zone. Cluster allocations fail when this happens. Also processes that may have
blocked on cluster allocations will never be woken up. Thanks to rwatson for
an overview of the issue and pointers to the mbuma paper and his tool to dump
out UMA zones.
Reviewed by: andre@
begun with a repo-copy of mac.h to mac_framework.h. sys/mac.h now
contains the userspace and user<->kernel API and definitions, with all
in-kernel interfaces moved to mac_framework.h, which is now included
across most of the kernel instead.
This change is the first step in a larger cleanup and sweep of MAC
Framework interfaces in the kernel, and will not be MFC'd.
Obtained from: TrustedBSD Project
Sponsored by: SPARTA
- Move sonewconn(), which creates new sockets for incoming connections on
listen sockets, so that all socket allocate code is together in
uipc_socket.c.
- Move 'maxsockets' and associated sysctls to uipc_socket.c with the
socket allocation code.
- Move kern.ipc sysctl node to uipc_socket.c, add a SYSCTL_DECL() for it
to sysctl.h and remove lots of scattered implementations in various
IPC modules.
- Sort sodealloc() after soalloc() in uipc_socket.c for dependency order
reasons. Statisticize soalloc() and sodealloc() as they are now
required only in uipc_socket.c, and are internal to the socket
implementation.
After this change, socket allocation and deallocation is entirely
centralized in one file, and uipc_socket2.c consists entirely of socket
buffer manipulation and default protocol switch functions.
MFC after: 1 month
jumbo mbuf clusters. To make the variable size clear they are named
MJUMPAGESIZE.
Having jumbo clusters with the native PAGE_SIZE is more useful than
a fixed 4k size according the device driver writers using this API.
The 9k and 16k jumbo mbuf clusters remain unchanged.
Requested by: glebius, gallatin
Sponsored by: TCP/IP Optimization Fundraise 2005
MFC after: 3 days
work by yar, thompsa and myself. The checksum offloading part also involves
work done by Mihail Balikov.
The most important changes:
o Instead of global linked list of all vlan softc use a per-trunk
hash. The size of hash is dynamically adjusted, depending on
number of entries. This changes struct ifnet, replacing counter
of vlans with a pointer to trunk structure. This change is an
improvement for setups with big number of VLANs, several interfaces
and several CPUs. It is a small regression for a setup with a single
VLAN interface.
An alternative to dynamic hash is a per-trunk static array with
4096 entries, which is a compile time option - VLAN_ARRAY. In my
experiments the array is not an improvement, probably because such
a big trunk structure doesn't fit into CPU cache.
o Introduce an UMA zone for VLAN tags. Since drivers depend on it,
the zone is declared in kern_mbuf.c, not in optional vlan(4) driver.
This change is a big improvement for any setup utilizing vlan(4).
o Use rwlock(9) instead of mutex(9) for locking. We are the first
ones to do this! :)
o Some drivers can do hardware VLAN tagging + hardware checksum
offloading. Add an infrastructure for this. Whenever vlan(4) is
attached to a parent or parent configuration is changed, the flags
on vlan(4) interface are updated.
In collaboration with: yar, thompsa
In collaboration with: Mihail Balikov <mihail.balikov interbgc.com>
The success of the cluster allocation is checked through a field in the
mbuf structure. This change is non-functional but properly silences code
inspection tools.
Found by: Coverity Prevent(tm)
Coverity ID: CID807
Sponsored by: TCP/IP Optimization Fundraise 2005
4k clusters in addition to 9k and 16k ones.
struct mbuf *m_getjcl(int how, short type, int flags, int size)
void *m_cljget(struct mbuf *m, int how, int size)
m_getjcl() returns an mbuf with a cluster of the specified size attached
like m_getcl() does for 2k clusters.
m_cljget() is different from m_clget() as it can allocate clusters
without attaching them to an mbuf. In that case the return value
is the pointer to the cluster of the requested size. If an mbuf was
specified, it gets the cluster attached to it and the return value
can be safely ignored.
For size both take MCLBYTES, MJUM4BYTES, MJUM9BYTES, MJUM16BYTES.
Reviewed by: glebius
Tested by: glebius
Sponsored by: TCP/IP Optimization Fundraise 2005
from there. All others get broken up and free'd individually to the mbuf
and cluster zones.
The packet zone is a secondary zone to the mbuf zone. There is currently
a limitation in UMA which prevents decreasing the packet zone stock when
the mbuf and cluster zone are drained and all their members are part of
packets. When this is fixed this change may be reverted.
based jumbo 9k and jumbo 16k cluster support.
All mbuf's with external storage attached are mandatory reference
counted. For clusters and jumbo clusters UMA provides the refcnt
storage directly. It does not have to be separatly allocated. Any
other type of external storage gets its own refcnt allocated from
an UMA mbuf refcnt zone instead of normal kernel malloc.
The refcount API MEXT_ADD_REF() and MEXT_REM_REF() is no longer
publically accessible. The proper m_* functions have to be used.
mb_ctor_clust() and mb_dtor_clust() both handle normal 2K as well
as 9k and 16k clusters.
Clusters and jumbo clusters may be obtained without attaching it
immideatly to an mbuf. This is for high performance cluster
allocation in network drivers where mbufs are attached after the
cluster has been filled.
Tested by: rwatson
Sponsored by: TCP/IP Optimizations Fundraise 2005
are string names for their respective UMA zones and malloc types, and
are passed into uma_zcreate() and MALLOC_DEFINE(). Export them
outside of _KERNEL in mbuf.h so that netstat can reference them.
Change the names to improve consistency, with each zone/type
associated with the mbuf allocator being prefixed mbuf_.
MFC after: 1 week
reporting - in my previous change, I missed the case where a mbuf
from the packet zone was freed back to the mbuf/packet keg, where
it was subsequently put into the mbuf zone and found not to contain
the expected trash. This change adds the necessary trash_dtor call inside
mb_fini_pack so that everything is correct.
Thanks for Bosko for finding the bug and showing me how secondary zones
work.
Approved by: re (dwhite)
the UMA "trash" allocator is used - this ensures that any writes to a freed
mbuf should provoke a panic.
Only enabled under INVARIANTS, of course.
Approved by: re (scottl)
statement from some files, so re-add it for the moment, until the
related legalese is sorted out. This change affects:
sys/kern/kern_mbuf.c
sys/vm/memguard.c
sys/vm/memguard.h
sys/vm/uma.h
sys/vm/uma_core.c
sys/vm/uma_dbg.c
sys/vm/uma_dbg.h
sys/vm/uma_int.h
do not need to perform an extra memory fetch in the Packet (Mbuf+Cluster)
constructor to initialize the reference counter anymore. The reference
counts are located in a separate memory region (in the slab header,
because this zone is UMA_ZONE_REFCNT), so the memory fetch resulted very
often in a cache miss. Additionally, and perhaps more significantly,
optimize the free mbuf+cluster (packet) case, which is very common, to
no longer require an atomic operation on free (to verify the reference
counter) if the reference on the cluster has never been increased (also
very common). Reduces an atomic on mbuf free on average.
Original patch submitted by: Gerrit Nagelhout <gnagelhout@sandvine.com>
all reserved, as the lisence makes clear), and strike the third clause
(now this is a 2-clause liberal BSDL as are the rest of files I hold
copyright over).
Ask uma_zcreate() to align mbufs to MSIZE bytes (otherwise dtom() breaks)
As it happens, uma_zalloc_arg() always returned mbufs aligned to MSIZE
anyway, but that was an implementation side-effect....
KASSERT -> CTASSERT suggested by: dd@
Approved by: silence on -net
so that they know whether the allocation is supposed to be able to sleep
or not.
* Allow uma_zone constructors and initialation functions to return either
success or error. Almost all of the ones in the tree currently return
success unconditionally, but mbuf is a notable exception: the packet
zone constructor wants to be able to fail if it cannot suballocate an
mbuf cluster, and the mbuf allocators want to be able to fail in general
in a MAC kernel if the MAC mbuf initializer fails. This fixes the
panics people are seeing when they run out of memory for mbuf clusters.
* Allow debug.nosleepwithlocks on WITNESS to be disabled, without changing
the default.
Both bmilekic and jeff have reviewed the changes made to make failable
zone allocations work.
nextpkt within the m_hdr was not being initialized to NULL for
!M_PKTHDR cases. *Maybe* this will fix weird socket buffer
inconsistency panics, but we'll see.
mbuma is an Mbuf & Cluster allocator built on top of a number of
extensions to the UMA framework, all included herein.
Extensions to UMA worth noting:
- Better layering between slab <-> zone caches; introduce
Keg structure which splits off slab cache away from the
zone structure and allows multiple zones to be stacked
on top of a single Keg (single type of slab cache);
perhaps we should look into defining a subset API on
top of the Keg for special use by malloc(9),
for example.
- UMA_ZONE_REFCNT zones can now be added, and reference
counters automagically allocated for them within the end
of the associated slab structures. uma_find_refcnt()
does a kextract to fetch the slab struct reference from
the underlying page, and lookup the corresponding refcnt.
mbuma things worth noting:
- integrates mbuf & cluster allocations with extended UMA
and provides caches for commonly-allocated items; defines
several zones (two primary, one secondary) and two kegs.
- change up certain code paths that always used to do:
m_get() + m_clget() to instead just use m_getcl() and
try to take advantage of the newly defined secondary
Packet zone.
- netstat(1) and systat(1) quickly hacked up to do basic
stat reporting but additional stats work needs to be
done once some other details within UMA have been taken
care of and it becomes clearer to how stats will work
within the modified framework.
From the user perspective, one implication is that the
NMBCLUSTERS compile-time option is no longer used. The
maximum number of clusters is still capped off according
to maxusers, but it can be made unlimited by setting
the kern.ipc.nmbclusters boot-time tunable to zero.
Work should be done to write an appropriate sysctl
handler allowing dynamic tuning of kern.ipc.nmbclusters
at runtime.
Additional things worth noting/known issues (READ):
- One report of 'ips' (ServeRAID) driver acting really
slow in conjunction with mbuma. Need more data.
Latest report is that ips is equally sucking with
and without mbuma.
- Giant leak in NFS code sometimes occurs, can't
reproduce but currently analyzing; brueffer is
able to reproduce but THIS IS NOT an mbuma-specific
problem and currently occurs even WITHOUT mbuma.
- Issues in network locking: there is at least one
code path in the rip code where one or more locks
are acquired and we end up in m_prepend() with
M_WAITOK, which causes WITNESS to whine from within
UMA. Current temporary solution: force all UMA
allocations to be M_NOWAIT from within UMA for now
to avoid deadlocks unless WITNESS is defined and we
can determine with certainty that we're not holding
any locks when we're M_WAITOK.
- I've seen at least one weird socketbuffer empty-but-
mbuf-still-attached panic. I don't believe this
to be related to mbuma but please keep your eyes
open, turn on debugging, and capture crash dumps.
This change removes more code than it adds.
A paper is available detailing the change and considering
various performance issues, it was presented at BSDCan2004:
http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf
Please read the paper for Future Work and implementation
details, as well as credits.
Testing and Debugging:
rwatson,
brueffer,
Ketrien I. Saihr-Kesenchedra,
...
Reviewed by: Lots of people (for different parts)
