Change 221669 by bz@bz_zenith on 2013/02/01 12:26:04
Run the initialization for polling earlier along with INTRs
so that we can put network interface into polling mode by default
if DEVICE_POLLING is compiled in and no interrupts are available.
MFC after: 3 days
Sponsored by: DARPA/AFRL
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
user. Kqueue now saves the ucred of the allocating thread, to
correctly decrement the counter on close.
Under some specific and not real-world use scenario for kqueue, it is
possible for the kqueues to consume memory proportional to the square
of the number of the filedescriptors available to the process. Limit
allows administrator to prevent the abuse.
This is kernel-mode side of the change, with the user-mode enabling
commit following.
Reported and tested by: pho
Discussed with: jmg
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Restore BIO_UNMAPPED and BIO_TRANSIENT_MAPPING in biodonne() when unmapping
temporary mapped buffer. That fixes double unmap if biodone() called twice
for the same BIO (but with different done methods).
Move mapping removal before calling bio_done() method. I believe that it is
very wrong to do anything to BIO after reporting completion. kib@ thinks
it was done for some forgotten now case when bio_done() method needed mapped
buffer. But 1) if BIO was sent as unmapped, then IMO done() should be called
in the same way; 2) IMO there is no guatantee that buffer will be mapped at
this point at all, for example, if all underlying stack supports unmapped
I/O, so bio_done() handler can not expect that.
- Take BIO lock in biodone() only when there is no completion callback set
and so we should wake up thread waiting in biowait().
- Remove msleep() timeout from biowait(). It was added 11 years ago, when
there was no locks used, and it should not be needed any more.
Move tq_enqueue() call out of the queue lock for known handlers (actually
I have found no others in the base system). This reduces queue lock hold
time and congestion spinning under active multithreaded enqueuing.
Introduce new function devstat_end_transaction_bio_bt(), adding new argument
to specify present time. Use this function to move binuptime() out of lock,
substantially reducing lock congestion when slow timecounter is used.
images compiled on the world with higher major version number than the
high version number of the booted kernel. Default to disable.
Sponsored by: The FreeBSD Foundation
Discussed with: bapt
MFC after: 1 week
devfs_iosize_max_clamp sysctl, which allows/disables SSIZE_MAX-sized
i/o requests on the devfs files.
Sponsored by: The FreeBSD Foundation
Reminded by: Dmitry Sivachenko <trtrmitya@gmail.com>
MFC after: 1 week
Refactor of /dev/random device. Main points include:
* Userland seeding is no longer used. This auto-seeds at boot time
on PC/Desktop setups; this may need some tweeking and intelligence
from those folks setting up embedded boxes, but the work is believed
to be minimal.
* An entropy cache is written to /entropy (even during installation)
and the kernel uses this at next boot.
* An entropy file written to /boot/entropy can be loaded by loader(8)
* Hardware sources such as rdrand are fed into Yarrow, and are no
longer available raw.
------------------------------------------------------------------------
r256240 | des | 2013-10-09 21:14:16 +0100 (Wed, 09 Oct 2013) | 4 lines
Add a RANDOM_RWFILE option and hide the entropy cache code behind it.
Rename YARROW_RNG and FORTUNA_RNG to RANDOM_YARROW and RANDOM_FORTUNA.
Add the RANDOM_* options to LINT.
------------------------------------------------------------------------
r256239 | des | 2013-10-09 21:12:59 +0100 (Wed, 09 Oct 2013) | 2 lines
Define RANDOM_PURE_RNDTEST for rndtest(4).
------------------------------------------------------------------------
r256204 | des | 2013-10-09 18:51:38 +0100 (Wed, 09 Oct 2013) | 2 lines
staticize struct random_hardware_source
------------------------------------------------------------------------
r256203 | markm | 2013-10-09 18:50:36 +0100 (Wed, 09 Oct 2013) | 2 lines
Wrap some policy-rich code in 'if NOTYET' until we can thresh out
what it really needs to do.
------------------------------------------------------------------------
r256184 | des | 2013-10-09 10:13:12 +0100 (Wed, 09 Oct 2013) | 2 lines
Re-add /dev/urandom for compatibility purposes.
------------------------------------------------------------------------
r256182 | des | 2013-10-09 10:11:14 +0100 (Wed, 09 Oct 2013) | 3 lines
Add missing include guards and move the existing ones out of the
implementation namespace.
------------------------------------------------------------------------
r256168 | markm | 2013-10-08 23:14:07 +0100 (Tue, 08 Oct 2013) | 10 lines
Fix some just-noticed problems:
o Allow this to work with "nodevice random" by fixing where the
MALLOC pool is defined.
o Fix the explicit reseed code. This was correct as submitted, but
in the project branch doesn't need to set the "seeded" bit as this
is done correctly in the "unblock" function.
o Remove some debug ifdeffing.
o Adjust comments.
------------------------------------------------------------------------
r256159 | markm | 2013-10-08 19:48:11 +0100 (Tue, 08 Oct 2013) | 6 lines
Time to eat crow for me.
I replaced the sx_* locks that Arthur used with regular mutexes;
this turned out the be the wrong thing to do as the locks need to
be sleepable. Revert this folly.
# Submitted by: Arthur Mesh <arthurmesh@gmail.com> (In original diff)
------------------------------------------------------------------------
r256138 | des | 2013-10-08 12:05:26 +0100 (Tue, 08 Oct 2013) | 10 lines
Add YARROW_RNG and FORTUNA_RNG to sys/conf/options.
Add a SYSINIT that forces a reseed during proc0 setup, which happens
fairly late in the boot process.
Add a RANDOM_DEBUG option which enables some debugging printf()s.
Add a new RANDOM_ATTACH entropy source which harvests entropy from the
get_cyclecount() delta across each call to a device attach method.
------------------------------------------------------------------------
r256135 | markm | 2013-10-08 07:54:52 +0100 (Tue, 08 Oct 2013) | 8 lines
Debugging. My attempt at EVENTHANDLER(multiuser) was a failure; use
EVENTHANDLER(mountroot) instead.
This means we can't count on /var being present, so something will
need to be done about harvesting /var/db/entropy/... .
Some policy now needs to be sorted out, and a pre-sync cache needs
to be written, but apart from that we are now ready to go.
Over to review.
------------------------------------------------------------------------
r256094 | markm | 2013-10-06 23:45:02 +0100 (Sun, 06 Oct 2013) | 8 lines
Snapshot.
Looking pretty good; this mostly works now. New code includes:
* Read cached entropy at startup, both from files and from loader(8)
preloaded entropy. Failures are soft, but announced. Untested.
* Use EVENTHANDLER to do above just before we go multiuser. Untested.
------------------------------------------------------------------------
r256088 | markm | 2013-10-06 14:01:42 +0100 (Sun, 06 Oct 2013) | 2 lines
Fix up the man page for random(4). This mainly removes no-longer-relevant
details about HW RNGs, reseeding explicitly and user-supplied
entropy.
------------------------------------------------------------------------
r256087 | markm | 2013-10-06 13:43:42 +0100 (Sun, 06 Oct 2013) | 6 lines
As userland writing to /dev/random is no more, remove the "better
than nothing" bootstrap mode.
Add SWI harvesting to the mix.
My box seeds Yarrow by itself in a few seconds! YMMV; more to follow.
------------------------------------------------------------------------
r256086 | markm | 2013-10-06 13:40:32 +0100 (Sun, 06 Oct 2013) | 11 lines
Debug run. This now works, except that the "live" sources haven't
been tested. With all sources turned on, this unlocks itself in
a couple of seconds! That is no my box, and there is no guarantee
that this will be the case everywhere.
* Cut debug prints.
* Use the same locks/mutexes all the way through.
* Be a tad more conservative about entropy estimates.
------------------------------------------------------------------------
r256084 | markm | 2013-10-06 13:35:29 +0100 (Sun, 06 Oct 2013) | 5 lines
Don't use the "real" assembler mnemonics; older compilers may not
understand them (like when building CURRENT on 9.x).
# Submitted by: Konstantin Belousov <kostikbel@gmail.com>
------------------------------------------------------------------------
r256081 | markm | 2013-10-06 10:55:28 +0100 (Sun, 06 Oct 2013) | 12 lines
SNAPSHOT.
Simplify the malloc pools; We only need one for this device.
Simplify the harvest queue.
Marginally improve the entropy pool hashing, making it a bit faster
in the process.
Connect up the hardware "live" source harvesting. This is simplistic
for now, and will need to be made rate-adaptive.
All of the above passes a compile test but needs to be debugged.
------------------------------------------------------------------------
r256042 | markm | 2013-10-04 07:55:06 +0100 (Fri, 04 Oct 2013) | 25 lines
Snapshot. This passes the build test, but has not yet been finished or debugged.
Contains:
* Refactor the hardware RNG CPU instruction sources to feed into
the software mixer. This is unfinished. The actual harvesting needs
to be sorted out. Modified by me (see below).
* Remove 'frac' parameter from random_harvest(). This was never
used and adds extra code for no good reason.
* Remove device write entropy harvesting. This provided a weak
attack vector, was not very good at bootstrapping the device. To
follow will be a replacement explicit reseed knob.
* Separate out all the RANDOM_PURE sources into separate harvest
entities. This adds some secuity in the case where more than one
is present.
* Review all the code and fix anything obviously messy or inconsistent.
Address som review concerns while I'm here, like rename the pseudo-rng
to 'dummy'.
# Submitted by: Arthur Mesh <arthurmesh@gmail.com> (the first item)
------------------------------------------------------------------------
r255319 | markm | 2013-09-06 18:51:52 +0100 (Fri, 06 Sep 2013) | 4 lines
Yarrow wants entropy estimations to be conservative; the usual idea
is that if you are certain you have N bits of entropy, you declare
N/2.
------------------------------------------------------------------------
r255075 | markm | 2013-08-30 18:47:53 +0100 (Fri, 30 Aug 2013) | 4 lines
Remove short-lived idea; thread to harvest (eg) RDRAND enropy into the
usual harvest queues. It was a nifty idea, but too heavyweight.
# Submitted by: Arthur Mesh <arthurmesh@gmail.com>
------------------------------------------------------------------------
r255071 | markm | 2013-08-30 12:42:57 +0100 (Fri, 30 Aug 2013) | 4 lines
Separate out the Software RNG entropy harvesting queue and thread
into its own files.
# Submitted by: Arthur Mesh <arthurmesh@gmail.com>
------------------------------------------------------------------------
r254934 | markm | 2013-08-26 20:07:03 +0100 (Mon, 26 Aug 2013) | 2 lines
Remove the short-lived namei experiment.
------------------------------------------------------------------------
r254928 | markm | 2013-08-26 19:35:21 +0100 (Mon, 26 Aug 2013) | 2 lines
Snapshot; Do some running repairs on entropy harvesting. More needs
to follow.
------------------------------------------------------------------------
r254927 | markm | 2013-08-26 19:29:51 +0100 (Mon, 26 Aug 2013) | 15 lines
Snapshot of current work;
1) Clean up namespace; only use "Yarrow" where it is Yarrow-specific
or close enough to the Yarrow algorithm. For the rest use a neutral
name.
2) Tidy up headers; put private stuff in private places. More could
be done here.
3) Streamline the hashing/encryption; no need for a 256-bit counter;
128 bits will last for long enough.
There are bits of debug code lying around; these will be removed
at a later stage.
------------------------------------------------------------------------
r254784 | markm | 2013-08-24 14:54:56 +0100 (Sat, 24 Aug 2013) | 39 lines
1) example (partially humorous random_adaptor, that I call "EXAMPLE")
* It's not meant to be used in a real system, it's there to show how
the basics of how to create interfaces for random_adaptors. Perhaps
it should belong in a manual page
2) Move probe.c's functionality in to random_adaptors.c
* rename random_ident_hardware() to random_adaptor_choose()
3) Introduce a new way to choose (or select) random_adaptors via tunable
"rngs_want" It's a list of comma separated names of adaptors, ordered
by preferences. I.e.:
rngs_want="yarrow,rdrand"
Such setting would cause yarrow to be preferred to rdrand. If neither of
them are available (or registered), then system will default to
something reasonable (currently yarrow). If yarrow is not present, then
we fall back to the adaptor that's first on the list of registered
adaptors.
4) Introduce a way where RNGs can play a role of entropy source. This is
mostly useful for HW rngs.
The way I envision this is that every HW RNG will use this
functionality by default. Functionality to disable this is also present.
I have an example of how to use this in random_adaptor_example.c (see
modload event, and init function)
5) fix kern.random.adaptors from
kern.random.adaptors: yarrowpanicblock
to
kern.random.adaptors: yarrow,panic,block
6) add kern.random.active_adaptor to indicate currently selected
adaptor:
root@freebsd04:~ # sysctl kern.random.active_adaptor
kern.random.active_adaptor: yarrow
# Submitted by: Arthur Mesh <arthurmesh@gmail.com>
Submitted by: Dag-Erling Smørgrav <des@FreeBSD.org>, Arthur Mesh <arthurmesh@gmail.com>
Reviewed by: des@FreeBSD.org
Approved by: re (delphij)
Approved by: secteam (des,delphij)
called. This probably should be fixed eventually, but for now it is
not needed to try to flush such vnodes from the buffer allocation
context.
Reported and tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (gjb)
really belong to it. Such vnodes, with the pointers to other vnodes
v_objects, are typically instantiated by the bypass filesystems.
Invalidating mappings of other vnode pages and the pages is wrong,
since reclamation of the upper vnode does not imply that lower vnode
is reclaimed too.
One of the consequences of the improper reclamation was destruction of
the wired mappings of the lower vnode pages, triggering miscellaneous
assertions in the VM system.
Reported by: John Marshall <john.marshall@riverwillow.com.au>
Tested by: John Marshall <john.marshall@riverwillow.com.au>, pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (gjb)
allocated, but the old table is kept around to handle the case of
threads still performing unlocked accesses to it.
Grow the table exponentially instead of increasing its size by
sizeof(long) * 8 chunks when overflowing. This mode significantly
reduces the total memory use for the processes consuming large numbers
of the file descriptors which open them one by one.
Reported and tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (marius)
mbufs, but return chain of free mbufs to a caller. Caller can either reuse
them or return to allocator in a batch manner.
- Implement sbdrop()/sbdrop_locked() as a wrapper around sbcut_internal().
- Expose sbcut_locked() for outside usage.
Sponsored by: Netflix
Sponsored by: Nginx, Inc.
Approved by: re (marius)
Add a SYSINIT that forces a reseed during proc0 setup, which happens
fairly late in the boot process.
Add a RANDOM_DEBUG option which enables some debugging printf()s.
Add a new RANDOM_ATTACH entropy source which harvests entropy from the
get_cyclecount() delta across each call to a device attach method.
This means we can't count on /var being present, so something will need to be done about harvesting /var/db/entropy/... .
Some policy now needs to be sorted out, and a pre-sync cache needs to be written, but apart from that we are now ready to go.
Over to review.
Looking pretty good; this mostly works now. New code includes:
* Read cached entropy at startup, both from files and from loader(8) preloaded entropy. Failures are soft, but announced. Untested.
* Use EVENTHANDLER to do above just before we go multiuser. Untested.
kmeminit() runs, so it contributes nothing to 'vm_kmem_size'; update a
comment to reflect that r254025 replaced the kmem submap with the kmem
arena.
Reviewed by: kib
Approved by: re (gjb)
Sponsored by: EMC / Isilon Storage Division
Contains:
* Refactor the hardware RNG CPU instruction sources to feed into
the software mixer. This is unfinished. The actual harvesting needs
to be sorted out. Modified by me (see below).
* Remove 'frac' parameter from random_harvest(). This was never
used and adds extra code for no good reason.
* Remove device write entropy harvesting. This provided a weak
attack vector, was not very good at bootstrapping the device. To
follow will be a replacement explicit reseed knob.
* Separate out all the RANDOM_PURE sources into separate harvest
entities. This adds some secuity in the case where more than one
is present.
* Review all the code and fix anything obviously messy or inconsistent.
Address som review concerns while I'm here, like rename the pseudo-rng
to 'dummy'.
Submitted by: Arthur Mesh <arthurmesh@gmail.com> (the first item)
not strlen as they are *not* strings.
Discovered by GSOC student, Mike Ma <mikemandarine@gmail.com> during his
fuse.glusterfs port to FreeBSD.
Final patch from mckusick@
Submitted by: mckusick@
Approved by: re (hrs)
MFC after: 2 weeks
is no sense to walk the whole dirty buffer queue. We are only
interested in, and can operate on, the buffers owned by the current
vnode [1]. Instead of calling generic queue flush routine, do
VOP_FSYNC() if possible.
Holding the dirty buffer queue lock in the bufdaemon, without dropping
it, can cause starvation of buffer writes from other threads. This is
esp. easy to reproduce on the big memory machines, where large files
are written, causing almost all dirty buffers accumulating in several
big files, which vnodes are locked by writers. Bufdaemon cannot flush
any buffer, but is iterating over the whole dirty queue
continuously. Since dirty queue mutex is not dropped, bufdone() in
g_up thread is starved, usually deadlocking the machine [2]. Mitigate
this by dropping the queue lock after the vnode is locked, allowing
other queue lock contenders to make a progress.
Discussed with: Jeff [1]
Reported by: pho [2]
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Approved by: re (hrs)
with the vnode shared-locked. If upgrade succeeded, the inactivation
can be done immediately, instead of being postponed.
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (glebius)
The r255797 was:
Increase the chance of the buffer write from the bufdaemon helper
context to succeed. If the locked vnode which owns the buffer to be
written is shared locked, try the non-blocking upgrade of the lock to
exclusive.
PR: kern/178997
Reported and tested by: Klaus Weber <fbsd-bugs-2013-1@unix-admin.de>
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (glebius)
atomically upgrade shared lock to exclusive. On failure, error is
returned and lock is not dropped in the process.
Tested by: pho (previous version)
No objections from: attilio
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (glebius)
negative timeout both before and after the conversion to sbintime_t.
For periodic kqueue timer, convert zero timeout into 1ms, to avoid
interrupt storm on fast event timers.
Reported and tested by: pho
Discussed with: mav
Reviewed by: davide
Sponsored by: The FreeBSD Foundation
Approved by: re (marius)
Otherwise, knote keeps a pointer to a vnode which could become invalid
any time.
Reported by: many
Tested by: Patrick Lamaiziere <patfbsd@davenulle.org>
Discussed with: jmg
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (marius)
Without these, if the timeout value passed is "large enough", the
value of the sum of it and other factors (e.g. current time as
returned by sbinuptime() or 'precision' argument) might result in a
negative number. This negative number is then passed to
eventtimers(4), which causes et_start() routine to load et_min_period
into eventtimer, making the CPU where the thread is stuck forever in
timer interrupt handler routine. This is now avoided rounding to
INT64_MAX the timeout period in case of overflow.
Reported by: kib, pho
Discussed with: kib, mav
Tested by: pho (stress2 suite, kevent7.sh scenario)
Approved by: re (kib)
seeing a stale fd_ofiles table once fd_nfiles is already updated,
resulting in OOB accesses.
Approved by: re (kib)
Sponsored by: EMC / Isilon storage division
Reported and tested by: pho
Reviewed by: benno
measurement of load caused by time-related events still using hardclock.
For example, without this change dummynet, scheduling events each hardclock
tick, was always miscounted as load of 1.
There is still aliasing with events delayed by the new precision mechanism,
but it probably can't be avoided without moving this sampling from using
callout to some lower-level code or handling it in some other special way.
Reviewed by: davide
Approved by: re (marius)
code could need to remove a kqueue from the filedesc list. Global
lock is already locked, which causes sleepable after non-sleepable
lock acquisition.
Reported and tested by: pho
Reviewed by: jmg
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Approved by: re (gjb)
context to succeed. If the locked vnode which owns the buffer to be
written is shared locked, try the non-blocking upgrade of the lock to
exclusive.
PR: kern/178997
Reported and tested by: Klaus Weber <fbsd-bugs-2013-1@unix-admin.de>
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (marius)
