| blob | 8454b64757390b1bdf885c27e9e6affc2a37d77a |
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 March 25, 2008
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
142 An
143 .Vt mbuf
144 is a basic unit of memory management in the kernel IPC subsystem.
145 Network packets and socket buffers are stored in
146 .Vt mbufs .
147 A network packet may span multiple
148 .Vt mbufs
149 arranged into a
150 .Vt mbuf chain
151 (linked list),
152 which allows adding or trimming
153 network headers with little overhead.
154 .Pp
155 While a developer should not bother with
156 .Vt mbuf
157 internals without serious
158 reason in order to avoid incompatibilities with future changes, it
159 is useful to understand the general structure of an
160 .Vt mbuf .
161 .Pp
162 An
163 .Vt mbuf
164 consists of a variable-sized header and a small internal
165 buffer for data.
166 The total size of an
167 .Vt mbuf ,
168 .Dv MSIZE ,
169 is a constant defined in
170 .In sys/param.h .
171 The
172 .Vt mbuf
173 header includes:
174 .Pp
175 .Bl -tag -width "m_nextpkt" -offset indent
176 .It Va m_next
177 .Pq Vt struct mbuf *
178 A pointer to the next
179 .Vt mbuf
180 in the
181 .Vt mbuf chain .
182 .It Va m_nextpkt
183 .Pq Vt struct mbuf *
184 A pointer to the next
185 .Vt mbuf chain
186 in the queue.
187 .It Va m_data
188 .Pq Vt caddr_t
189 A pointer to data attached to this
190 .Vt mbuf .
191 .It Va m_len
192 .Pq Vt int
193 The length of the data.
194 .It Va m_type
195 .Pq Vt short
196 The type of the data.
197 .It Va m_flags
198 .Pq Vt int
199 The
200 .Vt mbuf
201 flags.
202 .El
203 .Pp
204 The
205 .Vt mbuf
206 flag bits are defined as follows:
207 .Bd -literal
208 /* mbuf flags */
209 #define M_EXT 0x0001 /* has associated external storage */
210 #define M_PKTHDR 0x0002 /* start of record */
211 #define M_EOR 0x0004 /* end of record */
212 #define M_RDONLY 0x0008 /* associated data marked read-only */
213 #define M_PROTO1 0x0010 /* protocol-specific */
214 #define M_PROTO2 0x0020 /* protocol-specific */
215 #define M_PROTO3 0x0040 /* protocol-specific */
216 #define M_PROTO4 0x0080 /* protocol-specific */
217 #define M_PROTO5 0x0100 /* protocol-specific */
218 #define M_PROTO6 0x4000 /* protocol-specific (avoid M_BCAST conflict) */
219 #define M_FREELIST 0x8000 /* mbuf is on the free list */
221 /* mbuf pkthdr flags (also stored in m_flags) */
222 #define M_BCAST 0x0200 /* send/received as link-level broadcast */
223 #define M_MCAST 0x0400 /* send/received as link-level multicast */
224 #define M_FRAG 0x0800 /* packet is fragment of larger packet */
225 #define M_FIRSTFRAG 0x1000 /* packet is first fragment */
226 #define M_LASTFRAG 0x2000 /* packet is last fragment */
227 .Ed
228 .Pp
229 The available
230 .Vt mbuf
231 types are defined as follows:
232 .Bd -literal
233 /* mbuf types */
234 #define MT_DATA 1 /* dynamic (data) allocation */
235 #define MT_HEADER MT_DATA /* packet header */
236 #define MT_SONAME 8 /* socket name */
237 #define MT_CONTROL 14 /* extra-data protocol message */
238 #define MT_OOBDATA 15 /* expedited data */
239 .Ed
240 .Pp
241 If the
242 .Dv M_PKTHDR
243 flag is set, a
244 .Vt struct pkthdr Va m_pkthdr
245 is added to the
246 .Vt mbuf
247 header.
248 It contains a pointer to the interface
249 the packet has been received from
250 .Pq Vt struct ifnet Va *rcvif ,
251 and the total packet length
252 .Pq Vt int Va len .
253 Optionally, it may also contain an attached list of packet tags
254 .Pq Vt "struct m_tag" .
255 See
256 .Xr mbuf_tags 9
257 for details.
258 Fields used in offloading checksum calculation to the hardware are kept in
259 .Va m_pkthdr
260 as well.
261 See
262 .Sx HARDWARE-ASSISTED CHECKSUM CALCULATION
263 for details.
264 .Pp
265 If small enough, data is stored in the internal data buffer of an
266 .Vt mbuf .
267 If the data is sufficiently large, another
268 .Vt mbuf
269 may be added to the
270 .Vt mbuf chain ,
271 or external storage may be associated with the
272 .Vt mbuf .
273 .Dv MHLEN
274 bytes of data can fit into an
275 .Vt mbuf
276 with the
277 .Dv M_PKTHDR
278 flag set,
279 .Dv MLEN
280 bytes can otherwise.
281 .Pp
282 If external storage is being associated with an
283 .Vt mbuf ,
284 the
285 .Va m_ext
286 header is added at the cost of losing the internal data buffer.
287 It includes a pointer to external storage, the size of the storage,
288 a pointer to a function used for freeing the storage,
289 a pointer to an optional argument that can be passed to the function,
290 and a pointer to a reference counter.
291 An
292 .Vt mbuf
293 using external storage has the
294 .Dv M_EXT
295 flag set.
296 .Pp
297 The system supplies a macro for allocating the desired external storage
298 buffer,
299 .Dv MEXTADD .
300 .Pp
301 The allocation and management of the reference counter is handled by the
302 subsystem.
303 .Pp
304 The system also supplies a default type of external storage buffer called an
305 .Vt mbuf cluster .
306 .Vt Mbuf clusters
307 can be allocated and configured with the use of the
308 .Dv MCLGET
309 macro.
310 Each
311 .Vt mbuf cluster
312 is
313 .Dv MCLBYTES
314 in size, where MCLBYTES is a machine-dependent constant.
315 The system defines an advisory macro
316 .Dv MINCLSIZE ,
317 which is the smallest amount of data to put into an
318 .Vt mbuf cluster .
319 It is equal to the sum of
320 .Dv MLEN
321 and
322 .Dv MHLEN .
323 It is typically preferable to store data into the data region of an
324 .Vt mbuf ,
325 if size permits, as opposed to allocating a separate
326 .Vt mbuf cluster
327 to hold the same data.
328 .\"
329 .Ss Macros and Functions
330 There are numerous predefined macros and functions that provide the
331 developer with common utilities.
332 .\"
333 .Bl -ohang -offset indent
334 .It Fn mtod mbuf type
335 Convert an
336 .Fa mbuf
337 pointer to a data pointer.
338 The macro expands to the data pointer cast to the pointer of the specified
339 .Fa type .
340 .Sy Note :
341 It is advisable to ensure that there is enough contiguous data in
342 .Fa mbuf .
343 See
344 .Fn m_pullup
345 for details.
346 .It Fn MGET mbuf how type
347 Allocate an
348 .Vt mbuf
349 and initialize it to contain internal data.
350 .Fa mbuf
351 will point to the allocated
352 .Vt mbuf
353 on success, or be set to
354 .Dv NULL
355 on failure.
356 The
357 .Fa how
358 argument is to be set to
359 .Dv M_WAIT
360 or
361 .Dv M_DONTWAIT .
362 It specifies whether the caller is willing to block if necessary.
363 A number of other functions and macros related to
364 .Vt mbufs
365 have the same argument because they may
366 at some point need to allocate new
367 .Vt mbufs .
368 .Pp
369 Programmers should be careful not to confuse the
370 .Vt mbuf
371 allocation flag
372 .Dv M_DONTWAIT
373 with the
374 .Xr malloc 9
375 allocation flag,
376 .Dv M_NOWAIT .
377 They are not the same.
378 .It Fn MGETHDR mbuf how type
379 Allocate an
380 .Vt mbuf
381 and initialize it to contain a packet header
382 and internal data.
383 See
384 .Fn MGET
385 for details.
386 .It Fn MCLGET mbuf how
387 Allocate and attach an
388 .Vt mbuf cluster
389 to
390 .Fa mbuf .
391 If the macro fails, the
392 .Dv M_EXT
393 flag will not be set in
394 .Fa mbuf .
395 .It Fn M_ALIGN mbuf len
396 Set the pointer
397 .Fa mbuf->m_data
398 to place an object of the size
399 .Fa len
400 at the end of the internal data area of
401 .Fa mbuf ,
402 long word aligned.
403 Applicable only if
404 .Fa mbuf
405 is newly allocated with
406 .Fn MGET
407 or
408 .Fn m_get .
409 .It Fn MH_ALIGN mbuf len
410 Serves the same purpose as
411 .Fn M_ALIGN
412 does, but only for
413 .Fa mbuf
414 newly allocated with
415 .Fn MGETHDR
416 or
417 .Fn m_gethdr ,
418 or initialized by
419 .Fn m_dup_pkthdr
420 or
421 .Fn m_move_pkthdr .
422 .It Fn m_align mbuf len
423 Services the same purpose as
424 .Fn M_ALIGN
425 but handles any type of mbuf.
426 .It Fn M_LEADINGSPACE mbuf
427 Returns the number of bytes available before the beginning
428 of data in
429 .Fa mbuf .
430 .It Fn M_TRAILINGSPACE mbuf
431 Returns the number of bytes available after the end of data in
432 .Fa mbuf .
433 .It Fn M_PREPEND mbuf len how
434 This macro operates on an
435 .Vt mbuf chain .
436 It is an optimized wrapper for
437 .Fn m_prepend
438 that can make use of possible empty space before data
439 (e.g.\& left after trimming of a link-layer header).
440 The new
441 .Vt mbuf chain
442 pointer or
443 .Dv NULL
444 is in
445 .Fa mbuf
446 after the call.
447 .It Fn M_MOVE_PKTHDR to from
448 Using this macro is equivalent to calling
449 .Fn m_move_pkthdr to from .
450 .It Fn M_WRITABLE mbuf
451 This macro will evaluate true if
452 .Fa mbuf
453 is not marked
454 .Dv M_RDONLY
455 and if either
456 .Fa mbuf
457 does not contain external storage or,
458 if it does,
459 then if the reference count of the storage is not greater than 1.
460 The
461 .Dv M_RDONLY
462 flag can be set in
463 .Fa mbuf->m_flags .
464 This can be achieved during setup of the external storage,
465 by passing the
466 .Dv M_RDONLY
467 bit as a
468 .Fa flags
469 argument to the
470 .Fn MEXTADD
471 macro, or can be directly set in individual
472 .Vt mbufs .
473 .It Fn MCHTYPE mbuf type
474 Change the type of
475 .Fa mbuf
476 to
477 .Fa type .
478 This is a relatively expensive operation and should be avoided.
479 .El
480 .Pp
481 The functions are:
482 .Bl -ohang -offset indent
483 .It Fn m_get how type
484 A function version of
485 .Fn MGET
486 for non-critical paths.
487 .It Fn m_getm orig len how type
488 Allocate
489 .Fa len
490 bytes worth of
491 .Vt mbufs
492 and
493 .Vt mbuf clusters
494 if necessary and append the resulting allocated
495 .Vt mbuf chain
496 to the
497 .Vt mbuf chain
498 .Fa orig ,
499 if it is
500 .No non- Ns Dv NULL .
501 If the allocation fails at any point,
502 free whatever was allocated and return
503 .Dv NULL .
504 If
505 .Fa orig
506 is
507 .No non- Ns Dv NULL ,
508 it will not be freed.
509 It is possible to use
510 .Fn m_getm
511 to either append
512 .Fa len
513 bytes to an existing
514 .Vt mbuf
515 or
516 .Vt mbuf chain
517 (for example, one which may be sitting in a pre-allocated ring)
518 or to simply perform an all-or-nothing
519 .Vt mbuf
520 and
521 .Vt mbuf cluster
522 allocation.
523 .It Fn m_gethdr how type
524 A function version of
525 .Fn MGETHDR
526 for non-critical paths.
527 .It Fn m_getcl how type flags
528 Fetch an
529 .Vt mbuf
530 with a
531 .Vt mbuf cluster
532 attached to it.
533 If one of the allocations fails, the entire allocation fails.
534 This routine is the preferred way of fetching both the
535 .Vt mbuf
536 and
537 .Vt mbuf cluster
538 together, as it avoids having to unlock/relock between allocations.
539 Returns
540 .Dv NULL
541 on failure.
542 .It Fn m_getclr how type
543 Allocate an
544 .Vt mbuf
545 and zero out the data region.
546 .It Fn m_free mbuf
547 Frees
548 .Vt mbuf .
549 Returns
550 .Va m_next
551 of the freed
552 .Vt mbuf .
553 .El
554 .Pp
555 The functions below operate on
556 .Vt mbuf chains .
557 .Bl -ohang -offset indent
558 .It Fn m_freem mbuf
559 Free an entire
560 .Vt mbuf chain ,
561 including any external storage.
562 .\"
563 .It Fn m_adj mbuf len
564 Trim
565 .Fa len
566 bytes from the head of an
567 .Vt mbuf chain
568 if
569 .Fa len
570 is positive, from the tail otherwise.
571 .\"
572 .It Fn m_append mbuf len cp
573 Append
574 .Vt len
575 bytes of data
576 .Vt cp
577 to the
578 .Vt mbuf chain .
579 Extend the mbuf chain if the new data does not fit in
580 existing space.
581 .\"
582 .It Fn m_prepend mbuf len how
583 Allocate a new
584 .Vt mbuf
585 and prepend it to the
586 .Vt mbuf chain ,
587 handle
588 .Dv M_PKTHDR
589 properly.
590 .Sy Note :
591 It does not allocate any
592 .Vt mbuf clusters ,
593 so
594 .Fa len
595 must be less than
596 .Dv MLEN
597 or
598 .Dv MHLEN ,
599 depending on the
600 .Dv M_PKTHDR
601 flag setting.
602 .\"
603 .It Fn m_copyup mbuf len dstoff
604 Similar to
605 .Fn m_pullup
606 but copies
607 .Fa len
608 bytes of data into a new mbuf at
609 .Fa dstoff
610 bytes into the mbuf.
611 The
612 .Fa dstoff
613 argument aligns the data and leaves room for a link layer header.
614 Returns the new
615 .Vt mbuf chain
616 on success,
617 and frees the
618 .Vt mbuf chain
619 and returns
620 .Dv NULL
621 on failure.
622 .Sy Note :
623 The function does not allocate
624 .Vt mbuf clusters ,
625 so
626 .Fa len + dstoff
627 must be less than
628 .Dv MHLEN .
629 .\"
630 .It Fn m_pullup mbuf len
631 Arrange that the first
632 .Fa len
633 bytes of an
634 .Vt mbuf chain
635 are contiguous and lay in the data area of
636 .Fa mbuf ,
637 so they are accessible with
638 .Fn mtod mbuf type .
639 It is important to remember that this may involve
640 reallocating some mbufs and moving data so all pointers
641 referencing data within the old mbuf chain
642 must be recalculated or made invalid.
643 Return the new
644 .Vt mbuf chain
645 on success,
646 .Dv NULL
647 on failure
648 (the
649 .Vt mbuf chain
650 is freed in this case).
651 .Sy Note :
652 It does not allocate any
653 .Vt mbuf clusters ,
654 so
655 .Fa len
656 must be less than
657 .Dv MHLEN .
658 .\"
659 .It Fn m_pulldown mbuf offset len offsetp
660 Arrange that
661 .Fa len
662 bytes between
663 .Fa offset
664 and
665 .Fa offset + len
666 in the
667 .Vt mbuf chain
668 are contiguous and lay in the data area of
669 .Fa mbuf ,
670 so they are accessible with
671 .Fn mtod mbuf type .
672 .Fa len must be smaller than, or equal to, the size of an
673 .Vt mbuf cluster .
674 Return a pointer to an intermediate
675 .Vt mbuf
676 in the chain containing the requested region;
677 the offset in the data region of the
678 .Vt mbuf chain
679 to the data contained in the returned mbuf is stored in
680 .Fa *offsetp .
681 If
682 .Fa offp
683 is NULL, the region may be accessed using
684 .Fn mtod mbuf type .
685 If
686 .Fa offp
687 is non-NULL, the region may be accessed using
688 .Fn mtod mbuf uint8_t + *offsetp .
689 The region of the mbuf chain between its beginning and
690 .Fa off
691 is not modified, therefore it is safe to hold pointers to data within
692 this region before calling
693 .Fn m_pulldown .
694 .\"
695 .It Fn m_copym mbuf offset len how
696 Make a copy of an
697 .Vt mbuf chain
698 starting
699 .Fa offset
700 bytes from the beginning, continuing for
701 .Fa len
702 bytes.
703 If
704 .Fa len
705 is
706 .Dv M_COPYALL ,
707 copy to the end of the
708 .Vt mbuf chain .
709 .Sy Note :
710 The copy is read-only, because the
711 .Vt mbuf clusters
712 are not copied, only their reference counts are incremented.
713 .\"
714 .It Fn m_copypacket mbuf how
715 Copy an entire packet including header, which must be present.
716 This is an optimized version of the common case
717 .Fn m_copym mbuf 0 M_COPYALL how .
718 .Sy Note :
719 the copy is read-only, because the
720 .Vt mbuf clusters
721 are not copied, only their reference counts are incremented.
722 .\"
723 .It Fn m_dup mbuf how
724 Copy a packet header
725 .Vt mbuf chain
726 into a completely new
727 .Vt mbuf chain ,
728 including copying any
729 .Vt mbuf clusters .
730 Use this instead of
731 .Fn m_copypacket
732 when you need a writable copy of an
733 .Vt mbuf chain .
734 .\"
735 .It Fn m_copydata mbuf offset len buf
736 Copy data from an
737 .Vt mbuf chain
738 starting
739 .Fa off
740 bytes from the beginning, continuing for
741 .Fa len
742 bytes, into the indicated buffer
743 .Fa buf .
744 .\"
745 .It Fn m_copyback mbuf offset len buf
746 Copy
747 .Fa len
748 bytes from the buffer
749 .Fa buf
750 back into the indicated
751 .Vt mbuf chain ,
752 starting at
753 .Fa offset
754 bytes from the beginning of the
755 .Vt mbuf chain ,
756 extending the
757 .Vt mbuf chain
758 if necessary.
759 .Sy Note :
760 It does not allocate any
761 .Vt mbuf clusters ,
762 just adds
763 .Vt mbufs
764 to the
765 .Vt mbuf chain .
766 It is safe to set
767 .Fa offset
768 beyond the current
769 .Vt mbuf chain
770 end: zeroed
771 .Vt mbufs
772 will be allocated to fill the space.
773 .\"
774 .It Fn m_length mbuf last
775 Return the length of the
776 .Vt mbuf chain ,
777 and optionally a pointer to the last
778 .Vt mbuf .
779 .\"
780 .It Fn m_dup_pkthdr to from how
781 Upon the function's completion, the
782 .Vt mbuf
783 .Fa to
784 will contain an identical copy of
785 .Fa from->m_pkthdr
786 and the per-packet attributes found in the
787 .Vt mbuf chain
788 .Fa from .
789 The
790 .Vt mbuf
791 .Fa from
792 must have the flag
793 .Dv M_PKTHDR
794 initially set, and
795 .Fa to
796 must be empty on entry.
797 .\"
798 .It Fn m_move_pkthdr to from
799 Move
800 .Va m_pkthdr
801 and the per-packet attributes from the
802 .Vt mbuf chain
803 .Fa from
804 to the
805 .Vt mbuf
806 .Fa to .
807 The
808 .Vt mbuf
809 .Fa from
810 must have the flag
811 .Dv M_PKTHDR
812 initially set, and
813 .Fa to
814 must be empty on entry.
815 Upon the function's completion,
816 .Fa from
817 will have the flag
818 .Dv M_PKTHDR
819 and the per-packet attributes cleared.
820 .\"
821 .It Fn m_fixhdr mbuf
822 Set the packet-header length to the length of the
823 .Vt mbuf chain .
824 .\"
825 .It Fn m_devget buf len offset ifp copy
826 Copy data from a device local memory pointed to by
827 .Fa buf
828 to an
829 .Vt mbuf chain .
830 The copy is done using a specified copy routine
831 .Fa copy ,
832 or
833 .Fn bcopy
834 if
835 .Fa copy
836 is
837 .Dv NULL .
838 .\"
839 .It Fn m_cat m n
840 Concatenate
841 .Fa n
842 to
843 .Fa m .
844 Both
845 .Vt mbuf chains
846 must be of the same type.
847 .Fa N
848 is still valid after the function returned.
849 .Sy Note :
850 It does not handle
851 .Dv M_PKTHDR
852 and friends.
853 .\"
854 .It Fn m_split mbuf len how
855 Partition an
856 .Vt mbuf chain
857 in two pieces, returning the tail:
858 all but the first
859 .Fa len
860 bytes.
861 In case of failure, it returns
862 .Dv NULL
863 and attempts to restore the
864 .Vt mbuf chain
865 to its original state.
866 .\"
867 .It Fn m_apply mbuf off len f arg
868 Apply a function to an
869 .Vt mbuf chain ,
870 at offset
871 .Fa off ,
872 for length
873 .Fa len
874 bytes.
875 Typically used to avoid calls to
876 .Fn m_pullup
877 which would otherwise be unnecessary or undesirable.
878 .Fa arg
879 is a convenience argument which is passed to the callback function
880 .Fa f .
881 .Pp
882 Each time
883 .Fn f
884 is called, it will be passed
885 .Fa arg ,
886 a pointer to the
887 .Fa data
888 in the current mbuf, and the length
889 .Fa len
890 of the data in this mbuf to which the function should be applied.
891 .Pp
892 The function should return zero to indicate success;
893 otherwise, if an error is indicated, then
894 .Fn m_apply
895 will return the error and stop iterating through the
896 .Vt mbuf chain .
897 .\"
898 .It Fn m_getptr mbuf loc off
899 Return a pointer to the mbuf containing the data located at
900 .Fa loc
901 bytes from the beginning of the
902 .Vt mbuf chain .
903 The corresponding offset into the mbuf will be stored in
904 .Fa *off .
905 .It Fn m_defrag m0 how
906 Defragment an mbuf chain, returning the shortest possible
907 chain of mbufs and clusters.
908 If allocation fails and this can not be completed,
909 .Dv NULL
910 will be returned and the original chain will be unchanged.
911 Upon success, the original chain will be freed and the new
912 chain will be returned.
913 .Fa how
914 should be either
915 .Dv M_WAIT
916 or
917 .Dv M_DONTWAIT ,
918 depending on the caller's preference.
919 .Pp
920 This function is especially useful in network drivers, where
921 certain long mbuf chains must be shortened before being added
922 to TX descriptor lists.
923 .It Fn m_unshare m0 how
924 Create a version of the specified mbuf chain whose
925 contents can be safely modified without affecting other users.
926 If allocation fails and this operation can not be completed,
927 .Dv NULL
928 will be returned.
929 The original mbuf chain is always reclaimed and the reference
930 count of any shared mbuf clusters is decremented.
931 .Fa how
932 should be either
933 .Dv M_WAIT
934 or
935 .Dv M_DONTWAIT ,
936 depending on the caller's preference.
937 As a side-effect of this process the returned
938 mbuf chain may be compacted.
939 .Pp
940 This function is especially useful in the transmit path of
941 network code, when data must be encrypted or otherwise
942 altered prior to transmission.
943 .El
944 .Sh HARDWARE-ASSISTED CHECKSUM CALCULATION
945 This section currently applies to TCP/IP only.
946 In order to save the host CPU resources, computing checksums is
947 offloaded to the network interface hardware if possible.
948 The
949 .Va m_pkthdr
950 member of the leading
951 .Vt mbuf
952 of a packet contains two fields used for that purpose,
953 .Vt int Va csum_flags
954 and
955 .Vt int Va csum_data .
956 The meaning of those fields depends on the direction a packet flows in,
957 and on whether the packet is fragmented.
958 Henceforth,
959 .Va csum_flags
960 or
961 .Va csum_data
962 of a packet
963 will denote the corresponding field of the
964 .Va m_pkthdr
965 member of the leading
966 .Vt mbuf
967 in the
968 .Vt mbuf chain
969 containing the packet.
970 .Pp
971 On output, checksum offloading is attempted after the outgoing
972 interface has been determined for a packet.
973 The interface-specific field
974 .Va ifnet.if_data.ifi_hwassist
975 (see
976 .Xr ifnet 9 )
977 is consulted for the capabilities of the interface to assist in
978 computing checksums.
979 The
980 .Va csum_flags
981 field of the packet header is set to indicate which actions the interface
982 is supposed to perform on it.
983 The actions unsupported by the network interface are done in the
984 software prior to passing the packet down to the interface driver;
985 such actions will never be requested through
986 .Va csum_flags .
987 .Pp
988 The flags demanding a particular action from an interface are as follows:
989 .Bl -tag -width ".Dv CSUM_TCP" -offset indent
990 .It Dv CSUM_IP
991 The IP header checksum is to be computed and stored in the
992 corresponding field of the packet.
993 The hardware is expected to know the format of an IP header
994 to determine the offset of the IP checksum field.
995 .It Dv CSUM_TCP
996 The TCP checksum is to be computed.
997 (See below.)
998 .It Dv CSUM_UDP
999 The UDP checksum is to be computed.
1000 (See below.)
1001 .El
1002 .Pp
1003 Should a TCP or UDP checksum be offloaded to the hardware,
1004 the field
1005 .Va csum_data
1006 will contain the byte offset of the checksum field relative to the
1007 end of the IP header.
1008 In this case, the checksum field will be initially
1009 set by the TCP/IP module to the checksum of the pseudo header
1010 defined by the TCP and UDP specifications.
1011 .Pp
1012 For outbound packets which have been fragmented
1013 by the host CPU, the following will also be true,
1014 regardless of the checksum flag settings:
1015 .Bl -bullet -offset indent
1016 .It
1017 all fragments will have the flag
1018 .Dv M_FRAG
1019 set in their
1020 .Va m_flags
1021 field;
1022 .It
1023 the first and the last fragments in the chain will have
1024 .Dv M_FIRSTFRAG
1025 or
1026 .Dv M_LASTFRAG
1027 set in their
1028 .Va m_flags ,
1029 correspondingly;
1030 .It
1031 the first fragment in the chain will have the total number
1032 of fragments contained in its
1033 .Va csum_data
1034 field.
1035 .El
1036 .Pp
1037 The last rule for fragmented packets takes precedence over the one
1038 for a TCP or UDP checksum.
1039 Nevertheless, offloading a TCP or UDP checksum is possible for a
1040 fragmented packet if the flag
1041 .Dv CSUM_IP_FRAGS
1042 is set in the field
1043 .Va ifnet.if_data.ifi_hwassist
1044 associated with the network interface.
1045 However, in this case the interface is expected to figure out
1046 the location of the checksum field within the sequence of fragments
1047 by itself because
1048 .Va csum_data
1049 contains a fragment count instead of a checksum offset value.
1050 .Pp
1051 On input, an interface indicates the actions it has performed
1052 on a packet by setting one or more of the following flags in
1053 .Va csum_flags
1054 associated with the packet:
1055 .Bl -tag -width ".Dv CSUM_IP_CHECKED" -offset indent
1056 .It Dv CSUM_IP_CHECKED
1057 The IP header checksum has been computed.
1058 .It Dv CSUM_IP_VALID
1059 The IP header has a valid checksum.
1060 This flag can appear only in combination with
1061 .Dv CSUM_IP_CHECKED .
1062 .It Dv CSUM_DATA_VALID
1063 The checksum of the data portion of the IP packet has been computed
1064 and stored in the field
1065 .Va csum_data
1066 in network byte order.
1067 .It Dv CSUM_PSEUDO_HDR
1068 Can be set only along with
1069 .Dv CSUM_DATA_VALID
1070 to indicate that the IP data checksum found in
1071 .Va csum_data
1072 allows for the pseudo header defined by the TCP and UDP specifications.
1073 Otherwise the checksum of the pseudo header must be calculated by
1074 the host CPU and added to
1075 .Va csum_data
1076 to obtain the final checksum to be used for TCP or UDP validation purposes.
1077 .El
1078 .Pp
1079 If a particular network interface just indicates success or
1080 failure of TCP or UDP checksum validation without returning
1081 the exact value of the checksum to the host CPU, its driver can mark
1082 .Dv CSUM_DATA_VALID
1083 and
1084 .Dv CSUM_PSEUDO_HDR
1085 in
1086 .Va csum_flags ,
1087 and set
1088 .Va csum_data
1089 to
1090 .Li 0xFFFF
1091 hexadecimal to indicate a valid checksum.
1092 It is a peculiarity of the algorithm used that the Internet checksum
1093 calculated over any valid packet will be
1094 .Li 0xFFFF
1095 as long as the original checksum field is included.
1096 .Pp
1097 For inbound packets which are IP fragments, all
1098 .Va csum_data
1099 fields will be summed during reassembly to obtain the final checksum
1100 value passed to an upper layer in the
1101 .Va csum_data
1102 field of the reassembled packet.
1103 The
1104 .Va csum_flags
1105 fields of all fragments will be consolidated using logical AND
1106 to obtain the final value for
1107 .Va csum_flags .
1108 Thus, in order to successfully
1109 offload checksum computation for fragmented data,
1110 all fragments should have the same value of
1111 .Va csum_flags .
1112 .Sh STRESS TESTING
1113 When running a kernel compiled with the option
1114 .Dv MBUF_STRESS_TEST ,
1115 the following
1116 .Xr sysctl 8 Ns
1117 -controlled options may be used to create
1118 various failure/extreme cases for testing of network drivers
1119 and other parts of the kernel that rely on
1120 .Vt mbufs .
1121 .Bl -tag -width ident
1122 .It Va net.inet.ip.mbuf_frag_size
1123 Causes
1124 .Fn ip_output
1125 to fragment outgoing
1126 .Vt mbuf chains
1127 into fragments of the specified size.
1128 Setting this variable to 1 is an excellent way to
1129 test the long
1130 .Vt mbuf chain
1131 handling ability of network drivers.
1132 .It Va kern.ipc.m_defragrandomfailures
1133 Causes the function
1134 .Fn m_defrag
1135 to randomly fail, returning
1136 .Dv NULL .
1137 Any piece of code which uses
1138 .Fn m_defrag
1139 should be tested with this feature.
1140 .El
1141 .Sh RETURN VALUES
1142 See above.
1143 .Sh SEE ALSO
1144 .Xr ifnet 9 ,
1145 .Xr mbuf_tags 9
1146 .Sh HISTORY
1147 .\" Please correct me if I'm wrong
1148 .Vt Mbufs
1149 appeared in an early version of
1150 .Bx .
1151 Besides being used for network packets, they were used
1152 to store various dynamic structures, such as routing table
1153 entries, interface addresses, protocol control blocks, etc.
1154 In more recent
1155 .Fx
1156 use of
1157 .Vt mbufs
1158 is almost entirely limited to packet storage, with
1159 .Xr uma 9
1160 zones being used directly to store other network-related memory.
1161 .Pp
1162 Historically, the
1163 .Vt mbuf
1164 allocator has been a special-purpose memory allocator able to run in
1165 interrupt contexts and allocating from a special kernel address space map.
1166 As of
1167 .Fx 5.3 ,
1168 the
1169 .Vt mbuf
1170 allocator is a wrapper around
1171 .Xr uma 9 ,
1172 allowing caching of
1173 .Vt mbufs ,
1174 clusters, and
1175 .Vt mbuf
1176 + cluster pairs in per-CPU caches, as well as bringing other benefits of
1177 slab allocation.
1178 .Sh AUTHORS
1179 The original
1180 .Nm
1181 manual page was written by Yar Tikhiy.
1182 The
1183 .Xr uma 9
1184 .Vt mbuf
1185 allocator was written by Bosko Milekic.