(time grows downward)
thread 1 thread 2
------------|------------
dec ref_cnt |
| dec ref_cnt <-- ref_cnt now zero
cmpset |
free all |
return |
|
alloc again,|
reuse prev |
ref_cnt |
| cmpset, read
| already freed
| ref_cnt
------------|------------
This should fix that by performing only a single
atomic test-and-set that will serve to decrement
the ref_cnt, only if it hasn't changed since the
earlier read, otherwise it'll loop and re-read.
This forces ordering of decrements so that truly
the thread which did the LAST decrement is the
one that frees.
This is how atomic-instruction-based refcnting
should probably be handled.
Submitted by: Julian Elischer
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)
packet along with data, instead of in their own packet. When serving files
of size (packetsize - headersize) or smaller, this will result in one less
packet crossing the network. Quick testing with thttpd and http_load has
shown a noticeable performance improvement in this case (350 vs 330 fetches
per second.)
Included in this commit are two support routines, iov_to_uio, and m_uiotombuf;
these routines are used by sendfile to construct the header mbuf chain that
will be linked to the rest of the data in the socket buffer.
Changes from the original implementation:
- Fragmentation is handled by the function m_fragment, which can
be called from whereever fragmentation is needed. Note that this
function is wrapped in #ifdef MBUF_STRESS_TEST to discourage non-testing
use.
- m_fragment works slightly differently from the old fragmentation
code in that it allocates a seperate mbuf cluster for each fragment.
This defeats dma_map_load_mbuf/buffer's feature of coalescing adjacent
fragments. While that is a nice feature in practice, it nerfed the
usefulness of mbuf_stress_test.
- Add two modes of random fragmentation. Chains with fragments all of
the same random length and chains with fragments that are each uniquely
random in length may now be requested.
- Make m_prepend use m_gethdr instead of m_get where
appropriate
- Make m_copym use m_gethdr instead of m_get where
appropriate
- Add a call to m_fixhdr in m_defrag; m_defrag can't
deal with corrupted pkthdr.len counts.
MFC after: 3 days
When enabled, this causes m_defrag to randomly return NULL (following
its normal failure case so that extra memory leaks are not introduced.)
Code similar to this was used to find / fix a few bugs last week.
returning some additional room in the first mbuf in a chain, and
avoiding feature-specific contents in the mbuf header. To do this:
- Modify mbuf_to_label() to extract the tag, returning NULL if not
found.
- Introduce mac_init_mbuf_tag() which does most of the work
mac_init_mbuf() used to do, except on an m_tag rather than an
mbuf.
- Scale back mac_init_mbuf() to perform m_tag allocation and invoke
mac_init_mbuf_tag().
- Replace mac_destroy_mbuf() with mac_destroy_mbuf_tag(), since
m_tag's are now GC'd deep in the m_tag/mbuf code rather than
at a higher level when mbufs are directly free()'d.
- Add mac_copy_mbuf_tag() to support m_copy_pkthdr() and related
notions.
- Generally change all references to mbuf labels so that they use
mbuf_to_label() rather than &mbuf->m_pkthdr.label. This
required no changes in the MAC policies (yay!).
- Tweak mbuf release routines to not call mac_destroy_mbuf(),
tag destruction takes care of it for us now.
- Remove MAC magic from m_copy_pkthdr() and m_move_pkthdr() --
the existing m_tag support does all this for us. Note that
we can no longer just zero the m_tag list on the target mbuf,
rather, we have to delete the chain because m_tag's will
already be hung off freshly allocated mbuf's.
- Tweak m_tag copying routines so that if we're copying a MAC
m_tag, we don't do a binary copy, rather, we initialize the
new storage and do a deep copy of the label.
- Remove use of MAC_FLAG_INITIALIZED in a few bizarre places
having to do with mbuf header copies previously.
- When an mbuf is copied in ip_input(), we no longer need to
explicitly copy the label because it will get handled by the
m_tag code now.
- No longer any weird handling of MAC labels in if_loop.c during
header copies.
- Add MPC_LOADTIME_FLAG_LABELMBUFS flag to Biba, MLS, mac_test.
In mac_test, handle the label==NULL case, since it can be
dynamically loaded.
In order to improve performance with this change, introduce the notion
of "lazy MAC label allocation" -- only allocate m_tag storage for MAC
labels if we're running with a policy that uses MAC labels on mbufs.
Policies declare this intent by setting the MPC_LOADTIME_FLAG_LABELMBUFS
flag in their load-time flags field during declaration. Note: this
opens up the possibility of post-boot policy modules getting back NULL
slot entries even though they have policy invariants of non-NULL slot
entries, as the policy might have been loaded after the mbuf was
allocated, leaving the mbuf without label storage. Policies that cannot
handle this case must be declared as NOTLATE, or must be modified.
- mac_labelmbufs holds the current cumulative status as to whether
any policies require mbuf labeling or not. This is updated whenever
the active policy set changes by the function mac_policy_updateflags().
The function iterates the list and checks whether any have the
flag set. Write access to this variable is protected by the policy
list; read access is currently not protected for performance reasons.
This might change if it causes problems.
- Add MAC_POLICY_LIST_ASSERT_EXCLUSIVE() to permit the flags update
function to assert appropriate locks.
- This makes allocation in mac_init_mbuf() conditional on the flag.
Reviewed by: sam
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
of asserting that an mbuf has a packet header. Use it instead of hand-
rolled versions wherever applicable.
Submitted by: Hiten Pandya <hiten@unixdaemons.com>
were sometimes propagated using M_COPY_PKTHDR which actually did
something between a "move" and a "copy" operation. This is replaced
by M_MOVE_PKTHDR (which copies the pkthdr contents and "removes" it
from the source mbuf) and m_dup_pkthdr which copies the packet
header contents including any m_tag chain. This corrects numerous
problems whereby mbuf tags could be lost during packet manipulations.
These changes also introduce arguments to m_tag_copy and m_tag_copy_chain
to specify if the tag copy work should potentially block. This
introduces an incompatibility with openbsd which we may want to revisit.
Note that move/dup of packet headers does not handle target mbufs
that have a cluster bound to them. We may want to support this;
for now we watch for it with an assert.
Finally, M_COPYFLAGS was updated to include M_FIRSTFRAG|M_LASTFRAG.
Supported by: Vernier Networks
Reviewed by: Robert Watson <rwatson@FreeBSD.org>
o instead of a list of mbufs use a list of m_tag structures a la openbsd
o for netgraph et. al. extend the stock openbsd m_tag to include a 32-bit
ABI/module number cookie
o for openbsd compatibility define a well-known cookie MTAG_ABI_COMPAT and
use this in defining openbsd-compatible m_tag_find and m_tag_get routines
o rewrite KAME use of aux mbufs in terms of packet tags
o eliminate the most heavily used aux mbufs by adding an additional struct
inpcb parameter to ip_output and ip6_output to allow the IPsec code to
locate the security policy to apply to outbound packets
o bump __FreeBSD_version so code can be conditionalized
o fixup ipfilter's call to ip_output based on __FreeBSD_version
Reviewed by: julian, luigi (silent), -arch, -net, darren
Approved by: julian, silence from everyone else
Obtained from: openbsd (mostly)
MFC after: 1 month
header and return that length, was misguided.
The check itself didn't take into account the fact that the
mbuf pointer pased in may be null, and the function is
defined specifically for cases where the caller knows what it wants.
Rather than fix the check I'm removing it as phk suggested.
Submitted by: phk@freebsd.org
kernel access control.
Invoke additional MAC entry points when an mbuf packet header is
copied to another mbuf: release the old label if any, reinitialize
the new header, and ask the MAC framework to copy the header label
data. Note that this requires a potential allocation operation,
but m_copy_pkthdr() is not permitted to fail, so we must block.
Since we now use interrupt threads, this is possible, but not
desirable.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, NAI Labs
function. This permits conditionally compiled extensions to the
packet header copying semantic, such as extensions to copy MAC
labels.
Reviewed by: bmilekic
Obtained from: TrustedBSD Project
Sponsored by: DARPA, NAI Labs
of the inlines, like its cousin, m_free(). Also, make a small (first
step?) optimisation of m_free() to use the MBP_PERSIST{,ENT} interface
to hold the lock across frees when possible. The thing is that right
now, we can only do this easily for at most across one mbuf + one
cluster free, as the comment mentions (it also explains why). Anyway,
some basic tests revealed a 5-10% overall improvement. Some of the
results can be found here:
http://people.freebsd.org/~bmilekic/code/measure.txt
and a cluster in one shot.
o Introduce MBP_PERSIST and MBP_PERSISTENT control bits to mb_alloc();
MBP_PERSIST means "if you can allocate, then keep the cache lock
held on exit," and MBP_PERSISTENT means "a cache lock is alredy held
on entry, so allocate from the specified (already locked) cache."
They may be used in combination.
o m_getcl() uses the MBP_PERSIST/MBP_PERSISTENT interface so that it
doesn't drop the cache lock in between the mbuf and cluster allocations.
o m_getm(), which takes a size and allocates an mbuf + cluster "best fit"
chain, has been moved from uipc_mbuf.c to subr_mbuf.c and shown how to
use MBP_PERSIST/MBP_PERSISTENT to attempt to do a grouped allocation
without dropping the cache lock in between.
Why this is good: much less bus-locked lock acquires/drops when they're
not needed. Also, prototype for m_getcl():
struct mbuf * m_getcl(int how, short type, int flags);
"how" and "type" are self-explanatory. "flags" may be M_PKTHDR, in
which case m_getcl() will make the mbuf a pkthdr-mbuf.
While I'm in subr_mbuf.c:
o Every exported routine now has a nice comment with a description of
the expected arguments. Eventually, mbuf(9) needs to be re-vamped
but there's still more code to write/finalize before I get to that.
o internal macros have been changed a bit.
o consistently use 'short' for "type." This somehow slipped through
before (that 'type' was sometimes declared as int).
Alfred has been pushing for the MBP_PERSIST{,ENT} thing for almost a
year now. Luigi asked for m_getcl(), and will probably MFC that
part of this commit.
TODO [Related]: teach mb_free() about MBP_PERSIST{, ENT}.
set to zero. This field indicates the total space in the external buffer
and therefore should not be modified after the external buffer is added.
Add a comment warning that the mbufs returned by m_split() might be read-only.
Fix M_TRAILINGSPACE() to return zero if !M_WRITABLE(m).
Reviewed by: freebsd-net
Obtained from: Vernier Networks, Inc.
MFC after: 1 week
introduce a modified allocation mechanism for mbufs and mbuf clusters; one
which can scale under SMP and which offers the possibility of resource
reclamation to be implemented in the future. Notable advantages:
o Reduce contention for SMP by offering per-CPU pools and locks.
o Better use of data cache due to per-CPU pools.
o Much less code cache pollution due to excessively large allocation macros.
o Framework for `grouping' objects from same page together so as to be able
to possibly free wired-down pages back to the system if they are no longer
needed by the network stacks.
Additional things changed with this addition:
- Moved some mbuf specific declarations and initializations from
sys/conf/param.c into mbuf-specific code where they belong.
- m_getclr() has been renamed to m_get_clrd() because the old name is really
confusing. m_getclr() HAS been preserved though and is defined to the new
name. No tree sweep has been done "to change the interface," as the old
name will continue to be supported and is not depracated. The change was
merely done because m_getclr() sounds too much like "m_get a cluster."
- TEMPORARILY disabled mbtypes statistics displaying in netstat(1) and
systat(1) (see TODO below).
- Fixed systat(1) to display number of "free mbufs" based on new per-CPU
stat structures.
- Fixed netstat(1) to display new per-CPU stats based on sysctl-exported
per-CPU stat structures. All infos are fetched via sysctl.
TODO (in order of priority):
- Re-enable mbtypes statistics in both netstat(1) and systat(1) after
introducing an SMP friendly way to collect the mbtypes stats under the
already introduced per-CPU locks (i.e. hopefully don't use atomic() - it
seems too costly for a mere stat update, especially when other locks are
already present).
- Optionally have systat(1) display not only "total free mbufs" but also
"total free mbufs per CPU pool."
- Fix minor length-fetching issues in netstat(1) related to recently
re-enabled option to read mbuf stats from a core file.
- Move reference counters at least for mbuf clusters into an unused portion
of the cluster itself, to save space and need to allocate a counter.
- Look into introducing resource freeing possibly from a kproc.
Reviewed by (in parts): jlemon, jake, silby, terry
Tested by: jlemon (Intel & Alpha), mjacob (Intel & Alpha)
Preliminary performance measurements: jlemon (and me, obviously)
URL: http://people.freebsd.org/~bmilekic/mb_alloc/
something: offset into the first mbuf of the target chain before copying
the source data over.
Make drivers using m_devget() with a first argument "data - ETHER_ALIGN"
to use the offset argument to pass ETHER_ALIGN in. The way it was previously
done is potentially dangerous if the source data was at the top of a page
and the offset caused the previous page to be copied (if the
previous page has not yet been appropriately mapped).
The old `offset' argument in m_devget() is not used anywhere (it's always
0) and dates back to ~1995 (and earlier?) when support for ethernet trailers
existed. With that support gone, it was merely collecting dust.
Tested on alpha by: jlemon
Partially submitted by: jlemon
Reviewed by: jlemon
MFC after: 3 weeks
This work was based on kame-20010528-freebsd43-snap.tgz and some
critical problem after the snap was out were fixed.
There are many many changes since last KAME merge.
TODO:
- The definitions of SADB_* in sys/net/pfkeyv2.h are still different
from RFC2407/IANA assignment because of binary compatibility
issue. It should be fixed under 5-CURRENT.
- ip6po_m member of struct ip6_pktopts is no longer used. But, it
is still there because of binary compatibility issue. It should
be removed under 5-CURRENT.
Reviewed by: itojun
Obtained from: KAME
MFC after: 3 weeks
around, use a common function for looking up and extracting the tunables
from the kernel environment. This saves duplicating the same function
over and over again. This way typically has an overhead of 8 bytes + the
path string, versus about 26 bytes + the path string.
other "system" header files.
Also help the deprecation of lockmgr.h by making it a sub-include of
sys/lock.h and removing sys/lockmgr.h form kernel .c files.
Sort sys/*.h includes where possible in affected files.
OK'ed by: bde (with reservations)
we also reserve _adequate_ space for the mb_map submap; i.e. we need
space for nmbclusters, nmbufs, _and_ nmbcnt. Furthermore, we need to
rounddown, and not roundup, so that we are consistent.
Pointed out by: bde
The mbuf and mcluster free lists now each "own" a condition variable,
m_starved.
- Clean up minor indentention issues in sys/mbuf.h caused by previous
commit.
Don't use atomic operations for the stats updating, instead protect
the counts with the mbuf mutex. Most twiddling of the stats was
done right before or after releasing a mutex. By doing this we
reduce the number of locked ops needed as well as allow a sysctl
to gain a consitant view of the entire stats structure.
In the future...
This will allow us to chain common mbuf operations that would
normally need to aquire/release 2 or 3 of the locks to build an
mbuf with a cluster or external data attached into a single op
requiring only one lock.
Simplify the per-cpu locks that are planned.
There's also some if (1) code that should check if the "how"
operation specifies blocking/non-blocking behavior, we _could_ make
it so that we hold onto the mutex through calls into kmem_alloc
when non-blocking requests are made, but for safety reasons we
currently drop and reaquire the mutex around the calls.
Also, note that calling kmem_alloc is rare and only happens during
a shortage so drop/re-getting the mutex will not be a common
occurance.
Remove some #define's that seemed to obfuscate the code to me.
Remove an extranious comment.
Remove an XXX, including mutex.h isn't a crime.
Reviewed by: bmilekic
MCLGET macros in order to avoid incrementing the drop count twice.
Otherwise, in some cases, we may increment m_drops once in m_mballoc()
for example, and increment it again in m_mballoc_wait() if the
wait fails.
- Make sure that m_mballoc() really doesn't allow over nmbufs mbufs to
be allocated from mb_map. In the case where nmbufs-reserved space is not
an exact multiple of PAGE_SIZE (which it should be, but anyway...), we
hold nmbufs as an absolute maximum which need not ever be reached.
- Clean up m_clalloc(); make it more consistent in the sense that the first
argument `ncl' really means "the number of clusters ensured to be allocated"
and not "the number of pages worth of clusters to be allocated," as was
previously the case. This also makes it consistent with m_mballoc() as well
as the comment that preceeds it.
Reviewed by: jlemon
This is useful when doing copies of packet where some leading
space has been preallocated to insert protocol headers.
Note that there are in fact almost no users of m_copypacket.
MFC candidate.
allocation, as required.
If m_getm() receives NULL as a first argument, then it allocates `len'
(second argument) bytes worth of mbufs + clusters and returns the chain
only if it was able to allocate everything.
If the first argument is non-NULL, then it should be an existing mbuf
chain (e.g. pre-allocated mbuf sitting on a ring, on some list, etc.) and
so it will allocate `len' bytes worth of clusters and mbufs, as needed,
and append them to the tail of the passed in chain, only if it was able
to allocate everything requested.
If allocation fails, only what was allocated by the routine will be freed,
and NULL will be returned.
Also, get rid of existing m_getm() in netncp code and replace calls to it
to calls to this new generic code.
Heavily Reviewed by: bp
and function argument declarations. Make sure that functions that are
supposed to return a pointer return NULL in case of failure. Don't cast
NULL. Finally, get rid of annoying `register' uses.
mtx_enter(lock, type) becomes:
mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks)
mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized)
similarily, for releasing a lock, we now have:
mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN.
We change the caller interface for the two different types of locks
because the semantics are entirely different for each case, and this
makes it explicitly clear and, at the same time, it rids us of the
extra `type' argument.
The enter->lock and exit->unlock change has been made with the idea
that we're "locking data" and not "entering locked code" in mind.
Further, remove all additional "flags" previously passed to the
lock acquire/release routines with the exception of two:
MTX_QUIET and MTX_NOSWITCH
The functionality of these flags is preserved and they can be passed
to the lock/unlock routines by calling the corresponding wrappers:
mtx_{lock, unlock}_flags(lock, flag(s)) and
mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN
locks, respectively.
Re-inline some lock acq/rel code; in the sleep lock case, we only
inline the _obtain_lock()s in order to ensure that the inlined code
fits into a cache line. In the spin lock case, we inline recursion and
actually only perform a function call if we need to spin. This change
has been made with the idea that we generally tend to avoid spin locks
and that also the spin locks that we do have and are heavily used
(i.e. sched_lock) do recurse, and therefore in an effort to reduce
function call overhead for some architectures (such as alpha), we
inline recursion for this case.
Create a new malloc type for the witness code and retire from using
the M_DEV type. The new type is called M_WITNESS and is only declared
if WITNESS is enabled.
Begin cleaning up some machdep/mutex.h code - specifically updated the
"optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN
and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently
need those.
Finally, caught up to the interface changes in all sys code.
Contributors: jake, jhb, jasone (in no particular order)
kmem_free() for now. Kmem_malloc() and kmem_free() now have appropriate
assertions in place, and these checks aren't feasible until more of the
networking code is locked down. Also, the extra assertions here should
already be caught by the WITNESS code as lock order violations should
mutex operations on Giant be reintroduced here later.
The counters are incremented when a thread goes to sleep and decremented
either when a thread is woken up by another thread or when the sleep
times out. There existed a race where the sleep count could be decremented
twice resulting in an eventual underflow.
Move the decrementing of the "counters" to the thread initiating the sleep
and thus remedy the problem.
allocation routines are being called safely. Since we drop our relevant
mbuf mutex and acquire Giant before we call kmem_malloc(), we have
to make sure that this does not pave the way for a fatal lock order
reversal. Check that either Giant is already held (in which case it's safe
to grab it again and recurse on it) or, if Giant is not held, that no
other locks are held before we try to acquire Giant.
Similarily, add a KASSERT valid in the WITNESS case in m_reclaim() to
nail callers who end up in m_reclaim() and hold a lock.
Pointed out by: jhb
m_reclaim() and re-acquire it when m_reclaim() returns. This means that
we now call the drain routines without holding the mutex lock and
recursing into it. This was done for mainly two reasons:
(i) Avoid the long recursion; long recursions are typically bad and this
is the case here because we block all other code from freeing mbufs
if they need to. Doing that is kind of counter-productive, since we're
really hoping that someone will free.
(ii) More importantly, avoid a potential lock order reversal. Right now,
not all the locks have been added to our networking code; but
without this change, we're introducing the possibility for deadlock.
Consider for example ip_drain(). We will likely eventually introduce
a lock for ipq there, and so ip_freef() will be called with ipq lock
held. But, ip_freef() calls m_freem() which in turn acquires the
mmbfree lock. Since we were previously calling ip_drain() with mmbfree
held, our lock order would be: mmbfree->ipq->mmbfree. Some other code
may very well lock ipq first and then call ip_freef(). This would
result in the regular lock order, ipq->mmbfree. Clearly, we have
deadlock if one thread acquires the ipq lock and sits waiting for
mmbfree while another thread calling m_reclaim() acquires mmbfree
and sits waiting for the ipq lock.
Also, make sure to add a comment above m_reclaim()'s definition briefly
explaining this. Also document this above the call to m_reclaim() in
m_mballoc_wait().
Suggested and reviewed by: alfred
This is because calls with M_WAIT (now M_TRYWAIT) may not wait
forever when nothing is available for allocation, and may end up
returning NULL. Hopefully we now communicate more of the right thing
to developers and make it very clear that it's necessary to check whether
calls with M_(TRY)WAIT also resulted in a failed allocation.
M_TRYWAIT basically means "try harder, block if necessary, but don't
necessarily wait forever." The time spent blocking is tunable with
the kern.ipc.mbuf_wait sysctl.
M_WAIT is now deprecated but still defined for the next little while.
* Fix a typo in a comment in mbuf.h
* Fix some code that was actually passing the mbuf subsystem's M_WAIT to
malloc(). Made it pass M_WAITOK instead. If we were ever to redefine the
value of the M_WAIT flag, this could have became a big problem.
number of ext_buf counters that are possibly allocatable.
Do this because:
(i) It will make it easier to influence EXT_COUNTERS for if_sk,
if_ti (or similar) users where the driver allocates its own
ext_bufs and where it is important for the mbuf system to take
it into account when reserving necessary space for counters.
(ii) Facilitate some percentile calculation for netstat(1)
to accomodate the changes.
Here's a list of things that have changed (I may have left out a few); for a
relatively complete list, see http://people.freebsd.org/~bmilekic/mtx_journal
* Remove old (once useful) mcluster code for MCLBYTES > PAGE_SIZE which
nobody uses anymore. It was great while it lasted, but now we're moving
onto bigger and better things (Approved by: wollman).
* Practically re-wrote the allocation macros in sys/sys/mbuf.h to accomodate
new allocations which grab the necessary lock.
* Make sure that necessary mbstat variables are manipulated with
corresponding atomic() routines.
* Changed the "wait" routines, cleaned it up, made one routine that does
the job.
* Generalized MWAKEUP() macro. Got rid of m_retry and m_retryhdr, as they
are now included in the generalized "wait" routines.
* Sleep routines now use msleep().
* Free lists have locks.
* etc... probably other stuff I'm missing...
Things to look out for and work on later:
* find a better way to (dynamically) adjust EXT_COUNTERS
* move necessity to recurse on a lock from drain routines by providing
lock-free lower-level version of MFREE() (and possibly m_free()?).
* checkout include of mutex.h in sys/sys/mbuf.h - probably violating
general philosophy here.
The code has been reviewed quite a bit, but problems may arise... please,
don't panic! Send me Emails: bmilekic@freebsd.org
Reviewed by: jlemon, cp, alfred, others?
that should be better.
The old code counted references to mbuf clusters by using the offset
of the cluster from the start of memory allocated for mbufs and
clusters as an index into an array of chars, which did the reference
counting. If the external storage was not a cluster then reference
counting had to be done by the code using that external storage.
NetBSD's system of linked lists of mbufs was cosidered, but Alfred
felt it would have locking issues when the kernel was made more
SMP friendly.
The system implimented uses a pool of unions to track external
storage. The union contains an int for counting the references and
a pointer for forming a free list. The reference counts are
incremented and decremented atomically and so should be SMP friendly.
This system can track reference counts for any sort of external
storage.
Access to the reference counting stuff is now through macros defined
in mbuf.h, so it should be easier to make changes to the system in
the future.
The possibility of storing the reference count in one of the
referencing mbufs was considered, but was rejected 'cos it would
often leave extra mbufs allocated. Storing the reference count in
the cluster was also considered, but because the external storage
may not be a cluster this isn't an option.
The size of the pool of reference counters is available in the
stats provided by "netstat -m".
PR: 19866
Submitted by: Bosko Milekic <bmilekic@dsuper.net>
Reviewed by: alfred (glanced at by others on -net)
The variables "m_mclalloc_wid" and "m_mballoc_wid" were not in the
proper place. They should have been in uipc_mbuf.c and have been global,
not in mbuf.h and local per each file that uses mbuf.h.
Sorta bug fix:
In mbuf.h, the definitions of various things for KERNEL and not
KERNEL cases were very screwy. This fixes all of that which I could
find.
means that running out of mbuf space isn't a panic anymore, and code
which runs out of network memory will sleep to wait for it.
Submitted by: Bosko Milekic <bmilekic@dsuper.net>
Reviewed by: green, wollman
because in the case of mbuf clusters they only increment the reference
count rather than actually copying the data.
Add comments to this effect, and add a new routine called m_dup() that
returns a real, writable copy of an mbuf chain.
This is preliminary work required for implementing 'ipfw tee'.
Reviewed by: julian
into uipc_mbuf.c. This reduces three sets of identical tunable code to
one set, and puts the initialisation with the mbuf code proper.
Make NMBUFs tunable as well.
Move the nmbclusters sysctl here as well.
Move the initialisation of maxsockets from param.c to uipc_socket2.c,
next to its corresponding sysctl.
Use the new tunable macros for the kern.vm.kmem.size tunable (this should have
been in a separate commit, whoops).
SYSINIT_KT() etc (which is a static, compile-time procedure), use a
NetBSD-style kthread_create() interface. kproc_start is still available
as a SYSINIT() hook. This allowed simplification of chunks of the
sysinit code in the process. This kthread_create() is our old kproc_start
internals, with the SYSINIT_KT fork hooks grafted in and tweaked to work
the same as the NetBSD one.
One thing I'd like to do shortly is get rid of nfsiod as a user initiated
process. It makes sense for the nfs client code to create them on the
fly as needed up to a user settable limit. This means that nfsiod
doesn't need to be in /sbin and is always "available". This is a fair bit
easier to do outside of the SYSINIT_KT() framework.
This makes it possible to change the sysctl tree at runtime.
* Change KLD to find and register any sysctl nodes contained in the loaded
file and to unregister them when the file is unloaded.
Reviewed by: Archie Cobbs <archie@whistle.com>,
Peter Wemm <peter@netplex.com.au> (well they looked at it anyway)
the alloc is not M_DONTWAIT, then panic with "Out of mbuf clusters".
Callers that specify M_WAIT can't deal with getting a NULL buffer, so this
is a more graceful failure than randomly page faulting in the socket code
or elsewhere.
here, but kmem_malloc() is used and it takes the same "flags" as
malloc().
Use the mbuf allocation "flags" M_WAIT and M_DONTWAIT consistently.
There is really only one boolean flag, M_DONTWAIT, but the "flags"
were always treated as enum-like values, except in some places here
where the values are tacitly converted to boolean flags. Treat
them as enum-like values everywhere, except where we tacitly assume
that there are only two values in order to convert them to the
corresponding two kmem_malloc() "flags".