HardenedBSD/share/man/man4/ipfirewall.4

.Dd November 16, 1994
.Dt IPFIREWALL 4
.Os
.Sh NAME
.Nm ipfirewall ,
.Nm ipfw ,
.Nm ipaccounting ,
.Nm ipacct
.Nd IP packet filter and traffic accounting.
.Sh SYNOPSIS
.Fd #include <netinet/ip_fw.h>
.Ft int
.Fn setsockopt raw_socket IPPROTO_IP "ipfw/ipacct  option" "struct ip | struct ipfw" size

Ipfw options:
  IP_FW_ADD_BLK   - add entry to blocking chain. 
  IP_FW_ADD_FWD   - add entry to forwarding chain. 
  IP_FW_CHK_BLK   - check ip packet against blocking chain.
  IP_FW_CHK_FWD   - check ip packet against forwarding chain.
  IP_FW_DEL_BLK   - delete entry from blocking chain.
  IP_FW_DEL_FWD   - delete entry from forwarding chain.
  IP_FW_FLUSH     - flush all blocking & forwarding chain entries.
  IP_FW_POLICY    - define default ipfw policy.

Ipacct options:
  IP_ACCT_ADD     - add entry to accounting chain.
  IP_ACCT_DEL     - delete entry from accounting chain.
  IP_ACCT_FLUSH   - flush all accounting chain entries.
  IP_ACCT_ZERO    - zero all accounting chain entries.

Ipfw/ipacct entry structure:
  #define IP_FW_MAX_PORTS 10             

struct ip_fw {
  struct ip_fw *next;       
  struct in_addr src, dst; 
  struct in_addr src_mask, dst_mask;  
  u_short flags;                     
  u_short n_src_p, n_dst_p;  
  u_short ports[IP_FW_MAX_PORTS];  
  u_long p_cnt,b_cnt;  
}

Flags values for "flags" field:
  IP_FW_F_ALL  	- The entry should match all IP packets. 
  IP_FW_F_TCP     - The entry should match TCP packets.
  IP_FW_F_UDP     - The entry should match UDP packets.
  IP_FW_F_ICMP    - The entry should match ICMP packets.
  IP_FW_F_KIND    - Mask value to separate protocol kind.
  IP_FW_F_ACCEPT  - This entry is accepting ( see below )
  IP_FW_F_SRNG    - Source ports are range ( see below )
  IP_FW_F_DRNG    - Destination ports are range ( see below )
  IP_FW_F_PRN     - Print this entry ( see below )
  IP_FW_F_BIDIR   - This acct entry is bidirectional ( see below )
  IP_FW_F_MASK    - Mask to match all valid flag bits.

Kernel symbols to kvm_nlist():
  struct ip_fw *ip_fw_blk_chain - chain of forwarding entries.
  struct ip_fw *ip_fw_fwd_chain - chain of blocking entries.
  int           ip_fw_policy    - default policy.
  struct ip_fw *ip_acct_chain   - chain of accounting entries.

Options in the kernel configuration file:
  IPFIREWALL	   - enable ipfirewall.
  IPFIREWALL_VERBOSE - enable firewall output ( see below )
  DEBUG_IPFIREWALL   - enable extensive debugging output.
  IPACCT		   - enable ipaccounting.

.Sh DESCRIPTION
Ipfirewall (later ipfw) is a system facility,which allows filtering
of incoming and/or forwarding packets on the protocol+source/destination
address/ports base.
Ipaccounting (later ipacct) is a system facility,which allows counting
of incoming,outgoing and forwarding traffic by packet/byte count.
.Pp
Basic idea is that every packet checked against number of entries
in several chains.  There are 3 chains:
  Blocking - this chain defines whenever packet should be accepted
             ever for local delivery or for forwarding.
  Forwarding - this chain defines whenever packet should be accepted
               for forwarding only.
  Accounting - this chain defines types of packets , which should be

.Pp
Options to add/remove specific entries or to flush all entries described
above. Value passed to 
.Fn setsockopt
is a value of struct ip_fw for
entry. If an entry is added, it checked by such rules that when we start 
searching chain for matching entry the first matching is the best match,
[ or at least one of them :^) ].
 That means:
  * First in chain entries with specific protocol and small ranges
    of src/dst addresses and ports. 
  * Later go entries with wider ranges of ports and addresses.
  * Later entries matching every port for some address range.
  * Later universal entries matching any protocol.
.Pp
While deleting entry, every entry which is equal to that passed to 
.Fn setsockopt
will be removed.  Flush removes all entries.
Each entry has several fields by which packets are matched:


   struct ip_fw *next - next entry in chain.(Set internally)

   struct in_addr src - source address to be matched.
   struct in_addr src_mask  - source address mask.
           To match whole networks/subnets or address groups
           mask bits should be zeroed here and also
           in src_mask field. Valuable bits should be set
           in src_mask field.
   struct in_addr dst - destination address to be matched.
   struct in_addr dst_mask - destination address mask. 

   u_short flags  - flags field.See exact description of flags meaning
                    in description later.

   u_short n_src_p - number of source ports in "ports" array.
   u_short n_dst_p - number of destination ports in "ports" array. 
   u_short ports[] - ports array.Overall length currently defined
                     to reasonable maximum - 10,and could be changed.
                     The packet's src port can ever match one of
                     ports[0] ... ports[--n_src_p] numbers,or if
                     flag IP_FW_F_SRNG set take port[0] as bottom 
                     range value and ports[1] as top one.n_src_p should
                     be set to 2 then.If n_src_p equal to 0 , every port
                     match. The same rules apply to packet's dst port,
                     except that it matched against ports[n_src_p] ...
                     ... ports[n_src_p+n_dst_p--],or if IP_FW_F_DRNG set,
                     range is ports[n_src_p] to ports[n_srcp++].

   u_long p_cnt - packets count for ipacct entries.
   u_long b_cnt - bytes count for ipacct entries.

Packet matching proceeds in the following manner:

a) If packet entry protocol set to ALL, see c).

b) If entry protocol set to TCP/UDP/ICMP and packet protocol 
   different - no match, if packet protocol and entry protocol
   same - continue.
     
c) If source address pattern does not equal to packets sources address
   masked with src_mask, or destination pattern not equal to packets
   destination address masked with dst_mask - no match.
   If they does and protocol set to ALL/ICMP - got match.
   If they does and protocol set to TCP/UDP - continue.

d) If src port doesn't match or dst port doesn't match - all
   packet don't match. If they do - got match.
.Pp
In ipfw packet matched consequently against every chain entry.
Search continues untill first matching entry found.If IP_FW_F_ACCEPT
flag set - packet accepted.  If it is not set - packet denied.
If no matching entry found, all unmatched packets ever accepted or
denied depending on global policy value. It can be set with
IP_FW_POLICY raw socket option. The value for deny is 0, 
and 1 for accept.
.Pp
Entries can be added with IP_FW_F_PRN flag set.If kernel compiled
with IPFIREWALL_VERBOSE option,packets matching this entries will
be printed by kernel printf's.
.Pp
If some chain is empty,every packet accepted by this chain no
matter what default policy is.
.Pp
To check whenever or not packet denied by some chain , checking
options to setsockopt() can be issued. Then the argument is 
a buffer representing ip packet,thus it has to be 
struct ip + struct tcphdr .
Then setsockopt() return value 0 on accept or another on deny.
.Pp
Ipaccounting entries added the same way as ipfw ones.Packet checked
against all entries in chain and values of p_cnt and b_cnt in matching
entries rised.p_cnt rises by 1 and b_cnt by ip_len value of ip packet.
Thus all traffic size counted including IP headers.
.Pp
If IP_FW_F_BIDIR flag is set in accounting entry,packets counted are
those which match entry in standard way along with packets which match
entry while their source and destination addr/port pairs swapped.
.Pp
Zero option allows all accounting to be cleared.
.Sh DIAGNOSTICS

[EINVAL]  The IP option field was improperly formed; an option
          field was shorter than the minimum value or longer than
          the option buffer provided.An structural error in 
          ip_fw structure occurred (n_src_p+n_dst_p too big,
          ports set for ALL/ICMP protocols etc.)
.Sh SEE ALSO
.Xr setsockopt 2 ,
.Xr kvm_nlist 3 ,
.Xr kvm_read 3 ,
.Xr ip 4
.Sh BUGS
The ipfw/ipacct facilities are new and, although serious bugs have
been tracked, some less important ones are expected.
.Pp
This man page is mostly out of date and should be rewritten. 
.Sh HISTORY
 Ipfw facility has been initially written as package to BSDI
by Daniel Boulet <danny@BouletFermat.ab.ca>.
 It has been heavily modified and ported to
.Fx 2.0
by Ugen J.S.Antsilevich <ugen@NetVision.net.il>
 Ipacct facility written for
.Fx 2.0
by Ugen J.S.Antsilevich <ugen@NetVision.net.il>