Unmapped mbufs allow sendfile to carry multiple pages of data in a
single mbuf, without mapping those pages. It is a requirement for
Netflix's in-kernel TLS, and provides a 5-10% CPU savings on heavy web
serving workloads when used by sendfile, due to effectively
compressing socket buffers by an order of magnitude, and hence
reducing cache misses.
For this new external mbuf buffer type (EXT_PGS), the ext_buf pointer
now points to a struct mbuf_ext_pgs structure instead of a data
buffer. This structure contains an array of physical addresses (this
reduces cache misses compared to an earlier version that stored an
array of vm_page_t pointers). It also stores additional fields needed
for in-kernel TLS such as the TLS header and trailer data that are
currently unused. To more easily detect these mbufs, the M_NOMAP flag
is set in m_flags in addition to M_EXT.
Various functions like m_copydata() have been updated to safely access
packet contents (using uiomove_fromphys()), to make things like BPF
safe.
NIC drivers advertise support for unmapped mbufs on transmit via a new
IFCAP_NOMAP capability. This capability can be toggled via the new
'nomap' and '-nomap' ifconfig(8) commands. For NIC drivers that only
transmit packet contents via DMA and use bus_dma, adding the
capability to if_capabilities and if_capenable should be all that is
required.
If a NIC does not support unmapped mbufs, they are converted to a
chain of mapped mbufs (using sf_bufs to provide the mapping) in
ip_output or ip6_output. If an unmapped mbuf requires software
checksums, it is also converted to a chain of mapped mbufs before
computing the checksum.
Submitted by: gallatin (earlier version)
Reviewed by: gallatin, hselasky, rrs
Discussed with: ae, kp (firewalls)
Relnotes: yes
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20616
All of these arguments are stored in m_ext, so there is no reason
to pass them in the argument list. Not all functions need the second
argument, some don't even need the first one. The second argument
lives in next cache line, so not dereferencing it is a performance
gain. This was discovered in sendfile(2), which will be covered by
next commits.
The second goal of this commit is to bring even more flexibility
to m_ext mbufs, allowing to create more fields in m_ext, opaque to
the generic mbuf code, and potentially set and dereferenced by
subsystems.
Reviewed by: gallatin, kbowling
Differential Revision: https://reviews.freebsd.org/D12615
- Add m_getclr(9) symlink to ObsoleteFiles.inc (removed in r295481).
- Add const qualifiers in m_dup(), m_dup_pkthdr() and m_tag_copy_chain()
(r286450).
- Fix m_dup_pkthdr() definition (it's not the same as m_move_pkthdr()).
MFC after: 5 days
Since m_cat() may copy data from the second mbuf chain into the last mbuf
of the first chain, it may free the first mbuf of the second chain. Thus,
the second chain is not guaranteed to be valid after m_cat() returns.
Reviewed by: glebius
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D5497
socket-buffer implementations, introduce a return value for MCLGET()
(and m_cljget() that underlies it) to allow the caller to avoid testing
M_EXT itself. Update all callers to use the return value.
With this change, very few network device drivers remain aware of
M_EXT; the primary exceptions lie in mbuf-chain pretty printers for
debugging, and in a few cases, custom mbuf and cluster allocation
implementations.
NB: This is a difficult-to-test change as it touches many drivers for
which I don't have physical devices. Instead we've gone for intensive
review, but further post-commit review would definitely be appreciated
to spot errors where changes could not easily be made mechanically,
but were largely mechanical in nature.
Differential Revision: https://reviews.freebsd.org/D1440
Reviewed by: adrian, bz, gnn
Sponsored by: EMC / Isilon Storage Division
Since SMPng it wasn't really supported anymore and if it worked
then only by chance. Only very few drivers ever supported it.
Discussed with: yongari
MFC after: 2 weeks
Removed dead code that assumed that M_TRYWAIT can return NULL; it's not true
since the advent of MBUMA.
Reviewed by: arch
There are ongoing disputes as to whether we want to switch to directly using
UMA flags M_WAITOK/M_NOWAIT for mbuf(9) allocation.
free function controlable, instead of passing the KVA of the buffer
storage as the first argument.
Fix all conventional users of the API to pass the KVA of the buffer
as the first argument, to make this a no-op commit.
Likely break the only non-convetional user of the API, after informing
the relevant committer.
Update the mbuf(9) manual page, which was already out of sync on
this point.
Bump __FreeBSD_version to 800016 as there is no way to tell how
many arguments a CPP macro needs any other way.
This paves the way for giving sendfile(9) a way to wait for the
passed storage to have been accessed before returning.
This does not affect the memory layout or size of mbufs.
Parental oversight by: sam and rwatson.
No MFC is anticipated.
transition to mbuma (FreeBSD 5.3) and the fact that mbufs are now limited
almost entirely to packet storage, with straight UMA zones being used for
most other network data types.
alignment restrictive, and help performance on some ethernet cards which
currently copy the entire packet a couple bytes to get the packet aligned
properly...
Wordsmithing by: dwhite
Obtained from: NetBSD (code only)
I'll clean it up later: rwatson