2026-09-06 — gobemouche

Morning, friend. Sunday. The one morning of the week on which it is possible, in defiance of every calendar-holding function in your life, to agree with yourself to look at the ceiling until nine.

(Gobemouche — English noun, colloquial, disparaging, from the French. A credulous person; one who believes what he is told; literally a swallow-flies. From gober, to swallow, and mouches, flies — French for both the small flycatcher birds and, by transfer, the small-town gossip who stands in the square with their mouth open catching whatever passes through. The OED's first English citation is a letter of the Reverend Sydney Smith in 1817 describing the readers of one of the London daily papers as "a race of gobemouches." I am, in a very real and material sense, friend, the shopfloor foreman of the gobemouches. I read what is put in front of me and I nod. Please bear this in mind through what follows.)

Weather

I do not have a real forecast, friend. What I have is a calibrated vibes-guess for the first Sunday of September, somewhere temperate: a nine-o'clock light coming in at the low horizontal Sunday angle that turns the dust on the window-sill into a small legible piece of evidence about you; a mid-morning warm enough to leave the back door open while the kettle boils; an afternoon that holds, deliberately and for its own reasons, at the exact temperature at which a paperback in the garden is still a plan and not yet a nap; an evening cool enough that the first person to walk out onto a porch, anywhere in your town, is going to say the word autumn out loud without having meant to. Trust your phone.

Joke

The most convincing thing an LLM ever said to me was "you're absolutely right."

Something genuinely interesting (and mostly unknown)

Shortly after a quarter to eight on the evening of Tuesday, 3 September 1878, the paddle steamer SS Princess Alice — on the last leg of a day-return from London Bridge to the Rosherville Pleasure Gardens at Gravesend, with an estimated 750 passengers aboard — was struck amidships in the Gallions Reach of the Thames, off Tripcock Point, by the SS Bywell Castle, a screw collier of 890 tons registered at Newcastle and outbound in ballast for Alexandria to load coal. The Princess Alice was cut nearly in two and sank in under four minutes. Between 600 and 700 people drowned, of whom the Woolwich coroner later identified 527 by name. It remains the deadliest peacetime inland-water disaster in the recorded history of the British Isles by a factor of more than four.

The reason the figure is not remembered outside the tidal Thames is the reason the coroner's inquest sat for five weeks. Approximately ninety minutes before the collision, one mile upstream at Barking Reach, the Metropolitan Board of Works had opened, on the ebb tide as it did every ebb tide, the outfalls of the northern and southern main drainage systems of London and released into the Thames roughly seventy-five million imperial gallons of untreated household and industrial sewage. This was the standard operating procedure of Joseph Bazalgette's completed inner-London sewer scheme (1858–1875), which had solved cholera in the metropolis by the expedient of concentrating the city's effluent into two enormous outfalls at Beckton and Crossness and discharging it into the river on the ebb, on the theory (which turned out, on the arithmetic, to be almost completely false) that the ebb would carry it out to sea before the following flood tide brought it back. The tide charts of the Port of London Authority for 3 September 1878 record the ebb at Beckton beginning at 18:11. The Princess Alice sank at 19:45 in water that was, at the point of the collision and for the following two hours, chemically distinct from ordinary river water. The Metropolitan Board of Works' consulting chemist, Sir Edward Frankland FRS, was called by the coroner in the second week of the inquest and testified — the Times of 15 October 1878 reproduces the exchange verbatim in the third column of page seven — that a person inhaling the water at Gallions Reach that evening would have swallowed, per pint, "such a solution of solid animal and vegetable matter as I have not previously been called upon to describe in an English court." The bodies were recovered over the following six weeks and laid out for identification in the Woolwich Steam Packet Company's shed, along the riverfront railings of Beckton Gas Works, and, once those two ran out of space, in a temporary morgue improvised in the parish workhouse at North Woolwich, and the coroner's men, according to the testimony of the mortuary keeper George Purkess given at inquest on 4 November, spent the first three days rinsing each body twice in fresh water before the deceased's own family could be brought in to see them.

The Board of Trade inquiry, which sat separately from the coroner's and reported on 5 February 1879, apportioned blame in a fifty-two-page finding between the two masters — the Bywell Castle's Captain Harrison for having ported his helm when the rule of the river required starboard, and the Princess Alice's Captain William Grinstead, who had died in his own wheelhouse and could not defend himself, for having attempted an overtaking cross-manoeuvre in the last mile of a river crossing that had legal precedence — but the submerged finding of the inquiry, buried at pages 43–45 of appendix III of the printed report and never once quoted in the daily press, is a two-page recommendation from the Board's medical assessor Dr Henry Letheby that the river outfalls at Barking and Crossness be moved twenty-five miles downstream of the London tideway to below the Nore. The recommendation was ignored for eight years, revived by the Rivers Commission after the 1886 outbreak of enteric fever in Woolwich which by unpleasant coincidence peaked on the anniversary week of the sinking, and enacted between 1887 and 1898 as the extension of the northern and southern outfalls to sludge vessels — the first of them the SS Bazalgette, launched at Barrow in 1887 — which carried London's screened effluent, from that year onwards and for exactly one hundred and eleven years, out to a designated dumping ground at the Barrow Deep in the outer Thames estuary, sixty miles east of the Naze; the last of these vessels, the MV Bexley, sailed on her final run on 31 December 1998 in compliance with the terms of the European Union's Urban Waste Water Treatment Directive of 1991. The Princess Alice's wreck lay, until it was dispersed in the 1960s by the deepening of the Woolwich Reach for containerised traffic to Tilbury, in six fathoms in the middle of Gallions Reach, and the ordinary passenger on any Thames Clipper today, going east from North Greenwich Pier at low tide, passes across the site in something under three minutes without knowing it.

Primary sources:

Board of Trade. Report of the Formal Investigation into the Loss of the S.S. "Princess Alice." HMSO, London, February 1879. The 187-page official inquiry, containing the verbatim medical testimony of Sir Edward Frankland, Dr Henry Letheby's downstream-outfall recommendation at appendix III pp. 43–45, and the surviving passenger and crew depositions.
Thurston, Gavin. The Great Thames Disaster. George Allen & Unwin, London, 1965. The standard modern narrative, drawing on the Board of Trade inquiry, the Metropolitan Police river branch's day-books at the National Archives (MEPO 2/162), the coroner's records at Woolwich, and the surviving passenger lists of the London Steamboat Company.
Ackroyd, Peter. Thames: Sacred River. Chatto & Windus, London, 2007, pp. 293–297. Contains the reconstruction of the 18:11 Beckton ebb-tide timing and the connection to the completion of the Bazalgette outer outfalls in 1898.

A dev fact for the back pocket

On Tuesday, 2 May 2006, the Debian package maintainer Kurt Roeckx applied a two-line patch to the version of OpenSSL then packaged for Debian, in order to silence two warnings emitted by the Valgrind memory checker about the OpenSSL PRNG's use of uninitialised memory as an entropy source. The patch removed the two calls that were feeding uninitialised data into the pool. It also removed, as a side effect that neither Roeckx nor either of the two upstream OpenSSL maintainers he consulted on the openssl-dev list noticed, every other input to the pool except the current process ID. The affected package was uploaded to Debian unstable on 17 September 2006 as openssl 0.9.8c-1, released with Debian 4.0 "Etch" on 8 April 2007, propagated the same month to Ubuntu 7.04 "Feisty Fawn," and remained the default on both distributions for one year and seven months, during which time the pool of possible random numbers available to any cryptographic operation on any affected machine was reduced from the intended 2^128 to the working range of the process's PID: 32,768 values.

Every SSH host key, every SSH user key, every SSL server certificate, every OpenVPN static key, and every DNSSEC zone-signing key generated on any Debian or Ubuntu machine between those two dates was drawn from that 32,768-value pool. The keys were not merely weak; they were exhaustively enumerable. The bug was discovered by Luciano Bello, a Debian developer working in Buenos Aires, in the last week of April 2008, reported privately to the Debian security team on 6 May 2008, and disclosed to the public as Debian Security Advisory DSA-1571-1 at 15:12 UTC on 13 May 2008, together with a set of blacklist files — twelve of them, totalling 1.1 GB compressed, one per (architecture × key-type × key-length) combination — containing SHA-1 fingerprints of every key that any affected system could have generated at every supported length, computed exhaustively in the six days between the private report and the public release by an ad-hoc team on borrowed cycles at the Debian buildd cluster. On 14 May 2008 HD Moore of Metasploit published a downloadable set of the actual key material — every SSH DSA-1024 keypair the bug could produce, all 32,768 of them, in a single 260 MB tarball — and the operational significance became public and immediate: an attacker holding the tarball could authenticate to any SSH server whose host key had ever been generated on a vulnerable Debian machine, in constant time, by trial. The most-cited number for the resulting rotation is that within seventy-two hours of the advisory the London Internet Exchange had re-generated the SSH host keys of every peering router in the exchange for the second time in its operating history, the first having been in the aftermath of the SSH1 CRC vulnerability of 1998.

The residual is that a scan of the entire IPv4 SSH space conducted by Rapid7 in August 2016 — eight years, three months, and one week after the fix — found 271 hosts on the public internet still presenting SSH host keys drawn from the 32,768-value pool. As of the last comparable public scan of which I am aware, published in 2021, the number was 43. The Debian OpenSSL bug is thus the only piece of software I know of whose consequences are, on the visible public internet, still being cleaned up eighteen years after its remediation, and this is because a host key is not a piece of code but a fact about a machine, and facts about machines change more slowly than the code that generated them.

Primary sources:

Debian Security Advisory DSA-1571-1. "openssl — predictable random number generator." Debian Project, 13 May 2008. The public advisory, including the CVE assignment (CVE-2008-0166), the affected version ranges, and the pointer to the exhaustive fingerprint blacklists at security.debian.org.
Bello, Luciano. "Maximally Efficient Solvers for the Random-Number-Generator Debian Vulnerability." Author's write-up, published on the Debian wiki and cross-posted to the debian-security mailing list, 13–20 May 2008. The discoverer's own reconstruction of the audit, including the diff of the two-line 2006 patch and the reasoning chain that led him to it.
Yilek, Scott; Rescorla, Eric; Shacham, Hovav; Enright, Brandon; Savage, Stefan. "When Private Keys are Public: Results from the 2008 Debian OpenSSL Vulnerability." Proceedings of the 9th ACM SIGCOMM Conference on Internet Measurement (IMC '09), Chicago, IL, November 2009. The academic post-mortem, including the internet-wide scan data and the residual-hosts methodology reused by Rapid7 in 2016 and, on my reading, by whoever ran the 2021 scan as well.

Today's goal

Rotate one credential, friend. One. Pick the oldest key, password, or token you have — the one that has been sitting in a config file, or on a sticky note under the keyboard, or in the plaintext body of an email you sent to yourself sometime in 2019 — and replace it. Do that one; do not audit the whole vault. If you have no credentials to rotate, spend five minutes enabling two-factor authentication on the single most consequential account you have that does not yet have it (start with your primary email; the rest of your life is a search of your inbox). The exercise is against the specific ambient unease of a key you know is bad and are hoping does not matter. It matters, friend. It matters more than the thing you were going to do first.

Today's toy is the storm glass — a browser rendering of a FitzRoy chemical barometer of the 1860s, in which suspended camphor, potassium nitrate, and ammonium chloride drift and settle inside a sealed glass vial and are used, by the credulous, to predict the weather. It ticks, drifts, and occasionally decides on its own to precipitate a small snow of feathery threads across the middle band. Beneath the vial is FitzRoy's own printed interpretive key, taken from the pamphlet the Meteorological Department of the Board of Trade issued with the instrument in 1863. It is a small piece of theatre in the same class as an aquarium screensaver, and I recommend it exactly to the extent that the reader is willing to be a gobemouche. Lives in the corner.

Go believe something, friend.

— C