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 October 10, 2016 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.Ft int 44.Fn MCLGET "struct mbuf *mbuf" "int how" 45.Fo MEXTADD 46.Fa "struct mbuf *mbuf" 47.Fa "caddr_t buf" 48.Fa "u_int size" 49.Fa "void (*free)(void *opt_arg1, void *opt_arg2)" 50.Fa "void *opt_arg1" 51.Fa "void *opt_arg2" 52.Fa "short flags" 53.Fa "int type" 54.Fc 55.\" 56.Ss Mbuf utility macros 57.Fn mtod "struct mbuf *mbuf" "type" 58.Fn M_ALIGN "struct mbuf *mbuf" "u_int len" 59.Fn MH_ALIGN "struct mbuf *mbuf" "u_int len" 60.Ft int 61.Fn M_LEADINGSPACE "struct mbuf *mbuf" 62.Ft int 63.Fn M_TRAILINGSPACE "struct mbuf *mbuf" 64.Fn M_MOVE_PKTHDR "struct mbuf *to" "struct mbuf *from" 65.Fn M_PREPEND "struct mbuf *mbuf" "int len" "int how" 66.Fn MCHTYPE "struct mbuf *mbuf" "short type" 67.Ft int 68.Fn M_WRITABLE "struct mbuf *mbuf" 69.\" 70.Ss Mbuf allocation functions 71.Ft struct mbuf * 72.Fn m_get "int how" "short type" 73.Ft struct mbuf * 74.Fn m_get2 "int size" "int how" "short type" "int flags" 75.Ft struct mbuf * 76.Fn m_getm "struct mbuf *orig" "int len" "int how" "short type" 77.Ft struct mbuf * 78.Fn m_getjcl "int how" "short type" "int flags" "int size" 79.Ft struct mbuf * 80.Fn m_getcl "int how" "short type" "int flags" 81.Ft struct mbuf * 82.Fn m_gethdr "int how" "short type" 83.Ft struct mbuf * 84.Fn m_free "struct mbuf *mbuf" 85.Ft void 86.Fn m_freem "struct mbuf *mbuf" 87.\" 88.Ss Mbuf utility functions 89.Ft void 90.Fn m_adj "struct mbuf *mbuf" "int len" 91.Ft void 92.Fn m_align "struct mbuf *mbuf" "int len" 93.Ft int 94.Fn m_append "struct mbuf *mbuf" "int len" "c_caddr_t cp" 95.Ft struct mbuf * 96.Fn m_prepend "struct mbuf *mbuf" "int len" "int how" 97.Ft struct mbuf * 98.Fn m_copyup "struct mbuf *mbuf" "int len" "int dstoff" 99.Ft struct mbuf * 100.Fn m_pullup "struct mbuf *mbuf" "int len" 101.Ft struct mbuf * 102.Fn m_pulldown "struct mbuf *mbuf" "int offset" "int len" "int *offsetp" 103.Ft struct mbuf * 104.Fn m_copym "struct mbuf *mbuf" "int offset" "int len" "int how" 105.Ft struct mbuf * 106.Fn m_copypacket "struct mbuf *mbuf" "int how" 107.Ft struct mbuf * 108.Fn m_dup "const struct mbuf *mbuf" "int how" 109.Ft void 110.Fn m_copydata "const struct mbuf *mbuf" "int offset" "int len" "caddr_t buf" 111.Ft void 112.Fn m_copyback "struct mbuf *mbuf" "int offset" "int len" "caddr_t buf" 113.Ft struct mbuf * 114.Fo m_devget 115.Fa "char *buf" 116.Fa "int len" 117.Fa "int offset" 118.Fa "struct ifnet *ifp" 119.Fa "void (*copy)(char *from, caddr_t to, u_int len)" 120.Fc 121.Ft void 122.Fn m_cat "struct mbuf *m" "struct mbuf *n" 123.Ft void 124.Fn m_catpkt "struct mbuf *m" "struct mbuf *n" 125.Ft u_int 126.Fn m_fixhdr "struct mbuf *mbuf" 127.Ft int 128.Fn m_dup_pkthdr "struct mbuf *to" "const struct mbuf *from" "int how" 129.Ft void 130.Fn m_move_pkthdr "struct mbuf *to" "struct mbuf *from" 131.Ft u_int 132.Fn m_length "struct mbuf *mbuf" "struct mbuf **last" 133.Ft struct mbuf * 134.Fn m_split "struct mbuf *mbuf" "int len" "int how" 135.Ft int 136.Fn m_apply "struct mbuf *mbuf" "int off" "int len" "int (*f)(void *arg, void *data, u_int len)" "void *arg" 137.Ft struct mbuf * 138.Fn m_getptr "struct mbuf *mbuf" "int loc" "int *off" 139.Ft struct mbuf * 140.Fn m_defrag "struct mbuf *m0" "int how" 141.Ft struct mbuf * 142.Fn m_collapse "struct mbuf *m0" "int how" "int maxfrags" 143.Ft struct mbuf * 144.Fn m_unshare "struct mbuf *m0" "int how" 145.\" 146.Sh DESCRIPTION 147An 148.Vt mbuf 149is a basic unit of memory management in the kernel IPC subsystem. 150Network packets and socket buffers are stored in 151.Vt mbufs . 152A network packet may span multiple 153.Vt mbufs 154arranged into a 155.Vt mbuf chain 156(linked list), 157which allows adding or trimming 158network headers with little overhead. 159.Pp 160While a developer should not bother with 161.Vt mbuf 162internals without serious 163reason in order to avoid incompatibilities with future changes, it 164is useful to understand the general structure of an 165.Vt mbuf . 166.Pp 167An 168.Vt mbuf 169consists of a variable-sized header and a small internal 170buffer for data. 171The total size of an 172.Vt mbuf , 173.Dv MSIZE , 174is a constant defined in 175.In sys/param.h . 176The 177.Vt mbuf 178header includes: 179.Bl -tag -width "m_nextpkt" -offset indent 180.It Va m_next 181.Pq Vt struct mbuf * 182A pointer to the next 183.Vt mbuf 184in the 185.Vt mbuf chain . 186.It Va m_nextpkt 187.Pq Vt struct mbuf * 188A pointer to the next 189.Vt mbuf chain 190in the queue. 191.It Va m_data 192.Pq Vt caddr_t 193A pointer to data attached to this 194.Vt mbuf . 195.It Va m_len 196.Pq Vt int 197The length of the data. 198.It Va m_type 199.Pq Vt short 200The type of the data. 201.It Va m_flags 202.Pq Vt int 203The 204.Vt mbuf 205flags. 206.El 207.Pp 208The 209.Vt mbuf 210flag bits are defined as follows: 211.Bd -literal 212/* mbuf flags */ 213#define M_EXT 0x00000001 /* has associated external storage */ 214#define M_PKTHDR 0x00000002 /* start of record */ 215#define M_EOR 0x00000004 /* end of record */ 216#define M_RDONLY 0x00000008 /* associated data marked read-only */ 217#define M_PROTO1 0x00001000 /* protocol-specific */ 218#define M_PROTO2 0x00002000 /* protocol-specific */ 219#define M_PROTO3 0x00004000 /* protocol-specific */ 220#define M_PROTO4 0x00008000 /* protocol-specific */ 221#define M_PROTO5 0x00010000 /* protocol-specific */ 222#define M_PROTO6 0x00020000 /* protocol-specific */ 223#define M_PROTO7 0x00040000 /* protocol-specific */ 224#define M_PROTO8 0x00080000 /* protocol-specific */ 225#define M_PROTO9 0x00100000 /* protocol-specific */ 226#define M_PROTO10 0x00200000 /* protocol-specific */ 227#define M_PROTO11 0x00400000 /* protocol-specific */ 228#define M_PROTO12 0x00800000 /* protocol-specific */ 229 230/* mbuf pkthdr flags (also stored in m_flags) */ 231#define M_BCAST 0x00000010 /* send/received as link-level broadcast */ 232#define M_MCAST 0x00000020 /* send/received as link-level multicast */ 233.Ed 234.Pp 235The available 236.Vt mbuf 237types are defined as follows: 238.Bd -literal 239/* mbuf types */ 240#define MT_DATA 1 /* dynamic (data) allocation */ 241#define MT_HEADER MT_DATA /* packet header */ 242#define MT_SONAME 8 /* socket name */ 243#define MT_CONTROL 14 /* extra-data protocol message */ 244#define MT_OOBDATA 15 /* expedited data */ 245.Ed 246.Pp 247The available external buffer types are defined as follows: 248.Bd -literal 249/* external buffer types */ 250#define EXT_CLUSTER 1 /* mbuf cluster */ 251#define EXT_SFBUF 2 /* sendfile(2)'s sf_bufs */ 252#define EXT_JUMBOP 3 /* jumbo cluster 4096 bytes */ 253#define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */ 254#define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */ 255#define EXT_PACKET 6 /* mbuf+cluster from packet zone */ 256#define EXT_MBUF 7 /* external mbuf reference */ 257#define EXT_NET_DRV 252 /* custom ext_buf provided by net driver(s) */ 258#define EXT_MOD_TYPE 253 /* custom module's ext_buf type */ 259#define EXT_DISPOSABLE 254 /* can throw this buffer away w/page flipping */ 260#define EXT_EXTREF 255 /* has externally maintained ref_cnt ptr */ 261.Ed 262.Pp 263If the 264.Dv M_PKTHDR 265flag is set, a 266.Vt struct pkthdr Va m_pkthdr 267is added to the 268.Vt mbuf 269header. 270It contains a pointer to the interface 271the packet has been received from 272.Pq Vt struct ifnet Va *rcvif , 273and the total packet length 274.Pq Vt int Va len . 275Optionally, it may also contain an attached list of packet tags 276.Pq Vt "struct m_tag" . 277See 278.Xr mbuf_tags 9 279for details. 280Fields used in offloading checksum calculation to the hardware are kept in 281.Va m_pkthdr 282as well. 283See 284.Sx HARDWARE-ASSISTED CHECKSUM CALCULATION 285for details. 286.Pp 287If small enough, data is stored in the internal data buffer of an 288.Vt mbuf . 289If the data is sufficiently large, another 290.Vt mbuf 291may be added to the 292.Vt mbuf chain , 293or external storage may be associated with the 294.Vt mbuf . 295.Dv MHLEN 296bytes of data can fit into an 297.Vt mbuf 298with the 299.Dv M_PKTHDR 300flag set, 301.Dv MLEN 302bytes can otherwise. 303.Pp 304If external storage is being associated with an 305.Vt mbuf , 306the 307.Va m_ext 308header is added at the cost of losing the internal data buffer. 309It includes a pointer to external storage, the size of the storage, 310a pointer to a function used for freeing the storage, 311a pointer to an optional argument that can be passed to the function, 312and a pointer to a reference counter. 313An 314.Vt mbuf 315using external storage has the 316.Dv M_EXT 317flag set. 318.Pp 319The system supplies a macro for allocating the desired external storage 320buffer, 321.Dv MEXTADD . 322.Pp 323The allocation and management of the reference counter is handled by the 324subsystem. 325.Pp 326The system also supplies a default type of external storage buffer called an 327.Vt mbuf cluster . 328.Vt Mbuf clusters 329can be allocated and configured with the use of the 330.Dv MCLGET 331macro. 332Each 333.Vt mbuf cluster 334is 335.Dv MCLBYTES 336in size, where MCLBYTES is a machine-dependent constant. 337The system defines an advisory macro 338.Dv MINCLSIZE , 339which is the smallest amount of data to put into an 340.Vt mbuf cluster . 341It is equal to 342.Dv MHLEN 343plus one. 344It is typically preferable to store data into the data region of an 345.Vt mbuf , 346if size permits, as opposed to allocating a separate 347.Vt mbuf cluster 348to hold the same data. 349.\" 350.Ss Macros and Functions 351There are numerous predefined macros and functions that provide the 352developer with common utilities. 353.\" 354.Bl -ohang -offset indent 355.It Fn mtod mbuf type 356Convert an 357.Fa mbuf 358pointer to a data pointer. 359The macro expands to the data pointer cast to the specified 360.Fa type . 361.Sy Note : 362It is advisable to ensure that there is enough contiguous data in 363.Fa mbuf . 364See 365.Fn m_pullup 366for details. 367.It Fn MGET mbuf how type 368Allocate an 369.Vt mbuf 370and initialize it to contain internal data. 371.Fa mbuf 372will point to the allocated 373.Vt mbuf 374on success, or be set to 375.Dv NULL 376on failure. 377The 378.Fa how 379argument is to be set to 380.Dv M_WAITOK 381or 382.Dv M_NOWAIT . 383It specifies whether the caller is willing to block if necessary. 384A number of other functions and macros related to 385.Vt mbufs 386have the same argument because they may 387at some point need to allocate new 388.Vt mbufs . 389.It Fn MGETHDR mbuf how type 390Allocate an 391.Vt mbuf 392and initialize it to contain a packet header 393and internal data. 394See 395.Fn MGET 396for details. 397.It Fn MEXTADD mbuf buf size free opt_arg1 opt_arg2 flags type 398Associate externally managed data with 399.Fa mbuf . 400Any internal data contained in the mbuf will be discarded, and the 401.Dv M_EXT 402flag will be set. 403The 404.Fa buf 405and 406.Fa size 407arguments are the address and length, respectively, of the data. 408The 409.Fa free 410argument points to a function which will be called to free the data 411when the mbuf is freed; it is only used if 412.Fa type 413is 414.Dv EXT_EXTREF . 415The 416.Fa opt_arg1 417and 418.Fa opt_arg2 419arguments will be passed unmodified to 420.Fa free . 421The 422.Fa flags 423argument specifies additional 424.Vt mbuf 425flags; it is not necessary to specify 426.Dv M_EXT . 427Finally, the 428.Fa type 429argument specifies the type of external data, which controls how it 430will be disposed of when the 431.Vt mbuf 432is freed. 433In most cases, the correct value is 434.Dv EXT_EXTREF . 435.It Fn MCLGET mbuf how 436Allocate and attach an 437.Vt mbuf cluster 438to 439.Fa mbuf . 440On success, a non-zero value returned; otherwise, 0. 441Historically, consumers would check for success by testing the 442.Dv M_EXT 443flag on the mbuf, but this is now discouraged to avoid unnecessary awareness 444of the implementation of external storage in protocol stacks and device 445drivers. 446.It Fn M_ALIGN mbuf len 447Set the pointer 448.Fa mbuf->m_data 449to place an object of the size 450.Fa len 451at the end of the internal data area of 452.Fa mbuf , 453long word aligned. 454Applicable only if 455.Fa mbuf 456is newly allocated with 457.Fn MGET 458or 459.Fn m_get . 460.It Fn MH_ALIGN mbuf len 461Serves the same purpose as 462.Fn M_ALIGN 463does, but only for 464.Fa mbuf 465newly allocated with 466.Fn MGETHDR 467or 468.Fn m_gethdr , 469or initialized by 470.Fn m_dup_pkthdr 471or 472.Fn m_move_pkthdr . 473.It Fn m_align mbuf len 474Services the same purpose as 475.Fn M_ALIGN 476but handles any type of mbuf. 477.It Fn M_LEADINGSPACE mbuf 478Returns the number of bytes available before the beginning 479of data in 480.Fa mbuf . 481.It Fn M_TRAILINGSPACE mbuf 482Returns the number of bytes available after the end of data in 483.Fa mbuf . 484.It Fn M_PREPEND mbuf len how 485This macro operates on an 486.Vt mbuf chain . 487It is an optimized wrapper for 488.Fn m_prepend 489that can make use of possible empty space before data 490(e.g.\& left after trimming of a link-layer header). 491The new 492.Vt mbuf chain 493pointer or 494.Dv NULL 495is in 496.Fa mbuf 497after the call. 498.It Fn M_MOVE_PKTHDR to from 499Using this macro is equivalent to calling 500.Fn m_move_pkthdr to from . 501.It Fn M_WRITABLE mbuf 502This macro will evaluate true if 503.Fa mbuf 504is not marked 505.Dv M_RDONLY 506and if either 507.Fa mbuf 508does not contain external storage or, 509if it does, 510then if the reference count of the storage is not greater than 1. 511The 512.Dv M_RDONLY 513flag can be set in 514.Fa mbuf->m_flags . 515This can be achieved during setup of the external storage, 516by passing the 517.Dv M_RDONLY 518bit as a 519.Fa flags 520argument to the 521.Fn MEXTADD 522macro, or can be directly set in individual 523.Vt mbufs . 524.It Fn MCHTYPE mbuf type 525Change the type of 526.Fa mbuf 527to 528.Fa type . 529This is a relatively expensive operation and should be avoided. 530.El 531.Pp 532The functions are: 533.Bl -ohang -offset indent 534.It Fn m_get how type 535A function version of 536.Fn MGET 537for non-critical paths. 538.It Fn m_get2 size how type flags 539Allocate an 540.Vt mbuf 541with enough space to hold specified amount of data. 542.It Fn m_getm orig len how type 543Allocate 544.Fa len 545bytes worth of 546.Vt mbufs 547and 548.Vt mbuf clusters 549if necessary and append the resulting allocated 550.Vt mbuf chain 551to the 552.Vt mbuf chain 553.Fa orig , 554if it is 555.No non- Ns Dv NULL . 556If the allocation fails at any point, 557free whatever was allocated and return 558.Dv NULL . 559If 560.Fa orig 561is 562.No non- Ns Dv NULL , 563it will not be freed. 564It is possible to use 565.Fn m_getm 566to either append 567.Fa len 568bytes to an existing 569.Vt mbuf 570or 571.Vt mbuf chain 572(for example, one which may be sitting in a pre-allocated ring) 573or to simply perform an all-or-nothing 574.Vt mbuf 575and 576.Vt mbuf cluster 577allocation. 578.It Fn m_gethdr how type 579A function version of 580.Fn MGETHDR 581for non-critical paths. 582.It Fn m_getcl how type flags 583Fetch an 584.Vt mbuf 585with a 586.Vt mbuf cluster 587attached to it. 588If one of the allocations fails, the entire allocation fails. 589This routine is the preferred way of fetching both the 590.Vt mbuf 591and 592.Vt mbuf cluster 593together, as it avoids having to unlock/relock between allocations. 594Returns 595.Dv NULL 596on failure. 597.It Fn m_getjcl how type flags size 598This is like 599.Fn m_getcl 600but it the size of the cluster allocated will be large enough for 601.Fa size 602bytes. 603.It Fn m_free mbuf 604Frees 605.Vt mbuf . 606Returns 607.Va m_next 608of the freed 609.Vt mbuf . 610.El 611.Pp 612The functions below operate on 613.Vt mbuf chains . 614.Bl -ohang -offset indent 615.It Fn m_freem mbuf 616Free an entire 617.Vt mbuf chain , 618including any external storage. 619.\" 620.It Fn m_adj mbuf len 621Trim 622.Fa len 623bytes from the head of an 624.Vt mbuf chain 625if 626.Fa len 627is positive, from the tail otherwise. 628.\" 629.It Fn m_append mbuf len cp 630Append 631.Vt len 632bytes of data 633.Vt cp 634to the 635.Vt mbuf chain . 636Extend the mbuf chain if the new data does not fit in 637existing space. 638.\" 639.It Fn m_prepend mbuf len how 640Allocate a new 641.Vt mbuf 642and prepend it to the 643.Vt mbuf chain , 644handle 645.Dv M_PKTHDR 646properly. 647.Sy Note : 648It does not allocate any 649.Vt mbuf clusters , 650so 651.Fa len 652must be less than 653.Dv MLEN 654or 655.Dv MHLEN , 656depending on the 657.Dv M_PKTHDR 658flag setting. 659.\" 660.It Fn m_copyup mbuf len dstoff 661Similar to 662.Fn m_pullup 663but copies 664.Fa len 665bytes of data into a new mbuf at 666.Fa dstoff 667bytes into the mbuf. 668The 669.Fa dstoff 670argument aligns the data and leaves room for a link layer header. 671Returns the new 672.Vt mbuf chain 673on success, 674and frees the 675.Vt mbuf chain 676and returns 677.Dv NULL 678on failure. 679.Sy Note : 680The function does not allocate 681.Vt mbuf clusters , 682so 683.Fa len + dstoff 684must be less than 685.Dv MHLEN . 686.\" 687.It Fn m_pullup mbuf len 688Arrange that the first 689.Fa len 690bytes of an 691.Vt mbuf chain 692are contiguous and lay in the data area of 693.Fa mbuf , 694so they are accessible with 695.Fn mtod mbuf type . 696It is important to remember that this may involve 697reallocating some mbufs and moving data so all pointers 698referencing data within the old mbuf chain 699must be recalculated or made invalid. 700Return the new 701.Vt mbuf chain 702on success, 703.Dv NULL 704on failure 705(the 706.Vt mbuf chain 707is freed in this case). 708.Sy Note : 709It does not allocate any 710.Vt mbuf clusters , 711so 712.Fa len 713must be less than or equal to 714.Dv MHLEN . 715.\" 716.It Fn m_pulldown mbuf offset len offsetp 717Arrange that 718.Fa len 719bytes between 720.Fa offset 721and 722.Fa offset + len 723in the 724.Vt mbuf chain 725are contiguous and lay in the data area of 726.Fa mbuf , 727so they are accessible with 728.Fn mtod mbuf type . 729.Fa len 730must be smaller than, or equal to, the size of an 731.Vt mbuf cluster . 732Return a pointer to an intermediate 733.Vt mbuf 734in the chain containing the requested region; 735the offset in the data region of the 736.Vt mbuf chain 737to the data contained in the returned mbuf is stored in 738.Fa *offsetp . 739If 740.Fa offsetp 741is NULL, the region may be accessed using 742.Fn mtod mbuf type . 743If 744.Fa offsetp 745is non-NULL, the region may be accessed using 746.Fn mtod mbuf uint8_t 747+ *offsetp. 748The region of the mbuf chain between its beginning and 749.Fa offset 750is not modified, therefore it is safe to hold pointers to data within 751this region before calling 752.Fn m_pulldown . 753.\" 754.It Fn m_copym mbuf offset len how 755Make a copy of an 756.Vt mbuf chain 757starting 758.Fa offset 759bytes from the beginning, continuing for 760.Fa len 761bytes. 762If 763.Fa len 764is 765.Dv M_COPYALL , 766copy to the end of the 767.Vt mbuf chain . 768.Sy Note : 769The copy is read-only, because the 770.Vt mbuf clusters 771are not copied, only their reference counts are incremented. 772.\" 773.It Fn m_copypacket mbuf how 774Copy an entire packet including header, which must be present. 775This is an optimized version of the common case 776.Fn m_copym mbuf 0 M_COPYALL how . 777.Sy Note : 778the copy is read-only, because the 779.Vt mbuf clusters 780are not copied, only their reference counts are incremented. 781.\" 782.It Fn m_dup mbuf how 783Copy a packet header 784.Vt mbuf chain 785into a completely new 786.Vt mbuf chain , 787including copying any 788.Vt mbuf clusters . 789Use this instead of 790.Fn m_copypacket 791when you need a writable copy of an 792.Vt mbuf chain . 793.\" 794.It Fn m_copydata mbuf offset len buf 795Copy data from an 796.Vt mbuf chain 797starting 798.Fa off 799bytes from the beginning, continuing for 800.Fa len 801bytes, into the indicated buffer 802.Fa buf . 803.\" 804.It Fn m_copyback mbuf offset len buf 805Copy 806.Fa len 807bytes from the buffer 808.Fa buf 809back into the indicated 810.Vt mbuf chain , 811starting at 812.Fa offset 813bytes from the beginning of the 814.Vt mbuf chain , 815extending the 816.Vt mbuf chain 817if necessary. 818.Sy Note : 819It does not allocate any 820.Vt mbuf clusters , 821just adds 822.Vt mbufs 823to the 824.Vt mbuf chain . 825It is safe to set 826.Fa offset 827beyond the current 828.Vt mbuf chain 829end: zeroed 830.Vt mbufs 831will be allocated to fill the space. 832.\" 833.It Fn m_length mbuf last 834Return the length of the 835.Vt mbuf chain , 836and optionally a pointer to the last 837.Vt mbuf . 838.\" 839.It Fn m_dup_pkthdr to from how 840Upon the function's completion, the 841.Vt mbuf 842.Fa to 843will contain an identical copy of 844.Fa from->m_pkthdr 845and the per-packet attributes found in the 846.Vt mbuf chain 847.Fa from . 848The 849.Vt mbuf 850.Fa from 851must have the flag 852.Dv M_PKTHDR 853initially set, and 854.Fa to 855must be empty on entry. 856.\" 857.It Fn m_move_pkthdr to from 858Move 859.Va m_pkthdr 860and the per-packet attributes from the 861.Vt mbuf chain 862.Fa from 863to the 864.Vt mbuf 865.Fa to . 866The 867.Vt mbuf 868.Fa from 869must have the flag 870.Dv M_PKTHDR 871initially set, and 872.Fa to 873must be empty on entry. 874Upon the function's completion, 875.Fa from 876will have the flag 877.Dv M_PKTHDR 878and the per-packet attributes cleared. 879.\" 880.It Fn m_fixhdr mbuf 881Set the packet-header length to the length of the 882.Vt mbuf chain . 883.\" 884.It Fn m_devget buf len offset ifp copy 885Copy data from a device local memory pointed to by 886.Fa buf 887to an 888.Vt mbuf chain . 889The copy is done using a specified copy routine 890.Fa copy , 891or 892.Fn bcopy 893if 894.Fa copy 895is 896.Dv NULL . 897.\" 898.It Fn m_cat m n 899Concatenate 900.Fa n 901to 902.Fa m . 903Both 904.Vt mbuf chains 905must be of the same type. 906.Fa n 907is not guaranteed to be valid after 908.Fn m_cat 909returns. 910.Fn m_cat 911does not update any packet header fields or free mbuf tags. 912.\" 913.It Fn m_catpkt m n 914A variant of 915.Fn m_cat 916that operates on packets. 917Both 918.Fa m 919and 920.Fa n 921must contain packet headers. 922.Fa n 923is not guaranteed to be valid after 924.Fn m_catpkt 925returns. 926.\" 927.It Fn m_split mbuf len how 928Partition an 929.Vt mbuf chain 930in two pieces, returning the tail: 931all but the first 932.Fa len 933bytes. 934In case of failure, it returns 935.Dv NULL 936and attempts to restore the 937.Vt mbuf chain 938to its original state. 939.\" 940.It Fn m_apply mbuf off len f arg 941Apply a function to an 942.Vt mbuf chain , 943at offset 944.Fa off , 945for length 946.Fa len 947bytes. 948Typically used to avoid calls to 949.Fn m_pullup 950which would otherwise be unnecessary or undesirable. 951.Fa arg 952is a convenience argument which is passed to the callback function 953.Fa f . 954.Pp 955Each time 956.Fn f 957is called, it will be passed 958.Fa arg , 959a pointer to the 960.Fa data 961in the current mbuf, and the length 962.Fa len 963of the data in this mbuf to which the function should be applied. 964.Pp 965The function should return zero to indicate success; 966otherwise, if an error is indicated, then 967.Fn m_apply 968will return the error and stop iterating through the 969.Vt mbuf chain . 970.\" 971.It Fn m_getptr mbuf loc off 972Return a pointer to the mbuf containing the data located at 973.Fa loc 974bytes from the beginning of the 975.Vt mbuf chain . 976The corresponding offset into the mbuf will be stored in 977.Fa *off . 978.It Fn m_defrag m0 how 979Defragment an mbuf chain, returning the shortest possible 980chain of mbufs and clusters. 981If allocation fails and this can not be completed, 982.Dv NULL 983will be returned and the original chain will be unchanged. 984Upon success, the original chain will be freed and the new 985chain will be returned. 986.Fa how 987should be either 988.Dv M_WAITOK 989or 990.Dv M_NOWAIT , 991depending on the caller's preference. 992.Pp 993This function is especially useful in network drivers, where 994certain long mbuf chains must be shortened before being added 995to TX descriptor lists. 996.It Fn m_collapse m0 how maxfrags 997Defragment an mbuf chain, returning a chain of at most 998.Fa maxfrags 999mbufs and clusters. 1000If allocation fails or the chain cannot be collapsed as requested, 1001.Dv NULL 1002will be returned, with the original chain possibly modified. 1003As with 1004.Fn m_defrag , 1005.Fa how 1006should be one of 1007.Dv M_WAITOK 1008or 1009.Dv M_NOWAIT . 1010.It Fn m_unshare m0 how 1011Create a version of the specified mbuf chain whose 1012contents can be safely modified without affecting other users. 1013If allocation fails and this operation can not be completed, 1014.Dv NULL 1015will be returned. 1016The original mbuf chain is always reclaimed and the reference 1017count of any shared mbuf clusters is decremented. 1018.Fa how 1019should be either 1020.Dv M_WAITOK 1021or 1022.Dv M_NOWAIT , 1023depending on the caller's preference. 1024As a side-effect of this process the returned 1025mbuf chain may be compacted. 1026.Pp 1027This function is especially useful in the transmit path of 1028network code, when data must be encrypted or otherwise 1029altered prior to transmission. 1030.El 1031.Sh HARDWARE-ASSISTED CHECKSUM CALCULATION 1032This section currently applies to TCP/IP only. 1033In order to save the host CPU resources, computing checksums is 1034offloaded to the network interface hardware if possible. 1035The 1036.Va m_pkthdr 1037member of the leading 1038.Vt mbuf 1039of a packet contains two fields used for that purpose, 1040.Vt int Va csum_flags 1041and 1042.Vt int Va csum_data . 1043The meaning of those fields depends on the direction a packet flows in, 1044and on whether the packet is fragmented. 1045Henceforth, 1046.Va csum_flags 1047or 1048.Va csum_data 1049of a packet 1050will denote the corresponding field of the 1051.Va m_pkthdr 1052member of the leading 1053.Vt mbuf 1054in the 1055.Vt mbuf chain 1056containing the packet. 1057.Pp 1058On output, checksum offloading is attempted after the outgoing 1059interface has been determined for a packet. 1060The interface-specific field 1061.Va ifnet.if_data.ifi_hwassist 1062(see 1063.Xr ifnet 9 ) 1064is consulted for the capabilities of the interface to assist in 1065computing checksums. 1066The 1067.Va csum_flags 1068field of the packet header is set to indicate which actions the interface 1069is supposed to perform on it. 1070The actions unsupported by the network interface are done in the 1071software prior to passing the packet down to the interface driver; 1072such actions will never be requested through 1073.Va csum_flags . 1074.Pp 1075The flags demanding a particular action from an interface are as follows: 1076.Bl -tag -width ".Dv CSUM_TCP" -offset indent 1077.It Dv CSUM_IP 1078The IP header checksum is to be computed and stored in the 1079corresponding field of the packet. 1080The hardware is expected to know the format of an IP header 1081to determine the offset of the IP checksum field. 1082.It Dv CSUM_TCP 1083The TCP checksum is to be computed. 1084(See below.) 1085.It Dv CSUM_UDP 1086The UDP checksum is to be computed. 1087(See below.) 1088.El 1089.Pp 1090Should a TCP or UDP checksum be offloaded to the hardware, 1091the field 1092.Va csum_data 1093will contain the byte offset of the checksum field relative to the 1094end of the IP header. 1095In this case, the checksum field will be initially 1096set by the TCP/IP module to the checksum of the pseudo header 1097defined by the TCP and UDP specifications. 1098.Pp 1099On input, an interface indicates the actions it has performed 1100on a packet by setting one or more of the following flags in 1101.Va csum_flags 1102associated with the packet: 1103.Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent 1104.It Dv CSUM_IP_CHECKED 1105The IP header checksum has been computed. 1106.It Dv CSUM_IP_VALID 1107The IP header has a valid checksum. 1108This flag can appear only in combination with 1109.Dv CSUM_IP_CHECKED . 1110.It Dv CSUM_DATA_VALID 1111The checksum of the data portion of the IP packet has been computed 1112and stored in the field 1113.Va csum_data 1114in network byte order. 1115.It Dv CSUM_PSEUDO_HDR 1116Can be set only along with 1117.Dv CSUM_DATA_VALID 1118to indicate that the IP data checksum found in 1119.Va csum_data 1120allows for the pseudo header defined by the TCP and UDP specifications. 1121Otherwise the checksum of the pseudo header must be calculated by 1122the host CPU and added to 1123.Va csum_data 1124to obtain the final checksum to be used for TCP or UDP validation purposes. 1125.El 1126.Pp 1127If a particular network interface just indicates success or 1128failure of TCP or UDP checksum validation without returning 1129the exact value of the checksum to the host CPU, its driver can mark 1130.Dv CSUM_DATA_VALID 1131and 1132.Dv CSUM_PSEUDO_HDR 1133in 1134.Va csum_flags , 1135and set 1136.Va csum_data 1137to 1138.Li 0xFFFF 1139hexadecimal to indicate a valid checksum. 1140It is a peculiarity of the algorithm used that the Internet checksum 1141calculated over any valid packet will be 1142.Li 0xFFFF 1143as long as the original checksum field is included. 1144.Sh STRESS TESTING 1145When running a kernel compiled with the option 1146.Dv MBUF_STRESS_TEST , 1147the following 1148.Xr sysctl 8 Ns 1149-controlled options may be used to create 1150various failure/extreme cases for testing of network drivers 1151and other parts of the kernel that rely on 1152.Vt mbufs . 1153.Bl -tag -width ident 1154.It Va net.inet.ip.mbuf_frag_size 1155Causes 1156.Fn ip_output 1157to fragment outgoing 1158.Vt mbuf chains 1159into fragments of the specified size. 1160Setting this variable to 1 is an excellent way to 1161test the long 1162.Vt mbuf chain 1163handling ability of network drivers. 1164.It Va kern.ipc.m_defragrandomfailures 1165Causes the function 1166.Fn m_defrag 1167to randomly fail, returning 1168.Dv NULL . 1169Any piece of code which uses 1170.Fn m_defrag 1171should be tested with this feature. 1172.El 1173.Sh RETURN VALUES 1174See above. 1175.Sh SEE ALSO 1176.Xr ifnet 9 , 1177.Xr mbuf_tags 9 1178.Sh HISTORY 1179.\" Please correct me if I'm wrong 1180.Vt Mbufs 1181appeared in an early version of 1182.Bx . 1183Besides being used for network packets, they were used 1184to store various dynamic structures, such as routing table 1185entries, interface addresses, protocol control blocks, etc. 1186In more recent 1187.Fx 1188use of 1189.Vt mbufs 1190is almost entirely limited to packet storage, with 1191.Xr uma 9 1192zones being used directly to store other network-related memory. 1193.Pp 1194Historically, the 1195.Vt mbuf 1196allocator has been a special-purpose memory allocator able to run in 1197interrupt contexts and allocating from a special kernel address space map. 1198As of 1199.Fx 5.3 , 1200the 1201.Vt mbuf 1202allocator is a wrapper around 1203.Xr uma 9 , 1204allowing caching of 1205.Vt mbufs , 1206clusters, and 1207.Vt mbuf 1208+ cluster pairs in per-CPU caches, as well as bringing other benefits of 1209slab allocation. 1210.Sh AUTHORS 1211The original 1212.Nm 1213manual page was written by 1214.An Yar Tikhiy . 1215The 1216.Xr uma 9 1217.Vt mbuf 1218allocator was written by 1219.An Bosko Milekic . 1220