Version 9 (modified by 14 years ago) (diff) | ,
---|
Fedd Databases
There are several small databases that control how fedd
operates. They are the various access databases that control which fedids can access services and how they will be mapped into three-level names or local testbed parameters, and the database that maps local testbed names into service URIs.
All of the formats allow comments that begin with # and ignore blank lines.
AccessDBs
The primary role of the various fedd
accessDBs is to assign attributes to the fedids or three-level names that a requester may assert. Those attributes control the access to services. A requester with the proper attribute may access the service, one without is denied. The access component and experiment control component accessDBs assign the attributes necessary for their services implicitly in conjunction with specifying mapping information inherent to the function. The other components' accessDBs are simple attribute assignments.
Global AccessDB
The global accessDB is the simplest format of the DBs and does the simple funtion of mapping fedid to attributes. The format is a single line per fedid listing the fedid and a comma separated list of attributes.
fedid:xxx attr[,attr]
Lines are additive, so:
fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea split,allocate
and
fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea allocate fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea split
are equivalent.
Relevant attributes include:
- access
- The right to access resources on this testbed. Without an mapping to local resources in the access component's accessDB, this attribute is useless. It is specified implicitly by inclusion in that DB.
- allocate
-
The right to allocate local resources through the internal remote interface. A
fedd
running on boss to allocate resources based on the decisions of afedd
running on users would grant the usersfedd
this attribute. - create
-
The right to create an experiment on this
fedd
. Assigning this without an entry in the experiment control component's accessDB is not useful. Having the right without specifying a mapping to three-level name fill cause the creation to fail. - split
-
The right to run the experiment splitter program on this
fedd
. Because the service at http://users.isi.deterlab.net:23235 is open to anyone, this attribute is rarely allocated.
The accessDB options in the [splitter] and [allocate] sections of the configuration file allow the databases containing the allocate and split attributes to be specified in separate DBs. When the DB is specified in the [allocate] section, only the allocate attribute is recognized, and similarly with the [splitter] section. When specified in [globals] any attribute is respected.
The utility program fedid.py is useful in creating global accessDBs.
Access Component AccessDB
This is the most complex of the accessDBs. Its purpose is to configure how an access control component maps a three level name to a authorization attribute and local access control information. Usually the authorization attribute information is 'access' implying the ability to create new experiments. Other plug-ins are free to create and interpret other attributes.
In addition to that, a mapping can have local values, interpreted by the plug-in. By convention the attribute is unadorned and the optional plug-in dependent information is separated by a comman and enclosed in parentheses.
The general format looks like:
(testbed, project, user) -> attribute
With testbed dependent information:
(testbed, project, user) -> attribute, (some, testbed, stuff)
Emulab Mappings
The file is a text file. Each line containing a -> represents a rule for mapping from one or more three-level names to local testbed entities. The format for a three-level name to project info mapping has the format:
(testbed, project, user) -> attribute, (local_project, local_creation_user, local_service_user)
The triple on the left is the three-level name. One of those fields must be a fedid, given as fedid:xxxxx . The x's are the text of the hexadecimal representation of the fedid. After the fedid, the fields can either be strings or more fedids. Recall that a valid three-level name must be anchored in a fedid.
The local_project
is a project on the testbed, and local_creation_user
and local_service_user
are testbed users that are members of that project. The local_experiment_user
will run commands on behalf of fedd
that create the sub-experiment on this testbed and needs appropriate authority. The local_service-user
may be accessed by the experimenter once the experiment is in progress, and need not have the authority to create or delete experiments. The optional list of node types after the local_project
(separated by a colon) are the node types that the project can access.
An example of the simplest sort of access accessdb line is:
(fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, Deter, faber) -> access, (fed, foo, bar)
That line grants access to a user faber
in project Deter
from the testbed with the given fedid. That user will be mapped into local testbed project fed
as local testbed user foo
for experiment creation and bar
for service access. Notice that the strings to the left of the arrow refer to global three-level names and those on the right to local testbed entities.
Special fields can appear in either the three-level name or the local entity clauses. In the three level name, they determine how the rule matches. Two special fields are <any>
which matches any value and <none>
which explicitly indicates the field is absent. Here are examples with those rules:
(fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, <any>, bill) -> access, (fed, foo, bar) (fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, <none>, faber) -> access, (fed, baz, quux) (fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, <any>, <none>) -> access, (fed, foo, fred)
The first rule will match the bill
user from the given testbed regardless of the project that the testbed asserts (including no project). The second line matches the faber
user from the given testbed only if no project is asserted by that testbed. The third line matches any three-level name that originates from the given testbed and includes only a project (and any project will match).
Any three-level name that does not match a rule in this DB is denied access. That is, any three-level name matching a rule in this database is granted the access attribute.
On the local entity side, the special fields are <same>
which indicates that the corresponding value in the three-level name should be used as the local name and <dynamic>
which indicates that a new project or user should be allocated for this access. For example:
(fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, Deter, <any>) -> access, (fed, foo, <same>) (fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, <none>, faber) -> access, (<dynamic>, <dynamic>, <dynamic>)
The top line indicates that any user in the Deter
project of the given testbed will be mapped to the local fed
project using user foo
to create the experiment and an account with the same user name as the user in the three-level name for service access. (NB <same>
means "use the same string as in the three-level name" not "use the same string as for the other local user.") The administrator should guarantee that the mebers of the foo
group on his or her testbed are a superset of the users for which the given testbed will assert Deter
as a project in the three-level name. That may be done out of band. The next line indicates that user faber
without a project from the given testbed should be allocated a dynamic project. Currently if one field is <dynamic>
all three must be, a restriction that will be relaxed in the future.
If a rule with no <any>
fields matches, that rule is preferred to others with <any>
fields. If multiple <any>
fields match, the match with the fewest <any>
matches has precedence. An <any>
field cannot anchor a three-level matching rule (that it is, <any>
cannot match an initial fedid). That leaves one ambiguity - if (fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, Deter, faber) requests access from a testbed with only the rules:
(fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, Deter, <any>) -> access, (foo, <same>, <same>) (fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea, <any>, faber) -> access, (bar, <same>, <same>)
the outcome is determined by the setting of the project_priority option in the access section of the global configuration file. If it is true, the user will map to local project foo
, otherwise to bar
.
DRAGON Mappings
A dragon controller maps the three level name to a directory holding the X.509 certificates to use for the allocation. In OSCARS command-line terms, it maps the requester to the repo parameter. This line;
(fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea,emulab-ops, faber) -> access, (/usr/local/OSCARS/local_repo)
Maps the given three-level name to the keys stored in /usr/local/OSCARS/local_repo. That directory should be protected appropriately.
ProtoGENI Mappings
A dragon controller maps the three level name to an X.509 certificate used to make GENI calls, an optional password for that certificate, and a username and ssh_key with rights to move files around on ProtoGENI for configuration of nodes. ProtoGENI supports less autometed installation than Emulab.This line;
(fedid:ce90957dd5b7d20f9c3890c4599313b7f1cf31ea,emulab-ops, faber) -> access, (/usr/local/etc/fedd-config/protoGENI/ProtoGENI.pem, tied_user, /usr/local/etc/fedd-config/protoGENI/fedd_rsa, jhktsta)
Maps the given three-level name to a protogeni certificate called /usr/local/etc/fedd-config/protoGENI/ProtoGENI.pem
. The user tied_user
and the ssh key in fedd_rsa
in the given directory will be used for configuration. The password for the certificate is the last entry. Because it contains passwords, this file should be readable only by the fedd user.
Experiment Control Component AccessDB
Like the access component accessDB, the experiment control component accessDB performs double duty - it both determines what fedids can be used to create experiments, and it determines what three-level names a fedd
will assert on behalf of that user when acquiring access to other testbeds.
Lines in the accessDB have the following format:
fedid:xxx -> user fedid:xxx -> (project, user)
The top line says that the given fedid should be mapped to (this_testbed
, , user
), and the second line says that the given fedid should be mapped to (this_testbed
, project
, user
), where this_testbed
is the fedid of this fedd
.
If the same fedid appears on more than one line, any of the three-level name may be used on behalf of that requester. Currently fedd
tries them all in an arbitrary order until one is granted access at each testbed. During an embedding a requester may be granted access to different testbeds under different three-level names.
Any fedid not appearing in this databses is denied access to the experiment control facility; that is, a fedid in this DB is granted the create attribute. When an experiment is created, access to it is controlled independently of this database, so a fedid in this database does not automatically possess the right to terminate or interrogate an experiment they did not create. These rights are controlled by non-string attributes that are not specified in the databases.
To simplify creating this database, the fedd package supplies a utility script, exp_access_db.py.
Experiment Name Mapping DB
This DB controls how fedd
maps the user annotations of the experiment nodes to the URI of the fedd
controlling the testbed. It is a simple colon-separated text file where the annotation is on the left and the URI on the right. For example:
deter:https://users.isi.deterlab.net:23235 ucb:https://users.ucb.deterlab.net:23235