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
175 .Bl -tag -width "m_nextpkt" -offset indent
178 A pointer to the next
184 A pointer to the next
189 A pointer to data attached to this
193 The length of the data.
196 The type of the data.
206 flag bits are defined as follows:
209 #define M_EXT 0x0001 /* has associated external storage */
210 #define M_PKTHDR 0x0002 /* start of record */
211 #define M_EOR 0x0004 /* end of record */
212 #define M_RDONLY 0x0008 /* associated data marked read-only */
213 #define M_PROTO1 0x0010 /* protocol-specific */
214 #define M_PROTO2 0x0020 /* protocol-specific */
215 #define M_PROTO3 0x0040 /* protocol-specific */
216 #define M_PROTO4 0x0080 /* protocol-specific */
217 #define M_PROTO5 0x0100 /* protocol-specific */
218 #define M_PROTO6 0x4000 /* protocol-specific (avoid M_BCAST conflict) */
219 #define M_FREELIST 0x8000 /* mbuf is on the free list */
221 /* mbuf pkthdr flags (also stored in m_flags) */
222 #define M_BCAST 0x0200 /* send/received as link-level broadcast */
223 #define M_MCAST 0x0400 /* send/received as link-level multicast */
224 #define M_FRAG 0x0800 /* packet is fragment of larger packet */
225 #define M_FIRSTFRAG 0x1000 /* packet is first fragment */
226 #define M_LASTFRAG 0x2000 /* packet is last fragment */
231 types are defined as follows:
234 #define MT_DATA 1 /* dynamic (data) allocation */
235 #define MT_HEADER MT_DATA /* packet header */
236 #define MT_SONAME 8 /* socket name */
237 #define MT_CONTROL 14 /* extra-data protocol message */
238 #define MT_OOBDATA 15 /* expedited data */
244 .Vt struct pkthdr Va m_pkthdr
248 It contains a pointer to the interface
249 the packet has been received from
250 .Pq Vt struct ifnet Va *rcvif ,
251 and the total packet length
253 Optionally, it may also contain an attached list of packet tags
254 .Pq Vt "struct m_tag" .
258 Fields used in offloading checksum calculation to the hardware are kept in
262 .Sx HARDWARE-ASSISTED CHECKSUM CALCULATION
265 If small enough, data is stored in the internal data buffer of an
267 If the data is sufficiently large, another
271 or external storage may be associated with the
274 bytes of data can fit into an
282 If external storage is being associated with an
286 header is added at the cost of losing the internal data buffer.
287 It includes a pointer to external storage, the size of the storage,
288 a pointer to a function used for freeing the storage,
289 a pointer to an optional argument that can be passed to the function,
290 and a pointer to a reference counter.
293 using external storage has the
297 The system supplies a macro for allocating the desired external storage
301 The allocation and management of the reference counter is handled by the
304 The system also supplies a default type of external storage buffer called an
307 can be allocated and configured with the use of the
314 in size, where MCLBYTES is a machine-dependent constant.
315 The system defines an advisory macro
317 which is the smallest amount of data to put into an
319 It is equal to the sum of
323 It is typically preferable to store data into the data region of an
325 if size permits, as opposed to allocating a separate
327 to hold the same data.
329 .Ss Macros and Functions
330 There are numerous predefined macros and functions that provide the
331 developer with common utilities.
333 .Bl -ohang -offset indent
334 .It Fn mtod mbuf type
337 pointer to a data pointer.
338 The macro expands to the data pointer cast to the pointer of the specified
341 It is advisable to ensure that there is enough contiguous data in
346 .It Fn MGET mbuf how type
349 and initialize it to contain internal data.
351 will point to the allocated
353 on success, or be set to
358 argument is to be set to
362 It specifies whether the caller is willing to block if necessary.
363 A number of other functions and macros related to
365 have the same argument because they may
366 at some point need to allocate new
369 Programmers should be careful not to confuse the
377 They are not the same.
378 .It Fn MGETHDR mbuf how type
381 and initialize it to contain a packet header
386 .It Fn MCLGET mbuf how
387 Allocate and attach an
391 If the macro fails, the
393 flag will not be set in
395 .It Fn M_ALIGN mbuf len
398 to place an object of the size
400 at the end of the internal data area of
405 is newly allocated with
409 .It Fn MH_ALIGN mbuf len
410 Serves the same purpose as
422 .It Fn m_align mbuf len
423 Services the same purpose as
425 but handles any type of mbuf.
426 .It Fn M_LEADINGSPACE mbuf
427 Returns the number of bytes available before the beginning
430 .It Fn M_TRAILINGSPACE mbuf
431 Returns the number of bytes available after the end of data in
433 .It Fn M_PREPEND mbuf len how
434 This macro operates on an
436 It is an optimized wrapper for
438 that can make use of possible empty space before data
439 (e.g.\& left after trimming of a link-layer header).
447 .It Fn M_MOVE_PKTHDR to from
448 Using this macro is equivalent to calling
449 .Fn m_move_pkthdr to from .
450 .It Fn M_WRITABLE mbuf
451 This macro will evaluate true if
457 does not contain external storage or,
459 then if the reference count of the storage is not greater than 1.
464 This can be achieved during setup of the external storage,
471 macro, or can be directly set in individual
473 .It Fn MCHTYPE mbuf type
478 This is a relatively expensive operation and should be avoided.
482 .Bl -ohang -offset indent
483 .It Fn m_get how type
484 A function version of
486 for non-critical paths.
487 .It Fn m_getm orig len how type
494 if necessary and append the resulting allocated
500 .No non- Ns Dv NULL .
501 If the allocation fails at any point,
502 free whatever was allocated and return
507 .No non- Ns Dv NULL ,
508 it will not be freed.
509 It is possible to use
517 (for example, one which may be sitting in a pre-allocated ring)
518 or to simply perform an all-or-nothing
523 .It Fn m_gethdr how type
524 A function version of
526 for non-critical paths.
527 .It Fn m_getcl how type flags
533 If one of the allocations fails, the entire allocation fails.
534 This routine is the preferred way of fetching both the
538 together, as it avoids having to unlock/relock between allocations.
542 .It Fn m_getclr how type
545 and zero out the data region.
555 The functions below operate on
557 .Bl -ohang -offset indent
561 including any external storage.
563 .It Fn m_adj mbuf len
566 bytes from the head of an
570 is positive, from the tail otherwise.
572 .It Fn m_append mbuf len cp
579 Extend the mbuf chain if the new data does not fit in
582 .It Fn m_prepend mbuf len how
585 and prepend it to the
591 It does not allocate any
603 .It Fn m_copyup mbuf len dstoff
608 bytes of data into a new mbuf at
613 argument aligns the data and leaves room for a link layer header.
623 The function does not allocate
630 .It Fn m_pullup mbuf len
631 Arrange that the first
635 are contiguous and lay in the data area of
637 so they are accessible with
639 It is important to remember that this may involve
640 reallocating some mbufs and moving data so all pointers
641 referencing data within the old mbuf chain
642 must be recalculated or made invalid.
650 is freed in this case).
652 It does not allocate any
659 .It Fn m_pulldown mbuf offset len offsetp
668 are contiguous and lay in the data area of
670 so they are accessible with
672 .Fa len must be smaller than, or equal to, the size of an
674 Return a pointer to an intermediate
676 in the chain containing the requested region;
677 the offset in the data region of the
679 to the data contained in the returned mbuf is stored in
683 is NULL, the region may be accessed using
687 is non-NULL, the region may be accessed using
688 .Fn mtod mbuf uint8_t + *offsetp .
689 The region of the mbuf chain between its beginning and
691 is not modified, therefore it is safe to hold pointers to data within
692 this region before calling
695 .It Fn m_copym mbuf offset len how
700 bytes from the beginning, continuing for
707 copy to the end of the
710 The copy is read-only, because the
712 are not copied, only their reference counts are incremented.
714 .It Fn m_copypacket mbuf how
715 Copy an entire packet including header, which must be present.
716 This is an optimized version of the common case
717 .Fn m_copym mbuf 0 M_COPYALL how .
719 the copy is read-only, because the
721 are not copied, only their reference counts are incremented.
723 .It Fn m_dup mbuf how
726 into a completely new
728 including copying any
732 when you need a writable copy of an
735 .It Fn m_copydata mbuf offset len buf
740 bytes from the beginning, continuing for
742 bytes, into the indicated buffer
745 .It Fn m_copyback mbuf offset len buf
748 bytes from the buffer
750 back into the indicated
754 bytes from the beginning of the
760 It does not allocate any
772 will be allocated to fill the space.
774 .It Fn m_length mbuf last
775 Return the length of the
777 and optionally a pointer to the last
780 .It Fn m_dup_pkthdr to from how
781 Upon the function's completion, the
784 will contain an identical copy of
786 and the per-packet attributes found in the
796 must be empty on entry.
798 .It Fn m_move_pkthdr to from
801 and the per-packet attributes from the
814 must be empty on entry.
815 Upon the function's completion,
819 and the per-packet attributes cleared.
822 Set the packet-header length to the length of the
825 .It Fn m_devget buf len offset ifp copy
826 Copy data from a device local memory pointed to by
830 The copy is done using a specified copy routine
846 must be of the same type.
848 is still valid after the function returned.
854 .It Fn m_split mbuf len how
857 in two pieces, returning the tail:
861 In case of failure, it returns
863 and attempts to restore the
865 to its original state.
867 .It Fn m_apply mbuf off len f arg
868 Apply a function to an
875 Typically used to avoid calls to
877 which would otherwise be unnecessary or undesirable.
879 is a convenience argument which is passed to the callback function
884 is called, it will be passed
888 in the current mbuf, and the length
890 of the data in this mbuf to which the function should be applied.
892 The function should return zero to indicate success;
893 otherwise, if an error is indicated, then
895 will return the error and stop iterating through the
898 .It Fn m_getptr mbuf loc off
899 Return a pointer to the mbuf containing the data located at
901 bytes from the beginning of the
903 The corresponding offset into the mbuf will be stored in
905 .It Fn m_defrag m0 how
906 Defragment an mbuf chain, returning the shortest possible
907 chain of mbufs and clusters.
908 If allocation fails and this can not be completed,
910 will be returned and the original chain will be unchanged.
911 Upon success, the original chain will be freed and the new
912 chain will be returned.
918 depending on the caller's preference.
920 This function is especially useful in network drivers, where
921 certain long mbuf chains must be shortened before being added
922 to TX descriptor lists.
923 .It Fn m_unshare m0 how
924 Create a version of the specified mbuf chain whose
925 contents can be safely modified without affecting other users.
926 If allocation fails and this operation can not be completed,
929 The original mbuf chain is always reclaimed and the reference
930 count of any shared mbuf clusters is decremented.
936 depending on the caller's preference.
937 As a side-effect of this process the returned
938 mbuf chain may be compacted.
940 This function is especially useful in the transmit path of
941 network code, when data must be encrypted or otherwise
942 altered prior to transmission.
944 .Sh HARDWARE-ASSISTED CHECKSUM CALCULATION
945 This section currently applies to TCP/IP only.
946 In order to save the host CPU resources, computing checksums is
947 offloaded to the network interface hardware if possible.
950 member of the leading
952 of a packet contains two fields used for that purpose,
953 .Vt int Va csum_flags
955 .Vt int Va csum_data .
956 The meaning of those fields depends on the direction a packet flows in,
957 and on whether the packet is fragmented.
963 will denote the corresponding field of the
965 member of the leading
969 containing the packet.
971 On output, checksum offloading is attempted after the outgoing
972 interface has been determined for a packet.
973 The interface-specific field
974 .Va ifnet.if_data.ifi_hwassist
977 is consulted for the capabilities of the interface to assist in
981 field of the packet header is set to indicate which actions the interface
982 is supposed to perform on it.
983 The actions unsupported by the network interface are done in the
984 software prior to passing the packet down to the interface driver;
985 such actions will never be requested through
988 The flags demanding a particular action from an interface are as follows:
989 .Bl -tag -width ".Dv CSUM_TCP" -offset indent
991 The IP header checksum is to be computed and stored in the
992 corresponding field of the packet.
993 The hardware is expected to know the format of an IP header
994 to determine the offset of the IP checksum field.
996 The TCP checksum is to be computed.
999 The UDP checksum is to be computed.
1003 Should a TCP or UDP checksum be offloaded to the hardware,
1006 will contain the byte offset of the checksum field relative to the
1007 end of the IP header.
1008 In this case, the checksum field will be initially
1009 set by the TCP/IP module to the checksum of the pseudo header
1010 defined by the TCP and UDP specifications.
1012 For outbound packets which have been fragmented
1013 by the host CPU, the following will also be true,
1014 regardless of the checksum flag settings:
1015 .Bl -bullet -offset indent
1017 all fragments will have the flag
1023 the first and the last fragments in the chain will have
1031 the first fragment in the chain will have the total number
1032 of fragments contained in its
1037 The last rule for fragmented packets takes precedence over the one
1038 for a TCP or UDP checksum.
1039 Nevertheless, offloading a TCP or UDP checksum is possible for a
1040 fragmented packet if the flag
1043 .Va ifnet.if_data.ifi_hwassist
1044 associated with the network interface.
1045 However, in this case the interface is expected to figure out
1046 the location of the checksum field within the sequence of fragments
1049 contains a fragment count instead of a checksum offset value.
1051 On input, an interface indicates the actions it has performed
1052 on a packet by setting one or more of the following flags in
1054 associated with the packet:
1055 .Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent
1056 .It Dv CSUM_IP_CHECKED
1057 The IP header checksum has been computed.
1058 .It Dv CSUM_IP_VALID
1059 The IP header has a valid checksum.
1060 This flag can appear only in combination with
1061 .Dv CSUM_IP_CHECKED .
1062 .It Dv CSUM_DATA_VALID
1063 The checksum of the data portion of the IP packet has been computed
1064 and stored in the field
1066 in network byte order.
1067 .It Dv CSUM_PSEUDO_HDR
1068 Can be set only along with
1070 to indicate that the IP data checksum found in
1072 allows for the pseudo header defined by the TCP and UDP specifications.
1073 Otherwise the checksum of the pseudo header must be calculated by
1074 the host CPU and added to
1076 to obtain the final checksum to be used for TCP or UDP validation purposes.
1079 If a particular network interface just indicates success or
1080 failure of TCP or UDP checksum validation without returning
1081 the exact value of the checksum to the host CPU, its driver can mark
1091 hexadecimal to indicate a valid checksum.
1092 It is a peculiarity of the algorithm used that the Internet checksum
1093 calculated over any valid packet will be
1095 as long as the original checksum field is included.
1097 For inbound packets which are IP fragments, all
1099 fields will be summed during reassembly to obtain the final checksum
1100 value passed to an upper layer in the
1102 field of the reassembled packet.
1105 fields of all fragments will be consolidated using logical AND
1106 to obtain the final value for
1108 Thus, in order to successfully
1109 offload checksum computation for fragmented data,
1110 all fragments should have the same value of
1113 When running a kernel compiled with the option
1114 .Dv MBUF_STRESS_TEST ,
1117 -controlled options may be used to create
1118 various failure/extreme cases for testing of network drivers
1119 and other parts of the kernel that rely on
1121 .Bl -tag -width ident
1122 .It Va net.inet.ip.mbuf_frag_size
1125 to fragment outgoing
1127 into fragments of the specified size.
1128 Setting this variable to 1 is an excellent way to
1131 handling ability of network drivers.
1132 .It Va kern.ipc.m_defragrandomfailures
1135 to randomly fail, returning
1137 Any piece of code which uses
1139 should be tested with this feature.
1147 .\" Please correct me if I'm wrong
1149 appeared in an early version of
1151 Besides being used for network packets, they were used
1152 to store various dynamic structures, such as routing table
1153 entries, interface addresses, protocol control blocks, etc.
1158 is almost entirely limited to packet storage, with
1160 zones being used directly to store other network-related memory.
1164 allocator has been a special-purpose memory allocator able to run in
1165 interrupt contexts and allocating from a special kernel address space map.
1170 allocator is a wrapper around
1176 + cluster pairs in per-CPU caches, as well as bringing other benefits of
1181 manual page was written by Yar Tikhiy.
1185 allocator was written by Bosko Milekic.
projects / FreeBSD / releng / 8.1.git / blob