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
388 section) used allocation flags
392 These constants are kept for compatibility
393 and their use in new code is discouraged.
394 .It Fn MGETHDR mbuf how type
397 and initialize it to contain a packet header
402 .It Fn MEXTADD mbuf buf size free opt_arg1 opt_arg2 flags type
403 Associate externally managed data with
405 Any internal data contained in the mbuf will be discarded, and the
406 .Dv M_EXT flag will be set.
411 arguments are the address and length, respectively, of the data.
414 argument points to a function which will be called to free the data
415 when the mbuf is freed; it is only used if
423 arguments will be passed unmodified to
427 argument specifies additional
429 flags; it is not necessary to specify
433 argument specifies the type of external data, which controls how it
434 will be disposed of when the
437 In most cases, the correct value is
439 .It Fn MCLGET mbuf how
440 Allocate and attach an
444 If the macro fails, the
446 flag will not be set in
448 .It Fn M_ALIGN mbuf len
451 to place an object of the size
453 at the end of the internal data area of
458 is newly allocated with
462 .It Fn MH_ALIGN mbuf len
463 Serves the same purpose as
475 .It Fn m_align mbuf len
476 Services the same purpose as
478 but handles any type of mbuf.
479 .It Fn M_LEADINGSPACE mbuf
480 Returns the number of bytes available before the beginning
483 .It Fn M_TRAILINGSPACE mbuf
484 Returns the number of bytes available after the end of data in
486 .It Fn M_PREPEND mbuf len how
487 This macro operates on an
489 It is an optimized wrapper for
491 that can make use of possible empty space before data
492 (e.g.\& left after trimming of a link-layer header).
500 .It Fn M_MOVE_PKTHDR to from
501 Using this macro is equivalent to calling
502 .Fn m_move_pkthdr to from .
503 .It Fn M_WRITABLE mbuf
504 This macro will evaluate true if
510 does not contain external storage or,
512 then if the reference count of the storage is not greater than 1.
517 This can be achieved during setup of the external storage,
524 macro, or can be directly set in individual
526 .It Fn MCHTYPE mbuf type
531 This is a relatively expensive operation and should be avoided.
535 .Bl -ohang -offset indent
536 .It Fn m_get how type
537 A function version of
539 for non-critical paths.
540 .It Fn m_getm orig len how type
547 if necessary and append the resulting allocated
553 .No non- Ns Dv NULL .
554 If the allocation fails at any point,
555 free whatever was allocated and return
560 .No non- Ns Dv NULL ,
561 it will not be freed.
562 It is possible to use
570 (for example, one which may be sitting in a pre-allocated ring)
571 or to simply perform an all-or-nothing
576 .It Fn m_gethdr how type
577 A function version of
579 for non-critical paths.
580 .It Fn m_getcl how type flags
586 If one of the allocations fails, the entire allocation fails.
587 This routine is the preferred way of fetching both the
591 together, as it avoids having to unlock/relock between allocations.
595 .It Fn m_getclr how type
598 and zero out the data region.
608 The functions below operate on
610 .Bl -ohang -offset indent
614 including any external storage.
616 .It Fn m_adj mbuf len
619 bytes from the head of an
623 is positive, from the tail otherwise.
625 .It Fn m_append mbuf len cp
632 Extend the mbuf chain if the new data does not fit in
635 .It Fn m_prepend mbuf len how
638 and prepend it to the
644 It does not allocate any
656 .It Fn m_copyup mbuf len dstoff
661 bytes of data into a new mbuf at
666 argument aligns the data and leaves room for a link layer header.
676 The function does not allocate
683 .It Fn m_pullup mbuf len
684 Arrange that the first
688 are contiguous and lay in the data area of
690 so they are accessible with
692 It is important to remember that this may involve
693 reallocating some mbufs and moving data so all pointers
694 referencing data within the old mbuf chain
695 must be recalculated or made invalid.
703 is freed in this case).
705 It does not allocate any
712 .It Fn m_pulldown mbuf offset len offsetp
721 are contiguous and lay in the data area of
723 so they are accessible with
725 .Fa len must be smaller than, or equal to, the size of an
727 Return a pointer to an intermediate
729 in the chain containing the requested region;
730 the offset in the data region of the
732 to the data contained in the returned mbuf is stored in
736 is NULL, the region may be accessed using
740 is non-NULL, the region may be accessed using
741 .Fn mtod mbuf uint8_t + *offsetp .
742 The region of the mbuf chain between its beginning and
744 is not modified, therefore it is safe to hold pointers to data within
745 this region before calling
748 .It Fn m_copym mbuf offset len how
753 bytes from the beginning, continuing for
760 copy to the end of the
763 The copy is read-only, because the
765 are not copied, only their reference counts are incremented.
767 .It Fn m_copypacket mbuf how
768 Copy an entire packet including header, which must be present.
769 This is an optimized version of the common case
770 .Fn m_copym mbuf 0 M_COPYALL how .
772 the copy is read-only, because the
774 are not copied, only their reference counts are incremented.
776 .It Fn m_dup mbuf how
779 into a completely new
781 including copying any
785 when you need a writable copy of an
788 .It Fn m_copydata mbuf offset len buf
793 bytes from the beginning, continuing for
795 bytes, into the indicated buffer
798 .It Fn m_copyback mbuf offset len buf
801 bytes from the buffer
803 back into the indicated
807 bytes from the beginning of the
813 It does not allocate any
825 will be allocated to fill the space.
827 .It Fn m_length mbuf last
828 Return the length of the
830 and optionally a pointer to the last
833 .It Fn m_dup_pkthdr to from how
834 Upon the function's completion, the
837 will contain an identical copy of
839 and the per-packet attributes found in the
849 must be empty on entry.
851 .It Fn m_move_pkthdr to from
854 and the per-packet attributes from the
867 must be empty on entry.
868 Upon the function's completion,
872 and the per-packet attributes cleared.
875 Set the packet-header length to the length of the
878 .It Fn m_devget buf len offset ifp copy
879 Copy data from a device local memory pointed to by
883 The copy is done using a specified copy routine
899 must be of the same type.
901 is still valid after the function returned.
907 .It Fn m_split mbuf len how
910 in two pieces, returning the tail:
914 In case of failure, it returns
916 and attempts to restore the
918 to its original state.
920 .It Fn m_apply mbuf off len f arg
921 Apply a function to an
928 Typically used to avoid calls to
930 which would otherwise be unnecessary or undesirable.
932 is a convenience argument which is passed to the callback function
937 is called, it will be passed
941 in the current mbuf, and the length
943 of the data in this mbuf to which the function should be applied.
945 The function should return zero to indicate success;
946 otherwise, if an error is indicated, then
948 will return the error and stop iterating through the
951 .It Fn m_getptr mbuf loc off
952 Return a pointer to the mbuf containing the data located at
954 bytes from the beginning of the
956 The corresponding offset into the mbuf will be stored in
958 .It Fn m_defrag m0 how
959 Defragment an mbuf chain, returning the shortest possible
960 chain of mbufs and clusters.
961 If allocation fails and this can not be completed,
963 will be returned and the original chain will be unchanged.
964 Upon success, the original chain will be freed and the new
965 chain will be returned.
971 depending on the caller's preference.
973 This function is especially useful in network drivers, where
974 certain long mbuf chains must be shortened before being added
975 to TX descriptor lists.
976 .It Fn m_unshare m0 how
977 Create a version of the specified mbuf chain whose
978 contents can be safely modified without affecting other users.
979 If allocation fails and this operation can not be completed,
982 The original mbuf chain is always reclaimed and the reference
983 count of any shared mbuf clusters is decremented.
989 depending on the caller's preference.
990 As a side-effect of this process the returned
991 mbuf chain may be compacted.
993 This function is especially useful in the transmit path of
994 network code, when data must be encrypted or otherwise
995 altered prior to transmission.
997 .Sh HARDWARE-ASSISTED CHECKSUM CALCULATION
998 This section currently applies to TCP/IP only.
999 In order to save the host CPU resources, computing checksums is
1000 offloaded to the network interface hardware if possible.
1003 member of the leading
1005 of a packet contains two fields used for that purpose,
1006 .Vt int Va csum_flags
1008 .Vt int Va csum_data .
1009 The meaning of those fields depends on the direction a packet flows in,
1010 and on whether the packet is fragmented.
1016 will denote the corresponding field of the
1018 member of the leading
1022 containing the packet.
1024 On output, checksum offloading is attempted after the outgoing
1025 interface has been determined for a packet.
1026 The interface-specific field
1027 .Va ifnet.if_data.ifi_hwassist
1030 is consulted for the capabilities of the interface to assist in
1031 computing checksums.
1034 field of the packet header is set to indicate which actions the interface
1035 is supposed to perform on it.
1036 The actions unsupported by the network interface are done in the
1037 software prior to passing the packet down to the interface driver;
1038 such actions will never be requested through
1041 The flags demanding a particular action from an interface are as follows:
1042 .Bl -tag -width ".Dv CSUM_TCP" -offset indent
1044 The IP header checksum is to be computed and stored in the
1045 corresponding field of the packet.
1046 The hardware is expected to know the format of an IP header
1047 to determine the offset of the IP checksum field.
1049 The TCP checksum is to be computed.
1052 The UDP checksum is to be computed.
1056 Should a TCP or UDP checksum be offloaded to the hardware,
1059 will contain the byte offset of the checksum field relative to the
1060 end of the IP header.
1061 In this case, the checksum field will be initially
1062 set by the TCP/IP module to the checksum of the pseudo header
1063 defined by the TCP and UDP specifications.
1065 For outbound packets which have been fragmented
1066 by the host CPU, the following will also be true,
1067 regardless of the checksum flag settings:
1068 .Bl -bullet -offset indent
1070 all fragments will have the flag
1076 the first and the last fragments in the chain will have
1084 the first fragment in the chain will have the total number
1085 of fragments contained in its
1090 The last rule for fragmented packets takes precedence over the one
1091 for a TCP or UDP checksum.
1092 Nevertheless, offloading a TCP or UDP checksum is possible for a
1093 fragmented packet if the flag
1096 .Va ifnet.if_data.ifi_hwassist
1097 associated with the network interface.
1098 However, in this case the interface is expected to figure out
1099 the location of the checksum field within the sequence of fragments
1102 contains a fragment count instead of a checksum offset value.
1104 On input, an interface indicates the actions it has performed
1105 on a packet by setting one or more of the following flags in
1107 associated with the packet:
1108 .Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent
1109 .It Dv CSUM_IP_CHECKED
1110 The IP header checksum has been computed.
1111 .It Dv CSUM_IP_VALID
1112 The IP header has a valid checksum.
1113 This flag can appear only in combination with
1114 .Dv CSUM_IP_CHECKED .
1115 .It Dv CSUM_DATA_VALID
1116 The checksum of the data portion of the IP packet has been computed
1117 and stored in the field
1119 in network byte order.
1120 .It Dv CSUM_PSEUDO_HDR
1121 Can be set only along with
1123 to indicate that the IP data checksum found in
1125 allows for the pseudo header defined by the TCP and UDP specifications.
1126 Otherwise the checksum of the pseudo header must be calculated by
1127 the host CPU and added to
1129 to obtain the final checksum to be used for TCP or UDP validation purposes.
1132 If a particular network interface just indicates success or
1133 failure of TCP or UDP checksum validation without returning
1134 the exact value of the checksum to the host CPU, its driver can mark
1144 hexadecimal to indicate a valid checksum.
1145 It is a peculiarity of the algorithm used that the Internet checksum
1146 calculated over any valid packet will be
1148 as long as the original checksum field is included.
1150 For inbound packets which are IP fragments, all
1152 fields will be summed during reassembly to obtain the final checksum
1153 value passed to an upper layer in the
1155 field of the reassembled packet.
1158 fields of all fragments will be consolidated using logical AND
1159 to obtain the final value for
1161 Thus, in order to successfully
1162 offload checksum computation for fragmented data,
1163 all fragments should have the same value of
1166 When running a kernel compiled with the option
1167 .Dv MBUF_STRESS_TEST ,
1170 -controlled options may be used to create
1171 various failure/extreme cases for testing of network drivers
1172 and other parts of the kernel that rely on
1174 .Bl -tag -width ident
1175 .It Va net.inet.ip.mbuf_frag_size
1178 to fragment outgoing
1180 into fragments of the specified size.
1181 Setting this variable to 1 is an excellent way to
1184 handling ability of network drivers.
1185 .It Va kern.ipc.m_defragrandomfailures
1188 to randomly fail, returning
1190 Any piece of code which uses
1192 should be tested with this feature.
1200 .\" Please correct me if I'm wrong
1202 appeared in an early version of
1204 Besides being used for network packets, they were used
1205 to store various dynamic structures, such as routing table
1206 entries, interface addresses, protocol control blocks, etc.
1211 is almost entirely limited to packet storage, with
1213 zones being used directly to store other network-related memory.
1217 allocator has been a special-purpose memory allocator able to run in
1218 interrupt contexts and allocating from a special kernel address space map.
1223 allocator is a wrapper around
1229 + cluster pairs in per-CPU caches, as well as bringing other benefits of
1234 manual page was written by Yar Tikhiy.
1238 allocator was written by Bosko Milekic.