Fix/listing

Design.md

@@ -192,33 +192,23 @@ __. and ..__
 File creation is decoupled from open and write. A new entry is created in DBD. the metadata associated to the object notably contains:
  - a random content-md5 with char #32 set to '-'
  - Date and last-modified set to now()
- - owner-display-name and owner-id set to configured values
+ - owner-display-name and owner-id set to the values found in the bucket attributes
  - an empty locations array
- - Future: Do we have to remove the placeholder??
+ - Future: Do we have to remove or update the placeholder?
 
 This has the effect of creating instantly an empty file.
 
-
-## File removal
-
- - Future: If the object is the last with this prefix, create a placeholder
- - delete object entry from DBD bucket
- - delete all parts from sproxyd storage
-
-At this time, it's kind of broken when the file to remove is the last in the prefix.
-
 ### Directory creation
 
- - Future: Create a placeholder for the directory
-
-At this time, directory creation is badly implemented using internal states that can't be relied on.
+Directories are created as placeholders. A placeholder is an empty object named after the directory name with a trailing slash (```/```) with content-type set to ```application/x-directory```
 
 
-### Directory removal
+## File/Directory removal
 
- - Future: delete the placeholder if it is the last object with this prefix
-
-Today, it's almost broken
+ - create the placeholder and replace it if it already exists ! it is faster than a lookup
+ - if it is a directory removal, check if it is empty
+ - delete object entry from DBD bucket
+ - if it is a file delete all parts from sproxyd storage
 
 ### Open a file
 
@@ -255,20 +245,22 @@ When a commit is requested by the client, the object entry in DBD is updated wit
 
 On regular files (which are backed by real objects) getattrs requests the object's metadata from DBD.
  - Atime, mtime, ctime and chgtime are set to last-modified
- - uid/gid are left blank
+ - uid/gid are set to configured values
  - size is set to "content-length"
  - parts list is updated
 
 
-On directories, it's another story. A prefix/delimiter request is performed to check if the directory still exist. If it is the case, in memory structure is left unchanged.
+On directories, it's another story.
+ - Without a placeholder, a prefix/delimiter request is performed to check if the directory still exist. If it is the case, in memory structure is left unchanged.
+ - With a placeholder, placeholder content is loaded like regular files
 
 Regarding directories, there is something worth to mention. Directory content invalidation takes place here, in order to the client to get a fresh directory content. This invalidation is done using ganesha upcalls.
 
 
 
 ### setattrs
 
-Setattrs only applies on regular files.
+Setattrs only applies on regular files and directories, even without a placeholder.
 
 The following information is kept up to date by Scality NFS
  - FIXME: last-modified and Date are set to now()

README.md

@@ -37,21 +37,26 @@ $ docker build -t scality/nfsd .
 
 A configuration file must be provided, check the dbd and sproxyd urls and set the MYBUCKET shell var with the bucket name you want to be exported.
 
-OWNER_DISPLAY_NAME and OWNER_ID must be set in order to be able to read back from S3 data written through NFS. This information should be retrieved using a curl command on an existing object in the bucket
-```
-$ curl http://127.0.0.1:9004/default/bucket/$MYBUCKET/existing_object
-```
+User/group mapping is done using ```Anonymous_uid```/```Anonymous_gid``` parameters. By default these values are set to 0:0 (which is traditionally root:root). Existing and new files will be automatically owned by the user/group pointed by these values.
+
+Regarding the POSIX rights, by default directories have ```06777``` permission and regular files have ```0666```. The umask parameter is used to unset permission bits of files and directories (e.g. a bit set in the umask will unset the corresponding bit in the permission bitmap). Default umask is 0
 
+Most systems support set-group-ID on directories (and few systems also support set-user-ID). So these bits are set by default in order to be consistent regarding the newly created files belonging to the defined uid and gid.
+
+In the following configuration, objects belong to nobody:users with a umask of 02. This gives ```06775``` on directories and ```0664``` on files. This permits to all users belonging to the ```users``` group to access R/W the export. But worth to mention, it is not possible to alter the attributes of a file for which the owner doesn't match the uid of the running process. commands such as cp(1), chmod(1), chown(1), touch(1) may return EPERM
+
+(Note: ganesha fails to parse umask starting with multiple 0)
 
 ```
 $ mkdir conf logs
 $ MYBUCKET=mybucket
-$ OWNER_DISPLAY_NAME=ABC
-$ OWNER_ID=XYZ
+$ ANON_USER=nobody
+$ ANON_GROUP=users
+$ ANON_UMASK=02
 $ cat >conf/scality-nfsd.conf <<EOF
 SCALITY
 {
-	dbd_url = "http://127.0.0.1:9004/default/bucket";
+	dbd_url = "http://127.0.0.1:9000";
 	sproxyd_url = "http://127.0.0.1:8181/proxy/arc";
 }
 
@@ -70,12 +75,14 @@ EXPORT
 	# Could use CLIENT blocks instead
 	Access_Type = RW;
 
+	Anonymous_uid = $(id -u $ANON_USER);
+	Anonymous_gid = $(awk -F: '/^'$ANON_GROUP':/ { print $3 }' /etc/group);
+
 	# Exporting FSAL
 	FSAL {
 		Name = "SCALITY";
 		bucket = "$MYBUCKET";
-		owner_display_name = "$OWNER_DISPLAY_NAME";
-		owner_id = "$OWNER_ID";
+		umask = $ANON_UMASK;
 	}
 }
 EOF