1 .\" Copyright (c) 2000 FreeBSD Inc.
2 .\" All rights reserved.
4 .\" Redistribution and use in source and binary forms, with or without
5 .\" modification, are permitted provided that the following conditions
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.
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
25 .\" $FreeBSD: src/share/man/man9/mbuf.9,v 1.27.2.1 2003/05/28 13:53:18 yar Exp $
27 .Dd September 17, 2010
33 .Nd "memory management in the kernel IPC subsystem"
40 .Ss Mbuf allocation macros
41 .Fn MGET "struct mbuf *mbuf" "int how" "short type"
42 .Fn MGETHDR "struct mbuf *mbuf" "int how" "short type"
43 .Fn MCLGET "struct mbuf *mbuf" "int how"
45 .Ss Mbuf utility macros
47 .Fn mtod "struct mbuf *mbuf" "type"
48 .Fn M_ALIGN "struct mbuf *mbuf" "u_int len"
49 .Fn MH_ALIGN "struct mbuf *mbuf" "u_int len"
51 .Fn M_LEADINGSPACE "struct mbuf *mbuf"
53 .Fn M_TRAILINGSPACE "struct mbuf *mbuf"
54 .Fn M_PREPEND "struct mbuf *mbuf" "int len" "int how"
56 .Ss Mbuf allocation functions
58 .Fn m_get "int how" "int type"
60 .Fn m_getm "struct mbuf *orig" "int len" "int how" "int type"
62 .Fn m_getclr "int how" "int type"
64 .Fn m_gethdr "int how" "int type"
66 .Fn m_free "struct mbuf *mbuf"
68 .Fn m_freem "struct mbuf *mbuf"
70 .Ss Mbuf utility functions
72 .Fn m_adj "struct mbuf *mbuf" "int len"
74 .Fn m_prepend "struct mbuf *mbuf" "int len" "int how"
76 .Fn m_pullup "struct mbuf *mbuf" "int len"
78 .Fn m_copym "const struct mbuf *mbuf" "int offset" "int len" "int how"
80 .Fn m_copypacket "struct mbuf *mbuf" "int how"
82 .Fn m_dup "struct mbuf *mbuf" "int how"
84 .Fn m_copydata "const struct mbuf *mbuf" "int offset" "int len" "caddr_t buf"
86 .Fn m_copyback "struct mbuf *mbuf" "int offset" "int len" "caddr_t buf"
92 .Fa "struct ifnet *ifp"
93 .Fa "void (*copy)(volatile const void *from, volatile void *to, size_t len)"
96 .Fn m_cat "struct mbuf *m" "struct mbuf *n"
98 .Fn m_split "struct mbuf *mbuf" "int len" "int how"
100 .Fn m_unshare "struct mbuf *mbuf" "int how"
103 An mbuf is a basic unit of memory management in the kernel IPC subsystem.
104 Network packets and socket buffers are stored in mbufs.
105 A network packet may span multiple mbufs arranged into a chain
107 which allows adding or trimming
108 network headers with little overhead.
110 While a developer should not bother with mbuf internals without serious
111 reason in order to avoid incompatibilities with future changes, it
112 is useful to understand the mbuf's general structure.
114 An mbuf consists of a variable-sized header and a small internal
116 The mbuf's total size,
118 is a machine-dependent constant defined in
119 .In machine/param.h .
120 The mbuf header includes:
122 .Bl -tag -width "m_nextpkt" -compact -offset indent
124 a pointer to the next buffer in the chain
126 a pointer to the next chain in the queue
128 a pointer to the data
130 the length of the data
137 The mbuf flag bits are defined as follows:
140 #define M_EXT 0x0001 /* has associated external storage */
141 #define M_PKTHDR 0x0002 /* start of record */
142 #define M_EOR 0x0004 /* end of record */
143 #define M_PROTO1 0x0010 /* protocol-specific */
144 #define M_PROTO2 0x0020 /* protocol-specific */
145 #define M_PROTO3 0x0040 /* protocol-specific */
146 #define M_PROTO4 0x0080 /* protocol-specific */
147 #define M_PROTO5 0x0100 /* protocol-specific */
149 /* mbuf pkthdr flags, also in m_flags */
150 #define M_BCAST 0x0200 /* send/received as link-level broadcast */
151 #define M_MCAST 0x0400 /* send/received as link-level multicast */
152 #define M_FRAG 0x0800 /* packet is fragment of larger packet */
153 #define M_FIRSTFRAG 0x1000 /* packet is first fragment */
154 #define M_LASTFRAG 0x2000 /* packet is last fragment */
157 The available mbuf types are defined as follows:
160 #define MT_FREE 0 /* should be on free list */
161 #define MT_DATA 1 /* dynamic (data) allocation */
162 #define MT_HEADER 2 /* packet header */
163 #define MT_SONAME 8 /* socket name */
164 #define MT_FTABLE 11 /* fragment reassembly header */
165 #define MT_CONTROL 14 /* extra-data protocol message */
166 #define MT_OOBDATA 15 /* expedited data */
172 .Li struct pkthdr m_pkthdr
173 is added to the mbuf header.
174 It contains a pointer to the interface
175 the packet has been received from
176 .Pq Fa struct ifnet *rcvif ,
177 and the total packet length
180 If small enough, data is stored in the mbuf's internal data buffer.
181 If the data is sufficiently large, another mbuf may be added to the chain,
182 or external storage may be associated with the mbuf.
184 bytes of data can fit into an mbuf with the
190 If external storage is being associated with an mbuf, the
192 header is added at the cost of losing the internal data buffer.
193 It includes a pointer to external storage, the size of the storage,
194 a pointer to a function used for freeing the storage,
195 a pointer to an optional argument that can be passed to the function,
196 and a pointer to a reference counter.
197 An mbuf using external storage has the
201 The system supplies a default type of external storage buffer called an
203 Mbuf clusters can be allocated and configured with the use of the
210 is a machine-dependent constant.
211 The system defines an advisory macro
213 which is the smallest amount of data to put into a cluster.
214 It's equal to the sum of
218 It is typically preferable to store data into an mbuf's data region, if size
219 permits, as opposed to allocating a separate mbuf cluster to hold the same
222 .Ss Macros and Functions
223 There are numerous predefined macros and functions that provide the
224 developer with common utilities.
226 .Bl -ohang -offset indent
227 .It Fn mtod mbuf type
228 Convert an mbuf pointer to a data pointer.
229 The macro expands to the data pointer cast to the pointer of the specified type.
231 It is advisable to ensure that there is enough contiguous data in the mbuf.
235 .It Fn MGET mbuf how type
236 Allocate an mbuf and initialize it to contain internal data.
238 will point to the allocated mbuf on success, or be set to
243 argument is to be set to
247 It specifies whether the caller is willing to block if necessary.
252 a failed allocation will result in the caller being put
253 to sleep for a designated
254 .Va kern.ipc.mbuf_wait
258 A number of other mbuf-related
259 functions and macros have the same argument because they may
260 at some point need to allocate new mbufs.
262 Programmers should be careful not to confuse the mbuf allocation flag
268 They are not the same.
269 .It Fn MGETHDR mbuf how type
270 Allocate an mbuf and initialize it to contain a packet header
275 .It Fn MCLGET mbuf how
276 Allocate and attach an mbuf cluster to an mbuf.
277 If the macro fails, the
279 flag won't be set in the mbuf.
280 .It Fn M_PREPEND mbuf len how
281 This macro operates on an mbuf chain.
282 It is an optimized wrapper for
284 that can make use of possible empty space before data
285 (e.g. left after trimming of a link-layer header).
286 The new chain pointer or
294 .Bl -ohang -offset indent
295 .It Fn m_get how type
296 A function version of
298 for non-critical paths.
299 .It Fn m_getm orig len how type
302 bytes worth of mbufs and mbuf clusters if necessary and append the resulting
303 allocated chain to the
306 .No non- Ns Dv NULL .
307 If the allocation fails at any point,
308 free whatever was allocated and return
313 .No non- Ns Dv NULL ,
314 it will not be freed.
315 It is possible to use
319 bytes to an existing mbuf or mbuf chain
320 (for example, one which may be sitting in a pre-allocated ring)
321 or to simply perform an all-or-nothing mbuf and mbuf cluster allocation.
322 .It Fn m_gethdr how type
323 A function version of
325 for non-critical paths.
326 .It Fn m_getclr how type
327 Allocate an mbuf and zero out the data region.
330 The functions below operate on mbuf chains.
331 .Bl -ohang -offset indent
333 Free an entire mbuf chain, including any external
336 .It Fn m_adj mbuf len
339 bytes from the head of an mbuf chain if
341 is positive, from the tail otherwise.
343 .It Fn m_prepend mbuf len how
344 Allocate a new mbuf and prepend it to the chain, handle
348 It doesn't allocate any clusters, so
358 .It Fn m_pullup mbuf len
359 Arrange that the first
361 bytes of an mbuf chain are contiguous and lay in the data area of
363 so they are accessible with
365 Return the new chain on success,
368 (the chain is freed in this case).
370 It doesn't allocate any clusters, so
375 .It Fn m_copym mbuf offset len how
376 Make a copy of an mbuf chain starting
378 bytes from the beginning, continuing for
385 copy to the end of the mbuf chain.
387 The copy is read-only, because clusters are not
388 copied, only their reference counts are incremented.
390 .It Fn m_copypacket mbuf how
391 Copy an entire packet including header, which must be present.
392 This is an optimized version of the common case
393 .Fn m_copym mbuf 0 M_COPYALL how .
395 the copy is read-only, because clusters are not
396 copied, only their reference counts are incremented.
398 .It Fn m_dup mbuf how
399 Copy a packet header mbuf chain into a completely new chain, including
400 copying any mbuf clusters.
403 when you need a writable copy of an mbuf chain.
405 .It Fn m_copydata mbuf offset len buf
406 Copy data from an mbuf chain starting
408 bytes from the beginning, continuing for
410 bytes, into the indicated buffer
413 .It Fn m_copyback mbuf offset len buf
416 bytes from the buffer
418 back into the indicated mbuf chain,
421 bytes from the beginning of the chain, extending the mbuf chain if necessary.
423 It doesn't allocate any clusters, just adds mbufs to the chain.
426 beyond the current chain end: zeroed mbufs will be allocated to fill the
429 .It Fn m_devget buf len offset ifp copy
430 Copy data from a device local memory pointed to by
433 The copy is done using a specified copy routine
447 Both chains must be of the same type.
449 is still valid after the function returned.
455 .It Fn m_split mbuf len how
456 Partition an mbuf chain in two pieces, returning the tail:
460 In case of failure, it returns
462 and attempts to restore the chain to its original state.
463 .It Fn m_unshare mbuf how
464 Create a version of the specified mbuf chain whose
465 contents can be safely modified without affecting other users.
466 If allocation fails and this operation can not be completed,
469 The original mbuf chain is always reclaimed and the reference
470 count of any shared mbuf clusters is decremented.
471 As a side-effect of this process the returned
472 mbuf chain may be compacted.
474 This function is especially useful in the transmit path of
475 network code, when data must be encrypted or otherwise
476 altered prior to transmission.
480 When running a kernel compiled with the option
481 .Dv MBUF_STRESS_TEST ,
484 -controlled options may be used to create
485 various failure/extreme cases for testing of network drivers
486 and other parts of the kernel that rely on
488 .Bl -tag -width ident
489 .It Va net.inet.ip.mbuf_frag_size
494 into fragments of the specified size.
495 Setting this variable to 1 is an excellent way to
498 handling ability of network drivers.
499 .It Va kern.ipc.m_defragrandomfailures
502 to randomly fail, returning
504 Any piece of code which uses
506 should be tested with this feature.
511 .\" Please correct me if I'm wrong
512 Mbufs appeared in an early version of
514 Besides for being used for network packets, they were used
515 to store various dynamic structures, such as routing table
516 entries, interface addresses, protocol control blocks, etc.
520 man page was written by Yar Tikhiy.