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_args)"
53 .Fn MEXTFREE "struct mbuf *mbuf"
54 .Fn MFREE "struct mbuf *mbuf" "struct mbuf *successor"
56 .Ss Mbuf utility macros
57 .Fn mtod "struct mbuf *mbuf" "type"
58 .Fn M_ALIGN "struct mbuf *mbuf" "u_int len"
59 .Fn MH_ALIGN "struct mbuf *mbuf" "u_int len"
61 .Fn M_LEADINGSPACE "struct mbuf *mbuf"
63 .Fn M_TRAILINGSPACE "struct mbuf *mbuf"
64 .Fn M_MOVE_PKTHDR "struct mbuf *to" "struct mbuf *from"
65 .Fn M_PREPEND "struct mbuf *mbuf" "int len" "int how"
66 .Fn MCHTYPE "struct mbuf *mbuf" "u_int type"
68 .Fn M_WRITABLE "struct mbuf *mbuf"
70 .Ss Mbuf allocation functions
72 .Fn m_get "int how" "int type"
74 .Fn m_getm "struct mbuf *orig" "int len" "int how" "int type"
76 .Fn m_getcl "int how" "short type" "int flags"
78 .Fn m_getclr "int how" "int type"
80 .Fn m_gethdr "int how" "int type"
82 .Fn m_free "struct mbuf *mbuf"
84 .Fn m_freem "struct mbuf *mbuf"
86 .Ss Mbuf utility functions
88 .Fn m_adj "struct mbuf *mbuf" "int len"
90 .Fn m_align "struct mbuf *mbuf" "int len"
92 .Fn m_append "struct mbuf *mbuf" "int len" "c_caddr_t cp"
94 .Fn m_prepend "struct mbuf *mbuf" "int len" "int how"
96 .Fn m_copyup "struct mbuf *mbuf" "int len" "int dstoff"
98 .Fn m_pullup "struct mbuf *mbuf" "int len"
100 .Fn m_pulldown "struct mbuf *mbuf" "int offset" "int len" "int *offsetp"
102 .Fn m_copym "struct mbuf *mbuf" "int offset" "int len" "int how"
104 .Fn m_copypacket "struct mbuf *mbuf" "int how"
106 .Fn m_dup "struct mbuf *mbuf" "int how"
108 .Fn m_copydata "const struct mbuf *mbuf" "int offset" "int len" "caddr_t buf"
110 .Fn m_copyback "struct mbuf *mbuf" "int offset" "int len" "caddr_t buf"
116 .Fa "struct ifnet *ifp"
117 .Fa "void (*copy)(char *from, caddr_t to, u_int len)"
120 .Fn m_cat "struct mbuf *m" "struct mbuf *n"
122 .Fn m_fixhdr "struct mbuf *mbuf"
124 .Fn m_dup_pkthdr "struct mbuf *to" "struct mbuf *from"
126 .Fn m_move_pkthdr "struct mbuf *to" "struct mbuf *from"
128 .Fn m_length "struct mbuf *mbuf" "struct mbuf **last"
130 .Fn m_split "struct mbuf *mbuf" "int len" "int how"
132 .Fn m_apply "struct mbuf *mbuf" "int off" "int len" "int (*f)(void *arg, void *data, u_int len)" "void *arg"
134 .Fn m_getptr "struct mbuf *mbuf" "int loc" "int *off"
136 .Fn m_defrag "struct mbuf *m0" "int how"
138 .Fn m_unshare "struct mbuf *m0" "int how"
143 is a basic unit of memory management in the kernel IPC subsystem.
144 Network packets and socket buffers are stored in
146 A network packet may span multiple
151 which allows adding or trimming
152 network headers with little overhead.
154 While a developer should not bother with
156 internals without serious
157 reason in order to avoid incompatibilities with future changes, it
158 is useful to understand the general structure of an
163 consists of a variable-sized header and a small internal
168 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 */
243 .Vt struct pkthdr Va m_pkthdr
247 It contains a pointer to the interface
248 the packet has been received from
249 .Pq Vt struct ifnet Va *rcvif ,
250 and the total packet length
252 Optionally, it may also contain an attached list of packet tags
253 .Pq Vt "struct m_tag" .
257 Fields used in offloading checksum calculation to the hardware are kept in
261 .Sx HARDWARE-ASSISTED CHECKSUM CALCULATION
264 If small enough, data is stored in the internal data buffer of an
266 If the data is sufficiently large, another
270 or external storage may be associated with the
273 bytes of data can fit into an
281 If external storage is being associated with an
285 header is added at the cost of losing the internal data buffer.
286 It includes a pointer to external storage, the size of the storage,
287 a pointer to a function used for freeing the storage,
288 a pointer to an optional argument that can be passed to the function,
289 and a pointer to a reference counter.
292 using external storage has the
296 The system supplies a macro for allocating the desired external storage
300 The allocation and management of the reference counter is handled by the
303 The system also supplies a default type of external storage buffer called an
306 can be allocated and configured with the use of the
313 in size, where MCLBYTES is a machine-dependent constant.
314 The system defines an advisory macro
316 which is the smallest amount of data to put into an
318 It is equal to the sum of
322 It is typically preferable to store data into the data region of an
324 if size permits, as opposed to allocating a separate
326 to hold the same data.
328 .Ss Macros and Functions
329 There are numerous predefined macros and functions that provide the
330 developer with common utilities.
332 .Bl -ohang -offset indent
333 .It Fn mtod mbuf type
336 pointer to a data pointer.
337 The macro expands to the data pointer cast to the pointer of the specified
340 It is advisable to ensure that there is enough contiguous data in
345 .It Fn MGET mbuf how type
348 and initialize it to contain internal data.
350 will point to the allocated
352 on success, or be set to
357 argument is to be set to
361 It specifies whether the caller is willing to block if necessary.
366 a failed allocation will result in the caller being put
367 to sleep for a designated
372 A number of other functions and macros related to
374 have the same argument because they may
375 at some point need to allocate new
378 Programmers should be careful not to confuse the
386 They are not the same.
387 .It Fn MGETHDR mbuf how type
390 and initialize it to contain a packet header
395 .It Fn MCLGET mbuf how
396 Allocate and attach an
400 If the macro fails, the
402 flag will not be set in
404 .It Fn M_ALIGN mbuf len
407 to place an object of the size
409 at the end of the internal data area of
414 is newly allocated with
418 .It Fn MH_ALIGN mbuf len
419 Serves the same purpose as
431 .It Fn m_align mbuf len
432 Services the same purpose as
434 but handles any type of mbuf.
435 .It Fn M_LEADINGSPACE mbuf
436 Returns the number of bytes available before the beginning
439 .It Fn M_TRAILINGSPACE mbuf
440 Returns the number of bytes available after the end of data in
442 .It Fn M_PREPEND mbuf len how
443 This macro operates on an
445 It is an optimized wrapper for
447 that can make use of possible empty space before data
448 (e.g.\& left after trimming of a link-layer header).
456 .It Fn M_MOVE_PKTHDR to from
457 Using this macro is equivalent to calling
458 .Fn m_move_pkthdr to from .
459 .It Fn M_WRITABLE mbuf
460 This macro will evaluate true if
466 does not contain external storage or,
468 then if the reference count of the storage is not greater than 1.
473 This can be achieved during setup of the external storage,
480 macro, or can be directly set in individual
482 .It Fn MCHTYPE mbuf type
487 This is a relatively expensive operation and should be avoided.
491 .Bl -ohang -offset indent
492 .It Fn m_get how type
493 A function version of
495 for non-critical paths.
496 .It Fn m_getm orig len how type
503 if necessary and append the resulting allocated
509 .No non- Ns Dv NULL .
510 If the allocation fails at any point,
511 free whatever was allocated and return
516 .No non- Ns Dv NULL ,
517 it will not be freed.
518 It is possible to use
526 (for example, one which may be sitting in a pre-allocated ring)
527 or to simply perform an all-or-nothing
532 .It Fn m_gethdr how type
533 A function version of
535 for non-critical paths.
536 .It Fn m_getcl how type flags
542 If one of the allocations fails, the entire allocation fails.
543 This routine is the preferred way of fetching both the
547 together, as it avoids having to unlock/relock between allocations.
551 .It Fn m_getclr how type
554 and zero out the data region.
564 The functions below operate on
566 .Bl -ohang -offset indent
570 including any external storage.
572 .It Fn m_adj mbuf len
575 bytes from the head of an
579 is positive, from the tail otherwise.
581 .It Fn m_append mbuf len cp
588 Extend the mbuf chain if the new data does not fit in
591 .It Fn m_prepend mbuf len how
594 and prepend it to the
600 It does not allocate any
612 .It Fn m_copyup mbuf len dstoff
617 bytes of data into a new mbuf at
622 argument aligns the data and leaves room for a link layer header.
632 The function does not allocate
639 .It Fn m_pullup mbuf len
640 Arrange that the first
644 are contiguous and lay in the data area of
646 so they are accessible with
648 It is important to remember that this may involve
649 reallocating some mbufs and moving data so all pointers
650 referencing data within the old mbuf chain
651 must be recalculated or made invalid.
659 is freed in this case).
661 It does not allocate any
668 .It Fn m_pulldown mbuf offset len offsetp
677 are contiguous and lay in the data area of
679 so they are accessible with
681 .Fa len must be smaller than, or equal to, the size of an
683 Return a pointer to an intermediate
685 in the chain containing the requested region;
686 the offset in the data region of the
688 to the data contained in the returned mbuf is stored in
692 is NULL, the region may be accessed using
696 is non-NULL, the region may be accessed using
697 .Fn mtod mbuf uint8_t + *offsetp .
698 The region of the mbuf chain between its beginning and
700 is not modified, therefore it is safe to hold pointers to data within
701 this region before calling
704 .It Fn m_copym mbuf offset len how
709 bytes from the beginning, continuing for
716 copy to the end of the
719 The copy is read-only, because the
721 are not copied, only their reference counts are incremented.
723 .It Fn m_copypacket mbuf how
724 Copy an entire packet including header, which must be present.
725 This is an optimized version of the common case
726 .Fn m_copym mbuf 0 M_COPYALL how .
728 the copy is read-only, because the
730 are not copied, only their reference counts are incremented.
732 .It Fn m_dup mbuf how
735 into a completely new
737 including copying any
741 when you need a writable copy of an
744 .It Fn m_copydata mbuf offset len buf
749 bytes from the beginning, continuing for
751 bytes, into the indicated buffer
754 .It Fn m_copyback mbuf offset len buf
757 bytes from the buffer
759 back into the indicated
763 bytes from the beginning of the
769 It does not allocate any
781 will be allocated to fill the space.
783 .It Fn m_length mbuf last
784 Return the length of the
786 and optionally a pointer to the last
789 .It Fn m_dup_pkthdr to from how
790 Upon the function's completion, the
793 will contain an identical copy of
795 and the per-packet attributes found in the
805 must be empty on entry.
807 .It Fn m_move_pkthdr to from
810 and the per-packet attributes from the
823 must be empty on entry.
824 Upon the function's completion,
828 and the per-packet attributes cleared.
831 Set the packet-header length to the length of the
834 .It Fn m_devget buf len offset ifp copy
835 Copy data from a device local memory pointed to by
839 The copy is done using a specified copy routine
855 must be of the same type.
857 is still valid after the function returned.
863 .It Fn m_split mbuf len how
866 in two pieces, returning the tail:
870 In case of failure, it returns
872 and attempts to restore the
874 to its original state.
876 .It Fn m_apply mbuf off len f arg
877 Apply a function to an
884 Typically used to avoid calls to
886 which would otherwise be unnecessary or undesirable.
888 is a convenience argument which is passed to the callback function
893 is called, it will be passed
897 in the current mbuf, and the length
899 of the data in this mbuf to which the function should be applied.
901 The function should return zero to indicate success;
902 otherwise, if an error is indicated, then
904 will return the error and stop iterating through the
907 .It Fn m_getptr mbuf loc off
908 Return a pointer to the mbuf containing the data located at
910 bytes from the beginning of the
912 The corresponding offset into the mbuf will be stored in
914 .It Fn m_defrag m0 how
915 Defragment an mbuf chain, returning the shortest possible
916 chain of mbufs and clusters.
917 If allocation fails and this can not be completed,
919 will be returned and the original chain will be unchanged.
920 Upon success, the original chain will be freed and the new
921 chain will be returned.
927 depending on the caller's preference.
929 This function is especially useful in network drivers, where
930 certain long mbuf chains must be shortened before being added
931 to TX descriptor lists.
932 .It Fn m_unshare m0 how
933 Create a version of the specified mbuf chain whose
934 contents can be safely modified without affecting other users.
935 If allocation fails and this operation can not be completed,
938 The original mbuf chain is always reclaimed and the reference
939 count of any shared mbuf clusters is decremented.
945 depending on the caller's preference.
946 As a side-effect of this process the returned
947 mbuf chain may be compacted.
949 This function is especially useful in the transmit path of
950 network code, when data must be encrypted or otherwise
951 altered prior to transmission.
953 .Sh HARDWARE-ASSISTED CHECKSUM CALCULATION
954 This section currently applies to TCP/IP only.
955 In order to save the host CPU resources, computing checksums is
956 offloaded to the network interface hardware if possible.
959 member of the leading
961 of a packet contains two fields used for that purpose,
962 .Vt int Va csum_flags
964 .Vt int Va csum_data .
965 The meaning of those fields depends on the direction a packet flows in,
966 and on whether the packet is fragmented.
972 will denote the corresponding field of the
974 member of the leading
978 containing the packet.
980 On output, checksum offloading is attempted after the outgoing
981 interface has been determined for a packet.
982 The interface-specific field
983 .Va ifnet.if_data.ifi_hwassist
986 is consulted for the capabilities of the interface to assist in
990 field of the packet header is set to indicate which actions the interface
991 is supposed to perform on it.
992 The actions unsupported by the network interface are done in the
993 software prior to passing the packet down to the interface driver;
994 such actions will never be requested through
997 The flags demanding a particular action from an interface are as follows:
998 .Bl -tag -width ".Dv CSUM_TCP" -offset indent
1000 The IP header checksum is to be computed and stored in the
1001 corresponding field of the packet.
1002 The hardware is expected to know the format of an IP header
1003 to determine the offset of the IP checksum field.
1005 The TCP checksum is to be computed.
1008 The UDP checksum is to be computed.
1012 Should a TCP or UDP checksum be offloaded to the hardware,
1015 will contain the byte offset of the checksum field relative to the
1016 end of the IP header.
1017 In this case, the checksum field will be initially
1018 set by the TCP/IP module to the checksum of the pseudo header
1019 defined by the TCP and UDP specifications.
1021 For outbound packets which have been fragmented
1022 by the host CPU, the following will also be true,
1023 regardless of the checksum flag settings:
1024 .Bl -bullet -offset indent
1026 all fragments will have the flag
1032 the first and the last fragments in the chain will have
1040 the first fragment in the chain will have the total number
1041 of fragments contained in its
1046 The last rule for fragmented packets takes precedence over the one
1047 for a TCP or UDP checksum.
1048 Nevertheless, offloading a TCP or UDP checksum is possible for a
1049 fragmented packet if the flag
1052 .Va ifnet.if_data.ifi_hwassist
1053 associated with the network interface.
1054 However, in this case the interface is expected to figure out
1055 the location of the checksum field within the sequence of fragments
1058 contains a fragment count instead of a checksum offset value.
1060 On input, an interface indicates the actions it has performed
1061 on a packet by setting one or more of the following flags in
1063 associated with the packet:
1064 .Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent
1065 .It Dv CSUM_IP_CHECKED
1066 The IP header checksum has been computed.
1067 .It Dv CSUM_IP_VALID
1068 The IP header has a valid checksum.
1069 This flag can appear only in combination with
1070 .Dv CSUM_IP_CHECKED .
1071 .It Dv CSUM_DATA_VALID
1072 The checksum of the data portion of the IP packet has been computed
1073 and stored in the field
1075 in network byte order.
1076 .It Dv CSUM_PSEUDO_HDR
1077 Can be set only along with
1079 to indicate that the IP data checksum found in
1081 allows for the pseudo header defined by the TCP and UDP specifications.
1082 Otherwise the checksum of the pseudo header must be calculated by
1083 the host CPU and added to
1085 to obtain the final checksum to be used for TCP or UDP validation purposes.
1088 If a particular network interface just indicates success or
1089 failure of TCP or UDP checksum validation without returning
1090 the exact value of the checksum to the host CPU, its driver can mark
1100 hexadecimal to indicate a valid checksum.
1101 It is a peculiarity of the algorithm used that the Internet checksum
1102 calculated over any valid packet will be
1104 as long as the original checksum field is included.
1106 For inbound packets which are IP fragments, all
1108 fields will be summed during reassembly to obtain the final checksum
1109 value passed to an upper layer in the
1111 field of the reassembled packet.
1114 fields of all fragments will be consolidated using logical AND
1115 to obtain the final value for
1117 Thus, in order to successfully
1118 offload checksum computation for fragmented data,
1119 all fragments should have the same value of
1122 When running a kernel compiled with the option
1123 .Dv MBUF_STRESS_TEST ,
1126 -controlled options may be used to create
1127 various failure/extreme cases for testing of network drivers
1128 and other parts of the kernel that rely on
1130 .Bl -tag -width ident
1131 .It Va net.inet.ip.mbuf_frag_size
1134 to fragment outgoing
1136 into fragments of the specified size.
1137 Setting this variable to 1 is an excellent way to
1140 handling ability of network drivers.
1141 .It Va kern.ipc.m_defragrandomfailures
1144 to randomly fail, returning
1146 Any piece of code which uses
1148 should be tested with this feature.
1156 .\" Please correct me if I'm wrong
1158 appeared in an early version of
1160 Besides being used for network packets, they were used
1161 to store various dynamic structures, such as routing table
1162 entries, interface addresses, protocol control blocks, etc.
1167 is almost entirely limited to packet storage, with
1169 zones being used directly to store other network-related memory.
1173 allocator has been a special-purpose memory allocator able to run in
1174 interrupt contexts and allocating from a special kernel address space map.
1179 allocator is a wrapper around
1185 + cluster pairs in per-CPU caches, as well as bringing other benefits of
1190 manual page was written by Yar Tikhiy.
1194 allocator was written by Bosko Milekic.
projects / FreeBSD / releng / 7.2.git / blob