.\" $OpenBSD: mbuf.9,v 1.32 2008/04/13 14:13:14 jmc Exp $ .\" .\" Copyright (c) 2001 Jean-Jacques Bernard-Gundol .\" All rights reserved. .\" .\" Redistribution and use in source and binary forms, with or without .\" modification, are permitted provided that the following conditions .\" are met: .\" 1. Redistributions of source code must retain the above copyright .\" notice, this list of conditions and the following disclaimer. .\" 2. Redistributions in binary form must reproduce the above copyright .\" notice, this list of conditions and the following disclaimer in the .\" documentation and/or other materials provided with the distribution. .\" 3. The name of the author may not be used to endorse or promote products .\" derived from this software without specific prior written permission .\" .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR .\" IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES .\" OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. .\" IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, .\" INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT .\" NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, .\" DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY .\" THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT .\" (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF .\" THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. .\" .Dd $Mdocdate: April 13 2008 $ .Dt MBUF 9 .Os .Sh NAME .Nm mbuf .Nd Kernel memory management for networking protocols .Sh SYNOPSIS .Fd #include .Ft struct mbuf * .Fn m_copym2 "struct mbuf *m" "int off" "int len" "int wait" .Ft struct mbuf * .Fn m_copym "struct mbuf *m" "int off" "int len" "int wait" .Ft struct mbuf * .Fn m_free "struct mbuf *m" .Fn MFREE "struct mbuf *m" "struct mbuf *n" .Ft struct mbuf * .Fn m_get "int how" "int type" .Fn MGET "struct mbuf *m" "int how" "int type" .Ft struct mbuf * .Fn m_getclr "int how" "int type" .Ft struct mbuf * .Fn m_gethdr "int how" "int type" .Fn MGETHDR "struct mbuf *m" "int how" "int type" .Ft struct mbuf * .Fn m_prepend "struct mbuf *m" "int len" "int how" .Fn M_PREPEND "struct mbuf *m" "int plen" "int how" .Ft struct mbuf * .Fn m_pulldown "struct mbuf *m" "int off" "int len" "int *offp" .Ft struct mbuf * .Fn m_pullup "struct mbuf *n" "int len" .Ft struct mbuf * .Fn m_pullup2 "struct mbuf *n" "int len" .Ft struct mbuf * .Fn m_split "struct mbuf *m0" "int len0" "int wait" .Ft struct mbuf * .Fn m_inject "struct mbuf *m0" "int len0" "int siz" "int wait" .Ft struct mbuf * .Fn m_getptr "struct mbuf *m" "int loc" "int *off" .Ft void .Fn m_adj "struct mbuf *mp" "int req_len" .Ft void .Fn m_copyback "struct mbuf *m0" "int off" "int len" "caddr_t cp" .Ft void .Fn m_freem "struct mbuf *m" .Ft void .Fn m_reclaim "void" .Ft void .Fn m_copydata "struct mbuf *m" "int off" "int len" "caddr_t cp" .Ft void .Fn m_cat "struct mbuf *m" "struct mbuf *n" .Ft struct mbuf * .Fn m_devget "char *buf" "int totlen" "int off" "struct ifnet *ifp" \ "void (*func)(const void *, void *, size_t)" .Ft void .Fn m_zero "struct mbuf *m" .Ft int .Fn m_apply "struct mbuf *m" "int off" "int len" \ "int (*func)(caddr_t, caddr_t, unsigned int)" "caddr_t fstate" .Fn MEXTMALLOC "struct mbuf *m" "int size" "int how" .Fn MCLGET "struct mbuf *m" "int how" .Fn MEXTADD "struct mbuf *m" "caddr_t buf" "u_int size" "int type" \ "void (*free)(caddr_t, u_int, void *)" "void *arg" .Fn M_ALIGN "struct mbuf *m" "int len" .Fn MH_ALIGN "struct mbuf *m" "int len" .Fn M_READONLY "struct mbuf *m" .Fn M_LEADINGSPACE "struct mbuf *m" .Fn M_TRAILINGSPACE "struct mbuf *m" .Fn MCHTYPE "struct mbuf *m" "int type" .Pp .Bd -literal #define MLEN (MSIZE - sizeof(struct m_hdr)) #define MHLEN (MLEN - sizeof(struct pkthdr)) #define MINCLSIZE (MHLEN + 1) #define M_MAXCOMPRESS (MHLEN / 2) #define mtod(m,t) ((t)((m)-\*(Gtm_data)) struct m_hdr { struct mbuf *mh_next; struct mbuf *mh_nextpkt; caddr_t mh_data; u_int mh_len; short mh_type; u_short mh_flags; }; struct pkthdr { struct ifnet *rcvif; SLIST_HEAD(packet_tags, m_tag) tags; int len; int csum_flags; struct pkthdr_pf pf; }; struct pkthdr_pf { void *hdr; u_int rtableid; u_int32_t qid; u_int16_t tag; u_int8_t flags; u_int8_t routed; }; struct m_ext { caddr_t ext_buf; void (*ext_free)(caddr_t, u_int, void *); void *ext_arg; u_int ext_size; int ext_type; struct mbuf *ext_nextref; struct mbuf *ext_prevref; }; struct mbuf { struct m_hdr m_hdr; union { struct { struct pkthdr MH_pkthdr; union { struct m_ext MH_ext; char MH_databuf[MHLEN]; } MH_dat; } MH; char M_databuf[MLEN]; } M_dat; }; #define m_next m_hdr.mh_next #define m_len m_hdr.mh_len #define m_data m_hdr.mh_data #define m_type m_hdr.mh_type #define m_flags m_hdr.mh_flags #define m_nextpkt m_hdr.mh_nextpkt #define m_act m_nextpkt #define m_pkthdr M_dat.MH.MH_pkthdr #define m_ext M_dat.MH.MH_dat.MH_ext #define m_pktdat M_dat.MH.MH_dat.MH_databuf #define m_dat M_dat.M_databuf .Ed .Sh DESCRIPTION The .Nm functions provide a way to manage the memory buffers used by the kernel's networking subsystem. Several functions and macros are used to allocate and deallocate mbufs, but also to get, inject, remove, copy, modify, prepend or append data inside these mbufs. The size of an .Nm is MSIZE .Pq defined in Aq Pa sys/param.h . .Pp An .Nm structure is defined as an .Fa m_hdr structure followed by a union. The header contains the following elements: .Bl -tag -width foobarmoocow .It Fa mh_next A pointer to the next mbuf in the mbuf chain. .It Fa mh_nextpkt A pointer to the next mbuf chain (i.e., packet) in the queue. .It Fa mh_data Indicates the address of the beginning of data in the mbuf. .It Fa mh_len Indicates the amount of data in the mbuf. .It Fa mh_type Indicates the type of data contained in the mbuf (see below). .It Fa mh_flags Flags (see below). .El .Pp The .Fa mh_type variable can take the following values: .Pp .Bl -tag -compact -offset indent -width XXXXXXXXXXXXXXXXXX .It Dv MT_FREE the mbuf should be on the free list. .It Dv MT_DATA the data in the mbuf was dynamically allocated. .It Dv MT_HEADER the data contains a packet header. .It Dv MT_SONAME the data is a socket name. .It Dv MT_SOOPTS the data are socket options. .It Dv MT_FTABLE the data is a fragment reassembly header. .It Dv MT_CONTROL the mbuf contains extra-data protocol message. .It Dv MT_OOBDATA the data consists of out-of-banddata. .El .Pp The .Fa mh_flags variable can take the following values: .Pp .Bl -tag -compact -offset indent -width XXXXXXXXXXXXXXXXXX .It Dv M_EXT mbuf has associated external storage. .It Dv M_PKTHDR the mbuf is the first that forms a packet. .It Dv M_EOR end of record. .It Dv M_CLUSTER the external storage is a cluster. .It Dv M_PROTO1 protocol-specific. .It Dv M_BCAST packet send/received as link-level broadcast. .It Dv M_MCAST packet send/received as link-level multicast. .It Dv M_CONF packet was encrypted (ESP-transport). .It Dv M_AUTH packet was authenticated (AH or ESP). .It Dv M_AUTH_AH header was authenticated (AH). .It Dv M_TUNNEL header was IP-in-IP encapsulated by tunnel mode IPsec. .It Dv M_ANYCAST6 received as IPv6 anycast. .It Dv M_LOOP for mbuf statistics. .El .Pp An external cluster is used when the data to hold in the mbuf is large. The size of an external cluster is MCLBYTES .Pq also defined in Aq Pa machine/param.h . A cluster should be used when the size of the data reach MINCLSIZE (the minimum size to be held by an external cluster). .Pp The combination of the M_EXT and M_PKTHDR flags give four types of mbuf. When none of these constants are in use, the mbuf is a "normal" one, where the data part of the mbuf has the following elements: .Bl -tag -width foobarmoocow .It Fa m_dat buffer holding the data (size MLEN). .El .Pp When only M_PKTHDR is set, the data contained in the mbuf is a packet header. The data itself is contained in the mbuf (just like the previous case), but part of the mbuf is used to store a packet header. The data part has then the following elements: .Bl -tag -width foobarmoocow .It Fa m_pkthdr packet header, containing the length of the data, a pointer to the interface on which the data was received, checksum information and list of .Xr mbuf_tags 9 . .It Fa m_pktdat buffer holding the data (size MHLEN). .El .Pp The .Fa m_pkthdr.csum variable can take the following values: .Pp .Bl -tag -compact -offset indent -width XXXXXXXXXXXXXXXXXX .It Dv M_IPV4_CSUM_OUT IPv4 checksum needed. .It Dv M_TCPV4_CSUM_OUT TCP checksum needed. .It Dv M_UDPV4_CSUM_OUT UDP checksum needed. .It Dv M_IPV4_CSUM_IN_OK IPv4 checksum verified. .It Dv M_IPV4_CSUM_IN_BAD IPv4 checksum bad. .It Dv M_TCP_CSUM_IN_OK TCP/IPv4 checksum verified. .It Dv M_TCP_CSUM_IN_BAD TCP/IPv4 checksum bad. .It Dv M_UDP_CSUM_IN_OK UDP/IPv4 checksum verified. .It Dv M_UDP_CSUM_IN_BAD UDP/IPv4 checksum bad. .El .Pp When only M_EXT flag is set, an external storage buffer is being used to hold the data, which is no longer stored in the mbuf. The data part of the mbuf has now the following elements: .Bl -tag -width foobarmoocow .It Fa m_pkthdr a packet header, just like the previous case, but it is empty. No information is stored here .It Fa m_ext a structure containing information about the external storage buffer. The information consists of the address of the external buffer, a pointer to the function used to free the buffer, a pointer to the arguments of the function, the size of the buffer, the type of the buffer, and pointers to the previous and next mbufs using this cluster. .El .Pp When both the M_EXT and M_PKTHDR flags are set, an external storage buffer is being used to store the data and this data contains a packet header. The structure used is the same as the previous one except that the .Fa m_pkthdr element is not empty, it contains the same information as when M_PKTHDR is used alone. .Bl -tag -width Ds .It Fn m_copym "struct mbuf *m" "int off" "int len" "int wait" Copy an mbuf chain starting at .Fa off bytes from the beginning and continuing for .Fa len bytes. If .Fa off is zero and .Fa m has the M_PKTHDR flag set, the header is copied. If .Fa len is M_COPYALL the whole mbuf is copied. The .Fa wait parameter can be M_WAIT or M_DONTWAIT. It does not copy clusters, it just increases their reference count. .It Fn m_copym2 "struct mbuf *m" "int off" "int len" "int wait" The same as .Fn m_copym except that it copies cluster mbufs, whereas .Fn m_copym just increases the reference count of the clusters. .It Fn m_free "struct mbuf *m" Free the mbuf pointed to by .Fa m . A pointer to the successor of the mbuf, if it exists, is returned by the function. .It Fn MFREE "struct mbuf *m" "struct mbuf *n" Free the mbuf pointed to by .Fa m and use .Fa n to point to the next mbuf in the chain if it exists. See .Fn m_free . .It Fn m_get "int how" "int type" Return a pointer to an mbuf of the type specified. If the .Fa how argument is .Fa M_WAITOK , the function may call .Xr tsleep 9 to await resources. If .Fa how is .Fa M_DONTWAIT and resources are not available, .Fn m_get returns NULL. .It Fn MGET "struct mbuf *m" "int how" "int type" Return a pointer to an mbuf in .Fa m of the type specified. See .Fn m_get for a description of .Fa how . .It Fn m_getclr "int how" "int type" Return a pointer to an mbuf of the type specified, and clear the data area of the mbuf. See .Fn m_get for a description of .Fa how . .It Fn m_gethdr "int how" "int type" Return a pointer to an mbuf of the type specified after initializing it to contain a packet header. See .Fn m_get for a description of .Fa how . .It Fn MGETHDR "struct mbuf *m" "int how" "int type" Return a pointer to an mbuf of the type specified after initializing it to contain a packet header. See .Fn m_get for a description of .Fa how . .It Fn m_prepend "struct mbuf *m" "int len" "int how" Allocate a new mbuf and prepend it to the mbuf chain pointed to by .Fa m . If .Fa m points to an mbuf with a packet header, it is moved to the new mbuf that has been prepended. The return value is a pointer on the new mbuf chain. If this function fails to allocate a new mbuf, .Fa m is freed. See .Fn m_get for a description of .Fa how . .Pp .Fn m_prepend should never be called directly. Use .Fn M_PREPEND instead. .It Fn M_PREPEND "struct mbuf *m" "int plen" "int how" Prepend space of size .Fa plen to the mbuf pointed to by .Fa m . If a new mbuf must be allocated, .Fa how specifies whether to wait or not. If this function fails to allocate a new mbuf, .Fa m is freed. .It Fn m_pulldown "struct mbuf *m" "int off" "int len" "int *offp" Ensure that the data in the mbuf chain starting at .Fa off and ending at .Fa off+len will be put in a continuous memory region. .Fa len must be smaller or equal than MCLBYTES. The pointer returned points to an mbuf in the chain and the new offset for data in this mbuf is .Fa *offp . If this function fails, .Fa m is freed. .It Fn m_pullup "struct mbuf *n" "int len" Ensure that the data in the mbuf chain starting at the beginning of the chain and ending at .Fa len will be put in continuous memory region. To avoid being called again, .Fn m_pullup will attempt to copy .Fa "max_protohdr - len" bytes into the first mbuf. The .Fa len argument must be smaller or equal than MHLEN. If this function fails, .Fa n is freed. .It Fn m_pullup2 "struct mbuf *n" "int len" Just like .Fn m_pullup , ensure that the data starting at the beginning of the mbuf chain and ending at .Fa len will be put in continuous memory region. The .Fa len argument can be up to MCLBYTES. .Fn m_pullup2 will simply call .Fn m_pullup if .Fa len is smaller or equal to MHLEN. .It Fn m_split "struct mbuf *m0" "int len0" "int wait" Split an mbuf chain in two pieces, returning a pointer to the tail (which is made of the previous mbuf chain except the first .Fa len0 bytes). .It Fn m_inject "struct mbuf *m0" "int len0" "int siz" "int wait" Inject a new mbuf chain of length .Fa siz into the mbuf chain pointed to by .Fa m0 at position .Fa len0 . If there is enough space for an object of size .Fa siz in the appropriate location, no memory will be allocated. On failure, the function returns NULL (the mbuf is left untouched) and on success, a pointer to the first injected mbuf is returned. .It Fn m_getptr "struct mbuf *m" "int loc" "int *off" Returns a pointer to the mbuf containing the data located at .Fa loc bytes of the beginning. The offset in the new mbuf is pointed to by .Fa off . .It Fn m_adj "struct mbuf *mp" "int req_len" Trims .Fa req_len bytes of data from the mbuf chain pointed to by .Fa mp . If .Fa req_len is positive, the data will be trimmed from the head of the mbuf chain and if it is negative, it will be trimmed from the tail of the mbuf chain. .It Fn m_copyback "struct mbuf *m0" "int off" "int len" "caddr_t cp" Copy data from a buffer pointed to by .Fa cp back into the mbuf chain pointed to by .Fa m0 starting at .Fa off bytes from the beginning, extending the mbuf chain if necessary. The mbuf chain must be initialized properly, including setting .Fa m_len . .It Fn m_freem "struct mbuf *m" Free the mbuf chain pointed to by .Fa m . .It Fn m_reclaim "void" Ask protocols to free unused memory space. .It Fn m_copydata "struct mbuf *m" "int off" "int len" "caddr_t cp" Copy data from the mbuf chain pointed to by .Fa m starting at .Fa off bytes from the beginning and continuing for .Fa len bytes into the buffer pointed to by .Fa cp . .It Fn m_cat "struct mbuf *m" "struct mbuf *n" Concatenate the mbuf chain pointed to by .Fa n to the mbuf chain pointed to by .Fa m . The mbuf chains must be of the same type. .It Fn m_devget "char *buf" "int totlen" "int off" "struct ifnet *ifp" \ "void (*func)(const void *, void *, size_t)" Copy .Fa totlen bytes of data from device local memory pointed to by .Fa buf using the function .Fa func . The data is copied into an mbuf chain and a pointer to the head of it is returned. If .Fa off is non-zero, it means the packet is trailer-encapsulated and .Fa off bytes plus the type and length fields will be skipped before doing the copy. Returns NULL on failure. .It Fn m_zero "struct mbuf *m" Zeroize the data part of the mbufs in the mbuf chain pointed to by .Fa m . .It Fn m_apply "struct mbuf *m" "int off" "int len" \ "int (*func)(caddr_t, caddr_t, unsigned int)" "caddr_t fstate" Apply the function .Fa func to the data in the mbuf chain pointed to by .Fa m starting at .Fa off bytes from the beginning and continuing for .Fa len bytes. .It Fn mtod "struct mbuf *m" "datatype" Return a pointer to the data contained in the specified mbuf .Fa m cast to .Fa datatype . .It Fn MCLGET "struct mbuf *m" "int how" Allocate and add an mbuf cluster to the mbuf pointed to by .Fa m . On success, the flag M_EXT is set in the mbuf. See .Fn m_get for a description of .Fa how . .It Fn MEXTMALLOC "struct mbuf *m" "int size" "int how" Allocate external storage of size .Fa size and add it to the mbuf pointed to by .Fa m . On success, the flag M_EXT is set in the mbuf. See .Fn m_get for a description of .Fa how . .It Fn MEXTADD "struct mbuf *m" "caddr_t buf" "u_int size" "int type" \ "void (*free)(caddr_t, u_int, void *)" "void *arg" Add pre-allocated storage to the mbuf pointed to by .Fa m . On success, the flag M_EXT is set in the mbuf. .It Fn M_ALIGN "struct mbuf *m" "int len" Set the .Fa m_data pointer of the newly allocated mbuf with .Fn m_get or .Fn MGET pointed to by .Fa m to an object of the specified size .Fa len at the end of the mbuf, longword aligned. .It Fn MH_ALIGN "m" "len" Same as .Fn M_ALIGN except it is for an mbuf allocated with .Fn m_gethdr or .Fn MGETHDR . .It Fn M_READONLY "struct mbuf *m" Check if the data of the mbuf pointed to by .Fa m is read-only. This is true for non-cluster external storage and for clusters that are being referenced by more than one mbuf. .It Fn M_LEADINGSPACE "struct mbuf *m" Compute the amount of space available before the current start of data in the mbuf pointed to by .Fa m . .It Fn M_TRAILINGSPACE "struct mbuf *m" Compute the amount of space available after the end of data in the mbuf pointed to by .Fa m . .It Fn MCHTYPE "struct mbuf *m" "int type" Change the type of the mbuf pointed to by .Fa m to .Fa type . .El .Sh CODE REFERENCES The mbuf management functions are implemented in the files .Pa sys/kern/uipc_mbuf.c and .Pa sys/kern/uipc_mbuf2.c . The function prototypes and the macros are located in .Pa sys/sys/mbuf.h . .Sh SEE ALSO .Xr netstat 1 , .Xr mbuf_tags 9 .Pp "Networking Implementation Notes", .Pa /usr/share/doc/smm/18.net . .Rs .%A Jun-Ichiro Hagino .%T "Mbuf issues in 4.4BSD IPv6/IPsec support (experiences from KAME IPv6/IPsec implementation)" .%B "Proceedings of the Freenix Track: 2000 USENIX Annual Technical Conference" .%D June 2000 .Re