1.\" Copyright (c) 2000 FreeBSD Inc. 2.\" All rights reserved. 3.\" 4.\" Redistribution and use in source and binary forms, with or without 5.\" modification, are permitted provided that the following conditions 6.\" are met: 7.\" 1. Redistributions of source code must retain the above copyright 8.\" notice, this list of conditions and the following disclaimer. 9.\" 2. Redistributions in binary form must reproduce the above copyright 10.\" notice, this list of conditions and the following disclaimer in the 11.\" documentation and/or other materials provided with the distribution. 12.\" 13.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16.\" ARE DISCLAIMED. IN NO EVENT SHALL [your name] OR CONTRIBUTORS BE LIABLE 17.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23.\" SUCH DAMAGE. 24.\" 25.\" $FreeBSD: src/share/man/man9/mbuf.9,v 1.27.2.1 2003/05/28 13:53:18 yar Exp $ 26.\" $DragonFly: src/share/man/man9/mbuf.9,v 1.8 2007/08/18 18:58:20 swildner Exp $ 27.\" 28.Dd August 18, 2007 29.Dt MBUF 9 30.Os 31.\" 32.Sh NAME 33.Nm mbuf 34.Nd "memory management in the kernel IPC subsystem" 35.\" 36.Sh SYNOPSIS 37.In sys/param.h 38.In sys/systm.h 39.In sys/mbuf.h 40.\" 41.Ss Mbuf allocation macros 42.Fn MGET "struct mbuf *mbuf" "int how" "short type" 43.Fn MGETHDR "struct mbuf *mbuf" "int how" "short type" 44.Fn MCLGET "struct mbuf *mbuf" "int how" 45.\" 46.Ss Mbuf utility macros 47.Ft void * 48.Fn mtod "struct mbuf *mbuf" "type" 49.Fn M_ALIGN "struct mbuf *mbuf" "u_int len" 50.Fn MH_ALIGN "struct mbuf *mbuf" "u_int len" 51.Ft int 52.Fn M_LEADINGSPACE "struct mbuf *mbuf" 53.Ft int 54.Fn M_TRAILINGSPACE "struct mbuf *mbuf" 55.Fn M_PREPEND "struct mbuf *mbuf" "int len" "int how" 56.\" 57.Ss Mbuf allocation functions 58.Ft struct mbuf * 59.Fn m_get "int how" "int type" 60.Ft struct mbuf * 61.Fn m_getm "struct mbuf *orig" "int len" "int how" "int type" 62.Ft struct mbuf * 63.Fn m_getclr "int how" "int type" 64.Ft struct mbuf * 65.Fn m_gethdr "int how" "int type" 66.Ft struct mbuf * 67.Fn m_free "struct mbuf *mbuf" 68.Ft void 69.Fn m_freem "struct mbuf *mbuf" 70.\" 71.Ss Mbuf utility functions 72.Ft void 73.Fn m_adj "struct mbuf *mbuf" "int len" 74.Ft struct mbuf * 75.Fn m_prepend "struct mbuf *mbuf" "int len" "int how" 76.Ft struct mbuf * 77.Fn m_pullup "struct mbuf *mbuf" "int len" 78.Ft struct mbuf * 79.Fn m_copym "struct mbuf *mbuf" "int offset" "int len" "int how" 80.Ft struct mbuf * 81.Fn m_copypacket "struct mbuf *mbuf" "int how" 82.Ft struct mbuf * 83.Fn m_dup "struct mbuf *mbuf" "int how" 84.Ft void 85.Fn m_copydata "const struct mbuf *mbuf" "int offset" "int len" "caddr_t buf" 86.Ft void 87.Fn m_copyback "struct mbuf *mbuf" "int offset" "int len" "caddr_t buf" 88.Ft struct mbuf * 89.Fo m_devget 90.Fa "char *buf" 91.Fa "int len" 92.Fa "int offset" 93.Fa "struct ifnet *ifp" 94.Fa "void (*copy)(char *from, caddr_t to, u_int len)" 95.Fc 96.Ft void 97.Fn m_cat "struct mbuf *m" "struct mbuf *n" 98.Ft struct mbuf * 99.Fn m_split "struct mbuf *mbuf" "int len" "int how" 100.\" 101.Sh DESCRIPTION 102An mbuf is a basic unit of memory management in the kernel IPC subsystem. 103Network packets and socket buffers are stored in mbufs. 104A network packet may span multiple mbufs arranged into a chain 105(linked list), 106which allows adding or trimming 107network headers with little overhead. 108.Pp 109While a developer should not bother with mbuf internals without serious 110reason in order to avoid incompatibilities with future changes, it 111is useful to understand the mbuf's general structure. 112.Pp 113An mbuf consists of a variable-sized header and a small internal 114buffer for data. 115The mbuf's total size, 116.Dv MSIZE , 117is a machine-dependent constant defined in 118.In machine/param.h . 119The mbuf header includes: 120.Pp 121.Bl -tag -width "m_nextpkt" -compact -offset indent 122.It Fa m_next 123a pointer to the next buffer in the chain 124.It Fa m_nextpkt 125a pointer to the next chain in the queue 126.It Fa m_data 127a pointer to the data 128.It Fa m_len 129the length of the data 130.It Fa m_type 131the type of data 132.It Fa m_flags 133the mbuf flags 134.El 135.Pp 136The mbuf flag bits are defined as follows: 137.Bd -literal 138/* mbuf flags */ 139#define M_EXT 0x0001 /* has associated external storage */ 140#define M_PKTHDR 0x0002 /* start of record */ 141#define M_EOR 0x0004 /* end of record */ 142#define M_PROTO1 0x0010 /* protocol-specific */ 143#define M_PROTO2 0x0020 /* protocol-specific */ 144#define M_PROTO3 0x0040 /* protocol-specific */ 145#define M_PROTO4 0x0080 /* protocol-specific */ 146#define M_PROTO5 0x0100 /* protocol-specific */ 147 148/* mbuf pkthdr flags, also in m_flags */ 149#define M_BCAST 0x0200 /* send/received as link-level broadcast */ 150#define M_MCAST 0x0400 /* send/received as link-level multicast */ 151#define M_FRAG 0x0800 /* packet is fragment of larger packet */ 152#define M_FIRSTFRAG 0x1000 /* packet is first fragment */ 153#define M_LASTFRAG 0x2000 /* packet is last fragment */ 154.Ed 155.Pp 156The available mbuf types are defined as follows: 157.Bd -literal 158/* mbuf types */ 159#define MT_FREE 0 /* should be on free list */ 160#define MT_DATA 1 /* dynamic (data) allocation */ 161#define MT_HEADER 2 /* packet header */ 162#define MT_SONAME 8 /* socket name */ 163#define MT_FTABLE 11 /* fragment reassembly header */ 164#define MT_CONTROL 14 /* extra-data protocol message */ 165#define MT_OOBDATA 15 /* expedited data */ 166.Ed 167.Pp 168If the 169.Dv M_PKTHDR 170flag is set, a 171.Li struct pkthdr m_pkthdr 172is added to the mbuf header. 173It contains a pointer to the interface 174the packet has been received from 175.Pq Fa struct ifnet *rcvif , 176and the total packet length 177.Pq Fa int len . 178.Pp 179If small enough, data is stored in the mbuf's internal data buffer. 180If the data is sufficiently large, another mbuf may be added to the chain, 181or external storage may be associated with the mbuf. 182.Dv MHLEN 183bytes of data can fit into an mbuf with the 184.Dv M_PKTHDR 185flag set, 186.Dv MLEN 187bytes can otherwise. 188.Pp 189If external storage is being associated with an mbuf, the 190.Dv m_ext 191header is added at the cost of losing the internal data buffer. 192It includes a pointer to external storage, the size of the storage, 193a pointer to a function used for freeing the storage, 194a pointer to an optional argument that can be passed to the function, 195and a pointer to a reference counter. 196An mbuf using external storage has the 197.Dv M_EXT 198flag set. 199.Pp 200The system supplies a default type of external storage buffer called an 201.Dq mbuf cluster . 202Mbuf clusters can be allocated and configured with the use of the 203.Dv MCLGET 204macro. 205Each cluster is 206.Dv MCLBYTES 207in size, where 208.Dv MCLBYTES 209is a machine-dependent constant. 210The system defines an advisory macro 211.Dv MINCLSIZE , 212which is the smallest amount of data to put into a cluster. 213It's equal to the sum of 214.Dv MLEN 215and 216.Dv MHLEN . 217It is typically preferable to store data into an mbuf's data region, if size 218permits, as opposed to allocating a separate mbuf cluster to hold the same 219data. 220.\" 221.Ss Macros and Functions 222There are numerous predefined macros and functions that provide the 223developer with common utilities. 224.\" 225.Bl -ohang -offset indent 226.It Fn mtod mbuf type 227Convert an mbuf pointer to a data pointer. 228The macro expands to the data pointer cast to the pointer of the specified type. 229.Sy Note : 230It is advisable to ensure that there is enough contiguous data in the mbuf. 231See 232.Fn m_pullup 233for details. 234.It Fn MGET mbuf how type 235Allocate an mbuf and initialize it to contain internal data. 236.Fa mbuf 237will point to the allocated mbuf on success, or be set to 238.Dv NULL 239on failure. 240The 241.Fa how 242argument is to be set to 243.Dv MB_WAIT 244or 245.Dv MB_DONTWAIT . 246It specifies whether the caller is willing to block if necessary. 247If 248.Fa how 249is set to 250.Dv MB_WAIT , 251a failed allocation will result in the caller being put 252to sleep for a designated 253.Va kern.ipc.mbuf_wait 254.Xr ( sysctl 8 255tunable) 256number of ticks. 257A number of other mbuf-related 258functions and macros have the same argument because they may 259at some point need to allocate new mbufs. 260.Pp 261Programmers should be careful not to confuse the mbuf allocation flag 262.Dv MB_DONTWAIT 263with the 264.Xr kmalloc 9 265allocation flag, 266.Dv M_NOWAIT . 267They are not the same. 268.It Fn MGETHDR mbuf how type 269Allocate an mbuf and initialize it to contain a packet header 270and internal data. 271See 272.Fn MGET 273for details. 274.It Fn MCLGET mbuf how 275Allocate and attach an mbuf cluster to an mbuf. 276If the macro fails, the 277.Dv M_EXT 278flag won't be set in the mbuf. 279.It Fn M_PREPEND mbuf len how 280This macro operates on an mbuf chain. 281It is an optimized wrapper for 282.Fn m_prepend 283that can make use of possible empty space before data 284(e.g. left after trimming of a link-layer header). 285The new chain pointer or 286.Dv NULL 287is in 288.Fa mbuf 289after the call. 290.El 291.Pp 292The functions are: 293.Bl -ohang -offset indent 294.It Fn m_get how type 295A function version of 296.Fn MGET 297for non-critical paths. 298.It Fn m_getm orig len how type 299Allocate 300.Fa len 301bytes worth of mbufs and mbuf clusters if necessary and append the resulting 302allocated chain to the 303.Fa orig 304mbuf chain, if it is 305.No non- Ns Dv NULL . 306If the allocation fails at any point, 307free whatever was allocated and return 308.Dv NULL . 309If 310.Fa orig 311is 312.No non- Ns Dv NULL , 313it will not be freed. 314It is possible to use 315.Fn m_getm 316to either append 317.Fa len 318bytes to an existing mbuf or mbuf chain 319(for example, one which may be sitting in a pre-allocated ring) 320or to simply perform an all-or-nothing mbuf and mbuf cluster allocation. 321.It Fn m_gethdr how type 322A function version of 323.Fn MGETHDR 324for non-critical paths. 325.It Fn m_getclr how type 326Allocate an mbuf and zero out the data region. 327.El 328.Pp 329The functions below operate on mbuf chains. 330.Bl -ohang -offset indent 331.It Fn m_freem mbuf 332Free an entire mbuf chain, including any external 333storage. 334.\" 335.It Fn m_adj mbuf len 336Trim 337.Fa len 338bytes from the head of an mbuf chain if 339.Fa len 340is positive, from the tail otherwise. 341.\" 342.It Fn m_prepend mbuf len how 343Allocate a new mbuf and prepend it to the chain, handle 344.Dv M_PKTHDR 345properly. 346.Sy Note : 347It doesn't allocate any clusters, so 348.Fa len 349must be less than 350.Dv MLEN 351or 352.Dv MHLEN , 353depending on the 354.Dv M_PKTHDR 355flag setting. 356.\" 357.It Fn m_pullup mbuf len 358Arrange that the first 359.Fa len 360bytes of an mbuf chain are contiguous and lay in the data area of 361.Fa mbuf , 362so they are accessible with 363.Fn mtod mbuf type . 364Return the new chain on success, 365.Dv NULL 366on failure 367(the chain is freed in this case). 368.Sy Note : 369It doesn't allocate any clusters, so 370.Fa len 371must be less than 372.Dv MHLEN . 373.\" 374.It Fn m_copym mbuf offset len how 375Make a copy of an mbuf chain starting 376.Fa offset 377bytes from the beginning, continuing for 378.Fa len 379bytes. 380If 381.Fa len 382is 383.Dv M_COPYALL , 384copy to the end of the mbuf chain. 385.Sy Note : 386The copy is read-only, because clusters are not 387copied, only their reference counts are incremented. 388.\" 389.It Fn m_copypacket mbuf how 390Copy an entire packet including header, which must be present. 391This is an optimized version of the common case 392.Fn m_copym mbuf 0 M_COPYALL how . 393.Sy Note : 394the copy is read-only, because clusters are not 395copied, only their reference counts are incremented. 396.\" 397.It Fn m_dup mbuf how 398Copy a packet header mbuf chain into a completely new chain, including 399copying any mbuf clusters. 400Use this instead of 401.Fn m_copypacket 402when you need a writable copy of an mbuf chain. 403.\" 404.It Fn m_copydata mbuf offset len buf 405Copy data from an mbuf chain starting 406.Fa off 407bytes from the beginning, continuing for 408.Fa len 409bytes, into the indicated buffer 410.Fa buf . 411.\" 412.It Fn m_copyback mbuf offset len buf 413Copy 414.Fa len 415bytes from the buffer 416.Fa buf 417back into the indicated mbuf chain, 418starting at 419.Fa offset 420bytes from the beginning of the chain, extending the mbuf chain if necessary. 421.Sy Note : 422It doesn't allocate any clusters, just adds mbufs to the chain. 423It's safe to set 424.Fa offset 425beyond the current chain end: zeroed mbufs will be allocated to fill the 426space. 427.\" 428.It Fn m_devget buf len offset ifp copy 429Copy data from a device local memory pointed to by 430.Fa buf 431to an mbuf chain. 432The copy is done using a specified copy routine 433.Fa copy , 434or 435.Fn bcopy 436if 437.Fa copy 438is 439.Dv NULL . 440.\" 441.It Fn m_cat m n 442Concatenate 443.Fa n 444to 445.Fa m . 446Both chains must be of the same type. 447.Fa N 448is still valid after the function returned. 449.Sy Note : 450It does not handle 451.Dv M_PKTHDR 452and friends. 453.\" 454.It Fn m_split mbuf len how 455Partition an mbuf chain in two pieces, returning the tail: 456all but the first 457.Fa len 458bytes. 459In case of failure, it returns 460.Dv NULL 461and attempts to restore the chain to its original state. 462.El 463.Sh STRESS TESTING 464When running a kernel compiled with the option 465.Dv MBUF_STRESS_TEST , 466the following 467.Xr sysctl 8 Ns 468-controlled options may be used to create 469various failure/extreme cases for testing of network drivers 470and other parts of the kernel that rely on 471.Vt mbufs . 472.Bl -tag -width ident 473.It Va net.inet.ip.mbuf_frag_size 474Causes 475.Fn ip_output 476to fragment outgoing 477.Vt mbuf chains 478into fragments of the specified size. 479Setting this variable to 1 is an excellent way to 480test the long 481.Vt mbuf chain 482handling ability of network drivers. 483.It Va kern.ipc.m_defragrandomfailures 484Causes the function 485.Fn m_defrag 486to randomly fail, returning 487.Dv NULL . 488Any piece of code which uses 489.Fn m_defrag 490should be tested with this feature. 491.El 492.Sh RETURN VALUES 493See above. 494.Sh HISTORY 495.\" Please correct me if I'm wrong 496Mbufs appeared in an early version of 497.Bx . 498Besides for being used for network packets, they were used 499to store various dynamic structures, such as routing table 500entries, interface addresses, protocol control blocks, etc. 501.Sh AUTHORS 502The original 503.Nm 504man page was written by Yar Tikhiy. 505