*troff -ms
100 \s+2Current Research by The Computer Systems Research Group of Berkeley\s-2 \*(DT
Marshall Kirk McKusick Michael J Karels Keith Sklower Kevin Fall Marc Teitelbaum Keith Bostic
1 Summary
0
The release of 4.3BSD in April of 1986 addressed many of the performance problems and unfinished interfaces present in 4.2BSD [Leffler84] [McKusick85]. The Computer Systems Research Group at Berkeley has now embarked on a new development phase to update other major components of the system, as well as to offer new functionality. There are six major ongoing projects. The first is to develop an OSI network protocol suite and to integrate existing ISO applications into Berkeley UNIX. The second is to develop and support an interface compliant with the P1003.1 POSIX standard recently approved by the IEEE. The third is to refine the TCP/IP networking to improve its performance and limit congestion on slow and/or lossy networks. The fourth is to provide a standard interface to file systems so that multiple local and remote file systems can be supported, much as multiple networking protocols are supported by 4.3BSD. The fifth is to replace the 4.3BSD virtual memory system in order to support mapped files, larger memories, copy-on-write on fork and I/O, and better portability. The final project is to evaluate alternate access control mechanisms and audit the existing security features of the system, particularly with respect to network services. Other areas of work include multi-architecture support, a general purpose kernel memory allocator, disk labels, and extensions to the 4.2BSD fast filesystem.
Several of these projects are nearing completion, while others are in the prototype stage. We are planning to provide an informal test release sometime this summer to interested developers containing most of these changes. After incorporating feedback and refinements from the testers, they will appear in the next full Berkeley release, which is typically made about a year after the test release.
Recently Completed Projects
There have been several changes in the system that were included in the 4.3BSD Tahoe release. Multi-architecture support
Support has been added for the DEC VAX 8600/8650, VAX 8200/8250, MicroVAXII and MicroVAXIII.
The largest change has been the incorporation of support for the first non-VAX processor, the CCI Power 6/32 and 6/32SX. (This addition also supports the Harris HCX-7 and HCX-9, as well as the Sperry 7000/40 and ICL machines.) The Power 6 version of 4.3BSD is largely based on the compilers and device drivers done for CCI's 4.2BSD UNIX, and is otherwise similar to the VAX release of 4.3BSD. The entire source tree, including all kernel and user-level sources, has been merged using a structure that will easily accommodate the addition of other processor families. A MIPS R2000 has been donated to us, making the MIPS architecture a likely candidate for inclusion into a future BSD release. Kernel Memory Allocator
The 4.3BSD UNIX kernel used 10 different memory allocation mechanisms, each designed for the particular needs of the utilizing subsystem. These mechanisms have been replaced by a general purpose dynamic memory allocator that can be used by all of the kernel subsystems. The design of this allocator takes advantage of known memory usage patterns in the UNIX kernel and a hybrid strategy that is time-efficient for small allocations and space-efficient for large allocations. This allocator replaces the multiple memory allocation interfaces with a single easy-to-program interface, results in more efficient use of global memory by eliminating partitioned and specialized memory pools, and is quick enough (approximately 15 VAX instructions) that no performance loss is observed relative to the current implementations. [McKusick88]. Disk Labels
During the work on the CCI machine, it became obvious that disk geometry and filesystem layout information must be stored on each disk in a pack label. Disk labels were implemented for the CCI disks and for the most common types of disk controllers on the VAX. A utility was written to create and maintain the disk information, and other user-level programs that use such information now obtain it from the disk label. The use of this facility has allowed improvements in the file system's knowledge of irregular disk geometries such as track-to-track skew. Fat Fast File System
The 4.2 fast file sytem [McKusick84] contained several statically sized structures, imposing limits on the number of cylinders per cylinder group, inodes per cylinder group, and number of distinguished rotational positions. The new ``fat'' filesystem allows these limits to be set at filesystem creation time. Old kernels will treat the new filesystems as read-only, and new kernels will accomodate both formats. The filesystem check facility, fsck, has also been modified to check either type.
Current UNIX Research at Berkeley
Since the release of 4.3BSD in mid 1986, we have begun work on several new major areas of research. Our goal is to apply leading edge research ideas into a stable and reliable implementation that solves current problems in operating systems development. OSI network protocol development
The network architecture of 4.2BSD was designed to accommodate multiple network protocol families and address formats, and an implementation of the ISO OSI network protocols should enter into this framework without much difficulty. The current system implements the OSI Connectionless Network Protocol (CLNP), the End System-Intermediate System protocol (ES-IS), and OSI transport class 4 (TP-4). Device drivers for 802.3 are included, and device drivers for X.25 and possibly 802.5 interfaces are expected to be added. We will also incorporate into the Berkeley Software Distribution an updated ISO Development Environment (ISODE) featuring International Standard (IS) versions of OSI applications. ISODE implements the session and presentation layers of the OSI protocol suite, and includes an implementation of the file transfer protocol (FTAM) and the OSI Virtual Terminal protocol (VT). It is also possible that an X.400 implementation now being done at University College, London and the University of Nottingham will be available for testing and distribution.
This implementation is comprised of four areas.
Berkeley has been involved in the development of the IEEE POSIX standards since the last draft of the P1003.1 system interface standard. Since then, we have been parcipating in the working groups of P1003.2 (shell and application utility interface), P1003.4 (real-time and other system extensions), P1003.6 (security), P1003.7 (system administration), P1003.8 (transparent remote file access) and P1003.12 (process-to-process network interface).
The IEEE published the POSIX P1003.1 standard in late 1988. POSIX-related changes to the BSD system have included a new terminal driver, support for POSIX sessions and job control, expanded signal functionality, restructured directory access routines, and new set-user and set-group id facilities. We have an implementation of the POSIX driver with extensions similar to those in the old Berkeley terminal driver. A compatibility module provides binary compatibility with applications developed for the old Berkeley terminal driver. We also have an implementation of the 4.2BSD-based POSIX job control facility.
The P1003.2 draft is currently being voted on by the IEEE P1003.2 balloting group. Berkeley is particularly interested in the results of this standard, as it will profoundly influence the user environment. The other groups are in comparatively early phases, with drafts coming to ballot sometime in the 90's. Berkeley will continue to participate in these groups, and move in the near future toward a P1003.1 and P1003.2 compliant system. We have many of the utilities outlined in the current P1003.2 draft already implemented, and have other parties willing to contribute additional implementations. Improvements to the TCP/IP Networking Protocols
The Internet and the Berkeley collection of local-area networks have both grown at high rates in the last several years. The Bay Area Regional Research Network (BARRNet), connecting several UC campuses, Stanford and NASA-Ames has recently become operational, increasing the complexity of the network connectivity. Both Internet and local routing algorithms are showing the strain of continued growth. We have made several changes in the local routing algorithm to keep accommodating the current topology, and are participating in the development of new routing algorithms and standard protocols.
Work in collaboration with Van Jacobson of the Lawrence Berkeley Laboratory has led to the design and implementation of several new algorithms for TCP that improve throughput on both local and long-haul networks while reducing unnecessary retransmission. The improvement is especially striking when connections must traverse slow and/or lossy networks. The new algorithms include ``slow-start,'' a technique for opening the TCP flow control window slowly and using the returning stream of acknowledgements as a clock to drive the connection at the highest speed tolerated by the intervening network. A modification of this technique allows the sender to dynamically modify the send window size to adjust to changing network conditions. In addition, the round-trip timer has been modified to estimate the variance in round-trip time, thus allowing earlier retransmission of lost packets with less spurious retransmission due to increasing network delay. Along with a scheme proposed by Phil Karn of Bellcore, these changes reduce unnecessary retransmission over difficult paths such as Satnet by nearly two orders of magnitude while improving throughput dramatically.
The current TCP implementation is now available via the network and as a standard Berkeley distribution unencumbered by any commercial licensing. We are continuing to refine the TCP and IP implementations using the the NSF network, BARRNet, ARPANET/Milnet, and local campus nets as testbeds. In addition, we are incorporating applicable algorithms from this work into the TP-4 protocol implementation. Toward a Compatible File System Interface
The most critical shortcoming of the 4.3BSD UNIX system was in the area of distributed file systems. As with networking protocols, there is no single distributed file system that provides sufficient speed and functionality for all problems. It is frequently necessary to support several different remote file system protocols, just as it is necessary to run several different network protocols.
As network or remote file systems have been implemented for UNIX, several stylized interfaces between the file system implementation and the rest of the kernel have been developed. Among these are Sun Microsystems' Virtual File System interface (VFS) using vnodes [Sandburg85] [Kleiman86], Digital Equipment's Generic File System (GFS) architecture [Rodriguez86], AT&T's File System Switch (FSS) [Rifkin86], the LOCUS distributed file system [Walker85], and Masscomp's extended file system [Cole85]. Other remote file systems have been implemented in research or university groups for internal use, notably the network file system in the Eighth Edition UNIX system [Weinberger84] and two different file systems used at Carnegie Mellon University [Satyanarayanan85]. Numerous other remote file access methods have been devised for use within individual UNIX processes, many of them by modifications to the C I/O library similar to those in the Newcastle Connection [Brownbridge82].
Each design attempts to isolate file system-dependent details below a generic interface and to provide a framework within which new file systems may be incorporated. However, each of these interfaces is different from and incompatible with the others. Each addresses somewhat different design goals, having been based on a different version of UNIX, having targeted a different set of file systems with varying characteristics, and having selected a different set of file system primitive operations.
Our effort in this area is aimed at providing a common framework to support these different distributed file systems simultaneously rather than to simply implement yet another protocol. This requires a detailed study of the existing protocols, and discussion with their implementors to determine whether they could modify their implementation to fit within our proposed framework. We have studied the various file system interfaces to determine their generality, completeness, robustness, efficiency, and aesthetics. We have implemeneted a file system interface that we believe includes the best features of each of the previous implementations. This work and the rationale underlying its development have been presented to major software vendors as an early step toward convergence on a standard compatible file system interface. Briefly, the proposal adopts the 4.3BSD calling convention for file name lookup but otherwise is closely related to Sun's VFS and DEC's GFS. [Karels86].
Non-propietary client and server implementations of Sun's NFS protocol have been produced in this new framework at the University of Guelph. This NFS version has been tested successfully at the most recent ``Connectathon,'' and both the filesystem interface and NFS will be included in the Open Software Foundation's first system release, OSF/1.
One additional file system has been added under the new interface, a virtual-memory-resident file system (MFS). This file system is used for temporary file storage in /tmp, and greatly speeds ``I/O'' for creation and use of temporary files. Virtual Memory
The 4.3BSD virtual memory system is a direct descendent of the original Berkeley virtual memory system for 3BSD, released in 1979. The state of the art in virtual memory hardware and software has changed tremendously in that time. There are several problems with the current system:
Most of these problems have been addressed in the virtual memory system in the Mach system from Carnegie-Mellon University. That virtual memory system has been ported to 4.3BSD at the University of Utah, and work has begun to convert to the BSD mmap\^ () interface for mapped files and shared memory. That virtual memory system is expected to be the basis for the 4.4BSD system, and its integration and development at Berkeley will begin in the next few months. System Security
The invasion of the DARPA Internet by a quickly reproducing ``worm'' highlighted the need for a thorough review of the access safeguards built into the system. Until now, we have taken a passive approach to dealing with weaknesses in the system access mechanisms, rather than actively searching for possible weaknesses. When we are notified of a problem or loophole in a system utility by one of our users, we have a well defined procedure for fixing the problem and expeditiously disseminating the fix to the BSD mailing list. This procedure has proven itself to be effective in solving known problems as they arise (witness its success in handling the recent worm). However, we feel that it would be useful to take a more active role in identifying problems before they are reported (or exploited). We will make a complete audit of the system utilities and network servers to find unintended system access mechanisms.
As a part of the work to make the system more resistant to attack from local users or via the network, it will be necessary to produce additional documentation on the configuration and operation of the system. This documentation will cover such topics as file and directory ownership and access, network and server configuration, and control of privileged operations such as file system backups.
We are investigating the addition of access control lists (ACLs) for filesystem objects. ACLs provide a much finer granularity of control over file access permissions than the current discretionary access control mechanism (mode bits). Furthermore, they are necessary in environments where C2 level security or better, as defined in the DoD TCSEC [DoD83], is required. The POSIX P1003.6 security group has made notable progress in determining how an ACL mechanism should work, and several vendors have implemented ACLs for their commercial systems. Berkeley will investigate the existing implementations and determine how to best integrate ACLs with the existing mechanism.
A major shortcoming of the present system is that authentication over the network is based solely on the privileged port mechanism between trusting hosts and users. Although privileged ports can only be created by processes running as root on a UNIX system, such processes are easy for a workstation user to obtain; they simply reboot their workstation in single user mode. Thus, a better authentication mechanism is needed. At present, we believe that the MIT Kerberos authentication server [Steiner88] provides the best solution to this problem. We have integrated the current version of Kerberos into our current system and are beginning installation in a larger part of the Berkeley Computer Science Division. This integration includes the addition of the authentication mechanism in utilities such as telnet, login, remote shell, etc., as well as conversion aids. We plan to add support for telnet (eventually replacing rlogin), the X window system, and the mail system within an authentication domain (a Kerberos realm). We hope to replace the existing password authentication completely with the network authentication system. Other projects
There are numerous other projects in progress within CSRG and in other groups that are targetted for inclusion in 4.4BSD.
In addition to the system interface changes made for POSIX compliance, the system call interface is being updated in several other areas. Minor extensions to the socket interface facilitate use of OSI-specific features. Other calls are being consolidated and updated. Full compatibility with the older interfaces will be retained for one release cycle.
Support for the Hewlett-Packard 9000/300 series of workstations and servers has been added by the University of Utah. Another group has done a port to AT-style i386 systems and is adding support for various peripherals. Ports to the line of servers from MIPS and to the MIPS-based DEC workstations are in progress as well.
ReferencesCole, C.T., P.B. Flinn, A.B. Atlas, ``An Implementation of an Extended File System for UNIX,'' Usenix Conference Proceedings, pp. 131-150, June, 1985.
Department of Defense, ``Trusted Computer System Evaluation Criteria,'' CSC-STD-001-83, DoD Computer Security Center, August, 1983.