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 September 27, 2017 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 "char *buf" 48.Fa "u_int size" 49.Fa "void (*free)(struct mbuf *)" 50.Fa "void *opt_arg1" 51.Fa "void *opt_arg2" 52.Fa "int 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 saved in 420.Va ext_arg1 421and 422.Va ext_arg2 423fields of the 424.Va struct m_ext 425of the mbuf. 426The 427.Fa flags 428argument specifies additional 429.Vt mbuf 430flags; it is not necessary to specify 431.Dv M_EXT . 432Finally, the 433.Fa type 434argument specifies the type of external data, which controls how it 435will be disposed of when the 436.Vt mbuf 437is freed. 438In most cases, the correct value is 439.Dv EXT_EXTREF . 440.It Fn MCLGET mbuf how 441Allocate and attach an 442.Vt mbuf cluster 443to 444.Fa mbuf . 445On success, a non-zero value returned; otherwise, 0. 446Historically, consumers would check for success by testing the 447.Dv M_EXT 448flag on the mbuf, but this is now discouraged to avoid unnecessary awareness 449of the implementation of external storage in protocol stacks and device 450drivers. 451.It Fn M_ALIGN mbuf len 452Set the pointer 453.Fa mbuf->m_data 454to place an object of the size 455.Fa len 456at the end of the internal data area of 457.Fa mbuf , 458long word aligned. 459Applicable only if 460.Fa mbuf 461is newly allocated with 462.Fn MGET 463or 464.Fn m_get . 465.It Fn MH_ALIGN mbuf len 466Serves the same purpose as 467.Fn M_ALIGN 468does, but only for 469.Fa mbuf 470newly allocated with 471.Fn MGETHDR 472or 473.Fn m_gethdr , 474or initialized by 475.Fn m_dup_pkthdr 476or 477.Fn m_move_pkthdr . 478.It Fn m_align mbuf len 479Services the same purpose as 480.Fn M_ALIGN 481but handles any type of mbuf. 482.It Fn M_LEADINGSPACE mbuf 483Returns the number of bytes available before the beginning 484of data in 485.Fa mbuf . 486.It Fn M_TRAILINGSPACE mbuf 487Returns the number of bytes available after the end of data in 488.Fa mbuf . 489.It Fn M_PREPEND mbuf len how 490This macro operates on an 491.Vt mbuf chain . 492It is an optimized wrapper for 493.Fn m_prepend 494that can make use of possible empty space before data 495(e.g.\& left after trimming of a link-layer header). 496The new 497.Vt mbuf chain 498pointer or 499.Dv NULL 500is in 501.Fa mbuf 502after the call. 503.It Fn M_MOVE_PKTHDR to from 504Using this macro is equivalent to calling 505.Fn m_move_pkthdr to from . 506.It Fn M_WRITABLE mbuf 507This macro will evaluate true if 508.Fa mbuf 509is not marked 510.Dv M_RDONLY 511and if either 512.Fa mbuf 513does not contain external storage or, 514if it does, 515then if the reference count of the storage is not greater than 1. 516The 517.Dv M_RDONLY 518flag can be set in 519.Fa mbuf->m_flags . 520This can be achieved during setup of the external storage, 521by passing the 522.Dv M_RDONLY 523bit as a 524.Fa flags 525argument to the 526.Fn MEXTADD 527macro, or can be directly set in individual 528.Vt mbufs . 529.It Fn MCHTYPE mbuf type 530Change the type of 531.Fa mbuf 532to 533.Fa type . 534This is a relatively expensive operation and should be avoided. 535.El 536.Pp 537The functions are: 538.Bl -ohang -offset indent 539.It Fn m_get how type 540A function version of 541.Fn MGET 542for non-critical paths. 543.It Fn m_get2 size how type flags 544Allocate an 545.Vt mbuf 546with enough space to hold specified amount of data. 547.It Fn m_getm orig len how type 548Allocate 549.Fa len 550bytes worth of 551.Vt mbufs 552and 553.Vt mbuf clusters 554if necessary and append the resulting allocated 555.Vt mbuf chain 556to the 557.Vt mbuf chain 558.Fa orig , 559if it is 560.No non- Ns Dv NULL . 561If the allocation fails at any point, 562free whatever was allocated and return 563.Dv NULL . 564If 565.Fa orig 566is 567.No non- Ns Dv NULL , 568it will not be freed. 569It is possible to use 570.Fn m_getm 571to either append 572.Fa len 573bytes to an existing 574.Vt mbuf 575or 576.Vt mbuf chain 577(for example, one which may be sitting in a pre-allocated ring) 578or to simply perform an all-or-nothing 579.Vt mbuf 580and 581.Vt mbuf cluster 582allocation. 583.It Fn m_gethdr how type 584A function version of 585.Fn MGETHDR 586for non-critical paths. 587.It Fn m_getcl how type flags 588Fetch an 589.Vt mbuf 590with a 591.Vt mbuf cluster 592attached to it. 593If one of the allocations fails, the entire allocation fails. 594This routine is the preferred way of fetching both the 595.Vt mbuf 596and 597.Vt mbuf cluster 598together, as it avoids having to unlock/relock between allocations. 599Returns 600.Dv NULL 601on failure. 602.It Fn m_getjcl how type flags size 603This is like 604.Fn m_getcl 605but it the size of the cluster allocated will be large enough for 606.Fa size 607bytes. 608.It Fn m_free mbuf 609Frees 610.Vt mbuf . 611Returns 612.Va m_next 613of the freed 614.Vt mbuf . 615.El 616.Pp 617The functions below operate on 618.Vt mbuf chains . 619.Bl -ohang -offset indent 620.It Fn m_freem mbuf 621Free an entire 622.Vt mbuf chain , 623including any external storage. 624.\" 625.It Fn m_adj mbuf len 626Trim 627.Fa len 628bytes from the head of an 629.Vt mbuf chain 630if 631.Fa len 632is positive, from the tail otherwise. 633.\" 634.It Fn m_append mbuf len cp 635Append 636.Vt len 637bytes of data 638.Vt cp 639to the 640.Vt mbuf chain . 641Extend the mbuf chain if the new data does not fit in 642existing space. 643.\" 644.It Fn m_prepend mbuf len how 645Allocate a new 646.Vt mbuf 647and prepend it to the 648.Vt mbuf chain , 649handle 650.Dv M_PKTHDR 651properly. 652.Sy Note : 653It does not allocate any 654.Vt mbuf clusters , 655so 656.Fa len 657must be less than 658.Dv MLEN 659or 660.Dv MHLEN , 661depending on the 662.Dv M_PKTHDR 663flag setting. 664.\" 665.It Fn m_copyup mbuf len dstoff 666Similar to 667.Fn m_pullup 668but copies 669.Fa len 670bytes of data into a new mbuf at 671.Fa dstoff 672bytes into the mbuf. 673The 674.Fa dstoff 675argument aligns the data and leaves room for a link layer header. 676Returns the new 677.Vt mbuf chain 678on success, 679and frees the 680.Vt mbuf chain 681and returns 682.Dv NULL 683on failure. 684.Sy Note : 685The function does not allocate 686.Vt mbuf clusters , 687so 688.Fa len + dstoff 689must be less than 690.Dv MHLEN . 691.\" 692.It Fn m_pullup mbuf len 693Arrange that the first 694.Fa len 695bytes of an 696.Vt mbuf chain 697are contiguous and lay in the data area of 698.Fa mbuf , 699so they are accessible with 700.Fn mtod mbuf type . 701It is important to remember that this may involve 702reallocating some mbufs and moving data so all pointers 703referencing data within the old mbuf chain 704must be recalculated or made invalid. 705Return the new 706.Vt mbuf chain 707on success, 708.Dv NULL 709on failure 710(the 711.Vt mbuf chain 712is freed in this case). 713.Sy Note : 714It does not allocate any 715.Vt mbuf clusters , 716so 717.Fa len 718must be less than or equal to 719.Dv MHLEN . 720.\" 721.It Fn m_pulldown mbuf offset len offsetp 722Arrange that 723.Fa len 724bytes between 725.Fa offset 726and 727.Fa offset + len 728in the 729.Vt mbuf chain 730are contiguous and lay in the data area of 731.Fa mbuf , 732so they are accessible with 733.Fn mtod mbuf type . 734.Fa len 735must be smaller than, or equal to, the size of an 736.Vt mbuf cluster . 737Return a pointer to an intermediate 738.Vt mbuf 739in the chain containing the requested region; 740the offset in the data region of the 741.Vt mbuf chain 742to the data contained in the returned mbuf is stored in 743.Fa *offsetp . 744If 745.Fa offsetp 746is NULL, the region may be accessed using 747.Fn mtod mbuf type . 748If 749.Fa offsetp 750is non-NULL, the region may be accessed using 751.Fn mtod mbuf uint8_t 752+ *offsetp. 753The region of the mbuf chain between its beginning and 754.Fa offset 755is not modified, therefore it is safe to hold pointers to data within 756this region before calling 757.Fn m_pulldown . 758.\" 759.It Fn m_copym mbuf offset len how 760Make a copy of an 761.Vt mbuf chain 762starting 763.Fa offset 764bytes from the beginning, continuing for 765.Fa len 766bytes. 767If 768.Fa len 769is 770.Dv M_COPYALL , 771copy to the end of the 772.Vt mbuf chain . 773.Sy Note : 774The copy is read-only, because the 775.Vt mbuf clusters 776are not copied, only their reference counts are incremented. 777.\" 778.It Fn m_copypacket mbuf how 779Copy an entire packet including header, which must be present. 780This is an optimized version of the common case 781.Fn m_copym mbuf 0 M_COPYALL how . 782.Sy Note : 783the copy is read-only, because the 784.Vt mbuf clusters 785are not copied, only their reference counts are incremented. 786.\" 787.It Fn m_dup mbuf how 788Copy a packet header 789.Vt mbuf chain 790into a completely new 791.Vt mbuf chain , 792including copying any 793.Vt mbuf clusters . 794Use this instead of 795.Fn m_copypacket 796when you need a writable copy of an 797.Vt mbuf chain . 798.\" 799.It Fn m_copydata mbuf offset len buf 800Copy data from an 801.Vt mbuf chain 802starting 803.Fa off 804bytes from the beginning, continuing for 805.Fa len 806bytes, into the indicated buffer 807.Fa buf . 808.\" 809.It Fn m_copyback mbuf offset len buf 810Copy 811.Fa len 812bytes from the buffer 813.Fa buf 814back into the indicated 815.Vt mbuf chain , 816starting at 817.Fa offset 818bytes from the beginning of the 819.Vt mbuf chain , 820extending the 821.Vt mbuf chain 822if necessary. 823.Sy Note : 824It does not allocate any 825.Vt mbuf clusters , 826just adds 827.Vt mbufs 828to the 829.Vt mbuf chain . 830It is safe to set 831.Fa offset 832beyond the current 833.Vt mbuf chain 834end: zeroed 835.Vt mbufs 836will be allocated to fill the space. 837.\" 838.It Fn m_length mbuf last 839Return the length of the 840.Vt mbuf chain , 841and optionally a pointer to the last 842.Vt mbuf . 843.\" 844.It Fn m_dup_pkthdr to from how 845Upon the function's completion, the 846.Vt mbuf 847.Fa to 848will contain an identical copy of 849.Fa from->m_pkthdr 850and the per-packet attributes found in the 851.Vt mbuf chain 852.Fa from . 853The 854.Vt mbuf 855.Fa from 856must have the flag 857.Dv M_PKTHDR 858initially set, and 859.Fa to 860must be empty on entry. 861.\" 862.It Fn m_move_pkthdr to from 863Move 864.Va m_pkthdr 865and the per-packet attributes from the 866.Vt mbuf chain 867.Fa from 868to the 869.Vt mbuf 870.Fa to . 871The 872.Vt mbuf 873.Fa from 874must have the flag 875.Dv M_PKTHDR 876initially set, and 877.Fa to 878must be empty on entry. 879Upon the function's completion, 880.Fa from 881will have the flag 882.Dv M_PKTHDR 883and the per-packet attributes cleared. 884.\" 885.It Fn m_fixhdr mbuf 886Set the packet-header length to the length of the 887.Vt mbuf chain . 888.\" 889.It Fn m_devget buf len offset ifp copy 890Copy data from a device local memory pointed to by 891.Fa buf 892to an 893.Vt mbuf chain . 894The copy is done using a specified copy routine 895.Fa copy , 896or 897.Fn bcopy 898if 899.Fa copy 900is 901.Dv NULL . 902.\" 903.It Fn m_cat m n 904Concatenate 905.Fa n 906to 907.Fa m . 908Both 909.Vt mbuf chains 910must be of the same type. 911.Fa n 912is not guaranteed to be valid after 913.Fn m_cat 914returns. 915.Fn m_cat 916does not update any packet header fields or free mbuf tags. 917.\" 918.It Fn m_catpkt m n 919A variant of 920.Fn m_cat 921that operates on packets. 922Both 923.Fa m 924and 925.Fa n 926must contain packet headers. 927.Fa n 928is not guaranteed to be valid after 929.Fn m_catpkt 930returns. 931.\" 932.It Fn m_split mbuf len how 933Partition an 934.Vt mbuf chain 935in two pieces, returning the tail: 936all but the first 937.Fa len 938bytes. 939In case of failure, it returns 940.Dv NULL 941and attempts to restore the 942.Vt mbuf chain 943to its original state. 944.\" 945.It Fn m_apply mbuf off len f arg 946Apply a function to an 947.Vt mbuf chain , 948at offset 949.Fa off , 950for length 951.Fa len 952bytes. 953Typically used to avoid calls to 954.Fn m_pullup 955which would otherwise be unnecessary or undesirable. 956.Fa arg 957is a convenience argument which is passed to the callback function 958.Fa f . 959.Pp 960Each time 961.Fn f 962is called, it will be passed 963.Fa arg , 964a pointer to the 965.Fa data 966in the current mbuf, and the length 967.Fa len 968of the data in this mbuf to which the function should be applied. 969.Pp 970The function should return zero to indicate success; 971otherwise, if an error is indicated, then 972.Fn m_apply 973will return the error and stop iterating through the 974.Vt mbuf chain . 975.\" 976.It Fn m_getptr mbuf loc off 977Return a pointer to the mbuf containing the data located at 978.Fa loc 979bytes from the beginning of the 980.Vt mbuf chain . 981The corresponding offset into the mbuf will be stored in 982.Fa *off . 983.It Fn m_defrag m0 how 984Defragment an mbuf chain, returning the shortest possible 985chain of mbufs and clusters. 986If allocation fails and this can not be completed, 987.Dv NULL 988will be returned and the original chain will be unchanged. 989Upon success, the original chain will be freed and the new 990chain will be returned. 991.Fa how 992should be either 993.Dv M_WAITOK 994or 995.Dv M_NOWAIT , 996depending on the caller's preference. 997.Pp 998This function is especially useful in network drivers, where 999certain long mbuf chains must be shortened before being added 1000to TX descriptor lists. 1001.It Fn m_collapse m0 how maxfrags 1002Defragment an mbuf chain, returning a chain of at most 1003.Fa maxfrags 1004mbufs and clusters. 1005If allocation fails or the chain cannot be collapsed as requested, 1006.Dv NULL 1007will be returned, with the original chain possibly modified. 1008As with 1009.Fn m_defrag , 1010.Fa how 1011should be one of 1012.Dv M_WAITOK 1013or 1014.Dv M_NOWAIT . 1015.It Fn m_unshare m0 how 1016Create a version of the specified mbuf chain whose 1017contents can be safely modified without affecting other users. 1018If allocation fails and this operation can not be completed, 1019.Dv NULL 1020will be returned. 1021The original mbuf chain is always reclaimed and the reference 1022count of any shared mbuf clusters is decremented. 1023.Fa how 1024should be either 1025.Dv M_WAITOK 1026or 1027.Dv M_NOWAIT , 1028depending on the caller's preference. 1029As a side-effect of this process the returned 1030mbuf chain may be compacted. 1031.Pp 1032This function is especially useful in the transmit path of 1033network code, when data must be encrypted or otherwise 1034altered prior to transmission. 1035.El 1036.Sh HARDWARE-ASSISTED CHECKSUM CALCULATION 1037This section currently applies to TCP/IP only. 1038In order to save the host CPU resources, computing checksums is 1039offloaded to the network interface hardware if possible. 1040The 1041.Va m_pkthdr 1042member of the leading 1043.Vt mbuf 1044of a packet contains two fields used for that purpose, 1045.Vt int Va csum_flags 1046and 1047.Vt int Va csum_data . 1048The meaning of those fields depends on the direction a packet flows in, 1049and on whether the packet is fragmented. 1050Henceforth, 1051.Va csum_flags 1052or 1053.Va csum_data 1054of a packet 1055will denote the corresponding field of the 1056.Va m_pkthdr 1057member of the leading 1058.Vt mbuf 1059in the 1060.Vt mbuf chain 1061containing the packet. 1062.Pp 1063On output, checksum offloading is attempted after the outgoing 1064interface has been determined for a packet. 1065The interface-specific field 1066.Va ifnet.if_data.ifi_hwassist 1067(see 1068.Xr ifnet 9 ) 1069is consulted for the capabilities of the interface to assist in 1070computing checksums. 1071The 1072.Va csum_flags 1073field of the packet header is set to indicate which actions the interface 1074is supposed to perform on it. 1075The actions unsupported by the network interface are done in the 1076software prior to passing the packet down to the interface driver; 1077such actions will never be requested through 1078.Va csum_flags . 1079.Pp 1080The flags demanding a particular action from an interface are as follows: 1081.Bl -tag -width ".Dv CSUM_TCP" -offset indent 1082.It Dv CSUM_IP 1083The IP header checksum is to be computed and stored in the 1084corresponding field of the packet. 1085The hardware is expected to know the format of an IP header 1086to determine the offset of the IP checksum field. 1087.It Dv CSUM_TCP 1088The TCP checksum is to be computed. 1089(See below.) 1090.It Dv CSUM_UDP 1091The UDP checksum is to be computed. 1092(See below.) 1093.El 1094.Pp 1095Should a TCP or UDP checksum be offloaded to the hardware, 1096the field 1097.Va csum_data 1098will contain the byte offset of the checksum field relative to the 1099end of the IP header. 1100In this case, the checksum field will be initially 1101set by the TCP/IP module to the checksum of the pseudo header 1102defined by the TCP and UDP specifications. 1103.Pp 1104On input, an interface indicates the actions it has performed 1105on a packet by setting one or more of the following flags in 1106.Va csum_flags 1107associated with the packet: 1108.Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent 1109.It Dv CSUM_IP_CHECKED 1110The IP header checksum has been computed. 1111.It Dv CSUM_IP_VALID 1112The IP header has a valid checksum. 1113This flag can appear only in combination with 1114.Dv CSUM_IP_CHECKED . 1115.It Dv CSUM_DATA_VALID 1116The checksum of the data portion of the IP packet has been computed 1117and stored in the field 1118.Va csum_data 1119in network byte order. 1120.It Dv CSUM_PSEUDO_HDR 1121Can be set only along with 1122.Dv CSUM_DATA_VALID 1123to indicate that the IP data checksum found in 1124.Va csum_data 1125allows for the pseudo header defined by the TCP and UDP specifications. 1126Otherwise the checksum of the pseudo header must be calculated by 1127the host CPU and added to 1128.Va csum_data 1129to obtain the final checksum to be used for TCP or UDP validation purposes. 1130.El 1131.Pp 1132If a particular network interface just indicates success or 1133failure of TCP or UDP checksum validation without returning 1134the exact value of the checksum to the host CPU, its driver can mark 1135.Dv CSUM_DATA_VALID 1136and 1137.Dv CSUM_PSEUDO_HDR 1138in 1139.Va csum_flags , 1140and set 1141.Va csum_data 1142to 1143.Li 0xFFFF 1144hexadecimal to indicate a valid checksum. 1145It is a peculiarity of the algorithm used that the Internet checksum 1146calculated over any valid packet will be 1147.Li 0xFFFF 1148as long as the original checksum field is included. 1149.Sh STRESS TESTING 1150When running a kernel compiled with the option 1151.Dv MBUF_STRESS_TEST , 1152the following 1153.Xr sysctl 8 Ns 1154-controlled options may be used to create 1155various failure/extreme cases for testing of network drivers 1156and other parts of the kernel that rely on 1157.Vt mbufs . 1158.Bl -tag -width ident 1159.It Va net.inet.ip.mbuf_frag_size 1160Causes 1161.Fn ip_output 1162to fragment outgoing 1163.Vt mbuf chains 1164into fragments of the specified size. 1165Setting this variable to 1 is an excellent way to 1166test the long 1167.Vt mbuf chain 1168handling ability of network drivers. 1169.It Va kern.ipc.m_defragrandomfailures 1170Causes the function 1171.Fn m_defrag 1172to randomly fail, returning 1173.Dv NULL . 1174Any piece of code which uses 1175.Fn m_defrag 1176should be tested with this feature. 1177.El 1178.Sh RETURN VALUES 1179See above. 1180.Sh SEE ALSO 1181.Xr ifnet 9 , 1182.Xr mbuf_tags 9 1183.Sh HISTORY 1184.\" Please correct me if I'm wrong 1185.Vt Mbufs 1186appeared in an early version of 1187.Bx . 1188Besides being used for network packets, they were used 1189to store various dynamic structures, such as routing table 1190entries, interface addresses, protocol control blocks, etc. 1191In more recent 1192.Fx 1193use of 1194.Vt mbufs 1195is almost entirely limited to packet storage, with 1196.Xr uma 9 1197zones being used directly to store other network-related memory. 1198.Pp 1199Historically, the 1200.Vt mbuf 1201allocator has been a special-purpose memory allocator able to run in 1202interrupt contexts and allocating from a special kernel address space map. 1203As of 1204.Fx 5.3 , 1205the 1206.Vt mbuf 1207allocator is a wrapper around 1208.Xr uma 9 , 1209allowing caching of 1210.Vt mbufs , 1211clusters, and 1212.Vt mbuf 1213+ cluster pairs in per-CPU caches, as well as bringing other benefits of 1214slab allocation. 1215.Sh AUTHORS 1216The original 1217.Nm 1218manual page was written by 1219.An Yar Tikhiy . 1220The 1221.Xr uma 9 1222.Vt mbuf 1223allocator was written by 1224.An Bosko Milekic . 1225