NAME

pfpacket filter

SYNOPSIS

device pf
options PF_DEFAULT_TO_DROP

DESCRIPTION

Packet filtering takes place in the kernel. A pseudo-device, /dev/pf, allows userland processes to control the behavior of the packet filter through an ioctl(2) interface. There are commands to enable and disable the filter, load rulesets, add and remove individual rules or state table entries, and retrieve statistics. The most commonly used functions are covered by pfctl(8).
Manipulations like loading a ruleset that involve more than a single ioctl(2) call require a so-called ticket, which prevents the occurrence of multiple concurrent manipulations.
Fields of ioctl(2) parameter structures that refer to packet data (like addresses and ports) are generally expected in network byte-order.
Rules and address tables are contained in so-called anchors. When servicing an ioctl(2) request, if the anchor field of the argument structure is empty, the kernel will use the default anchor (i.e., the main ruleset) in operations. Anchors are specified by name and may be nested, with components separated by ‘/’ characters, similar to how file system hierarchies are laid out. The final component of the anchor path is the anchor under which operations will be performed.

SYSCTL VARIABLES AND LOADER TUNABLES

The following loader(8) tunables are available.
net.pf.states_hashsize
Size of hash tables that store states. Should be power of 2. Default value is 131072.
net.pf.source_nodes_hashsize
Size of hash table that store source nodes. Should be power of 2. Default value is 32768.
Read only sysctl(8) variables with matching names are provided to obtain current values at runtime.

KERNEL OPTIONS

The following options in the kernel configuration file are related to pf operation:
PF_DEFAULT_TO_DROP
Change default policy to drop by default

IOCTL INTERFACE

pf supports the following ioctl(2) commands, available through ⟨net/pfvar.h⟩:
DIOCSTART
Start the packet filter.
DIOCSTOP
Stop the packet filter.
DIOCSTARTALTQ
Start the ALTQ bandwidth control system (see altq(9)).
DIOCSTOPALTQ
Stop the ALTQ bandwidth control system.
DIOCBEGINADDRS struct pfioc_pooladdr *pp
Clear the buffer address pool and get a ticket for subsequent DIOCADDADDR, DIOCADDRULE, and DIOCCHANGERULE calls.
DIOCADDADDR struct pfioc_pooladdr *pp
Add the pool address addr to the buffer address pool to be used in the following DIOCADDRULE or DIOCCHANGERULE call. All other members of the structure are ignored.
DIOCADDRULE struct pfioc_rule *pr
Add rule at the end of the inactive ruleset. This call requires a ticket obtained through a preceding DIOCXBEGIN call and a pool_ticket obtained through a DIOCBEGINADDRS call. DIOCADDADDR must also be called if any pool addresses are required. The optional anchor name indicates the anchor in which to append the rule. nr and action are ignored.
DIOCADDALTQ struct pfioc_altq *pa
Add an ALTQ discipline or queue.
DIOCGETRULES struct pfioc_rule *pr
Get a ticket for subsequent DIOCGETRULE calls and the number nr of rules in the active ruleset.
DIOCGETRULE struct pfioc_rule *pr
Get a rule by its number nr using the ticket obtained through a preceding DIOCGETRULES call. If action is set to PF_GET_CLR_CNTR, the per-rule statistics on the requested rule are cleared.
DIOCGETADDRS struct pfioc_pooladdr *pp
Get a ticket for subsequent DIOCGETADDR calls and the number nr of pool addresses in the rule specified with r_action, r_num, and anchor.
DIOCGETADDR struct pfioc_pooladdr *pp
Get the pool address addr by its number nr from the rule specified with r_action, r_num, and anchor using the ticket obtained through a preceding DIOCGETADDRS call.
DIOCGETALTQS struct pfioc_altq *pa
Get a ticket for subsequent DIOCGETALTQ calls and the number nr of queues in the active list.
DIOCGETALTQ struct pfioc_altq *pa
Get the queueing discipline altq by its number nr using the ticket obtained through a preceding DIOCGETALTQS call.
DIOCGETQSTATS struct pfioc_qstats *pq
Get the statistics on a queue.
This call fills in a pointer to the buffer of statistics buf, of length nbytes, for the queue specified by nr.
DIOCGETRULESETS struct pfioc_ruleset *pr
Get the number nr of rulesets (i.e., anchors) directly attached to the anchor named by path for use in subsequent DIOCGETRULESET calls. Nested anchors, since they are not directly attached to the given anchor, will not be included. This ioctl returns EINVAL if the given anchor does not exist.
DIOCGETRULESET struct pfioc_ruleset *pr
Get a ruleset (i.e., an anchor) name by its number nr from the given anchor path, the maximum number of which can be obtained from a preceding DIOCGETRULESETS call. This ioctl returns EINVAL if the given anchor does not exist or EBUSY if another process is concurrently updating a ruleset.
DIOCADDSTATE struct pfioc_state *ps
Add a state entry.
DIOCGETSTATE struct pfioc_state *ps
Extract the entry identified by the id and creatorid fields of the state structure from the state table.
DIOCKILLSTATES struct pfioc_state_kill *psk
Remove matching entries from the state table. This ioctl returns the number of killed states in psk_killed.
DIOCCLRSTATES struct pfioc_state_kill *psk
Clear all states. It works like DIOCKILLSTATES, but ignores the psk_af, psk_proto, psk_src, and psk_dst fields of the pfioc_state_kill structure.
DIOCSETSTATUSIF struct pfioc_if *pi
Specify the interface for which statistics are accumulated.
DIOCGETSTATUS struct pf_status *s
Get the internal packet filter statistics.
DIOCCLRSTATUS
Clear the internal packet filter statistics.
DIOCNATLOOK struct pfioc_natlook *pnl
Look up a state table entry by source and destination addresses and ports.
DIOCSETDEBUG u_int32_t *level
Set the debug level.
DIOCGETSTATES struct pfioc_states *ps
Get state table entries.
If ps_len is non-zero on entry, as many states as possible that can fit into this size will be copied into the supplied buffer ps_states. On exit, ps_len is always set to the total size required to hold all state table entries (i.e., it is set to sizeof(struct pf_state) * nr).
DIOCCHANGERULE struct pfioc_rule *pcr
Add or remove the rule in the ruleset specified by rule.action.
The type of operation to be performed is indicated by action, which can be any of the following:
ticket must be set to the value obtained with PF_CHANGE_GET_TICKET for all actions except PF_CHANGE_GET_TICKET. pool_ticket must be set to the value obtained with the DIOCBEGINADDRS call for all actions except PF_CHANGE_REMOVE and PF_CHANGE_GET_TICKET. anchor indicates to which anchor the operation applies. nr indicates the rule number against which PF_CHANGE_ADD_BEFORE, PF_CHANGE_ADD_AFTER, or PF_CHANGE_REMOVE actions are applied.
DIOCCHANGEADDR struct pfioc_pooladdr *pca
Add or remove the pool address addr from the rule specified by r_action, r_num, and anchor.
DIOCSETTIMEOUT struct pfioc_tm *pt
Set the state timeout of timeout to seconds. The old value will be placed into seconds. For possible values of timeout, consult the PFTM_* values in ⟨net/pfvar.h⟩.
DIOCGETTIMEOUT struct pfioc_tm *pt
Get the state timeout of timeout. The value will be placed into the seconds field.
DIOCCLRRULECTRS
Clear per-rule statistics.
DIOCSETLIMIT struct pfioc_limit *pl
Set the hard limits on the memory pools used by the packet filter.
DIOCGETLIMIT struct pfioc_limit *pl
Get the hard limit for the memory pool indicated by index.
DIOCRCLRTABLES struct pfioc_table *io
Clear all tables. All the ioctls that manipulate radix tables use the same structure described below. For DIOCRCLRTABLES, pfrio_ndel contains on exit the number of tables deleted.
DIOCRADDTABLES struct pfioc_table *io
Create one or more tables. On entry, pfrio_buffer must point to an array of struct pfr_table containing at least pfrio_size elements. pfrio_esize must be the size of struct pfr_table. On exit, pfrio_nadd contains the number of tables effectively created.
DIOCRDELTABLES struct pfioc_table *io
Delete one or more tables. On entry, pfrio_buffer must point to an array of struct pfr_table containing at least pfrio_size elements. pfrio_esize must be the size of struct pfr_table. On exit, pfrio_ndel contains the number of tables effectively deleted.
DIOCRGETTABLES struct pfioc_table *io
Get the list of all tables. On entry, pfrio_buffer[pfrio_size] contains a valid writeable buffer for pfr_table structures. On exit, pfrio_size contains the number of tables written into the buffer. If the buffer is too small, the kernel does not store anything but just returns the required buffer size, without error.
DIOCRGETTSTATS struct pfioc_table *io
This call is like DIOCRGETTABLES but is used to get an array of pfr_tstats structures.
DIOCRCLRTSTATS struct pfioc_table *io
Clear the statistics of one or more tables. On entry, pfrio_buffer must point to an array of struct pfr_table containing at least pfrio_size elements. pfrio_esize must be the size of struct pfr_table. On exit, pfrio_nzero contains the number of tables effectively cleared.
DIOCRCLRADDRS struct pfioc_table *io
Clear all addresses in a table. On entry, pfrio_table contains the table to clear. On exit, pfrio_ndel contains the number of addresses removed.
DIOCRADDADDRS struct pfioc_table *io
Add one or more addresses to a table. On entry, pfrio_table contains the table ID and pfrio_buffer must point to an array of struct pfr_addr containing at least pfrio_size elements to add to the table. pfrio_esize must be the size of struct pfr_addr. On exit, pfrio_nadd contains the number of addresses effectively added.
DIOCRDELADDRS struct pfioc_table *io
Delete one or more addresses from a table. On entry, pfrio_table contains the table ID and pfrio_buffer must point to an array of struct pfr_addr containing at least pfrio_size elements to delete from the table. pfrio_esize must be the size of struct pfr_addr. On exit, pfrio_ndel contains the number of addresses effectively deleted.
DIOCRSETADDRS struct pfioc_table *io
Replace the content of a table by a new address list. This is the most complicated command, which uses all the structure members.
On entry, pfrio_table contains the table ID and pfrio_buffer must point to an array of struct pfr_addr containing at least pfrio_size elements which become the new contents of the table. pfrio_esize must be the size of struct pfr_addr. Additionally, if pfrio_size2 is non-zero, pfrio_buffer[pfrio_size..pfrio_size2] must be a writeable buffer, into which the kernel can copy the addresses that have been deleted during the replace operation. On exit, pfrio_ndel, pfrio_nadd, and pfrio_nchange contain the number of addresses deleted, added, and changed by the kernel. If pfrio_size2 was set on entry, pfrio_size2 will point to the size of the buffer used, exactly like DIOCRGETADDRS.
DIOCRGETADDRS struct pfioc_table *io
Get all the addresses of a table. On entry, pfrio_table contains the table ID and pfrio_buffer[pfrio_size] contains a valid writeable buffer for pfr_addr structures. On exit, pfrio_size contains the number of addresses written into the buffer. If the buffer was too small, the kernel does not store anything but just returns the required buffer size, without returning an error.
DIOCRGETASTATS struct pfioc_table *io
This call is like DIOCRGETADDRS but is used to get an array of pfr_astats structures.
DIOCRCLRASTATS struct pfioc_table *io
Clear the statistics of one or more addresses. On entry, pfrio_table contains the table ID and pfrio_buffer must point to an array of struct pfr_addr containing at least pfrio_size elements to be cleared from the table. pfrio_esize must be the size of struct pfr_addr. On exit, pfrio_nzero contains the number of addresses effectively cleared.
DIOCRTSTADDRS struct pfioc_table *io
Test if the given addresses match a table. On entry, pfrio_table contains the table ID and pfrio_buffer must point to an array of struct pfr_addr containing at least pfrio_size elements, each of which will be tested for a match in the table. pfrio_esize must be the size of struct pfr_addr. On exit, the kernel updates the pfr_addr array by setting the pfra_fback member appropriately.
DIOCRSETTFLAGS struct pfioc_table *io
Change the PFR_TFLAG_CONST or PFR_TFLAG_PERSIST flags of a table. On entry, pfrio_buffer must point to an array of struct pfr_table containing at least pfrio_size elements. pfrio_esize must be the size of struct pfr_table. pfrio_setflag must contain the flags to add, while pfrio_clrflag must contain the flags to remove. On exit, pfrio_nchange and pfrio_ndel contain the number of tables altered or deleted by the kernel. Yes, tables can be deleted if one removes the PFR_TFLAG_PERSIST flag of an unreferenced table.
DIOCRINADEFINE struct pfioc_table *io
Defines a table in the inactive set. On entry, pfrio_table contains the table ID and pfrio_buffer[pfrio_size] contains an array of pfr_addr structures to put in the table. A valid ticket must also be supplied to pfrio_ticket. On exit, pfrio_nadd contains 0 if the table was already defined in the inactive list or 1 if a new table has been created. pfrio_naddr contains the number of addresses effectively put in the table.
DIOCXBEGIN struct pfioc_trans *io
Clear all the inactive rulesets specified in the pfioc_trans_e array. For each ruleset, a ticket is returned for subsequent "add rule" ioctls, as well as for the DIOCXCOMMIT and DIOCXROLLBACK calls.
Ruleset types, identified by rs_num, include the following:
DIOCXCOMMIT struct pfioc_trans *io
Atomically switch a vector of inactive rulesets to the active rulesets. This call is implemented as a standard two-phase commit, which will either fail for all rulesets or completely succeed. All tickets need to be valid. This ioctl returns EBUSY if another process is concurrently updating some of the same rulesets.
DIOCXROLLBACK struct pfioc_trans *io
Clean up the kernel by undoing all changes that have taken place on the inactive rulesets since the last DIOCXBEGIN. DIOCXROLLBACK will silently ignore rulesets for which the ticket is invalid.
DIOCSETHOSTID u_int32_t *hostid
Set the host ID, which is used by pfsync(4) to identify which host created state table entries.
DIOCOSFPFLUSH
Flush the passive OS fingerprint table.
DIOCOSFPADD struct pf_osfp_ioctl *io
Add a passive OS fingerprint to the table. Set fp_os.fp_os to the packed fingerprint, fp_os.fp_class_nm to the name of the class (Linux, Windows, etc), fp_os.fp_version_nm to the name of the version (NT, 95, 98), and fp_os.fp_subtype_nm to the name of the subtype or patchlevel. The members fp_mss, fp_wsize, fp_psize, fp_ttl, fp_optcnt, and fp_wscale are set to the TCP MSS, the TCP window size, the IP length, the IP TTL, the number of TCP options, and the TCP window scaling constant of the TCP SYN packet, respectively.
The fp_flags member is filled according to the ⟨net/pfvar.h⟩ include file PF_OSFP_* defines. The fp_tcpopts member contains packed TCP options. Each option uses PF_OSFP_TCPOPT_BITS bits in the packed value. Options include any of PF_OSFP_TCPOPT_NOP, PF_OSFP_TCPOPT_SACK, PF_OSFP_TCPOPT_WSCALE, PF_OSFP_TCPOPT_MSS, or PF_OSFP_TCPOPT_TS.
The fp_getnum member is not used with this ioctl.
The structure's slack space must be zeroed for correct operation; memset(3) the whole structure to zero before filling and sending to the kernel.
DIOCOSFPGET struct pf_osfp_ioctl *io
Get the passive OS fingerprint number fp_getnum from the kernel's fingerprint list. The rest of the structure members will come back filled. Get the whole list by repeatedly incrementing the fp_getnum number until the ioctl returns EBUSY.
DIOCGETSRCNODES struct pfioc_src_nodes *psn
Get the list of source nodes kept by sticky addresses and source tracking. The ioctl must be called once with psn_len set to 0. If the ioctl returns without error, psn_len will be set to the size of the buffer required to hold all the pf_src_node structures held in the table. A buffer of this size should then be allocated, and a pointer to this buffer placed in psn_buf. The ioctl must then be called again to fill this buffer with the actual source node data. After that call, psn_len will be set to the length of the buffer actually used.
DIOCCLRSRCNODES
Clear the tree of source tracking nodes.
DIOCIGETIFACES struct pfioc_iface *io
Get the list of interfaces and interface drivers known to pf. All the ioctls that manipulate interfaces use the same structure described below:
If not empty, pfiio_name can be used to restrict the search to a specific interface or driver. pfiio_buffer[pfiio_size] is the user-supplied buffer for returning the data. On entry, pfiio_size contains the number of pfi_kif entries that can fit into the buffer. The kernel will replace this value by the real number of entries it wants to return. pfiio_esize should be set to sizeof(struct pfi_kif).
The data is returned in the pfi_kif structure described below:
DIOCSETIFFLAG struct pfioc_iface *io
Set the user settable flags (described above) of the pf internal interface description. The filtering process is the same as for DIOCIGETIFACES.
DIOCCLRIFFLAG struct pfioc_iface *io
Works as DIOCSETIFFLAG above but clears the flags.
DIOCKILLSRCNODES struct pfioc_iface *io
Explicitly remove source tracking nodes.

FILES

/dev/pf
packet filtering device.

EXAMPLES

The following example demonstrates how to use the DIOCNATLOOK command to find the internal host/port of a NATed connection:
#include <sys/types.h> 
#include <sys/socket.h> 
#include <sys/ioctl.h> 
#include <sys/fcntl.h> 
#include <net/if.h> 
#include <netinet/in.h> 
#include <net/pfvar.h> 
#include <err.h> 
#include <stdio.h> 
#include <stdlib.h> 
 
u_int32_t 
read_address(const char *s) 
{ 
	int a, b, c, d; 
 
	sscanf(s, "%i.%i.%i.%i", &a, &b, &c, &d); 
	return htonl(a << 24 | b << 16 | c << 8 | d); 
} 
 
void 
print_address(u_int32_t a) 
{ 
	a = ntohl(a); 
	printf("%d.%d.%d.%d", a >> 24 & 255, a >> 16 & 255, 
	    a >> 8 & 255, a & 255); 
} 
 
int 
main(int argc, char *argv[]) 
{ 
	struct pfioc_natlook nl; 
	int dev; 
 
	if (argc != 5) { 
		printf("%s <gwy addr> <gwy port> <ext addr> <ext port>\n", 
		    argv[0]); 
		return 1; 
	} 
 
	dev = open("/dev/pf", O_RDWR); 
	if (dev == -1) 
		err(1, "open(\"/dev/pf\") failed"); 
 
	memset(&nl, 0, sizeof(struct pfioc_natlook)); 
	nl.saddr.v4.s_addr	= read_address(argv[1]); 
	nl.sport		= htons(atoi(argv[2])); 
	nl.daddr.v4.s_addr	= read_address(argv[3]); 
	nl.dport		= htons(atoi(argv[4])); 
	nl.af			= AF_INET; 
	nl.proto		= IPPROTO_TCP; 
	nl.direction		= PF_IN; 
 
	if (ioctl(dev, DIOCNATLOOK, &nl)) 
		err(1, "DIOCNATLOOK"); 
 
	printf("internal host "); 
	print_address(nl.rsaddr.v4.s_addr); 
	printf(":%u\n", ntohs(nl.rsport)); 
	return 0; 
}

SEE ALSO

ioctl(2), altq(4), if_bridge(4), pflog(4), pfsync(4), pfctl(8), altq(9)

HISTORY

The pf packet filtering mechanism first appeared in OpenBSD 3.0 and then FreeBSD 5.2.
This implementation is derived from OpenBSD 4.5. It has been heavily modified to be capable of running in multithreaded FreeBSD kernel and scale its performance on multiple CPUs.

Recommended readings

Pages related to pf you should read also: