The web services listener of Progressive Robot hMailServer 6.3.3 through 6.3.7 counts every failed CardDAV, CalDAV or Autodiscover sign-in from a proxy it trusts - one on the same machine, or one named in WebServicesTrustedProxies, whose X-Forwarded-Proto it believes - against the proxy's own address, which allows a remote unauthenticated attacker who can reach the proxy to have that address banned from the server's mail listeners. The listener, off by default (WebServicesHttpPort and WebServicesHttpsPort 0), signs an account in with HTTP Basic over HTTPS for CardDAV from 6.3.3 and for CalDAV and Outlook Autodiscover from 6.3.4. Behind a TLS-terminating proxy, which the server trusts when it is on the same machine or named in WebServicesTrustedProxies, the listener takes the client's HTTPS from the proxy's X-Forwarded-Proto header, but it does not believe X-Forwarded-For and sees the proxy's address for every client. Each failed sign-in counted towards the auto-ban against that address, so that failed sign-ins through the proxy from anybody, under any user name, reaching MaxInvalidLogonAttempts (3 within 30 minutes by default) wrote an IP range that refused the proxy's address at the SMTP, POP3, IMAP and ManageSieve listeners for AutoBanMinutes (60 by default), and could write it again each time it lapsed. For a proxy on the same machine that address is the loopback address, so that a webmail on that machine, local scripts that submit mail and the other local clients of the mail listeners were shut out; for a proxy at another address that also carries mail, every mail client behind it was. Such a proxy forwarding to the listener's HTTPS port, re-encrypting, was counted against in the same way, and listing it in WebServicesTrustedProxies did not spare it. The auto-ban is on by default and AutoBanNeverBan, which spares an address, is empty by default; the listener itself went on serving the proxy.

Progressive Robot hMailServer 6.0.0-B1 through 6.3.7 on Windows hands OpenSSL the names of its own TLS files - the listeners' certificates, chains, private keys and Diffie-Hellman parameters, a port's client-certificate CA file (from 6.2.19), the object store's CA file (from 6.3.4) and the ACME client's keys and certificates - as Latin-1 bytes, which on code page 1252 are the code page's own for U+00A0 to U+00FF, and OpenSSL's file opening on Windows reads a name as UTF-8 first and opens the file at that reading when it is there, or creates it there when its folder is. When every byte of a configured name outside ASCII also reads as UTF-8 - for example U+00C3 U+00A9, the bytes C3 A9, which are U+00E9 in UTF-8 - the server used the file at that other name whenever one was there, which allows a local attacker who can create that file or its folder but cannot change the configured one, as any user can by default beside a file an administrator keeps under ProgramData, to have a port that logs on by client certificate (from 6.2.25) accept certificates from a certificate authority of the attacker's own and so sign in over the network as any mailbox without its password; to have the server trust such an authority for its object store and, intercepting that connection, read and modify the messages it stores; to receive the private keys the ACME client writes, the key of a certificate issued for the server's names among them; or to keep TLS ports from starting. The configuration is not the default: by default these files have names in ASCII, client certificates are off on every port, and the object store and ACME are off. The Linux builds are not affected.

Misinterpretation of SQL statement text in the database layer of Progressive Robot hMailServer 6.0.0-B1 through 6.3.7 allows a remote unauthenticated attacker to have a failed database write reported to the server as successful. The layer answers a failed statement as a success when it finds the marker [IGNORE-ERRORS], which is meant only for the server's own schema statements (its upgrade scripts and the database log's tables), and statements carry values from outside - an envelope address, the user name of a failed logon, a message's headers - in their text on MySQL/MariaDB and PostgreSQL, and in the batches of the database log on every engine. Up to 6.2.18 the marker was honoured anywhere in a statement, a value included, on MySQL/MariaDB and PostgreSQL. From 6.2.19 it was honoured only inside a comment, but the layer read a backslash inside a string literal as an escape, which SQL Server, SQL Server Compact and PostgreSQL with standardconformingstrings on (its default) do not, so that on those engines a value holding a backslash could make the layer take part of the statement's literals for statement text and find the marker there: on PostgreSQL from 6.3.3 for every statement carrying an outside value that holds a backslash, with no setting changed for a message delivered into a local mailbox (through an internationalised (SMTPUTF8) envelope sender) and for the IP range an auto-ban writes (named after the user name of the failed logon), and on SQL Server and SQL Server Compact for the database log, when the server logs to the database (logging is off, and the log device is files, by default). A statement run outside a transaction that then failed - a constraint, a deadlock, a lock or statement timeout, a full disk, and up to 6.2.18 a lost connection - was answered as a success, and its caller went on as if the row had been written: a message delivered into a local mailbox could be counted as delivered with no row to show it and no failure notice to its sender, and a message accepted with 250 could be left with no queue row and never delivered (up to 6.2.18, and on PostgreSQL from 6.3.3 when every recipient's address carries the marker too, which needs an address that takes any local part); an auto-ban whose IP range was not written could be logged and counted as applied, with no error reported, while the address stayed unbanned; and message metadata (written only when message indexing is on), entries of the database log and rows of features an administrator turns on could be lost without an error. On MySQL/MariaDB up to 6.2.18 the caller was handed whatever identifier the client library still held for the connection - 0, or that of an insert its last successful statement made, in whatever table - under which a queued message's recipients were then written, so that another message still in the queue could be delivered to them as well. This flaw gives a value no way to change what a statement does, and no privilege is gained.

SQL injection in the database layer of Progressive Robot hMailServer 6.0.0-B1 through 6.2.18, when the server uses MySQL/MariaDB or PostgreSQL, allows a remote attacker to alter SQL statements the server runs. On those two engines parameters are not bound: SQLStatement::GenerateFromCommand wrote each value into the statement's text as a quoted literal with a separate string replacement over the whole text per parameter, so a value holding the name of a parameter substituted after it had that name replaced, inside the value, by the later value with its own quotes, which ended the literal early and placed the later value in the statement as SQL. The row the server writes for each recipient of a queued message carries the recipient's address twice, so in the default configuration an authenticated user who may send mail reaches it by sending to an address of their choosing. A remote unauthenticated sender reaches the same statement through a message the server sends to the envelope sender, such as a delivery failure notice or an automatic reply, and, when greylisting or message indexing is enabled (both are off by default), through the greylisting record's envelope addresses or through the message metadata the server stores for IMAP SORT, which carries four header fields of a received message. An altered statement runs with the privileges of the server's database account; on PostgreSQL, whose client call runs every statement in the text it is given, further statements can follow it, while on MySQL/MariaDB, where one statement is run per call, the change is confined to the statement it is in. SQL Server and SQL Server Compact, where parameters are bound, are not affected.

The auto-ban of Progressive Robot hMailServer 6.0.0-B1 through 6.3.7 can be kept from banning an address after repeated failed logons, which allows a remote unauthenticated attacker who can connect from two or more addresses to go on guessing account passwords - over SMTP AUTH, IMAP and POP3, and over ManageSieve, CalDAV/CardDAV and the REST API where those are turned on (off by default) - without being banned. When the failed logons counted from one address reach MaxInvalidLogonAttempts (3 by default), the server clears that address's count, saves an IP range that refuses the address for AutoBanMinutes (60 by default) and drops the connection. The range is named "Auto-ban: " followed by the user name of the logon that reached the limit, and no two ranges may have the same name. The server chooses the name by checking candidates as written - the name, then the name with a number, a random suffix or a GUID added - and only then cuts it to the 100 characters the database column holds, so that with a user name of 90 characters or more every candidate is the same once cut. While one range named after such a user name stands, every later ban named after a user name with the same first 90 characters therefore fails at the database: no range is saved, the address is not refused, and, its count having already been cleared, it is given the next full set of attempts. A user name of 78 to 89 characters makes later bans fail in the same way: from the second, the third or the twelfth on, depending on its length and the database, or, from 78 to 85 characters, by chance. An attacker who has once had an address banned under such a name can in this way keep guessing passwords from any other address, ending each set of attempts with a logon under a name of that kind; each failed save is reported in the server's error log (as the database error HM5032, and from 6.2.19 also as HM6099). On SQL Server a user name holding characters outside the database's code page has the same effect at any length, because the check compared names in that code page; that save is refused by the server's own check of the name, with no database error, and is reported only from 6.2.19 (HM6099). A user name holding a NUL character, which the web services listener's HTTP Basic sign-in (off by default) carries from 6.3.3 and, on the Linux builds, IMAP LOGIN, offered as installed, from 6.3.0, makes bans fail at any length: on MySQL, MariaDB and PostgreSQL the statements that check the name and save the range end at the character, so that every ban named after such a name fails, from the first and from a single address, and, its count having already been cleared, the address is given the next full set of attempts; on SQL Server and SQL Server Compact the name is bound only as far as the character, so that while one range under the part before it stands, every later ban named after such a name fails. On SQL Server, and on MySQL and MariaDB for an account whose local part has nine characters or more, a guess at the account can itself carry a name of that kind, since those databases find an account under its address followed by spaces (and SQL Server also under one holding characters its comparison ignores), so that not even a MaxInvalidLogonAttempts of 1, which bans each failed logon under the name it gave, avoids it there.

The auto-ban of Progressive Robot hMailServer 6.2.0 through 6.3.3, with a MySQL or MariaDB database, can be kept from banning an address after repeated failed logons, which allows a remote unauthenticated attacker to go on guessing account passwords from a single address without being banned. The server makes its MySQL tables in MySQL's utf8 character set, which holds no character outside the Basic Multilingual Plane, sends its statements as UTF-8 over a utf8mb4 connection and leaves the database's default strict mode in place. From 6.2.0 it decodes the user name of a SASL PLAIN logon as UTF-8, so the name can hold such a character. When the failed logons counted from one address reach MaxInvalidLogonAttempts (3 by default), the server clears that address's count, saves an IP range named "Auto-ban: " followed by the user name of the logon that reached the limit, and drops the connection. If such a character is in the part of the user name that the range's name keeps (the name is cut to 100 characters), the database refuses the range: no range is saved, the address is not refused, and, its count having already been cleared, it is given the next full set of attempts. An attacker can in this way spend all but the last attempt of each set on password guesses, and make the last under such a name. On Windows, POP3 AUTH PLAIN reaches it on a server as installed, and IMAP AUTHENTICATE PLAIN where IMAP SASL authentication is turned on (off by default); on Windows and on Linux (from 6.3.0), SMTP AUTH PLAIN reaches it where plain-text authentication is allowed (off by default), and ManageSieve and CardDAV where they are turned on (off by default). Each failed save is reported in the server's error log. A database whose strict mode has been turned off stores the character as a question mark and makes the first such ban, but refuses every later ban named after the same user name, or after one holding other such characters in their place, as a duplicate while the first stands, since the server cannot look up a name holding such a character there.

The sign-in of Progressive Robot hMailServer 6.2.22-pre2 through 6.3.7 takes for an account names that are not its address but under which the database's own comparison finds it, while the per-name lockout and the compromised-account hold are kept by the name as the client gave it. This allows a remote unauthenticated attacker to go on guessing an account's password beyond the lockout's threshold, and anyone who has the password of an account held for review to sign in to it. The per-name lockout (from 6.2.22-pre2; AccountLockoutThreshold in hMailServer.ini, 0, off, by default) locks a name after repeated failed sign-ins. The hold (from 6.3.4, where the compromised-account detection is set to enforce, which is not the default) refuses an account judged to have been taken over. Both are asked, before the password, of the name the client gave: trimmed, in lower case and with a domain alias or the default domain applied. The account whose password the sign-in then checks is whichever one the database finds under that name. The databases find an account under spellings that this key counts apart. MySQL and MariaDB find it under one with accented letters in place of plain ones, and MariaDB 11.8, in a database made before 6.3.4, also under one with full-width letters, with a single letter in place of a pair of plain ones, or with characters its comparison ignores in any number. SQL Server and SQL Server Compact find it under one with full-width letters or with characters their comparison ignores in any number, and SQL Server also under one with a single letter in place of a pair of plain ones, with a character of Latin-1 in place of a digit and, for an address holding letters of Latin-1 outside ASCII, under those letters in the other case, and, under an accent-insensitive collation, with accented letters. PostgreSQL finds it under one holding a letter whose lower case under the database's locale is a plain letter, and, for an address holding letters outside ASCII, under those letters in the other case; on the build for musl (from 6.3.6) the server's own lower case folds nearly all of these too. On every database, a domain alias saved with white space at the end of its name gives such a spelling too, and on every database but PostgreSQL so does an account's bare local part where the default domain was saved with spaces at its end and no other white space (neither is the default). Each such spelling has a lockout threshold of its own, so the lockout limits the guesses at an account only to its threshold times the number of the account's spellings. That is far beyond any threshold for most addresses on MySQL and MariaDB, and without limit on SQL Server, on SQL Server Compact, the database the Windows installer uses by default, and before 6.3.4 on MariaDB 11.8. A locked name is got round under another spelling. A held account is signed in to under another spelling with its password over IMAP, POP3 and SMTP AUTH, and, where they are turned on (off by default), over ManageSieve, the REST API and the webmail. IMAP LOGIN, offered as installed, carries such spellings on MySQL, MariaDB and SQL Server Compact, and, on SQL Server and PostgreSQL, for some addresses; the sign-ins that carry a name as UTF-8 carry them for the others on SQL Server, and on PostgreSQL for those holding one of the plain letters another letter lowers to or a letter outside ASCII with another case. The same lookup mapped to accounts the names given by an OAuth 2.0 token, a client certificate, a Kerberos ticket and an OpenID Connect or SAML identity provider, where those are set up (none is by default), so that a name an issuer gave to one party could sign in to another's account; OAuth 2.0 tokens were mapped so from 6.2.0, before the per-name lockout existed. A name holding a NUL character, carried from 6.3.0 on Linux and 6.3.3 on Windows, failed the statement on MySQL, MariaDB and PostgreSQL.

The web services listener of Progressive Robot hMailServer 6.3.3 through 6.3.7 does not apply the server's IP ranges to the connections it accepts, which allows a remote unauthenticated attacker to go on guessing account passwords at the listener's sign-ins after the auto-ban has banned the attacker's address. The listener is off by default (WebServicesHttpPort and WebServicesHttpsPort 0). From 6.3.3 its CardDAV, and from 6.3.4 its CalDAV and its Outlook Autodiscover answer to a request that carries credentials, sign an account in with HTTP Basic over HTTPS. A failed sign-in there counts towards the auto-ban as a failed logon on any other protocol does: when the failures counted from one address reach MaxInvalidLogonAttempts (3 within 30 minutes by default), the server writes an IP range that refuses the address for AutoBanMinutes (60 by default), and the SMTP, POP3, IMAP and ManageSieve listeners refuse it. The web services listener asked no IP range when it accepted a connection (in 6.3.3 nothing, from 6.3.4 the country policy alone), so it went on serving such an address, whether its ban had been earned there or on another protocol, and also an address that an administrator's IP range allowing nothing holds. Each failed sign-in there closed only its own connection, and the next connection was accepted, during the ban and after it. The firewall rule and command the auto-ban can drive (AutoBanFirewall and AutoBanCommand, off by default) are told only the SMTP, POP3 and IMAP ports, so that the firewall rule and the packaged hook did not close the listener's ports, and the fail2ban jail packaged for Linux (off by default) lists only the mail and ManageSieve ports. The only limit left on the guesses at one account there was the per-name lockout (AccountLockoutThreshold in hMailServer.ini, 0, off, by default), which has a separate defect fixed in the same release. The flaw is reached on the listener's HTTPS port: behind a TLS-terminating proxy the listener sees the proxy's address for every client, and a ban never named the client there.

Incorrect default permissions in the Windows installer of Progressive Robot hMailServer 6.3.7 allow a local authenticated user to run code as the administrator who installs the product, when it is installed into a new program folder in a location whose permissions let every user write, such as the root of a drive. 6.3.7 fixed CVE-2026-107571 by giving the program folder, Bin and ControlPanel permissions that only administrators may change, but its setup lets Inno Setup make a new program folder first, with the permissions the folder above hands down, and replaces them only afterwards, tens of milliseconds later (33 to 139 milliseconds in the project's test runs). In that time any local account may make files and folders in the new folder, which keep that account as their owner, among them the Bin folder that setup then copies the programs into. Setup runs programs from Bin with the administrator's elevated rights, registering the service among them, before its hardening at the end of setup takes such a folder back, and those programs load a library placed in Bin in place of the system's. The default installation under Program Files, whose folders give ordinary users read and execute only, is not affected.

Unverified ownership in the Windows installer of Progressive Robot hMailServer 6.3.7 allows a local authenticated user to run code as the mail server's service account or as an administrator when setup installs into, or upgrades, a program folder that accounts other than the administrators could write before, such as one made beforehand at the root of a drive. 6.3.7 fixed CVE-2026-107571 by giving the program folder, Bin and ControlPanel permissions that only administrators may change, but setup refused nothing that was already in them, or in the folders that hold the event script, the add-ons and the database scripts: a program, a library, a script or a settings file that another account owns, or whose own permissions let another account change it, is kept with that owner and those permissions, which the hardening of the folders does not take away. Setup then runs programs from Bin with the administrator's elevated rights, and the service, which runs as LocalSystem or as its virtual account, runs them and the event script, so that a library such an account placed in Bin is loaded in place of the system's; the account can still change the file after the installation. The default installation under Program Files is not affected.

Inefficient algorithmic complexity in the JSON reader of Progressive Robot hMailServer 6.3.6 and 6.3.7 allows a remote attacker to hold the server's mail delivery threads or its REST API's worker threads, making mail delivery, the REST API and the webmail unavailable. 6.3.6 fixed CVE-2026-107574, in which reading a JSON object searched the member names already read for each new name, by keeping the names in a hash set; the set uses the standard library's hash, which takes no secret, so member names chosen to fall in one bucket of it still cost time growing with the square of their number. Against the released reader, one such object of 16 MB took about 5 seconds to read in the Linux build, and one of under 6 MB about 23 seconds in the Windows build, where objects of the same sizes with other names took under a tenth of a second. A remote unauthenticated attacker reaches the reader by mailing such an object as a TLS-RPT report, up to 16 MB once decompressed, to a hosted domain's report mailbox, where an administrator has set one; it is read on a mail delivery thread (ten by default), and those threads deliver all mail, local and outbound, so a stream of such reports stops all delivery for as long as it continues. It is also reached without an account through the reply to a password-protected message, with the message's link alone, where the REST API is on (it is off by default) and a public base URL for such links is set; and by any account through the webmail's routes granted 16 MB, and by an application's API key through the send API. Each such request holds one of the REST API's four worker threads, in the process that also serves SMTP, IMAP and POP3.

Incorrect default permissions in the Windows installer of Progressive Robot hMailServer 6.3.7 allow a local authenticated user to run code as the administrator who runs setup, when the server is installed into a program folder that is already there and that accounts other than the administrators could write, such as a folder made beforehand at the root of a drive, or when such an installation is upgraded or repaired. 6.3.7 fixed CVE-2026-107571 by giving the program folder, Bin and ControlPanel permissions that only administrators may change, but for a folder that is already there it does so only at the end of setup: setup first copies the programs into Bin, which every user may still write, and runs programs from Bin with the administrator's elevated rights, registering the service among them, before that hardening, so that a library another account places in Bin meanwhile is loaded in place of the system's. A handle an account opened on such a folder while it could write it keeps that access after the hardening. The default installation under Program Files, whose folders give ordinary users read and execute only, is not affected.

Allocation of resources without limits in the JSON reader of Progressive Robot hMailServer 6.3.2 through 6.3.7 allows a remote attacker to exhaust the server's memory and hold its threads, making mail delivery, the REST API and the webmail unavailable. The reader built a tree of values far larger than the text it read, copying each value into every level above it, and set no bound on the number of values, so a single JSON document of many small values, within the 16 MB the server reads, raised the service's memory by about 2 GB and held one thread for seconds to minutes. From 6.3.4 a remote unauthenticated attacker reaches the reader by mailing such a document as a TLS-RPT report, up to 16 MB once decompressed, to a hosted domain's report mailbox, where an administrator has set one; it is read on a mail delivery thread (ten by default), and those threads deliver all mail, local and outbound. From 6.3.4 it is also reached without an account through the reply to a password-protected message, with the message's link alone, where the REST API is on (it is off by default) and a public base URL for such links is set; and by any account, from 6.3.2 through the webmail's send and draft routes and from 6.3.4 through its other routes granted 16 MB and the send API; the webmail's send, draft and settings routes parsed the body again for each member read. The REST API serves every request on four worker threads, in the process that also serves SMTP, IMAP and POP3, so a few such documents at once make it unavailable and can exhaust the machine's memory and end the whole service. This is a separate defect from CVE-2026-107574, the reader's handling of duplicate member names, fixed in 6.3.6.

Improper link resolution in the Windows installer of Progressive Robot hMailServer 6.0.0 through 6.3.7 allows a local authenticated user to run code as the mail server's service account or as an administrator when the path to the program folder passes through a junction, a symbolic link or a mount point that the user controls or that every user may change, such as a junction made at the root of a drive, which takes the root's permissions, or through a folder that every user may rename. Setup installed the programs through any such link without first checking who controls it, or whether a folder above the program folder may be renamed, and the service is registered to run the programs by that path, so the programs are installed where a link another account controls leads, or that account can later make the path lead to a folder it controls; the service, which runs as LocalSystem or as its virtual account, and the programs administrators run from Bin are then started from there. 6.3.7's fix for CVE-2026-107571 does not cover this: its hardening of the program folder, Bin and ControlPanel does not change a folder reached through a link that the administrators do not alone control, and setup does not refuse such a path: it installs through the link and at most reports afterwards that the folders could not be hardened. The default installation under Program Files is not affected.

Progressive Robot hMailServer 6.3.4 through 6.3.7, with a MySQL or MariaDB database and TlsMode=verify-full, does not hand the database client the CA file that TlsCaFile names when the file's name contains a character outside ASCII on Linux, or above U+00FF on Windows, which allows an attacker who can intercept the connection to the database server to read and modify the traffic between hMailServer and its database, by presenting, with MariaDB Connector/C 3.4.2 to 3.4.10 as the client, a certificate from any authority, its own included, for any name, and otherwise a certificate for the database server's name issued by any certificate authority in the system's trust store. The server converted the name for the client with the C runtime's conversion in the C locale, which gives an empty string for such a name, and then set no CA file at all, without an error, while certificate verification stayed on, so the client - MariaDB Connector/C, bundled on Windows and the distribution's on Linux - verified the certificate of a database server over the network against the system's trust store instead of the configured CA; Connector/C 3.4.2 to 3.4.10 (bundled on Windows in 6.3.4 to 6.3.6) went further and, with no CA file, took a certificate within its dates whose issuer it could not find, for any name, with neither its chain nor its name checked and no password proof, when the server offered first a sign-in method that hashes the password, which an attacker in the middle chooses (MariaDB's CONC-853, fixed in Connector/C 3.4.11), and Connector/C 3.4 accepted from a MariaDB 11.4 or later server a certificate the store did not vouch for when that server proved it held the account's password; on a connection it treats as local (127.0.0.1, ::1, a Unix socket, a named pipe, and localhost on Windows), Connector/C 3.4.1 and later checked neither the certificate's issuer nor its name, so that a local attacker able to stand in for the database server could present any certificate within its dates of its own making. The settings line of hmailserver --check-database still named the configured file, and the application log said the certificate had been checked. With MariaDB Connector/C 3.4.1 to 3.4.9 as the client - bundled on Windows in 6.3.4 to 6.3.6, and the client of Linux distributions that have not patched it - the missing CA file also met the precondition of CVE-2026-107819, so that a certificate issued for any other name by a trusted authority was enough to obtain the database password. On Linux the file was found only when the server ran on glibc under a locale whose character set holds the name's characters (a UTF-8 locale, or a legacy 8-bit one the system has, in whose bytes the file is then named); with no locale, LCALL=C or a locale the system lacks, and on Alpine (musl), the connection failed instead. The configuration is not the default: TlsMode defaults to prefer, which checks no certificate.

Progressive Robot hMailServer 6.3.4 through 6.3.7, with a MySQL or MariaDB database, TlsMode=verify-full and no TlsCaFile, does not verify the database server's certificate's issuer or name when the connection is one that MariaDB Connector/C treats as local - Server= set to 127.0.0.1, ::1 or localhost (on Linux a Unix socket), to . on Windows (a named pipe), or to a comma-separated list of servers that holds such an element - which allows a local attacker who can stand in for the database server, for example by listening on its port while it is stopped or restarting, to receive the database sign-in, the password included, and to read and modify the traffic between hMailServer and its database. hMailServer applied verify-full by turning the client's certificate verification on without a CA file, but MariaDB Connector/C 3.4.1 and later, on a connection they treat as local and with no CA file, accept any certificate within its validity period that is self-signed or sent with its own root, for any name (Connector/C 3.4.2 to 3.4.10 also one whose issuer they cannot find), so the client accepted the certificate an impostor made, asked no proof of the password and allowed a switch to an authentication method that sends the password unhashed, while hmailserver --check-database said the certificate had been checked against the system's trust store and the application log that it had been checked. The Windows builds bundle Connector/C 3.4.9 (6.3.4 to 6.3.6) and 3.4.11 (6.3.7); on Linux the distribution's client is used, and Connector/C 3.3 and 3.4.0 and Oracle's MySQL client are not affected. The configuration is not the default: TlsMode defaults to prefer, which checks no certificate.

Progressive Robot hMailServer 6.2.25 through 6.3.7 on Windows, with a PostgreSQL database, the connection's sslmode require, verify-ca or verify-full, and a root certificate file named by PostgreSQLSslRootCert (from 6.2.25) or, with TlsMode=verify-full and neither PostgreSQL key set, by TlsCaFile (from 6.3.4), hands the file's name to libpq as bytes in the system's ANSI code page (up to 6.3.5 as Latin-1 bytes, which on code page 1252 are the code page's own for U+00A0 to U+00FF), and libpq looks for the file by those bytes and then hands the same bytes to OpenSSL, whose file opening on Windows reads a name as UTF-8 first and opens the file at that reading when one is there. When every byte of the name outside ASCII also reads as UTF-8 - for example U+00C3 U+00A9, the bytes C3 A9 on code page 1252, which are U+00E9 in UTF-8 - the database server's certificate was checked against the file at that other name whenever one was there, which allows a local attacker who can create that file but cannot change the configured one, as any user can by default beside a file an administrator keeps under ProgramData, or beside a folder directly under the system drive's root whose permissions the administrator restricted, and who can intercept the connection to the database server or stand in for it, to have hMailServer trust a certificate authority of the attacker's own and read and modify the traffic between hMailServer and its database, the database password included; the file alone, holding no authority that issued the database server's certificate, stops hMailServer from connecting to its database. The same applies in 6.3.4 through 6.3.7 to TlsCaFile when an Oracle MySQL client library has been put in place of the bundled MariaDB Connector/C, since that client opens the file through OpenSSL in the same way; the bundled MariaDB Connector/C opens the file in the code page and is not affected, and neither are the Linux builds. The configuration is not the default: by default no root certificate file is named.

Incorrect authorization in the COM administration API of Progressive Robot hMailServer 6.3.4 through 6.3.6 allows an account assigned the serveradmin role with a list of domains - a credential an administrator may create meaning to confine it to those domains - to act as the whole-server administrator over COM. Administrator roles are meant to confine such a credential to the domains its list names, and the REST API did not take it for the server administrator, but the COM API's server-administrator check read the administerserver permission as the whole server whatever list the grant carried. So the credential signed in through the Control Panel, or through a script run on the server, as the server administrator of every domain: it could read and change every domain's accounts and mail, rename any domain, map a domain alias onto any, write the server's settings and secrets, and create further server-administrator accounts. The COM API's domain-administrator check, asked with no domain, likewise let an account assigned the domainadmin role with a list of domains that leaves out the account's own domain administer that domain's ordinary accounts over COM. An account holding serveradmin with no list is the server administrator by design and is unaffected, as are the help desk, auditor and custom roles.

Missing authorization in the REST and COM administration APIs of Progressive Robot hMailServer 6.0.0 through 6.3.6 allows a remote authenticated user whose credential is confined to named domains to change mail routing and domain configuration beyond those domains on a multi-tenant server. A domain administrator, an account or API key given another role that manages accounts in named domains, or a full-scope API key issued for named domains could create a domain alias for any name (so that other tenants' mail for that name is diverted to an address the attacker controls, or denied), rename a domain to any name the server does not host, set a rule's bind-to-address, send-using-route or script-function action to reach one of the server's own addresses, a configured route's stored credentials or a function of the server's event script, and set the domain's relay host - which delivery then connects to with no address rule - to an address on the server's own network; and over the COM API any account could set the three rule actions on its own rules, where the script-function action runs at delivery. The authorization gate did not refuse these routes to a confined credential and the handlers were not given the caller; over COM the domain-alias save performed no authorization check at all, the domain rename required only a domain administrator, and a rule save required at most that the caller could change the rule's owner.

Not failing securely in the anti-virus attachment-stripping action of Progressive Robot hMailServer 6.0.0 through 6.3.6 allows a remote unauthenticated attacker to have a message that the server's virus scanner found infected delivered to the recipient's mailbox unchanged. With virus scanning enabled and its action set to remove attachments and deliver the text (the default action deletes the message), the server keeps one text part of an infected message and replaces every attachment with a note. A message whose single part is not text - a program or an archive sent as the whole message, a message carried whole, a file declared an attachment, or an S/MIME message signed as one part, whose content the recipient's mail client opens - had no part the action could pick out, and was delivered as it arrived, with the content the scanner had found and nothing but a line in the application log to show it. Any sender can write such a message.

Insufficient preservation of the original sender in the forwarding of Progressive Robot hMailServer 6.0.0 through 6.3.6 allows a remote unauthenticated attacker to have mail they write, naming in its From an address at a domain the server hosts, delivered to other receivers as passing that domain's DMARC policy. When the server sent on a copy of mail that came from outside - an account's forward, a mail rule's forward, a Sieve redirect, a list's copy, or an alias or list expanding to an outside address - it could give the copy an SMTP identity in the organisation of the domain the copy's From names, whatever the message's origin: with the Sender Rewriting Scheme or the rewrite of the envelope sender on forwarding enabled (both off by default), an address at the forwarding domain; for a list, its bounce address; for a message with an empty envelope sender, none, under the server's own host name; and an envelope sender the message came with at a subdomain of a hosted domain was kept. Where that identity's SPF record authorises the server, other receivers' SPF check of the copy passed for a domain aligned with the From, and DMARC passed with it, although the message had not passed DMARC when the server received it. Exploitation needs a forward, redirect, alias or list that carries the hosted domain's mail to an outside address, and one of: SRS or the envelope rewrite enabled; a list whose bounce address is in the domain's organisation; an envelope sender at a subdomain whose SPF record authorises the server; or, for an empty envelope sender, an SPF record that authorises the server under its host name. The DKIM route to the same result is CVE-2026-107567.

Missing authorization in the REST metrics route of Progressive Robot hMailServer 6.2.25 through 6.3.6 allows a remote authenticated user whose credential is confined to named domains to read the server-wide traffic history. GET /api/v1/metrics/history returns the server's traffic counters over the last 30 days; those counters are server-wide and carry no domain, which is why the report routes already refuse the same counters to a credential confined to named domains. The metrics route was not among the routes the server's authorization gate refuses a confined credential and its handler was not given the caller, so a domain administrator, an account given another role that holds viewconfiguration in named domains (the auditor role given a domain list among them), or an API key issued for named domains could read the whole server's traffic totals rather than its own domains'. The disclosure is of aggregate operational counters, not of any message or account.

Incorrect authorization in the REST administration API of Progressive Robot hMailServer 6.3.4 through 6.3.6 allows a remote authenticated user whose credential holds the manageroles permission in named domains - a full-scope API key issued for those domains, or an account assigned the serveradmin role with a list of domains - to obtain the server administrator's authority. An account's admin level gives it authority, and a role assigned to it replaces that authority, which is how an administrator narrows an account whose admin level is the server administrator's. The route that takes a role away (DELETE /api/v1/role-assignments/{address}) checked only that the credential takes in what the account held at the time, not what it would hold afterwards, so such a credential could take a narrowing role away from an account in its domains whose admin level is the server administrator's, returning it to the server administrator's authority, and, having set that account's password while it held the narrower role, sign in as the server administrator.

Missing authorization in the REST role-management routes of Progressive Robot hMailServer 6.3.4 through 6.3.6 allows a remote authenticated user whose credential is confined to named domains to modify a custom role that accounts in other domains hold, and to create custom roles and delete any that no account holds. The routes that create, update and delete a custom role require the manageroles permission, which only the serveradmin role holds; but a full-scope API key issued for named domains, and serveradmin assigned with a domain list, are serveradmin confined to those domains and hold manageroles. The server's authorization gate did not refuse the three role routes to a confined credential and the route named no domain, so a credential issued to one tenant could raise or lower the permissions of a custom role assigned across other tenants - granting another domain's role every permission over its mailboxes and settings, or removing what it needs. A custom role may not hold administerserver, administerdomain or manageroles, so the confined credential cannot make itself or another a server administrator by creating or rewriting a custom role.

Use of a less trusted source for the envelope sender in the external-account download of Progressive Robot hMailServer 6.0.0 through 6.3.6 allows a remote unauthenticated attacker to have mail they write in the name of a domain the server hosts sent on by the server with an envelope sender at that domain, so that it passes the domain's SPF and DMARC at other receivers. A message the server downloads from an external POP3 or IMAP account has no SMTP envelope of its own, and the server took the address in its From header as its envelope sender, whatever that address was. A forward or redirect of the downloaded message to an outside address then left with that address as its envelope sender, which the Sender Rewriting Scheme leaves alone for a hosted domain, so that the receiver's SPF check passed for the forged domain, aligned even under strict alignment, with no non-default setting, host name record or DMARC policy needed. An attacker needs to know a mailbox elsewhere that an account here collects and forwards to an outside address; an account holder who adds an external account of their own can collect from a server of their own, which matters where SmtpAuthenticatedSenderCheck holds signed-in senders to their own addresses.

Incorrect authorization in the Sender Rewriting Scheme (SRS) of Progressive Robot hMailServer 6.2.1 through 6.3.6 allows a remote attacker to relay mail through the server to outside addresses. With SRS enabled, the server rewrites the envelope sender of mail it forwards to a signed SRS address at the forwarding domain, and passes a bounce to that address back to the original sender; but it took every message to a valid SRS address, whatever its sender, as delivery to a local recipient, so the relay rules and the SMTP authentication they require were not applied, and the message was delivered to the outside address the SRS address encodes. Anyone holding a valid SRS address of the server, such as the recipient of a forwarded message, could have the server relay mail to the address it encodes for 21 days, and an account holder whose account forwards to an outside mailbox could obtain one for any address. SRS is off by default.

Insufficient resource pool in the ACME HTTP-01 challenge responder of Progressive Robot hMailServer 6.0.0 through 6.3.6 allows a remote unauthenticated attacker to cause a certificate issuance or renewal to fail. When the server obtains or renews a TLS certificate over ACME it runs an HTTP-01 challenge responder on port 80 for the certificate authority to read the challenge token; the responder served one connection at a time, accepting a socket, waiting up to five seconds to receive a request, answering it, and only then accepting the next. A client that connected and sent nothing held the single worker for the whole timeout, so enough idle connections during a validation keep the certificate authority's own request from being answered in time and the validation fails. Sustained across a certificate's renewals this leads to the certificate expiring, after which clients refuse the server's TLS; the mail service itself continues on its current certificate until then. No authentication is required to connect to the responder. Version 6.3.7 serves up to 64 connections at once and closes a connection past that, so that where a handful of idle connections, one after another, sufficed before, a client must keep 64 idle connections open at once throughout a validation to have the certificate authority's connection closed.

Insertion of sensitive information into a log file in the SMTP server of Progressive Robot hMailServer 6.0.0 through 6.3.6 allows a local user who can read the server's log, or anyone given a copy of it, to obtain credentials. With SMTP logging enabled, the server hid the credential of an AUTH PLAIN command only when it was sent after EHLO in one exact form, and wrote the rest of the SASL exchange as the client sent it: a PLAIN password sent before EHLO or with extra spacing, the OAuth 2.0 bearer token of AUTH XOAUTH2 and AUTH OAUTHBEARER (from 6.2.0), the SCRAM-SHA-256 client proof (from 6.2.0), which with the logged salt and iteration count allows offline password guessing, and the GSSAPI authenticator (from 6.3.2). SMTP logging is off by default; on Windows before 6.3.6 the log folder was readable by every local user.

Cross-site request forgery in the REST API of Progressive Robot hMailServer 6.0.0 through 6.3.6 allows a remote attacker to make an authenticated user's browser perform a state-changing request. The REST API accepts HTTP Basic authentication and answers a browser with a Basic sign-in box; a browser caches the credential once the box is answered and sends it with later requests to that origin, including a top-level form post from another site, and the server reads a request body whatever its declared content type. The cross-site protection covered only the cookie session (a header a cross-site form cannot set, and a SameSite=Strict cookie); HTTP Basic was not held to it. A page under an attacker's control, visited by a user whose browser holds a cached Basic credential for the REST API - typically an administrator who signed in through the box - could therefore cause a state-changing request such as creating a domain, an account or an API key, or changing a setting, with the victim's credential. The attacker cannot read the cross-origin response, so the forgery is blind.

Incorrect default permissions in the Windows installer of Progressive Robot hMailServer 6.0.0 through 6.3.6 allow a local authenticated user to run code as the mail server's high-privileged accounts when the product is installed in a folder whose permissions are inherited from a writable parent, such as the root of a drive. The installer places the server's programs in the Bin folder of the program folder and gives neither the program folder nor Bin any permissions of their own, so they inherit the parent's. Under the default location, Program Files, ordinary users receive only read and execute and the configuration is not affected; but the installer accepts any path without warning, and a program folder at the root of an NTFS volume inherits write access for a broad group - Authenticated Users (Modify) on a Windows client, Users (create files and folders) on Windows Server. A local user without administrator rights can then alter or replace the service program, which runs as LocalSystem or as the service's virtual account, or the programs an administrator runs elevated from Bin during setup and maintenance, or place a library in Bin that those programs load in place of the system copy, gaining code execution with those privileges. The default Program Files installation is not affected.