1.\" Copyright (c) 2000 FreeBSD Inc. 2.\" All rights reserved. 3.\" 4.\" Redistribution and use in source and binary forms, with or without 5.\" modification, are permitted provided that the following conditions 6.\" are met: 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. 12.\" 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 23.\" SUCH DAMAGE. 24.\" 25.\" $FreeBSD$ 26.\" 27.Dd April 18, 2011 28.Dt MBUF 9 29.Os 30.\" 31.Sh NAME 32.Nm mbuf 33.Nd "memory management in the kernel IPC subsystem" 34.\" 35.Sh SYNOPSIS 36.In sys/param.h 37.In sys/systm.h 38.In sys/mbuf.h 39.\" 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" 44.Fo MEXTADD 45.Fa "struct mbuf *mbuf" 46.Fa "caddr_t buf" 47.Fa "u_int size" 48.Fa "void (*free)(void *opt_arg1, void *opt_arg2)" 49.Fa "void *opt_arg1" 50.Fa "void *opt_arg2" 51.Fa "short flags" 52.Fa "int type" 53.Fc 54.Fn MEXTFREE "struct mbuf *mbuf" 55.Fn MFREE "struct mbuf *mbuf" "struct mbuf *successor" 56.\" 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" 61.Ft int 62.Fn M_LEADINGSPACE "struct mbuf *mbuf" 63.Ft int 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" 68.Ft int 69.Fn M_WRITABLE "struct mbuf *mbuf" 70.\" 71.Ss Mbuf allocation functions 72.Ft struct mbuf * 73.Fn m_get "int how" "int type" 74.Ft struct mbuf * 75.Fn m_getm "struct mbuf *orig" "int len" "int how" "int type" 76.Ft struct mbuf * 77.Fn m_getcl "int how" "short type" "int flags" 78.Ft struct mbuf * 79.Fn m_getclr "int how" "int type" 80.Ft struct mbuf * 81.Fn m_gethdr "int how" "int type" 82.Ft struct mbuf * 83.Fn m_free "struct mbuf *mbuf" 84.Ft void 85.Fn m_freem "struct mbuf *mbuf" 86.\" 87.Ss Mbuf utility functions 88.Ft void 89.Fn m_adj "struct mbuf *mbuf" "int len" 90.Ft void 91.Fn m_align "struct mbuf *mbuf" "int len" 92.Ft int 93.Fn m_append "struct mbuf *mbuf" "int len" "c_caddr_t cp" 94.Ft struct mbuf * 95.Fn m_prepend "struct mbuf *mbuf" "int len" "int how" 96.Ft struct mbuf * 97.Fn m_copyup "struct mbuf *mbuf" "int len" "int dstoff" 98.Ft struct mbuf * 99.Fn m_pullup "struct mbuf *mbuf" "int len" 100.Ft struct mbuf * 101.Fn m_pulldown "struct mbuf *mbuf" "int offset" "int len" "int *offsetp" 102.Ft struct mbuf * 103.Fn m_copym "struct mbuf *mbuf" "int offset" "int len" "int how" 104.Ft struct mbuf * 105.Fn m_copypacket "struct mbuf *mbuf" "int how" 106.Ft struct mbuf * 107.Fn m_dup "struct mbuf *mbuf" "int how" 108.Ft void 109.Fn m_copydata "const struct mbuf *mbuf" "int offset" "int len" "caddr_t buf" 110.Ft void 111.Fn m_copyback "struct mbuf *mbuf" "int offset" "int len" "caddr_t buf" 112.Ft struct mbuf * 113.Fo m_devget 114.Fa "char *buf" 115.Fa "int len" 116.Fa "int offset" 117.Fa "struct ifnet *ifp" 118.Fa "void (*copy)(char *from, caddr_t to, u_int len)" 119.Fc 120.Ft void 121.Fn m_cat "struct mbuf *m" "struct mbuf *n" 122.Ft u_int 123.Fn m_fixhdr "struct mbuf *mbuf" 124.Ft void 125.Fn m_dup_pkthdr "struct mbuf *to" "struct mbuf *from" 126.Ft void 127.Fn m_move_pkthdr "struct mbuf *to" "struct mbuf *from" 128.Ft u_int 129.Fn m_length "struct mbuf *mbuf" "struct mbuf **last" 130.Ft struct mbuf * 131.Fn m_split "struct mbuf *mbuf" "int len" "int how" 132.Ft int 133.Fn m_apply "struct mbuf *mbuf" "int off" "int len" "int (*f)(void *arg, void *data, u_int len)" "void *arg" 134.Ft struct mbuf * 135.Fn m_getptr "struct mbuf *mbuf" "int loc" "int *off" 136.Ft struct mbuf * 137.Fn m_defrag "struct mbuf *m0" "int how" 138.Ft struct mbuf * 139.Fn m_unshare "struct mbuf *m0" "int how" 140.\" 141.Sh DESCRIPTION 142An 143.Vt mbuf 144is a basic unit of memory management in the kernel IPC subsystem. 145Network packets and socket buffers are stored in 146.Vt mbufs . 147A network packet may span multiple 148.Vt mbufs 149arranged into a 150.Vt mbuf chain 151(linked list), 152which allows adding or trimming 153network headers with little overhead. 154.Pp 155While a developer should not bother with 156.Vt mbuf 157internals without serious 158reason in order to avoid incompatibilities with future changes, it 159is useful to understand the general structure of an 160.Vt mbuf . 161.Pp 162An 163.Vt mbuf 164consists of a variable-sized header and a small internal 165buffer for data. 166The total size of an 167.Vt mbuf , 168.Dv MSIZE , 169is a constant defined in 170.In sys/param.h . 171The 172.Vt mbuf 173header includes: 174.Bl -tag -width "m_nextpkt" -offset indent 175.It Va m_next 176.Pq Vt struct mbuf * 177A pointer to the next 178.Vt mbuf 179in the 180.Vt mbuf chain . 181.It Va m_nextpkt 182.Pq Vt struct mbuf * 183A pointer to the next 184.Vt mbuf chain 185in the queue. 186.It Va m_data 187.Pq Vt caddr_t 188A pointer to data attached to this 189.Vt mbuf . 190.It Va m_len 191.Pq Vt int 192The length of the data. 193.It Va m_type 194.Pq Vt short 195The type of the data. 196.It Va m_flags 197.Pq Vt int 198The 199.Vt mbuf 200flags. 201.El 202.Pp 203The 204.Vt mbuf 205flag bits are defined as follows: 206.Bd -literal 207/* mbuf flags */ 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 */ 219 220/* 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 */ 226.Ed 227.Pp 228The available 229.Vt mbuf 230types are defined as follows: 231.Bd -literal 232/* mbuf types */ 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 */ 238.Ed 239.Pp 240The available external buffer types are defined as follows: 241.Bd -literal 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 */ 254.Ed 255.Pp 256If the 257.Dv M_PKTHDR 258flag is set, a 259.Vt struct pkthdr Va m_pkthdr 260is added to the 261.Vt mbuf 262header. 263It contains a pointer to the interface 264the packet has been received from 265.Pq Vt struct ifnet Va *rcvif , 266and the total packet length 267.Pq Vt int Va len . 268Optionally, it may also contain an attached list of packet tags 269.Pq Vt "struct m_tag" . 270See 271.Xr mbuf_tags 9 272for details. 273Fields used in offloading checksum calculation to the hardware are kept in 274.Va m_pkthdr 275as well. 276See 277.Sx HARDWARE-ASSISTED CHECKSUM CALCULATION 278for details. 279.Pp 280If small enough, data is stored in the internal data buffer of an 281.Vt mbuf . 282If the data is sufficiently large, another 283.Vt mbuf 284may be added to the 285.Vt mbuf chain , 286or external storage may be associated with the 287.Vt mbuf . 288.Dv MHLEN 289bytes of data can fit into an 290.Vt mbuf 291with the 292.Dv M_PKTHDR 293flag set, 294.Dv MLEN 295bytes can otherwise. 296.Pp 297If external storage is being associated with an 298.Vt mbuf , 299the 300.Va m_ext 301header is added at the cost of losing the internal data buffer. 302It includes a pointer to external storage, the size of the storage, 303a pointer to a function used for freeing the storage, 304a pointer to an optional argument that can be passed to the function, 305and a pointer to a reference counter. 306An 307.Vt mbuf 308using external storage has the 309.Dv M_EXT 310flag set. 311.Pp 312The system supplies a macro for allocating the desired external storage 313buffer, 314.Dv MEXTADD . 315.Pp 316The allocation and management of the reference counter is handled by the 317subsystem. 318.Pp 319The system also supplies a default type of external storage buffer called an 320.Vt mbuf cluster . 321.Vt Mbuf clusters 322can be allocated and configured with the use of the 323.Dv MCLGET 324macro. 325Each 326.Vt mbuf cluster 327is 328.Dv MCLBYTES 329in size, where MCLBYTES is a machine-dependent constant. 330The system defines an advisory macro 331.Dv MINCLSIZE , 332which is the smallest amount of data to put into an 333.Vt mbuf cluster . 334It is equal to 335.Dv MHLEN 336plus one. 337It is typically preferable to store data into the data region of an 338.Vt mbuf , 339if size permits, as opposed to allocating a separate 340.Vt mbuf cluster 341to hold the same data. 342.\" 343.Ss Macros and Functions 344There are numerous predefined macros and functions that provide the 345developer with common utilities. 346.\" 347.Bl -ohang -offset indent 348.It Fn mtod mbuf type 349Convert an 350.Fa mbuf 351pointer to a data pointer. 352The macro expands to the data pointer cast to the pointer of the specified 353.Fa type . 354.Sy Note : 355It is advisable to ensure that there is enough contiguous data in 356.Fa mbuf . 357See 358.Fn m_pullup 359for details. 360.It Fn MGET mbuf how type 361Allocate an 362.Vt mbuf 363and initialize it to contain internal data. 364.Fa mbuf 365will point to the allocated 366.Vt mbuf 367on success, or be set to 368.Dv NULL 369on failure. 370The 371.Fa how 372argument is to be set to 373.Dv M_WAITOK 374or 375.Dv M_NOWAIT . 376It specifies whether the caller is willing to block if necessary. 377A number of other functions and macros related to 378.Vt mbufs 379have the same argument because they may 380at some point need to allocate new 381.Vt mbufs . 382.Pp 383Historical 384.Vt mbuf 385allocator (See 386.Sx HISTORY 387section) used allocation flags 388.Dv M_WAIT 389and 390.Dv M_DONTWAIT . 391These constants are kept for compatibility 392and their use in new code is discouraged. 393.It Fn MGETHDR mbuf how type 394Allocate an 395.Vt mbuf 396and initialize it to contain a packet header 397and internal data. 398See 399.Fn MGET 400for details. 401.It Fn MEXTADD mbuf buf size free opt_arg1 opt_arg2 flags type 402Associate externally managed data with 403.Fa mbuf . 404Any internal data contained in the mbuf will be discarded, and the 405.Dv M_EXT 406flag will be set. 407The 408.Fa buf 409and 410.Fa size 411arguments are the address and length, respectively, of the data. 412The 413.Fa free 414argument points to a function which will be called to free the data 415when the mbuf is freed; it is only used if 416.Fa type 417is 418.Dv EXT_EXTREF . 419The 420.Fa opt_arg1 421and 422.Fa opt_arg2 423arguments will be passed unmodified to 424.Fa free . 425The 426.Fa flags 427argument specifies additional 428.Vt mbuf 429flags; it is not necessary to specify 430.Dv M_EXT . 431Finally, the 432.Fa type 433argument specifies the type of external data, which controls how it 434will be disposed of when the 435.Vt mbuf 436is freed. 437In most cases, the correct value is 438.Dv EXT_EXTREF . 439.It Fn MCLGET mbuf how 440Allocate and attach an 441.Vt mbuf cluster 442to 443.Fa mbuf . 444If the macro fails, the 445.Dv M_EXT 446flag will not be set in 447.Fa mbuf . 448.It Fn M_ALIGN mbuf len 449Set the pointer 450.Fa mbuf->m_data 451to place an object of the size 452.Fa len 453at the end of the internal data area of 454.Fa mbuf , 455long word aligned. 456Applicable only if 457.Fa mbuf 458is newly allocated with 459.Fn MGET 460or 461.Fn m_get . 462.It Fn MH_ALIGN mbuf len 463Serves the same purpose as 464.Fn M_ALIGN 465does, but only for 466.Fa mbuf 467newly allocated with 468.Fn MGETHDR 469or 470.Fn m_gethdr , 471or initialized by 472.Fn m_dup_pkthdr 473or 474.Fn m_move_pkthdr . 475.It Fn m_align mbuf len 476Services the same purpose as 477.Fn M_ALIGN 478but handles any type of mbuf. 479.It Fn M_LEADINGSPACE mbuf 480Returns the number of bytes available before the beginning 481of data in 482.Fa mbuf . 483.It Fn M_TRAILINGSPACE mbuf 484Returns the number of bytes available after the end of data in 485.Fa mbuf . 486.It Fn M_PREPEND mbuf len how 487This macro operates on an 488.Vt mbuf chain . 489It is an optimized wrapper for 490.Fn m_prepend 491that can make use of possible empty space before data 492(e.g.\& left after trimming of a link-layer header). 493The new 494.Vt mbuf chain 495pointer or 496.Dv NULL 497is in 498.Fa mbuf 499after the call. 500.It Fn M_MOVE_PKTHDR to from 501Using this macro is equivalent to calling 502.Fn m_move_pkthdr to from . 503.It Fn M_WRITABLE mbuf 504This macro will evaluate true if 505.Fa mbuf 506is not marked 507.Dv M_RDONLY 508and if either 509.Fa mbuf 510does not contain external storage or, 511if it does, 512then if the reference count of the storage is not greater than 1. 513The 514.Dv M_RDONLY 515flag can be set in 516.Fa mbuf->m_flags . 517This can be achieved during setup of the external storage, 518by passing the 519.Dv M_RDONLY 520bit as a 521.Fa flags 522argument to the 523.Fn MEXTADD 524macro, or can be directly set in individual 525.Vt mbufs . 526.It Fn MCHTYPE mbuf type 527Change the type of 528.Fa mbuf 529to 530.Fa type . 531This is a relatively expensive operation and should be avoided. 532.El 533.Pp 534The functions are: 535.Bl -ohang -offset indent 536.It Fn m_get how type 537A function version of 538.Fn MGET 539for non-critical paths. 540.It Fn m_getm orig len how type 541Allocate 542.Fa len 543bytes worth of 544.Vt mbufs 545and 546.Vt mbuf clusters 547if necessary and append the resulting allocated 548.Vt mbuf chain 549to the 550.Vt mbuf chain 551.Fa orig , 552if it is 553.No non- Ns Dv NULL . 554If the allocation fails at any point, 555free whatever was allocated and return 556.Dv NULL . 557If 558.Fa orig 559is 560.No non- Ns Dv NULL , 561it will not be freed. 562It is possible to use 563.Fn m_getm 564to either append 565.Fa len 566bytes to an existing 567.Vt mbuf 568or 569.Vt mbuf chain 570(for example, one which may be sitting in a pre-allocated ring) 571or to simply perform an all-or-nothing 572.Vt mbuf 573and 574.Vt mbuf cluster 575allocation. 576.It Fn m_gethdr how type 577A function version of 578.Fn MGETHDR 579for non-critical paths. 580.It Fn m_getcl how type flags 581Fetch an 582.Vt mbuf 583with a 584.Vt mbuf cluster 585attached to it. 586If one of the allocations fails, the entire allocation fails. 587This routine is the preferred way of fetching both the 588.Vt mbuf 589and 590.Vt mbuf cluster 591together, as it avoids having to unlock/relock between allocations. 592Returns 593.Dv NULL 594on failure. 595.It Fn m_getclr how type 596Allocate an 597.Vt mbuf 598and zero out the data region. 599.It Fn m_free mbuf 600Frees 601.Vt mbuf . 602Returns 603.Va m_next 604of the freed 605.Vt mbuf . 606.El 607.Pp 608The functions below operate on 609.Vt mbuf chains . 610.Bl -ohang -offset indent 611.It Fn m_freem mbuf 612Free an entire 613.Vt mbuf chain , 614including any external storage. 615.\" 616.It Fn m_adj mbuf len 617Trim 618.Fa len 619bytes from the head of an 620.Vt mbuf chain 621if 622.Fa len 623is positive, from the tail otherwise. 624.\" 625.It Fn m_append mbuf len cp 626Append 627.Vt len 628bytes of data 629.Vt cp 630to the 631.Vt mbuf chain . 632Extend the mbuf chain if the new data does not fit in 633existing space. 634.\" 635.It Fn m_prepend mbuf len how 636Allocate a new 637.Vt mbuf 638and prepend it to the 639.Vt mbuf chain , 640handle 641.Dv M_PKTHDR 642properly. 643.Sy Note : 644It does not allocate any 645.Vt mbuf clusters , 646so 647.Fa len 648must be less than 649.Dv MLEN 650or 651.Dv MHLEN , 652depending on the 653.Dv M_PKTHDR 654flag setting. 655.\" 656.It Fn m_copyup mbuf len dstoff 657Similar to 658.Fn m_pullup 659but copies 660.Fa len 661bytes of data into a new mbuf at 662.Fa dstoff 663bytes into the mbuf. 664The 665.Fa dstoff 666argument aligns the data and leaves room for a link layer header. 667Returns the new 668.Vt mbuf chain 669on success, 670and frees the 671.Vt mbuf chain 672and returns 673.Dv NULL 674on failure. 675.Sy Note : 676The function does not allocate 677.Vt mbuf clusters , 678so 679.Fa len + dstoff 680must be less than 681.Dv MHLEN . 682.\" 683.It Fn m_pullup mbuf len 684Arrange that the first 685.Fa len 686bytes of an 687.Vt mbuf chain 688are contiguous and lay in the data area of 689.Fa mbuf , 690so they are accessible with 691.Fn mtod mbuf type . 692It is important to remember that this may involve 693reallocating some mbufs and moving data so all pointers 694referencing data within the old mbuf chain 695must be recalculated or made invalid. 696Return the new 697.Vt mbuf chain 698on success, 699.Dv NULL 700on failure 701(the 702.Vt mbuf chain 703is freed in this case). 704.Sy Note : 705It does not allocate any 706.Vt mbuf clusters , 707so 708.Fa len 709must be less than 710.Dv MHLEN . 711.\" 712.It Fn m_pulldown mbuf offset len offsetp 713Arrange that 714.Fa len 715bytes between 716.Fa offset 717and 718.Fa offset + len 719in the 720.Vt mbuf chain 721are contiguous and lay in the data area of 722.Fa mbuf , 723so they are accessible with 724.Fn mtod mbuf type . 725.Fa len 726must be smaller than, or equal to, the size of an 727.Vt mbuf cluster . 728Return a pointer to an intermediate 729.Vt mbuf 730in the chain containing the requested region; 731the offset in the data region of the 732.Vt mbuf chain 733to the data contained in the returned mbuf is stored in 734.Fa *offsetp . 735If 736.Fa offp 737is NULL, the region may be accessed using 738.Fn mtod mbuf type . 739If 740.Fa offp 741is non-NULL, the region may be accessed using 742.Fn mtod mbuf uint8_t + *offsetp . 743The region of the mbuf chain between its beginning and 744.Fa off 745is not modified, therefore it is safe to hold pointers to data within 746this region before calling 747.Fn m_pulldown . 748.\" 749.It Fn m_copym mbuf offset len how 750Make a copy of an 751.Vt mbuf chain 752starting 753.Fa offset 754bytes from the beginning, continuing for 755.Fa len 756bytes. 757If 758.Fa len 759is 760.Dv M_COPYALL , 761copy to the end of the 762.Vt mbuf chain . 763.Sy Note : 764The copy is read-only, because the 765.Vt mbuf clusters 766are not copied, only their reference counts are incremented. 767.\" 768.It Fn m_copypacket mbuf how 769Copy an entire packet including header, which must be present. 770This is an optimized version of the common case 771.Fn m_copym mbuf 0 M_COPYALL how . 772.Sy Note : 773the copy is read-only, because the 774.Vt mbuf clusters 775are not copied, only their reference counts are incremented. 776.\" 777.It Fn m_dup mbuf how 778Copy a packet header 779.Vt mbuf chain 780into a completely new 781.Vt mbuf chain , 782including copying any 783.Vt mbuf clusters . 784Use this instead of 785.Fn m_copypacket 786when you need a writable copy of an 787.Vt mbuf chain . 788.\" 789.It Fn m_copydata mbuf offset len buf 790Copy data from an 791.Vt mbuf chain 792starting 793.Fa off 794bytes from the beginning, continuing for 795.Fa len 796bytes, into the indicated buffer 797.Fa buf . 798.\" 799.It Fn m_copyback mbuf offset len buf 800Copy 801.Fa len 802bytes from the buffer 803.Fa buf 804back into the indicated 805.Vt mbuf chain , 806starting at 807.Fa offset 808bytes from the beginning of the 809.Vt mbuf chain , 810extending the 811.Vt mbuf chain 812if necessary. 813.Sy Note : 814It does not allocate any 815.Vt mbuf clusters , 816just adds 817.Vt mbufs 818to the 819.Vt mbuf chain . 820It is safe to set 821.Fa offset 822beyond the current 823.Vt mbuf chain 824end: zeroed 825.Vt mbufs 826will be allocated to fill the space. 827.\" 828.It Fn m_length mbuf last 829Return the length of the 830.Vt mbuf chain , 831and optionally a pointer to the last 832.Vt mbuf . 833.\" 834.It Fn m_dup_pkthdr to from how 835Upon the function's completion, the 836.Vt mbuf 837.Fa to 838will contain an identical copy of 839.Fa from->m_pkthdr 840and the per-packet attributes found in the 841.Vt mbuf chain 842.Fa from . 843The 844.Vt mbuf 845.Fa from 846must have the flag 847.Dv M_PKTHDR 848initially set, and 849.Fa to 850must be empty on entry. 851.\" 852.It Fn m_move_pkthdr to from 853Move 854.Va m_pkthdr 855and the per-packet attributes from the 856.Vt mbuf chain 857.Fa from 858to the 859.Vt mbuf 860.Fa to . 861The 862.Vt mbuf 863.Fa from 864must have the flag 865.Dv M_PKTHDR 866initially set, and 867.Fa to 868must be empty on entry. 869Upon the function's completion, 870.Fa from 871will have the flag 872.Dv M_PKTHDR 873and the per-packet attributes cleared. 874.\" 875.It Fn m_fixhdr mbuf 876Set the packet-header length to the length of the 877.Vt mbuf chain . 878.\" 879.It Fn m_devget buf len offset ifp copy 880Copy data from a device local memory pointed to by 881.Fa buf 882to an 883.Vt mbuf chain . 884The copy is done using a specified copy routine 885.Fa copy , 886or 887.Fn bcopy 888if 889.Fa copy 890is 891.Dv NULL . 892.\" 893.It Fn m_cat m n 894Concatenate 895.Fa n 896to 897.Fa m . 898Both 899.Vt mbuf chains 900must be of the same type. 901.Fa N 902is still valid after the function returned. 903.Sy Note : 904It does not handle 905.Dv M_PKTHDR 906and friends. 907.\" 908.It Fn m_split mbuf len how 909Partition an 910.Vt mbuf chain 911in two pieces, returning the tail: 912all but the first 913.Fa len 914bytes. 915In case of failure, it returns 916.Dv NULL 917and attempts to restore the 918.Vt mbuf chain 919to its original state. 920.\" 921.It Fn m_apply mbuf off len f arg 922Apply a function to an 923.Vt mbuf chain , 924at offset 925.Fa off , 926for length 927.Fa len 928bytes. 929Typically used to avoid calls to 930.Fn m_pullup 931which would otherwise be unnecessary or undesirable. 932.Fa arg 933is a convenience argument which is passed to the callback function 934.Fa f . 935.Pp 936Each time 937.Fn f 938is called, it will be passed 939.Fa arg , 940a pointer to the 941.Fa data 942in the current mbuf, and the length 943.Fa len 944of the data in this mbuf to which the function should be applied. 945.Pp 946The function should return zero to indicate success; 947otherwise, if an error is indicated, then 948.Fn m_apply 949will return the error and stop iterating through the 950.Vt mbuf chain . 951.\" 952.It Fn m_getptr mbuf loc off 953Return a pointer to the mbuf containing the data located at 954.Fa loc 955bytes from the beginning of the 956.Vt mbuf chain . 957The corresponding offset into the mbuf will be stored in 958.Fa *off . 959.It Fn m_defrag m0 how 960Defragment an mbuf chain, returning the shortest possible 961chain of mbufs and clusters. 962If allocation fails and this can not be completed, 963.Dv NULL 964will be returned and the original chain will be unchanged. 965Upon success, the original chain will be freed and the new 966chain will be returned. 967.Fa how 968should be either 969.Dv M_WAITOK 970or 971.Dv M_NOWAIT , 972depending on the caller's preference. 973.Pp 974This function is especially useful in network drivers, where 975certain long mbuf chains must be shortened before being added 976to TX descriptor lists. 977.It Fn m_unshare m0 how 978Create a version of the specified mbuf chain whose 979contents can be safely modified without affecting other users. 980If allocation fails and this operation can not be completed, 981.Dv NULL 982will be returned. 983The original mbuf chain is always reclaimed and the reference 984count of any shared mbuf clusters is decremented. 985.Fa how 986should be either 987.Dv M_WAITOK 988or 989.Dv M_NOWAIT , 990depending on the caller's preference. 991As a side-effect of this process the returned 992mbuf chain may be compacted. 993.Pp 994This function is especially useful in the transmit path of 995network code, when data must be encrypted or otherwise 996altered prior to transmission. 997.El 998.Sh HARDWARE-ASSISTED CHECKSUM CALCULATION 999This section currently applies to TCP/IP only. 1000In order to save the host CPU resources, computing checksums is 1001offloaded to the network interface hardware if possible. 1002The 1003.Va m_pkthdr 1004member of the leading 1005.Vt mbuf 1006of a packet contains two fields used for that purpose, 1007.Vt int Va csum_flags 1008and 1009.Vt int Va csum_data . 1010The meaning of those fields depends on the direction a packet flows in, 1011and on whether the packet is fragmented. 1012Henceforth, 1013.Va csum_flags 1014or 1015.Va csum_data 1016of a packet 1017will denote the corresponding field of the 1018.Va m_pkthdr 1019member of the leading 1020.Vt mbuf 1021in the 1022.Vt mbuf chain 1023containing the packet. 1024.Pp 1025On output, checksum offloading is attempted after the outgoing 1026interface has been determined for a packet. 1027The interface-specific field 1028.Va ifnet.if_data.ifi_hwassist 1029(see 1030.Xr ifnet 9 ) 1031is consulted for the capabilities of the interface to assist in 1032computing checksums. 1033The 1034.Va csum_flags 1035field of the packet header is set to indicate which actions the interface 1036is supposed to perform on it. 1037The actions unsupported by the network interface are done in the 1038software prior to passing the packet down to the interface driver; 1039such actions will never be requested through 1040.Va csum_flags . 1041.Pp 1042The flags demanding a particular action from an interface are as follows: 1043.Bl -tag -width ".Dv CSUM_TCP" -offset indent 1044.It Dv CSUM_IP 1045The IP header checksum is to be computed and stored in the 1046corresponding field of the packet. 1047The hardware is expected to know the format of an IP header 1048to determine the offset of the IP checksum field. 1049.It Dv CSUM_TCP 1050The TCP checksum is to be computed. 1051(See below.) 1052.It Dv CSUM_UDP 1053The UDP checksum is to be computed. 1054(See below.) 1055.El 1056.Pp 1057Should a TCP or UDP checksum be offloaded to the hardware, 1058the field 1059.Va csum_data 1060will contain the byte offset of the checksum field relative to the 1061end of the IP header. 1062In this case, the checksum field will be initially 1063set by the TCP/IP module to the checksum of the pseudo header 1064defined by the TCP and UDP specifications. 1065.Pp 1066For outbound packets which have been fragmented 1067by the host CPU, the following will also be true, 1068regardless of the checksum flag settings: 1069.Bl -bullet -offset indent 1070.It 1071all fragments will have the flag 1072.Dv M_FRAG 1073set in their 1074.Va m_flags 1075field; 1076.It 1077the first and the last fragments in the chain will have 1078.Dv M_FIRSTFRAG 1079or 1080.Dv M_LASTFRAG 1081set in their 1082.Va m_flags , 1083correspondingly; 1084.It 1085the first fragment in the chain will have the total number 1086of fragments contained in its 1087.Va csum_data 1088field. 1089.El 1090.Pp 1091The last rule for fragmented packets takes precedence over the one 1092for a TCP or UDP checksum. 1093Nevertheless, offloading a TCP or UDP checksum is possible for a 1094fragmented packet if the flag 1095.Dv CSUM_IP_FRAGS 1096is set in the field 1097.Va ifnet.if_data.ifi_hwassist 1098associated with the network interface. 1099However, in this case the interface is expected to figure out 1100the location of the checksum field within the sequence of fragments 1101by itself because 1102.Va csum_data 1103contains a fragment count instead of a checksum offset value. 1104.Pp 1105On input, an interface indicates the actions it has performed 1106on a packet by setting one or more of the following flags in 1107.Va csum_flags 1108associated with the packet: 1109.Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent 1110.It Dv CSUM_IP_CHECKED 1111The IP header checksum has been computed. 1112.It Dv CSUM_IP_VALID 1113The IP header has a valid checksum. 1114This flag can appear only in combination with 1115.Dv CSUM_IP_CHECKED . 1116.It Dv CSUM_DATA_VALID 1117The checksum of the data portion of the IP packet has been computed 1118and stored in the field 1119.Va csum_data 1120in network byte order. 1121.It Dv CSUM_PSEUDO_HDR 1122Can be set only along with 1123.Dv CSUM_DATA_VALID 1124to indicate that the IP data checksum found in 1125.Va csum_data 1126allows for the pseudo header defined by the TCP and UDP specifications. 1127Otherwise the checksum of the pseudo header must be calculated by 1128the host CPU and added to 1129.Va csum_data 1130to obtain the final checksum to be used for TCP or UDP validation purposes. 1131.El 1132.Pp 1133If a particular network interface just indicates success or 1134failure of TCP or UDP checksum validation without returning 1135the exact value of the checksum to the host CPU, its driver can mark 1136.Dv CSUM_DATA_VALID 1137and 1138.Dv CSUM_PSEUDO_HDR 1139in 1140.Va csum_flags , 1141and set 1142.Va csum_data 1143to 1144.Li 0xFFFF 1145hexadecimal to indicate a valid checksum. 1146It is a peculiarity of the algorithm used that the Internet checksum 1147calculated over any valid packet will be 1148.Li 0xFFFF 1149as long as the original checksum field is included. 1150.Pp 1151For inbound packets which are IP fragments, all 1152.Va csum_data 1153fields will be summed during reassembly to obtain the final checksum 1154value passed to an upper layer in the 1155.Va csum_data 1156field of the reassembled packet. 1157The 1158.Va csum_flags 1159fields of all fragments will be consolidated using logical AND 1160to obtain the final value for 1161.Va csum_flags . 1162Thus, in order to successfully 1163offload checksum computation for fragmented data, 1164all fragments should have the same value of 1165.Va csum_flags . 1166.Sh STRESS TESTING 1167When running a kernel compiled with the option 1168.Dv MBUF_STRESS_TEST , 1169the following 1170.Xr sysctl 8 Ns 1171-controlled options may be used to create 1172various failure/extreme cases for testing of network drivers 1173and other parts of the kernel that rely on 1174.Vt mbufs . 1175.Bl -tag -width ident 1176.It Va net.inet.ip.mbuf_frag_size 1177Causes 1178.Fn ip_output 1179to fragment outgoing 1180.Vt mbuf chains 1181into fragments of the specified size. 1182Setting this variable to 1 is an excellent way to 1183test the long 1184.Vt mbuf chain 1185handling ability of network drivers. 1186.It Va kern.ipc.m_defragrandomfailures 1187Causes the function 1188.Fn m_defrag 1189to randomly fail, returning 1190.Dv NULL . 1191Any piece of code which uses 1192.Fn m_defrag 1193should be tested with this feature. 1194.El 1195.Sh RETURN VALUES 1196See above. 1197.Sh SEE ALSO 1198.Xr ifnet 9 , 1199.Xr mbuf_tags 9 1200.Sh HISTORY 1201.\" Please correct me if I'm wrong 1202.Vt Mbufs 1203appeared in an early version of 1204.Bx . 1205Besides being used for network packets, they were used 1206to store various dynamic structures, such as routing table 1207entries, interface addresses, protocol control blocks, etc. 1208In more recent 1209.Fx 1210use of 1211.Vt mbufs 1212is almost entirely limited to packet storage, with 1213.Xr uma 9 1214zones being used directly to store other network-related memory. 1215.Pp 1216Historically, the 1217.Vt mbuf 1218allocator has been a special-purpose memory allocator able to run in 1219interrupt contexts and allocating from a special kernel address space map. 1220As of 1221.Fx 5.3 , 1222the 1223.Vt mbuf 1224allocator is a wrapper around 1225.Xr uma 9 , 1226allowing caching of 1227.Vt mbufs , 1228clusters, and 1229.Vt mbuf 1230+ cluster pairs in per-CPU caches, as well as bringing other benefits of 1231slab allocation. 1232.Sh AUTHORS 1233The original 1234.Nm 1235manual page was written by Yar Tikhiy. 1236The 1237.Xr uma 9 1238.Vt mbuf 1239allocator was written by Bosko Milekic. 1240