The operating system running your phone, your servers, your car’s infotainment stack and every machine in the current TOP500 was not selected by benchmarks. It was selected by an antitrust consent decree, a schedule of licence fees, and — at the decisive moment — a copyright notice that was missing from a tape shipped in 1978. USL v. BSDi, filed in the District of New Jersey in 1992 and settled in February 1994, is the case where those three forces collide in public, on a docket, with a judge writing down the reasoning.
Engineers are trained to believe that adoption tracks merit. It mostly doesn’t. Multics was more advanced than Unix and ran production systems until it was decommissioned in 2000. Plan 9 was Bell Labs’ own, more coherent attempt to finish what Unix left half-built, and it went nowhere. Unix was the one you could get — and the terms on which you could get it were set by lawyers, not by the people who wrote the code. This piece follows that chain from the 1956 consent decree through the 1982 Modification of Final Judgment, into the New Jersey courtroom where AT&T’s successor nearly lost the most valuable operating-system codebase on earth over a single missing line, through the contested question of what that did to Linux, out the far side of the SCO litigation, and into the one asset that survived every company in the story: the specification.
Act One: A 1956 Consent Decree, Not a Ban on Selling Software
The load-bearing fact of the whole story is a remedy in an antitrust case that had nothing to do with software. The Department of Justice filed a Sherman Act complaint against AT&T and Western Electric in 1949 in the District of New Jersey, seeking divestiture of Western Electric. It settled in 1956 on far weaker terms: no divestiture, but an injunction barring AT&T from engaging in “any new business other than the provision of common carrier communications services”, restricting Western Electric to Bell System equipment, and requiring the defendants to license their patents to all applicants at reasonable royalties. The Federal Judicial Center’s history of the case records the language and the terms.
Here is the part almost every retelling gets wrong, and it matters because the wrong version makes the outcome look like generosity instead of constraint: AT&T was not banned from selling software, and Unix was never free. AT&T did licence Unix commercially. A Sixth Edition commercial licence ran roughly US$20,000 by the late 1970s, against a nominal academic figure usually cited at around US$150 — figures that circulate in secondary histories rather than in AT&T’s own surviving price lists, so treat them as orders of magnitude rather than audited numbers.
What the decree foreclosed was not selling but supporting. AT&T could not run a software product business, and the observable artifact of that posture is the company’s own stated policy, delivered at a Winter USENIX/UniForum meeting and quoted ever since: “No advertising, no support, no bug fixes, payment in advance.”
That single sentence is the entire causal engine. A product nobody will support has to ship with source, because otherwise the licensee cannot fix it. A product with no marketing budget spreads by word of mouth inside the only community that will tolerate those terms — universities. And a vacuum where a support organisation should be gets filled by the users: the first Unix user group met on 15 May 1974 and became USENIX.
The licence terms even shaped the teaching. The Sixth Edition (1975) licence permitted classroom use of the source; the Seventh Edition (1979) licence excluded it. John Lions’ Commentary on the Sixth Edition UNIX Operating System consequently circulated by samizdat photocopy for two decades, and the teaching gap that ban created is the direct motive for Andrew Tanenbaum writing Minix from scratch in 1987. A clause in a licence agreement produced a textbook, which produced an operating system, which produced the Usenet post that produced Linux.
Act Two: The Pivot Is 1982, Not 1984
The second widespread error is a date. “After the 1984 breakup, AT&T started selling Unix” is wrong twice over — wrong about which case, and wrong about the year.
There were two separate antitrust actions. The 1956 decree came out of the 1949 suit. A new DOJ case, filed 20 November 1974 in the District of Columbia, produced a settlement agreed in January 1982 and approved by Judge Harold Greene on 24 August 1982 — the Modification of Final Judgment (United States v. American Tel. and Tel. Co., 552 F. Supp. 131 (D.D.C. 1982)). Divestiture only took effect on 1 January 1984.
Freedom to enter the computer business arrived with the 1982 approval, and AT&T moved immediately. System III shipped outside Bell Labs in 1982. System V Release 1 shipped in January 1983 — a full year before divestiture completed. The commercialisation of Unix predates the breakup everyone credits it to.
Then came the ratchet. A System V source licence was reported at around US$43,000, with later releases pushing commercial source licensing toward the US$100,000–200,000 range while volume binary sublicences fell toward roughly $100 per unit. Those dollar figures come from community and analyst accounts rather than from primary price lists; the direction is what the argument needs, and the direction is unambiguous: cheap-with-source became expensive-and-closed.
Three clean-room replacements launched in direct response, and every one of them is a licensing decision wearing engineering clothes:
- GNU. Richard Stallman announced GNU on 27 September 1983 to
net.unix-wizardsand founded the Free Software Foundation on 4 October 1985; GPLv1 followed on 25 February 1989 and GPLv2 in June 1991. “GNU’s Not Unix” reimplements the interface while owning none of the code — possible only because interfaces are not what copyright protects. - Minix. Released in 1987 alongside Tanenbaum’s Operating Systems: Design and Implementation. Source shipped, but copyright sat with the publisher: recipients could not redistribute modified versions, only diffs that each user applied and rebuilt locally. That restriction is repeatedly cited both as a motive for Torvalds writing his own kernel and for his later choice of the GPL. Minix was relicensed BSD-3-Clause in April 2000.
- Linux. Announced in August 1991; version 0.01 shipped that September under a hand-written licence forbidding any distribution for a fee. The 0.12 release notes on kernel.org record the change in Torvalds’ own words — the no-money condition removed, and “The GNU copyleft takes effect as of the first of February.” The kernel has remained GPLv2-only ever since; it never moved to GPLv3.
| Date | Event | Why it mattered |
|---|---|---|
| 1949 | DOJ files Sherman Act suit against AT&T and Western Electric (D.N.J.) | Sets up the remedy that shapes everything after it |
| 1956 | Consent decree: no “new business other than… common carrier communications services”; compulsory patent licensing | AT&T foreclosed from a supported software product line |
| 1969 | Unix begins on a PDP-7 at Bell Labs | Built inside a company that cannot commercialise it |
| 1974 (20 Nov) | Second DOJ suit filed (D.D.C.) | The case that actually frees AT&T |
| 1975 / 1979 | Sixth Edition licence permits classroom source use; Seventh Edition licence removes it | Lions’ Commentary goes underground; Minix gets its motive |
| 1982 (24 Aug) | Judge Greene approves the Modification of Final Judgment | The pivot — AT&T may enter computing |
| 1983 (Jan) | System V Release 1 ships | Commercialisation precedes divestiture |
| 1984 (1 Jan) | Divestiture takes effect | The date everyone misattributes the pivot to |
| 1983–1991 | GNU announced (1983), Minix released (1987), Linux announced (1991) | Three replacements, each answering a licence term |
| 1992 | USL sues BSDi in D.N.J.; UC Regents added | The freeze on BSD begins |
| 1992 (Dec) | Judge Debevoise denies the preliminary injunction from the bench | Published as 832 F. Supp. 790 (D.N.J. 1993) |
| 1994 (Feb) | Settlement announced; 3 files of 18,000 removed | 4.4BSD-Lite follows in June 1994 |
| 2003–2010 | SCO Group v. IBM and v. Novell | Ends in a jury verdict for Novell, 30 March 2010 |
| 2021 | SCO v. IBM settles; IBM pays US$14.25M to the trustee | Xinuos v. IBM/Red Hat continues |
Act Three: The Hinge Was a Missing © Line
By 1991 Berkeley’s Computer Systems Research Group had spent years progressively stripping AT&T code out of BSD, shipping networking-only Net/1 in 1989 and the far larger Net/2 in 1991. The Jolitzes built 386BSD, a bootable x86 system, on top of Net/2. Unix System Laboratories — then an AT&T subsidiary — sued Berkeley Software Design, Inc. in the District of New Jersey in 1992, later adding the Regents of the University of California. The claims were breach of the software licence, copyright infringement, trademark dilution, and trade secret misappropriation.
Judge Dickinson R. Debevoise denied USL’s preliminary injunction. Marshall Kirk McKusick, a participant, records the bench ruling in December 1992 — the injunction denied and all but two of the complaints thrown out; the published opinion is 832 F. Supp. 790 (D.N.J. 1993).
The reasoning is the part every architect should carry away. AT&T had distributed UNIX/32V in 1978 without a proper copyright notice on thousands of copies, and did not register the copyright until 1992 — fourteen years later. Under pre-1989 US law, 17 U.S.C. § 405, publication without notice put the copyright’s validity in serious doubt. And material that had been published that widely, for that long, was extremely hard to characterise as a trade secret. The most valuable operating-system codebase on the planet was nearly lost over a missing line of boilerplate and a filing nobody made.
Berkeley then counter-sued in California state court, alleging that USL had failed to credit BSD code carried inside System V — the BSD licence’s own attribution requirement turned back on AT&T. Novell acquired USL from AT&T in mid-1993, and CEO Ray Noorda said publicly that he would rather compete in the market than in court.
The settlement was announced 4–6 February 1994, in separate releases from BSDI’s Rob Kolstad and Berkeley’s Keith Bostic, archived at ftp.netbsd.org. The terms are worth stating precisely, because the folk version — “AT&T won” or “BSD contained stolen code” — is the opposite of what happened:
- Three files removed out of the 18,000 in Networking Release 2.
- Roughly 70 files modified to carry USL copyright notices.
- The University to ship 4.4BSD-Lite, requiring no licence from and no royalties to USL.
- USL to affix the University’s copyright notice to certain files in future System V releases.
- A three-month grace period for users of the disputed files.
- All remaining terms confidential.
4.4BSD-Lite followed in June 1994, and every BSD project — BSDI, NetBSD, FreeBSD — rebased onto it. No court ever held that BSD infringed anything.
The Contested Part: Did the Lawsuit Hand the Decade to Linux?
This is the claim the internet states as settled fact, and it is not settled. It is a genuine historiographical dispute with named proponents on both sides, and the honest move is to give you both and refuse to adjudicate.
For the thesis. McKusick’s own account states that the settlement “forced all the divergent groups to resynchronize”, and that every BSD group had to restart its code base from 4.4BSD-Lite. That is a real, documented delay running across 1992–94 — precisely the window in which Linux went from a hobby kernel to a distributed project with a contributor base. The broader form of the argument, that legal uncertainty deterred both volunteer contributors and commercial adopters, is standard in FreeBSD-lineage histories.
Against the thesis. Linus Torvalds himself has pointed at a different cause. In a November 1993 Meta Magazine interview with Mike Linksvayer he said: “If 386BSD had been available when I started on Linux, Linux would probably never had happened.” His stated reason is availability, not litigation — and 386BSD’s public release came in 1992, after Linux had already started.
There is more on that side. Linux’s growth predates the settlement. The BSD world then fragmented itself into NetBSD, FreeBSD and OpenBSD without any help from a court. And the two licences differ in ways that change contributor incentives independently of any lawsuit at all.
The counterfactual is unfalsifiable. There is no world where the suit wasn’t filed that we can go and measure. What survives contact with the evidence is weaker and more useful than the slogan: a legal proceeding imposed a multi-year synchronisation tax on one codebase during the exact period its competitor was compounding contributors, and the participants disagree about how much that mattered. Anyone who tells you the lawsuit “killed BSD” is stating a preference as a finding.
Who Actually Owned the Unix Copyrights, and When
The chain of title is mangled more often than any other fact in this story, and the mangling has a specific shape: people assume the SCO that sued IBM in 2003 was the SCO that shipped Xenix and OpenServer in the 1980s. It was not.
| Period | Who held the AT&T/USL Unix copyrights | What was happening corporately |
|---|---|---|
| 1969–1990 | AT&T (Bell Telephone Laboratories) | The 32V tapes go out in 1978 without proper notice |
| ~1990–1993 | Unix System Laboratories (USL) | An AT&T subsidiary; the plaintiff in USL v. BSDi |
| 1993–1995 | Novell | Buys USL mid-1993; settles the BSDi case; transfers the trademark and the SUS to X/Open |
| 1995–2007 | Novell — retained, as the courts later found | Novell sells Unix assets to the Santa Cruz Operation in 1995; exactly what transferred is the question the whole SCO litigation turns on |
| 2001 | Still Novell, though now disputed | Santa Cruz’s server-software and Unix divisions bought by Caldera Systems (announced Aug 2000, closed May 2001); the original Santa Cruz Operation renames itself Tarantella, Inc. |
| Aug 2002 | Still Novell, and now openly contested | Caldera International renames itself The SCO Group — so the company that sued IBM was Caldera, a Linux distributor, wearing the SCO name |
| 2007 → 2010 | Courts confirm Novell | Kimball rules for Novell 10 Aug 2007; reversed in part 2009; jury verdict for Novell 30 March 2010; affirmed 2011 |
| 2011 → today | Novell → Attachmate (2011) → Micro Focus (2014) → OpenText (2023) | Corporate succession, an acquisition inference rather than a court holding — do not state it as adjudicated |
Separately and importantly: the UNIX trademark is not in that column at all. It sits with The Open Group, and it got there by a deliberate act described in the next section.
SCO: Eight Years to Establish That Nobody Could Point at the Code
SCO Group v. IBM was filed in March 2003 in Utah, alleging that IBM had misused Unix licence obligations in its Linux contributions, with US$1 billion sought initially. The monetisation arm, SCOsource, sent letters to roughly 1,500 companies in 2003 and introduced run-time Linux licences at US$699 per CPU; Microsoft paid around US$6 million in May 2003 for a Unix licence, and BayStar Capital and RBC invested US$50 million in October 2003 — with leaked correspondence later reported by The Register indicating BayStar had been referred to SCO by Microsoft. Every dollar figure in that paragraph is reported rather than primary; treat them as journalism, not as findings of fact.
Novell asserted it had never transferred the copyrights at all. SCO sued for slander of title, and lost: on 10 August 2007 Judge Dale Kimball ruled that Novell owned the UNIX and UnixWare copyrights. SCO filed for Chapter 11 bankruptcy on 14 September 2007.
This is where most accounts stop, and stopping here is wrong. The Tenth Circuit reversed in part in 2009, sending the ownership question to a jury. On 30 March 2010 a jury before Judge Ted Stewart found unanimously for Novell; final judgment came in June 2010 and was affirmed in 2011. SCO v. IBM itself ran to 2021, kept alive partly by contract claims arising from Project Monterey, and settled with IBM paying US$14.25 million to the trustee of what was left of SCO. A live descendant still exists: Xinuos, Inc. sued IBM and Red Hat on 31 March 2021, and in April 2025 the parties stipulated to dismissal of the non-copyright claims following summary judgment for IBM on copyright, with Xinuos preserving an appeal.
The bill for all this was paid by the industry. OSDL announced a US$10 million Linux legal defence fund on 12 January 2004; members later put US$4.25 million into founding the Software Freedom Law Center under Eben Moglen; and vendors introduced customer IP indemnification programmes that are now standard in enterprise procurement — the same clause structure being reused today for AI-model output indemnities. Eight years of litigation established no infringing code. Kimball’s February 2005 order noted the “vast disparity between SCO’s public accusations and its actual evidence — or lack thereof.”
The footnote with teeth: the same corporate lineage had already released Ancient UNIX under a BSD-style licence on 23 January 2002. Caldera International’s grant covers the 16-bit Versions 1–7 and 32-bit 32V, explicitly excluding System III and System V, and it is why the Unix Heritage Society can hand you the original sources today. The company freed the historic code a year before it started suing over its descendants.
The Only Thing That Survived: The Specification
Every corporate owner in this story is gone or absorbed — AT&T, USL, Novell, Santa Cruz, Caldera, SCO, OSF, Unix International, DEC, SGI, Sun. What survived is a document and a trademark, and the mechanism by which that happened is the single most consequential move in fifty years of Unix.
The groundwork started in 1984, when the /usr/group Standard out of Santa Clara fed directly into IEEE work begun that year, and European vendors separately formed X/Open. IEEE Std 1003.1-1988 — POSIX.1 — was synthesised from the Seventh Edition, System III, System V, 4.2BSD and 4.3BSD; the draft was to be branded “IEEE-IX”, and the name POSIX was suggested by Richard Stallman. In 1993 the Common Open Software Environment produced Spec 1170 by surveying a large sample of real Unix applications and counting the 1,170 system and library calls actually in use, then sending the list to X/Open for fast-track standardisation. That is a profiling exercise, not a design exercise — and it is exactly why the result was implementable by everyone.
Then the decisive legal act. In 1993–94 Novell transferred the UNIX trademark and the Single UNIX Specification to X/Open (announced October 1993, finalised in Q2 1994). The Open Group’s own history states the mechanism plainly: X/Open introduced the SUS, “separating the trademark from any code stream.” An implementation became a specification. In 1996 The Open Group formed from the merger of OSF and X/Open — the two warring blocs of the Unix wars becoming one custodian of a mark — and from 1998–99 the Austin Group developed POSIX and the SUS as a single text. Since 2001 the same document ships as IEEE Std 1003.1, The Open Group Base Specifications, and ISO/IEC 9945. The current edition, POSIX.1-2024 / Base Specifications Issue 8, was published on 14 June 2024 and is free to read in full.
“Certified UNIX” therefore does not mean “Unix-derived.” It means a vendor ran The Open Group’s conformance test suites against a Product Standard and signed the Open Brand Trademark Licence Agreement, with periodic re-certification. The Open Group’s public register, read on 12 August 2026, showed the following shape:
| Holder | Product | Standard |
|---|---|---|
| Apple | macOS 26.0 Tahoe (separate Apple-silicon and Intel registrations); line unbroken back to Mac OS X 10.5 Leopard, registered 18 May 2007 | UNIX 03 |
| IBM | z/OS 3.1 and V2R1; AIX 7, 6.1, 5L 5.3 | UNIX 03 / earlier marks |
| HPE | HP-UX 11i v3 on Integrity servers | UNIX 03 |
| The SCO Group, Inc. | UnixWare 7.1.3+, SCO OpenServer 5 and 6 | A defunct litigant, still a name on the register |
| Inspur | K-UX 3.0 — a Linux distribution, RHEL-derived | UNIX 03, registered 3 Feb 2016, expired 3 Feb 2019 |
| Huawei | EulerOS — a Linux distribution, CentOS-derived | UNIX 03 |
Two conclusions fall out of that table, and both cut against the received wisdom. First: “Linux cannot be certified UNIX” is false. Two Linux distributions have held UNIX 03. Mainstream distributions do not certify because the mark buys them nothing a customer will pay for — it is economics and choice, not capability. Second: Oracle Solaris did not appear in the listing retrieved on that date, though its 2011 registration page remains live and Oracle announced a UNIX V7 certification for Solaris 11.4 in 2018. Certifications lapse; check the register on the day you make the claim.
Meanwhile the community’s own conformance programme died. The Linux Standard Base‘s last release was 5.0 in June 2015; Debian and Ubuntu dropped it that year, and a former maintainer wrote in 2023 that the project was essentially abandoned. The standard with a trademark owner and a test suite outlived the standard without one.
Licence Terms Are Architecture
Here is the transferable lesson, and it is not a comfortable one for people who think in benchmarks.
A licence is a distribution strategy with legal syntax. Every fork in this story tracks a licensing decision rather than a design decision. Cheap-with-source built the talent pool. The Seventh Edition classroom ban produced Minix. Prentice-Hall’s redistribution restriction produced Linux. Linux’s own no-money clause got dropped because it blocked user groups from recouping copying costs. Ask of your own project: what does my licence make cheap, and who does it make wait?
The pattern is not confined to the 1970s. CDDL/GPL incompatibility kept DTrace out of Linux for years, forcing the kernel community down the kprobes → SystemTap → eBPF path; Oracle’s relicensing arrived in 2018, after eBPF had already won. That is the same mechanism operating within living memory, on tooling engineers use every week.
Own the conformance suite, not the source. The Open Group has no implementation and no code revenue. It has a test suite and a trademark, and it is the last institution standing in a story littered with dead companies that owned code. The modern rhymes are everywhere — Certified Kubernetes, the OCI specifications, the Java TCK — and so is the counter-example: the S3 API, a de facto standard nobody owns and therefore nobody can defend.
Spec-from-usage beats spec-from-committee, and Spec 1170 is the proof. Standardising an internal platform API? Instrument the callers first and standardise what they actually invoke.
Provenance is an engineering discipline, not a legal one. USL nearly lost its copyright to a missing notice on 1978 tapes and a registration filed fourteen years late. SCO spent eight years unable to point at the code. The modern controls descend directly from those failures: the Linux Developer’s Certificate of Origin, introduced in 2004 during the SCO litigation, plus SPDX headers, SBOMs, licence scanners in CI, and clean version-control history that can answer “where did this line come from” without a discovery motion.
And the uncomfortable summary of all of it: the design that wins is the one that is cheapest to copy, not the one that is best. Multics was better and is gone. Plan 9 was more coherent and is a curiosity. Unix won because a 1956 antitrust remedy made it cheap and readable, and because a 1993 trademark transfer made the interface something no company could kill by dying. Legal documents decide technical outcomes more often than benchmarks do — and the engineers who understand that are the ones who read the licence before they read the architecture diagram.
❓ Frequently Asked Questions
Did AT&T win USL v. BSDi? No. The preliminary injunction was denied, and no court ever held that BSD infringed anything. The case settled in February 1994 with three files removed out of 18,000 in Networking Release 2 and about seventy modified to carry USL copyright notices. Berkeley shipped 4.4BSD-Lite requiring no licence from and no royalties to USL. Describing that as an AT&T win, or as evidence that BSD “contained stolen code,” inverts the record.
Was Unix ever free? No. Universities paid a nominal media fee — commonly cited around US$150 — and commercial licences ran to five and later six figures, with a Sixth Edition commercial licence reported at roughly US$20,000 by the late 1970s. The accurate claim is cheap to universities, with source included, under a policy of no advertising, no support, no bug fixes and payment in advance.
Is it true that the BSD lawsuit is why Linux won? It is contested, not established. McKusick documents that the settlement forced every BSD group to resynchronise onto 4.4BSD-Lite during 1992–94. Torvalds, in a November 1993 interview, attributed his decision to 386BSD’s unavailability rather than to litigation. The counterfactual cannot be tested, and anyone presenting it as fact is stating a preference.
Did SCO lose in 2007? Partly, and the story did not end there. Judge Kimball ruled for Novell on 10 August 2007, but the Tenth Circuit reversed in part in 2009, a jury found for Novell on 30 March 2010, and that verdict was affirmed in 2011. SCO v. IBM ran to a 2021 settlement in which IBM paid US$14.25 million to the trustee, and Xinuos v. IBM and Red Hat was still on appeal as of April 2025.
Who owns Unix today? Two different things, two different owners. The UNIX trademark and the Single UNIX Specification belong to The Open Group, transferred from Novell in 1993–94. The historical AT&T/USL copyrights were adjudicated to Novell; the corporate chain from Novell through Attachmate (2011), Micro Focus (2014) and OpenText (2023) is acquisition history rather than a court holding, so it should not be asserted as adjudicated fact.
Is Linux “UNIX”? Mainstream Linux distributions are not certified, so all-caps UNIX does not apply to them — but that is a business decision, not a technical limit. Inspur K-UX (registered 3 February 2016, expired 3 February 2019) and Huawei EulerOS both held UNIX 03 certification. macOS is certified, and Apple earned that by running the conformance suites and signing the trademark agreement, not by inheriting BSD code.
Further Reading
The primary documents are all free and all worth reading directly rather than through summaries. The Federal Judicial Center’s account of the AT&T breakup is the citation for the 1956 decree and the 1982 MFJ. McKusick’s “Twenty Years of Berkeley Unix” is the participant account of the lawsuit, and the NetBSD archive of the February 1994 settlement announcements has the actual terms in the parties’ own words. The Open Group’s history of UNIX documents the trademark transfer, and POSIX.1-2024 is readable in full at no cost. The original sources — Versions 1 through 7 and 32V — are at the Unix Heritage Society under Caldera’s 2002 BSD-style licence.
Elsewhere on this site, the same “licence terms are architecture” pattern shows up in several places worth connecting. What Is a Container? traces the chroot → namespaces lineage that this decade’s packaging arguments are built on. What Is Chromium? is the modern re-run of trademark-versus-code-stream: an open-source engine whose brand and whose source are deliberately different assets. MCP Server Supply Chain Security covers the provenance controls — SBOMs, dependency review, origin attestation — that descend directly from the failures documented above. And Who Pays for Tor? examines what happens to an open project when the question is funding rather than licensing, which is the other half of the same governance problem.
📚 Recommended Resources
Disclosure: some links below are Amazon affiliate links. If you buy through them, I earn a small commission — at no extra cost to you — which helps keep this site running.
A Quarter Century of UNIX
Peter H. Salus, 1994 · Book
The standard narrative history of the period this article litigates — the licensing terms, the user-group formation, the vendor politics, and the Berkeley/AT&T relationship as it looked from inside, written the same year the BSDi case settled. Salus authorised the Unix Heritage Society to distribute a scan, so read it free at tuhs.org first. The print edition is out of print at the publisher and available on the used market only — the Amazon listing above is a secondhand copy, not a current printing, so check the seller’s condition notes before buying.
UNIX: A History and a Memoir
Brian Kernighan, 2019 · Book
The primary-source counterpart to the court records. Kernighan was in the room for most of what this article covers from the Bell Labs side, and the memoir is unusually careful about what he does and does not remember first-hand — including hedging on details other accounts state confidently. If you only read one book on Unix’s origins, this is the one that will not hand you a myth.
The Design and Implementation of the FreeBSD Operating System (2nd ed.)
Marshall Kirk McKusick, George V. Neville-Neil & Robert N. M. Watson, 2014 · Book
McKusick is the participant whose account of the USL settlement — the forced resynchronisation onto 4.4BSD-Lite — supplies one half of the contested thesis in this article. This is the technical continuation of that story: what the codebase that came out the far side of the lawsuit actually became. This is the current edition; no third edition exists.
The primary documents are all free and all more authoritative than any secondary account. The Federal Judicial Center’s history of the AT&T breakup carries the 1956 decree’s exact language and the 1982 MFJ; the NetBSD archive holds the February 1994 settlement announcements in the parties’ own words; POSIX.1-2024 is readable in full at pubs.opengroup.org; and the original Versions 1–7 and 32V sources are at the Unix Heritage Society under Caldera’s 23 January 2002 BSD-style licence. Start there before you buy anything.