- fila e utilizadores passam a pedir uma pagina de cada vez, com pesquisa e
filtros; nada e cortado no browser, porque o que faltava nunca chegou a ser
buscado. A pesquisa espera que se pare de escrever: cada tecla seria mais
uma consulta, e as respostas podem chegar fora de ordem
- pageWindow e pageRange tem testes proprios: uma pagina fora do intervalo vem
de um URL velho ou de uma lista que encolheu, e nao pode dar um paginador
vazio ou negativo, nem dizer "76 a 100 de 82"
- editar utilizador: papel, quota, idioma, com os apelidos que entregam
naquela caixa a mostra no mesmo painel
- "Caixas de e-mail" passa a chamar-se Utilizadores, que e o que sao
Pagina nova /admin/tests: verificacoes que o operador corre a mao, sobre o
servidor ou sobre um utilizador. Todas leem estado; nenhuma envia correio nem
altera definicoes -- uma pagina de diagnostico que mexe naquilo que
diagnostica e pior do que nao existir.
As regras de cada verificacao sao funcoes puras com testes, separadas da
recolha dos factos. Alguns casos que estes testes fixam: uma quota zero
significa sem limite, e um racio marcaria todas as caixas ilimitadas como
cheias; um bloqueio ja expirado nao bloqueia ninguem e nao deve mandar
ninguem investigar; e uma resposta de ausencia ligada com o periodo ja
passado esta ligada sem fazer nada, que e o que parece avariado a quem a pos.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
As listas do painel respondiam com um LIMIT fixo e nenhum filtro: 200 caixas,
50 trabalhos na fila. Numa instalacao maior nao havia forma de chegar ao
resto, e a fila so podia ser inspeccionada pela cabeca -- justamente quando
esta acumulada e que alguem precisa de olhar para ela.
- parsePageParams com testes proprios: pagina e tamanho tem de ser inteiros
positivos, o tamanho e limitado, e um valor invalido cai no seguro em vez de
chegar ao SQL como NaN, LIMIT negativo ou OFFSET fraccionario
- o termo de pesquisa vai como parametro ligado, nunca interpolado; ha um
teste que tenta terminar a instrucao e largar a tabela
- a contagem usa o mesmo filtro da pagina, senao o numero de paginas descreve
um conjunto de linhas diferente do que e mostrado
- PATCH /admin/mailboxes/:id edita papel, quota, idioma, estado e liberta um
bloqueio; so um super admin pode conceder super admin, ou um admin de
dominio promovia-se para fora do seu proprio dominio
Apelidos por utilizador: um apelido nao pertence a um utilizador, pertence ao
dominio e leva uma lista de destinos que pode incluir enderecos noutros
servidores, e um catch-all e do dominio inteiro. O que pode ser mostrado
dentro de um utilizador sao os apelidos que entregam na caixa dele, e e isso
que /admin/mailboxes/:id/aliases responde.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
/admin/security mostrava "Nao foi possivel carregar os dados" porque
/api/v1/admin/tls respondia 503. A rota so era ligada quando o certificado
vinha do watcher do acme.json do Traefik, o unico provider que tinha metodos
de estado proprios.
Nesta instalacao o TRAEFIK_ACME_DIR nao esta definido e o STATIC_CERT_PATH
tambem nao, portanto o servidor cai no certificado autoassinado efemero -- ou
seja, o painel calava-se exactamente no estado em que o operador mais precisa
de ser avisado.
- ProviderStatus deriva o estado de qualquer port.CertProvider, lendo a folha
do certificado que ele realmente serve
- o painel passa a ser ligado sempre que ha JWT_SECRET
- passa a dizer quem emitiu e se e autoassinado, com um aviso proprio: nas
portas de correio um autoassinado e recusado por qualquer cliente que
verifique, o que pesa mais do que uma validade ainda longe
CheckSignatureFrom nao serve para detectar isto: tambem exige que o emissor
seja uma CA, coisa que uma folha descartavel nunca declara, e por isso dava a
fallback degradada como correctamente emitida.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
O crocha na lista de dominios lia-se da coluna dns_status, e nada lhe escrevia
depois do onboarding. A verificacao corre no servico de protocolos, que tem o
resolvedor, e a resposta ia directa para o browser sem ficar registada -- por
isso um dominio com 13 de 13 registos publicados e a resolver continuava
listado como PENDING, ultima verificacao "Nunca", ao lado da pagina que dizia
o contrario.
Quem verificou passa a escrever o que encontrou. Falhar a gravar nao
transforma uma verificacao boa num erro para o operador: fica no log.
A coluna aceita apenas PENDING, VERIFIED, PARTIAL, DRIFT e ERROR, e o
verificador tambem diz MISSING e UNKNOWN. Escrever qualquer uma dessas violava
a check constraint, a actualizacao era rejeitada e o crocha ficava velo na
mesma -- exactamente o sintoma que isto vem corrigir. Ha um teste so para essa
traducao.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Duas coisas apanhadas ao correr o reconcile contra a zona real depois de a
rota passar a funcionar.
1) Nao era idempotente: 0 criados, 4 corrigidos, 11 iguais -- e outra vez 4
corrigidos na passagem seguinte, para sempre, com o temporizador a fazer o
mesmo em ciclo. Os 4 eram os SRV. A Cloudflare devolve o content de um SRV
SEM a prioridade ("1 993 host" para um registo guardado como "0 1 993
host") e guarda-a em data. O adaptador so recorria a data quando o content
vinha vazio, portanto lia sempre um valor diferente do da spec.
Para SRV e TLSA a data passa a ser a autoridade.
2) A spec de cada dominio incluia o A e o AAAA do mail host. Um segundo
dominio servido pelo mesmo servidor aponta o MX para um host que nao lhe
pertence, por isso reconciliar o cienciaembarcada.com.br tentava escrever
mail.lucasrguerra.dev.br -- zona alheia. A Cloudflare respondia "an
identical record already exists" e a consola mostrava dois erros para um
dominio que estava correcto. O mesmo valia para o TLSA, que tambem tem o
nome do mail host.
Os registos de endereco passam a ser publicados so pela zona que contem o
host; o MX e os SRV, que apontam para ele, continuam a ser de cada dominio.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
O painel respondia "DNS reconciliation needs a provider token to be
configured" enquanto o token estava configurado e a funcionar: a varredura de
fundo estava a publicar registos com ele nesse preciso momento.
SetDnsReconciler so atribuia quando r.dns ja existia, e o main.go chama-o
antes de SetAdminDnsAPI. Escrevia num struct que ainda nao existia, sem erro
nem aviso, e a API era depois construida sem reconciliador nenhum.
Os testes anteriores nao apanharam isto porque construiam o adminDnsAPI
directamente -- justamente a fiacao que estava partida. O teste novo passa
pelos setters do Router nas duas ordens possiveis.
- o reconciliador passa a ser guardado no Router, logo a ordem deixa de contar
- so e ligado quando ha token, para a rota poder dizer que esta indisponivel
em vez de falar com a Cloudflare com um token vazio e reportar uma recusa
- a mensagem deixa de culpar o token, que nao era a causa
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Na lista de dominios, o botao de cada linha dizia "Adicionar" para um dominio
que ja estava la, e "Status" depois de clicado. Nao faz nem uma coisa nem
outra: escolhe sobre qual dominio as accoes de baixo actuam. Passa a dizer
Selecionar / Selecionado.
Removida tambem a string reconcileNotImplemented, que afirmava que a
verificacao de registos individuais ainda nao existia. Existe desde que o
endpoint de verify passou a consultar resolvedores publicos, e ja nao estava
a ser usada em lado nenhum.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
O painel listava os registos em falta e oferecia um botao para os criar, mas
esse botao chegava a um endpoint que so escrevia uma linha de auditoria e
respondia "REQUESTED". Nunca tocou no DNS. Quem tinha o token da Cloudflare
era o servico de protocolos, e a unica coisa capaz de o usar corria num
temporizador -- para um dominio so, o de MAIL_DOMAIN.
Por isso um dominio adicionado pela consola, como o cienciaembarcada.com.br,
nunca era reconciliado: os registos nao apareciam e o unico sintoma visivel
era a consola dizer que faltavam, para sempre.
- POST /api/v1/admin/dns/reconcile publica os registos de um dominio, agora,
e devolve o que foi criado, corrigido e o que ficou em conflito
- o temporizador passa a percorrer todos os dominios ativos
- conflitos sao reportados, nao resolvidos: um registo do mesmo tipo e nome
com outro valor pode ser correio da zona ou um servico de que ninguem se
lembra, e nao e deste servidor decidir sobrepo-lo
- o erro do fornecedor fica no log, nao na resposta: pode nomear a conta ou
a zona
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Tres defeitos na comparacao, todos encontrados contra a zona real. Cada
um fazia a consola declarar em falta um registo que existia e estava a
funcionar - o que e pior do que nao dizer nada, porque manda o operador
procurar um problema que nao ha.
MX. A especificacao guarda a prioridade num campo proprio, portanto o
valor desejado e so o host. A comparacao era feita contra "10 mail.x."
como texto, e por isso o MX de um dominio que recebe correio todos os
dias aparecia como inexistente.
TXT longo. Um valor acima de 255 bytes e publicado em varias strings, e
o resolver devolve-as entre aspas. Tirar as aspas e juntar tudo com
espacos punha um espaco no meio de uma chave DKIM: todos os registos
DKIM RSA da zona eram dados como em falta. As strings sao agora unidas
ponta a ponta, e os espacos dentro de cada uma sobrevivem.
Registos atras do proxy. Um CNAME proxied nao e publicado como CNAME -
o proxy responde A. Perguntar pelo CNAME devolvia resposta vazia, e o
registo era dado como em falta apesar de estar la a servir trafego. A
pergunta passa a ser do tipo que o proxy responde de facto.
A comparacao passa a ser por tipo em vez de uma normalizacao unica para
tudo, que era a origem comum dos tres.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Mostrava tambem o tamanho de cada pasta. Sao numeros que nao mudam e
que ninguem esta a espera: enchiam a barra sem dizer nada de util.
Agora aparece um numero apenas onde ha correio por ler, e nada nas
restantes.
Consequencia a saber: Rascunhos nunca tem mensagens por ler - nada
marca um rascunho como lido - portanto passa a nao ter numero nenhum.
Quem quiser saber quantos rascunhos tem, ve-o no cabecalho da pasta.
O tamanho continua no tooltip do link, e o cabecalho da lista continua
a dize-lo por extenso, portanto a informacao nao se perdeu - deixou
apenas de estar sempre no campo de visao.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
O assunto que chegou a uma caixa real: "Fora do escritorio: Re:
=?UTF-8?Q?Re=3A_Fora_do_escrit=C3=B3rio?=". Duas coisas erradas, e a
segunda maior do que parece.
Um assunto com acento viaja como palavra codificada (RFC 2047). Colei o
cabecalho em bruto, portanto aquilo foi lido tal e qual por uma pessoa.
E foi tambem o que derrotou a verificacao do "Re:" ja existente:
codificado, todo o assunto comeca por "=?", nunca por "Re:", e por isso
foi acrescentado outro por cima. Descodificar corrige os dois sintomas
de uma vez. Os prefixos acumulados ("Re: RE: Re:") passam a ser
reduzidos a um.
O que estava por tras: aquele assunto era "Re: Fora do escritorio" -
alguem tinha respondido a propria mensagem automatica, e ela respondeu
de novo. Faltava a supressao por remetente. O RFC 5230 seccao 4.1 pede
exactamente isto e propoe sete dias: sem ela, uma conversa de cinco
mensagens produzia cinco copias do mesmo aviso, e quem respondesse ao
aviso recebia outro - que e como uma pessoa e um respondedor automatico
acabam a falar um para o outro indefinidamente.
A supressao falha aberta: um registo que nao se consegue ler deixa a
resposta sair. Ser avisado duas vezes e uma falha menor do que nunca
saber que a caixa esta ausente, que e o proposito da funcionalidade. O
envio so e registado depois de a resposta estar mesmo na fila, para que
um envio falhado nao silencie a mensagem seguinte da mesma pessoa.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Chegava como conversa nova. Ia sem qualquer cabecalho de encadeamento,
por isso todos os clientes de correio a mostravam solta, sem relacao
com a mensagem que a tinha desencadeado - e quem escreveu tinha de
descobrir sozinho a qual dos seus emails aquilo respondia.
In-Reply-To nomeia a mensagem que esta a ser respondida e References
leva a cadeia inteira, que e o que permite a um cliente colocar a
resposta no sitio certo mesmo quando nunca viu as mensagens pelo meio
(RFC 5322 seccao 3.6.4). A cadeia e limitada em comprimento: uma
conversa longa faria este cabecalho crescer sem fim.
O assunto passa a ecoar o original - "Fora do escritorio: Re: Proposta
comercial" - porque e o que a pessoa le na sua lista, e porque ha
clientes que agrupam pelo assunto e nao apenas pelos cabecalhos. O
texto configurado fica como prefixo em vez de ser substituido, para
continuar a dizer o que a mensagem e. Um assunto que ja anuncia ser
resposta nao recebe um segundo "Re:".
Sem Message-ID no original nao se escreve nada: apontar para uma
mensagem inexistente faria os clientes mostrarem a resposta agarrada a
nada. Um Message-ID com quebra de linha e recusado por inteiro - viria
de um remetente remoto, e uma quebra ali terminaria o cabecalho e
deixaria o resto ser lido como mais cabecalhos.
Os cabecalhos que impedem dois respondedores automaticos de se
responderem um ao outro continuam todos la, com teste que o garante.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Encontrado a validar na produção, depois do deploy: o selector de pasta
de destino na tela de regras mostrava apenas INBOX, Archive, Junk e
Trash - exactamente sem as pastas que o utilizador tinha criado, que
são aquelas para que uma regra de arquivo costuma ser escrita.
A tela de regras escolhe um destino sem nenhuma mensagem aberta, por
isso não tem "pasta actual" para excluir e passava uma string vazia.
Uma pasta criada pelo utilizador não tem papel special-use, portanto
"role === current" era verdadeiro para todas elas de uma vez.
A exclusão passa a só se aplicar quando há de facto uma pasta actual.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
O lang-lint exige que a fonte deste repositório seja ASCII, e os testes
que escrevi usavam acentos em texto de exemplo. Isto teria feito o CI
vermelho.
Onde o acento era só texto de exemplo, foi substituído. Nos dois testes
onde o acento é o que está a ser testado - que um nome de pasta acentuado
sobrevive ao parser, e que uma pasta pode chamar-se "Relatorios de
Marco" - fica escrito como escape Unicode, que mantém o teste a
verificar exactamente a mesma coisa sem pôr o caractere na fonte.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Encontrado a correr o script guardado na caixa real antes de o pôr no
ar. O Google envia os relatórios DMARC de
noreply-dmarc-support@google.com, e a verificação só reconhecia a
palavra exacta ou um sufixo com "+": a caixa teria respondido
automaticamente a todos os relatórios diários que recebe.
O prefixo passa a ser comparado com os separadores que os remetentes
usam de facto - "+", "-", "." e "_" - e a lista cresce com os nomes
correntes deste tipo de endereço. Continua a responder a uma pessoa
cujo endereço apenas começa por uma dessas palavras, o que também tem
teste: "bouncer" é gente, "bounces-123" não é.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Regras. A tela pedia um tipo de correspondência do Sieve pelo nome -
:contains, :is, :matches (regex) - e colava o valor escrito
diretamente num script. Punha uma linguagem de padrões à frente de
quem só queria dizer "o que vem do banco vai para Faturas", e bastava
uma aspa no valor para o script deixar de fazer parse, desativando
silenciosamente todas as regras de uma vez.
O vocabulário passa a ser o de quem escreve: um campo, uma condição em
português, um valor e uma ação. Os curingas são gerados a partir de
"começa com" e "termina com" - ninguém os escreve, e um asterisco
escrito por acaso continua a ser um asterisco. A pasta de destino é
escolhida de uma lista das pastas que existem mesmo, em vez de ser
escrita à mão: antes uma regra podia nomear uma pasta inexistente e
nunca disparar, sem dizer porquê.
A ordem passa a ser visível e alterável, porque a primeira regra que
arquiva uma mensagem decide para onde ela vai.
Um script que esta tela não consegue exprimir - escrito à mão, ou por
um cliente de secretária sobre ManageSieve - é mostrado tal como está,
em vez de ser reescrito na coisa mais parecida que este formulário sabe
dizer, o que mudaria em silêncio o que o correio de alguém faz.
Pastas próprias. Existiam na base de dados e as regras podiam arquivar
nelas, mas a barra lateral listava só as seis fixas, portanto não havia
como ver uma - nem como criar. Criar fica onde a vontade aparece, na
própria barra; apagar avisa que leva as mensagens junto, porque leva.
Encontrado a escrever isto: o motor tratava "\\*" como barra seguida de
curinga em vez de um asterisco literal. Como é esta tela que gera os
curingas, uma regra sobre um assunto com "50% *" teria correspondido a
todos os assuntos começados por "50% ". Corrigido no motor, com testes
dos dois lados.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A metade que faltava. O motor já existia e nada o chamava; agora o
caminho de entrega consulta as regras da caixa que está a receber, e a
resposta de ausência é mesmo enviada.
Cada destinatário tem as suas próprias regras, avaliadas sobre a sua
própria cópia: duas caixas neste servidor que recebem a mesma mensagem
podem arquivá-la em sítios diferentes.
Nada disto pode impedir o correio de chegar. Um script que não faz
parse, ou uma base de dados que não responde, deixa a mensagem ser
entregue sem filtro em vez de a recusar - recusar diria ao remetente
que a mensagem falhou, e o servidor dele voltaria a tentar para sempre.
discard é aceitar e depois deitar fora, pela mesma razão: uma regra de
arquivo não é uma rejeição.
A resposta de ausência sai pela mesma fila, assinatura e limites de
envio que qualquer outra mensagem deste servidor - não é um caminho
especial - e só depois de a mensagem que a desencadeou estar guardada,
para não responder a correio que a caixa afinal não recebeu. Leva
Auto-Submitted: auto-replied, que é o que impede duas caixas ambas de
férias de se responderem uma à outra até uma esgotar a quota.
Pastas próprias. Só existiam as seis padrão, portanto uma regra que
dizia "arquiva isto em Faturas" nomeava uma pasta que não havia como
criar. Criar, renomear e apagar, com as reservadas protegidas: cada uma
delas significa alguma coisa para o protocolo ou para este servidor, e
deixar renomear a INBOX partia caminhos que nenhuma mensagem de erro
explica. Renomear não mexe nas mensagens nem nos UIDs, senão um cliente
IMAP perdia de vista tudo o que lá está dentro.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Primeira metade do que falta para a ausência e os filtros funcionarem:
o motor em si, como domínio puro e testável, ainda sem estar ligado à
entrega.
É um subconjunto deliberado do RFC 5228 mais o vacation do RFC 5230:
os testes e as ações que a interface gera e que um script escrito à
mão provavelmente usa - header com :is, :contains e :matches, anyof,
allof, not, if/elsif/else, fileinto, discard, addflag, keep, stop e
vacation. O que não entende é recusado em vez de ignorado: uma regra
que o utilizador escreveu e que silenciosamente não faz nada é pior do
que uma que diz que não consegue correr.
:matches é correspondência com curingas, não expressão regular. Em
Sieve todos os outros caracteres são literais, portanto "a.b@x.test"
tem de casar apenas com um ponto verdadeiro; compilá-lo como regex
faria casar também "axb@x.test" e entregaria a um autor de scripts toda
a gramática das expressões regulares, incluindo padrões que levam tempo
exponencial numa linha de assunto construída para isso. O algoritmo faz
backtracking iterativo, linear no caso comum, e há um teste com limite
de tempo que falha se voltar a ser exponencial.
A parte da ausência que realmente importa é quando não responder. O
RFC 3834 é a regra: responde-se à mensagem de uma pessoa e a mais nada.
Remetente de envelope vazio, auto-submitted, precedence bulk, qualquer
cabeçalho de lista, x-auto-response-suppress, endereços reservados como
mailer-daemon, e a própria caixa - nenhum recebe resposta. É assim que
uma caixa evita responder a uma lista à frente de toda a gente, ou
entrar num ciclo com outro respondedor automático.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Cc/Cco. Os dois rótulos do compositor eram o último texto em inglês
que restava, e "Bcc" tem nome próprio em português - quem não conhece
a abreviatura inglesa não adivinha o que o campo faz.
Larguras. A barra lateral tinha 244px fixos e nela não cabia "Console
de administração", que transbordava no telemóvel e no computador: o
botão passa a mostrar um rótulo curto e guarda o nome completo no
tooltip. O endereço da conta usava break-words e partia-se a meio do
domínio, deixando um "br" sozinho na linha de baixo; passa a truncar
com o valor inteiro no tooltip.
A lista tinha 404px fixos e o leitor 760px centrados, portanto num ecrã
largo os assuntos eram espremidos numa coluna estreita enquanto o
leitor ficava entre duas margens enormes. Ambos crescem agora com a
janela. O assunto ganha overflow-wrap: os Message-ID dos relatórios do
Google são uma única "palavra" sem espaços, mais longa que a coluna, e
sem sítio onde quebrar saíam da linha.
Imagens remotas. A escolha era reposta sempre que uma mensagem era
aberta, por isso a mesma newsletter tinha de ser desbloqueada em cada
leitura. A decisão passa a ser lembrada por remetente, com um segundo
botão para a retirar. Bloqueado continua a ser o que acontece por
omissão: uma imagem remota é um aviso de leitura que o remetente não
pediu licença para receber.
Assinatura. Era escrita num textarea - texto simples com quebras de
linha reais - e injetada no compositor com innerHTML, onde uma quebra
de linha é espaço em branco: todas desapareciam. Passa a ser um campo
rico, com negrito, itálico, links e imagens, guardado como HTML e
sanitizado nos dois sentidos, já que este valor é injetado como
marcação na origem do webmail. Assinaturas guardadas pelo campo antigo
continuam a funcionar: as suas quebras de linha são convertidas.
Mover entre pastas. Faltava a única ação corrente que o leitor não
tinha: arquivar uma mensagem para fora da caixa de entrada era
impossível. Enviados e Rascunhos nunca são destino - descrevem como a
mensagem veio a existir, não onde o leitor a arrumou - mas sair deles é
permitido. Pastas criadas pelo utilizador, que não têm papel
special-use, são destino como qualquer outra.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Um deploy recriava os contentores mas continuava a correr o código
antigo. "latest" é uma tag móvel, e o Docker considera satisfeita
qualquer tag que já tenha localmente: o compose pedia
protocols:latest, o host já tinha uma imagem com esse nome, e a que o
Release acabara de publicar nunca era descarregada. O deploy terminava
com sucesso, os contentores reiniciavam saudáveis, e nada do que
estava a correr mudava.
Foi assim que várias correções que passaram no CI nunca chegaram ao
site: a rota /api/v1/mail/delete respondia 404 em produção enquanto
existia no código e nos testes, e a pasta Reports existia na base de
dados - a migração corre a partir da imagem certa - sem existir em
nenhuma das interfaces.
pull_policy: always nos seis serviços que consomem imagens
publicadas. Os que usam imagens de terceiros com versão fixa
(postgres:16-alpine, redis:7-alpine, clamav) ficam como estão: aí a
tag identifica um conteúdo imutável e voltar a puxar não acrescenta
nada.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Todos estes foram reproduzidos na interface real, com a conta em
mail.lucasrguerra.dev.br, e não em teste.
Contadores das pastas. O badge caía para o total sempre que não havia
nada por ler, no mesmo lugar e no mesmo estilo do contador de não
lidas: uma caixa de entrada com catorze mensagens lidas mostrava "14",
e ler mais uma nunca movia o número - ele nunca esteve contando o que
o leitor pensava. Os dois números agora são visualmente distintos e o
de não lidas só aparece quando existe mensagem por ler.
Em Rascunhos o badge e o total são o mesmo número por definição, então
o sidebar imprimia-o duas vezes: "Rascunhos 1 1". A regra passou para
folderBadge, onde é testada, e o tamanho da pasta é renderizado num
único sítio.
Ações do leitor. Todas as mensagens ofereciam os mesmos quatro botões,
onde a pasta devia decidir. Um e-mail que o próprio utilizador enviou
oferecia "marcar como não lida" - uma flag que nada lê numa pasta onde
nada é entregue - e um rascunho por acabar oferecia Responder e
Encaminhar, mas nenhuma forma de o terminar. Abrir um rascunho agora
abre o compositor com o que estava escrito e adota o seu UID, para que
continuar a escrever substitua esse rascunho em vez de deixar o antigo
para trás. Em Lixeira o botão de apagar diz que apaga em definitivo,
porque é o que faz.
Anexos. t("mail.attachments") era chamado sem o parâmetro que a string
declara, e a interface mostrava literalmente "Anexos ({count})" nos
três idiomas. Na lista, que sabe que há anexos mas não quantos, passa a
nomear o facto em vez de imprimir uma contagem que não tem.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- Implemented tests for ingesting TLS RPT reports from emails and ensuring they are stored correctly in the database.
- Added tests to verify that the same report can be ingested multiple times without duplication.
- Created tests for backfilling reports that were previously delivered to the inbox.
- Developed WebSocket event handling tests to ensure events are correctly routed to subscribers based on mailbox ownership.
- Included tests for the mail API to validate session handling, mailbox isolation, and error mapping.
- Ensured that the API rejects unauthorized requests and handles malformed JSON gracefully.
- Added `MoveToTrash` method in `WebmailRepository` to move messages to the Trash folder instead of deleting them permanently.
- Introduced `handleDelete` method in `mailAPI` to handle message deletion requests from the client.
- Updated router to include a new route for message deletion.
- Enhanced the compose page to manage draft integrity, ensuring drafts are correctly associated with sent messages.
- Implemented UI changes to allow users to delete messages from the mail folder view.
- Added tests to verify the correct behavior of message deletion and draft handling.
- Updated database migration scripts to reconcile folder counters and ensure accurate unread counts.
- Added tests for TLS-RPT report parsing and webmail repository functionalities.
The previous commit put four literal non-ASCII characters into source, which
lang-lint rejects (PLAN.md 21.1: all code in English, ASCII-only, enforced by
CI rather than review). Nine lines failed.
The brand mark in the four rails and sign-in cards was a literal lambda, and
the domains screen's meta line joined DMARC, MTA-STS and DANE with a literal
middot. Both are now \u escapes, so the source is ASCII and the rendered page
is unchanged - the served sign-in page emits ce bb for the mark, and the
domains chunk carries the middot as \xb7.
globals.css goes with them. The linter does not scan .css today, so this was
passing, but the stylesheet was the only file in the tree using box-drawing
section rules and em dashes, and there is no reason for it to be the
exception - ASCII hyphens read the same in a comment.
Verified: lang-lint clean, 98 tests green, tsc clean, production build clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The webmail was indigo over glass and the console was green over glass, so the
two halves of one product did not look related, and neither followed a system
anything could be checked against. This replaces the interface with the
Nocturne design system from Claude Design: a near-neutral blue-grey ground,
Inter at weight 500, solid surfaces with a 1px inset edge instead of blur, and
a single blurple accent used as a line, a mark and a glow rather than a flood.
The palette is remapped at the source rather than renamed at every call site.
tailwind.config.js points slate, indigo, emerald, teal and the rest at the
Nocturne ramps, and globals.css carries the same values as variables, so a
utility class and a token cannot drift apart and no colour can enter the
interface from outside the system. The console's green is gone because it
resolves to the one accent; that was the point.
Structurally, both surfaces now share one rail - the same brand mark, the same
3px accent mark on the active item, the same account tile - with a segmented
control for crossing between them. Crossing into the console still goes to the
step-up rather than the admin sign-in, and leaving it is still a plain link.
Settings moves from eight stacked panels to a tab strip, so enabling a second
factor and writing a filter are no longer separated by everything else.
Nothing is sized by the length of its text any more. Labels wrap, tab strips
wrap instead of scrolling, subjects and queue diagnostics wrap instead of
truncating, and the message list's snippet is clamped to two lines rather than
one. That is what broke when the interface was read in Portuguese or Spanish.
Defects found while doing it:
- The queue and DMARC screens styled their status pills with "badge-danger",
"badge-warning" and "badge-verified" - three class names no stylesheet in
this project defines - so every verdict rendered as identical unstyled
text. They are Badge variants now, which exist.
- The DMARC screen printed the literal string "quarantine" as the
enforcement level whatever the domains publish. The report endpoint returns
no policy at all, so it was asserting a configuration nobody had made. It
shows the DMARC pass ratio it genuinely has.
- Half the mailboxes screen and two lines of the audit trace were hardcoded
strings, some Portuguese and some English, so they stayed in one language
whatever the interface was set to. They come from the bundle now, with the
new keys added to all three locales.
- Avatars printed email[0], so every account at one domain collapsed onto the
same letter. initialsFor takes two letters from the local part's own word
boundaries, behind tests.
Trivy reported one HIGH: nanoid 3.3.16, pulled in transitively by postcss,
vulnerable to a denial of service via an infinite loop in ID generation
(CVE-2026-67213). Pinned to ^3.3.18 through the overrides block the package
already uses for sharp and postcss.
Verified: 98 webmail tests green, tsc clean, production build clean, and the
compiled stylesheet serves the Nocturne values with no stock Tailwind indigo,
emerald or slate hex left in it. Trivy fs with the CI scanners, severities and
skip-dirs exits 0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The counts were computed inline in JSX where nothing could test them. The
sidebar matched a folder by a special-use role folders do not always carry, its
badges were fetched once and then froze - new mail did not raise them and
reading a message did not lower it - and the list header printed the size of
the loaded page, capped at 50, as the size of the folder. The rules now live in
mailCounts.ts behind tests, useFolders keeps them current from the server's
push and from this tab's own changes, and each filter chip carries its number.
Inline images never left the server: RenderMessage dropped every part a body
references as cid:, so mail written by any normal client rendered with broken
icons. They now come back as bounded data URIs, refusing image/svg+xml, and the
reader resolves them, sizes its frame from the message instead of a fixed
500px, and renders on the light surface mail assumes.
Reading mail no longer asks for a second factor - the mailbox is behind the
account's own password - while the console still requires one unconditionally,
now collected by a step-up that does not ask again for the password the session
in hand just proved. An admin sign-in issues the webmail session too, so
leaving the console is a link rather than another login.
Two defects found while testing the above:
- The admin sign-in checked the password and the second factor and never the
role, so any account that had enrolled a factor from its own settings could
obtain an admin-audience session.
- Session tokens carried nothing that varied below one second, so signing in
twice within the same second minted the identical token and collided with
the unique index on web_sessions.refresh_token_hash. A correct password
answered 500, and a double-clicked sign-in was enough to do it. Tokens now
carry a jti.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The endpoint added in #32 lives on the protocols service, but the proxy only
forwarded admin/dkim and admin/tls there - so admin/dns went to the auth
service, which has no such route and answers 401 from its admin guard. The
console would have reported "not authorised" for a request that was simply
sent to the wrong process.
It survived my own check because an unauthenticated probe of a nonexistent
admin path returns 401 too: the guard runs before routing, so 401 proves
nothing about whether a route exists. Comparing against a path I knew was fake
is what exposed it.
The list is now named and commented, and a test asserts every protocols-owned
area appears in it, because the failure mode reads as a session problem rather
than a routing one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three things the console and the mail client were getting wrong.
**Sent stayed empty and drafts would not save.** Both write through the same
persistence path, and both failed on the same constraint: spf_result,
dkim_result and dmarc_result are restricted to the RFC verdict vocabulary and
are nullable, so "not evaluated" has to be NULL. The compose path left them at
Go's zero value, which sends an empty string - satisfying neither the CHECK nor
the meaning - and the insert was rejected. Filing the Sent copy is deliberately
non-fatal, so it failed silently; saving a draft reported the error. Inbound
mail was unaffected because it always supplies a real verdict, which is also
why the existing test never caught it.
**DNS reconciliation could not verify anything.** The console posted to the
auth service, which has no resolver, so the endpoint recorded the request and
re-read the status already in the database. The machinery to do it properly
already existed - the background reconciler resolves every expected record
through public resolvers - it just was not reachable from the console.
Adds GET /api/v1/admin/dns/verify to the protocols service, which owns both the
resolver and the record spec, and wires the domains screen to it: every
expected record, its expected value, and whether it actually resolves. A
proxied record is reported as present rather than matching, because behind a
proxy the published answer is the provider's address and comparing values would
always disagree. Reconciliation still runs afterwards so the stored status and
the audit trail stay in step.
The spec is read, never provisioned: opening a screen must not create DKIM keys
as a side effect.
Verified: the persistence test fails without the fix with the exact constraint
violation (email_messages_dkim_result_check), 22 Go packages and 36 webmail
tests green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Fixing the recovery-code OOM removed one way to exhaust the container's
memory; it did not remove the reason that was possible. Argon2id is configured
for 64 MiB, and every password hash or verify reserves that for its duration.
Nothing limited how many ran at once, and verification is reached by an
unauthenticated login - so roughly twenty simultaneous sign-in attempts
reserve 1.28 GiB against a 512 MiB container and the process is killed. No
credentials needed, and account lockout does not help: it is per account, and
the requests need not share one.
Hashing now passes through a four-slot gate. Callers queue rather than being
refused, so every request stays correct and only gets slower under load, and
peak memory is four hashes regardless of traffic - about 256 MiB, comfortably
inside the limit.
This is the answer to "why not just remove the memory limits". The limit was
never the fault; unbounded per-request memory was. Removing it would move the
kill decision to the kernel, which picks by size and would take Postgres -
losing queued mail rather than one request.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Confirming a second factor returned "Upstream service is unreachable" - the
Next proxy's message for a backend it cannot reach. The backend was not
unreachable, it was dead: the process had just been OOM-killed.
generateRecoveryCodes hashed all ten codes with Promise.all. Argon2id here is
configured for 64 MiB per hash, so ten concurrently reserve 640 MiB against a
container limited to 512 MiB. Hashing is now sequential, peaking at 64 MiB and
taking a few seconds - the right trade for something that happens once per
enrolment. A timing-based test covers it, against the real hash rather than a
mock, because the cost being guarded is memory held concurrently and only the
real implementation reserves it.
The worse half is what the crash left behind. The enrolment row was marked
CONFIRMED before the codes were generated, so the account came out of this with
a second factor enabled, no recovery codes, and nobody having seen any: locked
behind an authenticator entry with no way back. That is how an account ends up
"assuming" it has a factor its owner cannot use.
The order is now verify, store the codes, then confirm. Anything that fails in
between leaves the enrolment pending and retryable, which is the recoverable
end of that failure.
Both bugs sat behind the one fixed in #28 - the code check failed first, so the
crash was never reached.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
An enrolment can end up confirmed against a secret the authenticator no longer
holds - the account is then locked behind codes nobody can produce, and the
only documented recovery was a DELETE against mfa_totp. That SQL was also
incomplete: it left the recovery codes valid and the sessions live.
reset-mfa removes the authenticator secrets and the recovery codes, and signs
the account's sessions out, because a live session would keep the console open
without the factor that was just removed. The password is untouched, so the
command alone grants nobody access.
The alternative people reach for - delete the mailbox and create it again -
takes its mail, folders and aliases with it, and none of those are the problem.
The docs now say so.
Also corrects the advice printed after create-admin, which still told the
operator to enrol from the webmail settings before the console would open. The
console has enrolled inline since #23.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Enrolling a second factor could reach a state where the authenticator showed
perfectly valid codes and every one of them was rejected, with the only
feedback being "TOTP verification failed".
The browser holds the enrolment secret in component state alone. Reloading the
settings page - or leaving and coming back - brought the "enable" button back,
and pressing it called the enrol endpoint again, which deleted the pending row
and stored a brand new secret. The entry already saved in the authenticator
then belonged to a secret the server no longer had. Nothing about that is
visible from either side: the app generates codes correctly, the server
compares them correctly, and they are for different keys.
Enrolment now resumes. The endpoint returns the pending secret if there is
one, so the QR shown after a reload is the same one already scanned and
rescanning is harmless. Starting over is still possible - ?reset=1, exposed in
the settings screen - because someone who deleted the entry from their app
needs a secret they can actually generate codes for. A pending row that can no
longer be decrypted is replaced rather than resumed, since nobody could ever
confirm it.
Also: the submitted code now has every non-digit stripped rather than just
surrounding whitespace. Authenticators display the code grouped as "123 456",
and a value typed or pasted with that space failed the six-character length
check before it was ever compared - which, to someone reading the right number
off their phone, looks like the server being wrong. The inputs sanitise as you
type for the same reason.
Ruled out first, rather than guessed at: the algorithm passes all six RFC 6238
reference vectors, base32 round-trips, and the deployment's clock is exactly in
sync with real time, so neither the implementation nor drift explains it.
Verified: the resume test fails without the change and passes with it; 47 auth
tests green against a real Postgres.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A fresh deployment published "v=DKIM1; k=rsa; p=" for its own domain. An empty
p= is not an absent key, it is a revoked one (RFC 6376 section 3.6.1): every
verifier reading it treats all signatures from that selector as invalid. On a
domain whose DMARC policy is p=quarantine with strict alignment - which this
reconciler generates - that quarantines all of its own outbound mail. Worse
than publishing nothing.
The chain: DkimRepository.Insert is an INSERT ... SELECT conditional on the
domain existing in the domains table, so on an empty database it inserts no
rows and returns no error. FindPublicKey then finds nothing, ensure() returns
an empty string with no error, and the reconciler faithfully published the
empty value. Every field along that path was behaving as written; nothing
checked that the result meant anything.
Two changes:
- The record builder omits a DKIM record when its key is empty, so the
revocation cannot be published regardless of why the key is missing.
- Provisioning logs when it produced no key, naming the likely cause. The
insert being conditional on an onboarded domain is exactly the case a new
operator hits, and "no key, domain probably not onboarded" is the difference
between a diagnosis and unsigned mail with no explanation. It stays
non-fatal: erroring here would abort the whole reconcile and leave a new
install with no A, MX or SPF records either.
Found while migrating to a new host: the fresh database had no domain row, so
the first reconcile revoked the domain's DKIM. The two published records have
been deleted from the zone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
SpfInclude derived the mechanism by reducing the relay hostname to its
organisational domain. That is unsound, and Brevo is the case that proves it:
the relay is smtp-relay.brevo.com, so the reduction yields
"include:brevo.com" - but brevo.com publishes the SPF of Brevo's own
corporate mail (Google Workspace and Zendesk). The sending ranges live at
spf.brevo.com.
Publishing the wrong include is worse than publishing none. It authorises
Google to send for the domain while leaving the relay unauthorised, so every
message the relay forwards fails SPF - and against the "-all" policy this
reconciler generates, is rejected rather than merely marked.
RELAY_SPF_INCLUDE now carries whatever the provider documents. A bare domain
gets the "include:" prefix; any other SPF term (ip4:, a:, mx:) passes through
untouched, since not every provider publishes an include. The old reduction
stays as a fallback for relays whose hostname does happen to match, because
publishing something plausible beats publishing nothing at all.
Caught before configuring a relay for real, by checking what brevo.com and
spf.brevo.com actually serve rather than trusting the heuristic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every message was resolved through MX and delivered over SMTP, including one
addressed to a mailbox on this very host. Two users on the same server could
not write to each other without a working outbound port 25 - which the
provider blocks, so on this deployment they could not write to each other at
all. It also made internal mail take a pointless round trip through the public
internet.
The worker now checks whether the recipient resolves to a local mailbox and,
if so, hands the message to the inbound path directly. The check runs ahead of
the relay branch on purpose: sending local mail out to a smarthost only to
have it delivered back is slower and has more ways to fail.
The shortcut is narrow. It relies on ResolveDeliveryTargets, which only
resolves addresses on an active local domain, and a test covers that a remote
recipient still goes out over SMTP - otherwise local delivery would swallow
mail meant for the internet. A resolver error is treated as retry rather than
as not-local: falling through to SMTP on an unknown answer would send to the
internet a message that may belong here. Local delivery is still
content-scanned; a virus between two accounts on one host is no less a virus.
This also surfaced a permissive test double: fakeAliasFanOutRepository claimed
every address was local, which was harmless while nothing consulted it while
routing and wrong the moment something did. It now hosts one address, as the
real repository does, and the DSN fan-out test names it explicitly.
Verified: the local-delivery test fails without the change (the recipient
triggers an MX lookup) and passes with it; 22 Go packages green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The autosave I had just added appended a new message on every debounce, so a
few minutes of typing would have left the Drafts folder holding a dozen copies
of the same unfinished mail. Caught before deploying.
The composer now returns the stored UID and sends it back as replace_uid on
the next save; the use case expunges the superseded draft, but only after the
replacement has committed, so a failure leaves the older draft in place rather
than losing both. Expunge is a soft delete matching the expunged_at column the
schema already uses, and it locates the message by mailbox as well as UID so a
caller cannot delete someone else's mail by guessing a number.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four faults, all in the path between the compose screen and the mailbox.
**Every message sent from the webmail had an empty body.** The HTTP handler
read a field named "body"; the composer has always sent "html". The two never
matched, so the payload handed to the queue contained headers and nothing
else. The handler now reads html and text, and still accepts body so an older
client is not broken. Sending was otherwise working, which is why this went
unnoticed: the mail left, it just said nothing.
**Bcc was discarded.** The field was absent from the request struct and from
ComposeInput, so a blind copy typed into the webmail reached nobody. It now
joins the envelope recipients and is deliberately kept out of the headers,
which is what makes it blind.
**HTML was labelled text/plain.** Even once the body was carried, the composer
produces markup and the message declared text/plain, so a recipient would see
raw tags. Messages with HTML now go out as multipart/alternative with a
plain-text part derived from the markup and ordered first per RFC 2046 5.1.4.
An empty or missing text alternative is itself a spam signal.
**Sent mail left no trace.** Nothing wrote to the Sent folder, so a delivered
message vanished from the sender's view entirely. A copy is now filed after
the message is durably queued - not before, or a send refused by the rate
limit would still leave a copy - reusing the inbound persistence path so it
gets a real IMAP UID and appears over IMAP too. A failure to file is logged
and swallowed: the mail is already on its way, and reporting failure would
invite the user to send it twice.
**Drafts were never saved.** There was no endpoint at all. The composer showed
"draft saved automatically" from a setTimeout with no request behind it, so
closing the tab lost the message the screen had just promised was safe. Adds
POST /api/v1/mail/draft, storing the draft as a whole RFC 5322 message in the
Drafts folder, and debounces the composer onto it.
The composer also navigated to Sent whether or not the send succeeded, which
is how a rejected message disappears silently. It now reports the error and
stays put.
**Delay notifications were being filed as spam.** A DSN this server writes is
injected through the same inbound path as mail from the internet, but without
a client IP - so SPF can only be "none" - and its bounce-shaped content scores
highly. The warning telling you your mail was delayed landed in your own Junk
folder. Messages the server generates itself now skip content scanning; the
exemption is narrow, and a test covers that ordinary mail is still scanned so
this cannot become a way past the filter. The DSN also declared
Auto-Submitted: auto-replied, which belongs to vacation responders; a delivery
report is auto-generated (RFC 3834 section 5), and filters read that header.
Not fixed here, because it is not a code fault: outbound port 25 is blocked by
the provider, which is what the delay notification was reporting
("i/o timeout" to Google's MX - a timeout, not a refusal, is the signature of
a block). Receiving is unaffected and queued mail keeps retrying. Documented
with the two ways out, including the relay path, which is verified working
code rather than another stub.
Verified: 22 Go packages pass, 5 new compose tests and 3 for the DSN handling.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The site returned 504 on every request while the web container was healthy and
Next was serving. The container sits on three networks - backend, proxy, and
the per-app network Coolify adds - and the label meant to tell Traefik which
one to use reached it as the literal text '${COOLIFY_PROXY_NETWORK:-coolify}'.
Coolify escapes $ to $$ when it rewrites compose labels, so the variable was
never interpolated, the label matched no network, and Traefik fell back to
picking one itself. Choosing 'backend' - internal: true, no route to the proxy
- produces exactly this: a gateway timeout in front of a working app.
The label has been wrong since it was added; container recreation during the
last deploy changed which network Traefik happened to pick and turned a latent
fault into an outage.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A pinned IMAGE_TAG solves 'latest never re-pulls' and introduces the mirror
problem: the pin is an instruction to run that exact image, so deploying
without bumping it re-pulls the old build and reports success. Redeploying
with 'ignore cache' does not help - that clears the build cache, and nothing
is built on the server.
Documents the two-step release and the asset-fingerprint check that tells you
which image is actually serving, rather than trusting a green deployment log.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The deployment guide still described enrolling a second factor from webmail
settings before the console would open. That detour is gone: the console
enrols one inline, and the console itself is now reached from a link inside
the app rather than from a chooser at the root.
Also records what the console does not do, since it previously implied
otherwise on screen: DNS records are not verified per record, attachments are
not uploaded, and the vacation responder, signature and Sieve rules are stored
but not yet consumed during delivery.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The interface was reporting things that were not true. Several screens were
mockups wired to nothing, and a reader with the console in Portuguese was
served a mixture of two languages.
Fabricated data, removed:
- The outbound queue was two invented jobs held in component state. Retry and
Freeze only edited that local array, so the screen showed the same two rows
on every deployment and reported success for actions that never happened.
"Freeze" named an operation the backend has never implemented. It now reads
/api/v1/admin/queue, which was already serving the real thing, and offers
retry and cancel, which exist.
- Every dashboard KPI fell back to an invented constant (1420 received, 890
sent, 3 queued, 18.4% disk), so an unreachable API produced a dashboard of
plausible traffic for a server that might have delivered nothing. Two of
those fields were never returned by the endpoint at all, so the fake disk
figure was what everyone always saw. DMARC did the same with 1250/98%/97%.
- The domains screen showed thirteen DNS records for example.com, all marked
VERIFIED, and posted {domain_id: "default-domain"} to endpoints expecting a
real id. DKIM rotation sends "domain", not "domain_id", so its 400 was
swallowed and a made-up TXT record displayed as though a key had been
issued: rotation had never once worked from this screen.
- The TLS panel asserted a Let's Encrypt RSA 4096 certificate and verified
DANE from static markup, on a deployment currently serving a self-signed
one. It now reads the cert watcher via /api/v1/admin/tls.
- The readiness scorecard printed "10 / 10 all checks passed" regardless of
what preflight reported, including failures.
- A message trace that found nothing invented a five-step delivery - TLS 1.3
accepted, SPF pass, spam score 0.8 - for a message that does not exist.
- Both sidebars showed a fixed admin@lambdamail.local with a hardcoded
"SUPER_ADMIN (2FA)", claiming a second factor whether or not one existed.
- Settings seeded a "[Boletim]" Sieve rule. Saving posts the whole list, so
the first edit wrote that invented filter into the account's real script.
The reconcile endpoint itself answered {records_verified: 13, status:
"VERIFIED"} for any domain id, including one that does not exist, without
resolving a name. It now reports what it does, which is record the request.
Surfaces:
The root was a chooser listing webmail and the console side by side, which
asked visitors a question they should not have to answer and advertised cookie
names and audiences to anonymous readers. It now redirects into the mail
client. Accounts that may administer the server get a link to the console from
inside the app - a link to /admin/login, so crossing over still costs the
second factor the admin audience requires - and the console gets a plain link
back, since the webmail session was never given up to get there.
Enrolment:
Both enrolment screens printed the base32 secret as text and left the operator
to type 32 characters into their phone. The enrol endpoint already returned an
otpauth:// URI; it simply was not drawn. Added a shared component that renders
it as a QR locally - the seed is already in the page, and sending it to an
image service would put it in another request log for nothing.
The recovery-code step rendered its own "sign in" button while the form's
submit button stayed visible below it: two identical buttons, and no way to
copy the codes it insisted you save. It is now terminal, with copy and
continue.
Language:
translate() had no interpolation, so "Undo Send ({seconds}s remaining)" showed
the placeholder and a second hardcoded countdown was appended after it. It now
substitutes, leaving an unmatched token visible rather than blanking it.
The pt-BR bundle had lost its diacritics - "Destinatario", "Ultimo erro",
"Acoes" - from the ASCII-source rule being applied to the translations, which
are data. Restored, and the three bundles are back in key parity at 270 each,
with roughly 120 strings that were hardcoded in the source now keyed.
Healthchecks:
web had none, despite being the only service users reach directly: a wedged
Next process kept its container running and Traefik kept routing to it. Added,
using node's own fetch since the image ships no curl. rspamd and clamav gained
the timeouts and start_periods whose absence would fail them during startup.
migrate stays without one and says why: it is a one-shot job, and a probe
would report unhealthy exactly when it has succeeded.
Verified: 36 webmail tests and 41 auth tests against a real Postgres, both new
admin/me cases confirmed to fail without the route (404). lang-lint clean,
production build clean, 0 vulnerabilities, compose validates.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The first deploy of the new interface reported success and served the
previous build: docker compose does not re-pull a tag it already has
locally, so a deployment sitting on "latest" keeps running whatever it
pulled the first time. Nothing in the output says so - the CSS was simply
the old one.
Pinning IMAGE_TAG to the sha- tag is what forces the pull, and it is what
makes a rollback possible at all. The deployment guide now says so, with
the reason.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The console demanded a second factor and refused to open without one, but
enrolment lived on the other surface - so an operator whose only account
was the administrator had no way in. The password was already verified at
that point, so the refusal now carries a short-lived grant that permits
enrolling a factor and nothing else, and the admin login screen walks
through it: secret, confirmation, recovery codes, sign in. Two tests cover
it, including that the grant opens no other endpoint.
The rewrite also undid work that was already in place:
- The postcss override was dropped and the advisory came back, along
with a critical one in vitest. postcss is a direct dependency now, so
the fix is the version plus an override scoped to Next's own copy;
vitest moves to 3.2.7. The lockfile needed regenerating for either to
take effect. Zero advisories now.
- 100 language-lint violations: the admin console had been rewritten
with its strings in Portuguese, hardcoded. That fails CI, and it also
puts those strings beyond the reach of the translation bundles - an
English or Spanish user would have read Portuguese. They are English
in the source again.
- The mail screen put a fixed 320px list beside the reader with no
breakpoint, so on a phone the viewport overflowed and the message was
a few pixels wide. One pane at a time below md, with a way back.
- The icon-only navigation toggles had nothing for a screen reader to
announce.
The layouts' own mobile navigation and every previously fixed behaviour
survived the rewrite - checked rather than assumed.
Also fixes a marker collision I introduced: the E2E helper locates the
repository by a migrations directory, and services/migrations satisfied
that one directory too early. It now requires the SQL to actually be
there.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Trivy failed the two images added in the previous commit. The Dockerfile
rule was straightforward - neither declared a user - and both now run
unprivileged, with the migrate base moved from v4.17.1 to v4.19.1 because
the older one sits on Alpine 3.19 with six HIGH advisories.
The rest is not fixable here. What remains in those two images is the
contents of golang-migrate and Rspamd themselves: 37 and 48 fixable
HIGH/CRITICAL findings that belong to those projects. Our layers add .sql
and .conf files and nothing else, so gating this pipeline on them would
gate it on somebody else's release cadence, and a blanket ignore rule
would quietly weaken the scan for the images we do control.
The image scan therefore covers the four images whose contents this
project authors, with the reason stated at the step and the upstream pair
recorded in OPERATIONS.md as something to watch. Both are still built in
CI, so a broken Dockerfile still fails.
Worth being clear: these two images were already running in the stack
before this change. Putting them in the matrix made their state visible;
it did not create it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The deployment failed on the migrate service. Coolify rewrites a relative
bind mount to a path under its own application directory, and when that
resolves to an empty directory golang-migrate finds no migrations and
exits 1 - with nothing useful in the deployment log to say why.
A deployment that pulls its images has no business also depending on
files being at a particular path on the host. Both mounts are now baked
into images built from the repository root:
migrations - the schema, so migrate always has it
rspamd - actions.conf and ratelimit.conf
The Rspamd one mattered for a quieter reason: Rspamd starts happily with
an empty local.d, so a mount that resolved to nothing would have left the
rate limits and score thresholds unapplied. The defence would have looked
configured and done nothing.
Both images join the CI and release matrices, building from the root
context. Verified by building them, checking their contents, and bringing
the whole stack up - Rspamd reports its nine rate directives loaded from
the image.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The first deployments died building on the target: four images, and the
webmail build alone was enough to sink it on a small arm64 host - the
build step was killed with no BuildKit error, which is what an exhausted
or overloaded machine looks like from the outside.
The images already exist. CI publishes all four to GHCR for amd64 and
arm64 on every merge, and they are the ones the suite ran against, so
rebuilding them on the deploy target is slower, heavier and less
trustworthy than pulling them.
docker-compose.yaml therefore carries no build definitions, and
docker-compose.build.yaml is the overlay that adds them for local
development and for CI - which does have to exercise the branch rather
than the published images. Both paths were run before committing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The deploy target is an Oracle Ampere host, and the default clamav/clamav
tag is Alpine-based and published for amd64 only - the deployment failed
pulling it with "no matching manifest for linux/arm64/v8".
The Debian variant of the same upstream image is built for amd64, arm64
and ppc64le and ships the same clamdcheck.sh the healthcheck uses.
Verified by pulling it, checking the manifest, and bringing the service up
healthy. Every other image in the stack was already multi-arch.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The deployment sets NODE_ENV=production, and npm omits devDependencies
when it is set - so "npm ci" in the build stage installed no typescript
and "npm run build" failed with exit 127, command not found. The first
Coolify deployment died there.
The build stage now asks for them explicitly rather than depending on the
environment it happens to run in. Reproduced by building all three images
with NODE_ENV=production before and after.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The auth service refuses to start with a JWT_SECRET under 32 characters,
which is the point - but .env.example still carried a 30-character value,
so the stack could not come up with the file it ships with. CI caught it
on the compose-up job: auth reported unhealthy and everything depending
on it stopped.
Reproduced locally against the same file before and after; the stack now
starts from a clean checkout.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three things stood between this and a real deployment.
A missing JWT_SECRET fell back to a constant that is published in this
repository, so any deployment that forgot the variable would have accepted
session tokens anyone could mint, including an admin one. The auth service
now refuses to start without a secret of at least 32 characters. The test
suites set it through vitest config, because ESM hoists imports and an
assignment inside a test file runs too late.
Session cookies were issued without the Secure attribute: it is driven by
NODE_ENV, which compose never set. Behind a TLS terminator that is a
downgrade waiting to happen.
Nothing created the first mailbox. A fresh install has no domain and no
account, and every route into the console needs one - so there was no
first login. There is now a create-admin command, which also provisions
the standard folders and the four system aliases. It is a deliberate
command rather than a bootstrap on boot: creating an administrator as a
side effect of starting a container is how a deleted account quietly comes
back.
Also: the production images were shipping compiled test files, since the
build compiled everything under src. A separate build config excludes
them while lint still type-checks them.
Documentation. README.md covers what this is, how it is put together, and
what is not finished - stated plainly, because a mail server that half
works loses mail. docs/DEPLOYMENT.md is the Coolify path in the order it
has to happen, starting with the three prerequisites no code can satisfy:
outbound port 25, the PTR record, and the DNSSEC delegation.
docs/OPERATIONS.md is the runbook - health states, certificates, backups,
reading a delivery failure, DKIM rotation, and an explicit list of what is
not automated. .env.production.example replaces guessing which of the dev
defaults are safe: none of them are.
docs/ was gitignored as a place for internal notes; only docs/internal/ is
now, so these guides ship with the code.
Every command the documentation tells an operator to run was executed
against a running stack: create-admin, preflight, the health endpoint, the
published mail ports, and the Secure cookie.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
DKIM rotation broke the domain it was run on. It retired the live key and
inserted a placeholder as ACTIVE - private_key_enc was the literal string
"dummy-encrypted-key" - so FindActiveKeys then failed for the whole
domain and outbound mail stopped being signed, with the real key already
gone. Reproduced against a database before changing anything.
The root cause runs deeper: the auth service derives its encryption key
with a plain SHA-256 of the master key while the protocols service uses
HKDF with its own info string, so nothing the auth service seals can ever
be opened by the signer. Rotation therefore moves to the protocols
service, which owns that vault and the real key generator. It retires the
previous key with a seven-day overlap in the same transaction rather than
deleting it, because mail already sent under it is still being verified.
Domain onboarding had three defects in one function. It built the alias
destination by interpolating the domain name into the SQL text, which was
injectable; it reused one placeholder for a CITEXT and a VARCHAR column,
which made the statement fail and happened to hide the injection; and it
ignored its scope argument entirely, so any DOMAIN_ADMIN could create
domains. It also provisioned postmaster, abuse and security, missing the
dmarc and tlsrpt aliases that the published records point at - PLAN.md
section 7.4b is explicit that reports bounce without them.
The TLS panel returned fixed values: acme_status "VALID" and 72 days to
expiry, whatever the process actually held. It now reads the certificate
watcher, and a compile-time assertion keeps the wiring from silently
regressing into "disabled". The audit log and the readiness scorecard were
fixtures too, and their field names did not match the API they now read.
Two tests failed and a third asserted a fabricated constant: one depended
on test order through a shared fixture that had a second factor enrolled,
one used a fixed domain name and so passed once per database, and one
expected "records: 13" from a function that never touches DNS. All three
are fixed, and the suite now passes twice in a row against one database.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
F4's acceptance criterion asks for the three translations to be 100%
complete with the language lint green. The bundles had parity but the
admin screens and the account area were still hardcoded English: 63
visible strings that no locale could reach.
They now come from a shared ui namespace, so the three bundles cannot
drift apart, and all three carry 126 keys.
The language lint also caught this branch's own Go tests: the RFC 2047
cases need accented output to be worth anything, and the delivery test
had Portuguese fixture text. The expectations are written as Go escapes
instead, which keeps the assertions and leaves the source ASCII, and the
fixture text is English.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sessions and password changes were the two things PLAN.md section 14.2
asks of the account area that were still missing, and the admin console
had no audit trail at all despite section 14.3 requiring one for every
mutation.
Session revocation is enforced on every request, not just recorded. A JWT
carries its own validity, so revoking a row without checking it would have
made "sign out this device" a button that did nothing until the token
expired eight hours later. Changing a password revokes every other session
for the same reason: a password change that leaves the attacker's session
alive has not locked anyone out.
Admin queries are scoped by role in the SQL rather than in the UI, as
section 14.3 requires: a DOMAIN_ADMIN calling the API directly sees only
its own domain, and a mailbox outside that scope answers 404 rather than
403 so the scope cannot be used to enumerate what exists elsewhere.
The two mutations that exist so far - enabling or disabling a mailbox, and
retrying or cancelling a queued delivery - write actor, IP and target to
audit_log, and there is an endpoint to read it back.
Covered by tests against a real database: a revoked token stops working
mid-session, a password change needs the current password and takes the
other sessions down with it, and the minimum length is enforced.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
F4 lists real-time updates and the interface had none: an open inbox only
changed when the user reloaded it.
Events come from the transactional outbox rather than from an in-process
callback, which is what the outbox is for (PLAN.md section 9.4): the row
is written in the same transaction as the message, so an event can never
describe a delivery that rolled back, and a crash between commit and
publish loses nothing because the row is still there on restart. The relay
claims rows with SKIP LOCKED so a second instance can help without
publishing anything twice.
The hub fans events out per mailbox, keyed by the mailbox the session
resolved to, so one account's arrivals are never visible to another even
though they share the process. A subscriber that stops reading is dropped
rather than allowed to stall the relay; it reconciles on reconnect.
The client reconnects with capped exponential backoff and treats every
successful connect as a resync, refetching the folder. Without that, a
socket that dropped for a minute would leave the list quietly stale, which
is worse than having no real-time updates at all.
The upgrade goes through a Next rewrite, not the API route handler: a
route handler sees a finished request and cannot upgrade a connection.
Verified against the running stack - the handshake returns 101 through the
Next server, and a message delivered on port 25 arrives as a frame on an
already-open socket without the page asking for it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The mail list rendered two hardcoded messages and the reader showed their
canned HTML; compose ran a countdown and then navigated away without ever
submitting anything. Both now use the API, including search, the unread
badges in the sidebar (which were the literals "3" and "1", so an empty
inbox still advertised unread mail), and marking a message read on open.
The undo window now means what it says: the message is held in the browser
and only submitted when the countdown elapses, so cancelling it means
nothing was ever queued rather than trying to recall it afterwards.
Message bodies render in an iframe with an empty sandbox - no scripts, no
same-origin - and remote references are stripped until the reader asks for
images, so a tracking pixel cannot fire on open. The previous sandbox
allowed popups.
The admin page titles and the language switcher labels now come from the
bundles. The labels live there rather than in the source because PLAN.md
R9 keeps non-ASCII out of code, and each language should be named the way
that language names it.
ADR-001 assigned the mail API to the TypeScript storage service; the
reasoning for putting it in Go instead is recorded in PLAN.md, which
.gitignore deliberately keeps out of the repository, so that revision
lives only in the working copy.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The mail screens had no backend at all: they rendered a hardcoded array,
and the storage service that ADR-001 assigned this to is still a health
stub. The API lands here instead, because the folder, UID, MODSEQ, flag
and blob logic already exists and is tested in this service for IMAP.
Reimplementing it in TypeScript would mean two implementations of the
same semantics and two chances to disagree about what a UID means.
Reads use a projected read model rather than hydrated aggregates, which
is the CQRS exception PLAN.md section 2 carves out for webmail queries.
Every query is scoped to the mailbox resolved from the verified session,
never to an identifier taken from the request, so one account cannot read
another's mail by guessing a UID - covered by a test that tries exactly
that and expects 404.
Sending reuses ProcessOutboundEmailUseCase, so mail written in the browser
gets the same treatment as mail submitted over 587: send limits, DKIM
signing, and the durable queue. The From address comes from the session
and never from the request body; letting the body choose it is how a
webmail becomes a relay for spoofed mail.
MIME is parsed server-side because this service already depends on a MIME
library. A message that will not parse still returns its raw text - an
unreadable message beats an inbox that refuses to open one.
Marking a message read updates the flag and the folder counter in one
transaction, and only on a real transition, so re-opening a message cannot
drive unread_count negative.
Verified against the running stack: a message delivered over port 25
appears in the list with its decoded subject and sender, opens, marks
itself read, is found by search, and a message composed in the webmail
reaches outbound_jobs with the session's address as the envelope sender.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two problems the mail screens exposed, both in the delivery path.
The subject, snippet, sender name and message-id columns have existed
since the first migration, but nothing ever wrote them: every row went in
with those fields NULL. A message list built on the database would have
shown blank subjects for every message. Delivery now parses them once per
message and stores them, decoding RFC 2047 encoded words so an accented
subject is readable rather than raw. Parsing failures are non-fatal: a
message with unreadable headers is still delivered, just with less
metadata, because refusing it would lose mail over a malformed Subject.
Worse, a message the spam filter routed to Junk failed delivery outright
when the mailbox had no Junk folder - the sender got 451 and retried
forever against a mailbox that would never grow one on its own. Delivery
now falls back to INBOX and says so in the log. Found by sending a real
message through port 25 against the running stack.
The SMTP handler also swallowed the cause of any unexpected failure: the
peer saw "temporary error" and the operator got nothing. It is logged now,
which is how the missing folder was diagnosed at all.
Also adds the JWT verifier the mail API needs. The tokens are minted by
the auth service and verified here under the same secret, so the test
fixtures are real tokens produced by that service: if the two
implementations drift, this fails before the webmail does. Challenge
tokens, wrong secrets, tampered payloads, expired tokens and the other
surface's audience are all rejected.
The dev seed gained the standard folder set, which is what a provisioned
mailbox has and what the fallback above should not have to compensate for.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ts-auth gains a Postgres service and migrations. The security regressions
this branch adds - wrong password, pass-the-hash, wrong TOTP code, lockout,
single-use recovery codes - are all decided in the database, so without one
they skip silently and CI would go green on an unauthenticated server.
webmail-build only ran "npm run build", so i18n.test.ts never executed
anywhere. It now lints and tests too, and the package gained the lint
script it was missing.
*.tsbuildinfo is ignored: tsc --noEmit leaves it behind and it is a local
build cache, not source.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three separate gaps, all of which made the F4 interface non-functional or
unprotected:
The admin console rendered for anyone. There was no middleware and no
guard in the layout, so /admin/dashboard returned 200 unauthenticated -
only its data calls failed. middleware.ts now redirects both surfaces to
their login when the session cookie is missing, expired, carries the wrong
audience, or is a pre-MFA challenge token. It reads the claims without
verifying the signature on purpose: the signing key belongs to the auth
service and must not reach the browser bundle. That is sound because this
is not the boundary - every /api/v1 request is verified for real upstream.
Nothing served /api/v1/*. next.config.ts had no rewrites and there were no
route handlers, so every call the pages made 404'd against the Next
server, while AUTH_SERVICE_URL sat unused. A catch-all route now proxies
to the auth service, forwarding the client address and preserving multiple
Set-Cookie headers, which logout needs to clear both surfaces. Session
cookies also move to Path=/, since a cookie pinned to /user or /admin was
never sent with the /api/v1 requests that needed it.
i18n existed only as files. The three bundles had perfect key parity and
real translations, but no library, no imports and no switcher: the UI was
hardcoded English and i18n.test.ts compared keys while the feature was
absent. There is now a provider, a hook, a language switcher, and
server-side Accept-Language negotiation that honours q-values and matches
a bare "pt" to pt-BR. Logins, both layouts and the settings screen consume
it; the admin pages still have English titles.
The settings screen invented app passwords in the browser with Math.random
and never stored them. It now calls the API, which returns the password
once and keeps only its hash.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The F4 auth service issued a valid session for any email and any password,
and accepted any non-empty string as a second factor. Reproduced against
the router before touching it: an invented password plus the code "000000"
returned a SUPER_ADMIN token that opened /api/v1/admin/dashboard.
Passwords. Hashing moves from PBKDF2 to Argon2id via hash-wasm, which is
what PLAN.md section 9 and the Go service already use - the previous
scheme meant a password set through the mail stack could never verify
here. Interoperability was checked in both directions before adopting the
library: a hash written by alexedwards/argon2id verifies in Node and one
written here verifies in Go. hash-wasm is pure WASM, so the multi-arch
Alpine images need no native build.
verifyPassword no longer falls back to comparing the password against the
stored hash. That fallback turned every stored hash into a working
credential: anyone with a database dump could authenticate by sending the
hash back. Anything that is not an argon2id PHC string is now refused.
Logins are decided against the database, with the failure counter and the
temporary lockout of PLAN.md section 5.2. An unknown address and a wrong
password give the same answer, and the unknown case still pays for a hash
so response time does not enumerate valid addresses.
Second factor. The TOTP secret is generated server-side, stored encrypted
and never accepted back from the client - the confirm step used to take
whichever secret the caller sent, so anyone could confirm with their own.
Recovery codes are Argon2id, single-use, and consumed under a guard so two
concurrent requests cannot spend the same one. App passwords are issued
and revoked through real endpoints instead of browser state.
Sessions gain a "purpose" claim. Without it the challenge token - handed
out after the password but before the second factor - satisfied every
check a real session did, which skipped MFA entirely.
The dev seed carried a hand-written password_hash whose digest was 24
bytes instead of 32, so it matched no password at all. Nothing noticed
while passwords went unverified; it is a real hash now.
Verified with 30 tests, 13 of them against a real Postgres: wrong password,
unknown account, pass-the-hash, wrong TOTP code, challenge token replay,
lockout, single-use recovery codes, and the surface isolation that F4's
acceptance criteria require.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Concurrency. release.yml calls ci.yml, so a push to main started the
called workflow and the standalone one, both landing in the concurrency
group ci-<ref>. One cancelled the other and every merge left a cancelled
run behind. main is now excluded from the push trigger: release.yml is
already the entry point there, and pull_request still covers changes
before they land.
Node 20 deprecation (23 warnings). Every action was pinned to a major
that still declares node20. Each replacement was checked by reading the
action.yml at the new tag and confirming "using: node24", and the inputs
in use were confirmed to still exist there:
checkout v4->v7, setup-node v4->v7, setup-go v5->v7,
upload-artifact v4->v7, download-artifact v4->v7,
setup-buildx-action v3->v4, build-push-action v6->v7,
login-action v3->v4, metadata-action v5->v6
download-artifact has a v8, but v7 is also node24 and matches the upload
side, so the pair stays on one major. trivy-action is a composite action
and has no Node runtime to deprecate.
Go cache (1 warning). setup-go looked for go.sum at the repository root
and skipped its cache: defaults.run.working-directory applies to run
steps, not to actions. cache-dependency-path now points at
services/protocols/go.sum.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The E2E suite has never run in CI. repoRoot() walked up looking for
PLAN.md, but .gitignore excludes PLAN.md on purpose ("Internal planning
docs - never published with the code"), so it is absent from every fresh
checkout and the helper always failed. All twelve E2E tests died on it,
in this run and identically in #18 before any of these changes.
The marker is now the migrations directory, which is committed and is
what every caller of repoRoot is ultimately looking for.
Reproduced by exporting the tree with git archive - which yields exactly
what actions/checkout puts on the runner - and running the suite there:
it fails with the old helper and passes with the new one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CI gains a security job (Trivy fs: vuln, secret and misconfig) and an
image job that builds each of the four services and scans the result. The
secret scanner earns its place here: a DKIM key or Cloudflare token
committed by accident is a full compromise of every domain served.
release.yml publishes on a merge into main and on v* tags. Four services
on two architectures is eight build jobs pushing by digest, then four
publish jobs joining each pair into one manifest list, tagged latest, sha
and semver. github.repository is lowercased first because GHCR rejects an
uppercase path and this repository is lucasrguerra/LambdaMail.
Both scanners failed on first run, so this commit also fixes what they
found rather than shipping a workflow that is red on arrival:
- All four images ran as root (DS-0002), which section 17 risk R5
explicitly argues against. They now run unprivileged; protocols keeps
its eight ports below 1024 through cap_net_bind_service on the binary
rather than through being root.
- A HIGH advisory in postcss, reached only through Next. "npm audit fix
--force" would have resolved it by downgrading Next from 15 to 9, so
it is pinned with an override instead.
- The Node images copied the build stage's node_modules wholesale,
shipping typescript, eslint and vitest into production, and the base
image's own bundled npm accounted for the rest of the CVEs. Runtime
dependencies now come from npm ci --omit=dev and npm is removed from
the final layer.
- compose-up could never have passed: the proxy network is external
because Coolify owns it on the deploy host, and nothing creates it in
CI, so compose refused to start the stack at all.
Verified locally: actionlint clean, all ten action refs confirmed to
exist, Trivy fs and all four image scans exit 0, the full stack reaches
Healthy on every container, and the containers run as uid 10001/1000 with
all eight privileged ports bound and SMTP answering on 25. The arm64
build itself is unverified - this machine has no buildx - and will first
be exercised by release.yml.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
"docker compose up" could not work, which means the F0 acceptance
criterion was not being met:
- ghcr.io/rspamd/rspamd:3.9 returns "denied" and clamav/clamav:1.3 does
not exist. Both were verified against the registries; the working
references are rspamd/rspamd:3.9 and clamav/clamav:1.4.
- Rspamd received RSPAMD_REDIS_SERVERS and RSPAMD_REDIS_PASSWORD as
environment variables it never reads. Confirmed in the container that
it does not expand ${VAR} in UCL: the password reached Redis as that
literal string. Greylisting and ratelimit keep their state in Redis,
so both were silently doing nothing while appearing configured. The
credential is now rendered at startup into override.d, a tmpfs, so it
touches neither the image nor a volume.
- The ratelimit module ships enabled with no rates defined, which
accepts everything. ratelimit.conf defines the per-IP, per-user and
per-recipient buckets of section 5.2.
Verified by bringing redis and rspamd up: the real password resolves, the
rates load, and a probe message returns "Action: greylist" with the state
written to Redis.
Services also gained an image: alongside build:, so the deploy host pulls
instead of compiling while "docker compose build" still works locally.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every fix here is a case where the code and the plan disagreed, and the
code was wrong.
Certificates (section 8.2, 8.4). The parser documented that it skipped
expired entries but never checked, so an expired certificate would be
served until every handshake failed. LastReload() only advanced when
acme.json changed - and it does not change for sixty days - which made
the timestamp risk R2 monitors indistinguishable from a watcher gone
blind. Expiry is now checked on every pass, LastReload records the check
and LastChange the swap, and the <7d/<24h alerts of section 8.4 exist.
ARC (section 5). buildArcSealer resolved the DKIM key at startup, before
runDnsReconciler had provisioned one. On a first boot ARC was disabled
until someone restarted the process, which the log even admitted while
offering no way for it to start. A LazySealer binds on first use instead.
Health (section 8.4). There was no degraded state at all. /health now
reports one - still 200, because restarting cannot fix a certificate
Traefik never issued.
Bounces. DSNs were sent From postmaster@lambdamail.local, a domain that
does not resolve, so the notification was itself undeliverable. They now
carry the configured mail host, a Message-ID, and Auto-Submitted so a
vacation responder cannot loop against them.
Fan-out atomicity (section 6.1). Persist ran one transaction per
recipient, so a failure halfway through an alias fan-out left earlier
recipients with a copy while the sender was told to retry - and the retry
delivered to them twice. PersistAll commits the whole delivery or none of
it. Verified against Postgres by reverting the repository to the old
behaviour and confirming the new test fails on it.
SMTP replies. The use case encoded replies in error text ("550 5.7.1 ...")
which the adapter re-parsed by prefix, so rewording a message could turn a
permanent refusal into a retry. SmtpRejection carries the decision and the
adapter renders it.
Repository tests. These only run with TEST_DATABASE_URL, which CI does
set - so the job was red: copy and expunge seeded a 68-character digest
into a CHAR(64) column, and three others used fixed digests that collide
with the UNIQUE index on any re-run. A shared testBlobDigest() helper
fixes both; the suite now passes twice in a row against one database.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
