, struct_len, bad); + ceph_decode_32_safe(p, end, value_len, bad); + + if (value_len == end - *p) { + info->xattr_info.xattr_value = *p;Quoted 9 lines
Quoted 4 lines
+ info->xattr_info.xattr_value_len = end - *p; + *p = end;
+ returninfo->xattr_info.xattr_value_len; + }The kernel uses tab indent almost exclusively. Can you fix the indention
above (and anywhere else I missed)?I'll fix this. newbie issues :P
Quoted 160 lines
Quoted 138 lines
fs/ceph/mds_client.c +4
-0+bad: + return -EIO; +} +/*
* parse extra results
*/fs/ceph/mds_client.c +2
-0@@ -570,6 +593,8 @@ static int parse_reply_info_extra(void **p,void *end, return parse_reply_info_readdir(p, end, info,features ); else if (op == CEPH_MDS_OP_CREATE) return parse_reply_info_create(p, end, info, features, s); + else if (op == CEPH_MDS_OP_GETVXATTR) + return parse_reply_info_getvxattr(p, end, info,features);
else
return-EIO;
}fs/ceph/mds_client.c +0
-0@@ -615,7 +640,7 @@ static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,
if (p != end)
goto bad;
- return 0;
+ return err;
bad:
err =-EIO;fs/ceph/mds_client.h +2
-0@@ -29,8 +29,10 @@ enum ceph_feature_type { CEPHFS_FEATURE_MULTI_RECONNECT, CEPHFS_FEATURE_DELEG_INO, CEPHFS_FEATURE_METRIC_COLLECT, + CEPHFS_FEATURE_ALTERNATE_NAME, + CEPHFS_FEATURE_GETVXATTR,
- CEPHFS_FEATURE_MAX =CEPHFS_FEATURE_METRIC_COLLECT, + CEPHFS_FEATURE_MAX =CEPHFS_FEATURE_GETVXATTR, };
/*fs/ceph/mds_client.h +2
-0@@ -45,6 +47,8 @@ enum ceph_feature_type { CEPHFS_FEATURE_MULTI_RECONNECT, \ CEPHFS_FEATURE_DELEG_INO, \ CEPHFS_FEATURE_METRIC_COLLECT, \ + CEPHFS_FEATURE_ALTERNATE_NAME, \ + CEPHFS_FEATURE_GETVXATTR, \ \ CEPHFS_FEATURE_MAX, \ }@@ -100,6 +104,11 @@ struct ceph_mds_reply_dir_entry { loff_t offset; };fs/ceph/mds_client.h +5
-0+struct ceph_mds_reply_xattr { + char *xattr_value; + size_t xattr_value_len; +}; +/*
* parsed info about an mds reply, including information about
* either: 1) the target inode and/or its parent directory and dentry,fs/ceph/mds_client.h +1
-0@@ -115,6 +124,7 @@ struct ceph_mds_reply_info_parsed { char *dname; u32 dname_len; struct ceph_mds_reply_lease *dlease; + struct ceph_mds_reply_xattr xattr_info;
/* extra */
union {fs/ceph/strings.c +1
-0@@ -60,6 +60,7 @@ const char *ceph_mds_op_name(int op) case CEPH_MDS_OP_LOOKUPINO: return "lookupino"; case CEPH_MDS_OP_LOOKUPNAME: return "lookupname"; case CEPH_MDS_OP_GETATTR: return "getattr"; + case CEPH_MDS_OP_GETVXATTR: return "getvxattr"; case CEPH_MDS_OP_SETXATTR: return "setxattr"; case CEPH_MDS_OP_SETATTR: return "setattr"; case CEPH_MDS_OP_RMXATTR: return "rmxattr";fs/ceph/super.h +0
-0@@ -1043,6 +1043,7 @@ static inline bool ceph_inode_is_shutdown(struct inode *inode)
/*xattr.c*/
int__ceph_setxattr(struct inode *, const char *, const void *, size_t, int);
+intceph_do_getvxattr(struct inode *inode, const char *name, void *value, size_t size);
ssize_t__ceph_getxattr(struct inode *, const char *, void *, size_t);
externssize_tceph_listxattr(struct dentry *, char *, size_t); extern structceph_buffer*__ceph_build_xattrs_blob(structceph_inode_info*ci);fs/ceph/xattr.c +0
-0@@ -918,6 +918,64 @@ static inline int __get_request_mask(struct inode *in) { return mask; }fs/ceph/xattr.c +23
-0+static inline bool ceph_is_passthrough_vxattr (const char *name) +{ + const char *vxattr_list[] = { + "ceph.dir.pin", + "ceph.dir.pin.random", + "ceph.dir.pin.distributed", + "ceph.dir.layout", + "ceph.dir.layout.object_size", + "ceph.dir.layout.stripe_count", + "ceph.dir.layout.stripe_unit", + "ceph.dir.layout.pool", + "ceph.dir.layout.pool_name", + "ceph.dir.layout.pool_id", + "ceph.dir.layout.pool_namespace", + "ceph.file.layout", + "ceph.file.layout.object_size", + "ceph.file.layout.stripe_count", + "ceph.file.layout.stripe_unit", + "ceph.file.layout.pool", + "ceph.file.layout.pool_name", + "ceph.file.layout.pool_id", + "ceph.file.layout.pool_namespace", + };fs/ceph/xattr.c static inline int __get_request_mask(struct inode *in) { · line 953
Quoted 10 lines
fs/ceph/xattr.c +10
-0+ int i = 0; + int n = sizeof(vxattr_list)/sizeof(vxattr_list[0]); + + while (i < n) { + if (!strcmp(name, vxattr_list[i])) + return true; + i++; + } + return false; +}This part, I'm not so crazy about. This list will be hard to keep in sync with the userland code and you're almost guaranteed to have a mismatch between what the kernel and userland supports since they're on
different release schedules.I think what we probably ought to do is run the
ceph_match_vxattrcall
first, and if it doesn't match the name and the name starts with "ceph.", then we'll send a GETVXATTR call to the MDS.
Quoted 6 lines
Quoted 2 lines
Sound ok?
I was trying to preempt a getvxattr call if the full xattr name was invalid.
Huh, why? That seems like the point of the whole bug. We need a way to hand vxattrs that the client doesn't recognize off to the MDS so it can satisfy the request.
Quoted 4 lines
I could just test for the prefixes: ceph.dir.pin, ceph.dir.layout and ceph.file.layout and send them over. Eventually, we'll get an ENODATA if xattr name validation fails on the MDS side.
Quoted 9 lines
The problem with doing a
ceph_match_vxattr()first is that it will succeed
for any of these vxattrs, but we'll land in the same place as the one you've
described in the tracker.
The very idea of implementing getvxattr is to bypass the existing "cached"
service mechanism for special vxattrs.
I don't get it. What's the benefit of calling out to the MDS to ask for ceph.file.layout.pool instead of just satisfying it from the info we have?
I'd prefer we just add a fallback for "ceph.*" vxattrs that for which we don't have a local handler.
Quoted 63 lines
Quoted 63 lines
Quoted 62 lines
fs/ceph/xattr.c +3
-0+ +/* check if the entire cluster supports the given feature */ +static inline bool ceph_cluster_has_feature(struct inode *inode,int
feature_bit)
+{
+ int64_t i;
+ structceph_fs_client*fsc =ceph_inode_to_client(inode); + structceph_mds_session**sessions =fsc->mdsc->sessions; + int64_tnum_sessions= atomic_read(&fsc->mdsc-Quoted 1 line
num_sessions);fs/ceph/xattr.c +17
-0+ + if (fsc->mdsc->stopping) + return false; + + if (!sessions) + return false; + + for (i = 0; i < num_sessions; i++) { + struct ceph_mds_session *session = sessions[i]; + if (!session) + return false; + if (!test_bit(feature_bit, &session->s_features)) + return false; + } + return true; +} +
ssize_t__ceph_getxattr(struct inode *inode, const char *name, void *value,
size_t size)
{fs/ceph/xattr.c +0
-0@@ -927,6 +985,13 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value, int req_mask; ssize_t err;
+ if (ceph_is_passthrough_vxattr(name)&&
+ ceph_cluster_has_feature(inode,CEPHFS_FEATURE_GETVXATTR)) {
+ err =ceph_do_getvxattr(inode, name, value, size); + return err;
+ }
+ dout("vxattr is not passthrough or kclient doesn't support GETVXATTR\n");
+
/* let's see if a virtual xattr was requested */ vxattr =ceph_match_vxattr(inode, name);
if (vxattr) {include/linux/ceph/ceph_fs.h +1
-0@@ -328,6 +328,7 @@ enum { CEPH_MDS_OP_LOOKUPPARENT = 0x00103, CEPH_MDS_OP_LOOKUPINO = 0x00104, CEPH_MDS_OP_LOOKUPNAME = 0x00105, + CEPH_MDS_OP_GETVXATTR = 0x00106,
CEPH_MDS_OP_SETXATTR =0x01105,
CEPH_MDS_OP_RMXATTR =0x01106,
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH 1/1] ceph: add getvxattr op
Quoted 309 lines · Jeff Layton
Quoted 307 lines · Milind Changire
responses below ...
Quoted 126 lines · Jeff Layton
Quoted 125 lines · Milind Changire
Problem:
Directory vxattrs like ceph.dir.pin* and ceph.dir.layout* may not be
propagated to the client as frequently to keep them updated. This creates vxattr availability problems.Solution:
Adds new getvxattr op to fetch ceph.dir.pin*, ceph.dir.layout* and ceph.file.layout* vxattrs.
If the entire layout for a dir or a file is being set, then it is expected that the layout be set in standard JSON format. Individual
field value retrieval is not wrapped in JSON. The JSON format also applies while setting the vxattr if the entire layout is being set in
one go.
As a temporary measure, setting a vxattr can also be done in the old
format. The old format will be deprecated in the future.URL: https://tracker.ceph.com/issues/51062
Signed-off-by Milind Changire <[email protected]>--- fs/ceph/inode.c | 51 ++++++++++++++++++++++++++++ fs/ceph/mds_client.c | 27 ++++++++++++++- fs/ceph/mds_client.h | 12 ++++++- fs/ceph/strings.c | 1 + fs/ceph/super.h | 1 + fs/ceph/xattr.c | 65 ++++++++++++++++++++++++++++++++++++ include/linux/ceph/ceph_fs.h | 1 + 7 files changed, 156 insertions(+), 2 deletions(-)fs/ceph/inode.c +31
-0@@ -2291,6 +2291,57 @@ int __ceph_do_getattr(struct inode *inode,struct page *locked_page, return err; } +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value, +size_t size) +{ +struct ceph_fs_client *fsc =ceph_sb_to_client (inode->i_sb); +struct ceph_mds_client *mdsc = fsc->mdsc; +struct ceph_mds_request *req; +int mode =USE_AUTH_MDS ; +int err; +char *xattr_value; +size_t xattr_value_len; + + req =ceph_mdsc_create_request (mdsc,CEPH_MDS_OP_GETVXATTR , mode); +if (IS_ERR (req)) { + err = -ENOMEM ; + goto out; + } + + req->r_path2 =kstrdup (name,GFP_NOFS ); +if (!req->r_path2) { + err = -ENOMEM ; + goto put; + } + +ihold (inode); + req->r_inode = inode; + err =ceph_mdsc_do_request (mdsc,NULL , req); +if (err <0 ) + goto put; + + xattr_value = req->r_reply_info.xattr_info.xattr_value; + xattr_value_len = req-Quoted 1 line
r_reply_info.xattr_info.xattr_value_len;
fs/ceph/inode.c +2
-0+ + dout("do_getvxattr xattr_value_len:%lu, size:%lu\n",
xattr_value_len, size);fs/ceph/inode.c +18
-0+ + err = xattr_value_len; +if (size ==0 ) + goto put; + +if (xattr_value_len > size) { + err = -ERANGE ; + goto put; + } + +memcpy (value, xattr_value, xattr_value_len); +put: +ceph_mdsc_put_request (req); +out: +dout ("do_getvxattr result=%d\n" , err); +return err; +} +/* * Check inode permissions. We verify we have a valid value for fs/ceph/mds_client.c +15
-0@@ -555,6 +555,29 @@ static int parse_reply_info_create(void **p,void *end, return -EIO; } +static int parse_reply_info_getvxattr (void **p,void *end, +struct ceph_mds_reply_info_parsed *info, +u64 features) +{ +u8 struct_v, struct_compat; +u32 struct_len; +u32 value_len; + +ceph_decode_8_safe (p, end, struct_v, bad); +ceph_decode_8_safe (p, end, struct_compat, bad); +ceph_decode_32_safe (p, end, struct_len, bad); +ceph_decode_32_safe (p, end, value_len, bad); + +if (value_len == end - *p) { + info->xattr_info.xattr_value = *p;Quoted 9 lines
fs/ceph/mds_client.c static int parse_reply_info_create(void **p, void *end, · line 576
Quoted 4 lines
fs/ceph/mds_client.c +4
-0+ info->xattr_info.xattr_value_len = end - *p; + *p = end; +return info->xattr_info.xattr_value_len; + }The kernel uses tab indent almost exclusively. Can you fix the indention
above (and anywhere else I missed)?I'll fix this. newbie issues :P
Quoted 162 lines
Quoted 138 lines
fs/ceph/mds_client.c +4
-0+bad: +return -EIO ; +} +/* * parse extra results
*/fs/ceph/mds_client.c +2
-0@@ -570,6 +593,8 @@ static int parse_reply_info_extra(void **p,void *end, return parse_reply_info_readdir(p, end, info,features ); else if (op == CEPH_MDS_OP_CREATE) return parse_reply_info_create(p, end, info, features, s); + else if (op == CEPH_MDS_OP_GETVXATTR) + return parse_reply_info_getvxattr(p, end, info,features); else return -EIO;}
fs/ceph/mds_client.c +1
-1@@ -615,7 +640,7 @@ static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg, if (p != end) goto bad; -return 0 ; +return err; bad:err = -EIO;fs/ceph/mds_client.h +10
-1@@ -29,8 +29,10 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT ,CEPHFS_FEATURE_DELEG_INO ,CEPHFS_FEATURE_METRIC_COLLECT , +CEPHFS_FEATURE_ALTERNATE_NAME , +CEPHFS_FEATURE_GETVXATTR , - CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT, + CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_GETVXATTR, };/* @@ -45,6 +47,8 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT , \CEPHFS_FEATURE_DELEG_INO , \CEPHFS_FEATURE_METRIC_COLLECT , \ +CEPHFS_FEATURE_ALTERNATE_NAME , \ +CEPHFS_FEATURE_GETVXATTR , \ \ CEPHFS_FEATURE_MAX, \ }@@ -100,6 +104,11 @@ struct ceph_mds_reply_dir_entry {loff_t offset; }; +struct ceph_mds_reply_xattr { +char *xattr_value; +size_t xattr_value_len; +}; +/* * parsed info about an mds reply, including information about * either: 1) the target inode and/or its parent directory and dentry,
fs/ceph/mds_client.h +1
-0@@ -115,6 +124,7 @@ struct ceph_mds_reply_info_parsed {char *dname;u32 dname_len;struct ceph_mds_reply_lease *dlease; +struct ceph_mds_reply_xattr xattr_info;/* extra */ union {fs/ceph/strings.c +1
-0@@ -60,6 +60,7 @@ const char *ceph_mds_op_name(int op)case CEPH_MDS_OP_LOOKUPINO :return "lookupino" ;case CEPH_MDS_OP_LOOKUPNAME :return "lookupname" ;case CEPH_MDS_OP_GETATTR :return "getattr" ; +case CEPH_MDS_OP_GETVXATTR :return "getvxattr" ;case CEPH_MDS_OP_SETXATTR :return "setxattr" ;case CEPH_MDS_OP_SETATTR :return "setattr" ;case CEPH_MDS_OP_RMXATTR :return "rmxattr" ;fs/ceph/super.h +1
-0@@ -1043,6 +1043,7 @@ static inline bool ceph_inode_is_shutdown(struct inode *inode)/* xattr.c */ int __ceph_setxattr(struct inode *, const char *, const void *,size_t ,int ); +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value,size_t size);
ssize_t__ceph_getxattr(struct inode *, const char *, void *,size_t);
externssize_tceph_listxattr(struct dentry *, char *, size_t); extern structceph_buffer*__ceph_build_xattrs_blob(structceph_inode_info*ci);fs/ceph/xattr.c +23
-0@@ -918,6 +918,64 @@ static inline int __get_request_mask(struct inode *in) {return mask; } +static inline bool ceph_is_passthrough_vxattr (const char *name) +{ +const char *vxattr_list[] = { +"ceph.dir.pin" , +"ceph.dir.pin.random" , +"ceph.dir.pin.distributed" , +"ceph.dir.layout" , +"ceph.dir.layout.object_size" , +"ceph.dir.layout.stripe_count" , +"ceph.dir.layout.stripe_unit" , +"ceph.dir.layout.pool" , +"ceph.dir.layout.pool_name" , +"ceph.dir.layout.pool_id" , +"ceph.dir.layout.pool_namespace" , +"ceph.file.layout" , +"ceph.file.layout.object_size" , +"ceph.file.layout.stripe_count" , +"ceph.file.layout.stripe_unit" , +"ceph.file.layout.pool" , +"ceph.file.layout.pool_name" , +"ceph.file.layout.pool_id" , +"ceph.file.layout.pool_namespace" , + };fs/ceph/xattr.c static inline int __get_request_mask(struct · line 954
Quoted 10 lines
fs/ceph/xattr.c +10
-0+int i =0 ; +int n =sizeof (vxattr_list)/sizeof (vxattr_list[0 ]); + +while (i < n) { +if (!strcmp (name, vxattr_list[i])) +return true; + i++; + } +return false; +}This part, I'm not so crazy about. This list will be hard to keep in sync with the userland code and you're almost guaranteed to have a mismatch between what the kernel and userland supports since they're on
different release schedules.I think what we probably ought to do is run the
ceph_match_vxattrcall
first, and if it doesn't match the name and the name starts with "ceph.", then we'll send a GETVXATTR call to the MDS.Sound ok?
Quoted 6 lines
Quoted 2 lines
I was trying to preempt a getvxattr call if the full xattr name was invalid.
Huh, why? That seems like the point of the whole bug. We need a way to hand vxattrs that the client doesn't recognize off to the MDS so it can satisfy the request.
File and dir layout vxattrs are server-side xattrs. So, it seemed prudent to delegate them to the server. The idea was also to help identify layouts and their inheritance for a dir hierarchy or a file at the server side and possibly avoiding client round trips to the server for learning about the same (and maintain consistency?)
Here's some discussion about the same:
https://tracker.ceph.com/issues/51062#note-3
Quoted 15 lines
Quoted 5 lines
I could just test for the prefixes: ceph.dir.pin, ceph.dir.layout and ceph.file.layout and send them over. Eventually, we'll get an ENODATA if xattr name validation fails on the MDS side.
Quoted 9 lines
The problem with doing a
ceph_match_vxattr()first is that it will succeed
for any of these vxattrs, but we'll land in the same place as the one you've
described in the tracker.
The very idea of implementing getvxattr is to bypass the existing "cached"
service mechanism for special vxattrs.
Quoted 6 lines
I don't get it. What's the benefit of calling out to the MDS to ask for ceph.file.layout.pool instead of just satisfying it from the info we have?
I'd prefer we just add a fallback for "ceph.*" vxattrs that for which we don't have a local handler.
See my explanation above.
Quoted 68 lines
Quoted 62 lines
Quoted 62 lines
Quoted 62 lines
fs/ceph/xattr.c +25
-0+ +/* check if the entire cluster supports the given feature */ +static inline bool ceph_cluster_has_feature (struct inode *inode,int feature_bit) +{ +int64_t i; +struct ceph_fs_client *fsc =ceph_inode_to_client (inode); +struct ceph_mds_session **sessions = fsc->mdsc->sessions; +int64_t num_sessions =atomic_read (&fsc->mdsc- >num_sessions); + +if (fsc->mdsc->stopping) +return false; + +if (!sessions) +return false; + +for (i =0 ; i < num_sessions; i++) { +struct ceph_mds_session *session = sessions[i]; +if (!session) +return false; +if (!test_bit (feature_bit, &session->s_features)) +return false; + } +return true; +} +
ssize_t__ceph_getxattr(struct inode *inode, const char *name,
void *value,
size_tsize)
{fs/ceph/xattr.c +7
-0@@ -927,6 +985,13 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value, int req_mask;ssize_t err; +if (ceph_is_passthrough_vxattr (name) && +ceph_cluster_has_feature (inode,CEPHFS_FEATURE_GETVXATTR )) { + err =ceph_do_getvxattr (inode, name, value, size); +return err; + } +dout ("vxattr is not passthrough or kclient doesn't support GETVXATTR\n"); +/* let's see if a virtual xattr was requested */ vxattr = ceph_match_vxattr(inode, name); if (vxattr) {include/linux/ceph/ceph_fs.h +1
-0@@ -328,6 +328,7 @@ enum {CEPH_MDS_OP_LOOKUPPARENT =0x00103 ,CEPH_MDS_OP_LOOKUPINO =0x00104 ,CEPH_MDS_OP_LOOKUPNAME =0x00105 , + CEPH_MDS_OP_GETVXATTR = 0x00106,Quoted 3 lines
Quoted 3 lines
include/linux/ceph/ceph_fs.h enum { · line 334
Quoted 2 lines
include/linux/ceph/ceph_fs.h +0
-0CEPH_MDS_OP_SETXATTR =0x01105 , CEPH_MDS_OP_RMXATTR = 0x01106,--
Jeff Layton <[email protected]>
Re: [PATCH 1/1] ceph: add getvxattr op
Quoted 317 lines · Milind Changire
Quoted 309 lines · Jeff Layton
Quoted 307 lines · Milind Changire
responses below ...
Quoted 126 lines · Jeff Layton
Quoted 125 lines · Milind Changire
Problem:
Directory vxattrs like ceph.dir.pin* and ceph.dir.layout* may not be
propagated to the client as frequently to keep them updated. This creates vxattr availability problems.Solution:
Adds new getvxattr op to fetch ceph.dir.pin*, ceph.dir.layout* and ceph.file.layout* vxattrs.
If the entire layout for a dir or a file is being set, then it is expected that the layout be set in standard JSON format. Individual
field value retrieval is not wrapped in JSON. The JSON format also applies while setting the vxattr if the entire layout is being set in
one go.
As a temporary measure, setting a vxattr can also be done in the old
format. The old format will be deprecated in the future.URL: https://tracker.ceph.com/issues/51062
Signed-off-by Milind Changire <[email protected]>--- fs/ceph/inode.c | 51 ++++++++++++++++++++++++++++ fs/ceph/mds_client.c | 27 ++++++++++++++- fs/ceph/mds_client.h | 12 ++++++- fs/ceph/strings.c | 1 + fs/ceph/super.h | 1 + fs/ceph/xattr.c | 65 ++++++++++++++++++++++++++++++++++++ include/linux/ceph/ceph_fs.h | 1 + 7 files changed, 156 insertions(+), 2 deletions(-)fs/ceph/inode.c +31
-0@@ -2291,6 +2291,57 @@ int __ceph_do_getattr(struct inode *inode,struct page *locked_page, return err; } +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value, +size_t size) +{ +struct ceph_fs_client *fsc =ceph_sb_to_client (inode->i_sb); +struct ceph_mds_client *mdsc = fsc->mdsc; +struct ceph_mds_request *req; +int mode =USE_AUTH_MDS ; +int err; +char *xattr_value; +size_t xattr_value_len; + + req =ceph_mdsc_create_request (mdsc,CEPH_MDS_OP_GETVXATTR , mode); +if (IS_ERR (req)) { + err = -ENOMEM ; + goto out; + } + + req->r_path2 =kstrdup (name,GFP_NOFS ); +if (!req->r_path2) { + err = -ENOMEM ; + goto put; + } + +ihold (inode); + req->r_inode = inode; + err =ceph_mdsc_do_request (mdsc,NULL , req); +if (err <0 ) + goto put; + + xattr_value = req->r_reply_info.xattr_info.xattr_value; + xattr_value_len = req-Quoted 1 line
r_reply_info.xattr_info.xattr_value_len;
fs/ceph/inode.c +2
-0+ + dout("do_getvxattr xattr_value_len:%lu, size:%lu\n",
xattr_value_len, size);fs/ceph/inode.c +18
-0+ + err = xattr_value_len; +if (size ==0 ) + goto put; + +if (xattr_value_len > size) { + err = -ERANGE ; + goto put; + } + +memcpy (value, xattr_value, xattr_value_len); +put: +ceph_mdsc_put_request (req); +out: +dout ("do_getvxattr result=%d\n" , err); +return err; +} +/* * Check inode permissions. We verify we have a valid value for fs/ceph/mds_client.c +15
-0@@ -555,6 +555,29 @@ static int parse_reply_info_create(void **p,void *end, return -EIO; } +static int parse_reply_info_getvxattr (void **p,void *end, +struct ceph_mds_reply_info_parsed *info, +u64 features) +{ +u8 struct_v, struct_compat; +u32 struct_len; +u32 value_len; + +ceph_decode_8_safe (p, end, struct_v, bad); +ceph_decode_8_safe (p, end, struct_compat, bad); +ceph_decode_32_safe (p, end, struct_len, bad); +ceph_decode_32_safe (p, end, value_len, bad); + +if (value_len == end - *p) { + info->xattr_info.xattr_value = *p;Quoted 9 lines
fs/ceph/mds_client.c static int parse_reply_info_create(void **p, void *end, · line 576
Quoted 4 lines
fs/ceph/mds_client.c +4
-0+ info->xattr_info.xattr_value_len = end - *p; + *p = end; +return info->xattr_info.xattr_value_len; + }The kernel uses tab indent almost exclusively. Can you fix the indention
above (and anywhere else I missed)?I'll fix this. newbie issues :P
Quoted 162 lines
Quoted 138 lines
fs/ceph/mds_client.c +4
-0+bad: +return -EIO ; +} +/* * parse extra results
*/fs/ceph/mds_client.c +2
-0@@ -570,6 +593,8 @@ static int parse_reply_info_extra(void **p,void *end, return parse_reply_info_readdir(p, end, info,features ); else if (op == CEPH_MDS_OP_CREATE) return parse_reply_info_create(p, end, info, features, s); + else if (op == CEPH_MDS_OP_GETVXATTR) + return parse_reply_info_getvxattr(p, end, info,features); else return -EIO;}
fs/ceph/mds_client.c +1
-1@@ -615,7 +640,7 @@ static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg, if (p != end) goto bad; -return 0 ; +return err; bad:err = -EIO;fs/ceph/mds_client.h +10
-1@@ -29,8 +29,10 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT ,CEPHFS_FEATURE_DELEG_INO ,CEPHFS_FEATURE_METRIC_COLLECT , +CEPHFS_FEATURE_ALTERNATE_NAME , +CEPHFS_FEATURE_GETVXATTR , - CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT, + CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_GETVXATTR, };/* @@ -45,6 +47,8 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT , \CEPHFS_FEATURE_DELEG_INO , \CEPHFS_FEATURE_METRIC_COLLECT , \ +CEPHFS_FEATURE_ALTERNATE_NAME , \ +CEPHFS_FEATURE_GETVXATTR , \ \ CEPHFS_FEATURE_MAX, \ }@@ -100,6 +104,11 @@ struct ceph_mds_reply_dir_entry {loff_t offset; }; +struct ceph_mds_reply_xattr { +char *xattr_value; +size_t xattr_value_len; +}; +/* * parsed info about an mds reply, including information about * either: 1) the target inode and/or its parent directory and dentry,
fs/ceph/mds_client.h +1
-0@@ -115,6 +124,7 @@ struct ceph_mds_reply_info_parsed {char *dname;u32 dname_len;struct ceph_mds_reply_lease *dlease; +struct ceph_mds_reply_xattr xattr_info;/* extra */ union {fs/ceph/strings.c +1
-0@@ -60,6 +60,7 @@ const char *ceph_mds_op_name(int op)case CEPH_MDS_OP_LOOKUPINO :return "lookupino" ;case CEPH_MDS_OP_LOOKUPNAME :return "lookupname" ;case CEPH_MDS_OP_GETATTR :return "getattr" ; +case CEPH_MDS_OP_GETVXATTR :return "getvxattr" ;case CEPH_MDS_OP_SETXATTR :return "setxattr" ;case CEPH_MDS_OP_SETATTR :return "setattr" ;case CEPH_MDS_OP_RMXATTR :return "rmxattr" ;fs/ceph/super.h +1
-0@@ -1043,6 +1043,7 @@ static inline bool ceph_inode_is_shutdown(struct inode *inode)/* xattr.c */ int __ceph_setxattr(struct inode *, const char *, const void *,size_t ,int ); +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value,size_t size);
ssize_t__ceph_getxattr(struct inode *, const char *, void *,size_t);
externssize_tceph_listxattr(struct dentry *, char *, size_t); extern structceph_buffer*__ceph_build_xattrs_blob(structceph_inode_info*ci);fs/ceph/xattr.c +23
-0@@ -918,6 +918,64 @@ static inline int __get_request_mask(struct inode *in) {return mask; } +static inline bool ceph_is_passthrough_vxattr (const char *name) +{ +const char *vxattr_list[] = { +"ceph.dir.pin" , +"ceph.dir.pin.random" , +"ceph.dir.pin.distributed" , +"ceph.dir.layout" , +"ceph.dir.layout.object_size" , +"ceph.dir.layout.stripe_count" , +"ceph.dir.layout.stripe_unit" , +"ceph.dir.layout.pool" , +"ceph.dir.layout.pool_name" , +"ceph.dir.layout.pool_id" , +"ceph.dir.layout.pool_namespace" , +"ceph.file.layout" , +"ceph.file.layout.object_size" , +"ceph.file.layout.stripe_count" , +"ceph.file.layout.stripe_unit" , +"ceph.file.layout.pool" , +"ceph.file.layout.pool_name" , +"ceph.file.layout.pool_id" , +"ceph.file.layout.pool_namespace" , + };fs/ceph/xattr.c static inline int __get_request_mask(struct · line 954
Quoted 10 lines
fs/ceph/xattr.c +10
-0+int i =0 ; +int n =sizeof (vxattr_list)/sizeof (vxattr_list[0 ]); + +while (i < n) { +if (!strcmp (name, vxattr_list[i])) +return true; + i++; + } +return false; +}This part, I'm not so crazy about. This list will be hard to keep in sync with the userland code and you're almost guaranteed to have a mismatch between what the kernel and userland supports since they're on
different release schedules.I think what we probably ought to do is run the
ceph_match_vxattrcall
first, and if it doesn't match the name and the name starts with "ceph.", then we'll send a GETVXATTR call to the MDS.Sound ok?
Quoted 6 lines
Quoted 2 lines
I was trying to preempt a getvxattr call if the full xattr name was invalid.
Huh, why? That seems like the point of the whole bug. We need a way to hand vxattrs that the client doesn't recognize off to the MDS so it can satisfy the request.
Quoted 6 lines
File and dir layout vxattrs are server-side xattrs. So, it seemed prudent to delegate them to the server. The idea was also to help identify layouts and their inheritance for a dir hierarchy or a file at the server side and possibly avoiding client round trips to the server for learning about the same (and maintain consistency?)
I don't see how this solution results in fewer calls to the MDS, and I don't agree that it's prudent to call up the MDS to satisfy a request for info that we already have.
Quoted 3 lines
Here's some discussion about the same:
https://tracker.ceph.com/issues/51062#note-3
I opened the bug and have been following it. I don't see any discussion about delegating getxattr for a bunch of existing vxattrs to the MDS. The bug is that you can set some vxattrs without being able to query them.
Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
I could just test for the prefixes: ceph.dir.pin, ceph.dir.layout and ceph.file.layout and send them over. Eventually, we'll get an ENODATA if xattr name validation fails on the MDS side.
The problem with doing a
ceph_match_vxattr()first is that it will succeed
for any of these vxattrs, but we'll land in the same place as the one you've
described in the tracker.
The very idea of implementing getvxattr is to bypass the existing "cached"
service mechanism for special vxattrs.
Quoted 9 lines
Quoted 6 lines
I don't get it. What's the benefit of calling out to the MDS to ask for ceph.file.layout.pool instead of just satisfying it from the info we have?
I'd prefer we just add a fallback for "ceph.*" vxattrs that for which we don't have a local handler.
See my explanation above.
I see your explanation, but it doesn't outline the benefit of sending these requests to the MDS instead of just satisfying them locally. What good does that do?
I also think you're missing another point: someone will need to keep this vxattr whitelist in sync, and it will be a maintenance headache. It'll also result in a worse experience for users.
Suppose we add some new ceph.dir.foo vxattr on the MDS and backport that to a point release. That attribute will not be accessible until someone also adds it to the whitelist in the client.
At that point, it may be *years* before users can access that vxattr. They'll not only have to wait for the MDS to get the patch that adds the new vxattr, but also for a client that has it in the whitelist.
If however, you do this the way I'm suggesting, once the MDS supports the new vxattr, it'll be usable, and we won't have to continually patch the kernel to support new attributes.
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH 1/1] ceph: add getvxattr op
Quoted 380 lines · Jeff Layton
Quoted 317 lines · Milind Changire
Quoted 309 lines · Jeff Layton
Quoted 307 lines · Milind Changire
responses below ...
Quoted 126 lines · Jeff Layton
Quoted 125 lines · Milind Changire
Problem:
Directory vxattrs like ceph.dir.pin* and ceph.dir.layout* may not be
propagated to the client as frequently to keep them updated. This creates vxattr availability problems.Solution:
Adds new getvxattr op to fetch ceph.dir.pin*, ceph.dir.layout* and ceph.file.layout* vxattrs.
If the entire layout for a dir or a file is being set, then it is expected that the layout be set in standard JSON format. Individual
field value retrieval is not wrapped in JSON. The JSON format also applies while setting the vxattr if the entire layout is being set in
one go.
As a temporary measure, setting a vxattr can also be done in the old
format. The old format will be deprecated in the future.URL: https://tracker.ceph.com/issues/51062
Signed-off-by Milind Changire <[email protected]>--- fs/ceph/inode.c | 51 ++++++++++++++++++++++++++++ fs/ceph/mds_client.c | 27 ++++++++++++++- fs/ceph/mds_client.h | 12 ++++++- fs/ceph/strings.c | 1 + fs/ceph/super.h | 1 + fs/ceph/xattr.c | 65 ++++++++++++++++++++++++++++++++++++ include/linux/ceph/ceph_fs.h | 1 + 7 files changed, 156 insertions(+), 2 deletions(-)fs/ceph/inode.c +31
-0@@ -2291,6 +2291,57 @@ int __ceph_do_getattr(struct inode *inode,struct page *locked_page, return err; } +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value, +size_t size) +{ +struct ceph_fs_client *fsc =ceph_sb_to_client (inode->i_sb); +struct ceph_mds_client *mdsc = fsc->mdsc; +struct ceph_mds_request *req; +int mode =USE_AUTH_MDS ; +int err; +char *xattr_value; +size_t xattr_value_len; + + req =ceph_mdsc_create_request (mdsc,CEPH_MDS_OP_GETVXATTR , mode); +if (IS_ERR (req)) { + err = -ENOMEM ; + goto out; + } + + req->r_path2 =kstrdup (name,GFP_NOFS ); +if (!req->r_path2) { + err = -ENOMEM ; + goto put; + } + +ihold (inode); + req->r_inode = inode; + err =ceph_mdsc_do_request (mdsc,NULL , req); +if (err <0 ) + goto put; + + xattr_value = req->r_reply_info.xattr_info.xattr_value; + xattr_value_len = req-Quoted 1 line
r_reply_info.xattr_info.xattr_value_len;
fs/ceph/inode.c +2
-0+ + dout("do_getvxattr xattr_value_len:%lu, size:%lu\n",
xattr_value_len, size);fs/ceph/inode.c +18
-0+ + err = xattr_value_len; +if (size ==0 ) + goto put; + +if (xattr_value_len > size) { + err = -ERANGE ; + goto put; + } + +memcpy (value, xattr_value, xattr_value_len); +put: +ceph_mdsc_put_request (req); +out: +dout ("do_getvxattr result=%d\n" , err); +return err; +} +/* * Check inode permissions. We verify we have a valid value for fs/ceph/mds_client.c +15
-0@@ -555,6 +555,29 @@ static int parse_reply_info_create(void **p,void *end, return -EIO; } +static int parse_reply_info_getvxattr (void **p,void *end, +struct ceph_mds_reply_info_parsed *info, +u64 features) +{ +u8 struct_v, struct_compat; +u32 struct_len; +u32 value_len; + +ceph_decode_8_safe (p, end, struct_v, bad); +ceph_decode_8_safe (p, end, struct_compat, bad); +ceph_decode_32_safe (p, end, struct_len, bad); +ceph_decode_32_safe (p, end, value_len, bad); + +if (value_len == end - *p) { + info->xattr_info.xattr_value = *p;Quoted 9 lines
fs/ceph/mds_client.c static int parse_reply_info_create(void **p, void *end, · line 576
Quoted 4 lines
fs/ceph/mds_client.c +4
-0+ info->xattr_info.xattr_value_len = end - *p; + *p = end; +return info->xattr_info.xattr_value_len; + }The kernel uses tab indent almost exclusively. Can you fix the indention
above (and anywhere else I missed)?I'll fix this. newbie issues :P
Quoted 162 lines
Quoted 138 lines
fs/ceph/mds_client.c +4
-0+bad: +return -EIO ; +} +/* * parse extra results
*/fs/ceph/mds_client.c +2
-0@@ -570,6 +593,8 @@ static int parse_reply_info_extra(void **p,void *end, return parse_reply_info_readdir(p, end, info,features ); else if (op == CEPH_MDS_OP_CREATE) return parse_reply_info_create(p, end, info, features, s); + else if (op == CEPH_MDS_OP_GETVXATTR) + return parse_reply_info_getvxattr(p, end, info,features); else return -EIO;}
fs/ceph/mds_client.c +1
-1@@ -615,7 +640,7 @@ static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg, if (p != end) goto bad; -return 0 ; +return err; bad:err = -EIO;fs/ceph/mds_client.h +10
-1@@ -29,8 +29,10 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT ,CEPHFS_FEATURE_DELEG_INO ,CEPHFS_FEATURE_METRIC_COLLECT , +CEPHFS_FEATURE_ALTERNATE_NAME , +CEPHFS_FEATURE_GETVXATTR , - CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT, + CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_GETVXATTR, };/* @@ -45,6 +47,8 @@ enum ceph_feature_type {CEPHFS_FEATURE_MULTI_RECONNECT , \CEPHFS_FEATURE_DELEG_INO , \CEPHFS_FEATURE_METRIC_COLLECT , \ +CEPHFS_FEATURE_ALTERNATE_NAME , \ +CEPHFS_FEATURE_GETVXATTR , \ \ CEPHFS_FEATURE_MAX, \ }@@ -100,6 +104,11 @@ struct ceph_mds_reply_dir_entry {loff_t offset; }; +struct ceph_mds_reply_xattr { +char *xattr_value; +size_t xattr_value_len; +}; +/* * parsed info about an mds reply, including information about * either: 1) the target inode and/or its parent directory and dentry,
fs/ceph/mds_client.h +1
-0@@ -115,6 +124,7 @@ struct ceph_mds_reply_info_parsed {char *dname;u32 dname_len;struct ceph_mds_reply_lease *dlease; +struct ceph_mds_reply_xattr xattr_info;/* extra */ union {fs/ceph/strings.c +1
-0@@ -60,6 +60,7 @@ const char *ceph_mds_op_name(int op)case CEPH_MDS_OP_LOOKUPINO :return "lookupino" ;case CEPH_MDS_OP_LOOKUPNAME :return "lookupname" ;case CEPH_MDS_OP_GETATTR :return "getattr" ; +case CEPH_MDS_OP_GETVXATTR :return "getvxattr" ;case CEPH_MDS_OP_SETXATTR :return "setxattr" ;case CEPH_MDS_OP_SETATTR :return "setattr" ;case CEPH_MDS_OP_RMXATTR :return "rmxattr" ;fs/ceph/super.h +1
-0@@ -1043,6 +1043,7 @@ static inline bool ceph_inode_is_shutdown(struct inode *inode)/* xattr.c */ int __ceph_setxattr(struct inode *, const char *, const void *,size_t ,int ); +int ceph_do_getvxattr (struct inode *inode,const char *name,void *value,size_t size);
ssize_t__ceph_getxattr(struct inode *, const char *, void *,size_t);
externssize_tceph_listxattr(struct dentry *, char *, size_t); extern structceph_buffer*__ceph_build_xattrs_blob(structceph_inode_info*ci);fs/ceph/xattr.c +23
-0@@ -918,6 +918,64 @@ static inline int __get_request_mask(struct inode *in) {return mask; } +static inline bool ceph_is_passthrough_vxattr (const char *name) +{ +const char *vxattr_list[] = { +"ceph.dir.pin" , +"ceph.dir.pin.random" , +"ceph.dir.pin.distributed" , +"ceph.dir.layout" , +"ceph.dir.layout.object_size" , +"ceph.dir.layout.stripe_count" , +"ceph.dir.layout.stripe_unit" , +"ceph.dir.layout.pool" , +"ceph.dir.layout.pool_name" , +"ceph.dir.layout.pool_id" , +"ceph.dir.layout.pool_namespace" , +"ceph.file.layout" , +"ceph.file.layout.object_size" , +"ceph.file.layout.stripe_count" , +"ceph.file.layout.stripe_unit" , +"ceph.file.layout.pool" , +"ceph.file.layout.pool_name" , +"ceph.file.layout.pool_id" , +"ceph.file.layout.pool_namespace" , + };fs/ceph/xattr.c static inline int __get_request_mask(struct · line 954
Quoted 10 lines
fs/ceph/xattr.c +10
-0+int i =0 ; +int n =sizeof (vxattr_list)/sizeof (vxattr_list[0 ]); + +while (i < n) { +if (!strcmp (name, vxattr_list[i])) +return true; + i++; + } +return false; +}This part, I'm not so crazy about. This list will be hard to keep in sync with the userland code and you're almost guaranteed to have a mismatch between what the kernel and userland supports since they're on
different release schedules.I think what we probably ought to do is run the
ceph_match_vxattrcall
first, and if it doesn't match the name and the name starts with "ceph.", then we'll send a GETVXATTR call to the MDS.Sound ok?
Quoted 6 lines
Quoted 2 lines
I was trying to preempt a getvxattr call if the full xattr name was invalid.
Huh, why? That seems like the point of the whole bug. We need a way to hand vxattrs that the client doesn't recognize off to the MDS so it can satisfy the request.
Quoted 6 lines
File and dir layout vxattrs are server-side xattrs. So, it seemed prudent to delegate them to the server. The idea was also to help identify layouts and their inheritance for a dir hierarchy or a file at the server side and possibly avoiding client round trips to the server for learning about the same (and maintain consistency?)
I don't see how this solution results in fewer calls to the MDS, and I don't agree that it's prudent to call up the MDS to satisfy a request for info that we already have.
Quoted 3 lines
Here's some discussion about the same:
https://tracker.ceph.com/issues/51062#note-3I opened the bug and have been following it. I don't see any discussion about delegating getxattr for a bunch of existing vxattrs to the MDS. The bug is that you can set some vxattrs without being able to query them.
Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
I could just test for the prefixes: ceph.dir.pin, ceph.dir.layout and ceph.file.layout and send them over. Eventually, we'll get an ENODATA if xattr name validation fails on the MDS side.
The problem with doing a
ceph_match_vxattr()first is that it will succeed
for any of these vxattrs, but we'll land in the same place as the one you've
described in the tracker.
The very idea of implementing getvxattr is to bypass the existing "cached"
service mechanism for special vxattrs.Quoted 9 lines
Quoted 6 lines
I don't get it. What's the benefit of calling out to the MDS to ask for ceph.file.layout.pool instead of just satisfying it from the info we have?
I'd prefer we just add a fallback for "ceph.*" vxattrs that for which we don't have a local handler.
See my explanation above.
I see your explanation, but it doesn't outline the benefit of sending these requests to the MDS instead of just satisfying them locally. What good does that do?
I also think you're missing another point: someone will need to keep this vxattr whitelist in sync, and it will be a maintenance headache. It'll also result in a worse experience for users.
Suppose we add some new ceph.dir.foo vxattr on the MDS and backport that to a point release. That attribute will not be accessible until someone also adds it to the whitelist in the client.
At that point, it may be *years* before users can access that vxattr. They'll not only have to wait for the MDS to get the patch that adds the new vxattr, but also for a client that has it in the whitelist.
Quoted 5 lines
If however, you do this the way I'm suggesting, once the MDS supports the new vxattr, it'll be usable, and we won't have to continually patch the kernel to support new attributes.
--
Jeff Layton <[email protected]>
hmm ... well, then I'll just have to remove layout and pin vxattrs from the list of vxattrs being supported by the client instead of maintaining the whitelist for getvxattr.
In the code, if the vxatrr is not found in the client supported list, the code will drop down to the else case with getvxattr support.
I think I finally got around to what you were trying to say.
kernel test robot
build botRe: [PATCH 1/1] ceph: add getvxattr op
In reply to Milind ChangireThank you for the patch! Perhaps something to improve:
Hi Milind,
Thank you for the patch! Perhaps something to improve:
[auto build test WARNING on ceph-client/for-linus]
[also build test WARNING on v5.16 next-20220112]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]
url: https://github.com/0day-ci/linux/commits/Milind-Changire/ceph-add-getvxattr-support/20220111-202533 base: https://github.com/ceph/ceph-client.git for-linus config: i386-allyesconfig (https://download.01.org/0day-ci/archive/20220112/202201122237.Lspc8TH5-lkp(a)intel.com/config) compiler: gcc-9 (Debian 9.3.0-22) 9.3.0
reproduce (this is a W=1 build):
# https://github.com/0day-ci/linux/commit/9e670d02ce9f9d6e1ac3e234a89d305c85302338
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Milind-Changire/ceph-add-getvxattr-support/20220111-202533 git checkout 9e670d02ce9f9d6e1ac3e234a89d305c85302338
# save the config file to linux build tree
mkdir build_dir
make W=1 O=build_dir ARCH=i386 SHELL=/bin/bash drivers/gpu/drm/vmwgfx/ drivers/iio/ fs/ceph/
If you fix the issue, kindly add following tag as appropriate
All warnings (new ones prefixed by >>):
fs/ceph/inode.c: In function 'ceph_do_getvxattr':
>> <command-line>: warning: format '%lu' expects argument of type 'long unsigned int', but argument 7 has type 'size_t' {aka 'unsigned int'} [-Wformat=]
include/linux/ceph/ceph_debug.h:5:21: note: in expansion of macro 'KBUILD_MODNAME'
5 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
| ^~~~~~~~~~~~~~
include/linux/dynamic_debug.h:134:15: note: in expansion of macro 'pr_fmt'
134 | func(&id, ##__VA_ARGS__); \
| ^~~~~~~~~~~
include/linux/dynamic_debug.h:152:2: note: in expansion of macro '__dynamic_func_call'
152 | __dynamic_func_call(__UNIQUE_ID(ddebug), fmt, func, ##__VA_ARGS__)
| ^~~~~~~~~~~~~~~~~~~
include/linux/dynamic_debug.h:162:2: note: in expansion of macro '_dynamic_func_call'
162 | _dynamic_func_call(fmt, __dynamic_pr_debug, \
| ^~~~~~~~~~~~~~~~~~
include/linux/printk.h:574:2: note: in expansion of macro 'dynamic_pr_debug'
574 | dynamic_pr_debug(fmt, ##__VA_ARGS__)
| ^~~~~~~~~~~~~~~~
include/linux/ceph/ceph_debug.h:19:2: note: in expansion of macro 'pr_debug'
19 | pr_debug("%.*s %12.12s:%-4d : " fmt, \
| ^~~~~~~~
fs/ceph/inode.c:2326:2: note: in expansion of macro 'dout'
2326 | dout("do_getvxattr xattr_value_len:%lu, size:%lu\n", xattr_value_len, size);
| ^~~~
<command-line>: warning: format '%lu' expects argument of type 'long unsigned int', but argument 8 has type 'size_t' {aka 'unsigned int'} [-Wformat=]
include/linux/ceph/ceph_debug.h:5:21: note: in expansion of macro 'KBUILD_MODNAME'
5 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
| ^~~~~~~~~~~~~~
include/linux/dynamic_debug.h:134:15: note: in expansion of macro 'pr_fmt'
134 | func(&id, ##__VA_ARGS__); \
| ^~~~~~~~~~~
include/linux/dynamic_debug.h:152:2: note: in expansion of macro '__dynamic_func_call'
152 | __dynamic_func_call(__UNIQUE_ID(ddebug), fmt, func, ##__VA_ARGS__)
| ^~~~~~~~~~~~~~~~~~~
include/linux/dynamic_debug.h:162:2: note: in expansion of macro '_dynamic_func_call'
162 | _dynamic_func_call(fmt, __dynamic_pr_debug, \
| ^~~~~~~~~~~~~~~~~~
include/linux/printk.h:574:2: note: in expansion of macro 'dynamic_pr_debug'
574 | dynamic_pr_debug(fmt, ##__VA_ARGS__)
| ^~~~~~~~~~~~~~~~
include/linux/ceph/ceph_debug.h:19:2: note: in expansion of macro 'pr_debug'
19 | pr_debug("%.*s %12.12s:%-4d : " fmt, \
| ^~~~~~~~
fs/ceph/inode.c:2326:2: note: in expansion of macro 'dout'
2326 | dout("do_getvxattr xattr_value_len:%lu, size:%lu\n", xattr_value_len, size);
| ^~~~---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org
kernel test robot
build botRe: [PATCH 1/1] ceph: add getvxattr op
In reply to Milind ChangireThank you for the patch! Perhaps something to improve:
Hi Milind,
Thank you for the patch! Perhaps something to improve:
[auto build test WARNING on ceph-client/for-linus]
[also build test WARNING on v5.16 next-20220112]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]
url: https://github.com/0day-ci/linux/commits/Milind-Changire/ceph-add-getvxattr-support/20220111-202533 base: https://github.com/ceph/ceph-client.git for-linus config: i386-randconfig-a006 (https://download.01.org/0day-ci/archive/20220113/202201130325.UBjYGkoh-lkp(a)intel.com/config) compiler: clang version 14.0.0 (https://github.com/llvm/llvm-project 244dd2913a43a200f5a6544d424cdc37b771028b)
reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/0day-ci/linux/commit/9e670d02ce9f9d6e1ac3e234a89d305c85302338
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Milind-Changire/ceph-add-getvxattr-support/20220111-202533 git checkout 9e670d02ce9f9d6e1ac3e234a89d305c85302338
# save the config file to linux build tree
mkdir build_dir
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=i386 SHELL=/bin/bash fs/ceph/
If you fix the issue, kindly add following tag as appropriate
All warnings (new ones prefixed by >>):
>> fs/ceph/inode.c:2326:55: warning: format specifies type 'unsigned long' but the argument has type 'size_t' (aka 'unsigned int') [-Wformat]
dout("do_getvxattr xattr_value_len:%lu, size:%lu\n", xattr_value_len, size);
~~~ ^~~~~~~~~~~~~~~
%u
include/linux/ceph/ceph_debug.h:35:45: note: expanded from macro 'dout'
# define dout(fmt, ...) pr_debug(" " fmt, ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
include/linux/printk.h:580:38: note: expanded from macro 'pr_debug'
no_printk(KERN_DEBUG pr_fmt(fmt), ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
include/linux/printk.h:132:17: note: expanded from macro 'no_printk'
printk(fmt, ##__VA_ARGS__); \
~~~ ^~~~~~~~~~~
include/linux/printk.h:450:60: note: expanded from macro 'printk'
#define printk(fmt, ...) printk_index_wrap(_printk, fmt, ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
include/linux/printk.h:422:19: note: expanded from macro 'printk_index_wrap'
_p_func(_fmt, ##__VA_ARGS__); \
~~~~ ^~~~~~~~~~~
fs/ceph/inode.c:2326:72: warning: format specifies type 'unsigned long' but the argument has type 'size_t' (aka 'unsigned int') [-Wformat]
dout("do_getvxattr xattr_value_len:%lu, size:%lu\n", xattr_value_len, size);
~~~ ^~~~
%u
include/linux/ceph/ceph_debug.h:35:45: note: expanded from macro 'dout'
# define dout(fmt, ...) pr_debug(" " fmt, ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
include/linux/printk.h:580:38: note: expanded from macro 'pr_debug'
no_printk(KERN_DEBUG pr_fmt(fmt), ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
include/linux/printk.h:132:17: note: expanded from macro 'no_printk'
printk(fmt, ##__VA_ARGS__); \
~~~ ^~~~~~~~~~~
include/linux/printk.h:450:60: note: expanded from macro 'printk'
#define printk(fmt, ...) printk_index_wrap(_printk, fmt, ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
include/linux/printk.h:422:19: note: expanded from macro 'printk_index_wrap'
_p_func(_fmt, ##__VA_ARGS__); \
~~~~ ^~~~~~~~~~~
2 warnings generated.vim +2326 fs/ceph/inode.c
2293
2294 int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
2295 size_t size)
2296 {
2297 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2298 struct ceph_mds_client *mdsc = fsc->mdsc;
2299 struct ceph_mds_request *req;
2300 int mode = USE_AUTH_MDS;
2301 int err;
2302 char *xattr_value;
2303 size_t xattr_value_len;
2304
2305 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
2306 if (IS_ERR(req)) {
2307 err = -ENOMEM;
2308 goto out;
2309 }
2310
2311 req->r_path2 = kstrdup(name, GFP_NOFS);
2312 if (!req->r_path2) {
2313 err = -ENOMEM;
2314 goto put;
2315 }
2316
2317 ihold(inode);
2318 req->r_inode = inode;
2319 err = ceph_mdsc_do_request(mdsc, NULL, req);
2320 if (err < 0)
2321 goto put;
2322
2323 xattr_value = req->r_reply_info.xattr_info.xattr_value;
2324 xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
2325
> 2326 dout("do_getvxattr xattr_value_len:%lu, size:%lu\n", xattr_value_len, size); 2327
2328 err = xattr_value_len;
2329 if (size == 0)
2330 goto put;
2331
2332 if (xattr_value_len > size) {
2333 err = -ERANGE;
2334 goto put;
2335 }
2336
2337 memcpy(value, xattr_value, xattr_value_len);
2338 put:
2339 ceph_mdsc_put_request(req);
2340 out:
2341 dout("do_getvxattr result=%d\n", err);
2342 return err;
2343 }
2344
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org