Put the kernelside rump kernel headers into <rump-sys> instead of
sprinkling them around the faction directories. Avoids having
to add a CPPFLAGS (or several) to pretty much every component
Makefile.

Leave compat headers around in the old locations.

The commit changes some autogenerated files, but I'll fix the
generators shortly and regen.

Add a COMMENT describing what each component roughly does.

"make describe" prints the comment.

Requested/inspired by Vincent Schwarzer on rumpkernel-users

branches: 1.6.6;
rename component.c -> shmif_component.c

rename rumpcomp_user.* -> shmif_user.*

branches: 1.4.4;
make the file monitoring hypercalls private to shmif

branches: 1.3.8; 1.3.18;
* make interface a cloner
* use SIOCSLINKSTR for supplying bus filename in case of cloned if

TODO: downing interface, unclone, and some tweaks for robustness

Include a pointer to the beginning of the buffer and add support
to the packet dumper. This helps in situations where the juicy
details are in a bus multiple generations old.

branches: 1.1.2; 1.1.4; 1.1.6; 1.1.8; 1.1.10;
Add a virtual ethernet interface which uses shared memory as the
bus instead of relying on the host kernel's tap and bridge. This
is much lighterweight approach which does not consume non-pageable
kernel resources.

rename component.c -> shmif_component.c

branches: 1.2.6; 1.2.10; 1.2.20; 1.2.24;
Use vmem for keeping track of unit numbers so that wildcard allocation
works correctly.

remember to commit this file too

this was moved to usr.bin ages ago

branches: 1.10.2; 1.10.6;
un-\240

add -h which prints only the bus header info

branches: 1.8.2; 1.8.4;
truncate pcap output file

Include a timestamp in the frame header. When converting to pcap,
it can give some idea of when packets were sent.

nb. it's the sending host's timestamp, not an observer timestamp
like in the typical pcap case.

Include a pointer to the beginning of the buffer and add support
to the packet dumper. This helps in situations where the juicy
details are in a bus multiple generations old.

Recognize ``-'' for stdout. allows piping output to tcpdump -r -

* include a magic number in the bus format
* simplify offset calculations by making them start from beginning of data

Make shmif memory access slightly more sane. Create a header which
is shared by the interface and the bus analyzer.

pass Wall

Add a little utility to convert shmif bus traffic to a pcap file
which can be then examined with tcpdump -r, wireshark, etc.

shmif: Fix logics for media change and status

- shmif_mediachange: Drop check for if_link_state. Otherwise,
there can be race b/w events in if_link_queue.

- shmif_mediastatus: Set ifm_status appropriately, by which
link state appears in ifconfig(8) "status:" line.

Pointed out by roy@, and taken from his codes in vether(4). Thanks!!

shmif: export LOCK_* definitions

They will be used by upcoming shmif_pcapin.

shmem(4): Fix typo in comment: AFT -> ATF.

Also fix grammar (if I understood correctly what this meant: rump
servers written in C, rather than set up via shell scripts around
rump_server invoking ifconfig).

No functional change intended.

shmif: support media

It enables to link-down shmif by ifconfig media none and link-up
again by media auto.

shmif: change behaviors about link states

- Change the link state to UP on ifconfig linkstr
- This behavior emulates physical devices
- Change the link state to UNKNOWN on ifconfig -linkstr just in case
- Reject sending/receiving packets if the link state is DOWN
- Permit to send/receive packets on UNKNOWN, which is required
to unbreak some ATF tests written in C

branches: 1.84.4; 1.84.10;
if_shmem(4): Use membar_acquire/release for lock acquire/release.

shmif: support ALTQ

if_attach and if_initialize cannot fail, don't test return value

These were originally made failable back in 2017 when if_initialize
allocated a softint in every interface for link state changes, so
that it could fail gracefully instead of panicking:

https://mail-index.NetBSD.org/source-changes/2017/10/23/msg089053.html

However, this spawned many seldom- or never-tested error branches,
which are risky to have around. And that softint in every interface
has since been replaced by a single global workqueue, because link
state changes require thread context but not low latency or high
throughput:

https://mail-index.NetBSD.org/source-changes/2020/02/06/msg113759.html

So there is no longer any reason for if_initialize to fail. (The
subroutine if_stats_init can't fail because percpu_alloc can't fail
either.)

There is a snag: the softint_establish in if_percpuq_create could
fail, potentially leading to bad consequences later on trying to use
the softint. This change doesn't introduce any new bugs because of
the snag -- if_percpuq_attach was already broken. However, the snag
can be better addressed without spawning error branches, either by
using a single softint or making softints less scarce.

(Separate commit will change the signatures of if_attach and
if_initialize to return void, scheduled to ride whatever is the next
convenient kernel bump.)

Patch and testing on amd64 and evbmips64-eb by maya@; commit message
soliloquy, and compile-testing on evbppc/i386/earmv7hf, by me.

branches: 1.81.10;
shmif: reduce the number of calls of cprng(9)

shmif: s/sc_uuid/sc_uid/

It's not UUID :-/

shmif: use cprng_strong64 instead of cprng_fast64 to generate a unique ID

shmif uses random bytes generated by cprng(9) as a unique device ID
between rump kernels to identify packets fed by itself and not receive
them. So if generated bytes are identical between shmif interfaces on
different rump kernels, shmif may drop incoming packets unintentionally.
This is one cause of recent ATF test failures of IPsec.

Fix it by using cprng_strong64 instead of cprng_fast64. This is a
workaround and we should also investigate why cprng_fast64 starts
failing on rump kernels, although using cprng_strong64 in initialization
itself is feasible.

Fix PR kern/54897

shmif: use cprng_strong32 to generate random bytes for a MAC address

cprng_fast32 sometimes returns indentical bytes, which look
"20:0e:11:33" in a MAC address, on different rump_server instances.
That leads MAC address duplications resulting in a test failure.

Fix it by using cprng_strong32 instead of cprng_fast32. However
we should rather fix cprng_fast32 (or rump itself) somehow.

The fix mitigates PR kern/54897 but test failures due to other
causes still remain.

Adopt <net/if_stats.h>.

branches: 1.76.6;
Add TX/RX offload capabilities to shmif(4). They are emulated in
software by ether_sw_offload_[tr]x().

For rump kernels, if_capabilities for shmemif(4) can be specified
by environmental variable RUMP_SHMIF_CAPENABLE:

setenv RUMP_SHMIF_CAPENABLE 0x7ff80 (all offload)
setenv RUMP_SHMIF_CAPENABLE 0x6aa80 (all TX)
setenv RUMP_SHMIF_CAPENABLE 0x15500 (all RX)

part of PR kern/53562
OK msaitoh

branches: 1.75.2;
Implement the BPF direction filter (BIOC[GS]DIRECTION). It provides backward
compatibility with BIOC[GS]SEESENT ioctl. The userland interface is the same
as FreeBSD.

This change also fixes a bug that the direction is misunderstand on some
environment by passing the direction to bpf_mtap*() instead of checking
m->m_pkthdr.rcvif.

branches: 1.74.2;
Fix compile error.

If if_initialize() failed in the attach function, free resources and return.

branches: 1.72.8;
Fix that rump.ifconfig shmifN destroy hangs up

rump.ifconfig shmifN destroy sometimes hangs up on closing a file
descriptor of kqueue under heavy load. It seems it happens because of
a race condition between closing a fd and kevent to the same fd on
another kthread for packet Rx. Fix it by executing the close operation
after the kthread left.

Fix handling return value of rumpcomp_shmif_watchwait

Move bpf_mtap and if_ipackets++ on Rx of each driver to percpuq if_input

The benefits of the change are:
- We can reduce codes
- We can provide the same behavior between drivers
- Where/When if_ipackets is counted up
- Note that some drivers still update packet statistics in their own
way (periodical update)
- Moved bpf_mtap run in softint
- This makes it easy to MP-ify bpf

Proposed on tech-kern and tech-net

branches: 1.69.2;
KNF. Remove extra spaces. No functional change.

Use curlwp_bind and curlwp_bindx instead of open-coding LP_BOUND

Introduce m_set_rcvif and m_reset_rcvif

The API is used to set (or reset) a received interface of a mbuf.
They are counterpart of m_get_rcvif, which will come in another
commit, hide internal of rcvif operation, and reduce the diff of
the upcoming change.

No functional change.

Prevent LWP migrations between CPUs during upper layer processing

This is a contract of psref(9) that is used by upper layer componenets,
e.g., bridge(4).

Introduce softint-based if_input

This change intends to run the whole network stack in softint context
(or normal LWP), not hardware interrupt context. Note that the work is
still incomplete by this change; to that end, we also have to softint-ify
if_link_state_change (and bpf) which can still run in hardware interrupt.

This change softint-ifies at ifp->if_input that is called from
each device driver (and ieee80211_input) to ensure Layer 2 runs
in softint (e.g., ether_input and bridge_input). To this end,
we provide a framework (called percpuq) that utlizes softint(9)
and percpu ifqueues. With this patch, rxintr of most drivers just
queues received packets and schedules a softint, and the softint
dequeues packets and does rest packet processing.

To minimize changes to each driver, percpuq is allocated in struct
ifnet for now and that is initialized by default (in if_attach).
We probably have to move percpuq to softc of each driver, but it's
future work. At this point, only wm(4) has percpuq in its softc
as a reference implementation.

Additional information including performance numbers can be found
in the thread at tech-kern@ and tech-net@:
http://mail-index.netbsd.org/tech-kern/2016/01/14/msg019997.html

Acknowledgment: riastradh@ greatly helped this work.
Thank you very much!

Put the kernelside rump kernel headers into <rump-sys> instead of
sprinkling them around the faction directories. Avoids having
to add a CPPFLAGS (or several) to pretty much every component
Makefile.

Leave compat headers around in the old locations.

The commit changes some autogenerated files, but I'll fix the
generators shortly and regen.

branches: 1.63.2;
Make shmif SIMPLEX

Add a sender field to a packet header on a shmif bus to identify
and ignore packets sent by itself.

This makes shmif work with bridges.

ok pooka@

branches: 1.62.2;
Count packets in if_shmem

ok @pooka

Add missing __diagused in rump code

branches: 1.60.2;
kill sprintf

rename rumpcomp_user.* -> shmif_user.*

Remove busversion.

In recv, align data after ether_header at _ALIGN_BYTES.

branches: 1.56.2; 1.56.4;
sauce previous with a comment

Ignore too short packets. This condition is true especially for the
first packet on a new bus (len == 0). Everything else in the stack
besides bpf seemed to be able to deal with one, though.

Three cheers for the mighty C type system that doesn't recognize the
difference between an enum and an integer. Due to wrong parameter order,
the lock backoff routine ended up sleeping 1 million seconds instead
of 1ms. On the plus side, we got some sort of idea how often the shmif
spinlock backoff routine is hit during a standard test run.

Make hypercall calling conventions consistent: iff a hypercall can fail,
it returns an int containing the error value.

Reduce hypercalls related to reading to essentially an amalgamation
of readv and preadv. ditto for writing. Hypercalls are so seldomly
used that it doesn't justify 3x the calls for syntactic sugar.

add a private hypercall to map the bus into memory

Improve the time-related hypercalls so that's it's possible to
sleep until an absolute time on the host's monotonic clock (should
something like that be supported).

* remove the unused "opaque" parameter
* fix some typos in the previous

make the file monitoring hypercalls private to shmif

fix debug printf formats

Pass up all multicast addresses, not just broadcast. Among other
things, makes IPv6 work over this interface.

Do not assume that O_FOO flags have matching values in the rump kernel
and hypervisor.

branches: 1.44.4; 1.44.8;
First step of random number subsystem rework described in
<20111022023242.BA26F14A158@mail.netbsd.org>. This change includes
the following:

An initial cleanup and minor reorganization of the entropy pool
code in sys/dev/rnd.c and sys/dev/rndpool.c. Several bugs are
fixed. Some effort is made to accumulate entropy more quickly at
boot time.

A generic interface, "rndsink", is added, for stream generators to
request that they be re-keyed with good quality entropy from the pool
as soon as it is available.

The arc4random()/arc4randbytes() implementation in libkern is
adjusted to use the rndsink interface for rekeying, which helps
address the problem of low-quality keys at boot time.

An implementation of the FIPS 140-2 statistical tests for random
number generator quality is provided (libkern/rngtest.c). This
is based on Greg Rose's implementation from Qualcomm.

A new random stream generator, nist_ctr_drbg, is provided. It is
based on an implementation of the NIST SP800-90 CTR_DRBG by
Henric Jungheim. This generator users AES in a modified counter
mode to generate a backtracking-resistant random stream.

An abstraction layer, "cprng", is provided for in-kernel consumers
of randomness. The arc4random/arc4randbytes API is deprecated for
in-kernel use. It is replaced by "cprng_strong". The current
cprng_fast implementation wraps the existing arc4random
implementation. The current cprng_strong implementation wraps the
new CTR_DRBG implementation. Both interfaces are rekeyed from
the entropy pool automatically at intervals justifiable from best
current cryptographic practice.

In some quick tests, cprng_fast() is about the same speed as
the old arc4randbytes(), and cprng_strong() is about 20% faster
than rnd_extract_data(). Performance is expected to improve.

The AES code in src/crypto/rijndael is no longer an optional
kernel component, as it is required by cprng_strong, which is
not an optional kernel component.

The entropy pool output is subjected to the rngtest tests at
startup time; if it fails, the system will reboot. There is
approximately a 3/10000 chance of a false positive from these
tests. Entropy pool _input_ from hardware random numbers is
subjected to the rngtest tests at attach time, as well as the
FIPS continuous-output test, to detect bad or stuck hardware
RNGs; if any are detected, they are detached, but the system
continues to run.

A problem with rndctl(8) is fixed -- datastructures with
pointers in arrays are no longer passed to userspace (this
was not a security problem, but rather a major issue for
compat32). A new kernel will require a new rndctl.

The sysctl kern.arandom() and kern.urandom() nodes are hooked
up to the new generators, but the /dev/*random pseudodevices
are not, yet.

Manual pages for the new kernel interfaces are forthcoming.

branches: 1.43.2;
Report vmem(9) errors out-of-band so that we can use vmem(9) to manage
ranges that include the least and the greatest vmem_addr_t. Update
vmem(9) uses throughout the kernel. Slightly expand on the tests in
subr_vmem.c, which still pass. I've been running a kernel with this
patch without any trouble.

Use VMEM_ADDR_MIN and VMEM_ADDR_MAX.

Introduce a couple of new constants, VMEM_ADDR_MIN (the least possible
address in a vmem(9) arena, 0) and VMEM_ADDR_MAX (the maximum possible
address, currently 0xFFFFFFFF). Modify several boundary conditions so
that a vmem(9) arena can allocate ranges including VMEM_ADDR_MAX.
Update documentation and tests.

These changes pass the tests in sys/kern/subr_vmem.c. To compile the
and run the test program, run "cd sys/kern/ && gcc -DVMEM_SANITY -o
subr_vmem ./subr_vmem.c && ./subr_vmem".

Rename slightly misleading KTHREAD_JOINABLE to KTHREAD_MUSTJOIN.

Update copyright statements.

no functional change.

make the if-else logic more obvious

After my change to the "interface accepts this packet" logic
yesterday the CARP test stopped working, since CARP depends on
IFF_PROMISC (which was previously always accidentally enabled).
While making the interface honor IFF_PROMISC, also make it compare
the received frame's address against ifp->if_sadl instead of a
local enaddr value we cached when the interface was created.

Don't assume rump kernel PAGE_SIZE and host page size are the same.

Support bpf. shmif_dumpbus(1) can be used for much the same effect,
but sometimes it's just more convenient to run tcpdump live.

Pass packet up if it's *for* us, not if it's from someone else.
This fixes a rather curious forwarding/redirect/etc. storm which
happened when there were >2 shmif kernels on the same shmbus with
ip forwarding set on. (at least it stress-tested other code ;)

branches: 1.33.2;
Allow creation with NULL busname (to be later set with SIOCSLINKSTR).

Support destroy in shmif.

Use vmem for keeping track of unit numbers so that wildcard allocation
works correctly.

"fix" completely crackheaded code w/ uninitialized use,
missed as usual by the wonderful -g -O0

* make interface a cloner
* use SIOCSLINKSTR for supplying bus filename in case of cloned if

TODO: downing interface, unclone, and some tweaks for robustness

Prefault bus pages in driver attachment. This seems to work around
weird corruption I've been seeing (most likely a host kernel bug).

For more details, see thread at:
http://mail-index.netbsd.org/tech-kern/2010/08/17/msg008749.html

* fix off-by-wrap case where current datagram aligns exactly with
the end of the bus
* clarify the "can we still use the device's next pointer" calculation
and move it to its own routine
* sprinkle dprintf

simplify, improve, etc.

reset pktsize each loop

reset wrap where necessary

Move the lockops together with the interface -- they are needed
only at runtime.

Take kernel lock before passing data to if_input. This is in line
with IPL_NET interrupts generally not being MPSAFE.

Don't do timestamping with buslock held. it's the little things ...

Include a timestamp in the frame header. When converting to pcap,
it can give some idea of when packets were sent.

nb. it's the sending host's timestamp, not an observer timestamp
like in the typical pcap case.

Include a pointer to the beginning of the buffer and add support
to the packet dumper. This helps in situations where the juicy
details are in a bus multiple generations old.

further simplify & fix calculations

* include a magic number in the bus format
* simplify offset calculations by making them start from beginning of data

Make shmif memory access slightly more sane. Create a header which
is shared by the interface and the bus analyzer.

* remove some magic numbers from the code
* include bus version number in header just in case we want to examine
bus traffic at some point in the future

Change bus header a little: reserve only 32bits for the lock and
use 32bit atomic ops to handle it. Begin data from 0x10 instead
of 0x14.

* use atomic ops instead of __cpu_simple_luck
* this interface is un-IFF_SIMPLEX

* print interface number, backend path and ethernet address for dmesg
* make kassert more demanding

* increase bus size so that we don't wrap so quickly when NFS
flushes writes
* verbosize debug print a bit

branches: 1.10.2; 1.10.4;
fix comment

use autogenerated prototype

printf -> rumpuser_dprintf
(can't use printf while holding spinlock)

Filter out ENETRESET from ether_ioctl() since we aren't interested
in multicast hugging.

branches: 1.6.2; 1.6.4;
let drivers define DPRINTF

* allow to specify PROT_READ/PROT_WRITE when mmapping a file
* add msync

branches: 1.4.2;
Create receive thread only in if_start() to make sure it happens
after the we have been attached to the ethernet framework.
XXX: if_stop is still unimplemented

add missing atomic.h include

don't require the ifnum return pointer.

Add a virtual ethernet interface which uses shared memory as the
bus instead of relying on the host kernel's tap and bridge. This
is much lighterweight approach which does not consume non-pageable
kernel resources.

rename rumpcomp_user.* -> shmif_user.*

OpenBSD has kqueue

Use kqueue on *BSD, not just NetBSD.

branches: 1.12.2; 1.12.4;
On Linux, pass the inotify descriptor up so that we can correctly read
it in the wait() routine.

Include necessary headers on Linux

Translate error codes between the hypervisor and rump kernel.

fix non-NetBSD implementation

Make hypercall calling conventions consistent: iff a hypercall can fail,
it returns an int containing the error value.

don't build in kernel build.

add a private hypercall to map the bus into memory

some more headers are required by the third implementation

include event.h only where used

be a bit less sloppy with errno

* remove the unused "opaque" parameter
* fix some typos in the previous

make the file monitoring hypercalls private to shmif

rename rumpcomp_user.* -> shmif_user.*

branches: 1.4.4; 1.4.6;
Make hypercall calling conventions consistent: iff a hypercall can fail,
it returns an int containing the error value.

add a private hypercall to map the bus into memory

* remove the unused "opaque" parameter
* fix some typos in the previous

make the file monitoring hypercalls private to shmif

Remove shlib_version files and just use Makefile SHLIB_MAJOR/MINOR,
with the default provided by Makefile.rump (they're all 0.0 anyway)

branches: 1.1.2; 1.1.4; 1.1.6; 1.1.28; 1.1.46;
Add a virtual ethernet interface which uses shared memory as the
bus instead of relying on the host kernel's tap and bridge. This
is much lighterweight approach which does not consume non-pageable
kernel resources.

Make shmif buildable when DRPINTF enabled

use _KERNEL_RCSID only in _KERNEL, _RCSID elsewhere

header massage, mostly improves portability

branches: 1.9.4; 1.9.14; 1.9.18;
Update copyright statements.

no functional change.

branches: 1.8.2; 1.8.6;
make this happy to compile standalone

branches: 1.7.2;
Fix hopefully last off-by-one: if we fill the bus, we must also
advance the "first" pointer. This problem triggered only if the
bus was filled in the first round, since the first pointer is at
the end-of-bus only for the bootstrap round.

* fix off-by-wrap case where current datagram aligns exactly with
the end of the bus
* clarify the "can we still use the device's next pointer" calculation
and move it to its own routine
* sprinkle dprintf

branches: 1.5.2;
simplify, improve, etc.

Move the lockops together with the interface -- they are needed
only at runtime.

introduce a sleep to trying to grab the bus

Include a timestamp in the frame header. When converting to pcap,
it can give some idea of when packets were sent.

nb. it's the sending host's timestamp, not an observer timestamp
like in the typical pcap case.

Include a pointer to the beginning of the buffer and add support
to the packet dumper. This helps in situations where the juicy
details are in a bus multiple generations old.

Put the kernelside rump kernel headers into <rump-sys> instead of
sprinkling them around the faction directories. Avoids having
to add a CPPFLAGS (or several) to pretty much every component
Makefile.

Leave compat headers around in the old locations.

The commit changes some autogenerated files, but I'll fix the
generators shortly and regen.

branches: 1.1.4; 1.1.6; 1.1.10; 1.1.12;
rename component.c -> shmif_component.c

Protect __KERNEL_RCSID. It isn't available when compile rumpkernels under e.g. Linux.

Merge the [pgoyette-compat] branch

Add missing RCSIDs.

branches: 1.2.2;
Unconditionally include <unistd.h>, as we use ftruncate() unconditionally
too.
Reported (with patch) by Antonio Huete Jimenez in PR 53153, slightly
modified by me.

branches: 1.1.4; 1.1.6; 1.1.10; 1.1.34;
rename rumpcomp_user.* -> shmif_user.*

branches: 1.1.4; 1.1.6; 1.1.10;
rename rumpcomp_user.* -> shmif_user.*

shmif: export LOCK_* definitions

They will be used by upcoming shmif_pcapin.

branches: 1.11.22; 1.11.28;
make constant unsigned

branches: 1.10.20;
Make shmif buildable when DRPINTF enabled

Bump SHMIF_VERSION

Make shmif SIMPLEX

Add a sender field to a packet header on a shmif bus to identify
and ignore packets sent by itself.

This makes shmif work with bridges.

ok pooka@

branches: 1.7.4;
header massage, mostly improves portability

branches: 1.6.4; 1.6.14; 1.6.18;
Update copyright statements.

no functional change.

branches: 1.5.2; 1.5.4; 1.5.6; 1.5.10;
Move the lockops together with the interface -- they are needed
only at runtime.

Include a timestamp in the frame header. When converting to pcap,
it can give some idea of when packets were sent.

nb. it's the sending host's timestamp, not an observer timestamp
like in the typical pcap case.

Include a pointer to the beginning of the buffer and add support
to the packet dumper. This helps in situations where the juicy
details are in a bus multiple generations old.

* include a magic number in the bus format
* simplify offset calculations by making them start from beginning of data

Make shmif memory access slightly more sane. Create a header which
is shared by the interface and the bus analyzer.