redirect router: taint-enforce filenames

[exim.git] / doc / doc-docbook / spec.xfpt

diff --git a/doc/doc-docbook/spec.xfpt b/doc/doc-docbook/spec.xfpt
index 4d02bdc32e8c339ae29431a5fabfd913fa8a34ac..1d6fa536b45352f7fdb1fe31acc71a2d5a45b3f4 100644 (file)
--- a/doc/doc-docbook/spec.xfpt
+++ b/doc/doc-docbook/spec.xfpt
@@ -6362,7 +6362,7 @@ All other options are defaulted.
 .code
 local_delivery:
   driver = appendfile
-  file = /var/mail/$local_part
+  file = /var/mail/$local_part_verified
   delivery_date_add
   envelope_to_add
   return_path_add
@@ -6370,7 +6370,17 @@ local_delivery:
 # mode = 0660
 .endd
 This &(appendfile)& transport is used for local delivery to user mailboxes in
-traditional BSD mailbox format. By default it runs under the uid and gid of the
+traditional BSD mailbox format.
+
+.new
+We prefer to avoid using &$local_part$& directly to define the mailbox filename,
+as it is provided by a potential bad actor.
+Instead we use &$local_part_verified$&,
+the result of looking up &$local_part$& in the user database
+(done by using &%check_local_user%& in the the router).
+.wen
+
+By default &(appendfile)& runs under the uid and gid of the
 local user, which requires the sticky bit to be set on the &_/var/mail_&
 directory. Some systems use the alternative approach of running mail deliveries
 under a particular group instead of using the sticky bit. The commented options
@@ -9815,6 +9825,12 @@ newline at the very end. For the &%header%& and &%bheader%& expansion, for
 those headers that contain lists of addresses, a comma is also inserted at the
 junctions between headers. This does not happen for the &%rheader%& expansion.
 
+.new
+.cindex "tainted data"
+When the headers are from an incoming message,
+the result of expanding any of these variables is tainted.
+.wen
+
 
 .vitem &*${hmac{*&<&'hashname'&>&*}{*&<&'secret'&>&*}{*&<&'string'&>&*}}*&
 .cindex "expansion" "hmac hashing"
@@ -12192,6 +12208,13 @@ When the &%smtp_etrn_command%& option is being expanded, &$domain$& contains
 the complete argument of the ETRN command (see section &<<SECTETRN>>&).
 .endlist
 
+.new
+.cindex "tainted data"
+If the origin of the data is an incoming message,
+the result of expanding this variable is tainted.
+See also &$domain_verified$&.
+.wen
+
 
 .vitem &$domain_data$&
 .vindex "&$domain_data$&"
@@ -12385,6 +12408,25 @@ Global address rewriting happens when a message is received, so the value of
 because a message may have many recipients and the system filter is called just
 once.
 
+.new
+.cindex "tainted data"
+If the origin of the data is an incoming message,
+the result of expanding this variable is tainted.
+
+&*Warning*&: the content of this variable is usually provided by a potential
+attacker.
+Consider carefully the implications of using it unvalidated as a name
+for file access.
+This presents issues for users' &_.forward_& and filter files.
+For traditional full user accounts, use &%check_local_users%& and the
+&$local_part_verified$& variable rather than this one.
+For virtual users, store a suitable pathname component in the database
+which is used for account name validation, and use that retrieved value
+rather than this variable.
+If needed, use a router &%address_data%& or &%set%& option for
+the retrieved data.
+.wen
+
 .vindex "&$local_part_prefix$&"
 .vindex "&$local_part_suffix$&"
 .cindex affix variables
@@ -12451,6 +12493,14 @@ When an address is being routed or delivered, and a
 specific suffix for the local part was recognized, it is available in this
 variable, having been removed from &$local_part$&.
 
+.new
+.vitem &$local_part_verified$&
+.vindex "&$local_part_verified$&"
+If the router generic option &%check_local_part%& has run successfully,
+this variable has the user database version of &$local_part$&.
+Such values are not tainted and hence usable for building file names.
+.wen
+
 .vitem &$local_scan_data$&
 .vindex "&$local_scan_data$&"
 This variable contains the text returned by the &[local_scan()]& function when
@@ -20528,6 +20578,19 @@ is not the case when the file contains syntactically valid items that happen to
 yield empty addresses, for example, items containing only RFC 2822 address
 comments.
 
+.new
+.cindex "tainted data" "in filenames"
+.cindex redirect "tainted data"
+Tainted data may not be used for a filename.
+
+&*Warning*&: It is unwise to use &$local_part$& or &$domain$&
+directly for redirection,
+as they are provided by a potential attacker.
+In the examples above, &$local_part$& is used for looking up data held locally
+on the system, and not used directly (the second example derives &$home$& via
+the passsword file or database, using &$local_part$&).
+.wen
+
 
 
 .section "Forward files and address verification" "SECID125"
@@ -20753,6 +20816,11 @@ It must be given as
 .code
 list1:   :include:/opt/lists/list1
 .endd
+.new
+.cindex "tainted data" "in filenames"
+.cindex redirect "tainted data"
+Tainted data may not be used for a filename.
+.wen
 .next
 .cindex "address redirection" "to black hole"
 .cindex "delivery" "discard"
@@ -22312,6 +22380,14 @@ If &%file%& or &%directory%& is set for a delivery from a redirection, it is
 used to determine the file or directory name for the delivery. Normally, the
 contents of &$address_file$& are used in some way in the string expansion.
 .endlist
+.new
+.cindex "tainted data" "in filenames"
+.cindex appendfile "tainted data"
+Tainted data may not be used for a file or directory name.
+This means that, for instance, &$local_part$& cannot be used directly
+as a component of a path.  It can however be used as the key for a lookup
+which returns a path (or component).
+.wen
 
 
 .cindex "Sieve filter" "configuring &(appendfile)&"
@@ -23749,6 +23825,12 @@ directories are also controllable. See chapter &<<CHAPenvironment>>& for
 details of the local delivery environment and chapter &<<CHAPbatching>>&
 for a discussion of local delivery batching.
 
+.new
+.cindex "tainted data" "in pipe command"
+.cindex pipe "tainted data"
+Tainted data may not be used for the command name.
+.wen
+
 
 .section "Concurrent delivery" "SECID140"
 If two messages arrive at almost the same time, and both are routed to a pipe