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
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 .Fa "struct mbuf *mbuf"
48 .Fa "void (*free)(void *opt_arg1, void *opt_arg2)"
54 .Fn MEXTFREE "struct mbuf *mbuf"
55 .Fn MFREE "struct mbuf *mbuf" "struct mbuf *successor"
57 .Ss Mbuf utility macros
58 .Fn mtod "struct mbuf *mbuf" "type"
59 .Fn M_ALIGN "struct mbuf *mbuf" "u_int len"
60 .Fn MH_ALIGN "struct mbuf *mbuf" "u_int len"
62 .Fn M_LEADINGSPACE "struct mbuf *mbuf"
64 .Fn M_TRAILINGSPACE "struct mbuf *mbuf"
65 .Fn M_MOVE_PKTHDR "struct mbuf *to" "struct mbuf *from"
66 .Fn M_PREPEND "struct mbuf *mbuf" "int len" "int how"
67 .Fn MCHTYPE "struct mbuf *mbuf" "u_int type"
69 .Fn M_WRITABLE "struct mbuf *mbuf"
71 .Ss Mbuf allocation functions
73 .Fn m_get "int how" "int type"
75 .Fn m_getm "struct mbuf *orig" "int len" "int how" "int type"
77 .Fn m_getcl "int how" "short type" "int flags"
79 .Fn m_getclr "int how" "int type"
81 .Fn m_gethdr "int how" "int type"
83 .Fn m_free "struct mbuf *mbuf"
85 .Fn m_freem "struct mbuf *mbuf"
87 .Ss Mbuf utility functions
89 .Fn m_adj "struct mbuf *mbuf" "int len"
91 .Fn m_align "struct mbuf *mbuf" "int len"
93 .Fn m_append "struct mbuf *mbuf" "int len" "c_caddr_t cp"
95 .Fn m_prepend "struct mbuf *mbuf" "int len" "int how"
97 .Fn m_copyup "struct mbuf *mbuf" "int len" "int dstoff"
99 .Fn m_pullup "struct mbuf *mbuf" "int len"
101 .Fn m_pulldown "struct mbuf *mbuf" "int offset" "int len" "int *offsetp"
103 .Fn m_copym "struct mbuf *mbuf" "int offset" "int len" "int how"
105 .Fn m_copypacket "struct mbuf *mbuf" "int how"
107 .Fn m_dup "struct mbuf *mbuf" "int how"
109 .Fn m_copydata "const struct mbuf *mbuf" "int offset" "int len" "caddr_t buf"
111 .Fn m_copyback "struct mbuf *mbuf" "int offset" "int len" "caddr_t buf"
117 .Fa "struct ifnet *ifp"
118 .Fa "void (*copy)(char *from, caddr_t to, u_int len)"
121 .Fn m_cat "struct mbuf *m" "struct mbuf *n"
123 .Fn m_fixhdr "struct mbuf *mbuf"
125 .Fn m_dup_pkthdr "struct mbuf *to" "struct mbuf *from"
127 .Fn m_move_pkthdr "struct mbuf *to" "struct mbuf *from"
129 .Fn m_length "struct mbuf *mbuf" "struct mbuf **last"
131 .Fn m_split "struct mbuf *mbuf" "int len" "int how"
133 .Fn m_apply "struct mbuf *mbuf" "int off" "int len" "int (*f)(void *arg, void *data, u_int len)" "void *arg"
135 .Fn m_getptr "struct mbuf *mbuf" "int loc" "int *off"
137 .Fn m_defrag "struct mbuf *m0" "int how"
139 .Fn m_unshare "struct mbuf *m0" "int how"
144 is a basic unit of memory management in the kernel IPC subsystem.
145 Network packets and socket buffers are stored in
147 A network packet may span multiple
152 which allows adding or trimming
153 network headers with little overhead.
155 While a developer should not bother with
157 internals without serious
158 reason in order to avoid incompatibilities with future changes, it
159 is useful to understand the general structure of an
164 consists of a variable-sized header and a small internal
169 is a constant defined in
174 .Bl -tag -width "m_nextpkt" -offset indent
177 A pointer to the next
183 A pointer to the next
188 A pointer to data attached to this
192 The length of the data.
195 The type of the data.
205 flag bits are defined as follows:
208 #define M_EXT 0x0001 /* has associated external storage */
209 #define M_PKTHDR 0x0002 /* start of record */
210 #define M_EOR 0x0004 /* end of record */
211 #define M_RDONLY 0x0008 /* associated data marked read-only */
212 #define M_PROTO1 0x0010 /* protocol-specific */
213 #define M_PROTO2 0x0020 /* protocol-specific */
214 #define M_PROTO3 0x0040 /* protocol-specific */
215 #define M_PROTO4 0x0080 /* protocol-specific */
216 #define M_PROTO5 0x0100 /* protocol-specific */
217 #define M_PROTO6 0x4000 /* protocol-specific (avoid M_BCAST conflict) */
218 #define M_FREELIST 0x8000 /* mbuf is on the free list */
220 /* mbuf pkthdr flags (also stored in m_flags) */
221 #define M_BCAST 0x0200 /* send/received as link-level broadcast */
222 #define M_MCAST 0x0400 /* send/received as link-level multicast */
223 #define M_FRAG 0x0800 /* packet is fragment of larger packet */
224 #define M_FIRSTFRAG 0x1000 /* packet is first fragment */
225 #define M_LASTFRAG 0x2000 /* packet is last fragment */
230 types are defined as follows:
233 #define MT_DATA 1 /* dynamic (data) allocation */
234 #define MT_HEADER MT_DATA /* packet header */
235 #define MT_SONAME 8 /* socket name */
236 #define MT_CONTROL 14 /* extra-data protocol message */
237 #define MT_OOBDATA 15 /* expedited data */
240 The available external buffer types are defined as follows:
242 /* external buffer types */
243 #define EXT_CLUSTER 1 /* mbuf cluster */
244 #define EXT_SFBUF 2 /* sendfile(2)'s sf_bufs */
245 #define EXT_JUMBOP 3 /* jumbo cluster 4096 bytes */
246 #define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */
247 #define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */
248 #define EXT_PACKET 6 /* mbuf+cluster from packet zone */
249 #define EXT_MBUF 7 /* external mbuf reference (M_IOVEC) */
250 #define EXT_NET_DRV 100 /* custom ext_buf provided by net driver(s) */
251 #define EXT_MOD_TYPE 200 /* custom module's ext_buf type */
252 #define EXT_DISPOSABLE 300 /* can throw this buffer away w/page flipping */
253 #define EXT_EXTREF 400 /* has externally maintained ref_cnt ptr */
259 .Vt struct pkthdr Va m_pkthdr
263 It contains a pointer to the interface
264 the packet has been received from
265 .Pq Vt struct ifnet Va *rcvif ,
266 and the total packet length
268 Optionally, it may also contain an attached list of packet tags
269 .Pq Vt "struct m_tag" .
273 Fields used in offloading checksum calculation to the hardware are kept in
277 .Sx HARDWARE-ASSISTED CHECKSUM CALCULATION
280 If small enough, data is stored in the internal data buffer of an
282 If the data is sufficiently large, another
286 or external storage may be associated with the
289 bytes of data can fit into an
297 If external storage is being associated with an
301 header is added at the cost of losing the internal data buffer.
302 It includes a pointer to external storage, the size of the storage,
303 a pointer to a function used for freeing the storage,
304 a pointer to an optional argument that can be passed to the function,
305 and a pointer to a reference counter.
308 using external storage has the
312 The system supplies a macro for allocating the desired external storage
316 The allocation and management of the reference counter is handled by the
319 The system also supplies a default type of external storage buffer called an
322 can be allocated and configured with the use of the
329 in size, where MCLBYTES is a machine-dependent constant.
330 The system defines an advisory macro
332 which is the smallest amount of data to put into an
334 It is equal to the sum of
338 It is typically preferable to store data into the data region of an
340 if size permits, as opposed to allocating a separate
342 to hold the same data.
344 .Ss Macros and Functions
345 There are numerous predefined macros and functions that provide the
346 developer with common utilities.
348 .Bl -ohang -offset indent
349 .It Fn mtod mbuf type
352 pointer to a data pointer.
353 The macro expands to the data pointer cast to the pointer of the specified
356 It is advisable to ensure that there is enough contiguous data in
361 .It Fn MGET mbuf how type
364 and initialize it to contain internal data.
366 will point to the allocated
368 on success, or be set to
373 argument is to be set to
377 It specifies whether the caller is willing to block if necessary.
378 A number of other functions and macros related to
380 have the same argument because they may
381 at some point need to allocate new
384 Programmers should be careful not to confuse the
392 They are not the same.
393 .It Fn MGETHDR mbuf how type
396 and initialize it to contain a packet header
401 .It Fn MEXTADD mbuf buf size free opt_arg1 opt_arg2 flags type
402 Associate externally managed data with
404 Any internal data contained in the mbuf will be discarded, and the
405 .Dv M_EXT flag will be set.
410 arguments are the address and length, respectively, of the data.
413 argument points to a function which will be called to free the data
414 when the mbuf is freed; it is only used if
422 arguments will be passed unmodified to
426 argument specifies additional
428 flags; it is not necessary to specify
432 argument specifies the type of external data, which controls how it
433 will be disposed of when the
436 In most cases, the correct value is
438 .It Fn MCLGET mbuf how
439 Allocate and attach an
443 If the macro fails, the
445 flag will not be set in
447 .It Fn M_ALIGN mbuf len
450 to place an object of the size
452 at the end of the internal data area of
457 is newly allocated with
461 .It Fn MH_ALIGN mbuf len
462 Serves the same purpose as
474 .It Fn m_align mbuf len
475 Services the same purpose as
477 but handles any type of mbuf.
478 .It Fn M_LEADINGSPACE mbuf
479 Returns the number of bytes available before the beginning
482 .It Fn M_TRAILINGSPACE mbuf
483 Returns the number of bytes available after the end of data in
485 .It Fn M_PREPEND mbuf len how
486 This macro operates on an
488 It is an optimized wrapper for
490 that can make use of possible empty space before data
491 (e.g.\& left after trimming of a link-layer header).
499 .It Fn M_MOVE_PKTHDR to from
500 Using this macro is equivalent to calling
501 .Fn m_move_pkthdr to from .
502 .It Fn M_WRITABLE mbuf
503 This macro will evaluate true if
509 does not contain external storage or,
511 then if the reference count of the storage is not greater than 1.
516 This can be achieved during setup of the external storage,
523 macro, or can be directly set in individual
525 .It Fn MCHTYPE mbuf type
530 This is a relatively expensive operation and should be avoided.
534 .Bl -ohang -offset indent
535 .It Fn m_get how type
536 A function version of
538 for non-critical paths.
539 .It Fn m_getm orig len how type
546 if necessary and append the resulting allocated
552 .No non- Ns Dv NULL .
553 If the allocation fails at any point,
554 free whatever was allocated and return
559 .No non- Ns Dv NULL ,
560 it will not be freed.
561 It is possible to use
569 (for example, one which may be sitting in a pre-allocated ring)
570 or to simply perform an all-or-nothing
575 .It Fn m_gethdr how type
576 A function version of
578 for non-critical paths.
579 .It Fn m_getcl how type flags
585 If one of the allocations fails, the entire allocation fails.
586 This routine is the preferred way of fetching both the
590 together, as it avoids having to unlock/relock between allocations.
594 .It Fn m_getclr how type
597 and zero out the data region.
607 The functions below operate on
609 .Bl -ohang -offset indent
613 including any external storage.
615 .It Fn m_adj mbuf len
618 bytes from the head of an
622 is positive, from the tail otherwise.
624 .It Fn m_append mbuf len cp
631 Extend the mbuf chain if the new data does not fit in
634 .It Fn m_prepend mbuf len how
637 and prepend it to the
643 It does not allocate any
655 .It Fn m_copyup mbuf len dstoff
660 bytes of data into a new mbuf at
665 argument aligns the data and leaves room for a link layer header.
675 The function does not allocate
682 .It Fn m_pullup mbuf len
683 Arrange that the first
687 are contiguous and lay in the data area of
689 so they are accessible with
691 It is important to remember that this may involve
692 reallocating some mbufs and moving data so all pointers
693 referencing data within the old mbuf chain
694 must be recalculated or made invalid.
702 is freed in this case).
704 It does not allocate any
711 .It Fn m_pulldown mbuf offset len offsetp
720 are contiguous and lay in the data area of
722 so they are accessible with
724 .Fa len must be smaller than, or equal to, the size of an
726 Return a pointer to an intermediate
728 in the chain containing the requested region;
729 the offset in the data region of the
731 to the data contained in the returned mbuf is stored in
735 is NULL, the region may be accessed using
739 is non-NULL, the region may be accessed using
740 .Fn mtod mbuf uint8_t + *offsetp .
741 The region of the mbuf chain between its beginning and
743 is not modified, therefore it is safe to hold pointers to data within
744 this region before calling
747 .It Fn m_copym mbuf offset len how
752 bytes from the beginning, continuing for
759 copy to the end of the
762 The copy is read-only, because the
764 are not copied, only their reference counts are incremented.
766 .It Fn m_copypacket mbuf how
767 Copy an entire packet including header, which must be present.
768 This is an optimized version of the common case
769 .Fn m_copym mbuf 0 M_COPYALL how .
771 the copy is read-only, because the
773 are not copied, only their reference counts are incremented.
775 .It Fn m_dup mbuf how
778 into a completely new
780 including copying any
784 when you need a writable copy of an
787 .It Fn m_copydata mbuf offset len buf
792 bytes from the beginning, continuing for
794 bytes, into the indicated buffer
797 .It Fn m_copyback mbuf offset len buf
800 bytes from the buffer
802 back into the indicated
806 bytes from the beginning of the
812 It does not allocate any
824 will be allocated to fill the space.
826 .It Fn m_length mbuf last
827 Return the length of the
829 and optionally a pointer to the last
832 .It Fn m_dup_pkthdr to from how
833 Upon the function's completion, the
836 will contain an identical copy of
838 and the per-packet attributes found in the
848 must be empty on entry.
850 .It Fn m_move_pkthdr to from
853 and the per-packet attributes from the
866 must be empty on entry.
867 Upon the function's completion,
871 and the per-packet attributes cleared.
874 Set the packet-header length to the length of the
877 .It Fn m_devget buf len offset ifp copy
878 Copy data from a device local memory pointed to by
882 The copy is done using a specified copy routine
898 must be of the same type.
900 is still valid after the function returned.
906 .It Fn m_split mbuf len how
909 in two pieces, returning the tail:
913 In case of failure, it returns
915 and attempts to restore the
917 to its original state.
919 .It Fn m_apply mbuf off len f arg
920 Apply a function to an
927 Typically used to avoid calls to
929 which would otherwise be unnecessary or undesirable.
931 is a convenience argument which is passed to the callback function
936 is called, it will be passed
940 in the current mbuf, and the length
942 of the data in this mbuf to which the function should be applied.
944 The function should return zero to indicate success;
945 otherwise, if an error is indicated, then
947 will return the error and stop iterating through the
950 .It Fn m_getptr mbuf loc off
951 Return a pointer to the mbuf containing the data located at
953 bytes from the beginning of the
955 The corresponding offset into the mbuf will be stored in
957 .It Fn m_defrag m0 how
958 Defragment an mbuf chain, returning the shortest possible
959 chain of mbufs and clusters.
960 If allocation fails and this can not be completed,
962 will be returned and the original chain will be unchanged.
963 Upon success, the original chain will be freed and the new
964 chain will be returned.
970 depending on the caller's preference.
972 This function is especially useful in network drivers, where
973 certain long mbuf chains must be shortened before being added
974 to TX descriptor lists.
975 .It Fn m_unshare m0 how
976 Create a version of the specified mbuf chain whose
977 contents can be safely modified without affecting other users.
978 If allocation fails and this operation can not be completed,
981 The original mbuf chain is always reclaimed and the reference
982 count of any shared mbuf clusters is decremented.
988 depending on the caller's preference.
989 As a side-effect of this process the returned
990 mbuf chain may be compacted.
992 This function is especially useful in the transmit path of
993 network code, when data must be encrypted or otherwise
994 altered prior to transmission.
996 .Sh HARDWARE-ASSISTED CHECKSUM CALCULATION
997 This section currently applies to TCP/IP only.
998 In order to save the host CPU resources, computing checksums is
999 offloaded to the network interface hardware if possible.
1002 member of the leading
1004 of a packet contains two fields used for that purpose,
1005 .Vt int Va csum_flags
1007 .Vt int Va csum_data .
1008 The meaning of those fields depends on the direction a packet flows in,
1009 and on whether the packet is fragmented.
1015 will denote the corresponding field of the
1017 member of the leading
1021 containing the packet.
1023 On output, checksum offloading is attempted after the outgoing
1024 interface has been determined for a packet.
1025 The interface-specific field
1026 .Va ifnet.if_data.ifi_hwassist
1029 is consulted for the capabilities of the interface to assist in
1030 computing checksums.
1033 field of the packet header is set to indicate which actions the interface
1034 is supposed to perform on it.
1035 The actions unsupported by the network interface are done in the
1036 software prior to passing the packet down to the interface driver;
1037 such actions will never be requested through
1040 The flags demanding a particular action from an interface are as follows:
1041 .Bl -tag -width ".Dv CSUM_TCP" -offset indent
1043 The IP header checksum is to be computed and stored in the
1044 corresponding field of the packet.
1045 The hardware is expected to know the format of an IP header
1046 to determine the offset of the IP checksum field.
1048 The TCP checksum is to be computed.
1051 The UDP checksum is to be computed.
1055 Should a TCP or UDP checksum be offloaded to the hardware,
1058 will contain the byte offset of the checksum field relative to the
1059 end of the IP header.
1060 In this case, the checksum field will be initially
1061 set by the TCP/IP module to the checksum of the pseudo header
1062 defined by the TCP and UDP specifications.
1064 For outbound packets which have been fragmented
1065 by the host CPU, the following will also be true,
1066 regardless of the checksum flag settings:
1067 .Bl -bullet -offset indent
1069 all fragments will have the flag
1075 the first and the last fragments in the chain will have
1083 the first fragment in the chain will have the total number
1084 of fragments contained in its
1089 The last rule for fragmented packets takes precedence over the one
1090 for a TCP or UDP checksum.
1091 Nevertheless, offloading a TCP or UDP checksum is possible for a
1092 fragmented packet if the flag
1095 .Va ifnet.if_data.ifi_hwassist
1096 associated with the network interface.
1097 However, in this case the interface is expected to figure out
1098 the location of the checksum field within the sequence of fragments
1101 contains a fragment count instead of a checksum offset value.
1103 On input, an interface indicates the actions it has performed
1104 on a packet by setting one or more of the following flags in
1106 associated with the packet:
1107 .Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent
1108 .It Dv CSUM_IP_CHECKED
1109 The IP header checksum has been computed.
1110 .It Dv CSUM_IP_VALID
1111 The IP header has a valid checksum.
1112 This flag can appear only in combination with
1113 .Dv CSUM_IP_CHECKED .
1114 .It Dv CSUM_DATA_VALID
1115 The checksum of the data portion of the IP packet has been computed
1116 and stored in the field
1118 in network byte order.
1119 .It Dv CSUM_PSEUDO_HDR
1120 Can be set only along with
1122 to indicate that the IP data checksum found in
1124 allows for the pseudo header defined by the TCP and UDP specifications.
1125 Otherwise the checksum of the pseudo header must be calculated by
1126 the host CPU and added to
1128 to obtain the final checksum to be used for TCP or UDP validation purposes.
1131 If a particular network interface just indicates success or
1132 failure of TCP or UDP checksum validation without returning
1133 the exact value of the checksum to the host CPU, its driver can mark
1143 hexadecimal to indicate a valid checksum.
1144 It is a peculiarity of the algorithm used that the Internet checksum
1145 calculated over any valid packet will be
1147 as long as the original checksum field is included.
1149 For inbound packets which are IP fragments, all
1151 fields will be summed during reassembly to obtain the final checksum
1152 value passed to an upper layer in the
1154 field of the reassembled packet.
1157 fields of all fragments will be consolidated using logical AND
1158 to obtain the final value for
1160 Thus, in order to successfully
1161 offload checksum computation for fragmented data,
1162 all fragments should have the same value of
1165 When running a kernel compiled with the option
1166 .Dv MBUF_STRESS_TEST ,
1169 -controlled options may be used to create
1170 various failure/extreme cases for testing of network drivers
1171 and other parts of the kernel that rely on
1173 .Bl -tag -width ident
1174 .It Va net.inet.ip.mbuf_frag_size
1177 to fragment outgoing
1179 into fragments of the specified size.
1180 Setting this variable to 1 is an excellent way to
1183 handling ability of network drivers.
1184 .It Va kern.ipc.m_defragrandomfailures
1187 to randomly fail, returning
1189 Any piece of code which uses
1191 should be tested with this feature.
1199 .\" Please correct me if I'm wrong
1201 appeared in an early version of
1203 Besides being used for network packets, they were used
1204 to store various dynamic structures, such as routing table
1205 entries, interface addresses, protocol control blocks, etc.
1210 is almost entirely limited to packet storage, with
1212 zones being used directly to store other network-related memory.
1216 allocator has been a special-purpose memory allocator able to run in
1217 interrupt contexts and allocating from a special kernel address space map.
1222 allocator is a wrapper around
1228 + cluster pairs in per-CPU caches, as well as bringing other benefits of
1233 manual page was written by Yar Tikhiy.
1237 allocator was written by Bosko Milekic.