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_collapse "struct mbuf *m0" "int how" "int maxfrags"
141 .Fn m_unshare "struct mbuf *m0" "int how"
146 is a basic unit of memory management in the kernel IPC subsystem.
147 Network packets and socket buffers are stored in
149 A network packet may span multiple
154 which allows adding or trimming
155 network headers with little overhead.
157 While a developer should not bother with
159 internals without serious
160 reason in order to avoid incompatibilities with future changes, it
161 is useful to understand the general structure of an
166 consists of a variable-sized header and a small internal
171 is a constant defined in
176 .Bl -tag -width "m_nextpkt" -offset indent
179 A pointer to the next
185 A pointer to the next
190 A pointer to data attached to this
194 The length of the data.
197 The type of the data.
207 flag bits are defined as follows:
210 #define M_EXT 0x0001 /* has associated external storage */
211 #define M_PKTHDR 0x0002 /* start of record */
212 #define M_EOR 0x0004 /* end of record */
213 #define M_RDONLY 0x0008 /* associated data marked read-only */
214 #define M_PROTO1 0x0010 /* protocol-specific */
215 #define M_PROTO2 0x0020 /* protocol-specific */
216 #define M_PROTO3 0x0040 /* protocol-specific */
217 #define M_PROTO4 0x0080 /* protocol-specific */
218 #define M_PROTO5 0x0100 /* protocol-specific */
219 #define M_PROTO6 0x4000 /* protocol-specific (avoid M_BCAST conflict) */
220 #define M_FREELIST 0x8000 /* mbuf is on the free list */
222 /* mbuf pkthdr flags (also stored in m_flags) */
223 #define M_BCAST 0x0200 /* send/received as link-level broadcast */
224 #define M_MCAST 0x0400 /* send/received as link-level multicast */
225 #define M_FRAG 0x0800 /* packet is fragment of larger packet */
226 #define M_FIRSTFRAG 0x1000 /* packet is first fragment */
227 #define M_LASTFRAG 0x2000 /* packet is last fragment */
232 types are defined as follows:
235 #define MT_DATA 1 /* dynamic (data) allocation */
236 #define MT_HEADER MT_DATA /* packet header */
237 #define MT_SONAME 8 /* socket name */
238 #define MT_CONTROL 14 /* extra-data protocol message */
239 #define MT_OOBDATA 15 /* expedited data */
242 The available external buffer types are defined as follows:
244 /* external buffer types */
245 #define EXT_CLUSTER 1 /* mbuf cluster */
246 #define EXT_SFBUF 2 /* sendfile(2)'s sf_bufs */
247 #define EXT_JUMBOP 3 /* jumbo cluster 4096 bytes */
248 #define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */
249 #define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */
250 #define EXT_PACKET 6 /* mbuf+cluster from packet zone */
251 #define EXT_MBUF 7 /* external mbuf reference (M_IOVEC) */
252 #define EXT_NET_DRV 100 /* custom ext_buf provided by net driver(s) */
253 #define EXT_MOD_TYPE 200 /* custom module's ext_buf type */
254 #define EXT_DISPOSABLE 300 /* can throw this buffer away w/page flipping */
255 #define EXT_EXTREF 400 /* has externally maintained ref_cnt ptr */
261 .Vt struct pkthdr Va m_pkthdr
265 It contains a pointer to the interface
266 the packet has been received from
267 .Pq Vt struct ifnet Va *rcvif ,
268 and the total packet length
270 Optionally, it may also contain an attached list of packet tags
271 .Pq Vt "struct m_tag" .
275 Fields used in offloading checksum calculation to the hardware are kept in
279 .Sx HARDWARE-ASSISTED CHECKSUM CALCULATION
282 If small enough, data is stored in the internal data buffer of an
284 If the data is sufficiently large, another
288 or external storage may be associated with the
291 bytes of data can fit into an
299 If external storage is being associated with an
303 header is added at the cost of losing the internal data buffer.
304 It includes a pointer to external storage, the size of the storage,
305 a pointer to a function used for freeing the storage,
306 a pointer to an optional argument that can be passed to the function,
307 and a pointer to a reference counter.
310 using external storage has the
314 The system supplies a macro for allocating the desired external storage
318 The allocation and management of the reference counter is handled by the
321 The system also supplies a default type of external storage buffer called an
324 can be allocated and configured with the use of the
331 in size, where MCLBYTES is a machine-dependent constant.
332 The system defines an advisory macro
334 which is the smallest amount of data to put into an
339 It is typically preferable to store data into the data region of an
341 if size permits, as opposed to allocating a separate
343 to hold the same data.
345 .Ss Macros and Functions
346 There are numerous predefined macros and functions that provide the
347 developer with common utilities.
349 .Bl -ohang -offset indent
350 .It Fn mtod mbuf type
353 pointer to a data pointer.
354 The macro expands to the data pointer cast to the pointer of the specified
357 It is advisable to ensure that there is enough contiguous data in
362 .It Fn MGET mbuf how type
365 and initialize it to contain internal data.
367 will point to the allocated
369 on success, or be set to
374 argument is to be set to
378 It specifies whether the caller is willing to block if necessary.
379 A number of other functions and macros related to
381 have the same argument because they may
382 at some point need to allocate new
389 section) used allocation flags
393 These constants are kept for compatibility
394 and their use in new code is discouraged.
395 .It Fn MGETHDR mbuf how type
398 and initialize it to contain a packet header
403 .It Fn MEXTADD mbuf buf size free opt_arg1 opt_arg2 flags type
404 Associate externally managed data with
406 Any internal data contained in the mbuf will be discarded, and the
413 arguments are the address and length, respectively, of the data.
416 argument points to a function which will be called to free the data
417 when the mbuf is freed; it is only used if
425 arguments will be passed unmodified to
429 argument specifies additional
431 flags; it is not necessary to specify
435 argument specifies the type of external data, which controls how it
436 will be disposed of when the
439 In most cases, the correct value is
441 .It Fn MCLGET mbuf how
442 Allocate and attach an
446 If the macro fails, the
448 flag will not be set in
450 .It Fn M_ALIGN mbuf len
453 to place an object of the size
455 at the end of the internal data area of
460 is newly allocated with
464 .It Fn MH_ALIGN mbuf len
465 Serves the same purpose as
477 .It Fn m_align mbuf len
478 Services the same purpose as
480 but handles any type of mbuf.
481 .It Fn M_LEADINGSPACE mbuf
482 Returns the number of bytes available before the beginning
485 .It Fn M_TRAILINGSPACE mbuf
486 Returns the number of bytes available after the end of data in
488 .It Fn M_PREPEND mbuf len how
489 This macro operates on an
491 It is an optimized wrapper for
493 that can make use of possible empty space before data
494 (e.g.\& left after trimming of a link-layer header).
502 .It Fn M_MOVE_PKTHDR to from
503 Using this macro is equivalent to calling
504 .Fn m_move_pkthdr to from .
505 .It Fn M_WRITABLE mbuf
506 This macro will evaluate true if
512 does not contain external storage or,
514 then if the reference count of the storage is not greater than 1.
519 This can be achieved during setup of the external storage,
526 macro, or can be directly set in individual
528 .It Fn MCHTYPE mbuf type
533 This is a relatively expensive operation and should be avoided.
537 .Bl -ohang -offset indent
538 .It Fn m_get how type
539 A function version of
541 for non-critical paths.
542 .It Fn m_getm orig len how type
549 if necessary and append the resulting allocated
555 .No non- Ns Dv NULL .
556 If the allocation fails at any point,
557 free whatever was allocated and return
562 .No non- Ns Dv NULL ,
563 it will not be freed.
564 It is possible to use
572 (for example, one which may be sitting in a pre-allocated ring)
573 or to simply perform an all-or-nothing
578 .It Fn m_gethdr how type
579 A function version of
581 for non-critical paths.
582 .It Fn m_getcl how type flags
588 If one of the allocations fails, the entire allocation fails.
589 This routine is the preferred way of fetching both the
593 together, as it avoids having to unlock/relock between allocations.
597 .It Fn m_getclr how type
600 and zero out the data region.
610 The functions below operate on
612 .Bl -ohang -offset indent
616 including any external storage.
618 .It Fn m_adj mbuf len
621 bytes from the head of an
625 is positive, from the tail otherwise.
627 .It Fn m_append mbuf len cp
634 Extend the mbuf chain if the new data does not fit in
637 .It Fn m_prepend mbuf len how
640 and prepend it to the
646 It does not allocate any
658 .It Fn m_copyup mbuf len dstoff
663 bytes of data into a new mbuf at
668 argument aligns the data and leaves room for a link layer header.
678 The function does not allocate
685 .It Fn m_pullup mbuf len
686 Arrange that the first
690 are contiguous and lay in the data area of
692 so they are accessible with
694 It is important to remember that this may involve
695 reallocating some mbufs and moving data so all pointers
696 referencing data within the old mbuf chain
697 must be recalculated or made invalid.
705 is freed in this case).
707 It does not allocate any
711 must be less than or equal to
714 .It Fn m_pulldown mbuf offset len offsetp
723 are contiguous and lay in the data area of
725 so they are accessible with
728 must be smaller than, or equal to, the size of an
730 Return a pointer to an intermediate
732 in the chain containing the requested region;
733 the offset in the data region of the
735 to the data contained in the returned mbuf is stored in
739 is NULL, the region may be accessed using
743 is non-NULL, the region may be accessed using
744 .Fn mtod mbuf uint8_t + *offsetp .
745 The region of the mbuf chain between its beginning and
747 is not modified, therefore it is safe to hold pointers to data within
748 this region before calling
751 .It Fn m_copym mbuf offset len how
756 bytes from the beginning, continuing for
763 copy to the end of the
766 The copy is read-only, because the
768 are not copied, only their reference counts are incremented.
770 .It Fn m_copypacket mbuf how
771 Copy an entire packet including header, which must be present.
772 This is an optimized version of the common case
773 .Fn m_copym mbuf 0 M_COPYALL how .
775 the copy is read-only, because the
777 are not copied, only their reference counts are incremented.
779 .It Fn m_dup mbuf how
782 into a completely new
784 including copying any
788 when you need a writable copy of an
791 .It Fn m_copydata mbuf offset len buf
796 bytes from the beginning, continuing for
798 bytes, into the indicated buffer
801 .It Fn m_copyback mbuf offset len buf
804 bytes from the buffer
806 back into the indicated
810 bytes from the beginning of the
816 It does not allocate any
828 will be allocated to fill the space.
830 .It Fn m_length mbuf last
831 Return the length of the
833 and optionally a pointer to the last
836 .It Fn m_dup_pkthdr to from how
837 Upon the function's completion, the
840 will contain an identical copy of
842 and the per-packet attributes found in the
852 must be empty on entry.
854 .It Fn m_move_pkthdr to from
857 and the per-packet attributes from the
870 must be empty on entry.
871 Upon the function's completion,
875 and the per-packet attributes cleared.
878 Set the packet-header length to the length of the
881 .It Fn m_devget buf len offset ifp copy
882 Copy data from a device local memory pointed to by
886 The copy is done using a specified copy routine
902 must be of the same type.
904 is still valid after the function returned.
910 .It Fn m_split mbuf len how
913 in two pieces, returning the tail:
917 In case of failure, it returns
919 and attempts to restore the
921 to its original state.
923 .It Fn m_apply mbuf off len f arg
924 Apply a function to an
931 Typically used to avoid calls to
933 which would otherwise be unnecessary or undesirable.
935 is a convenience argument which is passed to the callback function
940 is called, it will be passed
944 in the current mbuf, and the length
946 of the data in this mbuf to which the function should be applied.
948 The function should return zero to indicate success;
949 otherwise, if an error is indicated, then
951 will return the error and stop iterating through the
954 .It Fn m_getptr mbuf loc off
955 Return a pointer to the mbuf containing the data located at
957 bytes from the beginning of the
959 The corresponding offset into the mbuf will be stored in
961 .It Fn m_defrag m0 how
962 Defragment an mbuf chain, returning the shortest possible
963 chain of mbufs and clusters.
964 If allocation fails and this can not be completed,
966 will be returned and the original chain will be unchanged.
967 Upon success, the original chain will be freed and the new
968 chain will be returned.
974 depending on the caller's preference.
976 This function is especially useful in network drivers, where
977 certain long mbuf chains must be shortened before being added
978 to TX descriptor lists.
979 .It Fn m_collapse m0 how maxfrags
980 Defragment an mbuf chain, returning a chain of at most
983 If allocation fails or the chain cannot be collapsed as requested,
985 will be returned, with the original chain possibly modified.
993 .It Fn m_unshare m0 how
994 Create a version of the specified mbuf chain whose
995 contents can be safely modified without affecting other users.
996 If allocation fails and this operation can not be completed,
999 The original mbuf chain is always reclaimed and the reference
1000 count of any shared mbuf clusters is decremented.
1006 depending on the caller's preference.
1007 As a side-effect of this process the returned
1008 mbuf chain may be compacted.
1010 This function is especially useful in the transmit path of
1011 network code, when data must be encrypted or otherwise
1012 altered prior to transmission.
1014 .Sh HARDWARE-ASSISTED CHECKSUM CALCULATION
1015 This section currently applies to TCP/IP only.
1016 In order to save the host CPU resources, computing checksums is
1017 offloaded to the network interface hardware if possible.
1020 member of the leading
1022 of a packet contains two fields used for that purpose,
1023 .Vt int Va csum_flags
1025 .Vt int Va csum_data .
1026 The meaning of those fields depends on the direction a packet flows in,
1027 and on whether the packet is fragmented.
1033 will denote the corresponding field of the
1035 member of the leading
1039 containing the packet.
1041 On output, checksum offloading is attempted after the outgoing
1042 interface has been determined for a packet.
1043 The interface-specific field
1044 .Va ifnet.if_data.ifi_hwassist
1047 is consulted for the capabilities of the interface to assist in
1048 computing checksums.
1051 field of the packet header is set to indicate which actions the interface
1052 is supposed to perform on it.
1053 The actions unsupported by the network interface are done in the
1054 software prior to passing the packet down to the interface driver;
1055 such actions will never be requested through
1058 The flags demanding a particular action from an interface are as follows:
1059 .Bl -tag -width ".Dv CSUM_TCP" -offset indent
1061 The IP header checksum is to be computed and stored in the
1062 corresponding field of the packet.
1063 The hardware is expected to know the format of an IP header
1064 to determine the offset of the IP checksum field.
1066 The TCP checksum is to be computed.
1069 The UDP checksum is to be computed.
1073 Should a TCP or UDP checksum be offloaded to the hardware,
1076 will contain the byte offset of the checksum field relative to the
1077 end of the IP header.
1078 In this case, the checksum field will be initially
1079 set by the TCP/IP module to the checksum of the pseudo header
1080 defined by the TCP and UDP specifications.
1082 On input, an interface indicates the actions it has performed
1083 on a packet by setting one or more of the following flags in
1085 associated with the packet:
1086 .Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent
1087 .It Dv CSUM_IP_CHECKED
1088 The IP header checksum has been computed.
1089 .It Dv CSUM_IP_VALID
1090 The IP header has a valid checksum.
1091 This flag can appear only in combination with
1092 .Dv CSUM_IP_CHECKED .
1093 .It Dv CSUM_DATA_VALID
1094 The checksum of the data portion of the IP packet has been computed
1095 and stored in the field
1097 in network byte order.
1098 .It Dv CSUM_PSEUDO_HDR
1099 Can be set only along with
1101 to indicate that the IP data checksum found in
1103 allows for the pseudo header defined by the TCP and UDP specifications.
1104 Otherwise the checksum of the pseudo header must be calculated by
1105 the host CPU and added to
1107 to obtain the final checksum to be used for TCP or UDP validation purposes.
1110 If a particular network interface just indicates success or
1111 failure of TCP or UDP checksum validation without returning
1112 the exact value of the checksum to the host CPU, its driver can mark
1122 hexadecimal to indicate a valid checksum.
1123 It is a peculiarity of the algorithm used that the Internet checksum
1124 calculated over any valid packet will be
1126 as long as the original checksum field is included.
1128 When running a kernel compiled with the option
1129 .Dv MBUF_STRESS_TEST ,
1132 -controlled options may be used to create
1133 various failure/extreme cases for testing of network drivers
1134 and other parts of the kernel that rely on
1136 .Bl -tag -width ident
1137 .It Va net.inet.ip.mbuf_frag_size
1140 to fragment outgoing
1142 into fragments of the specified size.
1143 Setting this variable to 1 is an excellent way to
1146 handling ability of network drivers.
1147 .It Va kern.ipc.m_defragrandomfailures
1150 to randomly fail, returning
1152 Any piece of code which uses
1154 should be tested with this feature.
1162 .\" Please correct me if I'm wrong
1164 appeared in an early version of
1166 Besides being used for network packets, they were used
1167 to store various dynamic structures, such as routing table
1168 entries, interface addresses, protocol control blocks, etc.
1173 is almost entirely limited to packet storage, with
1175 zones being used directly to store other network-related memory.
1179 allocator has been a special-purpose memory allocator able to run in
1180 interrupt contexts and allocating from a special kernel address space map.
1185 allocator is a wrapper around
1191 + cluster pairs in per-CPU caches, as well as bringing other benefits of
1196 manual page was written by Yar Tikhiy.
1200 allocator was written by Bosko Milekic.
projects / FreeBSD / releng / 10.2.git / blob