求助！为什么会Server is currently unavailable or down for maintenance!!

a project model for the FreeBSD Project.7z

A project model for the FreeBSD Project Niklas Saers Copyright © 2002-2005 Niklas Saers [ Split HTML / Single HTML ] Table of Contents Foreword 1 Overview 2 Definitions 2.1. Activity 2.2. Process 2.3. Hat 2.4. Outcome 2.5. FreeBSD 3 Organisational structure 4 Methodology model 4.1. Development model 4.2. Release branches 4.3. Model summary 5 Hats 5.1. General Hats 5.1.1. Contributor 5.1.2. Committer 5.1.3. Core Team 5.1.4. Maintainership 5.2. Official Hats 5.2.1. Documentation project manager 5.2.2. CVSup Mirror Site Coordinator 5.2.3. Internationalisation 5.2.4. Postmaster 5.2.5. Quality Assurance 5.2.6. Release Coordination 5.2.7. Public Relations & Corporate Liaison 5.2.8. Security Officer 5.2.9. Source Repository Manager 5.2.10. Election Manager 5.2.11. Web site Management 5.2.12. Ports Manager 5.2.13. Standards 5.2.14. Core Secretary 5.2.15. GNATS Administrator 5.2.16. Bugmeister 5.2.17. Donations Liaison Officer 5.2.18. Admin 5.3. Process dependent hats 5.3.1. Report originator 5.3.2. Bugbuster 5.3.3. Mentor 5.3.4. Vendor 5.3.5. Reviewers 5.3.6. CVSup Mirror Site Admin 6 Processes 6.1. Adding new and removing old committers 6.2. Adding/Removing an official CVSup Mirror 6.3. Committing code 6.4. Core election 6.5. Development of new features 6.6. Maintenance 6.7. Problem reporting 6.8. Reacting to misbehaviour 6.9. Release engineering 7 Tools 7.1. Concurrent Versions System &#40;CVS&#41; 7.2. CVSup 7.3. GNATS 7.4. Mailman 7.5. Perforce 7.6. Pretty Good Privacy 7.7. Secure Shell 8 Sub-projects 8.1. The Ports Subproject 8.2. The FreeBSD Documentation Project References List of Figures 3-1. The FreeBSD Project's structure 3-2. The FreeBSD Project's structure with committers in categories 4-1. Jørgenssen's model for change integration 4-2. The FreeBSD release tree 4-3. The overall development model 5-1. Overview of official hats 6-1. Process summary: adding a new committer 6-2. Process summary: removing a committer 6-3. Process summary: adding a CVSup mirror 6-4. Process summary: A committer commits code 6-5. Process summary: A contributor commits code 6-6. Process summary: Core elections 6-7. Jørgenssen's model for change integration 6-8. Process summary: problem reporting 6-9. Process summary: release engineering 8-1. Number of ports added between 1996 and 2005 Foreword Up until now, the FreeBSD project has released a number of described techniques to do different parts of work. However, a project model summarising how the project is structured is needed because of the increasing amount of project members. [1] This paper will provide such a project model and is donated to the FreeBSD Documentation project where it can evolve together with the project so that it can at any point in time reflect the way the project works. It is based on [Saers, 2003]. I would like to thank the following people for taking the time to explain things that were unclear to me and for proofreading the document. Andrey A. Chernov <ache@freebsd.org> Bruce A. Mah <bmah@freebsd.org> Dag-Erling Smørgrav <des@freebsd.org> Giorgos Keramidas<keramida@freebsd.org> Ingvil Hovig <ingvil.hovig@skatteetaten.no> Jesper Holck<jeh.inf@cbs.dk> John Baldwin <jhb@freebsd.org> John Polstra <jdp@freebsd.org> Kirk McKusick <mckusick@freebsd.org> Mark Linimon <linimon@freebsd.org> Marleen Devos Niels Jørgenssen<nielsj@ruc.dk> Nik Clayton <nik@freebsd.org> Poul-Henning Kamp <phk@freebsd.org> Simon L. Nielsen <simon@freebsd.org> Chapter 1 Overview A project model is a means to reduce the communications overhead in a project. As shown by [Brooks, 1995], increasing the number of project participants increases the communication in the project exponentionally. FreeBSD has during the past few year increased both its mass of active users and committers, and the communication in the project has risen accordingly. This project model will serve to reduce this overhead by providing an up-to-date description of the project. During the Core elections in 2002, Mark Murray stated “I am opposed to a long rule-book, as that satisfies lawyer-tendencies, and is counter to the technocentricity that the project so badly needs.” [FreeBSD, 2002B]. This project model is not meant to be a tool to justify creating impositions for developers, but as a tool to facilitate coordination. It is meant as a description of the project, with an overview of how the different processes are executed. It is an introduction to how the FreeBSD project works. The FreeBSD project model will be described as of July 1st, 2004. It is based on the Niels Jørgensen's paper [Jørgensen, 2001], FreeBSD's official documents, discussions on FreeBSD mailing lists and interviews with developers. After providing definitions of terms used, this document will outline the organisational structure (including role descriptions and communication lines), discuss the methodology model and after presenting the tools used for process control, it will present the defined processes. Finally it will outline major sub-projects of the FreeBSD project. [FreeBSD, 2002A, Section 1.2 and 1.3] give the vision and the architectural guidelines for the project. The vision is “To produce the best UNIX-like operating system package possible, with due respect to the original software tools ideology as well as usability, performance and stability.” The architectural guidelines help determine whether a problem that someone wants to be solved is within the scope of the project Chapter 2 Definitions 2.1. Activity An “activity” is an element of work performed during the course of a project [PMI, 2000]. It has an output and leads towards an outcome. Such an output can either be an input to another activity or a part of the process' delivery. 2.2. Process A “process” is a series of activities that lead towards a particular outcome. A process can consist of one or more sub-processes. An example of a process is software design. 2.3. Hat A “hat” is synonymous with role. A hat has certain responsibilities in a process and for the process outcome. The hat executes activities. It is well defined what issues the hat should be contacted about by the project members and people outside the project. 2.4. Outcome An “outcome” is the final output of the process. This is synonymous with deliverable, that is defined as “any measurable, tangible, verifiable outcome, result or item that must be produced to complete a project or part of a project. Often used more narrowly in reference to an external deliverable, which is a deliverable that is subject to approval by the project sponsor or customer” by [PMI, 2000]. Examples of outcomes are a piece of software, a decision made or a report written. 2.5. FreeBSD When saying “FreeBSD” we will mean the BSD derivative UNIX-like operating system FreeBSD, whereas when saying “the FreeBSD Project” we will mean the project organisation. Chapter 3 Organisational structure While no-one takes ownership of FreeBSD, the FreeBSD organisation is divided into core, committers and contributors and is part of the FreeBSD community that lives around it. Figure 3-1. The FreeBSD Project's structure Number of committers has been determined by going through CVS logs from January 1st, 2004 to December 31st, 2004 and contributors by going through the list of contributions and problem reports. The main resource in the FreeBSD community is its developers: the committers and contributors. It is with their contributions that the project can move forward. Regular developers are referred to as contributors. As by January 1st, 2003, there are an estimated 5500 contributors on the project. Committers are developers with the privilege of being able to commit changes. These are usually the most active developers who are willing to spend their time not only integrating their own code but integrating code submitted by the developers who do not have this privilege. They are also the developers who elect the core team, and they have access to closed discussions. The project can be grouped into four distinct separate parts, and most developers will focus their involvement in one part of FreeBSD. The four parts are kernel development, userland development, ports and documentation. When referring to the base system, both kernel and userland is meant. This split changes our triangle to look like this: Figure 3-2. The FreeBSD Project's structure with committers in categories Number of committers per area has been determined by going through CVS logs from January 1st, 2004 to December 31st, 2004. Note that many committers work in multiple areas, making the total number higher than the real number of committers. The total number of committers at that time was 269. Committers fall into three groups: committers who are only concerned with one area of the project (for instance file systems), committers who are involved only with one sub-project and committers who commit to different parts of the code, including sub-projects. Because some committers work on different parts, the total number in the committers section of the triangle is higher than in the above triangle. The kernel is the main building block of FreeBSD. While the userland applications are protected against faults in other userland applications, the entire system is vulnerable to errors in the kernel. This, combined with the vast amount of dependencies in the kernel and that it is not easy to see all the consequences of a kernel change, demands developers with a relative full understanding of the kernel. Multiple development efforts in the kernel also requires a closer coordination than userland applications do. The core utilities, known as userland, provide the interface that identifies FreeBSD, both user interface, shared libraries and external interfaces to connecting clients. Currently, 162 people are involved in userland development and maintenance, many being maintainers for their own part of the code. Maintainership will be discussed in the Maintainership section. Documentation is handled by The FreeBSD Documentation Project and includes all documents surrounding the FreeBSD project, including the web pages. There were during 2004 101 people making commits to the FreeBSD Documentation Project. Ports is the collection of meta-data that is needed to make software packages build correctly on FreeBSD. An example of a port is the port for the web-browser Mozilla. It contains information about where to fetch the source, what patches to apply and how, and how the package should be installed on the system. This allows automated tools to fetch, build and install the package. As of this writing, there are more than 12600 ports available. [2] , ranging from web servers to games, programming languages and most of the application types that are in use on modern computers. Ports will be discussed further in the section The Ports Subproject. Chapter 4 Methodology model 4.1. Development model There is no defined model for how people write code in FreeBSD. However, Niels Jørgenssen has suggested a model of how written code is integrated into the project. Figure 4-1. Jørgenssen's model for change integration The “development release” is the FreeBSD-CURRENT ("-CURRENT") branch and the “production release” is the FreeBSD-STABLE branch ("-STABLE") [Jørgensen, 2001]. This is a model for one change, and shows that after coding, developers seek community review and try integrating it with their own systems. After integrating the change into the development release, called FreeBSD-CURRENT, it is tested by many users and developers in the FreeBSD community. After it has gone through enough testing, it is merged into the production release, called FreeBSD-STABLE. Unless each stage is finished successfully, the developer needs to go back and make modifications in the code and restart the process. To integrate a change with either -CURRENT or -STABLE is called making a commit. Jørgensen found that most FreeBSD developers work individually, meaning that this model is used in parallel by many developers on the different ongoing development efforts. A developer can also be working on multiple changes, so that while he is waiting for review or people to test one or more of his changes, he may be writing another change. As each commit represents an increment, this is a massively incremental model. The commits are in fact so frequent that during one year [3] , 85427 commits were made, making a daily average of 233 commits. Within the “code” bracket in Jørgensen's figure, each programmer has his own working style and follows his own development models. The bracket could very well have been called “development” as it includes requirements gathering and analysis, system and detailed design, implementation and verification. However, the only output from these stages is the source code or system documentation. From a stepwise model's perspective (such as the waterfall model), the other brackets can be seen as further verification and system integration. This system integration is also important to see if a change is accepted by the community. Up until the code is committed, the developer is free to choose how much to communicate about it to the rest of the project. In order for -CURRENT to work as a buffer (so that bright ideas that had some undiscovered drawbacks can be backed out) the minimum time a commit should be in -CURRENT before merging it to -STABLE is 3 days. Such a merge is referred to as an MFC (Merge From Current). It is important to notice the word “change”. Most commits do not contain radical new features, but are maintenance updates. The only exceptions from this model are security fixes and changes to features that are deprecated in the -CURRENT branch. In these cases, changes can be committed directly to the -STABLE branch. In addition to many people working on the project, there are many related projects to the FreeBSD Project. These are either projects developing brand new features, sub-projects or projects whose outcome is incorporated into FreeBSD [4]. These projects fit into the FreeBSD Project just like regular development efforts: they produce code that is integrated with the FreeBSD Project. However, some of them (like Ports and Documentation) have the privilege of being applicable to both branches or commit directly to both -CURRENT and -STABLE. There is no standards to how design should be done, nor is design collected in a centralised repository. The main design is that of 4.4BSD. [5] As design is a part of the “Code” bracket in Jørgenssen's model, it is up to every developer or sub-project how this should be done. Even if the design should be stored in a central repository, the output from the design stages would be of limited use as the differences of methodologies would make them poorly if at all interoperable. For the overall design of the project, the project relies on the sub-projects to negotiate fit interfaces between each other rather than to dictate interfacing. 4.2. Release branches The releases of FreeBSD is best illustrated by a tree with many branches where each major branch represents a major version. Minor versions are represented by branches of the major branches. In the following release tree, arrows that follow one-another in a particular direction represent a branch. Boxes with full lines and diamonds represent official releases. Boxes with dotted lines represent the development branch at that time. Security branches are represented by ovals. Diamonds differ from boxes in that they represent a fork, meaning a place where a branch splits into two branches where one of the branches becomes a sub-branch. For example, at 4.0-RELEASE the 4.0-CURRENT branch split into 4-STABLE and 5.0-CURRENT. At 4.5-RELEASE, the branch forked off a security branch called RELENG_4_5. Figure 4-2. The FreeBSD release tree The latest -CURRENT version is always referred to as -CURRENT, while the latest -STABLE release is always referred to as -STABLE. In this figure, -STABLE refers to 4-STABLE while -CURRENT refers to 5.0-CURRENT following 5.0-RELEASE. [FreeBSD, 2002E] A “major release” is always made from the -CURRENT branch. However, the -CURRENT branch does not need to fork at that point in time, but can focus on stabilising. An example of this is that following 3.0-RELEASE, 3.1-RELEASE was also a continuation of the -CURRENT-branch, and -CURRENT did not become a true development branch until this version was released and the 3-STABLE branch was forked. When -CURRENT returns to becoming a development branch, it can only be followed by a major release. 5-STABLE is predicted to be forked off 5.0-CURRENT at around 5.3-RELEASE. It is not until 5-STABLE is forked that the development branch will be branded 6.0-CURRENT. A “minor release” is made from the -CURRENT branch following a major release, or from the -STABLE branch. Following and including, 4.3-RELEASE[6], when a minor release has been made, it becomes a “security branch”. This is meant for organisations that do not want to follow the -STABLE branch and the potential new/changed features it offers, but instead require an absolutely stable environment, only updating to implement security updates. [7] Each update to a security branch is called a “patchlevel”. For every security enhancement that is done, the patchlevel number is increased, making it easy for people tracking the branch to see what security enhancements they have implemented. In cases where there have been especially serious security flaws, an entire new release can be made from a security branch. An example of this is 4.6.2-RELEASE. 4.3. Model summary To summarise, the development model of FreeBSD can be seen as the following tree: Figure 4-3. The overall development model The tree of the FreeBSD development with ongoing development efforts and continuous integration. The tree symbolises the release versions with major versions spawning new main branches and minor versions being versions of the main branch. The top branch is the -CURRENT branch where all new development is integrated, and the -STABLE branch is the branch directly below it. Clouds of development efforts hang over the project where developers use the development models they see fit. The product of their work is then integrated into -CURRENT where it undergoes parallel debugging and is finally merged from -CURRENT into -STABLE. Security fixes are merged from -STABLE to the security branches. Chapter 5 Hats Many committers have a special area of responsibility. These roles are called hats [Losh, 2002]. These hats can be either project roles, such as public relations officer, or maintainer for a certain area of the code. Because this is a project where people give voluntarily of their spare time, people with assigned hats are not always available. They must therefore appoint a deputy that can perform the hat's role in his or her absence. The other option is to have the role held by a group. Many of these hats are not formalised. Formalised hats have a charter stating the exact purpose of the hat along with its privileges and responsibilities. The writing of such charters is a new part of the project, and has thus yet to be completed for all hats. These hat descriptions are not such a formalisation, rather a summary of the role with links to the charter where available and contact addresses, 5.1. General Hats 5.1.1. Contributor A Contributor contributes to the FreeBSD project either as a developer, as an author, by sending problem reports, or in other ways contributing to the progress of the project. A contributor has no special privileges in the FreeBSD project. [FreeBSD, 2002F] 5.1.2. Committer A person who has the required privileges to add his code or documentation to the repository. A committer has made a commit within the past 12 months. [FreeBSD, 2000A] An active committer is a committer who has made an average of one commit per month during that time. It is worth noting that there are no technical barriers to prevent someone, once having gained commit privileges to the main- or a sub-project, to make commits in parts of that project's source the committer did not specifically get permission to modify. However, when wanting to make modifications to parts a committer has not been involved in before, he/she should read the logs to see what has happened in this area before, and also read the MAINTAINER file to see if the maintainer of this part has any special requests on how changes in the code should be made 5.1.3. Core Team The core team is elected by the committers from the pool of committers and serves as the board of directors of the FreeBSD project. It promotes active contributors to committers, assigns people to well-defined hats, and is the final arbiter of decisions involving which way the project should be heading. As by July 1st, 2004, core consisted of 9 members. Elections are held every two years. 5.1.4. Maintainership Maintainership means that that person is responsible for what is allowed to go into that area of the code and has the final say should disagreements over the code occur. This involves involves proactive work aimed at stimulating contributions and reactive work in reviewing commits. With the FreeBSD source comes the MAINTAINERS file that contains a one-line summary of how each maintainer would like contributions to be made. Having this notice and contact information enables developers to focus on the development effort rather than being stuck in a slow correspondence should the maintainer be unavailable for some time. If the maintainer is unavailable for an unreasonably long period of time, and other people do a significant amount of work, maintainership may be switched without the maintainer's approval. This is based on the stance that maintainership should be demonstrated, not declared. Maintainership of a particular piece of code is a hat that is not held as a group. 5.2. Official Hats The official hats in the FreeBSD Project are hats that are more or less formalised and mainly administrative roles. They have the authority and responsibility for their area. The following illustration shows the responsibility lines. After this follows a description of each hat, including who it is held by. Figure 5-1. Overview of official hats All boxes consist of groups of committers, except for the dotted boxes where the holders are not necessarily committers. The flattened circles are sub-projects and consist of both committers and non-committers of the main project. 5.2.1. Documentation project manager The FreeBSD Documentation Project architect is responsible for defining and following up documentation goals for the committers in the Documentation project. Hat held by: The DocEng team <doceng@FreeBSD.org>. The DocEng Charter. 5.2.2. CVSup Mirror Site Coordinator The CVSup Mirror Site Coordinator coordinates all the CVSup Mirror Site Admins to ensure that they are distributing current versions of the software, that they have the capacity to update themselves when major updates are in progress, and making it easy for the general public to find their closest CVSup mirror. Hat currently held by: John Polstra <jdp@FreeBSD.org>. 5.2.3. Internationalisation The Internationalisation hat is responsible for coordinating the localisation efforts of the FreeBSD kernel and userland utilities. The translation effort are coordinated by The FreeBSD Documentation Project. The Internationalisation hat should suggest and promote standards and guidelines for writing and maintaining the software in a fashion that makes it easier to translate. Hat currently available. 5.2.4. Postmaster The Postmaster is responsible for mail being correctly delivered to the committers' email address. He is also responsible for ensuring that the mailing lists work and should take measures against possible disruptions of mail such as having troll-, spam- and virus-filters. Hat currently held by: David Wolfskill <dhw@FreeBSD.org>. 5.2.5. Quality Assurance The responsibilities of this role are to manage the quality assurance measures. Hat currently held by: Robert Watson <rwatson@FreeBSD.org>. 5.2.6. Release Coordination The responsibilities of the Release Engineering Team are Setting, publishing and following a release schedule for official releases Documenting and formalising release engineering procedures Creation and maintenance of code branches Coordinating with the Ports and Documentation teams to have an updated set of packages and documentation released with the new releases Coordinating with the Security team so that pending releases are not affected by recently disclosed vulnerabilities. Further information about the development process is available in the release engineering section. Hat held by: the Release Engineering team <re@FreeBSD.org>, currently headed by Murray Stokely <murray@FreeBSD.org>. The Release Engineering Charter. 5.2.7. Public Relations & Corporate Liaison The Public Relations & Corporate Liaison's responsibilities are: Making press statements when happenings that are important to the FreeBSD Project happen. Being the official contact person for corporations that are working close with the FreeBSD Project. Take steps to promote FreeBSD within both the Open Source community and the corporate world. Handle the “freebsd-advocacy” mailing list. This hat is currently not occupied. 5.2.8. Security Officer The Security Officer's main responsibility is to coordinate information exchange with others in the security community and in the FreeBSD project. The Security Officer is also responsible for taking action when security problems are reported and promoting proactive development behaviour when it comes to security. Because of the fear that information about vulnerabilities may leak out to people with malicious intent before a patch is available, only the Security Officer, consisting of an officer, a deputy and two Core team members, receive sensitive information about security issues. However, to create or implement a patch, the Security Officer has the Security Officer Team <security-team@FreeBSD.org> to help do the work. Hat held by: the Security Officer <security-officer@FreeBSD.org>, currently headed by Colin Percival <cperciva@FreeBSD.org>. The Security Officer and The Security Officer Team's charter. 5.2.9. Source Repository Manager The Source Repository Manager is the only one who is allowed to directly modify the repository without using the CVS tool. It is his/her responsibility to ensure that technical problems that arise in the repository are resolved quickly. The source repository manager has the authority to back out commits if this is necessary to resolve a CVS technical problem. Hat held by: the Source Repository Manager <cvs@FreeBSD.org>, currently headed by Peter Wemm <peter@FreeBSD.org>. 5.2.10. Election Manager The Election Manager is responsible for the Core election process. The manager is responsible for running and maintaining the election system, and is the final authority should minor unforseen events happen in the election process. Major unforseen events have to be discussed with the Core team Hat held only during elections. 5.2.11. Web site Management The Web site Management hat is responsible for coordinating the rollout of updated web pages on mirrors around the world, for the overall structure of the primary web site and the system it is running upon. The management needs to coordinate the content with The FreeBSD Documentation Project and acts as maintainer for the “www” tree. Hat held by: the FreeBSD Webmasters <www@FreeBSD.org>. 5.2.12. Ports Manager The Ports Manager acts as a liaison between The Ports Subproject and the core project, and all requests from the project should go to the ports manager. Hat held by: the Ports Management Team <portmgr@FreeBSD.org>, 5.2.13. Standards The Standards hat is responsible for ensuring that FreeBSD complies with the standards it is committed to , keeping up to date on the development of these standards and notifying FreeBSD developers of important changes that allows them to take a proactive role and decrease the time between a standards update and FreeBSD's compliancy. Hat currently held by: Garrett Wollman <wollman@FreeBSD.org>. 5.2.14. Core Secretary The Core Secretary's main responsibility is to write drafts to and publish the final Core Reports. The secretary also keeps the core agenda, thus ensuring that no balls are dropped unresolved. Hat currently held by: Joel Dahl <joel@FreeBSD.org>. 5.2.15. GNATS Administrator The GNATS Administrator is responsible for ensuring that the maintenance database is in working order, that the entries are correctly categorised and that there are no invalid entries. Hat currently held by: Ceri Davies <ceri@FreeBSD.org> and Mark Linimon <linimon@FreeBSD.org>. 5.2.16. Bugmeister The Bugmeister is the person in charge of the problem report group. Hat currently held by: Ceri Davies <ceri@FreeBSD.org> and Mark Linimon <linimon@FreeBSD.org>. 5.2.17. Donations Liaison Officer The task of the donations liason officer is to match the developers with needs with people or organisations willing to make a donation. The Donations Liason Charter is available here Hat held by: the Donations Liaison Office <donations@FreeBSD.org>, currently headed by Michael W. Lucas <mwlucas@FreeBSD.org>. 5.2.18. Admin (Also called “FreeBSD Cluster Admin”) The admin team consists of the people responsible for administrating the computers that the project relies on for its distributed work and communication to be synchronised. It consists mainly of those people who have physical access to the servers. Hat held by: the Admin team <admin@FreeBSD.org>, currently headed by Mark Murray <markm@FreeBSD.org> 5.3. Process dependent hats 5.3.1. Report originator The person originally responsible for filing a Problem Report. 5.3.2. Bugbuster A person who will either find the right person to solve the problem, or close the PR if it is a duplicate or otherwise not an interesting one. 5.3.3. Mentor A mentor is a committer who takes it upon him/her to introduce a new committer to the project, both in terms of ensuring the new committers setup is valid, that the new committer knows the available tools required in his/her work and that the new committer knows what is expected of him/her in terms of behaviour. 5.3.4. Vendor The person(s) or organisation whom external code comes from and whom patches are sent to. 5.3.5. Reviewers People on the mailing list where the request for review is posted. 5.3.6. CVSup Mirror Site Admin A CVSup Mirror Site Admin has accesses to a server that he/she uses to mirror the CVS repository. The admin works with the CVSup Mirror Site Coordinator to ensure the site remains up-to-date and is following the general policy of official mirror sites. Chapter 6 Processes The following section will describe the defined project processes. Issues that are not handled by these processes happen on an ad-hoc basis based on what has been customary to do in similar cases. 6.1. Adding new and removing old committers The Core team has the responsibility of giving and removing commit privileges to contributors. This can only be done through a vote on the core mailing list. The ports and documentation sub-projects can give commit privileges to people working on these projects, but have to date not removed such privileges. Normally a contributor is recommended to core by a committer. For contributors or outsiders to contact core asking to be a committer is not well thought of and is usually rejected. If the area of particular interest for the developer potentially overlaps with other committers' area of maintainership, the opinion of those maintainers is sought. However, it is frequently this committer that recommends the developer. When a contributor is given committer status, he is assigned a mentor. The committer who recommended the new committer will, in the general case, take it upon himself to be the new committers mentor. When a contributor is given his commit bit, a PGP-signed email is sent from either Core Secretary, Ports Manager or nik@freebsd.org to both admins@freebsd.org, the assigned mentor, the new committer and core confirming the approval of a new account. The mentor then gathers a password line, SSH 2 public key and PGP key from the new committer and sends them to Admin. When the new account is created, the mentor activates the commit bit and guides the new committer through the rest of the initial process. Figure 6-1. Process summary: adding a new committer When a contributor sends a piece of code, the receiving committer may choose to recommend that the contributor is given commit privileges. If he recommends this to core, they will vote on this recommendation. If they vote in favour, a mentor is assigned the new committer and the new committer has to email his details to the administrators for an account to be created. After this, the new committer is all set to make his first commit. By tradition, this is by adding his name to the committers list. Recall that a committer is considered to be someone who has committed code during the past 12 months. However, it is not until after 18 months of inactivity have passed that commit privileges are eligible to be revoked. [FreeBSD, 2002H] There are, however, no automatic procedures for doing this. For reactions concerning commit privileges not triggered by time, see section 1.5.8. Figure 6-2. Process summary: removing a committer When Core decides to clean up the committers list, they check who has not made a commit for the past 18 months. Committers who have not done so have their commit bits revoked. It is also possible for committers to request that their commit bit be retired if for some reason they are no longer going to be actively committing to the project. In this case, it can also be restored at a later time by core, should the committer ask. Roles in this process: Core team Contributor Committer Maintainership Mentor [FreeBSD, 2000A] [FreeBSD, 2002H] [FreeBSD, 2002I] 6.2. Adding/Removing an official CVSup Mirror A CVSup mirror is a replica of the official CVSup master that contains all the up-to-date source code for all the branches in the FreeBSD project, ports and documentation. Adding an official CVSup mirror starts with the potential CVSup Mirror Site Admin installing the “cvsup-mirror” package. Having done this and updated the source code with a mirror site, he now runs a fairly recent unofficial CVSup mirror. Deciding he has a stable environment, the processing power, the network capacity and the storage capacity to run an official mirror, he mails the CVSup Mirror Site Coordinator who decides whether the mirror should become an official mirror or not. In making this decision, the CVSup Mirror Site Coordinator has to determine whether that geographical area needs another mirror site, if the mirror administrator has the skills to run it reliably, if the network bandwidth is adequate and if the master server has the capacity to server another mirror. If CVSup Mirror Site Coordinator decides that the mirror should become an official mirror, he obtains an authentication key from the mirror admin that he installs so the mirror admin can update the mirror from the master server. Figure 6-3. Process summary: adding a CVSup mirror When a CVSup mirror administrator of an unofficial mirror offers to become an official mirror site, the CVSup coordinator decides if another mirror is needed and if there is sufficient capacity to accommodate it. If so, an authorisation key is requested and the mirror is given access to the main distribution site and added to the list of official mirrors. Tools used in this process: CVSup SSH 2 Hats involved in this process: CVSup Mirror Site Coordinator CVSup Mirror Site Admin 6.3. Committing code The committing of new or modified code is one of the most frequent processes in the FreeBSD project and will usually happen many times a day. Committing of code can only be done by a “committer”. Committers commit either code written by themselves, code submitted to them or code submitted through a problem report. When code is written by the developer that is non-trivial, he should seek a code review from the community. This is done by sending mail to the relevant list asking for review. Before submitting the code for review, he should ensure it compiles correctly with the entire tree and that all relevant tests run. This is called “pre-commit test”. When contributed code is received, it should be reviewed by the committer and tested the same way. When a change is committed to a part of the source that has been contributed from an outside Vendor, the maintainer should ensure that the patch is contributed back to the vendor. This is in line with the open source philosophy and makes it easier to stay in sync with outside projects as the patches do not have to be reapplied every time a new release is made. After the code has been available for review and no further changes are necessary, the code is committed into the development branch, -CURRENT. If the change applies for the -STABLE branch or the other branches as well, a “Merge From Current” ("MFC") countdown is set by the committer. After the number of days the committer chose when setting the MFC have passed, an email will automatically be sent to the committer reminding him to commit it to the -STABLE branch (and possibly security branches as well). Only security critical changes should be merged to security branches. Delaying the commit to -STABLE and other branches allows for “parallel debugging” where the committed code is tested on a wide range of configurations. This makes changes to -STABLE to contain fewer faults and thus giving the branch its name. Figure 6-4. Process summary: A committer commits code When a committer has written a piece of code and wants to commit it, he first needs to determine if it is trivial enough to go in without prior review or if it should first be reviewed by the developer community. If the code is trivial or has been reviewed and the committer is not the maintainer, he should consult the maintainer before proceeding. If the code is contributed by an outside vendor, the maintainer should create a patch that is sent back to the vendor. The code is then committed and the deployed by the users. Should they find problems with the code, this will be reported and the committer can go back to writing a patch. If a vendor is affected, he can choose to implement or ignore the patch. Figure 6-5. Process summary: A contributor commits code The difference when a contributor makes a code contribution is that he submits the code through the send-pr program. This report is picked up by the maintainer who reviews the code and commits it. Hats included in this process are: Committer Contributor Vendor Reviewer [FreeBSD, 2001] [Jørgensen, 2001] 6.4. Core election Core elections are held at least every two years. [8] Nine core members are elected. New elections are held if the number of core members drops below seven. New elections can also be held should at least 1/3 of the active committers demand this. When an election is to take place, core announces this at least 6 weeks in advance, and appoints an election manager to run the elections. Only committers can be elected into core. The candidates need to submit their candidacy at least one week before the election starts, but can refine their statements until the voting starts. They are presented in the candidates list. When writing their election statements, the candidates must answer a few standard questions submitted by the election manager. During elections, the rule that a committer must have committed during the 12 past months is followed strictly. Only these committers are eligible to vote. When voting, the committer may vote once in support of up to nine nominees. The voting is done over a period of four weeks with reminders being posted on “developers” mailing list that is available to all committers. The election results are released one week after the election ends, and the new core team takes office one week after the results have been posted. Should there be a voting tie, this will be resolved by the new, unambiguously elected core members. Votes and candidate statements are archived, but the archives are not publicly available. Figure 6-6. Process summary: Core elections Core announces the election and selects an election manager. He prepares the elections, and when ready, candidates can announce their candidacies through submitting their statements. The committers then vote. After the vote is over, the election results are announced and the new core team takes office. Hats in core elections are: Core team Committer Election Manager [FreeBSD, 2000A] [FreeBSD, 2002B] [FreeBSD, 2002G] 6.5. Development of new features Within the project there are sub-projects that are working on new features. These projects are generally done by one person [Jørgensen, 2001]. Every project is free to organise development as it sees fit. However, when the project is merged to the -CURRENT branch it must follow the project guidelines. When the code has been well tested in the -CURRENT branch and deemed stable enough and relevant to the -STABLE branch, it is merged to the -STABLE branch. The requirements of the project are given by developer wishes, requests from the community in terms of direct requests by mail, Problem Reports, commercial funding for the development of features, or contributions by the scientific community. The wishes that come within the responsibility of a developer are given to that developer who prioritises his time between the request and his wishes. A common way to do this is maintain a TODO-list maintained by the project. Items that do not come within someone's responsibility are collected on TODO-lists unless someone volunteers to take the responsibility. All requests, their distribution and follow-up are handled by the GNATS tool. Requirements analysis happens in two ways. The requests that come in are discussed on mailing lists, both within the main project and in the sub-project that the request belongs to or is spawned by the request. Furthermore, individual developers on the sub-project will evaluate the feasibility of the requests and determine the prioritisation between them. Other than archives of the discussions that have taken place, no outcome is created by this phase that is merged into the main project. As the requests are prioritised by the individual developers on the basis of doing what they find interesting, necessary or are funded to do, there is no overall strategy or priorisation of what requests to regard as requirements and following up their correct implementation. However, most developers have some shared vision of what issues are more important, and they can ask for guidelines from the release engineering team. The verification phase of the project is two-fold. Before committing code to the current-branch, developers request their code to be reviewed by their peers. This review is for the most part done by functional testing, but also code review is important. When the code is committed to the branch, a broader functional testing will happen, that may trigger further code review and debugging should the code not behave as expected. This second verification form may be regarded as structural verification. Although the sub-projects themselves may write formal tests such as unit tests, these are usually not collected by the main project and are usually removed before the code is committed to the current branch. [9] 6.6. Maintenance It is an advantage to the project to for each area of the source have at least one person that knows this area well. Some parts of the code have designated maintainers. Others have de-facto maintainers, and some parts of the system do not have maintainers. The maintainer is usually a person from the sub-project that wrote and integrated the code, or someone who has ported it from the platform it was written for. [10] The maintainer's job is to make sure the code is in sync with the project the code comes from if it is contributed code, and apply patches submitted by the community or write fixes to issues that are discovered. The main bulk of work that is put into the FreeBSD project is maintenance. [Jørgensen, 2001] has made a figure showing the life cycle of changes. Figure 6-7. Jørgenssen's model for change integration Here “development release” refers to the -CURRENT branch while “production release” refers to the -STABLE branch. The “pre-commit test” is the functional testing by peer developers when asked to do so or trying out the code to determine the status of the sub-project. “Parallel debugging” is the functional testing that can trigger more review, and debugging when the code is included in the -CURRENT branch. As of this writing, there were 269 committers in the project. When they commit a change to a branch, that constitutes a new release. It is very common for users in the community to track a particular branch. The immediate existence of a new release makes the changes widely available right away and allows for rapid feedback from the community. This also gives the community the response time they expect on issues that are of importance to them. This makes the community more engaged, and thus allows for more and better feedback that again spurs more maintenance and ultimately should create a better product. Before making changes to code in parts of the tree that has a history unknown to the committer, the committer is required to read the commit logs to see why certain features are implemented the way they are in order not to make mistakes that have previously either been thought through or resolved. 6.7. Problem reporting FreeBSD comes with a problem reporting tool called “send-pr” that is a part of the GNATS package. All users and developers are encouraged to use this tool for reporting problems in software they do not maintain. Problems include bug reports, feature requests, features that should be enhanced and notices of new versions of external software that is included in the project. Problem reports are sent to an email address where it is inserted into the GNATS maintenance database. A Bugbuster classifies the problem and sends it to the correct group or maintainer within the project. After someone has taken responsibility for the report, the report is being analysed. This analysis includes verifying the problem and thinking out a solution for the problem. Often feedback is required from the report originator or even from the FreeBSD community. Once a patch for the problem is made, the originator may be asked to try it out. Finally, the working patch is integrated into the project, and documented if applicable. It there goes through the regular maintenance cycle as described in section maintenance. These are the states a problem report can be in: open, analyzed, feedback, patched, suspended and closed. The suspended state is for when further progress is not possible due to the lack of information or for when the task would require so much work that nobody is working on it at the moment. Figure 6-8. Process summary: problem reporting A problem is reported by the report originator. It is then classified by a bugbuster and handed to the correct maintainer. He verifies the problem and discusses the problem with the originator until he has enough information to create a working patch. This patch is then committed and the problem report is closed. The roles included in this process are: Report originator Maintainership Bugbuster [FreeBSD, 2002C]. [FreeBSD, 2002D] 6.8. Reacting to misbehaviour [FreeBSD, 2001] has a number of rules that committers should follow. However, it happens that these rules are broken. The following rules exist in order to be able to react to misbehaviour. They specify what actions will result in how long a suspension the committer's commit privileges. Committing during code freezes without the approval of the Release Engineering team - 2 days Committing to a security branch without approval - 2 days Commit wars - 5 days to all participating parties Impolite or inappropriate behaviour - 5 days [Lehey, 2002] For the suspensions to be efficient, any single core member can implement a suspension before discussing it on the “core” mailing list. Repeat offenders can, with a 2/3 vote by core, receive harsher penalties, including permanent removal of commit privileges. (However, the latter is always viewed as a last resort, due to its inherent tendency to create controversy). All suspensions are posted to the “developers” mailing list, a list available to committers only. It is important that you cannot be suspended for making technical errors. All penalties come from breaking social etiquette. Hats involved in this process: Core team Committer 6.9. Release engineering The FreeBSD project has a Release Engineering team with a principal release engineer that is responsible for creating releases of FreeBSD that can be brought out to the user community via the net or sold in retail outlets. Since FreeBSD is available on multiple platforms and releases for the different architectures are made available at the same time, the team has one person in charge of each architecture. Also, there are roles in the team responsible for coordinating quality assurance efforts, building a package set and for having an updated set of documents. When referring to the release engineer, a representative for the release engineering team is meant. When a release is coming, the FreeBSD project changes shape somewhat. A release schedule is made containing feature- and code-freezes, release of interim releases and the final release. A feature-freeze means no new features are allowed to be committed to the branch without the release engineers' explicit consent. Code-freeze means no changes to the code (like bugs-fixes) are allowed to be committed without the release engineers explicit consent. This feature- and code-freeze is known as stabilising. During the release process, the release engineer has the full authority to revert to older versions of code and thus "back out" changes should he find that the changes are not suitable to be included in the release. There are three different kinds of releases: .0 releases are the first release of a major version. These are branched of the -CURRENT branch and have a significantly longer release engineering cycle due to the unstable nature of the -CURRENT branch .X releases are releases of the -STABLE branch. They are scheduled to come out every 4 months. .X.Y releases are security releases that follow the .X branch. These come out only when sufficient security fixes have been merged since the last release on that branch. New features are rarely included, and the security team is far more involved in these than in regular releases. For releases of the -STABLE-branch, the release process starts 45 days before the anticipated release date. During the first phase, the first 15 days, the developers merge what changes they have had in -CURRENT that they want to have in the release to the release branch. When this period is over, the code enters a 15 day code freeze in which only bug fixes, documentation updates, security-related fixes and minor device driver changes are allowed. These changes must be approved by the release engineer in advance. At the beginning of the last 15 day period a release candidate is created for widespread testing. Updates are less likely to be allowed during this period, except for important bug fixes and security updates. In this final period, all releases are considered release candidates. At the end of the release process, a release is created with the new version number, including binary distributions on web sites and the creation of a CD-ROM images. However, the release is not considered "really released" until a PGP-signed message stating exactly that, is sent to the mailing list freebsd-announce; anything labelled as a "release" before that may well be in-process and subject to change before the PGP-signed message is sent. [11]. The releases of the -CURRENT-branch (that is, all releases that end with “.0”) are very similar, but with twice as long timeframe. It starts 8 weeks prior to the release with announcement of the release time line. Two weeks into the release process, the feature freeze is initiated and performance tweaks should be kept to a minimum. Four weeks prior to the release, an official beta version is made available. Two weeks prior to release, the code is officially branched into a new version. This version is given release candidate status, and as with the release engineering of -STABLE, the code freeze of the release candidate is hardened. However, development on the main development branch can continue. Other than these differences, the release engineering processes are alike. .0 releases go into their own branch and are aimed mainly at early adopters. The branch then goes through a period of stabilisation, and it is not until the Release Engineering Team> decides the demands to stability have been satisfied that the branch becomes -STABLE and -CURRENT targets the next major version. While this for the majority has been with .1 versions, this is not a demand. Most releases are made when a given date that has been deemed a long enough time since the previous release comes. A target is set for having major releases every 18 months and minor releases every 4 months. The user community has made it very clear that security and stability cannot be sacrificed by self-imposed deadlines and target release dates. For slips of time not to become too long with regards to security and stability issues, extra discipline is required when committing changes to -STABLE. Figure 6-9. Process summary: release engineering These are the stages in the release engineering process. Multiple release candidates may be created until the release is deemed stable enough to be released. [FreeBSD, 2002E] Chapter 7 Tools The major support tools for supporting the development process are CVS, CVSup, Perforce, GNATS, Mailman and OpenSSH. Except for CVSup, these are externally developed tools. These tools are commonly used in the open source world. 7.1. Concurrent Versions System &#40;CVS&#41; Concurrent Versions System or simply “CVS” is a system to handle multiple versions of text files and tracking who committed what changes and why. A project lives within a “repository” and different versions are considered different “branches”. 7.2. CVSup CVSup is a software package for distributing and updating collections of files across a network. It consists of a client program, cvsup, and a server program, cvsupd. The package is tailored specifically for distributing CVS repositories, and by taking advantage of CVS' properties, it performs updates much faster than traditional systems. 7.3. GNATS GNATS is a maintenance database consisting of a set of tools to track bugs at a central site. It supports the bug tracking process for sending and handling bugs as well as querying and updating the database and editing bug reports. The project uses one of its many client interfaces, “send-pr”, to send “Problem Reports” by email to the projects central GNATS server. The committers have also web and command-line clients available. 7.4. Mailman Mailman is a program that automates the management of mailing lists. The FreeBSD Project uses it to run 16 general lists, 60 technical lists, 4 limited lists and 5 lists with CVS commit logs. It is also used for many mailing lists set up and used by other people and projects in the FreeBSD community. General lists are lists for the general public, technical lists are mainly for the development of specific areas of interest, and closed lists are for internal communication not intended for the general public. The majority of all the communication in the project goes through these 85 lists [FreeBSD, 2003A, Appendix C]. 7.5. Perforce Perforce is a commercial software configuration management system developed by Perforce Systems that is available on over 50 operating systems. It is a collection of clients built around the Perforce server that contains the central file repository and tracks the operations done upon it. The clients are both clients for accessing the repository and administration of its configuration. 7.6. Pretty Good Privacy Pretty Good Privacy, better known as PGP, is a cryptosystem using a public key architecture to allow people to digitally sign and/or encrypt information in order to ensure secure communication between two parties. A signature is used when sending information out many recipients, enabling them to verify that the information has not been tampered with before they received it. In the FreeBSD Project this is the primary means of ensuring that information has been written by the person who claims to have written it, and not altered in transit. 7.7. Secure Shell Secure Shell is a standard for securely logging into a remote system and for executing commands on the remote system. It allows other connections, called tunnels, to be established and protected between the two involved systems. This standard exists in two primary versions, and only version two is used for the FreeBSD Project. The most common implementation of the standard is OpenSSH that is a part of the project's main distribution. Since its source is updated more often than FreeBSD releases, the latest version is also available in the ports tree. Chapter 8 Sub-projects Sub-projects are formed to reduce the amount of communication needed to coordinate the group of developers. When a problem area is sufficiently isolated, most communication would be within the group focusing on the problem, requiring less communication with the groups they communicate with than were the group not isolated. 8.1. The Ports Subproject A “port” is a set of meta-data and patches that are needed to fetch, compile and install correctly an external piece of software on a FreeBSD system. The amount of ports have grown at a tremendous rate, as shown by the following figure. Figure 8-1. Number of ports added between 1996 and 2005 Figure 8-1 is taken from the FreeBSD web site. It shows the number of ports available to FreeBSD in the period 1995 to 2005. It looks like the curve has first grown exponentionally, and then since the middle of 2001 grown linerly. As the external software described by the port often is under continued development, the amount of work required to maintain the ports is already large, and increasing. This has led to the ports part of the FreeBSD project gaining a more empowered structure, and is more and more becoming a sub-project of the FreeBSD project. Ports has its own core team with the Ports Manager as its leader, and this team can appoint committers without FreeBSD Core's approval. Unlike in the FreeBSD Project, where a lot of maintenance frequently is rewarded with a commit bit, the ports sub-project contains many active maintainers that are not committers. Unlike the main project, the ports tree is not branched. Every release of FreeBSD follows the current ports collection and has thus available updated information on where to find programs and how to build them. This, however, means that a port that makes dependencies on the system may need to have variations depending on what version of FreeBSD it runs on. With an unbranched ports repository it is not possible to guarantee that any port will run on anything other than -CURRENT and -STABLE, in particular older, minor releases. There is neither the infrastructure nor volunteer time needed to guarantee this. For efficiency of communication, teams depending on Ports, such as the release engineering team, have their own ports liaisons. 8.2. The FreeBSD Documentation Project The FreeBSD Documentation project was started January 1995. From the initial group of a project leader, four team leaders and 16 members, they are now a total of 44 committers. The documentation mailing list has just under 300 members, indicating that there is quite a large community around it. The goal of the Documentation project is to provide good and useful documentation of the FreeBSD project, thus making it easier for new users to get familiar with the system and detailing advanced features for the users. The main tasks in the Documentation project are to work on current projects in the “FreeBSD Documentation Set”, and translate the documentation to other languages. Like the FreeBSD Project, documentation is split in the same branches. This is done so that there is always an updated version of the documentation for each version. Only documentation errors are corrected in the security branches. Like the ports sub-project, the Documentation project can appoint documentation committers without FreeBSD Core's approval. [FreeBSD, 2003B]. The Documentation project has a primer. This is used both to introduce new project members to the standard tools and syntaxes and acts as a reference when working on the project. References [Brooks, 1995] Frederick P. Brooks, 1975, 1995, 0201835959, Addison-Wesley Pub Co, The Mythical Man-Month: Essays on Software Engineering, Anniversary Edition (2nd Edition). [Saers, 2003] Niklas Saers, 2003, A project model for the FreeBSD Project: Candidatus Scientiarum thesis. [Jørgensen, 2001] Niels Jørgensen, 2001, Putting it All in the Trunk: Incremental Software Development in the FreeBSD Open Source Project. [PMI, 2000] Project Management Institute, 1996, 2000, 1-880410-23-0, Project Management Institute, Pennsylvania, PMBOK Guide: A Guide to the Project Management Body of Knowledge, 2000 Edition. [FreeBSD, 2000A] 2002, Core Bylaws. [FreeBSD, 2002A] 2002, FreeBSD Developer's Handbook. [FreeBSD, 2002B] 2002, Core team election 2002. [Losh, 2002] Warner Losh, 2002, Working with Hats. [FreeBSD, 2002C] Dag-Erling Smørgrav and Hiten Pandya, 2002, The FreeBSD Documentation Project, Problem Report Handling Guidelines. [FreeBSD, 2002D] Dag-Erling Smørgrav, 2002, The FreeBSD Documentation Project, Writing FreeBSD Problem Reports. [FreeBSD, 2001] 2001, The FreeBSD Documentation Project, Committers Guide. [FreeBSD, 2002E] Murray Stokely, 2002, The FreeBSD Documentation Project, FreeBSD Release Engineering. [FreeBSD, 2003A] The FreeBSD Documentation Project, FreeBSD Handbook. [FreeBSD, 2002F] 2002, The FreeBSD Documentation Project, Contributors to FreeBSD. [FreeBSD, 2002G] 2002, The FreeBSD Project, Core team elections 2002. [FreeBSD, 2002H] 2002, The FreeBSD Project, Commit Bit Expiration Policy: 2002/04/06 15:35:30. [FreeBSD, 2002I] 2002, The FreeBSD Project, New Account Creation Procedure: 2002/08/19 17:11:27. [FreeBSD, 2003B] 2002, The FreeBSD Documentation Project, FreeBSD DocEng Team Charter: 2003/03/16 12:17. [Lehey, 2002] Greg Lehey, 2002, Greg Lehey, Two years in the trenches: The evolution of a software project. Notes [1] This goes hand-in-hand with Brooks' law that “adding another person to a late project will make it later” since it will increase the communication needs Brooks, 1995. A project model is a tool to reduce the communication needs. [2] Statistics are generated by counting the number of entries in the file fetched by portsdb by April 1st, 2005. portsdb is a part of the port sysutils/portupgrade. [3] The period from January 1st, 2004 to December 31st, 2004 was examined to find this number. [4] For instance, the development of the Bluetooth stack started as a sub-project until it was deemed stable enough to be merged into the -CURRENT branch. Now it is a part of the core FreeBSD system. [5] According to Kirk McKusick, after 20 years of developing UNIX operating systems, the interfaces are for the most part figured out. There is therefore no need for much design. However, new applications of the system and new hardware leads to some implementations being more beneficial than those that used to be preferred. One example is the introduction of web browsing that made the normal TCP/IP connection a short burst of data rather than a steady stream over a longer period of time. [6] The first release this actually happened for was 4.5-RELEASE, but security branches were at the same time created for 4.3-RELEASE and 4.4-RELEASE. [7] There is a terminology overlap with respect to the word "stable", which leads to some confusion. The -STABLE branch is still a development branch, whose goal is to be useful for most people. If it is never acceptable for a system to get changes that are not announced at the time it is deployed, that system should run a security branch. [8] The first Core election was held September 2000 [9] More and more tests are however performed when building the system &

TD的安装和使用

TD的开发和使用，使用此文档可快速安装Td，还有其他资源，稍后上传

TD服务器搭建及配置指南(TestDirector).docx

。。。

停机维护静态html页面

在系统后台更新维护期间，用户按照原地址访问的时候会出现404， 如果不告诉用户当前系统处在维护状态， 用户会以为自己的网络有问题，亦或是自己记错了网站的地址。 因为给用户体验不好。

操作系统(内存管理)

文将对 Linux™ 程序员可以使用的内存管理技术进行概述，虽然关注的重点是 C 语言，但同样也适用于其他语言。文中将为您提供如何管理内存的细节，然后将进一步展示如何手工管理内存，如何使用引用计数或者内存池来半手工地管理内存，以及如何使用垃圾收集自动管理内存。 为什么必须管理内存 内存管理是计算机编程最为基本的领域之一。在很多脚本语言中，您不必担心内存是如何管理的，这并不能使得内存管理的重要性有一点点降低。对实际编程来说，理解您的内存管理器的能力与局限性至关重要。在大部分系统语言中，比如 C 和 C++，您必须进行内存管理。本文将介绍手工的、半手工的以及自动的内存管理实践的基本概念。 追溯到在 Apple II 上进行汇编语言编程的时代，那时内存管理还不是个大问题。您实际上在运行整个系统。系统有多少内存，您就有多少内存。您甚至不必费心思去弄明白它有多少内存，因为每一台机器的内存数量都相同。所以，如果内存需要非常固定，那么您只需要选择一个内存范围并使用它即可。 不过，即使是在这样一个简单的计算机中，您也会有问题，尤其是当您不知道程序的每个部分将需要多少内存时。如果您的空间有限，而内存需求是变化的，那么您需要一些方法来满足这些需求： 确定您是否有足够的内存来处理数据。 从可用的内存中获取一部分内存。 向可用内存池（pool）中返回部分内存，以使其可以由程序的其他部分或者其他程序使用。 实现这些需求的程序库称为 分配程序（allocators），因为它们负责分配和回收内存。程序的动态性越强，内存管理就越重要，您的内存分配程序的选择也就更重要。让我们来了解可用于内存管理的不同方法，它们的好处与不足，以及它们最适用的情形。 回页首 C 风格的内存分配程序 C 编程语言提供了两个函数来满足我们的三个需求： malloc：该函数分配给定的字节数，并返回一个指向它们的指针。如果没有足够的可用内存，那么它返回一个空指针。 free：该函数获得指向由 malloc 分配的内存片段的指针，并将其释放，以便以后的程序或操作系统使用（实际上，一些 malloc 实现只能将内存归还给程序，而无法将内存归还给操作系统）。 物理内存和虚拟内存 要理解内存在程序中是如何分配的，首先需要理解如何将内存从操作系统分配给程序。计算机上的每一个进程都认为自己可以访问所有的物理内存。显然，由于同时在运行多个程序，所以每个进程不可能拥有全部内存。实际上，这些进程使用的是 虚拟内存。 只是作为一个例子，让我们假定您的程序正在访问地址为 629 的内存。不过，虚拟内存系统不需要将其存储在位置为 629 的 RAM 中。实际上，它甚至可以不在 RAM 中 —— 如果物理 RAM 已经满了，它甚至可能已经被转移到硬盘上！由于这类地址不必反映内存所在的物理位置，所以它们被称为虚拟内存。操作系统维持着一个虚拟地址到物理地址的转换的表，以便计算机硬件可以正确地响应地址请求。并且，如果地址在硬盘上而不是在 RAM 中，那么操作系统将暂时停止您的进程，将其他内存转存到硬盘中，从硬盘上加载被请求的内存，然后再重新启动您的进程。这样，每个进程都获得了自己可以使用的地址空间，可以访问比您物理上安装的内存更多的内存。 在 32-位 x86 系统上，每一个进程可以访问 4 GB 内存。现在，大部分人的系统上并没有 4 GB 内存，即使您将 swap 也算上， 每个进程所使用的内存也肯定少于 4 GB。因此，当加载一个进程时，它会得到一个取决于某个称为 系统中断点（system break）的特定地址的初始内存分配。该地址之后是未被映射的内存 —— 用于在 RAM 或者硬盘中没有分配相应物理位置的内存。因此，如果一个进程运行超出了它初始分配的内存，那么它必须请求操作系统“映射进来（map in）”更多的内存。（映射是一个表示一一对应关系的数学术语 —— 当内存的虚拟地址有一个对应的物理地址来存储内存内容时，该内存将被映射。） 基于 UNIX 的系统有两个可映射到附加内存中的基本系统调用： brk： brk() 是一个非常简单的系统调用。还记得系统中断点吗？该位置是进程映射的内存边界。 brk() 只是简单地将这个位置向前或者向后移动，就可以向进程添加内存或者从进程取走内存。 mmap： mmap()，或者说是“内存映像”，类似于 brk()，但是更为灵活。首先，它可以映射任何位置的内存，而不单单只局限于进程。其次，它不仅可以将虚拟地址映射到物理的 RAM 或者 swap，它还可以将它们映射到文件和文件位置，这样，读写内存将对文件中的数据进行读写。不过，在这里，我们只关心 mmap 向进程添加被映射的内存的能力。 munmap() 所做的事情与 mmap() 相反。 如您所见， brk() 或者 mmap() 都可以用来向我们的进程添加额外的虚拟内存。在我们的例子中将使用 brk()，因为它更简单，更通用。 实现一个简单的分配程序 如果您曾经编写过很多 C 程序，那么您可能曾多次使用过 malloc() 和 free()。不过，您可能没有用一些时间去思考它们在您的操作系统中是如何实现的。本节将向您展示 malloc 和 free 的一个最简化实现的代码，来帮助说明管理内存时都涉及到了哪些事情。 要试着运行这些示例，需要先 复制本代码清单，并将其粘贴到一个名为 malloc.c 的文件中。接下来，我将一次一个部分地对该清单进行解释。 在大部分操作系统中，内存分配由以下两个简单的函数来处理： void *malloc(long numbytes)：该函数负责分配 numbytes 大小的内存，并返回指向第一个字节的指针。 void free(void *firstbyte)：如果给定一个由先前的 malloc 返回的指针，那么该函数会将分配的空间归还给进程的“空闲空间”。 malloc_init 将是初始化内存分配程序的函数。它要完成以下三件事：将分配程序标识为已经初始化，找到系统中最后一个有效内存地址，然后建立起指向我们管理的内存的指针。这三个变量都是全局变量： 清单 1. 我们的简单分配程序的全局变量 int has_initialized = 0; void *managed_memory_start; void *last_valid_address; 如前所述，被映射的内存的边界（最后一个有效地址）常被称为系统中断点或者 当前中断点。在很多 UNIX® 系统中，为了指出当前系统中断点，必须使用 sbrk(0) 函数。 sbrk 根据参数中给出的字节数移动当前系统中断点，然后返回新的系统中断点。使用参数 0 只是返回当前中断点。这里是我们的 malloc 初始化代码，它将找到当前中断点并初始化我们的变量： 清单 2. 分配程序初始化函数 /* Include the sbrk function */ #include void malloc_init() { /* grab the last valid address from the OS */ last_valid_address = sbrk(0); /* we don't have any memory to manage yet, so *just set the beginning to be last_valid_address */ managed_memory_start = last_valid_address; /* Okay, we're initialized and ready to go */ has_initialized = 1; } 现在，为了完全地管理内存，我们需要能够追踪要分配和回收哪些内存。在对内存块进行了 free 调用之后，我们需要做的是诸如将它们标记为未被使用的等事情，并且，在调用 malloc 时，我们要能够定位未被使用的内存块。因此， malloc 返回的每块内存的起始处首先要有这个结构： 清单 3. 内存控制块结构定义 struct mem_control_block { int is_available; int size; }; 现在，您可能会认为当程序调用 malloc 时这会引发问题 —— 它们如何知道这个结构？答案是它们不必知道；在返回指针之前，我们会将其移动到这个结构之后，把它隐藏起来。这使得返回的指针指向没有用于任何其他用途的内存。那样，从调用程序的角度来看，它们所得到的全部是空闲的、开放的内存。然后，当通过 free() 将该指针传递回来时，我们只需要倒退几个内存字节就可以再次找到这个结构。 在讨论分配内存之前，我们将先讨论释放，因为它更简单。为了释放内存，我们必须要做的惟一一件事情就是，获得我们给出的指针，回退 sizeof(struct mem_control_block) 个字节，并将其标记为可用的。这里是对应的代码： 清单 4. 解除分配函数 void free(void *firstbyte) { struct mem_control_block *mcb; /* Backup from the given pointer to find the * mem_control_block */ mcb = firstbyte - sizeof(struct mem_control_block); /* Mark the block as being available */ mcb->is_available = 1; /* That's It! We're done. */ return; } 如您所见，在这个分配程序中，内存的释放使用了一个非常简单的机制，在固定时间内完成内存释放。分配内存稍微困难一些。以下是该算法的略述： 清单 5. 主分配程序的伪代码 1. If our allocator has not been initialized, initialize it. 2. Add sizeof(struct mem_control_block) to the size requested. 3. start at managed_memory_start. 4. Are we at last_valid address? 5. If we are: A. We didn't find any existing space that was large enough -- ask the operating system for more and return that. 6. Otherwise: A. Is the current space available (check is_available from the mem_control_block)? B. If it is: i) Is it large enough (check "size" from the mem_control_block)? ii) If so: a. Mark it as unavailable b. Move past mem_control_block and return the pointer iii) Otherwise: a. Move forward "size" bytes b. Go back go step 4 C. Otherwise: i) Move forward "size" bytes ii) Go back to step 4 我们主要使用连接的指针遍历内存来寻找开放的内存块。这里是代码： 清单 6. 主分配程序 void *malloc(long numbytes) { /* Holds where we are looking in memory */ void *current_location; /* This is the same as current_location, but cast to a * memory_control_block */ struct mem_control_block *current_location_mcb; /* This is the memory location we will return. It will * be set to 0 until we find something suitable */ void *memory_location; /* Initialize if we haven't already done so */ if(! has_initialized) { malloc_init(); } /* The memory we search for has to include the memory * control block, but the users of malloc don't need * to know this, so we'll just add it in for them. */ numbytes = numbytes + sizeof(struct mem_control_block); /* Set memory_location to 0 until we find a suitable * location */ memory_location = 0; /* Begin searching at the start of managed memory */ current_location = managed_memory_start; /* Keep going until we have searched all allocated space */ while(current_location != last_valid_address) { /* current_location and current_location_mcb point * to the same address. However, current_location_mcb * is of the correct type, so we can use it as a struct. * current_location is a void pointer so we can use it * to calculate addresses. */ current_location_mcb = (struct mem_control_block *)current_location; if(current_location_mcb->is_available) { if(current_location_mcb->size >= numbytes) { /* Woohoo! We've found an open, * appropriately-size location. */ /* It is no longer available */ current_location_mcb->is_available = 0; /* We own it */ memory_location = current_location; /* Leave the loop */ break; } } /* If we made it here, it's because the Current memory * block not suitable; move to the next one */ current_location = current_location + current_location_mcb->size; } /* If we still don't have a valid location, we'll * have to ask the operating system for more memory */ if(! memory_location) { /* Move the program break numbytes further */ sbrk(numbytes); /* The new memory will be where the last valid * address left off */ memory_location = last_valid_address; /* We'll move the last valid address forward * numbytes */ last_valid_address = last_valid_address + numbytes; /* We need to initialize the mem_control_block */ current_location_mcb = memory_location; current_location_mcb->is_available = 0; current_location_mcb->size = numbytes; } /* Now, no matter what (well, except for error conditions), * memory_location has the address of the memory, including * the mem_control_block */ /* Move the pointer past the mem_control_block */ memory_location = memory_location + sizeof(struct mem_control_block); /* Return the pointer */ return memory_location; } 这就是我们的内存管理器。现在，我们只需要构建它，并在程序中使用它即可。 运行下面的命令来构建 malloc 兼容的分配程序（实际上，我们忽略了 realloc() 等一些函数，不过， malloc() 和 free() 才是最主要的函数）： 清单 7. 编译分配程序 gcc -shared -fpic malloc.c -o malloc.so 该程序将生成一个名为 malloc.so 的文件，它是一个包含有我们的代码的共享库。 在 UNIX 系统中，现在您可以用您的分配程序来取代系统的 malloc()，做法如下： 清单 8. 替换您的标准的 malloc LD_PRELOAD=/path/to/malloc.so export LD_PRELOAD LD_PRELOAD 环境变量使动态链接器在加载任何可执行程序之前，先加载给定的共享库的符号。它还为特定库中的符号赋予优先权。因此，从现在起，该会话中的任何应用程序都将使用我们的 malloc()，而不是只有系统的应用程序能够使用。有一些应用程序不使用 malloc()，不过它们是例外。其他使用 realloc() 等其他内存管理函数的应用程序，或者错误地假定 malloc() 内部行为的那些应用程序，很可能会崩溃。ash shell 似乎可以使用我们的新 malloc() 很好地工作。 如果您想确保 malloc() 正在被使用，那么您应该通过向函数的入口点添加 write() 调用来进行测试。 我们的内存管理器在很多方面都还存在欠缺，但它可以有效地展示内存管理需要做什么事情。它的某些缺点包括： 由于它对系统中断点（一个全局变量）进行操作，所以它不能与其他分配程序或者 mmap 一起使用。 当分配内存时，在最坏的情形下，它将不得不遍历 全部进程内存；其中可能包括位于硬盘上的很多内存，这意味着操作系统将不得不花时间去向硬盘移入数据和从硬盘中移出数据。 没有很好的内存不足处理方案（ malloc 只假定内存分配是成功的）。 它没有实现很多其他的内存函数，比如 realloc()。 由于 sbrk() 可能会交回比我们请求的更多的内存，所以在堆（heap）的末端会遗漏一些内存。 虽然 is_available 标记只包含一位信息，但它要使用完整的 4-字节 的字。 分配程序不是线程安全的。 分配程序不能将空闲空间拼合为更大的内存块。 分配程序的过于简单的匹配算法会导致产生很多潜在的内存碎片。 我确信还有很多其他问题。这就是为什么它只是一个例子! 其他 malloc 实现 malloc() 的实现有很多，这些实现各有优点与缺点。在设计一个分配程序时，要面临许多需要折衷的选择，其中包括： 分配的速度。 回收的速度。 有线程的环境的行为。 内存将要被用光时的行为。 局部缓存。 簿记（Bookkeeping）内存开销。 虚拟内存环境中的行为。 小的或者大的对象。 实时保证。 每一个实现都有其自身的优缺点集合。在我们的简单的分配程序中，分配非常慢，而回收非常快。另外，由于它在使用虚拟内存系统方面较差，所以它最适于处理大的对象。 还有其他许多分配程序可以使用。其中包括： Doug Lea Malloc：Doug Lea Malloc 实际上是完整的一组分配程序，其中包括 Doug Lea 的原始分配程序，GNU libc 分配程序和 ptmalloc。 Doug Lea 的分配程序有着与我们的版本非常类似的基本结构，但是它加入了索引，这使得搜索速度更快，并且可以将多个没有被使用的块组合为一个大的块。它还支持缓存，以便更快地再次使用最近释放的内存。 ptmalloc 是 Doug Lea Malloc 的一个扩展版本，支持多线程。在本文后面的 参考资料部分中，有一篇描述 Doug Lea 的 Malloc 实现的文章。 BSD Malloc：BSD Malloc 是随 4.2 BSD 发行的实现，包含在 FreeBSD 之中，这个分配程序可以从预先确实大小的对象构成的池中分配对象。它有一些用于对象大小的 size 类，这些对象的大小为 2 的若干次幂减去某一常数。所以，如果您请求给定大小的一个对象，它就简单地分配一个与之匹配的 size 类。这样就提供了一个快速的实现，但是可能会浪费内存。在 参考资料部分中，有一篇描述该实现的文章。 Hoard：编写 Hoard 的目标是使内存分配在多线程环境中进行得非常快。因此，它的构造以锁的使用为中心，从而使所有进程不必等待分配内存。它可以显著地加快那些进行很多分配和回收的多线程进程的速度。在 参考资料部分中，有一篇描述该实现的文章。 众多可用的分配程序中最有名的就是上述这些分配程序。如果您的程序有特别的分配需求，那么您可能更愿意编写一个定制的能匹配您的程序内存分配方式的分配程序。不过，如果不熟悉分配程序的设计，那么定制分配程序通常会带来比它们解决的问题更多的问题。要获得关于该主题的适当的介绍，请参阅 Donald Knuth 撰写的 The Art of Computer Programming Volume 1: Fundamental Algorithms 中的第 2.5 节“Dynamic Storage Allocation”（请参阅 参考资料中的链接）。它有点过时，因为它没有考虑虚拟内存环境，不过大部分算法都是基于前面给出的函数。 在 C++ 中，通过重载 operator new()，您可以以每个类或者每个模板为单位实现自己的分配程序。在 Andrei Alexandrescu 撰写的 Modern C++ Design 的第 4 章（“Small Object Allocation”）中，描述了一个小对象分配程序（请参阅 参考资料中的链接）。 基于 malloc() 的内存管理的缺点 不只是我们的内存管理器有缺点，基于 malloc() 的内存管理器仍然也有很多缺点，不管您使用的是哪个分配程序。对于那些需要保持长期存储的程序使用 malloc() 来管理内存可能会非常令人失望。如果您有大量的不固定的内存引用，经常难以知道它们何时被释放。生存期局限于当前函数的内存非常容易管理，但是对于生存期超出该范围的内存来说，管理内存则困难得多。而且，关于内存管理是由进行调用的程序还是由被调用的函数来负责这一问题，很多 API 都不是很明确。 因为管理内存的问题，很多程序倾向于使用它们自己的内存管理规则。C++ 的异常处理使得这项任务更成问题。有时好像致力于管理内存分配和清理的代码比实际完成计算任务的代码还要多！因此，我们将研究内存管理的其他选择。 回页首 半自动内存管理策略 引用计数 引用计数是一种 半自动（semi-automated）的内存管理技术，这表示它需要一些编程支持，但是它不需要您确切知道某一对象何时不再被使用。引用计数机制为您完成内存管理任务。 在引用计数中，所有共享的数据结构都有一个域来包含当前活动“引用”结构的次数。当向一个程序传递一个指向某个数据结构指针时，该程序会将引用计数增加 1。实质上，您是在告诉数据结构，它正在被存储在多少个位置上。然后，当您的进程完成对它的使用后，该程序就会将引用计数减少 1。结束这个动作之后，它还会检查计数是否已经减到零。如果是，那么它将释放内存。 这样做的好处是，您不必追踪程序中某个给定的数据结构可能会遵循的每一条路径。每次对其局部的引用，都将导致计数的适当增加或减少。这样可以防止在使用数据结构时释放该结构。不过，当您使用某个采用引用计数的数据结构时，您必须记得运行引用计数函数。另外，内置函数和第三方的库不会知道或者可以使用您的引用计数机制。引用计数也难以处理发生循环引用的数据结构。 要实现引用计数，您只需要两个函数 —— 一个增加引用计数，一个减少引用计数并当计数减少到零时释放内存。 一个示例引用计数函数集可能看起来如下所示： 清单 9. 基本的引用计数函数 /* Structure Definitions*/ /* Base structure that holds a refcount */ struct refcountedstruct { int refcount; } /* All refcounted structures must mirror struct * refcountedstruct for their first variables */ /* Refcount maintenance functions */ /* Increase reference count */ void REF(void *data) { struct refcountedstruct *rstruct; rstruct = (struct refcountedstruct *) data; rstruct->refcount++; } /* Decrease reference count */ void UNREF(void *data) { struct refcountedstruct *rstruct; rstruct = (struct refcountedstruct *) data; rstruct->refcount--; /* Free the structure if there are no more users */ if(rstruct->refcount == 0) { free(rstruct); } } REF 和 UNREF 可能会更复杂，这取决于您想要做的事情。例如，您可能想要为多线程程序增加锁，那么您可能想扩展 refcountedstruct，使它同样包含一个指向某个在释放内存之前要调用的函数的指针（类似于面向对象语言中的析构函数 —— 如果您的结构中包含这些指针，那么这是 必需的）。 当使用 REF 和 UNREF 时，您需要遵守这些指针的分配规则： UNREF 分配前左端指针（left-hand-side pointer）指向的值。 REF 分配后左端指针（left-hand-side pointer）指向的值。 在传递使用引用计数的结构的函数中，函数需要遵循以下这些规则： 在函数的起始处 REF 每一个指针。 在函数的结束处 UNREF 第一个指针。 以下是一个使用引用计数的生动的代码示例： 清单 10. 使用引用计数的示例 /* EXAMPLES OF USAGE */ /* Data type to be refcounted */ struct mydata { int refcount; /* same as refcountedstruct */ int datafield1; /* Fields specific to this struct */ int datafield2; /* other declarations would go here as appropriate */ }; /* Use the functions in code */ void dosomething(struct mydata *data) { REF(data); /* Process data */ /* when we are through */ UNREF(data); } struct mydata *globalvar1; /* Note that in this one, we don't decrease the * refcount since we are maintaining the reference * past the end of the function call through the * global variable */ void storesomething(struct mydata *data) { REF(data); /* passed as a parameter */ globalvar1 = data; REF(data); /* ref because of Assignment */ UNREF(data); /* Function finished */ } 由于引用计数是如此简单，大部分程序员都自已去实现它，而不是使用库。不过，它们依赖于 malloc 和 free 等低层的分配程序来实际地分配和释放它们的内存。 在 Perl 等高级语言中，进行内存管理时使用引用计数非常广泛。在这些语言中，引用计数由语言自动地处理，所以您根本不必担心它，除非要编写扩展模块。由于所有内容都必须进行引用计数，所以这会对速度产生一些影响，但它极大地提高了编程的安全性和方便性。以下是引用计数的益处： 实现简单。 易于使用。 由于引用是数据结构的一部分，所以它有一个好的缓存位置。 不过，它也有其不足之处： 要求您永远不要忘记调用引用计数函数。 无法释放作为循环数据结构的一部分的结构。 减缓几乎每一个指针的分配。 尽管所使用的对象采用了引用计数，但是当使用异常处理（比如 try 或 setjmp()/ longjmp()）时，您必须采取其他方法。 需要额外的内存来处理引用。 引用计数占用了结构中的第一个位置，在大部分机器中最快可以访问到的就是这个位置。 在多线程环境中更慢也更难以使用。 C++ 可以通过使用 智能指针（smart pointers）来容忍程序员所犯的一些错误，智能指针可以为您处理引用计数等指针处理细节。不过，如果不得不使用任何先前的不能处理智能指针的代码（比如对 C 库的联接），实际上，使用它们的后果通实比不使用它们更为困难和复杂。因此，它通常只是有益于纯 C++ 项目。如果您想使用智能指针，那么您实在应该去阅读 Alexandrescu 撰写的 Modern C++ Design 一书中的“Smart Pointers”那一章。 内存池 内存池是另一种半自动内存管理方法。内存池帮助某些程序进行自动内存管理，这些程序会经历一些特定的阶段，而且每个阶段中都有分配给进程的特定阶段的内存。例如，很多网络服务器进程都会分配很多针对每个连接的内存 —— 内存的最大生存期限为当前连接的存在期。Apache 使用了池式内存（pooled memory），将其连接拆分为各个阶段，每个阶段都有自己的内存池。在结束每个阶段时，会一次释放所有内存。 在池式内存管理中，每次内存分配都会指定内存池，从中分配内存。每个内存池都有不同的生存期限。在 Apache 中，有一个持续时间为服务器存在期的内存池，还有一个持续时间为连接的存在期的内存池，以及一个持续时间为请求的存在期的池，另外还有其他一些内存池。因此，如果我的一系列函数不会生成比连接持续时间更长的数据，那么我就可以完全从连接池中分配内存，并知道在连接结束时，这些内存会被自动释放。另外，有一些实现允许注册 清除函数（cleanup functions），在清除内存池之前，恰好可以调用它，来完成在内存被清理前需要完成的其他所有任务（类似于面向对象中的析构函数）。 要在自己的程序中使用池，您既可以使用 GNU libc 的 obstack 实现，也可以使用 Apache 的 Apache Portable Runtime。GNU obstack 的好处在于，基于 GNU 的 Linux 发行版本中默认会包括它们。Apache Portable Runtime 的好处在于它有很多其他工具，可以处理编写多平台服务器软件所有方面的事情。要深入了解 GNU obstack 和 Apache 的池式内存实现，请参阅 参考资料部分中指向这些实现的文档的链接。 下面的假想代码列表展示了如何使用 obstack： 清单 11. obstack 的示例代码 #include #include /* Example code listing for using obstacks */ /* Used for obstack macros (xmalloc is a malloc function that exits if memory is exhausted */ #define obstack_chunk_alloc xmalloc #define obstack_chunk_free free /* Pools */ /* Only permanent allocations should go in this pool */ struct obstack *global_pool; /* This pool is for per-connection data */ struct obstack *connection_pool; /* This pool is for per-request data */ struct obstack *request_pool; void allocation_failed() { exit(1); } int main() { /* Initialize Pools */ global_pool = (struct obstack *) xmalloc (sizeof (struct obstack)); obstack_init(global_pool); connection_pool = (struct obstack *) xmalloc (sizeof (struct obstack)); obstack_init(connection_pool); request_pool = (struct obstack *) xmalloc (sizeof (struct obstack)); obstack_init(request_pool); /* Set the error handling function */ obstack_alloc_failed_handler = &allocation_failed; /* Server main loop */ while(1) { wait_for_connection(); /* We are in a connection */ while(more_requests_available()) { /* Handle request */ handle_request(); /* Free all of the memory allocated * in the request pool */ obstack_free(request_pool, NULL); } /* We're finished with the connection, time * to free that pool */ obstack_free(connection_pool, NULL); } } int handle_request() { /* Be sure that all object allocations are allocated * from the request pool */ int bytes_i_need = 400; void *data1 = obstack_alloc(request_pool, bytes_i_need); /* Do stuff to process the request */ /* return */ return 0; } 基本上，在操作的每一个主要阶段结束之后，这个阶段的 obstack 会被释放。不过，要注意的是，如果一个过程需要分配持续时间比当前阶段更长的内存，那么它也可以使用更长期限的 obstack，比如连接或者全局内存。传递给 obstack_free() 的 NULL 指出它应该释放 obstack 的全部内容。可以用其他的值，但是它们通常不怎么实用。 使用池式内存分配的益处如下所示： 应用程序可以简单地管理内存。 内存分配和回收更快，因为每次都是在一个池中完成的。分配可以在 O(1) 时间内完成，释放内存池所需时间也差不多（实际上是 O(n) 时间，不过在大部分情况下会除以一个大的因数，使其变成 O(1)）。 可以预先分配错误处理池（Error-handling pools），以便程序在常规内存被耗尽时仍可以恢复。 有非常易于使用的标准实现。 池式内存的缺点是： 内存池只适用于操作可以分阶段的程序。 内存池通常不能与第三方库很好地合作。 如果程序的结构发生变化，则不得不修改内存池，这可能会导致内存管理系统的重新设计。 您必须记住需要从哪个池进行分配。另外，如果在这里出错，就很难捕获该内存池。 回页首 垃圾收集 垃圾收集（Garbage collection）是全自动地检测并移除不再使用的数据对象。垃圾收集器通常会在当可用内存减少到少于一个具体的阈值时运行。通常，它们以程序所知的可用的一组“基本”数据 —— 栈数据、全局变量、寄存器 —— 作为出发点。然后它们尝试去追踪通过这些数据连接到每一块数据。收集器找到的都是有用的数据；它没有找到的就是垃圾，可以被销毁并重新使用这些无用的数据。为了有效地管理内存，很多类型的垃圾收集器都需要知道数据结构内部指针的规划，所以，为了正确运行垃圾收集器，它们必须是语言本身的一部分。 收集器的类型 复制（copying）： 这些收集器将内存存储器分为两部分，只允许数据驻留在其中一部分上。它们定时地从“基本”的元素开始将数据从一部分复制到另一部分。内存新近被占用的部分现在成为活动的，另一部分上的所有内容都认为是垃圾。另外，当进行这项复制操作时，所有指针都必须被更新为指向每个内存条目的新位置。因此，为使用这种垃圾收集方法，垃圾收集器必须与编程语言集成在一起。 标记并清理（Mark and sweep）：每一块数据都被加上一个标签。不定期的，所有标签都被设置为 0，收集器从“基本”的元素开始遍历数据。当它遇到内存时，就将标签标记为 1。最后没有被标记为 1 的所有内容都认为是垃圾，以后分配内存时会重新使用它们。 增量的（Incremental）：增量垃圾收集器不需要遍历全部数据对象。因为在收集期间的突然等待，也因为与访问所有当前数据相关的缓存问题（所有内容都不得不被页入（page-in）），遍历所有内存会引发问题。增量收集器避免了这些问题。 保守的（Conservative）：保守的垃圾收集器在管理内存时不需要知道与数据结构相关的任何信息。它们只查看所有数据类型，并假定它们 可以全部都是指针。所以，如果一个字节序列可以是一个指向一块被分配的内存的指针，那么收集器就将其标记为正在被引用。有时没有被引用的内存会被收集，这样会引发问题，例如，如果一个整数域中包含一个值，该值是已分配内存的地址。不过，这种情况极少发生，而且它只会浪费少量内存。保守的收集器的优势是，它们可以与任何编程语言相集成。 Hans Boehm 的保守垃圾收集器是可用的最流行的垃圾收集器之一，因为它是免费的，而且既是保守的又是增量的，可以使用 --enable-redirect-malloc 选项来构建它，并且可以将它用作系统分配程序的简易替代者（drop-in replacement）（用 malloc/ free 代替它自己的 API）。实际上，如果这样做，您就可以使用与我们在示例分配程序中所使用的相同的 LD_PRELOAD 技巧，在系统上的几乎任何程序中启用垃圾收集。如果您怀疑某个程序正在泄漏内存，那么您可以使用这个垃圾收集器来控制进程。在早期，当 Mozilla 严重地泄漏内存时，很多人在其中使用了这项技术。这种垃圾收集器既可以在 Windows® 下运行，也可以在 UNIX 下运行。 垃圾收集的一些优点： 您永远不必担心内存的双重释放或者对象的生命周期。 使用某些收集器，您可以使用与常规分配相同的 API。 其缺点包括： 使用大部分收集器时，您都无法干涉何时释放内存。 在多数情况下，垃圾收集比其他形式的内存管理更慢。 垃圾收集错误引发的缺陷难于调试。 如果您忘记将不再使用的指针设置为 null，那么仍然会有内存泄漏。 回页首 结束语 一切都需要折衷：性能、易用、易于实现、支持线程的能力等，这里只列出了其中的一些。为了满足项目的要求，有很多内存管理模式可以供您使用。每种模式都有大量的实现，各有其优缺点。对很多项目来说，使用编程环境默认的技术就足够了，不过，当您的项目有特殊的需要时，了解可用的选择将会有帮助。下表对比了本文中涉及的内存管理策略。 表 1. 内存分配策略的对比 策略 分配速度 回收速度 局部缓存 易用性 通用性 实时可用 SMP 线程友好 定制分配程序 取决于实现 取决于实现 取决于实现 很难 无 取决于实现 取决于实现 简单分配程序 内存使用少时较快 很快 差 容易 高 否 否 GNU malloc 中 快 中 容易 高 否 中 Hoard 中 中 中 容易 高 否 是 引用计数 N/A N/A 非常好 中 中 是（取决于 malloc 实现） 取决于实现 池 中 非常快 极好 中 中 是（取决于 malloc 实现） 取决于实现 垃圾收集 中（进行收集时慢） 中 差 中 中 否 几乎不 增量垃圾收集 中 中 中 中 中 否 几乎不 增量保守垃圾收集 中 中 中 容易 高 否 几乎不 参考资料 您可以参阅本文在 developerWorks 全球站点上的 英文原文。 Web 上的文档 GNU C Library 手册的 obstacks 部分 提供了 obstacks 编程接口。 Apache Portable Runtime 文档 描述了它们的池式分配程序的接口。 基本的分配程序 Doug Lea 的 Malloc 是最流行的内存分配程序之一。 BSD Malloc 用于大部分基于 BSD 的系统中。 ptmalloc 起源于 Doug Lea 的 malloc，用于 GLIBC 之中。 Hoard 是一个为多线程应用程序优化的 malloc 实现。 GNU Memory-Mapped Malloc（GDB 的组成部分） 是一个基于 mmap() 的 malloc 实现。 池式分配程序 GNU Obstacks（GNU Libc 的组成部分）是安装最多的池式分配程序，因为在每一个基于 glibc 的系统中都有它。 Apache 的池式分配程序（Apache Portable Runtime 中） 是应用最为广泛的池式分配程序。 Squid 有其自己的池式分配程序。 NetBSD 也有其自己的池式分配程序。 talloc 是一个池式分配程序，是 Samba 的组成部分。 智能指针和定制分配程序 Loki C++ Library 有很多为 C++ 实现的通用模式，包括智能指针和一个定制的小对象分配程序。 垃圾收集器 Hahns Boehm Conservative Garbage Collector 是最流行的开源垃圾收集器，它可以用于常规的 C/C++ 程序。 关于现代操作系统中的虚拟内存的文章 Marshall Kirk McKusick 和 Michael J. Karels 合著的 A New Virtual Memory Implementation for Berkeley UNIX 讨论了 BSD 的 VM 系统。 Mel Gorman's Linux VM Documentation 讨论了 Linux VM 系统。 关于 malloc 的文章 Poul-Henning Kamp 撰写的 Malloc in Modern Virtual Memory Environments 讨论的是 malloc 以及它如何与 BSD 虚拟内存交互。 Berger、McKinley、Blumofe 和 Wilson 合著的 Hoard -- a Scalable Memory Allocator for Multithreaded Environments 讨论了 Hoard 分配程序的实现。 Marshall Kirk McKusick 和 Michael J. Karels 合著的 Design of a General Purpose Memory Allocator for the 4.3BSD UNIX Kernel 讨论了内核级的分配程序。 Doug Lea 撰写的 A Memory Allocator 给出了一个关于设计和实现分配程序的概述，其中包括设计选择与折衷。 Emery D. Berger 撰写的 Memory Management for High-Performance Applications 讨论的是定制内存管理以及它如何影响高性能应用程序。 关于定制分配程序的文章 Doug Lea 撰写的 Some Storage Management Techniques for Container Classes 描述的是为 C++ 类编写定制分配程序。 Berger、Zorn 和 McKinley 合著的 Composing High-Performance Memory Allocators 讨论了如何编写定制分配程序来加快具体工作的速度。 Berger、Zorn 和 McKinley 合著的 Reconsidering Custom Memory Allocation 再次提及了定制分配的主题，看是否真正值得为其费心。 关于垃圾收集的文章 Paul R. Wilson 撰写的 Uniprocessor Garbage Collection Techniques 给出了垃圾收集的一个基本概述。 Benjamin Zorn 撰写的 The Measured Cost of Garbage Collection 给出了关于垃圾收集和性能的硬数据（hard data）。 Hans-Juergen Boehm 撰写的 Memory Allocation Myths and Half-Truths 给出了关于垃圾收集的神话（myths）。 Hans-Juergen Boehm 撰写的 Space Efficient Conservative Garbage Collection 是一篇描述他的用于 C/C++ 的垃圾收集器的文章。 Web 上的通用参考资料 内存管理参考 中有很多关于内存管理参考资料和技术文章的链接。 关于内存管理和内存层级的 OOPS Group Papers 是非常好的一组关于此主题的技术文章。 C++ 中的内存管理讨论的是为 C++ 编写定制的分配程序。 Programming Alternatives: Memory Management 讨论了程序员进行内存管理时的一些选择。 垃圾收集 FAQ 讨论了关于垃圾收集您需要了解的所有内容。 Richard Jones 的 Garbage Collection Bibliography 有指向任何您想要的关于垃圾收集的文章的链接。 书籍 Michael Daconta 撰写的 C++ Pointers and Dynamic Memory Management 介绍了关于内存管理的很多技术。 Frantisek Franek 撰写的 Memory as a Programming Concept in C and C++ 讨论了有效使用内存的技术与工具，并给出了在计算机编程中应当引起注意的内存相关错误的角色。 Richard Jones 和 Rafael Lins 合著的 Garbage Collection: Algorithms for Automatic Dynamic Memory Management 描述了当前使用的最常见的垃圾收集算法。 在 Donald Knuth 撰写的 The Art of Computer Programming 第 1 卷 Fundamental Algorithms 的第 2.5 节“Dynamic Storage Allocation”中，描述了实现基本的分配程序的一些技术。 在 Donald Knuth 撰写的 The Art of Computer Programming 第 1 卷 Fundamental Algorithms 的第 2.3.5 节“Lists and Garbage Collection”中，讨论了用于列表的垃圾收集算法。 Andrei Alexandrescu 撰写的 Modern C++ Design 第 4 章“Small Object Allocation”描述了一个比 C++ 标准分配程序效率高得多的一个高速小对象分配程序。 Andrei Alexandrescu 撰写的 Modern C++ Design 第 7 章“Smart Pointers”描述了在 C++ 中智能指针的实现。 Jonathan 撰写的 Programming from the Ground Up 第 8 章“Intermediate Memory Topics”中有本文使用的简单分配程序的一个汇编语言版本。 来自 developerWorks 自我管理数据缓冲区内存 （developerWorks，2004 年 1 月）略述了一个用于管理内存的自管理的抽象数据缓存器的伪 C （pseudo-C）实现。 A framework for the user defined malloc replacement feature （developerWorks，2002 年 2 月）展示了如何利用 AIX 中的一个工具，使用自己设计的内存子系统取代原有的内存子系统。 掌握 Linux 调试技术 （developerWorks，2002 年 8 月）描述了可以使用调试方法的 4 种不同情形：段错误、内存溢出、内存泄漏和挂起。 在 处理 Java 程序中的内存漏洞 （developerWorks，2001 年 2 月）中，了解导致 Java 内存泄漏的原因，以及何时需要考虑它们。 在 developerWorks Linux 专区中，可以找到更多为 Linux 开发人员准备的参考资料。 从 developerWorks 的 Speed-start your Linux app 专区中，可以下载运行于 Linux 之上的 IBM 中间件产品的免费测试版本，其中包括 WebSphere® Studio Application Developer、WebSphere Application Server、DB2® Universal Database、Tivoli® Access Manager 和 Tivoli Directory Server，查找 how-to 文章和技术支持。 通过参与 developerWorks blogs 加入到 developerWorks 社区。 可以在 Developer Bookstore Linux 专栏中定购 打折出售的 Linux 书籍。 关于作者 Jonathan Bartlett 是 Programming from the Ground Up 一书的作者，这本书介绍的是 Linux 汇编语言编程。Jonathan Bartlett 是 New Media Worx 的总开发师，负责为客户开发 Web、视频、kiosk 和桌面应用程序。您可以通过 johnnyb@eskimo.com 与 Jonathan 联系。