Repository navigation
Releases
EmailZit 0.12.0
Mail that a receiver refuses can leave another way, a receiver that refuses everything says so by itself, the nightly backup can take the customers' mail along, and three people can answer one info@ without three copies of it.
This release comes from one question: what does a mail server owe the person who runs it? It owes them a way out when a receiver turns the whole address range of their hosting provider away, a word when that happens, a copy of the mail, and the folder sharing every office needs. All of it is off or empty until somebody switches it on.
Delivery routes
- One recipient domain can leave another way (Settings → Delivery routes): through a relay the receiver accepts, or straight to the domain past a smart host that carries everything else. Large providers block whole address ranges of hosting providers — the answer then says that part of that network is on their block list — and no record and no setting on the server lifts such a block. A route is the only remedy a server operator has.
- The relay's password is kept sealed and written root-only for Postfix, and mail of a route with a sign-in goes out on a connection that must be encrypted. A route for a domain this server receives mail for itself is refused: it would take the mail of your own customers off the machine.
Deliverability
- A new page groups the outgoing mail by receiver for the last day, week or month: delivered, deferred, refused, and the last answer that receiver gave. Message by message, a receiver that refuses everything looks like a few lines among thousands — which is how such a block goes unnoticed for days.
- A receiver that refused or deferred at least five messages and delivered none is reported to the administrators, with what it said and what can be done about it, and reported again when the mail flows.
Notices without mail
- The same notices now also go to a webhook (a JSON POST with product, server, key, subject, text and time) and to Telegram. A mail server that cannot send mail cannot report by mail that it cannot send mail, and that is exactly the case that matters. HostZit sends the same fields, so one receiving end understands both products.
The customers' mail in the nightly backup
- The nightly run can export every active customer and put the export on the SFTP target, sealed with the passphrase of the backups. The import on the customer page opens a sealed file by itself, so a restore needs the panel and the passphrase and nothing else.
- It writes a full copy of every customer every night, so it is off by default: a server with a lot of mail belongs in a backup program (BackupZit backs up
/var/vmail).
Shared folders
- A folder of a mailbox can be opened for other mailboxes of the same customer, to read or also to write: the
info@that three people answer, without a password all three know and without a forwarder that makes three copies and loses every reply. The other mailbox sees it in its mail program undershared/<address>/. - Only inside one customer, and not by politeness: every customer is its own system user with a mail directory nobody else can read.
Old mail in Junk and Trash
- With a number of days the server deletes old mail out of Junk and Trash every night, for every mailbox, with Dovecot's own tool, so the index and the quota stay right. Mail programs do that only for the mailboxes that are opened, which is why those two folders fill the disk of a hosting server. Both spans are 0 until somebody sets them.
One click from HostZit or a billing system
POST /api/v1/customers/{id}/ssonow makes the customer's panel login on the first click, as cPanel does, with the username the customer shares with HostZit. The session of a one-time link is not sent to the first password change, and while the customer never chose a password they choose one without the current one; a login staff switched off stays off. Thanks to the HostZit team for this one.
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.12.0/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
EmailZit 0.11.5
A new server no longer reports a failed postfix in the log of its first start, and the note of a domain reads in the language of the console.
Both come from the HostZit team's joint test J1: one line in the log of a first boot and one note on a domain page that was never translated. Nothing of this changes how mail is accepted or delivered.
Changes
- An interrupted write of the mail configuration keeps the files it wrote. The package of a new server stops the mail helper while that very write is running, so the write counted as failed and was rolled back — on a new server the rollback removes
/etc/postfix/main.cfagain, under a postfix that is starting at that moment. The log of a first start carried apostfix: fatal: open /etc/postfix/main.cfand a failed unit for about three seconds, until the next start wrote everything again. The end state was always right; the log line was not. An interrupted write is no longer treated as a failed one, so the files stay and the next start finds its work done. - A start that finds the mail configuration unchanged brings up the mail services that are not running. This belongs to the change above: files that stay mean that nothing has to be written, and the services the interrupted start never reloaded would otherwise wait for the next restart of the server.
- The note of a domain says in the language of the console which records HostZit kept. When HostZit already holds own records of the customer with the same names, the domain page said so in a sentence that was built together with HostZit's own message and therefore stayed English. The sentence is now a text of the console, translated in all 14 languages, and it names the records HostZit kept (at most eight, then "…"); HostZit's message is shown only when it sends no list. The same holds for "HostZit is not reachable." and "HostZit refused the records:".
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.11.5/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
Thanks to the HostZit team for both findings.
EmailZit 0.11.4
A domain kept in HostZit's DNS takes its verification record back once it is verified, and "Check now" asks the DNS the same reliable way as the hourly check.
The joint test J1 on two live servers: a hosting customer got its mail on EmailZit, its domain was verified from the public DNS, and the customer's website sent through the relay without waiting. Three things were tidied up along the way. Servers without a linked HostZit are affected only by the second one.
Changes
- The verification record leaves HostZit's zone after the verification. EmailZit did not write a domain's records again once the domain was verified, so the
_emailzitTXT record stayed in HostZit's zone. Now the records are written again after a verification by the button, by the hourly check, or when staff mark a domain verified or not verified by hand. - "Check now" asks like the hourly check: the name servers of the domain's zone, with the absolute name. Before, it asked the server's own resolver with a relative name, so an answer that resolver remembered from before the record existed could keep the button saying "not found" for up to an hour.
- The page of a domain whose records EmailZit keeps in HostZit no longer also asks to put them at a DNS provider.
- The guide says in all 14 languages that the record leaves HostZit's zone after the verification.
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.11.4/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
EmailZit 0.11.3
A domain proves itself in its own DNS without a cache getting in the way, and a "try again later" says what it waits for.
The HostZit team ran the joint test J3 on two live servers: a website whose domain has no mail here sent over the relay, and the message was delivered — but only on the fourth try, fifteen minutes after the first question, although the record had been in the public DNS for five of them. The reason was the question EmailZit asked, and this release fixes it. Servers without a linked HostZit are not affected.
Changes
- The question to the DNS carries the absolute name. Without the final dot a resolver also asks the name with the server's own search domain appended (
_emailzit.example.com.mail.example.net). An authoritative name server refuses such a question, that refusal is not a "there is no such name", and EmailZit then asked the system resolver — whose negative answer is remembered for up to an hour. A record published a minute ago could stay invisible to the check for that hour. - What the name servers of the zone say is believed, "there is no such name" included. The system resolver is asked only when the zone cannot be asked at all: no reply, a refusal, or a name server that answers one question and refuses the next (the reason the fall-back exists at all).
- The name servers of the zone have their own deadline (8 of the 20 seconds a question may take), so one that stalls leaves time for the system resolver instead of eating the whole question.
- A "try again later" of the relay names what it waits for. While the record of a website's domain is not in the public DNS the answer is now
4.7.1 The domain is being checked in its public DNS; try again later; the old text about a hosting server that cannot be asked is kept for exactly that case. The owner of the website reads this text in HostZit, so it has to name the right thing. - The log says why a proof is not there yet: one line per question with the domain, the record, what the DNS answered and how much of the fifteen minutes is left.
- The guide says in all 14 languages what a 451 of the relay waits for.
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.11.3/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
Thanks to the HostZit team for the joint test and for finding this.
EmailZit 0.11.2
Several linked HostZit servers again keep their records when one of them is down, and the record a website domain has to show is now in the panel.
The HostZit team retested EmailZit 0.11.1 against HostZit 0.10.1 in a lab with two hosting servers and sent three findings and three questions. This release answers all of them. A server with one HostZit, or none, sees only the new list on the HostZit page.
Changes
- One HostZit that cannot be reached no longer keeps the others from taking the records. In the loop over the links a transport error ended the loop, so the records of a customer whose zone a later link holds were never written while a stopped or renamed HostZit stood in front of it. A link that does not answer is now skipped like a link that does not hold the zone, and its note stays only while no link took the records.
- A HostZit older than 0.10.1 gets the mail records without the proof record. Such a HostZit does not know the name
_emailzitand refuses the whole set because of it, which left a domain without MX, SPF, DKIM and DMARC after an upgrade of EmailZit alone. The set is now written once more without that record, and the domain page says to update that HostZit or to set the record at the DNS provider of the domain. - The panel shows the record a website domain has to prove itself with. Settings → HostZit lists, under every link, the domains of websites that have no mail on this server, each with its record (
_emailzit.<domain>TXTemailzit-relay=…), whether it is in the public DNS and when it was last seen there. A customer whose DNS is not on HostZit can be given that record; a domain shown as missing is the website mail the relay refuses. - A record inside a zone is asked at the name servers of that zone. A website on a name inside a zone (
shop.example.comin the zoneexample.com) has no name servers of its own, so the check used to fall back to the system resolver, which can still answer from what it remembered. The zone above the name is now found and asked, for the proof of the relay as well as for the verification and the DKIM record of a mail domain. - A new server writes its mail configuration even when the first try is lost. The mail helper writes the Postfix and Dovecot configuration when it starts, and the package of a new server restarts the helper while that write runs. One lost write left the server without
/etc/postfix/main.cf: Postfix was skipped, Dovecot and Rspamd stayed down, and no mail flowed until an administrator saved the mail settings by hand. The write is now tried again. - The guide says in all 14 languages where the record of such a website domain is shown, and what an older HostZit means for the records of a domain.
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.11.2/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
Thanks to the HostZit team for the retest and the findings.
EmailZit 0.11.1
A security release: the public DNS, not another server, decides what the relay and a domain are allowed to do.
The HostZit team read the code of 0.11.0 and ran joint tests with two HostZit and two EmailZit servers in a lab. Six findings came back, all of them the same mistake in different places: EmailZit trusted what the other server said instead of what the DNS shows. This release answers all six. Servers that run both products for their own customers see no difference; a server that links a HostZit of a client gets the fences it was missing.
Changes
- A zone inside HostZit is no longer a proof of a domain. Any hosting account can create a zone for a name that is not theirs, and such a zone does nothing in the public DNS. EmailZit now writes the verification record into the zone as well and verifies the domain only from the public DNS, at the name servers of the domain itself — right after the write and once an hour. A HostZit that really holds the delegated zone makes the record visible in seconds, so nothing changes for honest customers. The same rule now decides the DKIM signing: a domain is signed when its record really is in the DNS, not when HostZit accepted it.
- Websites of a linked HostZit may only send with domains that are proven. A domain that has mailboxes here is accepted only when the HostZit of that relay login hosts the domain as well, for the same customer; a customer an administrator assigned to the link by hand needs no such question. A domain that has no mail here must show the record EmailZit asked that HostZit to publish in the public DNS: while it is not visible the answer is 451 (the website's message waits in its queue), afterwards 550, and a proof counts for seven days. Each link has to make its own record appear, so one HostZit can neither pass with the record of another nor take its proof away.
- A relay login works only from the addresses of its own link. The port is open for the addresses of all linked HostZit servers together, so without this the login of one hosting server could be used from another.
- The scope of a new link belongs to the pairing code. Settings → HostZit → Make a pairing code asks which customers the new link serves, and the link has that from its first second — a HostZit never sees more than the administrator meant, not even in the seconds between the pairing and a change on the settings page. The first link of a server starts at "all customers"; with a link already there, a new link (also one added by hand) starts with none until its customers are chosen. Pairing again with a server that is already linked leaves its customers alone.
- The records of a domain that no HostZit holds are offered again every six hours, so a zone created later, or a write that was refused while something else was in the way, is picked up without anybody clicking.
- The DKIM check asks the system resolver when a name server of the domain answers the test question and then refuses the record itself — one such name server used to leave the check undecided for ever.
- The guide says in all 14 languages how a domain verifies itself from the public DNS, what the pairing code decides, and what the relay demands.
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.11.1/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
Thanks to the HostZit team for the review and the joint tests.
EmailZit 0.11.0
A package is now the plan of one mailbox, staff can open a mailbox to help a customer, and EmailZit links several HostZit servers at once.
Changes
- A package is the plan of one mailbox. Mail is sold per mailbox, so a customer can hold several plans at the same time — for example three Mail 10, two Mail 100 and one Mail 25. Every mailbox has the size and the sending limits of its own plan; the customer's page says how many mailboxes of each plan the customer has and how many are in use, and a new mailbox can be made while a plan still has a free one. Staff can move a mailbox to another plan later, and it then gets that plan's size. The customer keeps a base package for what belongs to the whole customer: domains, aliases, the storage of all mailboxes together, the sending limit of the customer and the recipients per message.
- The REST API writes an order with
PUT /api/v1/customers/{id}/mailbox-packages({"packages":[{"package":"mail-10","mailboxes":3},{"package":"mail-100","mailboxes":2}]}), and every answer about a customer now carriesmailbox_packageswith what is paid for and what is in use. Existing servers keep what they had: every mailbox gets its customer's package, and the customer gets as many mailboxes of it as that package allowed. - Staff can open a mailbox, the way a hosting panel does: the webmail of that mailbox opens for half an hour without its password, and the password and the second factor of the mailbox stay untouched. It is for the question customers ask most often — did that message arrive. Every time is written to the audit log, and the mail user's own page says when it last happened.
- Several HostZit servers at once. EmailZit keeps a list of links: pairing with a new server adds a link, pairing again with the same address replaces only that link, and ending a link leaves every other link with its token and its relay login. Each link has its own relay login for the websites of that HostZit, and the check of a sender asks the HostZit the login belongs to.
- Each link serves all customers or only the ones chosen for it. A HostZit of a single client, linked to a shared EmailZit, then sees nothing of the other customers: its token answers "no such customer" for them, it cannot open their one-time sign-in links, its relay carries only their domains, and DNS records are written only for them. A customer that link creates over the API belongs to it.
- The certificate of a linked HostZit is checked against the public key the pairing code carried, instead of being trusted blindly, and a token the pairing made stops working as soon as its link is gone.
- A domain is signed only while its DKIM record really is in the DNS. EmailZit looks the record up at the name servers of the domain at every start and then every hour; until it is published the mail goes out unsigned, because a signature whose key cannot be found fails at the receiver. When HostZit keeps the zone, EmailZit writes the record itself and signs at once.
- When the relay port cannot be opened because something else holds it, the log says so once instead of every half minute.
- The guide describes the plans, opening a mailbox, the links and the DKIM rule, in all 14 languages.
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.11.0/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
EmailZit 0.10.0
One code links EmailZit and HostZit, and the websites of a HostZit on another server send their mail through EmailZit.
Changes
- Linking HostZit with one pairing code. Settings → HostZit → Make a pairing code shows a code once; paste it into HostZit. The code works once and only for ten minutes, EmailZit keeps only its hash and never writes it to a log. The two servers then hand each other a token of their own, each with the least permissions and bound to the addresses of the other server, so nobody copies a token or an address list by hand. The page says which server is linked, since when, and from which addresses its token works; Unlink on either side revokes both tokens. A link typed in by hand keeps working.
- Mail from the websites of a HostZit on another server. Until now only websites on the same server could send through EmailZit. The pairing now also hands over a relay login, and EmailZit's own submission server takes that mail on port 2587 after STARTTLS and that sign-in. The port is open only while such a link exists and only for the addresses of that HostZit, and the login works nowhere else — not for a mailbox, the webmail or a mail program.
- Every relayed message is checked: the sender domain must be a mail domain of this server or a domain HostZit hosts for one of its customers (asked once and then remembered for ten minutes; an unknown domain is refused with 550, and while HostZit cannot be asked the answer is 451, so the website tries again later), and the From address must use that same domain. The customer's package limits count this mail together with the mailboxes. The message then goes the way of a website on this server: DKIM signature, outgoing spam check, delivery log.
- The customer sees the mail of their websites on their page — refused with the reason, taken with what the delivery log says about it — and the weekly report counts it.
- A linked HostZit no longer takes the ports 80 and 443. Only a web server installed on the same machine does, and when that web server is gone the ports come back by themselves. Before, a HostZit on another server could leave autoconfig, Autodiscover and the MTA-STS policy unanswered.
- The hint under an API token's allowed addresses now says to enter every address of the other server, IPv4 and IPv6.
- The guide describes the pairing and the relay, in all 14 languages.
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.10.0/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
EmailZit 0.9.0
The domain page leads through the set-up, step by step.
Changes
- Setting a domain up is now guided. The domain page says which of the four steps the domain is on — prove that the domain is yours, make the first mailbox, point the host names here and let the certificate cover them, put the records at the DNS provider — and how many of the required records are already in the DNS. With HostZit as the DNS provider EmailZit writes the records itself, and that step is skipped.
- It says what is in the way, which is where most of the time is lost: host names the server's certificate does not cover yet (publish
_mta-stsonly afterwards, or other servers cannot read the policy), host names that point at a proxy in front of the DNS (Cloudflare's orange cloud, for example) instead of at this server, and an MX record that still points at the old provider — bring the mail over first, change the MX last. - The DNS check asks the name servers of the domain itself. Before it asked the server's own resolver, which remembers that a name did not exist and then reports a record as missing although it is in the DNS.
- All records at once as lines of a zone file, with a long DKIM value in quoted pieces.
- A test message from the domain: the panel sends a short, fixed message from
postmaster@of the domain to an address you type, so the receiver's verdict on SPF, DKIM and DMARC becomes visible. One message per domain and minute. - The guide describes the four steps and the traps, in all 14 languages.
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.9.0/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).
EmailZit 0.8.3
Backups keep working when a stray file sits next to them, and a restored server takes over the mail settings.
Changes
- A file EmailZit cannot read no longer costs the server its backup. An old certificate an administrator had put aside in the server's TLS folder belonged to root; the panel could not read it and the whole backup failed with "permission denied" — so the server had no backup at all. Only the files EmailZit itself needs (
secret.key, the panel certificate and its key) must be readable now; anything else in the folder is left out, named in the backup's manifest and written to the log. - A restore now takes over the mail settings. After restoring a backup on a new server, Postfix and Dovecot kept the name of the machine they were installed on until an administrator opened Settings → Mail and saved. The restore now marks the data directory and the panel writes the mail configuration of the backup at its next start, as the backup page has always promised.
- Spam lists: an entry with braces (
{or}) in the address is refused. Such an entry ends up in a Sieve script, where${…}would be a variable if a later script asked for the variables extension. Found by the fuzzer; no address uses these characters.
Checked before this release: every panel route against every role and every API route against every token scope; tests with the race detector; fuzzing of the parsers of untrusted input; gosec, staticcheck, govulncheck (Linux and Windows), Semgrep, Trivy, Gitleaks; OWASP ZAP against a fresh appliance; Trivy, Lynis, testssl.sh and Nuclei against the appliance image; upgrade from 0.8.2 and installation with install.sh on a fresh Debian 13 server; a backup of the public server e1.maxzit.com restored on a fresh appliance (customers, domains, mailboxes, aliases, DKIM keys and the mail settings all came over).
Install or update
curl -fsSLO https://github.com/Max-Zit/emailzit/releases/download/v0.8.3/install.sh
sudo sh install.sh --hostname mail.example.comExisting servers update from the panel (Updates and server). Appliance images are attached below (check them against SHA256SUMS-appliance).