]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
14 years agoAdd new memory days for Russian Federation.
osa [Thu, 20 May 2010 10:51:22 +0000 (10:51 +0000)]
Add new memory days for Russian Federation.
Fix date and description.

Obtained from: http://pda.lenta.ru/news/2010/05/19/baptism/
http://www.garant.ru/doc/busref/spr_prazdn/

Thanks to: maxim

14 years agoWith r208341 in place WARNS can be increased.
marius [Thu, 20 May 2010 09:00:11 +0000 (09:00 +0000)]
With r208341 in place WARNS can be increased.

14 years agoUpdate to a config.h created by a file 5.03 configure script. This causes
marius [Thu, 20 May 2010 08:56:50 +0000 (08:56 +0000)]
Update to a config.h created by a file 5.03 configure script. This causes
file.1 to contain the correct version number and SIZEOF_LONG_LONG to be
defined as appropriate, which is crucial for 64-bit big-endian ELF files
to be handled correctly on big-endian systems.

PR: 146387
Reviewed by: delphij
MFC after: 3 days

14 years agoWhen waiting for the busy page, do not unlock the object unless unlock
kib [Thu, 20 May 2010 08:51:01 +0000 (08:51 +0000)]
When waiting for the busy page, do not unlock the object unless unlock
cannot be avoided.

Reviewed by: alc
MFC after: 1 week

14 years agoUpgrade to 9.6.2-P2, which addresses the following;
dougb [Thu, 20 May 2010 08:15:06 +0000 (08:15 +0000)]
Upgrade to 9.6.2-P2, which addresses the following;

   Named could return SERVFAIL for negative responses
   from unsigned zones.

14 years agoo DragonFly 2.6.0 added.
maxim [Thu, 20 May 2010 08:07:25 +0000 (08:07 +0000)]
o DragonFly 2.6.0 added.

Submitted by: pluknet

14 years agoRevert r208324. We still need it for installworld as $PATH is overriden.
jkim [Thu, 20 May 2010 06:51:01 +0000 (06:51 +0000)]
Revert r208324.  We still need it for installworld as $PATH is overriden.

14 years agoRename an argument from "exp" to "expect" since the former makes FlexeLint
phk [Thu, 20 May 2010 06:18:03 +0000 (06:18 +0000)]
Rename an argument from "exp" to "expect" since the former makes FlexeLint
uneasy, in case anybody think it might be exp(3) in libm.

This also makes it consistent with other archs.

14 years agoFix some way-past-brucification complaints from FlexeLint.
phk [Thu, 20 May 2010 06:16:13 +0000 (06:16 +0000)]
Fix some way-past-brucification complaints from FlexeLint.

14 years agoAdd support to background fsck to delete zero-length directories.
mckusick [Thu, 20 May 2010 06:05:40 +0000 (06:05 +0000)]
Add support to background fsck to delete zero-length directories.

14 years agoo OpenBSD 4.7 added.
maxim [Thu, 20 May 2010 05:03:02 +0000 (05:03 +0000)]
o OpenBSD 4.7 added.

14 years agoRemove makewhatis(1) from bootstrap tools. We no longer support source
jkim [Thu, 20 May 2010 00:13:01 +0000 (00:13 +0000)]
Remove makewhatis(1) from bootstrap tools.  We no longer support source
upgrades from versions prior to 6.0 on head.

14 years agoRegen for r208321.
jkim [Thu, 20 May 2010 00:10:11 +0000 (00:10 +0000)]
Regen for r208321.

14 years agoAdd an option file for WITHOUT_MAN_UTILS to regenerate src.conf(5).
jkim [Thu, 20 May 2010 00:07:21 +0000 (00:07 +0000)]
Add an option file for WITHOUT_MAN_UTILS to regenerate src.conf(5).

14 years agoAdd a new build option, MAN_UTILS. This option lets you control building
jkim [Wed, 19 May 2010 23:56:26 +0000 (23:56 +0000)]
Add a new build option, MAN_UTILS.  This option lets you control building
utilities and related support files for manual pages, which were previously
controlled by MAN.  For POLA, the default depends on MAN, i.e., WITHOUT_MAN
implies WITHOUT_MAN_UTILS and WITH_MAN implies WITH_MAN_UTILS.  This patch
is slightly improved by me from:

PR: misc/145212

14 years agoCorrectly identify some twinax cables, which report
gallatin [Wed, 19 May 2010 20:00:15 +0000 (20:00 +0000)]
Correctly identify some twinax cables, which report
a media type of 1.

14 years agoAdd constants for the optional EOI suppression support in local APICs and
jhb [Wed, 19 May 2010 19:52:41 +0000 (19:52 +0000)]
Add constants for the optional EOI suppression support in local APICs and
EOI registers in I/O APICs.

14 years agoUpdate mxge firmware to latest available from Myricom.
gallatin [Wed, 19 May 2010 19:44:00 +0000 (19:44 +0000)]
Update mxge firmware to latest available from Myricom.

14 years agoThis change does the following for the scripts that run up through
dougb [Wed, 19 May 2010 19:03:19 +0000 (19:03 +0000)]
This change does the following for the scripts that run up through
FILESYSTEMS (the default early_late_divider):
1. Move sysctl to run first
2. Move as many BEFOREs to REQUIREs as possible.
3. Minor effect, move hostid_save from right before mdconfig to right
   after.

A lot of the early scripts make use of sysctl one way or another so
running this first makes a lot of sense given that system-critical
values are often placed in sysctl.conf.

My original purpose for working on this was that while doing some
debugging on other stuff I noticed that the order of execution was
different in the first pass through the early scripts and the second.
In practice that doesn't matter because the scripts are not executed the
second time. However this _can_ result in problems if the difference in
the rcorder moves a script from the late section to the early section in
the second pass (which would mean the script would not get executed).
So, I wanted to make the order of execution of the scripts in the early
section more deterministic.

In the course of debugging the ordering problems I noticed that moving
the BEFOREs to REQUIREs prevented the changes in order from the first
pass to the second pass without having to make any substantial changes.
(Of course it's no secret that I think BEFORE should be avoided as much
as possible, but this is a good example of why.)

Reviewed by: silence on freebsd-rc@
MFC after: 8.1-RELEASE

14 years agoFix a race between ngs_rcvmsg() and soclose() which closes the control
attilio [Wed, 19 May 2010 15:06:09 +0000 (15:06 +0000)]
Fix a race between ngs_rcvmsg() and soclose() which closes the control
socket while it is still in use.
priv->ctlsock is checked at the top of the function but without any
lock held, which means the control socket state may certainly change.
Add a similar protection to ngs_shutdown() even if a race is unlikely
to be experienced there.

Sponsored by: Sandvine Incorporated
Obtained from: Nima Misaghian @ Sandvine Incorporated
<nmisaghian at sandvine dot com>
MFC after: 10 days

14 years agoffs_mount: accept and drop userland-only options that can be passed from
avg [Wed, 19 May 2010 09:32:11 +0000 (09:32 +0000)]
ffs_mount: accept and drop userland-only options that can be passed from
loader(8)

In r193192 loader(8) has grown an ability to pass root mount options
from fstab via vfs.root.mountfrom.options.  Unfortunately, some options
that can be present in fstab are for userland only and lead to root
mounting failure when seen by kernel.
Rather than teaching loader about FFS-specific options that should be
filtered out, ffs_mount recognizes those options as valid, but ignores
and deletes[1] them.

[1] is suggested by jh.

PR: kern/141050
Reported by: many
Reviewed by: jh, bde
MFC after: 4 days

14 years agomdoc: consistently spell our email addresses <foo@FreeBSD.org>
uqs [Wed, 19 May 2010 08:57:53 +0000 (08:57 +0000)]
mdoc: consistently spell our email addresses <foo@FreeBSD.org>

Reviewed by: ru

14 years agoGarbage collect unused variable, missed in r200840
uqs [Wed, 19 May 2010 08:52:59 +0000 (08:52 +0000)]
Garbage collect unused variable, missed in r200840

Found by: clang static analyzer

14 years ago- Fix terminating argument to execl(3), sizeof(NULL) != sizeof(char *)
uqs [Wed, 19 May 2010 08:52:51 +0000 (08:52 +0000)]
- Fix terminating argument to execl(3), sizeof(NULL) != sizeof(char *)
  might be true on some systems [1]
- Rewrite gotos to use return directly
- Some spelling fixes
- Reduce unneeded/non-standard Makefile settings

Requested by: bde, imp [1]
Explanation provided by: bde [1]
Reviewed by: imp

14 years ago - Don't immediately re-run softdepflush if we didn't make any progress
jeff [Wed, 19 May 2010 06:18:01 +0000 (06:18 +0000)]
 - Don't immediately re-run softdepflush if we didn't make any progress
   on the last iteration.  This can lead to a deadlock when we have
   worklist items that cannot be immediately satisfied.

Reported by: uqs, Dimitry Andric <dimitry@andric.com>

 - Remove some unnecessary debugging code and place some other under
   SUJ_DEBUG.
 - Examine the journal state in softdep_slowdown().
 - Re-format some comments so I may more easily add flag descriptions.

14 years agoCorrect a typo.
nwhitehorn [Wed, 19 May 2010 01:37:47 +0000 (01:37 +0000)]
Correct a typo.

Pointy hat to: me

14 years agoFix our version of IPv6 address representation.
alfred [Wed, 19 May 2010 00:35:47 +0000 (00:35 +0000)]
Fix our version of IPv6 address representation.

We do not respect rules 3 and 4 in the required list:

1. omit leading zeros

2. "::" used to their maximum extent whenever possible

3. "::" used where shortens address the most

4. "::" used in the former part in case of a tie breaker

5. do not shorten one 16 bit 0 field

6. use lower case

http://tools.ietf.org/html/draft-ietf-6man-text-addr-representation-04.html

Submitted by: Kalluru Abhiram @ Juniper Networks
Obtained from: Juniper Networks
Reviewed by: hrs, dougb

14 years agoSwitch to C99 exact-width types.
marcel [Wed, 19 May 2010 00:23:10 +0000 (00:23 +0000)]
Switch to C99 exact-width types.

14 years agoA few changes:
jfv [Wed, 19 May 2010 00:03:48 +0000 (00:03 +0000)]
A few changes:
  When not defining header split do not allocate mbufs,
  this can be a BIG savings in the mbuf memory pool.

  Also keep seperate dma maps for the header and
  payload pieces when doing header split. The basis
  of this code was a patch done a while ago by
  yongari, thank you :)

  A number of white space changes.

MFC: in a few days

14 years agoRestore the previous VESA mode after BIOS POST, just in case.
jkim [Tue, 18 May 2010 22:30:55 +0000 (22:30 +0000)]
Restore the previous VESA mode after BIOS POST, just in case.

14 years agoProvide missing members for Book-E pmap (and fix build).
raj [Tue, 18 May 2010 21:23:51 +0000 (21:23 +0000)]
Provide missing members for Book-E pmap (and fix build).

14 years agoRemove unnecessary pointer increment. A wrong pointer may be passed to
jkim [Tue, 18 May 2010 18:28:17 +0000 (18:28 +0000)]
Remove unnecessary pointer increment.  A wrong pointer may be passed to
free(9) and it can cause kernel panic when there are multiple graphics
controllers in the system.

Tested by: Brandon Gooch (jamesbrandongooch at gmail dot com)
MFC after: 3 days

14 years agoAdd a sentence to the man page explaining that the -d option
bcr [Tue, 18 May 2010 18:14:12 +0000 (18:14 +0000)]
Add a sentence to the man page explaining that the -d option
can only be used when ntpd is compiled with DEBUG support.

PR:             docs/138206
Submitted by:   Oliver Pinter (oliver dot pntr at gmail dot com)
MFC after:      5 days
Approved by:    roberto

14 years agoAdd a device description.
rpaulo [Tue, 18 May 2010 17:01:07 +0000 (17:01 +0000)]
Add a device description.

14 years agoThe page queues lock is no longer required by vm_page_set_invalid(), so
alc [Tue, 18 May 2010 16:40:29 +0000 (16:40 +0000)]
The page queues lock is no longer required by vm_page_set_invalid(), so
eliminate it.

Assert that the object containing the page is locked in
vm_page_test_dirty().  Perform some style clean up while I'm here.

Reviewed by: kib

14 years agoRetry reads that fail with EINTR. This fixes a problem
kientzle [Tue, 18 May 2010 14:11:38 +0000 (14:11 +0000)]
Retry reads that fail with EINTR.  This fixes a problem
with bsdtar failing on SIGINT.

14 years agoOnly use the cache after the early stage of loading. This is
rdivacky [Tue, 18 May 2010 08:55:23 +0000 (08:55 +0000)]
Only use the cache after the early stage of loading. This is
because calling mmap() etc. may use GOT which is not set up
yet. Use calloc() instead of mmap() in cases where this
was the case before (sparc64, powerpc, arm).

Submitted by: Dimitry Andric (dimitry andric com)
Reviewed by: kan
Approved by: ed (mentor)

14 years agoAllow the experimental NFSv4 client to use cached attributes
rmacklem [Tue, 18 May 2010 05:18:21 +0000 (05:18 +0000)]
Allow the experimental NFSv4 client to use cached attributes
when a write delegation is held. Also, add a missing
mtx_unlock() call for the ACL debugging code.

MFC after: 5 days

14 years agoFix Sibyte SMP kernel breakage caused by r208249.
neel [Tue, 18 May 2010 05:12:54 +0000 (05:12 +0000)]
Fix Sibyte SMP kernel breakage caused by r208249.

We need to include the header file that provides declaration of the
smp_topo_none() function.

14 years agoAdds the file I forgot to add... that handles
rrs [Tue, 18 May 2010 04:08:58 +0000 (04:08 +0000)]
Adds the file I forgot to add... that handles
the mpwait.S for RMI

Approved by: JC

14 years agoAdds JC's cleanup patches that fix it so
rrs [Tue, 18 May 2010 04:02:34 +0000 (04:02 +0000)]
Adds JC's cleanup patches that fix it so
we call an platform dependant topo function as
well as clean up all the XLR specific ifdefs around
smp platform init.

Obtained from: JC

14 years ago - Round up the journal size to the block size so we don't confuse fsck.
jeff [Tue, 18 May 2010 01:45:28 +0000 (01:45 +0000)]
 - Round up the journal size to the block size so we don't confuse fsck.

Reported by: Mikolaj Golub <to.my.trociny@gmail.com>

 - Only require 256k of blocks per-cg when trying to allocate contiguous
   journal blocks.  The storage may not actually be contiguous but is at
   least within one cg.
 - When disabling SUJ leave SU enabled and report this to the user.  It
   is expected that users will upgrade SU filesystems to SUJ and want
   a similar downgrade path.

14 years agoAdd a sanity check for a negative args.fhsize to the experimental
rmacklem [Mon, 17 May 2010 23:55:38 +0000 (23:55 +0000)]
Add a sanity check for a negative args.fhsize to the experimental
NFS client.

MFC after: 5 days

14 years agoPrevent unloading a kld for a driver that has subinterfaces (vlan and/or
jhb [Mon, 17 May 2010 19:51:34 +0000 (19:51 +0000)]
Prevent unloading a kld for a driver that has subinterfaces (vlan and/or
wlan interfaces) from being automatically reloaded via devd shutdown
event handlers.
- Revert part of my previous changes to call ifn_stop on subinterfaces
  when an interface is detached.  It is better to destroy the interfaces
  first so that an 'ifconfig foo0.blah down' doesn't result in ifconfig
  auto-loading if_foo.ko.  The ifconfig command will not be invoked if
  foo0.blah is gone when ifn_stop() is called.  Furthermore, it is not
  necessary to explicitly invoke ifn_stop() after the subinterface is
  destroyed as devd will already do that.
- Pass -n to ifconfig when destroying interfaces so that destroying a
  cloned interface does not kldload any drivers.

Reviewed by: dougb
MFC after: 4 days

14 years agoIgnore failures from removing multicast addresses from the parent (trunk)
jhb [Mon, 17 May 2010 19:36:56 +0000 (19:36 +0000)]
Ignore failures from removing multicast addresses from the parent (trunk)
interface when tearing down a vlan interface.  If a trunk interface is
detached, all of its multicast addresses are removed before the ifnet
departure eventhandlers are invoked.  This means that all of the multicast
addresses are removed before the vlan interfaces are removed which causes
the if_delmulti() calls in the vlan teardown to fail.

In the VLAN_ARRAY case, this left vlan interfaces referencing a no longer
valid parent interface.  In the !VLAN_ARRAY case, the eventhandler gets
stuck in an infinite loop retrying vlan_unconfig_locked() forever.  In
general the callers of vlan_unconfig_locked() do not expect nor handle
failure, so I believe it is safer to ignore the errors and tear down as
much of the vlan state as possible.

Silence from: net@
MFC after: 4 days

14 years agoUse better type for siginfo (volatile sig_atomic_t instead of int).
pjd [Mon, 17 May 2010 19:13:49 +0000 (19:13 +0000)]
Use better type for siginfo (volatile sig_atomic_t instead of int).

Pointed out by: jh

14 years agoSlightly improve wording.
ivoras [Mon, 17 May 2010 11:18:33 +0000 (11:18 +0000)]
Slightly improve wording.

14 years agoRemoved ar(1)'s support for compressed archives. This change removes
kaiw [Mon, 17 May 2010 09:37:59 +0000 (09:37 +0000)]
Removed ar(1)'s support for compressed archives. This change removes
ar(1)'s dependencies on compressor libraries -lz, -lbz2 and -llzma and
fixes building HEAD on some versions of FreeBSD[78]. Option -j and -z
is now accepted but ignored.

Compressed ar(1) archives are not useful without a ld(1) that can read
them. Also, the current ar(1) compression scheme prevents random
access of archive members and needs to be redesigned anyway.

Submitted by: kientzle (original patch)
Reviewed by: delphij
Discussed on: -current mailing list

14 years agoFix build.
delphij [Mon, 17 May 2010 07:06:54 +0000 (07:06 +0000)]
Fix build.

14 years agoMake mvs_if.c to not be always linked statically into the kernel.
mav [Mon, 17 May 2010 03:51:57 +0000 (03:51 +0000)]
Make mvs_if.c to not be always linked statically into the kernel.
Link it same way as mvs.c.

14 years agoOn entry to pmap_enter(), assert that the page is busy. While I'm
alc [Sun, 16 May 2010 23:45:10 +0000 (23:45 +0000)]
On entry to pmap_enter(), assert that the page is busy.  While I'm
here, make the style of assertion used by pmap_enter() consistent
across all architectures.

On entry to pmap_remove_write(), assert that the page is neither
unmanaged nor fictitious, since we cannot remove write access to
either kind of page.

With the push down of the page queues lock, pmap_remove_write() cannot
condition its behavior on the state of the PG_WRITEABLE flag if the
page is busy.  Assert that the object containing the page is locked.
This allows us to know that the page will neither become busy nor will
PG_WRITEABLE be set on it while pmap_remove_write() is running.

Correct a long-standing bug in vm_page_cowsetup().  We cannot possibly
do copy-on-write-based zero-copy transmit on unmanaged or fictitious
pages, so don't even try.  Previously, the call to pmap_remove_write()
would have failed silently.

14 years agoTeach gpart about bootcode on APM.
nwhitehorn [Sun, 16 May 2010 22:21:33 +0000 (22:21 +0000)]
Teach gpart about bootcode on APM.

14 years agoPull OF_quiesce() out of the MI Open Firmware layer and entirely into
nwhitehorn [Sun, 16 May 2010 22:01:43 +0000 (22:01 +0000)]
Pull OF_quiesce() out of the MI Open Firmware layer and entirely into
PPC ofw_machdep.c, in recognition of its state as a machine specific hack.

Requested by: marius

14 years agoallocate ipv6 flows from the ipv6 flow zone
kmacy [Sun, 16 May 2010 21:48:39 +0000 (21:48 +0000)]
allocate ipv6 flows from the ipv6 flow zone

reported by: rrs@

MFC after: 3 days

14 years agoOn SIGINFO print current values on stderr.
pjd [Sun, 16 May 2010 21:06:26 +0000 (21:06 +0000)]
On SIGINFO print current values on stderr.

14 years agoReorganize slightly in preparation for making lzma and bz2 support conditional.
kientzle [Sun, 16 May 2010 20:43:17 +0000 (20:43 +0000)]
Reorganize slightly in preparation for making lzma and bz2 support conditional.

14 years agoIt is not necessary (and in some cases harmful) to hardcode ata_kauai's
nwhitehorn [Sun, 16 May 2010 20:31:31 +0000 (20:31 +0000)]
It is not necessary (and in some cases harmful) to hardcode ata_kauai's
IRQ to 39 on K2 devices, as well as Shasta ones.

Reported by: Andreas Tobler

14 years agoEnable smu(4) to report fan speeds on late-model Powermac G5s.
nwhitehorn [Sun, 16 May 2010 19:57:38 +0000 (19:57 +0000)]
Enable smu(4) to report fan speeds on late-model Powermac G5s.

14 years agoFix userland build by making io_task available only for the kernel and by
pjd [Sun, 16 May 2010 19:44:08 +0000 (19:44 +0000)]
Fix userland build by making io_task available only for the kernel and by
providing taskq_dispatch_safe() macro.

MFC after: 1 week

14 years agoThis pushes all of JC's patches that I have in place. I
rrs [Sun, 16 May 2010 19:43:48 +0000 (19:43 +0000)]
This pushes all of JC's patches that I have in place. I
am now able to run 32 cores ok.. but I still will hang
on buildworld with a NFS problem. I suspect I am missing
a patch for the netlogic rge driver.

JC check and see if I am missing anything except your
core-mask changes

Obtained from: JC

14 years agoCorrect an error of omission in r202897: Now that amd64 uses the direct map
alc [Sun, 16 May 2010 19:25:56 +0000 (19:25 +0000)]
Correct an error of omission in r202897: Now that amd64 uses the direct map
to access the message buffer, we must explicitly request that the underlying
physical pages are included in a crash dump.

Reported by: Benjamin Kaduk

14 years agoRelocate interrupt sense setting for K2 SATA from the ATA driver to the
nwhitehorn [Sun, 16 May 2010 17:55:09 +0000 (17:55 +0000)]
Relocate interrupt sense setting for K2 SATA from the ATA driver to the
OFW PCI layer and read the sense directly from the device tree instead
of guessing.

MFC after: 1 week

14 years agoThis adds back the Iterator to the sctp
rrs [Sun, 16 May 2010 17:03:56 +0000 (17:03 +0000)]
This adds back the Iterator to the sctp
code base. We now properly have ONE thread
that services all VNET's. Also we purge out
the old timer based iterator code which had
multiple LOR's and other issues.

MFC after: 3 days

14 years agoAdd a comment about the proper use of vm_object_page_remove().
alc [Sun, 16 May 2010 16:54:05 +0000 (16:54 +0000)]
Add a comment about the proper use of vm_object_page_remove().

MFC after: 1 week

14 years agoOn PowerMac11,2 and (presumably) PowerMac12,1, we need to quiesce the
nwhitehorn [Sun, 16 May 2010 15:56:59 +0000 (15:56 +0000)]
On PowerMac11,2 and (presumably) PowerMac12,1, we need to quiesce the
firmware in order to take over control of the SMU. Without doing this,
the firmware background process doing fan control will run amok as we
take over the system and crash the management chip.

This is limited to these two machines because our kernel is heavily
dependent on firmware accesses, and so quiescing firmware can cause
nasty problems.

14 years agoOn SMP G5 systems, sometimes the power-mode-data property is only found
nwhitehorn [Sun, 16 May 2010 15:21:13 +0000 (15:21 +0000)]
On SMP G5 systems, sometimes the power-mode-data property is only found
on CPU 0, so look there if it is not otherwise available.

14 years agoAdd support for the U4 PCI-Express bridge chipset used in late-generation
nwhitehorn [Sun, 16 May 2010 15:18:25 +0000 (15:18 +0000)]
Add support for the U4 PCI-Express bridge chipset used in late-generation
Powermac G5 systems. MSI and several other things are not presently
supported.

The U3/U4 internal device support portions of this change were contributed
by Andreas Tobler.

MFC after: 1 week

14 years agoAllow to configure UMA usage for ZIO data via loader and turn it on by
pjd [Sun, 16 May 2010 15:14:59 +0000 (15:14 +0000)]
Allow to configure UMA usage for ZIO data via loader and turn it on by
default for amd64. On i386 I saw performance degradation when UMA was used,
but for amd64 it should help.

MFC after: 3 days

14 years agoAdd task structure to zio and use it instead of allocating one.
pjd [Sun, 16 May 2010 15:12:34 +0000 (15:12 +0000)]
Add task structure to zio and use it instead of allocating one.
This eliminates the only place where we can sleep when calling zio_interrupt().
As a side-effect this can actually improve performance a little as we
allocate one less thing for every I/O.

Prodded by: kib
MFC after: 1 week

14 years agoAdd the ofw_bus_if.h dependency introduced with r207585.
marius [Sun, 16 May 2010 12:39:17 +0000 (12:39 +0000)]
Add the ofw_bus_if.h dependency introduced with r207585.

14 years agoThe whole point of having dedicated worker thread for each leaf VDEV was to
pjd [Sun, 16 May 2010 11:56:42 +0000 (11:56 +0000)]
The whole point of having dedicated worker thread for each leaf VDEV was to
avoid calling zio_interrupt() from geom_up thread context. It turns out that
when provider is forcibly removed from the system and we kill worker thread
there can still be some ZIOs pending. To complete pending ZIOs when there is
no worker thread anymore we still have to call zio_interrupt() from geom_up
context. To avoid this race just remove use of worker threads altogether.
This should be more or less fine, because I also thought that zio_interrupt()
does more work, but it only makes small UMA allocation with M_WAITOK.
It also saves one context switch per I/O request.

PR: kern/145339
Reported by: Alex Bakhtin <Alex.Bakhtin@gmail.com>
MFC after: 1 week

14 years agoUse setresuid/setresgid to drop privileges
kevlo [Sun, 16 May 2010 08:03:24 +0000 (08:03 +0000)]
Use setresuid/setresgid to drop privileges

14 years agoFix deadlock between zfs_dirent_lock and zfs_rmdir
mm [Sun, 16 May 2010 07:46:03 +0000 (07:46 +0000)]
Fix deadlock between zfs_dirent_lock and zfs_rmdir

OpenSolaris onnv revision: 11321:506b7043a14c

Approved by: pjd, delphij (mentor)
Obtained from: OpenSolaris (Bug ID 6847615)
MFC after: 3 days

14 years agoFix perfomance problem with ZFS prefetch caching [1]
mm [Sun, 16 May 2010 07:16:28 +0000 (07:16 +0000)]
Fix perfomance problem with ZFS prefetch caching [1]
Add statistics for ZFS prefetch (sysctl kstat.zfs.misc.zfetchstats)

Partial import of OpenSolaris onnv revision 10474:0e96dd3b905a

Reported by: jhell@dataix.net (private e-mail) [1]
Approved by: pjd, delphij (mentor)
Obtained from: OpenSolaris (Bug ID 68599976868951)
MFC after: 3 days

14 years agoCorrect compilation error introduced in last commit.
mjacob [Sun, 16 May 2010 06:40:05 +0000 (06:40 +0000)]
Correct compilation error introduced in last commit.

X-MFC: 208119
MFC after:      1 week
Sponsored By:   Panasas
Pointy Hat to: Me
Noticed by: Rob

14 years agoDisable bypass for the vop_advlockpurge(). The vop is called after
kib [Sun, 16 May 2010 05:00:29 +0000 (05:00 +0000)]
Disable bypass for the vop_advlockpurge(). The vop is called after
vop_revoke(), the v_data is already destroyed.

Reported and tested by: ed

14 years ago- fixes a bug that it didn't initialize the ratectl after association;
weongyo [Sat, 15 May 2010 21:18:15 +0000 (21:18 +0000)]
- fixes a bug that it didn't initialize the ratectl after association;
  so ni_txrate returned 0 which is a invalid result.
- The fourth argument of ieee80211_ratectl_tx_complete() could be not
  NULL.

Reported by: Gustau P?rez <gperez at entel.upc.edu>
Tested by: Gustau P?rez <gperez at entel.upc.edu>,
Ian FREISLICH <ianf at clue.co.za>
MFC after: 3 days

14 years agoWhap. Hook up some wires that were forgotten a few months ago and restore
mjacob [Sat, 15 May 2010 20:26:10 +0000 (20:26 +0000)]
Whap. Hook up some wires that were forgotten a few months ago and restore
the zombie device timeout code and the loop down time code and the fabric
hysteresis code.
MFC after: 1 week
Sponsored By: Panasas

14 years agoFix a mismerge in r206001.
marius [Sat, 15 May 2010 19:46:16 +0000 (19:46 +0000)]
Fix a mismerge in r206001.

PR: 146614
Approved by: jfv (implicit)
MFC afer: 3 days

14 years agoChange the commented msgs examples in profile/csh.login from -f to -q.
jilles [Sat, 15 May 2010 17:49:56 +0000 (17:49 +0000)]
Change the commented msgs examples in profile/csh.login from -f to -q.

Starting something that wants input on login seems strange and can be
dangerous. In some configurations, causing output can be bad, but it is not
as dangerous.

I do not expect this msgs invocation to be uncommented often.

PR: conf/96015
MFC after: 4 days

14 years agoApply a patch that has been lingering in my inbox for far too long:
phk [Sat, 15 May 2010 10:31:11 +0000 (10:31 +0000)]
Apply a patch that has been lingering in my inbox for far too long:

On a soekris Net5501, if you do a watchdog -t 16, followed by a watchdog
-t 0 to disable the watchdog, and then after some time (16s) re-enable
the watchdog the box reboots immediatly. This prevents also to stop and
restart watchdogd(8).

This is because when you stop the watchdog, the timer is not stoped,
only the hard reset is disabled. So when the timer has elapsed, the C2
event of the timer is set.

But when the hard reset is re-enabled, the event is not cleared and the
box reboots.

The attached patch stops and resets the counter when the watchdog is
disabled and do not disable the hard reset of the timer (if the timer
has elapsed it's too late).

Submitted by:  Patrick Lamaizière

14 years agoDo not attempt to render a logrecord with length byte, until we have
phk [Sat, 15 May 2010 08:57:16 +0000 (08:57 +0000)]
Do not attempt to render a logrecord with length byte, until we have
decompressed all the bytes required.

14 years agoSmall changes preparing for MFC, need to conditionalize
jfv [Fri, 14 May 2010 22:18:34 +0000 (22:18 +0000)]
Small changes preparing for MFC, need to conditionalize
the buf_ring_free call, and lem is missing the WOL change
put into em.

14 years ago A few minor fixes:
jfv [Fri, 14 May 2010 22:00:37 +0000 (22:00 +0000)]
 A few minor fixes:
   - add a moderation value to the Link vector
   - allow disabling HW RSC on the 82599 if LRO
     is not enabled.
   - correct error in the stats code
   - change optic type on the 82598 DA device

Thanks to Andrew Boyer for the changes.

14 years agoYet another potential dereference of a dead provider.
mjacob [Fri, 14 May 2010 21:27:39 +0000 (21:27 +0000)]
Yet another potential dereference of a dead provider.

Sponsored by:   Panasas
MFC after: 1 week

14 years agoFix an issue with the dynamic pcpu/vnet data allocators.
bz [Fri, 14 May 2010 21:11:58 +0000 (21:11 +0000)]
Fix an issue with the dynamic pcpu/vnet data allocators.

We cannot expect that modspace is the last entry in the linker
set and thus that modspace + possible extra space up to PAGE_SIZE
would be contiguous.  For the moment do not support more than
*_MODMIN space and ignore the extra space (*).

(*) We know how to get it back but it'll need testing.

Discussed with: jeff, rwatson (briefly)
Reviewed by: jeff
Sponsored by: The FreeBSD Foundation
Sponsored by: CK Software GmbH
MFC after: 4 days

14 years ago - Add versioned symbols to liblzma
delphij [Fri, 14 May 2010 20:40:37 +0000 (20:40 +0000)]
 - Add versioned symbols to liblzma
 - Use default SHLIB_MAJOR.

Approved by: mm
X-MFC with: lzma library MFC

14 years agoDocument the led(4) interface to the identification LEDs.
marius [Fri, 14 May 2010 20:11:30 +0000 (20:11 +0000)]
Document the led(4) interface to the identification LEDs.

MFC after: 3 days

14 years ago- Enable DMA write parity error interrupts on Schizo with a working
marius [Fri, 14 May 2010 20:00:21 +0000 (20:00 +0000)]
- Enable DMA write parity error interrupts on Schizo with a working
  implementation.
- Revert the Sun Fire V890 WAR of r205254. Instead let schizo_pci_bus()
  only panic in case of fatal errors as the interrupt triggered by the
  error the firmware of these and also Sun Fire 280R with version 7
  Schizo caused may happen as late as using the HBA and not only prior
  to touching the PCI bus (in the former case the actual error still is
  fatal but we clear it before touching the PCI bus).
  While at it count and export non-fatal error interrupts via sysctl(9).
- Remove unnecessary locking from schizo_ue().

14 years agoRemove duplicate
dougb [Fri, 14 May 2010 19:14:59 +0000 (19:14 +0000)]
Remove duplicate

14 years agoHide the creation and population of the temproot
dougb [Fri, 14 May 2010 19:12:06 +0000 (19:12 +0000)]
Hide the creation and population of the temproot

14 years agoIf controller received bad frames make sure to update newly added
yongari [Fri, 14 May 2010 17:39:28 +0000 (17:39 +0000)]
If controller received bad frames make sure to update newly added
RFA. Also drop frames that have either CRC error or alignment
error. Normally bad frames are not received at all. But controllers
running in promiscuous mode will receive bad frames. 82557 will
also receive bad frames to receive VLAN oversized frames.

While I'm here mark RNR condition if driver happen to see RNR in
RFA status and restart RU to receive frames again. Because driver
checks all received frames in RX loop, RNR condition could be set
in the middle of RX processing. Just relying on RNR interrupt was
not enough.

This change fixes "Memory modified after free" issue when fxp(4)
is running as a member of if_bridge(4).

Tested by: Larry Baird <lab <> gta dot com>
MFC after: 5 days

14 years agoDont' allow dma map load deferring. fxp(4) is not able to handle
yongari [Fri, 14 May 2010 16:58:37 +0000 (16:58 +0000)]
Dont' allow dma map load deferring. fxp(4) is not able to handle
EINPROGRESS.

14 years agoMake sure to check that the active provider pointer points to something before
mjacob [Fri, 14 May 2010 16:56:18 +0000 (16:56 +0000)]
Make sure to check that the active provider pointer points to something before
dereferencing the pointer.

Sponsored by:   Pansas
MFC after: 1 week

14 years agoController updates RFA via DMA so driver needs synchronization.
yongari [Fri, 14 May 2010 16:55:13 +0000 (16:55 +0000)]
Controller updates RFA via DMA so driver needs synchronization.
Add missing BUS_DMASYNC_POSTWRITE and BUS_DMASYNC_PREREAD.

14 years agoDocument the 'short preamble' capability for 802.11bg.
bz [Fri, 14 May 2010 15:06:12 +0000 (15:06 +0000)]
Document the 'short preamble' capability for 802.11bg.

Reviewed by: sam
MFC after: 4 days

14 years agoSync apmd(8) with DragonflyBSD, bringing WARNS to 3
uqs [Fri, 14 May 2010 14:26:56 +0000 (14:26 +0000)]
Sync apmd(8) with DragonflyBSD, bringing WARNS to 3

Reviewed by: ed (partial, long time ago)

14 years agofsirand(8): make WARNS=3 clean
uqs [Fri, 14 May 2010 14:26:49 +0000 (14:26 +0000)]
fsirand(8): make WARNS=3 clean

- Drop bogus quad_t cast for di_gen, it is a 32bit type
- Print di_gen with leading zeros, to get consistent output
  Before this change, amd64 would print:

  ino 18 gen 616ca2bd
  ino 19 gen ffffffff95c2a3ff
  ino 20 gen 25c3a3d5
  ino 21 gen 8dc1472
  ino 22 gen 3797056b
  ino 23 gen 1d47853a
  ino 24 gen ffffffff82d26995

  After the change

  ino 18 gen 616ca2bd
  ino 19 gen 95c2a3ff
  ino 20 gen 25c3a3d5
  ino 21 gen 08dc1472
  ino 22 gen 3797056b
  ino 23 gen 1d47853a
  ino 24 gen 82d26995

PR: bin/139994 (sort of)
Reviewed by: mckusick

14 years agoRemove trailing white space. No functional changes.
dougb [Fri, 14 May 2010 04:53:57 +0000 (04:53 +0000)]
Remove trailing white space. No functional changes.

14 years agoWording fixes.
brueffer [Fri, 14 May 2010 01:45:05 +0000 (01:45 +0000)]
Wording fixes.

PR: 143454
Submitted by: Warren Block <wblock@wonkity.com>
MFC after: 3 days